305 A Obstructions

1150

305 (IX.1 1) Obstructions

A. History The theory of obstructions aims at measuring the extensibility of mappings by means of algebraic tools. Such classical results as the +Brouwer mapping theorem and Hopf’s extension and tclassification theorems in homotopy theory might be regarded as the origins of this theory. A systematic study of the theory was initiated by S. Eilenberg [l] in connection with the notions of thomotopy and tcohomology groups, which were introduced at the same time. A. Komatu and P. Olum [L] extended the theory to mappings into spaces not necessarily +n-simple. For mappings of polyhedra into certain special spaces, the +homotopy classification problem, closely related to the theory of obstructions, was solved in the following cases (K” denotes an m-dimensional polyhedron): K”+‘+S” (N. Steenrod [SI), Knt2 4s” (J. Adem), Kntk*Y, where ni( Y)=0 for i <n and n < i < n + k (M. Nakaoka). There are similar results by L. S. Pontryagin, M. Postnikov, and S. Eilenberg and S. MacLane. Except for the special cases already noted, it is extremely difflcult to discuss higher obstructions in general since they involve many complexities. Nevertheless, it is significant that the idea of obstructions has given rise to various important notions in modern algebraic topology, including cohomology operations (- 64 Cohomology Operations) and characteristic classes (- 56 Characteristic Classes). The notion of obstruction is also very useful in the treatment of cross sections of fiber bundles (- 147 Fiber Bundles), tdiffeomorphisms of differentiable manifolds, etc.

B. General

Theory for an n-Simple

Space Y

The question of whether two (continuous) mappings of a topological space X into another space Y are +homotopic to each other cari be reduced to the extensibility of the given mapping: (X x {O))U(X x {l})+ Y to a mapping of the product X x 1 of X and the unit interval I= [0, 11 into Y. Therefore the problem of classifying mappings cari be treated in the same way as that of the extension of mappings. Let K be a tpolyhedron, L a subpolyhedron of K, and R” = LU K” the union of L and the +n-skeleton K” of K. Let Y be an tarcwise connected n-simple space, and ,f’ be a mapping of L into Y. Denote by O”(f’) the set of mappings of I?” into Y that are extensions off’,

and by @“(jr) the set of thomotopy classes of mappings in %“(f’) relative to L. The set &O(f’) consists of a single element ‘because of the arcwise connectedness of Y, @(f’) is nonempty, and @‘(f’) (na 2) may be empty. Let ,f” be an element of @“(f’). If we consider the restriction off” to the boundary Oni1 of an oriented (n + l)-cell (r”+’ of K, then f”: $‘+l -t Y determines an element c(f”, on+‘) of the thomotopy group 7~,,(Y) (- 202 Homotopy Theory). This element gives a measure of obstruction for extending f” to the interior of C?+I. We obtain an (n+ 1)-tcocycle c”+‘(f”) of the tsimplicial pair (K, L) with coefficients in n,J Y), called the obstruction cocycle off”, by assigning c(fn,a”+‘) to each (n+ l)-ce11 a”+‘. This obstruction cocycle c”“(f”) is the measure of obstruction for extending f” to R”“. A necessary and sufflcient condition for the extensibility is given by c”“(f”) ==0. Clearly, c”+l (j”“) is uniquely determined for each element f of @“(f’). The set of a11 c”+‘(f”) with S”E@“(~‘) forms a subset o”+‘(f’) of the group of cocycles Z”‘l(K, L; n,(Y)). @“‘l(f) is nonempty if and only if o”+l(,f’) contains the zero element 0. LetKn=KxI,Lu=(KxO)L(LxZ)U (K x 1). Given two mappings SO, fi : K-t Y satisfying f. 1L =f, 1L, we cari defïne a natural mapping F’ : La -) Y such that an element F” of @“(F’) corresponds to a thomotopy h”-’ relative to L connecting f. ( K”-’ with fi 1K”-‘. Given an element F”E@“(F’), we have the element P’(F”) of Z”“(Km, L”; n,(Y)), which we identify with Z”(K, L; n,(Y)) through the natural isomorphism of chain groups of the pair (Ko, L”) to those of the pair (K, L). Thus we cari regard c”+‘(F”) as an element of Z”(K, L; 7c”(Y)), which is denoted by d”(f,, h”-‘,f,), and cal1 it the separation (or difference) cocycle. If ,jo ( R”-l =SI ) i?‘, we have the canonical mapping F”: L” U (Ko) * Y, and the separation cocycle is denoted simply by d”(fo,fi). The set of :separation cocycles corresponding to elements of @“(Fr) is considered to be a subset of Z”(K, L; n,( Y)) and is denoted by o”(.fO,fi). A necessary and sufflcient condition for h”-’ to be extensible to a homotopy on l?” is d”(f,,h”-‘,f,)=O. Therefore a necessary and suffccient condition for f0 1R” =fi ( R” (rel L) (i.e., relative to L) is O~o”(f,,f,). Givenf;,f;:K”-tYwithfO[L= .fTIL, then d”(&‘,h”-‘,f;) (~o’(f;,f;)) is an element of Z”(K”, L; 7c”(Y)): which is also considered to be a cochain of the pair (K, L). In this sense, we ca11 @(SO, h” -‘, ,f,“) the separation (or deformation) cochain over (K, L). The coboundary of the separation cochain d”(fl, h”-‘,,f;) coincides (except possibly for sign) with c”“(&‘)-c”“(f,“). For a tïxed fo E a>“( f’), any n-cochain dn

1151

305 c Obstructions general, if On( fo, fi) is nonempty, it is a coset of H”(K, L; x,(Y)) factored by the subgroup O”(f,, fo). Combined with the existence theorem on separation cochains, this cari be utilized to show the following theorem. Assume that O”(f ‘) is nonempty. The set of a11 elements @‘(f ‘) that are extensions of an element of @-l (f ‘) is put in one-to-one correspondence with the quotient group of Hn(I?‘, L; A,( Y)) modulo On( fo, fô) by pairing the obstruction On( fô, f “) with each f” for a fixed fô. Among such elements of @“( f ‘), the set off” that are extensible to @+’ is in oneto-one correspondence with the quotient group of H”(R”+1 ,L; ~n(Y))=H”(KL; G(Y)) modulo the subgroup On( fô”, f$+‘), assuming that fo is extended to fo”+’ (fkst classification theorem).

of the pair (K, L) with coefficients in n,(Y) is expressible as a separation cochain d”= d”(fl,f;) wheref/EQ”(f’) is a suitable mapping such that fil R”-’ =fr” 1R”-’ (existence theorem). Therefore if we take an element f”-’ of @-I(f’) whose obstruction cocycle c”(f”-‘) is zero, the set of a11 obstruction cocycles c”“(J”) of a11 such ~“E@“(S’) that are extensions off”-’ forms a subset of O”“(f’) and coincides with a coset of Z”+‘(K, L; n,(Y)) factored by B”+‘(K, L; rcL,( Y)). Thus a cohomology class ?+r(f”-~)EH”+~(K, L; rrn( Y)) corresponds to an f”-’ E Qn-l (f’) such that c”(f”-‘) = 0, and ?“+i(f”-‘) = 0 is a necessary and sufftcient condition for f”-’ to be extensible to I?“” (first extension theorem). For the separation cocycle, d”(f,, h”-‘,fJ~ H”(K, L;a,(Y)) corresponds to each homotopy h”-’ on I?n-Z such that d”-‘(f,, h”-‘,f,)=O, and 6”( f& h”-‘, fi) = 0 is a necessary and sufficient condition for h”-’ to be extensible to a homotopy on R” (tirst homotopy theorem). The subset of H”+‘(K, L, K,( Y)) corresponding to on+’ (f ‘) is denoted by On+’ (f ‘) and is called the obstruction to an (n + 1)dimensional extension off ‘. Similarly, the subset On( fo, fi) of H”(K, L, n,( Y)) corresponding to o”( fo, fi) is called the obstruction to an ndimensional homotopy connecting f. with fi. Clearly, a condition for f' to be extensible to Rn+’ is given by 0 E O”+l (f ‘), and a necessary and sufftcient condition for f. 1K” = fi 1K” (rel L) is given by OeO”(fo, fi). A continuous mapping <p: (K’, L’)+(K, L) induces homomorphisms of cohomology groups ‘p*: H”+‘(K, L; n,( Y))+H”+‘(K’, L’, n,(Y)), H”(K, L; w,( Y))+H”(K’, L’; n,(Y)). Then for f’:LtY, O”“(f’ocp)~rp*O”+‘(f’), and for f,, fi : K + Y such that f. 1L = fi ) L, On( f, o cp,f, o cp)~ p*O”( fo, fi). Therefore we also lïnd that the obstruction to an extension and the obstruction to a homotopy are independent of the choice of subdivisions of K, L, and consequently are topological invariants. Let fo, fi, and fi be mappings K-* Y such that f. 1L = fi 1L = f, 1L. Given homotopies h~;l:f,)~“-l~fi)Rn-l(relL),h;;l:fl)~-l~ f2 1R”-’ (rel L), then for the composite h”,;’ = h;;’ o ht;‘, we have

C. Primary

Obstructions

and for the inverse homotopy f. 1R”-’ of h”,;‘, clearly d”(f,,h;o’,fo)= -d”(fo,h&‘,f,).

h;;’ : f, 1Z?‘-l T

**Therefore O”(fO,fO) forms a subgroup of
**

H”(K, L, nnn(Y)) that is determined by the homotopy class off0 1em1 relative to L. In

Assume that H’+‘(K, L; ni( Y))= H’(K, L; ni( Y)) =O,whereO<i<p(e.g.,rq(Y)=O,O<i<p).In this case, by consecutive use of the lïrst extension theorem and the lïrst homotopy theorem, we cari show that each @(f ‘) (i <p) consists of a single element and OP+‘( f ‘) also consists of a single element Pc1 (f ‘) E HP+‘(K, L; xP( Y)). The element CP+’ (f ‘), called the primary obstruction off ‘, vanishes if and only if f’ cari be extended to RP+1 (second extension theorem). When Hi+’ (K, L; ai(Y)) = 0 for i > p (for example, when ai(Y) = 0 for p ci < dim(K -L)), f’ is extendable to K if and only if the first obstruction off’ vanishes (third extension theorem). Correspondingly, if H’(K, L; zi( Y))= Hi-’ (K, L; rci( Y)) = 0 (0 < i < p), then for any two mappings fo, fi : K-+ Y, f. (L = fi 1L, Op(fo, fi) consists of a single element dP(fo, fi) E HP(K, L; rcp( Y)), which we cal1 the primary difference off0 and fi. This element vanishes if and only if f0 1l@’ = fi 1&’ (rel L) (second homotopy theorem). Moreover, when H’(K, L; rci( Y)) = 0 (i > p), the primary difference is zero if and only if f. E fi (rel L) (third homotopy theorem). Assume that the hypotheses of the second extension theorem and second homotopy theorem are satistïed. If we assign to each element f P of @‘(f ‘) the primary difference of f P and the tïxed element f{, then Gi”( f ‘) is in one-to-one correspondence with HP(RP, L; rcp( Y)) by the lïrst classification theorem (second classification theorem). Similarly, assume that the hypotheses of the third extension theorem and third homotopy theorem are satisfîed. Iff,:K+ Y, f’=& 1L, then homotopy classes relative to L of extensions f off’ are put in one-to-one correspondence with the

305 D Obstructions elements of HP(K, L; xp( Y)) by pairing with ,f (tbird classification theorem). dp(,f;fO)

1152

D. Secondary

Obstructions

For simplicity, assume that ni(Y) = 0 (i< p and p < i < q). If the primary obstruction CP+I (f’)~ H P+I (K, L; 7~,,(Y)) off’: L-t Y vanishes, we cari detïne 04+’ (,f’) c H4+’ (K, L; 7rIq(Y)), which we cal1 the secondary obstruction off’. When Y = SP, q = p + 1, p > 2, the secondary obstruction Op”(f’) coincides with a coset of HpfZ(K, L; Z,) modulo the subgroup Sq2(HP(K, L; Z)), where Sq’ denotes the +Steenrod square operation [S]. In this case, if L = KP, then Op”(,f’) reduces to a cohomology class, Sq’(i*)-‘f’*(o) with i: L+K, where o is a generator of HP(SP, Z) (in this case (i*)m’f’*(a) # 0 is equivalent to ?“(f’)=O) [S]. Moreover, if Sq2f’*(a) = 0, then there exists a suitable extension fp+2:lZp+2 + Y = SP of ,f’. The set of obstruction cocycles of all such fp” defïnes the tertiary obstruction Op+3( f ‘), which coincides wit h a coset of Hp+3(K, L; Z,) modulo the subgroup SqZ(HP”(K, L; Z,)). By using the tsecondary cohomology operation 0 of J. Adem, it cari be expressed as @((i*)-‘f’*(o)) (- 64 Cohomology Operations). All the propositions in this article remain true if we take +CW complexes instead of polyhedra K.

ferential equations by reducing the operations of differentiation and integration into algebrait ones in a symbolic manner. The idea was initiated by P. S. Laplace in his Théorie analytique des probabilités (18 12), but the method has acquired popularity since 0. Heaviside used it systematically in the late 19th Century to solve electric-circuit problems. The method is therefore also callec. Heaviside calculus, but Heaviside gave only a forma1 method of calculus without bothering with rigorous arguments. The mathematical foundations were given in later years, tïrst in terms of +Laplace transforms, then by ap:$ying the theory of tdistributions. One of the motivations behind L. Schwartz’s creation of this latter theory in the 1940s was to give a sound foundation for the forma1 method, but the theory obtained has had a much larger range of applications. Schwartz’s theory was based on the newly developed theory of +topological linear spaces. On the other hand, J. Mikusinski gave another foundation, based only on elementary algebrdic notions and on Titchmarsh’s theorem, whose proof has recently been much simplifïed. In this article, we fïrst explain the simple theory established by Mikusinski [2] and later discuss its relation to the classical Laplace transform method.

References [l] S. Eilenberg, Cohomology and continuous mappings, Ann. Math., (2) 41 (1940), 231-251. [2] P. Olum, Obstructions to extensions and homotopies, Ann. Math., (2) 52 (1950) l-50. [3] P. Olum, On mappings into spaces in which certain homotopy groups vanish, Ann. Math., (2) 57 (1953) 561-574. [4] N. E. Steenrod, The topology of fibre bundles, Princeton Univ. Press, 1951. [5] N. E. Steenrod, Products of cocycles and extensions of mappings, Ann. Math., (2) 48 (1947), 290-320. [6] E. H. Spanier, Algebraic topology, McGraw-Hill. 1966.

B. The Operational

Calculus

of Mikusibski

306 (X11.20)

Operational Calculus

A. General Remarks calculus” in the usual for solving tlinear dif-

The term “operational sense means a method

The set % of all continuous complex-valued functions a = {a(t)} defïned on t 2 0 is a tlinear space with the usual addition and scalar multiplication. %?is a +Commutative algebra with multiplication a. b detïned by the tconvolution {S;u(t-s)b(s)ds}. Th e ring W has no +zero divisors (Titchmarsh’s theorem). (There have been several interesting proofs of Titchmarsh’s theorem since the tïrst demonstratilon given by Titchmarsh himself [3]. For example, a simple proof has been published by C. RyllNardzewski (1952).) Hence we cari construct the tquotient field -2 of the ring %. An element of 2 is called a Mikusitiski operator, or simply an operator. If we deiïne a(t) = 0 for t < 0 for the elements {a(t)} in ??, then V? is a subalgebra of %VI, which is the set of all locally integrable (locally L,) functions in (-a, a) uhose +support is bounded below. Here we identify two functions that coincide almost everywhere. The algebra J& has no zero divisor, and its quotient tïeld is also 2. The unity element for multiplication in %, denoted by 6 = b/b (b #O), plays the role of the +Dirac &function. It is sometimes called the impulse function. The operator 1= { 1) ~‘6 is the

1153

306 C Operational sented by

Calculus

function that takes the values 0 and 1 according as t < 0 or t > 0. This operator is Heaviside’s function and is sometimes called the unit function. Usually it is denoted by l(t) or simply 1. The value l(0) may be arbitrary, but usually it is detïned as 1/2, the mean of the limit values from both sides. The operator I is an integral operator, because, as an operator carrying a into I. a, it yields t l.a=

a(s)ds

m

+C

M(ni)

of "

i=1Zj7Je

= the integral 1

of a over [0, t].

is 0

where we assume that Âo, 1,). . . , i, exhaust the roots of the equation o(L) = 0, 1, is a multiple root of degree 1, and a11 other roots are simple (WI= n - 1). The above formula is called the expansion theorem.

More generally, the operator {t”-‘/I(n)} (ReÂ. > 0) gives the Âth-order integral. The operator s = 611, which is the inverse operator of 1, is a differential operator. If a E +Z is of tclass C’ , then we have s~a=a’+a(+0)6=a’+{a(+O)}/{l}. Similarly, if UEW is of class C”, we have

C. Limits

of Operators

**S”.a=a’“)+a’“-“(0)6+a(“-2)(0)s
**

+ . +a(O)s”-1.

(2)

The operator U+S. a cari be applied to functions a that are not differentiable in the ordinary sense, and considering the application of s to be the operation of differentiation, we cari treat the differential operator algebraically in the tïeld 02. In particular, we have s. 1 = 6, and this relation is frequently represented by the formula dl(t)/dt=S(t). A rational function of s whose numerator of lower degree than its denominator is an telementary function of t. For example, we have the relations l/(s-~)“={t”-‘ea’/(n-l)!}, (3) is

A sequence a, of operators is said to converge to the limit a = b/q if there exists an operator q( # 0) such that q. a, EV and the sequence of functions q. a, converges tuniformly to b on compact sets. The limit a is determined uniquely without depending on the operator q. Based on this notion of limits of operators, we cari construct the theory of series of operators and differential and integral calculus of functions of an independent variable i whose values are operators. They are completely parallel to the usual theories of elementary calculus (- 106 Differential Calculus; 216 Integral Calculus; 379 Series). A linear partial differential equation in the function V(X, t) of two variables, in particular its initial value problem, reduces to a linear ordinary differential equation of an operational-valued function of an independent variable x. For a given operator w, the solution (if it exists) of the differential equation v’(1) = w. <p(Â) with the initial condition C~(O)= 6 is unique, is called the exponential function of an operator w, and is denoted by ~(1) = e”“. If the power series j. I”w”/n!

s/(s2 +/32) = {cos~t}.

(4)

(6)

The solution of an ordinary linear differential equation with constant coefficients C:=,, a,<p(‘)(t) =f(t) under the tinitial condition C~“)(O) = yi (0 < i <n - 1) is thus reduced to an equation in s by using formulas (1) and (2), and is computed by decomposing the following operator into partial fractions: (5)

converges, the limit is identical to the exponential function e”‘“. However, there are several cases in which eAWexists even when the series (6) of operators does not converge. For example, for w = - &, we have e -d= {(Â/2&P)exp( -L’/4t)}, , (7)

and for w = -s, we have e-“S=h”=s.H,(t),

**whereLi=~,+l~o+~,+2~l + . .. +w~-,-~,
**

0 < r < n - 1. The general solution is represented by (5) if we consider the constants yo, . . . ,yn-i or fio, . ,jI-i as arbitrary parameters. If the rational function in the right-hand side of (5) is M@)/L)(s) and the degree of the numerator is less than that of the denominator, then the right-hand side of (5) is explicitly repre-

(8)

where the function H,(t) takes the values 0 and 1 according as t < 1 or t > 1. H,(t) belongs to the ring @ and is called the jump function at Â. For f(t)e%, we have ~A~u(t)~=u(t-4~9 and hence we cal1 (8) the translation operator

306 D Operational

1154 Calculus changing the variable from s to p and multiplying the former transform by p.

(or shift operator). For w = -s, the series (6) does not converge, but if we apply the forma1 relation e~“s=C~o(-is)“/n! to f(t), we have a forma1 Taylor expansion

E. Relation

to Distributions

The solution of linear tdifference equations are represented by rational functions of h”. The power series C a,h” of operators always converges. This fact gives an explicit example of a representation by forma1 power series. Note that the operators em” and e-“JS play an essential role in the solution of the +wave equation

**azq -=- a2azp ax2 at2
**

and the +heat equation

For f EV?, an operator of the form h’. sk. f is identifïed with a distribution of L. Schwartz with support bounded from below. We cari identify with a Schwartz distribution the limit of a sequence f, (or a suitable equivalence class of sequences) of operators of the form h” sk .f such that f., fn+l, . are identical in the interval (-II, n). The notions of Schwartz distributions and of MikusiBski operators do not include each other, but both are generalizations of the notions of functions and their derivatives. For formulas and examples Appendix A, Table 12.

**azq ~=~~ dacp 5X2 at
**

The operator (7) converges to 6 for /2+0, and this gives a tregularization of the Dirac 6function.

References [l] 0. Heaviside, On operators in physical mathematics 1, II, Proc. Roy. Soc. London, ser. A, 52 (1893), 504-529; 54 (1893), 105-442. [2] J. Mikusihski, Operational calculus, Pergamon, 1959. (Original in Polish, 1953.) [3] E. C. Titchmarsh, The zeros of certain integral functions, Proc. London Math. Soc., 25 (1926), 283-302. [4] K. Yosida, Operational calculus: A theory of hyperfunctions, Springer, 1984. (9)

D. Laplace

Transform {f(t)} E %?,the limit

For every function lim Pe-“‘f(l)dl= B-+m s o

m eeAsf(,l)dA s0

always exists (in the sense of the limit of operators), and as an operator coincides with the original function {f(t)}. Therefore, if the usual Laplace transform (- 240 Laplace Transform) of the function f(n) exists and (9) is a function g(s), then as a function of the differential operator s, g(s) is the operator that is given by the inverse Laplace transform ,f(t) of g(s). Formulas (4) and (7) are indeed typical examples of this relation, where the left-hand side is the usual Laplace transform of the right-hand side. In the practical computation of (5), we cari compute the Laplace inverse transform of the right-hand side. However, if we took the Laplace transform as the foundation of the theory, it would not only be complicated but also be subject to the artifïcial restriction caused by the convergence condition on Laplace transforms. In the theory of operational calculus, the transform m

dP)=P

**307 (XIX.1 3) Operations Research
**

A. General Remarks

e-P’f(t)dt I 0

(10)

Operations research in the most general sense cari be characterized as the application of scient& methods, techniques, and tools to the operations of systems SO as to provide those in control with optimum solutions to problems. This definition is due to Churchman, Ackoff, and Arnoff [l]. Operational research began in a military context in the United Kingdom during World War II, and it was quickly taken up under the name operations research (OR) in the United States. After the war it evolved in connection with industrial organization, and its many techniques have found expanding areas of application in the Uniied States, the United Kingdom, and in other industrial countries. Nowadays OR is used widely in industry for solving practical problems, such

is sometimesusedinstead of the Laplace transform itself. But the difference is not essential; we obtain the latter transform merely by

as planning, scheduling,inventory, transportation, and marketing. It also has various important applications in the fïelds ojagricul-

1155

307 c Operations

Research

ture, commerce, economics, education, public service, etc., and some techniques developed in OR have influenced other lïelds of science and technology.

B. Applications Applications of OR to practical problems are often carried out by project teams because knowledge of disparate aspects of the problems are required, and interdisciplinary cooperation is indispensable. The following are the major phases of an OR project: (i) formulating the problem, (ii) constructing a mathematical mode1 to represent the system under study, (iii) deriving a solution from the model, (iv) testing the mode1 and the solution derived from it, (v) establishing controls over the solution and putting it to work (implementation). When the mathematical mode1 that has been constructed in phase (ii) is complicated and/or the amount of data to be handled is large, a digital tcomputer is often utilized in phases (iii) and (iv).

C. Mathematical

Models

[2]

Typical mathematical models and tools that appear frequently in OR are: (1) Optimization mode1 (- 264 Mathematical Programming). This mode1 is characterized by one or more real-valued functions, which are called objective functions, to be minimized (or maximized) under some constraints. According to the number of objective functions, the types of objective functions, and the types of constraints, this mode1 is classilïed roughly as follows: (i) Single-objective model, which includes linear, quadratic, convex, nonlinear and integer programming models (- 215 Integer Programming, 255 Linear Programming, 292 Nonlinear Programming, 349 Quadratic Programming); (ii) multi-objective model; (iii) stochastic programming mode1 (- 408 Stochastic Programming); (iv) dynamic programming mode1 (- 127 Dynamic Programming); (v) network flow mode1 (- 281 Network Flow Problems). (2) Game-theoretic mode1 (- 173 Game Theory). Game theory is a powerful tool for deriving a solution to practical problems in which more than one person is involved, with each player having different objectives. (3) Inventory mode1 (- 227 Inventory Control). It is necessary for most lïrms to control stocks of resources, products, etc., in order to carry out their activities smoothly; various inventory models have been developed for such problems. Mathematically, optimization techniques (- 264 Mathematical Program-

ming), Markovian decision processes (- 127 Dynamic Programming, 261 Markov Processes), and basic tprobability theory are used to construct models for these problems. (4) Queuing mode1 (- 260 Markov Chains H). In a telephone system, calls made when a11 the lines of the system are busy are lost. The problem of computing the probability of loss involved was lïrst solved in the pioneering article on tqueuing theory by A. K. Erlang in 1917. For systems in which calls cari be put on hold when a11lines are busy, one deals with the twaiting time distribution instead of the probability of 10s~. In the 1930s F. Pollaczeck and A. Ya. Khinchin gave explicit formulas for the characteristic function of the waiting time distribution. In many situations, such as in telephone systems, air and surface traffic, production lines, and computer systems, various congestion phenomena are often observed, and many kinds of queuing models are utilized to analyze the congestion. Mathematically, almost a11 such models are formulated by using Markov processes. For practical uses, approximation and computational methods are important as well as theoretical results. (5) Scheduling mode1 (- 376 Scheduling and Production Planning). Network scheduling is used to schedule complicated projects (for example, construction of buildings) that consist of a large number of jobs related to each other in some natural order. PERT (program evaluation and review technique) and CPM (critical path method) are popular computational methods for this mode1 (- 281 Network Flow Problems). Job shop scheduling is used when we have m jobs and n machines and each job requires some of the machines in a given order. The mode1 allows us to lïnd an optimal order (in some certain sense) of jobs to be implemented on each machine. (6) Replacement model. There are two typical cases. One is the preventive maintenance model, which is suitable when replacements are done under a routine policy because a replacement or a repair before a failure is more effective than after a failure. Probabilistic treatments are mainly used, and this mode1 resembles those for queues and tMarkov processes. The other is a mode1 for deciding whether to replace a piece of equipment in use. In this case, we need to compare costs of both used and new equipment, and evaluations of various types of present cost are important. (7) Simulation. This is a numerical experiment in a simulated mode1 of a phenomenon which we want to analyze. Simulation is one of the most popular techniques in OR. (8) Other models. Besides the models listed above, many problems are formulated by way of various other models in OR. In modeling,

307 Ref. Operations

1156 Research weak operator topology, but not with respect to the strong operator topology. The operation from 2 x J to &? of taking the product, (A, B)-tAB, is continuous with respect to the uniform operator topology, is continuous with respect to the strong operator topology when the first factor is restricted to a set bounded in the operator norm, but is not contmuous on .-A x %. It is continuous with respect to the weak operator topology when one of the factors is iïxed (i.e., it is separately continuous). The set D is a TBanach space with respect to the operator norm, or, more precisely, a +C*algebra. It is a +locally convex topological linear space with respect to the strong or weak operator topology. The Banach space g is the +dual of the Banach space ~4, of all tnuclear operators in S, (- 68 Compact and Nuclear Operators 1). The weak* topology in 33 as the dual of %, is called the <T-weak topology.

tprobability theory, and mathematical tstatistics, especially, tMarkov chain, tmultivariate analysis, tdesign of experiments, tregression analysis, ttime series analysis, etc. often play important roles.

References [l] C. W. Churchman, R. L. Ackoff, and E. L. Arnoff, Introduction to operations research, Wiley, 1957. [2] H. M. Wagner, Princip]es of operations research, Prentice-Hall, 1975.

308 (X11.19)

Operator Algebras

A. Preliminaries Let 5 be a +Hilbert space. The set of +bounded linear operators on @ is denoted by .S(sj) = J. It contains the identity operator 1. The notions of opera.tor sum A + B, operator product AB, and +adJoint A* are defïned on it. A subalgebra of U(sj) is called an operator algebra. In this article we consider mainly von Neumann algebras. For C*-algebras - 36 Banach Algebras G-K. Any tHermitian operator A (i.e., an operator such that A = A*) has the property that (Ax, x) is always real for any XE$~. If (Ax,x)>O for any x, A is called positive, and we Write A 3 0. When Hermitian operators A and B satisfy A ~ B 3 0, we Write A > B. Thus we introduce an ordering A > B between Hermitian operators. A set {A,) of positive Hermitian operators is said to be an increasing directed set if any two of them A,, A, always have a common majorant A,, that is, A,< A, and A, < A,. If a Hermitian operator A satisfies (Ax, x) = sup(A,x, x) for such a set, it is called the supremum and is denoted by sup A,. The supremum sup A, exists if and only if the sup(A,x, x) is finite for any x, and then A, converges to A with respect to the weak and strong operator topologies.

C. Von Neumann

Algebras

B. Topologies

in .#I

Various topologies are introduced in % = .a(%): the +uniform operator topology, the +strong operator topology, and the tweak operator topology (- 251 Linear Operators). These topologies are listed above in order of decreasing îïneness. The operation in % of taking the adjoint, A-t A*, is continuous with respect to the uniform operator topology and

A subset ./H of .S is called a *-subalgebra if it is a subalgebra (i.e., A, BE.~ implies i.A+pB, ABE.~) and contains the adjoint ,4* of any AEM. The commutant .d’ of a subset .d of Z8 is the set of operators that commu1.e with both A and A* for AE.~. The commutant is a *subalgebra, and .d’ = SZZ”‘. A von Neumann algebra .,&! is a =-subalgebra of %Ythat is delïned by one of the following four equivalent conditions: (i) ,@ is a *subalgebra of 8 containing 1, closed under the weak operator topology; (ii) & is a *subalgebra of Ua,containing 1, closcd under the strong operator topology; (iii) .&Y is the commutant of a subset of 9 (or, equivalently, .4? = &“‘); (iv) .4? is a *-subalgebra of .%Y containing Z, closed under the uniform operator topology, and, as a Banach space, coinciding with the conjugate space of some Banach space. Note that a *-subalgebra of g, closed under the uniform operator topology, is a C*-algebra. Von Neumann a1gebra.s are also called rings of operators or W*-algebras. The latter term is usually used for a C*-.algebra *-isomorphic to a von Neumann algebra in contrast to a concrete von Neumann algebra. The study of these algebras was :started by J. von Neumann in 1929. He showed the equiva1 lente of conditions (i)-(iii) (von Neumann% 1 density theorem), and established a foundation for the theory named after him [l] The notion of von Neumann algebras cari be regarded as a natural extension of the notion oFmatrix algebras in a lïnite-dimensional space, and therein lies the importance of the theory. The

0 A. A trace t on a von Neumann algebra . EkfE = Ed restricted to the subspace E$j is called the induced von Neumann algebra of & on Etij and is denoted also by dE. 0 A. the von Neumann algebra generated by A.. satisfying (A.4? is a weight satisfying t(UAU*) = t(A) for U unitary . and Traces A state <pof a C*-algebra & is a complexvalued function on ~-4 that is (1) complex linear: q(A+B)=cp(A)+cp(B).e. 1951). This is called the GNS construction after its originators 1. BE .) and sj. called its predual and denoted by J&. AJ whenever A. EdE cari be regarded as a von Neumann algebra on E4j. BE. J&‘) by ~2.. a spatial *-isomorphism. and an induction. The *-isomorphism AE &+A @ 1 E & @ 1.. with respect to the weak or strong operator topology (Kaplansky’s density tbeorem (Ann. the mapping A E J%’+ EA E &fE is a *homomorphism and is called the induction of dl ont0 AE. 5.= A). The linear span of a11 normal states of a von Neumann algebra A is a norm-closed subspace of its dual A*. Math. 1956). and AE 4. has a unique extension to a linear functional on !JJl. ‘!R. 5. and Write J?~. where E is a projection operator belonging to the tenter ZJ‘= J?! n 4’ of 4.)(f. of %m%. is constructed by detïning the inner product (q(A)..(f..M1. and when we denote the set of elements of operator norm < 1 in d (resp. a normal *-representation z. which is additive and homogeneous (q(A + B) = q(A)+rp(B) and &A)=~~I(A) for a11 A. and by completion. is an increasing net of positive elements of J%’ and A = sup A. A weigbt <p on a von Neumann algebra Jz’ is a function deiïned on the positive elements of A. then EJE = {EAE 1A Idi’} is a *-subalgebra of %? closed with respect to the weak operator topology. a representation ne (i.. For any AicB(Sji). This gives a com- plete description of a11 possible normal representations (i. and (3) normalized: 11 cp\( = 1 (equivalent to ~(1) = 1 if I E &).)) of LX~. is dense in A. has the property that the linear span !IX. The following theorem is of use in the theory of von Neumann algebras: Given a *subalgebra d of LB containing 1.. Math.. there exists a triplet (sj. z.. M. into a dense subset of $j. Q A.2). then its extension to &’ is a positive linear functional for which the triplet (!$. such that (q(B’). with Ai~Mi is denoted by A1 0 . Sakai (Pacifie J. Its kernel is of the form E&i!. there exists a Hilbert space Ç&. its closure & with respect to the weak or strong operator topology is von Neumann algebra. &ZJ. in a. B’E’%. and 1..2) is the tcompletion of their ttensor product as a complex linear space equipped with the unique tinner product satisfying Ch Q.. and . normal if q(A) = sup <p(A. States. (2) positive: q(A*A)>O for AE &. (resp. Naïmark.).(&‘)&. which we denote by the same letter <p.L&...e.g. D.(A)@) = q(AB). is the closure of K.. a bijective isometric linear) mapping U from !& to $j2 such that UAU*=nA for a11 Ae. Canonically associated with a normal semilïnite weight <p.)=(f.. 0 fi) = A.e. a *-homomorphism into B(&. with positive real or infinite values. The space $J. is called an amplification.e.. @))=cp(B*B). where q is the quotient mapping. For any positive linear functional <p on d. but since its elements operate exclusively in the closed subspace E!& we cari regard it as an algebra of operators’on E!-j. of &!. where 1 is the trivial von Neumann algebra consisting solely of complex multiples of the identity operator. q(B)) = @?*A) in the quotient of J& by its left ideal {Alq(A*A)= 0}..&. A1 is likewise the closure of ~2... If E is a projection operator in a von Neumann algebra 4. a normal *-homomorphism into some B) of a von Neumann algebra. and semihite if the left ideal ‘!Rq.)* for a11fi.+Y2and is called the tensor product of JZ!.+. there exists a unique operator in a(!& 0 !&) denoted by A. M. The tensor product 5jj. and a complex linear mapping q from ‘9&.and a vector 5.) (unique up to the unitary equivalente) of a Hilbert space &. @ A2f2 for a11fic!&. E.. If E is a projection operator in A’.f. because 4’ turns out to be the dual of A. A. is a bounded increasing net in A. a *-isomorphism n from A1 into &Y2 is called spatial if there exists a unitary (i. For two von Neumann algebras . q( 1)) is given by the GNS construction. CO= 0 and cc + a = CO).g.+. such that <p(A) = (n. The restriction of cp to positive elements of W.+.Jz’~c @ji) (i = 1.. In this sense.Lgl Og. Siegel. Any normal *-homomorphism is continuous in the strong and weak operator topologies and is a product of an amplification. For von Neumann algebras Ai c %(!&). ~(CA)=~(A) for A. Weigbts. xc. Gel’fand..(A)&.&’ and I > 0 with the convention 0. Then Z. and q&A)t@)=q(AB). In the latter case. and a *-homomorphism R is called normal if supanA. which we cal1 the reduced von Neumann algebra of &’ on ES. It is said to be faitbful if it does not vanish except for rp(O)=O.. consisting of a11 elements A~dt’ for which ‘p(A*A) is finite. is defined by n.1157 308 D Operator Algebras fourth condition of the defmition was given by S. It is not a von Neumann algebra because it does not contain I. where B. C~C. If cp is finite (i. = ~(sup. @ & of two Hilbert spaces Gi (i = 1. gis!&.) whenever A.

.. (i... Then K is a linear space. p) be a tmeasure space. AE. and III. and lïnally McDuff (Ann.. Murray and von Neumann (1943). of type II and an equivalence relation by delïning y([) as equivalent when x(c) and Thus we obtain a space of equivalence classes which we denote by $. Let K be a set of these functions having the following properties: (i) Ilx([)ll is measurable for x([)EK. Sakai (the fïfth). + E. the numerical function (x(c). Every Abelian von Neumann algebra is lïnite. (i. Since the discovery of two nonisomorphic examples of factors of type II. We consider functions x(5) on 9JI whose values are in 5j([) for each <. Given a von Neumann algebra A. Proc.. tïnite) normal trace t such that t(A)#O. Zeller-Meier (the eighth and ninth). respectively. is isomorphic to the algebra B(5) of a11 bounded operators on an n-dimensional Hilbert space sj. There also exist unique central projections E. McDuff showed that there exist countably many nonisomorphic examples of factors of type II 1. In contrast to this.308 E Operator 1158 Algebras finite).xz([)... After 1967. A von Neumann algebra is of type II if it is semitïnite and contains no Abelian projection.. D. AE. A factor of type 1. J. . The von Neumann Classification A von Neumann algebra for which a semifmite normal trace does not exist is called a purely infinite von Neumann algebra or von Neumann algebra of type III. J. $5 is a Hilbert space with the inner product (x. = 1.. t(A*A)= t(AA*) for a11 A EA).. are properly infmite. y([)) is measurable any x(LJEK. and G.. Factors A von Neumann algebra whose tenter consists exclusively of scalar multiples of the identity operator is called a factor. a von Neumann algebra & is called semifinite (resp. + E. 1963). for example.1. if it is properly infinite. (iii) there is a countable family {x1 (<). A properly infinite von Neumann algebra d is characterized by the property that . Purely infinite &? and . the set {xl([). Dixmier and E. We cal1 each function in K a measurable vector function. A nonzero projection operator E in 4 is called Abelian when &E is Abelian. E.. and AE. We cal1 & a von Neumann algebra of type 1 (or discrete) when it contains an Abelian projection E for which 1 is the only central projection P covering E (i.... ). in the tenter d of .d. Lance (the sixth and seventh).e. by F..(l).. ME. . The two decompositions cari be combined. Schwartz (1963) (the third example. 1957). and III. the condition on the corresponding . (n = CO. classification of factors has been a central problem in the theory of von Neumann algebras. Math. (3. y) = (x(i)> y(i)) 440 s . Japan Acad. there exist mutually orthogonal projections E. Ching (the fourth). .. and Es + Ei = 1. We introduce in the set of measurable vector functions x(i) with F.e.. II. } of functions in K such that for each fixed [c!IR. E. is f-mite.. II. 1970) showed the existN:nce of uncountably many nonisomorphic examples of factors of type II. are von Neumann algebras of type 1. type III. and Ei such that -AE. After the discovery of the third to ninth nonisomorphic examples of factors of type II. great progress was made in the investigation of isomorphism classes of factors. of type 11 and not fïnite).2.M and J&?@ g(sj) for any separable 5j is *isomorphic. and we have uncountably many nonisomorphic examples of factors of types II. Functional Anal. Let (‘%R. 1969) and Sakai (J. type II. (ii) if for a function y([).. (If some of the projections E are 0. Von Neumann’s reduction theory (. is properly infinite. For type III factors Section 1. E. by J.e. .Section G) reduces the study of arbitrary von Neumann algebras on a separable Hilbert space more or less to the study of factors. II. If there are no central projection operators E #O such that &E is finite. in A and for a11 positive A in A (equivalently.I there is a semifïnite (resp.) G. A is called properly infinite. then yak.} is dense in b(c).%(4j) for inlïnite 5j. finite) if for each positive Hermitian operator A (#O) in . Tbe Integral Direct Sum and Decomposition into Factors The Hilbert spaces considered in this section are a11 tseparable. W. This decomposition is unique. A fïnite von Neumann algebra is also characterized as a von Neumann algebra in which the operation of taking the adjoint is continuous with respect to the strong operator topology on bounded spheres (Sakai.. x. A von Neumann algebra of type 1t is called of type II 1 if it is fïnite and of type II. Factors are classified into types 1. E < P). with each [E~JI we associate a Hilbert space $j([).M such that E.AE is to be waived..

there exists a bounded continuous function F(z) (depending on A.) of !$. The Gel’fand representation is an example.. and consider an Abelian von Neumann algebra d on 4j. and an antiunitary involution J. Martin. Hugenholtz.#Yc &‘. deiïned on a dense subset r#I. Tomita’s theory was perfected by Takesaki [ 131. Mautner. In the mathematical foundations of statistical mechanics.Az’ of its closure defines a positive self-adjoint operator A. P. called modular automorphisms. 1.. and j.& (the group Aut A of a11 *automorphisms of the von Neumann algebra A modulo the subgroup Int A of a11 inner *-automorphisms). If. The most important ingredient of this theory lies in certain one-parameter groups of *-automorphisms of a von Neumann algebra. independent of the weight) and belongs to the tenter of Out .Section D) by the relation S. Kubo. 8. The principal results of the Tomita-Takesaki theory are (1) if x E A.) 1. B). T. IIA(c)JI is bounded. B = (kT)-‘.. Schwinger and accordingly known as the KMS condition. A(i)~(i) is also measurable. on 0 < Im z < /? holomorphic in 0 < Im z < b and such that F(t) = cp(Aa.(A)= J. then @‘(A) E A$4Aii’ E &! for a11 real t.AJ. 1950). (.. If we take as d the tenter 3 of A. Connes [ 191. then almost a11 the A(c) are type 1 factors (F. @ Ai).) Then a Hilbert space 5(i) cari be constructed SO that b is represented as the integral direct sum of a(<). BE %+. and if we take as d a maximal Abelian von Neumann algebra contained in .. Van Daele [15].. Tomita succeeded in 1967. C.+&4)=~(A*). the antilinear operator S. Operators in d are a11 decomposable and are called diagonalizable. An operator function A(c) that associates with each c E YJI a bounded linear operator A([) on sj([) is called measurable if for any measurable x(c). Some of the basic definitions and results of the Tomita-Takesaki tbeory are as follows. showing that the group of modular automorphisms (up to inner automorphisms) is intrinsic to the von Neumann algebra (i. Tomita-Takesaki Tbeory Motivated by the problem of proving the commutant theorem for tensor products (i. for every pair A. fl X.. PL)and represent d as the set of bounded measurable functions on ‘!LX. If <p is a normal semifinite faithful weight on A. Powers reported his results [ 171 on nonisomorphism of the one-parameter family . A von Neumann algebra & on $j whose elements are a11 decomposable is characterized by . N. Winnink [ 141 at about the same time that Tomita’s work appeared in 1967. and M.. is a conjugate-linear isomorphism of A onto X. A(c) transforms a function in 6 to a function in & and thus defïnes a bounded linear operator on b. (This is possible in different ways. moreover... It was a fortunate coincidence that this condition was formulated in a so-called C*-algebra approach to statistical mechanics by R. let fi be a Hilbert space. called a modular operator.EX. Ann. This operator is called the integral direct sum (or direct integral) of A(c). then almost a11 the . each one-parameter group of modular automorphisms being intrinsically associated with a faithful semifïnite normal weight of the algebra. in generalizing the theory of Hilbert algebras. M. = J. A given one-parameter group a. A weight rp on A is said to satisfy the KMS condition at fi (a real number) relative to a one-parameter group of *-automorphisms a.&‘. called modular automorphisms (see below). (A1 @ AZ)’ =A. which remained unsolved for algebras of type III up until that time. Structure Type III and Classification of Factors of At the Baton Rouge Conference in 1967. coincides with a group of modular automorphisms a: if and only if cp satistïes the KMS condition at p= -1 relative to ut. this condition characterizes equilibrium states of a physical system for a given one-parameter group of automorphisms (of a C*-algebra) describing the time-development of the system.#Y(<) are factors (Von Neumann% reduction tbeory). Haag. and (2) if AEM. n %$.L. Math. R. of &Z if. and an operator on J3 that cari be reduced to this form is called decomposable. H. The A([) yielded by the decomposition of operators A in A generate a von Neumann algebra A([) on B(l). then j. after years of effort. The original proofs of the TomitaTakesaki theory have been simplified considerably by the work of M. Rieffel [16] and A. Generally. previously developed’only for semifinite von Neumann algebras. who also showed that modular automorphisms satisfy (and are characterized by) a condition originally introduced in statistical mechanics by the physicists R. Then we construct a measure space (YJI. and this defines a continuous one-parameter group of *-automorphisms @ of .(B)) and F(t + il?)= ~(O#S)A). is tclosable and the polar decomposition S. (In statistical mechanics.1159 308 1 Operator Algebras which is called the integral direct sum (or direct integral) of fi([).e.e. where k is the Boltzmann constant and T the absolute temperature of the system. Deeper insight into the signifïcance of modular automorphisms is also provided by the work of A. and J.

It is called the hypertïnite factor. = e -‘T for some faithful normal trace z. & is of type III. Res. is isomorphic to the original von Neumann algebra .. (where 0 <n < 1) and 0 (type III.e. modulo inner automorphisms. the crossed product of . heG. of the type II.. of factors of type III (now called Powers’s factors). (which are then Powers’s factors) has been demonstrated by Connes [22]. such as ITPFI and the von Neumann algebra generated by any representation of canonical commulation (or anticommutation) relations on a separable Hilbert space..M (by *-automorphisms us. If d is properly infinite. with a one-parameter group of trace-scaling *-automorphisms 0. is ergodic. J. The restriction 8.) A von Neumann algebra on a separable Hilbert space is called approximately fïnitedimensional (or approximately finite or byperfinite) if it is generated by an increasing sequence of finite-dimensional *-subalgebras. The classification of approximately finite-dimensional factors is almost complete.. G of a vo n Neumann algebra . The S-set S(. the asymptotic ratio set r. In particular. is periodic with period .. are classifïed exactly by the isomorphism classes of the ergodic groups 8c of *-automorphisms of commutative von Neumann algebras 3. Any such factor is a Krieger’s factor.Section J) and its application to the following structure theorem for von Neumann algehras of type III. to the tenter 3 of A’ is of special importance. Th’z uniqueness of approximately lïnite-dimensional factors of type II. < 1. &! is of type III 1 if M is a factor.).(R) induced by the translations of the real line R. 1) (the examples of Powers) as well as for Â = 1. 1) (type III. or the set {O.#) of a factor of type III on a separable Hilbert space is either the set of all nonnegative reals (type III. Connes [19] introduced two invariants. In that case. any von Neumann algebra 4? of type III cari be written as the crossed product of a von Neumann algebra X of type II. The work of Araki and Woods shows that there exists only one ITPFI of type III. 1973).lr with 0. (which is then the tensor product of the hyperfïnite factor with a($)) and of type III.k’ together with the J. Examples of such factors have been extensively studied by Krieger.L. (The excluded case does not occur for &X’ of type Il 1. if 0.. A von Neumann algebra on a separable Hilbert space is approximately finitedimensional if and only if it is injective (. dual group as a one-parameter group of *automorphisms which is trace-scaling. Approximately fïnite-dimensional factors of type III.(G. AE&? and i(h). This class of von Neumann algebras includes many important examples.). and. z o 8. Using the Tomita-Takesaki theory. 1973) expressed independently a certain class of factors of type III as a kind of crossed product of semifïnite von Neumann algebras with their injective endomorphisms (automorphism in the case of Connes).N) and the p-set p(&‘) of the von Neumann algebra .M) (the set of a11 real t for which the modular automorphism @’ is inner).. A structural analysis of factors of type III. if & is aperiodic and not isomorphic to the one-parameter group of *-automorphisms of L. had been distinguished. gE G) is the von Neumann algebra N generated by 1he operators n(A).M) = r. The isomorphism class of d is determined by the isomorphism class of . Inst. Le. Crossed Products The crossed product U410.X. The crossed product of a von Neumann algebra . M is a factor if and only if 8. defined on the Hilbert space L.. who has also shown [23] that isomorphism of a Krieger’s factor is equivalent to weak equivalente of the associated nonsingular transformation of the standard measure space. which had been constructed by von Neumann in 1938 in terms of an iniïnite tensor product of factors of type 1 (abbreviated as ITPFI). a crossed product OFa commutative von Neumann algebra with a single *-automorphism. These analyses led Takesaki [21] to the discovery of a duality theorem for crossed products of von Neumann algebras with locally compact groups of theik *-automorphisms (. for each LE(O.308 J Operator 1160 Algebras conjugacy class of Q.&‘) (the intersection of the spectra of all modular operators) and the T-set T(. along with the same number of factors of type II 1..e.). i. is not smooth. the set of a11 integral powers of 3.36 Banach Algebras H). Araki and E. $j) of all . Takesaki (Acta Math. In fact the uniqueness of an approximately finite-dimensional factor of type II. and A is of type III. Math. namely the S-set S(.. and Araki (Publ. given independently by Connes [ 193. Woods [18] in terms of two invariants. when J& is an ITPFI.&’ (acting on a Hilbert space sj) and a locally compact Abelian group G relative to a continuous action c( of G on . Woods [20] has shown that the classification of ITPFIs of type III.LZ with the group of modular automorphisms @’ is a von Neumann algebra M of with a canonical action 0. i.. of 0. has been known since the work of von Neumann. Sci.. A systematic classifïcation of ITPFIs was subsequently given by H.!j-valued Lz- . while there exist continuously many ITPFIs of type III.log A. 0 <IL < 1. proved the equality S(. Prior to Powers’s work only three different kinds of factors of type III.(. 0 < 3.(d) and the relation T(d) = 2z110gp(J&)l -l.

[S] J. 1962). where the second factor a@. Zsido. von Neumann. North-Holland. and is intrinsic to the von Neumann algebra & (i. t+b(A)=(q+. monotone. P> t(g). [l l] M. T. 187-221. T. G] @e is isomorphic to & 0 @L. 1971. [8] S. [14] R. [4] J. [lo] S. 1961.(B) = p(p)Bp(p)* for BE&” and pee..representations are all of type I. [ 171 R. Sakai [S]). n ‘%s reflects certain properties of the von Neumann algebra & for O< c(< l/2 [24. Duality for crossed products of von Neumann algebras. generated by TX. Sakai. If this representation is a tfactor representation. V’i4 is called the natural positive cone. Math.e. III. Powers. Schwartz. homogeneous of degree l/2. Lectures on von Neumann algebras. 1981. Examples of groups that are not of type I are known (. Springer. W. 215-236. Springer. W*-algebras. A *-representation x+ TX gives rise to a von Neumann algebra A. Natural Positive Cone [l] J. For a C*-algebra &. References K. M. Theory of operator algebras I. U(g)s(cp)= 5((PogP). t($))).e. 128. bijective homeomorphism. On the equilibrium states in quantum statistical mechanics. Robinson. For example. A tgroup of type I is a group whose factor. 1979.1161 308 Ref. 1968. 1970. Fell. satisfying the relations U(g)ALJ(g)*=g(A). Collected works II. Every normal positive linear functional $ on & has a unique representative <(+) in this cone (i. 5 (1967).. Topological Groups). Nakagami and M. Tomita’s theory of modular Hilbert algebras and its applications. Dixmier. Gordon & Breach. 15 (1974). Takesaki. L.(G)) is the algebra of all bounded linear operators on L. Winnink. Pedersen.423 . Hugenholtz. connected semisimple Lie groups and connected nilpotent Lie groups are of type I (Harish-Chandra. 69 (1977). The duality theorem of Takesaki [2t] asserts that [A 0. Topological Algebras Groups and von Neumann Consider a unitary representation g+ Ug of a locally compact Hausdorff group G (.(G). Pergamon. A C*-algebra is called a C*-algebra of type I if its *-representations are always of type I. 1979. Functional Anal..25]. Math. It is known that this class is exactly the class of GCR algebras (. A new approach to the Tomita-Takesaki theory of generalized Hilbert algebras. [2] W. Phys. It is also known that a separable non-type I C*-algebra has a representation of type II and a general non-type I C*-algebra has a representation of type III (J. J3). C*-algebras and W*-algebras. Kaplansky. Springer. N. Rieffel and A.27]. Van Daele. Representations of uniformly hyperfinite algebras and their asso- The closure Vu of the set of vectors A. Amer. independent of the weight rp). Math. Springer. [ 121 Y. In particular. x E &. M. Benjamin. Von Neumann algebras.q(A) for all positive A in ‘%+. 378393. [ 161 M. Arveson. and M. It is a self-dual closed convex cone. (G.36 Banach Algebras H). 1953. 1961. Editura Academiei/Abacus Press. Trans. [ 131 M. C*-Algebras and von Neumann Algebras Let a C*-algebra & be given.. Lecture notes in math. gEAut A. Bratteli and D. Academic Press.. [6] I. 1976. The group of all *automorphisms of & has a natural unitary representation U(g). Pacific J. Math. Gauthier-Villars.. Operator algebras and quantum statistical mechanics. Amer. The canonical action 6( of the dual G on Jlr is defined by 62. G. Math. Dixmier. Proc. Sot. 731. M. C*-algebras and their automorphism groups.. Les C*-algbbres et leurs rep& sentations. 1964.. all its representations generate tinjective von Neumann algebras if and only if & is tnuclear [26. Takesaki. J. 1979. 1975. and the mapping l is a concave. Lecture notes in math.437 Unitary Representation El.. Springer.(A)<($). A bounded approach to Tomita-Takesaki theory. Operator on G (relative to the Haar measure) Algebras functions by where t: E L. Springer. The type of this representation is defined according to the type of &. where p(p) is defined by CAPM (9) = (9. Haag. [9] J.(G)). van Daele. [3] 0. 1967. Sot. An invitation to C*-algebras. [7] G. Rings of operators. Takesaki. [ 151 A. the type of this representation is defined according to the type of the von Neumann algebra -&’ generated by U. J. Glimm. Ann. K. gE G. 1979. Stratila and L. Comm.

although it must be transformed to the true anomaly or to the eccentric anomaly when we compute the coordinates of the aster- .. [ 191 A. Duke Math. Stratila. Canad.) The position of the orbital plane is determined by the inclination angle i to the ecliptic and the longitude R of the ascending node. (4) 6 (1973). R. 25 (1973).7) Orbit Determination A. The period T of one revolution. Krieger. north Fig. . Sci. Choi and E. 26 (1977) 443-446. Inst. and eclipsing binaries mined by similar methods.. 223 (1976) 19-70. while the argument w of perihelion. Woods. [25] A. 1231 W. Of these three anomalies the mean anomaly can be derived directly from Kepler’s elements. although orbits of meteors and visual. w. Ann. of the perihelion passage.55 Celestial Mechanics) and is described by Kepler’s orbital elements a. [28] R. Fast. Effros. we use the true anomaly u. To express the position of the asteroid on the ellipse as a function of time. Une classification des facteurs de type III. The mean motion is a fundamental frequency in the solution of the Hamilton-Jacobi equation and is obtained by differentiating the energy constant -p/2a with respect to an action variable fi. J. 4 (1974): 127-156. e. J. [18] H. 1981. Duality for crossed products and the structure of von Neumann algebras of type III. 1203 E. comets. A classilication of factors. 1 Orbital elements 309 (XX. Res. Math. Ecole Norm. for example. (2) to determine orbital elements from observed positions of the celestial bodies. Publ. Araki and E. we adopt as one of the main parameters the perihelion distance 4 = a( 1 -e) instead of the semimajor axis a. Academic Press. the eccentric anomaly E. Connes. 1983. Sup. General Remarks The purposes of the theory of orbit determination are (1) to study properties of orbits of celestial bodies.. The classification of factors is not smooth. Q. 24. Classification of injective factors. Pacific J. shows the direction of the major axis. The elliptic orbit is fixed by the initial conditions of the motion or the integration constant. and artificial satellites in the solar system. which is the angular distance of the asteroid from the perihelion. J. 966 102. (A) 4 (1968) 51-130. satellites. asteroids. and the mean anomaly M = n(t -to). Nuclear C*algebras and injectivity: The general case. Modular theory in operator algebras. [29] S. Fundamentals of the theory of operators I. Kadison and J. Connes. Connes. 50 (1974). with p a constant depending on the mass of the asteroid. Some properties of modular conjugation operators of von Neumann algebras and a non-commutative Radon-Nikodym theorem with a chain rule. On ergodic flows and the isomorphism of factors. Acta Math. [26] M... Inst. Kepler’s Orbital Elements Consider. Math. 1333252. Math. Ann. [24] H. 104 (1976) 733115. V.I. Caracterisation des espaces vectoriels ordonnes sous-jacents aux algebres de von Neumann. Indiana Univ. Math. (Fig. Ringrose.3 of the tHamilton-Jacobi equation (. [21] M. an asteroid moving on an ellipse with one focus at the sum. J. [22] A.. 43 (1976) 309-322. 3099354.309 A Orbit Determination ciated von Neumann rings. photo- The size and shape of the ellipse are determined by the semimajor axis (half I:he tmajor axis) a and the teccentricity e.. (Sometimes. Separable nuclear C*-algebras and injectivity. [27] M.. can be deter- 1162 B. Takesaki. Ann. i. (2) 86 (1967). Woods.. 1). and t. Araki. and (3) to compute their predicted positions utilizing the orbital elements. measured from the ascending node to the perihelion. Sci. Math. Ann. or mean motion n = 2n/T. metric. Fourier. Editura Academiei/Abacus Press. which is the mean angular velocity. is computed by Kepler’s third law u2a3 = p. Choi and E. and then the position of the asteroid on the orbit is determined by the time t. 138-171. Celestial bodies to which the theory is applied are mainly planets. Ann. Math.. Math. 131 (1973) 249-310. Effros.

the law of conservation of area1 velocity. On the other hand. it is possible to determine the orbital elements that can. Orbit Determination An astrometric observation of a celestial body usually consists of measurements of two coordinates (right ascension and declination) on the celestial sphere. for 100 days) are found. If the topocentric distance of the celestial body is known.. It should be remarked that for artificial satellites distance measurements have been made by radar. The approximate orbital elements can be computed if the launching conditions of the satellites are known. tGauss at the beginning of the 19th century to find the orbit of Ceres. The orbital elements of the ellipse thus defined at each moment.1163 309E Orbit Determination equation (1) Because of the perturbations. Solving this equation. the first asteroid to be discovered. E. or amounts of secular perturbations. called osculating elements. However. and velocity determinations have been made by measuring the Doppler effect. Therefore. C. and similar methods can be developed for parabolic and hyperbolic orbits. information on atmospheric density and the gravitational potential of the earth are derived. Therefore. tsecular.. by three sets of observations made at three moments at short intervals. Therefore. we obtain an expression for E as a function of M: E = M +“~i~JJne)sinnM. the orbital elements can be computed directly from observations. and variations of the mean orbital elements. During a time interval shorter than the period of one revolution. A method for orbit determination was worked out by C. When such additional observations are made. errors in the determined values often are very large. Although the topocentric distances are not known. where J. when gravitational interactions from other bodies cannot be disregarded. the orbital elements are found to be variable by computing the tperturbations by the tmethod of variation of constants. are variable with time. Kepler’s E-esinE=M holds between E and M. it is necessary to observe the initial conditions of motion. In this manner. although at every moment the instantaneous velocity and position of the asteroid determines an ellipse. observed positions should be corrected by subtracting the effects of periodic perturbations computed from approximate orbital elements already known. and the perturbations computed from them. From them.e. the orbit deviates from the fixed ellipse. to determine orbital elements for artificial satellites. the osculating elements change very little in a few weeks. is the tBesse1 function of order n. oid. Osculating Improvement Elements and Orbit For the ttwo-body problem the orbit is a fixed and invariable ellipse. special methods have had to be developed. . The perturbations are expressed as sums of periodic. mean orbital elements can be derived every day. To compute perturbations that cause this change of osculating elements. mean orbital elements are derived by the method of orbit improvement. Therefore we can assume that the area of the triangle made by the sun and the two positions of the asteroid observed at different moments is proportional to the corresponding time interval. and therefore Kepler’s orbital elements are constants. i. On the other hand. This method is called the indirect method. and Kepler’s second law. in practical computations. those data are compared with the respective values that follow from the initial observations. Artificial Satellites D. we often solve equation (1) directly by numerical methods or by using tables. for a certain period (say. the periodic as well as secular perturbations become very large after a few hours because the period of revolution may be as short as two hours. since the distance is not usually known. to derive six orbital elements. By using the observations thus corrected. Using this property of the orbit we can derive the topocentric distance and then the orbital elements. F. if the intervals are very short. holds approximately. However. and long-periodic terms. the osculating elements at the initial moment. we know that orbits of asteroids are planar. three sets of observations should be made at three moments separated by appropriate time intervals. for artificial satellites moving around the earth. then the method of least squares is used to improve the estimation of the orbital elements. the variations of the osculating elements are usually very small. and it becomes necessary to carry out observations at distant moments also. Since the periods of revolution of asteroids are of the order of a few years.be identified with the osculating elements observed at the mean moment. However.

Escobal. as well as data regarding their internal constitutions. x-. Bahnbestimmung der Planeten und Kometen.Section F). and 1x1. ixAiy=l. On the other hand. Kakutani. (2) 41 (1940)). XVy+XAy=X+J>.4) Ordered Linear Spaces A.(xAy) iXAiy=i(XVJl) (i>O). [2] A. and absolute value of the element x. negative part. 1929. and Ix-yl=xvy-XAY. and x v( -x) are called respectively the positive part. For photometric binaries radial components of velocities are derived by measuring the Doppler effect. such as their masses. if an element x of E is an upper bound of {x. the tjoin and +meet of x. Theoria motus.309 F Orbit Determination For satellites of other planets. Among them we cite H. +Banach spaces.. liXl= Jillxl. 2 (44) (1937)). L.X+/TX-= 0. Ix I= x+ + x . H. R. Order Limits Given a subset {xZ} of E. 310 (X11. D. 1965. densities. The theory of vector lattices was presented in a lecture by F. The central notion is Bana’zh lattices (. Springer. and H. b6E with a<b. References [I] G. Definitions F. Bohnenblust.~. Nakano [3]. [4] P. and rtopological linear spaces. C. Macmillan. Masses of planets can be computed by Kepler’s third law when the orbital elements of satellites are known.(ii)x>yan’i/13O(iis a real number) j ix > 3. ixviy=i(xvy). 1963. For Sections B through E. Vulikh [S]. b]. Schaefer [S]. E forms a UaHice under this order 2. although the exact estimation of the distances to binaries is often impossible. in addition. are generalizations of Euclidean spaces. and gravitational potentials of the planets can be determined from their secular perturbations. can be derived from the observed orbital elements. The following identities hold: x= X+ -x (Jordan decomposition). If. C. Dover. y are denoted by x v y and x A y respectively. The determination of orbits. where the leading idea has been to generalize the distance in Euclidean spaces in various ways. F. Math.} and any upper bound JJ of {xZ} satisfies the relation y 2 x. 1164 Section I).. iXViy=i(XAy). Vector lattices have been used in lattice-theoretic treatments of integration (- The last relation means that E is a tdistributive lattice. Wiley. Kantorovitch (Mat. (iGo) and (XVy)AZ=(XAz)V(yAZ). For any x. the set {xla<x<h} is called an interval and is denoted by [a. Birkhoff [2]. measurements of two coordinates with respect to the centers of the planets are made. then it is called the least upper bound (or su- . Stracke. B. 1961. S. A real tlinear space E is said to be an ordered linear space if E is supplied with an torder relation > with the following two properties: (i)xayax+z>y+z. V. +spectral resolution. and sizes. G. For a.r) v 0. Freudenthal. Riesz at the International Congress of Mathematicians in 1928 [l] and has been developed by many authors. Riesz (Ann. the elements x v 0. YE E. methods similar to those for satellites can be applied. [3] C. Sb. A subset of E is called (order) bounded if it is included in an interval. A linear subspace I of E is called an ideal (or order ideal) of E if x E I and Ivl<lxl imply ~61. History Many spaces used in functional analysis. such as +Hilbert spaces. H. Dubyago. Binaries In the study of visual binaries. F. and for eclipsing binaries important information. An element e of E is said to be a unit or an Archimedean unit if for any x E E there exists a natural number n such that x < ne. ( -. Gauss. Methods of orbit determination. but the theory has been extended to the case where E is a +locally convex topological linear space with the SI ructure of a vector lattice [S-S]. we assume E to be a vector lattice. For x E E. B. generalizing the order concept for real numbers has led to spaces of another kind: ordered linear spaces and vector lattices. and are denoted respectively by x+. The following relations are obvious: (x+z)v(y+z)=(xvy)+z. and tergodic theory. we call E a vector lattice (Riesz space or lattice ordered linear space).

Amsterdam.eE) is said to be order convergent to x if there exists a nonincreasing sequence {u”} (u. or VoI x.&(~)) -xlQy/1(1=1. called the dual lattice of the vector lattice E and denoted by Eb. F) is a complete vector lattice. We give two examples. or A\.e.. be a linear space of functions defined on a set R ordered pointwise.) = J(o-limx.-xxyll+O. + py. In this case x is called the order limit of {x. there exists a subsequence {x. x E E. u. the following is one of the most fundamental: In a Banach lattice E. A/ad. C) and boundedness in S with weaker ones and still obtain the same situation. Proc. m. For order convergent sequences {x.3. The tRadonNikodym theorem in A@. . to mean (T. imply Ix = o-lim &. First.. Furthermore if a normed vector lattice E is complete relative to the norm.t.XI < u. Examples of Vector Lattices Iflc4 = sup f(Y). form vector lattices under pointwise ordering (168 Function Spaces). F. This space. If there exists a bijective mapping defined on a vector lattice E onto E.2.. If E is a-complete. similarly.} and an element y of E satisfying the relations IX. F) is the set of all (order) bounded tlinear functionals on E. E.} (x.. Among relations between order convergence and norm convergence in Banach lattices. is defined to be we define o-lim infx.(. We say that E is Arcbimedean if the relations 0 < nx < y (n = 1. A e-complete vector lattice is always Archimedean. The greatest lower bound (or infimum) of {x~}.} and is denoted by sup..x)>(T. then E can be represented as the set of real numbers such that the Archimedean unit of E is represented by the number 1 (H.x~E). The examples in Section D are Banach lattices (for PEA@. ) imply x = 0..}. Let E..x.). M..x y. Freudenthal..x) for any XEH.. that is linear and order isomorphic. norm convergence of a sequence {x”} to x is equivalent to relative uniform star convergence of {x. Any set bounded relative to the order is bounded relative to the norm. and let A@.). is equivalent to x = o-lim sup x. C) be the set of all finite o-additive tset functions defined on C. We say that E is complete (Gcomplete) if any (countable) subset of E that is bounded above has a least upper bound. We can replace the conditions of finiteness in A@.x#. F) be the set of order bounded linear mappings of a vector lattice E into a vector lattice F. 11x. then S is a complete vector lattice under the ordering just defined. and L.294 Numbers). let C be a a-algebra of subsets of a space 0.x.2. = 0 and Ix. we can show the following relations: o-lim(lx. Banacb Lattices A linear space E is called a normed vector lattice if E is a vector lattice having the structure of a tnormed space satisfying 1x1~ 1yl* llxll< Ilyll. x = o-lim x. and .x.) :: (o-lim y. then L?(E. o-lim supx.) and o-lim(x. for any subsequence {x. a representation of E.?= lim i.. is a complete vector lattice. F).} and { y.x. A sequence {x.} bounded above. Dual Spaces Let !2(E.}. i. C) and the tspectral resolution theorem of symmetric operators in S can be extended to theorems of tspectral representations in general vector lattices. define ‘pi > ‘pz to mean cpl(x)~cp. . but the converse does not hold in general.xy. For a linear functional. we have D. and the order dual of . With these definitions. denoted by inf. The second example is an ordered space consisting of all tbounded symmetric operators Ton a Hilbert space H. and tfunction spaces. However.1165 310 F Ordered Linear Spaces premum) of {x.)} of {x. E E) such that A. however. if A is a commutative tW*-algebra of operators on H and S is the set of tsymmetric operators belonging to A.-x/l+Oand /Iy.. InBanachlattices. F) is called a positive operator if cp> 0. for any sequence {x. and l. IYI GX Sequence spaces. Z) is a complete vector lattice if we define pi 2~~ to mean pLt(S)>pLz(S) for any SEC.} and is denoted by x=o-limx. If F is the set of real numbers R.. is defined dually. Then A(R.(i))} of {x+. C) we define 11~11 =IM4).) = (o-lim x. qpzE 2(E. but the others are complete.(x)(x>O.. An element cpE C(E. . For f6Eb and x 2 0. If E is Archimedean. Any Archimedean vector lattice can be extended to a complete vector lattice in the same way as the real numbers are constructed from the rational numbers by Dedekind cuts (. For any cpi. such as C. we call E. then f?(E. this space is not a vector lattice. x...) + p(o-lim y. we call E a Banacb lattice. = o-lim infx. then the relations x = o-lim x. > T. If E has an Archimedean unit and is simple (which means that E and (0) are the only ideals of E). where order boundedness means that any bounded (in the sense of the order) subset of E is mapped into a bounded set of F. where we define T. to x. such as c..-yll -0 imply lIx. A”Vm>nxm. these two concepts of boundedness coincide. In general. If F is complete. 39 (1936)). Among these spaces c and C are not u-complete.).

Niiro (Nat. London Math. (2) 42 (1941). we say that E is an abstract M space. Sawashima (Sci. a null function if N(q) = 0 holds. Frobenius (S.) and If] < C. and AL..lh where I. we state the integrals of Daniel-Stone and of Banach. Schaefer [8]. and lim. AL spaces. and scalar multiplication.e. respectively. and a set A is a null set if its tcharacteristic function is a null function.“=. 108 (1968)). Karlin (J. we consider the following three conditions: (M) x. or (L. (1 d n. (L). Duke Math. 26 (1950). Miyajima (J. the set of all real-valued continuous functions defined on a compact Hausdorff space Q... 6 (1940)). In this case the eigenvalues of A on the spectral circle are the kth roots of unity for some k multiplied by r(A). 30 (1955)). 1948). F. and abstract L space.. by definition. I. G. and AL. Integrals Based on Ordering H.2. = 0 imply lim . Tokyo. E is the same as the norm dual of E. i. Mech. Let 3 be the set of equivalence classes of k. Sot..fl means . original in Russian. f. Sci. Ann.. 1.f.. > 0 for all i.+. Univ. Then its spectral radius r(A) belongs to a(A). A square matrix A = (ai. A. 8 (1959)).l. Pap.Section E). Krein and M. The spectral circle of A is the circle of radius r(A) having the origin as its center. i. < n).% IhI imply E(lfl)~C. Gen. . for every function cp on S admitting fco as values. spaces (l/p + l/q = l). Ando (J. For these extensions. j) of order II is called positive if ai. F. Math.. Ann. Akad. j > 0 for all i and j. Hokkaido Univ. Preuss. F. Fat. H.> 0 implies E(f)>O.e. S. 64 (1907)) and G. Theorem (Perron-Frobenius). If E satisfies one of the conditions (M). Moreover. Cal!.. A function cp is. on a tmeasure space.. Math. Spectral Properties of Positive Operators The n-dimensional real vector space E. ser.. i.310 G Ordered 1166 Linear Spaces spanned by a strictly positive element. written AM. respectively. Since a positive matrix of order II corresponds to a positive operator in E. each of which is a simple root of the eigenequation of A.. P. Ochanomizu Univ. A positive linear functional on & satisfies M. Daniell’s condition (iii)‘: fi >f2 > . sup{ 1~11Ada}. is a vector lattice under pointwise order (. Sawashima and F. or an abstract L.. (iii) A . A is called irreducible if there exists no permutation matrix P such that P-‘AP= . I. y>O*~jxvy~~ = max(llxll. Here the representation of a Banach lattice means a representation of a vector lattice preserving the norm (Kakutani.. Let us begin with a set (5 of realvalued functions defined on an abstract space S and assume that 6 is a vector lattice with respect to the usual order relation. Bohnenblust. J.“=. . H. see the following articles: M. 27 (1980)). spaces are AL spaces. H. Let A be a positive square matrix. respectively. a functional N(q) as follows: are square matrices of order n. . and for this spectrum r(A) there exists an eigenvector x 3 0. Z.. We denote by cr(A) the tspectrum of A and by r(A) the tspectral radius of A. and L. Ilvll).4 + IIYII. we define. and AL. H. extensions of this theorem to positive operators in ordered linear spaces have been studied by many mathematicians.. It corresponds to a positive operator in E. (ii) positivity. Stone’s condition (iii) if and only if it satisfies P. Assume further that A is irreducible and the order of A is greater than 1. . space.. E(f+ g) = E(f) + E(g). addition. Math.fv( -. (.e.). An AM space with a Kakutani unit is represented by C(n). &. Lotz (Math. we can define addition and the scalar multiplication for such functions except on a null set.)=O. Rep.f. The duals of AM spaces. If the unit ball of an AM space has a +greatest element. and A.. -Nf. G. Wiss. An element x in E. Rutman (Amer. 13 (1957)). 1908 and 1912) established the following remarkable result on the spectral properties of positive matrices. 30 (1979)) and S... Y~O=~IX+YII = Il.. Sot. Univ. Since each function of so = {‘p 1N(q) < +co} takes finite values except on a null set.I x~Y=wlx+Yllp= llxllP+ IIYIIP (1 < p < co). where A. Perron (Math. Transl. WI x. Sci.e. AM spaces with the Kakutani units. Fuc. AL. spaces are represented by L. Bonsall (J. As applications of ordered linear space theory.E(f. Niiro and I. The AL spaces and AL. Abstract M Spaces and Abstract L Spaces For a Banach lattice E. i. 0.Section D).f). lf.... Then r(A) > 0 and the eigenspace of A for r(A) is a 1-dimentional subspace Here we put N(q)= +co when for a function Q there are no functions { fn} such that Iv]< C.~& (n= 1. B. Tokyo. it is called the Kakutani unit of E. A functional on e satisfying both conditions (i) and (ii) is called a positive linear functional. T. Assume further that a functional E(f) defined on e satisfies the following conditions: (i) additivity. Next. Educ. is called strictly positive if x. . J. the dual (in any sense) of a Banach lattice is also a Banach lattice. 1966).

Since 2 is an tabstract L-space. Let us denote now by L! the closure of (%in 3. the concepts of measurable functions. 1957. . Furthermore. The integral L thus defined is. Nakano. S is a locally compact Hausdorff space and & is the set of continuous functions with compact supports.{F(x(s))+f(x(l-s))}=Jx(s)ds is called the Banach integral of x(s). Harper & Row. and a result corresponding to Fubini’s theorem has been obtained [9].. replacing the sequence in it by a tdirected family of functions 02~ Specifically if. Publ. revised edition. Atti de1 Congresso Internazionale dei Matematici.. an extension of the functional E on @. Colloq.. Amer. US.=O. Introduction to the theory of partially ordered vector spaces. 1) by using the tHahn-Banach extension theorem [12]. x(s)ds References [l] F. Springer. the relation between L and the Lebesgue integral with respect to this measure is known [9]. gs(Z imply lim. c~-=%ld--vcp))is called the Daniell-Stone integral of rp. Sur la decomposition des operations fonctionnelles lineaires.$) =O. For this integral.. replacing condition (iii’) for a positive linear functional E(f) on 6 by condition(iii”): lim. (4) j 1 ds = 1. we define.) = 0 [lo]. [3] H. or an integral considered as a mapping from a vector lattice into another vector lattice (or from an ordered set into another ordered set).. andf. then we can prove immediately that jx(s)ds has the following properties: (1) ~{UX (s) + by(x)}ds=ajx(s)ds+bjy(s)ds. in the Daniel&Stone integral. then almost all results in this section (e. Normed linear spaces. [2] G. .. M. Amer. (3) jx(s+s. such as an integral for more general functions with values in a vector lattice. The construction of the Daniell-Stone integral and the Banach integral opened avenues to several other abstract integrals based on the order relation. Also.. a). Banach defined an integral by using methods similar to Daniell’s. Math. 1) and p(x) = in$ M(x. . Then any function cp belonging to f? is said to be Daniell-Stone integrable. Maruzen. Lebesgue’s convergence theorem is easily proved. +co) of the function defined . Day. H.{F(x(s))+F(x(l-s))}. Peressini. Nat. where x(s) is considered as the periodic extension to (-co. Ordered Linear originally Spaces with respect to the relation cpN $ defined as N(cp . Daniel1 (Ann. as a functional on 2. Banach also defined another integral for all real-valued bounded functions on [0. For discussions of these and other abstract integrals . 143-148. [S] B. S. Erg. The Daniell-Stone integral introduced above is due to M. Clll. L. and the Daniell-Stone integral L(q) can be constructed from E(f) on [0. E(f. where sc is an arbitrary real number.g. If necessary. [7] A. 1946) have defined a more general integral than the Daniell-Stone with a condition analogous to (iii’). Sot. and measure can be defined by using L and 2. 1) and 2I be the family of all finite sets of real numbers c(= (c(r) c(~. If we write Jx(s)ds for F(x). where a and b are real constants. Math. and (E is included in 5 (by identifyingfandgof&whenE(lf-gl)=O). Lattice theory. he defined the set f! of Daniell-Stone integrable functions by L={cplF(q)= -I(--cp)}. (2) x(s) > 0 implies Jx(s)ds>O. Acad. His definition is as follows: Let 5 be the set of all real-valued bounded functions on [0. there exists a linear functional F on 5 satisfying F(x)<p(x). Also. [6] I. Then.Furthermore. 1948. Math. Furthermore. measurable sets.[13-151. CL. (ii).. N.Jds=Jx(s)ds.). by the Hahn-Banach extension theorem. Then 3 is a Banach lattice with the norm N. Sci. and (iii’). 1950.. Modulared semi-ordered linear spaces. Sot. [4] M. Bologna. 1967. Riesz.1167 310 Ref. 3. 1967. Partially ordered linear topological spaces. lf. Namioka. Vulikh. Math.f. 1958. Ordered topological vector spaces. McShane (Proc. we can add the property (5) jx(l s) ds = j x(s) ds by defining x(s)ds=. C. Indeed. for x(s) E $j and c(E 2I. the Daniell-Stone integral L(q) is represented by the Lebesgue integral of cp on a certain measure space. the Hahn-Banach extension theorem) hold trivially. Gronigen. Stone [9]. s Then F(x)= s or . 1928. Mem.. and L(q) = N(cp+)N(cp-)(cp+=&ld+d. Birkhoff. Bourbaki [ 1 l] and E.l<g. then a functional E(f) on a satisfying conditions (i) and (ii) is proved to satisfy the condition (iii’). J. if the function takes values in a complete vector lattice. (2) 19 (1917-1918)) originally defined the upper integral F(cp) by using E(f) on & satisfying conditions (i).

T. 8.X or supX. while x>y means that x 2 y and x # y. (Chelsea.543-547. N. and if the subset has a lower bound. ch. Banach. Imp. then the ordering < . Springer. [ 111 N. Sunouchi. x <x. Press. 1964). A set X with an ordering between its elements is called an ordered set (partially ordered set or semiordered set). Linear operators. When a is an upper bound of X and u~X. 7. In an ordered set A. Banach. [ 131 S. Sot. J. 1963).c. Order-preserving maps and integration processes. most of the terminology associated with the inequality of numbers is carried over to general ordering.87~89. I (1958). 45p49. 1963. Schaefer. 535-538.518~523. ch. and b a successor of a. ch. M. 17 (1941). 1937. Definitions A subset of an ordered set X of the form {x 1a < x < h} is denoted by (a. More generally.1. If there is a least element in the set of upper bounds of X. 1306. Springer. [ 151 T. order relation. 1932. Warsaw. [ 171 K. l-9.483-490. Functional analysis. it is also an ordering. [lo] S. Notes on integration. Amer. ch. 34 (1948). If a universal proposition concerning the ordering is true. y. 1168 is called a total (or linear) ordering. When u<c<b or b<c<u. In particular. h). Ordering The concept of ordering is abstracted from various relations. The dual concept of an upper bound is a lower bound of the subset. Hildebrandt. and propositions concerning an ordering are defined by replacing the ordering with its dual. 16 (1940). ActualitCs Sci. x < y and y $ x imply x = y. such as the inequality relation between real numbers and the inclusion relation between sets. 4477455. [ 121 S. its dual is the least element (or minimum element) and is denoted by min X. 336-342. 117a. then its dual is also true. and > is the dual of <. Math. then a is called a predecessor of b. M. Saks (ed. 6. Tokyo.u. 1965. this principle is called the duality principle for ordering. c is said to lie between a and b. j. H. h). second edition. Yosida. Banach lattices and positive operators. an element a is called an upper bound of a subset X if x < (2 for every element x of X. Acad. Izumi. b satisfying a < h is called a quotient of X and is denoted by b/u.50-52. The binary relation > is called the dual ordering of <.X or inf X. second edition. S. the duals of concepts. ThCorie des opkrations linkaires. A pair of elements a. 30-32.. l-4. {x 1x < a}. An abstract integral. IV. H. 21~25.311 A Ordering [S] H. 1953. Studies. If for an arbitrary pair of elements x. Its dual is the greatest lower bound (or infimum) and is denoted Isy g. Bourbaki. 1965. H. Hermann: ch. Nat. Bull. Orihara. and A is called a totally ordered (or linearly ordered) set. 1959. McShane. conditions. and a set of the form (a. 1974. A subset of an ordered set X is also an ordered set with respect to the same ordering as in X. 128la. Proc. Schwartz. the relation between the elements of X. Z. 59 (1953). Elkments de mathkmatique. B. 320-330 (Dover. or simply order) if the following three laws hold: (i) the reflexive law. We sometimes write x < y as y > . Theory of the Integral. semiordering. or {x 1x > a} is called an interval. [16] N. Math.b.). A set bounded both from above and from below is simply said to be bounded. Wiley. 1967. Princeton Univ. IV. 50-58. is called an ordering (partial ordering. ch. For example. Matuyama. X is said to be bounded from above (or bounded above). it is said to be bounded from below (or bounded below)..12) Ordering A. 9. S(c) = {x 1x cc} is called the segment of X determined by c. Incidentally.539%542. then a is called the greatest element (or maximum element) of X. x < y is equivalent to the statement x < y or x = y. Integration. The Lebesgue integral in abstract spaces. denoted by < or other symbols. x<y means that xdy and xfy. V-X. and (iii) the transitive law. 35 (1949). Stechert. 1343. 5. 1969. I-111. III (1971). When u < b and there is no element lying between a and b. In this manner. Nakamura. When an upper bound exists. it is called the least upper bound (or supremum) of X and is denoted by 1. Acad. according to the definition of <. (ii) the antisymmetric law. Ann. x < y and y < z imply x <z. Suppose that we are given a set X = {x. A totally ordered set A is said to be dense if for any pair of distinct elements a and b in A there exists a third element c lying between a and b. Ind. Sci. Stone. y of an ordered set A either x < y or y d x must hold. Proc. 1244b. l-4. where A is of the form {i& 1C(n) ). then sup X is also written 311 (11. [ 141 E. If the ordered set X is the image q(A) of a set A under a +mapping cp. [9] M. and G. US I-111. 18 (1942). 11 l139. Dunford and J. Integration in abstract spaces.53-56.b. Such an element a (if it exists) is unique and is denoted by max X.

.. To define a mapping F from a well-ordered set X into a set Y. . If {b 1b 2 u} fl B # 0 for every element a of A. of ordered sets. holds for all 1 E A.)..A of its subsets. . If the greatest (least) element exists...(. The prin- An ordered set (or in general a preordered set (. For a. it may be written as sup.. F.< p(b).<. of ordered sets. and (ii) P(x) is true if P(y) is true for all y satisfying y < x.<u. are elements of the Cartesian P = &. Ordinal Sum and Ordinal Product Suppose that 2I = {A..)....4. Direct Sum and Direct Product Let S be a set that is the tdisjoint union of a family {A.) is true for the least element x0 of X. Directed Sets C. Then P(x) is true for all x in X. min. respectively. where S(x) is the segment of X determined by x. such a subset B is itself a directed set.< (b. B. for some SEA and ad b with respect to the ordering in A. cp is called a dual isomorphism (or anti-isomorphism).)..}. But in general. If {b 1b > u} c B for some element a of A. we define (a. of ordered sets. its dual is a minimal element. By the chain condition.<. it is the only maximal (minimal) element. . cp(l) or sup ~(1) and called simply the supremum of cp(J. D. b E S. An infinite sequence { 4. and the minimal condition to the descending chain condition. to mean that a. is an ordered set. the maximal condition is equivalent to the ascending chain condition. and for each element x of X there is given a method to associate an element G(f) of Y uniquely with each mapping f: S(x)+ Y with domain S(x).312 Ordinal Numbers).. Order-Preserving Mappings A mapping rp: A-A’ of an ordered set A into an ordered set A’ is called an order-preserving mapping (monotone mapping or order homomorphism) if a . similar conventions hold for inf. An element a of a set X is called a maximal element if a <x never holds for any element x of X. define adb to mean that a. and its ordering is called a well-ordering. and product (b. } of elements of an ordered setXsuchthata. we may use the following principle: Suppose that F(x. The dual condition is called the maximal condition.). Let B be a subset of a directed set A. G.. and suppose that each A. A’ is said to be order homomorphic (order isomorphic) to A when there exists an order homomorphism (order isomorphism) cp such that A’ = q(A)...is called an ascending chain. . A.. etc. The following theorem is called the principle of transfinite induction: Let P(x) be a proposition concerning an element x of a wellordered set X such that (i) P(x.1169 311 G Ordering ciples of induction are often used for proving propositions or giving definitions concerning ordinal numbers (. such a subset B is also cotinal in A. If a mapping cp: A-+ A’ gives an order isomorphism of A to the dual of A’.. then the set X is called a well-ordered set. a. where X is the set of all natural numbers. The condition that B is colinal in A is equivalent to the condition that A -B is not residual in A.< b always implies q(a) . and the condition that X has no ascending chain is called the ascending chain condition. if every nonempty subset of X has a least element. then rp is called an order isomorphism. sup.. equivalently. Moreover. The definition of the mapping F by this principle is called a definition by transfinite induction. then B is said to be cofinal in A.JA. cp(l) and is called the supremum of cp(i) for all 1 that satisfy C(1).. a maximal (minimal) element is not necessarily unique.Section H)) in which every finite subset is bounded from above is called a directed set. bEA. The notions dual to those of ascending chain and ascending chain condition are descending chain and descending chain condition. . When (al)le. If a totally ordered set X satisfies the minimal condition or. Under the taxiom of choice. E. < b.}.<a.. The ordered set S obtained in this way is called the direct sum (or cardinal sum) of the family {AA). of a family {A.. . Then there exists a unique mapping F: X-+ Y satisfying F(x)= G(F 1S(x)) for all x. if rp is bijective and qp-’ is also an order-preserving mapping from A’ onto A.) is defined for the least element x0 of X. } is a family of mutually disjoint ordered sets and is itself an . we mean either the ascending or the descending chain condition. Chain Conditions An ordered set X is said to satisfy the minimal condition if every nonempty subset of X has a minimal element. When there is no danger of misunderstanding. then B is said to be residual in A. tMathematica1 induction is a special case of this principle.. . The ordered set P obtained in this way is called the direct product (or cardinal product) of the family {A. max. .

It can be applied to X = I”I. 1243b. The ordering <. Birkhoff. The least transtinite ordinal number is o. ye A E 9I and x d y holds with respect to the ordering in A. the set a’ = { 5 ( 5 <a} is also an ordinal number. another definition was given by J. X. 2 = (0. Ind. English translation.. ThCorie des ensembles. in fact. if A is well-ordered. The set w = (0. and y = (y. In particular. . A similar situation was found concerning the definition of tcardinal numbers. General Remarks Let A g B mean that two tordered :sets A. 1968. B. the empty set. X. a twell-ordering of the ordinal numbers. . are ordinal numbers. 1967. The ordered set S obtained in this way is called the ordinal sum obtained from 2I and is denoted by C. Let [X]=X/be the tquotient set of set X by this equivalence relation.) References [l] N.. According to this definition. is called a transfinite ordinal number. X is then called the ordinal product. and the ordering of the finite ordinal numbers c’aincides with the usual ordering of the natural numbers. Sot. . which defines an tordering of the ordinal numbers. X. 1948. 1. then the relation z is an tequivalence relation.311 H Ordering ordered set. an ordinal number was first defined as the order type of a twell-ordered set (Cantor [2]). ch. . and let [xl. AddisonWesley. By defining (x. which is given in Section D. which are finite sets.} of A has a least element whenever x=(x. . like w.52 Categories and Functors.)*x. Theory of sets.. This ordinal number is called the ordinal number of A. [2] G.. a preordering of pairs (x. we have A <: B. Lattice theory. it was found that a contradiction occurs if order type defined in this way are considered to form a set. Preordering A relation R between elements of a set X is called a preordering (or pseudoordering) if it satisfies the reflexive law and the transitive law. I. ElCments de mathkmatique. is often used to denote the ordinal product obtained from X. Publ.2. and the class to which an ordered set A belongs is called the order type of A. Also . are called finite ordinal numbers. For any ordinal number a. . or (ii) for A and B satisfying x E A E (II. (For further topics . The finite ordinal numbers are identified with the natural numbers (including 0). Math. [y] be the equivalence classes determined by x. B are torder isomorphic.13) Ordinal Numbers A. } of all natural numbers is also an ordinal number. . and (ii) p E CLimplies /3 c 0:. which led to a new definition of cardinal numbers using ordinal numbers. 1940..) We also write c(E p as a c /?. ActualitCs Sci..)R (x. Colloq. y) of real numbers is obtained.). Amer. for example. B. lower-case Greek letters denote ordinal numbers. are ordered sets. Therefore ttransfinite induction can be applied to ordinal numbers. B} and A <B.y. 1. . The ordering in X defined by setting x < y when xc < y. but not necessarily the antisymmetric law. 409 Structures. are two distinct elements of X. the ordinal sum is denoted by AS-B. An ordinal numb’cr that is an infinite set. From a preordering R an equivalence relation . with the ordering 1 < 2 < . Hence.references to 381 Sets. ye BE 2t. When A. which is stated in Section B. = B. 1 = {0}. introduced by defining CI< fl to mean either R <p or a = /Y.. of a family of ordered sets indexed by an ordered set A. l}.2}. An equivalence class under this relation is called an order type.} is called the lexicographic ordering in X. then an ordering < can be defined in [X] by [x] < [y] 0 xRy. which is denoted by 0. Definitions A set tl is called an ordinal number if it satisfies the following two conditions: (i) tl IS a wellordered set with the tbinary relation E as its ordering. 3 = (0. and the subset {i 1x2 # y. <x. second edition.. Then an ordering < can be defined in the disjoint union S = u X (X E %) as follows: x <y in S means that either (i) there exists an A satisfying x. Suppose that X is a subset of the Cartesian product n. The least ordinal number is 0.. However. = A. yeX. Also.. of indices. revised edition. 1170 312 (11. is a tlinear ordering and. and is the tsuccessor of a. These ordinal numbers. if ?I = {A. (Throughout this article. For every well-ordered set A. is an ordinal number. AB .). for the least element p of { 11 xi # y. the ordering in this ordinal product is called the lexicographic ordering in the Cartesian product AxBx .y. Bourbaki.. there exists one and only one ordinal number order isomorphic to A. Historically. X. Hermann. There exists at most one H. 3. von Neumann [3].can be defined in X by x-yo(xRy and yRx).

Let R be the set-valued function of ordinal numbers.p=cr. then for any c( there exists a /I that satisfies f(p) = /3 > t(. cf(a) is a regular cardinal number.B (or a/?).~= cr... For any set A of ordinal numbers. /3 are delined by translinite induction on /I and have the following properties: cr+o=a. wo=o. . (a.(b+y) =cr.yz+. Inaccessible Ordinal Numbers which is called the n-adic normal form for a. In particular. In fact. Cardinal Numbers C. An ordinal number a with the property that . All finite ordinal numbers are cardinal numbers. then tl <f(cd. Such an ordinal number E is called an c-number. it is called Cantor’s normal form. is both the smallest uncountable cardinal number and the smallest uncountable ordinal number. and E. Since f(a) = ob is a normal function. where ‘$3(M) denotes the tpower set of M. it is weakly inaccessible. then tl+ b is the ordinal number of the tordinal sum A + B. We say that f is strictly monotone when CL < /3 implies f(a) <f( /I). (alepb alpha) or 0. A finite ordinal number is called an ordinal number of the first number class.@?. and o is the least transfinite cardinal number.1171 312 E Ordinal Numbers of c( and is denoted by ordinal number that is the tpredecessor of tl. With the taxiom of choice. Let f be an ordinal number-valued function of ordinal numbers. We say that f is continuous when f(y) = sup { f(t) 15 < y} for each limit ordinal number y.. The value of this function corresponding to a is denoted by K. this definition is equivalent to the one given here. D.fi+a. y c R(a). @‘=(c@)~.c( ) tlY=sup{c(~l~<y}. and for the power we assume that tl > 0. V=suP{~. An ordinal number is said to be regular when cf(cc)= c( and singular when cf(cr) <a. and K. (5 13~ (5 < q E A)} is an ordinal number and is sup A. The first ordinal number that is colinal to CI is The cofinality cf(a) of a always satisfies cf(a) < c(.yn. A regular ordinal number LXis said to Lx strongly inaccessible when CI> w and the following condition is satisfied: If x.?) and the left distributive law cr.5 implies c(< 5 is called an initial number or a cardinal number.+RB”. a+Y=suP{~+515<r}. =f(BJ and put B=sup(&In<c4. and the power ~8 of ordinal numbers CL. respectively. A strongly inaccessible ordinal number is usually defined as a regular number c(> w such that /I < c( implies ‘p(p) < tl. If f is a normal function. If a regular ordinal number t( is strongly inaccessible.y. any ordinal number c( can be written uniquely in the form cc=lrSl. The concept of ordinal number of the third (or higher) number class is defined similarly. ct . and an ordinal number tl satisfying K. under the tgeneralized continuum hypothesis. the power satisfies the laws c@+~=cx~~~~. called the cofinality cf(cc). c( is said to be weakly inaccessible. Here y is a limit ordinal number. the tsupremum of A. For any ordinal number CC. l<i. and there exists a mapping of x onto y. when there exists a monotone function f that satisfies c(= sup{ f(t)’ 1l< 8). rg=&. O<Yi<7c./?). There exists one and only one monotone function that maps the class of ordinal numbers onto the class of transfinite cardinal numbers. If CI and bare the ordinal numbers of the well-ordered sets A and B. A strictly monotone continuous function is called a normal function. Sum. is called an ordinal number of the second number class. cr. The first limit ordinal number is w. and Power The sum cc+p. therefore any regular ordinal number is a cardinal number. = w. cc+B’=(a+8)‘. then ye R(a)./I+cc. Moreover. y are a pair of sets such that XE R(a). When tl = e+ is regular and /I is a limit ordinal number./I is the ordinal number of the tordinal product BA.. strong . there exists an E that satisfies O’ = E. The sum and product thus defined satisfy the associative laws (~(+/I)+y=cc+(B+y). defined by R(O)=@ and R(a)=~{~(R(~))~~<a} (by ttranstinite induction). E. when n = o.. Under the axiom of choice. When n > 1. < CI<K. and all the other ordinal numbers are called isolated ordinal numbers.515<r}. If f is strictly monotone. This unique CC is called the cardinality (or cardinal number) of the set M and is denoted by M. Bl>DZ>. CCO=1.yl+K82. it can be shown that for each set M there exists one and only one cardinal number TV satisfying M -CC.<n. Product. We say that /I is cofinal to u. K. and it is a normal function. it suffices to define /I& < o) by PO =f(cc + 1X A+. Let M-N mean that a one-to-one correspondence exists between the two sets M and N. the product a.B”>O. A transfinite ordinal number without a predecessor is called a limit ordinal number.

Usually we assume further that f has a certain regularity. . Contributions to the founding of the theory of transfinite numbers. (3) . .>y (N1 f(x.. . ci. (. y. . . . (sometimes called integration constants). 1932.” If the left-hand side . and which contain the partial derivatives a~/&. co). ci. If we can eliminate ci.f of (1) contains y(“) explicitly or aflay # 0. and if further f is a polynomial in y. and a holomorphic function if x. then (1) is said to be linear. and (py(xo. 1 (1923) 199-208. Math. to be constants in cp(x. 1967. . y. . then we say that the order of (1) is n.c. To find a solution of (1) is to solve or integrate it. and let (x. . . y. c. . then we obtain an nth-order differential equation of the form (1).. Here each equation of the system has a form similar to (I). Klairaut differential equations. y”. 1968.“) =0 obtained from a general solution cp= 0 by giving particular i. von Neumann.Cl.“) # 0. and their derival:ives. . y. Hermann. c. &ED. . (Gesammelte Abhandlungen. Zur Einfiihrung der transfiniten Zahlen. . y(“). second edition.) [3] J. . . . ay/dx.1). but each left-hand side contains y.” to the arbitrary co. y”” for the first n derivatives of y.. y. . Open Court. y. such as being of class c’ (r = 0. CT. 1. Math.) Also .y. . from these n equations and cp=O. Ordinarily. y’. In particular.. I. . Ann. .312 Ref.c~. c. . . We consider most frequently a first-order system of the form Yj=. English translation. c. y’.. They are called singular solutions (for example. .“. AddisonWesley.. Then we obtain a system of n equations in the variables x.. (Collected works I.Y. Consider ci. .“) = 0 defines an timplicit function y(x) of class C’ satisfying the condition y(x.320 Partial Differential Equations).ci’.) be a function of the n + 2 variables x. . y’. . The highest order of derivatives in the left-hand sides is called the order of the system of differential eaquations. and weak inaccessibility are 1172 References [l] N. c. . . .. or tcomplex analytic. A set of n functions y. we say that the degree of (1) is m. 313 (X111.references to 381 Sets. English translation. 2 .Y. Pergamon. . B. we omit the word “ordinary. .lstants is values co called a particular solution. c. . y. Theorie des ensembles. ch. if f is a linear form in y. which are equations similar to (1) but in which y is a function of two or more variables x1. We write y’. Conversel:y. General Remarks since we are concerned only with ordinary differential equations.. . Ind. Springer. Here we assume that the function f in the left-hand side of (1) is a real (complex) function of the n + 2 variables x. . Then the equation cp(x.) = 0 and differentiate cp n times with respect to x.. c. a solution of an nth-order differential eqLLation can usually be written in the form dX. treal analytic. Systems of Differential Equations A set of n differential equations containing n unknown functions y..cT.. of x is called a solution if the functions satisfy the given system of differential equations.Y. . 1243b. Cantor. n.O)=O. of class C’ in a domain D. Actualites Sci.252 Linear Ordinary IDifferential Equations.fi(x. . Let CP(X.. . . c.Y.. of a variable x is called a system of ordinary differential equations. y”“. which contains n arbitrary constants cl. y(“) and is defined in a given domain of R”+’ (C”+‘). i=l . c. Table 14. . y are complex. (4 Let x be a real (complex) variable and y a real (complex) function of x.C~. Szeged. y’. ) y’“‘) = 0 (1) (which holds identically with respect to x). 291 Nonlinear Problems).. Elements de mathematique.. . . A relation among x. . y(“) that is of degree m with respect to y(“). y. A function y = F(x) that satisfies (1) is called a solution of (1). In this article. [2] G.. the term differential equation refers to an ordinary or partial differential equation. y’. A solution containing n arbitrary constants of the form (2) of an nth-order differential equation is called a general solution.. . 4 differential equation that is not linear is :said to be nonlinear (. 1961.. cy. Assume that y = F(x) is a differentiable function of class C” if x... Theory of sets.. and a solution cp(x. cp(xo. y’. . . y.2) Ordinary Differential Equations A. Ordinal Numbers inaccessibility equivalent. . . 1915. Ordinary differential equations may be contrasted to partial differential equations.) = y.. 3.~3. 49 (I 897). c. Beitrage zur Begrtindung der transtiniten Mengenlehre II.~o. . Appendix A. y. x2. c. y. . c. 207-246. . . y are real. Bourbaki. .). is called an ordinary differential equation for the function y = F(x). Acts Sci. Some equations admit solutions that are not particular solutions. .c”)=o.

b) be an open interval and D a domain of R”. methods of successive approximation. 316 (Initial Value Problems).).. are considered to be functions of ( y . .c. .. .cp.e. . Appendix A. Hille.. . .. Ordinary differential equations..) = f. . . . power series methods (assuming that the solution can be expanded in a power series C a.. (yf. c. (4) represents a family of curves of class C’ in the x.. .. Historically. . A solution containing n parameters analogous to (4) is called a general solution of (3). E 1 we have a((pi. c1 . .references to 316 Ordinary Differential Equations (Initial Value Problems). Let Yi=(Pi(X~C1~~~~9Cnh i=l. As may be imagined by the interpretation in this section. . . (4) be functions of class C’ defined for (x. y:) gives the speed and the direction of a stationary flow in R”+l at each point).1). 126 Dynamical Systems). (. in particular theorems on the existence and uniqueness of solutions. we find a stationary flow of which the speed and the direction are given at each point).. . however.y. References [l] W... is called the initial value problem (316 Ordinary Differential Equations (Initial Value Problems)). c. . yi) is the tangent vector (in the terminology of physics. The problem of finding this solution.). yi)~ 3(x..y. i. . . . if a solution with a property A is given. [3] E.(x.. . . .. Methods of Integration We have different methods of solving differential equations..)#O in D. satisfy a system of differential equations of the form (3).431 Transformation Groups. .)(i=1.2 ).-space of D under the mapping yi = cpi(x. C. for some special types of differential equations (. .. and a solution obtained from a general solution by giving particular values to the n parameters is called a particular solution.. y. Kaplan. . S. . are real. yt) of 3(x. . numerical methods.y.. . yr. finding explicit solutions of various kinds of differential equations has been the main object of the theory. We assume that for each fixed x.space R”+’ containing n parameters c1 . y. .e. . a general system of equations can be transformed to a system of the form (3). methods using tLaplace transforms or tFourier transforms.)E I x D. = qi(x. Recently. C. . .. . cl. then (1) is equivalent to a system of equations of the form (3). ~7. . n) for a fixed x0 E 1.~. c. Table 14. .. . there exists in general one and only one particular solution passing through the point (x. Table 14. and we have y. has been recognized. fi = y. 1968. Lie gave theoretical foundations for this method by using Lie transformation groups (. . (~7. . the solution of (3) for which yi(x. c1 ..) be the image in the y. Addition-Wesley.). Wiley. Linear Differential Equations differential equa- D. Hartman.2. . . c. . and let 3(x. [2] P. Lectures on ordinary differential equations. . for which (y. . 4.. . 1964. . .1173 314 A ODES (Asymptotic Behavior of Solutions) Ifwe put y=y.(x. where fr = y. . for example.5) Ordinary Differential Equations (Asymptotic Behavior of Solutions) A.. . This method is useful A system of linear ordinary tions can be written as x’ = A(t)x. . y.. and finding its coefficients).~(“-~)=y~ and solve (1) with respect to y(“) to get y(“)=f. To solve differential equations by a finite number of integrations is called the method of quadrature. . tperturbation methods. ....“) for x~EI.. . 314 (X111.... . Addison-Wesley. In this way. On the other hand. 1958. fn =$ In an analogous way. . y’=y. (3) can be interpreted as follows: Let I = (a. By solving (3) we find the family of curves of class C’ in R”+’ (in the terminology of physics.i. y.2. etc. . . . topological and analytic studies of differential equations are applied to find their solutions (. . . The Geometric Interpretation Whenx. substituting the series for y in (1).y.. 315 (Boundary Value Problems)... Also .) (i = 1.Appendix A. the importance of qualitative studies. . There are many other methods. y. and if the uniqueness of the solution having the property A and the existence of solutions having the properties A and B can be shown. Then for every x = x0 E I. . (1) . .303 Numerical Solution of Ordinary Differential Equations). then the given solution necessarily has the property B. n.. y. For example.111). ..314 Ordinary Differential Equations (Asymptotic Behavior of Solutions). . .) defined in a neighborhood of every point ( yi.(x -a)“. .) = yp for x = x0.y. Therefore (3) is called the normal form of differential equations. ci . Ordinary differential equations.)/a(c.

1.=limI.(t). Suppose in addition that B(t)+0 as t-co. such that for any ci. We begin with the particular case of (I). Poincart: in 1880.. . it suffices to transform the matrix A into a tJordan canonical form. he succeeded in sharpening those results previously obtained. C. at least theoretically. C. For any nontrivial solution x(t) of (4). According to tFloquet’s theorem.314 B ODES (Asymptotic 1174 Behavior of Solutions) solutions of (2). The interval I can always be taken to be O< t < co. If /l(t) is a continuous function oft defined on an open interval I.(t)}.. Kneser. and M. 0. . and the fact that all solutions of (1) tend to zero as t+ co is equivalent to the tasymptotic stability of the solution x = 0. To study the asymptotic properties of the (21 where the matrices A(t) and B(t) satisfy ~“IIA’(s)llds<co andj”IlB(s)llds< co. Hukuhara (. that is. Consider the linear system x’ = A. ck # 0. The following asymptotic properties of a solution x(t) as t+ co are considered: (i) boundedness of limsup t-’ log[x(t)l.(t). J. The study of the tasymptotic expansions of solutions when the coefficient A(t) is an analytic function oft was initiated by H. A. Thus. (3) is transformed into a system with constant coefficients by means of a suitable transformation x = P(t)y.. . Asymptotic Integration Suppose that A(t) is bounded. B. R. Trjitzinsky. Here we assume that A(t) need not be analytic. and those with zero real parts are of simple type. (3) where t is a real independent variable. The theory has been almost completed by W.I. x = (x1. Next consider the linear system x’= [A(t)+ B(t)]x. Then there exists a tfundamental system of solutions of (4). .. where A. but Hukuhara established a unified method of treating the problems arising in these two different types of investigations. Thus all solutions of (2) are bounded if and only if every eigenvalue of A has a real part not greater than zero. by applying a suitable transformation of the independent variable if necessary.x”) is an n-dimensional complex vector function oft. etc.. Furthermore. and the number of distinct type numbers does not exceed n. . Constant Coefficients Coefficients and Periodic A sharp estimate of the term o(t) was given by Hukuhara. and A(t) is an n x n matrix whose elements are complex-valued functions oft. Spith.(t) is a periodic matrix function of period w. The question naturally arises as to how the solutions behave as t approaches either one of the endpoints of I. J..254 Linear Ordinary Differential Equations (Local Theory)). Hukuhara. where P(t) is a nonsingular periodic matrix of period w. Horn. all solutions of (2) tend to zero as t + cc if and only if every eigenvalue of A has negative real part. His work was continued by F. (iii) convergence of solution: lim x(t). Conversely. A. Let A. Perron initiated a new direction of research by weakening the regularity conditions on the coefficients. <p. the corresponding blocks in the Jordan canonical form are all 1 x 1 matrices. C. the question of the asymptotic properties of the solutions. . any solution of (1) is continuously differentiable for t E I.(t)x. the limit p=lim t-‘log(x(t)l exists and is equal to the real part of one of the eigenvalues of A.x. Malmquist. where A(t) is a constant matrix: x’ = Ax. (iv) integrability: Jmlx(s)lPds< co. Then the type number x(x) is finite for any nontrivial solution x(t) of(l). The methods used in these two lines of investigation were originally distinct. then there exists a nontrivial solution x(t) of (4) satisfying lim tf’log Ix(t)1 = p. since the structure of the solution space of (2) is completely determined by the Jordan canonical form of A. On the other hand. Consider the linear system x’=[A+B(t)]x. (ii) boundedness of solution: limsupIx(t)l< co. Let /* 1 < /** < . if at least one eigenvalue of A has real part p. J. The fact that all solutions of (1) are bounded is equivalent to the tstability of the solution x = 0. . This work was continued by J. Lettenmeyer. be the real parts of the eigenvalues of A. the information on the asymptotic behavior of the solutions of the periodic system (3) can be derived from the corresponding theory for the system with con. {xl(t). (4) where A is a constant matrix and B(t) is a matrix function such that j:” IlB(s)ll ds-+O as t-+ co. and others. . We call x(x)= limsupt-‘loglx(t)l the type number (or Lyapunov characteristic number) of the solution x(t). that is.(t) and I. and others in the direction of removing assumptions on the structure of A(t) and extending the domain where the expansions are valid.stant coeflicients (2). ..

. Suppose. where K is a positive constant. y’= F. or IMjk(tZ)-Mjk(tl)(<K for all t.) If limsupS’p[A(s)]ds< 00. x). then limsupRertrA(s)ds<co. have negative real parts. are mutually distinct and Mjk(t) = Re r0 [S(s) .)+0 as 1x1-+0 uniformly with respect to t. . of the so-called Dini-Hukuhara theorem.. and if x(to)$S. When limlx(t)l suppose that F(t. x) = (aQ(t. . x. N. In the nonlinear system x’ = F(t.) is to the origin. then the periodic solution p(t) is asymptotically stable. In the case of general A(t). and that satisfies f(t.. t. Suppose that fx(t. IfliminfRertrA(s) ds > -co. If all the tcharacteristic exponents of the tvariational system of (7) with respect to p(t). (6) where gj is an eigenvector to 5. any solution x(t) of (6) tends to zero as t-co.(t. x). . the boundedness of all solutions of (1) implies the boundedness of all solutions of (5). then there exists an E > 0 such that for any solution x(t) satisfying (x(t.. p(t))y with F. x) is an n-vector function that is continuous for t > 0. it is known that not all solutions of (5) are bounded even if all solutions of (1) tend to zero as t+ 00 and if the matrix B(t) is such that j” IIB(s)ll ds < cc and B(t)-+0 as t -+ co.x)/lxl-0 as 1x1-+0 uniformly with respect to t.or Mjk(t)-+--a as t+oo and Mjk(tZ)-Mjk(tl)<K for t. we have (x(t)-p(t+c)l+O as t+ 00 for a suitable choice of c (asymptotic phase). if A is a matrix such that its k eigenvalues have negative real parts and the other n-k eigenvalues have positive real parts. then Re s’ tr A(s) ds is bounded.+. If an autonomous system x’ = F(x) has a periodic solution p(t) and the corresponding variational system y’ = F. Then for every eventually nontrivial solution x(t) of (6) that tends to zero as t+co. due to L. if at least one eigenvalue of A has real part p < 0.A. then under the assumption that s” IIB(s)ll ds< 00. If all solutions of (1) are bounded. the zero solution x(t) = 0 of (6) is asymptotically stable. Nonlinear Consider equations Differential Equations differential a system of nonlinear of the form xj(t) = exp (S 0 Aj(s)ds [<j+o(l)]. Levinson proved the following theorem: Assume that I. [l/1 + hA(t)JJ . <t2. Then (5) has a fundamental system of solutions {xi(t). and if A has an eigenvalue whose real part is positive. The following inequalities often provide useful information about the asymptotic behavior of solutions of (1): where A is an II x n constant matrix and f(t. Conversely.. if there exists a nontrivial solution x(t) of (1) such that limx(t)=O.)ES. Boundedness and ConvergenCe of Solutions Consider again the linear system (5) satisfying s”IIA’(s)ll ds< co and j”IIB(s)ll ds< 00. . but if there is no such solution.1 characteristic exponents with negative eigenvalues. x’ = Ax + f(t.(p(t))y has n .. of A corresponding D. then all solutions of (1) tend to zero as t+ co.x)/lxl+O as 1x1-+0 and t-co. if limSfp[A(s)] ds exists.n. Then if all eigenvalues of A have negative real parts.)--(to)1 <E for some to and tl. 0) = 0. PC= lim t-‘log)x(t)l exists and equals the real part of one of the eigenvalues of A. Suppose that f(t.n. Suppose that f(t. then the zero solution of (6) is tunstable. x) =(iYfj(t. . (7) where p[A(t)] =lim. then there exists a solution x(t) of (6) such that lim t-’ loglx(t)l =p. E. provided that x(t. However. moreover. (p[A(t)] was introduced by Lozinskii. then 1imRej’tr A(s)ds= -co. Suppose that all eigenvalues of A(t) have nonpositive real parts and that the eigenvalues of A = lim A(t) whose real parts vanish are simple.k). Cesari.l x 1~ A.. x)/ax.(t. x) is of period w with respect to t and has continuous partial derivatives with respect to x. if A(t) is periodic or satisfies 1iminfReptr A(s)ds> -co. . > j=l.II/h. provided that~“IIA(s)llds<cc. then (1) has a solution x(t) with the property that limsuplx(t)l >O. then there exists a k-dimensional manifold S containing the origin with the following property: For to sufficiently large. Then all solutions of (5) are bounded.. Ix(t)1 tends to a finite limit as t-co.)> -Kfor t.(s)] ds satisfy either Mjk(t)+ co as t-t cc and for each pair(j. then all solutions of (1) are bounded.1175 314 E ODES (Asymptotic Behavior of Solutions) be the eigenvalues of A(t) and A = lim A(t). It can be shown that every solution of (1) tends to a finite limit as t+ co. and if limS’p[A(s)]ds= -co. .(t.Mj&)-M. respectively.<t. This result is a generalization.(t)} such that f exists for every solution x(t) of(l).. then for every solution x(t) of (I).). x(t) cannot remain in the vicinity of the origin no matter how close x(t.. In this case. . that (7) has a solution p(t) of period o. x)/ax.

-*[1+0(l)]. and the relations considered are called boundary conditions. such that q(t)< k. If q(t) is negative. >Y’“-” (ai). Let a.1 y y’ .(t)=<?[l x. When y = 1.(t)lxlYsgnx = 0 is also oscillatory. consider (1) Finally consider the nonlinear equation (9) x”+q(t)l~I~sgnx=O. We can consider boundary value problems in the same way for systems of differential equations. and is not oscillatory if q(t)< l/4?. then (8) has a fundamental system of solutions satisfying xl(t)=f’f[l x. Coppel. The problem of finding solutions of (1) satisfying these relations is called a boundary value problem of (1). 1955. (9) is oscillatory Lf q(t) > (1 + E)/4t2 for some t: > 0. When k = 2 and u.. and if q’(t) 2 0. [4] W.(t)= -em’[l +0(l)]. where 11is a positive constant and q(t) is a positive function. if either q’(t) > 0 and lim q(t) < x or q’(t) < 0 and lim q(t) > 0. Equations The aforementioned results can be specialized to the case of higher-order scalar (or single) ordinary differential equations. . and ~“I[q-3’2(s)q’(s)]‘+(1/4)q-5i2(s)q’2(s)l ds<m. Coddington and N. If (9) is oscillatory and if ql(t)>q(t). x1(t)= 1+0(l). third edition. then all solutions of (9) converge to zero as t+m. J. then all solutions x(t) of (9) are bounded together with their derivatives x’(t). Heath.. Wiley. Stability and asymptotic behavior of differential equations. Springer.(t)=l-‘o(l).4) Ordinary Differential Equations (Boundary Value Problems) A. and {” ?~~“~(s)ds= co. ak be points in an interval I c R and several relations between nk values YCaib Y’taiX f. then (8) has a fundamental system of solutions {x1(t). Sci. The number of linearly independent solutions x(t) of (8) satisfying ~“lx(s)12ds< CC plays an important role in teigenvalue problems. 1970. A necessary and sufficient condition for equation (9) with y # 1 to be oscillatory is as follows: l” sq(s)ds = co if~>l. Sur les points singuliers des kquations diffkrentielles 1inCaires. Scalar Differential 1176 Behavior of Solutions) Equation (9) is said to be oscillatory if every solution of (9) that is continuable to t = co has arbitrarily large zeros. >y’“‘)=O. Hokkaido Univ. for t+ co.(t)=ie”[l+o(l)].. Bellman. i= 1. . domaine rCel. then the equation x”+q. S. [3] L. x>(t)= 1+0(l). 1953. In particular.(t)=f’f[l +0(l)]. [6] E. If q’(t)>O. SIAM Rev.g. called a twopoint boundary value problem. IA4’(t)Mm3’2(t)l Q k. (8) References [1] M. has been a main subject of study. W. McGraw-Hill. the problem. e. xZ(t)=em”[l x.. detailed and deep results have been obtained for second-order linear differential equations of the form x”+q(t)x=O. [S] P. Fat. j” q-1’2(s)ds = c(j. a2 are the endpoints of I. lim q(t) = xz and either q”(t)>0 or q”(t)<O.. (I) 2 (1934). Stability theory of differential equations. 315 (X111. +0(l)]. Then x(t) and x’(t) are bounded for every solution x(t) of (8).. Levinson. Hukuhara. if j”lq(s)+ 1 Ids< a. +0(l)].11ds < co. [2] R. then (8) has a fundamental system of solutions x. 17 (1975).(t)= +0(l)]. Ordinary differential equations. and that 1[x] is of +limit circle type at infinity if q(t) > 0. x. [7] J. It is known that the ordinary differential operator ![x] =x” + q(t)x is of +limit point type at infinity if there exist a positive function M(t) and positive constants k. x2(t)=. x.M(t).314 F ODES (Asymptotic F. 1964. x2(0= Cl +dl)l. Suppose that q(t)+c>O as t+m and j”Iq’(s)I ds < m.. then all solutions of (9) are bounded. 1388. k. On the generalized EmdenFowler equation. General Remarks equation in the real Consider the differential variable x f(x .. . then (8) always has both bounded and unbounded monotone solutions. 339% 360. 1965. Asymptotic behavior and stability problems in ordinary differential equations. Wong. x2(t)} satisfying.~“sYq(s)ds=~ifO<y<l. k. The same is true if q(t) is a positive periodic function of period w such that wfiq(s)ds<4. A. . McGraw-Hill. Theory of ordinary differential equations. and if 1” 1q(s) . Cesari. Much sharper results can often be derived through direct analysis of scalar equations themselves. If J”slq(s)lds< co. -ie-“[1+0(l)]. tMathieu’s equation. Hartman. A.

. b]. fi.}.*[y]=O... If 1= 0 is not an eigenvalue. . and there ex: adjacent zeros of q.y”-l’(b).b].. b] with weight functi We say that the problem (4) is self-adjoint if L[y] = L*[y] and the conditions Ui[y] =O.ly+J Ui[y] =0 are equivalent to y = c G(x. I q. . .n. . .. .~. zeros in a <x <b.. . Given a function f(x) and complex constants yl. m*. n.. i = 1. Y’) . i=1. The boundary value problem where G(x. YCYI =O* i=l . 5. ! defined in an open interval -0 b < co and pk is of class Cnmk. . ..... r are continuous and p(x) > ( [a. b) the tParseva1 equality if S is a function of class C” sal 0. (5) admits nontrivial solutions only for special values of 1. A) such that the conditions L[y] =.a linear ordinary tor L defined by w.. then the system {cp.. h. have the relation G(x. m..1177 315 c ODES (Boundary Differential Equations differential opera- Value ProbIN B.. Consider.. we have J. L. (4) is called a Sturm-Liouville pro1 r are real-valued functions deli and CI. 5. ui” CYI =o. s Ia 5M5) dt.m*. is said to be an adjoint boundary condition of U. The function G(x.i=l.here pk(x) is a complex-valued function of Aclass C”-’ defined on a compact interval a<x<bandp. y...bL[y]pdx= fiyL*[y*]dx. if for any function y of class C” satisfying Ui [ y] = 0.. . Then (i) the eigenvalues ing sequence tending to +co.. the linear boundary value problem defined by LCYl=f(x). y. . .U.. 5. m.+a.+. m*. When f(x) = 0. there exists a unique function G(x.(x) associated with A.by Ui[Y] = f j=l M. (2) is a two-point boundary value problem. uiCYl=“. b]. otherwise it is called inhomogeneous.9n... The boundary value problet order equation (P(4Y’)’ + M4 + W))Y = 0. Y. for example.. yi = 0.m. and any function y* of class C” satisfying Vi* [ y*] =O. . i=l. m*.y’j-“(a)+ t j=l Iv. <.. Such values of 1 are called the eigenvalues (or proper values) of (5).cp.(x)#Oforanyxe[a. . 5. i = 1.. L natural way operators in the H consisting of functions that are grable in a XX < b. m. .f on [a. .(a. I) is called the Green’s function of (5). i=l .. Linear C0nsider. 5) = G(x. The boundary value problem containing a parameter 1 LCYI = lY. Nonlinear Differential Equ: Boundary value problems for 1 ferential equations are very dif sults are obtained only for equ form.i=l.. and hence for expansion f=a. VCYI=Yi. We define a system of linear boundary operators u l.. orthonormal set in the Hilbert consisting of functions that arc grable on (a.390 Spectral Analysis of 01 C... i = 1.... .} i. then the Fourier expansion uniformly to . ( tion q..[y]=O. . For I function G(x. A) of (5) and th tion G*(x. l)f(<)d<. then (5) is equivalent to When the coefficients pO. are equivalent to the conditions U. b). For any value of 1 that is not an eigenvalue. Y” =f(x. aY(4 + PJY’(4= 0. i=l . . YY@) + 6Y’ L*cYl=o. and the corresponding solutions $0 are called the eigenfunctions (or proper functions) of (5). A set of m* linear boundary conditions 7JF [y] = 0.(x) a zero ( (iii) the set of eigenfunctions is set on [a. value problem (3) is said to be an adjoint of boundary uYl=o. are real constants. . . the SI equation Y(X) = a * W. and the gen based on operator theory in H (. Let L*[y] be a formally tadjoint differential operator of L[y]. the problem is called homogeneous. A) = C? the assumption that (5) is selfthe following four proposition has only real eigenvalues whit or countably infinite discrete s functions corresponding to tw values are orthogonal to each is an torthonormal set of eiger that no eigenfunction is linear of {rp.. 5.i=l. &II) of L*[y] =ly. 0).

. j= 1. k #j. b. . . y. Math. the problem has a solution. Ordinary differential equations.uj)(D %$x) fj(x. For example. . . These conditions are called initial conditions. . y’(u) = b’. initial values. .y”) = 0.. . . y.< ok(x)..)) ~0 for yj=pj(x) and o. i=l. n.y. .). y’“-‘1) the conditions y(u)= b. Then a system of functions (yl(x). satisfy (x-aj)(D’gj(x)-4(x.. where i and k are constants. . S. .). Since.) [S] N. Then (6) admits a solution y(x) such that y(a) = A. 1969. y. The statement that a solution satisfies initial conditions yi(u) = bi. .n. under suitable conditions. -co<y’< +co. The following I”& Yt Y’)l c MU +y?. . j = 1. . ... of Math. i=l. Ungar.Y. . A. Aml:r. . y. and B. and w(u)<A<o(a) and o(b)<B<ti(b). If in addition f is an increasing function with respect to y. appears in the theory of fluid dynamics. . 1975.. . Schwartz. (Original in Russian. II. [6] B.(~)~ yk<<wk(x)..).. General Consider equations Remarks a system of ordinary differential . Suppose that the fi(x.Y. Addison-Wesley. . b. b.. 316 (X111. y’“-1’) j= 1. and Q(X) <y(x) < W(X) for a <x .(x)<Y. Y(0) = Y’(O) = 0. the solution is unique. . (Original in Russian... . is reduced to Hukuhara’s problem by a suitable change of variables. u”(X) >f(x. Levinson. Moreover. ~1. The boundary value problem y”’ + 2yy” + 2/l&2 .. n. References [l] E. Introduction to spectral theory: Self-adjoint ordinary differential operators. y’“-‘)(n)= b’“-‘) constitute initial conditions.< b. Hartman. . It is known that if I > 0.. The problem of finding a solution such that yj(uj) = bj. L. y(b) = theorem has been proved: Suppose that f(x. y’. A.. The following result is a generalization of +Perron’s theorem: Let Q~(x). n. Theory of ordinary differential equations. Lectures on ordinary differential equations. 1963. Consider the system of differential equations Y.. y’bbk (x+m). in general. Dunford and J.. If f and its derivatives dflay(‘) are continuous.) as the coordinates of a point space R”+‘. Sot. W(X).315 Ref.~. 1955. then there exists a unique solution of (1) satisfying the conditions yi(u) = bi. satisfy (x . 1964. i= 1. M.. b(“-‘) are initial values. . we can transform a differential equation of higher order into a system of differential equations of the form (1) by introducing new dependent variables.. k = 1. . y(b) = B. 133 E./ n. T. and their derivatives aJ/ay. . means that the integral curve represented by it passes through the point (a. n.Y. Linear operators II..) conditions y(a) = A.(x:l) is a solu- . Coddington and N. we4 o’(x)) and a” <f(x. 1970. The problem of solving y’“‘=f(x y(uj)=bj..ldx=fi(x. the solution is unique.y. . . .3) Ordinary Differential Equations (Initial Value Problems) A. 6... and the values a. (1) >y >y’ . co(x)<y<~(x). Levitan and I. Hille. . b’. A. . y’) is continuous for a < x <b. Sargsyan (Sargsjan). Suppose that the d are continuous. . i= l. called Hukubara’s problem. y.. Cauchy first gave a rigorous proof for the existence and uniqueness of solutions: Iffi. are continuous in a neighborhood Iof a point (a. Hukuhara to the study of singular points of ordinary differential equations. and satisfy gj(uj) < bj < wj(aj).. < 1. . . . . all de& nitions and theorems concerning the system (1) can be interpreted as applying to a higherorder equation. Transl. .))~Oforyj=wj(x)andw.Y. . . .. n. . The problem of finding solutions that satisfy initial conditions is called an initial value problem (or Cauchy problem) If we consider (x. . Wiley. ODES (Boundary and boundary 1178 Value Problems) [4] M. . 1967. then a in the (n + I)-dimensional solution of (1) represents a curve in this space called a solution curve (or integral curve). Interscience. Monographs.=&. $Q(x). for the equation y’“)=f(x. McGraw-Hill. oj(x). 1968. . y. .. Then there exists a solution y(x) such that yj(uj)= bj and gj(x) <y(x) < ai( This theory was applied by M. [Z] P. n. Naimark (Neumark). and the values a. y. b. be continuous and right and left differentiable functions and gj(x) < Wj(x) for a <x <b. . Linear differential operators I. the solution is obtainable by the method of successive approximations.). O’(X)) for a < x < b. then there exists a uni#que solution satisfying given initial conditions. reduces to the initial value problem when the uj coincide.. . j = 1. b. . dy. b.) are continuous for c(Q x < p and wL(x) < y. 1954. . . n. k#j. and that if 0 < 1.. .

We have Perron’s theorem: Let w(x) and W(X) be continuous functions that are right differentiable in CI< x <p and satisfy w(x) <a(x). .) Then for any (a. The Lipschitz condition: Ilf(x. and then define yk(x).thenDisclosedinR and there exists. Consider the scalar equation y’ =f(x. such that a& and (bk-b)/&-a) +f(u. y. Perron’s C..)togetherwith IIyII’=y:+ . z) defined for (x. we say that qz is a prolongation or extension of vi. {(u. E. The solution curve of a nonextendable solution tends to the boundary of D as x tends to any one of the ends of its interval of definition. y. Let R be an open set in R”+‘. ..y. for example y.. Then a necessary and suficient condition for there to exist a unique solution curve of (1) going from any point of D to the right is that there exist a (?-function rp(x. z) such that (x. y)) for llyll =4x).)}.). a curve that reaches the boundary of R. one using Caucby polygons and one using Qixedpoint theorems for function spaces. and its limit is a solution of (1) satisfying y(a) = b. initiated by C. which state uniqueness conditions. If (1) admits at most one solution satisfying a given condition.(x)) satisfied the integral equation just given. LB0 aconstant. y).Iyl<co byR.1179 316 E ODES (Initial Value Problems) tion of (1) if and only if Yitxlcbi+ s II xf. y) is continuous in a domain D of R”+l. There are two methods of proving this theorem.(x). recursively by K(X) = b + cf(x. .A sufficient condition for D: E < x < fi. respectively. we call this condition a uniqueness condition. y) and (x. If Z.(x) = b.f=(f. are known. k = 1. A necessary and sufficient condition for (1) to admit a solution curve going from any point (a. When the fi are not continuous. S(y) < w(x) to possess the property in the statement of Perron’s theorem is given by D’w(x)> S(f(x. . Let S(y) be a continuous tsubadditive and positively homogeneous function and w(x) a function continuous and right differentiable on CL <x < /?. The equations (1) are then written as the single equation Y’ = f(X. the method of successive approximation. there exists a nonextendable solution that is an extension of the solution. every solution curve is prolonged to the right to the boundary of Q. Given a solution of(l). y. Picard.(x)) to be a solution of (1) for the initial value problem y. B. Let y = (pl (x) and y = Q~(x) be solutions of (1) defined in the intervals Ii and I.n. then there exists a solution curve passing through any point of D. Then equation (1) admits a solution satisfying y(a) = b and defined in an interval Ix -al < min(r.f. D a closed set in a. p/M) (existence tbeorem). . and f a continuous function in D. Various kinds of uniqueness theorems. Okamura gave a necessary and sufficient condition for uniqueness: Suppose that f is continuous in D.Yl(x)~. cp(x. and We state main theorems for differential equations in the real domain in Sections C-F and in the complex domain in Section G. g(x) < y < W(X). . Then {y. Suppose that D’@(x) < f(x. (Dfo denotes the tright derivative of 0. The fact that the interval of definition is a 6 x </l can be expressed by saying that if we denote the set co<x<B.y. b)cD there exists a solution defined on a <x < fl and satisfying y(u) = b.y)-f(x. + yi.. We use the vectorial notation: y = (yl. Uniqueness Theorems Continuity does not imply uniqueness of the solution. This method is as follows: We choose a suitable function. .~~~Y”(x))dx~ i=l ..(x)} is uniformly convergent. D. . b) in D to the right is that there exist a sequence of points in D. . we define (YlM .. Z)E D and satisfying the conditions cp(x. .~S(x)). z) = 0 for y = z. y. A continuous function w(x) is said to be a .(a) = bi if (y.T y.-. cl2 and P~(x)=Q)~(x) for x~l.. From this theorem we deduce that if f(x.. .b.. Hukuhara and M.. Perron’s theorem was generalized by M.. Equations Domains in the Real and Complex is one of the simplest.(x. 2 . Moreover.. is often used to prove the existence of solutions. b). z) > 0 for y # z.. and let f be a continuous function defined on D : CI< x < 8. y. E. H. When f is continuous and satisfies the Lipschitz condition..z)ll <LIIy-zll.. . among solution curves going from a point in D to the right. . and that Ilf(x.Y). Assuming the continuity off.(x))dx. Existence Theorems Theorem Suppose that f(x. y) is continuous for Ix-al < r and ((y-b\\ <p.y)ll GM there.g(x)) and D’?S(x)>f(x. Nagumo.

1). b. 4. n. all have complex values.y)J =o(r(x)) as x+c1+0. j = 1. b) and taking y/(x-a) as G. satisfy the same system with the initial condition zj(u) = -JJa. A function satisfying this inequality is called a right superior function of (1) with respect to S(y). Substituting the latter series into both members. (p(x.(x . If (1) has a unique solution. G. and finally that S(f(x. Assuming that f is continuous at (a. Then 5(C) is a continuum of the tfunction space C( [a. then (p(x.b)k and y = C c. and such that there exists a solution curve passing through any point in a <x < . a.z(x)emLIX-‘ldx IS (I .) means (4$/c%+) (x.&J. If f further admits continuous derivatives af/a. we can successively determine the coefficients c. 2) is continuous with respect to (x. proceeds for the scalar equations as follows: Let f(x. j = 1. C. We have C. b. zjk = 8cpj/i?bk. q(x) < y<Cp(x). are also continuous. Theorems stating such facts are called comparison theorems. y) = 1: ujk(x a)j(y . 2) in its region of definition. then (1) has a unique solution that is holomorphic at x = a and takes the value b at x =: a. S(&a)) < w(a) implies S(q(x)) d w(x) for x > a if both S(q(x)) and w(x) are defined. Let C be a icontinuum in D. a. the condition D’w(x)> S(f(x. . I). which says that the intersection of the set of points belonging to the members of s(C) and a hyperplane x = 5 is a continuum. b. we obtain Nagumo’s condition (X-U)S(f(X. j = 1. Caratheodory proved the existence of solutions of (1) under the less restrictive assumption that f is continuous with respect to y for any fixed x and measurable with respect to x for any fixed y. b. ykfb. . We have the . /l]). then ~(x. y)) for S(y) = w(x) implies that w(x) is a right majorizing function of (1).(x. then any solution of y’= F(x. If F(x. y. Equations in the Complex Domain F. y. G(x. and let g(C) denote the set of solutions intersecting C. Then for two solutions pi. This theorem was extended by Hukuhara as follows. y) is continuousandboundedinD:a<x</$ ~~yll<co. a.). Cauchy proved the theorem by using majorant series. n. 2 lc. n. . Peano proved the following theorem: With the same notation and assumption as in Perron’s theorem. . b). Suppose that G(x. . and let y(x) be a solution of (1) such that y(a) = b. b).$.. This theorem can be proved by utilizing the method of successive approximations and fixed-point theorems. y)) for llyll =0(x). Let (p(x..a)“. and zj = a(pj/au. we can derive from comparison theorems general uniqueness theorems. G.O.O)=O. :.. y) such that Y(U) = b for w(a) < b < ~(a). S(y))> S(f(x. Equations Containing Parameters of the solution of We assume uniqueness Y’ = f(x.y) is continuous for a<x<fl and O<y<r(x). Let z(x) be a function such that z(u)=b and Ilzi(x)-J. y. called the method of majorants. wjl = arpj/dil. a solution of y’= G(x. Then ~(x. i.316 F 1180 Value Problems) If the derivatives aflay. A) is continuously differentiable with respect to A. .t~)n of dY M where f is a continuous function of (x.. %=(l-(x-a)/r)(l-(Y-b)/p) 1 satisfying Y(a) = b. . it suffices that D+w(x):>S(f(x. b.atisfy the system These differential systems are called the variational equations of (1). where G?fjlay..(x . y) is a right superior function of (1). 4..Y*))~S(Y. there exist a maximum solution cp and a minimum solution cp of y’ = f(x. a. This method. b. . a.Y. From this theorem we can deduce the Kneser-Nagumo theorem. 1) denote the solution of (2) satisfying y(a) = b. .following theorem: If f is holomorphic at (a..z(x))ll <E(X). by the method of undetermined coefliciems. Suppose that f is continuous and satisfies a Lipschitz condition.). In order for w(x) to be a right majorizing function. Then we obtain lIzi(yi(x)ll <eLixma’ X.. (2) We assume that the variables x. Suppose that f(x. . consider the solution Y = C C.. IJQ of (1) such that S(q.)-f(X. yJ)< Gk Sty. b. y)). Y. and moreover. one of which we state.-Yy. which gives approximate solutions of (1). It was proved by Hukuhara that if C is in the hyperplane x = a.. we have vi EV)~. a. Conversely. Assumingthatlfl<Mforlx-u(<randly-bb(<p. -y2)). satisfy the system of linear ordinary differential equations and the initial condition ODES (Initial right majorizing function of (1) with respect to S(y) if for any solution p(x).I for any n. y) such that y=o(r(x)) as x+cc+O vanishes identically. 1) is a continuously differentiable function of (x. 1=(4. then (1) admits a solution connecting the two hyperplanes x = a and x = p and passing through the boundary of the set of points belonging to the members of g(C).

). x0.u) be a tmeasure space. Benjamin-Cummings. If IIf II* = 1. Hille. and we write {f. is normalized. 1978.21) Orthogonal A. Let (X. b) #O.. [9] R. Akademische Verlag.(X) if and only if the system is total in L*(X). . y = ~(x. 1964. then the solution of (2). Coddington and N.. y. Sot.} eO(X). .)the References [l] E. with 0 < t < t. b). b. If f(x. Univ. Okamura.e. b). If the measure has a tdensity function q(x) with respect to the Lebesgue measure and (1 g) =jxf(x)S(x)q(x)dm(x)=O. x0. Math.(x) with respect to the norm in L*(X).f. if (3) has a solution y = v(t) defined in 0 < t < t. If we can approximate any function f o R arbitrarily closely by a finite linear combination of the f. J. sec. is holomorphic on C except possibly at x = a. ) is said to be an orthogonal system (or orthogonal set). 1) is continuous with respect to (x. then f is said to be normalized. Nagumo. If (cp. Wiley. then { fn} is said to be complete in L*(X). . Then q(x) is holomorphic at a. Mem. y) f 0. and p(a) = b. fi(a. Fat.fn) = 0 for all n implies q(x) = 0 almost everywhere (a. 1955. b. =(q. Orthogonal Functions Systems O<t<t. y). y) and holomorphic with respect to y. such that ak+a and ~(a. Japan. (A) 24 (1941).g)=jxf(x)g(x)dp(x) and the norm Ilf 11 =(Jf)“*. If {f. Ein neuer Existenz Beweis fiir die Integrale der Differentialgleichung y’ = f(x. and let R be a subset of L..)b.) = ye is holomorphic with respect to (x. Suppose that x is a real variable and y is a complex vector. E. b). then y =&(t)) is a solution for 0 < t<t. 24 (1943). y) is holomorphic at (x(t)..} be a set of linearly independent functions in L*(X). 15 (1945). if any pair of functions in the set are orthogonal. I. then the series Czl cnfn(x) is called an orthogonal series.(x)} (n = 1. [8] E.. y.(X-up. [S] 0. [6] M. then the equation y’=f(x. yO) is holomorphic with respect to yO.(x -a)“. Differentialgleichungen reeller Funktionen. Proc. Abraham and J. 1). 1958. If X is a subset of a Euclidean space and p is the tLebesgue measure m.(a. y.(x)} is said to be orthonormal. [4] H. is holomorphic at (a. Ann. [3] P. If(Xg)=O. A). then the solution y =9(x. we define the inner product (l. pt. We have the following uniqueness theorem: Suppose that f(x. [7] M.x.b)=O.. and this solution can be expanded into a tPuiseux series: y= 5 C. and we write { fn} E ON(X) if each f. Phys. Foundations of mechanics. An orthogonal system {f. the analytic continuation of a solution continues to be a solution if it does not encounter any singularity off. McGraw-Hill. satisfying y(0) = b. Kamke. By a ttheorem of identity.)/llfn11*. &). Ordinary differential equations. Ordinary differential equations in the complex domain. Coll.} is complete in L.(n=1.1181 317 A Orthogonal Functions which shows that Z C. Tokyo. then c. Condition necessaire et suffisante remplie par les equations differen- .(X). 1) and holomorphic with respect to (y. yO) of (1) satisfying y(x. Hartman. then (1) has a solution (p(x) holomorphic on C and y(t)=&(r)) for 317 (X. Sur la theorie des equations differentielles ordinaires. Let {f.} E O(X).) = y. x0. and satisfying y(0) = b. The orthogonal set {f. then we simply say that they are orthogonal. a. 1) is holomorphic with respect to (ye. Kyoto.2. If f is continuous with respect to (x.} is total in R. a. {a. If f(x. Sci. satisfying y(x. then we say that f and g are orthogonal on X with respect to the measure p. Theory of ordinary differential equations.. 1930. then ~(x.. If the series converges to q(x) tin the mean of order 2. Marsden. Let {f. If fi (a.} E O(X). v(t)) for 0 < t < t. then the solution ~(x. Sci.2.-Math. . For complexvalued functions A g on X belonging to the tfunction space &(X). A.(x -a)” is a majorant series of 1 c. vol. y) is holomorphic at (a. x0.. VII. Let C be a curve having the point a as one of its ends and q(x) be a solution with the following properties: y. we say that {f.}. second edition.Wecallthec. and x(O) = a. Uber die Lage der Integralkurven gewiihnlicher Differentialgleichungen. Wiley. Conversely. and there exists a sequence of points on C. 1) is holomorphic at (a. and if f(x. A. we say that they are orthogonal with respect to the weight function q(x). Levinson. of Y’ = X’WW)~ Y) (3) tielles ordinaires sans points de Peano. II=0 If f is holomorphic at (a. If a solution q(x) is holomorphic on a smooth curve x = x(t).. 1976.y) admits a unique solution such that y-b as x-a. [2] E.) at (a. ye. Hukuhara. where fi and fi are holomorphic at (a. A). y. Suppose that f= fi/fi. Perron. 5. b). ye.). A set of functions {f. and f. Univ.

then u~pnP-2 < A p (1 .(x) = 0 if x is expandable in two ways. . are Fourier coefficients of (P(X)E L. (1) If {f. b) (for orthogonality of eigenfunctions .(x)1 < M. {J. for any sequence {c. (5) Suppose that {f. as its Fourier coefflcients and such that expansion coefficients or Fourier coefficients of q(x) with respect to {f.(p.e. b) instead of O(X).}.fJf.} spans the same subspace as { 9”). (2) { 1. then c. and (F. l/p + l/p’ = 1. (4) Set r”(x) = -1 or 1 according as the nth digit of the binary expansion of x (0 <x < 1) is 1 or 0. Then {JA(x)y.(x).~+~(x) .27t) (. rvP+l(x). on E. This is called Rademacher’s system of orthogonal functions.315 Ordinary Differential Equations (Boundary Value Problems) B). converges a. .e. then C c. of the orthogonal expansion of a function of cpEL. Examples of Orthogonal Systems We assume that X is a finite interval (a. and set w~(~)=r~~+~(x)r. When the orthonormal system is the trigonometric system.(x)} E O(u.e. For this purpose we set fi(x)=gl(x)/llgl 11.(x)l =GM(const. b) and that functions on X are real-valued.:p<2). (6) If for some positive E we have C lc. this is called the Hardy-Littlewood theorem. n). Then {r. b) or ON(a. {sinnx} EO(0.} by forming suitable linear combinations of the g”. Equality in the Bessel inequality for all (peL. (C. (3) If {f. The factor log* n cannot be replaced by any other monotone increasing factor w(n) satisfying 0 <w(n) = o(log* n) (RademacherMen’shov theorem).e. ..ll~ where cp. is any teigenvalue. The system is not complete. if (C Ia Ip)l/p < co (1 < p < 2). then we have the +Bessel inequality C. then there exists a function q(x) which has the II. If. (4) If the orthogonal expansion of a function cp~L. where {cz} is the decreasing rearrangement of { Ic.317 B Orthogonal 1182 Functions where 1 < p < 2. This procedure is called Schmidt orthogonalization or GramSchmidt orthogonalization. If {g. and conversely. . for any orthonormal system {cp. If the c.). l)-summability a. (5) Let the binary expansion of n be n = 2’1+ 2’1+ .f”(x) converges a. and let y. .(x)l.(x) be solutions of y”(x) + 1. as its Fourier coefficients and such that *lp(x)lqdx$Aq f u.I}.and there are many theoI*ems for this system analogous to those for the 1rigono- .}. of the partial sums s*“(x) (n = 1. Conversely. Tandori proved that if c. If. there exists a function b(x) which has the a. b) such that its orthogonal expansion CcJJx) diverges everywhere. b). we can construct an orthonormal system {f.e. t-c).and Cc&x) converges ta.). In this case C. this is called the Hausdorff-Young theorem.(x)=&4C:Zi (g.Izlog2 n < co. and r.‘O as n+ co. K. I }.(x)} E O(0. cJJx) is called the orthogonal expansion of cp with respect to {f”}. (Paley’s theorem).} EON@. “=l s L1 s If 4 > 2 and x uzqnq-* < co. b) and x cf log’ n < co. then it is t(C.159 Fourier Series). (2) We can construct a complete orthonormal system {f. <v. Rademacher’s system is useful for constructing various counter-examples.. A(x)y(x) = 0 satisfying the condition y. 1).) of its expansion.12-‘< 00. then Cc”f”(x) converges a. We write O(a. ON(X).(X) are linearly independent. sin nx} E 0(0. Then {w”(x)} is a complete orthonormal system called Walsh’s system of orthogonal functions.(x)} EON(X). This system is interpreted as a system of characters of the group of binary numbers. Orthogonal theorem. where 1. When the system is trigonometric. Riesz’s theorem).} such that C jc. (7) If we set s*(x) = sup. n > 2. + 2”~ (vl < v2 < .(x)} eON(u. .j’< //cpI/‘. then //s*II~<A~(CC:~V~-*)~/~ (q>2).2. then I: Ic.“=l lc.(x)} and a function (P(x)EL~(u.} c L.e.} in L. cos nx.(X) with respect to {f.(b) = 0.JO and C c.. (3) Suppose that A(x) is positive and continuous. b) is tsummable by Abel’s method on a set E. Then: (i) If the a. but it is interpreted as a tsample space of coin tossing. Systems on the Real Line C.(x)}. there is a function cpE L. 1C:=1 cJ.@. (ii) Let { uz} be the decreasing rearrangement of { 1 a.e.l’ < co. then (1) {cosnx} EO(0. b) is equivalent to convergence a.(X) (the tParseva1 identity) is equivalent to completeness of {f.(x)} E ON(a..(X) that has the c.(u) = y. l)-summable a.(a..(X). as its Fourier coeffkients.“=. b).*qnq-2 n=1 This is called the tRiesz-Fischer B.L(x)=c~~C4lllcp. are the Fourier coeffkients of q(x) with respect to {f.

Appendix A. 1<1<2”‘. =o. 336 Polynomial Approximation).Appendix A. P. The orthonormal system x:(x) (1 <k < 2”. b). [6] G. This system is complete in L$‘)(a. Table 2O.19 Analog Computation F. This system of orthogonal polynomials is useful in the expansions of approximate eigenfunctions of atoms analogous to hydrogen. Methods of mathematical physics I.. Vorlesungen iiber Orthogonalreihen. Furthermore. Convergence problems of orthogonal series. D.Appendix A. Alexits. These polynomials appear as eigenfunctions of the Schrodinger equation for harmonic oscillators. We orthogonalize {x”} by Schmidt orthogonalization and obtain polynomials p.1/2)/2m) The particular case m = 0 gives the Laguerre polynomials. If we set tl = /J in the Jacobi polynomials. Orthogonal polynomials.VI). Laguerre polynomials are used in tnumerical integrations of a Gaussian type in (0. Orthogonal functions. if LY =p =O. . (3) Setting (p(x)=emXZ (or emXzp)in (-co. Orthogonal Functions metric system. Table 2O.) defined on (a. Furthermore.(x)P. Warsaw.(x)=c.(x) also appear in the best approximation problem (.II.Appendix A. References [ 1) S. Hermite polynomials are special cases of parabolic cylinder functions (.Appendix A.167 Functions of Confluent Type). however. lfk.VI).(x) =x’( 1 -x)p (. Sot. 1959. We call {p”(x)} the system of orthogonal polynomials belonging to the weight function q(x). (k.+l(t)-P”(~)P”+l(x) t-x plays an important role. (6) In the interval [0.(a. then we get the tLegendre polynomials. cc). Several important special functions in classical mathematical physics are given by orthogonal polynomials: (l)Settingcp(x)=(l-~)“(l+x)~(cc>-l.Appendix A. Szegii. revised translation of the German edition of 1960.(x)} is a complete orthonormal system in the ordinary L. Publ. 1/2”). Table 20. [4] F. Springer. 1961. Table 2O. we get x!m = J2m.(x) = ( -ll)“eXZ(d”e-X2/dx”). 11. Kaczmarz and H. Pergamon. Theorie der Orthogonalreihen. although they are sometimes defined in [O. we get the ultraspherical polynomials (or Gegenbauer polynomials) (Appendix A. Table 20. the system (mp. = -3. we get the Chebyshev polynomials T. Interscience. They are also connected with probability integrals and are used in mathematical statistics. [3] R.P> -1) in [ -1. we get so-called orthogonality sum. b) space. Math.V). modulo constant factors (. 1939. it is customary to normalize them as L. Sansone. (4) Replacing the integral by a finite sum -. co). of functions f such that fiIf(x)[*cp(x)dx< In other words.(L)P. 1951). which is defined to be the space ~0.VII) and their application to the mean square approximation . [S] G. (2) If we set q(x) = x4evX in (0. Interscience.1).(x) can be determined so that the sign of the coefftcient of the highest power of x is positive. we get the Jacobi polynomials. velocity distribution functions of molecules in gas theory. co). Steinhaus. (Regarding orthogonal polynomials with respect to a finite sum (. b) and that the inner product of functions X g on (a. Courant and D. l] with respect to q. and if tl = fi = -l/2. Table 20) Suppose that we are given a weight function cp(x)30 (q(x) > 0 a. ~&j(m)g(m) m the definition of inner profor a finite duct.. 1~ m) is called Haar’s system of orthogonal functions. The Haar expansion of the continuous function f(x) converges to f(x) uniformly. Tricomi. 1935 (Chelsea. b) is defined by (Jg)=~~f(x)g(x)cp(x)dx.1183 317 Ref.(x) of degree n.(x) = (e”/n!)(d”/dx”)(x”e-“) (. Colloq. 1955.(x) = cos(n arc cos x). we get the Sonine polynomials (or associated Laguerre polynomials) with appropriate constant factors. and so on. Here the sign of p.e.. Amer. the Christoffel-Darboux formula &P.) Since orthogonal polynomials with respect to a finite sum are often called simply orthogonal polynomials by engineers. [2] G. k/2”) XE((l1)/2”. 11. xE((k-11/2)/2m. associated Laguerre polynomials appear in the solutions of the Schrodinger equation for the behavior of hydrogen atoms.(x)}. 1953. we get Hermite polynomials H. Concerning the convergence problem of the orthogonal expansion by {p. x~((k- 1)/2”. Table 2O. The T. In this case. revised English edition. one must be careful not to confuse these with the ordinary ones. Orthogonal Polynomials (. set If c(is a positive integer m. G. Hilbert.

On the other hand. k=l i=1. When a driving force term q(t) is present in the right-hand side of (2). This phenomenon is called self-excited vibration.vibrations.2 . t) = u(/.. are the parametrically sustained vibrations caused by periodic variation of a parameter of the vibrating system. In this case. C. the prevention of vibrations and the generation of stable sustained oscillations.ns/a). the existence of which has recently been confirmed. As examples of practical applications of the theory of oscillations.) is said to be in free harmonic oscillation if the coordinates can be expressed as xi = 2 Ai. the equation becomes d2xldt2+2edxJdt+n2x=0. and its reciprocal the frequency. (1) -COSCG s which represents do cp(t)@sin otdt > due to . x. .cos(n. the period is 2744 n is the circular frequency.. Actual vibrating systems contain more or less nonlinear elements.163 FunctionalDifferential Equations). tunnel diodes).Z&/C.. Each of these simple harmonic oscillations is called a normal vibration. Linear Oscillation Periodic solutions of tlinear differential equations have been studied in detail for a long time. The theory of vibrations has its origin in the study of mechanical vibrations.290 Non- . Electric wires and pantographs for use in high-speed electric railways must be designed to prevent unwanted parametrically sustained vibrations. The difference between the greatest and least values of f(t) (globally or in an interval) is the amplitude. t) = 0. other than forced oscillations and self-excited . a parametron is an electric element utilizing parametrically sustained vibration. which give rise to various kinds of oscillations different from those described by the linear theory (. Nonlinear Oscillation u(0. Typical examples are the free vibration of a simple pendulum with small amplitude and an electric circuit composed of a selfinductance and capacity (without resistance). 27c/(~ is called the period. Exactly periodic oscillations are studied in the theory of tperiodic solutions of differential equations. where the number of degrees of freedom is infinite.). sin(knx/l) k is given by a series cos(knnt/l). the solutilsn takes on the additional term B. The solution is given by x = A cos(nt + CC). and the extremal values in the intervals decrease to zero in a geometrical progression with the common ratio u = exp( . Besides being caused by some special kinds of circuit elements (e. Among sustained vibrations.t + a. t) 3x2 ’ If E<O in (2) (negative resistance)> the solution (3) increases in amplitude. too.1 4) Oscillations A. but its nomenclature has been used also for electric circuits. If a resisting force proportional to the velocity is acting. investigations concerning the free oscillation of the earth.e fundamental vibration (corresponding to k = 1) and simply harmonic motions of frequencies equal to multiples of the fundamental frequency. The tperiod of a solution f(t) is called the period of the oscillation. . Perhaps the simplest case of such an oscillation is represented by the differential equation d2xJdt2+n2x=0. Here the amplitude is A. x becomes zero at a fixed interval n/a. resulting i:n an automatic generation of oscillation. a=Jn2-Ei (3) n>E.. A vibration or oscillation is a phenomenon that repeats periodically.318 A Oscillations 1184 318 (XX. such a situation often occurs when the vibrating system has time delay characteristics (. the forced oscillation where the restitutive force is proportional to the displacement from the equilibrium position. which is just the superposition of tb. A system of m tdegrees of freedom (x. This is called harmonic oscillation or simple barmanic motion. either exactly or approximately. we have the vibration of a string: wzn2 at= i?=u(x. As a limiting case. General Remarks The solution 1 A. whose solution x= AC”‘cos(ot+cc). so that a small disturbance is amplified. (2) is not periodic.g. This phenomenon is called damped oscillation with damping ratio u and logarithmic decrement log u = . m. in engineering. we mention. and x is the initial phase. and in geophysics. However..

1949. For example.) Also . Press. 1961. . 1945). the transition from one to the other taking place abruptly. Ordinary differential equations. (Original in Russian. This is called relaxation oscillation. 290 Nonlinear Oscillation. Addison-Wesley. second revised edition. Pontryagin. 1962. E.references to 287 Nonlinear Lattice Dynamics.1185 318 Ref.) [3] L. 1937. 291 Nonlinear Problems. ChaIkin. Macmillan. II. References [l] Lord Rayleigh. two nearly stationary states occur alternately. Princeton Univ. The theory of sound. 1894. d2x/dt2 1 -x2) dx/dt +x = 0 (E > 0) represents a stable sustained oscillation such that for large values of E. Theory of oscillations. Andronov and C. (Original in Russian. Oscillations E( linear Oscillation). [2] A. I. S. A. 1896 (Dover.

.

.

Paradoxes of the Continuum The problem of the continuum is important in both mathematics and philosophy. Let W= {0. it becomes very hard to develop even an ordinary theory of real numbers (. M cannot be an element of M. 1952. it became popular in set theory.. N. This is contradictory. at each moment the arrow stands still. is not equal to the nth digit of the nonterminating decimal fraction representing the real number defined by the nth expression in the last-described enumeration. we obtain an enumeration of those which do. If we adopt this theory. for any natural number n. among which the following two are best known (1) Assume that Achilles and a tortoise start simultaneously from the points A and B. by definition. On the other hand. Principia mathematics. This is contradictory. B. From the specified enumeration of all expressions in the English language.3 12 Ordinal Numbers A) of all tordinal numbers.” Then we have before us a de& nition of a real number in the interval (0. A4 must be an element of M. The expressions in the English language can be enumerated by the device that is applied to the usual enumeration of the algebraic equations . This is contradictory. “paradox” and “antinomy” often mean the same thing. But if M is of the first kind. indicates that the definition of a set should be restrictive. Paradoxes in Set Theory (1) The Russell Paradox (1903). by its definition. R < R. l] by means of an expression in the English language.2. so M cannot be an element of M. Russell. “Cretans are always liars. 11. the tortoise advances to a point B. (3) The Richard Paradox (1905). being an element of w is less than Q But s2 is an ordinal number. Kleene. In other words. Russell suggested tramified type theory. . w. swhich has twenty-one syllables. . Hence. Denote the set of all sets of the first kind by M. This real number. This is contradictory. together with the Burali-Forti paradox. Every set is either a set of the first kind or of the second kind.156 Foundations of Mathematics). A contradiction between a proposition and its negation is called an antinomy if both statements can be supported by logically equivalent reasoning. Since the kind of reasoning employed in this paradox is very simple and is often utilized in mathematics.319 A Paradoxes 1188 319 (1. Consider the following expression: “The greatest real number represented by a proper nonterminating decimal fraction whose nth digit. Whitehead and B. by definition. respectively. 1966. The following paradox was given by Berry (1906): “The least natural number not nameable in fewer than twenty-two syllables” is actually named by this expression. When Achilles reaches the point B. Cambridge Univ. 1. (2) The Burali-Forti Paradox (1897). [2] S. Therefore the arrow can never move. then M must be an element of M. References [l] A. [3] E. but since M is an element of M.3) Paradoxes A. There are several paradoxes of Zeno concerning the continuum. . General Remarks A statement that is apparently absurd but not easily disproved is called a paradox. This realization led to the development of +axiomatic set theory. with integral coefficients.. Let R be the ordinal number of B! Then every ordinal number. must differ from every real number definable by an expression in the English language.. In practical use. and any set that contains itself as an element is called a set of the second kind. } be the +well-ordered set (. the tortoise advances further to a point B. Thus Achilles can never overtake the tortoise. C. however. We classify sets into two kinds as follows: Any set that does not contain itself as an element is called a set of the first kind. On the other hand. Mendelson. M is a set of the first kind. If M is a set of the first kind. The Epimenides paradox is a traditional ancient Greek paradox of this kind: Epimenides (a Cretan) said. Van Nostrand. When Achilles reaches the point i?. 1925. (2) A flying arrow occupies a certain point at each moment. if M is a set of the second kind. this paradox. however. To remove this paradox from set theory. Introduction to metamathematics. Introduction to mathematical logic. by striking out those which do not define a real number in the interval (0.” C. Achilles running after the tortoise. second edition. Press. Van Nostrand.

~+.‘. A partial differential equation is called linear if it is a linear relation with respect to z and its partial derivatives.291 Nonlinear Problems). Xl the system of ordinary on the curve x=x(n). j+k<n- 1. y: F(~.ax. the integral hypersurface of the equation is z-fp(x.. X. . At+ W. The definition of a system of partial differential equations is similar to that of a system of tordinary differential equations.k=O. (5) is a linear partial differential equation. and if P.-. P.. the equation &Y)~+%Y)s+ + % C(x. ...P.. differential equation is called a solution of the partial differential equation..Y. y. Example 1.~+. pjk are holomorphic functions of 1 in a neighborhood of 1= 0 and take the respective values x0. .~=a2z~axay.. .P~~. y) of (5) that satisfies aj+kdx. Y=Y(a)v j.. .fn-J is a general solution of (1) (.t=azzlay2 when2 is a function of x and y.z. . If F vanishes for a system of values x0. Obtaining such a solution for a given partial differential equation is called solving this equation. . . +(-l)“Po. and the dependent variable z. x.. x2. x2 -dx.. Characteristic Manifolds We consider a partial differential the nth order of two independent x. pi at 1= 0..dy”-PP.. y = y(L). .) The torder of the highest derivative appearing in a partial differential equation is called the order of the partial differential equation.. x. .. . General A partial equation l>. j+k<n-1 For a system of partial differential equations.. equation variables of y)q + F(x.) XJ’O. .. The same procedure is applicable to solving other systems of partial differential equations.> azz azz G’ax.‘“’ > Inotherwords. (The partial differential equation becomes an ordinary differential equation if the number of independent variables is one. .~on)=O.. where pjr = g$. x.~. . .f. f.+ 1 has an integral then we have . y”. . . R are functions of independent variables x1.. . .. . Usually we write pi for az/axi.x. If PI..y)p p.Section C).. P2. P. x. . . y”.1189 320 B Partial Differential Equations 320 (XIII. and p = &/ax....+x”g=o 1 2 ” is equivalent to solving differential equations dx.. we associate a manifold defined by a real parameter 1.ax. X2.) that satisfies the given partial dXl>X2. its tpartial derivatives. With this equation. For example. x for x1.j+k<n)andis tholomorphic in a neighborhood of this system of values.-iaren-lindependent tintegrals of (2). X (2) .. ...dxdy”-‘+ . Example 2. and by analogy to the case n = 2. then for an arbitrary function a’.dx”#O (7) dx.. . .. . are functions of n independent variables xi.+P”~+R~=0. B. p$(j.. Y) .. X..1 9) Partial Differential A.o. A partial differential equation is quasilinear if it is a linear relation with respect to the highestorder partial derivatives. x =x(n). Find a (6) and consider the following problem: solution cp(x.2. .+P2. .) = 0. >.k=0. . l. if x.. Pjk=Pjk(A)t n-l. xi.n.. We call this problem the Cauchy problem for equation (5). . ... .1. n-l.. and y for x2 when n = 2..p~o=zo... 1 2 ” (4) which is an equation of type (1).iff.1.. . A partial differential equation is called nonlinear if it is not linear (. . and the independent variables x1.PI~. +P”&=R 2 ” hypersurface V(z. . r=aZZlaxZ. Then solving the partial differential equation x. we define solutions in the same manner. 2 = @(jr. and if the quasilinear partial differential equation (Lagrange’s differential equation) =o that involves a function z of independent variables x1. q = &lay..> Equations Remarks differential equation is a functional a2 X”‘Z. . j.~+p2~+. x2. Y) asap =Pjk(l). From this we can obtain a general solution by the method of example 1. A function z = x.x..k=O. Y)Z = G(x.2 .

a. it is said to be null-bicharacteristic.(x. because z. The integral curve (x(t). There are n curves on which the left-hand side of (7) vanishes on the integral surface z = cp(x. . pi). then we can obtain all the solutions by differentiations and eliminations. We cannot apply this theorem when the left-hand side of (7) vanishes at (x0. Cauchy’s existence theorem cannot be applied on the characteristic manifold. Then we get Vk Y. grad q(x)) = 0 holds for x in a neighborhood of S or merely for x E S. 5’) #O. Let (9) be a complete solution: V(x. and (10).Y=Y(A) j. b)/D(x. y”. Classification of Solutions for (x0. Thus (i)-(iii) exhaust all the cases. Evidently P. (9). n.. Such a solution is called a general solution of the partial differential equation (8) ‘of the first order. (ii) When V = 0. k=O. we get a solution that does not contain arbitrary constants.) This is Cauchy’s existence theorem. The zeros (x. we get a solution z of (8) that contains an arbitrary function cp instead of arbitrary constants..320 C Partial Differential 1190 Equations C. (8) A solution of (8) that contains two arbitrary constants is called a complete solution. b as functions of x. av av -+q2=o.... uniqueness of the solution fails. At such a point. The phase function q(x) can be obtained at least locally by using the bicharacteristic strip. y from (8).. denoted by P. These complete solutions can be tr. p$). a. b are constants. #3V/aa = 0. we obtain a particular solution of (8). P. b) = 0. Y. 5’) is simple. 0.. The foregoing considerations can be extended. if this constant is zero. av av aa+ab’P’(L’)=O. y”.. In particular. P. P(l)} (P(i) = { Pjk(l)}) of the parameter 1 is called a characteristic manifold of equation (5). z.. aV/ab do not vanish simultaneously. then we have a unique holomorphic solution of this Cauchy problem in a neighborhood of (x0. whose form is assumed to be b = q(a). in (9). and aV/ab = 0.. Furthermore (x0. Hence we have the following three cases: (i) When a.. we get c3Vab avaa avab aaax+abax=“’ --+-aaay ab a)=o.. 5’) is called simple if grads Pm(xo. or a complex vector 5” # 0 is called a characteristic direction at the point x0 if Pm(xo. we get (9) aXjayk x=x(i). y”.)&!?m First. we consider a partial differential equation of the first order of two independent variables: Fb. y). Differentiating this. y) on these curves.(x. is called the principal part of P. Let S be a hypersurface defined by q(x)=0 with gradcp(x)#O. Therefore we assume that there is only one such relation between a and b. or it may be infinite.(x. (iii) When aI’/&. z. and by obtaining a complete solution of (8) we can get all the solutions of (8). ay (10) (x. b.znsformed into each other by tcontact transformations. Suppose that (x0. D). The number of compleie solutions may be more than 1. d = 0. a and b are constants and the solution z becomes a complete solution. Its homogeneous part of order m. The manifold {X(n)> Y(A). S is called a characteristic surface if P.(x. P.. to the space of higher dimensions R” or C”. b are all functions of x and y. y”. y) vanishes because of ( 11). We associate the values j+k pjk(. j+k<n.). y. avaa (11) The coefficients are assumed smooth and real in R” or holomorphic in C”.(x. Let P be a linear partial differential operator of order m: Eliminating a and b from (9) and (lO). 5) is called the bicharacteristic strip of Pm.. solving equation (8) is equivalent to getting three functions a. pi). If we regard a. we get the original equation (8). These curves are called characteristic curves of (5).grad. to some extent. z of x. 5) are called the characteristic set. ( = . If there are two such relations.. y. A real vector t”( #O). z.. and q(x) is called a phase function. g+pg=o. (Here we use the notation l$ = aF/apjk. t(t)) that satisfies 2 = grad. p. Furthermore. 5) (5 # 0) of P. We call this solution a singular solution of (8). and there may be several solutions through the point (x0. By specializing this function cp.(x. the Jacobian D(a. we get a complete solution. 4 cp(4) = 0. with each point I. This means that there exists a functional relation between u and b. 5) is constant along the bicharacteristic strip. Therefore we can replace (9) and (10) by (9) and (11). If we get one complete solution of (8).. 5’) = 0. (12) If we solve (12) for the unknowns a and z.

this deli- We call this system the characteristic differential equation or Charpit subsidiary (auxiliary) equation of the partial differential equation (8) of the first order. z.. from (13).1191 320 D Partial Differential Equations Moreover. I. . Instead. .. z) on S.an-r+l. as in the case when II = 2. . y. Such a solution is called a general solution of (15).an-. . . A curve on S whose tangents are all generating lines of (T) is a characteristic curve. k..xz.+l take fixed values. If we write We call (13) a complete solution of (15). 4. an-r+l)=O.. . Then we get av ~+Pi~=09 1 av Pi’z? aZ i=l. a. . n .. x.. z. System (18) determines not only x. . a. then we have a complete solution of (15). . fk.an-r+l: jJa. The set of these tsurface elements (x.q=aq~ay. . a. G. .. Such a solution is called a singular solution of (15). . p. . p=acpfax. we generally obtain exactly one relation between x1. . . (ii) When we can eliminate the constants a 1. there exists at least one functional relation among a.a. . .:a. . and (17). due to J. . z) on the integral surface S and the direction cosines of a tangent plane at that point. then the characteristic curve is given by the system of ordinary differential equations: dx dy -=-= P Q dz -= Pp+Qq -dp X+pZ _ -dp y+qz’ (18) (16) Then there exist numbers that 1.2 ). We have the same classification as in the case n = 2: (i) When we regard a. But these general solutions are not essentially different.a.. . This characteristic manifold is considered as a part of the integral surface with infinitesimal width.. The characteristic strip is represented by the equations x = x(n). . A solution z = cp(x. y. .x. av -=o. . . (16). D. Take an equation that contains n-r + 1 arbitrary constants a1. . an-r+l from (13) and (14).. they are contained in the general solutions. then it is a cdmplete solution. y) of a general partial differential equation is called a singular solution if Cauchy’s existence theorem cannot be applied on any curves on the manifold formed by z = cp(x. a2. we get all the solutions of (15) from a complete solution (13) of (15).. and in this case we call it the characteristic strip. p.. y. an-r+l)=O.p=~(l). (15) We can regard equation (8) as a relation between the point (x. .. Therefore the tangent planes at all points of the surface form a one-parameter family. y). . we have surface g=azdx dl and dz=pdx+qdy. u”-~+~. . q) is the characteristic manifold. t) = 0. . n. we obtain r partial differential equations of the first order: nition of a general solution is not applicable since we cannot successfully define a general solution by using the number of arbitrary functions contained in a solution. This is a solution of (15) that contains exactly k arbitrary functions fi. aan-r+l we get a solution that does not contain an arbitrary constant.) I. The tangent plane at a point M on the integral surface S is tangent to this cone (T) along one generating line G of (T).... . aal av .. by eliminating a. . . .. In this case. we now use the following definition... . ax dl I azdY dy dl (19) . an-r+1 as constants. . . We might think that there are n-r general solutions corresponding to k = 1. . if k = n-r + 1..u”-~+~ from equations v=o. .. .) -=o.. (17) Hence.az. y)... .1. an-....+1: V(xl.~=y(l).. Cauchy’s Method j=l. .. A. such I=1 . . . 1. ..r. and z.. . .2 . We assume that there exist exactly k ( < n-r) relations among a. . f2.. . (13) Differentiate this equation assuming that a.. . On the integral z = z(x. r. .. (14) If we eliminate a... . y. In particular. . For the partial equation of the second order F(x.n-r+l. s. r.q=q(l)containing a parameter... (iii) If not all of the i3V/&zi vanish.. They envelop a cone (T) whose vertex is (x. y. . . . z but also p and q. Darboux: A solution of a general partial differential equation is called a general solution if by specializing its arbitrary functions and constants appropriately we obtain a solution whose existence is established by Cauchy’s existence theorem. . j=l. Now we consider the case where the number of independent variables is n.z=z(l).

Generally.x) = q(x) is obtained (at least locally) as follows. Laplace’s equation has a primitive u=-f 1 r ~ ( x+iy Z-r solution E. Then consider a homogeneous partial differential equation f(D.D. if we choose a particular solution u = Y-’ of Laplace’s equation.+~=O. C2. D. .f. . . t2..Jp = cpXr(xO) (1 <j < n). . . by 0. Such a solution is called a primary solution of (20). such as Laplace’s equation. a system of h equations i=l. issuing from (x. .51.. 5) is homogeneous of degree 1 in 5. In particular.(t. . t2. 5’) at t=O be (x(t.. For +Laplace’s equation Au=~. for the equation (0: -0:)~ = 0. 2. (20) We can obtain a homogeneous equation from an inhomogeneous partial differential equation by transformation of the dependent variable. . we consider the solution u=F(O. x)= u(t. . . w becomes the homogeneous partial differential equation (0: .. A basic equation is an equation. is called a primitive solution of the original equation.. u:“‘. . &I= tf -t2t3. +x. We denote the differential operator a/ax. x1.D.) of ah a% ah p+2+~-~p=o’ av we have a solution i a% (22) ‘40.. u =f(cr.. to).f(tr.. then u= [(x-5)2+(y-~)2+(z-1)~‘]-“2 x F(5. Let the solution of the differential equations (23) where r2 = x2 + y2 + z2.0. are functions of x. The equation (0: .. 5. For example. Then specializing . Furthermore. OS). Equation (22) has another primitive solution of the type u=$(t+f. for the twave equation Equation (19) is called the strip condition.. that has a primary solution and a primitive solution...tt> x. f is an arbitrary function.). ll$!.. .) is a thomogeneous polynomial of m independent variables tr.)u = 0 corresponds to the homogeneous polynomial .D.. U(m). 0 Such a solution. 5) + jc tjf<.x. y..)u =0 by the transformation of the dependent variable u = eX”w.320 E Partial Differential 1192 Equations Second. u = F(x. 8) satisfies the equation of a characteristic curve of (22): d5i z- --. Dfw = D. x. The equations (18) evidently satisfy this condition.x1. and y is a particular solution of (22). For example.+. .)u=O. for ewation (20). by Euler’s identity the righthand side is identically zero. For example. A solution of an equation that has the same charactzristic curves as a basic equation can be obtained from a particular solution of the basic equation by integrations and additions.O. <(t. [“)).f(tl.. h.) and u = F(x r -x2).. s). up. The solution ~(t.)=O. x. .. x) is-obtained by quadrature along these curves (characteristic strips) from the relation du ~ = -ftt> dt x..) O. and F is an arbitrary function of Oi. the solution u(t. x0. lGi<n..I)/(x + iy)).. where CJis an arbitrary function. (21) where 0.&. Z) =0 in the interior of the domain of integration... y.e. when f(t. u(2). u(l). OY F. i. which is a specialization of the primitive solution u = r-‘f((z -. Homogeneous Equations Partial Differential > Assume that . . (24) we have a primary solution u = F(z + ix cos c(+ iy sin a). x. ..uy.. for m functions . where c1is a parameter... which is the product of partitular solutions and some function that contains an arbitrary function. there are two primary solutions. . x0.D.D. v> 0 d5 dv 4 is a solution of Au + 47rF(x.. .. ) up. Determined Systems difvari- a2u a*u a2u The general form of a system of partial ferential equations in two independent ables is Fi(X.5. This means that u is constant along the characteristic strip.

+uy. there is a trend to construct a general theory for tdifferential operators regardless of the classical types of differential equations (. L. u(x.uyv.g. Example 1.. An example of a determined system is the Xauchy-Riemann equation L is assumed to be a linear partial differential operator with constant coefficients. We explain here only two of them: the fundamental solution and the local existence of solutions.. b-function. . namely. the existence of a local solution. . ux. G.. a fundamental solution (or kernel) of the Cauchy problem for a hyperbolic equation is called a tRiemann function.. z2 =x2 + iy. Sometimes a fundamental solution of the Cauchy problem for a parabolic equation is called a Green’s function. uw .. For example. General Theory of Differential Operators In recent developments of the theory of partial differential equations. y) a fundamental kernel (or elementary kernel) of L..xl+~y‘y*=o. x) that satisfies LE = 0 (t>O) and E(t.ux2> which can be reduced to ux.. If we put &t. v) = 0 that can be regarded as a solution of this underdetermined system. unique continuation of solutions.y. (For example. A simple example of an overdetermined system is 4 =fk YX u. y) itself is called a fundamental solution. there exists a functional relation w(u. and an underdetermined system if h < m. consider a fundamental solution of the Cauchy problem concerning the future behavior of a differential operator L =. Malgrange proved that any linear differential operator with constant coefficients has a fundamental solution [4]. Sometimes E(x.112 Differential Operators).1193 320 H Partial Differential H. A fundamental solution of the 3dimensional Laplacian A=$+. we take a property that is satisfied by some equation of classical type (e.x.P(a/ax).a/at . Then E(x.=. then we solution (or elementary is a linear differential the equation ux-l&=0. y). Y) = G-Y). which can be further reduced to two determined equations Au = 0 and Au = 0. The Cauchy-Riemann differential equations for a holomorphic function f(z. ux2=vy*. Y).+$ x2 3 is E(x)= -1/47cr.+v. The system is called a determined system if h = m. If we take a fundamental solution (fundamental kernel) E and add to it an arbitrary solution of the equation Lu = 0. an telliptic differential equation) and proceed to characterize all equations that have this property.. then we call this distribution E(x. = 03 which is also an overdetermined system. Also. where 6(x) is the tDirac call E(x) a fundamental solution) of L. Fundamental Equations Solutions u(1). where r=Jm. y). = 0. ucrn)of the independent variables x and y. u. 189 Green’s Operator). Ehrenpreis and B. A necessary and sufficient condition for the existence of a solution of this system is & = f. This freedom of the fundamental solution (fundamental kernel) can be used to construct tGreen’s functions of the boundary value problem of elliptic equations and of the mixed initial-boundary value problem for parabolic equations. uy. An example of an underdetermined system is a(4 4 ~ = u. uy. = 0. On the other hand.y.x) (t>O) and E”(t. y) is a fundamental kernel of L.y. ux..x)=O (t<O).) There are several basic problems in this general theory: the existence of a fundamental solution.=o.. are u. Let L be a differential operator with constant coefficients and E(x) be a fundamental solution of L. an overdetermined system if h > m. for u(x. For example.x)= E(t.=uy.x)I. a distribution E(t.= -%. in general it is a distribution. The fundamental solutions (fundamental kernels) relative to the Cauchy problem are also defined as in this section. the differentiability and analyticity of solutions. thypoellipticity is a property of classical tparabolic and elliptic equations.x) is a fundamental solution (or kernel) of the differential operator L. z2) = u + iv of two complex variables z i =x i + iy. then we get another fundamental solution (fundamental kernel) of L. A Green’s function is a fundamental solution (fundamental kernel) that satisfies given boundary conditions (188 Green’s Functions. A Riemann function actually is not always a function. and the propagation of smoothness.= .. ux2x2 + uy. If a tdistribution E satisfies the equation LE(x)=c?(x). if L operator and E satisfies L-w.=o.vy . that is.=6(x). =f(x. ak Y) If this equation holds. then E”(t. LE” = s(t.ux*y. x).

x)=E(t. Springer.f must be real analytic. l-38. x) (t 20) that satisfies LE=O (t>O). [9] I.) [ 1 l] J. Locally convex spa. and M. Hilbert.) [lo] S. Treves [ 181 and by R. Gordon & Breach. Schechter. Ann. Courant and D. [S] S. then . Lectures on partial differential equations. sec. Linear partial differential equations with constant coefficients. and if k is odd. 1973. Math. 1966. [ 173 L. 1954. [2] R. Holden-Day. Univ.x)=(1/4~t)d(r--t) (t>O). vol. Interscience. Magenes. Actually. III. Methods of mathematical physics II. Springer. 1954. internat. then this equation has no Cl-solution. [ 151 F. Cours d’analyse mathCmatique. If L is a linear differential operator with constant coefficients. A fundamental solution of the Cauchy problem for the future of the 3dimensional wave operator i.x) (t>O). Kyoto Univ. (Note that. then we have a solution LI that is the tconvolution of E and S: u = E *f: On the other hand. If the coefficients of L and f are holomorphic in a neighborhood of this point and if the homogeneous part of highest order does not vanish. CarathCodory. Math. it is not locally solvable. Cambridge Univ. 7 (1938). 1965. Existence of Local Solutions Given a linear differential operator L and the equation Lu =A we have the problem of determining whether this equation always has a solution in some neighborhood of a given point. An example of a smooth linear partial differential equation without solution. Calculus of variations and partial differential equations of the first order. Table 15. Example 2. r=p x1 + xi + xi. 1 (1962)). 1968. 1927. second (edition. [ 141 F. Dunod. Lions..e. Lewy. fourth edition. and (a/at)E(O. Therefore. F. Gauthier-Villars. A fundamental solution for L is given by E(t. Partial differential equations. no solution exists even in the distribution sense. 1953. Springer. I. 1967. Hadamard. 1968. 1969. 1964. Fefferman [ 191.. Friedman. L. Petrovskii.459-510. Springer. l-74.f is a function (or distribution) that is zero outside of a compact set. Lions and E. Bull. [ 121 J. Treves. Beals and C. 1967.ces and linear partial differential equations. (Original in Russian. Bers. a distribution E(t. HGrmander.) For linear differential operators L. Linear partial differential operators. Nirenberg and F.rtieller Differentialgleichungen im Gebiete der nichtanalytischen Funktionen. 1932. Nirenberg and F. a study by L.~)= 6(x). On local solvability of linear partial differential equations I. if I. 1962. This result has been developed and completed by L. Mizohata (J. II.. [S] J. ser. Interscience. G. References [l] E. fast. Le probltme de Cauchy et les kquations aux d&iv&es partielles 1inCaires hyperboliques. Hiirmander gives some necessary conditions and also some sufficient conditions for the local existence of a solution of the equation Lu =f for sufficiently general f [4]. L is locally solvable. Interscience. E is a fundamental solution of L. (Original in Russian. 1. L. Moscou. Pergamon.320 I Partial 1194 Differential Equations the neighborhood of the origin. 1963. Goursat. Lewy proposed the following example [3]: where f is a real function of x3. In . Pure Appl. I. (2) 66 (1957). L. Comm. the results mentioned at the beginning of this section are no longer applicable. Partial differential equations of mathematical physics. E(O. He showed that if this equation has a solution that is of class C’. A.. 1911. III.x)=O. is given by E(t. The operator considered by S. and . John. [IS] L. [ 161 F. John. [ 131 A. 155-158. II. [7] I. 23 (1970). uber das Cau#:hysche Problem fiir ein System linearer pa. Theory of partial differential equations. 1961. Holt. Probl6mes aux limites non homog&nes et applications. the problem of local solvability becomes extremely difficult. 1969.Appendix A. Partial differential equations. [3] H. Press.V). Petrovskii. II. Treves. since the coefficients of L are now complex-valued. Partial differential equations. (Original in Japanese. Treves. [4] L. then there exists a solution that is holomorphic in a neighborhood of the given point (+Cauchy-Kovalevskaya existence theorem). However. Equations diffirentielles opCrationelles et problkmes aux lirnites. = 0 (t < 0) (. Hermann. Mizohata. Sobolev.. Math. 1964.) [6] C. 1971. G. Rinehart and Winston. if j is of class C” but not real analytic. for linear partial differential operators with multiple character‘istics. is even. H. serves as a standard model in this problem.

x) and wk(x) (0 <k <m . we give two examples of initial value problems for tpartial differential equations. consider the initial value problem 2 2 !?+!?+a”=() 2 a2 ay2 az2 values This is called the Cauchy-Kovalevskaya existence theorem (for linear partial differential equations). or Cauchy data. 225-292. As in (II) we choose one of the independent variables as the principal variable and regard the others as parameters. Xi = xi (2 G i G n) if as/ax. x) = (0. We suppose that S is a tnoncharacteristic surface. B. Math. 91(1973).. Y.).. = (as/ax.O) that satisfies L Cul = 44 4.x). initial data. S is thus the initial surface. -= a% anqx) wk(x) on S.y” (t.+.4 If the function w(y. h(x..+v. Ann. D) be a linear partial differential operator of order m: 4x.x)=w. = 0 of two independent variables x and y. Beals and C.(x) are of class C” in a neighborhood of x = 0. x) =(O. that is. s.) # 0 on S. Indeed..4 = W(Y. 4.O). Then there exists a unique solution u(x) of class C” in a neighborhood of x = 0. Let L be a linear partial (t. . In fact.(x). 321 (X111.. We consider the Cauchy problem 4x. O<k<m-1.1) be the functions of class C” in a neighborhood of x = 0 and on S. 482-498. then u = cp(x + y) is a solution of this equation that satisfies ~(0.. also for initial conditions on a hypersurface.. Ann.u.. Conditions to determine initial values are called initial conditions. The Cauchy-Kovalevskaya theorem asserts the local existence of solution when the initial values are of class C”.21) Partial Differential Equations (Initial Value Problems) A. v.WD=. this problem can be reduced to the Cauchy problem (1) by taking account of the fact that h(x. . . Let a(x. Fefferman.. For example. the solution of this problem can never exist in (or even on one side x > 0 of) any neighborhood of x = y = z = 0. If the function q(y) is of class C’. The Cauchy-Kovalevskaya Existence Theorem for a System of Partial Differential Equations in the Normal Form The Cauchy-Kovalevskaya existence theorem (1) is extended for more general systems of . alal DQ=axa.+. (I) Consider the partial differential equation u..e.<m. where the coefftcients ayOy. J. x) are functions of class C” (i. We may consider initial value problems not only for initial conditions on a hyperplane t = a. $(o. Fourier. DMx) = u(x).+cc. = S(X). On local solvability of linear partial differential equations. Cardoso and F. called an initial surface (. .e. When we assign a value a to the principal variable t.Section C).. by the change of variables X. The problem of finding a solution of (1) under given initial conditions is called an initial value problem or Cauchy problem. the initial value problem does not always have a solution. (II) We denote a point of R”+’ (or C”‘t) by x.+v. 24 (1974). Let u(x) and y(x) (0 < k < m .aq’ lLYl=a. Y. s. Inst. -=o.1) are of class C” in a neighborhood of (t. If the functions u(t.1195 321 B PDEs (Initial Value Problems) [19] R. 1 .. y) = q(y). A necessary condition of local solvability for pseudodifferential equations with double characteristics... General Remarks where the coefficients a. Treves. Its characteristic polynomial is First.. but with the initial au 40. [20] F. Hadamard noted that if the initial values are not of class C”. # 0 on S. treal analytic functions or holomorphic functions) in a neighborhood of (t.) #O on S. z) is not of class C” in any neighborhood of y = z = 0. (1) O<k<m-1.. as/ax. respectively. x) of class C” in a neighborhood of (t. (1) where n is the outward normal direction of S. differential operator of order m: Let S: s(x) = 0 be a regular surface (i.1 < m 1 a. x=(x1.. s. then there exists a unique solution u(t.O). axto..) # (0)) of codimension 1. the values of the dependent variables (unknown functions) and their derivatives are called initial values.D)= I. . Ivl=v. x) = (0..

3) showed that if F satisfies the +Lipschitz condition I Fb. Sk-t. .q. S.X. qk)isa real-valued function of class Cz in a neighborhood of x = u. u. For a single partial differential in the normal form D... .. S. x) of class c’” in a neighborhood of (t.y changing variables from x to S. (t.41. and (5) partial differential equations in the normal form studied by Kovalevskaya. xk) does not vanish and b. . . ak) = 0.(x) and then S(x) so that the Jacobian d(S. 411 on the same hypersurface. vol.O) that satisfies the equations and the initial values ikU.si=Si). if F. p. S(a... then there is at most one solution u of (4) of class C’ in a neighborhood of a that satisfies u = q(x) on the hypersurfa’ce S(x) = 0. the equation where x=(x.O.-.. .).. -$CO> -U)= wik(x).x. in a neighborhood of (O. . ---=Fi iitpr t.../ ~ CX.. x) = (0. Taking the other functions S.. = ui that satisfy b = ~(a.+v.I where l~i. yk) and is of class C2.+v. By choosing S.lyi = di at yi = hi. .. . Ynj/L?tYOi. 4’) . . ur. For the uniqueness of solutions of this Cauchy problem. the initial conditions are trans- . .. we change the variables x to s.u. (x). . S are of class C’ and satisfy (5).sr . u = c. S(x. cp(x.x. C.u . . skml (s.~p. yk) is also a function of class C2 such that cp= c.j~m.p. =o. Y.. %. p=(pl.)/ i?(x. = di. u = b.. . . u.. . u... .). Quasilinear Equations of the Second Order Consider weassumethatF(x. y. =O. . We assume that the initial conditions are u = q(x) on S(x) = 0 and u’..... Y.. q. Haar (Atti de/ Congress0 Internazionule dei Matematici. ). .+. . _.v.. .F(x. . . . x. Then there is a solution u of (2) in a neighborhood of x = a and yi = h.)/8(x.O). .. we consider the first order ?u where more general equations of ?u Fx i I... .=. y... . .-. ci = (@/dXi).y... q.)..l ?xin.. . . 0 <k < pi ~ 1) are of class C” in a neighborhood of x = 0. aJ... and pi= i?‘u/axi.._ . x.. and that ‘p(y.x.Ivl=v. When s. A.. We assume that Fi (1~ id m) are functions of class C” with respect to arguments t. pi = ci.. To obtain this result he studied the partial differential inequality (3) Next.) does not vanish.(t.s. . u. &p/c. c2. .Then weget (6) then the solution of the initial value problem for (2) is unique. Bologna. .. 1928. Consider [?P’U. xk). If the functions wik(x) (1 < i . Single Equations of the First Order equation given 1 didm.321 C PDEs (Initial 1196 Value Problems) Suppose that F(x. we obtain a normal form solved for &/as by condition (5) (this condition means that the bypersurface S(x) = 0 is not tangent to the tcharacteristic curves)..< I-‘J>and x =(x1. yi = hi.. .=S.< m.. pk) is a realvalued function of class C2 in a neighborhood of xi = “i. ) Xk) are functions of class C2 in a neighborhood of x. Furthermore.. that satisfies u(u. .(x) and then S(x) so that the +Jacobian a(& S. ... S.-. q.. These facts can be proved in the following way. .. .S.(x). then there exists a unique solution U. . . O<k<pi-1 Then there exists a solution u of (4) of class C2 in a neighborhood of x = a that satisfies u = q(x) on the hypersurface S(x) = 0.dX. . / . yk) = y)( y1 . Sk-.u. x). xkr u. . ..

..Sk-I). t)~ .... Sk-l)). ..“~~ujk”~“~.t)l~C(1+151)L. P PC=1 .. which is more general than differential equations of the normal form.k-1... .. Sk-J= cp(x(O.. (7) Jvl=v. .. . . conditions . N). S) # 0. Value Problems) formed u=cp*(s. .)y”fik(& t) + Cj(& t)..’ This has a normal form if nk = m (k = 1.) Suppose that Q(S.1197 321 E PDEs (Initial to . = $*(s 1. G. au Gzqo3 84... the preceding theory applies to this equation.O)=f(x). . we consider the following simple example of a linear equation.S)f$ a with initial u=cp*(s. . if there is a unique solution for sufliciently smooth initial values that depends on the initial values continuously in a x (27rilJ1 .+. tb j=l. >1 -b* 0 as new unknowns. Thus.q . then the condition is that these functions satisfy the inequalities Iv1”(5. To give such a condition we consider the following system of ordinary differential equations. .O)=g(x) is given by X+C* u(x. who considered the following system of partial differential equations.‘. . . Then we can formulate a necessary and sufficient condition for the well-posedness of the initial value problem for the future (the problem is regarded as specifying a mapping that assigns to the initial values on t = 0 the values u(x.. . (10) .. Its solution satisfying the initial conditions v(x. N’). t) is determined as the value of a linear operator from C(R2) to C(R’) on xt-space. we get another system: qu= acp*las P when so = 0. et sx We take as the space of initial values the ttopological linear space composed of all functions whose derivatives up to a sufficiently large order are bounded on the whole space R” and equipped with the topology determined by the tseminorms that are the maximums of derivatives up to a given order on the whole space. which are given by a tFourier transformation on the x-space of system (8): It is obvious from this expression that if we regard f(x) and g(x) as elements of the tfunction space C(R) of continuous functions of x E R with the topology of tuniform convergence on compact sets. (9) If the tfundamental system of solutions of (9) is vjj)(& t) (i = 1. = qi.N’...a% @=” s for v(x.. if the coefficients are of class C”. or more precisely.+v. (27ti{.““(t)a”o+v’+“‘+*“uk~x~t) a0a. s 1. .c)=~(f(X+c~)+~(X-c~))+~ _ s(Wa. .. . i uj(xt t. . . ... Continuous Initial Values Dependence of Solutions on the First. . + cj(x. Now.sk-l)... . as0 as. -ZQ(S. Then equation (6) added to the initial conditions u = cp*(s.. (8) E. suitable sense. t) for t > 0).. N’. . If such continuous dependence on the initial values is established. . Petrovskii.. j = 1. then v(x. j=l.. The wave equation a% . O<t<T. sk-i). ._ aqo... . (au/at)(x. we say that the initial value problem is well posed (properly posed or correctly posed).. and set au/as. . Systematic research on the well-posedness of Cauchy problems was initiated by I. t) is the simplest thyperbolic equation. ax: +Bj(xlr. where x E R” and 0 < t < T. q. qO=$*(Sl.+v.. N. and we take as the range space a similar space on the xl-space.%-I) at so =0 is equivalent to the system in the normal form of partial differential equations of the first order tg=g ... tb 0 z a II-2 uj(x. (The coefficients are all assumed to be functions oft only.N’.xn9t)9 j=l VO<?lk. taking the derivatives a 0 at nj-1 uj(x.

In this case. Japan. very few results about the global existence of solutions. in general. if the function F in equation (2) in the normal form satisfies the Lipschitz condition with constants A and B that are independent of x. The method of proof of this theorem is as follows: First. however. Mizohata (J. are continuous. For nonlinear equations there are. it is also well posed for the past. then the solution of class C’ satisfying the prescribed initial conditions is unique (Holmgren’s uniqueness theorem. In particular. the theorem does not hold in general. PliS (1954) where all coefficients are of class C”. then even when some of the eigenvalues are complex. Every system of partial differential equations of higher order of normal form can be reduced to a system of the first order of normal form. m = 2.1. and so on. because we have a counterexample due to A. Math.:hey are special types of quasilinear systems. F. with coefficients of class C”. where the a. Let K : Q(X) = 0 be a regular . Shirota.. S. Mizohata. The same method can be applied to nonlinear systems of the first order if . Carleman (1939) about the system: has only distinct roots for 5 ~0. If the characteristic equation of (W. We write u(x.5)+ . are of class C2 and the b. and we can take T arbitrarily large provided that (10) is satisfied.. Calderon showed that Carleman’s result can be extended to the case n > 2 (Amer. 1901). Kumanogo discuss the uniqueness theorem for equations of double characteristics or of parabolic type. where Pj(x. 5) are functions of (x. equation (7) is said to be of hyperbolic type (. y) of order at most k . there is at most one solution whose derivatives of order up to m . <)= ~.1 (n is the number of independent variables) is not tcharacteristic for the given equation of order m.y. An example that is not of a normal form and is well posed is a parabolic equation (. if the coefhcients are of class C”. t) is a homogeneous polynomial of degree j of 5 = (lr. then we get a global existence theorem. Hormander [4] for an extension to the complex coefftcient case. j=I If a linear partial differential equation of the first order of normal form has coefficients of class c’“. y. as the hyperplane on which the initial values are assigned and proved the existence of a solution on &r < Ix1 1<Ed.321 F PDEs (Initial 1198 Value Problems) A. Hence Petrovskii’s theory guarantees global existence of solutions. Calderon proved this except for the cases k b 4. y. then the Cksolution of the Cauchy problem is unique in a neighborhood of x = a. This result can be extended to systems of equations under similar assumptions. respectively. and H. .. Uniqueness of Solutions ik+ 5 qx. if an analytic manifold S of dimension n .5)+h’(x.. if the number of independent variables is 2 and the coefftcients are all real. with h I and h’ homogeneous in 5 of degree m and m 1. 5. we prove the existence for 1x1<Ed and sufficiently small sI by Picard’s tsuccessive iteration method. n = 2. T. If we omit the assumption about the eigenvalues.1 coincide with given functions on the manifold S. D) = &~. 80 (1958)). P..) with real coefhcients and B is a differential operation in (x.a.325 Partial Differential Equations of Hyperbolic Type). G. The proof of this fact relies on the CauchyKovalevskaya existence theorem. Sot.. The uniqueness problem for the initial value problem is in genera1 very diflicult even for linear equations when the coefficients are not of class C”. If system (7) is of normal form and is well posed for the future. In PetrovskiTs theory it is sufficient to assume that the coefftcients in (7) are continuous and bounded.327 Partial Differential Equations of Parabolic Type). This is the condition obtained by Petrovskii. their derivatives are Holder continuous. bpy = 0.) are all distinct. Therefore. Moreover. and the coefficients of B are bounded and continuous. where a certain topological difficulty arises.. there is only one solution for the original initial value problem with continuous partial derivatives up to the order of the equation. there is at most one solution of class C’ for the initial value problem. He proved that if the eigenvalues of the matrix (a. For example.u~(x)Da be a linear partial differential operator of order m. .. Then we regard x = E. Math.(x)5”=h(x. 11 (1959)) succeeded in obtaining the proof for the exceptlonal cases. y) of class C’.. Construction Expansion of Solutions by Asymptotic Let a(x. then we have a result of T. S. J.. We assume that the coefficients of 4(x.()lk-j=O. Consider the following linear partial differential equation of the kth order: where C and L are constants independent of 5. See L..

. . Sci. R. but the initial values W&C)(0 6 k < m . and it can be uniformized (Leray. . D) satisfies Levi’s condition (a(~.)=x2. US. . We assume that K is a simple characteristic. D.e. Let S : s(x) = 0 be a regular surface and T be a subvariety of codimension 1 of S. . j = 1. Next we consider the Cauchy problem (1) when the initial surface S (x1 = 0) is noncharacteristic... j=O. D) is well decomposable). Then the equation a(x.(x). i. but characteristic on T. D)u(x) = 0 (Mizohata.)=$ Ll C”jl =ki2 LC”j-k+Ilt where (13 + h’k 4 and L. Math. Math. Then the solution u(x) is ramified around a characteristic surface K that is tangent to S on T.. Mizohata constructed tnull solutions. . . Leray and by L.T. Girding. the solution in general may have essential singularities along UfZ1 Ki. G. For example. h(x. In this case. Pures Appl. Bull. 85 (1957). . . 0. 60. where fo(t) is a function with a pole or a logarithmic singularity at t = 0 chosen so that u(x) satisfies the initial conditions. In fact. this is solved by transforming this problem into tintegrodifferential equations. When the multiplicity of characteristic roots is more than 1.. if wk(x) (0 < k <m . Lax. Math. 42 (1956).1) has a pole along T. Garding. . ) be a sequence of functions satisfying df. tl. Consider the Cauchy problem (1) of Section A. Now. j=O The coefficients uj(x) are obtained by solving successively the equations u(x) = f a+ Gi(x)lOg i=l Pi(x) + H(x). 0) = 0 has m distinct roots. Pure Appl. 61. ~(0. . Publ. Then the Cauchy problem (1) has a unique holomorphic solution on the universal covering space of V . Then there exist p characteristic surfaces Ki (1 < i < p) originating from T. D) be a differential operator with holomorphic coefficients in a complex domain. put A(t) = tP+m+j/r(p+ m + j) (p>O). 0 obtained by the preceding process. D) is an operator with analytic coeflicients. We suppose more generally that the initial values y(x) are multivalued functions ramified around T. Sot./dt(t)=&(t).()=h.. (1 < i < m) (Y. j(x) by solving successively the equations (13) on each K. See Hamada. Pures Appl.. . Such a method of solution is closely related to the method discussed in this section. S. J. h. . De Paris (J. . Mizohata.ufzl Ki. but if a(x. + mJ. By using this fact.u=O..(x) is obtained by solving the THamilton-Jacobi equation h(x. Kyoto Univ. We factor h(x. 1 (1962). Nat. 1. France. Bull. Suppose that S is noncharacteristic on S . Math.ojj((Pi(X))Ui. If a(x.. . Leray. fundamental solutions and singularities of solutions for hyperbolic equations have been studied (Hadamard [S. Kotake.O) = 0 has p distinct roots (p = m. J. .p”.. Ad. Then there exist m characteristic surfaces Ki: qi(x) = 0 (1 < i < m) originating from T. (2 < k < m) are differential operators of order k depending only on a and rp. where V is a neighborhood of x = 0.. Hamada. T. 51 (1972)). Inst. 24 (1957). . 1 (1962)). Math. By this method of asymptotic expansion of solution. Math. We assume that The solution is with essential singularities along x2 = 0. . 5)‘~ .(x. D. Define u(x) by u(x)=0 for q(x)<0 and u(x)=~~o~((p(x))uj(x) for q(x) >.x. Let a(x.u(O.=. u(O. Proc. J. Lax. Ludwig. . In fact. Then u(x) is a null solution of a(x. cp. 22 (1960).. > where F. 1. D)u(x) = 0 has a formal solution in the form w.(x) (1 < i < m) and H(x) are holomorphic functions in a neighborhood of x = 0 and pi (1 <i < m) are integers > 0. Sci. . even if the initial values have only poles. rp. the Cauchy problem (1) has a unique solution in the form u(x)= 2 fj(cP(x))uj(x). 1. and Leray. . and Wagschal (J. .(x. the solution does not yield essential singularities along urZ1 Ki. 5 (1969). and Leray.)=O. Math. + . . We assume that the equation n.1) have singularities on a regular subvariety T (x1 =x1 = 0) of S. Thus we can determine the coefficients ui.) = 0 and (ah/&(x. q=))#(O) on K. we set u(x) in the form U(X) = C& C. Kotake. 55 (1976)). Res.325 Partial Differential Equation of Hyperbolic Type). Duke Math.(p. C.. In order to obtain this solution.2. 51 (1972)). See J. x2. Kyoto UC. France. 92 (1964)). The Cauchy problem in the case when the initial surface has characteristic points had been studied by J. Let fj(t) (j= 0.x. . this formal solution is convergent.x. Sot. Comm. Wagschal. rp. hi(O. Math. .. J. Math.1199 321 G PDEs (Initial Value Problems) surface of codimension 1.)= 0. Courant and P. . Pures Appl. where hi (1 <i<s) are irreducible polynomials of degree mi in 5 with holomorphic coefficients. This situation occurs quite generally.5. consider the Cauchy problem Dfu-D. the situation is not the same.j(x).

Cambridge University Press. [lo] F. 1973. and by integrating it we get a solution 0(x.. y) describe laws of static or stationary phenomena such as the field of universal gravitation. [6] J.P> 4)= 0. PDEs (Initial References 1200 Value Problems) sume that a problem is mathematcally well posed. [7] C. [l] R. Lectures on partial differential equations. and the steady flow of electricity or heat. John. For ordinary differential equations. P. 1935. In many cases. and there are cases where it is impossible.abolic equations like u. Hilbert. + uYY= 0 for u(x.20) Partial Differential Equations (Methods of Integration) A. but tinitial value problems are not (323 Partial Differential Equations of Elliptic Type). Springer. we as- Fb. -& F. (Original in German. (Original in Russian. Courant and D. (3) and if we find p and q from (1) and (3). Calculus of variations and partial differential equations of first order. 1953. 1965. the electrostatic field. we often obtain the desired solution by first finding the general solution containing several arbitrary constants and then specializing those constants to satisfy prescribed additional conditions. or very difficult. however. initial value problems or mixed problems with both boundary conditions and initial conditions are well posed (. For the rest of this article we explain fundamental and typical methods of integration (Appendix A. 1966. q. 322 (X111. fourth edition. = 0 for u(x. Hiirmander. the solution (i) exists. second edition. [S] J. guess the possible solutions. For this equation.. [ 1 l] J. Lecons sur la propagation des ones et les equations de l’hydrodynamique. By contrast. _ K=F. [4] E. methods of solution are rather specific and are classified according to the types of additional conditions. u.u. 7. Gordon & Breach. By carefully examining problems in physics and engineering. G. Linear partial differential operators. 1963. Goursat..) [3] L. -dp F. Interscience. then du = p dx + q dy is an iexact differential form. II. we usually obtain many well-posed and important problems. 8. Interscience. for thyperbolic equations like utf . 1927. =0 for u(x. Gauthier-Villars. Cours d’analyse mathematique. 1949). A problem is said to be mathematically well posed (properly posed or correctly posed) if. . au P=z’ au q =3 (1) we consider a system of ordinary differential equations called tcharacteristic differential equations: dx dy du pFp+qF.325 Partial Differential Equat tons of Parabolic Type). Table 15). 1954. t) which control the change (in reference to time) of the various stationary phenomena. Hermann. the steady flow of incompressible fluids without vortices. Petrovskii. Hadamard. Springer. Dieudonne. 1911. 1962. to find a suitable specialization of these functions so that the given additional conditions are fulfilled. under assigned additional conditions. Elements d’analyse. Y. In these problems. Partial differential equations. telliptic equations like +Laplace’s equation u. usually the data are suf& ciently smooth functions. 1932. Caratheodory.+qF. q) = a. Hermann. In these cases. 4. 1971-1978. y>u. t) and +par. [2] I. part (iii) of the above definition (the continuous dependence of the solutions on the data) follows often from assumptions (i) and (ii). a. Methods of mathematical physics II. for partial differential equations. h) . and verify it directly.) [9] F. Gauthier-Villars.) [S] S. and anarbitrary of this sysconstant G(x. Hadamard. In the formal general solution of a partial differential equation we have arbitrary functions instead of arbitrary constants.u. Le probleme de Cauchy et les equations aux derivees partielles lintaires hyperboliques. the magnetostatic field. Treves. 1965. The situation is more complicated. Theory of partial differential equations. 1903 (Chelsea. The Lagrange-Cbarpit For the partial order differential Method equation of the first The methods of integrating partial differential equations are not as simple as those for tordinary differential equations. second edition. (Original in Japanese. III. y. For example.321 Ref. Mizohata.+pF.’ (2) If we obtain tem containing a in the form at least one tintegral p. Holden-Day. Linear partial differential equations with constant coefficients. 1971. and (iii) depends continuously on the assigned data. General Remarks B.. For this reason. (ii) is uniquely determined. ‘boundary value problems are well posed. u.

Then du = a dx + am1dy is an exact differential form.y.g’(a) = 0. .. we obtain a solution involving an arbitrary function.. by using orthogonal coordinates. For example.. in the equation F(xl. u. In particular. g(u)) = 0 and Q’. this series is a desired solution if the coefficients a.. which is smooth in the circle rz = x2 + yz < 1 and takes the value of a given continuous function g(0) on the circumference I = 1. is continuous in y 2 0. both sides must be equal to the same constant o?.. a. we often obtain satisfactory results. y). n=l By virtue of the uniqueness of the solution of an elliptic equation.Pl. bn can be chosen so that the series converges uniformly for 0 < r < 1 and can be differentiated twice term by term for 0 < r < 1. C.. Separation Superposition of Variables and the Principle of The method of separation of variables applied after a suitable transfbrmation of variables is often successful.xiT (1’) we can use the tcharacteristic differential equations to find a complete solution and a general solution (the Lagrange-Charpit method. we obtain by this process a relation $‘(y)/$(y)= cp”(x)/cp(x). from which we get a particular solution u = ~e+~ sin v(x . we obtain a new In fact..P”)=O... 82 Contact Transformations). when the equation is linear and homogeneous.’ = 1 in u(x. this is the unique desired solution. Here. Such a solution is called a tgeneral solution of (1). Mixed Problems For linear homogeneous equations of hyperbolic or parabolic type. for example. and hence from the original equation we get q = a-‘.. when we have n independent variables x 1. setting b = g(a) and eliminating u from the two equations cD(x. A solution of (1) containing two arbitrary constants is called a tcomplete solution. consider pq= 1. we can use polar coordinates to rewrite the equation in the form Au=u. and by integrating it a complete solution u = ax + a-’ y + b is obtained. The boundary value problem in the preceeding paragraph is well posed. by forming a linear combination of particular solutions that correspond to various values of a parameter v. A general solution is found by eliminating a from u = ax + a-l y + g(a)andO=x-a-‘y+g’(a). For example.sinnB). a solution of uY. and if we have g(e)=?+ z (u. by a superposition consisting of a linear combination and a limiting process). . For example.1201 322 D PDEs (Methods of Integration) = 0 of (1). x. Likewise...x”. which is the tfundamental solution of the heat equation. Concerning the equation uf + u. = u. and coincides with a bounded continuous function rp(x) on y = 0 can be obtained by a superposition of the solution (4) such as au pi=. /? u=L%x+JiTy+p For tlinear equations.0~) containing a parameter v. r” sin no. This name refers to the fact that the function (4) can be used to obtain solutions of the heat equation under some initial conditions. mixed problems fre- .. Since the right-hand side is independent of x and the left-hand side is independent of y. it is often effective to write the solution as a product u = &x)$(y).cosnB+ b. n=l The simplest and most useful method is the separation of variables. by setting u = q(x)+ $(y) we obtain rp’(x)’ +$‘(y)’ = 1 or q’(x)’ = 1 -G’(y)‘.+~+~uee=O r r and apply the method of separation of variables to obtain particular solutions I” cos no. D. Hence it is reasonable to suspect that by a superposition of these particular solutions we can obtain the desired solution: u&y)=?+ $$ (u. From this we get a (complete) solution involving two arbitrary constants t(. . by integrating the solution P2Ycos vx with respect to the parameter v between the limits -co and 00 (namely. Since the characteristic differential equations are dx dy -=-z-cP 4 du 2Pcl -dp 0 =-dq 0 ’ solution m u= e?‘Ycosvxdv=&exp( s-m -g) (Y>O)> (4) we have p = a (constant). tpolar coordinates.u. More exactly. or tcylindrical coordinates according to the form of boundary. + uyY = 0. concerning the boundary value problem for Laplace’s equation Au = u. From the theat equation u.cosnB+b. + @.u. we obtain a new solution (the principle of superposition).. Next.sinn@r”. = 0 that is a function of class Cz in y > 0. .

. Then a solution corresponding to the general case is given by U(x.t)=“z A. (the corresponding y(x) is sin vx). Furthermore. Consider the case where f(t) is the unit impulse function: f(t)= o’ I. x > 0.u. t) (7) satisfying an initial condition u =f(x).sinnt)sinnx. the behavior of the string is described by the solution of utf .. t) that fulfills all the boundary and initial conditions.cp(x)exp( .. the method described in this section is applicable to solving the following nonhomogeneous equation. in more general cases.. but there are some direct methods that are more effective. p(x) = 1.u = Oforx=O.int) (n = 1. t) in terms of the system of eigenfunctions {sin nx}. Here. . the desired particular solution can be obtained by solving the teigenvalue problem (5). l-1. But in the special case T(x) = 1. Such problems can be reduced I:O problems of the first type in the manner described in the previous paragraph. err > 0). (iv) T(O)u(O)= T(l)u(l). ut = 0 for t = 0 is assigned. under homogeneous boundary conditions: (i) u = 0 for x = 0 and x = 1 in the case of two fixed ends. II=1 withv=v. Disregarding the initial condition for a while.andu. These v are called teigenvalues of(S). ut = g(x) for t = 0. I= rc. i. when 7’(x) = p(x) = 1. t) describes the oscillation of a string that is at rest until t = 0 and moves under the effect of an external force represented by f(x. and if we set u .. suppose that we have two fixed ends. (6) which is obtained by a superposition of these particular solutions. t) and B. . we reduce the problem to determining u. If we can determine the coefficients a. t)= into (7). For example. If the uniqueness of the solution of the mixed problem is proved. In this case. . . We consider u(x.iwt). For mixed problems of the second type. b. u’ are finite at x = 0 and x = 1 (regularity condition). we must find the solution of the equation =f(x. nontrivial solutions y(x) exist only for some discrete values of v.(t) (n = 1.v=Oforx:=Oandx=1.iwt).=Ofort=O. problems in which both initial and boundary conditions are assigned. (v) u. when an oscillating string is at rest until t =: 0. ). sin nx exp( . a similar method can be applied by setting u = u(x)exp( .4. classified into two types. and f(x) = “El a. t) be a solution corresponding to this case and vanishing for t Q 0. which characterizes the motion of a string under the influence of an external force f(x. furthermore. t) = 0 (t > 0).vxx =f(x. + a.. it is not necessary to look for any solution other and let U(x. t .u=Oforx=/(a. homogeneous boundary conditions are assigned. Then u(x.>O. (5) under the boundary condition for the first type of mixed problem and the initial condition u= u.. T(O)u’(O)= T(/)u’(l) (periodicity condition).B.2. we find a particular solution fulfilling only the boundary condition by setting u(x. and for t > 0 its right end is fixed and its left end moves subject to an assigned rule.t) for t>O cc 1 (a.. This is called a transient problem. = 0 under the boundary condition ~(0. = 0 for x = 0 and x = 1 in the case of two free ends. the homogeneous initial condition u =: 0.z)f(z)dt. Again. . t) = 0). For the first type. f(x. = f (x.cosnt+ II=. (ii) u.3. u(l. t) for t > 0. For example. ). = 0 for t = 0. where the two ends are tied elastically to the fixed points.(t)sinnx Except when the solution is trivial (i. (iii) -u. y(x) must satisfy (T(x)y’)‘+v2p(x)y=0. b. y(O)=y(l)=O.322 D PDEs (Methods 1202 of Integration) than the one obtained by combining the method of separation of variables and the principle of superposition. These problems are. in vibration problems of a nonhomogeneous string between 0 <x < I. but the boundary condition is nonhomogeneous.(t)sinnx. u(x.+a. . sin nx. y(x) f 0 must hold. The method of separation of variables is applicable to problems of this type also. g(x)= f nb. If we now choose an arbitrary function B(x. t)= y(x)exp ivt.e. ’ t<O. the first being Duhamel’s method. When the external force varies with a harmonic oscillation over time as in f (x. t): 4. we expand the unknown u and the function f(x. t]l= u(x.B(x. t) = . For example. then the series (6) is a solution of the mixed problem in question.. and by substituting u=“gl a.sinnx. the values of v for which functions of this kind exist are only l/2. Problems of this kind often appear in electrical engineering. t) = f (t). quently appear.2. and the corresponding solutions y(x) are called teigenfunctions for the eigenvalues v.e. then u satisfies 4. so that this series converges uniformly and is twice differentiable term by term. t). Also. That is. t>o. particular solutions are sin nx exp int. I= rr. . . .

In the complex Hilbert space.. Sometimes ‘L is also called the adjoint operator of L. Denoting the Laplace transform of a solution u(x. if we substitute these solutions u and v into (8). of ~ sufficiently small radius E with center (x0. consider a boundary problem for L(u)=g+e the transposed operator of L. Green’s Formula and Application Fundamental Solutions of Given a linear partial differential operator L[ul=~a.}+Duu. the operator L* defined above is usually called the formal adjoint operator to distinguish it from the one defined by (L[u]. and choose a tfundamental solution of M(v) = 0 having an adequate singularity at a point (x0. y. Then.+(Cv).dxp. -. P=(PI.y)~+E(x. and set 4% Y) = (w~)lw l/P + &.}+Euv.. p) of this boundary value problem for the ordinary differential equation. .y)~ Lw=A(x.}-u{(Bv). of the circle. (9) where L is a path in c(> a0 parallel to the imaginary axis.)2+(y-yy. we call the operator L*[v] GC( -1)Pl +~~~+P~DP(Zp(x)v) P the adjoint operator of L. v(l.p~+-+pnpXp . for a circle of radius r with center at the origin.Y)~+2B(x.-p%=o axay ay2 with the boundary condition v(O. Then. and s the arc length. For example.. If we can verify that the function u(x. This is called the Bromwich integral. y) thus obtained is a solution of L(u) =f fulfilling the assigned additional conditions. then the desired solution is given by (8) where P=v{Au. Q=v{Bu.. t)=& s L -ePdp.p)=p s0 mf(t)e-p’dt. We can apply this formula for solving a nonhomogeneous equation L(u) =f as follows.+Bu. YX (10) We explain in more detail the specific case where the number of independent variables is p=((x-xx. we obtain an explicit representation of u(x.1203 322 E PDEs (Methods of Integration) m s ue-Pldt=!&@.251 Linear Operators). y.}-u{(Av).P”). we consider the circumference K.+Cu.. D(L) being the domain of the definition of L (. and that satisfies L(u)=f in the interior of the circle (f is bounded and continuous in the interior of the circle) and is equal to a given continuous function g on the circumference C. .y) w +D(x. 0 The second method is based on application of the tLaplace transformation. = 0 by e-p’ and integrate the result with respect to t from 0 to co. y) that is continuous in the interior and on the circumference C. y) is the desired solution.ww -. and the operator t~[~]~~(-l)P~+~~~+P~DP(ap(x)v) P and dD denotes the boundary curve of the domain D. P 4x9 PI E.+(Bu).)2)1’2. that this and only this function u(x. the adjoint operators are more appropriate than the transposed operators. Assume that there exists a solution of L(u) =f satisfying the assigned additional condition. t) by 2..y)u ay and its adjoint partial differential operator a2(Av) + 2 aw4 M(v)=axZ + awd ~ P we multiply both sides of u. 4x.(x)m4 P Dp. yo). p) = 0. under the assumption of uniqueness of solutions.) of D and fulfilling a suitable boundary condition.v)=(u.. . we have v... taking account of the initial condition.‘L*[v]) ay' for which &f(a)=a’“+fi. In this ~ case.) contained in the interior of the first circle. These trans- posed or adjoint operators are often used to represent (at least locally) the solutions u of L[u]=“fI The problem is. In this case. then we see.. ax2 ay2 for all ueD(L). to find a function u(x.u.y)aU+F(x. For the linear partial differential operator &+c(x. n the internal normal of aD at a point of aD.a(w + Fv ax ay ’ we have tGreen’s formula: (vL(u)-uM(v))dxdy If for p = LX + i/3 (tl > clo) we can find a solution v(x.

)ds + s C. Goursat. . Schwartz. by a change of independent variables.. and let r be the distance from x0 to x. Partielle der mathematischen Let K be the n-dimensional ball with radius R. Webster and Differentialgleichungen Hilbert.l)-dimensional sphere).). d’analyse math&maIII.y)u aY2 ay =i QFiii ’ and 4x. center x0. 325 Partial Differential Equations of Hyperbolic Type.). a2/axf. Y) where x=(x1. Teubner. y. As fundamental singularities for =fb% (1) M(~)=!?!-!?! andM(v)=?-??! ay ax2 ay2 ax2 of parabolic type and of hyperbolic type. Partial differential equations. in (lo). [6] A. Physik. (. Ix-Xol>Y-y. Ix-Xol<Y-Yo. If the quadratic form C aij&lj in 5 is tpositive definite at every point x of a domain G. II. In this case.U+c(x.24) Partial Differential of Elliptic Type A. 10. i. By letting right-hand -U(Xo.(x)a. Then for any function u(x) of class C2 we have for n>2. Lions. Gauthier-Villars.. for n=2. this equation (or the operator L) is said to be elliptic (or of elliptic type) in G. G.Yo) E tend to zero. boundary R (an (n . 1966.e. I. enclosed by the circumferences K. 323 (X111. revised edition. we must take those given respectively by @.uv. Y>Y. and C. We apply formula (8) to the domain 0. the first term on the side yields 2n 1 a1ogp --pdO= s 0 2n ap -u(x. gods.(x.y)g+b2(x. is called Laplace’s operator (or the Laplacian).. Courant and D.322 Ref. Hermann. { 0. Methods of Interscience.. 1953. h satisfies M(u) = 0 in the interior of the circle enclosed by C. then we get vfdxdy= s Kr (vu. the equation is transformed locally into the canonical form 2 2 $+‘“+b. .ikY). G. Cours tique. Table 15). Y4Y0. If. For n=2. Equations diff&entielles opCrationelles et problkmes aux limites. to apply Green’s formula it is necessary to find a solution v of M(v) = 0 possessing a fundamental singularity as the logarithmic singularity. Friedman. l/2. a so-called fundamental solution. x.. Fundamental Solutions References [l] R. As stated in the previous paragraph. SzegG. Rinehart and Winston. 327 Partial Differential Equations of Parabolic Type).. a...(x))‘> 0 is the condition of ellipticity.). . denoted by A. Therefore sentation of u. aji = aij.y)(. 1961. mathematical physics. [3] A. 1930. and area S. B. Holt. 1927. 1204 PDEs (Methods of Integration) [S] J. =. General Remarks Equations by the logarithmic (x.(x)--(a..Y. 1962. L. [2] E. Y) = exp(-$fJ$). singularity of u at the point we have an explicit repre- Suppose that we are given a tlinear order partial differential equation second- and it is easily verified that this is the desired solution. [4] L.. The simplest examples of elliptic equations are Au = 0 (Laplace’s differential equation) and Au =f(x) (Poisson’s differential equation) (Appendix A. Springer. The operator x1=. . and vanishes on C.323 Partial Differential Equations of Elliptic Type. 1969. Mathematics for the physical sciences.

‘Jaor’-“(n>. because the integral equation in rp as above has a unique solution expressible by Neumann series if the domain G is small enough (. 5) is called a fundamental solution (or elementary solution) of (1) or of L if u(x)= sG E(x. t. y. Now. where a(t)=det(@(l. The maximum principle is one of the most powerful tools available for the treatment of elliptic equations of the second order with real coefficients. The third method is due to E.<O or if c(x)<0 and f(x) < 0 in (I). we have a representation of u(xO) by replacing Au by f in the integrals just given. n’ = (n .345 Pseudodifferential Operators). 5) is a solution of the equation L[E] = 0 having a singularity at x = 5 of the form -o. the Dirichlet problem for Au = 0 has been studied in detail (. concerning the solutions of the equation Au + cu =0 (c > 0. constant). df(5. r/). w. Let G be a tregular domain in the plane.1205 323 C PDEs of Elliptic Type Thus. A fundamental solution E(x. 5)f(5)& provides a solution of (1) for any foci with compact support. The second is that of J. Next. Hadamard [ 11. c)] =x(x.. is the tBesse1 function of order v. if u is a solution of Poisson’s equation Au=f(x). y) satisfies a tHolder The Dirichlet problem for . there are three different methods of constructing fundamental solutions. c). 5)= To obtain r2-n { -lo. then the solution u of Au =f(x. then the solution of the Dirichlet problem is unique.120 Dirichlet Problem). u(x) defined by then the function then E(x. a solution u does not attain its local negative minimum in G. 5)= ! 10g(~(xi-&)z)l’2=Iog~. In particular. If c(x)<0 andf(x).‘Jaologr (n=2). if we put (~(Xi-~i)z~-n’*=~2-n.)) and r= (Caij(~)(xi-&)(xj-~j))1/2 and (a’j) is the inverse matrix of (a. of L[u] =f(x) we set where f(x. We call the problem of finding a solution u of the given elliptic equation in G that is continuous on G U I’ and takes the assigned continuous boundary values on T the first houndary value problem (or Dirichlet problem). This is called the strong maximum principle (. Furthermore.2)/2. 5) = . ?be&> condition. we obtain the following tintegral equation of Fredholm type in q(x): If we denote the tresolvent ir(x. A function E(x. which uses the geodesic distance between two points x and 5 with respect to the Riemannian metric ~aij(x)dxidxj. a solution na3. where V(x. Thus Levi’s method enables us to construct a fundamental solution locally by successive approximation. n>3.y(x.Y)’ -& KkYi G ss 5.3)oro. and J. <) as a function of the variables x satisfies A V(x. the following relation holds: Writing L[V(x. Levi [Z] and is as follows: Let Let G be a bounded domain with boundary I. n=2. we have the following criterion: If there exists a function w(x) > 0 of class C2 in G and continuous on GU r such that L[w(x)] < 0. concerning the uniqueness of the solution. Roughly.189 Green’s Operator).. From this it follows that the solution of the Dirichlet problem for (1) is unique if c(x) < 0. The first and most general is to use pseudodifferential operators (. w. The Dirichlet Problem n-2 c#.). y) vanishing on I is given by U(X. E.[3. This is important in applications of elliptic equations to geometry. 5) = 0 except at x = 5.4] for Hopf ‘s maximum principle and Giraud’s theorem). C. Consider now the more general case (1).=27FW42) represents a tparticular solution of Au =f(x). If we denote tGreen’s function of A relative to the Dirichlet problem in G by K(x. <)$@)/w” is seen to be a fundamental solution of L. n=2. 5) and put of this equation by where I is the tgamma function.

he showed the following alternatives: Either L[u] =. we say simply that the equation is of elliptic type. or L[u] = 0 has nontrivial solutions vanishing on r (in this case the number of linearly independent solutions is finite)...l))d5d~~y<l ss G Under these assumptions. Assume now that and that f satisfies a HGlder Moreover.f admits a unique solution vanishing on r for every A or L [u] = 0 has nontrivial solutions vanishing on I’ (in this case.. y) + u. Y.. Then we have the following inequality (Schauder’s estimate) for any UEC*+‘(G): II4 is a quasilinear Au =f(x. y. for a is a positive solution u(x). C~. Then p is a solution Pk Y) -+ ss 5.(x) and c(x) are continuous and aij and S satisfy HGlder conditions. y.P1l. u. then there exists a unique solution u(x) vanishing on r. In (l). II.y)l<A. y)~ G.(X> y)K. suppose that aU.P*. y. ‘1) be Green’s function for the Dirichlet problem relative to A. if for any values of u...(x) and c(x) are zero.. y) be a function that coincides with the prescribed boundary value on r.). y) coinciding with h(x.168 Function Spaces). elliptic equation. pij this quadratic form is positive definite. when the b. the number of linearly independent solutions is finite). In fact. we say that the equation is elliptic at u(x).u. v)dtdv {4x. the equation of tminimal surfaces (.. Schauder and others. y)= h(x. are positive constants depending on L. Picard solved this equation by the method of successive approximation [7].(x. q)[ be C.I+lK.G~~1~Il~c~lIlo. 63 (. Therefore we have the following alternatives: Either L[u] =f has a unique solution for any f and any given boundary values on r. {u.): F(xl. G. uy) is also elliptic. Let K(x. Y..L[h]) and vanishing on r. . h/ax.f vanishing on r we set D.. pij = a2 @xi axj. 5..inequality (9) in Section H). More general equations of type (1) have been studied by J.~. p.. Schauder proved. y). Y. 5)~ c x R”.(x. u.G+II4l2CPIJ) +~2ll4o.. y.(x.. let h(x.334 Plateau’s Problern) where H(x. first. 5. Jp-II.. with prescribed boundary values reduces to the problem in the previous paragraph.x. lqh. of L defined as the smallest number > 1 such that (LK+L’(IK. VI + 04 G of the integral equation =f(x> YX Y. . The equation is called quasilinear if F is linear in pij..(x. If. where pi = au/axi.Pn. and cxbut independent of u.< B. 5. and c are Holder continuous of exponent t( (0 < c(< 1) uniformly on G and that r is of class C’+“.~. vb(5. let If(X> Y>u’. Specifically. Assume finally that where IIf IL stands for the norm in the function space C?(S) (.Pijr. u. y) on r satisfies AU = f(X. we define the functions u. E. &lay). U. Let K(x. and c and f satisfy HGlder conditions.Pnn)=o. The inequality (2) is one of the most important a priori estimates in the theory of elliptic equations [S. II and h are continuously differentiable. and the boundary r of G is of class C’. and let the supremum of If(x. aij(x) and f(x) satisfy Hiilder conditions. Now suppose that r consists of a finite number of +Jordan curves whose tcurvatures vary continuously. K... that when the h.323 D PDEs of Elliptic 1206 Type for any (x.. Next.?) + C(X> Y)K(X. K. d = (-1/W YW.(aF/apij)~i~j definite form. (x. and K.j=. let h(x. . the problem is reduced to finding a solution u satisfying an equation similar to the one for L[u] (replacing f by f. To find a solution of L[uj =. (2) 411 <Llu-u’l+L’(lp’-pl+Iq’-ql). Y. Then if we put u = h(x. depends only on the ellipticity constant i. y). y) successively as the solutions of coinciding with h on I-. bi. y)} is uniformly convergent. q) be Green’s function in G and in the region defined by lu-h(x. and the limit 11(x. Furthermore. Moreover. Quasilinear Consider equation Partial Differential partial Equations differential the second-order in u(x. More precisely. x. y. H(x. P’> d-m. 5. p. y) be the tharmonic function taking the assigned boundary values on I-.. 5. 2+a. For example. Y>%P? condition in (x.l <B. <. Starting from u.

Riemann treated the simplest case... y) and 0(x. His method is to estimate the maximum norms of u and of its first derivatives. and their derivatives of order 2 always satisfy HGlder conditions. q and AC .. y. and finally to use the Leray-Schauder fixed-point theorem [ 111. there exists a unique solution u of Au = f (x.~).. assuming the existence of the minimum of J.2BL consider the equation au au =f x. Nagumo (Osaka Math. of the which is a linear second-order self-adjoint elliptic equation. there exists a solution of the Dirichlet problem: A. c(x). ax.e.. For the general nonlinear equation Fb l.<f(x. Thus we have a mapping u-v. y) = u(x. u) satisfies a HGlder condition in ~(x.considered as a curve in xyuspace represented by the parameter of arc length is of class C3+’ (CL > 0). AZ. b. axay ay* where A.$=o. and even quasilinear elliptic systems have been treated in detail in several recent works [S. Let w(x. +2xbi(x)gu+c(x)u’+2f(x)u I dx. then assuming some differentiability condition on aij(x). u. and C are functions of x....y)). ax..... the case Au = 0. N. Suppose that f(x..u.. The proof of this theorem is carried out in the following way: For any function u satisfying lu(x.l~A. y) both be continuous on G U r and of class C2 in G.Pij. p. This result. u.w(x.tg)& n 1. 91. C are of class C*. and h. hi(x). 1 J n where we assume ~u~~(x)~~<~>O.. E.u... u(x.) .x. Concerning the Dirichlet problem for the second-order semilinear elliptic partial differential equation i~~aij(x)~=f(x.. and that Awaf(x. . are limited.u. P. where aij(x) = Sj... . we have the linear equation in v: A[u]$+2B[u. Ural’tseva [9] and the Schauder estimate (2). aij(XThm&.. y. y. y). given a continuous function rp on r such that w < cp< O. Moreover.(x. J. any plane having 3 common points with this curve has slope less than a fixed number A. u) such that wk Y) .. it is known that the solution is unique within the region mentioned above. Y). When f does not contain p and q. Dirichlet problem for this equation is unique. B. A@..< 4x.Pl. > We can obtain the solution v taking the boundary value cp on I-. u. in A. we have the tEulerLagrange equation u-f(x)=O..Y&. lu. Under the following conditions.. any quasilinear . He proved..~) I 1 " au au =f x.P”. i. the existence of the solution of Au = 0 with assigned boundary values. Even when FU < 0 does not hold. to apply a result of 0.Y)l~maxlcpI~ lu. Under the boundary conditions imposed on u. ~1.. J..y)). Finally. 4 by u(x.~.l<A. 6 (1954)) establishes a general existence theorem. B. F.(x.-. Y).. Y) in G. a work by M. Quasilinear equations in divergence form f&"i(x.. (4) elliptic Then. Serrin [lo] treated the Dirichlet problem and established the existence and the uniqueness of solutions for some classes of equations of type (5) containing the minimal surface equation. > ( or more generally. y) < u <0(x. y). ax.BZ > 0. if there exists a function u that makes J minimum. respectively.1207 323 E PDEs of Elliptic Type Furthermore.Pll. Ladyzhenskaya and N.. the conclusion remains the same if we can reduce the equation to this case by a suitable change of variables.~.. Y) < 0(x.. y) = cp on r. Applying the Vixed-point theorem in function space to this mapping. and C. the following method is known. and the boundary value cp along I.u. B.. if X(6JF/apij)5i5j > 0.-.. A. the solution Pd = 0.. ax.y. < 0. equation $. B.u. &+qu.. G is tconvex.W.> ( (5) a% +c2u=o. Relation Consider to the Calculus the bilinear form of Variations replacing u.(x) = c(x) = 0. This method can be applied when G is small and the values of h. we have the desired solution v(x.

and let called Dirichlet’s principle. Green’s formula yields +(c(x)u. under the assumptions that G is a domain of class C’.(x) = . Also. particularly for F=(1+p:+..f. the solutions of the (Ldm= c JG D”f(xP”g(x) dx. in order that there exist at least one solution u of the second or third boundary value problem relative to L[u] =.+.. R.=ug+(p-b)u.323 F PDEs of Elliptic 1208 Type second boundary value problem are determined uniquely up to additive constants. we define the following inner product: The problem of finding the solution u(x) of the equation L[u] =f continuous on the closed domain G and satisfying B[u] = cp on the boundary S of G is called the second boundary value problem (or Neumann problem) when b = 0. Courant.(x) = b. was used by D..-~ 21 Then if the boundary S of G is of class C’ and .dx. cp are continuous.1 ..L it is necessary and sufficient that fvdxsG cpvdS=O. IY[v... Haar. x. then the function that minimizes the integral ^I J= x . . also 121. When u satisfies L[u] =f (MEL.i-a. If F(x.q)+ ua”. uijcos(voxj))2)‘~2.p. j=l i= 1.168 Function Spaces)...). n dx. The Second and Third Problems Boundary Value Let G be a domain in R” with a smooth boundary consisting of a finite number of hypersurfaces. and the third boundary value problem (or Robin problem) when /I $0. and at least one of c and p is not identically 0. lal<m where alnl DE= ax. v0 is the outer normal of unit length at the point x E S. With respect to this inner product.xj)/a. . x..) satisfies C(ZZF/(ipii?pj)~i[j>O (and F has some regularity).xi)[h.(&I. . @dS=(f. u.1. Rado.. and cp and /J’ are continuous [3. Weyl. Assume that G is such a domain. ax?’ Icrl=a. . in boundary value problems. let M be the tadjoint operator of L. c GO.. Hilbert. H. . F. Method of Orthogonal Projection (6) where a=(~~=. and v is the conormal defined by cos(vxi)= -f aijcos(v. In general.. the coefficients of L and f satisfy Holder conditions.pr. ./~X. T. where b=i=. Furthermore.(~~~. s s av Taking on u: where b. let B[u] be the boundary operator defined by where v is any solution of M[v] =0 with the boundary condition B’[u] = 0. When c = 0 and p = 0. 0.(G) whose partial derivatives in the sense of tdistributions up to order m belong to I&(G).Z. (. . This is also an elliptic partial differential equation in u... G. f cos(v. For elements f and g in H”(G).+p~)1’2inthecaseoftheminima1 surface equation.cp). . Here the necessity is easily derived from Green’s formula. Then the second and third boundary value problems admit one and only one solution. .?I..u. p > 0.. the case of tminima1 surfaces) has been studied by A. The theory of +Hilbert spaces is applicable to the boundary value problems in Section F. and others. G... the condition that the solutions must satisfy at the boundary is called the boundary condition. account of the boundary condition . Serrin. . We assume that the boundary S of G is expressed locally by a function with iHolder continuous first derivatives (G is then called a domain of class C’l”). . . ss with the given boundary condition satisfies the Euler-Lagrange equation (of type (4)) (pi = au/Zxi) and the boundary condition as well..) and q(x) is an arbitrary element of H’(G). In general. H”(G) is a Hilbert space. let H”(G) be the space of functions in L. and others to show the existence of solutions for linear self-adjoint elliptic equations. The case where F is a function of p alone (in particular. Nikodym.g 1 ... Giraud used the notion of fundamental solution to reduce the second and third boundary value problems relative to L[u] =f to a problem of integral equations.

and we call such a u(x) a weak solution. for an elliptic operator L defined in an open set G. H.S is the norm in Hk(S). . aatqav + flu = 0. where v is the normal of unit length at the boundary. Now. hence the L2 a priori estimates play a fundamental role: If u E H”(G). In particular.. (7) where 6 and c are positive constants. If t is large. Under these estimates the boundary value problem is said to be coercive. 15-51) and others... we see that u(x)~C~(@ by tSobolev’s theorem [12].l). we can see that u satisfies the boundary condition a au/h + /Iu = 0.S+ Il”llb (9) where K is a constant determined by (L. this boundary condition means that the solutions belong to the closure fib(G) of g(G) in Hb(G) (. if we apply Green’s formula. where the Bj(x..D)u(x) = cpj(X). I-. there exists a unique solution u(x)~Hl(G) [12. In this treatment. Girding studied the Dirichlet problem for strongly elliptic operators [14]. D) are differential operators at the boundary and f and {cpj} are given functions. following Weyl. G). denoting its inverse..t1) is a one-to-one mapping from the domain B(L)= {u~H~(G)~a&@v+~u=O} onto L.168 Function Spaces). we introduce the complex parameter 1 and consider the boundary value problem (L + nr) [u] =L f~ z. Elliptic Equations of Higher Order The differential operator of order m: . ll”llm~K(llLull +. Fujiwara and N. In general. . Giirding’s inequality unfix. Next. In this case. and /?. is a tcompact operator when it is regarded as an operator in L..t for a large t. by replacing c(x) by c(x) . the theory of interpolation of function spaces are also used [lS] (. by G. Now. If we put m=2b.(G). . and L.(G) considered above. b . These problems are formulated as follows: Uul =fM Bj(X. when s> n/2.51). Using the notion of function space. the boundary S. Under certain algebraic conditions (Shapiro-Lopatinskii conditions) on (L. M. . j=l. Shimakura (J. plays an important role.(x)<’ #0 a aujav+pu=O.. and others. This equation can be regarded as an equation in H’(G).(G). m is even. is called a strongly elliptic operator. A.(c). studies 33. there are works by F. In this case. Agmon. Thus. Princeton Univ. In other words. Schechter [17]. Nirenberg [16]. s c>O. we can show that for any f(x)~&(G).)[u]=GJ + sBu(pdS= -(Jcp). z dW>cW.we is called an elliptic operator ({ # 0). L. for a solution of this functional equation. In applications. Pures Appl. . then the solution U(X)E H’(G) belongs to H’+‘(G) when ME H”(G) (s = 0. the problems are treated also in H”(G). {Bj}). which acts on the equation from the left. XES. u(x) is a genuine solution. where L* is the tadjoint operator of L. we have (I+@+ t)G. 49 (1970)) and others. . This means that u(x) is a solution of L[u] =f in the sense of distributions. 1954. and mj are the orders of Bj (compare (9) with (2)). since the solution u(x) contained in L2(G) (hence also contained in g(L)) satislies the equation and the boundary condition. a m Thus the problem is reduced to finding U(X)E H’(G) satisfying this equation for all (PE H’(G).2 . if we take (P(X)E 9(G). E. it can be shown that if we assume smoothness of the coefficients. Math. For systems of such equations. ). Such a treatment may be called the method of orthogonal projection. the boundary value condition is stated as @u/&j =fj(x) (j = 0. if u(x) satisfies J!. the problem is reduced to the displayed equation in L.(G) into H’(G). Press. Rellich’s theorem yields that G. Variational general boundary value problems have been treated by D. is a continuous mapping from L. L*[q]) = (J rp). then u(x) belongs to H”+” on every compact set in G (Friedrich’s theorem [ 1.189 Green’s Operator).[u] =f(x) and f(x) belongs to H” on any compact set in G.. Douglis. If necessary. Thus. we have (u. II .b(=m/% (8) Conversely. if Re. 1. (L .. General boundary value problems for elliptic equations of higher order have been considered by S. a. Browder (Ann. Ilk.13].1209 323 H PDEs of Elliptic get Type if &=. math.iI lIBjullm-mj-(1/2). So we can apply the tRiesz-Schauder theorem (. In particular. Bj.168 Function Spaces). 1. since G is bounded and G.

whatever the boundary values may be.196 Hilbert). .n 6 > 0. we have the inequality . Watanabe (Tohoku Math J. The Unique Continuation Theorem Since all the solutions of Laplace’s equation Au = 0 are analytic. Elliptic the Index Pseudodifferential Operators and In a linear elliptic equation Lu =f.I. Ann. for any weak solution UE H’(G) of the equation Lu = 0 and for any two points x and y in G’. Moser [24]): Let L be of the form Lu= i.(x)~=. Also. The index of noncoercive boundary value problems has also been studied by Vainberg and Grushin. we cannot affirm the unique continuation property for general elliptic equations. Pure Appl. F. x2) moves once around the unit circle in the positive direction. See also the work of K. C. Then. v2) on the unit circle x: +x$ = 1. suppose the vector (v. In particular. then any solution u of the elliptic equation F = 0 is analytic on the domain of existence (1900. P. Hilbert conjectured that when F(x. Aronszajn [20].(x).t)(p=p. Seeley (Topics in pseudodifferential operators. .(x). and others. is equal to 2 .345 Pseudodifferential Operators) is said to be elliptic. and then H.r. PliS(Comm. Heinz. real analytic) in G [19]. Sb. then u(x) vanishes identically in this domain. This fact is also proved by applying +Holmgren’s uniqueness theorem (. Grushin [:22] calculated the index i of the tcoercive boundary value problem for an elliptic operator by showing that i is equal to the index of some elliptic pseudodifferential operator on the boundary. A counterexample was given by A. let G’ be any subdomain of G whose distance from 8G is not smaller tha. U. and such that the ellipticity condition (3) holds at almost everywhere in G with some Ia 1.. Carleman for secondorder elliptic partial differential equations L[u] =0 with Cl-coefficients in the case of two independent variables.s=p.. 14 (1961)). <)I > c(l+l~l)” for all XER” and ltl>c-‘.e. Example [23]: Given a real vector field (vi. The notion of ellipticity can be extended to operators on a manifold.321 Partial Differential Equations (Initial Value Problems)). Analyticity 1210 Type of Solutions K.335375).9) with symbol p(x. V. tharmonic functions (i. it follows that if u(x) vanishes on an open set in a domain. 4. This result was further extended by I. real analytic) in an open set G and that u is a distribution solution in G. was extended to second-order linear elliptic equations with C2coefficients in the case of any number of independent variables by C.= 1. and is particularly useful in the calculation of the +index of elliptic operators.. v2(x) rotates 1 times around the origin as the point x = (xi.. Hence the linear elliptic operators are hypoelliptic (resp. solutions of Au = 0) are (real) analytic in the domain of existence. Bernshtein.237 K-Theory H). analytically hypoelliptic) (. Then the index of the boundary value problem ( $+$ 2 >44 =m. This conjecture was proved by S. Then u is also of class C” (resp. 1968.. 23 (1971)). The unique continuation theorem. The Giorgi-Nash-Moser Result Let us state the following result (J. R. 1 2 x:+x. and finally by N.. p.E. Singer determined the index of a general elliptic operator on a manifold in terms of certain topological invariants of the manifold (. Vainberg and V. cl . and others.21. Calderon [21] and others in the direction of establishing the uniqueness of the Cauchy problem. Hilbert’s 19th problem. provided there exists a positive constant c such that Ip(x. J. M.. Rado.(x). axj 1 j=) f a a. L.M. 5 (47).t=p. [)E$‘. This unique continuation theorem can be extended to linear elliptic partial differential equations with analytic coefficients in view of the analyticity of solutions. v1(4~+.. (1939)). first established by T. of class Lm(G). G. R. E.. The theory of elliptic pseudodifferential operators has been widely applied to the study of elliptic differential equations. Atiyah and I. However.. where aij= aji are real-valued. M. A pseudodifferential operator P(x..112 Differential Operators).) is analytic in the arguments. 101 (1929)). Petrovskii to a general system of nonlinear differential equations of elliptic type (Mat. Miiller.s. y. This research was extended by A.. (.q=p. Lewy proposed a method of extending this equation to a complex domain so that it can be regarded as a thyperbolic equation (Math.r=p. suppose that all thaoefficients and f are of class C” (resp. Math.323 I PDEs of Elliptic I. B. it is to be remarked that even if we assume that the coefficients are of class C”.

WI where f and g are given functions. of the shape of G. Mizohata and R.. Weyl’s method is based on the minimax principle [26]. N. G’. [6] J. . (1) on Z2U&. Press. Press and Oxford Univ. 5) -n’m dS. Hormander (Acta Math. C. N(T) is no more than O(T”im) if L is of degenerate elliptic type (C. Klasse der Siichs. 24 (1907). v. Weinberger. the Dirichlet problem (1) and (12) (with g = 0) has a weak solution UE LP(G) for any f~ LP(G). 257-282. let Z. denote by v(x) = (vl (x).. Bolley and J. Akad. I. +a~. where b(x) is defined by b(x)=-i$ v. Scuola Norm. Kyoto Univ. H. if L is of constant coefftcients. Maximum principles in differential equations. J. Nordin. Piss. F.. at which b(x)>O. Vishik and V. Math. Suppose that the coefficients of L and the boundary I of G are smooth enough. 88 (1936)) studied the behavior of the trace of the Green’s function of zl-L as lzl+ co in the complex plane (. Baouendi and C. 5) is the tprincipal symbol of L and S”-’ is the unit sphere on R”.1211 323 Ref. 121 (1968)) treated the case of compact manifolds without boundary and obtained the best possible error estimate. Mat. Kyoto Univ. The value of b(x) is closely related to the regularity near the point XE I if L is degenerate at x in the normal direction (. Goulaouic.jEl c ” a2aij(X). where G is a bounded domain in R” with smooth boundary. i. 1967. 275-317. L.327 Partial Differential Equations of Parabolic Type G).(x)) the unit outer . Arch. 9 (1969)) P. see also S.. be the sets ofxsI\Z.[27]). it holds that N(T)=CT”‘“+an C =(2x)-” sG dx error term.. Grushin [29]. 1923.Wesetq=p/(p-l). G. Prentice-Hall. The regularity of solutions is also discussed in [28]. [2] E. McKean and I. U Z3 satisfying L[u] u=g =f in G. 10 (1972)). Protter and H. 34 (1969)). M. Math. P. Rend. M. Shimakura (J. 1970.Wealso Put c*(x)=cw~l~+ n abi(x) . Asymptotic Distribution of Eigenvalues I (11) Let L be an elliptic operator on G of order m with smooth coefficients realized as a selfadjoint operator in L2(G) under a nice boundary condition. Arima (J. 38 (1934). Circ. u). Sulle equazioni lineari totalmente elliptiche alle derivate parziali. Formula (10) was at first established by H. Degenerate elliptic equations of type (1) have also been investigated from the probabilistic viewpoint (. [S] J. Let Z. IV-1 (1974)). Letl<p<co.O < c(i 1) depend only on (n. . [3] C. and Z. Ark.. C is independent of the boundary condition and. normal vector to G. 1 (1967)) treated the case of manifolds and discussed the geometric meaning of this formula. At XE I. as T+ 4x. Leipzig.. S. Schauder. Hadamard. Weyl [25] for the case of L = Laplacian. Anal.also M. respectively. Math. Differential Geometry. References [l] J. Partial differential equations of elliptic type. Equations of Degenerate Elliptic Type An operator L of the form (1) is said to be degenerate at x0 E G in the direction 5 E R” if 5 is a null vector of the matrix (a.. [4] M. Palermo. and hence it is often called Weyl’s formula. Lectures on Cauchy’s problem in linear partial differential equations. 4 (1964)). Miranda. Camus (Ann. The general boundary value problems for degenerate elliptic equations of higher order have been treated by M. . ~0. T. Umber lineare elliptische Differentialgleichungen zweiter Ordnung.(x)bi(x)-j$ y}. Singer (J. Carleman (Ber. In general. Mat. Numerische Abschltzungen in N. Schauder.. E. Minakshisundaram (Canad. Yale Univ. Math. Sup.-Phys. Springer. and others. 1 (1947)) discussed this formula in connection with the heat equation. Let N( T)( T> 0) be the number of eigenvalues of L smaller than T. H. Also. We have the following existence theorem [28]: If(i)eitherc<OonGorc*<OonGandif (ii) pc + qc* < 0 on G. Math. S. axiaxj (13) s S”-l where a(x..6) and are independent of the particular choice of (I. Wiss.. (10) Then the Dirichlet problem for equation (1) is to find a function u(x) defined on G U C. Rational Mech. Then.115 Diffusion Processes). PDEs of Elliptic Type where A and tl (A > 0. i. . V.. Levi.(x’)). Z. L is said to be of degenerate elliptic type if (aij(x)) is nonnegative definite at any x E G and if L is degenerate at some point of G in some direction. Moser proved that the above inequality is a corollary to a Harnacktype inequality (. and =O. be the set of x E I at which L is not degenerate in the normal direction.

Aronszajn. 1967. Amer. 0. Press. t:Original in Russian. 1930. Ural’tseva. Van Nostrand. [16] S. 299-325. 5 (1934) 34442. Agmon. Math. 1964. Multiple integrals in the calculus of variations. Cambridge Univ.)=lku(x. Agmon. Magenes. 6 (1953). 6233727. 35-92.. Schechter. B. 2 (1967). . Pure Appl. Pure Appl. Garding. (Original in Russian. Pure Appl. General boundary value problems for elliptic equations..) [23] I.. Grushin. Pure Appl. Math.. 1955. Stand. . P. Nonlinear Partial Differential and Their Characteristic Strips E. On the differentiability of solutions of linear elliptic equations.x.324 A PDEs of First Order elliptischen linearen Differentialgleichungen. Philos. Friedrichs. 71 (1912) 441-479. A. John. Sot. dt i=l. [26] R. A. ~=QW is called a characteristic curve of (1:1(. 80 (1958). Ecole Norm. Courant and D.. Serrin. Mizohata. Math. Quasilinear Partial Differential and Their Characteristic Curves Equations Suppose that we are given a tquasilinear partial differential equation 2 pitxTu)&yQ(x. 543-568. Lions and E. For example. 1973. 1969. [20] N. Ann. The theory of partial differential equations. Springer.e.) [13] L.. a homogeneous function of degree k. 12 (1959). i. Math. [28] 0. V. 55-72.. II. Estimates near the boundary for solutions of elliptic partial differential equations satisfying general boundary conditions I. Grushin. A unique continuation theorem for solutions of elliptic equations or inequalities of second order. Lectures on elliptic boundary value problems. 1 (1967).. [24] J. [ 151 K. London. Therefore the solution u = u(x) is a function such that u@x. Radkevich. Comm. Calderon. Topologie et equations fonctionnelles.). N. Cl23 S. Berlin. Gauthier-Villars. . The problem of Dirichlet for quasilinear elliptic equations. Lecon sur quelques problemes aux limites de la theorie des equations differentielles. Ann. Morrey. [21] A. Dunod. Math.. 16-36. Remarks on strongly elliptic partial differential equations.. ... Math. u’ = It u). 1966. Comm. Math. 17 (1964). . II.) [29] M. Vishik and V. USSR-Sb. Doughs. 36 (1957) 235-249.. xn) @=e’>O). 1971. Boundary value problems for elliptic equations degenerate on the boundary of a domain. [8] C. u = u”ek* (a solution of xf = xi. Uniqueness in the Cauchy problem for partial differential equations.22) Partial Differential Equations of First Order A. Comm.. J. Second-order equations with non-negative characteristic form. Vekua.quations We denote the value of au/ax.. Trans. Math. 1962. N. 2 = Pi(X. Math. On Harnack’s theorem for elliptic differential equations.. 1956. Das asymptotische Verteilungsgesetze der Eigenverte linearer partieller Differentialgleichungen.. 1973. I. Nirenberg. Pure Appl. by pi and define the surface element (or hypersurfaoe element) . the: characteristic curve of C. 649-675. Oleinik and E. Uniformly nonelliptic problems I. B. 51 (1934) 45-78. [9] 0. lx. Moser. Nirenberg. 14 (1961) 577-591.. I u)9 x=(xl. Vainberg and V. A. [22] B. Weyl. 1212 [25] H.=‘=. Math. [27] S. (A) 264 (1969). 423-454.. xiau/axi = ku is xi = xpe’. 11 l133. Pure Appl. USSR-Sb. [lS] J. u).. A necessary and sufficient condition for u = u(x) to be a solution of (1) is that the characteristic curve passing through any point on the hypersurface u = u(x) (in the (n + l)dimensional xu-space) always be contained in this hypersurface. 12 (1959) 457-486. R. Dirichlet’s problem for linear elliptic partial differential equations. 1 (1953). 1965.. Studia Math. Hilbert.320 Partial Differential Equations. (Original in Japanese. Academic Press. . Math. Problemes aux hmites non homogtnes et applications I-III. Interscience. Comm. [ 1 l] J.. Sup. Methods of mathematical physics I. 322 Partial Differential Equations (Methods of Integration)). II. . and L. Plenum.n. Schauder. 9 (1969).. Roy.(t). Picard. [14] L.) 324 (X111. System von Differentialgleichungen erster Ordnung vom elliptischen Typus und Randwertaufgaben.. Plane waves and spherical means applied to partial differential equations. (Original in Russian. 1965. [7] E. J. L. u = u(t) of the system of ordinary differential equations dx. 1968. 8 (1955). Comm. Linear and quasilinear elliptic equations.) [lo] J. (1) A curve defined by a solution xi = ~c. 1968. Interscience. . Ladyzhenskaya and N. (Original in Russian. 413-493. [ 191 F. [ 171 M. V. Leray and J. V.

‘~((5. then the system (3) is called a complete system. G) and call it also the Poisson bracket. q’ = q./q. Fj] = 0 (i. u. . I C. In this case. . Y i=l.CGKl..x. we have. p) = 0.Plr. When F.Theequations of the characteristic strip are x’ = q.. we usually use the notation (F.)x. define a differential operator X by x= t PJx)L “=. A solution of the partial differential equation (2) is. p of class C”. . In this case. %)I = i$I g CF. p). k.. this definition of characteristic curve coincides with the one mentioned in Section A. z’ = 2pq. .p. For quasilinear equations. (3) A set (x(t). . q=qo at x=0).. .=1p. . by the (2n + 1)-dimensional vector (x. x. p’ = p. If F. b are constants). ..z = wYo)+2wYoYw’(Yo)+ ~(Yob2/w(Yo))2.=q). (4) We call k differential operators Xi = Z. furthermore. u. pidxi=O Here. . u. . .. Xi] is a differential operator of first order. (2) X&=0.ri+PiF.U. y = yo. we define the Lagrange bracket CF> Gl by CF.j= .. . .(x)a/ax.z = 0 (where x.u. in general.). au PY=aXy. . G(x. u = u(t) is called a characteristic curve of (2). x=(x. .Cl = .). y’ = p..=p. An example of a nonlinear partial differential equation of first order is pq . G] = [Xi..=x..)/(W ‘(Y~))~. This bracket has the following properties: CF.. G are functions of x and p only.X..1213 324 D PDEs of First Order pendent when the rank of the matrix (Pj) is equal to k. and we see that [F. p=po. . and a tsingular solution is z = 0.(x) of class C” such that is called a characteristic strip of equation (2). an r-dimensional tdifferentiable manifold consisting of surface elements satisfying the relation dui i=l that is. and therefore the characteristic strip is given by y = y. . Putting zo= W(yo) (an arbitrary function) for x = 0. p) = (x i. X2]“. is called an r-dimensional union of surface elements... . u. z=zo. . . Consider k partial differential equations involving one unknown function u(xl. . then F and G are differential operators F = Xi u and G = X2 u with respect to u (for pv = au/ax..Gil. surface elements belonging to an (n .=y. as initial values. + (p.Pn)=O. F. $= -(F. u.. Complete Systems of Linear Differential Equations Partial For functions P.(x)(i= 1.) XJ-=o. . [X. p) of x. u. q = q. u(t).. from these expressions yields a general solution z = z(x. . k) mutually inde- If a common solution u(x) of these equations exists. Furthermore.. called the commutator of the differential operators Xi. A necessary and sufficient condition for the system (3) to be a complete system is that there exist k3 functions n. pl.1)-dimensional union of surface elements satisfying F(x. z = zo + 2PoX +(Pd/qo)x2v P = PO +(Po/qo)x. and the curve x=x(t).x. y).u=z.+ W(y.) has the maximum number n -k of independent tintegrals. . Involutory Systems For two functions F(x. . the righthand side of the third relation vanishes. formed by the set of all characteristic strips possessing.. pi= aulaxi.p. (i= 1. it is also a solution of [F. . G do not contain u and are homogeneous linear forms with respect to p. a tcomplete solution is 4az = (x + ay + b)2 (where a. Xi or the Poisson bracket. . Consider the partial differential equation F(x l. If a system of k independent linear partial differential equations involving one unknown function f(x).p)=O. +x (if we take x as an independent vari- D.)) of class C”. CF.p(t)) of surface elements depending on a parameter t and satisfying the system of ordinary differential equations & dt du -F&Y zziz n PiFpi. p.. The elimination of y. . .(x...Gl=iI (E($+P~:) able and impose an initial condition y = y. x.).

. Fjl = 0 can be derived from Fj = 0.Hxi..) is equal to r).u. we put FXr = pi. .x..u.. is tcompletely integrable. a) (a =(u~+~. we regard the equation itself as an involutory system.. from the equations thus obtained.acp/aX)=o. n) are mutually independent. that is. say. . i=l.u. xn. j is called an involutive (or involutory) system. n. That is. then the Euler-Lagrange equations are equivalent to Hamilton’s differential equations dxi/dt = H. by finding a partial derivative. . When the equatians (4) are mutually independent. In . 41~ 0 for all i.P)=~.)) from (4) and (4’). It does not concp explicitly.n..aining ones from the displayed equation. .... 6xi along a?[...p. Furthermore. this equation has the Then. the original system has no solution. . a dependent variable cp. we have x ..p). An involutory system (4) can be extended to an involutory system consisting of n independent equations by adding n-k suitable equations F~+~(x...l-1. ‘. k). . . which is called the normal form of the partial differential equation of first order..=O. . the rank of (aFJ@. the equations of the characteristic curve of this equation are dx. This method of integrating an involutory system is called Jacobi’s second method of integration. x’) dt. the system of ttotal differential equations -=pY(x. Then. a necessary and sufficient condition for them to have in common a solution with n-k degrees of freedom (a solution that coincides with an arbitrary function on an adequate manifold of dimension n-k) is that the system (4) be a system of equations involving unknowns p and equivalent to an involutory system.Ifasolution u(x) of this equation is given as an implicit since F(t. and suppose that the family is ttransversa1 to an r-dimensional manifold rU (r < n) (that is. E. . .+. =O. . in particular. ... . dx x1=1.p).’ 2a. . if 2I consists of only one point (r =O). .$x. If we then have more than n + 1 equations.. ’ gives formally a partial differential equawith independent variables u.. between the equations. Relation to the Calculus of Variations Consider a partial differential equation of first order F(x. x.. for the integral J= which is a complete system of linear partial differential equations for A and the & (i = 1. . . .‘. Y) for which the Fj are independent (i. .. . .. A curve represented by a solution of the Euler-Lagrange equations is called a stationary curve.. say adax.-dimensional tx-space such that passing through every point of G there is one and only one curve in this family.. Fk+l =ak+l. . u) = 0.e. F. ak+. = 0 for the differentials 6t.H.)#O.)=O. llL au Under the assumption that det(F. 1214 function by ‘p(xI. . xi = tpi(t. F6t -C F.. . if we have an involutory system of n + 1 independent equations F. When k = 1. n) such that Fl satisfies the system of equations CFiafl =O. we take independent equations and add them to the original system.. a stationary curve passing through this point is transversal to IX). Moreover. ‘1 F(t. .. we get a partial differential equation of the form arplat+H(t. x. x.. i=l. . Now. a complete solution of (4).. x1. Setting pi= @/ax. i= 1..F.P)=~~+~.. . . n. .. as a function of the rem.. consider the tEuler-Lagrange differential equations dFXi/dt-F+=O. x. ... .. . . We always treat a system by extending it to an involutory system. i=l....x.. Now consider a family of stationary curves in a domain G of the (n + l). p. . ~=H”i(t.324 E PDEs of First Order 1..n. dpildt = . = a. if we put v=l. if we find that py as functions of (x. . we can find successively f= Fl (I= k + 1.X. then we find a complete solution by eliminating pl. . . . . we obtain a system Fj = 0 (j = 1. ..u.. which are called Hamilton’s differential equations. a. p.. and all [F. and solve these relations with respect to XI in the form. and we can find u as a solution containing essentially n-k + 1 parameters c.. . Otherwise. F. (4’) That is.. ax. .(x.-= This tion and tain form acpiau1. Therefore. dpi z= -ft&x..x. u. x’) = zy=‘=l pi Hpi . . A system (4) such that [Fi.u. p).a).

Methods of mathematical physics II. (3) Equation (3) is also called the equation of a vibrating string. = FPi/FP. . F. <) for any n-tuple of real numbers 5 = (1) is called (51.. . In particular. 1953. a curve that is an envelope of a l-parameter family of characteristic curves of the partial differential equation (2) is a solution of the Monge equation. Caratheodory. 197 1. .(t.. . partial differential equation in n + 1 t. x) of G. According to the theory of lirst-order par- .~.x. .e.. ~. This equation is called the Monge differential equation. s.5)=1’‘f a&l-i=l (for example.<) satisfies the characteristic equation at p” : H(t’.‘=O. . H(t. English translation. A. x.. 1962.1~=. Petrovskii.5)1=c>o. u.. .<) is called a characteristic hyperplane of (1) at p” if the direction (A.. Hilbert. Table 15).. u. + &(x” . an integral curve that is not a characteristic curve is a tline of regression of the surface generated by the family of characteristic curves tangent to the integral curve under consideration (which is an integral surface of F(x. H(t. n au Consider the partial differential By eliminating p and t between dxi --=Fpp dt :=A PiFpi.) [3] E. . we obtain Mb i i.<) = 0. a that are functions in (t. Calculus of variations and partial differential equations of the first order. Lectures on partial differential equations. 1911. Conversely. (Original in Russian. x. . and F = 0 when FP. G. the equation of a vibrating membrane.-ax.5”)#(0. -a. x) = 0 in tx-space is a characteristic hypersurface of (1) if at each point of S the tangent hyperplane of S is a characteristic hyperplane of (I). . or 3. then the equation aV/&+H(t.) of the second order.l(t . Gauthier-Villars. st. p) = 0 with coefficients aoi. Springer. or the equation of sound propagation according as n = 1. x0) in tx-space and having normal direction (A. (t. x. Teubner.to) + . .5)-n. Second-Order A tlinear variables Linear Hyperbolic Equations holds. p) = 0). 1927.Appendix A.2. 1954. John. Variationsrechnung und Partielle Differentialgleichungen Erste Ordnung..1. . fourth edition. all integral curves coincide with characteristic curves. If F is linear in pi. ...x.. x0.. [2] I. When n = 2.. . . Another example is References [l] R. 1967. if we denote by V(t.fax. i. and F(x.X. (2) has two distinct real roots 1= I. au which describes the propagation of electric current in a conducting wire with leakage and is called the telegraph equation (. [S] C. x) is said to be hyperbolic (or of hyperbolic type) (with respect to the t-direction) in txspace if the characteristic equation of equation (1) considered at each point of tx-space.(t.--au=O. This equation is called the HamiltonJacobi differential equation or the canonical or eikonal equation.25) Partial Differential Equations of Hyperbolic Type A. The Monge Differential Equation equation (2). III.(t. Interscience. . .) = 0 everywhere on S. x=(x. regularly hyperbolic if these two roots are separated uniformly. that is.x.-- au at i=l axi 1 a. Cours d’analyse mathematique. x. 1935. and the curve represented by its solution is called an integral curve of the equation. Interscience.. by eliminating p between dxJdx. (t?. #O). A hyperplane .)=o.la~. a solution of this equation is equal to the value of the integral J for a family of stationary curves transversal to an adequate ‘?I. II. second edition.. . [4] F. Holden-Day. . A typical example the wave equation of hyperbolic equations is a% as a% qU=c7tZ-@-. .xz) = 0 passing through a point p” = (to. that is. In the (n + 1)-dimensional tx-space. ax. Partial differential equations. a2u z-c . l. Goursat.ln.1215 325 A PDEs of Hyperbolic Type this case. x) the value of the integral J along the stationary curve from 2I to any point (t. <). quasilinear.a2u @-2%=0.a~. Courant and D. . A hypersurface S : s(t. A characteristic curve is also an integral curve. 0).aV/ax)=O 325 (X111.j=l aij&tj=O.

The Caucby Problem Important for the hyperbolic equation (1) is the KZauchy problem. . A tsmooth curve y in tx-space is called timelike if the tangent vector of y at each point p on y belongs to D+(p) or D-(p).x) is continuous in the following sense: If a sequence of initial functions {uOk(x).. For the wave equation l(3). x0) (..) and their derivatives up to the Ith order tend to 0 uniformly on every compact set in the hyperplane t = 0. x) also tends to 0 uniformly on every compact set in each hyperplane t = constant. .Section D) is determined. E(t) law for the wave = constant. Moreover. whose boundary consists of the part of C(p”) with t > to (t < to) and the interior of the ellipsoid on the hyperplane t = constant.325 B PDEs of Hyperbolic 1216 Type are of class C’. there exists a proper subdomain of Go such that the solution depends only on the initial data on the subdomain. Suppose that (1) is regularly hyperbolic and the coefficients are bounded and sufficiently smooth (i.)~. If the solution of the Cauchy problem has such a property. for the wave equation (3) with n = 3. If the coefficients of (1) are bounded functions.> -Hi.x)=u. Consider the set of points that can be connected with the point p” by a timelike curve.[)=O. We call its closure an emission. . i. the following proposition is svalid: The values of the solution u at a point p” = (to. An emission is a conical body surrounded by characteristic hyperplanes in some neighborhood of the vertex p”.(x. x0) depend only on the initial data on a domain Go (domain of dependence) of the initial hyperplane.151=1 (I4(LT5)L B.a dx. x(z) of a system of ordinary differential equations dt drdi dz=Ht. or that diffusion of waves does not occur. in this case) passing through p” and the initial hyperplane.x.H dz .x).. x0.e. .x.5N. tial differential equations. Now if (1) is hyperbolic.(t. this correspondence {u. The Energy Inequality The energy conservation equation (3).. of the initial hyperplane induces a change of values of the solution only in some neighborhood of the forward emission 9+(Qo) (domain of influence). i. . the intersection of emissions 9+(p”) and the hyperplane t = constant is always compact.to)Z = & (xi . the emissions 9+(p”) are contained in a conical body independently of the situation of p”. then there exists a unique solution u = u(t.. We have the following dual proposition: A change in the initial conditions in a neighborhood of a point Q. In some special cases.. . since the intersection of any hyperplane t = constant and the cone C(p”) is an (n. Lax = max.2.() = 0} has as its envelope a cone C(p”) with the vertex p”. x) that satisfies (1) in t > 0 and the initial conditions 40.to) + . -cc <xi< co (1 <i<n). such that if the functions uo(x) and ur(x) in (4) C.t a point p” = (to.i..(x). Moreover. a conical body D+(p”) (k(p’)) is determined. . the set of all characteristic hyperplanes at p” : {l(t .= 1. Huygens’s principle is valid only for odd n > 1. In other words.ui(x)}+u(~. + 5.-Hen> dz d5.e. which is determined as the intersection of the backward emission %(p”) and the initial hyperplane. the solution for the Cauchy problem a. (4) where the functions uo(x) and ui(x) are given on the initial hyperplane t = 0.(x). of class C’ with v sufficiently large). . solution curves t = t(z).e. by the initial data in a neighborhood of the cone with vertex p” : (t . If the coefficients of the equation a. It follows that the domain of dependence and the domain of influence are bounded. A. x) of class C? in the domain 0 < t < co. 4 = u. and a subset 9+(p”) (9-(p’)) of the closure for which t 2 to (t < to) a forward (backward) emission. in a neighborhood of the intersection of bicharacteristic curves (lines.l)dimensional ellipsoid or two points for n = 1.x. &/&(0.. namely.(x)} (k. Then for the Cauchy problem the following theorem holds.. it is said that Huygens’s principle is valid.%’ H(t. the Cauchy problem for regularly hyperbolic equations is twell posed in the sense of Hadamard [2]. then the sequenc’e of corresponding solutions u. the problem of finding a function u = u(t. For example.. -xi) = 0 1H(t”. For dependence of the solution on initial data. depending on the dimension n + 1 of the tx-space..re bounded. as can be seen from the solution formula (12).(x).. a characteristic hypersurface of S is generated by so-called bicbaracteristic curves. Theorem (C): There exists a positive integer 1(= [n/2] + 3). . where IW. u.

the problem of solving an equation L[u] =f(t. (4”) Then the following inequality holds: Ejk)(u. In general. and the equality L(t. is found explicitly.do is in general not a function in the ordinary sense.7. the integral co A1”+4”2c$y)dy R.) and dw is the surface element of the unit sphere lol= 1 in x-space. y) can be analyzed using the asymptotic expansion with (n even). x. let q(x) be a function of class C’ with sufficiently large v and with compact support. 7. or more generally. Now. x. We call (5) the energy inequality (J. y)=@t-7)6(x-y) a/at. x. 7. the energy of the solution is also bounded. while if the energy of the initial functions is bounded. the hypothesis 1= [n/2] + 3 in theorem (C) can be replaced by a weaker condition I = [n/2] + 2. x. Therefore. a/ax)R(t. In fact. 7.. x. . Now. y. 7. we can infer from formula (7) that the Cauchy problem (1) with initial condition (4’) can be reduced to the one for initial conditions with parameters Y. &4/&(7. 7. even though the initial functions are of class C’.x)= W. (6) where n is.=i (xi . Then for a function q(x) of class C’ (with v sufficiently large) with compact support.=lR&t. w) is a function of (t. Then. when R. there is an example for which no global solution of class C? exists. R&t.ykdy)dy (9) 47.(s) of a real variable s by x. then R(t. x. by (7) and the delinition of R.7. w)dw defined for t > 7 to t < 7. x) of (1) on G(7) is defined by following equality holds [7]: q(x)= m s -cc A:“+q”2cp(y)dy X. since the tprinciple of superposition is valid because (1) is linear. to the Cauchy problem with initial conditions on the hyperplane t = 7: is a solution of the Cauchy problem (1) with initial condition (4’). to < 7<t’. Schauder [S]). w: u(r. In other words.x)=O. o)do can be considered a tdistribution on (t. . 7. x. x)= q(x) (4’) for arbitrary 7.1)-times differentiable by definition. s (ml=1 (7) Ef’(u. assigning it the value 0 there. x. y. 7.7. y.(s)= ls1*/4(2zi)“-‘q! = -sqlogIs1/(2ni)“q! (n odd).y)=~*(7. au/at(7. 7.. (5) where the constant C is independent of u.aja7. R(t. y) = A$‘+q)‘2 ~l+i R&t. x. w) of class C’. y)-space. the solution in the Cauchy problem for hyperbolic equations may not be of class C’. Let the coefficients of a hyperbolic equation (1) be bounded functions. y) in this sense is called a fundamental solution or Riemann function of the Cauchy problem. since x. the . and v increases with q.the dimension of the x-space and q is a positive integer such that q + n is even. a/ay)R(t. is defined by (8). 7. x. y) is a fundamental solution of L in the sense of distribution theory.x. y. 7. G(7)) where AYis the tLaplacian with respect to the variables y=(y.o. If we extend the function ~. where L* is the tadjoint operator of L and 6 is tDirac’s 6function. the integral ~. The kernel R(t. x. G(7)) < CE$‘(u. the Cauchy problem (1) with initial condition (4”) has a unique solution R&t.(t. The problem of solving (1) under the condition (4).xt)‘} with the hyperplane t = r (t < t’).1217 325 D PDEs of Hyperbolic Type is generalized to the so-called energy inequality for hyperbolic equations. But we denote it by R(t. y). y) formally. Af+@‘2 & R. and understand that a linear operator u(t.Ww 5 -52 s Iwj=1 (8) D. which plays an essential role in deducing the well-posedness of the Cauchy problem and the properties of the domain of dependence of the solution. x. y. The k( > l)th-order energy integral of the solution u(t.x)= 0. 7.y. x. and let G(r) be the intersection of the conical body K = {(t.((x-Ykw~. (8) yields a solution formula of the Cauchy problem (1) with initial condition (4’) as a functional of the initial functions.x)=&((x-y)o). w) for q chosen large enough so that theorem (C) can be applied to (1) with initial condition (4”). y. . y. Since in (8).We define a function x. x) under (4). The fundamental solution R(t. is valid. 7. .do is not necessarily of class Cnfq as a function of (t. Moreover. x. G(t”)).(s) is (q . X. 7. For the wave equation (3). but if we take I= [n/2] + 1. can be reduced. Representation Formulas the Cauchy Problem for Solutions of We consider solution formulas that represent solutions of the Cauchy problem explicitly as functionals of the initial functions. by transforming the unknown function u and applying tDuhamel’s method. x) l&&t tl)Z > C.x.+i R.

respectively.. x) are determined by substituting u by U(t.(l) of the characteristic equation L(i.)2-(X2-52)2 solution). AN(t) are purely imaginary for all real 5 = (r. . .+. and dw. x = (xi. If the functions A in (13) and U =x0(x) + tu. x) = u. and we have the following important result: If the coefficients of (1) are of class C” (resp..lal=cc. x) depends only on the smoothness of the initial conditions on a neighborhood of the intersection of the initial hyperplane and all bicharacteristic curves passing through p. D. If (14) is hyperbolic.-plane is a sphere (xi.x3)=~at 1 6 sQ +L t uI(t. . Behavior of the fundamental solution R(t. x2) and radius t.$-cc. and f (11) (Poisson’s ukx. 4 1 (n-2)! Q(x. s [w/=1 linear differential equation in t.~)=~k(x). x) and aij(t. u.--. 1 a .~2~:3)dw. ' j<n. a2u/axidxj. then the fundamental solution R(t. in the 5. y)..--. r. . I&) # (0.axp' alal (d’Alembert’s u(t.. Second-Order Equations A second-order -=A Nonlinear Hyperbolic respect to the sequence of functions {x&s)}. . (12) 471 s n. y) is of class C” (real analytic) in (t. the Cauchy problem for (14) with initial condition (15) is well posed in the sense of Hadamard. xg) element of t is hyperbolic in the sense of Girding if the following two conditions are satisfied: (i) the partial derivative aN/atN appears in L. and therefore we can write the solution formula explicitly. x2.(x+t)+u. In general.).1 . x. (x) determined by (4) are sufftciently smooth with respect to t. .u.2n +k at uo(51 a 52)d51@2 s C~Jt2-(X1-51)2-(x2-52)2 ~1(51>52)d51dL s C~Jt2-(X1-r. is a disk in the with center (x.325 E PDEs of Hyperbolic 1218 Type E. This fact is called Huygens’s principle in the wider sense. (0. The solution formula for (3) with initial condition (4’) for n > 3 is u(t. (. a theorem analogous to theorem (C) holds for the Cauchy problem for (14) with initial conditions solution). . . the smoothness of the solution u at a point p = (t. x. (15) that is. and 3 are. ... y) near discontinuous points has also been investigated [2]. (13) is called hyperbolic in a neighborhood of a function U(t. Solutions of the Cauchy problem (3) with initial condition (4) for n = 1. O<k<n-1. it) = 0 are bounded functions of real variables 5 = (<i. and LJ2u/dxiaxj. at”-2 s () 1 W” F. . where cr=(ai. . .) with constant where o. . the Cauchy problem for (13) with initial condition (4) has a unique solution in some neighborhood of the initial hyperplane under the condition that equation (13) is hyperbolic in a neighborhood of U. x) in the partial derivatives of A with respect to a2u/itax.and (14) u(t. respectively. is the surface D.~)d~. L&J. T. When L is a homogeneous equation of the Nth order. Xl >x*1 solution). nonlinear differential equation a5 at2 t. condition (ii) is equivalent to the following condition: (ii’) 1. = 2Jrr”/I(n/2) is the surface area of the unit sphere of n-dimensional space. where aoi(t. In the language of the Cauchy problem. (<). =o. . the fundamental solution can be constructed. Higher-Order An Nth-order n + 1 variables coefficients Hyperbolic Equations cp(x+zo)dw.x. where C. L. akujatk(o. . x) except for points that are on bicharacteristic curves of (1) passing through the point (z. 2. cc. .x. real analytic). .. t2ii3-space with center and radius t.O). The principal part (consisting of the highest-order terms) of a hyperbolic equation is also hyperbolic..(x-t) 2 1 X+t +j sx f Ul(W5 (10) c-1 - a fi 6X ax. x) if the linear equation of the form (1) obtained from (13) is hyperbolic. . if the Cauchy problem . (Kircbhoff’s 5. initial value problems for nonlinear equations have only local solutions.. T) =an-2 t (t2-~2)(n-3”2~Q(~. (ii) the real parts of the roots i = E. For the wave equation (3).x2. x. Conversely.--- au au a2u a% ~at axi ataxi'axialG 1 <i.

then (14) must satisfy the hyperbolicity conditions (i) and (ii) in the sense of Girding. lo]. In other words. . x) and each 5 # 0. 5. Petrovskii showed that the Cauchy problem for a system that is hyperbolic in this sense is well posed [lo]. Weakly Hyperbolic Operators We adopt the following definition of hyperbolicity: a linear differential operator of Nth order G. Friedrichs. A system of first-order linear differential equations x. .(t. .(t. .lSI=l.1. i. (16) is said to be regularly hyperbolic. Theorem (C) holds for the Cauchy problem for a regularly hyperbolic equation (16) with initial conditions (15). x)ns(io” =o . . . Huygens’s principle in the wider sense is valid. since the influence of the lower-order terms in the equation is taken into account in the definition of hyperbolicity. are higher-order linear differential operators of the form (16). .<N =o is called hyperbolic in the sense of Petrovskii the characteristic equation (16) if aN+ 1 a.. discontinuity of the solution is carried over only along bicharacteristic curves.(t. where the L. Thus a linear equation with variable coefficients lem. since for them the energy inequalities are valid most naturally. x.l=N has N distinct purely imaginary roots (called characteristic roots) A. are separated uniformly. the Cauchy problem for (17) is well posed if and only if (18) is hyperbolic in the sense of Girding. and weakly hyperbolic otherwise. studied symmetric hyperbolic systems of equations.&(L x.x). A..perbolic Type for (14) with initial condition (15) is well posed. is called symmetric hyperbolic (in the sense of Friedrichs) if the matrices A. 1<i<l.(t. Systems of Hyperbolic For systems of equations Equations .) . an Nth-order homogeneous equation remains hyperbolic for any addition of lower-order terms if and only if the characteristic equation has N distinct purely imaginary roots for any real 5 =(li. hyperbolicity in the sense of Petrovskii and symmetric hyperbolicity due to Friedrichs. .<i<l.) # (0. that is. We take up two important types. 0. Mizohata [ 123). in the case of constant coefficients. 5)1= c > O holds. For the domain of dependence of the solution. . In this case the equation is called hyperbolic in the strict sense.&.j#k 15th calculated formally using the matrix of differential operators in the system. There were some imperfections in his argument. nj)th order. We call a system of linear differential equations l.1219 325 H PDEs of Hy. Girding’s conditions for hyperbolicity cannot be generalized to the case of variable coefficients.$ Lij[Uj]=O. 0. the inequality lim (t. 5) for each point p = (t. well-posedness of the Cauchy problem is equivalent to hyperbolicity in the sense of L.+/. . 5) . if the characteristic roots 1. . several types of hyperbolicity are formulated in connection with the well-posedness of the Cauchy prob- is called hyperbolic if the Cauchy problem for L[u] =0 with initial condition (15) is well posed in Hadamard’s sense. x) is positive definite. a system of hyperbolic the sense of Petrovskii) differential equations if the determinant (17) (in a. a result analogous to the case of the second-order equation can be obtained using an energy inequality [9.. . A typical example is provided by Maxwell’s equations. In the case of constant coefficients. Moreover. For this system it has been shown that the Cauchy problem is well posed and the domain of dependence of the solution is bounded [13]. However. x) (0 < i . which have been corrected by others (. 0). H.S. .< n) are symmetric and A. is hyperbolic in the sense of Petrovskii as a single equation of N( = $. observing that the energy inequality played an essential role in Petrovski% research. . If the coefficients are of class C”. this definition is equivalent to the previous one. K. x. In the second-order case. A hyperbolic operator L is called strongly hyperbolic if L remains hyperbolic for any addition of lowerorder terms. Garding [ 141.e. .

This class of functions was used efficiently in the studies of the Cauchy problem for tweakly hyperbolic partial differential equations: in j=O.1)th order among lower-order terms of L (Mizohata and Y. 5) (P.(t. Gevrey Classes Classically. Mizohata [ 161).aiacaa~.x.2 where L. D. A=A. 0. where L. Kyoto Uniu. Thus.. I. the functions of tclass s (s > 1) of Gevrey (.+lal=nj c a&(t. Petkov [20]).. Sot. Cdl.X. Moscow Math. Kasahara and M. Anal. if there exist positive rational numbers q and r (q 2 r) such that 11=o . are real.+lal+q&+r(fil<p. Pisa.-ii(t. . j = 1. 2) is at most p.+lal=N rr. M. For higherorder equations. Menikoff.&..+/al+qB. x). As for operators with variable coefficients. and A such that ~~p.QN-1 X)2y way: (‘<ER”\O) fora.I.325 I PDEs of Hyperbolic 1220 Type for a..58 C”-Functions and QuasiAnalytic Functions G.<).jik l/z. i. L. namely. i. Lax [ 151. x.. x. of the char- We assume that the multiplicities acteristic roots are constant. Yamaguti (Mem. Then L is hyperbolic if and only if it satisfies E. even the principal part is not necessarily hyperbolic. 0. A.(t. for a weakly hyperbolic operator with variable coefficients. Oleinik [ 191 studied the Cauchy problem with nonconstant multiplicities for second-order equations. strongly hyperbolic operators have been characterized by K. then it is necessary for the well posedness of the Cauchy problem that A necessary condition for hyperbolicity is that all characteristic roots of LN(tr x.2.laO[’ > associated with (17) is not clear in general (references in [20]). 1977..t))‘i. denotes the homogeneous part of (N . J. x...h’-1. it is known that not only regularly hyperbolic operators but also some special classes of not regularly hyperbolic operators are strongly hyperbolic (V. 2 ( Et+&g$ in the following If: l<l=i. 1.5)l=C>O. Sci.A’a’lcrl!“}. Ann. In the case of constant coefficients. a. the suffrcieni. Math. if the multiplicity of characteristic roots at (F. . i=l .-. which corresponds to Levi’s condition in the case of constant multiplicity.X._. 1976. n a2LN a3. Ohya [17]).there exist constants C.(t.. s. Chazarain [ 1 S] has studied weakly hyperbolic operators with characteristic roots of arbitrary constant multiplicity.r)=Ij(~.x. Levi’s condition. forall (t. . Ohya. On the other hand.l.7aresatisfied. (1960)). condition in the case of multiplicity 2 is given by some conditions related to the subprincipal symbol L. The Cauchy problem for a weakly hyperbolic system of equations is more complicated. Math. J.() = 0 be real for any (t. a. 5) = AN + is decomposed LN(t.e. and to the tPoisson brackets (A. because of the essential difficulty that the matrix structure Assume that the ij(t.I(a/ax)“cp(x)I G C. Amer.+r]/I]+s(l+q)<.x. x.-. N-s. Let us consider the hyperbolicity of an operator L which is not regularly hyperbolic under the assumption that the multiplicities of the characteristic roots are constant and at most 2. the characteristic polynomial L. . 1977).5)-n.. 168 Function Spaces B (14) were introduced into the stud& of the fundamental solution for the heat equation: $JR”)= {cp(x)~C?(R”)]for any compact subset K of R” and any multi-indices cc. lvrii.. or more precisely. -- i a2L. I”. (“~ER”\O) L. (1 <s d N) are the homogeneous lower-order terms of order N -s of L (Ivrii and V.T]xR”=Q 1.. [O.. Hormander. Ya. +C m=. 5) c a.e. 1976). E. Sup. j=l. J. Scuola Norm.(t.

0.a/ax). Let max 14iakvi=p. Mixed problems for second-order hyperbolic equations are considered in [6]. The theory of lacunas of fundamental solutions of hyperbolic operators. x) =0 in 2).co)xfi={(t. which are compatible with the boundary conditions.. a/ax. x) is holomorphic. the Cauchy problem is well posed in yj@). In regard to mixed problems for hyperbolic equations of higher order. These formulas provide means of obtaining topological criteria for lacunas. where is a (pseudo)differential operator with a. has been developed further in a paper [25] by M. (iii) L is regularly hyperbolic. 0 < k < N . j=l. then for any s such that 1 <s < p/q. x.0)\W(A. b. K. 0) and all the E(Lk. A typical example of such a mixed problem is provided by the case L = IJ (n = 2) and B[u] = u(t. The integrands are closed trational (n .1.x)It~[O. Thus we have DpE(L.. 00) x a). . and define the wavefront surface W(A.x. respectively. In [25].2 . 0. O<k<N-1. Grothendieck (Publ.&x)=(27q” s ~(5 -ice)-leix(C-ic8)dg.x) in on (0.(t. R. x) is the restriction of an entire function to P(E(L. .&.1221 325 K PDEs of Hyperbolic Type where vi is constant for any (t.a/ax) is a (tpseudo) differential operator of order at most q. provided that all a. (20) = const s t. Girding. = {(x’. The problem of finding a function u(t. dip is said to be a weak (strong) lacuna of L if E(L. The sufficiency directly follows from (20’) and the necessity follows from a theorem of A. 5) is real and distinct.x~fi}. x) in the form E(L. 8. 0) by the union of all K(A. x. x. Then L has a fundamental solution E = E(L.. Leray and Ohya [22]). = 0) is not characteristic for L or Bj.i?/i% alax’. Jo* (x~r)“PGr(Kk45b qso (20’) . In [25] it is shown that x belongs to a weak lacuna for all E(Lk. Ed&.(-l)j-‘5jd51/\.. . x) of L and f(t. . x)* and t.) 1x’ER”-I. a/ax) be a linear hyperbolic operator of N th order defmedin[O.. 5) ER x R”.) + M(t) be a hyperbolic polynomial with respect to the vector &R” .) if and only if ac(* = 0. Mixed Initial-Boundary Value Problems Let Q be a domain in R” with a sufficiently smooth boundary I-.co)xr. a/at.0. (21) Bj[u]=O aku/atk(o... &(t.x’.. transported to the origin. x) belong to y!:\(Q)..1)-forms on (n . J.co)xQ (O. apt. where E(L. HES. 5 DBE(L:. Let L(t) = &(I. a/ax.a/at. Here. and contained in the union of the origin and the half-space x. x) satisfying the conditions L[u]=O where c is sufficiently large and the integral is taken in the sense of tdistribution. Lacunas for Hyperbolic Operators when xeK(A.x)=uk(x).q=mk-IpI-n is the degree of homogeneity of the left-hand side. x. and L. 0) for 5 #O.Ad~A . which describes the vibration of membranes with a fixed boundary. 0.th order defined in a neighborhood of [0.1)-dimensibnal complex tprojective space and are integrated over certain thomology classes c(* = a(A. there exists a unique solution u(t. 0 > 0. a/ax’. and that {u~(x)J.. By the hyper- = const Jx. F. a/at. a/ax) being strictly hyperbolic (pseudo) differential operators associated with L.. and let Bj(t. x)E Cm( [0. f?). where J&(C) is the principal part of L. Math. Bronshtein 1231. Atiyah. that they are locally holomorphic outside W. The mixed problem (21) is said to be well posed if for any initial data uk(x)~Cm(iZ).x. j = 1. 0.). initiated by Petrovskii [24]. x.. and Zf=. 1966) which implies that the rational forms which appear in (203 span all the tcohomology classes in question. a/at. (iv) Nj<N-1 and Nj#Nkifj#k. x. D. 0) and.x’. x). be the ttangent conoid of A at 5.co). the Herglotz-Petrovskii-Leray formulas are generalized to any nonstrict L(t). We denote the tprincipal parts of L and Bj by L. This result was proved even for the case of arbitrary nonconstant multiplicity of characteristic roots by M.~~~~-~ are given in $. If we suppose that L(t. Then it can be shown that the singular supports of E(L. a/at.) and Bjo(t. 8. . we make the following four assumptions: (i) Sz = R”. 0) are contained in W(A. 0) = K(A.(t. b. X. x) 5)q5flLrm-k45)> q>o. Let A. 0.andw=~j”=.2. a/ax). > 0). let L(t. this means that LN(0) # 0 and L(< + t0) # 0 when IIm t 1is sufficiently large.L(R”) (Ohya [21]... Bott. is called a mixed initial-boundary value problem. be linear differential operators of N. moreover. denoted by K = K(L. Let Y c g be a maximal connected open set. vi = N. J. The convex hull of the support of E.. co) x r. Inst.@).o.is a cone depending only on the real part Re A of the complex hypersurface A: L(l) = 0. (ii) r = {x Ix. &*.Z(t. x.

an equation for the L. Then w of (22) is an approximate solution of q u = 0 for large k.N. 1838). w in the form (22) can be prolonged to a domain containing the caustic satisfying the asymptotic solution w(t.. the acoustic problem + ikmn3Ai’( (23) where Ai is the Airy function Ai(r i ei(zf+f3/3)dt s m q u=O u=O in (0. In the treatment of mixed problems for L and Bj not satisfying the uniform Lopatinski condition. co) x I # cp. Under the condition that the principal curvatures of the caustic are positive. and R. Ludwig [31]).Ifuj. Consider.x’. and w+ represents a reflected wave propagating in the direction vcp + = vq .V0j+Aqnj= we have -ilJujml.(t. x. x’. the wellposedness of mixed problems depends not only on the properties of the Lopatinski determinant but also on the shape of the domain (M. i. if M’ hits the boundary I transversally.uj on (0. co) x I. Then w + w+ is an approximate solution..x)k”‘-j j (D. Kreiss [27]. for example. k) = eik(e(x)-‘){Ai( L. Asymptotic Solutions . which is the same as that of the Cauchy problem for L[u] = 0 (T. H.(‘. k). Sot. q w=O(kmN).(t. asymptotic solutions play an important role.x.+Ax)--f) jio uj(t. w of the form (22) cannot be an asymptotic solution near the caustic. and vj+ satisfy the transport equations and vj+ = . A. c0) x r. (p+=‘p. The function R defined by W. x. Balaban [26].. >O. x)k -j such that lVcp+l’=l inR. In order to explain some properties of phenomenon governed by hyperbolic equations.x’. x.j=C~. Sakamoto [28]). { {x + IVq(x) 1IE R} 1x E supp w} has an envelope.x’.i. k) . G53 .~. is called a Lopatinski Bj satisfy the uniform if u-1 =o.2(Vcp v)v.~)=0 has p roots K: with ImK.= . x. We say that L and Lopatinski condition $ (t. 5’). An analogous result holds in the case of a domain R with a compact boundary I. k) k)}.O.x. and g~(.j(t. x’. co)xR. t’ER”-’ where cp is a smooth function satisfying eikonalequationJV~~Z=l.325 L PDEs of Hyperbolic bolicity 1222 Type in K Imi<O. and the number p is independent of (t. and it represents a wave propagating in the direction Vq. satisfy the transport equations 22+2Vo. The asymptotic behavior of high-frequency solutions near the caustic was first considered by G.[‘)) and C is a xj’. A necessary condition for the well-posedness of the mixed problem (21) is that the number of boundary conditions coincide with this integer p.k2’3p(x))g1(t.. however.l.kz’3p(x))go(t. provided that L and Bj satisfy the uniform Lopatinski condition at every point of I.. If asymptotic solution (22) has a caustic. x)k-‘. When supp w fl(0. where v is the unit inner normal of I. B Airy (Trans. .k)=Cg. These asymptotic solutions show that the high-frequency waves propagate approximately according to the laws of tgeometric optics. the mixed problem (21) is well posed.. 0. x’) and (A.. and inf IR(t.&t’)l=c>O.eWe+W-O j$o uj’ 0. k) be a function defined in (0. For general domains. of L. we can construct w+(t. Ikawa [30]).lmA-<(l K’l+l4=1 When the uniform Lopatinski condition is satisfied. (r. and (21) represents a phenomenon with a finite propagation speed. and N-p roots ic: with Im ~~~ < 0. At’) = det where L’ = n&i (K contour enclosing all determinant.e. co) x R with parameter k > 1 of the form w(t. on (0. the well-posed problems have been characterized for operators with constant coefficients when fi=R’!+ (Sakamoto [29]). Cambridge Philos. Concerning the reflection of grazing rays by Let w(t.

N . For more general singularities of initial data. Interscience.. E. the solution also has singularities for t > 0. 345-392. WFu(.. J. 7 (1954). ~(O))E iJkWF(d} (J. 0.=Rfl{]xl<R}gooutof&inafixed time. B. Chazarain [18]). Symmetric hyperbolic linear differential equations. (N. [17] S. R. Guillemin and Sternberg. Amer. Luneberg [32]. 1973. and all the broken rays according to the geometric optics starting from~. 51. Schauder. the solution becomes smooth in R.e. Res. Sot. Mizohata. 1953..) [9] J. L.1. The propagation of singularities is more complicated in mixed problems because of the reflections of singularities at the boundary. Sot. Fund.). Lecture notes. 5 Y). Math. Girding. the reflected wave can be constructed by the superposition of asymptotic solutions of the type (23). Leray.. John. t) is contained in {(x(t). Hadamard. Hyperbolic differential equations. Institute for Advanced Study. Courant and K. G. Princeton. [ 1l] 1. [3] J. continuous derivatives of sufficiently high order to suffer jump discontinuities across I. 1948. Duke Math.. Uber das Cauchysche Problem fur Systeme von partiellen Differentialgleichungen. Interscience. It should be noted that the tpropagation of singularities cannot be derived from the Huygens principle in the wider sense. ts. 1952. Math.. Krzyianski and J. Sur la condition d’hyper- . have. 1968. x. [S] S. Math. [2] J..1223 325 Ref. 513(dtj/df)(s) = X. Assume that uk. 85 (1951) l-62. Hermann. [(t))~ T*(R”) 1 Cdxjldt) ts) =tanllatj) -(aW~xi)(s. Melrose [33] showed the following: Suppose 8 = CR c {x 11 x I< R} for some R > 0. Le probleme de Cauchy et les equations aux d&i&es partielles lineaires hyperboliques. Hermann.. For the tacoustic problem. Linear hyperbolic equations with constant coefficients. [ 151 P. (Original in Japanese. Cambridge. Ohya. Petrovskii. Sci. Friedrichs. [ 161 S. on either side of a sufticiently smooth (n . 1954. Quasilineare Differentialgleichungen zweiter Ordnung vom hyperbolischen Typus. Mat. Hilbert.) [13] K. Acta Math. Supersonic flow and shock waves. J. [S] J. Plane waves and spherical means applied to partial differential equations. and across these the partial derivatives of u have jump discontinuities (Courant and Hilbert [ 11). 1903. (x(O). k =o 91.. Then the solution u has continuous partial derivatives of sufficiently high order everywhere except on the characteristic surfaces of L issuing from F. Schauder. Anal.1)-dimensional manifold F. Math. [6] M. Applications of functional analysis in mathematical physics.. [7] F. 14 (1977)). R. Math. 24 (1935) 213-246. 815-870. Hadamard.i. Sur la condition de E. Math. Interscience. Surveys. Lax. 1950. 1965. even for regularly hyperbolic operators of second order we cannot determine the location and the type of singularities of the solutions for initial data with singularities directly from the singularities of the fundamental solution Nt. Appl. it is known that the twavefront propagates along the tbicharacteristic strips satisfyingqt--L. 2 (44) (1937). Kyoto Univ. Amer. 4 (1968) 5 11-526. Interscience.. Math. When the initial data have singularities. K.S. References M.) [ 121 S. Suppose that the multiplicities of characteristic roots of L are constant. (Original in Russian. D. Math. x. 1955. Petrovskii. Lecons sur la propagation des ondes et les equations de l’hydrodynamique. 1 (1961) 109-127. Mizohata. .. J.1=1.x. [lo] 1. for sufficiently large t.(t. 1932.. 1 c141 L . The methods of construction of asymptotic solutions of the forms (22) and (23) are also applicable to Maxwell equations or more general hyperbolic systems (R. Mizohata and Y. Inst. Some remarks on the 1 Cauchy problem. N-s. Sb. 1963. Studia Math. Friedrichs. (Original in Russian. Methods of mathematical physics II. 0. which is a property of hyperbolic equations quite different from the properties of parabolic ones. G. Asymptotic solutions of oscillatory initial value problems. Das Anfangswertproblem einer quasilinearen hyperbolischen Differentialgleichung zweiter Ordnung. Pure Appl. Lectures on partial differential equations.. Ludwig and Granoff.. [4] R.2 .Vq)=O. Sobolev.. PDEs of Hyperbolic Type strictly convex obstacles. The theory of partial differential equations..thatis. 6 (1936). Courant and D.. 24 (1957) 627-646. 1962. Pub]. Propagation of Singularities Let L be a hyperbolic operator with C” coetficients and consider the Cauchy problem L[u] = 0 with initial condition (15). Comm. Then for initial data with singularities in [l] R. gemischte Randwertaufgaben. Levi concernant des equations hyperboliques.. 162-189.

F.. Ggrding. 14 (1974). 326 (X111. 23 (1970). Sot. Ann. Chazarain. Ann. 289-370. Amer. Singularities and energy decay in acoustical scattering. 1976.. Math. 253-279. Mem. 93118. J. 277-298. Press. l-70. Systkmes IinCaires. Mixed problems for hyperbolic equations I. If there exist points where the flow is subsonic as well as points where it is supersonic. Sot. An example is the equation ( > ‘-7 ~2 azq G-2 2uv azq -+ c axay I-( ~2 azq -=o ~2> ay2 (1) of 2-dimensional stationary flow without rotation of a compressible fluid without viscosity. Melrose. 17 (1980). [31] D. 0. 569-586. 1944. 29 (1974). Inst. 349-373. Math. Ivrii and V. K.326 A PDEs of Mixed 1224 Type bolicitC pour les Cquations g caractCristiques multiples II. CBRM (1964). Petkov. Equation (1) is of elliptic type if q < c. Inst. On the mixed problem for a hyperbolic equation. [32] R. Brown Univ. (Original in Russian. Comm. Mat. Japan. M. 124 (1970). where cp is the velocity potential.. Pure Appl. Lacunas for hyperbolic differential operators with constant coefficients I.e. Almost all research so far has been on equation (2) or slight modifications of it. or tparabolic) of the equation depends on the location of the point x. D. 40 (1971).. [25] M. 16 (1964).S. i. Fourier. Surveys. u= &p/ax and v = acp/ay are the velocity components. (N. 1974. [ 191 0. Atiyah.. Math. The study of general equations of mixed type. Kreiss. C. Leray and Y. Fourier. The linearized equation takes the form $-Iqx)IZ=O. Ludwig. even when they are linear. Math. Bott. Ohya. Duke Math. 268% 286. Comm. [22] J. and c is the local speed of sound. On the diffusion of waves and the lacunas for hyperbolic equations. 24 (1974). &-well posedness for hyperbolic mixed problems with constant coeflicients. Balaban.27) Partial Differential of Mixed Type A.. OpCrateurs hyperboliques g caractkristiques de multipliciti: constante. II. [30] M. [23] M. Oleinik. Sakamoto. Sakamoto. Pure Appl. 10 (1976) 4. 24 (1974). Mathematical theory of optics. If the type varies as the point x moves. (1) :IS of mixed type.. [28] R. 105144.. [26] T. 145-206. and L. 10 (1970). On the mixed problems for the wave equation in an interior domain II. (Original in Russian. Russian Math. 112 (1971).403-417. (2) which is called Cbaplygin’s differential equation. 23 (1970). [20] V. Ikawa. Ohya. Japan. 83-84. 45 (1979j. 17 (59) (1945). Bronshtein. B.. the equation is said to be of mixed type.). Math. Math. 173-202. Comm. Uniform asymptotic expansions at a caustic. The parametrix of the Cauchy problem for hyperbolic operators with characteristics of variable multiplicity.. J. the flow is subsonic. Luneberg.) [24] I. Math. [27] H. 63104. B.. 215-250. J.. Tricomi’s The simplest Differential equation Equation of the form (2) is (3) . R. Initial boundary value problems for hyperbolic systems. 19 (1966). [lS] J. Osaka J. Petrovskii.) [Zl] Y. II. Ya. Appl. Cbaplygin’s Differential Equation It is difficult to solve equation (1) directly since it is nonlinear. 203-223. Functional Anal. hyperboliques non stricts. Sb. Propagation des singularit& pour une classe d’opkrateurs B CaractCristiques multiples et r&solubilit& locale.e.. Equation (2) is hyperbolic for :c > 0 and elliptic for x < 0. A. i. and of hyperbolic type if q > c. Math. Pure Appl. Le problkme de Cauchy pour les Cquations hyperboliques & caracttristique multiple. 43-59. The study of equations of mixNed type has become important with the development of high-speed jet planes. The type (telliptic. General Remarks Equations Let A [u(x)] = 0 be a tquasilinear second-order partial differential equation. On the Cauchy problem for weakly hyperbolic equations. J. which is a known function of the speed q =: (u’ + u2)1/2 of the flow. [33] R. However. J. we can linearize it by taking q and 0 = arc tan(o/u) as independent variables (the hodograph transformation). 134 (1973). Kyoto Univ. Kyoto Univ. Math. thyperbolic. aY2 XW)20. [29] R. Acta Math. 109-189. the flow is supersonic. 2jdme Colloque sur 1’Analyse Fonctionnelle. G. is much more difficult and less developed than the study of equations of nonmixed type. Necessary conditions for the correctness of the Cauchy problem for nonstrictly hyperbolic equations.

Atti Acad. both methods utilize energy integrals in the proof of the uniqueness of solutions. and cr that takes given values on u and on one of the two characteristic curves. Didenko. We seek a solution of (3) in the domain D bounded by AC. Friedrichs [4]. Pure Appl.. [7] V. [S] A. Akad. After Tricomi. 0. Sot. Amer. Friedrichs. Dokl. 1959.. 1. 144 (1962). Amer. P. 1970. M. Math. Tricomi proved the existence and uniqueness of the solution of his problem under some conditions on the shape of Q and the smoothness of the boundary values. 55 (1949). An initial value problem for a class of equations of mixed type. AC and BC are two tcharacteristic curves of (3) and e is a Jordan curve connecting A and B. Some systems of differen- tial equations of mixed type (in Russian)... 709-712. The study of equations of mixed type that are of more general form than (2) by means of Friedrichs’s theory is an open problem. Q Y A 0 where z=(zl. Bers. PDEs of Mixed Type which is called Tricomi’s differential equation. Sot. There is. Transl. V. C References x B 0 Fig. Friedricbs’s Theory For the study of equations of mixed type it would of course be most convenient if there existed a general theory of boundary value problems independent of the type of the equation. Nauk SSSR. Smirnov. Comm. Sulle equazioni lineari alle derivate parziali di 2” ordine di typo misto. respectively: on the common boundary x = 0 of the elliptic domain and the hyperbolic domain of the equation. 11 (1958) 333-418. Dokl. Nauk SSSR. Wiley. constructing such a general theory is considered very difftcult. [8] A. [2] L. because the twell-posedness of boundary conditions as well as the analytic properties of solutions are quite different according to the type. Equations of the mixed type. (Original in Russian.1225 326 Ref. G. say on AC. This is called the singular initial value problem. 1964.. However. Bitsadze. Bergman [3] obtained an integral formula for the solution under the condition that zi(y) and zl(y) are real analytic. Naz. 1958. Symmetric positive linear differential equations. 1 D. This boundary value problem is called the Tricomi problem. [3] S. Akad. Math. We can also consider problems such as finding a solution of (3) (or of (2)) satisfying the initial conditions partial differential equations. Pergamon. 136 (1961) 274-276. Using this observation. the following equations are treated in [S. Bull. [6] V. Mathematical aspects of subsonic and transonic gas dynamics. Boundary value problem for higher-order equations of mixed type (in Russian). (5) 14 (1923) 133-247. who noticed that although the methods of solving the tCauchy problem and the tDirichlet problem are quite different. [4] K.. For example. On the problem of Equations of mixed type in multi-dimensional domains (in Russian). The first contributor to the solution of this difficult problem was K. 901-902. much research has been done on his and similar problems for equations of form (2) [Z]. Zhegalov. BC. Further Studies Work on equations of more general type than (2) or (3) has appeared (not all depending on Friedrichs’s theory).) . I. 165-174. (Original in Russian. however. Lincei. Tricomi. Bergman. Tricomi considered the following boundary value problem for (3): In Fig. Dokl. he succeeded in constructing a unified theory that enables us to treat various types (including the mixed type) of linear equations in a single schemean admissible boundary value problem for a symmetric positive system of first-order linear [l] F. 110 (1956). Nauk SSSR. Bitsadze. Akad. a difftculty in Friedrichs’s theory since it does not give a unified procedure for reducing a given boundary value problem for a given equation to an admissible boundary value problem for a symmetric positive system of partial differential equations. Equations of mixed type. 0. V.. Math. 1978. F.) [9] M. 6. zn) and G(y) and K(y) are symmetric matrices.7]. E. S.

(7) with ac= b’. respectively. Thus we are led to the problem of expanding a given function q(x) in a tFourier series.. x and t are sometimes called the variables of space and of time. 0) = q(x). aji. constructed by superposition of the particular t = constant. are chosen so that C.115 Diffusion Processes). we can assume without loss of genera. = sin fi (x -a)exp( -.(x. we can expect that a solution to the following problem exists: Find a continuous function u(x. It has the single family of tcharacteristics According to J. There are four typical problems with regard to equation (8). t) that satisfies equation (3) for a < x < b.1. CPM> t=0. Partial Differential Type in Two Variables Equations of Parabolic B. are the roots of sin & (b -a) = 0. and the c.r. and the boundary conditions ~(a. then C.u. Equation (3) was one of the first treated in the theory of partial differential equations. equation (7) can be reduced to the: form by an appropriate change of variables 5 = U(x. ‘~~~U(X. -cc t>0. t) = cp(x). t) = 0.“=r c. considered as a function of the distance x measured along the rod and the time t. (4) +d(x. t) is the required solution.=Au-$0. t) ==u(b. t)u=g. t). t). If e’ <O in this region. to be posed (9) . t). from the outset. (5) 327 (X111. Here the i. I (2) Partial differential equations of parabolic type are important because of their connection with various phenomena in the physical world. where .u. x. they include not only equations that govern the flow of heat but also those that describe diffusion processes (. t) = cp(4. t)&+e(x.‘)~+b(x.t)~+c(x. t)+y+f(. where aij = dent variables (x. On such physical grounds. where l$ u(x. C. t) of the region under consideration. takes values given by q(x). then it can also be represented by superposition of particular solutions e -(n-xPW of (3) as u(x. Consider a finite rod with constant temperature 0 at its ends x = a and x = b.I = x1=.(x. This equation is said to be parabolic (or of parabolic type) if and only if the quadratic form x aijtitj in 5 is positive definite at each point (x. General Remarks linear partial dif- Consider a second-order ferential equation for an unknown function u of (n + 1) indepen.f)u-~-=8. t > 0. t)!$ for the temperature u(x. In fact. if v(x) is continuously differentiable. The Equation The l-dimensional of Heat Conduction case of the heat equation is We are concerned mainly with the partial differential equation of parabolic ‘:ype in two variables: a(x.327 A PDEs of Parabolic 1226 Type solutions u.26) Partial Differential Equations of Parabolic Type A. t) that satisfies equation (3) for t > 0 and that. The temperature distribution in an infinite rod is given by a continuous function u(x. all /’ (8) with a > 0. a*/ZxF is the Laplacian over the space variables. t) = (x1. t)$+2b(x. for t = 0. Thermodynamics suggests that the initial temperature q(x) (q(a)=cp(b)=O) prescribed at t =0 is sufficient to determine the distribution of heat u(x.u.lity that our equation takes the canonical form 2 a(x. t)g+c(x. t) in a rod. t) in the rod at all later times t >O. t).“=r c. Fourier the answer to this problem is expressed in a series X:1 c. t) = I-i 1 2fi 3c (p(c()e-(a-xw&. taken If q(x) is bounded. z = V(x. The most widely studied of the parabolic equations is the equation of heat conduction (or the heat equation): L[u. a<x<b. In the region where Ial + ICI >O.

the uniqueness of the solution is established in the following way: Let the curve fi and B^E in Fig. Here we impose the hypothesis on the curves x = cpl(t) and x= q2(t) that they are nowhere tangent to a characteristic (9). c = constant. t) = h(t) for given functions g(t) and h(t) is a problem of this type. and the line integral on the right is evaluated in the counterclockwise direction over C. We call (21) Green’s formula for the partial differential equation of parabolic type (3). t) that satisfies (8) for t > 0 and the initial condition limtlo u(x. Green’s Formula The tadjoint of the differential in (3) is given by operator L[u] (14) Integration by parts yields the identity (20) a solution of (8) that takes prescribed values on part of the boundary of that region.e-“n’cp. Gevrey [3] showed that if such a solution does exist.. M. au z+H~=O for x=b. t) = h(t). t) is continuous closed region (ABED). (11) exists if where g’(t) is bounded and only if (12) is of class C” and satisfies 1k’“‘(t)1 < M(n!)‘/r” for positive constants M and r.(x). (16) where G is the region bounded by the closed curve C. It corresponds to the problem of heat conduction in an infinite rod. and satisfies equation (3) in (ABED) except on AB. cp. For instance. in some neighborhood of a given curve C nowhere tangent to a characteristic.<t<t. is also a mathematical formulation of the problem of heat conduction in a rod [4]. vanishes on A^D. t)= fJ c. t) = q(x). t) such that u(x. this formula is used to establish the uniqueness of the solution of (3) and to derive an integral representation for it. and continuous. Consider the equation This problem. a solution t>o. 1 be such that no characteristic meets either of them in more in the than one point.(t)=constant. lg u. d(b)+HvnW=O s b a dxh(x)dx. t) satisfying the conditions (18) and the equation (19) The fourth type of problem is to find a function u(x. For example. (13) D. n=1 CT. (15) for sufficiently small h. Green’s formula (21).bA4 = 0. t)= g(t) and u. If u(x. The third type of problem is to find in a region of the form a<x<b. of (8) that satisfies the conditions (14’) ljLyu(x. $. t) = cpb). <x. As in the case of partial differential equations of elliptic type (. The problem of heat conduction mentioned in Section B corresponds to the particular case ‘pi(t) = constant.(x. or of a linear combination of u and au/an. the functions cpl(t) and q2(t) must satisfy the tH61der condition with exponent c(> l/2: I~i(t+h)-~i(t)l~clhl”. In the second type of problem we are required to find in a region of the form cPl(t)~xGcpz(tX t. h = constant > 0.. for which condition (15) is automatically fulfilled. a solution u that possesses prescribed values of u and au/an. . then it vanishes identically.(xO. x. the problem of finding a solution u(x. a solution u(x. If p(x) is of class C’.(x) is a normalized function (~~q$(x)dx = 1) that satisfies the boundary conditions d(a) ..1221 327 D PDEs of Parabolic Type The first consists of determining.= ah au ---= ax2 at 0 (10) where cp.323 Partial Differential Equations of Elliptic Type). 4x9 t) = s(t). According to E. posed for equation (lo). and the segment AB. (17) in the region a < t <b. Holmgren. along C. then the solution to this problem can be expanded as u(x. B^E. H = constant > 0. g-hu=O for x=a. applied to the region G=(ABQP) and the functions $ = 1.

to) as hJ0. 1 E. a) E R” x R” and t < 8. Formula (23) shows that u(xO. t. For equation (3). The function (22) is called the fundamental solution of (3) because it plays the same role as the fundamental solution logr (r=((a-x)2+(~-y)2)“2) of Laplace’s equation a2upx2+a2Ulay2=0. t) be a solution (3) and Let we of e-%(x.B. More generally. t)dx Since the integral in the left-hand side of this equality approaches u(xO. 4 . t). t)=(4n(b-t))-““exp I-&$}.. General Second-Order Parabolic Type Consider the equation be written as F. Similarly. = . Let D be a region (bounded or unbounded) of points x whose boundary is a smooth hypersurface S. (24) at ) (27) where A(t) is a second-order telliptic operator with parameter t..)e s-(x. the following function E (called the Gauss kernel) is a fundamental solution of equation (2): E(a. (16’) derivative in the ) where a/&r is the directional . + h. be its fLaplace transform with respect to t.cpw. we lhave m e-‘*u.A(r)u Equations of (I) with f=: 0. to). t) be a solution cc u(x. This idea can also be applied to the solution of parabolic equations with constant coefficients in (n + 1) variables. 1. t.. Applying Green’s formula to the region (PABQMP) and the functions cp= u(x.x.(x. ~u(x. we can establish the basic representation formula t) U(x.$ ------. t)]fI$ be a particular solution. co) equation (27) together with the conditions lpi(x.327 E PDEs of Parabolic cp= u2. A). The Laplace Transform Method Let u(x. by extending Green’s formula suitably. we obtain +1 ‘72 emA*u(x. + h. the following maximum principle holds: In the region (ABED) of Fig. x. e-%(x./ Q Fig.. Then u is identically equal to K on the segment QP and in the region (ABQP). where h is a positive number and M is a point with coordinates (x0. the desired solution of (3) can be derived by inverting the Laplace transform (25). XES.t)=(P(x). we are able to prove uniqueness theorems for the four problems stated in Section C for more general linear parabolic equations. t)dt s0 iu(x. t)dx 4% to) =s. x. t). to) is determined in terms of the particular solution (22) if we know the values of u and au/dx on the part PABQ of the boundary of the region (ABED).q(x)+ 6’6) 4x. p 1 ------.. We find in view of (3): a*u s = wx.. 1”)= s cl To obtain a representation for solutions proceed as follows: Let u(x. Once the solution of (26’) has been found. Utilizing integration by parts. = PASQ s--- u(x. t) = q(x).u(x. (25) t)dt. which can (23) for solutions u of (3). x. of (3) for t >O and 1>0. suppose that L[u] 2 0 and that u takes its maximum value K at an interior point M. XED. t) that satisfies in D x (0.-x)2/4h PC! provided that lim. t)=f(x. t)/an+h(x. yields 1228 Type where (x. such as (1). t)u(x. to. In a similar way. rj = CJ(x. s0 t)dt = [C”‘u(x. t. various versions of the maximum principle are known for equation (1) [S]. We pose the following initial boundary value problem for (27): Find a function u(x. t). t)= 0 and lirntlo u(x. PABQ +( ug-+J)dt F. t) U(x.

the inequality (29) holds. Wp(x. where u = or/@+ 1) with LY in (29) and B’= sup lo(x)I.2. 5ED.t) ) dxdt=O for any function rp(x. t) of (27) and (16’) as the limit of u. & t) 2 0. t) is expressed in the form U(t. called the fundamental solution of the linear parabolic equation (27) with boundary conditions (16’).k where 4cp. . and h(x. for any bounded solution u(x.115 Diffusion Processes). y. the inequality Iv(x)-u(y)l<A’B’Ix--yl” (30) j. 1) but independent of the particular choice of (aij). Furthermore.(x. t) depends only on 5. = k6 fork=0 . t) = 2 p=1 m e-“Vp.)v=lv-$ with 1= l/6.t) as m-co. holds for any (x. t) of (28) in (to. Nash proved that. x. That is. Thus U(& Z.(l). . t) = R. dx. R. wz = U([.7]: (i) There exists a u = U(& T. t) of (27) in the distribution sense is a genuine solution (. t) is a generalization of the function (24). 7. UC% t) = CPK) w 09x2 w D s (q(x) is the function given in (16’). m) in the usual sense. m-l witha=tfm.q starting from the initial value cpat t =O. t) satisfies (27) in D x (0.S)E(t~. BytheLaplace transform method as described above. jDUW(~ +A(t)*cp(x. XER”} and /I = a/(2cr+ 2). J. 1. T. } (A$p + &Qp= 0) and a sequence {p. < t < s. then the fundamental solution U(<. the theory of l-parameter semigroups of linear operators (. z. Then we obtain a solution u(x. x. if A is tself-adjoint.378 Semigrbups of Operators and Evolution Equations) can be applied to establish similar fundamental results. x. t) (xES). co) and vanishing outside a compact subset of D (A(t)* is the adjoint of the partial differential operator A(t) and dx = dx. t) of (27) satisfying (16’) with f(x.. As a corollary to this theorem. t-z) (t > 7). R. if the aij do not contain t and if v(x) is a bounded solution in R” of the elliptic equation obtained by replacing au/at by 0 in (28). w) U(z. where B=sup{ lu(x.(A)} of measures on the real line for which the following hold: (1) The fundamental solution U(& x. t) (x.x. .1229 327 G PDEs of Parabolic Type outward tconormal direction at (x.-I). It can be shown that a weak solution of the parabolic equation (27) is a genuine solution.$=lv. (this regularity assumption can be relaxed). t) 2 0. We put R. we can associate with $ a unique solution v of A(t.In particular. Nash’s Results Let us consider a parabolic equation is a solution of (27) and satisfies (16’). co) x R” and for any (x. Besides the properties (i) and (ii). t.Wpp(4. and further SDU(<. 4&. andofu[9]. The constant . . if u(x. Therefore this theory is of considerable significance from the point of view of the theory of probability (. (2) The solution u(x. put t.. the constants u and A are positive.(5. t) is locally summable and (28) where aij = aji are real-valued functions of class C” and equal to constants outside a fixed compact set of R” for all t 2 t. . t)l 1t > t. Then. For t > 0. any solution u(x. as a function of x and t.. then there exist a sequence of teigenfunctions {$p(x. x. . . The following results are known [6. t > ‘t 2 0) that. x. and t. f= 0. y) ER” x R”. G.. t) of class Cz in D x (0.(x.-.-. s -cc. depending on (n. w. t. s) such that x ER”. Let m be a large positive integer. By iterating this procedure m times. t) (z < w < t).x. . the fundamental solution satisfies UK. t) = 0 is expanded as e-“‘l(l. y ER”. x. In this inequality. when the coefficients of A(t) are infinitely differentiable. (ii) The function u(x. Instead. SDU(<. The Laplace transform is not suitable for solving problems (27) and (16’). t) dx = 1 under some additional assumptions. . t) defined by If the function h in the boundary conditions (16’) and the coefficients of A are independent oft. satisfies equation (27) and the homogeneous boundary conditions (16’) with rp= 0. z.. Suppose that there exists a constant 12 1 such that 1-‘J<[’ ~~aij(x~t)5ilj~al~12 for all (4X. we have a function u. and t-t and is written as U(t. then u(x. x. 2) 1 p = 1.oO) X R” x R”.125 Distributions and Hyperfunctions). GZ.

H.323 Partial ferential Equations of Elliptic Type L). Press. The unit circle 1x1~ 1 is the tnatural boundary of f(x). Bull.+a. 555102. Methods of mathematical physics. 10 (1914). 1969. Russian Math. Wewritealso1a(=cc. and 0. The number of partitions of n. 105-147.. Maximum principles in differential equations. Math.. II. 17-3 (1962). M. Linear equations of the second order of parabolic type. Interscience. (Original in Russian 1962. Prentice-Hall. 1964.p. 1953. is denoted by p(n) and is called the number of partitions of n. . Oleinik. Petrovskii if and only if there exists a positive number 6 such that Rei. D. Bochner. S.) [S] J.. t.6) Partitions of Numbers A partition of a positive integer n is. 1. [3] M. . France. Nash. an expression of n as the sum of positive integers. Les equations aux derivees partielles du type paraboliques. Lions.!+2+1= 2+1+1+1=1+1+1+1+1. Japan. Hilbert. u. 4 (1964)). H. Equation (27) is p-parabolic if --A(t) is strongly elliptic of order p.i 0. which is closely related to f(x). F. Let us denote by {2. [9] J. (Original in Russian. Eidel’man. References [l] L. 80 (1958)... 9 (1913) 3055471. [2] A. p-parabolic equations are known to be thypoelliptic if the coefficients are of class C” [ 111. the roots 1 of the equation 1”” +c’ a. where the order of the summands is ignored and repetition is permitted. G. The igenerating function of p(n) is f(x)=l+ 5 p(n)x”= ft=1 5 (l-X3jl?. A. where a = (a .j a. we can define the p-parabolic systems of equations [lo]. which is holomorphic in Ix I< 1. 1963.In (31). Amer. The Dedekind eta function. [7] A. [lo] S.. Protter and H. Gevrey.+. Prentice-Hall.. I. Euler (1748) obtained the following formula (called the pentagonal number theorem because n(3n .j(x.) [ 1 l] S. Math. S. 1967. J... Math. t) in the region under consideration and for any <E R”. Partial Differential Parabolic Type Equations of p- Let p and m be given positive integers..j)‘s such that pj + 1aI = pm and that 0 <j < m. (33) 328 (V.321 H PDEs of Parabolic 1230 Type Dif[4] R. J. Pures Appl. 2) (. John. Fundamental solutions of parabolic differential equations and boundary value problems. Math. Ann. n=1 Hence ~(7 + l)= exp(rci/l2)n(z). Math. Contributions to the theory of partial differential equations. 1 <k<m. 85 (1957). and F. Let us consider an equation for an unknown function u of (n + 1) independent variables (x. Partial differential equations of parabolic type. Friedman. 1962.< -bl&‘. 931-954.t). Hypoellipticitt ties Cquations paraboliques. Parabolic systems. Kalashnikov. t) of the type z+c a. Courant and D. Sot. [S] M.(x. The integer p is then seen to be even. Bin. Mizohata. t)&$=.143. Princeton Univ.i is the summation taken over the (aj) such that pj+lal<pm and O<j<m.(x. Ito. 1954. S.f (31) (a/ax. Continuity of solutions of parabolic and elliptic equations.j is the summation over the (a. Bers. ( n=1 I for any (x.1)/2 .Therefore. D. Springer. t)(g)“v= a. is defined by the following formula for the complex variable t taking values in the upper half-plane: q(r)=exp(rriz/l2) fi (1 -exp(2ainT)). We say that the equation (31) is p-parabolic (or of p-parabolic type) in the sense of I. (32) where Ch.. L. J. Kyoto IJniu. The mixed initial boundary value problems are investigated in detail also by Eidel’man [lo] and by R.p . 15-50. C.. North-Holland. A. L. 1961.j(X.t. Math. For example p(5) = 7 since5=4+1=3+2=3+1+1=.) and aa/axa=(a/ax. Weinberger. Studies. Similarly. The heat equation (2) is 2-parabolic in this sense. Arima (J. Eidel’man obtained precise estimates of the fundamental solutions and of their derivatives for p-parabolic equations [lo]. . 27 (1957). Surveys. [6] S. Equations differentielles operationnelles et problemes aux limites. [)}r=:=. . A’ depends only on (n.p(n) equals the number of tconjugate classes of the tsymmetric group of order n and is closely related to the trepresentation theory of this group.

l/z)= . 111. J./$ n(r). then where E is a 24th root of unity. ZZO). Later S. From this formula we can calculate p(n) successively.nis(c.h)= -$+A ( . Ramanujan (1918) proved the following inequalities.[t] . G. L. Iseki (1952).k&ev($$$J Xf(exp(y--z)). c. introduced by Hardy and Ramanujan. k).. Postnikov [S] succeeded in proving by means of an elementary function-theoretic method. The generating function f(x) varies greatly: namely. letting r+ 1-O in x = r exp(2aip/q). 6.+A > If we make substitutions a = h’. . W. in fact P. Newman (1951).1943). It is known that q(r) is a tcusp form of weight -l/2 with respect to the full tmodular group F( 1) [9.r)). Siegel (1955) gave a simple proof of the formula q( . A direct proof of the transformation formula and the reciprocity law was given by K. L 27ri s rX”+l f(x)dx ’ After.+. G. (h. Hardy and S. Hence. d are integers satisfying ad bc= 1 and c>O. we can infer from the pentagonal number theorem that n( .1231 328 Partitions number): function of Numbers f(x) of p(n): is a tpentagonal fJl-x4 = 1 + 2 II=1 ( -l)yXn(3n-w +X”(3~+lYZ)~ =W. According to Cauchy’s integral formula. The theory was improved by Rademacher who expanded p(n) into the . Hardy and Ramanujan thus obtained + O@xp(DJtt)). Subsequently they obtained Here the symbol ((t)) in the sum denotes the function that is 0 for integral t and t . we have the reciprocity law for Dedekind sums: s(h. This follows easily from the Jacobi-Biebler equality where W. With regard to the Dedekind sum s(h.. c = k.l/2 otherwise ([ ] is the tGauss symbol). for instance p(lO)=42 and p(lOO)= 190. p(n) can be represented as an integral: ~(4 -W4~4exp~~J2n/3).. d) + 7ci(a + d)/l2c). +2. which threw light on recent additive number theory. if a. Erdiis (1942) and D. we can deal with the integral by the tcircle method../I formula [ 121.k)+s(k. b = (hh’ + 1)/k. A.k)=l. fl./@ q(r). known as the CI.) is a pentagonal number.292. C. Hardy and Ramanujan proved the following transformation formula for the generating I ( 1937. hh’= -l(modk). MacMahon obtained in this way the values of p(n) for n up to 200. and d = -h in the Hardy-Ramanujan transformation formula. we obtain where the contour I is taken inside the unit circle around the origin. A. Multiplying both sides of Euler’s formula by the generating function of p(n) and comparing the coefficients. H. k) was given by Rademacher in the form By using the ttransformation formula for 9functions. where p and q are fixed integers.=k(3k1)/2 (k=O.l/r)= . where where o. then the E appearing in the transformation formula of ~(7) is seen to be equal to exp( . is defined by = 1+ -f q+(Z”+Z-“) “=I (lql< 1. where A and B are suitable constants: (A/n)e’&~(n)<(B/n)e~@J”. P. it follows from the transformation formula that f(x) w exp(w2/6q2(l . Nevertheless. By making use of a remarkable identity..569. =exp(nis(h. The size of p(n) increases rapidly with n. Iseki (1957) gave another proof by using a new method. . k)) and the value of s(h.

Markham. and p(llm+ 6) = 0 (mod 11). Ayoub. [2] G. Iseki. Many problems arise when we put additional conditions on the nj. Nat. 1963. The transformation formula for the Dedekind modular function aml related functional equations. [3] G. J. Collected paper:. J. Etienne Pascal (discoverer of the curve called Pascal’s tlimacon). 4 (1952). p(7m+5)=0 (mod7). however. 653-662. 1971.. and in 1642 invented an adding ma. After hearing of Toricelli’s experiments in 1646. Ramanujan observed that p(5m + 4) = 0 (mod5). Proc. Wright (1934). Math. Introduction to the analytical theory of numbers (in Russian). [9] M. The partition problem can also be extended to the case of an algebraic number field of finite degree (Rademacher. G. Immediately before his entry.. Amer. and these letters proved to be the beginning of the theory of tprobability.. and in this study he formulated and used tmathemat- References [1] R. Springer. An introduction to the theory of numbers. Schoenfeld (1944) S.328 Ref. Tata Inst. Mitsui (1957)).38) Pascal.chine. Pascal formulated the principle stating that pressure.. Hardy and E.. Wright. M. [12] S. 329 (Xx1. Blaise Blaise Pascal (June 19. As a youth. Rademacher.exp( c h(IllOdk. Between 1652 and 1654. K. For instance. [4] K. 1954. Cambridge Univ. he became interested in the theory of lluids and on his own began to conduct experiments. Sot. L. Duke Math. Pascal had conducted research on tPascal’s triangle. G. fourth edition.. where ‘h(n)= W. Surveys. Abh. H. J. Hardy and S. Meinardus (1953). Concerning games of chance. 2. Rademacher (1942) and Newman studied these cases by using q(z). Atkin proved that if 24n = 1 (mod5”7b11c) then p(n)-0 (mod5”71+tb~2)llc) (Glasgow Math. Akad. At present. 1970. A proof of a transformation formula in the theory of partitions. Knopp. . [lo] T. Math. this is the best result.(z)= 5 z” n:=on!T(v+n+l)’ Rademacher (1943) had developed an ingenious proof by taking “Ford’s circle” as the contour r. By me:ans of this principle. Asymptotic formulae in combinatory analysis. of Srinivasa Ramanujan.+. Iseki. Letn=n. but subsequently he began to devote himself to religion. Postnikov. Mitsui. is transmitted throughout the fluid. 8 (1967)). A. Mitsui (1978)). [ 1 l] H. Deutsch. we may require that the nj satisfy certain congruence relations (L. therefore. On the partition problem in an algebraic number field. 24 (1957). Ramanujan.beapartitionofn. Math.k)=I . Moscow. [S] A. that Rademacher (1954) proved further where L. no. Ramanujan. H. Math. A. He entered the Abbey Port-Royal of the Jansenist sect. [Sj H. Lectures on analytic number theory.. 1623-August 19. 19541955. Pascal was preoccupied with social affairs.2&m/k). Res. Rademacher. [7] H. An introduction to the analytic theory of numbers. Uber Modulfunktionen und Partitionenprobleme. In 1640. Partitions series 1232 of Numbers [6] S. 1662) was born in Clermont-Ferrand in southern France..+n. He lost his mother when still an infant and was brought up by his father. Iseki (1959)) or are powers of integers (E. 1973. Iseki (1959)) or are powers of primes (T. London Math. he discovered tPascal’s theorem on conic: sections.. he and Fermat exchanged correspondence about games of chance. Sot. More generally. L.. he demonstrated a remarkable ability for mathematics. 0. I. Clarendon Press. a vacuum cannot exist.(h. Tokyo J. 1927 (Chelsea. Topics in analytic number theory. Wiss. 1962). where he remained until his death. M. Press. S. he explained various phenomena concerning fluids such as the atmosphere and laid the foundations for hydrostati’cs. when applied at any point within a contained liquid. Hua (1942). A.. (2) 17 (19lQ 75-115.. Allg.+n. Petersson. 1965. Math. Sot. Modular functions in analytic number theory. Japan. 14-26. Berlin Kl. Fund. J. under the influence of Desargues. this research put to rest the prevailing opinion that nature abhors a vacuum and that. G. 1 (1978).

n factorial. we have (a.= x(x . 330 (11. k. the barycenter and area of the figure enclosed by a cycloid and straight lines parallel to its base. Netto. in which he carried on a dispute with the Jesuits. second edition. Wiley. (n). Lehrbuch der Combinatorik.). He also indicated a way to obtain the sum of the mth powers of the consecutive terms of an arithmetic progression. As a generalization. [2] P. he determined the area enclosed by a tcycloid and its base. he did not abandon mathematics. Pascal. The polynomial (x). [3] J. Press. Teubner.. where the sum is extended over all nonnegative pi with Z pi = n. 1968.8) Permutations Combinations and (multinomial theorem). 1915-1916 (Chelsea.. . (x -k + 1) in x of degree k is called the Jordan factorial of degree k. we have the recursive relation and in general which leads to many identities involving binomial coefficients. Oeuvres I-XIV. 1958). A subset of R is called a k-subset if it contains exactly k elements. we have a k-array or kpermutation of elements of Q.1233 330 Ref. . 1954. Pascal. and it is verified that x = 0n (x). led tLeibniz to discover the fundamental theorem of calculus. as was noticed by Pascal. the series is conver- [l] E. in particular in a tp-adic number field. As an example of binomial coeflicients with noninteger arguments. no. Permutations and Combinations ical induction. For generating function (1 + 2)” = x:0 any complex number x.1) . . Chevalier (ed. Cambridge Univ.). The number of such arrays is (r& = n(n . Pascal also formulated clear ideas about axioms. de la Facultt des Lettres de l’Univ. While in Port-Royal. = n!. he published Lettres provinciales (1657). Riordan. 1960). L’Oeuvre mathematique de Pascal. gent for IzI < 1. The recursive relation allows us to compute the values of n easily 0k References for small integers n. [3] KGkiti Hara./n! in terms of the Jordan factorial (x).. II. 1981. however. The number of ways of choosing k elements. Hachette. Combinatorial identities.. Oeuvres completes. For integral values x. Combinatory analysis I. and we have (a + l~)“=~~=~ 0 i a n-kbk (binomial theorem)../k!. from a set of n elements is Let there be given a set R of n elements. If we choose k distinct elements of R and arrange them in a row. His book Penstes shows his deep involvement with religion.+a~~=~&a~~. we have (-j’2)=(-1)“2-‘“0 References of R is i = (n).. Bibliothtque de Pleiades. . This is ais0 the number ofnonnegative integral solutions of &xi = k. The number of k-subsets (or kcombinations) binomial (-i)k~j=~+[p'). MacMahon. In any case. (1 + z)” are polynomials. (n -k + 1).. 19041914. and the volume of the figure obtained by rotating it around these lines. A.. . The arrangement of these values in a triangular form: 1 1 1 1 2 1 1 3 3 1 1 4 6 4 1 is called Pascal’s triangle. allowing repetition. The study of the methods used by Pascal to obtain these results. In particular. is the number of permutations of R. The same results hold in a tcomplete field with tvaluation. Mem. [2] J. 1927 (Chelsea. 21. which were forerunners of differential and integral calculus. d’Osaka. The 0 X coefficients 0n are defined by the z”. Brunschvicg (ed.+. B. Gallimard. [l] L. In 1658. . and with an intuitive idea of limits obtained the formula j: x”dx = amf’/(m + 1).a~~ .1) . B.

then ‘i-t T in the generiilized sense if and only if for some (or equivalently all) [sp(T) we have [gp(T. created by L. including a complete convergence proof of perturbation series. (iii) Suppose that T is a +Fredholm operator. Stability of Basic Properties 331 (XII. then T+ A is self-adjoint (RellicbKato theorem) [l. and def T . the main concern was to find solutions as power series in a parameter K that could be regarded as small. 115 (1938). A.[S] and [7]. In the former we deal with a family of operators T(lc) directly. since the latter (resp. of b for which (*) holds with some a is called the ‘r-bound of A. the perturbation method was developed as an approximation device in classical and quantum mechanics. Y). Y. in addil. but is usually more general than the former in such problems. Another major topic in the perturbation theory for linear operators is the perturbation of continuous spectra. Y).1 3) Perturbation Operators A. and A is isymmetric. [4]). We also use C(X.68 Compact and Nuclear operators F).~allull. In the perturbation theory for linear operators. General Remarks of Linear (1) Let TE C(X. . Y). more generally. (1) C(X. n+cg. r(T)).3]. r(T) (the graph of T). Continuity and Analyticity Closed Operators of F. we also have nul(T+ A) < nul T anlj def( T-t A)<defT. the former) requires roughly the constancy of the domain of the “square root” of T(K) (resp. are linear operators unless other specifications are made. N) = 0. N) = sup dist(u.7]). In this article X. Y) becomes a +metric space by a distance function a(S. If either A is r-compact or the inequality (*) holds with constants a. a(T) (the spectrum of T). T-compactness of A implies Tboundedness (and in Hilbert spaces 11 AlI T = 0). Rayleigh and E.. which was initiated by K. B(X). Y)) to denote the set of all +closed linear operators (resp. is said to converge to T in the generalized sense. N) is called the gap between hi and N [S]. In the latter case when: bp + a < p. uthf. (i) A is said to be relatively bounded with respect to T (or simply T-bounded) if D(A) 3 D(T) and there exist a. p(T) (the resolvent set of T). (ii) A is said to be relatively compact with respect to T (or T-compact) if D(A) 3 D(T) and A is compact from D(T) with the graph norm of T to Z (. When T. The infimum. denoted by I/A 11 T. T) (the resolvent of T). Important notions for characterizing the smallness of A E . T)d2&r(S).331 A Perturbation 1234 of Linear Operators B.N)=max[?j(M. the domain of T(ti)). It is closely related to scattering theory and is discussed more fully in 375 Scattering Theory.) for sufficiently large n and llR([. sectorial) forms t(K). Y) (resp. The following notations defined in 251 Linear Operators are used without further explanation: D(T).251 Linear Operators). In the perturbation theory of eigenvalues and eigenfunctions. Y). (2) Let TEC(X. then T+ AE C(X.)-R([. Historically. This generalized convergence coincides with the norm convergence if T. we deal with associated semibounded Hermitian (or. C. In this article we discuss problems in the operator formulation.amilies of In order to handle unbounded operators. r(T))<&. while in the latter. For problems in Hilbert spaces there are two general frameworks in which to formulate perturbation situations: the operator formulation and the form formulation. was laid down by F. Y) be Tbounded. X = Y is a Hilbert space. TEB(X. . 6(0. Ilull =1 8(M. Kato [2. which are important in application:<.M)]. b > 0 such that (*) ~~Aull. where for closed subspaces M and N we put &M. A standard reference in this field is [S] (also . b satisfying bp + a < p for a certain positive number p determined by T. Schriidinger.for all ueD(T). N). .4(X.+bllTu//. T)/I +O.6(N. T) having the property that &r(S). Ann. R(T). If X = Y and p(T) # 0. T. For the form formulation . Rellich [l] and T. T. T is +self-adjoint. (i) If II All T < 1 (or if A is :r-compact). X). and let AEA{X. A(X. are complex Banach spaces and T. we are concerned more generally with the behavior of spectral properties of linear operators when the operators undergo small change. and R([. then 7’+ A is a Fredholm operator and ind( T+ ‘4) = ind T (for ind T. (ii) If. nul T. Z) relative to Tare the following. it is necessary to introduce generalized notions of convergence and analyticity of families of closed operators. 81. Y). Most of the material presented in this article is taken from [S:]. however.. 6(M. B(X.-+ T in this metric.N). The latter is applicable only when there is semiboundedness (or a sectorial property) inherent in the problem. The foundation of the mathematical theory. 0. This is called norm resolvent convergence. all tlinear operators) from X to Y and C(X) = C(X. Friedrichs (Math.ion.[6.

In particular.)+ R(<.(K) @ T2(1c). Y). Namely. #‘))(I -P(O)) is the reduced resolvent.} is an isolated eigenvalue of T(O) and that m = dim P(O)X < co. the family discussed in (III) is a self-adjoint holomorphic family if T is selfadjoint and T(") is symmetric. of bounded operators is well known (. IKI<6.Then T(K) is decomposed as T(K)= T. (3) Let D c C be a domain. L. (pk(lc))} =O. of closed operators [l. + IC” T(“) + .e.P)X treduce T and give rise to the decomposition T= TIxl @ Tlx. where H(‘)u. the totality of eigenvalues of T(K) near A. In particular. i..37 Banach Spaces K).. T. b. T) strongly for some (or equivalently all) [ with Im 5 #O. (iii) T(K) U(K) = V(K).e. (I) if X = Y and if [E~(T(K)) for all ICED. A of c(T) is by a simple and A(o(T)\ A) the operator (1) Separation of the spectrum. H(‘h. including the case of a degenerate lo (m > l). = (I . . . D. . Then dim P(rc)X = m. Suppose that a bounded subset separated from the rest of c(T) closed contour I. there exists 6 > 0 such that I KI < 6 implies (T(K)).uED(T). is expressed by one or several power series of K~/” with a suitable integer p > 0. which is roughly the strong convergence of resolvents. it may happen that the coefficients kj and uj of formal power series can be computed up to a certain j. The closed subspaces X. with /uoll = 1 and where S = lirn<+ R(c.(K) in this case is just the finite-dimensional eigenvalue problem det {lhj. I KI < 6. Suppose now that A = {A.X). the power series A(K) = C)Ljd can be explicitly cofnputed as A1 =(H(%A. 51. in which the situation becomes more complicated due to the splitting of eigenvalues. and let T(")cA(X.-t T strongly in the generalized sense if R(c. Then T(K)u = Tu + KT% + . (III) Let TE C(X. +A. = PX and X. uo) +(SH(‘h. This notion is generalized to a family T(K)E C(X. Thus the problem for T.1235 331 D Perturbation of Linear Operators r c p (2) When X = Y. . Moreover. there is also the notion of strong convergence in the generalized sense [S].).[S]. and the problem of determining the spectrum of T(K) inside r is reduced to the problem of determining the spectrum of Tl(~) (I KI < 6). defined near K~. -( T(K)c~~(K).(i.(IV)IfX=Y is a Hilbert space and if T(K) is self-adjoint for real JC. P E B(X) and P* = P). when T” and T are self-adjoint operators in a Hilbert space. . Thus the separation of the spectrum discussed in (1) is applicable to T(K). wherea. .). then T(K) is holomorphic in D if and in D. This follows from the upper continuity of compact components of the spectrum with respect to the metric d of C(X) [S].. T(K) is said to be bolomorphic in D if at each K~ E D there exist a Banach space Z and U(JC)EB(Z. The notion of analytic&y (holomorphic property) of a family T(K)EB(X. c > 0. .. Y). (P. . (ii) U(K) is one to one and onto D(T(lc)). . If H(K) = H(O)+ &P holomorphic family + . This is called strong resolvent convergence. . u. such that (i) U(K) and V(K) are holomorphic at K~ as families of bounded operators. is a self-adjoint described in example (IV) in (3) of Section C and if m = 1. This theory is called asymptotic perturbation theory. u E D( T) defines a holomorphic family of type(A)inD={?cIIicI<(b+c)-‘}. (II) If only if R([. T.5]. corresponding to the projection R(i. Let us mention several special cases.u. V(K)EB(Z.. . Suppose that A and r are as in (1) with T replaced by T(O). KED. Then which is independent of r.. is a projection (i. Y).. . T(K)) is holomorphic D(T(K)) is independent of K and if T(K)u is holomorphic in D for every UE D( T(K)) then T(K) is holomorphic in D (holomorphic family of type (A) [S]). The case of degenerate 1. Perturbation of Isolated Eigenvalues Let TE C(X). W)di. We assume that OED and regard T(O)= T(0) as the unperturbed operator. (3) Even when a problem cannot be handled by means of analytic perturbation. =(H%. = Tl @ T2.T(K) is said to be a self-adjoint family. a( TJ = c(T) fl {inside of r} and a(T. ..e. In particular. (2) Let T(K) be holomorphic in D. KED. The power series for the associated eigenvectors U(K) = C &uj can also be computed. . Y)suchthat D(T("))ID(T) and ~~T(")u~~~c"-'(~~IuII+~IITuII). This series is known as the RayleighSchriidinger series. If T(K) is a self-adjoint family. In many such cases it can be shown under general assumptions that an asymptotic expansion such as n(lc)=n. a base {cpl (K). = I. In particular. Estimates for O(K) can also be given.5]. The strong convergence in the generalized sense mentioned in (2) of Section C is used here.(IC)} of P(K)X can be constructed in such a way that the (Pi are holomorphic in { llcl<6’<6} [3.)=o(T)rl {outside of r}. we can take p = 1 so that the eigenvalues are holomorphic near 1. r c p( T) lies inside (outside) of r). For details. K+O(K) is valid as long as the coefficients involved can be computed legitimately [2. This provides a rigorous foundation for the perturbation method in many important practical problems. can be treated similarly. The totality {lj(~)} of solutions of this equation. Y). The perturbation theory discussed in this subsection is called analytic (or regular) perturbation theory.

. and D. von Neumann.375 Scattering Theory. (ii) If HO and H act in different Hilbert spaces X0 and X. Rend. Circ. 113 (1’937). Ind. M. Proc. 34 (1958)). Japan. (1) The essential spectrum (. Ann. Kato. In fact. Perturbation theory for linear operators...375 Scattering Theory). Pacific J. if H = HO + K with compact K.390 Spectral Analysis of Operators E) is stable under compact perturbation.). J. especially for some concrete problems.. References [l] F.. . An eigenvalue 1. Rellich. 7 (1957). in relation to resonance poles (or poles of the holomorphic continuation of the resolvent or the scattering matrix). F. belongs to B. 27 (1909)). More generally.(H) = a. 0. then there exist a unitary operator U and a compact operator K such that H = UH. Math. Combes. In some problems. (1963). certain parts of continuous spectra tend to be stable under perturbation. of HO which is embedded in the continuous spectrum may diffuse into the continuous spectrum in the presence of a perturbation. B.. if X is separable and if u~(H)=a.331 E Perturbation E. Voigt (1977). H(K). . U-’ + K (J. Rather. Namely. Mat. studying the mode of change under perturbation is not usually a tractable problem. J. Nauk SSSR. On the perturbation l.. For the topics mentioned in (2) and (3) - c71. Math. Palermo. Also . has no eigenvalues near Lo but may have a continuous spectrum highly concentrated around /I. Some problems of resonance can be treated by the technique of dilation analyticity. In such a case. 462-484. J. J..heory of closed linear operators. HO) is compact for some (or equivalently all) [e p(H) fl p(H. T. Comm. Math.) and q(H) are unitarily equivalent for any smooth strictly increasing real function cp (M. Kato. Sh. it is proved that the first few terms of the perturbation series for L(K) that are still computable are related to the real part of the resonance. Among generalizations we mention the following two. 15 (1965)). Moreover. Functional Anal. All the results mentioned above are proved by scattering-theoretic methods. 27.. (3) A vast quantity of results in the spectral theory of the Schrodinger operators appearing in the tschrodinger equation in quantum mechanics can be obtained by perturbation methods. [2] T.(H. [5] T. Stijrungstheorie der Spektralzerlegung I-V. Math. 6 (1951) 145-226. On the convergence of the perturbation method. Perturbation theory for absolutely continuous spectra is closely related to the study of generalized wave operators in scattering theory. (2) The detailed structure of continuous spectra is hard to analyze.) (H. Univ. either by the wave operator approach or by the abstract stationary approach (. (2) The absolutely continuous spectrum (390 Spectral Analysis of Operators E) is stable under perturbation by the ttrace class.. T. 117 (1940). then the same conclusion as in (i) holds (D... D. Voiculescu (1979) have extended these results to normal operators and m-tuples of commutative self-adjoint operators. For B. respectively. Aguilar and J. [3] T. mat. Springer.. with K E B. ser. T. Pearson. Proc.9].. then the absolutely continuous parts of cp(H. not discussed in this article. Akad. Friedrichs. Sikonia (1971) J. Izv. Perturbation 1236 of Linear Operators of Continuous Spectra the existence and the completeness of the latter implies the unitary equivalence of absolutely continuous parts. Pure Appl. then the absolutely continuous parts of HO and H are tunitarily equivalent. Pac$c J. For continuous spectra. C). Math. with I/K lip < E for any p > 1 and E> 0 (S... In this section we discuss only self-adjoint operators and let H=jldE(I). Kato. and if there exists J E B(X. Phys.) be self-adjoint operators in a Hilbert space X. 116 (1939) 5555570. if H = HO + K. Tokyo.390 Spectral Analysis of Operators I.. H).. it suffices to assume that R(<. 4 (1952).. any self-adjoint operator H in a separable Hilbert space can be changed into H + K with a pure point spectrum by adding a K EB.X).). 361-406. Fat. HO)~B1 for some [E p(H) fl p(H. 113 (1937) 677-685.. 28 (1978)). I.68 Compact and Nuclear Operators. 33 (1957)). I. 22 (1971)). 7. Birman. 229 (1935)).. (i) If R(<. Kuroda.ich is also effective in other problems of spectral analysis (J.) (M. X) such that JD(H. 118 (1942). and the study of this stability has been a major topic in perturbation theory (also .R([. Kato. H) . . Kato. Japan Acad. ActualitCs Sci. Math. Weyl. 356-382. (i) can be derived from (ii). H. sec. and 11 lip to be used below .[S.. and in particular o. .(H.) c D(H) and such that the closure of HJ-JH.R([.(H. then CT~(H)=O. On the perturbation of continuous spectra. esp. Japan Acad.. 1 (1948). This phenomenon of spectral concentration is studied.(X. 1966. [4] K. Berg (1971) W. 600619. Some Other Topics (1) For the perturbation theory for semigroups of operators and evolution equations. IC#O. a technique wh. 3233337. Sot.). Sci. Comm. Conversely. Namely. Rosenblum.

. C41). V&e represented 2/a in the following infinite product: A formula combining Machin’s representation of rr and the power series Arc tan x = x (1/3)x3 +(1/5)x5 . W. Thus rc can be defined as 332 (V1. Since the 17th century. The approximate value of K up to 50 decimals is 3. In 1882. obtained the first 21. 1975.g.7) Pi (71) (W. H. in 1873 W.(Z.6. In the 17th and 18th centuries. Lindemann proved that x is a ttranscendental number using Euler’s formula eRi= -1.1416.3. Simon.. using information in Cl].2.. is called Machin’s formula and was often used for calculating an approximate value of n. &L+!. obtained by the metrical theory of continued fractions. = 4 Arc tan l/5 . In 1761. According to the Rhind Papyrus. 70 (1951). 1978. which is valid for almost all irrational numbers (e.. Table 6). F. L. Brouncker.6.. Sot. Wallis) [6] N. Equation of evolution. Adriaen van Roomen obtained nk 355/l 13. K. - .230 partial denominators of the regular continued fraction representation of ?I and described how their numerical evidence tallies with theoretical results. Pitman. [9] H. 1979. [7] M. Tanabe. (Original in Japanese.. P. 1975. J. Dunford and J. 2n -L 4 lZn=4z hold.. . Euclid gave no statement about the numerical value of 71. Gregory. Archimedes obtained 3%~ rc< 35 by calculating L. In 5th~century India. the initial letter of aspz~srpo~ (perimeter). (J. Various numerical results obtained by electronic computers are not formally published. C. 195-211. Linear operators I-III. Shanks obtained an approximate value of 7cup to 707 decimals. and an approximate value up to l. 1979. the Japanese mathematicians T. Lambert used Brouncker’s expression of a in a continued fraction to prove that R is irrational. Ferguson calculated 710 digits of x and found that Shanks’s value was correct only up to the 527th digit. G.Arc tan l/239 The symbol 7chas been used since W.5. Methods of modern mathematical physics I-IV.14.. Choong et al.4..OOO decimals has been obtained.Appendix B.3. Trans. The computation of an accurate approximate value of n has been made easier by the recent development of computing machines. In Sth-century China.OOO.) be the perimeter of a regular n-gon circumscribed about (inscribed in) a circle of radius 1. the notation =1-1/3+1/5-l/7+ The ratio of the circumference of a circle to its diameter in a Euclidean plane is denoted by A. Seki. Leibniz) (J. Schwartz.4. Euler..197l.5. (J. Matunaga computed rc to 50 decimals. Math. Beckmann [Z] gives a detailed and humorous historical account of the calculation of n from ancient times up to the present computer age. These values were obtained by methods similar to those of Archimedes. L.. Machin). 1958. The fact that this ratio is a constant is stated in Euclid’s Elements. . Reed and B. many formulas that represent rc as a sum of infinite series or as a limit have been used to obtain more accurate approximate values. F. [S] T.7. (. 3 has been used from antiquity. Amer.1237 332 Pi (7~) Using this formula. T.141592653589793238462643383279502884197 16939937510.>n>l. No improvement of his approximation was obtained until 1946 when D. [3].. (4/3)4 was used in ancient Egypt. Jones (1675-1749) and L. and I. By utilizing this formula. . and Y. F. Academic Press. Let L. and in 16th-century Europe. Wiley-Interscience. 7c 1 12 32 52 -=4 1+2+2+2+. Takebe..As an approximate value of n. however.) 7=1. In 3rd-century China Liu Hui used x + 3.1610) calculated rt to 35 decimals. The following are representations of x known in those days: n 2. F. 1972. 1963. 83 Continued for Fractions) . Then the relations L. Fundamental properties of Hamiltonian operators of Schrodinger type. Tsu Chung-Chih mentioned 2217 as an inaccurate approximate value and 355/l 13 as an accurate approximate value of n. Kato. some being deposited in the UMT repository of the editorial office of the journal Mathematics of coniputation. Aryabhatta obtained rc+ 3. van Ceulen (1540.

Noordhoff. 2): (1) Every q. Dp are the subdomains constructed from q.18) Plane Domains A. Some typical examples of domains are as . B. Continued fractions. (6) The complement of D with respect to the complex sphere is a connected (not necessarily arcwise connected) closed set. (2) Half-plane: Re z > 0. contains all qv except for a finite number of v. Comp. In a domain D. second edition. consists only of the boundary points of D. 1238 follows: (1) Circular domain: Iz . .. Pz. Khinchin. W. (5) Slit domain: a domain obtained by excluding a Jordan arc I from a domain D. The intersection n 0.000 decimals. it can always be continuously deformed to a point. a tconnected open set) in the tcomplex plane or on the tcomplex sphere is called a plane domain. [4] A. (2) For v-00. Comp. Wrench. Rational approximations to rc. (4) Annular domain: r < 1 z .. (3) Angular domain: c(< arg(z -c) < p. The tclosure of such a domain is called a closed plane domain. Qn-l such that D (Q. and q. Choong.. A sequence {qy} of cross cuts mutually disjoint except for their endpoints is called a fundamental sequence of cross cuts if it lsatisties the following two conditions (Fig.. where all points on F (except an endpoint lying on the boundary of D) are contained in D.1 suitable mutually disjoint A boundary point P of a domain D is called accessible if there exists a sequence of points P.(D.O<y<l/2(n=l. Caratheodory [2]. 76-99. and C. Daykin. 25 (1971) 3877392. Pi (7-r) References [1] D. Jordan domains are simply connected. Domains in the Complex Plane A +domain (i. If D is a domain on a complex sphere. lie completely in D. for the domain obtained by removing x = l/(n + l). there exist n . The thomology group H. separates qy-1 and qv+l on D. 1 Let the domain D be bounded by a smooth Jordan curve. E. {q\} of cross cuts are equivalent if every D.. contains all qk except for a finite number of v.. A curve in 1) converging to the point P from the interior of the angular domain D' is called a Stolz’s path or a nontangential path ending at the point P. (5) For every Jordan curve C in D.O < y < 1. the sequence qv tends to a point on the boundary. If D is an n-ply connected domain. This notion is due to C. Let n 3 2 be an integer and D a plane domain.)fromth~zsquare 0 <x < 1. Calculation of rt to 100. For example. and denote by D. 1963. the Jordan arc I is called the slit of the domain. the subdomain of D separated by qy that contains qy+l. Boundary Elements 333 (11. that is.. D . Math. tending to P such that the line Isegments P. [2] P. Math. either the interior or the exterior of C is contained in D. the boundary points with x = 0. Two fundamental sequences {q”}. and every 0.. (2) Every Jordan curve in D is thomotopic to one point. a Jordan arc whose two endpoints lie on the boundary of D is called a cross cut of D. D. and a plain domain is called simply a domain.. 16 (1962).c I< R. we consider only subsets of the complex plane (or sphere). Let {qv} be a fundamental sequence of cross cuts. In this article.Q has exactly two tconnected components.332 Ref. [3] K. The boundary element of a cross cuts Q. Fig. Rathbone. R.. Z) is identical to Z”-’ if and only if the complement of D in the complex sphere has n connected components. 1971. The tinterior of a +Jordan curve J in the complex plane is a domain called a Jordan domain. (3) The tmonodromy theorem holds in D. . each of the following three conditions is equivalent to the condition that D is simply connected: (4) The boundary of D consists of a single tcontinuum or a single point. or Im z > 0. A history of pi. Let D be a simply connected domain. Here D. Y. This condition determines an equivalence relation. under which the equivalence class of fundamental sequences of cross cuts is called a boundary element. Beckmann.c I< r. Shanks and J.e.. Then D is said to be n-ply connected or multiply connected without specifying n.. each of the following three conditions is equivalent to the condition that D is tsimply connected: (1) For every cross cut Q of D. For a domain D.2. Take an angular domain D’ with vertex at the point P and the initial parts of the two sides of D’ lying in D. and let P be a boundary point of D. U U Qnml) is simply connected. Golem Press..O < y < l/2 are all inaccessible (Fig. as above. . In this case. 1).

‘)/2 is the +mean curvaand R. A closed domain is usually considered to be the union of a domain and the set of all its boundary elements. Fig. i. r=a2. or (1 +$)r-22pqs+(l +$)t=O. [2] C. Meusnier (1776). Cambridge Univ. A. with n = n(u. Math. If every infinite subsequence of {G.1239 334 B Plateau’s Problem multiply connected domain is defined similarly for each isolated component of the boundary. A. Elements of the topology of plane sets of points. and R. which is a secondorder tquasilinear partial differential equation of elliptic type and whose geometric interpretation had already been given by M.33) Plateau’s Problem A. The union K of such domains G is called the domain kernel of the sequence {G. Let D be the finite domain bounded by C.. there exists a domain G such that every closed domain containing the origin and contained in G is contained in G.} has the same domain kernel K.} be a sequence of domains containing the origin 0. 323-370. for all v.. u satisfying E = . If there is no neighborhood of the origin contained in G. H. Since +Beltrami’s form satisfies A. We consider surfaces z = z(x. u. second edition. The +Euler-Lagrange differential equation for the functional J[z] is References [1] M. aZ with the condition that z = z(x. 73 (1913). Caratheodory. let a surface be expressed in vector form x = x(u. For example. It is a problem of the tcalculus of variations.207 Ideal Boundaries). a soap membrane bounded by a given closed space curve takes condition for a minimal surface becomes Ax = 0 (with A the +Laplace operator) provided that isothermal parameters u. 1951.77 Conformal Mappings). B. are the curvature. Let its +first fundamental form be dx2 = Edu2 + 2F du dv + G dv2 and its tsecond fundamental form be -drdn=Ldu2+2Mdudv+Ndv2. Uber die Begrenzung einfach zusammenhangender Gebiete. then we say that the sequence {GY} converges to K. where H = (R. the 334 (X. we put K = (0). By equating to zero the +lirst variation of the areal functional based on infinitesimal displacement in the normal direction. each point of a slit domain. the shape of a minimal surface. t = d2zldy2. Ann. for all v. s = a2z/axay. This experiment was performed by the Belgian scientist J. Then under suitable assumptions on the smoothness of z(x. y). To formulate the problem more generally. v) by means of parameters u. Origin Because of surface tension. y) having common boundary r. The notion of domain kernel is important in considering the limits of a sequence of conformal mappings (.x = Hn. hence Plateau’s problem is that of determining the minimal surfaces bounded by given closed space curves. a surface of the least area. Various notions of tideal boundary come from considering suitable boundary elements for various purposes (. Press. If a suitable neighborhood of the origin is contained in G. C. Formulation Let F be a tsimple closed curve in xyz-space such that its projection C on the xy-plane is also a simple closed curve. we obtain the EulerLagrange equation in the form 2H=(NE-2MF+LG)/(EG-F’)=O. except for a finite number of v. y) has r as its boundary. Plateau (1873) to realize minimal surfaces. Domain Kernels P=z’ q=&’ Let {G. 2 aZ C.2/ax2. determines two distinct boundary elements on each side. u) the unit normal vector.e.} (Carathtodory). Newman.’ + ture of the surface +radii of principal second differential R. except for the endpoint of F lying on the boundary of the domain. the problem is to minimize the tfunctional J[z]= JJ /mdxdy.

275 Minimal Submanifolds). Let q(u. Douglas. Bernshtein (1910) from the viewpoint of a tboundary value problem of the first kind for the elliptic partial differential equation in the previous section. when r is assumed merely to be trectifiable. Courant. where +genus and orientability are assigned as the topological structure of the surface to be found.is a simple closed curve with bounded curvature. the problem can be easily generalized to an n-dimensional Euclidean space R” (. In general. N. the existence proof of the solution of the former for a Jordan domain implies +Riemann’s mapping theorem together with W. On the other hand. by Rad6. Previously. u). F. For instance. . R. the vector function that minimizes the Dirichlet integral among functions with fixed boundary values is harmonic. He further discussed the general case where r can bound a surface with finite area. Riemann. Specifically. At present.77 Conformal Mappings). Based on these facts. New Developments Among recent contributions to the study of Plateau’s problem. Weierstrass. C. u). the following remarkable with components fj(u. Subsequently. Garnier (1928) investigated the existence of a solution by the limit process when l. (3) An analytic vector s(w) is representable in terms of the boundary values of its real part. On the other hand.e. Rad6 (1930). C. In this formulation. A. by introducing a new functional depending on boundary values instead of the area1 functional. J. Courant. and others discussed the case where the given space curve r is a polygon. Then isothermality is expressed by the condition that the analytic vector ~(w)=x(u. u) (j = 1. However. The Generalized Case Up to now we have been concerned with Plateau’s problem in the case of a single simple closed curve. F = 0 are chosen.]. Douglas (193 I) succeeded in giving a satisfactory result for the existence problem. Douglas transformed Dirichlet’s functional with harmonic argument functions into a functional whose arguments are boundary functions. (2) Dirichlet’s functional for a scalar function f(u. then Poisson’s integral formula 5(w)=eie+w de + iIm s(w) b(Q)e’*-w 237s 0 with the boundary function b(0) =: Re g(e”) can be used.i= J-1) satisfies r(w)’ = 0 (Weierstrass). CarathCodory’s result on boundary correspondence (. ss 2n 2n h(4 . The existence of a solution of Plateau’s problem can be discussed by starting . . u) be a harmonic vector conjugate to X(U. E. and others treated the generalized case of a fmite number of boundary curves. n) by D[f] = &. B Morrey (1948) generalized the problem by replacing the ambient space R” by an n-dimensional +Riemannian manifold and gave the existence proof in considerable generality [6]. Schwarz.bka2 0 o 4sin2(0-(p)/2 dOdq.Z)dudu. Relation to Conformal Mappings There is a notable relation between Plateau’s problem and conformal mapping when the dimension of the space is 2. The variational equation of D[f] is Af= 0. and for an n-dimensional vector function f(u.lse also. and Douglas): (1) The first method is to minimize directly the areal functional jj Jwdu do. .u)+~~(u. Courant (1937) gave another existence proof by reducing Plateau’s problem to the IDirichlet principle [3]. the methods of discussing the existence of solutions of Plateau’s problem can be classified into the following three sorts (represented. if the domain of w is the unit disk IwI < 1. and Plateau’s problem is to determine the minimal surface with a preassigned boundary. of D. u) representing a minimal surface is harmonic (the components of this vector are tharmonic functions of u. R. the existence of a solution was first shown by the limit process by T. a minimal surface is defined as a surface with everywhere vanishing mean curvature. the vector x(u.u)isdefined byD[f]=~~(f. The variational equation of the areal functional becomes H = 0.334 c Plateau’s 1240 Problem from the variational problem of minimizing D[f]. Haar (1927) dealt with the minimal problem for the functional J[z] by a tdirect method in the calculus of variations. . Osgood and C. The problem is further generalized from the case of fixed boundary to the case where the boundary is merely restricted to lie on a given manifold [3]. respectively. On the other hand. Namely. F.u) (w=u+iv.. i. by starting from the problem of minimizing Douglas’s functional 1 2n G. in connection with the +monodromy group concerning a second-order linear ordinary differential equation. D[f.2+f. Existence of a Solution The existence of a solution of Plateau’s problem was discussed by S. The existence of a solution has been shown in this c. u) we can prove the existence of solution Plateau’s problem satisfactorily.

Next. Douglas. [7] R. One of them is connected with the final result of Douglas (1939) on the existence of solution surfaces. Oeuvres I-XI. Henri Henri Poincare (April 29. 1903. 1966. 319-329. Science et hypothese. A new uniqueness theorem for minimal surfaces. [4] J. His paper on the tthree-body problem won the prize offered by the king of Sweden in 1889. Poincare. The methods he developed in his threevolume Mkanique cdeste (1892. [6] C. His achievements center on analysis and applications to theoretical physics and astronomy. 550569. Anal. his work covered many fields of mathematics. J. the surface is also of class Cm*” up to the boundary. Poincare opened the road to talgebraic topology and made suggestive contributions to the ttheory of relativity and tquantum theory. 1966. J. C. H. Arch. C. Hildebrandt. Osserman. [S] R. Osserman. C. After graduating from the Ecole Polytechnique. He asserted that science is for science’s own sake [4]. Osserman (1970) and R. Les mithodes nouvelles de la mecanique celeste l-111. GauthierVillars. but as subsets of R”. Amer. Almgren. Arch. algebraic geometry. Plateau’s problem. which is a surface of least area. is an immersion. Federer (1969). C. Benjamin. 1892-1899. Poincare. References [l] T. These points are called branch points. 1950.e. Nitsche. who sought to minimize the tHausdorff measure among general classes of geometric objects. [2] J. Springer. Anal. . Ann. 1975. Trans. Further developments in connection with Plateau’s problem have emerged in the work of E. Math. There are also some recent results on the number of solutions of Plateau’s problem. Osserman also gave examples of generalized minimal surfaces in R” (n > 3) with true branch points. In addition.. F. Gauthier-Villars. Hildebrandt (1969) and others proved that if the boundary is of class Cm+. Solution of the problem of Plateau. Poincare. PoincarC. La valeur de la science. C. The mapping of a 2-dimensional manifold with boundary into R” defining Douglas’s solution of the Plateau problem for a finite number of simple closed curves is a tminimal immersion with the possible exception of isolated points where it fails to be an immersion. then at the University of Paris in 1881. B. 1916-1956. Springer. [2] H. Henri results have emerged. not as parametrized manifolds. For instance. Math. 335 (XXl. Multiple integrals in the calculus of variations. Courant. Poincare. spectral theory. i. It was then proved by R. He was made a member of the Academic des Sciences in 1887 and of the Academic Francaise in 1908. teau’s problem. Morrey. Flammarion. [3] R. and topology. Fleming.39) PoincarQ. Math. 52 (1973). The existence and regularity of solutions of Plateau’s problem from this point of view have been discussed by H. However. Van Nostrand. Federer. S. 1933.1241 335 Ref. D.1899) began a new epoch in celestial mechanics. R. Reifenberg (1960) and others. 1914. Rado. Interscience. C. Rational Mech. [S] S. [3] H. is free of branch points. [lo] H. Lewy (1951) proved that if the boundary of a minimal surface is analytic. then the surface is analytic up to the boundary. conformal mapping and minimal surfaces. [9] J. A survey of minimal surfaces. In this connection. Subsequently. Vorlesungen iiber Minimalflachen. he taught at the University of Caen in 1879. [4] H. 33 (1931) 263-321. m > 1. Almgren. W.. H. Gulliver (1973) that for n = 3 the mapping of Douglas’s theorem. and his popular philosophical works concerning the foundations of natural science and mathematics exhibit a lucid style. His tuniformization of analytic functions by means of the theory of tautomorphic functions in 1880 is especially notable. we mention the question of boundary regularity. Nitsche. Dirichlet’s principle. On the problem of Plateau. France. J. 1912) was born in Nancy. Gulliver also dealt with an analogous problem for surfaces of prescribed mean curvature. An invitation to varifold geometry. [ 1 l] F. He died in Paris. 1854-July 17. Springer. 47-82. (2) 91 (1970).. Erg.. 1969. A proof of the regularity everywhere of the classical solution to Pla- References [l] H.. and others [lo]. Rational Mech. Geometric measure theory. 35 (1969). including arithmetic. Flammarion. Nitsche (1973) proved the uniqueness of solutions for analytic boundaries of ttotal curvature at most 4a. Boundary behavior of minimal surfaces. Springer. 1969. Sot..

f(x) . can uniformly approximate each continuous function on [0.. + t (a. x”} on [a.*(f)<c..a... 11 with arbitrary precision if and only if C pi’ = --co (Miintz’s theorem). then for every E> 0 there exists a polynomial P. We have the following generalization of(i): Let p. 1908.. = co. . the modulus of continuity of ktb order is defined by B. are n + 1 arbitrary distinct points of A (Haar’s condition) (Math.(x)1 C. General Remarks Approximation On the existence of polynomial approximations.20) Polynomial A. xnml + . holds for all values of 0. B.-ix”-‘-.(f.(x)=cos(narccosx) is the Chebyshev polynomial of degree n. M. is called a Chebyshev system (or unisolvent system). 11.. H. Xpz. Degrees of Approximation Continuity and Moduli of For a continuous function . the system of functions {‘pk}k=o. .(x).P”(x)1 attains its maximum (these points are called deviation points) and (f(xi)-p.. q.(x) = 2-(“m1)T.(x)=C”. .. For any given n there is a best approximation of Jr(x) by a linear combination of q. Ann.(x)}. with real coefhcients such that max Ix”--a.(x) that is not identical to the function f~C[a. let P. k=l (1) such that the inequality If(o)-P..(x)=if . C161. cos nx} on [0.(x) is a trigonometric polynomial of the form (I). . i = 0. Then x” . Best Approximations Let qo(x). which is formulated in the following two forms: (i) Iff(x) is a function that is continuous in the finite interval [a.f(x) if and only if there are at least n + 2 distinct points xi < . < x.f(x) defined on [a.. we can assume that the given function f(x) is defined in the segment [0...h.(x) = a.c. p2.336 A Polynomial 1242 Approximation Science et mtthode. where ck is indepen- . be a sequence of positive numbers such that limp. .x*. Consider the Bernshtein polynomial -xp. b]. cosx. consider the polynomial P._. The . x. P”(0) = u. b]. . b].cos k0 + bk sin kO).(Xi+ I)<0 (i= 1.C”xtA for t <(b -a)/k. sin nx} on [0. . several mlethods have been developed to find it (.. ~JX). . n + 1) (Chebyshev’s theorem).. 78 (1918)).. Poincare. For example.(x) be a linear combination of q. . + a. b].. Then linear combinations of x0 = 1.(x) converges to f(x) tuniformly.(x) of degree n = n(s) such that the inequality If(x) . ‘p... xi. { 1.(Xi))(f(Xi+. .P. For a Chebyshev system {qk(x)} on [a. l/(n+ 1)) (Jackson’s theorem [I]). n) is not zero. a function P.hlf(x)-~‘.T(x) by a linear combination of (cp. where x0.=. we mention two simple ones. b].1 -1 GXCl takes its smallest value.(x) is the best approximation for . Flamis called the best approximation of . Thus we do not always have a unique best-approximation polynomial [S] H. then corresponding to every positive number E there exists a trigonometric polynomial of degree n = PI(E).cp.. For such an approximation to be unique it is necessary and sufficient that the determinant of the matrix (qk(xi)) (k. b]. (ui i:< the ordiPut E. Then E.(x)} satisfies this condition..+~ of [a.300 Numerical Methods).)-P. .. we can apply tFejer’s theorem on +Fourier series. To prove version (i) of the theorem.*(f)= info“. b]. Then P. For any given continuous function f(x). . and conversely. nary modulus of continuity.w. (ii) If . when the set 4 c R” (n > 2) contains three nonintersecting arcs emanating from a common point. be a sequence of linearly independent continuous functions on a bounded closed domain A in R”.max. 1. Since the best approximation is desired for numerical computation. Of the many direct proofs now available for Weierstrass’s theorem. However.(O)1 <F.. . Xk(l Ii=000 Then B.--a.. rr] and {sin x..(x). xp’. The second form of Weierstrass’s theorem follows from the first.. where T. x + a.(x).>. C. x. where If(x) . sets { 1.. In particular. where f is a continuous periodic function of period 271 and P.(x). .P. ._. marion. Stone obtained a theorem that generalizes Weierstrass’s theorem to the case of functions of several variables. rr] are Chebyshev systems. .(x)l.2.(x) attains inf max I.P”(x)1 GE holds throughout the interval [a. . h.. If {cp. we have Weierstrass’s approximation theorem.. 336 (X.f(0) is a continuous function of period 2n. To prove (ii). but such an approximation is not always unique. A admits no Chebyshev system.

119 If(x) .(x)dx.(x). . U..(f. For the approximation of fec( [ -1. Griinwald [ 111). we can conclude that there is a continuous function defined in [ -1.(x)l.(x) the one that gives the best mean square approximation to f (i. l] and any polynomial degree n. . q. Although the partial sum s. then f(z) is approximated uniformly by polynomials on any compact set in E (Runge’s theorem). Consequently. B. . There is no linear operation that gives the best trigonometric approximation.P. 11) by polynomials.(f. and n.(x) of degree at most n such that for any Interpolation Polynomials xsc-1. G.(x) also gives the best approximation up to a constant factor. . there exists a polynomial P. E..L-4= 1/2)(x-xj)) &jiof(xJ 1 2n sin((x . tFejCr means e”(x)). there is a continuous function for which &(Xx) diverges everywhere (J.‘(x)1 < nmax. Sunouchi and C. For L. 1T. f(‘)(x) is the rth derivative of f(x). and his results were developed by J.(x). we have the Bernshtein inequality max.c =-s for x E [ -1. If xk = 2nk/(2n + 1) (k=O..x)) diverges everywhere (P. For example.(x) of the Fourier series of f(x). and this result cannot be improved in general. For the proof of these theorems. However. t) have been carried out by S. When E contains no interior point. then among all linear combinations of q. Restating these facts for the algebraic polynomial case. .e. then If(x)-a. 2rc( j + l)/ (2n + l)]. If(x)--o. the saturation phenomenon of approximation often appears.norm) by trigonometric polynomials is given by the partial sum s.1243 336 F Polynomial Approximation dent off: The best possible coefficient clr has been determined by J. and t(x)=(l/n)(m+(lxl/n)).(x)l=O(n-‘) if and only if the tconjugate function J(x)ELip 1.(x)1 =o(n-‘) if and only if f(x) is constant (M. Approximation by Fourier Expansions If {q. Moreover.=i U. x) converges uniformly to f(x) (D. . ci. xln and arbitrary numbers c.. observe the arithmetic means of s. 61. .(t) is a step function that has the value 2nj/(2n + 1) in [2nj/(2n + l). estimation of the magnitude of the derivative of the polynomial of degree n plays an essential role. .(f(‘). the least square approximation (or best approximation with respect to the L. Zamansky [7]. P.(x)l<A(logn)w. For example. Erdiis [lo]. Stechkin obtained the following results: fcLipcr(O<a<l)(84ContinuousFunctions A).. b) and f is any function in &(a.(x) of the Fourier series of f(x). V. l] for which the tLagrange interpolation polynomial and its arithmetic mean are both divergent everywhere if we take as nodes the roots of the Chebyshev polynomial of degree n. b). N.. (1~ p < co). where uk =ji f(x)cp. In approximation with a linear combination of cpc(x).(x) (i.. [ 123).e.P”(X)IG WW)‘~.(x) of the Fourier series of a continuous function f(x) converges almost everywhere to f(x). The Case of a Complex Domain If a given function f(z) is holomorphic in a bounded tsimply connected domain E in the complex plane and continuous in E. the polynomial approxima- . q.n-‘).(x) of degree n and the Markov inequality Since the trigonometric system is a Chebyshev system. Zygmund [4]. S. given 2n + 1 distinct points x. t) in terms of If(x) .(x)/ =O(nP). Trigonometric [S. Jackson Cl]). then U. We also have theorems evaluating w. 1T.. . Runge.. cZn. then the interpolating trigonometric polynomial is given by sin((n+ M. is a constant not depending on J x. Walsh and M.(x)1 for any trigonometric polynomial T. 2n).*(f) and ~~(f.(x)} is an torthonormal system of functions in &(a. xi. the one for which the integral attains its minimum) is the Fourier polynomial ~{=oakpk(x). . Watari [S]). there is always a unique trigonometric polynomial of degree n with prescribed values ck at the points xk. If(x)-a.(f. .l. . s. x) resembles the partial sum s. L. Bernshtein [3] and A.. Keldysh (e. there exists a continuous function for which (l/n)(C. . Given any continuous function f(x) with period 27r. If f(x) is continuous and of tbounded variation.xj)/2) f(t) sin((n + 1/2)(x sin((x. Further investigations on the relation between E. If where q. the trigonometric polynomial that coincides with f(x) at the points xk is called the trigonometric interpolation polynomial with nodes at xk. F. . This theorem was first studied by C. Favard [2]. G. .g. but in the case of uniform approximation we have If(x)-s. (f”‘i +4X where M.t)/2) t)ldrp (t) 3.(x) of 0 ” ’ D.(f.. Marcinkiewicz [9])...

.. we can show that there exists a polynomial T. S. In this case..f(z) is holomorphic in Iz( <p (p> l). If the method of evaluating the degree of approximation using the absolute value I. .(z-z. Assuming that S(z) satisfies certain additional conditions.(z) of degree n (n = 1.(z) coincides with the sum of the first n terms of iNewton’s interpolation formula. zy). then n.n.(z) converges to f(z) uniformly on D.)-f(z.f’(z) . If D is a closed . tion of a continuous function defined in E was given by M. On the other hand. When zp’ is independent of the choice of n.(z) with the (n+ 1)st roots of 1 as nodes converges to j’(z) uniformly in IzI < 1. Generally. E. there exists the largest number p with the following property: .l 1 I =:IT.).. and p(D) coincides with the tcapacity and ttranstinite diameter of D [ 151. be a given set of real or complex numbers.(z-z?$).(z) does not always converge to j(z). For real variables also.(Z)} is +overconvergent).f(z) n.). . P. which is the ordinary (real) Chebyshev nomial. Mergelyan obtained the following theorem [ 131: A necessary and sufhcient condition for an arbitrary function continuous on a compact set E and holomorphic inside E to be approximated on E uniformly by polynomials is that the set E does not divide the complex plane.(z) on D for R > p (Bernshtein and Walsh [12]).f‘(z) is holomorphic on D..1)st roots of I as zp). Unifying these two results.. If(z) . If R < p.(x)=cos(narccosx)/2”-‘.(z)] is replaced by methods using a +curvilinear integral or tsurface integral. where M is a constant independent of II and z. P. y) with a pole at infinity.). Suc- exists. Lagrange’s Interpolation Formula For each n (n = 0. and so on.z”~‘+. The polynomial n. In particular.(z) = z” and P.f’(z1)(z2=z.))l(z. . Lavrent’ev.(z) can be calculated by +finite differences... .rc. If D is simply connected and j’(z) is single-valued and holomorphic on D.. H.(z) of degree n such that max.f(z) is single-valued and holomorphic at every interior point of D. when D = [ -1. Furthermore.. we still obtain similar results.) (z. ).. z!$:!. there exist no such polynomials P.(z) is unique and is called the best approximation polynomial (in the sense of Chebyshev). and f(z) a continuous function on D.. then P.z)If‘(z)n.(z) of degree n such that min ~~~~z”+a. then P.336 G Polynomial 1244 Approximation cessive coefficients of the polynomtal P...)+.(z). However.[17].f(z) is single-valued and holomorphic on D. ) such that I. Let D.)+a. n + 1).. +a.+a..(z)] < M/R” for some R > 1 (actually {Z.2. in this case there exist a number M that does not depend on n and a number R > 1 such that If(z) .)(z-z. and let f(z) be an arbitrary function. Sewell [ 141 proved the existence of a constant r such that I. Moreover. if we take f(z) = l/z and the (n $. This is called Lagrange’s interpolation polynomial and is given by T. On the degree of approximation of polynomials to f‘(z) on a simply connected domain.P. as explained below. If . I..(z)]. there are examples of divergent P.(z) is called Newton’s interpolation polynomial and is given by P.-~.(z) is called a Chebyshev polynomial of degree n with respect to the domain D. When .). Chebyshev Approximation Let D be a bounded closed subset of the complex plane. Then there exists a polynomial n. then there exists a polynomial rr. there exist polynomials P.(z)=a. =.(z)lPldzl (p>O). 11.( z) of degree n that minimizes the integral ~eu(. Then there is a unique polynomial of degree n that coincides with f(z) at each point zp) (k = 1. we have T. Moreover. Similar statements are valid for functions of a real variable. where u(z) is a given positive continuous function on C.(z-z.f‘(z) .. a.n..(z)1 $ M/R”.. y) = log R > 0.(z-z. if .#z. N. The sequence P. 1. For example. Let D be a closed domain in the complex plane with a boundary C that is a rectifiable Jordan curve.(z)( attains the infimum E. For new results and applications of Chebyshev polynomials . be the locus G(x.(z-z. G.(z)1 < M/R” for ZE D.. let z(i”).1. W.f(z) n.(z)1 < M/n’R”. Convergence of Newton’s interpolation polynomial is closely connected to convergence of +Dirichlet series. the limit poly- w(z)=(z-zyy.(z) converges to j(z) only at the point 1. there are the following results: Let D be a closed bounded set whose complement K is connected and regular in the sense that K possesses a iGreen’s function C(x. where ao=S(zl).+a. al=(~(z. by approximating f(z) = z” by polynomials of degree n . A.(j).

p. we say that {p..(z). [3] S. Approximation on a Curve by Orthogonal Polynomials Let C be a rectifiable Jordan curve in the complex plane.~~ . The first step in this process is the orthogonal development of f(x) by Chebyshev polynomials {T. Publ.. etc. z. for example by its tcontinued fraction expansion.(x. N). . If JCPk(4~ldzI = Sk.(z) = a.1245 336 Ref. 66 (1949). B] by a (polynomial) function q(x) can be evaluated by means of the least square approximation. Amer.) = f Mi (i = 1. Math. 11 (1959) 480-488. and s. = ak . N.(x)l of such an approximation. On the determination of the class of saturation in the theory of approximation of functions II.f(xi)T(ui) (k= 1. Sur la divergence des polynomes d’interpolation. This Chebysbev interpolation is actually given by a. Numerical Approximation of Functions The accuracy of the approximation of a given function f(z) by the partial sums of its tTaylor expansion C a. Sci. where N = n+ 1 and the ui=(xi-(A + 8)/2)/((8-A)/2) (i=l. Math. Given a compact domain D and a holomorphic function on D. Unger. 2. .(x. p. and so on. Since in this case 1 ak=2aR2k+’ c f(z)zkldzl =L fo. The problem of existence and determination of such polynomials for any given domain was proposed and first solved by G. Polynomial Approximation Jordan domain and if f(z) is single-valued and holomorphic in D. is often useful. the tTaylor expansion of f(z) coincides with the orthogonal expansion of f(z).@(xi) = + M. = N ml Ck. Sup. Szeged. .(z) such that the orthogonal expansion of f(z) with respect to p.(x)l is estimated by a constant multiple of T. the best approximation with respect to the uniform norm.n).(x) of f(x) (. 1956. [2] J.z. 8 (1937) 131-135. (u)l. The theory of approximation. and set f(xi) 9.~“.. Watari. If(z) .. if there exist orthogonal polynomials p. } with respect to the curvilinear integral on C = 3D or the surface integral on D. s References [1] D. 1947.. Ecole Norm.p. . Cambridge. Moreover. Duke Math. I. A computer can perform the division very quickly. 19-93...rc. Given a holomorphic function f(z) on D.(z). T. In particular. Roughly speaking. becomes sufficiently small.) [6] A. Sci.(z) is given by the orthogonalization of the system { 1.rp. such as tBessel’s inequality. Zamansky.Section B). Sur l’ordre de meilleure approximation des fonctions continues par des polynomes de degre don& Mtm. [9] J. and Walsh.. Jackson.4 (1912).) = f (M . Marcinkiewicz. we must determine coefficients ck that satisfy the conditions of Chebyshev’s theorem.cp. 243-265.ak and M. Zygmund. N) are the roots of T. Repeat this process until Aa. . + a. 47-75. jisan orthogonal system. . Carleman.a a. Sunouchi and C. 61 (1937).2. Smooth functions. and let pk(z)~tL2(C). If we denote the nth partial sum of this series by s. [S] N.(z)} belongs to the domain D.. u=(x-(A+B)/2)/((BA)/2). Sot.then{l. TBhoku Math.dz.cp.(z) is the least square approximation by a linear combination of pa(z). J... Achieser.(x) = C~+c. Bernshtein (Bernstein). Sur les meilleurs procedes d’approximation des certains classes des fonctions par des polynomes trigonometrique. then s.(u). are all valid here as in the theory of general torthogonal systems. z’. Math. Trigonometric series..(z). sir.(x)l < K 1T. Let M be the extremum of the error If(x) .. Szego. Theory of approximation. Favard.. 12 (1945)..(x)} : %(x)=%oak&(u).. if wetakelzl=RasC.rp. 3. where u(z) is a given positive continuous function on D. the tRiesz-Fischer theorem.(x -x0)” decreases rapidly as the distance Ix-x01 increases. and the rational approximation of a function.(x) = C ak K(u) satisfying f(xi) .+. z + . we set ak = JcfWd4ld z I an d consider the formal series Ckm. Colloq.Mi) with respect to Aa. Ann. . 27ci s =zk+’ ical calculation of functions. 2099224. The error If(x) .f(xi). Belgique. [4] A. Generalizations were given by G. Acad. Then solve the linear equation @“(xi) cp. This and other results. J. To get the best approximating polynomial p(x) = P. J. Roy. [7] M. [8] G. l-104. (Original in Russian.. .. I.(z) converges to f(z) uniformly on D.(z)l < M/R” for some R > 1 on D. 1930. then {PAZ)} is called an orthonormal system on C.. 1959.(z)lPdS. + a. The second method is best suited to numer- . ser.. The accuracy of the approximation of f(x) defined on a compact interval [A. Acta Sci..(u): If(x) . . . then there exists a polynomial n. Faber. Consider a function cP. Zygmund. ak = 2N-’ CE. Bull.(z) of degree n that minimizes the integral rSDu(z)lf(z)-n. Class de saturation de certains proctdts d’approximation des series de Fourier des fonctions continues et applications a quelques probltmes d’approximation.

and therefore R[X] is an integral clomain. The approximation of functions. X. Ginn... This expression is called a polynomial in X and Y over R. X. Math.. terminates) F(X.] is called the polynomial ring in m variables (on m inde- . If c(. B. .][X.. I. . X.X (1) A polynomial f(X)eR [X] can be regarded as a function of a commutative ring R’ containing R into itself such that c-f(c). Math. 1942. Rice.. .. For given J gcR [X] (degg > l).. Mergelyan (Mergeljan). If R is an tintegral domain. . X..) denote the element of R (or S) obtained by substitution of x1.. x2.. An element of R[X. [ 171 S. Griinwald..a. Y] can then be expressed as Car” X’Y”. . Ann. we define the degree of 0 to be indefinite. respectively. over R.~~l’~‘X~~ (2) (C denotes a finite sum for nonnegative integral vi beginning with vr = v2 = . (Original in Russian.. 3 (1962) 287-293.. is called a monomial. . .. such asaX. the number n is called the degree of the polynomial f(X) and is denoted by degf: If a... .J = 0.. .. . Walsh.. xm for X. then the latter inequality is an equality. R[X.+u. Math. Sot. 1969. 1964. 1246 . . Analytic function theory II. the ring obtained by adjoining Y to R [Xl.). c(. J.) (or of F(X..f(X) of the form f(X)=a. R. [ 121 J...a. . Interscience.]=R[X. .X. the degree of this term. Colloq. X. II. f(X)+g(X) and f(X)g(X) are the functions such that c++f(c)+g(c) and L Hf(c)g(c). X. 1962. x1. 11-17. An expression .X+.a of degree 0 is called the constant term of F. Erdos. Degree of approximation by polynomials in the complex domain.~. 1966. the polynomial (1) is called a manic polynomial. Sewell. . Press. is called a polynomial in a-variable X over R.= 0) is called a polynomial in m variables X. X.)=~a.f divided by g. In this sense... .. Amer.369 Rings of Polynomials).4) Polynomials A.mX. studies 9.a.. c(~. 198 1. in F(X. [ 131 S. Polynomial Rings in R[X] are Addition and multiplication defined by Let R be a commutative +ring and a 0. This q is called the integral quotient off by g. . [ 161 J. Interpolation and approximation by rational functions in the complex domain. then (xi.. X. Math.. X... Some theorems and remarks on interpolations. X. math. For these inequalities and for convenience elsewhere.. Generally. [ 151 E. [l l] G.. N. . braic equation F(X. elements of R. a polynomial consisting of a single term. #O. The totality of polynomials in X over R forms a commutative ring with respect to ordinary addition and multiplication (whose definition will be given later). . over R. Y] denote the ring R[X][Y]. . Hille. . and let F(sc. . In the case of one variable.cQ. Tchebysheff systems with applications in analysis and statistics.. Szeged.. .Xz X.Ym + . and its element . if a. be elements of R (or a commutative ring S containing R).l. .. The term ~.~. called a zero point (in S) of the polynomial X.degg(X)}.~. . The same fact remains true in the general R[X] if g(X) is manic. J. I.. The greatest degree of terms is called the degree of the polynomial 8’. Acta Sci.. It holds that deg(f(X) + g(X)) < max{degf(X). Sneddon..Y2. E. Ann. It is called the ring of polynomials (or the polynomial ring) of X over R and is denoted by R[X]. degf(XMX)~ degf(X)+degg(X). 12 (1950). Karlin and W.337 A Polynomials [lo] P. We call each summand a term of the the coefficient and vr + v2 Pob-omiaL aY. Addison-Wesley. Polynomials in Several Variables Let R[X.)=O). 337 (111. + v. it is called a root of F(X.. Publ. ..) = 0).. (2) 37 (1936) 908-918. Amer. . On the representation of functions by series of polynomials on closed sets. and r is called the remainder of ..) [14] W. If a polynomial F in X. Assume that R is a field.. Now let c(~. Uber Divergenzerscheinungen der Lagrangeschen Interpolationspolynome der stetigen Functionen.. X.. a. (. fifth edition. “‘> a. = v. rER [X] such thatf=gq+randdegr<deggorr=O(division algorithm). .. L. ser.+u. is composed of terms of the same degree n. we can find unique 4. 1935.) (or a solution of the algeF(X. .lX~.cc. Polynomials in One Variable C. = 1. . Princeton Univ. It is also called a polynomial in x1. a 1. We say that we adjoin X to R to obtain R[X]....) is F(a. Sot. Transl. then F is called a bomogeneous polynomial (or form) of degree n. 1951. ...X.. namely.

Accordingly. W. Symmetric Polynomials Let f(X. . If k is a tunique factorization domain. c(. ... c(. we define the (formal) derivative off with respect to Xi as L(X... These two theorems have been generalized in many ways and given precise formulations.. to find the greatest common divisor (f..a(aE k).. A polynomial f(X)ck[X] of degree n is said to be reducible over k if f is divisible by a polynomial of degree v < n in k[X] (v # 0).a. The map f H cTf~aXi is called the (formal) derivative with respect to Xi. Dorge.... then af/aX is denoted by df/dX.... ablesx. so that after a finite number of steps we attain ry-l = rvq. . J. If f(X) is a polynomial (1) in I [X] such that for a prime element p in I.. X. Inaba). apply the division algorithm repeatedly to obtain f=9q1+r1. In particular.-.) by assigning appropriate inktox.-. Xyi-’ ... If df/dX = 0 for an irreducible polynomial f(X) in k[X]. X. .... Since k[X] and hencek[X... G.. Also.. then f(X) is irreducible over the field of quotients of Z (Eisenstein’s theorem)... . it is said to be irreducible over k. D. Then using the division algorithm.+l(rv+l = 0).tl is equal to f(cc)..) (. This is called the remainder theorem.... a. Irreducible Polynomials Let k be a field. E. q(X)Ek[X].. . we can obtain an irreducible polynomial in X i.. . Derivatives Given a polynomial degg>degr.. X. . then f(X) is said to be inseparable. Ifkisafield. . g E k[X]. a. . if m = 1. 110 (1892)).. In Hilbert’s irreducibility theorem. If f(N) = 0. otherwise. etc.. an irreducible polynomial f(X) is inseparable if and only if we can write f(X) = g(XP)....... . X. .vm Xi’. f(X)E k[X]. rek. then f(X) is divisible by X-cc in k[X]. = .. _.. of k in the above expression is called a rational expression m tli.. we can define the concepts concerning divisibility (such as a divisor.. f(cr) = r. Factorization Let k be an integral domain. The Remainder Theorem Let k be an integral domain..>.... A polynomial f is a tprime element of k[X] if and only if f is irreducible over k. Any polynomial of degree 1 is irreducible. . The usual rules of tderivatives also hold for the formal derivative. . F. . by any infinite field that is Vinitely generated over its tprime field (K.. . we get f(X)=@-cMX)+r. the remainder of f(X) divided by X . over a field k. (provided that d~l. If a polynomial (2) in m variables over an The tfield of quotients of the polynomial hng X.~“)ZO). X. . . . Polynomials and Alternating X. E. H. with elements c(i. an expression f(ccl. 4X.) by another polynomialg(X. that is.X/J... Franz. (Hilbert’s irreducibility theorem.. a... and let g(X)= X . a multiple. a product of primitive polynomials is primitive (Gauss’s theorem).. X.] are integral domains.] over a field k is denoted by 4X. When k is of characteristic p # 0. Every polynomial over k can be uniquely expressed as a product of some primitive polynomials and an element of k..67 Commutative Rings). ..v. . k[X] is a tprincipal ideal domain. Furthermore.. then k[X] and k[X. .) and is called the field of rational expressions (or field of rational functions) in variables Xi. rl=r2q3+r3.. Reine Angew.. . f 0 (modp). . Rational Expressions Therefore. the algebraic number field may be replaced. XP (0 <p <m) from the polynomial values F(X1.1247 337 I Polynomials into Primes talgebraic number field k is irreducible. .>degr. . then every irreducible polynomial f(X) ( # 0) is separable.. .. =a... for example.X. = (L g). g) off. otherwise. f(X) is separable. . I. If ... A polynomial over k is said to be primitive if the greatest common divisor of all the coeflicients is equal to 1...] are unique factorization domains... X.) = ~Viav. X.... over an integral in varidomain I. . ..67 Commutative Rings L). g=rlq2+r2.. This algorithm is applied to Euclid rings (. . It can be written as a quotient of one polynomial f (Xi. that is.) # 0. we can use the Euclidean algorithm. X.. Math. Xvm and denote it by af/aX.) obtained by replacingx. X... If the tcharacteristic of the field k is 0.)/s(ccl. Its element is called a rational expression in X1. Then r. Let I be a unique factorization domain.) be a polynomial X. over k. =a. . then so are k[X] and k[X.$0 (modp’)..... .-O(modp) but a..

X. 1959. 1963.-X. an expression ... .). . .. ...f(X. cr.) is invariant under every permutation of Xi. 8.. we have a:D = a~a~+18a. .28) Potential Theory A. PQ2-“&dQX n>3. a. Interscience... .~~-.+(-l)“~~~“~-lS. Because of this close relation between potentials and harmonic functions. we have the relations S. are the roots of an ialgebraic equation a.)=(X. the measure p is taken to be a nonnegative +Radon measure with compact tsupport. sometimes potlential .gn are equal. Thus the totality of symmetric polynomials of X. Concerning the elementary symmetric polynomials and the s. .X” + a. .. (Original in Russian. 338 (X. Thenwehavec..) be denoted by (. ..-.. Hermann.... Godement. . = 0 of degree n. If r*i. [S] S. ... n = 2. of the equation... X. xn if f(X.. is identical with the ring I[a. X.a.. ...) c3u/i?x.) gives a field of force in the n-dimensional (n > 2) Euclidean space R” Given a point P in R” and a measure p. .l)krr. respectively.(X -X.=a~-4a:~. They are called the logarithmic potential and Newtonian potential.. .Y.-4a.. X. cc. we have .+. these are symmetric polynomials of X.. G. .)i~asymmetric polynomial of Xi.n+2 ). . X. ch. for every element cp of the polynomial ring 1248 gives a criterion for the condition tl at some of xi.). [3] A. . second edition. Then the polynomial p is invariant under even permutations of Xi. X..S.X.-i) be the product of n(n.a.-Xx. .+X. Kurosh... can be uniquely expressed as a polynomial ‘p(n. Newtonian Potential In dynamics.. are called elementary symmetric polynomials (or elementary symmetric functions).2r~~. . we have a.) (X. .+X..= CXiXj=X...-ais. X. L.).-~. . some authors mean by Newtonian potential the function u(P)=JPQmldp(Q) in R3... X...X. a potential means a function u of n variables xi. it is called an alternating polynomial (or alternating function). Also.X.. any harmonic function defined on a domain in R” can be expressed as the sum of a potential of a single layer and a potential of a double layer (defined in the next paragraph). then D(! i . .. 1962.X. Because of its particular expression.30. gz + 3a. a. Obviously.(~.ands. X. z..X3+.-4rr.2.+ (-l)“va. 1970. an of I is called a symmetric (alternating) function of s1i.=O(v=1.f(X. Actualites Sci.a. 02. .) by every +odd permutation of X. Discriminants The square D(X. (X.. X. . u1 .f(ai.2. with elements x. . X... .) of the simplest alternating polynomial p is a symmetric polynomial. For instance. an alternating polynomial f is divisible by the simplest alternating polynomial p. Moreover.. Springer. Diophantine geometry. -X. If the characteristic of I is different from 2. Let p(X .l....) on= X.r.... .. H owever.--27u~u~.X. .s.. Elements de mathematique.=O(~=n+l. X. s2 = 0: . it is called a symmetric polynomial (or symmetric function) of X. o..f=ps. Let the coefficient of X”-k in the expansion of (X ..]. such that -grad u = -(fh/dx.a2a.. . .). [2] N. . Cours d’algebre..)= 0 are typical examples of potential fu actions.337 J Polynomials ... x2. Usually.a. 1966.) is called the discriminant of the equation. . . Ind.u~-4u~u. For Hilbert’s irreducibility theorem. If f(X... = g. mz.) is symmetric (alternating).. X.. .. ..) [4] R.s.. ch. where s stands for a symmetric polynomial.. = C Xi’ (v = 1.X.X. . Hence p is an alternating polynomial of Xi. if n = 2. It is called the simplest alternating polynomial of these variables. p is also called the difference product of Xi.) = p2(X. . the functions u(P) given by the integrals u(P) = su(P)= slog PQ dn(Q).)(X.+~~S.)(X. . 4. ...-. a. Algebre. . and p becomes -p under odd permutations. 1102b.. For example. it can be written as ... if n = 3.+X.. X”-’ + + a. +.a.) by replacing Xi. 1955. Hermann. Lang.=CXi=X. Conversely. 0. for s.) obtained from f(X.) (Newton’s formulas).a.. Algebra I.-Xx. u. and gi..). These potentials are tsuperharmonic in R” and harmonic outside the support of n.... and it is therefore a polynomial in 0i. Bourbaki. Chelsea..cp(o. . . J. .. x. van der Waerden. . . ..X. s3 = 0: .+2~:+4a. X. It can be expressed in terms of coefftcients a. Conversely.D = a: -4a.) is transformed into -. X. Lectures on general algebra.+(--l)“cT&.1)/2 differences between Xi. every symmetric polynomial of X... . X..~. X. ICr.X. . References [1] B. . . D(cc. . seventh edition. This is called the fundamental theorem on symmetric polynomials.

the relations shown in Fig. PL)for any P with S.. The integral u(p)= p&&(Q) s s Q is called the potential of a double layer (or double distribution). B. and some restrictions are called for.193 Harmonic Functions and Subharmonic Functions S. then the derivative at P in the direction of any fixed tangent line at PO has a finite limit as P tends to PO along the normal line. is the support of p. (2) Ugaheri’s maximum principle (dilated maximum principle): There is a constant c > 0 such that SwpenW... (6) Continuity principle: If @(P. v. when R = R”. Ninomiya studied (2) and (6). (5) For any compact set Kc rZ and any p with S. where (a)+(b) means that (a) implies (b). If the support of p is contained in a surface S and dp = p da with density p and surface element da. Ugaheri. (p. and lirnsup@(P.1249 338 c Potential Theory theory means the study of harmonic functions (. Q) = @(Q. G. and N. The second condition is valid. @(P.). (3) A variation of Ugaheri’s maximum principle: Given any compact set K c a. then it is also continuous in R... p) or @p(P). when n = R”. there exists a constant c which may depend on K such that suhK @VT P) G c whspW. (3) as P approaches P. p) satisfies the following principles: (1) Frostman’s maximum principle (first maximum principle): suppen@(P.) + u(P. &W) s exists for measures p. and (c)+(d) means the negation of (c)+(d).) at P. Then the Newtonian potential u of /. p) is bounded from above on S. and i are nonnegative Radon measures with compact support in R. When = W’. Q) a real-valued function on the product space R x n. The function 6(P./*) as P tends to the point at infinity.(p~ 14 for any P. If p satisfies the tHiilder condition at PO ES.f satisfies Poisson’s equation Au= -4np. @(P... Therefore. Q)= (p(PQ) satisfies (2) in R”.~@(P. @ is a tlower semicontinuous function on n x R satisfying -co < @ d co. Choquet. where S.~. c K. p)+sup. 1 & * (5) ?H (6) (2) The classical notion of potentials is generalized as follows: Let R be a space supplied with a measure p ( > 0) and @(P.-. T. then (1) holds also. the kernel @(P.. then each partial derivative of u has a finite limit as P tends to a point of S. If p is continuous on S. (For the representation by potentials of superharmonic functions . The definition of If the continuity principle holds for a general kernel @(p. p is of class C2 on S. Q) = PQ-” (0 < tl < n) is called a potential of order CL (sometimes of order n-m) or a Riesz potential. P) is called the adjoint kernel of @. p).. the value is called the mutual energy of p and v. Ohtsuka established . We assume that fi is a tlocally compact Hausdorff space. along the normal line. c K. then it is bounded on n also.. p) is bounded above on S. p) for any pL. Generalized Potential (1)s Fig. P) G c supp+.. If p is continuous on S. If. and the tdirectional derivative of u at a point P in the direction of the normal line to S at PO tends to -2Tcp(P. When the integral j@(P. the potential with the kernel @(P. then it is bounded on K also. Then @(P. if @(P. p) of a potential and if for any p there is a sequence {Pk} of points in fi .)+ .~)<s~p~. and along which @(Pk.z= s s 3% PQ I p=p. and let @(P. then the limits at PO of u from the two directions along the normal line at PO exist and are 27-cp(P. In particular. it is called the potential of p with kernel @ and is denoted by @(P. (4) Upper boundedness principle: If @(P.. p) < su~.@(Q. Q)dp(Q) is well defined at each point PER.193 Harmonic Functions and Subharmonic Functions). p) is continuous as a function on S.) Suppose that the measure p of R3 satisfies dp = pdz with sufficiently smooth density p and volume element dz. then u is continuous in the whole space. p) is called the energy of p. then the potential u of p is called the potential of a singli layer (or simple distribution). In particular. C.) and -2np(P. as P moves on the line.. co) and satisfying ~(0) = co. for instance. that has an accumulation point on S. Ugaheri showed that for any nonnegative decreasing function q(r) defined in [0. Generally. The Maximum Principle Principle and the Continuity da(Q) Let s2 = R”. and p. p) is tsubharmonic in R”-S..) + u(P. p) be the kernel of the Newtonian potential.S. M.. 1 hold among principles (l)-(6). v 2 0.@. potential given above may be too general. the directional derivative jumps by -47cp(P. further..

)). Cartan [4] for the Newtonian kernel.v)>Ofor any p. Next.. Aronszajn and K. although their meaning is not t:le same as here..e.v. Some characterizations for a kernel to be of positive type or to satisfy the energy principle were given by Ninomiya. c K and p(K) = 1. which was studied by N.P)+(v. Smith.~)-(3~.) in R in this inequality [18]. For instance.PO} with respect to the original topology of 0 or there exists a measure p such that lim inf@(P. Denote by L the class of measures i. then liminf@(P. Moreover. The class of unit distributions can be topologized by the vague topology. are contained in a fixed compact set and p. where S. and W(M) to be w for the empty set @.. means that the fine topology is stronger than the vague We are concerned with determining when a family of potentials corresponding to a class of measures {p.v)l (GEE). p . I”).e. Choquet and Ohtsuka generalized these de& nitions and results [ 1S]. First. p .v. for instance. If the equality (p .p-v). vaguely to pLo.. is continuous in the wider sense and both @ and 6 satisfy the continuity principle. are perfect.. If PO is an isolated point of X U {PO} with respect to the topology weakest among those stronger than both the original topology and the fine topology. The notion of thinness was .e. (For results on the convergence of sequences of potentials .e. w. then equality holds quasi-everywhere (q.v)l (1cL). we say that the property holds quasi-everywhere (q.p) as PEX-{PO} tends to P.. and strong convergences are equivalent for any sequence {p.. set U:(X) = W(K) for an arbitrary set X in R and inf. are also used in the theory of +capacity. (For literature on (l)-(6) and other related principles .(X) = co (resp.) in R. Thin Sets A set X c R is called thin at PO when either PO is an isolated point of the set X U {. Further.. the values of the measures of M are bounded. continuous.) D.’ by the seminorms ~--~-[(1. Q. for a nonempty compact set K in sl. When a property holds except on a set X such that W.v) = 0 always implies LL= v..Section L) and a certain additional condition is of positive type..) G. moreover. then the kernel is said to satisfy the energy principle.} with indices w in a directed set converges. The examples in Sections F. This notion is used to give conditions for E to be tcomplete with respect to the strong topology [ 1 I.v)-2(1(. The strong topology is defined on E by the seminorm J(p-v.&>@(P. @ may have co as its value) and @ is finite outside the diagonal set of fi x R. the weak topology is defined on E by the seminorms p--v-l&p)-(1. Topologies on Classes of Measures Let C. be the space of continuous functions with compact support in n and MO’ be the class of measures in n. nearly everywhere (n.nf(p. A symmetric kernel is called positive definite (or of positive type) if (p . For the Newtonian kernel. define W(K) to be . This topology was introduced by H. converges. and define the fine topology on M.[ 181.. Using them.338 D Potential 1250 Theory topology. If all S. A subclass M of Mi is relatively compact with respect to the vague topology if on every compact set in R. F. I.e.v) = (~.} of measures with bounded energies. The terms q. He proved that the fine. p) > @(P.e. 3L~L. where vague<fine. Convergence of Sequences of Potentials E. where G is an open set in a.) (resp. The tseminorms pv+lJfd/r-Jfdvl (feC. It is the weakest topology that makes each @(P. and studied the cases where tconvolution kernels on a locally compact topological group are consistent or perfect [ 111. when the kernel is of positive type. p) is finite for all ~LEM$. This topology coincides with the original topology in R. Fuglede called a consist’ant kernel satisfying the energy principle perfect. The converse is true in a special case [S].) define the vague topology on A4:.e. We note that (6) does not necessarily hold in general. he showed that a symmetric kernel that satisfies Frostman’s maximum principle or the domination principle (..p). B. 181..[ 181. The Energy Principle Denote by E the class of all measures of finite energy.(X) = supxi G K(G). and n. Cartan [S] showed that vague<fine<weak<strong on E. If. PQ -a (0 <a <n) in R” and the Bessel kernel. WI: now define the notion q. T. weak. and J show that the potentials with kernels satisfying a weak condition such as (6) possess a number of important properties. Fuglede [ 161 called a kernel of positive type consistent when any +Cauchy net with respect to the strong topology that converges vaguely to a measure converges strongly to the same measure. such that (I. q(X) = co).p. that (6)-t(S) and (S)+(6) in general and proved (S)+(6) in the special case where @ is continuous on R x 0 in the wider sense (i.. then X is thin with respect to the adjoint kernel &. which induces a topology on fi itself. vcE.

Assume that R is a metric space with distance p. X = {P 1@(P.1251 338 L Potential Theory Problem introduced in 1940 by M. This criterion was obtained by N. it is quasicontinuous in R.G is continuous. Wiener in 1924 and utilized to give a condition for a boundary point PO of a domain D in R3 to be tregular with respect to the tDirichlet problem for D. K. a such that q(r) <acp(( 1 + 6)r) in 0 < r < r.. If a > 0. Brelot (1941) called a set A polar when there is a measure p for which @(P. on K and @(P. This result is called Evans’s theorem (or the Evans-Selherg theorem) (48 Capacity) in the special case where A is compact. Let q(r) be a positive decreasing function. X) by ri(n. Consider ri(D. p) in R. Then a necessary and sufficient condition for X to be thin at P.Section L). Kishi).e. G) as the capacity of G. Equilibrium Mass Distributions A unit measure p supported by a compact set K is called an equilibrium mass distribution on K if @(P. Principle A kernel is said to satisfy the sweeping-out principle (or balayage principle) if for any compact set K and measure p there exists a measure v supported by K such that @(P. the balayage principle implies . For any i general kernel. p) = co} is a G. and f is finite upper semicontinuous on K.. When every compact subset of X is thin at P. p) < a in R. and the minimizing problem is equivalent to finding the projection of v to {p E E 1S. we find that every potential is quasicontinuous in R (M. X is called internally thin at PO. the problem is called conditional.. Suppose that whenever the potential of a measure p with kernel @ is continuous as a function on S. In this situation. J.(o. Q)) as the kernel @(P. Q). v) < @(P. is regular if and only if the complement of D is thin at PO. P. there is a measure p such that the set of points where the Newtonian potential of p is equal to co coincides with A and p(R”-.infKcG where G is an open set. When @ is symmetric and of positive type. v) (v E E) is equal to IIp-vII’llvll*. This terminology originated in the classical process for the Newtonian potential in which the exterior of K is covered by a countable number of balls and the masses inside the balls are repeatedly swept out onto the spherical surfaces. P. v) = Q(P. Frostman’s maximum principle is equivalent to the equilibrium principle. we say that we sweep out p to K. p) n. The Gauss Variational H. Brelot and investigated in detail in [3]. Conversely.e. 0)) as defined in 48 Capacity C. K supports a nonzero measure of finite energy. The kernel is said to satisfy the equilibrium principle if there exists an equilibrium mass distribution on every compact set. The Sweeping-Out A function f in R is called quasicontinuous if there is an open subset G in n of arbitrarily small capacity such that the restriction off to R . on K and cD(P. p) on S. K) ( = V(K. When @ is not symmetric. set for which the value of ri. and take cp(p(P.A) =0 (Choquet [7]). p) n. Among many results obtained for this problem [ 181. where Xj= {PEXI sj. When we find such a v. I.<rp(p(P.. p) . on K and f(P) > @(P. quasicontinuity depends on the definition of capacity. Polar Sets Given a compact set K and a function f on K. the following is typical: If CDis symmetric. Suppose there exist positive numbers r. When an additional condition is imposed on p. this projection is equal to the measure obtained by the balayage of v to K (.. Corresponding to tinner and outer capacities.. Quasicontinuity L. and p/a is called a capacitary mass distribution. is xgl sj/We(Xj)< cc (s> 1) [19].. l/a can be regarded as a kind of capacity... although the method using Gauss variation is not applicable. X) is infinite. K) for an arbitrary set X. the Gauss integral with f(P) = @(P. and define ri. p) = cc on A. When @ is symmetric. then there exists a p such that f(P) <@(P. p) is equal to a constant a n. and finding v is called a sweepingout process (or balayage). Naturally. @ is then said to satisfy the quasicontinuity principle. Assuming in addition that CDis positive symmetric and taking l/U@. swG. the same relations hold for some p if @ is positive and 6 satisfies the continuity principle (Kishi [ 15]). c K}.))<sj+‘}.. A necessary and sufficient condition for X to be internally thin at PO is c S’/14qXj) < co.2 jf& for a measure p such that S. In some cases. Then for any p. 6. inner and outer capacitary mass distributions and their coincidence can be discussed [ 111.e.(n.. the problem of minimizing the Gauss integral (p.. c K is called the Gauss variational problem. A similar result is valid for a nonsymmetric kernel if the continuity principle is assumed (Choquet [6]). given a Gh set A of TNewtonian outer capacity zero in R” (n > 3).

Hunt obtained a relation between this principle and the representation of GTf in the form s: P. a point P is called an internally (externally) irregular point of X if the inner (outer) balayage mass distribution to X of the unit measure cp at P is different from cp. His result is important in the theory of +stochastic processes (.2 d n <n) in R” and the Yukawa potential with kernel aPQm’ exp( --im) in R3 satisfy the complete maximum principle.}~~~ is sub-Markovian. Other Principles A kernel @ is said to satisfy the uniqueness principle if p = v follows from the equality @(P. we now introduce weak topologies in the space C of continuous functions in R and in the class A4. v). G. or. for . The complete maximum principle has been defined and studied for G* (Deny). satisfies the domination principle and 6 satisfies the continuity principle. if @ satisfies the lower envelope principle on every compact set considered as a space.LLLEEand any v implies the same inequality in Q. In contrast to the domination principle. the Dirac measure at the origin.fdt with a tsemigroup P. 181. @(P.dt and cl0 semigroup {4. jL) = min(Q(P. A Hunt kernel satisfies the domination prinlsiple and the balayage principle to all open sets. p)=@(P. X is thin (internally thin) at P if and only if P is an externally (internally) irregular point of X (Cartan [:5]). Similarly. and it satisfies the complete maximum principle if and only if {LY. the domination principle implies Frostman’s maximum principle [ 171. (IC. in Q. and ICYis a singular kernel satisfying the domination principle such that ICY* E. Potentials of order CI (n . 0. where cp is a continuous exponential function. Potentials with Distribution Kernels A function f in R” is called slowly increasing in the sense of Deny if there exists a positive . It has a unique decomposition IC= cp. This principle implies both Frostman’s maximum principle and the domination principle. then the same inequality holds in 0. then @ satisfies the domination principle or the inverse domination principle under some additional conditions (Kishi). A positive linear mapping G of M.. and a > 0 (Cartan and J..p)<O(P. + ICY). the balayage principle.. ICYis equal to 0 ‘or a Hunt kernel satisfying the complete maxi-mum principle. for PEE and an)i I’. In a special case. and w2 [S]. continuous in the wider sense. < 1. It is called a Hunt kernel when there exists a vaguely continuous such that K = s. v) is valid on . equivalently. A diffusion kernel G on a locally compact Abelian group induces a convolution kernel K if G is translation-invariant. @ is said to satisfy the inverse domination principle if the inequality @(P. Ninomiya and Kishi studied this principle. implies the same inequality in 0 for p E E. then @ satisfies the lower envelope principle on every compact set considered as a space. The linear mapping G* of C. For a convolution kernel K satisfying the domination principle. of Radon measures of general sign with compact support.338 M 1252 Theory N. 1950). Then c’ satisfies the balayage principle if and only if G* satisfies the domination principle (Choquet and Deny). Deny. For the equilibrium and domination principles for nonsymmetric kernels . p E M. v)). Relations between this principle and some other principles were studied by Kishi [ 141.. there is a i such that @(P. symmetric.(P. and finite outside the diagonal set. By means of the bilinear form Sf& (f~ C. Corresponding to inner and outer capacitary mass distributions. [12]. p) < @(P. and in the class M of measures of general sign with not necessarily compact support. p). The Fourier transform of such a sernigroup has a closed connection with a negativedefinite function [l].v) on S. which is valid n. A kernel is said to satisfy the complete maximum principle if the inequality @. A. A kernel @ is said to satisfy the lower envelope principle if given p and v. With respect to the Newtonian potential. Conversely.e. While all principles discussed so far are global. we can examine inner and outer balayage mass distributions and inner and outer Gauss variational problems and their coincidences [S. p) d @(P. into M that is continuous with respect to these two weak topologies is called a diffusion kernel. which asserts that if the inequality O(P. we introduce weak topologies in C’. If Q. The converse is true if @ is positive. Diffusion Kernels Potential the domination principle (also called Cartan’s maximum principle). into C that is determined by Ifa’Gp = l G*fdp is called a transposed mapping. = ICYfor every XE&. c:. C’hoquet and Ohtsuka made a local study.).261 Markov Processes). Convolution Kernels M.S. the inequality is valid for all relatively compact open sets w. v) + u on S. any v. Ida.[14]. We can define the balayage principle for G and the domination principle for G* as in the case where kernels are functions. P.20 =E..

J.. Foundations of potential theory. The notion of Dirichlet space was introduced by A. in R” for any TE W. On the theory of potentials in locally compact spaces. Caracterisation du principe de domination pour les noyaux de convolution non-born&. C. Furthermore. 208215. third edition. (For axiomatic potential theory . Sci. Brelot. 23 (1963). Suppose that n is a locally compact Abelian group and D is a Dirichlet space such that U. aUT/axj =fj is defined a. Acad. Acta Math.~~=. [6] G. Sur les ensembles efliles. Acad. 107-183. = 2(n . 165-187. [S] H. Lunds Univ. Springer. then the corresponding capacity is the Lebesgue measure.193 Harmonic Functions and Subharmonic Functions. [2] A. Sci. Sur les fondements de la theorie fine du potentiel. For every TE W the function FK x FT is slowly increasing in the sense of Deny. Maximum principles in the potential theory. Forst. Les potentiels d’tnergie linie.e. 244 (1957).. Kellogg. 103 (1960). Sci. 404. Then we call D special and characterize it in terms of a real-valued continuous function on 0 [2]. and for any compact set K c fi. 5 2 0 a Radon measure in 0. Q. Bull. functions are assumed to be complex-valued. called the K-potential of T. Elements de la theorie classique du potentiel. (2) 68 (1944). Grenoble (N. Ann. Univ.1253 338 Ref. FT.. A Hilbert space D consisting of locally t-integrable functions is called a Dirichlet space if C. Kishi. Acad. In the case of the Newtonian kernel. Kishi. There exists a distribution U = UT that satisfies FU = FK x FT. These results are due to Deny [9]. Ito. Brelot. R. R. 45 (1959). and problems of equilibrium. 160661609.60112. the method of projection is applicable. Paris. Cartan. Sem. Dirichlet Spaces In this section. Mat. [12] M. 3 (1935). let f be a function in R” that is absolutely continuous along almost every line parallel to each coordinate axis and whose partial derivatives are square integrable.ull= Ilull for every UED and ysR. [ 111 B. 1975. If in addition p > 0. respectively. Paris. D. An existence theorem in potential theory. However. A tdistribution K is called a distribution kernel if the tFourier transform FK of K is a function k > 0 and both k and l/k are slowly increasing in the sense of Deny. Deny. Fuglede.. Potential theory on locally compact Abelian groups. References [l] C.. then u is called a pure potential. Potentiel d’equilibre et capacite des ensembles avec quelques applications a la theorie des fonctions. Proc. Choquet. and capacity can be examined. if tDirac’s distribution 6 is taken as a distribution kernel.afi/ax. s(xj-Yj)~(Q)PQ-“dz(Q). 82 (1950). 1929..2)7c”‘*/I(n/2). the lower envelope principle. cp)= j cpdp holds for every .and IITll*=~. where l/c. the equilibrium principle. and C. Beurling. Suites de potentiels. For any pure potential.compact support is tdense in W. UT=(n-2) where xi. Given such a distribution K. Then f is equal to the potential of some TE W with Newtonian kernel up to an additive constant. [lS] M. Frostman.. T= -c. the family of Newtonian potentials of measures with finite energy is not a Hilbert space (Cartan). the relations Iv(P)-u(Q)l<lu(P)-u(Q)1 and Iu(P)l<lu(P)l for uoD and a function u always imply u~D and (IuI( 6 l/uII. I? D is dense in both C.. Nagoya Math. [7] G. Noyaux de convolution et balayage sur tout ouvert. balayage. Nat. [ 131 0. Springer. Every ordinary potential of a double layer is a special case of UT. and the complete maximum principle hold [Z]. These A are square integrable. Dirichlet spaces. 1974.u(x) = u(x Y)ED and IIU. [lo] 0. Lecture notes in math. Beurling and J. 139-215. there exists a constant A(K) such that jKIu(dc<A(K)IIuII for every u E D. 12-36. The family of functions of class C” with. Math. j=l k Q. Deny. n D. Conversely. Choquet and J. Potentiels sur un ensemble de capacite nulle.S. Acta Math. Theorie g&ale du balayage en potentiel newtonien. T2) = j kFT. Centre de Documentation Universitaire. 244 (1957). 57 (1975).. Sci.) integer 4 such that jf(P)(l + OP2)-4dr(P) < co. 221-280. Since W is complete. Medd. A function u E D is called a potential if there exists a Radon measure p such that (u. C. Nagoya Math. Choquet. Let 0 be a locally compact Hausdorff space. 27 (1966). J. the space of continuous functions with compact support. yj are components of P. 1965. dr. [14] M. and D. the balayage principle. [4] M.). [3] M.C3=~Slfil~dz. For instance. Springer. 167197. PotentiaI Theory cpE C. the family W of distributions T for which FT are functions and 1)TII * = Jk(FT)‘dr < co (this is called the energy of T) is a tHilbert space with inner product (‘J’i . [9] J. J. US. Nagoya Math. [8] G. Deny.. Berg and G. 1707-1710. 133137. 22 (1946).

n Im a. does not always follow from the existence of lim. p. Inst. Osaka City Univ. the condition a. If a.. A power series that represents a holomorphic function is called a holomorphic function element.+. any function f(z) holomorphic in a domain can be represented by a power series in a neighborhood of each point a of the domain.f(z). c. Etude sur la theorie du potentiel pris par rapport au noyau symetrique.al. On thin sets in potential theory. Puiseux./c.. A power series tconverges absolutely and uniformly on every compact subset inside its circle of convergence and defines there a single-valued complex function. a.. we encounter a tsingularity on the circumference for at least one radius. Even Cesaro summability of x:. (0) = . The existence of lim. Foundations of modern potential theory. J. (A) 8 (1957). may be replaced by a. For a given power series P. We call R the radius of convergence and the circle lz ---al = R (sometimes Iz -.n)” is called a power series (in one variable). c. z = R is a singularity (Vivanti’s theorem).t”lk (k a fixed natural number) is called a Puiseux series. we have Abel’s continuity theorem: If the radius of convergeno: of f(z) = C. 1254 339 (X1.6>0 (YStolz’s path). converges to A (Tauber’s theorem.. Conversely. coverges (or is t(C. Landkof./lc. Mandelbrojt. = 0( l/n) cannot be weakened (Littlewood). Since the series is ttermwise differentiable. J. be elements of a tfield K and z be a variable. Princeton. vol. Study.2) Power Series A. and its value at co is defined to be cO. [ 181 M.339 A Power Series [ 161 N. H. B.(. 302-313. For a power series with the circle 01‘ convergence IzI = R. Also R=lim(c. for Adv. Funtions. that the radius of convergence R of Z c. 1. = O( l/n) or n Re a.z -” is called a power series with center at the point at infinity. the argument TV of the singularity on JzJ = R nearest z = R is given in t. Sem. Littlewood).he following way.t”.larg(l-z)l<(n/2)-6}. On potentials in locally compact spaces. every power series can be P(h) = lim sup m. for simplicity. 1972. A. K. a. If we perform tanalytic continuations of a power series from its center along radii of its circle of convergence. Ohtsuka.. the hypothesis on the a. 1477179. Weierstrass defined an analytic function as the set of all elements that can be obtained by tanalytic continuations starting from a given function element (-. General Remarks written in the form C~oc. The value of R is given by R = l/lim sup .(z . Sci. Hiroshima Univ. When t is a local canonical parameter. are real and nonnegative.a)“.z” equals 1. and such a t is called a local canonical pirameter.. Suppose. Abel’s Continuity Theorem As a property of the power series on the circle of convergence. S. [19] M.l (Cauchy-Hadamard formula) with the conventions 0= l/co. we can determine a unique real number R (O< R < co) such that P converges if [z--al < R and diverges if R < lz . In ‘Tauber’s theorem. a series of the form Q = C$.“=O a.$I~ (P(h) . the function is actually a holomorphic function of a complex variable. Inst.. [ 171 N. The theorems concerning additional sufficient conditions for the validity of the converse of Abel’s theorem are called theorems of Tauberian type (or Tauberian theorems). k)-summable (k > -1)) to /f. then 1: a. We assume that K is the field of complex numbers. Ohtsuka.. after the French mathematicians A. 25 (1961) 135-352.. Besides the series Z$. provided that the limit on the right-hand side exists. Lau:rent and V. 1957. Anal. and put Let a and c 02 c 1. Hardy and J. may be bounded from above but not necessarily from below (G. Here.“ea. SuIIicient condi- . 1897). t” is called a Laurent series and a series C.~” is equal to 1 and xg. By putting z-a = t when its center a is a finite point and z-i = t when its center is co. c 2.. E. a series of the form Z.“= _ mc. 1937). Ninomiya.al < R) the circle of convergence of P. A series of the form P = zzo cn(z . Polytech.. Power series are sometimes called Taylor series. if all the c.198 Holomorphic Functions). = o( l/n) and f(z)+A when z approaches 1 along a curve ending at z = 1. In particular. co = l/O. n-m Then c( is obtained from cos a= P. f(z) does not necessarily lead to the convergence of C.1)/h (S. Springer.“= -m c. Such a representation is called the Taylor expansion of f(z) at a (or in the neighborhood of a). The converse of this theorem is not always true. then f(z)+A when z approaches 1 in any sector {zI]zl<l.

z”n with radius of convergence 1.) an increasing sequence of natural numbers) is 1 and lim inf n. the point of A lying on it and nearest to a is called a vertex. . (1) converges for any z with lzl # 1. If the radius of convergence of Csoanzln (with (1. where the si are the number of nonzero a. If R.z” being 1. we also have the following result: When the radius of convergence of Za. then f(z) is bolomorphic at points not lying on IzI = R. the series ~E. (A. .-=s 1-z” Z” implies x a.‘. n-rm a. Lambert Series A series of the form (1) is called a Lambert series. Take any half-line starting at a... Let {a} and {/I} be the set of vertices of the star regions with respect to the origin of X a. it converges uniformly on any compact setcontainedinlzl<lorlzl>l. if the tbranch in Iz( -CR* of the analytic function f(z) determined by P is single-valued meromorphic but the branch in Iz I < R’ with R’ > R* has singularities other than poles. we have where p and rp are the tMijbius function and tEuler function.<... By applying this theorem to Z a. 1942). an+p “’ %+p+l . Hardy and Littlewood (1921) showed that this theorem of Tauberian type is equivalent to the tprime number theorem.) (E./m./Z. (Hadamard’s theorem.+m(ln+l -1. There is a detailed study of Lambert series by K. then IzI= 1 is the natural boundary of Z a.=limsupl~(. 1. are different valued of 1. P. If the na.. . R* can be computed in the following way. which holds for IzI < min(R. Polya.z” and C b. and moreover..+I n ./n = co. Bore1 (1896). . R. = s. to be tsummable for various summation methods are also known.<m. I.. Polya). -A. A generalization of Fabry’s theorem was obtained by G.is divergent. 1). When a point a and a set A of points in the complex plane are given.1255 339 D Power Series tions for Z a. 11.z” has IzI = 1 as its natural boundary (A. Fabry (1896) showed that with the radius of convergence of Z~oa..zb’ is equal to 1. If C a.. If R is the radius of convergence of Z a.=1 lim 5 (1-z)na. Singularities of Power Series Given a power series P = C a.z” with respect to the origin. then IzI= 1 is the natural boundary (Hadamard’s gap tbeorem).... The latter condition was weakened to lim inf... Fredholm).z” converge or diverge simultaneously for IzI # 1.. (1) and the power series x a.. Knopp (1913).u.)/1.z”. Let the a. 1892). 1892). a n+p a“+I %+2 ..z”. is convergent.. is increasing. Then the star region determined by the origin and the set {x/3} is contained in the star region of Za. Fatou. E.. are real and bounded from below. E./I. and limI.z” and Cd.z” (Hadamard’s multiplication theorem. It is known that Fabry’s last condition above is in a sense the best possible (Polya.. an+p+l .z”.)/& > 0 by E. These theorems are called gap theorems because they concern power series with gaps in their exponents. C. be positive numbers and b a natural number greater than 1.-. then we have the reciprocity relation Then the sequence of numbers I. contained in the interval (mi(l -O). the set of centers of the function elements obtained by analytic continuations along half-lines starting at the origin is called the star region of C c.z”.If~a. is the sum extending over all divisors of n.z” is 1. if there exist a suitable sequence of natural numbersm. Hurwitz. D(P)= ‘.b. then R* is called the radius of meromorphy of P and IzI = R* (sometimes Iz I < R*) is called the circle of meromorpby of P. an+zp The following are some results concerning conditions for the coincidence of the circle of convergence of a power series and its tnatural boundary.-. respectively. = 0.. Put Z.-. respectively.ad = A.andanumberB(O<B<l) such that lim s.=R*.. D. then IzI = 1 is its natural boundary (Weierstrass.>0. For a power series x c. As special cases of this relation. Z-l-O.mi(l +0)).. If the radius of convergence of Ego u. then IzI = 1 is its natural boundary.= -Jl). .+. . by a suitable choice of the sequence {E. . the set of points that can be joined with a by a segment disjoint from A is called the star region determined by a and A. Regarding the natural boundary of a power series. it can be shown that if lim.

then all boundary points of the domain of overconvergence are accumulation points of the zeros of S(n. Sem. L’ultraconvergence dans les series de Taylor. and interaction between nearest neighboring sites is taken into consideration. Hermann. Ostrowski. Ostrowski [7. then in a s&iciently small neighborhood of a point on IzI = 1 where f(z) is holomorphic. Uber eine Eigenschaft gewisser Potenzreihen mit unendlichvielen verschwinden der Koefhzienten.2. 340 (XVII. Actualites Sci. Gauthier-Villars. [6] H.2. Introduction in Probabilistic methods are often very useful in the rigorous treatment of the mathlematical foundations of statistical mechanics and also in some other problems related to statistical mechanics. [lo] R. Akad. Ostrowski. .2. cw A power series is completely determined by its coefficients. Wiener. ) converges uniformly. [9] G. Bieberbach. king Models References For the general theory of power series. If logn. R. (2) Markov statistical mechanics.“. (2) 33 (1932). then we can find an integer y such that the c.) has a subsequence {A. = { +l.) is overconvergent. The king model was proposed by II. Fortgesetzte Untersuchungen iiber die Abschnitte von Potenzreihen. Ind. Ising [ 11 to explain the phenomena of phase transitions of a ferromagnet. z) (k = 1. l-100. Press. Uber vollstandige Gebiete gleichmassiger Konvergenz von Folgen analytischer Funktionen.. but little is known about the relations between the arithmetical properties of its coefficients and the function-theoretic properties of the function represented by the series.f(z) = C a. The Taylor series.) and S(n. =O(n. is naturally divergent for IzI > 1. where S(n. .. For overconvergence.. Pitt. 1 (1922). 1955. in which either a + or spin is put on each site of a crystal lattice. 1931.. A known result is that if the power series C c. Math.1 7) Probabilistic Methods Statistical Mechanics A. Tauberian theorems. S.. Acta Math.. f(z) = C. Landau.339 E Power Series E. On the other hand. R. .y” (n > 1) are all integers (Eisenstein’s theorem. but a suitable subsequence S(n. . any power series for which overconvergence takes place can be represented as the sum of a power series having a lacunary structure and a power series whose radius of convergence is greater than 1.. then the power series has a lacunary structure and overconvergence takes place for S(n. z).. . Overconvergence If the radius of convergence of .+. Bourion. S(&. . z) (k = 1. Darstellung und Begriindung einiger neuerer Ergebnisse der Funktionentheorie. [3] L.z” with rational coefficients represents a branch of an talgebraic function..z) (Ostrowski 1256 [2] P. the sequence of partial sums S( 1. Springer. For power series of several variables .z’n has a lacunary structure if the sequence { 1. ).Oa. Table lO.IV. if the zeros of a subsequence S(n. For power series expansions Appendix A. Analytische Fortsetzung.) may still be convergent.-B.. Tauberian theorems..(l + 0) (0 > 0). [S] N. Clarendon Press. Preuss.z) (k=1. Wiss.. Abh. Math..21 Analytic Functions of Several Complex Variables. 5577565. S] called this phenomenon overconvergence and proved the following result: By definition. For theorems of Tauberian type. If this situation occurs. 1852). Univ. G. Also . [S] A. Oxford Univ. 41 (1918) 253-270. 1932. Hamburg.370 Rings of Power Series. Jentsch. and (5) the +Boltzmann equation. > E. [4] S. z) (k = 1. z). ) has no accumulation point on IzI = 1. second edition. S(2. Bourion [9] gave a unified theory of these results using tsuperharmonic functions. [l] E. we explain here (I) Ising models. (3) percolation processes. Jentsch [lo] showed that all singularities of a power series on its circle of convergence are accumulation points of the zeros of the partial sums. Mandelbrojt. z) = C:=O uvzy. -1)’ be the totality of spin configurations in V. (4) random Schrodinger equations. 1929. for formal power series . As examples of such methods. This result includes Hadamard’s gap theorem as a special case.2. Dienes.references to 198 Holomorphic Functions.2.} such that A. . G. Each ele- . (1921). 3277350. . B. [7] A. For singularities.. Let I’ be a cube in the d-dimensional integer lattice space Zd and X. 1937. Les singularit& des fonctions analytiques represent&es par une strie de Taylor. H. A. 1958.z” is 1.. z) (k = 1. Springer.. Conversely. Ann.

It is known that any Gibbs state of the Ising model is a reversible stationary state of the stochastic Ising model.e. D. T is the tabsolute temperature). On the other hand when an external field is absent (h = 0) and d 2 2. j) means that (i.(2) has been carried out by Onsager [4]. In the following we assume h = 0. h # 0) there is only one Gibbs state for any fi. 63. a phase transition occurs. and if 1> 1. on V which minimizes the free energy F. we mention some known facts in this field.spin on each site on Zd. (variational principle). t + At] with probability ci(qt)At + o(At). for a sufficiently low temperature.. infinite interacting particle systems. at present the probabilistic delinition of Gibbs states given by Dobrushin [2] and Lanford and Ruelle [3] is prevalent. In particular. Physically.healthy one. However.h(a) = exp(-8Wd) c . Since for each Zd-homogeneous (i. then S-i is a stationary state. 0 < t < cc. j) is a nearest neighboring pair of sites and h stands for the parameter of an external field. If pLt= p for all t > 0.. a (limiting) Gibbs state can also be defined for the infinite domain V= Zd by the above mentioned variational principle. (3) In the 2dimensional case.1257 340 D Probabilistic Methods in Statistical Mechanics ment of XV is denoted by c= {ci}ieV (ci= $1 or -1). Percolation Stochastic Ising models. This process on X is called a spin-flip model. Then there exists a unique state g{.x. For an initial distribution p we denote by pt the distribution at time t. < co) such that if A < 1. there exists a critical value /l. every Gibbs state is Z*homogeneous [S. Let X= { +l. Let Zd be the d-dimensional integer lattice space. i. let us consider a random motion of spins which evolves while interacting with neighboring spins. The most important result is the following: There exists a critical value 1.. (2) For d > 2.(d) but it occurs for every fi > b.(i) changes to -v]. Example 1. Putting + or .ew-&424)’ aEXv. q(i)= +l for infinitely Processes many i] = 1. Stochastic Ising models. +l. For each state p. Finally. Example 2.e. A stochastic Ising model was proposed by Glauber [8] to describe the random motion in a ferromagnet upon contact with a heat bath. We explain this field by looking at a typical class of processes.(d) such that the phase transition does not occur for any b < b. Denote by -1 the configuration at which all sites are healthy and by 6-r the unit point mass at -1. h). a Gibbs state is an equilibrium state. Free energy plays an important role in the study of the ergodic properties of these models. (4) For d 2 3. C. there is another stationary state p satisfying p[qEX.. -lJzd be the totality of spin configurations and an element of X be denoted by r={~(i)}~. which has made rapid progress during the last decade. Then the flip rate {ci(~)j is defined by the potential of the Ising model. and many models occurring A percolation process is a mathematical model which describes the random spread of a fluid . The calculation of b. p is called a stationary state.6-i is a unique stationary state. Contact processes. biology. (0 < 1.(d). For the configuration qt at time t.&)= v-a lim -444 1 VI is well defined. Markov Statistical Mechanics in physics.e. -1 and r(j)=+l}=k. We suppose that a spin configuration 0 has a potential of the type ‘V(a)= 1 Oiiaj+hC CT. q. The process is described in terms of a collection of nonnegative functions ci(q) defined for iEZd and q E X.” is called a Gibbs state on V of the Ising model with parameter (p. The field of investigation of stationary states and statistical or ergodic properties of these processes is called Markov statistical mechanics. It is known that if an external field is present (i.(i) in the time interval [t. gj. translationally invariant with respect to Z”) state p the mean free energy 1 . the free energy is defined by where /l= l/kT (k is the tBoltzmann constant. and the converse is also valid. = ki Here + 1 denotes an infected individual and -1 denotes a. The flip rate of the contact process is given by 1 ci(T) if if q(i)= q(i)= #{jllj-il=l. A tprobability measure on Xv is called a state on V..j)cV isV where (i. the phase transition never occurs. A contact process was introduced by Harris [ 1 l] to investigate the spread of infection.~d (v(i)= +l or -1). there are at least two Gibbs states. and sociology are formulated as a class of infinite-dimensional tMarkov processes. (1) In the l-dimensional case. there is a Zd-inhomogeneous Gibbs state for sufficiently large p [7]. in which various physical quantities are calculated. the mean free energy is a nondecreasing functional along the distributions pLt. (i.

.x. (2) p. where {. Let E. w)} is a +Poisson random measure with mean measure dy and cp~:x)is a nonnegative measurable function satisfying v(x) = 0(1x1-“-*) as Ixl+‘co. This system of equations is considered to be a model describing the motion of quantummechanical particles in a random medium. dl is cp(X-Y)ddY. In the l-dimensional case.40. n/z as 2.X. and define the critical percolation probability pH = infj p 1O(p) > 0).402 Statistical Mechanics)..)}l(xi-xj.. The quantity N(J) can be identified with a limiting state density of A(w). Let H(p) be the probability that the fluid spreads infinitely far. Usually.y. Since A(w) and its shifted operator A(w(.. and denote by N(A) the mean of E.) at -u(t~x. 0) and has the following asymptotic form at A= 0: i”‘210gN(i)+ -y..x. It can be used to describe phenomena such as the penetration through a’ porous solid by a liquid or the spread of an infectious disease [ 141.. the problem is to investigate various spectral properties of the self-adjoint operators A(w). It is assumed that this system of potentials forms a spatially homogeneous random field with the tergodic property.=2sin(x/lS) for the triangular lattice. Let 1. . the theory of tlarge deviation for Markov processes plays a crucial role [20]. Then N(1) is a nondecreasing function vanishing on (--co. then it is proved that A(w) has only a pure point spectrum [ 173. as V tends to R” regularly.~(w)} is the set of eigenvalues of A(w) in a smooth bounded domain V in R” with a Dirichlet boundary condition and 1VI is the volume of V. w). d = the diameter of the hard spheres). Here we describe the latter only. An approach to the Boltzmann equation in the spatially homogeneous case can also be based on a tmaster equation. 3. therefore the corresponding operators are of the form A(w)= -A+Uc. M. 8:).2 sin(x/l8) for the honeycomb lattice.. Each bond b is open with probability p and closed with probability 1 -p independently of all other bonds. Mathematically rigorous. w) implies that the spectral structures are independent of each sample w a. P) and A denotes the +Laplacian in R”. if their structures of A(o) are measurable with respect to @. B} be a countable connected +graph with a set of sites (tvertices) S and a set of bonds (tedges) B. Several rigorous results have been obtained. In multidimensional cases. E.s. w). Boltzmann Equation where w denotes a random parameter in a probability space (52. F... Random Scbriidinger Equations Random Scbriidinger equations are tschrtidinger equations in R” with random potentials U(x. To obtain the above asymptotils behavior of N(1).J. he showed that solutions of the BBGKY hierarchy converge to those of the Boltzmann hierarchy for small time under the Boltzmann-Grad limit (N + co. t>O. ~~~. For a gas of hard spheres this is determined by the master equation a --utt. x1 /.4. asymptotic behavior at the left edge of the mean of the resolutions of the identity for a certain A(w) have been investigated [19].gR3. if the potentials u(.. 0. x. discussions were given by 0.(x. = {S.(O. and (3) pH = 1 . Ntl’+ 1.~)ld~. namely.w)= r JR" In the kinetic theory of gases the tBoltzmann equation is derived from the tliouville equation by considering the BBGKY hierarchy of particle distribution functions for N particles and then by taking the limit N --* co under certain conditions (. the limit function of E.w). the above assumption on the potentials U( . through a medium. Then it is known that (1) pH = l/2 for the square lattice [16]. Kac [22] considered a Poisson-like process describing the random time evolution of the n-tuple of the velocities of n particles. Mathematically. Lanford [21] for a gas of hard spheres. w) are functionals of a strongly ergodic +Markov process. (1) where S2 is the 2-dimensional unit sphere. It is assumed that the random potential for this A(w) takes the form U(x.340 E Probabilistic 1258 Methods in Statistical Mechanics where {n(dy.cs) be the continuous kernel for the resolution of the identity for A(o). where y1 is the first eigenvalue of --A with a Dirichlet boundary condition on the ball in R” with unit volume. the process is identified as a site percolation process or a bond percolation process... +x)) are tunitarily equivalent. and each eigenfunction decays exponentially fast [ 1S]. Suppose that a fixed point o is the source of a fluid which flows from o along open bonds only.

. more precisely. based on Kac’s work [22]. Smythe and J. (Original in Russian. 235-241. Lanford III and D. One-dimensional random Schrodinger operator with purely point spectrum. Moreover. y'=y-(y-x. 11 (1977). P.(x. M. Ann. Appl.. 1981. 2 (1974).. 149-164. Kesten. “-m s S(.. 5 (1970). Advances in Math. Liggett. Ising. Theory Prob. ibid. a process of this type describes the time evolution of the velocity of a particle interacting with other similar particles.1)1. 13 (1968). 671. Phys. where u. Methods in Statistical Mechanics the uniform x. A. Holley and T. 1968. The propagation of chaos is the stage corresponding to the tlaw of large numbers. [17] I. on Random Fields.). 194-215. [4] L. Let u be a positive constant and S(G) denote the (3n . J. on S(. if {u. Ruelle. 246-290.} is said to have Roltzmann’s property or to be chaotic (or u-chaotic to stress u).} x'=x+(y-x.1259 340 Ref. Ann.X)= s SxR' (44 X’ML Y’) . Math. McKean [24]. and u(t) is the solution of the following Boltzmann equation with u(O) = u: &U(f. a sequence {u. Comm.. A. (Original in Russian.. Harris. 74 (1980). Phys. 197-224. References [l] E..) [8] R. which was also discussed by Kac [23] and McKean [25]. 253-264. 83-94.rp. Comm. 13 (1969). Phys. Observables at infinity and states with short range correlations in statistical mechanics.(x)Wdx s for each m > 1 and (pkE C. McKean [26] introduced a class of tMarkov processes associated with certain nonlinear evolution equations including the Boltzmann equation.= fi k=l R3 cp. Time-dependent statistics of the Ising model. On the absence of nontranslationally invariant Gibbs states for the two-dimensional Ising model. 41-59. such a Markov process was constructed by solving a certain tstochastic differential equation [27]. 198-206.. Problems of Information Transmission.44 x)u(t. 1978. 969-988. 73 (1980). [ 121 R. [9] R. Markov processes with a large number of locally interacting components: Existence of a limit process and its ergodicity. Dobrushin. Springer. Molchanov. J. Percolation processes I: Crystals and mazes. Cambridge Philos. In the case of the spatially homogeneous Boltzmann equation of Maxwellian molecules without cutoff. Comm. L. E. Interaction of Markov processes. Kac’s assertion is that the Boltzmann equation is to be derived from the master equation via the propagation of chaos. Firstpassage percolation on the square lattice.=xj-(xj-xi. Pastur. This implies the existence of probability measurevalued solutions of the equation. .)dx. Probability. 598. The critical probability of bond percolation on the square lattice equals l/2. [ 161 H. Esztergom.. Z.. 31 (1925). 187-248. North-Holland. Dobrushin. Proc.z)z.) [lo] F. C. L. Glauber. Appl. then {un(t)} is also u(r)chaotic. Functional Anal. Beitrag zur Theorie des Ferromagnetismus. Phys. Lecture notes in math.JG) . (Originals in Russian. Crystal statistics I: A twodimensional model with an order-disorder transition.. M. Phys. [2] R. The survival of contact processes. Probability.) [3] 0.x. The propagation of chaos was verified by Kac [22]. [ 131 T.. Theory Prob. 53 (1957).. Spitzer. Appl.(x..Z)Z. Z)Z. The description of a random field by means of conditional probabilities and conditions of its regularity. Sot. Onsager. and others for a considerably wide class of nonlinear equations of Boltzmann type (with cutoff). Gibbs state describing coexistence of phases for a three-dimensional Ising model. 629641. Higuchi.. R. Liggett. L. (Original in Russian.1)-dimensional sphere with center 0 and radius ./&) for each n > 1. Aizenman. 17 (1972) 582-600. 7 (1971). 65 (1944). 1972.(R’). Given a symmetric probability density u... [15] R. S. E. M.)u.(O) = u.(t) is the solution of the master equation (1) with u. Probabilistic distribution on S2 and x. Goldseid. [14] S. Translation invariance and instability of phase coexistence in the two dimensional Ising system. Markov processes with many locally interacting components: The reversible case and some generalizations. The stochastic evolution of infinite systems of interacting particles.) .xi.. [7] R. 1977./&. Y. Math. Springer. Broadbent and J. Math. Hammersley..dx. The next stage is the tcentral limit theorem or fluctuation theory. 117.. 4 (1963) 294307. l-10. [6] Y. 1971. = xi + (Xi . if there exists a probability density u on R3 such that lim c~l(x. Phys. Dobrushin. Conf.. Contact interactions on a lattice. and L. T. Rev. Lecture notes in math. [S] M.} is a u-chaotic sequence.149. J. Math. 1 <k < m. [ 1 l] T. 1977. 6 (1978). H. Proc. Wierman.

Anal. we regard a Yopological space T as a measurable space endowed with the topological a-algebra B(T) on r. F(x) = 1. probability measure appears usually as the tprobability distribution of a irandom variable (. 379-400. 47(z)= cei@3x)dcqx).2. Rational Mech. P.e. Interscience. US. Lecture notes in physics 38. Acta Phys.270 Measure Theory. which is an n x n matrix whose (i. Some probabilistic aspects of the Boltzmann-equation. %“) an n-dimensional (probability) distribution.. 1975. Acad. where !S” is the galgebra of all +Borel sets in R”.270 Measure Theory). 28 (1975). 525-565. Time evolution of large classical systems. (The covariance matrix is also called the variance matrix or the variancecovariance matrix. JR” (1) where (z. x E R”). etc. F(x) = P( {w 1X(w) <x}). Springer. For probability measures on topological spaces . General Remarks A probability measure 0 on a tmeasurable space (S. Donsker and S.J’)-element is m. Several different quantities characterize the properties of probability distributions in one dimension: the mean (or mathematical expectation) m = s?w x0(x). A distribution function is characterized by the following properties: (1) It is monotone nondecreasing. Let X be an n-dirnensional random variable with probability distribution Q defined on a tprobability space (a. J.2) Probability Measures A. Japan.. Wahrscheinlichkeitstheorie und Verw. EE % ‘. 46 (1978). x) denotes the +scalar product of z and x (z. Proc. d = E(X -m)‘. P). [22] M. F(x)=0 and lim. Asymptotics for the Wiener sausage. Nakao. The structure of eigenfunctions of one-dimensional unordered structures. the moment matrix. 69101. which is an n-dimensional vector whose ith component is given by mi = JxidO(x). The characteristic function of @ is the +Fourier transform 4” defined by ZER”. (2) it IS right continuous. 1271 H. Math. 56 (1966). (Original in Russian. the variance 0’ = jYZ IX HI(‘~@(x). Math. Pure Appl. xkd@(x). 3 (1977). McKean. Comm.he distribution of X is given by Q(E) = P( { Q 1X(W)E E)).) [19] S. I1 l-139. Sci. [21] 0. Math. the following quantities are frequently used: the mean vector.( . 3. W’). D. Tanaka. [23] M. Characteristic Functions Consider a probability measure @ defined on a measurable space (R”. (3) lim. USSR-Izv. Arch. Varadhan. and the characteristic quantities of @ defined above are given in terms of X((u) as follows: m = E(X). Speed of approach to equilibrium for Kac’s caricature of a Maxwellian gas.. When @ is an n-dimensional distribution. E.) The covariance matrix and the moment matrix are ipositive definite and symmetric. 5) with Q(S) = 1 (. Comm. 435-455. P).105. 1959. Lanford III. The moments and the moments about the mean are connected by the relation p. the standard deviation 0..342 Probability Theory). [20] M. Probability and related topics in physical sciences.. Fluctuations in the kinetic theory of gases...341 A Probability 1260 Measures B. Unless stated otherwise.. Then I.j=jxixjd@(x). 12 (1978).~]). ). 8”=G1J(R”)). the kth absolute moment flk=JTw Ixlkd@(x). [25] H. and its (cumulative) distribution function F defined by F(x)=@(( -a. Austrica Supp. 1973. the kth moment about the mean pLk=JTc(x-m)kdG)(~). On the spectral distribution of the SchrGdinger operator with random potential. 1261 H. A class of Markov processes associated with nonlinear parabolic equations. A one-to-one correspondence exists between a l-dimensional distribution Q. . Probabilistic treatment of the Boltzmann equation of Maxwellian molecules. 21 (1966). the covariance matrix. Quantities Distributions Characterizing Probability [IX] S. In probability theory.. which is an n x n matrix whose (i. Math. Kac. j)-element is crij = J(xi-mi)(xj-mj)d@(x). = XL=0 r 0k cc. McKean. Let X(w) be a real random variable on a tprobability space (Q. Nat. 341 (XVll. 1907~1911. the +o-algebra generated by the +open subsets of T. Pure Appl. McKean. R. 28 (1975)..m)k (r = 1. 1241 H.. Molchanov. Gebiete.-. 1978. The quantities listed above are defined only under some integrability conditions. 10. 67. 347-367.‘%. P. Z. C. G) is defined to be a imeasure on (S. i. From this probabilistic background we often call a probability measure on (R”. Springer. the kth moment c(~= s?. P. Similar statements hold for the multidimensional case. Hence the distribution of an R”-valued random variable is a probability measure on (R”.-. etc. Kac. S. A.

and @‘3 is continuous and tsingular. . xdz. Table 22. i 0. . f(t. Let @ be an absolutely continuous distribution. we have j. The following l-dimensional distributions are absolutely continuous: the normal distribution (or Gaussian distribution) N(p. a. the binomial distribution Bin(n. z.2.-3a.a$ a. . is called a purely discontinuous distribution. .. . 4) with parameters m and q. aD. then the left-hand side of (2) is equal to Q(I). Given a l-dimensional distribution @ with fik < +co. .192 Harmonic Analysis).1261 341 D Probability Measures Then the Fourier transform of Q..b)= t $ Cu..=l.2 0. . . and the negative binomial distribution NB(m. By virtue of the tLebesgue decomposition theorem. the bypergeometric distribution H(N. a. i. A counterpart to Bochner’s theorem holds for any positive definite sequence as well (THerglotz’s theorem)..2.342 Probability Theory). the multiple hypergeometric distribution. Any continuous positive deiinite function cp on R” such that ~(0) = 1 is the characteristic function of an n-dimensional probability distribution (TBochner’s theorem) (.. p) with parameters n and p.a. t+.@. b. n). the negative multinomial distribution.@. . . and p.a.< b. The set D of all discontinuity points of @ is at most countable... we have necessarily exist for all n-dimensional distributions but does exist for a number of useful probability distributions @.x))d@(x) (zcR”) does not where Q1 is purely discontinuous. @ is called a continuous distribution.. if D is a lattice. zCp)of the n-dimensional space R” and any complex numbers a.. b) denotes the modified function of [a.+u. Q. n. . etc.+a. x..W1.Appendix A... b. b. (For characteristic functions of typical probability distributions .b). . We now list some frequently used ldimensional lattice distributions (for explicit data . For general information about criteria that can be used to decide whether a given function is a characteristic function .. .. we denote by yk the coefficient of (i~)~/k! in the TMaclaurin expansion of logrp(z)..-6af. If the distribution function of @ is a continuous function.) E R”. . D.[8]. 1) is ~ called the standard normal distribution).. n. where 81 denotes the boundary of I.+u. Table 22): the unit distribution with @( (0)) = 1. The characteristic function is often useful for giving probability distributions explicitly. the Poisson distribution P(1) with parameter i. The characteristic function cp of an ndimensional probability distribution has the following properties: (i) For any points z(l).+u. If an n-dimensional interval I = [a. and x=(x. y2 = CQ-a: = c?. t=u or b. the .z’k’)uja.. This function f(x) is called the probability density of @..] defined by ai6xi<bi(i=1. which can also be written as E(ei@.. . (p= 1. (iii) ~(0) = 1. Specific Distributions Given an n-dimensional distribution @‘.~. y.ER. The moments and semi-invariants are connected by the relations y1 = CI~.. . is also called the characteristic function of X. (ii) cp(zCk)) converges to ~(0) as zCk)-+O. u. In particular.. p) with parameters n and p. . . l/2.. . up..+2a:.. and then f(z) uniquely determines @.e.) The moment generating function defined by f(z)=lexp( -(z. indicator (2) where f(t.=a. b] defined by 1. is tabsolutely continuous with respect to tLebesgue measure.%~o. We call yk the (kth order) semi-invariant of 0.. every probability distribution can be expressed in the form @=u.fl dz”’ .. Equation (2) is called the inversion formula for the characteristic function cp.~z. A complex-valued function cp of z E R” is called tpositive definite if it satisfies the inequality in (i). The following k-dimensional lattice distributions are used frequently: the multinomial distribution M(n. n) is an interval of continuity for the probability distribution Q..X)) (. bl.. . p) with parameters N. The following properties play a fundamental role in the study of the relationship between probability distributions and characteristic functions: (i) the correspondence defined by (1) between the n-dimensional probability distribution @ and its characteristic function cp is one-to-one. @(al) = 0. y4=a4-3a:-4alag +12$a. .. the geometric distribution G(p) with parameter p. CT*)with mean p and variance g2 (sometimes N(0. (ii) For any up. a point a with @({u}) > 0 is called a discontinuity point of Q.Appendix A. @ is called a lattice distribution. When O(D) = 1. a. Then there exists a unique (up to Lebesgue measure zero) measurable nonnegative function f(x) (xe R”) such that Q(E) = JEf(x)dx..

Sn.q)=NB(m. ‘Pk (k= 1. uk (k = 1. * Qz has a density f(x) = JR”fi (x y)dF.d* r(~l. and @. Such convergence is called weak convergence in probability theory. be tindependent random variables with distributions @...~+t)d~l(~)d~Z(y) is called the composition (or convolution) of @. )there exists(x3. for every k. + v2. ok).F. Some of the distributions listed above have the reproducing property: P(Ir)*P(I.+P2. ).... and @ are probability measures on a tcomplete separable metric space It should also be noted that the probability measures on a complete separable metric space consti- . (6) Iirk. P(%.f(x)d~.. The mean concentration defined by . Therefore. Convergence of Probability Distributions Nh.. (2) At every continuity point of the distribution function F(x. Qz. . . etc. Furthermore... C) with mean vector p=(pIrpz. II) be the 1-dimensiona. . @I. +X.. has a density fr(x). and let P(I.I?‘) = 0). normal distribution with mean m and variance u. etc.. the kth order semi-invariant of the convolution of two distributions is equal to the sum of their kth order semi-invariants. * F2(x) = Ja”F.(x). n). %(F)< W’).. the Fdistribution F(m. (4) For all open G c R”. the exponential distribution e(a) with parameter a. Yk= P(R.. )and(cr. Let X.c2. )suchthat@(cc. . Ok(E) = D(E). +I.. the distribution function F./l. we can use it to QF....pJ and covariance matrix I= (a. Yu. If l-dimensional distributions Qk. c).p). /I]. When R is a topological space.) (k= 1. lim.xJ = F(x. o1 + is also useful for similar purposes.). and is denoted by Q. the uniform distribution U(a. V.~:+~22). the gamma distribution T(p. where xE is the indicator function of the set E.) (Fk is the distribution function of Q.(y).2.) of @.@‘2)=max(8s.37 Banach Spaces). of The concept of convergence of distributions plays an important role in limit theorems and other fields of probability theory. F.. Suppose that we are given a family of distributions @ = {@(a.. then we say that ct. the t-distribution t(n). (3) For every continuity set E of @ (namely. and ‘pz.(y). ) to converge to @ weakly. )* @(x. respectively. N(~c. and X.).a)* C(P. E. we have Qk= N(m.. For a sequence of ndimensional distributions (I$ (k = 1.& .. )= @(cc.2). /I’~.1))2 dx s cc Caucby distribution C(p. a l-dimensional distribution function I>O. where p is a metric defined in the following way: Given any n-dimensional distributions @r. q).. )} indexed with parameters cc. lim sup. &(x1. called the Levy distance. (1) For every continuous function with compact support. (F(x + I) .2). If @. r(P2./?I . The metric p.). (D.q)* NB(m. is @I * Qz When Fi is the distribution function of Qi (i= 1.(n:-y)dF... Convolution Given any two n-dimensional distributions Or. and @)2with characteristic functions ‘p.. p. i?in(n. the Dirichlet distribution. n). we consider convergence of probability measures on 0 with respect to the tweak topology introduced in the space of measures on Q (. These are known.341 E Probability 1262 Measures random function c&j variables.. a set such that Q(E . * Qz is expressed in the form F. the +X2 distribution x’(n).. P(i.rr:)=N(~l+~2.. .u (+median) and a../I. lim. Levy [6]).+m.. the beta distribution B(p. c~. has a reproducing property. i.. then @. the n-dimensional distribution @(E) = SR2n~E(. --mcx<‘x max (F(x+i)-F(x-1)). as Cramer’s theorem and Raikov’s theorem. *Qz.(z) (p*(z).+n.q). is called the maximal concentration function Since it satisfies the relation (i= 1.NB(m. etc.p)=Bin(n. Linnik proved a similar fact (the decomposition theorem) for a more general family with reproducing property by using the theory of analytic functions [9]. Iffor(a.. QJl)=- g2).*F2(I)<QFI(I) study the properties of sums of independent F (P. .p)* Bin(n..(x)=Sa...a) be the distribution obtained through translation by a of the Poisson distribution with parameter J. /I) on an interval [x.2) exist such that N(m. .. .. lim. then cp is the product of ‘pr and cp2 : q(z) = ‘p. Prokhorov in metric spaces [lo]. V. . a) with parameters ...2)..& ... we put ~~~=inf{~(@~i(F)<@~(F~)+ E for every closed F} (F” is the s-neighborhood of F) and define p(@1.2. was introduced by Levy [6] in one dimension and by Yu. Each of these conditions except (2) is still necessary and sufficient for @)nto converge weakly to @ when @.a)=Y..~. Let N(m. .a.. each of the following conditions is necessary and sufftcient. Then the distribution of X.f(x)d~. @I =O. x. such as the k-dimensional normal distribution N( p.%)= Given F(x).... * F2 of @. lim inf.2) for some mk..4. there are several k-dimensional absolutely continuous distributions. and OI. u) =Ql em. (5) For all closed F c R”. the Z-distribution Z(m.)*C(/*.. If q(z) is the characteristic function of the convolution of two probability distributions a.)=uP.(G) 2 O(G)..x.)= P(1. &..F(x . *Y2. . W.

. Let (pk be the characteristic function of an n-dimensional probability distribution ak. and lim..) is a positive semidefinite matrix. Let @ and Y be probability distributions with distribution functions F and G . . 1.4’1 cpqzpzq ei(zsx)-l -$A$ n(dx) . The characteristic function of a l-dimensional infinitely divisible distribution can be written in the form cp(z)=exp ( m + 1 4% z). n(k). * @‘k (= @tk).J?. and K(u) is a nondecreasing bounded function such that K( -co)=O. and n(dx) is a measure on R” such that n({O})=O and -n(dx) < co.} converges weakly to @ [7] (Levy’s continuity theorem).) Let CDand Y be n-dimensional distributions.2. . It is called Kolmogorov’s canonical form. (cp. A family @JaEA) of probability measures on a complete separable metric space is said to be tight if for every E> 0 there exists a tcompact set K = K(E) such that @. We can give a criterion for the convergence of probability measures in terms of their characteristic functions. . where m is a real constant.Jz) = q(z). n(k). . . o is a nonnegative constant. For the characteristic function of an infinitely divisible ndimensional distribution.. the use of tLaplace transforms as a substitute for Fourier transforms provides a powerful tool. probability distributions. If the probability distributions of the sums X. If a l-dimensional infinitely divisible distribution @ satisfies JR1 x2 d@(x) < co. Formula (4) is called L&y’s canonical form. xjd@. Let Xki. G(u) is a nondecreasing bounded function with G(-co)=O. A family Qa (aeA) is tight if and only if it is ttotally bounded with respect to the topology induced by the Levy distance. (For infinitely divisible distributions on a homogeneous space . Both normal distributions and Poisson distributions are infinitely divisible. @z. (4) where rn~:R”.. and assume that the distribution of Xki belongs to a(~&.Eu(s) such that @=@. > > z=(z. There are many results on the relation between these functions.@(dx) <a. If the sequence { (pk} converges pointwise to a limit function cp and the convergence of (Pi is uniform in some neighborhood of the origin.. Formula (3) is called Khinchin’s canonical form.Y(E)=(D(IE)(1E={~~l~~ E}) for every set E.-+O as k+m. co). (5) 5 -a’ An n-dimensional probability distribution @ is called infinitely divisible if for every positive interger k. then Qk weakly converges to CD. Infinitely Divisible Distributions + m (eizu -1-izu)$dK(u) > .@. .(K’) <E for all aeh. there exists a probability distribution Qk such that @ = Qk * @‘r * . . Ifforsome1>O.(x)=J:.1 y’ dG(u) > irz-gz2 -02 T (3) where y is a constant. This condition is sufficient but not necessary.z)=exp(iuz)lizu/( 1 + u’).. . . .2. . lim. @. ).2.. then the limit distribution is infinitely divisible. . A(u. If all absolute moments exist. Let @r. For any probability distribution concentrated on [0. z)( 1 + uz)/uz at u = 0 is defined to be . . . and the value of A(u. then cp is also the characteristic function of an n-dimensional probability distribution CJand the sequence {@.. z) - t P. where s. and their convergence [ 14161. z&R”. . v is a nonnegative constant.5 Additive Processes. ~~l/?[lu=cc forBj=S”“~lIxljd~(x)<co. Hence a tight family Qa (a E A) has a weakly convergent subsequence. . Then $ converges weakly to @ if and only if for every z. The method of moment-generating functions is also useful. we say that @‘E U(E). Suppose that ak and CDare n-dimensional probability measures with characteristic functions (pk and rp.z2/2.. *Q2* . q. be independent random variables for every k. i= 1. If an n-dimensional distribution @ satisfies the condition slxl . .. *$. Then CDis an infinitely divisible distribution if and onlyifforeverya>Owecanlind@l. the canonical form is as follows: q(z) =exp ( + J( R” i(m. = CyLkj Xki converge to a probability distribution as k+ co. and @ be l-dimensional distributions.. we say that @ and Y are equivalent.1263 341 G Probability Measures tute a complete separable metric space with respect to the Levy distance. The method of probability generating functions is available for the study of arbitrary probability distribution concentrated on the nonnegative integers [14]. . then its characteristic function is given by cp(z)=exp ( imz-iz’ G.xjd@(x) (j= 0. i = 1. .

the distributions of the sums B[‘(Cbl X.e. {o)~nl n.z)= tan(lrr/2) (ctf l). In any ldimensional symmetric distribution. Let {X. Furthermore.Appendix A.} such that for suitably chosen constants B. (For the stable distribution with exponent l/2 ...z)cp(iL. the general form was obtained by L&y [6]. c > 0. Table 22). The ratio y.i)/b). The characteristic function of a quasistable distribution has the canonical representation d4 = exp $(z)? I. By the delinition we see that all stable distributions are infinitely divisible. > 0 and real I. + j. If for suitably chosen constants A. Kolmogorov’s Extension Theorem $(z)=imz-c~zlz(l+i~(z/~zl)cr~(z. with identical distribution @ and for any positive numbers i. i. IflI < 1.. z)= Let R = RT. the convolution @I * @)2 is equivalent to @.t. which implies ~(z)=(-~~+i(z/~z~)cJ~z~~. we can define quasistable distributions..2. putting q(z)= expll/(z).(x) for every 0 <a < I @ is an L-distribution if and only if there exists a sequence of independent random variables {X. In the l-dimensional case. We associate with RT the o-algebra dT generated by the cylinder sets. w(z.~. and (3) tj(jLz) = q(z) for every z. We can characterize stable distributions in terms of their characteristic functions q(z) as follows: For every pair 1.c0 lzla for a symmetric stable distribution. and the stable distributions with exponent a = 1 are the Cauchy distributions. which B. Semistable distributions are another generalization of stable distributions.0) <p < F(l.F(m + x) = F(m-xx). @ is said to be quasistable if to every b. there exists a i = i(l.<t. converge to @ and sup. A quasistable distribution with c(# 1 is obtained from a stable distribution by tran:jlation. Given a probability . 1. has the distribution 0. there correspond a positive number b and a real number i such that we have the relation F((x--I. and Q(z.. The quantity cP such that F([. .. where m is a real number. = /lx/a3 is used as a measure of departure from symmetry of a distribution and is called the coefficient of skewness. yZ expresses the degree of deviation from the normal distribution. b. but quasistable distributions with a = 1 are not. it is called symmetric. wherec.}.&) > 0 such that cp(iz) = cp(1.P(Ix. In particular./B.2 is called a median. We have +(z) = . A necessary and sufficient condition for a distribution to be quasistable is that its characteristic function q(z) satisfy the relation cp(h.z)=+(~.z)= cp(bz)e@ (y = i -2. we have I/J(. We call @ a stable distribution if for every pair of distributions al.< < t.)/b.The parameter x is called the exponent (or index) of the stable distribution. . Then the following three statements are equivalent: (1) CDand Y are equivalent.. and A. -oo<c. and B. If y2 f 0. every distribution equivalent to @ is also stable. and n = 1.>O. N. Quasistable distributions are L-distributions.)* F( (x . > 0. every moment of odd order about the mean (if it exists) is equal to zero. A distribution function F(x) is called unimodal if there exists one value x =: a such that F(x) is convex for x <u and concave for x > a. The Shape of Distributions Let F(x) be a l-dimensional distr bution function. The parameter u is called the exponent of the quasistable distribution. A l-dimensional probability distribution D is called an L-distribution if the distribution function F of @ is the convolutioc. where T is an arbitrary index set. >O. (2) G(x) = F(ix) for every x. and characteristic functions cp and $.X. of F(x/a) and some other distribution function F. A distribution is called semistable if its characteristic function q(z) satisfies the relation tj(qz) = q”$(z) for a positive number q (# l).. i. where q(z) = exp(+(z)). Also in this case.‘(& X. If @ is stable. l<k<n.<co..3 is called the coefficient of excess.. > 0 and A. . converge to a distribution. If F satisfies the relation 1 . We can restate this characterization as follows: @ is stable if and only if for every pair of independent random variables X.) be a sequence of independent random variables with identical distribution.z)cp(b. the distributions of the sums B. Generalizing stable distributions. such that (i. H.z)+$(~“~z). .a)).z). the limit distribution is a quasistable distribution (L&y). For the normal distribution. we have y.2X2)/).) (0 < p < 1) is called the quanfile of order p of F. The stable distributions with exponent x= 2 are the normal distributions. Let F be the distribution function of a l-dimensional distribution @....(W) denotes the tth coordinateofru. -&).)-A. @)2equivalent to @.O<x<2. the ratio 1~~ = p3/a4 ... V.I>~)~o as n+ a for every c > 0. All L-distributions (and hence quasistable distributions) are unimodal [ 181. the quantity [1. = y2 = 0. and X.O < n < 2.)-A.341 H Probability 1264 Measures (2/n)loglzl (SL= 1).i2)/b2) = F( (x .(w)6 E.~~Jw)EE. there exists a positive number i. where X.EkE93(R1). Gnedenko and A. Kolmogorov 1171 called stable distributions also.

* as the a-algebra generated by tE the system of half-spaces {xeT. cp is continuous and positive definite. If I(z) = 7.*)).. But it is known that there is no probability measure on T= T* (topological dual of T) which corresponds to cp. [3] J.1265 341 Ref. is the characteristic functional of a probability measure on T.nl. dt)= s exp(i(x. [4] M. Thus every functional cp on T. . consistency condition and hence.” I be the tproduct measure of k copies of a given probability measure Q on R’.e. Characteristic Function& Dimensional Spaces on Intkite- Contrary to the finite-dimensional case. English translation. Schwartz’s spaces Y(R) and 9(R”) are nuclear. which is positive definite. IIoL. Foundations of the theory of probability.11 tll’). Let T be a real tvector space endowed with the topology z defined by a system of tHilbertian seminorms { 11.e. Grundbegriffe der Wahrscheinlichkeitsrechnung. Gauthier-Villars. Chelsea. Let (7. are independent identically @distributed random variables. is the probability measure of a Gaussian twhite noise on Y’.(( each of which is HS-dominated by some Il’l/a... [2] H.Bz. Stochastic processes.(~)=R. a tbounded linear operator on T such that xi I/ Ueill * < co. we have the following. Then the family nk E Z. A tHilbert-Schmidt operator U is. TT t))d@(x).*. then (T. Princeton Univ. k E N} satisfies the Pn. 11. let @)n1. 7) is called a tnuclear space.. +.. for a Hilbert-Schmidt operator U. if we are given a family of finitedimensional probability measures {Qs} which satisfies the consistency condition (6).} c A. An introduction to probability theory and its applications II.*. Mathematical methods of statistics. 11. Springer. i.* be the topological dual of (T. Kolmogorov’s extension theorem is generalized to the case where the component spaces are tstandard measurable spaces (270 Measure Theory) instead of R’. : R S2-+RS~ is the natural projection.*)) such that Q*( U. EE ds. Bochner’s theorem does not necessarily hold in infinite-dimensional spaces.. Cramer. Probability Measures measure @ on (RT.23jT) such that Qs(E)=@(rr~‘(E)). Doob. [S] W. For example. which is called the Sazonov topology. Press. Kolmogorov.c(E A}. ~(0) = 1. where rcSis the natural tprojection Q: RT-+RS. by Kolmogorov’s extension theorem. third edition. 7) is a nuclear space. 1950. 1933. Wiley. for any finite S c T. and also to the case where product spaces are replaced by tprojective systems [ 193. the set of all r-continuous real valued linear functionals on T). and continuous with respect to the Sazonov topology. Probability theory.11) be an infinite-dimensional tHilbert space and q(t) = exp( . UER’. This theorem is useful in constructing tstochastic processes. then Kolmogorov’s extension theorem [ 1) asserts that there exists a unique probability measure CDon (RT. For example. ~(0) = 1.. Then the topology l(7) coincides with the topology induced by the system of seminorms 11.B(T. by definition. where U are Hilbert-Schmidt operators. 1937. EE 23’. determines a probability measure on RZ. 1963.II.-. SUp{(~i Ileill’)“‘. M. Levy. we can define a fmitedimensional tmarginal distribution Qs for any finite subset S of T by (D. and continuous with respect to the topology I(z) [23]. (c T) are finite and if EE @I.. (we(RZ. c S.270 Measure Theory) and T.(E)=@(x. J. References [ 1] A. 1946. L. . N. {ei}:aorthonormal} < co. let (T. s. and p(O)= 1. define where A~. For a probability measure 0 on (T. As special cases of the foregoing theorem. tET.//c by Iltllu= IIUtll.n2. then every positive definite r-continuous functional cp with ~(0) = 1 is the characteristic functional of a probability measure on T.11) be a Hilbert space. if and only if cp is positive definite. T. then d( T.. Conversely. Define a new topology I(7) of T by all Hilbertian seminorms (1.x(t)<a}. 1966. Bochner’s theorem is generalized to infinite-dimensional spaces as follows. Van Nostrand. Let T.. b( T. Wiley.‘(E)).. by any tcomplete orthonormal system {ei} (this quantity does not depend on the choice of {ei}). If (T. r = 11.* (Minlos [24]). neZ. Theorie de l’addition des variables aleatoires..*. which is denoted by CD’.. Loeve. teT. fe 9. C(EA.(O).).n. bT). Taz) = 1 for some sequence {cc. Feller. jj “. The measures {Qs} satisfy the following consistency condition: If S. Thus X.* is the topological dual of (7’.*) of T. where @* is the touter measure (.* (Sazonov [25]).n2 . Define a seminorm Il. Define a Bore1 structure which is called the characteristic functional of @ A functional cp on T is the characteristic functional of a probability measure @ on (T. 1953.cDZ)). 7) (i. [6] P. The probability measure on Y’ with the characteristic functional exp( -jFm lf(s)l’ds).

II.) [26] M. 4th Berkeley Symp. Minlos. (Original in Russian. Tamarkin. 22 l-223. for example. Harmonic analysis and the theory of probability. In this work. D. Limit distributions for sums of independent random variables. 3 (1962). Nelson. Theory of Prob. Ya. Lukacs.403-419. M. whose theory was established at about the same time [ 11. Ann. tcombinations. 1955. are defined on (Q 23. Regular probability measures on function space.). Bayes. Ann. Theory of Prob. P. Parthasarathy. Appl. Princeton Univ. L. The definition of a priori probaoility due to Laplace provoked a great deal of argument when it was applied. [23] A. 1964.. independence. Stat. Math.. Press. Finally.. P) in terms of measure theory. though similar considerations had been made before him for special probllzms. Characteristic functions. Linnik. Statist. 1958. L&y. Billingsley. Academic Press.. (Original in Russian. de Buffon. A. A. Selected Transl. California Publ. Surveys. Sazonov. Kolmogorov. Wiley. where R is the set of all possible outcomes of the phenomenon. 155-214. A. Univ. Prob. S.. the main concern of mathematicians lies not in the intuitive or practical meaning of probability but in the logical setup governing probability. (2) 69 (1959). Ever since probability theory was given . V. Since the 19th century. V. V. [ 131 P. Vilenkin. L. de Fermat in the 17th century. Academic Press. Widder. A remark on characteristic functionals. Appl. [9] Yu. T. Math. V. Calcul des probabilitCs. Univ. The problem of moments. Addison-Wesley. 1960. 1960.) [22] Yu. Generalized functions IV. and Prob. N. [ 161 D. [2l] 1. 1956. 1201 E. Laplace not only systematized but also greatly extended previous important results by introducing new methods. A note on the papers of R. This elementary theory of probability was later enriched by the work of scholars such as Jakob Bernoulli [2]. (Original in Russian. Wiley. Gel’fand and N. 291-3 13. For example. Kolmogorov. 3 (1958). I (1956). P). [S] E. GauthierVillars. of California Press. From this viewpoint the mathematical model of a random phenomenon is given by a probability measure space (Q23. 1943. [ 171 B. Sazonov.. de Moivre [3]. 1961. R. such as the use of tdifference equations and tgenerating functions. Shohat and J. etc. 2 (1957). N. [ 151 J. 4 (I 959). [I91 S. A. Pascal and P. 207-236. Laplace completed the classical theory of probability in his book Thhrie analytique des prohahilitds (I 8 12). The method of characteristic functionals. Nowa’rlays. Theory of Prob. the theory of probability has been extensively applied to the natural sciences and even to the social sciences. Interscience. (Original in Russian. Stat. 1959. 1959. Daniel Bernoulli. second edition. Convergence of random processes and limit theorems in probability theory.. Laplace transform. Generalized random processes and their extension to a measure. R.342 A Probability 1266 Theory [7] P. 523-531. Probability measures on metric spaces. However. An introduction to probability theory and its applications I. Probability and related topics in physical sciences. [ 121 K. 1957. Math. V. [ 141 W. Feller. Prokhorov. History The origin of the theory of probability goes back to the mathematical problems connected with dice throwing that were discussed in letters exchanged by B. Oliver & Boyd. 342 (XVII. 1961. 1954. Math. Gnedenko and A. of California Press. these arguments are concerned with philosophical rather than mathematical aspects. Unimodality of infinitely divisible distribution functions of class L. 630-643. Proc. All probabilistic concepts. Prokhorov. Prob. 1967. 6 (I 978). (Original in Russian.1) Probability Theory 1 A. Hafner. Minlos and V. 188-192. Lagrange. Wiener [S]. (Original in Russian. Kolmogorov [6]. Such a measure-theoretic basis of probability theory is due to A. Legendre. and J. Convergence in distribution of stochastic processes. (Original in Russian. Applications of harmonic analysis. [I I] L. Convergence of probability measures. and tbinomial coefficients. Sot.. LeCam. M. 1925. A. Univ.) [IO] Yu. which is a mathematical model of mass phenomena [S]. These problems were concerned primarily with concepts such as tpermutations. Bore1 concerning the strong law of large numbers [“I and in the rigorous definition of Brownian motion by N. Math.) [I81 M. Kac. 1964.) [25] V. von Mises advocated an empirical theory of probability based on the notion of Kollektiv (collective). Bochner. and %3is a o-algebra consisting of all sets E for which P(E) is defined. Yamazato. 1949.. Amer. Amer. Appl. such as random variables. P(E) represents the probability that an outcome belonging to E be realized. Sot. 1941. 1959. Math. Decomposition of probability distributions. 1968.) [24] R. in the work of E.

For a sequence {E. and tstatistical mechanics.1267 342 B Probability Theory solid foundations by Kolmogorov. Because of the obvious one-to-one correspondence between measurable events and bmeasurable sets (i. In applications a stochastic process is used as the mathematical model of a random phenomenon varying with time. ) of events. .2.) = 0.e. intersection or product event) of E. A family {b.} (I E A) is independent if every finite subfamily is independent..... for almost all w. (Ieh) are mutually independent or that the family {E. tstationary processes. is the set U.). P) is called a probability space. . impossible Given an infinite family {E. the event E is said to occur almost certainly (almost surely (abbrev.. We assume that an event is always a measurable one.. The complementary event (resp.E. . The triple (a.s. N) of events. we say that the events E. we have 0~ P(E) < 1 for any event E.flEi. we say that the events E. Z” P(E. the family {E. (P2) for every sequence {E.) = 1. a number of sufficient conditions for depen- .. which is concerned with the evaluation of P(lim sup” E. . .) occur. If E is an event and E is the b-measurable set corresponding to E.} (n = 1. . . . (n = 1.). . . we have P(E)= f P(E. Given a finite sequence {E. and tGaussian processes.IfEflF=@.. II=1 This property is called the additivity of probability.)=fiP(E.} (LoA) of events. . ~ event. and (ii) if the events E. whole space Q). .). there are stochastic processes which can be defined by tstochastic differential equations. and lim inf.2. Probability Spaces Let fi be an tabstract space and 23 be a toalgebra of subsets of R. the correspondence of each event E with the set E of all sample points w satisfying E). where n.2. and P(n) = 1.). ) are mutually independent or not. The superior limit event (inferior limit event) is the set of all w for which infinitely many events among E.} (A E A) of events is independent. E. depends on w in general. Therefore P(liminf. Moreover.E. i.) < co implies that P(lim sup” E. A probability measure (or probability distribution) over R(B) is a set function P(E) defined for EE B and satisfying the following conditions: (Pl) P(E) 20. 8.. . E. For a finite or infinite family {E.2. In the same way as functions are often defined by differential equations. the sum event (resp. it has made tremendous progress. P(E. we have (i) whether the events E.thenwesaythatE and F are mutually exclusive or that they are exclusive events. a.2.e. or with probability 1). which correspond to functions in analysis. E.2. j=l B. if {E. The Borel-Cantelli lemma. .} (n = 1. . tMarkov processes and tMarkov chains.2. occur for all n after some number n.} (n = 1. p ( UE” ” =CW. . ) are mutually independent. the probability of the event E.) is the probability that the events E.. reads as follows: Given a sequence {E.} (AE A). We say that a condition E(W) involving a generic sample point w is an event. it is called a measurable event or random event if the set E of all sample points satisfying E(W) belongs to 23. The space R (resp.) = cc imples that P(lim sup” E. ) of pairwise disjoint sets in 8.2. The following types of stochastic processes have been investigated extensively: tadditive processes. occur. P(0) = 0. tstochastic filtering. N) is independent if every subsequence satisfies P(EilflEi2fl. . The concept of independence of events can be generalized to a family {Bk} (noA) of o-subalgebras of b as follows. (n = 1. . . space of elementary events. E 23.} (n = 1. sure event) is the complementary set EC (empty set 0.} (n = 1.E.. . . the sets lim sup.) is the probability that in& nitely many events among E. ) is a sequence of pairwise exclusive events and E is the sum event of E. and P(lim sup.2. The tergodic theory that originated in statistical mechanics is now regarded as an important branch of probability theory closely related to the theory of stationary processes. each element w of a) is called the basic space. or sample space (resp.. Frequently.(n. .. applications of part (ii) are greatly hampered by the requirement of independence. Z. . respectively. . a b-measurable set itself is frequently called an event. ) of events. (all events except finitely many E. we call P(E) or Pr(s) the probability that the event E occurs.} (n = 1. n > (P3) P(a) = 1.) N) are mutually independent or that the sequence {E. tBrownian motion and tbranching processes are important special stochastic processes.S. since we encounter only measurable events in the theory of probability. The theory of stochastic processes and stochastic differential equations can be applied to tstochastic control. are called the superior limit event and inferior limit event.). If P(E) = 1. tmartingales. sample point or elementary event). By the definition of P. (n = 1. in particular.} (1 E A) of o-subalgebras of events is said to be independent if for every choice of E.. The most important concept in today’s probability theory is that of tstochastic processes.

} (3. . . is called the mean. . If X. N) is independent. N) (X..E( Y))) is called the covariance of X and Y. Given an infinite family {X. V(X + Y) = V(X) + V(Y) for mutually independent random variables X and Y. The latter definition of independence of random variables is compatible with the previous definition of independence of c-subalgebras of d... that is.23. are random variables. .. .} (A E A) of random variables. . we say that the random variables X.} (1 E A)) is defined similarly. . (n = 1. If the X.. F(u) = 0. (n = 1. Random Variables Let (0.} (n = 1. an n-dimensional random variable X =(X1 . Since a random variable X is a 2% measurable function... but the converse is false in general. then the independence of the family {X.. is measurable is called the (ralgebra generated by {X.andom variables. . .. we can speak of the tintegral of X relative to the measure P on 8. and that.... A).) (n = 1. N).} (1. C. Each element of this class is said to be measurable with respect to the family {X.. When X and Y have finite variances.dimensional random variables (n = 1. . .. E A) of r. . On the other hand.2..] denotes the a-subalgebras of 23 generated by the sets {w I X. Y) d 1 in all cases. Given a family {X. (n = 1. The standard deviation of X is the nonnegative square root a(X) of the variance. the value E((XE(X))( Y.)-dimensional random variable X=(X. the k. is called the marginal distribution of the /-dimensional distribution @.e. the set {w 1X(w) < u} is in 23).(WIC A. or expected value of X.(o) :Gun}).-dimensional distribution Qn of X. denoted also by M(X) or m.] } (1 E A) in the previous sense. X.. e)-valued random variable if it is measurable.} (1eA) and is denoted by 23 [X. i.{ (3.342 C Probability 1268 Theory pendent or that the sequence {X.e. .. E(X.2.E(X))Z) is called the variance of X. . N) are mutually inde- It follows that E(aX + b Y) = uE(X) + bE( Y). If X is integrable relative to P. . The Chung-Erdik theorem [9] is quite useful in this connection.-) dimensional random variables.. dent events to have the same conclusion as (ii) have been discovered. if the relation P({~IX.(k.(w)~a.(~)EA. . F(u) = 1. b.2.. lim. N) of random variables. It also follows from the definition that -1 :C p(X. the integral of . V(aX + b) = u2 V(X) for any real numbers u. N). (n= 1. C5)is called an (S. (n = 1. the smallest o-algebra with respect to which every X.2.. “‘> N) ({X. . the independence of the family {X.2. P) be a probability space.X over A is denoted by E(X.-. in particular. Let ‘23’ be the o-algebra of all +Borel subsets of the real line R. X. X.E A) in the latter sense is equivalent to the independence of the family {b [X. A random variable is a real-valued function X defined on R that is %-measurable (i.)=P({wIX. . I i E A].E(X))’ is integrable. X. The point function F defined by W=P({~IX(4d~}). for every set A of e. UER. we say that the random variables are mutually independent or that the family is independent if every finite subfamily is independent.) is the joint random variable of X. V(X) = E((X . E(X Y) := E(X)E( Y). N). B1) such that The measure @ is called the (l-dimensional) probability distribution of the random variable X or simply the distribution of X. N) and that the (I-dimensional) distribution @ of X is the joint distribution (or simultaneous distribution) of X. X. with A. u. a mapping X from (Q. we say that the I( = Cr=. .2. if b [X. Q). the mapping X =(X1.. the set {w I X((U)E A} belongs to d.N)}) holds for every choice of 1 -dimensional +Borel sets A.A)) are k. k. the correlation coefficient of X and Y is defined by E((X-E(X))(Y-E(Y))) p(x’ Y)={E((X-E(X))2)E((Y--.2. X.) for l-dimensional Bore1 sets in equation (1). denoted by a’(X).2. . Given a finite sequence {Xn} (n = 1.E(Y))z)}“2. Y)=O. for every real number a. Then each random variable X induces a probability measure @ on (R.) from s1 into R” is said to be an n-dimensional (or R”-valued) random variable.~.) induces its n-dimensional probability distribution (or simply n-dimensional distribution) and its n-dimensional distribution function F(u..-) dimensional Bore1 sets A. is a monotone nondecreasing and right continuous function such that lim.(k. X. it is enough to take k.2.2.(~=~. (A..... N) are k. . (n= 1. expectation. . an arbitrary l-dimensional Bore1 set. The independence of X and Y implies that p(X. If (X . X. If X and Y are two random variables for which E((X E(X))(Y-E(Y))) exists. 23) into another imeasurable space (S. The variance is important because of the well-known Chebyshev inequal- . The function F is called the cumulative distribution function (or simply the distribution function) of the random variable X.} (n E A) of random variables. More generally. usually denoted by E(X). If the X.}. Similarly. .

These values are denoted respectively by P(E I F) and P(E I F’)... P. 1”)= 0. Convergence of Random Variables If P(lim.=i:X. The same fact holds when Y is a multidimensional random variable.e. we can prove that there exists a nice version of P( YE E 1%. almost sure convergence does not in general imply convergence in the mean.. = X. independent is convergent almost surely if and only if there exists a sequence of independent random variables Xi. B[X] and 5 are mutually indepent. ) are mutually independent. The definition of P(E 1 Y) or P(E ( Y = y) is also the same as in the case of the conditional expectation. It follows from the definition that the conditional expectation has the following properties. Lkvy [lo] proved that if the X. X.s. (vii) if 8 is then lim. For example. However. . the sequence {X.e. . co). X2. Let 3 be a o-subalgebra of 23 and Y a real random variable. n is convergent. a o-subalgebra of ‘& then E(E(X I 5) I 6) = E(X 18). In this case.} is said to converge in probability to X. X$. does not always satisfy the conditions of a probability measure. we also write E(X I Y) for E(X I 8) and call it the conditional expectation of X relative to Y.Y)‘).If X is a random variable . (vi) if lim.. (n = 1. P( YE E I 5) or P( Y-‘(E) 15) is the conditional probability of the occurrence of the event YE E under &. When 5 is generated by a random variable Y.. an arbitrary version of P( YE E 1g). X. . 5) a completely additive set function which is tabsolutely continuous with respect to P. . Conditional Expectation Probability and Conditional Let (Q. 8. random variables converging in distribution may even be defined on different probability spaces. by the tRadonNikodym theorem.1269 342 E Probability variable with finite vari- Theory ity: If X is a random ance e2. On the other hand.. then E(X I 3) = E(X)..-X. According to the foregoing definition...2... = X. if E( 1X. almost surely (a. X. When X is the indicator function (i. (n = 1. However. E s defines on (Q. and convergence in probability implies convergence in distribution. In particular.. The famous three-series theorem of Khinchin and Kolmogorov [t l] claims that the series x.. F’. .1 < Y and Y is an integrable random variable. there is a tBore1 measurable function f such that E(XI Y)=f(Y(w)).2. . P(IX-E(X)12c)<02/c2 for every positive with finite mean. @. . viewed as a function of E.. (viii) if X2 is integrable and Y is any @measurable random variable. with X.. the sequence {Xn} is said to converge in distribution to X... E(xE I 5) is called the conditional probability of E relative to 3 and is denoted by P(E ( 5). the icharacteristic function) xE of a set E in b. denoted by E(X 13). E(X. . P(E 15) is the simple function which takes the values P(E fl F)/P(F) on F and P(E n FC)/P(FC) on 8”.. either almost sure convergence or convergence in the mean implies convergence in probability. if the ranlim. . Finally.I>&)=O for every positive number E. if 5 = {F.. then E((X JW IS))‘) G Et@‘. Note that the sequence of random variables converging in distribution may not converge in any ordinary sense..). D. (iv) if X and 8 are mutually independent..(iii) E(E(X 13)) = E(X).)which is a probaf(w)dP for every EEG. . Therefore. dom variables X. . 1 EK)> ” 1 w.. such that each of the three series c PGL z X).) = 1. the function p(E)= X(w)dP.) n E.} is said to converge almost everywhere (almost certainly. EE5. I 5) = E(X. (n = 1.. number c. “=I k=l. the sequence {X. . and we write E(XI Y= y) for f(y).. the sequence r.2. P) be a probability space and 5 a r~subalgebra of %3. then E(X 1%) 2 @(ii) E(aX+bY1~)=aE(XJ~)+bE(YJ~).Iflim. up to a set of P-measure zero: (i) if X > 0. On one hand. a} with 1 > P(F) >O. 1%).P(IX. Since P( YE E I 8) is determined except on a P-null set depending on E.. there is an g-measurable function f such that p(E)= sE This function is unique up to a set of Pmeasure zero and is called the conditional expectation (or conditional mean) of X relative to 5. i. .2. (v) if X is $-measurable. then E(XI 8)=X and E(XYI 5) = XE( Y I 5). or with probability l)toX. is convergent almost everywhere if and only if it is convergent in distribution (or in probability). For a given positive number p the sequence {X”} is said to converge in the mean of order p to X. and if m lim f(x) d@“(X) = ~mfwwx) O2 n-m s --oo s for every continuous function f with compact support..X. co) have distributions @. respectively. with IX.

X. 1937. English translation. . J. 27 (1909). ThCorie analytique des probabilitts. [lo] P. L. X. [S] R. Differential space. When X is a random variable subject to the distribution with continuous probability density f(x). [3] A.. We then apply Bayes’s formula to reevaluate the probability of each hypothesis Ei knowing that some event E has occurred as the result of a trial. [Z] J. and we often set P(E. 1933. A history of the mathematical theory of probability from the time of Pascal to that of Laplace. ThCorie de I’addition des variables alkatoires. Suppose that the probabilities on the right-hand side of the formula are given. Bayes’s Formula Let E. called the tail oalgebra of { &}. [7] E. Borel. > 0 for infinitely many n’s is a symmetric event. S. =0} is {Lx”+l. Zero-One Laws In probability theory there are many theorems claiming that an event with certain properties has probability 0 or 1. sequence of independent o-subalgebras of 23. 72 (1952). for every n. a tail event.186. a is called a symmetric event concerning {X”} if occurrence or nonoccurrence of c( is invariant under every finite permutation of X. L. 179.. Kolmogorov’s zero-one law: Every tail event concerning a sequence of independent random variables has probability 0 or 1. Kolmogorov (Kolmogoroff ). . This version is called a regular conditional probability of YE E under 3 or the conditional probability distribution of Y under 5. this is written as P. ErdGs. Wiener. . I El = P(EJP(E I EJ P(E. {lim.. . Then the oalgebra 2 = nk Un.)+. although this has caused a great deal of criticism. Amer. Ars conjectandi. Vorlesungen au. .. This is why P(E.. Palermo.342 F Probability 1270 Theory [6] and the Hewitt-Savage zero-one law. Springer. . Hewitt-Savage zero-one law: Every symmetric event concerning a sequence of independent and identically distributed random variables has probability 0 or 1.) (P(Ei I E)) is called the a priori (a posteriori) probability. Trans. Laplace. 1949). be pairwise exclusive events.)P(E~E. Math. fW W IX =x01 ..(E) X). Kolmogorov’s zero-one law is a special case where ‘&. 1953. [4] P. X. Bayes’s formula is extended to the following form: References [l] I.f(xo I El = jZmP(EIX=x)f(x)dx’ where f(x 1E) is the conditional probability density of the random variable X under the assumption that the event E has occurred. Macmillan. Kolmogorov’s zero-one law can be extended as follows: Let g. . n= 1. . On the application of the Borel-Cantelli lemma. . P( YE E 1X=x). Erg. is the o-algebra generated by X. Kolmogorov’s zero-one law . Paris.. E. 247-271 [S] N. bility measure in EEL’ for every weR and that such a version is unique almost surely. Bernoulli... occurrence or nonoccurrence of CI depends only on }. P(A) = 0 or 1 for every A E 2. 2 (1923). Wahrscheinlichkeitsrechnung und ihre Anwendung in der Statistik und theoretischen Physik.2. [9] K. Sot. Chelsea.. G. Math. . Grundbefriffe der Wahrscheinlichkeitsrechnung.. Rend. be a.(E 1X=x) are interpreted similarly. For example. L&y.(E 13). we have P(E. F. Chung and P. 1713. E. the determination of the values of a priori probabilities is sometimes difficult.. and P. is trivial. Mat. For example.s dem Gebiete der angewandten Mathematik I..} if for every n. Les probabilitks dCnombrables et leur application arithmetiques.) = l/n in practical applications. E. 1812.. Wiley. . 1865 (Chelsea. Foundations of the theory of probability. Phys. However.e. .)’ where P(E. we mention two famous examples. 1931.k s. Here. CI is called a tail event concerning {X. i. ) be an event concerning a sequence of random variables {X”}. . Circ. the event that xi+ X. Doob.)P(E~E.. Franz Deuticke. The conditional probability distribution can be defined not only for real random variables but also for every random variable which takes values in an tanalytic measurable space. Stochastic processes. Such theorems are called zero-one laws. and assume that one of them must occur. usually represent n unknown hypotheses. von Mises.. In practical applications E. This is called Bayes’s formula.. de Moivre.. If E is another random event. The doctrine of chances. Gauthier-Villars. Todhunter. 131-174.) is the probability of the event Ei and P(E 1EJ is the conditional probability of E under the assumption that the event Ei has occurred. Math.+P(E. 1718. Let a = cc(X. 1950. and P(E I X =x0) is the conditional probability of E relative to X. [ 1 l] J. [6] A. P. P( YE E 1X).

When the dimension of the space spanned by r + 1 points is r. By convention. and there necessarily exists a P’ that contains r + 1 arbitrary given points in a projective space. An introduction to probability theory and its applications. Given two sets P. Construction of Projective Geometry We construct projective geometry axiomatically [4]. so we can identify every line with a point range. we say that points of M lie in a general position. otherwise they are dependent. 1966. 1968. respectively. and the relation (p. q2.~~} and {po. Then P’U P” is the projective space of the lowest dimension which contains P’ and P”. F}. then we say that the line 1 contains the point p.14) Projective Geometry A. We call P’ a projective line and Pz a projective plane. then S is called a subspace. and distinct points q. We call a projective geometry satisfying axiom (IV) a finite-dimensional projective geometry. 1963. the empty set is a (-1)-dimensional projective space. When two lines 1. If P is of dimension n. Let M. II. N be subspaces of P. Benjamin.P~. Consider sequences of subspaces of the type P 2 Pnml $ . The number n of the longest sequence is called the dimension of P.1271 343 B Projective Geometry [12] M. 343 (VI. we write P” instead of P. I. and for a pair of points PE M. Van Nostrand.1)-dimensional subspace in P” a byperplane. and we call it the set spanned by M and N. Wiley. 1)~ F holds for a point p and a line 1. q E N consider the set p U q of all points on the line that contains p and q. Each subspace S of P. Lines and points are projective spaces of dimensions 1 and 0. q E N} is denoted by MU N. q2 necessarily intersect (Fig. Lo&e. we say that these points are independent. [14] L. Feller. In this case a line 1E Q is represented as a subset of P. If any r+ 1 points of a given subset M of P” are independent for each r < n. there exists a one-to-one correspondence between the set of lines and the set of point ranges. which from this point on will be the sole object of our consideration. 1966. and pz. we say that they intersect at p. p2 and the line containing q. gives a finite-dimensional projective geometry. Then the line containing pl. if we denote the intersection of P’ and P” by P’n P”. Addison-Wesley. second edition. together with the set of lines of P contained in S. Q and a trelation I c P x Q. these points are said to be collinear. Probability. We call P a projective space. 1). pz. Q. 1957. Points and lines are subspaces. and (III) are called general projective geometry and projective geometry. respectively. (II) Suppose that we are given noncollinear points po. Lamperti. [IS] J. q2} are collinear triples. The set { p U q 1p E M. + =P 1 = z PO # 0. Let S be a subset of P and p. and when several lines contain the same point. The space P’ necessarily contains r + 1 independent points. pz be any two distinct points of S. 1)~ I means that the point p belongs to the set 1. The set of all points that are contained in a line is called the point range with the line as its base. Introduction Projective geometry is the most fundamental of classical geometries and one of the first examples of axiomatized mathematics. Fig. Now we impose the following axiom: (IV) There exist a finite number of points such that any subspace that contains all of them contains P. (II). We call P’U P” and P’ n PS the join and the intersection of P’ and P”. Breiman. we put 0 U M = M and p U p = p. Now suppose that {P~. If (p. 1 (III) Every line contains at least three distinct points. and 1. Cl33 W.. For ‘p we impose the following axioms: (I) There exists one and only one line that contains two given distinct points. it is unique if the points are indepen- . On the other hand. respectively. In projective geometry.. Probability. We call each element of P a point and each element of Q a line. these lines are said to be concurrent. and so S is a projective space. When several points are contained in the same line. consider the triple !@= {P. If the line that contains p1 and p2 is always contained in S. B. . contain a point p. pl. By convention. third edition.. M. The ‘$3 that satisfy axioms (I) and (II) and axioms (I). Probability theory. where 0 is the empty set. We call each 2-dimensional subspace a plane and each (n . . then it is the projective space of highest dimension that is contained in both of them.

gz4. onto P$ from P$-rm’.. However. we can construct a fundamental figure c’ on P2. The projective space PG obtained by the principle of duality. Now for arbitrary subspaces P. we call the figure that consists of these four points and the six straight lines gij = p.~. P.Uq. thus obtained is called a perspective mapping. Under the assumption that P’ and P” do not have points in common.). If six points qi (I d i < 6) on a line I are points of intersection of six sides gi2.4r g34. and then cutting it by P2. we take Ponrm’ that C. (pl U p2) n (ql U q2) are collinear.Zr are dual to each other. g23 of a complete quadrangle with I. ea’sh pi is called a vertex. Assuming that P’ and P” have points in common the operation of constructing P’n P” from P’ and P” is called cutting P” by P’. If for two fundamental figures Z and c’ there exist a finite number of fundamental figures F. In this case. By projecting Z from P. or more generally. Each set that consists of the totality of subspaces of an arbitrary demension in the same P’ or a subset of it is called a P’figure. each of which is independent and fies pi # q. of all hyperplanes that contain a Porn2 in P” is called a pencil of hyperplanes. 2). 4 92% 9. when n = 2.343 c Projective 1272 Geometry have no points in common with ttem and project each point of P. we say that C and L” are in perspective and denote the relation by C X c’. and a fundamental figure Z in the space P. q3.3). that points satisU qi Here we introduce projective coordinates Consider Desargues’s theorem: Suppose p.U~. p2. 2 points(~.). The proposition or the figure thus obtained is said to be dual to the original one. each pencil of hyperplanes of a subspace of an arbitrary dimension in P”. there exist projective geometries for which it does not hold. the operation of constructing P’U P” from P’ and P” is called projecting P” from P’. we call P? P4 P3 AIYi!kL 91 q1 Fig. the center of projection (Fig.. too. The set C. it is called a pencil of lines and a pencil of planes. U pj (1 :. The converse is also true.+P. i d j < 4) a complete quadrangle p. (i = 1. then the three Fig. These mappings are extended to those of fundamental figures. and P’ and . If a one-to-one correspondence P.J is represented as the composite of a finite number of perspective mappings. so we assume Desargues’s theorem for n=2. gi3. is called the dual space of P”.2. When four points pi (1 < i < 4) in P” lie on the same plane and in general position. and Pz. If P’U P” = P’ and P’ f’ P‘= P”.U~. Prier+‘.. In projective geometry. by regarding the hyperplanes of P” as points of Pi.mrml. We call the latter the dimension theorem (or intersection theorem) of projective geometry.p2p3p4. g.~F. Projective Coordinates in P”. This is assured because propositions dual to axioms (I))(IV) hold. then r + s = t + u.~C’. p3 and ql. (1 < i < I) such that C. (0 < r < n).thenwesaythatCandZ’are projectively related to each other and denote this by C. then the pencil is determined uniquely by these two. . Each pencil of hyperplanes of P”. In P”. In such cases it is impossible to introduce coordinates. This operation is called projection of Z from PO onto P2. dent.. P. and the Pzm2 common to them is called the center of C.(~. respectively. we interchange P’ and Pnmrml (O<r < n) and also interchange contains and is contained (and related terms). F.3) are concurrent. If the three lines pi (i = 1.. 1 . q. 9. then its dual proposition is also true (duality principle). 3). are two sets of in P”. is called a linear fundamental figure of P” or simply a fundamental figure. and each gij is called a side.)n(q.Uq. The one-to-one correspondence PC-P.2. This theorem holds for n > 3 generally. Pn--’ that contain the same Ponrml is called the star with center P. Suppose that in a proposition or a figure in P”.. When n = 2 and 3. If a pencil of hyperplanes contains two distinct hyperplanes of P”.)n(q. if a proposition is true.. . the set Zr of all Prier. then we call it a projective mapping. we call these non-Desarguesian geometries. and we call P. Suppose that we are given spaces P. 3 Fig.c’ (Fig.

. pm. we put pa = A n (p U A*) and call it the component of p on A. . E K) with respect to variable coordinates z”. they are called coordinates of a line.. we shall omit u. pm. When we are given a fixed triple [p. the system {f&.(O~cr. is excluded from K(p. O} a projective coordinate system of P”.pm. ALEK are parameters. for brevity.P. . a. we can show that if there are given three fixed distinct points on a line 1.} by 0 and call { 5. . Now we represent the point whose coordinates are (x0. .... X. . x1 E K and x1(x0)-l = 5 homogeneous coordinates of p.. . . pl). 5 Fig. each of a. pi. When p is contained in A.. U a. More generally. pm. For any point p of P” not contained in A. s constitute a quadrangular set of six points and call s the sum of px and py with respect to [po. In particular. X 1.t such that p. the set of points on 1 not equal to pm is called a point range of the number system. We call the field a Staudt algebra. . . 6). . pm.Ua. For each point p on 1 we call the element 5 = B(p) of K the inhomoge- neous coordinate of p with respect to this frame. pY different from pm on I. uA] is a frame of A. . For A=a. The elements of the ordered set (<I.~n). . . p. Therefore each hyperplane is uniquely determined by the ratio of X0. and an abstract algebra isomorphic to it is called a coefficient field of P”.. Fig.90 Coordinates B. provided that we exclude pa.<cr. Then & = [aEO. when coordinates are introduced.PJ a frame on a line 1 of P” with coefhcient field K. . .pI] (Fig. .x”) such that xi(xo)-i = r’ are called the homogeneous coordinates of p. it is necessary and sufficient that there exist an element 1# 0 of K such that ya = x”l (c(= 0. and those of the set (x0. xi. .. pm. then any pair of distinct points on 1 determines uniquely a point on I such that the six points thus obtained constitute a quadrangular set.~U. a=. In this case we denote {Q...p. pJ a frame (or projective frame) of I. xi) such that x0.. py.. We call the set of three points [po. X. . . .. .P. For any two points px... x”) as homogeneous coordinates of p with respect to 5. . 6 A point range of the number system constitutes a tlield (which may be noncommutative) with respect to the previously defined sum and product. . On a line 1 we fix three mutually distinct points p.l).1273 343 c Projective Geometry them a quadrangular set of six points (Fig. A projective mapping of 1 onto itself that leaves invariant each of three distinct points po.Ls (0 <CI i n. U . <‘. we call the pair (x0. In order for (x0. when 1. pJ on a line I. and we put 5’ = O(pi). a point z is contained in the space spanned by Y + 1 independent points xg (0 < b < r) in P” if and only if z” = &..1 . a necessary and sufftcient condition for points z to be on the line that passes through two distinct points x and y is that za = x”A + yap (0 < cz < n).). t constitute a quadrangular set of six points is called the product of px and py with respect to [po. a. If n = 2.. we denote by A* the space spanned by the remaining fundamental points. u] of ordered n + 2 points in a general position is called a frame (or projective frame) of P”. pm. Denoting by CP~. Hereafter. pl.. Under a certain condition. Suppose that isomorphisms &a : K(a=. = a. and pm the supporting point. 5).} is determined by one of the &. x”) are homogeneous with respect to sAO.pm. y’) to be homogeneous coordinates of the same point. pJ (Fig. p. In P”. .P~... . hyperplane coordinates of the hyperplane. 4). the equation of a hyperplane is represented in the form ~~=o X. we fix (0. Is E K).) the Staudt algebra that is determined by a frame [p..< n). the origin.x1. p.za = 0 (X. We call X0. . The quadrangular property is invariant under projective mappings. Since the supporting point p. x.. X. For any point p that is not contained in A*.. . x”) simply by x.. . and if n = 3. . we denote by pi the component of p on a.. x’) and (y”. we call each isomorphism O:K(pe. and we call p.. . the point.. x1) such that x1 #O as the homogeneous coordinates of P. In conformity with these results.<. we take the point s such that pm. U ai (1~ i . plane coordinates of a plane.. We denote by K(p. p. p1 the unit point. . (0 <CL <n) is called a fundamental point... a+K are assigned for each pair CI. px. fi (0 < c(< /l< n). 5”) are called the inhomogeneous coordinates of p with respect to 3. .pi)+K a coordinate system of 1. x1 . from the point range.. p. and u is called a unit point. provided that coordinates of p (x1 .. . we now introduce coordinates in P”. as before. we define (0. On the other hand. Also.pm. By Desargues’s theorem. p1 on 1 is necessarily an tinner automorphism of the held K(P~. pm.p. . .]. (For coordinates of P’ in P” . A set 5 = [a.

he origin. We call proposition (2) the fundamental theorem of projective geometry and proposition (3) the theorem of Pappus. respectively. and I. Then cp induces a one-to-one correspondence between the set of r-dimensional subspaces of P” and the set of r-dimensional subspaces of P”. Utilizing such isomorphisms. p2.)n(. then 3.). any other correlation is obtained as a composite of z0 and a collineation.P31=:> A cPI>P3. A collineation necessarily a projective transformation. If 7 o 7 is an identity. (3) Given two distinct lines I. If the coeffkient field is the real number field.. is a projective mapping.3. then the three points (p2 U q3) fl(p3 J q2). we consider a frame such that pl.)-+K for the Staudt algebra K(p. The totality of projective transformations of P” constitutes a tnormal subgroup of a(P”). For the complex number field. Next. cp(pi) (i = 1. q. Now. The totality of collineations of P” constitutes a ttransformation group and is called the group of collineations of P”. we denote it by K(P”). p3 that are collinear.. we call 7 an involutive correlation.< r < n . the supporting point. The former is obtained as a composite of a projective transformation and an automorphism of the coeffkient field. can be expressed as cpp’3 CpP”.P4. it naturally induces a mapping Pl+P”..2.Uq. we denote it by O(P”) and call it the group of projective transformations. suppose that we are given two projective spaces P” and P” that are subspaces of a space PN (n < N).p4].P4. pz. p2. then the isomorphism 6 of the Staudt algebra K(q. pJ on a line in a space. x!) (i = 1. Specifically. and p3 are.3) that lie on I. pm. we call cp a collineation. if we assign an isomorphism Q. If we denote the inhomogeneous coordinates of pi with respect to a general frame by (~0.P*~PIl> IP. we can determine homogeneous coordinates in an arbitrary subspace of P” by a frame on it. where p. we call it a projective collineation in the wider sense. there exists a unique projective transformation Moreover. If we denote a correlation by zO. P31 =CP3. These three propositions are mutually equivalent.2. then ~(P”)/B(P”)z’%(K)/~(K).and(~.)n(p. (2) Given frames 5 and 5’ of P”. Then. we call the projective space a real (complex) projective space.3) and qi (i= 1. For four collinear points pi (1 < i < 4) in P”.:K(p. which we also denote by r. If z is a correlation.Uq.. and if P’z P” in P”.4).343 D Projective 1274 Geometry Transformations sending 5 onto 5’. Suppose that cp: P”+ P” is a collineation in the wider sense and 0 . The inhomogeneous coordinate 1 of p4 with respect to this frame is called the anharmonic ratio (cross ratio or double ratio) of these four points and is denoted by [p1. . Suppose that the coefficient field is a commutative field whose characteristic is not 2. and I.7~Uq~)are collinear. p3 are distinct and p4 # pl. if we interchange four points. In classical geometry. it is necessary and suffkient that all automorphisms of the coeffkient field be inner automorphisms. Hence in order for all collineations to be projective transformations. when a collineation in the wider sense cp: Pn-P is a projective mapping.) In this case.P1rP*l =cP4>P3.P*>P41=l-A .2. l. respectively.p. Then zor is a collineation. this is not necessarily true.p3. Projective A one-to-one correspondence cp between the point sets of two projective spaces P” and P” is called a collineation in the wider sense if for any three points pl. It is isomorphic to the group of tinner automorphisms J(K) of the is not coefficient field K of P”. however. The totality of projective transformations that leave invariant a frame 5 of P” constitutes a subgroup B)l. Suppose that the coeffkient field is commutative. (When Desargues’s theorem holds.. If P” = P. if P” = P”. A projective collineation is also called a projective transformation.2. D. contained in a plane in P” and two sets of three distinct points pi (i= 1. we consider the following three propositions: (1) The coefficient field of P” is commutative.3) are also collinear and vice versa. then (~~Uq.J. we call q a correlation. If the coefficient field is the real (complex) number field.p. any two projective spaces of the same dimension can be identified with subspaces of a projective space of higher dimension. and the unit point.p2..1.Pz.+l(s). then collineations are always projective transformations.. if we denote the group of tautomorphisms of K by ‘%(K). then we have the order of the i = CP2rPli P4. ql) on an arbitrary line onto K can be uniquely determined so that 0-l o 8. only these cases were studied.

when K is commutative. then x.-’ there necessarily exist q-dimensional generating spaces. Then the equation of the quadric hypersurface Q. If it is a line we call it a generating line. If the points of intersection of a line passing through a point x with Q”. we extend the definition of correlation and include the case where T. A projective transformation is not defined on its singular subspace. K). Each point on Q. Q: is a truled surface covered by two families . l/2.e. its singular subspace is contained in Qt-‘. In particular. 6(P”)g PGL(n + 1. p4 are said to be harmonically separated from each other. When 1= -1. When I2 -A + 1= 0. which is singular of the first species (i. Each projective transformation X+X is expressed with respect to homogeneous coordinates xa (0 < CI< n) of P” as respect to some frame by (X. we can consider the anharmonic ratio of four hyperplanes of a pencil of hyperplanes. these four points are said to be an equianharmonic range of points. if T= (ti) is a tregular matrix (t. (Here also. Quadric Hypersurfaces PZO.. we say that x and y are mutually conjugate. then the linear transformation is a correlation. If 7* is singular of the hth species. p4 are called harmonic conjugates with respect to pl.(x) the polar of x with respect to Q. We call a subspace contained in Q. p2 and pa.) The condition that z* is an involutive correlation is given by T* = + ‘T. K) with the coefficient field K by its center {pllp~K\{O}}. the group of projective transformations 6( P”) of P” is isomorphic to the factor group PGL(n + 1. = (t$) is not regular. and the converse is also true. z. Extending the definition of projective transformations. with an arbitrary square matrix that is not necessarily regular. The correlation r* is a null system if and only if any point x of P” is contained in the hyperplane r*(x). it is called a singular projective transformation. E. We call points on this singular subspace singular points of Q. and let Q.1)/2 according as II is even or odd. For a polar system r*. A singular projective transformation of the hth species is the composite of the projection of P” onto some Pneh with the singular subspace as its center and a regular projective transformation of Pnmh. det(t. z2 is a harmonic range of points. We call the polar of a point on Q. if the coefficient field is an talgebraically closed field. the involutive correlation r* is called a null system.‘. We call Phml the singular s&space of this transformation. When T is regular. If 7* is regular or singular of the hth species. zl. there are the following two exceptions: when 1= -1. these four points are called a harmonic range of points. For the dual of these. and when 1 is a root of 12--1+ 1 =O. and the points p3. and x the pole of z. When a point x lies on the polar of a point y. y. we call the corresponding quadric hypersurface regular or singular of the hth species. Also.-‘.-‘. we call the transformation represented by (l).-’ a generating space. these six values are different.“) #O. The anharmonic ratio is a quantity that is invariant under projective transformations. for each regular Q.-’ be the totality of points x contained in z*(x). EK).). When T* =‘T*. y. = -IT*. however. Conversely. we call the involutive correlation z* a polar system. Then. and 2. Q”. The concept of the anharmonic ratio can be extended further to the case of four elements of fundamental figures in general. We put q = (n ..1275 343 E Projective Geometry In general. the set of points x that are contained in hyperplanes z*(x) constitutes a quadric hypersurface (or hyperquadric).-’ at that point. n + 1 hyperplanes ZFzo tixa = 0 (0 < a < n) have a space P*-’ in common. So there is a one-to-one correspondence between projective transformations and tequivalence classes of the regular matrices T=(t. If (1) is singular of the hth species.-’ is conjugate with itself. or pl. a projective transformation. that is. If the coordinates of a point are denoted by (x”) and hyperplane coordinates with Let 7* be a polar system. and when T is not regular.-‘. Therefore.+(x) a polarity with respect to Q!-l.) with the tequivalence relation T-~T(IEK\{O}). if the frank of T is n + 1 -h. When T. it is called a regular projective transformation. is called a cone.-l the tangent hyperplane of Q. p2. We call 7.. then (1) determines a projective transformation.’ and 7*(x) are denoted by zl. K) of the tgeneral linear group GL(n+ 1.2)/2 or (n . Qn2-l with just one singular point).-’ is given by (3) For such a correlation 7* we call a relation between the set of points x of P” and the set of hyperplanes z. then we say that the projective transformation is singular of the hth species.(x) with respect to Q.

Q. the aim of projective geometry is to study properties that are invariant under the group of projective transformations. Let K be an arbitrary field. respectively.-i. its tdual lattice is also an n-dimensional projective geometry. l) E F:. this means that the images of I and f are conjugate with respect to Qt. So we may consider projective geometry and irreducible complemented modular lattices as having the same mathematical structure.-’ and Q. Projective Program Geometry and the Erlangen From the standpoint of the tErlangen program of F. we can reconstruct various classical geometries.. consider the projective space P” whose coefftcient field is the real number field. right or left) linear space I”‘+‘(K) over K. Fp(L) = {P. Moreover. If L is an irreducible complemented modular lattice of finite height. and this is the principle of duality. we call Qi a conic. In order that six points pi (1 < i < 6) in a plane lie on Qi. Projective Geometry and Modular Lattices +~01~23~~02~13+~03~12. For an arbitrary natural number n. in this case we have ‘@ % ‘@(L(v)) and L e L@(L)). (~2 U ~3) n(ps U PA> and (~3 U ~4) f(p6 Up. Q. In these theories. the above theory has to be greatly modified.-‘.l that leaves or thee2n-1 invariant is a tnon-Euclidean tconformal geometry according as the transformation space is the set of inner points of Q. We say that each point of Qt is the image of the line corresponding to it in P3. Analytic Geometry Representations of Projective F. we call it a pencil of tonics and a pencil of quadrics. fix an imaginary regular quadric hypersurface Q. Geometrically. then the geometry belonging to the subgroup of (li(P”. Two lines 1 and 7 intersect if and only if (1. The totality of subspaces of each dimension in general projective geometry ‘p constitutes a tcomplete tmodular lattice L(‘p) with respect to the inclusion relation. Then the geometry that belongs to this group is taffine geometry.T) = 0. H. 1) = 0 holds. linear line congruences (linear line complexes) that are families of lines dependent upon two (three) parameters are of great interest. It is the set of all &’ that pass through the intersection of Q. if p < I and (p. Conversely. we put (1 1)=10’t23_102t13+103T12 G. it is necessary and sufficient that the three points (pi Up. suppose that L is a modular lattice with tminimum element @. In particular. We thus obtain Euclidean geometry. then there exists a one-to-one correspondence between the points on the regular quadric hypersurface Q: defined by (1. commutative or noncommutative. if we assign some regular quadric hypersurface Qt-‘. quadric hypersurfaces play a fundamental role. Projective of generating lines. invariant.-’ or the whole Q. tLattices (lattice-ordered sets) and projective geometry are intimately related. I) = 0 and the lines in P3. and denote by P the totality of elements p tprime over @ (i. and fix a hyperplane 17.-’ in P”. When the coefficient field is noncommutative. If we regard these Iij as homogeneous coordinates of P5. Denoting by Iij and Fj (0 < i <j < 3) the tPliicker coordinates of two straight lines 1 and i in P3.343 F 1276 Geometry Utilizing various subgroups of this group. For example.-” in II.)fl (~4 U ~4. Quadric hypersurfaces and sets of lines in P3 are important objects of study in both projective and algebraic geometry. F} is a general projective geometry. then p(L) is a linite-dimensional projective geometry. Let G(P”) be the subgroup of projective transformations formed by all projective transformations that leave II.e. The totality of linear subspaces in it constitutes an irreducible complemented modular lattice P”(K) with . If a lattice L is an ndimensional projective geometry. tatomic elements) and by Q the totality of elements 1 prime over atomic elements. then there exists one and only one Qi passing through these points. If pi (1 < i < 5) are live points in a general position in a plane. Given two hypersurfaces Q. The set of all such &’ is called a pencil of quadric hypersurfaces. and consider the geometry that belongs to the subgroup of G(P”) leaving this Q. The dual of the last theorem is called Brianchon’s theorem. Therefore the line passing through the images of 1 and 7 is a generating line of Q$. Similarly.. we consider an (n + 1)-dimensional (for the noncommutative case. (4) Then (I. The image of a pencil of lines in P3 is a generating line of Qt.-2 invariant. In the cases n = 2 and 3.-‘. Then. If ?p is a finite-dimlensional projective geometry. Klein. and Qi” is covered by two families of k-dimensional generating spaces.) be collinear (Pascal’s theorem). then it is an tirreducible complemented modular lattice of finite theight. we consider another &’ such that the polar of an arbitrary point x with respect to &’ belongs to the pencil of hyperplanes determined by polars of x with respect to Qn2-l and Q’-‘.

Sci. x#(O.. the complex number field. Ginn. [2] W. Tits [9]. English translation. . We put I([x]. Let B be the subgroup of G consisting of all upper triangular matrices (i.. B. . + (x”)’ = 1 in the (n + 1)-dimensional Euclidean space En+’ by identifying the end points of each diameter.. if K is the real number field R. Iyanaga and K. x”)~x=~K. II. Math. When K is a ttopological field (e. Lecture notes in math. Amer. [S] E. Benjamin. A homeomorphism f of A onto A’ is called a pseudoconformal transformation if there exists a tbiholomorphic mapping f of an open neighborhood of A in X onto an open neighborhood of A’ in x’ such . In algebraic geometry the tdirect product of two projective spaces is important. Tits. matrices whose entries below the principal diagonal are all zero). Geometric algebra.g.V~WK) and Q=~~(~xl.... 1910-1938. [8] R.e. II. it can be shown that an n-dimensional projective geometry over K is isomorphic to P”(K). 171-196. Univ. and let G be the general linear group of degree n over k. The theory of buildings has deep connection with algebraic groups.. Vandenhoeck & Ruprecht. Points of P”(K) correspond to (right or left) l-dimensional linear subspaces. Hence P”(R) is compact. [9] J. Sot.13 Algebraic Groups R). Methods of algebraic geometry I. revised edition. Tokyo.(0). D. Let A and A’ be subsets (with relative topology) of tcomplex manifolds X and X’ of dimension n. Van Nostrand. Pedoe. Affine geometry and projective geometry. 1957. and by [x] the equivalence class that contains x. Buildings of spherical type and finite BN-pairs. 1967. N) consisting of a group G and its subgroups B. N satisfies the axioms of a BN-pair or Tits system (. Moreover. Tits’s Theory of Buildings Projective Geometry) (Generalization of 344 (Vll.22) Pseudoconformal A. Schreier and E. the totality of r-dimensional subspaces in P”(K) constitutes a TGrassmann manifold. 1951.. N) forms a Tits system called type (A. We denote by P”(K) the factor set of P under the foregoing equivalence relation. Colloq. Foundations of projective geometry. [7] A. If we fix a basis of I/“+‘(K).. Fat. Publ.. V. Springer. We call each element of P”(K) a point and each element of Q a line. 1962. i.O.O . 0).. We call it a right or left projective space. called a “building. 1967. we call it a biprojective space. Einfiihrung in die analytische Geometrie und Algebra.e.” was introduced by J.C~l)lCxl~ [y] EP”(K)}. then the elements x and y are called equivalent. [3] 0. [4] 0. Projective geometry I. we write x N y. If there exists a nonzero element 1 of K such that y = xl. respectively. Young.1277 344A Pseudoconformal Geometry respect to the inclusion relation.<cc<‘n}. if K is a topological field we can regard P”(K) as a thomogeneous space of the topological group B(P”(K)). Spemer. Birkhoff. 1961..-. Cambridge. matrices such that each column and each row contain just one nonzero entry). In particular. 1974. Projective geometry of n-dimensions. Definitions Geometry In a situation when a triple (G. Therefore projective geometries can be completely classified by means of the natural number n and the field K except when n = 2 and the geometry is non-Desarguesian. the real number field. Chelsea. [l] G. Artin. Lectures in projective geometry. J. 1947. Hodge and D. W. Thus by means of Tits’s theory of buildings the relationships among projective geometry and other geometries have been clarified [9]. Matsuzaka. then P”(R) is homeomorphic to the factor space obtained from the n-dimensional hypersphere S” :(x0)’ + . Since the group of projective transformations 0(P”(K)) acts ttransitively on P”(K). References x”~+Y”P.e. Lattice theory. 14 (1967). we may define the topology of P”(K) as the factor space P”(K) = p/ N. a new geometric object. 386. Let N be the subgroup of G consisting of all monomial matrices (i. G consists of all nonsingular square matrices of degree n with entries in k. or the tquaternion field). The corresponding theory of buildings of the type above is nothing but the projective geometry. B. His theory contains projective geometry as a particular case. Veblen and J. [6] S. Seidenberg. Conversely. Similar facts hold for the cases of the complex and quaternion number fields. Xl . We may restate this fact as follows: We consider a space P”= V-“+‘(K) .[y])={[z]lz’= The Tits system corresponds to a projective geometry in the following case.. then we can represent P”= {x=(x’. Then (G. Let k be any commutative field. Hartshorne. Interscience. I. Then these points and lines and the natural inclusion relation satisfy axioms (I)-(IV) and give an ndimensional projective geometry.). and P”(K) gives rise to an n-dimensional projective geometry.

H(M) is also called the CR (Cauchy-Riemann) structure of M. the smooth (or real analytic) boundaries of bounded domains in C”. to pseudoconformal geometry on hypersurfaces we can apply the methods of differential geometry as well as those of the theory of functions of several complex variables. Chern and J.M is called the bundle of holomorphic tangent vectors of M and is denoted by H(M). L.X is an involutive linear automorphism ofjhe tangent space T. Tanaka (1965) generalized the method of Cartan for the case n 2 3 and ob- 2n + I 2 5. Burns and S.X* T.4. defined only up to a multiplier. a) for U. (xEM) defines a tline bundle E over M. T. is the quadratic form on H.. we can prove that the group A(M) of all pseudoconformal automorphisms of a nondegenerate real analytic hypersurface M in a complex manifold X of dimension n is a Lie transformation group of dimension not exceeding n* + 2n. then f is clearly a CRequivalence. and obtained a criterion for two hypersurfaces in C2 to be pseudoconformally equivalent.M. if there is a bundle isomorphism cp of Y to Y’ such that ‘p*w’= o. i. Equivalence Problem Cartan studied the equivalence problem for the case n = 2. Put H. that y(x)=f(x) for x6. most studies in pseudoconformal geometry so far have concentrated mainly on the investigation of smooth hypersurfaces in a complex manifold-more specifically.X of X at x induced by the complex structure of X. A diffeomorphism f: M+M’ is called a CR-equivalence if the tdifferential mapping rf:f: TM + TM’ off preserves the CR-structures.: T.S*“” or the tangent sphere bundle of a rank one isymmetric space . Let M’ be a smooth hypersurface in a complex manifold X’. If there exists such a mapping . Moser [4]. Then the union of E. S. Nagano [6] and later H. M is called a nondegenerate hypersurface.I. Shnider [S] classified all sirnply connected compact homogeneous strictly pseudoconvex hypersurfaces M wrth dim M = B. . Morimoto and T. where TM denotes the ttangent bundle of M.M defined by L. Let M be a smooth hypersurface in a complex manifold X with the +almost complex structure tensor J. u) = dO(u. In particular. Pseudoconformal geometry is a geometry that studies geometric properties invariant under the pseudoconformal equivalence. E. If the Levi form is nondegenerate at every point of M.structed a similar bundle B over M and a Cartan connection on B in a different way fro-m that of Chern and Moser. By using this solution of the equivalence problem. H.et H be the subgroup of SU(p + 1.+z.q+ 1) such that if M and M’ are pseudoconformally equivalent. However.M = T. obtained a similar result and gave the first proof of this result. Rossi [7] tried to generalize this result and obtained a partial classification of simply connected compact homogeneous hypersurfaces with dimension > 5 D. A is said to be pseudoconformally equivalent to A’. J. O)EC”+’ fixed. C.M for x E M. Jacobowitz [S] con. q + 1) lea.)* = m*]z~+z~+z~\* (m> 1) or the double covering of such a surface.e.. Classification Cartan (1932) classified all simply connected homogeneous hypersurfaces in C*.~. then M and M’ are pseudoconformally equivalent.z.ln connections in some fiber bundle over the hypersurfaces. Conversely.(u. If f: M-M’ is a pseudoconformal transformation.M n J. However. According to the Cartan-Tanaka-Chern-Moser result. then M is pseudoconformally equivalent to either (1) S3 or its quotient by the action of a root of unity or (2) the hypersurface given in the 2-dimensional projective space by the equation in homogeneous coordinates:(z. We do not know whether B and Y actually coincide.344 B Pseudoconformal 1278 Geometry tained a criterion in terms of +Cart. In particular.+z. Let E. H.ving the point (l. The Levi form L. where Y’ is the corresponding principal fiber bundle over M’ and u’ is the Cartan connection on Y’. then there is a bundle isomorphism cp 01“ Y to Y’ preserving the Cartan connections: v*w’ = w. They proved that A4 is pseudoconformally equivalent to . be the annihilator of H. where fI is a nonvanishing section of E in a neighborhood of x. The union of all H. In fact. at XEM. Such a hypersurface is called homogeneous. we can construct functorially a principal fiber bundle Y over M with structure group H and a Cartan connection w on Y with values in the Lie algebra of SU(p+ l. independently of Tanaka [3]. N. Poincare [1] studied perturbations of the boundary of the unit ball in C2 that are pseudoconformally equivalent.O. UGH. he did not publish the proof of his result until S. if the Levi form is definite. then M is called strictly pseudoconvex.M in TX*(M). A.z. Let M be a real analytic hypersurface in C”+’ (n > 1) whose Levi form has p positive and q negative eigenvalues (p + q = n). he proved that if M is a compact homogeneous strictly pseudoconvex hypersurface with dim M = 3. Cartan [2] studied the equivalence problem of hypersurfaces in C* and gave the complete list of all simply connected hypersurfaces on which the group of pseudoconformal automorphisms acts transitively.

.e.+ 1z.2) for which we denote by H(M. then the universal covering space fi of M is also spherical. then (A) implies (D). and D.Section B). Let B c Rk be a small open ball around 0.. -+D. He then showed that the second largest dimension for A(M) is equal to n2 + 1 except when n = 3 and the index r = 1. I. and Wells (1978). be bounded domains in C” with smooth boundary aD. On the other hand. We know that the only compact simply connected spherical M is S2”+1. When D. such that fl . Bell and E. is real analytic for i = 1. Let U= {x E VI --E < p(x) < E} for small E > 0 such that Z? is compact and aU is smooth. then f extends holomorphically past the boundary. is biholomorphically equivalent to D. We do not know whether (B) implies (C) in general. When does (A) imply (B) and when does (B) imply (C)? Let M be a compact connected strictly pseudoconvex real hypersurface in a complex manifold X of dimension n + 1.. then M is the afflne part of a real hyperquadric in P. Bell generalized the result of Fefferman in the case when one of D. a real hypersurface M in a complex manifold X of complex dimension n + 1 is called spherical if at every point p E M. V<n. by using the theorem of Fefferman. i. are strictly pseudoconvex..+. and M2 are real analytic and if D. S. If M is a homogeneous spherical hypersurface. it seems difficult to extend Cartan’s classification of all simply connected homogeneous hypersurfaces to higher dimensions. He also obtained a similar result in the nonhomogeneous case. We do not know whether (A) implies (C) when M. If M is spherical. and D. Relations to Other Equivalences E.2) be strictly pseudoconvex. and D. (b) If aD.. Let g(u) be the open set in Cm(U) of strictly plurisubharmonic functions $ with dt+b A & A (d&b)” #O on u. The hyperquadric p+l= a&+. Naruki [ 123 obtained the same result. (i = 1. Deformations of Domains Let D. where L. are strictly pseudoconvex. we can prove that (A) implies (C) when M. are real analytic and D... (C) M. is spherical.12+. and M. and M2 are real analytic and if D. It is clear that (C) implies (D) and that (D) implies (A). +D. I. and D.4(M) = 1 1( = n2 + 1). S. though we know that (A) implies (B). for which dim . Shnider. such that U n aD and V fl aD’ are pseudoconformally equivalent.) the CR structures. there is a neighborhood of p in X such that U n M is pseudoconformally equivalent to an open submanifold of S’“+‘. K..= Mi (i = 1. be a proper holomorphic mapping. and D. Fefferman [lo] proved that (A) implies (D) when D. is a certain subgroup of SU(n + 1. we do not know whether (A) implies (D) or not. then f extends smoothly up to the boundary D. (B) Ml is CR equivalent to M. aD’. Pinchuk [ 133 proved the following: Let D. Let cp be a smooth strictly tplurisubharmonic function defined on a neighborhood V of M such that M = {x~VIcp(x)=O} anddq#Oon M. is biholomorphic. = {(z 1. (II) S*“+l S2”+l flR”+‘. =D. Burns and Shnider classified all homogeneous domains M in S’“+‘: M is pseudoconformally equivalent to (I) or (II) of the following: (Ia) S2”+l V fl S’“‘l. and D.. Yamaguchi (1976) treated a hypersurface M in a complex manifold of dimension n with a large automorphism group A(M). 0 < m < n. respectively.C (. where U. and f(fi) is a homogeneous domain in S2n+1.12}. C. Then we can find a holomorphic mapping f:D+D’ such that J(x)= f(x) for XED fl U. l)/(center). D’ be strictly pseudoconvex domains in C” with simply connected real analytic boundaries aD. is pseudoconformally equivalent to M2. Under the additional assumption that M is homogeneous. (a) If D. On the other hand. we can prove that (B) is equivalent to (D).. 2. and let f: D.<dim. and D. then there is a covering into mapping f: a-+S”‘+l. D.. Combining this theorem with Fefferman’s result we see that for two domains as above. Concerning the tproper holomorphic mappings rather than diffeomorphisms.1279 344 E Pseudoconformal Geometry or the unit circle bundle of a homogeneous negative line bundle over a homogeneous algebraic manifold.+1)EC”+111m(z. . 0. Burns and Shnider (1979) proved the following theorem: Let Mi = dD. He showed that if dim A(M)= n2 + 2n. he showed that if dim A(M) = n2 + 1. Let f: U+C” be a nonconstant holomorphic mapping from a connected neighborhood U of a point PE aD in C” into C” such that f (U n aD) c aD’. At present. where V is a complex vector subspace of C”+’ with O. Ligocka [ 1 l] proved that if M. are not strictly pseudoconvex and Mi is not real analytic. z.. We consider the following propositions (A)-(D) for these domains: (A) D. then M is a real hyperquadric in the n-dimensional complex projective space P. D is biholomorphically equivalent to D’ if and only if aD is locally pseudoconformally equivalent to au. : D. there are neighborhoods U and V of a point p ELID and q E do’. As remarked by Burns.C (except when n = 5 and r = 2)..)>lz. We denote by p( u x B) the set of $ cCm( u . are pseudoconvex.. are pseudoconvex.. (D) There is a diffeomorphism f: fil -0. is strictly pseudoconvex. (Ib) Qn+lL.

+A. Pura Appl. 26 (1974) l-65. Tanaka. [l l] S. Burns and S.. 14 (1963).. Math.. 1 (1932). of the unit open ball B. Scuola Norm. 131-145. n + 1).x. 1855220. Moser. 109-123. Inventiones Math. 17-90.(x)= 6).+. [S] H.. S. II. Real hypersurfaces in complex manifolds. (3) Let M be a real hypersurface in C”+’ with H(M) the bundle of holomorphic tangent vectors to M... Mat. in C”. also bounds a relatively compact region D. Les fonctions analytiques de deux variables et la representation qconforme. J. This result gives.=x. 13 l190. If M bounds the relatively compact region D in X then M. Studies.d. Rend. Math. can be extended continuously to a function 7: 0. there is a constant c > 0 such that If(zl)-f(zz)I<cIz. = B.... More precisely. 62 (1976). J. He also proved that every proper holomorphic mapping f: B. B). F.dl if and only if t.x:+B... 33 (1976) 2233246. Ann. Ligocka.={xcUl$. Then either f is a constant mapping or there is a biholomorphic automorphism 7: B. if n > 1. graded Lie algebras and Cartan connections. He considered the equiv- . ~2~4.cR small enough. Shnider.) c aD. (k = 1. Japan. 133 (1974) 2199271. After introducing these notations. Henkin (1973) proved that every proper holomorphic mapping f: D.(x)E~( U) for all t E l?.. If there is a Cl-mapping f: U fl o1 +C” such that f is holomorphic on U fl D. Webster considered the relation between pseudo-Hermitian manifolds and pseudoconformal geometry and proved that for n > 2 the ellipsoid E given by the equation A. [lo] C.y. Non-degenerate hypersurfaces in complex manifolds admitting large groups of pseudoconformal transformations I... 6 E R small enough. Sur la geometric pseudoconforme des hypersurfaces de l’espace de deux variables complexes I. then there is a holomorphic mapping 7: U’-+ C” of a neighborhood U’ of U f’aD. =t. 0) of dimension 2n + 1 is a Lie transformation group of dimension not exceeding (n + 1)‘.+. (2) H. 59 (1973). be strictly pseudoconvex domains in C” with C” boundary aD.wherez.344 F Pseudoconformal 1280 Geometry alence problem of pseudo-Hermitian manifolds by applying Cartan’s method of equivalence. 6. Inventiones Math.. Yamaguchi. Rice Univ. +D. Burns.2+. In particular. into C” such that y(x) =f(x) for XE U n & n U’. (ii) The group of CR-automorphisms of M reduces to the identity only. Jacobowitz. References [l] H. Mat. Ann. Bell and E. Spherical hypersurfaces in complex manifolds. Mf.Z+. There are arbitrary small perturbations of the unit sphere in C”+’ that admit no pseudoconformal transformations other than the identity. such that y(x) =f(x) for XE U n B.+iy.. into a strictly pseudoconvex D. by virtue of Fefferman’s theorem. [2] E. taking t E B. Poincare. S. There exists an open dense set V c Y( U x B) with cpE V and a set of tsecond category &? c V such that for every $ E%?. -~~lr/~ for every ~1.d is a compact connected strictly pseudoconvex hypersurface in X. D. .. Acta Math. .. Japan. 11 (1932).. 55-96. and f( U n aD. Fefferman. Inventiones Math. [7] H.. Nagoya Math. is necessarily an automorphism of B. He proves.+B. Pisa. t) = $. Circ..= l. Nagano. Morimoto and T. Homogeneous strongly pseudoconvex hypersurfaces. [4] S..+B. Chern and J. of a strictly pseudoconvex domain D..(k=l.. G. Cartan. there exist smooth families of deformations of the unit ball in C”+’ of arbitrary high dimension. 57 (1980). 2899300. (i) Mt. [6] A. Sup. Webster (1978) called the pair (M. 23 (1907). Induced connections on hypersurfaces in C”. aD. Rossi...n+l)ispseudoconformally equivalent to the hypersphere S2”” if and only if A. that the group of all pseudo-Hermitian transformations of the nondegenerate pseudo-Hermitian manifold (M. 0) a pseudoHermitian manifold..+B. On nondegenerate real hypersurfaces. A simplification and extension of Fefferman’s theorem on biholomorphic mappings. Shnider. This result is related to the implication (B) Z-(C) in Section D. among other things. is CRequivalent to Mfl. The Bergman kzrnel and biholomorphic mappings of pseudoconvex domains. a necessary and sufficient condition for an ellipsoidal domain to be biholomorphitally equivalent to the unit ball. For $E Y[$. 2 (1976). and Wells (1978) proved the following theorem. We take a real nonvanishing l-form 0 that annihilates H(M). x B) such that +(x. 43 (1977). Let U be a neighborhood of a point psaD. 3333354. For(1. [S] D. On pseudoconformal transformations of hypersurfaces. ti~B and 6.EY(UxB)wesetM. Inventiones Math. + 0. Sot. M. [3] N. [9] K.. = 6. Alexander [14] proved the following: Let U be a connected neighborhood of a point p6S “-i in c” and f: U-42” a holomorphic mapping such that f( U n S’“-‘) c S2”-‘.. Palermo. M. Topics Related Geometry to Pseudoconformal (1) Pinchuk (1975) proved the following: Let D. 2833289..y:+. J.

s. Pseudo-Hermitian structure on a real hypersurface. p and any compact set K c R”.B...(l +It3m+d’=‘-p’fl’. 28 (1976) 117-122. 137146. DX) = c lal<m Differential operators (1) with coefficients of class g (. Nirenberg [ 11. UEY.. On holomorphic mappings of real analytic hypersurfaces. we here list the main results of the theory of pseudodifferential operators: (i) Pseudolocal property.. then P has the pseudolocal property u E 9” * sing supp Pu c sing supp u [3]..Set H-“JH”. On extendability of isomorphisms of Cartan connections and biholomorphic mappings of bounded domains. and set 345 (X111.(R”)}. there Choosing the Hormander class SF. Then by means of the tFourier inversion formula. H” =H”. 5) is defined by (2).) with the symbol p(x. M. 5)p(x.. 0.. then p(x. Therefore. which is a Banach space provided with the norm ~Ju//. [14] H. J. Then there exist P*=p*(x.r)l~c. 26 (1977). Pseudodifferential Operators Operators Pseudodifferential operators are a natural extension of linear partial differential operators.u)=(u.= II~lls. (ii) Algebra of pseudodifferential operators. Proper holomorphic mappings in C”. for a general function p(x.& a*/i?x. -m<s-cm Then where t2([) denotes the tFourier transform of u(x) (. The theory of pseudodifferential operators grew out of the study of singular integral operators. PER”. For any 1 < r < co and real s. But this representation of Pu(x) has a meaning even if p(x. j= 1.x = C. are continuous mappings of Y into Y. and Q = q(x..160 Fourier Transform H).+mzsuch that (Pu. Indiana Math. and 3=pj(x. r) is not a polynomial in <. 25-41.@are independent of K..(G). J.z. the operator (1 . J. Math. 5) is said to be of class S:. The term “pseudodifferential operator” first appeared in Kohn and Nirenberg [ 11. P* is the forma1 ad- . If for any pair of multiindices E. [ 131 S.fi. The operator P of class SF6 in general does not have the local property u E Y’ * supp Pu c supp u.1281 345 A Pseudodifferential Operators [ 121 I.(1-A)s’*~).. the pseudodifferential operator P =p(x. ...(G) for real numbers m. Let P be a tlinear partial differential operator of the form p = p(x. in the case0<6<p<l anda< asamodelclass.DX)~Sz.‘. In particular. and others. Let P=p(x.n H”. D. P*u) for u. Hiirmander [2]..A)s” can be uniquely extended to be a mapping of Y’ into Y’ by the relation ((1-A~‘*u. (2) lI4. 0.l. but if p > 0. H-=O=H-W. exists a constant Cb.2. [lS] S.33) Pseudodifferential A. When 0 = R” and constants Ca.168 Function Spaces B(13)) belong to s.2 is defined as a pseudodifferential operator of class Sf. Alexander. USSRSb. Naruki.. 0. J. Differential Geometry. Webster.) E S. we denote SEd(R”) simply by Szd.= j/(1 -A)““uII. and developed rapidly after 1965 with the systematic studies by J. L.‘.D&Sz. i.D&S~. 13 (1978).*.2. UEY’. tMW5.(n) and is often denoted by P = p(x. VEY. the tSobolev space H”. 5) be a Cm-function defined in R x R”. Pinchuk. 34 (1978).(G) (c Cz(R”)). Operators of class SF. H”= Hsg2 is a Hilbert space with the norm aa(x) (1) and let u(x) be a function of class C. p.-m<s<m u fp’.)E SF..e. and 6 with p > 0 and 6 > 0 in the following way: Let p(x. A symbol class is determined in accordance with various purposes. but it is always required that the corresponding operators have essential properties in common with partial differential operators. The complex power (1 -A)@ of lA = 1 . Kohn and L. v) = (u. XEK. for any real s. I.l#).’ is defined by HS*‘={u~Y’l(l-A)S’*ueL. The operator P defined by (2) is called a pseudodifferential operator (of order m) of class SE. Thus. TBhoku Math.ei by the symbol (1 + l (I*)‘/*.. Hormander [3] defined a symbol class SF. 503-519.K such that lD~DrsP(x. Pu(x) can be written in the form z%(x) =(27pZ s R” exp(ix..

wehave Ip(x. More refined and useful classes of pseudo- c41. 1.x(y)) = (ax. we say that a point (x0. with 0~ 6 <p < 1. such that lD. Then for any P = p(x. Asharp form of Girding’s inequality has been proved by Hkmander [6]. For P E Srd and any real s there exists a constant C. I< I”” (I 5 I> R).. An important fact is that the relation smgsuppu = Proj. Tokyo.andR>O. D. u. Grushin (Functional Anal.)withujEHm”. 1968) and Kumano-go [4]. H.u)> whereu=(u. PI. . if there exist a(. Functional Anal. right) parametrix is a sufficient condition for P to be thypoelliptic if p > 0. such that llp4dCsllulls+. and Cm’ < Jdet(a. 0. Then for P = p(x. <)D&(x. +m. where det(Z. the equation Pu = f ~9’ is locally solvable). . such that aPhu~C~(R)..4]. Dx)eSEd in R. the cotangent bundle of R minus its zero section.)E SE8 in R:.I (resp.x(y)) denotes the determinant of the Jacobian matrix (3. the analysis on T*(R)\O..pD. we say that p*(x. and HGrmander [S].-(p--G)N-+-co as N+co. Kumano-go (J. 5) satislies the following conditions: (i) for some C...) belong to S:.. (resp.D1516’p’-p’a’. (ii) for any c(. when m = s = 0.)ap.. Kato (Osaka J. Then we easily see that WF(u) is a closed conic subset of 7’*(R)\O... 5’) of T*(R) 10 does not belong to the wave front set (or the singular speclrum) of u. 5) = (iD. US. has been obtained by Hiirmander [3]. 17 (1970)) when 0 < 6 <p < 1. then for any integer N we have and (4) Hence the operator the sense class SFn is an algebra in where m. 4x0) #O. Let x(y)=(x. m2). denoted by WF(u). j. respectively: P*k and i)-gP:(x> 0 c3. we havem-(p--6)N+-m and m. 5) and qAx. V. 0.. 4 1:1970)). . D. 76 (19791). An operator P = p(x. () have asymptotic expansions in the sense of (3) and (4).~(x.) of P in the class Sp. which enables us to resolve sing supp u on T*(R)\O. there exists an operator Q = 9(x. j. onto R: such that ax. and Il~ll~~~=C. (x.x(y))l<C for a constant C>O. the existence of a left (resp. For a distribution u E s’(O).(x. of sing suppu. we set POX. and otherswhen0<6<p<landfi<l. UEW+” (5) 1512R. Vaillancourt [S].t. Melin (Ark. b(xO) 20. (iii) H”-boundedness. (vi) Parametrix. Mat.(y)/ayj).realm’(<m)..z~~)l/l~~~1”>0. Furthermore. 9 (1971)). Sci. Lax and L. Pseudodifferential operators of multiple symbol have been defined by K. The estimate (5). Kumano-go pt]. Nirenberg [7]. and sharpened by A.(y). and A. Cordes (J. Assume that the symbol .. For a given PsSE.j=l (6) . A general sufficient condition for the existence of a parametrix for an operator of class Sz. 5) E SFd. Nat. . 5) and q(x. .to(n). For a differential operator P.) there exists a constant C such that WPu. (iv) A sharp form of Gkding’s inequality.r’.. > O.. k = 1..<). 5’)~ T*(R)\0 if lim. . n. of P. 13 (1976)).. .#) we can define the wave front set of u E g’(n). In particular. in particular. Fat. was obtained by HGrmander [3]. from which we can perform a so-called microlocal analysis. such that (Qw)(y)=(Pu)(x(y))for w(y)=u(x(y))~Y. Ip(~~. This fact enables us to define pseudodifferential operators on C”-manifolds [3. Fefferman and D. an operator E E SE. and Q = P. P has the micro-pseudolocal property: u~Y’ 3 WF(Pu)c WF(u). Sci. <). -Cll~Ilf~-~~-~.345 A Pseudodifferential joint 1282 Operators if P. DX) E S. P. As the sharp form of the pseudolocal property of an operator PE SFdr if 0 < 6 < p < 1. Let P=p(x. to).~(x> O/p(x> 01 G Co. 5). we call it a parametrix of P. Acad. = max(m... and PE S. Univ. Math.lo(0) is said to be microlocally elliptic at (x0. Appl.(y)) be a Cm-coordinate transformation from R..I) is of of class Se r”. C.p $6 < p < 1. k= 1.x.. T. if 0 < fi<p<l.. 18 (1975)). b there exists a constant C. P.~~. ME CW).<)l> C. H. Then. <) = (pj. 5) = WD$P(X...41... 0.(y)/Zyjeg..7(x. Phong (Proc.. is called a left (resp. V. Calder6n and R. . If E is a left and right parametrix . H. right) parametrix of P if EP . WF(u) (the projection of WF(u) on a) holds. D. Let p(x. I) be a Hermitian symmetric and nonnegative matrix of P~.D. Friedrichs (Courant Inst. Then there exists a parametrix Q = q(x. By means of operators of class S. Assume that 0 < 1 . which is microlocally elliptic at (xc’. PE .!=~ll~~ll~~~~ (v) Invariance under coordinate transformations.

1981).. A*A and AA* are pseudodifferential operators. F. 94 (1972)). In this case. y)l <C(l + lel)m-P’S’+(l-P)(‘“‘+‘u’) for some fixed m and p.(A) and the symbol aAm= u.+. y) of A is of class C” outside W. constitute a system of local coordinate functions of R” x (RN ~0) x R” in a conic neighborhood of C. Strauss (Comm. Let A. Sci.e. If d. The following theorem is due to Egorov [ 111: Let P and Q be pseudodifferential operators of class Sz . The conic Lagrange manifold A+. Let A be a Fourier integral operator such that A.. Y)) 6 YMY) dY de..@=@ uJ. respectively. Then the adjoint of A is a Fourier integral operator such that the associated conic Lagrange manifold is the graph of the inverse transformation x-l. then the composed operator A. Ikawa (Pub. 14 (1978)) on the exponential decay of solutions..e.& y)#O for 6~0. Res. Press.. Ohya (Publ.~(A) = a.. Hence for any Fourier integral operator A.. i.y)ER”xR”l38#Osuchthat(x. Math. C. 0.. are linearly independent at every point of C.. Yu. . .(acp(x. and others. There have been detailed studies of the case where the d.. Math. Sci. Here ~1. = A. S.. N. j=l. S. one can construct a Fourier integral operator A such that A. e.. 4 (1968)) Hormander (J. 32 (1977)) on tweakly hyperbolic equations. . the canonical 2-form e = Cjdtj A dxj . Let a. Taylor (Princeton Univ. Analyse Math. In fact. Lax (Comm. F.~)~ 5 =ba e.(A) is the graph of the identity mapping of T*(R”)\O. Mizohata and Y. Atiyah and R.. given a conic Lagrange manifold A in T*(R” x R”)\O and a function a.y)E C. For recent developments in the theory of pseudodifferential operators and its applications . and W.~@-’ is called the local symbol of A.< 1.-JR”) if and only if A. Surveys.~). be a system of local coordinates in A. J. and any triple of multi-indices c(. then the kernel distribution k(x. These.. Pure Appl. Let J denote the Jacobian determinant m6. The function cp is called the phase function and a the amplitude function. a Fourier integral operator A is a pseudodifferential operator of class 8: .cp(x. = A. such as M. Friedrichs and P.y)Id. B. 30 (1977)).. 8. Res.Y. J. and Nirenberg and Treves [ 161. Ralston. to T*(R” x R”) ~0.. 19.a(x. . l/2 < p . If the equality PA= AQ .. the cotangent bundle of R” x R” minus its zero section.Cjdqj A dyj vanishes on A. Pure Appl.e#O} and W={(x. Bott (Ann. J. and let A be a Fourier integral operator such that the associated conic Lagrange manifold A. and the multiplicative group of positive numbers acts on A. Beals (Duke Math.y). <). and for rp(x.. M.(A) is the graph of a homogeneous canonical transformation x. 5. ff= The function a. 86 (1967)) on the tLefschetz fixed-point formula... Kumanogo (Comm. is a smooth manifold in R” x (RN\O) x R”. y) a real-valued function of class C” for 0 # 0 and homogeneous of degree 1 in 0 there. Pure Appl. Beals and Fefferman [ 173 on local solvability theory. C.={(x. The image @C. Math.. S.cp(x. A. V. y. Inst.. and the mapping 0: c. Conversely. A pseudodifferential operator of class 8: i-JR”) is a particular type of Fourier integral operator. if the associated conic Lagrange manifold is the graph of xi. .1283 345 B Pseudodifferential Operators differential operators have been defined by R. J. Alinhat and M. .y)=O.(A) = A and u.-JR”) with the symbols p(x. satisfying the (7) s RN+” x4x. . Morawetz. is an immersion of C.+.~ is the restriction of a to C. be another such operator. Math. 1981).. Treves (Amer.. acp/aeN. Math...rp(x. Inst. Let C. The theory of pseudodifferential operators has found many fields of application. M. 30 (1975)) on subelliptic operators. A is also a Fourier integral operator and the associated conic Lagrange manifold is the graph of the composed homogeneous canonical transformation xix. 18 (1965)) on symmetrizable systems. p. 5) and q(x.}. Math. on it. 42 (1975)) Hormander [S]...8. Egorov (Russian Math. Fourier Integral Operators A Fourier integral operator A: C$‘(R”)+GS’(R”) is a locally finite sum of linear operators of the type Af(~)=(2n)-‘“+~)” exp(icp(x. y)/&?j). D. Math.BD... 22 (1969)).2 . ... 8.@. Bouendi (Amer.(A) is the graph of a homogeneous canonical transformation x of T*(R”)\O. d.y) is a Cm-function inequality pp. J. 102 (1980)) on uniqueness of the Cauchy problem. .d . y) of the Fourier integral operator A. Those Fourier integral operators whose associated conic Lagrange manifolds are the graphs of homogeneous tcanonical transformations of T*(R”)\O are most frequently used in the theory of linear partial differential equations.a.Kumano-go [4]. Treves (Plenum. ~Y)-+. 8. V.. together with adae.?(A) essentially determine the singularity of the kernel distribution k(x. and others. . Here a(x. Hormander [6]. is a conic Lagrange manifold.

Kumano-go (Comm. Let us consider a phase function of the form cp(x. Then one can find a Fourier integral operator A such that the function 9(x.345 Ref. y) = S(x. The above-stated transformation theorem of Egorov has been studied in detail with reference to systems of pseudodifferential equations with analytic coefficients [ 19](274 Microlocal Analysis). 24 (1974)). + p(t. E(t.p(t. Then by (7) the Fourier integral operator A = A. Nirenberg and Treves. x. Pseudodifferential 1284 Operators function independent of y of class S. Some suflicient conditions for boundedness have been obtained by Eskin (Math. q). 5) with (x. 5) of Egorov’s theorem satisfies the relation 9(x. X. Hiirmander and J. [12. USSR-.. Math. Chazarain (Inoentiones Math.. SF.” and applied these operators to the derivation of highly accurate asymptotic formulas for spectral functions of elliptic operators.u)= {(x> 5)= T(y.1~k. 1 (1976)). USSR-Sb. x. [loI> 1.g. <).. let (x. Thus by means of (9) we have WF(u(O)c {(x. ‘Then (X(t. which was originally published in 1965. 3 (1967)).(x”. 5 exists in g” ([0. Egorov (Math. 11 (1970)) applied his theorem and the corollary stated above to the study of hypoellipticity and local solvability for pseudodifferential operators of principal type..S(t. Nirenberg. Kohn and L. E = V. <)-p(x(x. homogeneous of degree 1 in r for /<I> 1.l. d. Fujiwara [ 181.S(x.S+p(t. and others. 325 Partial Differential Equations of Hyperbolic Type M). z(t.. 5 = V. y. these results were completed by Beals and Fefferman [ 171. <)E&?’ ([0. II).<)~S~. E Sp.. To]. ..S)=O on [O.325 Partial Differential Equations of Hyperbolic Type L. J. A calculus of Fourier integral operators in R” was given in Kumano-go [4]. q). T is defined by y = V$(x. p = 1. J. where p.(xo. 5) is a real-valued ?-function.P(~. x.%. 5) be an amplitude (8) Let T be the canonical transformation with the tgenerating function S(x. On the other hand. 5)) belongs to the class S. T]. 7’1. a)) be the bicharacteristic strip defined by +Hamilton’s canonical equation dxJdt = V. i.y ‘5 in R: x R. 4. = x. which correspond to Fourier integral operators in the theory of distributions.n algebra pseudo-differential operators.rl)~‘~F(u)}.5’)..15] constructed a general global theory of Fourier integral operators making use of Maslov’s theory [14].. K. we have WF(A.S(t. The propagation of wave front sets by means of a Fourier integral opera&r is described as follows. 4) of the eikonal equation a.. 3 (1976)) used a type of Fourier integral operator in deriving the energy estimates and constructing the fundamental solutions for strict hyperbolic operators.. G. Comm. UEY. 4 (1978)). 4) l(~~d~WF(uo))> (10) which is the fundamental result in the study of the propagation of wave front sets as solutions of general hyperbolic equations (-. The works of Egorov. J.) with some To > 0. H6rmander (Acta Math. Duistermaat [9.13. USSRSb. x.. x. is defined by A. y. Assume that m = 1. 121 (1968)) introduced the term “Fourier integral operators. S:.e. y. The theory of Fourier integral operators has its origin in the asymptotic representation of solutions of the wave equation (. 1 References [l] J. x. D. Y. 5) -pl(x. 18 (1965). Consider the Cauchy problem Lu = 0 on [0. q). 5). Nirenberg and Treves [ 161 obtained decisive results concerning local solvability for linear partial differential operators of principal type. such that p(x. Pure Appl. 5) .. o(R”). of . x R. 5) = (XQ. y. 5)M:c W(5M. ultEO = uo. 5) =(X(t.2p(R”). 0. Asada and Fujiwara (Japan. Sl.. then 9(x. q). e. Then for the Fourier integral operator A. q)) can be solved by means of the relations y = V. 5)~ B”( [O. the Fourier integral operator has come to be recognized as a powerful tool in the theory of linear partial differential operators... Partial DiJff: Eq.) for a real-valued symbol p(t.u(x)=(27c-“‘2 s R” exp(iS(x.For a small 0 < Tg To the solution S(t. = (Y. Using Egorov’s theorem and the same corollary..) such that the solution u(t) is found in the form u(t) = E. Math. The boundedness of Fourier integral operators in the spaces L*(R”) (or the space H”) has been studied in several cases. d</dt = . W> Y. 5) ..~)I(y. Eskin (Math. holds.V. X. T]. By virtue of this research.~O)#O at (x0. 0. and that p1 (x. also.to)=O. and let a(x.x. Hiirmander [9]. as a family of canonical transformations with a parameter tc [0.5.14]).. A. (9) Next consider a hyperbolic operator L = 0.)..(R”) and drp. An interesting application of the global theory of Fourier integral operators appeared in J. 269-305.nere exists an amplitude function e(t.(t)u.=. Sato and gave rise to the concept of tquantized contact transformations. 5. Then t. I. and HGrmander motivated the theory of hyperfunctions developed by M.V. T]. T] with the initial condition SI. q).

Nirenberg. and the number who prefer Oj is nji = n . the number who prefer Oi is nij. Treves. Pure Appl.. Math. Nat. Comm. . the method of comparing them two at a time in every possible way is called paired comparison. II. [8] L. J. Oj) is pij. Hormander. Fourier integral operators I.. Using pz = @(pi . The experimental method in the Bradley-Terry model . . Furthermore. Proc.459-510./n as estimates of the true pij. jI6] L. 287. Kumano-go.=p. Of the 12judges who compare this pair. 1973. 13 (1960). Math. Math. Egorov. Acad.. P. Pseudo-differential operators. 129-209. In the following sections we describe the basic statistical procedures used in sensory testing. Xi. It consists chiefly of statistical methods to deal with psychological measurements and of theories dealing with mathematical models concerning learning processes. Springer. On local solvability of linear partial differential equations. we can obtain the estimates pi. A class of bounded pseudo-differential operators. On canonical transformations of pseudo-differential operators (in Russian). Hormander. (2) 97 (1973). Let Q(x) be the standardized normal distribution function and pij = @(pi pj). 79-183. P. V. 51(1975). Fourier integral operators. 1973. we can test the hypothesis thatp. 83 (1966). [ 151 J. 23 (1970). Math. Vaillancourt. In this comparison it is assumed that the strengths of the stimuli Oi. Fourier integral operators II.. Sensory Tests A measurement wherein human senses are taken as the gauge is called a sensory test. Lecture notes in math. 96-99. . 18 (1965) 501-517. Ann.. Duistermaat. 11851187.nij. and the examining circumstances must be controlled. 473-492.= n. social attitudes.) [S] A. Comm. Comm. Duke Math. a sharp form of Girding’s inequality. Math. [3] L. 128 (1971). Suppose that the probability that Oi is preferred to Oj for a pair (Oi. (1) Thurstone-Mosteller Model. Math. Nauk.. and Oi is preferred when Xi > Xi. J..). Calderon and R.=.17) Psychometrics A. C. [ 181 D. US. 359-443. 69 (1972). Comm. Uspekhi Mat. Proc. [ 131 D. On the boundedness of integral transformations with highly oscillatory kernels. Gauthier-Villars. The Weyl calculus of pseudo-differential operators. and mental abilities. Pure Appl. 25 (1969). J. Ann. Fujiwara. Symp. Pseudo-differential operators and non-elliptic boundary problems. Exact and asymptotic solutions of the Cauchy problem. The panel of judges must be composed appropriately. Asymptotic solutions of oscillatory initial value problems. Math. Duistermaat and L. On stability for difference schemes. . Pure Appl. 128 (1972).. D. [14] V. [9] L. Amer. Math.. D. Pseudo-differential operators and hypoelliptic equations. Theorie des perturbations et mtthodes asymptotiques.. Hiirmander. Hormander. Lax. Paired Comparison When there are t objects (treatments or stimuli in some cases) O. Japan Acad. [19] H. [6] L. it is assumed that the joint probability distribution of Xi and Xj is the 2dimensional tnormal distribution with pi and 0’ as mean and variance of Xi. Various methods of psychological measurements are applied. Pure Appl. Lecture notes.. 473-508. Acta Math. B. Pure Appl. 10 (1967). 1974.=p. Sot. 627-646. 19 (1966). 1981. Beals and C. General Remarks Psychometrics is a collection of methods for drawing statistical conclusions from various psychological phenomena which are expressed numerically or quantitatively.1285 346 C Psychometrics [2] L. 346 (XVlll. 1972.fij) and p. (2) The Bradley-Terry Model. On local solvability of linear partial differential equations I.. MIT Press. [7] P.. 183-269.. Ludwig. [ 1 l] Yu. Pure Math.. Nirenberg and F.. and . Oj to the senses are random variables Xi. O. [4] H. O. 235236.p as correlation coefficient of Xi and Xj. Fefferman. [ 171 R. Hyperfunctions and pseudo-differential equations... I-38. [ 121 P. (Original in Japanese. Maslov. Hormander. Hormander. 482498. Lax and L. Pseudo-differential operators. 138183. Komatsu (ed. 24 (1957). Sci. Comm. Acta Math. 32 (1979). Using p. Proc. Courant Institute. [lo] J. The following are typical mathematical models of this method. There is no loss of generality in assuming that 2a2( 1 -p) = I and & pi = 0.

O. a psychological scaling is given by estimating the parameters. the model of factor analysis is formulated as follows: Let zjk be the standardized score of the jth test achieved by the kth subject. n of them examine Oi first and 0.) = 1./(zT+~L~). order effect. and duo-trio test are sensory difference tests. Each pair Oi.280 Multivariate Analysis). meaning. etc. D. (iv) Cov(fi. the hypothesis that A and B are different and that the judge has no ability to determine the difference between them is tested by using the tbinomial distribution. respectively. Scaling (I) One-Dimensional Case.s. The Pair Test. (ii) V(J)= 1 and Cov(f. vj)=O.. we can test the appropriateness of the models.346 D Psychometrics is the same as in the Thurstone-Mosteller model. Factor Analysis Though factor analysis can be considered to be a method to deal with multivariate data in general (. 1. rci= 1) such that P~~=z. . .280 Multivariate Analysis).f..)=O for i# i’.. etc. triangle test.k + uju.. and Duo-Trio (2) Multidimensional Scaling (MDS).. while the method by Shepard and Kruskal is called the nonmetric MDS. Usually it is assumed that (i) V(z. O.presentsthe magnitude of the ith common factor (ability) in the kth subject and aji is the size of contribution of the ith factor to the score of the jth test. and error. MDS was first developed by Torgerson (1958) and refined further by Shepard (1962) and Kruskal(l964). +BIBD.fi. The mark given by the kth judge on his preference of Oi to 0. (3) Scheffk’s Model. can also be applied to paired comparisons.k=l... Multidimensional scaling is a collection of methods to deal with data consisting of many measurements on many objects and to characterize the mutual distance (dissimilarity). no preference.2. or closeness (affinity). &. or 3-point) scale. r. . It is postulated that.. It has seen useful applications in the analysis of people’s attitude and perception and their characterizations by means of a few numbers or points in a space of low dimension. Under the assumption that a psychological phenomenon is a random variable with some distribution law and the parameters of the distribution law determine psychological scales. it was initiated by Spearman (1927) and developed further by Thurston (1945) in order to measure human abilities from test scores.f. Oi is moderately preferable to Oj..re. and discriminant analysis (. Mosteller of stimuli 1286 model is a method for scahng a set by means of observable proportions. there exist parameters ni for Oi (rci>O. Triangle test: A judge is requested to select two samples of the same kind out of A. j#j’. then it is assumed that it can be represented as a linear combination of r common factors and one specific factor as zjk = aj. vj.k + aj2f2k + + aj.jka (1) E.. associated with O. . it has had close connections with psychometric studies. Significance tests for these effects and estimates of various parameters are given by using statistical tlinear models. especially principal component analysis.. Oj) marks one of the seven points 3. canonical correlation analysis.i=l. In all the above cases. B..3.... next.. A. Oi is slightly preferable to Oj. by representing those objects by a small number of indices or by points in a small-dimensional Euclidean space.. N.. Oj is slightly preferable to Oi. p.)=O. 2. A judgment is recorded on a 7-point (or 9-. Triangle Test Test. F. Mathema+. -1. 5-. The methods are as follows. Duo-trio test: A judge is first acquainted with a sample A and then is requested to choose from A and B the one he has seen in the previous step.. The pair test. Oj is presented to 2n judges. and the remaining n examine the pair in the opposite order. k= l. +PBIBD. Pair test: A judge is requested to designate a preference between the paired samples A and B.ically. Historically. The method developed by Torgerson and also the INDSCAL method by Carrel and Chang (1970) are called metric multidimensional scaling. (iii) V(vj) = 1 and Cov(vj. which can be regarded as the sum of a main effect. j = 1. The Thurstone- wherefi. . deviation of subtractivity. The former is applied when the data are represented in continuous scales and the latter when the data are in discrete nominal or ordinal scale. Some of them require judgments concerning a particular attitude that is considered unidimensional. 0. Techniques of multidimensional scaling are closely related and sometimes actually equtvalent to various methods of multivariate analysis. N. In the 7-point scaling system a judge presented with the ordered pair (Oi. Historically.. is denoted by X. Obtaining the tmaximum likelihood estimator of ni. in both theoretical developments and applications. that Oi is strongly preferable to Oj. Psychometric scaling methods are procedures for constructing scales for psychological phenomena. .

2. If the response probability is path independent. . Hence. YJ. the manner of their correspondences depends on each trial. Lute’s P-model [9]. each of which corresponds to either response A.) that influence the ensuing behavior. xi=. t). 7ti = 1). The Bush-Mosteller model assumes the response probability to be path independent. and oh. Under the condition of event commutativity. ... .. We can consider the stimulus as a set composed of m elements.).-~ = Ej. or A. . respectively (both positive).XA( 1 . where F. then &. . = 1. responses. . if A./m.... Example (3). the condition of event commutativity is satisfied.i..) (v = i. As a result of the nth trial. among which those often used are the MINRES by Harman (1967).j. it is called path independent. test scores. (2) Determination of factor loadings. .. which is the estimation of the aji: A number of methods have been proposed.Nf(Pl) can be obtained. . . and R is the correlation matrix of the Z’S G... Then the recursive formula for P. The recursive formula is expressed as fi(P. gl).. Y. .. &‘” = Ei.) = aiPn + (1 . . and A.. A necessary and sufftcient condition for Ei and Ej (ifj) to be commutative is that either fi or fj be an tidentity operator or li = S. 8.).. . . is expressed as un/(un + ub). Furthermore.. In this model. . is of the form Pn+l =f(Pn.Y.). PI). and they are frequently represented by recursive formulas for response probabilities. = J. n).. If the formula can be written as f=f(P.. (3) Estimation of factor scores fi.. . The response Then it follows that and where hi” is called the communality of thejth variable zj and uf is called the specificity. and assume that P. Assume that a sequence of trials is done in order to study some given behavior and that on each trial particular events occur (stimuli. we write P”=F(n. Then the behavior itself is modified by such a sequence of trials. Learning models refer to such processes of behavior modification. Other linear models have been proposed in which the response probability is either quasi-independent of path [8] or path dependent [lo]. Example (2).. Suppose that on the nth trial s ( <m) elements are sampled.) elements correspond to A. Z. the response probability of A.. A are the matrices of factor scores.2.j.. In the recursive formula J two events Ei and Ej (i #j) are said to be commutative if~.. and that an event Ei occurs on the nth trial with Pr(& = Ei) = xi (i = 1. + {Xx Y. assume that J n+l = Jn +X.-~. =fylfp . called d-trial path dependent. In a linear model. .. &I ... Bush-Mosteller model [6].cc&.. . (3) Nonlinear Models.XA( 1 . on the nth trial be u. then the response probability is Gl.k(Pn) (ij. 6$. otherwise.1) trials (& Ni = n .. When the recursive formula can be expressed as f=f(P.. the response probability is path independent and 8” = (X. t. k). reinforcements. .).X. fk(Plb where f. we set Y. among which X. N. = 0. =Ek)=Jj..(P.2. we set Y.. j. ) with respective probabilities P. & &. is reinforced. . and A.k= 1. the explicit formula can be written P.(P. and the remaining XL ( = s . the response probability is said to be quasiindependent of path [ 111. ... CI?“.1287 346 G Psychometrics f(P.. respectively. N. . = F(N. . Learning Theory (1) General Description. .. . . . P.. then we have P.. we obtain the recursive formula Pn+l = P..(Pn) =I&. Problems of factor analysis are classified into three types: (1) Estimation of communality: There are several methods of determining communality or initial estimates of it when some iterative procedure is used. the varimax by Kaiser (1958). . In nonlinear models the recursive formula cannot be written as a linear function of P.i.. . where Ni is the frequency of occurrence of Ei in the first (n . c?“= I$.f. elements correspond to A. Let the response strengths of A.. etc.(R) = . If any two events are commutative. .. By making f explicit with respect to n. . . If J. Assume that two mutually exclusive response alternatives A.)}/m. . and 1 -P. . If both event commutativity and path independence of response probability are satisfied. Example (1). elements correspond to A. on the nth trial. . . In the special case d = 0.&“-. &. occur on the nth trial (n = 1. factor loadings. .Y. For simplicity we write f(P..: Usually factor scores are estimated after factor loadings have been determined.. (2) Linear Models. Thurstone proposed p=ZR-‘A and Harman F=ZA(A’A)-‘. . the recursive formula is written as a linear function of P”. Here ai (0 < ai < 1) represents the degree of ineffectiveness of Ei for learning and Izi (0 < Ai < 1) is the tfixed point of fi. It follows easily from (1) that any orthogonal transformation of the scores does not affect the model. the explicit formula P.. G?~ = E.)=~. and the tmaximum likelihood based on the normal model by Lawley and Maxwell.1). Estes’s stimulus-sampling model [7].. CF..

The abilities of man. 308-33!). 1954. Bush and W. 5 (1940). Sternberg. R. Guttman. [ 131 R. Hanania. Psychometrika. 187-240. 27 (1962). 23 (1958). Therefore we have Pn+~ =P. Kaiser.)= /&u. Wiley. [ 171 H. This model is nonlinear. Audley and A. Stanford Univ. R. Kendall. 87-94. R. Psychol. H.). Wiley. the recursive formula of u. Estes (eds. the recursive formula for vi can be expressed as cpi(uL) = /j’iuL (/$ > 0). D. 1967.l{Pn+bi(l -4)) (bi=Pi’lPA gn = Ei.). +(l -Pl)exp(& N. R. K. Bush. Press. The statistical analysis of the learning process. if we assume ~i(u)>O for u>O and cpi(cu)=ccpi(u) for u > 0 and c > 0. N. Macmillan. [2] J. Lute. [ 191 D. Lute. The analysis of proximities: Multidimensional scaling with an unknown distance function. 1927. E. . Wiley. For fitting a model and experimental data. The method of paired comparisons. third edition. Stanford Univ. Foundations of multivariate analysis. 53-68. Brit. 1982. A path-dependent linear model. Kruskal. 1959. Guilford. [20] K. [ 141 L. Multiple rectinear prediction and the resolution into components. Psychometrika. Studies in Mathematical Learning Theory. Bush and W. Maxwell. [ 111 S. Psychometrics strengths u. Rank correlation methods. [S] M. second edition. [3] M. Theory and method of scaling. Carrol and J. Studies in Mathematical Learning Theory. The varimax critlzrion for analytic rotation in factor analysis. Sternberg. R. Chang. 125-219. trial number of first success or last error. A. with /ii > 0. I. Here. . N. and L$ depend on each trial. [4] W. 1963. expected response probabilities and various other statistics deduced from the model (total error. [S] R. 1963. F. [6] R. J. Hafner. R. K. 1959.. and the response probability is path independent. 1288 References For sensory tests. 1962. D.and E. Modern factor analysis. Psychometrika. Mosteller. H. but we can generalize to the case of more than two alternatives. of Chicago Press. Other nonlinear models in which the response probability is either quasi-independent of path [S] or path dependent [9] have also been proposed. [9] R. R. Jonckheere. K.+~ changes independently from ub.. J. = P. [ 121 J. Psychometrika. In a similar way. Griffin.). 1963. S. For MDS. A generalization of the Bush-Mosteller model with some significance tests. Takeuchi. Estes (eds. 283-319. Press. [7] W. 1958./{P. Handbook of Mathematical Psychology II.. 35 (1970).logb. Harman. [ 151 J. B. Psychometrika. J. For learning theory. Wiley. Shepard. Psychometric methods. Hence it is clear that the events are commutative. David. Component and pattern models with Markovian interpretations. 115-129. Stochastic learning theory. Estes. Yanai. R.. H. Here we have taken up only the case in which the number of response alternatives is 2. Stochastic models for learning. 1959. P. Analysis of individual difference in multidimensional scaling via an N-way generalization of EckartYoung decomposition. and sequential statistics such as length of response +run or +autocorrelation between responses) are used. Buch and F. Butterworth. we obtain P. Estimation methods have also been devised for the parameters involved. Under the assumption that the response strength is path independent and that u. [ 1S] H. N. By making the recursive formula explicit. is written as u. 9 (1956). 29 (1964). and B. 1955. &” = Ei. Statist. Univ. Torgerson.9-52. [l] H.)}. Galanter (eds. [lo] S. 75-99. Lawley and A.). H. [ 163 C. B. Individual choice behavior: A theoretical analysis. Psychometrika. McGraw-Hill. Nonmetric multidimensional scaling: A numerical method. Mukherjee. Ci. Factor analysis as a statistical criterion.346 Ref.+~ = cpi(u. l-120. then cpi(u. 24 (1959). Wiley. R.. Spearman. For factor analysis.

.

The +class number P. General Remarks Let k = Q(&). Ideal Classes Let k be an imaginary quadratic tunitsofkare +l. L. m E 1 (mod4). for m<O. we define x(l)= 1. 59 Class Field Theory. w) is a +minimal basis of k.83 Continued Fractions. w is the number of roots of unity in k.. The symbol x(n) for nE Z is called the Kronecker symbol for k. + o~(w.e. Siegel (Acta Arith.67. where (i) x:(n)= (. M. and if m=2. In particular. L. that is.43. If (n. Stark independently proved that these nine numbers are the only ones for which h(d)= 1 (Baker.14 Algebraic Number Fields). m’s 1 (mod4).8. In 1934. for a table of the fundamental unit of k for m < 100 .3 (mod 4). (4) x(n)= 1 if and only if n= N(a) (modd) for some integral ideal a of k such that (a.12) Quadratic Fields A. Muthematica. If (n. k is called a real (imaginary or complex) quadratic field (. and x(n)=*1 if(n.1. The +discriminant d of k is given by d = m in case m z 1 (mod 4) and d = 4m in case m E 2. H. If (m/p) = -.(2)X(m)=X(n)ifm=n (modd). Ed is the positive fundamental unit (> I) of k. then (2) = pp’(p # p’) and N(p) = N(p’) = 2. The Kronecker symbol for k has the following four properties: (1) x(n) = 0 ii‘ (n. +Gauss that h(d)-+ cxzas Id I+ co. (Property (4) shows that a quadratic field provides a class field. Any quadratic field k is obtained from Q by adjoining a square root of a square-free integer (i.then(p)= pp’ in o (p # p’) and N(p) = N(p’) = p. then x(n) = (n/jml).rk.1 if (p) = p in o. d) # 1. 163. If m = 5 (mod 8) then (2) = p and N(p) =4. (iii) x(p) = 0 if (p) = c ’ in O. and (m/p) is the +Legendre symbol. 1 (1935)) proved h(d)=1 holdsfor(d(=3. 13 (1966). p’ are prime ideals of o. where d is the discriminant of k. The mapping (r: a+x’ is an tautomorphism of the field k. That is. (ii) Letp#2and(p.[ 11). That is. =( 1 + fl)/2) field. +Dirichlet ods as follows: hlog+. (ii) x:(n) = (. (3) x(mn)=X(m)X(n). The Kronecker Symbol Quadratic 347 (V. d) # 1. There exists a unit E” that is the smallest one among the units (> 1) of k. where d is the discriminant of k. C. y) of +Pell’s equation x2 dy’ = + 4 by using continued fractions (.3 (mod4). an integer #O. (ii) x(p)= . G. the symbol x can also be defined using the +Jacobi symbol as follows: If m 3 1 (mod4). The fundamental unit ~~=(~+y&)/2 (> 1) can be calculated by finding a minimal positive integral solution (x. and _+ 1 in all other cases. L. and finally A. then x(n) =C#. Here p. Sta. More precisely.l)(nz-‘)/B for e=3. The conjugate element of an element a = a + b& (a. 1 with no square factor) m:k=Q(fi).3 (mod 4). +iincasem=-1. Baker and H. N means the +norm. This conjecture was proved by H.lf(m/p)=l. If m = 1 (mod 8). ’ is a ifundamental unit of k. Prime Ideals The decomposition of a prime number p in the tprincipal order D of k is given as follows: (i) Let p 1d.3. Heilbronn (1934). b E Z. we define X(-n+=(sgnd)X(n). and (iii) x:(4=(1) (mod 4). Then p is decomposed in o in the form (p) = p2. m’ E 3 (mod 4).11. F. and (iv) x(n) = ni x(pi)‘i for n = ni p. C. and x is the Kronecker symbol.= -idi I 1 h=Gldir I I x(v)r h of k was calculated by (1840) by analytical meth- &)logsinF for m>O.7. B. bE Q) of k over Q is given by a’ = n -b&.d)= 1. and let d be the discriminant of k.m)=l.(d)) = 1. Let (1 + fi)/2 (O= t fi for mz 1 (mod4).) (n*~lm+(n-1)/2 Any tcxtension field of the rational number field Q of degree 2 is called a quadratic field.m'c3 Then (1. for e=3. in case nr = . Heilbronn and E.19.d)=1. then (p)=p in o and N(p)=p’. any talgebraic integer x of k has the unique expression x = u + bw with a. for m = 2. (iii) Let 2/ld. Michigan Math . It was conjectured from the time of C.et k be a real quadratic field.4. then x(n)= xz(n)(n/lm’j) for d=2em’.347 A 1290 Fields D. Denote by h(d) the class number of the imaginary quadratic field with discriminant d.I)(“-‘)‘~ for e = 2.. We define the symbol x for k as follows: (i) x(p)= 1 if (p)=pp’ (pfp’) in 0. For a negative integer -n. Any unit E of k can be uniquely expressed in the form F:= + a$ (no Z). + uo. *E. Linfoot proved that there can be at most one more such d. IJnits E. If m is positive (negative).i 3>0. The fl.

232. each ideal class is divided into two ideal classes in the narrow sense. d) = 1. and B”= B. . A’.sdis the fundamental unit of k = a(.) = 1. . EJ to be a character system for some genus is that ~~=fland~~~~. and 9 is a subgroup of the ideal class group (r of k.1. .d)= 1. t) are uniquely determined. 14 (1967)).. .235.2).35. The tconductor of H is said to be the conductor of kJQ or simply the conductor of k. we define x1(4. A”= A3.. Each coset of (E modulo 43 is called a genus of k. G.40. B2 = E. (i) H. . .2. A*. the principal genus is 8 = {E. Then (Ed. . . 21 developed the arithmetic of quadratic fields systematically by introducing the Hilbert norm-residue symbol (..115. . For each genus. the Kronecker symbol x is expressed by x(a)=g $ for(d)= 1. Norm Residues A quadratic field k is the tclass field over Q for an ideal group H.20. (This notion is a special case of the notion of ideal classes in the narrow sense of algebraic number fields. = p forpId. . where A4 = E. it follows that n = N(a) (mod d) for an integral ideal a of k is a necessary and sufficient condition for x1(n) . . A’}.(n)= 1 to hold for ncZ with (n. =x.. In order that xl(n)= . E.I.. For real quadratic fields. = 1. Put &i = xi(N(a)) (i = 1. d) # 1. A3B}. B. The conductor f = n. . . t). From this it follows that there are t . History The arithmetic of quadratic fields was originally developed in terms of the theory of binary quadratic forms with rational integral coeficients by Gauss and Dirichlet [2].) In the cases (i) m < 0 and (ii) m > 0. 0 for (a. the usual classification of ideals and classification of ideals in the narrow sense are identical. A’B}.))/(log Jli. 29 (1975)) that h(d) = 2 holds only for IdI = 15. Since x1 xz . By means of the tHilbert normresidue symbol. For example. it is necessary and sufficient that C = C: --d = C: hold for some ideal class C.) Et)=(l.91. . W. . we have 2-i (log(h(d)log&. The theory was then translated into the terms of ideal theory by J. . Dedekind [2] (. Also. and the class number formula was obtained by Dirichlet as a formula for binary quadratic forms. Each ideal class C such that C” = C is called an ambig class of k. . .24.. it is not yet determined whether there exist infinitely many d with h(d) = 1 (.13. xt is equal to the Kronecker symbol.52. 1. When m > 0. A./$ (E. The set fi of all ideal classes of k such that (El . .andfz=2’for2jd. A necessary and sufficient condition for (cl. AB. t). . . N(E.’ .. > 1).(n)= 1 for neZ with (n. In order that an ideal class C belong to the principal genus. f. . . (d)) = 1.88. for pi = 2.[1.) = . where . . Baker and Stark proved independently (Ann.. Table 4). Then each ambig ideal is uniquely expressed in the form a = p.348 Quadratic Forms M).. Math.5. and they form an Abelian group of type (2. and each genus is uniquely determined by its character system. We call (cl. .1291 347 H Quadratic Fields J.“*. (2) 94 (1971). the p-conductor f. Each coset of G modulo H is called an ideal class in the narrow sense. For example. h= Q(n) 8.187.. . 148. = 1. . There are 2’-’ ambig classes. However. ..(n) as follows: For pi # 2.. it is necessary and sufficient that n = N(a) (modd) for an integer tl of k (where N(a) = a~‘). for m < 0.2). Hilbert [4. Math.. . p. That is. and the ambig classes are {E.. we define x.. and a/$ is an Abelian group of type (2.1 tinvariants of the ideal class group 6 of k that are powers of 2. . we identify xi with XT in the definition of the Kronecker symbol. . . A’B. t=3. . Let (pi) = pf (i = 1. Later the arithmetic of quadratic fields assumed the aspect of a simple example of class field theory (. Hence there are 2’-’ genera of k. . pt be the set of all prime numbers dividing d. Each ideal a of k with au= a is called an ambig ideal of k. of k = Q(Ji)isgivenbyf=dform>Oandf= dp.p#2. .267. . . . and 427. the theory of genera for quadratic fields explained in Section F was first developed by Gauss in terms of binary quadratic forms. . We call an ideal class in the narrow sense simply an ideal class. N(E. m’) = 1). x. . .. . x.(n) = (n/pi). CL= {E. Comp. . d) = 1.123. where a is an integral ideal with (a. Each ambig class contains exactly two ambig ideals of the form p . EJ the character system of this genus. Let p1 .. R. Ed)is uniquely determined for the id&al class C containing a and does not depend on the choice of a. Genera Let G be the group of all (tfractional) ideals of k. .) 1) is called the principal genus of k. F. and let H be the group of all tprincipal ideals (a) of k such that N(a) > 0.d=2’m ((2. ch. For n EZ with (n.5. for k = we have d= -2’.14 Algebraic Number Fields G..403.Appendix B.59 Class Field Theory).2.6]). A3.. the values ci = Xi(N(a)) (i = 1.51. pft(a) by some a~ Q and vi = 0. B.

IXVIII (1897). R. . Vorlesungen iiber Zahlentheorie I. i. . [4] D. then a form of rank r is equivalent to the form CI=l x”. . if the rank of Q is Y. On the other hand. y) is nondegenerate (i. with an n x m matrix P). we have A(Q) . Jones. x.pandq are uniquely determined by Q (Sylvester’s law of inertia).e. a. L.e. .{ 0}.. .256 Linear Spaces). . 1894 (Chelsea. 1967. Then if Q is (equivalent to Q’. R). b-’ E (K*)2. Math. we say that Q is equivalent to Q’ over K (resp. Real Quadratic Forms Now let K be the field of real numbers R. Hasse.63-363. then Q is equivalent to a form Cf=. 1969). 4.. Hence. [3] H. If Q is of rank r.) [S] B. W. Y) is a tsymmetric bilinear form on V.. Verein. This gives rise to a mapping X-Q(I) of I/ into K.x? over K (ui # 0.. Viewig. 1. Given an element p in K*. We say that Q is nondegenerate if B(x. The matrix A = (uik) is the matrix of the quadratic form Q. the discriminant of Q. . G. Q(x) = 0 for some nonzero vector :r. of America and Wiley. Number theory. Assoc. Using the notation for the tinner product of vectors.x. 1932.. i = 1. a. fourth edition. For a vector x in V whose coordinates are x1. Another important special case is that of n = m. General Remarks A quadratic form Q is a quadratic homogeneous polynomial with coefficients in a +field K. Vorlesungen iiber Zahlentheorie. In particular. we consider the quadratic form Q’ defined by Q’(~~. Consider a linear substitution xi = Cjm& PijxJ (i. then Q must come from a quadratic form (-. herausgegeben und mit Zusgtzen versehen von R. x. .) The rank of the matrix A is called the rank of Q.1 5) Quadratic Forms A.A(Q’). Putting uik = ski = cik/2 (i < k). Hirzel. Equivalent forms have the same rank. instead of 1Al. This gives rise to an equivalence relation. 1969). 7).A. 1927 (Chelsea.)= c 1biQkSn cikxjxk. y) is called the symmetric bilinear form associated with Q. we say that Q represents Q’ over K (resp. (Sometimes. 1950. we call ( -l)n(n-‘)‘22”l Al. The arithmetic theory of quadratic forms.e. Such a mapping satisfies the following two conditions: (i) Q(ax)=a*Q(x) ((1E K). (Original in Russian. We call (p. Complex Quadratic Forms [l] Z.. Math. Generally. Any quadratic form Q represents 0 by taking the zero matrix as P. 175-546.ifIPI#O(IP( is an invertible element of R). In particular. if 1A I# 0). R). . Dedekind. III. G. H. . 1 C. 348 (III. hence over C two forms of the same dimension are equivalent if and only if they are of the same rank. Hilbert. 1966. B. Die Theorie der algebraischen Zahlkcrper. by the expression “Q represents 0 over K. Landau. Borevich and I. If the coefficients cik belong to the field of real (complex) numbers. and (ii) Q(x + Y) . the equivalence of the forms is equivalence over K. (Gesammelte Abhandlungen I. Two quadratic forms of the same dimension . Then Q represents p if and only if Q’ represents 0. Let V be an n-dimensional vector space over K... We assume that the tcharacteristic of K is not 2. written Q(xl. B(x. then it represents any element of K*. Jber. 1950.Q(Y) = B(x. q) the signature of Q. [6] E. Springer.x~+~)=Q(x)--px~+~. Conversely. Vorlesungen iiber Zahlentheorie. Springer. Shafarevich.).Q(x) . Then we get a new quadratic form Q’(x’) with the matrix ‘PAP.Here. n). we put Q(x) = Q(x 1.Ax)=‘xAx.y)=‘xAy. we call Q a real (complex) quadratic form. = cii (i= 1. If each pij belongs to the field K (to a subring R of K that contains the unit element of K).” we usually mean the nontrivial representation of 0 over K by Q.y)=(x. we assume that the coefficients of the quadratic forms and the coordinates of linear transformations are all contained in the field K. x = Px’. When we specify a field K. The problem of representing numbers by a quadratic form (representation problem) is the particular case corresponding to m = 1. if a mapping Q: V+K satisfies these two conditions.) [2] P. we write a-b if a. and the determinant 1 A 1 is the discriminant of Q. Deutsch. Then the discriminant of Q’ is given by IPI’(AI. lxi+j(p+q=r). 2-‘B(x. If Q is nondegenerate and represents 0. . we have Q(x) = x& uitxixk. Academic Press. we can write Q(x) If K is the field of complex numbers C. A basic problem in the theory of quadratic forms is to determine the exact conditions under which a given quadratic form Q represents another quadratic form Q’. denoted by A(Q). for elements a and b in K* = K . then it is equivalent to the form cfL’=. Chelsea. II. 1964.347 Ref. xf 1 x4= . Dirichlet.. Quadratic References 1292 Fields =(x.

then W=Z/2Z. where 1 <r<n. x. Next. @ Q. =O.. is called the Witt decomposition of Q (E. D. then Q represents 0. then W~(Z/2Z)+(Z/2Z)ifq=l(mod4). A(Q)-A(Q’). are equivalent and Q1 @ Qz and Q.+i. x. Q is negative definite if and only if -Q is positive definite.. Each of the following conditions is necessary and sufficient for a form Q to be positive definite: (i) for any nonzero real vector x we have Q(x) > 0. ..... B(x.x. and another form Qz with variables x”+i.. We say that x is isotropic if B(x.+i = . y) = 0 for all x. We say that Q and Q’ belong to the same type if Q. The quadratic form xi x2 +x3x4 + .OQ.x.e.. (0. .+Q. . r)). suppose that Q and Q’ are nondegenerate quadratic forms over the tp-adic number field K. and Qb are equivalent and denote the set of types of nondegenerate quadratic forms over K by W. and let C*(Q) denote C(Q) if n is even and C’(Q) if n is odd. . . Quadratic Forms over a General Field K The following facts are valid on any field K whose characteristic is not 2. Any nondegenerate quadratic form Q(x. is called the direct sum of Q. where if Q. . . x.ifK=R. (Here M.. q). x) = 0. since some authors call the number p-q or p or q the index of Q. @ Qb be Witt decompositions of Q and Q’..e.(~~+~. Then P is an torthogonal matrix. and they have the same Minkowski-Hasse character x. To make the distinction clear.. Let N. . a. . . . N. . .Q.) is equivalent to the direct sum of a kernel form N. E. otherwise it is an indefinite quadratic form. A form with n variables is called a positive (negative) semidefinite quadratic form if its signature is (r. . A linear transformation x’+x = 6%’ that leaves invariant the unit form Cy=i XT is an orthogonal transformation. . Quadratic Forms over Finite p-adic Number Fields Fields and Let Q and Q’ be nondegenerate quadratic forms over the tlinite field F. n=O (mod2).. if Q has rank not less than 5. =x. .W~Z/4Z if q = 3 (mod4). Here a.+1. are also equivalent. Thus a subspace W is totally singular if and only if it is totally isotropic (i. then Qz and Q. Let B be the symmetric bilinear form associated with Q. we get a new quadratic form Qi @ Qz or Q.).. . . are uniquely determined by Q up to equivalence. then W is a finite group. n=O (mod2) and p-q=0 when K=R. . . .). if K = R and (p. if the rank of Q is not less than 3.+. Two forms Q and Q’ are equivalent with respect to an orthogonal transformation if and only if the corresponding matrices have the same eigenvalues. Any quadratic form can be transformed into a diagonal form CG1 a. The number r is called the index of Q. A quadratic form with the signature (n.A(Q’). n)) is called a positive (negative) definite quadratic form. The matrix of Qi 0 Qz is the direct sum of the matrices of Qi and Q2.) Also. (0. They are equivalent if and only if they have the same rank and A(Q) . we have Q. Q~(x~. An element x in I’ is said to be singular with respect to Q if Q(x) = 0. N. x) = 0 (characteristic of K # 2)... 0) (resp. respectively. nz 0 (mod2) and A(Q).x. is the total matrix algebra of degree t over K and D(K) is the unique tquaternion algebra over K. x.~Q.xf via an orthogonal transformation. 0) (resp.(x. . If Qi and Q. 0 Qo. are the teigenvalues of the matrix of the form. Witt [S]). and Q2.) and a form Q&.. The index I of Q is the dimension of a maximal totally singular subspace of K In particular.(x. We define the sum of the types of Q and Q’ as the type of Q @ Q’.(x2. and this gives W the structure of a commutative group. . . Moreover. The structure of the Witt group depends on K. @ Q.definedbyQ.1 when K=F9. A subspace IV of V is said to be totally singular if all the elements in IV are singular. if K = F4. are equivalent (Witt’s theorem). ..ifKisa tlocal field with a tnon-Archimedean valuation. q) is the signature of Q. Given a quadratic form Q i with variables xi. Then x is singular with respect to Q if and only if B(x. then the index r = min(p..x.)= . Then x(Q) = 1 or -1 according as C*(Q) = M. Here we must be careful.thenWzZ. and W z Z/22 if q is a power of 2. n> 2r. .(K) or M. If K = C. . then Q represents 0 in K. and Q. where 1 is defined as follows: Let C(Q) be the tclifford algebra of Q. Q is called a definite quadratic form if it is either positive or negative definite. . called the Witt group.)+Q.1. . The decomposition N. .)=0 only if xa..1293 348 E Quadratic Forms are equivalent if and only if they have the same signature. Necessary and sufficient conditions for a nondegenerate Q to be a kernel form are: n = the rank of Q = 0 (mod 2) when K = C. y E W).+... A(Q). i.+. (ii) all the tprincipal minors of the matrix of Q are positive.. . .. we also call our r the index of total isotropy. and the number p-q the index of inertia. >x.(K) @ D(K). The type of a kernel form is the identity element of this group W.#O. and x(Q)= 1 when K is a p-adic number field. They are equivalent if and only if they have the same rank. + x+~x~~ is called the kernel form and is denoted by N.

In particular.. 4.l</<m-I). and the index of inertia j. Quadratic Fields Forms over Algebraic Number Let K be an algebraic number field of finite degree. are the eigenvalues of the matrix of the Hermitian form.. T~+~x. Reduction of Real Quadratic Forms G.) The value H(x) is a real number. H.r+i>O(l .. if we have. = 3. then it is the set of invariants of a quadratic form over K (Minkowski-Hasse theorem)..m. Let S be a positive definite form in m variables. As for quadratic forms. of Q over K. U. if the system {n. i. xv.k=l p-completion of K. Then we can write Q(x) = S[x] = ‘xSX. a. uik = a. if a property concerning K holds if and only if it holds for all K. 4) is called the signature of H. K. be the Let A be an m x m matrix and X an 112 xn matrix.4]. (iii) A -( -l)(n2+j.. The notion of Hermitian forms can be generalized as follows: Suppose that K is a tdivision ring with an involution u+u (LIE K) (i.)of K. The matrix A of H is a +Hermitian matrix whose principal minors are real numbers. ai&xk is called a Hermitian form if aik E C. (p. for all p [3.. A} satislies conditions (i)-(iv). p be an +Archimedean or nonArchimedean +prime divisor of K.r. Xi are the complex conjugates of aikr xi. In this section we put K = R and define two forms to belong to the same class if they are equivalent over the ring of rational integers. we say that the Hasse principle holds for the property. Hence the invariants with respect to equivalence over K of a nondegenlzrate quadratic form Q over K are n = the rank of Q.. Hermitian Forms Forms 1294 An expression H(x) = x7. If we apply a linear transformation P(x’) =x. where a. Any Hermitian form is equivalent to a form Cp=i Xixi -C&. Any Hermitian form can be transformed via a unitary transformation into a diagonal form Cr=.... Each of the following conditions is necessary and sufficient for H to be positive definite: (i) H(x)>0 for any nonzero complex vector x. Two examples of such K having involutions that differ from the identity mapping are a separable quadratic extension K of a field I.-0.. Let K be an algebraic number field of finite degree. (ii) npxp = 1 (this is equivalent to the tproduct formula of normresidue symbols). if K is the field of rational numbers. Q and Q’ are equivalent over K if and only if they are equivalent over K. ah = ha). Quadratic forms Q and Q’ over K are said to be of the same class if they are equivalent over the tprincipal order o in K. 1.jl.4]. 2. and its matrix is a iunitary matrix. = -1 and (iv) xp. (Here Zik.<k<m.. Here the following properties hold: (i) xP= 1 for all but finitely many p. 5: 6 (mod 8) [3.+~. and Q represents 0 in K (i..(i~=l. rank. and indefinite Hermitian forms as we did for quadratic forms over the field of real numbers. we obtain a Hermitian form with respect to x’ whose matrix is given by ‘PAP.=. I. the Minkowski-Hasse character xP for +prime divisors p of K. an element u in K such that H(x) = a + a. A = the discriminant of Q. for any given vector x whose coordinates belong to K.. whereq. uik =uki.e. In general. respectively. For example. 7 (mod 8). while it is > 2 for m > 8.. aixixi. and +sesquilinear form associated with H. . and Q and Q’ be nondegenerate quadratic forms over K.!12 in K. Then a Hermitian form H over K is defined by H(x)= t riuikxk. In particular. for all nonArchimedean prime divisors p of K and (ii) they are equivalent over K.i for each +realinfiniteprimedivisorp.~E K.%. We identify the form Q with its matrix S = (sij).348 F Quadratic F.e. Q and Q’ are said to be of the same genus if(i) they are equivalent over the principle order oy in K. A genus is decomposed into a finite number Iof classes. negative definite. A linear transformation that leaves the Hermitian form Cy=. n-5 is a linear mapping of K onto itself such that Z = u. for all p. for all p. The definition of a semidefinite Hermitian form is given in the same manner as for a quadratic form. Conversely. for all the Archimedean prime divisors p of K. On the other hand. Then S is said to be a reduced quadratic form if S[gJ > skk and s. We define the notions of positive definite. where S is the matrix of the quadratic form Q. We put A[X] ='XAX.. we define the notions of the matrix of H and the discriminant... the number of classes in the genus of X:1 x2 is 1 for m < 8. Then Q represents Q’ over K if and only if Q represents Q’ over K. if j. Yixi invariant is called a unitary transformation. i = 1 ifj.. there exists a nonzero vector :c whose coordinates belong to K such that Q(x) = 0) if and only if it represents 0 in K.isan . (ii) all the principal minors of the matrix of H are positive. Class and Genus of a Quadratic Form where xig K. and a tquaternion algebra K over a field L. then we have a Witt decomposition for H.. x y.

T) be the limit of the ratio @. .. Let A(S. and in those cases the order of the product Q. pt.1295 348 K Quadratic Forms arbitrary vector whose coordinates gl. I]. Let a. = 1 otherwise). M(S) is called the measure of genus of S. the following inequalities hold: 0 < r. where S. We put Mb% T) = ‘WlJ-)+‘w.($ T) the number of the solutions of the congruence equation S[X] = T (mod q).(S. J. t) > 0. Minkowski. be the domain formed by the matrices X such that S [X] E B. T) is also known. Furthermore. if we take the identity matrix E(“‘) of degree m as S.-’ Mb% T) M(S) > 4dS> T)= For the case m = 3. n= 1: M(S)=- 1 1 E(S. which is denoted by H(S). and let B. . T)&cr. T) = ~A.= l/2 if m = n > 2. putting n = 1. Let O(S) be the set of all real m x m matrices W for which S[ W] = S.. ...)/u(B) of the volumes of B and B.. the genus of E(“‘) contains only one class. regarding the number of possible ways to express a natural number t as a sum of m squares.. r11r22. A special case of Siegel’s theorem was proved by H. If q is a prime power p”. Math. III]. Tamagawa used the theory of tadelized algebraic groups and proved that the tTamagawa number t(SO(n. T) be the number of rational integral solutions for the equation S[X] = T. --IS1 =r*. pt.@. Given a natural number D. Hence the number of classes of quadratic forms with rational integral coefficients and discriminant fD is also finite. respectively (m 2 n).(S. T.where. Also. Batemann.(S. is a domain in the space of dimension mn formed by all m x n matrices. J. ..8. g. let us consider a domain B in the Euclidean space of dimension n(n + 1)/2 formed by the set of n x n symmetric matrices containing T.@.J) E(Sl) E(S.. and T(S) is a discrete subgroup with a finite number of generators. In particular.(S. The set of positive definite quadratic forms H such that S-’ [H] = S forms a variety of dimension n(m .P. T) have been calculated.. sot.. it is known that if t is odd and A(Ec3). Amer..) = 1.) +. as the domain B shrinks toward the point T. m . L.T). Units Let S be a symmetric matrix of degree m with rational coordinates.. 71 (1951)). pt. . Then Siegel’s theorem states: CL. where c(m) depends only on m. gm are integers such that (gk. S)) of the special ortho- .. T = t ( = a natural number). we obtain from Siegel’s theorem the number of ways in which we can express t as the sum of m squares [6. Except for a finite number of p. but it was C. The set of all symmetric matrices of degree m forms a linear space of dimension m(m + 1)/2 in which the subset ‘$3 formed by the positive definite symmetric matrices is a convex open subset. Siegel [6] who proved it in its general form.<r.3. Siegel’s result was generalized by Siegel himself to the case where the form S is indefinite [6.(S.. T) for sufficiently large a (where E. G.~=2ifm=n+l orm=n>2 and E= 1 otherwise. Let S be an indefinite quadratic form whose signature is (n. f2r. Jacobi for the case where m=4. . the following formula was obtained by C.. The thomogeneous space o(syr(s) is of finite measure with respect to a tHaar measure defined on the space. For a reduced form R = (rkl). for a natural number q. II] and where the coefficients of the forms are elements of an algebraic number field of finite degree [6. Then B. but otherwise it is infinite (except for the case m = 2. O(S) is a tLie group. the numbers cr. is a complete system of representatives of the classes in the genus of S. T(S) is a finite group if S is definite. the subset ‘% formed by all reduced positive definite symmetric matrices is a convex cone whose boundary consists of finitely many hypersurfaces and whose vertex is the origin.” q-mn+“(n+1)‘zA4(S. . with rational r). An element of T(S) is called a unit of S. Moreover. S. For m = 2.. then the ratio ht. 1 < rz2 < ..)+E(S. . T. then the formula in Siegel’s theorem is related to the problem of expressing natural numbers as sums of m squares. and T(S) be the subset of O(S) consisting of the integral matrices.)+. Hence. K. On the other hand.n). The infinite product of the left-hand side of this equation does not converge absolutely if either m = n = 2 or m = n + 2. The explicit form of tl. there are only a finite number of definite or indefinite reduced quadratic forms with rational integral coefficients whose discriminant is +D. Any class of positive definite quadratic forms contains at least one (and generally only one) reduced form. details . and E(S) be the order of the group of units r(s). Minkowski-Siegel-Tamagawa Theory Let S and T be rational integral positive definite symmetric matrices of degree m and n. T) takes a constant value SC. we denote by A. n = 1. We say that S is reduced if H(S) fl % # 0. then t + 7 (mod 8) (for Trans. k<l. < r.n).rmm<c(m)~R~. is considered to be the natural order of the primes p. .

[S] E. These series are called theta series. Math. . Hecks showed that A(n) = A. Hermann. Eichler. W. D(Q) = -4A(Q)).fi)/2. II. + u. Springer. and . [4] 0. For D < 0.=r>O. c. d are integers such that ad . Bourbaki. Theta Series Let Q(xi. y) = ax* + bxy + cy*. there is a one-to-one correspondence between the tideal classes of k and the classes of quadratic forms with discriminant d (when D < 0. ue) be the smallest positive integral solution of t* -Du’ = 4. The correspondence is given in the following manner: If a is an ideal in k with a basis x. if m = 2k.x+a. . we again have a one-to-one correspondence between the classes of ideals of this order and the classes of quadratic forms [3]. . s.n)#1. where N is the absolute +norm. we consider the classes of positive definite forms). [6] C.4. 230263. y are rational integers.a. If Tm z > 0. Theorie der quadratiljchen Formen in beliebigen Kiirpern. Uber die analytische Theorie der quadratischen Formen I. where . x2. For a complex number z..): where (D/n) is the +Kronecker symbol (here D =f2d. n=O Moreover. [2] N. the order w.) = II. Binary Quadratic Coefficients Forms with Integral With respect to the numbers else is known. and E is a character mod N.=(1+ $)/2 if d = 1 (mod 4). Springer. Wiley. He also showed that from this fact. Then the following formula holds (Dirichlel.sD. Math.*. Introduction to quadratic forms. we put Fk Q) = c expWiQ(x. -cu (t+bu)/2 >. 38 (1937). and 2 otherwise. where xi. we can introduce the notion of proper equivalence as follows: Q and Q’ are properly equivalent if the matrix of Q is transformed to the Imatrix of Q’ by a linear transformation P whose determinant is 1.(n) is a number-theoretic function of n determined by the genus of Q. the theory concerning Q is closely related to the arithmetic theory of the iquadratic field Q(3) = k. we have the following transformation formula: = ~(d)(cz + d)kF(z.13 Algebraic Groups P) [lo]. Let (t. u) be an integral solution of +Pell’s equation t* Du* = +4. Using the theory of modular forms.348 L 1296 Quadratic Forms plained shortly).bc = 1. of the units is known: it is 6 if D = -3. T. 176 (1937).y).. be the class number in I:he finer sense of the forms of discriminant D. h.(n) + O(nk/*). c z 0 (mod N). When f= 1. The arithmetic theory of quadratic forms. 1952. Q) = t . x. b. we put D(Q) = b* -4ac and call it the discriminant of Q (i.. If f > 1. 9. Ann. E. We also have M. [3] M.ifwetakea. y) = ax2 + bxy + cy* with discriminant D are given by + (t-bu)P ( au gonal group is 2. Q. 1963.:=s+tw.f That is. Witt. Jones. 1272a. x. 4 if D = -4. 37 1:1936). J. Ind.. run over all the integers. Then the units of the form Q(x.t>O (r.. Actualites Sci. Let d be the +discriminant of k and put D=df*. and let h. Q) is a tmodular form with respect to the tcongruence subgroup of level N. We let w=$/2 if d=O (mod4).)4. 31-44. 11. we must replace the ring of integers o by the +order of the tconductor . . ch.and(D/n) = (d/n) if (J n) = 1). then we get a relation between the classification of the forms and the classification in the finer sense of the ideals in k Suppose that D > 0 is not a square... the series converges and represents an tentire function of z. In other words. Reine Angew. If we denote by A(n) the number of integral solutions of the equation Q(x. When D > 0. 212-291 (Gesammelte Abhand- . in the correspondence for the casef=l.4(n)e2ni"z.e. When D(Q) is not a square. c) = 1. where ~1. III. Quadratische Formen und orthogonale Gruppen. Algebre. 1950. Given a form Q(x.weput(D/n)=Oif(f. then the corresponding form is given by Q(x. Elements de mathematique. . . teQ). 1959. F(z. if we consider the ring formed by the elements x +fjiw (x. (2) 36 (1935).b. O’Meara. Siegel. Let (to.y)= N(a)-‘N(a. N is a natural number determined by Q. L. L. x. we have F(z. put eD =(t. Then. . 527-606... Siegel’s theory in this section can be deduced (. the meaning of w is ex- References [1] B. Q is said to be primitive if (a. little Now put m = 2. xm) be a positive definite form with integral coefficients.

1297

349 c Quadratic

Programming

lungen, Springer, 1966, vol. 1,326-405,410443,469-548). [7] C. L. Siegel, On the theory of indefinite quadratic forms, Ann. Math., (2) 45 (1944), 577-622. (Gesammelte Abhandlungen, Springer, 1966, vol. 2,421-466.) [S] E. Hecke, Analytische Arithmetik der positiven quadratischen Formen, Kgl. Danske Videnskab. Selskab, Mat.-Fys. Medd., XIII, 12 (1940). (Mathematische Werke, Vandenhoeck & Ruprecht, 1959,789-918.) [9] G. L. Watson, Integral quadratic forms, Cambridge Univ. Press, 1960. [lo] T. Tamagawa, Adtles, Amer. Math. Sot. Proc. Symp. Pure Math., 9 (1966), 113-121. For binary quadratic forms, [ 1 l] P. G. L. Dirichlet, Vorlesungen tiber Zahlentheorie, Vieweg, fourth edition, 1894 (Chelsea, 1969).

When D is nonnegative definite, any feasible solution of the above system of equalities gives an optimal solution of the primary problem (Q), and when D is positive definite, the solution is unique. When D is not nonnegative definite, the optimal solution of(Q), if it exists, is one of the feasible solutions of(C). The last line of(C) implies that the solution must be a basic solution of the linear system of equalities, and it also restricts the possible combinations of the basic variables. Since there exist only a finite number of possible combinations of the basic variables, the quadratic programming problem can be solved in a finite number of steps, if it has an optimal solution.

B. Duality The dual problem of(Q) is the following. (QD) Minimize w = b’y + +x’Dx under the condition A’y + Dx > c and x > 0, y 2 0. If D is nonnegative definite, the following theorem holds. Theorem: If the primary problem (Q) has a solution x = x*, then the dual problem has a solution x = x* and y = y*, and max z = min w. A more general form of the quadratic programming problem can be given as follows. (Q) Maximize z = c’x -$x’Dx under the condition XE V and b-Axe W, where V and Ware closed convex cones in R” and R”, respectively. Then the dual problem is expressed as follows. (QD) Minimize w = b’y - +x’Dx under the condition XE V, YEW* and A’y+ Dx-CE V*, where V* and W* are the dual cones of V and W. The above theorem holds for both (Q) and

**349 (X1X.3) Quadratic Programming
**

A. Problems A quadratic programming problem is a special type of mathematical programming (- 264 Mathematical Programming) where the objective function is quadratic while the constraints are linear. A typical formulation of the problem is as follows. (Q) Maximize z = c’x -4x’Dx under the condition Ax < b and x > 0, x E R”. Let the Lagrangian form for this problem be q(x, 1) = c’x -)x’Dx + I’(b - Ax).

Then, from the general properties of Lagrangian forms, the following theorem can be proved. Theorem: If x =x* is an optimal solution of the problem (Q), there exists a vector 1* satisfying the conditions -Dx*+c<L*, b’l* = (c - Dx*)‘x*. b*>O;

(QW

C. Algorithms Various algorithms have been proposed for quadratic programming [ 1,2,4], most of which are based on condition (C). Wolfe [4] proposed a method based on the simplex method for linear programming. If we introduce the artificial vectors 5 and q, we can find a feasible solution of(C) by solving the following linear programming problem. (LQ) Maximize z = -l’Gl’q under the condition that Ax+u-k=b, x20, y’u=O, y>o, x%=0. A’y+Dx-v+q=c; uao, v>o, 520, q>o;

Moreover, if the matrix D is nonnegative definite, the above conditions are also sufftcient for x =x* to be optimal. The second condition can be shown to be equivalent to b*‘(b-Ax*)=0 and x*‘(A’)L*+Dx*-c)=O.

By introducing the slack vectors u > 0 and v > 0, the conditions can be expressed as x20, (C) y’u=O, yao, Ax+u=b, x%=0. u>o, v>o;

A’y+Dx-v=c;

(LQ) can be solved by applying the simplex algorithm with the only modification being

349 Ref.

1298 Programming B. Classification The subset defined by equation (111may be empty; for example, x2 + y2 + z2 + 1 = 0. In this article, we consider only quadric surfaces that are not the empty set. When a quadric surface F without tsingular points has a center or centers, we say that F is central. If we choose a suitable rectangular coordinate system, the equation of a central quadric surface is written in one of the following forms:

Quadratic

that the last line of the condition restricts the possible changes in the basic variables. When D is positive definite, we can always obtain a solution if there is a feasible solution of the original problem, and Wolfe proposed a moditication of the foregoing algorithm for the case when D is nonnegative definite that tells whether or not it has an optimal solution, and gives it if it has one. Some other algorithms are also effective when D is positive or nonnegative definite, but when D is not nonnegative definite, no simple effective method has been found to reach the optimal solution even when its existence has been established.

(2)

(3)

**References [l] E. M. L. Beale, On quadratic programming, Naval Res. Logistic Quart., 6 (1959)
**

221-243. (4)

(5)

x2

a2

A quadratic programming procedure, Naval Res. Logistic Quart., 4 (1957), 79-85. [3] W. S. Dorn, Duality in quadratic programming, Quart. Appl. Mat., 18 (1960) 1555

162. [4] P. Wolfe, The simplex method for quadratic programming, Econometrica, 27 (1959), 382.-398.

[2] C. Hildreth,

y2

bd

(6)

(7)

**350 (VI.1 0) Quadric Surfaces
**

A. Introduction A subset F of a 3-dimensional Euclidean space E3 is called a quadric surface (surface of the second order or simply quadric) if F is the set of zeros of a quadratic equation C(X, y, z)=O, where the coefficients of G are real numbers. The equation G(x, y, z) =0 is written as

ax2+hy2+cx2+d+2fyz+2yzx+2hxy+2f’x + 2y'y + 2h’z = 0. (1)

When the equation takes the form (2), (3), (4), (5), or (6), we call the quadric surface an ellipsoid, hyperboloid of one sheet, hyperboloid of two sheets, elliptic cylinder (or elliptic cylindrical surface), or hyperbolic cylinder (hyperbolic cylindrical surface), respectively. When the equation takes the form (711,the surface coincides with a pair of parallel planes. If a = b in (2), (3), (4), or (5) the surface is a tsurface of revolution with the z-axis as the axis of revolution. In this case, we call the surface an ellipsoid of revolution, hyperboloid of revolution of one sheet, hyperboloid of revolution of two sheets, or circular cylinder (or circular cylindrical surface), respectively. If a = b = c for an ellipsoid of revolution, then t ne surface is a sphere with radius a. If we choose a suitable recl.angular coordinate system, the equation of a noncentral surface of the second order is, written in one of the following forms:

In general, a straight line intersects a quadric surface at two points. If it intersects the surface at more than two points, then the whole straight line lies on the surface. Suppose that we are given a quadric surface and a point 0. Suppose further that we are given a straight line passing through the point 0 and intersecting the quadric surface at two points A and A’. If A0 = OA’ for all such straight lines, then the point 0 is called the center of the quadric surface.

(8)

(9)

(10)

1 When the equation takes the form (8), (9), or (lo), we call the surface an elliptic paraboloid, hyperbolic paraboloid, or parabolic cylinder (or parabolic cylindrical surface), respectively. If a

1299

350 D Quadric quadric

Surfaces cones (3’)

= b in (8), the surface is called an elliptic paraboloid of revolution. Among these, (2), (3), (4), (8), and (9) are sometimes called proper quadric surfaces, and the others degenerate quadric surfaces. Equations (2)-( 10) are called the canonical forms of the equations of these surfaces (a, b, c in canonical forms should not be confused with a, b, c in (1)). The planes x=0, y=O, and z=O in surfaces (2), (3), and (4) and the planes x = 0 and y = 0 in surfaces (8) and (9) are called principal planes of the respective surfaces; and lines of intersection of principal planes are called principal axes. For a surface of revolution, positions of principal planes and principal axes are indeterminate. We call a, b, c in equations of canonical form the lengths of the principal axes, or simply the principal axes. If F is a hyperboloid of one sheet or a hyperbolic paraboloid, there are two systems of straight lines lying on F; two straight. lines belonging to the same system never meet (and are not parallel), and two straight lines belonging to different systems always meet (or are parallel). If F satisfies (3), these systems of straight lines are given by

~+y’-~=, a2 b2 c2

and

-x’-y’+“=o

a2 b2 c2 asymptotic cones of (3) and (4), respectively.

(4’)

C. Poles and Polar

Planes

If F satisfies (9), then two such systems are given by

(1 and ,U are parameters). We call these straight lines generating lines of the respective surfaces. A hyperboloid of one sheet and a hyperbolic paraboloid are truled surfaces described by these generating lines. When a quadric surface has singular points, they are double points. The set of double points of a quadric surface F is either a single point 0, a straight line I, or a plane rc. In the second case, F consists of two planes passing through 1 or I itself, and in the third case, F coincides with rc. In the first case, we say that F is a quadric conical surface (or quadric cone) with vertex 0. Its equation is written in the form Ax’ + By’ + Cz* = 0 (ABC # 0). When A, B, C are of the same sign, F consists of only one point 0. Otherwise, we can assume that A, B>O, C= -1. In this case, if A=B, F is called a right circular cone, and if A #B, F is called an oblique circular cone. Given hyperboloids (3) and (4), we call the

Suppose that we are given a straight line S passing through a fixed point P not contained in a quadric surface F, and S intersects the surface at two points X, Y. The locus of the point Q that is the tharmonic conjugate of P with respect to X and Y is a plane. We call this plane rc the polar plane of P with respect to the quadric surface F, and P the pole of the plane rc. If the polar plane of a point P contains a point Q, then the polar plane of Q contains P. In this case, we say that the two points P and Q are conjugate to each other with respect to the quadric surface. When the point P is on the quadric surface, the tangent plane at P is regarded as the polar plane of P. If the polar plane (with respect to a quadric surface) of each vertex of a tetrahedron is the face corresponding to that vertex, we call this tetrahedron a self-polar tetrahedron. If the polar planes (with respect to a quadric surface) of four vertices of a tetrahedron A are four faces of a tetrahedron B, the same property holds when we interchange A and B. We say that such tetrahedrons are polar tetrahedrons with respect to the quadric surface. Suppose that we are given a quadric surface and two planes. If the pole (with respect to the quadric surface) of one plane is on the other plane, these two planes are said to be conjugate with respect to the surface. When we are given two tpencils of planes in tprojective correspondence, the locus of lines of intersection of two corresponding planes is generally a hyperboloid of one sheet or a hyperbolic paraboloid. In particular, if the axes of these pencils of planes intersect, the locus is a quadric conical surface, and if the axes are parallel, the locus is a quadric cylindrical surface (i.e., an elliptic or hyperbolic cylinder). When there exists a projective correspondence between two straight lines not on a plane, the locus of lines joining corresponding points is a quadric surface (M. Chasles).

D. Surfaces of the Second Class A surface F in E3 is called a surface of the second class if it admits two tangent planes

350 E

1300 Surfaces and satisfy

Quadric

passing through an arbitrary straight line L provided that F fl L = 0. This surface can be represented as the set of zeros of a homogeneous equation of the second order in +plane coordinates ui, u2, us, u4. It is possible that a surface of the second class degenerates into a conic or two points. In general, quadrics are surfaces of the second class, and vice versa. As in the case of quadrics, we can define poles, polar planes, and polar tetrahedrons with reference to surfaces of the second class. Four straight lines joining corresponding vertices of two tetrahedrons polar with respect to a surface of the second class are on the same quadric. We say that two such tetrahedrons are in hyperboloid position. E. Confocal Quadrics

$=&

~-jk

If Pi, P2; Qi, Q2 are corresponding then Pi Q2 = P2Q, (J. Ivory). F. Circular Sections

points,

When the intersection of a plane and a quadric is a circle, the intersection is called a circular section. In general, circular sections are cut off by two systems of parallel planes 1:hrough a quadric. The point of contact on the tangent plane parallel to these is an +umbilical point of the quadric. G. Quadric Hypersurfaces

**A family of central quadrics represented by the following equations is called a family of confocal quadrics: 2+yz+;2=1
**

a+k b+k cfk ’ a>b>c>O, (11)

where k is a parameter. For a quadric belonging to this family, any point on the ellipse x’/(a - c) + y’/(b -c) = 1, z = 0 or the hyperbola x’/(a-b)-z’/(b-c)=,l, y=O is called a focus. This ellipse and hyperbola are called focal conies of the quadric. Given an ellipsoid F and a point X(x, y, z) not contained in the principal plane, we can draw three quadrics F’, F”, F”’ passing through X and confocal with F. These surfaces F’, F”, F”’ intersect each other and are mutually perpendicular. One of them is an ellipsoid, another one a hyperboloid of one sheet, and the third a hyperboloid of two sheets. Let k,, k,, k, be the values of the parameter k in (11) corresponding to these three surfaces. Then the coordinates x, y, z of the point X are given by

A subset F of an n-dimensional Euclidean space E” is called a quadric hypersurface (or simply hyperquadric) if it is the set of points (x, , ,x,) satisfying the following equation of the second degree:

where aik, bi, c are all real numbers. We can assume without loss of generality that the matrix A = (aik) is symmetric. Assume that A is not a zero matrix. In the case n = 2, F is a conic, and in the case n = 3, it is a quadric surface. The theory of classification of quadric surfaces can be generalized to t.he ndimensional case as follows: Let r(A*) = r* be the rank of the (n + 1) x (n + 1) matrix

A*=

x= (a+W(a+k2)(a+k3) J(b-a)(c-a) ’ J (b+Mb+Mb+k,) (a-b)(c-b) ’

=

y=

z= (c+Wc+Mc+k,) J

(a-c)(b-c)

We call k,, k,, k, the elliptic coordinates of the point X. Two points (x, y, z), (x’, y’, z’) are called corresponding points if they belong to confocal quadrics of the same kind, x2 y2 ;+r+;=

r2 (x)+(4.)+(21)2=1,

and put r(A) = r. Then we have the following three cases: (I) r = r*; (II) r + 1 = r*; and (III) r + 2 = r*. Corresponding to each case, equation (12) can be simplified (by a coordinate transformation in E”) to the following canonical forms, respectively: (I) i

i=l

I b, 1 [ 1

al, b, ... .

a,,

.

a,,

b,

. ..

arm

b,

bn

c

A

;

b,

. .

b,

bn ’ c

liXF =o,

22

1,

(II)

i

i=l

liXf + 1 =o,

r2

(III)

c+k

i

i=l

&XT + 2x,+i = 0,

a+k

b+k

1301

350 Ref. Quad& Surfaces

where (A,, . . . . 1,,0, . . . . 0) (with n-r zeros) is proportional to the teigenvalues of the matrix A. In general, we have 1 < r < n. In the cases where r = n in forms (I) and (II) and r + 1= n in (III), the hypersurface is called a properly (n l)-dimensioaal quadric hypersurface, and in other cases,a quadric cylindrical hypersurface. In cases (I) and (II), the quadric cylindrical hypersurface is the locus of (n - r)-dimensional subspaces passing through each point of a properly (r - l)-dimensional quadric hypersurface and parallel to a fixed (n - r)-dimensional subspace. In case (III), the quadric cylindrical hypersurface is the locus of (n - r - l)-dimensional subspaces passing through each point of a properly r-dimensional quadric hypersurface and being parallel to a fixed (n-r - l)dimensional subspace. For form (I) with li > 0 (i= 1, . . . , n), a properly (n - 1)-dimensional quadric hypersurface reduces to a point in E”; for form (II) with Izi> 0 (i = 1, . . . , n) it becomes the empty set. Suppose that we are given a quadric hypersurface F that is neither a point nor empty. Then the system {A,, .. . , A,} associated with F in its canonical equation is unique up to order (and signature in form (III)). Suppose that we are given a quadric surface F and a point P on F. Suppose further that if a point X other than P is on F, then the whole straight line PX lies on F. In this case, the hypersurface is called a quadric conical hypersurface (or simply quadric cone). For example, for case (I), we can take P = 0 (the origin), and the hypersurface is a quadric cone. In cases(I) and (II), the hypersurface is symmetric with respect to the origin. In these cases,a hypersurface is called a central quadric hypersurface, and in case (III), it is called a noncentral quadric hypersurface or parabolic quadric hypersurface. When we cut a parabolic quadric hypersurface by a (2-dimensional) plane containing the x,+,-axis, the section is a parabola. If Izi< 0 (i = 1, . .. , r) in form (II), then the surface is called an elliptic quadric hypersurface, and if there are both positive and negative numbers among the Li, the surface is called a hyperbolic quadric hypersurface. The section of an elliptic quadric hypersurface by a plane is always an ellipse. The section of a hyperbolic quadric hypersurface by a plane is an ellipse, a hyperbola, or two straight lines. In general, the section of a quadric hypersurface by a subspace is a quadric hypersurface on that subspace.

H. Quadric Hypersurfaces in an Affine Space

transformation of coordinates in E”. If we regard E” as an n-dimensional taffrne space over the real number field and reduce (12) to the simplest form by a coordinate transformation in the affme space, we have the following canonical forms corresponding to cases (I), (II), and (III) discussed in Section G: (I)

(II)

(S&ix:i=l w$x:-

i

j=s+l i j=s+l

x;=o,

x:+1=0,

(III)

(S,&i=l

i

j=s+l

x;+2x,+1=o,

**where 0 <s < r and r-s = t. The terms properly
**

(n - l)-dimensional, cylindrical, conical, parabolic, elliptic, and hyperbolic can be defined in

terms of this affine classification. For example, a cone is of type (I), a parabolic hypersurface is of type (III), an elliptic hypersurface is of type (II) (0, r), a hyperbolic hypersurface is of type (II) (s, t) (s, t > 0), and type (II) (s, 0) represents the empty set. A necessary and sufficient condition for a (nonempty) hypersurface to be represented by two canonical forms N(s, t), N’(s’, t’) is that (i) N = N’ and (ii) if N =(I) or N = (III), then s = s’, t = t’ or s = t’, t = s’, and if N = (II), then s = s’, t = t’.

I. Quadric Hypersurfaces in a Projective Space

Suppose that we are given a field K of characteristic not equal to 2 and an n-dimensional tprojective space P” over K. A subset F of P” is called a quadric hypersurface (or simply hyperquadric) if F is represented by a homogeneous equation of the second degree &=,,aikxixk = 0, where (x,, x1, .. . , x,) are homogeneous coordinates in P” and aik E K, A = (aik) is a nonzero symmetric matrix. The problem of classifying such surfaces is reduced to that of tquadratic forms or, equivalently, to that of symmetrix matrices in K. Two symmetric matrices A and B are equivalent if there exists a regular matrix T such that B =‘TAT (- 348 Quadratic Forms). In particular, when K is an talgebraically closed field or a treal closed field, a simple result is obtained. If K is an algebraically closed field, then the equation of the quadric hypersurface is reduced to the canonical form &, xf = 0, where r = r(A) = the rank of A. When K is a real closed field, then the canonical form is xi=,, x? - &+i xj’ = 0.

References

In Section G we considered a quadric hypersurface defined by (12) in an n-dimensional Euclidean space E” and transformed the equation to canonical form by an orthogonal

[l] G. Salmon, A treatise on the analytic geometry of three dimensions, Hodges, Figgis & Co., seventh edition, 1928 (Chelsea, 1958).

351 A Quantum

1302 Mechanics tor onto the eigenspace spanned by teigenvectors belonging to the eigenvalue a,. Suppose that A =Zna,P,. Then the hypothesis on measurement in quantum mechanics is given as follows. When an observable A is observed in a state +, one of the eigenvalues a, is found with probability proportional to (II/, P.$)‘. When an eigenvalue a, is.once observed, a :state jumps from $ to an eigenstate P,,t+bwhich belongs to the eigenvalue a,. Quantum mechanics predicts only a probability pn with which a certain value a, is found when an observable A is observed. This probability, given by (+, P,,$), is not changed even if $ is replaced by e’?$, 0 < 0 < 27~. Therefore ei”$ represents the same state as $. The set of e”$, 0 <B < 27t, for a fixed ti (11$/I = 1) is called a unit ray. If Pn is 1 -dimensional and P,,cp := cp, //cpII = 1, then ($, Pn$) = I($, (p)l’ is called the transition probability between the two states. The expectation value (or expectation) of an operator A in a state $, usually nsormalized to f,“‘p=,l, is defined to be (A)=(I), A$)= n% n A general self-adjoint operator A can be written as A = JA dP(i). When A is observed in a state $, the probability for a value to be found between /I, and 1, (&>A,: 1, included and I, excluded if P(L) is right continuous) is ($(P(I,)-P(i,)), $) (- 390 Spectral Analysis of Operators). The quantity (cp, All/) is called the matrix element of A between the two states cp and $. A state $ can be viewed as a functional $(A) = (A) on the set of all observables A (its value being the expectation), which is linear in A, positive in the sense $(A*A)>O for any operator A, and normalized: $( 1) = 1. If 0 < 1< 1 and $(A)=@,(A)+(l-1)$,(A) for all observables A, then the state Ic, is called a mixture of states $i and ti2 with weights /z and (1 - 1). If a state is not a nontrivial (i.e., I # 0, 1, $i # ti2) mixture, it is called pure. The state (A) = (+, All/) on the set of all self-adjoint operators A given by a vector $ is pure in this sense. If sup, $(A,) = +(A) whenever A, is an increasing net of positive operators with A as its limit, then + is called normal. Any normal positive linear functional on the set of all self-adjoint operators can be described by a trace-class positive operator p, called the density matrix, as (A) = tr(Ap). If {&} is a complete orthonormal set, where each & is an eigenvector of p belonging to the eigenvalue A,,, then (A) = ~ CnUh Atin). Commutation Relations variables operators

[2] G. Salmon and W. Fiedler, Analytische Geometrie des Raumes I, Teubner, fourth edition, 1898. [3] H. F. Baker, Principles of geometry III, Solid geometry, Cambridge Univ. Press, 1922. [4] 0. Schreier and E. Sperner, Einfiihrung in die analytische Geometrie und Algebra II, Vandenhoeck & Ruprecht, 1951; English translation, Projective geometry of ndimensions, Chelsea, 1961. [S] M. Protter and C. Morrey, Analytic geometry, Addison-Wesley, 1975.

**351 (Xx.23) Quantum Mechanics
**

A. Historical Remarks

+Newtonian mechanics (classical mechanics) successfully explained the motion of mechanical objects, both celestial and terrestrial, on a macroscopic scale. It failed, however, to explain blackbody radiation, which was discovered in the last decade of the 19th century. M. Planck introduced a hypothesis of discrete energy quanta, each of which contains an amount of energy E equal to the frequency of the radiation v multiplied by a universal constant h (called Planck’s constant). He applied this hypothesis to derive a new formula for radiation that gives predictions in good agreement with observations. A. Einstein proposed the hypothesis of the photon as a particlelike discrete unit of light rays. Assuming that many physical quantities, including energy, have only discrete values, N. H. Bohr explained the stability of electronic states in atoms. As illustrated in these examples, quantum mechanics is applied to study the motion of microscopic objects, including molecules, atoms, nuclei, and elementary particles.

B. Quantum-Mechanical

Measurement

Fundamental differences between the new mechanics and classical mechanics are due to the facts that many physical quantities, for example, energy, can take only discrete values in the microscopic world, and that states of microscopic objects are disturbed by observation. A (pure) state at a certain time is expressed by a unit vector $ in a +Hilbert space X, and observables, or physical quantities, are expressed by +self-adjoint operators in such a space. Let a, (n = 1,2, . . . ) be teigenvalues of an observable A, and let P,, be a tprojection opera-

C. Canonical

In quantum mechanics, canonical are represented by the self-adjoint

1303

351 D Quantum

Mechanics and the Schriidinger

Qk (coordinates) and Pk (momenta), k = 1, . . . , N, which satisfy the canonical commutation relations

D. Time Evolution Equation

where 1 is the identity operator, [A, B] denotes the commutator AB - BA, h = h/(24, and the relation is supposed to hold on a certain dense domain of vectors. Self-adjoint operators Qk and Pk satisfying the above relations are unique up to quasi-equivalence under a suitable domain assumption, e.g., if &Q: + C, Pt is essentially self-adjoint on a dense domain invariant under multiplication of the Q’s and P’s and on which the above relations are satisfied. Under such an assumption, Qk and Pk are unitarily equivalent to a direct sum of the Schriidinger representation on L,(RN,dx, . . . dx,), where Qk is multiplication by the kth coordinate xk and Pk is the differentiation - ih(3/3x,) (Rellich-Dixmier theorem). The above Schrodinger representation is called the position representation (or qrepresentation). The formulation using the function space L, of real variables pk, k = 1,2, . . , N, on which the operators Pk act as multiplications by pkr is called the momentum representation (or p-representation). If Hermitian operators A and B satisfy the canonical commutation relations in the form (A$, B$)-(B$, A$) = ih(lC/, +), then the following Heisenherg uncertainty relation holds for the expectation:

The time t of an observation is fixed in the foregoing discussion. A state changes, however, as the time t changes, in such a way that the transition probability between states is preserved. By Wigner’s theorem (- Section H), this time evolution of states can be implemented by unitary operators U, defined by the transformation of vectors $-U,$= $,. Furthermore, under some continuity assumption, such as that of (cp, U,$), U, can be made a continuous one-parameter group. By Stone’s theorem, U, = eeiHtlh for a self-adjoint operator H. This operator is called the Hamiltonian operator (or simply Hamiltonian) determined by the structure of a system. An infinitesimal change in $ corresponding to an infinitesimal change in t can be generated by this operator H as follows:

This equation is called the time-dependent Schriidinger equation. A state $ changes but observables do not change with time in the Schriidinger picture above. The other picture, known as the Heisenberg picture, is equally possible. In this picture, the state is expressed by a timeindependent vector, while operators A vary with time as follows: A+ U** AU, = A(t). Rates of change of operators A(t) can be calculated by means of the equation F=fCH, A(t),,

This gives the uncertainty in observations, which means that two observables A and B cannot simultaneously be observed with accuracy. This is another important property of microscopic motion that cannot be found in macroscopic motion. In a direct sum of the Schrddinger representation of the canonical commutation relations, the unitary operators

which is called the Heisenherg equation of motion. When time t changes, the expectation value of an operator A in a state II, changes in both pictures according to d(A)/dt=i([H,A])/h. According to classical analytical dynamics, a change of a dynamical quantity that is a function of tcanonical variables qi (positions) and pi (momenta) is given by dA/dt = -(H, A),

U@)=expi~a,Q,,

k

V(b)=expiCb,P,

k

with real parameters uk and 4, k = 1, . . , N, satisfy the following Weyl form of the canonical commutation relations: U(a)U(a’)= U(a)V(b)= U(a+a’), V(b)U(a)exp ( V(b) V(b’) = V(b + b’), -iCa,b,

k

. >

Conversely, any pair of families of unitary operators U(a) and V(b) satisfying these relations and depending continuously on parameters a and b are unitarily equivalent to those obtained as above (von Neumann uniqueness theorem).

where H is a THamiltonian function and the parentheses ( , ) denote the tPoisson bracket. A replacement of the Poisson bracket (A, B) by [A, B]/ih transforms this classical equation into the quantum-mechanical equation above. It should be noticed that the mathematical structure of the Poisson bracket is similar to that of commutator. In this transition from classical to quantum mechanics the correspondence principle can be used. This requires that the laws of quantum mechanics must lead

351 E

Quantum Mechanics

1304

to the equations of classical mechanics in the classical situation, where many quanta are involved and h can be regarded as infinitesimally small in the commutation relation. The correspondence principle suggests that Hamiltonian operators in quantum mechanics can be obtained from Hamiltonian functions H(p,, qk) of canonical variables pk and qk in classical mechanics after replacing pk and qk by the operators Pk and Qk in the Schriidinger representation (up to uncertainty of about the order of operators). This process of moving from canonical variables and the Hamiltonian function in classical mechanics to canonical operators and the Hamiltonian in quantum mechanics is called quantization. Taking a system of s particles and letting xk, y,, and zk be the Cartesian coordinates of the kth particle, we usually write the equation of motion as

a* -rfl gAk ih---= at

+ ~(X,,Y,,Zl, . . ..~.,Y,,Z,)

> *>

which is a second-order partial differential equation. Here rn,, is the mass of the kth particle; Ak is the tLaplacian of xk, ykr and zk; and V is a real function called the potential energy. This equation is the time-dependent Schriidinger equation. The partial differential operator on the right-hand side is called the (s-body) Schriidinger operator and $(x1, y,, z,, . , x,, y,, z,, t) is called a wave function. The probability of finding a particle in the volume dx,dy,dz, bounded by xkr xk + dx,, y,, yk + dy,, and zk, z,+dz, is proportional to ($(x,,y,,z,, . . ..x 83y s1z S’ t)l*. Usually \+I’ is normalized so that its integral over the whole space is 1. We sometimes call $ the probability amplitude. When $ is given by e--iEf’h+I, . . . ,zJ, the expectation value of an operator A in a state $, (A)=j+*A$dx, . ..dz., does not depend on time. When this is the case, II, is called a stationary state. A real value E and a function &,, . . , ZJ are found by solving an teigenvalue problem H~J = Ep. This equation is the timeindependent Schriidinger equation, which is a second-order partial differential equation. Since the Hamiltonian H stands for the energy of this system, the eigenvalues E are the energy values that this system can take. When a potential function V is given, it is a nontrivial matter to prove that the (s-body) Schrodinger operator with the given V is essentially self-adjoint on the set of, for example, all P-functions with compact supports so that its closure H defines mathematically the continuous one-parameter group of unitaries U t =e-ietih for the time evolution of the quan-

turn system of s particles with the given interaction potential V. If V satisfies an estimate 11 V$jj <illH,,ll/l] +p]/$ll (called the Kato perturbation on Ho) for some nonnegative 1< 1 and p > 0 and for all $ in a dense domain on which If,, is essentially self-adjoin& where Ho denotes the Schrodinger operator with V=O (called the free Hamiltonian or the kinetic energy term), then Ho + V is essentially selfadjoint on the same domain. For the case where V consists of Coulomb interactions between electrons and Coulomb potentials on electrons by fixed nuclei, for example, such an estimate and hence the essential selfadjointness of Hamiltonians for atoms and molecules were established first by T. Kato (Trans. Amer. Math. SOL, 70 (195 1)). For a l-body Schrodinger operator (or 2body Schrijdinger operator after the center of mass motion has been separated out), the point spectrum is that of the particle trapped by the potential, and the state represented by its eigenvector is called a bound state. The eigenvalue is nonpositive for a reasonable class of potentials V (for example, if V(x) (xER~) is continuous and 8(1x)-‘-‘) as J:cJ+oo for some E> 0), and its absolute value is called the binding energy. The eigensolutions of the Schriidinger equation are what have been called stationary states above. There are also stationary solutions that do not correspond to the point spectrum and hence are not square integrable. They are used in the stationary methods of scattering theory (- 375 Scattering Theory).

E. Some Exact Solutions Schradinger Equation (1) Harmonic

for t:he l-Body

oscillator. First consider the case in which the space is of 1 dimension, so that the Laplacian A is (d/dx)‘. Let m be the mass of the particle and V(x) = ma? x2/2 for a positive constant w (called the angular frequency). The Hamiltonian

H = -(h2/2m)(d/dx)’ has simple eigenvalues E, = hw(n + (l/2)), with a complete functions &,(x)=c,H,(q)e-q2’2,

+ mw2.c2/2

n=0,1,2 orthonormal

)..., set of eigen-

q=(mw/h)1’2x, and c, is

where H,(q) is a tHermite polynomial the normalization constant: H,(q)=

rm

k=O

1 (-l)kn!(2q)“-2k/{(n-2k)!k!},

c, = {22”(n!)2nh/(mw)}-114,

functions f.(xj). If V(x) (x E R3) depends only on r = 1x I= (zkl (xj)*)“’ (called a central potential).“(x) = (1 . there are no point spectra. F.X*)+(dl+m(XZ .(0. For each 1 and m.“(x) is an tassociated polynomial: P.. is replaced by n. (e-‘“‘*f)(x) ’ (3) Separation of angular dependence for central potential.~::‘)(s)s’+’ cnl= -{(n-I-e-‘/*. .&~ = r-‘u.l=. 1950..) obtained as the roots E < 0 of one of the following equations: {(V/E){(V/E)-l}“*= l}1~Z=tan{a(-2mE)1~2/(2h)}. according as I= 0. Kac (Proc. Statist..l-l. P.2.<a/2and V(x)=Oforlxl> a/2 (x ER). Let l’(x)=VforIxl. R. of the tLaguerre and L:(x) is the @h derivative where A and B are self-adjoint . where V is a sum of a locally integrable function bounded below and a Kato perturbation on Lo. . and the corresponding eigenfunction is n.. Wiley. is the principal quantum number. Consider H = Lo + V (form sum). Probability and related topics in the physical sciences.L~~. where E denotes the expectation for the Wiener process. P-wave. 1959). Mod.. cp (x1 = r sin Qcos rp. where n=l+ 1. x. (4) Hydrogen-type atom. 20 (1948)) has given the solution of the Schrijdinger equation as an integral of eiLih over all possible paths q(t). .<t.cp). The wave function $(x) with the angular dependence &.<t. Let q(t) (tsR) be Gaussian random variables with mean 0 and covariance E(q(t)q(s))=2-‘exp(-It-sl) (called the oscillator process)..(R.. Il/.P. and its multiplicity is n2..e2 .x2/2). where L = L(q. n in E.. .+. Let V(r) = -Ze’/r (2 > 0).. functions (for spatial dimension < 3). $02dx) (j=l . Consider the l-body Schriidinger operator H = Ho + V (form sum). 4) (4 = (d/dt)q(t)) is the classical Lagrangian for the Hamiltonian system.. . ‘x -(h*/(2m))u”(r)+{h*l(l+ lItill= s0 14r)12dr<=b where the angular function &. + n.. and L.I(r)I. (2) One-dimensional square-well potential. operators 1)![2Z/(na(n+l)!)]3/(2n)}“2.Z2/(2an2) with eigenfunctions I... If V is a sum of L.=rsinesincp.1. . with nonnegative integers n. there are eigenvalues .(R dr-‘4rb 1)/(2mr) + V(r)-E}u(r)=O. where V is the sum of a locally integrable function bounded below and a Kato perturbation on Ho.(Q.... Mathematical reformulation of the formula in terms of the Wiener measure has been given by M. for almost all x. For any J(x) in L. . m=l. . Here P. 1+2. uJr) = ~. .=rcosB) as N4= k. Feynman (Rev. + . 2nd Berkeley Symp. Math. Phys. If V<O and The eigenvalue is determined by n.. Let Lo be the Hamiltonian for a l-dimensional harmonic oscillator with m = w = h = 1 and I++~be the eigenfunction $o(x) = n-li4exp( . n)andt..1305 351 F Quantum polynomial Mechanics When the space is of r dimensions..) -1+1.. . 1=0. If V>O.ix x V: I. . is an eigenfunction of the square L* of the orbital angular momentum L = . This integral is called the Feynman path integral.<t. . n. 0. and the integer m is the orbital magnetic quantum number.t(cos O)&mr. cp) is called the S-wave.. . then all eigenvalues E and a basis for eigenfunctions $(x) can be obtained in terms of the polar coordinate r. corresponding to the different possible values of 1 and m..2. . Dwave.. -1. 1.l)‘/dx’+“)/(2’1!). x. Let b(t) (t 20) be the Wiener process and q(t)= hb(t)/(2m)“‘. The above path integral formulas are closely related to the Trotter product formula e-t(A+B)= lim (em tA/ne-rB’n)n n-m (t&O).. Path Integrals then there are N eigenvalues (N = 1. -cot{a(-2mE)“*/(2h)}. Probability. The above equation for u(r) is called the radial equation. This is called the Feynman-Kac formula.. . For any L. The nonnegative integer 1 is the azimuthal quantum number. s = 2Zr/(na). c. = c.=(-1)“{(21+l)(l-m)!/(47r(l+m)!)}”Z.. then the right-hand side is continuous in x for t > 0.2 Legendre ).

To solve this difficulty. y’. The Klein-Gordon equation can also be considered to be the classical wave equation of matter and can be second-quantized.3). ~~i(i~ :). and h/2 is called the absolute value of the spin. (k = 1.) angular S =z c is the intrinsic where the energy eigenvalues _+Jm2c4fp2c2.3) and E by ih8/& in the relativistic identity E2=m2c4+p2c2. obey this equation. (The same formula holds without the bdundedness assumption when t E iR. The positron theory is introduced. Therefore we say that electrons are particles of spin h/2. E are There are four independent eigensolutions u(l). uc3). M. Y. thus providing evidence that the Dirac equation is correct. p + mc’p. The coefficients y” can be so determined that every component of $ also satisfies the KleinGordon equation. The existence of negative energy states. The momen- bounded below and A + B is essentially selfadjoint. The matri Y S2 = Sz + S. so that (o/c)’ can be neglected. L is not conserved. 1. ii)> %=(:A (The y’ are called Dirac’s quantity y matrices. The Dirac equation has +plane wave solutions and the ckr called Pauli by ox=c A). The Dirac Equation The Schradinger equation is not relativistically invariant. where yk = .311 Second Quantization). Wave functions of free particles are believed to satisfy this equation. an electron-position pair must be created. Sixteen linearly independent matrices are obtained by repeated multiplication of five matrices. Nishina and 0.2. forces us to give up considering the Dirac equation as an equation of one electron. v=O. If gamma rays are absorbed to excitl: an electron from a negative energy state into a positive energy state. however.2. these two components become independent of each . and the Dirac equation is considered as the classical field of electron waves and is second-quantized (-. Any 4 x 4 matrix can be expressed as a linear combination of these sixteen matrices. Thus the yv are found to be 4 x 4 matrices satisfying the commutation relations yUyY+yVy’=2gfiY (p. However. Relativity requires the equal handling of space and time.ihr x V.3). the total angular momentum J = L + (h/2)a can be conserved when c~is a vector whose components can be given as (: i). states of the electron can be expressed in 1:erms of twocomponent wave functions. The absence from the vacuum of a negatively charged electron in a negative energy state could then be expected to manifest itself as a positively charged particle (positron) with positive mass and kinetic energy. y2. When the speed of an electron is very small. Motions of particles with zero spin. where B=(i ol).351G Quantum 1306 Mechanics ber of electrons in the normal state of the vacuum. it has to be taken into account in order to obtain a mathematically complete set. the neutron for example have spin. Dirac assumed that the $ of a free electron is expressed in terms of a tspinor with four components satisfying a linear differential equation that automatically implies the Klein-Gordon equation. where g”‘=O for p#v and go’= -gkk=l (k= 1. This was predicted in the theory of light spectra.) G. ay=(:’ spin matrices. IL(=). Dirac proposed the hypothesis (Dirac’s hole theory) that all the negative energy states are filled up by an infinite num- tum of the eI&tron.2is diagonal and is equal to s(s+ l)h21 (I is the identity matrix). Two of them are of positive energy and the other two are of negative energy. Many particles besides the electron.’ + .2. pi mesons (n) for example.2. Klein calculated the cross section of Compton scattering (the Klein-Nishina formula) by use of the Dirac equation and found good agreement with observations. y3 and the identity matrix. A. also called the spin. For the electron s = l/2. which include the four matrices y”. We can rewrite the Dirac equation as ih&+h/& = He. The Dirac equation (8) (x0 = ct) satisfies these requirements.3). are given “1>. and uc4). This approximation is called the Pauli approximation. P. and y” =/I Since H cannot commute with the orbital angular momentum of an electron L = r x p = . The Klein-Gordon equation is obtained by replacing pk by (h/i)a/dx. H = cu. If the spin-orbit term that appears in the Pauli approximation is also neglected. Although the negative energy case is physically undesirable.8tlk (k = 1. where c is the speed of light. because u has four components.S.

where J is the sum of the orbital angular momentum L and the spin angular momentum S (J = L + S) and eigen- values of J* are given by J (J+ l)h*. Two identical particles. H. @I D. If T commutes with H or H is invariant under the transformation x-+x’= TX. and the traceless tensor according to 0:. each of which is equal to the dimension of the corresponding representation of the group {T}. the electron. When a system consists only of fermions of the same kind with spin h/2 and a Hamiltonian of this system does not include terms depending on spins. A system consisting of N identical particles can be classified by the irreducible representations of the tsymmetric group S. the pseudovector according to DC.-. or between identical bosons. then T$(x) = q(x) = $(T-‘x) satisfies the same Schriidinger equation as $. neutron. cannot be distinguished.. There are degeneracies of the energy eigenvalues. two of them cannot occupy the same state (this law is called the Pauli principle). then the wave functions are just products of spin and orbital parts. This implies the selection rule J’-+J’+ 1. both of which are either fermions of the same kind or bosons of the same kind. where M ranges from L to -L by unit steps.60 Classical Groups I). H is invariant under the 3-dimensional rotation group. where the transformation of the function is defined by t.e. When the particles are fermions.1307 351 H Quantum satisfy the SchrGdinger Mechanics other and individually equation. i. Let these coordinates be denoted together as x. The transformation of every observable obeys a certain rule under the transformation of coordinates.1. Application Groups of Representation Theory of Lie A symmetry (with active interpretation) is a bijective mapping of pure states (represented by unit rays in a Hilbert space) preserving transition probabilities between them. States with + are called even states and those with -> odd states. which should be tunitary on L.-space relative to the Lebesgue measure dx if T leaves the measure invariant. contains as a factor a representation equivalent to DJpv. of N elements.T$ give a (generally infinite-dimensional) representation of this group. energy levels of atoms and nuclei can be classified by 0:. for example. states are classified by the irreducible representations DL of the trotation group (258 Lorentz Group). states are labeled by the irreducible representations DJ of the rotation group. Irreducible representations of this group are written as 0: .=D.. many of which can occupy the same state. If the transformation of an observable is given by DJ. called fermions. For example. For example. the z-component of L.(J’>l)iftheelectric dipole transition dominates (J = 1). When H is spherically symmetric. J’. In electromagnetic transitions in atoms or nuclei. which form a projective representation. More general selection rules can be obtained in the same way for general multipole transitions. a rotation of the coordinate system or a permutation of the labels of the particles. each of which belongs to a different M. Therefore the Hamiltonian should be invariant under permutations between identical fermions. of which only one can occupy a given state.. and proton are fermions. To obtain matrix elements of observables between two stationary states. The square of the sum L of all orbital angular momenta has eigenvalues L(L + l)h’.+. as. The scalar is transformed according to D:. a positive integer. J’.+D. D.. only the transition O+ 1 is possible. while the photon and pi meson are bosons. There are 2L + 1 degenerate states. where L must be 0 or a positive integer. There is a class of particles. then a matrix element of this observable between the states belonging to DJv and DJ. Representation theory is useful in determining general formulas of transition strengths. so that only totally antisymmetric states are permissible for fermions. or a halfinteger.. we obtain the 3-dimensional orthogonal group (. Each J must be zero. Even when there is an interaction between the orbital angular momentum and the spin angular momentum. Eigenstates + of the Schriidinger equation are functions of the coordinates of each particle. When J’= 0. The set of transformations x-+x’= TX forms a tgroup {T}.V(x’) = $(x). Adding inversions to the pure rotations. There is another class of particles.+D. Suppose that an operator T operates on the x. group representation theory is useful. then the decomposed stationary state can be labeled by an tirreducible representation. In order to make wave functions totally .. where f corresponds to the characters of the inversion relative to the origin. the vector according to 0. If the representation for each eigenvalue is decomposed into irreducible ones. Furthermore a connected Lie group of symmetries is implemented by unitaries. called bosons. Wigner’s theorem says that any symmetry can be implemented by either a unitary or an antiunitary mapping of the underlying Hilbert space as a mapping of unit rays onto themselves. vanishes unless the tensor product representation DJ @ DJ. and the corresponding transformations $. .

The classification of the spectral terms of a polyatomic molecule is related to its symmetry. masses. . and the physical relationship among all inequivalent representations of a C*-algebra was first discussed by K. antisymmetric. whereas when A =O. which can be zero or a positive integer.. K. van Hove (Physica. then the eigenstate of these N nucleons is classified by its irreducible representations I). The right-hand side of this equation gives all possible states of the atom. Haag and D. The up state of the isospin corresponds to the proton and the down state corresponds to the neutron. where J’= min(J. Friedrichs (Mathematical aspects of the quantum theory offields. the corresponding state has twofold degeneracy. orbital wave functions with a total spin u/2 should be limited to those corresponding to the tYoung diagram [2Nm”. are neglected. Analogous to ordinary spin. called the nucleon. I. Danske Vid. If a Hamiltonian of N nucleons is invariant under any transformation belonging to SU(2).. J2) corresponding to the decompositionD.) l. l”] = T(2. described by the set of all transformations that interchange identical atoms. a physical observable is an element of a C*-algebra ‘QI and a state is a functional cp on VI (its value q(A) is .. because these two particles have very similar natures except for their charges.. Fys.. If these two atoms are identical. Haag (Mat. When a state of N nucleons has an isospin T= v/2. Interscience. To construct such a wave function.. depending on the character of the reflections. 5 (1964)).. are classified by the irreducible representations of the group Td (which is generated by adjoining reflection symmetry to the ttetrahedral group T). In the approximation of many-body problems by means of independent particles. an atom with two electrons carrying the angular momenta J. and possibly the interaction responsible for the proton-neutron mass difference. a point first emphasized by K. The C*-Algebra Approach J. Stationary states can be classified by the absolute value A of the angular momentum of the diatomic system around this axis. This invariance is called charge symmetry. If this Hamiltonian is also independent of spin. Consider transformations belonging to the tspecial unitary group SU(2) in the 2-dimensional space spanned by the proton state and the neutron state. the orbital-spin wave function of this state must correspond to the Young diagram [2Nm”. Medd. As an approximation. 29 (1955)). Phys. Selsk. stationary states of methane molecules CH. two states (labeled by &) arise. 1953). For example. the molecular states can be classified further as even and odd according to the character of reflections with respect to the plane containing the center of mass and perpendicular to the axis. In C*-algebra approach. etc..10DJ2=DJ. it is invariant under unitary transformations in the 4-dimensional space spanned by the four internal states of the nucleon: up and down spins. and JZ has 25’ + 1 different angular momentum states. isospin can be introduced to describe the two states of nucleans. l”]. 48 (1947)). Kastler (J. Segal (Ann Math.2. 0.l). E. the wave function of the total system is constructed by multiplying the wave functions of the individual particles. Therefore the states of N nucleons can be classified by the irreducible representations of the group U(4) (Wigner’s supermultiplet theory).+JZ+DDJ..J.hysical examples illustrating this point were given by L.. For example. and p.351 I Quantum 1308 Mechanics and neutrons may be taken to be invariant under the interchange of protons and neutrons. When A > 1.-J. the Hamiltonian of a system consisting of protons The uniqueness of operators satisfying the canonical commutation relations (representations of CCR’s) up to quasi-equivalence (Section C) no longer holds if the number of canonical variables become infinit~e (a so-called system of infinitely many degrees of freedom).+ D. holds in the nucleus and elementary particles if electromagnetic and weak interactions. The use of C*-algebras in physics was first advocated by I. This invariance. since the isospin wave function multiplied by the orbital-spin wave function is a totally antisymmetric wave funcl:ion.+JZ-1+. up and down isospins. Charge Symmetry The proton and the neutron can be considered to be different states of the same particle. . Level structures of a crystal are classified by the irreducible representations of its symmetry groups. called isospin invariance. it is useful to consider the reduction to irreducible parts of the ttensor products of representations of the groups attached to the individual particles. Math. 18 (1952)) and R. Polyatomic Molecules The group generated by the 2-dimensional rotations about the axis connecting two atoms and the reflections with respect to the planes containing this axis is used to classify states of diatomic molecules. where T stands for the total isospin of each state.

P. weakly modular (also called orthomodular) tatomic lattice satisfying the covering law. and are physically relevant if states in K. Foundation of Quantum Mechanics Hilbert spaces and the underlying field of complex numbers.g. a... equivalently. n. A$). and E. The set of all mixtures of vector states ($. where weak modularity means cr\(c’vb)=b and bv(b’Ac)=c whenever b < c. 11 cpII= 1 or equivalently cp(1) = 1 if 1 &I. such as behind the moon (the behind-the-moon argument). called questions. However. where it would form a Boolean lattice. e. equilibrium states of a infinitely extended system with different temperatures (. it is mathematically more convenient to consider the states of the particle without any compensating object (in the same way that an infinitely extended gas is more convenient for some purposes than a finitely extended gas surrounded by walls). together with their order structure (logical implications) and associated lattice structure (join.1309 351 L Quantum Mechanics the expectation value of the observable A E 2I when measured in that state) that is linear. torthocomplemented. all physically relevant information is in the vacuum representation. are physically relevant. Wightman.(A). then we should either forbid nontrivial superpositions of states of the two classes or. A E 2l.c>b)isb or q v b. such that TX. The tGNS construction associates with every state cp a Hilbert space H. when considered as states on the algebra of observables in the form of expectation functionals ($. while the vacuum vector would be unchanged.e. a particle with spin h/2 can also be discussed in the vacuum sector if we consider a state of this particle in the presence of its antiparticle at a far distance. C.. In quantum field theory. i. are weak limits of states in K. in the weak topology contains K. One approach of this kind focuses attention on the set of all observables that have only two possible measured values 1 (yes) and 0 (no). meet. and orthocomplementation as logical sum.. and hence there are various attempts to find axioms for quantum mechanics that imply the usual mathematical structure and at the same time allow direct physical interpretation. superselection sectors explained below (. the vacuum state can be taken to be pure (by central decomposition if necessary) and in the associated GNS representation (called the vacuum sector) all vectors can be assumed to be physically relevant pure states. of ‘u by bounded linear operators on H. product. to H. a representation n. such that rp(A) = (Q. which constitute a mathematical background for quantum mechanics. and normalized.(A) T for all As2l. 88 (1952)).. in H. L.) ail<Ei.. which can be obtained as weak limits of states in the vacuum sector by removing the compensating particle to spatial infinity and which produce inequivalent representations called superselection sectors. are not immediately discernible from physical observations. and a unit cyclic vector R. and negation). Under 360” rotation a vector representing a state of a particle with spin h/2 acquires a factor -1 (. a(A)+) for any fixed faithful representation 71 of 2l is weakly dense in the set of all states of Cu. information on any state cp can be obtained by measurements only up to a neighborhood in the weak topology: 1rp(A. Wick. Conversely any such lattice is a di- .(A)n. all states of K..402 Statistical Mechanics).i=l.. a point emphasized by Haag and Kastler as a foundation of the algebraic viewpoint in the formulation of quantum theory. Because actual measurement can be performed only on a finite number of observables (though chosen at will from an infinite number of possibilities) and only with nonzero experimental errors. (K. The lattice of all orthogonal projections (corresponding to all closed subspaces of a Hilbert space) in quantum mechanics is a tcomplete. Wigner (Phys. restrict observables to those leaving the subspace spanned by vectors in each class invariant so that the relevant linear combinations of vectors. positive in the sense q(A*A) > 0 for any A E ‘%!I.150 Field Theory) and equilibrium or ground states with broken symmetry.(A) = 7c. physically contains K2) if the closure of K. This is called the univalence superselection rule and has been pointed out by A. then anycbetweenqvbandb(qvb>. This is called quantum logic in contradistinction to the situation in classical physics. If the 360” rotation is not to produce a physically observable effect.258 Lorentz Group).. of states can describe measured information on states at least equally well as another set K. S.). Abundant disjoint states occur for a system of infinitely many degrees of freedom.. From another viewpoint. Rev.. for example. A nontrivial linear combination (superposition) of these two would then be changed to a vector in a different ray. In principle. G. and the covering law means that every b # 0 possesses an atom p under it (p < b) and that if an atom q satisfies q A b = 0. Two states cp and $ (or rather 7crn and 7tJ are called disjoint if there is no nonzero mapping T from H. are actually mixtures (rather than superpositions) of states in two classes and are invariant under the 360” rotation. A set K...

Weyl. For a situation where a pair of (correlated) particles in states a and b are created and their spins (1 or -1. while the convex cone of all states (without normalization) of Jordan algebras and C*-algebras have been characterized in terms of a certain class of projections associated with faces of the cone. Emch. 75 (1980)) has given a characterization allowing direct physical interpretation by replacing P-projection with a measure p and the functions A.. It is also possible to characterize the set of all states equipped with the convex structure (mixtures) geometrically. and E. Pauli. irreducible tJordan algebra over the field of reals (the positivity of a state cp is defined by ~(a*) > 0) has been characterized as a transitively homogeneous self-dual cone in a finite-dimensional real vector space (a cone V is transitively homogeneous if the group of all nonsingular linear transformations leaving V invariant is transitive on the topological interior of V) by E. von Neumann. Moscow Math. or the so-called spin balls (the set of all normalized states being a ball) linearly generated by the identity and yj (j = 1. called P-projections. G. by E. and others (Actu Math. Bellissard and B.. 140 (1978). . [ 111 G. 1958. ~(a) > 0. (Original in German. In finite-dimensional cases.I < I. Tomonaga. Due to some features of quantummechanical measurement not in conformity with common sense. van der Waerden. 1!)71.e. References [l] P. Fourier. . up or down spin) measured at positions distant from each other. Messiah. Wigner. Inst. P. so that hidden variable theories of this type have been rejected. Wiley. 1962. Sot. this type of characterization extends to the “natural” positive cones of vectors (A. W. there have arisen hidden variable theories that are deterministic and reproduce the quantum-mechanical prediction. Phys. complex. Mathematical foundations of quantum mechanics. The set of all states (without the normalization condition) of a finite-dimensional. Connes. Iochum. Shultz. J. Dover. then the set of all probability measures on the lattice (i. 1972. of hidden variables 1.. formally real. b)-E(u.(L)dp(L) for a probability rect sum of irreducible ones. Princeton Univ. the expectation value E(u. 1949. North-Holland. b)= A. (Original in German. 1958. Inst. 12 (1963). 28 (1978)). Math. Araki (Commun.e. General principles of quantum mechanics. The theory of groups and quantum mechanics.1 W. 1959. 13 (1965)). [2] A.. L. M. Algebraic methods in statistical mechanics and quantum field theory. [7] B. p( 1) = 1) is exactly the restriction to questions of states p(A)= tr(pA) given by the density matrices p (Gleason’s theorem). A. 1980. [S] H. Fourier. n) satisfying yj o yk = 0 ifj # k and yj2=1. Academic Press. Group theory and quantum mechanics. II. Quantum mechanics I. b) for the product would be given in a hidden variables theory by E(u.) [4] S. Clarendon Press. [S] J. fourth edition. 36 (1934)) as direct sums of the following irreducible ones: the Jordan algebra (with the product A o B = (AB + BA)/2) of all Hermitian n x n matrices over the real. b)+E(u’. i. can be obtained as the lattice of subspaces V satisfying (T/l)’ = V in a vector space over a (generally noncommutative) field with an antiautomorphic involution *. an additional requirement is needed to restrict the underlying field and its * to be more familiar ones. Vinberg (Trans. or quaternion field. 1962. Functional integration and quantum mechanics. In this approach. If that is done. B. North-Holland. 144 (1980)).) [9] E. van der Waerden. L. F. 1928. such as real. all 3 x 3 Hermitian matrices over octanions. J. equipped with a nondegenerate Hermitian form. Academic Press. [lo] B. Springer. 1967. Wigner (Ann. assignment of expectation values 0 <p(a) < 1 for all elements a in the lattice such that p(Vi ai) = xi ~(a~) if ai I aj for all pairs i # j. 1932.). Dover. b’)+E(u’.) [6] B. 1955. In infinite-dimensional cases. Ann. Principles of quantum mechanics.1966. von Neumann. Dirac.(l)B. Press. Math. M. 1959. Sources of quantum mechanics. each of which. 1974.. Group theory and its application to the quantum mechanics of atomic spectra. P. [ 121 T. Springer. Quantum mechanics I. Press. Alfsen. Quantum theory and beyond. Bastin (ed. . Cambridge Univ. 1961. b’)1<2. complex. (Original in French. Jordan.351 Ref.. Simon. 1979. if of dimension (length of longest chain) > 3.[ < 1 and 1B. 24 (1974). Ann. representing spins and hence satisfying 1A. where the relevant Jordan algebras were completely classified earlier by P. Then the following Bell’s inequality holds: IE(u. or quaternion fields and their usual conjugations *. II. and B. Quantum 1310 Mechanics notion of filtering corresponding to quantummechanical measurement. This contradicts both quantum-mechanical predictions and experimental results.) C3. (Original in German.

Beurling and L. Morrey [6]).C. .) A p-conformal function is represented as the composite go h of an analytic function g with a p-conformal mapping h. he determined the quasiconformal mappings between two given domains. in the theory of tpartial differential equations of elliptic type (M.1 . A 1-quasiconformal mapping is conformal. which has led to the precise definition for quasiconformality itself. C. almost everywhere in D with the measurable coefficient p. Consider an orientation-preserving topological mapping f of a domain D on the z( =x + iy)-plane. including the theory of Teichmiiller spaces (.e.. 352 (XI. and The inverse mapping of a K-quasiconformal mapping is also K-quasiconformal.. Let f be a K-quasiconformal mapping of lzJ < 1 onto 1w I< 1. 0.e. A. Principal Properties and Results B.1311 352 C Quasiconformal Mappings [ 131 B. Griitzsch noticed that Picard’s theorem still holds under the weaker condition.lf?l) is bounded. e. 375 Scattering Theory. 1976. its Jacobian is positive a.416 Teichmiiller Spaces). then f is conformal. Bers [S] (. a closed Jordan domain with four specified points on the boundary. Ahlfors (. and (If.1 + Ifsl)ldzl and minor axis of length (If. called the modulus of Q and denoted by mod Q. then f is called a Kquasiconformal mapping of D. if it can be defined. B. Every quasiconformal mapping is ttotally differentiable a. The maximal dilatation (1 + ll~llm)/(l-ll~ll. Teichmiiller.1 I&l)ldzl. Virtanen. is K. we call it a Kquasiregular function or K-pseudoanalytic function. If in the above statements f is not necessarily topological but merely a continuous function satisfying the same requirements. We cannot speak of the history of quasiconformal mappings without mentioning the discovery of extremal length by A. K. which dispense with continuous differentiability. second edition. is uniquely determined. i. These applications are explained in Sections C and D. When the value of K is irrelevant to the problem considered.1 + I.367 Riemann Surfaces). Sometimes f is called.I-.e. S.-quasiconformal mapping fi. are due to Ahlfors [3]. Benjamin. Lavrent’ev [Z]) and especially in the problem of moduli of Riemann surfaces. Kakutani. The quasiconformality off is defined as follows.l . These notions are also defined for mappings between tRiemann surfaces. that is. and for this reason they have been utilized for the type problem or the classification of open Riemann surfaces (Ahlfors. Also . A. Mori [4].quasiconformal. When the ratio K(z)=(lf. 377 Second Quantization.-quasiconformal mapping fi with a K. (1) (the geometric delintion) Let Q be a curvilinear quadrilateral. Griitzsch (1928) introduced quasiconformal mappings as a generalization of conformal mappings. providing the smallest sup K. If K = 1. f is called quasiconformal. This is equivalent to: (2) (the analytic definition) f is absolutely continuous on almost every line segment parallel to the coordinate axes contained in D (this condition is often referred to as ACL in D) and satisfies the inequality lfr:. for short a p-conformal mapping. those closest to conformality L. Definitions The current definitions of quasiconformality. d’Espagnat. Quasiconformal mappings have important applications in other fields of mathematics. I.. If modf(Q) < K mod Q for any curvilinear quadrilateral Q in D. The K-quasiconformal mapping f satisfies the so-called Beltrami differential equation VI. The ratio (> 1) of the sides of I. History Mappings H.g. Toki. K.e. The composite mapping fi ofi of a K. and 386 S-Matrices. Y. Quasiconformal mappings have less rigidity than conformal mappings. Then f extends to a homeo- . K-quasiconformal mappings are simply said to be quasiconformal.1 5) Quasiconformal A.references to 150 Field Theory.l+ l&1)/( If. (almost everywhere).fMlf. and let the interior of Q be mapped conformally onto a rectangular domain I. (If in addition //pII m <(K . which are not conformally equivalent to each other.)does not exceedK. The image of an infinitesimal circle (dzl = constant is an inlinitesimal ellipse with major axis of length (If.I <K+l K-1lf I I almost everywhere in D with some constant K > 1. Conceptual foundations of quantum mechanics. Let f(z) be a continuously differentiable homeomorphism with positive Jacobian between plane domains. V.143 Extremal Length). we call it a p-conformal function. where the (-1.I)6 K a.l)/(K + l). l)-form p&dz-’ is independent of the choice of the local parameter z.

then f0 is called an extremal quasiconformal mapping in 9 Let R={(x. This shows that any family of Kquasiconformal mappings of ]z] < 1 onto ]w] < 1 is tnormal. Then there exists an orientation-reversing quasiconformal mapping of II onto R* which keeps every point of L fixed if and only if some constant C exists satisfying I& -5. . R’= {(x’.I@(z)ldxdy of @ in (1) is finite.) attains the infimum of K(f) for all fog. (a) Boundary Correspondences and Extensions. Q* such that R U L U a* = C U {m}. the following theorem holds: Let p(x) be a realvalued monotone increasing continuous function on R such that lim. there exists a p-conformal mapping of D onto a plane domain A which is unique up to conformal mappings of A [S]. real analytic.352 C Quasiconformal 1312 Mappings continuous (resp.~0 homeomorphic closed Riemann surfaces 19. t E R. c. Theorem of quasiconformal reflection (Ahlfors [lo:]). On the other hand. & on L such that [3~6. . the extension and reflection of quasico. Then the unique extremal quasiconformal mapping for 9 is the aftine mapping x’=(a’/a)x.<2. the uniqueness theorem is as follows: If the norm II@11= jJ. then the extremal quasiconformal mapping in 9 exists but is not always determined uniquely (K.?formal mappings are made essential use of.m of the boundary ]z] = 1. [. The space of functions p has the structure of a Banach space with L&-norm. Lehto and Virtanen [7] and Ahlfors [S].]z. t)) is a family of p depending on the parameter t with IIp(t)li ~ <k < 1 and p(t) is with some constant K > 1 everywhere on R. complex analytic). y’=(b’/b)y (Grotzsch Cl]). Then there exists a quasiconformal mapping of the upper half-plane y > 0 onto itself with boundary correspondence x H p(x) if and only if 1 k4x-+t)-Ax) -<-------<p P /4+/4--t) for some constant p > 1 and for all x.P of orientation-preserving homeomorphisms of R onto S.0. More precisely. w of R. If we define 9 as the family of all quasiconformal automorphisms f of D satisfyingf(e”)=g(e”).. which have opened up a new way to study +Teichmiiller space. the uniqueness does not hold in general (Strebel [ 131). < 1. Then 9 contains a unique extremal quasiconfcmrmal mapping. Ahlfors and Beurling characterized the correspondence between ]z] = 1 and 1WI = 1 induced by f [9]. What amounts to the same thing. Otherwse. Consider again a p-conformal mapping 9 of the unit disk D: ]z] < 1 onto itself which induces a topological autom0rphis. Let K(f) denote the maximal dilatation of a quasiconformal mapping f: Suppose that a family 9 = { jJ of quasiconformal mappings is given.. a necessary and sufficient condition is proved for the Beltrami . Let L denote a curve which passes through cc and divides C U {a} into two domains Q.and16isthebest coefficient obtainable independently of K (Mori). If {p(t) = ~(z. Ahlfors [3]). Concerning the dependence of p-conformal mapping on p. For the proofs of these important results. S and a thomotopy class . S satisfy the dilferential equation (I?w/Z)/(C~W/~~Z)=[(K .f(O) = 0. then fr@) is also continuous (continuously differentiable.O<y’<b’} be a pair of rectangular domains. a classical result concerning the tconformal mapping of surfaces asserts the existence of a solution of Beltrami’s differential equation fi=A!fZ.l)/(K + l)]‘o/I@. complex analytic) in t. ]/I[. If.I d C for anyzree points <. Next suppose that we are given 1.y’)IO<x’<a’. Strebel [ 133).O<y<h}. If some f0 E$ exists such that K(j. Let 9 be the family of all quasiconformal mappings of R onto R’ which map each vertex to a vertex with (O. the Teichmiiller mapping is the unique extremal quasiconformal mapping in 9. (b) Mapping Problem.]<l. then the Holder condition holdsfor]z. As to the Teichmiiller mapping. and the space of mappings f@ also has the structure of a Banach space with respect to a suitable norm.0)form @ such that the respective local coordinates z. (1) morphism of ]z] < 1 onto (WI d 1.O). (c) Extremal Quasiconformal Mappings. When p is real analytic and the derivatives of functions are defined in the usual manner.]<l. For further properties and bibliography . continuously differentiable. at which @ #O (Teichmiiller [ 121.y)(O<x<a.(0.O). +m p(x) = k 00. The extremal mapping f satisfying equation (1) is sometimes referred to as the Teichmiiller mapping. Given a measurable function p in a simply connected domain D with llpll I. either R and S are conformally equivalent to each other or else R admits an essentially unique analytic (2. This turns out to be a generalization of Grotzsch’s extremal affrne mapping. Ahlfors and Bers [ 1 l] obtained the following important result: Denote by f p a p-conformal mapping of the whole finite plane onto itself that preserves 0 and 1. real analytic. furthermore.

Faber mijglichst konforme Abbildungen von schlichten Bereichen.1313 352 Ref.) [3] L.. A. quasiconformal mappings were applied only to the ttype problem of simply connected Riemann surfaces and to the classification of Riemann surfaces of infinite genus (. Rickman. This is a generalized definition because the system may not be equivalent to a Beltrami equation.( f ).367 Riemann Surfaces). as they are under conformal mappings. respectively. 84 (1957) 78-84. Quasiconformal extension of higher-dimensional half-spaces have been studied by Ahlfors and L. this last result can be extended to the extremal quasiconformal mapping between arbitrary Riemann surfaces. Generalization to Higher Dimensions Let f be a continuous ACL-mapping of a subdomain G of R” into R” whose Jacobian matrix is denoted by f’(x). Furthermore. Leipzig. V. Similar Notions The term quasiconformal was used differently by Lavrent’ev. The smallest K > 1 for which this inequality is true is called the outer dilatation off and is denoted by K. Amer. Sot.. E. 114-120. [4] A.. thenfis said to be K’-quasiregular. F. Ahlfors. They utilized the fact that if the density and its reciprocal are bounded in a steady flow of a 2-dimensional tcompressible fluid. Trans. and that if in addition the supremum of the tMach number is smaller than 1. References [1] H. (Original in Russian. 1 for which the inequality det f ‘(x) < K. Math. 3 (1954) 1-58. Amer. then the mapping of the physical plane to the potential plane (the plane on which the values of the tvelocity potential and the tstream function are taken as coordinates) is quasiconformal. Hamilton [14]. Grbtzsch.. and is absolutely continuous (0. Mori. The latter is an entirely different notion from the one discussed in this article. Bers used the term pseudoanalytic to describe a certain function related to linear partial differential equations of elliptic type. Noordhoff. It is worth remarking that the investigation of quasiconformal mappings is intimately connected with the recent development of the theory of tKleinian groups via Teichmiiller spaces. J. Math. When n = 2. For n > 3 the following properties also still hold: A quasiregular mapping is discrete. If max(K.(f). [min. If f is quasiregular. and there is a similar term quasi-analytic.(f))<K’. Applications In the earlier stage of development of this theory. the operator norm and the determinant off’ are denoted by Ilf’ll and det f ‘.e. [S] L. Reich and Strebel in [ 151). D. and VHidlH [ 181). Ber. compact subset and has properties similar to those of analytic functions. Lavrent’ev. Carleson [15]. Then f is said to be quasiregular if all the partial derivatives off are locally of class L” on G and if there exists a constant K > 1 such that (11 f ‘Il(x))n < K. open. 84 (1957).207208. Moreover. On quasi-conformality and pseudo-analyticity. particularly to those concerning the behavior of fluids. 84 (1932). On a theorem of Mori and the definition of quasi-conformality. Martio. Sot. totally differentiable a.. However. 1962. Wiss. An orientation-preserving mapping is called K-quasiconformal (J. Akad. Analyse Math. This application is based on the fact that it is often possible to find a quasiconformal mapping with the prescribed boundary correspondence even when no equivalent conformal mapping exists and the fact that the classes 0. 56-77. (. =i 1 f ‘(x + y) I”] holds almost everywhere in G is called the inner dilatation off and is denoted by K. then the smallest K >. Analytic transformations in the theory of functions of several variables are called pseudoconformal by some mathematicians. 1963. This function is pseudoanalytic in the sense of Section B on every relatively . On quasiconformal mappings. then the mapping from the physical plane to the thodograph plane is pseudoanalytic. as follows: A topological mapping f= u + iv is called quasiconformal with respect to a certain system of linear partial differential equations when u and u satisfy the system. Varational methods for boundary value problems for systems of elliptic equations. Sachs. S. detf’(x) almost everywhere in G. Quasiconformal Mappings coefficient p of a quasiconformal mapping of 9 to be extremal (R. The theory of quasiconformal mappings was also applied by Lavrent’ev [16] and Bers [2] to partial differential equations. VHisala [ 173) if it is a K-quasiregular homeomorphism.. these definitions agree with those given in Section B. Trans. and O. it is reduced to a quasiconformal mapping if the system is uniformly elliptic. S. E.367 Riemann Surfaces) of Riemann surfaces are invariant under quasiconformal mappings.( f ).K. [2] M. Bers. Verh.

Mathematical aspects of subsonic and transonic gas dynamics. Teichmiiller. 1965. [ 133 K. Lecture notes in math. Wiss. [9] A. Fenn. Martio. Wiley. Ahlfors and L. 229. Amer. V. Rickman. Preuss. second edition. Comment. Academic Press... 22 (1939). Lehto and K. V&ill. Sot. Math. 36 (1979) 15-30. I. Trans. and J. V. [l l] L. Math. [ 151 L.352 Ref. Quasiconformal 1314 Mappings [6] C. 39 (1964). On the solutions of quasilinear elliptic partial differential equations. Math. 1971. l-40. Akad. 77789. [18] 0. Abh.. 43 (1938). Bers. English translation. 109 (1963) 291-301. Bers. [7] 0. V. 1974. 36 (1962). Springer. Hamilton. Vatsala. 1973. 126-166.. J. [ 191 L. Strebel.. Acad. [ 173 J. A new proof of a fundamental inequality for quasiconformal mappings. Ann. Definitions for quasiregular mappings. 448 (1969). The boundary correspondence under quasiconformal mappings. Extremal quasiconformal mappings with prescribed boundary values. [lo] L. Quasiconformal reflections.. Beurling and L. 3066323. Math. Van Nostrand. 96 (1956). Trans. 1958. Acta Math. [ 121 0. Extremale quasikonforme Abbildungen und quadratische Differentiale. Lectures on quasiconformal mappings. Analyse Math. Ahlfors. [S] L. Springer. Ahlfors. (2) 72 (1960) 385-404. Riemann’s mapping theorem for variable metrics.. 1966. Lectures on n-dimensional quasiconformal mappings. Bers. Ann.. Ahlfors et al. Acta Math. B. Helv. [16] L. Quasikonforme Abbildungen.-Nat. Contributions to analysis.. Sci. 125-142. Ahlfors. 1-197. V. Morrey. Amer.. Kl. S. [ 141 R. Quasiconformal mappings in the plane. Springer. Virtanen. Sot. V. S. Zur Frage der Eindeutigkeit extremaler quasikonformer Abbildungen des Einheitskreises. . Math. 138 (1969) 3999406.

.

2.) 0 D( j.. For the basis we can write and satisfies an orthogonality relation. (j.) and D( j.)! I)! (pl)vJ(.-m.)!(j:. respectively.1.) @ (D( j. The problem is to decompose this tensor representation into the direct sum of irreducible representations. A$‘) in tquanturn mechanics..)!(j+m)!(j-m)! -in. but of a different appearance. Y ( +ml)!(j. -v)! X (j. Irreducible Representations Dimensional Rotation Group of the Three- x (jlW2m2 Ij. Each component is i times the infinitesimal rotation around the respective axis and is the generator of the infinitesimal rotation for every irreducible component $..1/2. The 3jsymbol multiplied by (-l)jz+j.-j-v)!(j. can be written as the sum of products of four Wigner coeffkients. Another concrete expression for the same coeffkients. j.-j2-m3(2j3 + l)-1’2 B. (D( j. They satisfy torthogonality relations. Wigner introduced the 3jsymbol. Decomposition of the tensor product of two irreducible representations D(j.Ij. 1. j).hj ALj23).& -4 Irreducible representations of the group SO(3) of 3-dimensional rotations can be obtained from irreducible representations D(j) (j = 0. ) of its tuniversal covering group SU(2) of 2 x 2 matrices with determinant 1. ef) = W(cdab.-v)!(j-j. and two corresponding sets of basis vectors.) into irreducible components leads to WI)0 D(j2)=~W9 for m. . ) of SU(2) is the 2j-fold tensor product A 0 . This is invariant under cyclic permutations of 1. x J xc (2j+ l)(j.-L. +j2.ilj2j3 ( ml m24 > =( -l)j. ef) = W(badc.-j.)!(j+j. The representation D(j) (j = 0.+v)! >. 0 A of AESU(~) restricted to the totally symmetric part of the 2j-fold tensor product space. j2.j1.+m2)! v!(j. y-.+m. the coefficients are given by an of resulton Racah algebra is a systematic method of calculating the tmatrix element (I/I. where A is a dynamical quantity and $ and $’ are irreducible components of the state obtained by combining n tangular momenta. The addition of two angular momenta leads to a +tensor representation D(j. J(j.-m. The transformation coefficient for the two ways of reduction is written in the form =JGj12+ 1)(&3+ 1) Wl.m)-fold u multiplied by a positive normalization constant (m=j.+v)!(j-jl-m. There are two ways. j. given by .) 0 D( j.+m. through the 2-fold covering isomorphism SO(3) g SU(2)/{ +I} (. The vectors ti(jm) in each irreducible represen- . to reduce the tensor product of three irreducible representations..m.60 Classical Groups I). + m2 + m3 =0 and zero otherwise..-j21. -j) defines an orthonormal basis of the representation space of D(j). 3 and is multiplied by ( -l)ilij2+j3 under transpositions of indices as well as under the simultaneous sign change of all the m’s. = ( -l)e+fmomd W(ebcf. j. Let u = (A) and u=(y) be a basis for the complex 2dimensional space on which SU(2) operates..+j.i. ef) = ( -l)o+b+c+d =mFm Wlmlh4j 2 . which we shall denote by Wm). The angular momentum j has x-.) and D(j. . Here W(abcd. and the coefficients are called the ClebscbGordan coefficients or Wigner coefficients. was obtained earlier by Wigner.3/2. The symmetrized tensor product of (j + m)-fold u and (j .p to overall phase factor (a complex number modulus 1). +j.j2m21jlj2jm). +j2+j+ -j.2. (ad) IX) =( -l)‘+/-“-‘W(aefd. ef).-jl is the Vcoeffkient of Racah.) @ O(j.)). .. called the Racah coefficient. The 6jsymbol {$} is related to the Racah coefficient ~ by W(abdc.) of two tirreducible representations of the 3-dimensional rotation group.j2. By a suitable choice of the ing arbitrary phase (which may depend .-j)!(j+j.25) Racah Algebra A.m2)(j. . W has the following symmetry properties: W(abcd.jI..fe) j=j. z-components j. General Remarks tation space are determined only u. +j. ef) = W(acbd.)) 0 D( j.353 A Racah Algebra 1316 353 (XX.

C(. while (a(ill T@IIGL’~‘) depends on the dynamics of the system. Princeton Univ..k+l. Condon and G. Clebsch-Gordan coefficients for j. Rose. b] = ab .(j. jk) For an irreducible component of the tensor product of two irreducible tensors. Pseudorandom Numbers Tables of numbers generated by random mechanisms have been statistically tested and .. Spectroscopic and group-theoretical methods in physics (Racah memorial volume). Theory of atomic spectra.“. Rose. = 512. A. Angular momentum in quantum mechanics. Cohen. De-Shalit.j. . Wiley.j.j. 26 (1963). When Vk) and Uck) operate only on the state vectors in the subspaces H.1. theoretically. of the total space (tHilbert space) H = H. M.. and M..j. is defined as the matrix element between basis vectors of CNjd x WJI x CWA x W4)l and CWJ x %)I x VU) x W4)l: (j.j.)(cCj. Edmonds. 1959. [3] U. Cj. Press. S. Elementary theory of angular momentum. Blatt. Morita. P.j. j. and fsampling or trandomization techniques in statistics. and H. Rev. In this formula the Clebsch-Gordan coefhcients are determined from group theory. Tsukada.)jmlj. Sambutsky. Phys.5) Random Numbers as A.1317 354 B Random Tensors Numbers C. S. 24 (1952) 249-257. 1959. Phys. x Hz.. Morita. See [6] and [7] for explicit formulas of Clebsch-Gordan coefficients and [S] for Racah coefficients. the 9j-symbol..(j. 3. Group theory and its application to the quantum mechanics of atomic spectra. It is a basic tool for the tMonte Carlo method.(j. and components of different CIare assumed to be orthogonal. U. T. k . B. 1957. Here [a. R. Shortley. it satisfies The 9j-symbol can be written as a weighted sum of the products of the three w’s.j.. T. G. General Remarks the matrix (aj. and M. Random numbers used in practice are pseudorandom numbers (. 1935 (reprinted with corrections. 64-74.( -l)q qk U. References [ 1] E. [S] L.jm) =(-1)j1’j.Section B).j. tsimulation of stochastic phenomena in nature or in society. Yamada. 1968. Wigner. their scalar product ( Tck). 1959. Theoret. Fano and G. -k) that transforms in the same way as the basis of D(k) under rotation of coordinates is called an irreducible tensor of rank k. Racah. That is.)j. Mod.. North-Holland. [4] A. Prog. Uck)) = C. J. Cambridge Univ. Some properties of the Racah and associated coefficients. [2] M. qkl =qT.(j. and I. Academic Press.) The last factor. [S] F. . 1951). Press.ba.U(k))la. and 712. .~~ U(k2)IIalj. Bloch. C. respectively. Irreducible tensorial sets. [TV+’ 0 U’kz’]: 354 (XVl. Suppl.. Biedenharn.jl) A sequence of numbers that can be regarded as realizations of independent and identically distributed trandom variables is called a sequence or table of random numbers. The matrix element of this quantity between two irreducible components can be written in the form where CL is a parameter to distinguish multiple components with the same j.+ij. [6] E. E.j. . [7] M.)j. E. the definition of random numbers leads to an algorithmic approach to the foundations of probability =J(2k+1)(2j+1)(2jf+1) x )$rjl I/ T(“~)lla’j.. Academic Press.” has the matrix element (a1C(2jlj2jml(T(k).jll[T’kl’@ can be written Uck2)]@lldj. Irreducible A dynamical quantity qk (q = k.~kl==(kTq)(kfq+l) CL.j.-jW(j. Talmi. R. H.)jm) =JGj12+ l)Ch+ 1)(2j13+ 1)(2L4+ 1) CL21.

D.. A modified sequence xk = k2Cr (mod 1) is known to be random for any irrational s( in the sense that the tserial correlation N-’ C.}~X. The sequence thus generated might be cyclic with a short period.detine &(xIY)= min(log. xLxL+.N=n”(n=a computer word length). numbers generated by certain simple algorithms can be viewed practically as a sequence of random numbers. Each of them is written in terms of simple computer instructions. For infinite sequences. Kolmogorov-Chaitin Random Sequences Complexity and Finite As Shannon’s entropy is a quantity for measuring the randomness of random variables. this is called a sequence of pseudorandom numbers. 1. uk. . .. Weyl considered sequences f(k) = kcc (mod l). otherwise the mixed congruence method. . 2. which is approximated by the discrete distribution on (0. Martin-Liif [6] discussed a relation between complexity and randomness. Among them the use of +order statistics and acceptance-rejection techniques have wide applicability. though they can be used for some special purposes. (4) H.. We repeat this process and obtain a sequence of pseudorandom numbers. as well as by observing the overall properties. for no n). Statistical Tests published. tL by the test at the level 1 --2-M-C. k= 0.) have been devised. Random numbers with distribution function F( .) . . The points (u.. that is. multinomial sample. ukl+r-r). Then there exists a constant C independent of L and M such that implies the acceptance of the decimal sequence 5. (1) The middle-square method was proposed by von Neumann.n)=x The function A is said to be asymptotically optimal if for any B there exists a constant C such that K. (3) Observe the rank orders of a set of random numbers.. . For the generation of uniform pseudorandomnumberson{O.. E. . y E Y and a partial recursive functionA:Yx{l.(x 1y) < K. which is known to exist.+r. and test the randomness of their permutations (test the number of runs up and down). (2) The +Fibonacci sequence {u.(x) y) is simply denoted by K(x 1y) and is called the Kolmogorov-Chaitin complexity of x given y. . To check uniform random numbers on (0. computation tricks avoiding the direct computation of F-‘( .l. The lengths from initial values to the terminal cycles are empirically checked. the distance of the empirical distribution function from that of the theoretical one is tested by the +KolmogorovSmirnov test.(x 1y) + C for any x E X and ye Y. This condition on the complexity is satisfied by at least (1 -2-M) 10L sequences iamong the decimal sequences of length L. . +Goodness of tit can be tested by the +chi-square test.1) (N >> 1).“=. I). For typical distributions. i co (if A(y.l) into subintervals. however. and u0 that make the cycle maximum for given n and s are determined by number theory.354 c Random 1318 Numbers C. K. where c( is an irrational number and k = 1.} defined by uk+r --uk-r + uL (modn”) is apparently regular.l/4 converges to 0 uniformly in 1 as N+co.). then the frequency of random numbers falling into these is a. P. the minimum k such that uk = uO. and the constants a. possibly after many repetitions.. 1. whose values are uniformly distributed on the interval (0.l) the following tests are used: (1) Divide (0. such as uniformity of the frequency of patterns of subintervals in which a set of random numbers falls. the following algorithms are used. Collective and Infinite Random Sequences For finite sequences. To generate random numbers on a large scale. the procedure is called the multiplicative congruence method. The cycle.nlA(y. = auk + c (mod n’) or (mod ns k 1). can be used.) are obtained by transforming uniform distributions by F’( . the Kolmogorov-Chaitin complexity [4. They are not independent. We square an integer of s digits of radix (or base) n and take out the middle s digits as the next term. Consider any statistical test for the randomness on the set of (say) finite decimal sequences which is effective in the sense that it has a finite algorithm. If c = 0.N-l}. n)=x). c. . the notion of randomness is obscure by nature. For constructive objects x E X. (2) For a set of random numbers. . he on a small number of parallel hyperplanes in the /-dimensional cube. For an asymptotically optimal A..5] is that of individual objects based on logic instead of probability. Good choice of the constant a makes the sequence quite satisfactory. but it is uniformly distributed.. 2.. Distribution of random numbers that are easily generated and suitable for general use is the continuous uniform distribution on the interval (0. N . electronic devices based on stochastic physical phenomena.2. (3) The congruence method [3]: Define a sequence by uk+. independence can be tested by observing the frequency of transitions of subintervals in which a pair of consecutive numbers falls. For digital computers.l). . such as thermoelectron noise or radioactivity.

For a class Ic/ of selection functions. The definition of random sequences.“‘. W.. Almost all real numbers are normal to any r. 13.. fi. on Large-Scale Digital Calculating Machinery. (xy)z = x(yz) (associative law). tZ. in the sequence x1.1319 355 A Real Numbers [4] A. Uniform distribution of sequences. On normal numbers. 661-672. there exists one and only one number w E R. 12. Information and Control.< z. Church [7]. Furthermore. 3. [7] A. = (b.. . . The art of computer programming II. 1. 10.obey the usual laws (with the single exception of division by zero). . Furthermore. a definition of infinite random sequences has been given by A. where{n. A second classified bibliography on random number generation and testing. ye R. Mathematical methods in large-scale computing units. PER. R. A selection function is a (0. Church. there exists a unique number 1 (unity)ER such that lx = x for every XER (existence of tunity element). 16 (1973). Algorithmic information theory. infinite sequence <i. called their sum and denoted by x + y. Information Theory.. [9] W. 3. Rev. By definition. . Schmidt [9] proved that the normality to base r implies the normality to base p if and only if logr/logp is rational. the relation < obeys the transitive law: x < y and y < z imply x . and the same thing holds for any q-subsequence with (PE $. Amer.. D.. . then x is said to be normal to base r. .(p(~. (ii) Order and . 1. 1974. 121-149. A real number whose decimal expansion is random in the above sense is normal to base 10. Axioms for the Real Numbers ClOl. . . . an infinite decimal sequence is called a @collective if each of the numbers 0. one and only one of the following three relations holds: x < y. ch. 9. e. Owing to properties (i) and (ii). .B. Proc. clearer definition is possible. r . . 6. Statist. 10 (1960). If Nn(x.. &i)=l}. Harvard Univ. 350359.[x] = Z x.. 8. Addison-Wesley. Math. . Math. 46 (1978). . second edition.. .<. . Wiley-Interscience. x = y. a necessary and sufftcient condition for a selection function cp (for which (p(l. . On the concept of a random sequence. von Mises. for each x. 47 (1940) 130-135. . Niederreiter. . . Subsequences of normal sequences. References [l] D.xn. 9 (1966). [lo] T. IBM J.)/n+r-k as n-+cc for every k and every B. Pacific J. all tfour arithmetic operations.. for which x + y = y + x (commutative law) and (x + y) + z = x + (y + z) (associative law) hold. 662-664. [6] P. Normal Numbers Let x . . Martin-LGf. . let N&x. For a selection function rp and an infinite sequence 5 1. Res. Lehmer. F. No one has so far been able to prove or disprove the normality of such irrational numbers as rr. . [ 2. and (x + y)z = xz + yz (distributive law) hold.10) Real Numbers A. . 7. 1981. . E. Bk) be the number of occurrences of the block B. Math. Champernowne [S] constructed a normal number given by the decimal expansion 0. Also.. . (2) Order properties: (i) For each x. there exists one and only one number -XER for which x+(-x) = 0.. 1951. [S] L.}={n. The set R of all real numbers has the following properties: (1) Arithmetical properties: (i) For each pair of numbers x. 21 (1977). for each x # 0 (x ER) there exists one and only one number x-l ER for which xx-i = 1.102.. . IEEE Trans. Bull. . Based on the notion of collectives by R.. With x < y meaning x < y or x = y. 5. 2nd Symp. which makes R ttotally ordered.52. Kamae.. &. . 141-146. 11. [3] D. . [2] E. IT-14 (1968). . or x > y. in other words. For any ordered set B. Also.1.. Schmidt. [S] G. 2. Sot.4. called their product and denoted by xy. 1}-valued function on the set of (say) finite decimal sequences such that {n. N.9 be the r-adic expansion of the fractional part of a real number x. Logical basis for information theory and probability theory. Develop. Press. . Israel J. &) of numbers 0. Kolmogrov. Knuth. a random sequence is a $collective for the class $ of recursive selection functions. J. H. 89. Int. there exists a unique number 0 (zero) such that x + 0 = x for every x E R (existence of tzero element). the cp-subsequence is defined as &. G. Kuipers and H.. (ii) For each pair of numbers x.9 appears in it with a limiting relative frequency of l/10. . &) depends only on L) to have the property that the normality implies the {cp}-collectiveness has been obtained in 355 (11. however. y E R.<n.. there exists one and only one number w E R. for which xy = yx (commutative law). fi. For the converse.. R is a ttield. Sowey. Chaitin. Almost all real numbers are random in their decimal expansions. en-r)= l} is an infinite set for any d<1.1.. 602-619. . .

(a. R is an tordered field. and u. we call +m and --co positive infinity and negative infinity. h 3 x (i.. (4) If for two sequences {a. ccj)=R.. Any fundamental sequence of real numbers is convergent (completeness of real numbers). (3) Continuity property: If nonempty subsets A and B of R. In particular. A real number that is not rational is called an irrational number.* -co) and call co (-co) the limit of an. . R is a ttopological space (+order topology) that satisfies the keparation axioms T. satisfying 1a. Let {u. Assume that for each arbitrary positive number E there exists a number n. k n. For each cut (A. then we write (I. } generated by 1 can be identified with the group Z of integers.. If for each infinite interval (a. and the set Q of rational numbers is dense.E (-co. w) (U. (2) For each pair of positive real numbers a and b with a < b.}.e.. denoted as before by lima.b)={xlu<x<b). a > -m for all XER. Let {a. T4. a.} is a convergent sequence or that a..b]={xlu<x<bJ. To extend the concept of intervals. XER}. Intervals For two numbers (u. with respect to arithmetic operations and ordering. b) as an topen base.a)) there exists a number n.} be a sequence of real numbers. x = sup A = infB). This property of R is called Dedekind’s axiom of continuity (294 Numbers). The subset {m/n I m. we write B. Further..satisfyR=AUBand AflB=@ (empty set).} is called a fundamental sequence or Cauchy sequence. B) of sets is called a cut of R. If for any arbitrary positive E there (exists a natural number n. [a.xER}.. by properties (l)-(3).. any (finite or infinite) open interval is homeomorphic to R. For a set with properties 1 and 2 of Section A.E(U. [a.~)={xIu<x.<b. (a...co)={xIa<x. there exists a number XER (necessarily unique) such that for every u E A. Topology of R With the collection of all its open intervals (a. Writing fco for x.}. x E R with x > 0 is called a positive number.. . b) is called an open interval and [u. we define (-co. . . B) of R. In R every (finite or infinite) interval (including R itself 11 is tconnected.. A necessary and sufficient condition for a subset F of R to be tcompact is that F be bounded and closed (Weierstrass’s theorem). The topology of R may :also be . any finite closed interval is compact.) be a kequence of real numbers. The subset of all positive integers { 1.+ +m (a.<u*z:. Specifically. The set R of all real numbers is determined uniquely up to an isomorphism.. and (-co.) = 0. + 2. [u. and call them (finite) intervals. b E R with a < h. with a < h for each pair a E A andhEB.a)) for all n>n. then there exists one and only one number c E R with lim a. such that U. Properties of Real Numbers (1) For each pair of positive numbers u and h > a. a.db. and for every b E B.. or properties 1 and 5 of this section. b) = jxIx<b. it can be proved that property 3 of Section A is equivalent to property 3.2. x > -io. n> n. in other words. there exists a natural number n with u < nh (Archimedes’ axiom). T. We also say that {a. n. of which a and b are their left and right endpoints.bl <E for all n > n. converges to b. or properties 1 and 4.. and x E R with x < 0 a negative number. = c (principle of nested intervals). respectively..}. n E Z.bl={xIa<x<b}.xER}. and x<y and 0 <z imply xz < yz. such that la.355 B Real Numbers arithmetical properties are related by: x < y implies x+z<y+z for any ZER. and call them infinite intervals.<. WewriteIxI=xifx>OandIxI= -x if x < 0. . b] a closed interval. (a.} be a sequence of real numbers.b)={xIu<x<b}. the ‘least upper bound of A: a = sup A (tgreatest lower bound of A: b = infA) exists.-+b) and call b the limit of {a. n # 0) of R forms the subfield of R generated by 1. 1320 (5) Let {a. R is a locally compact space satisfying the second tcountability axiom.. respectively. we have <u. } may be identified with the set N of all natural numbers. The symbols x and --cu are introduced as satisfying cc > x. (3) For any subset A in R tbounded from above (below). its subgroup {0. and call 1x1 the absolute value ofx.(-m. = lim b.db. there exists a rational number x such that a <x <b (denseness of rational numbers).. then {a. co) (( -:~. In particular.. C. D. Then we write lim. = h (or a..<. R forms an additive Abelian group. a <x. + 1. It can be identified with the field Q of all rational numbers. -a.b]={xIxdb.xER}.6 n I< E for all m. lim(b. then the pair (A. {b.

Actualitts Sci. In particular. called recursive functions by them and now called primitive recursive functions after S. Hermann. and (po. M. Post and A. R as a topological Abelian group (with respect to addition) is isomorphic to the topological Abelian group R+ of all positive real numbers with respect to multiplication. . Cantor. Mayer & Miler. Also .1932. the term natural number is used to mean a nonnegative integer. following the suggestion given by J. Bourbaki. (ii) if p lies to the left of q. English translation. Furthermore. we give it by utilizing the idea of introducing schemata. Academic Press. Turing defined the notion of computable functions by introducing the concept of Turing machines. pq = p’q’ (pq and p’q’ are tcongruent)o cp(q) cp(p)=cp(q’)-a. cp(pl)= 1.+ab. A function whose tdomain and trange are both the set of natural numbers (0. Topologie gtntrale. Kleene later improved Godel’s definition and developed the theory of general recursive functions [2]. Then the following problem naturally arises: How shall we define a finitary method? In other words. 1932. Then q(p) is called the coordinate of the point p. Xl. there exist topological mappings f : R +R+ withf(x+y)=f(x)f(y) and g:R++R with g(xy)=g(x)+g(y). say from left to right. Ind. ch..87 Convergence). General topology pt. 356 (1.1321 356 A Recursive Functions defined by the notion of convergence (. L.x. and E. Elements of R/Z are called real numbers mod 1. for variables ranging over the natural numbers.. Dieudonne. Weierstrass. a natural extension of the notion of primitive recursive functions. 189441927. were found to be equivalent. z. third edition./b. a. 1. instead of giving the definition of recursive functions in the original style (the Herbrand-Godel-Kleene definition). A Euclidean straight line with a fixed frame is called a real line (identified with R by the mapping cp) and is usually denoted by the same notation R or R’. Kleene [2] (the definition is given in Section B). These notions. any proper closed subgroup I of R is discrete and isomorphic to the additive group Z of integers. then J g are uniquely determined and are written f(x)=aX.b. g(b)= 1. and (iii) for two line segments pq and p’q’ (where p and p’ are to the left of q and q’. introduced independently and almost simultaneously.422 Topological Abelian Groups). } is called a number-theoretic function. + b. 1930. Church and Kleene defined I-calculable functions using the l-notation (Church [4]). [S] K. Gesammelte Abhandlungen. References [l] N. That is. how shall we characterize a number-theoretic function that is effectively computable. Foundations of modern analysis. enlarged and corrected printing. 1960. for some a>0 we have T={neln~Z}.+b. [4] J. [2] G.9) Recursive Functions A. C. Elements de mathematique III. then q(p) < cp(q). pi be two distinct points on 1. 1969. . A. respectively). g(x)=log. or provided with an algorithm of computation? Gijdel defined the notion of general recursive function by introducing a formal system for the elementary calculation of functions.. 1143c. Akademische Verlag. 1. To be precise. General Remarks E. Herbrand. Dedekind. Then there exists one and only one bijection cp from the set L of all points of 1 to R satisfying (i) ‘p(pJ=O. Hence R is a ttopological field (regarding the characterization of R as a topological group or a topological field . Nombres reels.-*a/b (where b #O. with p. Gesammelte mathematische Werke I-III. pJ the frame of the line 1. 1966.. the quotient group R/Z as a topological group is isomorphic to the rotation group of a circle (l-dimensional ttorus group).2.-+a + b. Hence such functions are now simply called recursive functions.+a and b. then a.-mb. We employ the letters x. Hilbert (1926) and K. b. y. Addison-Wesley. Here.references to 381 Sets. Iff(l)=a. #O). Braunschweig. Let PO. Gesammelte Abhandlungen l-7. In this article.. 1960. situated to the left of pl. 4. a. Godel introduced an efficient method of arithmetizing metamathematics based on representing certain linitary procedures in metamathematics by primitive recursive functions. [3] R. . The Real Line Let 1 be a Euclidean straight line considered to lie horizontally. Godel [l] considered certain number-theoretic functions. Regarding R as a topological Abelian group (with respect to addition).. Springer. and a.b. x2. Arithmetic operations in R are all continuous: If a.

(III) (IV 4)(X .. by k-fold recursions for every positive interger k PI.Y)=o) for the definition that satisfies vx. 0)... ... Now..b). . Primitive 1322 Functions Recursive Functions five definition Similarly. if 3yR( y).. pyR( y) is undefined. otherwise. .)=0.) that results from the application of 51 to functions tjl.R(x.. Robinson (1947).. we always transform it by replacing the predicates contained in it (if any) by corresponding representing functions... .). of cp from any function II. Thus. 1936) inveslzigated in detail functions that are definable: in general. it equals z. pi (the (i + 1)st prime number. General Recursive Functions Then we call P a primitive recursive predicate if its representing function cp is primitive recursive. x. (1.. XJ of natural numbers. min(a. . a’b. (II).x* ...) if cp takes only 0 and 1 as values and satisfies P(x. . x. . . The facts. . $[. .“‘. There are further investigations on primitive recursive functions by R. + (-..... A function is called primitive recursive if it is definable by a finite series of applications of the operations (IV) and (V) ($.. The following functions and predicates are examples of primitive recursive ones: a + b. when cp is definable by applying a primitive recursive operation to * I. i<y A function cp is said to be primitive recursive uniformly in $1.y)=O). $I if it is definable by a finite series of applications of (IV) and (V) starting from $1.x.x. . .. .. (a). .356 B Recursive B.(x..x.. A general recursive predicate is a predicate x. . . Almost all results mentioned in this section were given by Gadel [ 11... We say that a function cp(x.... y) holds. Then an operation R is called primitive recursive if the function or the predicate a($. .“> x. . otherwise. a primitive recursive function is general recursive. . otherwise.. la-bJ. including the ones concerning relativization. . x.$h. .)=~(x..1 X..x2 where /.. . are already-introduced functions) starting from functions each of which is given by (I). x. in addition to those used to define the primitive recursive functions. . . tjr and from functions each of which is given by (I). . p. VyY. . 4)(X.) = xi (1 . . &. . . (II).. definitions by cases.) X&l >. y)=O) is not necessarily defined for each n-tuple (x 1. Pr(a) (a is a prime number). x..1 x. such that its representingfunction is general recursive. a=b... .> *. . .411 Symbolic Logic). . the following operations are primitive recursive: the finite product n:y<rr the logical connectives 1. the bounded quantifiers Sly. . I+m>O) is primitive recursive in +. . $. Then cpis primitive =~y<I..X.b). $J of the definition of primitive recursive functions as follows: A function is called primitive recursive in $1.. Consider the following schemata: (I) (11) q(x) = x’ (=x+1). where li/( ) is a constant natural number if n = 1. (the exponent of pi of the unique factorization of u into prime numbers if a #O. p1 = 3. or (III). . ) / x. For a predicate R(y) on the natural numbers.). a function is called a general recursive function (or simply recursive function) if it is definable by a series of applications of schemata including a new schema (VI) dXI> . Whenever we are given a concept or a theorem. . a ( b (a divides b). P&er (1934)..Q. Q. Note that a function definable by a +double recursion is not necessarily primitive recursive. y). = 2. Given the functions G1. The following operation is also primitive recursive: cp(Y. that are valid for pri- . R. x2.. Put cp(x. we define the relativization (with respect to +..<i<n).. .y) is the least y such that y<z and R(x. . ). ..x1. . . . 4)(X. Ptter (1935. u < b. py($(xl..)=q (4 a given natural number). VXJY(~(X... . .. .). .x. v . and the bounded p-operator pyYcz defined :as follows: py.. recursive in $.. ... . a”. C. z) The following p-operator is used to define general recursive functions by extending primitive recursive functions. . . pyR( y) is the least y such that R(y). . x. . Generally./ x. .. . or (III).@(y.1)> =x(y. . . ...x. .. . $I and predicates Q..). .:c. . X”)O’p(Xl. Q.)=~Y(lcl(x. A. by (definition.maO. a!.) is the representing function of a tpredicate P(x 1. max(a. if there exists such a number y.x. and the finite sum ZYcZ is a primitive recursive operation in this sense. and others. . Q 1.x.

. . .y) such that x~Eo!lyR(x. XJ or I.y. . y) that is primitive recursive in y(z. .. A natural number e for which (3) and (4) hold is said to define cp recursively in Y or to be a Giidel number of q from Y.. q(l). . .. x. A set E of natural numbers is recursively enumerable if and only if there is a primitive recursive predicate R(x. . . Recursive Enumerability .(z. . y) and a primitive recursive function V(y) such that given any general recursive function rp(x. y)} is an example of a set that is recursively enumerable but not recursive. . .). D. The empty set is also considered recursively enumerable. . . the definition seems to be satisfactory.y.) (ii) P(x..). x. x. . . a natural number e can be found such that Vx. (x. wherex.(e. JII (abbreviated Y) be any given functions.. . . a. . . $J when cpis general recursive uniformly in Y... B.I.. .y). . where U(y) is the primitive recursive function mentioned in Kleene’s normal form theorem. . We can relativize Kleene’s normal form theorem with respect to them as follows: For each n. x. x. It is known that in this definition “general recursive” can be replaced by “primitive recursive” (.. . . the set C is said to be complete for the class of recursively enumerable sets. x.x. Furthgmore. which asked whether the sets that are recursively enumerable but not recursive have the same tdegree of (recursive) unsolvability as that of C..97 Decision Problem).. .. . 6) and l-WI... . A recursively enumerable set E is general recursive if and only if there is a general recursive predicate R(x. A number-theoretic predicate P(x 1. . . ....R(x. . In particular. we can construct a predicate x. For each n..) such that E(x. Y such that given any function cpthat is general recursive in Y. any function with a computation procedure or algorithm can be assumed to be general recursive... . If S is a consistent system such that primitive recursive predicates are decidable within S and the predicates PfA (for any formula A.x.3yT. x. . .. q(2).. Therefore..)= ment: Every effectively calculable function is a general recursive function. Church (1936) proposed the following state- A set {q(O). and the concept of effectiveness used in tsemi-intuitionism is clarified using general recursive functions (. . and it has the following remarkable property: For every recursively enumerable set E.lP(xl..xx. a Gijdel number e of cpfrom Y can be found independently of Y (except for I and the respective numbers of variables of $I) . Rosser. . Post [6]). x.Let $. . ~~.)o~y..3y~‘(e. x... . . A. 3y.x. .Vx.. . Friedberg (1957) and A. .. dx.. various decision problems have been negatively solved (. as mentioned at the beginning of this article. .) is said to be decidable within S if there is a formula P(a.x. .. we can construct a primitive recursive predicate T.. (3) (4) UWGy(e. So Church’s thesis and its converse provide the exact definition of the notion of effectively computable functions. (1) (2) 4$x. x. .. .x. .1323 356 D Recursive Functions mitive recursive functions are also valid for general recursive functions. traditional descriptive set theory can be reinvestigated from this point of view. x.... In this sense.x. ....x. a natural number e can be found such that VX . .. a.)... . Though this notion is somewhat vague and intuitive. . . Utilizing this. . then a necessary and sufficient condition for P to be decidable within S is that P is a general recursive predicate (A..y) (Kleene [2]). (Here “general recursive” can be replaced by “primitive recursive. . . Mostowski. Muchnik (19561958)..x.y). ..x.) of natural numbers (the symbol k means provable in S).) of S such that for each n-tuple (x1. .. Post’s problem... . yl..xl. .. y.Y)).xl. . .x..) (with no tfree variables other than the distinct variables a. . x. y) such that x&lZo3yR(x... y) means that y is the GSdel number of a proof of A@ . 1947). . .“) We call a set E a recursive set if the predicate x E E is general recursive..22 Analytic Sets).. .. Any natural number e for which (1) and (2) hold is said to define q recursively or to be a Giidel number of a recursive function cp. .. was negatively solved simultaneously by R.I enumerated by a general recursive function cp(allowing repetitions) is called a recursively enumerable set. . M... x.(e.) is called a recursively enumerable predicate if there is a general recursive predicate R(x. Now let S be a tformal system containing ordinary number theory.. 1936). there is a primitive recursive function such that XE E 9 (P(X)E C. x. .y)). The set C = (x j3yT. designate the numerals corresponding to x1. .Vx. The converse of this is evidently true by the definition of recursiveness... . Generally.)-l-PP(x. . . in S.y) (Kleene [S]. .)) are primitive recursive. PfA(xl.. . . .. . . . . .1= UWT. a predicate E(x 1... Kleene’s Normal Form Theorem.

. . We assign to each function cp(a:+a natural number called an index (which plays the same role as a Godel number) in such a way that it reflects the manner of application of the schemata used to introduce q(a). and $. b)= h(b). . . cp(a) is called a genera1 recursive function if it is defined for all values of the argument a..) is called a partial function if it is not necessarily defined for all ntuples (xi. .. x”) e x(x. (III) cp(a. . = used for the definition of q should be replaced by 2) starting from functions given by (I). If a predicate R(x.) is expressed as U(.(~. .... and (IX) cp(u.) N $4%X. (VIII) cp(rj. 6) = a’.) is (partial) recursive uniformly in u. a primitive recursive function . x. and 47(x..”(e.356 E Recursive Functions E.. 111.. .. 47(x. = U(pyrl-“m(e.pl.. Consider a functional (simply called a function) of a given finite number of such variables of types taking natural numbers as values....y))) is a partial function. .:‘(~. x.u.. cl. On the partial recursive functions.x. $(x1. We call a function cp(a) partial recursive if it is detinable by a finite series of applications of the schemata (I)-(VIII) (= is employed instead of = in (IV)-(VI) and (VIII)). .. . a$. . x....) (a Godel number of 4” trom Y). x. then a Giidel number e of cp is found independently of s(i.. For any given natural number e. . Now suppose that 2. ‘/j.x..) can be found such that. The notion of partial recursive functions appeared first in the theory of tconstructive ordinal numbers of Church and Kleene (1963).. a. x.. We call such a functional ‘p(c~i. .x.j objects to natural numbers are objects of type j-t 1. . Therefore. c) N {u} (6.) (partial) recursive. so is the other..). Partial Recursive Functions and put 1324 A function cp(x . . In particular. Note that for the case of types < 1. (I) cp(a.. be number-theoretic functions of one variable. is a list of variables from which a is obtained by changing the order of two variables of the same type).. x.)( {e)“‘(x. x. . y. x are given functicns of the indicated variables.. . . . When a natural number e is a Godel number of 47(x. E. .y)).y. . The natural numbers are the objects of type 0. x. b is any list (possibly empty) of variables that are mutually distinct and different from the other variables Iof the schema. x. . n.. x. y) is general recursive. . . .. b) = xh da. . and also general recursive functions in the present sense are equivalent to the corresponding notions (introduced via relativizat. and the values are the same. y.))(x. ..) = U(PYT..). .) is sometimes written as {e) (x.Snm(z. x. x. . . primitive recursive functions. and (VI) (in each schema. da’. etc. (1) For natural numbers m. or ai. Kleene introduced and investigated the recursive functionals of variables of arbitrary (finite) types [lo. . (II).) is a partial recursive function of the variables of z and of xi. Then there F. {z) (x. and with it we can develop a theory of recursive functions of variables of two types. x. for any natural numbere.. We say that such a function is partial recursive (partial recursive (uniformly) in ‘I’) and that a natural number e defines cp recursively ((uniformly) in Y) or is a GSdel number of a partial recursive function 47 (from ‘I’)... . .... Denote variables ranging over the type-j objects by r’. . 6) = a. y. aj-‘. .). (VII) (p(ai. .. x. . and they are also definable by a finite series of applications of the schemata (IV).) or x(x. 61. For two partial functions $(x1. partial recursive functions.) . (VI do. . xi. Y)) (or 2 U(IIJ~. . . (II) cp(6) = q (q is a natural number). 6). . a.y)). .) and x(x.y. (2) For any partial recursive function $(z. A function cp is called a primitive recursive function if it is definable by a finite series of applications of the following schemata (I)-(VIII). Extension of Recursive Functions Number-Theoretic Functionals to Let xi. b)=a’(a).. If cp(xi.x. we write cp(a) with an index e as {e}(e).)-{S..x. .x.)). Partial recursive functions can be defined in the Herbrand-Godel-Kleene style as a natural extension of general recursive functions. the following two theorems. and (III). (P(x.. where a is a variable of type 0.. (IV db) = 44x(b)... . y) is partial recursive. Extending the notion of recursive functionals. . 6). . then pyR(x.. (VI) da) = $(a.1y7. c) (c is a finite list of variables of arbitrary types)... x.. . in general.-“m(e..x. 1 I a.. x. xn) means that if either $(x. J. given by Kleene [3].. are most useful. xi.. .) of natural numbers.xn) is defined for x.). bj.x. The following theorem is important: Let r be the maximum type of a. . b) = ccj(iccj-‘~(aj. x1.x... .je}(y.x. x. These notions can be relativized also with respect to any given functions. x.. Yaw. x. . and the oneplace functions from type.x. . b)) (j&’ designates that 1 is a function of the variables cc’-‘).. range over the set NN of all number-theoretic functions of one variable. (V). .x.ion with uniformity) in the ordinary sense already described.. 6. a natural number e can be found such that {e} (x1. I.

.g.. Kripke. to tgenera1 recursive predicates in $i.. . .e. K-recursiveness can be defined. as in the case of general recursiveness. However. . Kino. and introduced the notion of admissible sets. y. and definability in both quantifier forms. . the first nonconstructive ordinal. and defined ICrecursive functions using schemata similar to the abovementioned (I)-(VI).31. Although research following a similar line had also been done by M.e. T. by various equivalent methods. e. cc.. The fine structures of these properties are currently the objects of intensive research.. Kleene succeeded in establishing a theory of hierarchies that essentially contains classical descriptive set theory as an extreme case [S. Every infinite cardinal is admissible. E. and others. V<. be variables ranging over %>‘32.5>%. For example. . In fact. Given an admissible ordinal K.. the equation calculus... .’ predicate is admissible (Section H). x) = 0) instead of (IX) is partial recursive. i.. Davis. and others. a.. . The least admissible ordinal is w. schemata. = K. etc. . An admissible ordinal K has the closure properties required for the construction of the calculus. and infinitary logic are closely related. A predicate P(a 1.).32].p. not all the partial recursive functions of variables of types > 2 can be obtained by applying schemata (I))(VIII) and (IX’). x. Takeuti considered functions with a fixed infinite cardinal IC as a domain and a range. Using this.. yielding the notions of k-degrees and K-recursive enumerability. Takeuti introduced a notion of primitive recursiveness for functions from a segment of the ordinal numbers to ordinal numbers. .. .. . . /l< K and /I =f(a) is computable. Sets or functions are described by tpredicates. for every n 2 1. $r (la 0) if it is expressible syntactically by applying a finite number of logical symbols: +. . Platek investigated recursion theory in a still wider setting.1325 356 H Recursive recursive predicates M. there are also investigations by A.. A. be variables ranging over the set N of natural numbers. An admissible set is a transitive s-model of a certain weak set theory.0 2. G.. Mostowskii. . . ra2. Subsequently Kripke observed that the assumption that IC is a cardinal is not necessary... . r>2. v. Takeuti and A. the first ordinal not expressible as the order type of a A. In connection with recursive functions of ordinal numbers. the set NN of all tnumber-theoretic functions with one argument. Vx. and whenever a.) are also valid for K-recursive functions. Machover. w. . and the next is the ordinal wi of Church and Kleene. b. When 1= 0. . For each infinite cardinal IC there are K+ admissible ordinals of power K.a. In particular. parametrization theorem. . al. and an ordinal K is admissible if and only if there exists an admissible set A such that A n 0. he constructed a model of set theory in ordinal number theory.)(m. . Every function definable using (IX’) $(a)px($(a. Moschovakis and others [14-271.3x. 10. Most of the elementary properties of general recursive functions (e.. .g. &. (‘-l. a. a. C. Tug& S. Early treatments of recursive functions of ordinal numbers dealt only with functions on infinite cardinals. the normal form theorem. not on a segment of ordinal numbers.. The notions of degrees of unsolvability and recursive enumerability can also be generalized. a. set theory. they are called simply analytic and arithmetical. it was Kleene who succeeded in bringing the theory to its present form. respectively. Hierarchib Utilizing the theory of trecursive functions. Recent developments have shown that generalized recursion theory. Let a. consider the case I= 0.n>O. For brevity. 35. Levy. Moldestad. M. . . . and others in abstract computation theory [20&23]. i. Further developments have been pursued by J. V& it is called arithmetical in $i . Let $i. . and denote by a a finite list of variables (a. $r (I > 0).q’-‘). wo3pVf-2N(e a w (-1 3tf-2) 2 99. In addition to the abovementioned. where 0. a. . Every arithmetical predicate P(a) is expressible in a form contained .. respectively. He dealt with functions defined. when P is expressible without function quantifiers 35. . J. . which we classify as follows. then /?=f(cc) is computable in fewer than K stages. but on a set. and introduced the notion of admissible ordinals.11/# 2 0) be number-theoretic functions. H. there are some investigations by Y. and GI. A. S..... . Recursive Functions of Ordinal Numbers G. 1.m+n>O) with variables of two ttypes is called analytic in $i. N Functions are two primitive such that @)(a)woV(‘-13v]‘-2M(e. enumeration theorem. sets on which a well-behaved recursion theory can be developed. N. is the class of all ordinal numbers. Fenstad.. .

3crVa?lxR(a. . x. the theory of arithmetical and analytic hierarchies for sets (C = 0) effective descriptive set theory [28].~~~: is.u. $t.Section F) as S(Z.A(x. also. L$~v~~~~.. Kleene has extended the series of L. . Vx3xR(a.. For each k > 0. . &). VxR(a. where k is the number of function quantifiers prefixed.(x).) . A predicate belongs to AT if and only if it is general recursive (an analog of +Suslin’s theorem).e.... x. In this theorem. Y). determine the arithmetical hierarchy of degrees of recursive unsolvability. n. YX 1326 table (a): and their dual forms. of course. l7.356 H Recursive Functions in the following (4 R(a): 3xR(a. where the suffix k refers to the number of quantifiers prefixed. .. 1. . there exists a C~+.A(a. or jCi[C].cc.5()03aVxA(x. and in contrast to this. . For example./J.(a).ita(2”. 1= 0.x). (1) Va3/IVxR(a. a) (5) . we can classify the forms of all analytic predicates by the table(b): 04 A(a). . x. to the case m = 0).e. given a general recursive predicate R(a.. we can consider hierarchies of predicates which are arithmetical or analytic in a finite number of functions in C. . Similarly.” (or the class of such predicates) is denoted by A:. }.-. W.(a) is a Z. by using . Given a set C ( c NN) of functions with one argument. table (a) gives the classification of the arithmetical predicates in a hierarchy. we have a natural number e such that Vx~yR(cc. and the theorem on complete form.. it is the family {P/PEC..) predicate that is not expressible in its dual form L$‘+. x. a.“) (hierarchy theorem). we have 3a. there exists in Cf (or fl. Lk+l(a)03xTlL~(a.“‘~.x. and Z or I7 shows that the outermost quantifier is existential or universal.. k = 0. all the general recursive predicates in Cf exhaust Ai+.) (2) (3) (4) -3aA(lt(a(t)). a. This hierarchy is called the arithmetical hierarchy. For each k 2 1. .29]. When m = 0. there exists a complete predicate with respect to &?(Z7. The hierarchy given by table (b) is called the analytic hierarchy. We now restrict our consideration to predicates for natural numbers (i. . . . gard LL[C] as a class of predicates (or sets) P. x. For 1> 0 (namely. iIa.a. &VxR(a. ..l each form of predicate in (b) (or the class of all predicates reducible to that form)..y)oVx3yS(sc. These notations have been given by J.l (k > l). For Ci. Z7. respectively.“) an enumerating predicate that specifies every predicate in Cf (L$... Vx 3yR(a. in neither Cf not TZ. y) (. For k 3 1. . . z.wL. we can tuniformly trelativize the above results with respect to $i. (Post’s theorem). we have the enumeration theorem. Now let {Z[. when predicates are arithmetical or analytic in $i. . a. there is a primitive recursive predicate S such that 3ctR(a. there is a tprimitive recursive predicate S(sc.) and its dual hold. we can take Tt(z. 3xA(x)-=3aA(a(O)).P. y).3’)). .. Thus L. Such a hierarchy is called a C-arithmetical or C-analytic hierarchy and denoted by {Zf[C].x.. Addison [28. a. . .a.y) (enumeration theorem).“. Define the predicates L. that is. Vx3aAIx. . where A is arithmetical and each R is general recursive. xl. The NNarithmetical hierarchy and the NN-analytic hierarchy for sets correspond respectively to the finite Bore1 hierarchy and the !orojective hierarchy in the tspace of irrational numbers. a C. CC:+. x). when we re4. x). 3x. . and its degree is prbperly higher than that of L.a. denote by Ai the (class of) predicates expressible in both forms CL and 17. . obtain such an expression we first the given predicate into its tprenex form and then contract successive of the same kind by the formula 3x.“[C]}.x).+. In order to transform normal quantifiers where each R is tgeneral recursive. $t. .. denote by Zi. . LX.Lt(a(t)). .~W[}k(r =O.a)o3a3xS(a(x). 2. y).“). For each k > 0... 1) be the corresponding hierarchy relative to $1. respectively. x. . Concerning the function quantifiers. for Z7: and m = n = 1. A predicate that is expressible in both forms Cf and I7.oCi predicate which is of the highest +degree of recursive unsolvability among the L’f+i predicates. x).) and its “‘dual form. x. Using these facts.x).(@+.f CCI~...-.(a)* a = a. the hierarchy theorem. ~.~~~~~. by L. primitive) recursive function for its variable (theorem on complete form). . Therefore. a.a)93yS(y.” Each form in (a) (or the class of all predicates with that form) is denoted by Cf or II.) (hence.. Addison called the theory of those hierarchies classical descriptive set theory.“) predicate with only one variable such that any &“(@) predicate is expressible by substituting a suitable general (or more strictly.“(Z7. y) such that. o3xA((x).c~. <[E C. 1.. That is.2. . For any general recursive predicate R. . .z. . 3x VyR(a. .

. [3] S... A function or predicate is said to be hyperarithmetical if it is recursive in H. Inst. . H~. Studies. <‘-‘). Rekursive Funktionen. Recursive functionals and quantifiers of finite types I. for yo0.. [2] S. Amer.. . 50 (1944). Markov. For an arithmetical predicate A(a. t’-‘) 03n’-1V<r-ZR(a’. Springer. [ 1 l] S. When t = r + 1. Post. 3a’V<‘-‘R(a. A. Sot. Amer. Clarke [30] has published a detailed review of the general theory of hierarchies.~(u). table (c) gives the classification of the predicates of order r + 1 into the hierarchy. 53 (1943). [4] A.. . 108 (1963). C. Princeton Univ. and symbols of the tpredicate calculus with quantification consisting only of variables of types cr. b). [S] R. Math. . denote them by Y) if P is syntactically expressible in terms of variables of finite types.y. <r-2)./?. for a II: predicate P(u.. and the theorem on complete form (Kleene [lo]). [lo] S. Amer. l’-‘) such that 30’V~r-zP(a’. Trans. 42 (1954). A predicate that is hyperarithmetical relative to ZT. we have the enumeration theorem. a.3tl. I+$ (I > 0) (for brevity.. 1962). 106-142. We say a predicate P(d) is of order Y in completely defined functions $i. Sot. computability. Kleene.x). [lo]. Kleene. Math. J. L. 173-198. &. Symbolic Logic. otherwise E(a) = 1..nyp VxR(u. 1957).. C. These concepts and the results mentioned below can be relativized to any given functions or predicates. there is a primitive recursive predicate R(a’. the hierarchy theorem. Spector [34]). . Va’3<‘-‘R(a. a(x)) (G. . n’-i. References [1] K. and Y are functions of variables of type 0.2. 5y~0.~Y(u)~H.. Kleene. is defined for each y E 0. Recursively enumerable sets of positive integers and their decision problems. Monatsh. <‘-‘). 41-73. each predicate P(a’) of order r + 1 (r > 0) is expressible in one of the following forms: (4 B(a). . Math. 150-l 55. C. Peter. 1952. a’. LX’.cr’. Phys.) is hyperarithmetical if and only if it is general recursive in E (Kleene [lo]). . . Thus a hierarchy of degrees is uniquely determined by constructive ordinal numbers. Akademische Verlag. [9] A. In fact. predicates that are general recursive in ‘I”. . x) (Spector [35]). Trudy Mat.. This hierarchy is called the hyperarithmetical hierarchy of degrees of recursive unsolvability. Introduction to metamathematics.~~.a. and it is of a properly higher degree than that of Hz when z<. A. C. decidability. Bull. 1941. D.. Press. Math. 38 (1931). Hermes.81 Constructive Ordinal Numbers) of notations for the constructive ordinal numbers as follows [6]: H.. where B is of order r and each R is general recursive. . for any l7: predicate P. predicates (k>O) is of A:+i (Kleene [32]). 195 1. a predicate P(d) is of order 1 (of order 2) in Y if and only if P is arithmetical (analytic). Van Nostrand. A function &z. 3a’V/I’Zll’-‘R(a. Uber formal unentscheidbare SHtze der Principia Mathematics und verwandter Systeme I. 1. cr’. we have Vx 3yP(x. Kleene extended his theory of hierarchy to the case of predicates of variables of any type by utilizing the theory of general recursive functions with variables of finite types 0. Theory of algorithms (in Russian). 91 (1959). Math. Recursive predicates and quantifiers. [7] S. (6) Using these equivalences. a’.. y) a 3tlmEHYp VxP(x. C. 1965. a) is always a n: predicate (Kleene [33]). VC(‘38rVS’-1R(a. H. We have theorems similar to (2)-(4) and the following theorem and its dual for I > 2: For any given general recursive predicate P(a’. Ann. there is a general recursive predicate R such that P(u).1327 356 Ref. Steklov. General recursive functions of natural numbers. [12] H. then Hy and Hz are of the same degree (C. a).H~(a. This H. where Y.). [6] E. The calculi of lambdaconversion. Math. <‘-‘). 3 (1938). The predicates of order 0 in Y are exactly the general recursive ones in Y. Let a’ be a list of variables of types < t. Church. 284-3 16. If lyl=lzl (lyl is the ordinal number represented by y). Conversely. On notation for ordinal numbers.8r. When t > 1. t’-‘). . Kleene. As to tuniformization. Kleene.(a)a = a. Recursive functionals and quantifiers of finite types II. for 3. 112 (1936). Sot. Enumerability. (. C. .(a)o3xT. 727-742. for some y E 0. Trans. Denote by Hyp the set ( c NN) of all hyperarithmetical functions a. Trans. 3aacHypA(a. Sot. <‘-‘). = (~1 (no). but the converse does not hold in general (Addison and Kleene.. Ann. l-52. Let E be an object of type 2 defined by: E(a) = 0 if 3x@(x) =0). VI-i. [S] S. A necessary (Kleene [31]) and sufficient (Kleene [32]) condition for a predicate to be hyperarithmetical is that it be expressible in both one-function quantifier forms Ai (an effective version of Suslin’s theorem). Amer. Math.. Godel. Kleene. for the predicates P(a’+‘) in each form. Kreisel. Recursive Functions the system S. . a.

N. 1964. 20 (1955).. (When a polygon in a Euclidean plane bounds a tconvex cell. 29 (1964) 161-162. Transfinite recursions on admissible ordinals. Gandy. 1331 S. Trans. 1978. Fenstad. Amer. 313-320. [14] J.) A necessary and sufficient condition for a regular n-gon to be geometrically constructible is that n be decomposable into the product of prime numbers n = 2”p. Springer. Sot.) are different +Fermat numbers (. General recursion theory.2 . E. McGraw-Hill. Symbolic Logic. Amer. Kripke... Math. Math. Symbolic Logic. Sot. [23] J. 67 (1961). [30] D. Proc. [34] C. 151-163. Fenstad. [ 181 G. Primitive recursive set functions. Theory of recursive functions and effective computability. Kleene.). Math. [27] P.(m>O). North-Holland. Hinman. Math. Recursion theoretic hierarchies. (2) Its vertices are all surrounded alike. Kleene. [22] J. Mem. Thus regular polygon sometimes means the convex cell bounded by a regular polygon as described above. XIII (1971). [ 171 M. 1). [ 191 T. Rogers.. Fund. Spector. Math. Hyperarithmetical quantifiers. and take a point on the line perpendicular to the plane at the center of the polygon. J. 1980. Math.. When the number of vertices (which equals the number of sides) is n. E. Fund. J. That is. J. we call it a regular polyhedron: (1) Each face of 3.. Moldestad. Fund. Tug&. Fenstad. Sot. Regular Polyhedra Consider a regular polygon on a plane. Bull. and G. C. Computation in higher types. it is called a regular n-gon. Springer. 48 (1959-1960).. [20] J. Descriptive set theory. C. W. C. 226-252. Amer. 46 (1958-l 959) 3377357.357 A Regular 1328 Polyhedra [ 13) H. [32] S. Recursive well-orderings.179 Geometric Construction). 46 (19581959) 123-135.. Symposia in Pure Math. On the partial recursive functions of ordinal numbers. Symbolic Logic. is a regular polygon. Sot. Jensen and C. Sacks (eds. Gandy. 1975. Kleene.. [21] J. [35] C. Separation principles in the hierarchies of classical and effective descriptive set theory. 16 (1964). Japan. Arithmetical predicates and function quantifiers. 143-176. Machover.). R. Clarke. Hierarchies of predicates of finite types. 14 (1959) 23-40. 0. Spector.. J. Springer. Quantification of numbertheoretic functions. [25] Y. 1977. The set of points on all half-lines joining this point and points on the polygon (considered as a convex cell) is called a regular polyhedral angle having this point as vertex (Fig. we obtain a regular polyhedral angle. N. Sacks (eds. R. Moschovakis. these regular polyhedral angles are all congruent to each other.. and G. Generalized recursion theory. Intinitary logic and admissible sets. We call the center of these circles the center of the regular n-gon. Karp. 1980. 575-578. North-Holland. Bull. Addison. which is a 2-dimensional cell. The n vertices of a regular n-gon are obtained by dividing a circle into n equal parts. 0. North-Holland. [3 l] S. Hierarchies of numbertheoretic predicates. by the projection of 3 from each vertex of 3. North-Holland. Regular Polygons A +polygon in a Euclidean plane bounding a tconvex cell whose sides and interior angles are all respectively congruent is called a regular polygon. A. 1974. l-31. Generalized recursion theory II. Math. Math. [24] Y. W. J. [29] J. E. [ 161 S. 61 (1955) 193-213. Barwise. 1967. Barwise. 79 (1955) 312-340. Admissible sets and structures. Springer. B.p. From (2) we see that the number of edges emanating from each vertex of 3 is . 1978. [ 151 R.6) Regular Polyhedra A. E. Fig. 30 (1965). 357 (Vl. Compositio Math. B. E. Amer.. 1 When a iconvex polyhedron 5 in E3 satisfies the following two conditions.. 1974. 34 (1969). Symbolic Logic. There exist a circle (circumscribed circle) passing through all the vertices of a regular n-gon and a concentric circle (inscribed circle) tangent to all the sides. [26] J. Sot. Addison.. Some consequences of the axiom of constructibility. The theory of transfinite recursion.wherethepi(i=1. and all faces of 3 are congruent to each other. Takeuti. this 2-cell is sometimes called a convex polygon. G. [28] J. Moschovakis. A formalization of the theory of ordinal numbers. Elementary induction on abstract structures. 295-317..

8). 6 Regular dodecahedron. The tetrahedron is dual to itself. 4 Regular icosakedron. 6) (see also see Table 1). From a given regular polyhedron. there exist concentric circumscribed and inscribed spheres whose center is the center of symmetry of 5 and is called the center of ‘& Drawing tangent planes to the circumscribed sphere at each vertex. Corresponding to these polyhedra. and dodecahedrons (Fig. cubes or hexahedrons (Fig. icosahedrons (Fig. For a regular polyhedron 5. We say that the given regular polyhedron and the one obtained in this way are dual to each other. we can obtain a regular polyhedron dual to the given one (Fig.I \ \ \ (1) r=%otEtan2 p 2. respectively. Fig. BR . 3). C. 2 Regular tetrahedron. 4). hedron are dual to each other. Fig. 5 Regular hexahedron (or cube). Fig. f3 the magnitude of the dihedral angle at each edge. Then the following relations hold (we assume that each face is a regular p-gon and q faces meet at each vertex): I’ ’ N-f’ ‘\ \ /’ ’ . we have finite subgroups of O(3) called regular polyhedral groups (. In a regular polyhedron. Regular Polyhedra in 3-Dimensional Number Vertices 4 6 12 8 20 of (see Table 2).’ .1329 357 c Regular Polyhedra independent of the vertex. Fig. octahedrons (Fig. 3 Regular octahedron. 5). 2).15 1 Finite Groups). I Fig. and R and Y the radii of circumscribed and inscribed spheres. as are the icosahedron and dodecahedron. It has been known since Plato’s time that there are only five kinds of regular polyhedra: tetrahedrons (Fig. The octahedron and hexaTable 1.[ @ ‘\ ‘\ \+--. let a be the length of an edge. 8 Fig. Higher-Dimensional Cases It is possible to generalize these considerations to higher dimensions to define inductively Euclidean Number Edges 6 12 30 12 30 Space E 3 of Number Faces 4 8 20 6 12 of Number of Faces around a Vertex 3 4 5 3 3 Figure Regular tetrahedron Regular octahedron Regular icosahedron Regular hexahedron Regular dodecahedron Face Equilateral triangle Equilateral triangle Equilateral triangle Square Regular pentagon .=tanptanq AR Fig. we can obtain another one by taking as vertices the centers of all the faces of the given polyhedron (Fig. 7).

Anschaulithe Geometrie. Y. . For example. Press. [2] D. Armand Cohn. x divides y. y) is sometimes written as xRy. Y be two sets and x. b) is true or false for each pair (a. (1) 1330 l/2 l/G lhGi77 2fi Table 3.2” R/a G/4 a/2 1/a fi(ti 4 20 2/3 138” 11’22. Hadamard. When n = 4 we have 6 kinds of regular polyhedra (Table 3). n > 4. Regular complex polytopes.’ Number n-hedron (2n . . Coxeter. Rademacher. 1974. Numerical of sine 2ti/3 1 2ti/3 2/G e 7OO31’43. b) in the Zartesian product X x Y. [3]E. Regular Polyhedra Table Number Faces 4 6 8 12 2. Regular b: dual to each other Polyhedra in n-Dimensional Regular Polyhedron Euclidean in R” . b: dual to each other regular polyhedra in E”. 1906. i. A proposition R(x.2)-hedron n-hedron n+l 2n 2” Space (n > 5) Number Vertices n+1 2” 2n of Duality a b Figure Regular Regular Regular (n + I)-hedron Zn-hedron 2”-hedron Kind Regular Regular Regular a: dual to itself. M.4” 116O33’54.357 Ref. Let X. S. 1973. For n > 5 we have only 3 kinds (Table 4) (. S. In its wider sense the term relation means inary relation (n = 1. Springer. x . General Remarks References [ 11 J. Dower.411 Symbolic Logic G). second edition.70 Complexes).2) Relations A.3. 1952. For a given relation R. Cambridge Univ. [S] H. 358 (11. respectively. third edition. Methuen. A relation R(x.e.6” 90” 109O28’16. 1948. . Springer. we define its inverse Then R is the relation R -’ by yRm’xoxRy. . 1934.2. Lecons de gkomttrie ClCmentaire II. Vorlesungen iiber die Theorie der Polyeder.8” 2fi 4ti +I> r/a ti/12 Values for Eqs. but in this article we restrict ourselves to its most ordinary meaning.. y is called a relation or a binary relation if it can be determined whether R(a. y) containing x. Regular Polyhedra in 4-Dimensional 3-Dimensional Euclidean Polyhedra Number 5 8 16 24 120 600 Space E4 Number Vertices 5 16 8 24 600 120 I of Duality a b a b Regular Figure Regular Regular Regular Regular Regular 5-hedron 8-hedron 16-hedron 24-hedron 120-hedron 600-hedron Kind Tetrahedron Cube Tetrahedron Octahedron Dodecahedron Tetrahedron a: dual to itself: Table 4. [4] H.425-427.y is even. Cohn-Vossen. Geometry and the imagination. y be two variables taking their values in X. Hilbert and S. if both X and Y are the set of rational integers. Coxeter. Chelsea. Regular polytopes. ) (. then the following propositions are relations: x < y. 1932. Steinitz and H. M. to the case n = 2. English translation. S.

: X x Y+ Y are the kanonical projections. namely. and Y the final set of I. and r. = (G. Then the correspondence I =(G.21) Relativity A. Y). and there are some doubts about the limit of its applicability.3 11 Ordering). Studying these results. if R and R-’ are identical). and this theory has even become a guiding principle for developing new theories in physics. Einstein and is composed of special relativity and general relativity. References See references to 38 1 Sets. For x E X.Given a correspondence I = (G. which extended Galileo’s relativity principle of tNewtonian mechanics to telectromagnetism and radically revised the concepts of space and time.y)~Go(y. symmetric. it was believed that electromagnetic waves propagate through the ether. Y. W. o r. X.orl)andthelaw(r. Suppose that we are given a relation xRy (xoX. which consists of elements (x. R is called antisymmetric if xRy and yRx imply x = y. X. is called the graph of the relation R. inverse relation of R-‘. transitive. If G is the graph of a relation R. respectively. A relation xRy (x E X. Conversely. and the inverse relation of x is a divisor of y is y is a multiple of x.1331 359 B Relativity assume that to any x belonging to the domain A of I there corresponds one and only one ye Y. in 1905 Einstein proposed the special theory of relativity. z) E G2. I is called a one-to-one correspondence. but it is hard at present to test theoretical results experimentally. but all these attempts were unsuccessful (A. a hypothetical medium. Y) is called a correspondence from X to Y. there exists a unique relation R with the graph G given by xRyo (x. Effects due to general relativity other than those just described have been studied to a considerable extent. Z)E Go there exists ys Y satisfying (x. then G-’ is the graph of the inverse relation R-‘. A. 359 (XX. The set X is called the initial set of the correspondence I. E. X) is denoted by I--’ and is called the inverse correspondence of I. Let I be a correspondence from X to Y. Y) by its graph G. For a subset G of X x Y. Its conclusions about the solar system are compatible with observed facts that are regarded as experimental support for the theory. a correspondence I determines a relation R. We have (I-‘)-’ = I. I(x) = {y} for any XE A. X. X. A number of experimenters tried to find the motion of the earth relative to the ether. Michelson. For given sets A and B. and . and antisymmetric relation is called an tordering (.orl)-l=r. If I and I-’ are both univalent correspondences. The domain of the correspondence I is the range of r-i. B. where prx :X x Y-+X and pr. A relation R is called reflexive if xRx holds.l. for any subset G of X x Y. Z) is denoted by I.o(r. By extending special relativity. the inverse relation of x < y is y > x. Morley). and conversely. y) of the Cartesian product X x Y satisfying xRy.. Its principal part is a new theory of gravitation containing Newton’s theory as a special case. and transitive relation is called an equivalence relation (. X. Given a correspondence I = (G. Y). y) 1xRy}. are called the domain and range of the correspondence I. Almost all the conclusions of special relativity theory are now confirmed by experiments. YE Y). R is called symmetric if xRy o yRx (namely. y) E Gi and (y. Then I’ is called a univalent correspondence. Y)E G. Correspondences For a subset G of the Cartesian product X x Y. the sets A = prx G and B = prr G. Y) and I. Z).381 Sets C). the triple I = (G. A reflexive. X. We have the associative law (r. Then the set G = {(x. A reflexive.x)~G-‘. We define a subset G of X x Z by: (x. Special Relativity In Newtonian are described mechanics.135 Equivalence Relations). the correspondence (G-l. Y. a univalent correspondence with domain A and range B is called a tmapping (or tfunction) with domain A and range B (. y E Y) determines a correspondence r = (G. natural phenomena in a 3-dimensional Euclidean r. y)~ G} is denoted by G(x) or I(x). and we say that any element y of G(x) corresponds to x by I. Suppose that we are given correspondences I1 = (Cl. A reflexive and transitive relation is called a tpreordering. In the example above.lo The theory of relativity is a system of theoretical physics established by A. the set {ye Yl (x. we define a subset G-‘of YxXby(x.) o rl = r. and vice versa. o I1 and is called the composite of I. History B. Einstein established (1915) the general theory of relativity. R is called transitive if xRy and yRz imply xRz. Toward the end of the 19th century.

To illustrate this conclusion we consider electromagnetic theory. Cartesian coordinates. y. A. Einstein solved successively the problems of the Lorentz-Fitzgerald contraction. xi’=x’. x’. y’. Einstein started with This quantity transforms as an talternating tensor of degree 2 under Lorentz transformations. are i=1. z). Here (x.2. however. b gob V” Vb. it may safely be said that special relativity is a theory of invariants with respect to the Lorentz group.b 1332 where a. Even if the magnetic field does not exist in one inertial system. and physical laws are written in tensor form and are invariant under Lorentz transformations of coordinates. (1) Such transformations form a l-dimensional subgroup of Go with u as a parameter. y. = 0. the special principle of relativity is a generalization of the Newton-Galileo principle of relativity. To summarize mathematically. z) and x’ = (ct’. the factor group G/G. i=l. The group of motions in Minkowski spacetime is called the inhomogeneous Lorentz group. On the other hand.3.b where cj and ci are constants.359 c Relativity space considered independent of time.. Relativity and Electromagnetism xi’=~c:x~+-ci. In special relativity. The transformations with ci = 0 are usually called Lorentz transformations. or < 0. 2. namely.x3)=(ct. and the group G composed of these transformations is called the homogeneous Lorentz group or simply the Lorentz group. This space was introduced by H. A nonzero vector I’ is called timelike.x2. This is the mathematical expression of the special principle of relativity. where V2 = C. I C. y. a. z) are spatial t is time. and along this line of thought. Historically. the electric charge density p and the electric current density J are unified into a tcontravariant vector with . and Fresnel’s dragging coefficient. . By means of it Minkowski gave an ingenious geometric interpretation to special relativity. u+v w=ix@+ Elements ~~~0’~ fj:xO’=xO of G not belonging --a.= L. null (or lightlike). Lorentz. Minkowski and is called Minkowski space-time. IuI<c. to G. and c is the speed of light. L. or spacelike according as V2 > 0.2. electric and magnetic fields are considered in relativity to form one physical quantity with components Both T (time reversal) and S (space reflection or parity transformation) have aroused much interest among physicists. 0 CgabCqCjb=gijr o. If Go denotes the tconnected component of the identity element of G. the proper Lorentz group. In view of this. (ii) Principle of invariance of the speed of light: The speed of light in a vacuum is the same in all inertial systems and in all directions. the field can arise in another system that moves uniformly relative to the original system. The electric field E is usually represented by a “polar vector” and the magnetic field H by an “axial vector” in a 3-dimensional Euclidean space. (x0. In special relativity. A frequently used element of Go is x-vt t -(v/2)x Lu:xr=Jjq7 Y’=Y. In like manner. t’=*p’ z’=z. it is postulated that space and time cannot be separated but are unified into a 4-dimensional pseudo-Euclidean snace with the tfundamental form ds2=Cgahdxadxb=c2dt2-dx2-dy2-dz2. b=O. From these assumptions Einstein derived (1) as the transformation formula between inertial systems x = (ct. We call G.3. xiI. These are important concepts in special relativity.. the dilation of time as measured by moving clocks. in all coordinate systems that move relative to each other with uniform velocity. Since the transformation (1) tends to a Galileo transformation in Newtonian mechanics as c + co. x. and the composition law of the subgroup is given by L. -xi. z’) that move relative to each other with uniform velocity u along the common x-axis.2) and of order 4. This was the first step in special relativity. Its elements can be written as the following two postulates: (i) Special principle of relativity: A physical law should be expressed in the same form in all inertial systems. irrespective of the motion of the light source. the transformation formula (1) was first obtained by H. 1. the aberration of light. x’. but his theoretical grounds were not satisfactory. under the assumption of contraction of a rod in the direction of its movement in order to overcome the difftculties of the ether hypothesis. is an Abelian group of type (2. a physical quantity is represented by a ttensor (or a scalar or a vector) in Minkowski space-time. 3 and x. the Doppler effect.

1964). V. can be written in tensor form: where Cg’“gja=Sji. S. respect to Lorentz s=(s0. Fitch. The special principle of relativity also showed its validity in the electron theory of P. 1956. N. that is. von Eotvos (1890) and others. Similar results have been obtained for time reversal (J. and accordingly even a particle at rest has energy mc2 (rest energy). momentum. the equation of motion of a particle in a gravitational field is given by D. 1957).s1. Dirac (1928) and the quantum electrodynamics of S. and accordingly fictitious forces due to acceleration (such as centrifugal force) cannot be distinguished from gravitational force.R/2 = 0. This conclusion has been verified experimentally by studies of nuclear reactions and has become the basis of the development of nuclear power. however. respectively. This had been shown with high accuracy by the experiments of R.s2. Here 6/&s stands for tcovariant differentiation with respect to the Special relativity electromagnetic . and the gravitational equation satisfied by gij can be expressed as a geometric law of the manifold. Yang. the +Maxwell equations. H. Then outside the source of a gravitational field. . Usually (2) and (3) are called the exterior and interior field equations. became the motif in the development of tunified field theories. in agreement with the first principle. Cromin. Though special relativity originated in studies of electromagnetic phenomena. Tomonaga (1943) and others.1333 359 D Relativity transformations: part of general relativity is a theory of gravitation founded on the general principle of relativity and the principle of equivalence. The first principle is an extension of the special principle of relativity to accelerated systems in general. and R be the +Ricci tensor and the tscalar curvature formed from gij. with the conversion formula given by E = mc’.130 Electromagnetism). Christenson. M. General Relativity has its origin in studies of phenomena. Here the energy-momentum tensor Tj is a tsymmetric tensor representing the dynamical state of matter (energy. Such a vectors in Minkowski space-time is sometimes called a four-vector as distinguished from an ordinary vector such as J. of the Lorentz group G) is violated in the decay of elementary particles (T. the equation of motion for a charged particle in an electromagnetic field can be expressed as where e and m are the charge and mass of the particle. Since a physical quantity is represented by a tensor on the space-time manifold. The second principle claims that gravitational and inertial mass are equal. J. It requires that a physical law should be independent of the choice of local coordinates in a 4-dimensional tdifferentiable manifold representing space and time (space-time manifold). This shows the equivalence of mass and energy. o. It has. namely. and R.g. while the central if the particle mass is so small that its effect on the field is negligible. C. the corresponding space-time structure is altered. Lee and C. If the electromagnetic field Fij and the current four-vector s are thus defined. it has gradually become clear that the theory is valid also for other phenomena. which was not seen in the older physics. Gravitational phenomena are thus reduced to properties of the geometric structure of the space-time manifold. Jzlc). A. J. L. D. R. gij must satisfy Gij = R. physical laws are expressed in tensor form. and s is the arc length along the particle trajectory (. Wu et al. the basic equations of electromagnetism. The tfundamental tensor gij of this manifold represents the gravitational potential. flat Minkowski space-time is changed into a curved 4-dimensional manifold with +pseudo-Riemannian metric of tsignature (1. respectively. This idea.Jxlc. One interesting result is that the energy of a particle moving with uniform velocity u is given by E = rnc’/Jw. Let R. and inside the source.3). Next.b a In the same way. and stress). = 0.. Turlay.lc. W. Now the gravitational law proposed by Einstein is an analog of the +Poisson equation in Newtonian mechanics. (2) where K is the gravitational constant. been shown that the invariance for space reflection (namely for the coset SG. Starting from these two principles. If a gravitational field is produced by matter.s3) = (P. Einstein was led to the following conclusion.

The metric tensors are derived from the complex potential E. The dynamical evolution of space-time structure has been studied by means of the +Cauchy problem of general relativity. the fundamental form is approximately given by Newton’s gravitational potential cp as 1334 out on such problems as the gravitational field of a spinning mass. E. Choosing appropriate dynamical variables. Experimental verification of the theory of general relativity has been obtained by detection of the following effects: the shift of spectral lines due to the gravity of the earth and of white dwarfs. Stimulated by the discoveries of neutron stars and black holes and by the big-bang theory of the universe in the 1960s numerous studies of general relativity have been carried . In the static and weak field limits. Hoffman. Solutions of Einstein’s Equations The isometric symmetry of space-time is described by tKilling vectors. It is generally accepted that these results are experimental verifications of general relativity. this metric reduces to the Schwarzschild metric. L. (5) where V represents divergence in a flat space. respectively. and (2) and (3) reduce in this limit to Laplace’s and Poisson’s equations. The gravitational waves transport energy and momentum. The solutions of (5) can be obtained using techniques developed for the soliton problem. It should be noted that (2) has wave solutions. and the decrease of the orbital period of a binary star system due to the emission of gravitational waves has been observed. Einstein. a particle in a gravitational field moves along a timelike tgeodesic in the spacetime manifold. When a = 0. Infeld. The latter thyperbolic equations may also be written in Hamiltonian form. (6) withp2=r2+a2cos20andA=r2-2mr+ a*. the space-time structure of black holes. Newton’s theory of gravity is valid in the limit $7/c* << I. In the interior equation (3). the generation of gravitational waves. which have no counterpart in Newton’s gravitational theory. two Killing vectors.e. A typical example of such a problem is the dynamics of a spatially homogeneous 3- -(1+(p/cZ)(dx2+dy2+dz2). One example of stationary and axially symmetric solutions is the Kerr metric. the dynamical process of gravitational collapse. This metric solution represents a gravitational field around a spinning mass with mass M = mc2/G and angular momentum J= Mac. singularity) is not assumed a priori as in (4). Experiments to detect gravitational waves generated in the universe have been planned. Applying a Backlund transformation to the Kerr metric. the global structure and dynamics of the universe. the path of light is represented by a null geodesic. (2) reduces to the Ernst equation: (&+&*)V*&=2V&. in general. From this point of view. Time delay and the advance of the perihelion have been observed in a binary system of neutron stars. 1938). the matter producing a gravitational field is represented by a tensor Tj of class Co. The concept of gravitational waves suggests the existence of a quantum of the gravitational field (graviton).VE. the deflection of light or radio waves passing near the sun. in which case equation (2) reduces to an telliptic partial differential equation on a 3-dimensional manifold. A comparison of the theoretical predictions and the observations is generally favorable. however. which is written as ds’=c*dt*-F(asin’edrp-cdt)* 2 -P2 -(r2+u2)sin2Bd~2. All the observational data are compatible with the theoretical results. and B. This fact implies that gravitational effects propagate with the velocity of light. and the gravitational mass is decreased by the emission of the waves. If the space-time is axially symmetric as well as stationary. but the phenomena in the universe are so complex that the effects of general relativity cannot always be isolated. All these solutions belong to the space-time metric with. In other words.. the time delay of radar echo signals passing near the sun. the equation of motion of a material particle (i. but is derived as a result of (2) (A. and so on. the detection of the graviton is far from feasible. equation (2) or (3) is divided into constraint equations in terms of the initial data and evolution equations in terms of the dynamical variables. an infinite series of stationary and axially symmetric solutions can be derived. Similarly. But there is also a way of representing it by singularities of a solution of the exterior equation (2). and the advance of the perihelion of Mercury. whose equation is formally obtained from (4) by replacing the right-hand side of the second equation by zero. The stationary metric is characterized by a timelike Killing vector.359 E Relativity arc length s along the particle trajectory.

1972. References [l] A. this has been studied as a cosmological model. Global Structure of Space-Time Following the advances of modern differential geometry. These are called Robertson-Walker metrics and are considered to describe a realistic expanding universe. W. Ellis. The metric functions obey the Einstein field equation (2) or (3). Space. The space-time with a constant scalar curvature R is called de Sitter space.) [3] A. [5] L. The event horizon is the boundary (the set of points) in space-time from which one can escape to infinity. The singularity whose existence is implied by this theorem means that the space-time manifold is geodesitally incomplete (the space-time is complete if every geodesic can be extended to arbitrary values of its affine parameter). Press. For some Cauchy problems relevant to cosmology and gravitational collapse. using the Kruskal coordinates. K. The mathematical model of space-time is a connected 4-dimensional THausdorff C”-manifold endowed with a metric of signature (1.3). (Original in German. In the case of spherically symmetric collapse. 1973. and that the energy-momentum tensor satisfies the condition dimensional manifold. an introduction to current research. Theories relativistes de la gravitation et de l’electromagnetisme. Thorne.n/2 < q < p < n/2. the metric takes the form ds* =c2dt2-a(t)*{dx2 +f(x)* x (de* +sin* Od#)}. To study the global structure at infinity. S. Singularities in space-time are one of the major problems concerning the global structure of the manifold. In the black-hole structure of spacetime. 1973. that a Cauchy surface exists. this assumption is verified and the final metric is given by the Schwarzschild metric. tanh t/4m = UJV or v/u. Relativity A sufficient condition for occurrence of a singularity is that there be some point p such that all the null geodesics starting from p converge to p again. for the proof of the singularity theorem it is presumed that the space-time is free of closed nonspacelike curves. Hawking and G.1335 359 Ref. or sinh x. [2] H. R. Freeman. q coordinate values in . Lichnerowicz. tanp=t+r. and J. 1956. time. Gordon & Breach. Press. S. Misner. [7] S. and reduces to the Minkowski space if R = 0. By means of this mapping. 1950. matter. are assigned finite p. 1955. Dewitt and B.). 1923. The causal structure of space-time is also related to the global structure of the manifold. Dewitt (eds. where f(x) = sin x. where andfi=secpsecq. 1962. The large scale structure of space-time. F. Witten (ed. W. F. “2. The meaning of relativity. In addition to this condition. Wheeler. including infinity. For example. a tconformal mapping of the metric is used. black holes have been introduced as the final state of gravitational collapse. If we assume that singularities do not exist in the exterior of the event horizon. [6] C. If the 3-dimensional space is isotropic as well as homogeneous. Black holes. Dover. the Minkowski metric is written in the form ds* = O* &*. Wiley. Masson. there exists a closed surface called an event horizon in an asymptotically flat space. the inevitable occurrence of a singularity has been proved (singularity theorem). In this regard. which are related to the coordinates of (6) by ( > _ k-1 e r/*m = u2 for any timelike vector Vi. Einstein. Weyl. In order to clarify the global structure of the solutions of Einstein’s equations. a stationary black-hole structure is uniquely described by the Kerr metric [6]. The maximal extension of the Schwarzschild metric is given as 32m3 ds* = --e-*i*“‘(dv* r -du*) .). [4] C. Cambridge Univ. Gravitation. However. fifth edition. Princeton Univ. Gravitation. it is not known whether this assumption is true in more general gravitational collapse. A. or the boundary of the set of points that one can see from the infinite future. manifestly coordinate-independent techniques to analyze space-time properties have been applied to general relativity. The metric allows the physical description of local causality and of local conservation of energy and momentum. tanq=t-r. . maximally analytic extension of the solutions has been studied. x.r*(dB* + sin’ Odq2). Then the black hole is a region from which no signal can escape to the exterior of the event horizon. all points.

The solution was due to Tartaglia.8) Renaissance Mathematics Toward the middle of the 13th century. However. W.). This constitutes a famous episode in the history of mathematics. Cardano (1501~ 1576). Vorlesungen der Mathematik II. In 1543. [ 101 A.1603). G. the English philosopher Roger Bacon (1214-1294) attacked Aquinian philosophy in his Opus majus. Padua. and some problems of quadrature.). N. then physical quantities. iiber Geschichte second edition. insisted on the importance of experimental methods in science. the forms of the numerals became fixed. and Florence. and L. Held (ed. Tomonaga. Hawking and W. the German priest Nicolaus Cusanus (1401.. A system of numeration was imported from Arabia to Europe in the 13th century. Press. Teubner. life (trans. M. [3] G. Dover. the indomitable proponent of this theory. Copernicus studied at the Universities of Bologna. Cardano. 1663. 1980. for example. Algebra was subsequently systematized by the French mathematician F. Introduction Group The concept of renormalization was introduced by S. and the importation of Arabian science into Europe during the Renaissance that prepared for this development. by the time of S. left manuscripts in which he wrote about mechanics. Cambridge Univ. Bourbaki. S. especially in Italy. in the domains of the arts and literature. the all-encompassing genius born in the same century. Opera. Da Vinci’s contemporary. In the 15th century. and Ferrara. scholastic theology and philosophy were at their height with the Summu theologiae of Thomas Aquinas (1225‘?-1274). the German painter A. Israel (eds. 360 (Xx1. Renaissance Mathematics 1336 [S] M. and F. The Renaissance flourished first in Italy. can be obtained as finite quantities by letting . Low in order to overcome the difficulty of divergence in field theory. the convergence of infinite series. Lyon. since the Greeks were able to solve equations only of degrees 1 and 2. the translation of the Greek texts of Euclid and Archimedes into European languages. Stoner). and more advanced mathematics began to be studied in European universities. and with the development and acceptance of printing. Copernicus (147331543) published his heliocentric theory (1543). Gell-Mann. N. influenced by Arabian mathematics. Schwinger. Cardan (G. and perspective. Stevin (1548?? 1620?) it took the definite form of a decimal system. By the end of the 15th century. Carmeli. to whom acknowledgment was made. was also born in the 16th century. Cardano). Its content. Viete (1540. Ferrari (1522-I 565). The book of my 1963.360 Ref. Padua. Elements mathematiques. one of the first printed books on mathematics. the mass WI of an electron. Cardano published the solution of equations of the third degree in his book Ars magna (1545). [2] J. L. Pacioli (1445?% 15 14) published Summa de arithmetica. Plenum. although publication of the method was against his will. an Einstein centenary survey. General relativity and gravitation I. References [l] M. During the same period. but in the latter half of the century. 1977. The best known result of l6th-century mathematics is the solution of algebraic equations of degrees 3 and 4 by the Italian mathematicians Scipione del Ferro. it was the invention of printing in the 15th century. 1979. E. Galileo studied at the University of Pisa and taught at the Universities of Pisa. the German scholar Regiomontanus (143661476) wrote the first systematic treatise on trigonometry independent of astronomy. J.1557) G. wrote a textbook on perspective. then in other European countries in the 15th and 16th centuries. 361 (Xx. but what is historically more important is the fact that essential progress beyond Greek mathematics was made by mathematicians of this period. In 1494. includes practical arithmetic and doubleentry bookkeeping. If the upper bound of the momentum is limited to a finite cutoff value A. Newer ideas in mathematics and the natural sciences dominated the 17th century. 10 [4] N. Group theory and general relativity. Cantor.32) Renormalization A. J. Hermann. and strongly urged the study of mathematics. 1900. McGraw-Hill. Leonardo da Vinci (145221519). Tartaglia (1506. practical mathematics (influenced by the Arabians) had become popular in Europe.1464) discussed infinity. d’histoire des 1960. The book enjoyed wide popularity.1642). vols. Galilei (called Galileo) (1564. geometric optics. Diner (1471I1528). [9] S. General relativity. II.

= m. g(p)) defined by Here g(n) is the solution of the equation ~.)G~2N’~ 0 The irreducible m&+B(g)$-ZNy. The pdependence of these functions can be determined by means of the following renormalization conditions: rA2N)=iArA2N). A second kind expresses the response of the renormalized Green’s function to the change in the mass and coupling constant. m/A. This cancellation is usually performed in each order of the perturbation expansion. satisfies the renormalization ( P$+P$-2NY > d’2N’({Pi}. the foregoing solution shows that high-energy phenomena can be described by the lowenergy phenomena whose coupling constant is given by g. > Since the lected in equation sequently inhomogeneous term can be negthe high-energy region..S(“)p-‘(g’)dg’ =log1. Since dimensional analysis yields Th2N)(lpi. and conits solution is =exp So. = Z. = g( co). can be renormalized by the transformation ‘pO= Zj’2q. in terms of a parameter p. and consequently the quantity dCzN)({pi}. In particular. and rni = m2 + am’. B. the Green’s functions are indefinite because they depend on A. Renormalization Statistical Physics Group Theory in The renormalization group technique has proved to be powerful in statistical physics.(g) Green’s function rihZN) satisfies for example. and is expressed by the Callan-Symanzik equation [4. and this transformation is a certain (semi) group. the relevant Hamiltonian is said to be renormalizable. and where Z is determined by Zm.(g) GfN)=AGgN).. All the divergences are taken into the renormalization constants Z. The inhomogeneous term is defined by Cl-31 A typical method t6 resolve the ultraviolet divergence is to add subtraction terms in the Lagrangian so that they cancel the divergence.. This is the first kind of renormalization group. g) = 14-2NTR (pi.. This dependence on the cutoff A corresponds to the response for the scale transformation of length../am. pipj+l -46. called a renormalization group. i. > where /?(g) = Zm. Otherwise it is called unrenormalizable. Renormalization Group in Field Theory where fl= p dg/dp and y = fp d log Z. g.5] m&+/?(g)$+2Ng. when gm = 0. the renormalized Green’s function and coupling constant g are also functions of p. p.. Z. alogZ.(p) and Z. so that the renormalized quantities cp.~(NlogZ.) 1 [ s x rA2N’(Pi. m. the foregoing becomes homogeneous. This circumstance is called asymptotic freedom.). Z. This is called the renormalization method using subtraction.PL)= l./4g)=0. and m are finite. These renormalization constants can be calculated by means of a perturbation method. -2N y(g’M-W&’ 9 r(Pi.. If the coefficients jl and y are calculated perturbationally up to a certain order. Several kinds of renormalization group have been used in field theory. g). and y. s(4).am/am. The Lagrangian density a AGjfN)=Zmo-G~2N) am0 +Zm. This indeterminacy is usually expressed as Z. called the renormalization point. ag/am.e. Since the cutoff A is arbitrary insofar as it is finite.*g.(g) = +Zrn. Since the renormalization constants depend on the continuous parameter p. the renormalized Green’s function is obtained up to the same order by solving the foregoing renormalization equation. . as well as in the statistical mechanics of phase transition.( p)/dp. When the addition of a finite number of subtraction terms cancels the divergence. and 6m2. g. C.(p). but the requirement of circumventing the divergences alone is not sufficient to determine them explicitly. m.1337 361 C Renormalization Group equation A go to infinity after summing all divergent terms. high-energy phenomena are described by the asymptotically free field.

Petermann. This transformation R.=... -TT.. h’~h. Comm. References [1] E. = RbXO. I. The fundamental idea is to eliminate some degrees of freedom. Introduction to the theory of quantized fields. The fixed point K* is given by the solution of K* =fb(K*). Jr..% =X0* + CjhjQj. Wilson./A.. 1541-1547. 499-520. [4] C. Renormalization group and the Kadanoff scaling picture. On the microscopic foundation of scaling laws. Helv.)-” near the critical point T.. The common fundamental structure of these renormalization group techniques is explained as follows.-d. There are many different ways of carrying out this idea explicitly..8* is the solution of G[X*] =O. La normalisation des constantes dans la thkorie des quanta. such as the cp4-model.. G... I8 (1970). and critical exponents are usually calculated in power series of E-d. DiCastro. R. Z... If the operator K thus defined has a negative eigenvalue ii. Stueckelberg and A. K=<-‘.. and the physical quantity Q. . 23 (1971).. The transformation of X is expressed as dX/dl= G [X]. G. Rev. Renormalization is useful in evaluating these critical exponents systematically. D2 (1970). Interscience. Phys.. [6] C. 19. The renormalization group method can be applied to other many-body problems. are renormalized.. Phys. Similarly. By the normalization we obtain the scaling law we have h”lh. Broken scale invariance in scalar field theory. Phys. The original Hamiltonian X0 is transformed into Z1 by means of this elimination process and by some scale transformation that preserves the phase space volume. 69-74.. Comm. by introducing a field hj conjugate to the relevant operator Qj. R. = 1. Bogolyubov and D. h.. Similarly we have X1 = R.. we have (i) momentum-space renormalization group theories [6&10] and (ii) real-space renormalization group theories [IO. and then to evaluate critical exponents from their asymptotic behavior near the fixed point. V.. we consider a Hamiltonian of the form 2 =X* + wQ and expand G[X] as G[X* + wQ] = wKQ + 0(w2).. Phys.. Symanzik. Renormalization group and critical phenomena. +O(R.. is the case where a single interaction parameter K is transformed into a new parameter K’ by K’=f. A=(df. The first few terms are calculated explicitly for specific models. small-momentum parts inside each cell are integrated or eliminated. Roughly classifying these into two groups.. as ~=~o+~. particularly in studies of phase transitions and critical phenomena [6&10].. A generator G is defined by G = lim. ). Low. Nuovo Cimento. one can estimate critical exponents [IO. Acta. the corresponding physical quantity Qi becomes irrelevant after repeating the renormalization procedure.. Phys. hj/t’+ . Small distance-behaviour analysis and Wilson expansions.... such as the Kondo effect [ 121. where v is called the critical exponent of 5. 95 (1954). Math. Phys. Jona-Lasinio.. This renormalization operation is written as R... Lett. 26 (1953). In most cases.l)/(b .. That is. denotes the critical dimension. By applying the Bore1 sum method to these &-expansions.e.. [3] N.. Thus. 322-323. The multiplicative renormalization group and the critical behavior in d = 4 -E dimensions. 49-86. where d.: i. C.. Cell-Mann and F.=R. Rev.. . namely...P.. .. . 5 (1972). long-wave parts.1). 111.~~~-.. and consequently the remaining slowly fluctuating parts.(K). = R.. from the eigenvalues of K..= t.& /. The simplest example of R. The correlation length < diverges like t(T)-(TT.R.=R$F. [S] K. Quantum electrodynamics at small distances. The fixed point A‘*=R. . Small distance behaviour in field theory and power counting.. 227-246. [7] K. e’ = h. where d is the dimensionality of the system and q is the exponent describing the deviation of the singular behavior of C(R) from classical theory... The correlation exponent v defined by 5 -(K-K*)-’ is given by the Wilson formula v = log b/log A. Rev. .. corresponding to a positive eigenvalue lLj > 0 becomes relevant.. N. First the momentum space or real space is divided into cells and rapidly fluctuating parts.. Math. 29A (1969). Shirkov. . G. namely.) is found to have the scaling property . the correlation function C(R) for the distance R behaves like C(R)-Rm(dm2+q)~exp(-~R). 1300-1312. Phys./dK).) The critical exponent of the specific heat defined by C-t P is given by the formula c(= 2 -d/l.h2.. we study the Hamiltonian . This is called the c-expansion. )-td’il.361 Ref. is constructed perturbationally. Callan.. to find recursion formulas for interaction parameters... hj .. E.. 1 I]. )~b-~f(b”lh .f‘(t.fV 1. = exp(lG)..f(l . h. has the (semi) group property R. C. Other scaling exponents { qj} can be obtained via the formula ‘pj = ?.. The free energy per unit volume f[X] =f(h... [Z] M. DiCastro and G. In order to find critical exponents from the asymptotic behavior of G near Z*.59. Renormalization 1338 Group By taking Q1 as the energy operator. s’?~=R..~+~2~2+. .. Lett.

x’) and for any x. Phase transitions and critical phenomena VI. 249 Lie Groups. 42 (1975). General Remarks For a mathematical system A. . If a G-set M has no proper G-subset (i.. Fisher. 42 (1975). Kadanoff. In this case. analytic. S. For UE G. ye M there exists an element a E G satisfying ax = y. Recursion equations in gauge field theories. Phys. P. Then we have a condition (ab)x = a(bx).1 6) Representations A. which is called the left (right) regular representation of G. Ann. 3174-3183. Rep. 3184-3205. G is said to operate on M from the right. x E M).e. 100 (1976).. Lett.. the group of all tpermutations of a set M (. 362 (IV. Giving a reciprocal permutation representation of G in M is equivalent to giving the structure of a right G-set to M.. Critical exponents in 3. lx = x (a. G. 359394. R satisfies the condition that a E G. 1976. B.) [ 123 K. A (reciprocal) permutation representation is said to be faithful if it is injective. called the quotient G-set of M by R. x E N implies ax EN. Sov. Wilson and M.e. 240-243. G. 248 Lie Algebras. Gsurjection. X’E M B4 (1971). be G. We denote by 6. Phys.13 Algebraic Groups. [ 131 L. the corresponding G-set is also said to be faithful. [S] K. We call a left G-set simply a G-set. then. Notes on Migdal’s recursion formulas.60 Classical Groups. if the product axcM of aEG and XCM is defined and satisfies this condition. 47 (1975).e. Wilson.99 dimensions. JETP. x E M) is satisfied. 423 Topological Groups. If an equivalence relation R in a G-set M is compatible with the operation of G (i. If a G-set M has no nontrivial quotient Gset. which is called a G-subset of M. B4 (1971). A mapping f: M+ M’ of G-sets is called a G-mapping (G-map) if the condition f(ax) = af(x) (a~ G. y) implies R(ax. 422 Topological Abelian Groups. A reciprocal permutation representation of G in M is an tantihomomorphism G-+G. x’) if and only if x = x’. For representations of other algebraic systems . In this article. A permutation representation of a group G in M is a homomorphism G-+Gw. 1975. 413-418. [lo] C. if the only equivalence relations compatible with the operation are R(x. Rev. G-injection. ay)). Then we have a faithful permutation representation (reciprocal permutation representation). which becomes a homomorphism if we define the multiplication in G. one different from M itself and the empty subset). The renormalization group: Critical phenomena and the Kondo problem. then G is said to operate on M from the left. the operation of G on M is said to be transitive.. Phys. and M is called a left then the operation of G on M and the corresponding permutation representation are said to be primitive. Permutation Representations of Groups R(x. Migdal. the induced permutation ac:x-+ax (xa) is called the left (right) translation by a. Mod. then for any two elements x. and G-bijection are defined naturally. and the corresponding permutation representation is also said to be transitive. 1 is the identity element. A. we can take G itself as M and deline the left (right) operation by the multiplication from the left (right). i. Giving a permutation representation of G in M is equivalent to giving the structure of a left G-set to M. Academic Press. 773-840.. (Original in Russian. [9] K. R(x. 437 Unitary Representations. In general.. JETP. Two permutation representations are said to be similar if there exists a G-bijection between the corresponding G-sets. Phys. 61 Clifford Algebras. The inverse mapping of a G-bijection is also a G-bijection. Phase transitions in gauge and sublattice systems. 69 Compact Groups. Phys.42 Boolean Algebras. Domb and M. Kogut. For topological. Sov. we consider the representations of tgroups and tassociative algebras. For specific groups . by the right notation x(fg) = (xf)g. Rev. In particular. 75-200. Wilson and J.. as before. Phase-space cell analysis of critical behavior. [ 1 l] A. and algebraic groups . II. G. If the product xa E M of a E G and x E M is defined and satisties the conditions x(ab) = (xa)b and xl =x.190 Groups B). 28 (1972). then the quotient set M/R forms a G-set in the natural way. 231 Jordan Algebras.). then N forms a Gset. Green (eds. The renormalization group and the E expansion. Rev. If a subset N of a G-set M satisfies the condition that a E G. and M is called a right G-set. E.1339 362 B Representations G-set.. Phys. Phys. We denote by aM the permutation of M corresponding to a~ G and write Q(X) = ax (x E M). a mapping from A to a similar (but in general “more concrete”) system preserving the structure of A is called a representation of A. 12C (1974). 743-746.

For instance. for any left A-module M we can define a linear representation p of A in M (with M viewed as a K-module via K + A) by putting p(a)x = ax. Every G-set M is the direct sum of a family {M. M is said to be k-ply transitive. In particular. and fix any element XE M.e. a permutation representation of a group G in M is said to be of degree n. this case has been studied in detail (. for any subgroup H of G.. the corresponding representation of G in M assigns the identity mapping I. Then an A-homomorphism M-M’ is precisely a Khomomorphism f: M -+ M’ satisfying the condition fop(a)=p’(a)of(a~A).362 C Representations Let M be a transitive G-set. and conversely. This Amodule is called the representation module of p. the canonical injection G+ 6. Furthermore. they are called the direct product of G-sets and the direct sum (i. which is called the zero representation. Let c?~(M) be the tassociative algebra over K consisting of all Kendomorphisms of M. This representation p is called the linear representation associated with M. For example.+ M.. Conversely. Here an equivalence class of R is precisely a left coset of the stabilizer (stability group or isotropy group) H. and let GL(M) be the group of all tinvertible elements in G. If p is injective. For convenience. I. the direct product G-set Mk = M x . Since K is commutative. where we assume M # (0). .Sections G. M is said to be simply transitive. if M = K. and each M. become G-sets in the natural way.. is a faithful permutation representation. x E M). and the tdirect sum CIE.. If M is tsimple as an A-module. the case where K is an tintegral domain and M is a tfree module over K of finite rank is also important. A reciprocal linear representation is an antihomomorphism A-P&~(M). in this case we say that p and p’ are similar (isomorphic or equivalent) and write p g p’. If M has n elements. Let p. For a G-set M. to every e E G. the restriction of a linear representation of K[G] in M to G is a linear representation of G. We now consider the linear representation of associative algebras. and similarly for reciprocal linear representations. When G is a group of permutations of M. the representation module of p coincides with the original one.(M). A linear representation of a group G in M is a group homomorphism G+GL(M). A reciprocal linear representation of A corresponds to a right Amodule. x M (k times) contains a G-subset M’k)={(xl.. the structure of a K-module in M obtained by the canonical homomorphism K + A coincides with the original one. M’.151 Finite Groups G). we introduce the structure of a left A-module into M by defining ax=p(a)x (as A. Conversely. G/H is a transitive left G-set. 1340 be extended uniquely to a linear representation of the tgroup ring K [G] in M.. p’ be linear representations of A in K-modules M. . disjoint union) of G-sets. Linear Representations Associative Algebras of Groups and Let K be a tcommutative ring with unity element and M be a K-module. This can . p and the corresponding M are said to be faithful.} of transitive G-subsets. M. Thus the study of (reciprocal) linear representations of a group G in M can be reduced to the study of (reciprocal) linear representations of the group ring K [G] in M. the Cartesian product n. respectively. this is sometimes called a homomorphism from p to p’. p is said to be irreducible (or simple). the linear representation associated with the left A-module A is faithful. Hence we have a G-bijection G/H. We always assume that A has a unity element and the homomorphisms are unitary.. this is called the (left) regular representation of A. this representation is called the unit representation of G (over K). if the operation of a group G on M is trivial: crx =x (0~ G. Thus the study of (reciprocal) linear representations of A is equivalent to the study of left (right) A-modules. M.1 Given a commutative ring K with unity and a linear representation p of a K-algebra A in a K-module M. respectively.IfM(k) is transitive. A linear representation of the algebra A in M is an algebra homomorphism A+c?~(M).. an A-isomorphism is a Kisomorphism f: M + M’ satisfying the condition fop(a)o =$(a) (uEA).xk)Ii#jimpliesxi#xj}. For a family {Mnjnci\ of G-sets. If M is transitive and the stabilizer of each point of M consists of the identity element alone. is called an orbit (or system of transitivity). If we view G as a G-set.xEM). we can write Ix=xi (~EK.” (Note that a group ring has a canonical basis-the group itself-and allows a more detailed investigation. which we call simply “algebras. we can also consider a linear representation in the trivial space M = {O}.XE M). A transitive G-set is called a homogeneous space of G. Let M be the representation module of a linear representation p of an algebra A. Let A be an associative algebra over K. Though we shall mainly treat the case where K is a field and M is a finite-dimensional tlinear space over K. the mapping f: G+ M defined by f(a) = ax is a Gsurjection and induces a G-bijection f: G/R + M. = {u E G 1ax = x}. A homomorphism from an irreducible representation p to p must be an isomorphism or the zero C.

relative to this basis is given by qi./M..‘. A linear representation is said to be reducible if it is not irreducible. T. e. 6) (go G). . of linear representations pl. M'.) over K:(&)(tj) =(&r nijtj). if we put n. If p is never similar to the direct sum of two nonzero linear representations. In this case. .i+. . e. . then by (e i.+. .c. Thus a linear representation of A in K". n’. and the corresponding representations are similar. T. Coefficients Representations and Characters of Linear Hence giving the finite-dimensional linear representations over a field K is equivalent to giving the matrix representations over K. . The number p > 0 of pi similar to p’ is called the multiplicity of p’ as an irreducible component of p. + . we define the linear representation P 0 P’ in M 0 ~4’ by (P 0 p’M=p(d 0 p’kd u-+T(u)= 0 T. we introduce the structure of a left (right) A-module using the inner product ( . The residue classes of e. . We always assume that K is a field. A right (left) A-module M is regarded as a linear space over K..+.. Matrix Representations Let K” be the K-module consisting of all ntuples (ci) of elements in a commutative ring K. then p is said to be indecomposable. . .) of M over K such that (er.i) forms a basis of Mi over K (1 < i < r). p. . D. of the representation modules M. are uniquely determined by p up to their order and similarity (JordanHSlder theorem). .. over K. this means that M is tindecomposable as an A-module.(a) Z-n(a) . ) as follows: (x. .. Then the matrix representation corresponding to p relative to the basis (er . and called the direct sum of representations. . . Ti(a) is an ni x nj matrix and 7$1) = 0 for i > j.-. is called a matrix representation of A over K. + M. .cM.368 Rings G). K) is the group of all n x n invertible matrices. A matrix representation of a group G over K of degree n is a homomorphism G+GL(n. p is said to be completely reducible (or semisimple). E.. e.) is a +basis of a K-module the K-isomorphism K"+ M given by the assignment (&)-C~=i ei&. .> e. We take a basis (ei...-. . If A is a semisimple ring. . . A linear representation p over K of a K-algebra A is said to be of degree n if its representation module M is of dimension n over K. the linear representation p corresponding to a matrix representation a-*(Aij(u)) is given by P(akj= t i=l eilij(a). .(K). p’ of a group G in M.. a homomorphism A-t M. this is called the tensor product of representations p and p’. . . . i. where GL(n. (~1 T. . The sequence { Mi} is a tcomposition series if and only if each pi (hence Ti) is irreducible. In particular. T’ be matrix representations of degree n. . Explicitly. = m. where.(K) of all n x n matrices (1.. M.e. Suppose thatasequence{O}=M. e mi form a basis of the quotient space Mi/Mi-. In its dual space M*.(a) 0 1 UCA. then such a homomorphism is a scalar multiplication. For linear representations p. Then a homomorphism from T to T’ is an n’ x n matrix P satisfying PT(a)= T'(a)P (as A).cM./M. . We consider the linear representations of an algebra over a field K. In this case. x*a) . = dim M. If M is tsemisimple as an Amodule. x*) ((x. . Then a K-module is a linear space over K. The representation p is completely reducible if and only if it is similar to the direct sum of its irreducible components (admitting repetition). we have a bijective correspondence between the matrix representations of A of degree n and the linear representations of A in M. + p. .K).=M of A-submodules of M is given. The linear representations associated with a submodule and a quotient module of M as an A-module are called a subrepresentation and a quotient representation. if K is an talgebraically closed field and M is finite-dimensional. pr is written p1 + . .1341 3623 Representations Therefore T and T' are similar if and only if n = n’ and PT(a)P-' = T'(a) (a~ A) for some n x n invertible matrix P. If M. ax*) = (xa. The linear representation associated with the direct sum M. any linear representation of A is completely reducible. .m.. it suffices that this equation is satisfied by all a E G.) has the form r T. We also say that p contains p’ p times or p’ appears p times as an irreducible component of p.(a) .. .. An irreducible representation p’ similar to some pi is called an irreducible component of p. p is similar to the matrix representation I T. pl. Let T. For a representation of a group G. cY~(K")is identified with the K-algebra M. The converse also holds (. and the matrix representation corresponding to the linear representation pi associated with M. .(a) homomorphism (Schur’s lemma). and n is called its degree. respectively. .

* (XE M. F. The determinant of the reduced representation is called the reduced norm of A. .(u) = tr T(u) (a~ A). For a fixed X*E M* the assignment x+ pX. . If the matrix representation u-(&(u)) corresponds to M relative to this basis..X* gives an A-homomorphism M + A*. Scalar Extension of Linear Representations Let K. and fix a homomorphism 0: K +I?.)bea basis of M over K.. for a semisimple algebra A.~~L).Section F) is called an absolutely irreducible character.iEL). a coefficient of a linear representation p of a group G can be regarded as a function on G taking values in K. we can regard (K[G])“= L[G]:g@i~=gl(gEG. For any finite-dimensional semisimple algebra and group ring of a finite group. and let a+ T(u)=(>. XCM. If K is algebraically closed and p is irreducible (or more generally. any irreducible representation of A is similar to some subrepresentation of the coregular representation of A. Then i. x* and is determined by its values at generators of A as a linear space. . The reciprocal linear representation associated with the right A-module A is called the right regular representation of A.. then p g p’ is equivalent and the different irreducible characto &=x&7’. Let (el. and its character is equal to the reduced character. the mapping g+‘p(g)-’ (ysG) is called the contragredient representation of p. The linear mapping ‘p(a) is the ttransposed mapping of p(a). x* E M*) for a given linear representation p. where A* is considered as a left A-module. For the linear representation p associated with M.~~(u)) be the matrix representa- 1342 tion that corresponds to p with respect to this basis. e.. . each A. where UEA.~: (1 < i. .*) is the dual basis.. x* E M*. Then xP is a function on A that is uniquely determined by p and belongs to AZ. The character of p is equal to the sum of the characters of the irreducible components of p taken with their multiplicities. we define a ilinear form p. Frobenius and I. then M” has the natural structure of a left Au-module.IfMisaleft A-module. x E M.e.. . . are irreducible representations of A such that pi and pj are not similar unless i = j. In particular. If pl.~~E)(277 Modules L). tabsolutely irreducible).. We take a matrix representation T corresponding to p and put x. j < n). Then any irreducible representation is similar to one and only one of pl... and is denoted by ‘p. where (eT. For an algebra A over K. Then p z p’ implies Af = A$.b~A. pV. in A* is direct.x*))..~. e.p (a E A.x*) (a~ A).) be a basis of the representation module M of p over K. xp is called the character of p. ters are linearly independent. and its transposed representation (i. any irreducible representation of a finitedimensional semisimple algebra or a finite group is an irreducible component of the regular representation. Let p be a linear representation of A over K and M be its representation module. AZ. For any x E M. If p is the representation associated with M..~(u. the statement holds irrespective of the characteristic of K. it can be viewed as a function on the set of all tconjugate classes of G.e. the representation associated with M* is called the transposed representation (dual representation or adjoint representation) of p. For a linear representation p of a group G. The character of an irreducible representation is called an irreducible character (or simple character).29 Associative Algebras H). We shall treat finite-dimensional representations exclusively.. If we consider absolutely irreducible characters only.. then {n. Let(e. Moreover. we have (M*)* = A4 as an A-module. therefore we have dim AZ = nz (G.. .X* of an irreducible representation p generates an Asubmodule of A* isomorphic to M. For a group G. The direct sum of all absolutely irreducible representations of A is called the reduced representation of A.* E A* on A by p&u) = (ax. . If M is finitedimensional over K. p. the sum AZ. Let AZ be the subspace of A* generated by all coefficients p.} is linearly independent.u E L). and A* can be decomposed into the direct sum of Ap*. The sum of all absolutely irreducible characters of A is called the reduced character (or reduced trace) of A. . J. In particular.. . the scalar extension A” of the K-module A has the natural structure of an algebra over L:(uOi)(bO~)=ahO3. = P?. x*EM*. (I< i < r) be the irreducible representations associated with the minimal left ideals of the +simple components Ai of A. the regular representation and the coregular representation are similar (. the representation associated with the left A-module A*) is called the coregular representation of A. . In particular. The character of an absolutely irreducible representation (. is canonically identified with A*.T.. If K is of characteristic 0. L be commutative rings with unity element.362 F Representations zz (ux. Schur). In other words. .~~K. We denote by M” the scalar extension a*(M) = M OK L of a K-module M relative to rr:xiO/*=xOi”~(x~M. the linear representation p” over L associated with M” is called the scalar extension of p relative to a:p”(u@l)=p(u)@l. let the p. + + A. (a @ 1”)(x @ p) = ax @ 1. In addition. This is called the coefficient of p relative to x. For a linear representation p of a group G. Therefore any nonzero coefficient pX. the character of p can be regarded as a function on G. then the matrix representation corre- .

In particular. Then for a p of A and its represenlinear representation tation module M. . Let K be a field. an equivalent condition is that the scalar extension p’ to the talgebraic closure I? is irreducible. . for example.{0}. p2 over K. if e: K -+L is an isomorphism. If g is divisible by the characteristic of K. Recent research has clarified considerably the structure of this group [19]. we have a modular representation (. Such a representation is called an ordinary representation. For simplicity. The linear representation of G over K is equivalent to the linear representation of the group ring K [Cl. U a. To such a representation corresponds a matrix representation T such that T(u) has exactly one nonzero entry in each row and column for every UE G... Consequently.‘uu. 1945). . each degree is a divisor of the order g of G.(a)“). the scalar extensions p”. The set S(K) of talgebra classes over K. M is called the induced representation and is denoted by pG. For an arbitrary field K. and any irreducible representation over any field extension of K is realizable in K. Each irreducible representation appears as an irreducible component of the regular representation with multiplicity equal to the degree. then K. .. 0 1) over L is given by a @ 1 -@. If m is an ideal of K and 0: K + K/m (tresidue class ring) is the canonical homomorphism. then using the partition of G into the cosets G = a 1H U . concerning which we have already stated the general facts.. Then the multiplicities of all irreducible components of pL are the same. Linear Representations of Finite Groups Let G be a finite group of order g. p” is called the conjugate representation of p relative to 0. We shall consider finitedimensional representations exclusively.) ’ 1 I where we define T(b) = 0 for b# H. we assume that K is tperfect and L = K. any scalar extension of an irreducible representation over K is irreducible. Let A be linitedimensional over K. ML. Let p be a linear representation of a subgroup H of G and M be its representation module. If p is a prime ideal of K and o: K + K.‘aul) . then the construction of p” from p is called the reduction modulo m (. H we can write the matrix representation corresponding to pG as a+ T(a. L a field extension. the scalar extension pL of an irreducible representation p to a tseparable algebraic extension L of K is completely reducible. For linear representations pl. We assume that K is a field. The induced representation from a representation of degree 1 of a subgroup is called a monomial representation. We fix a splitting field K for G and assume that K is of characteristic 0. e. . Brauer.sm. if the field K is a splitting field for the group ring K[G]. Then the linear representation of G associated with the K [Clmodule K[G]6&. respectively. If K is the ring Z of rational integers. If the characteristic of K is zero or more generally not a divisor of g. each of which is represented by a (central) simple component of the group algebra K [G] of G.‘uu. for such a field K. The exponent of G is the smallest positive integer it satisfying u” = 1 for every element UE G. (tlocal ring) is the canonical homomorphism.. T(a. Another equivalent condition is that every endomorphism of the representation module M of p must be a scalar multiplication.Section I). In addition. p1g pz is equivalent to pf g pg.) . this multiplicity is called the Scbur index of p. we can regard pL as an extension of the mapping p. We can also consider the “completion of representation” with respect to p. . then K is called a splitting field for G. and 0: K +L the canonical injection. The conjugate representation relative to the automorphism a:i-+X (complex conjugation) of the complex number field is called the complex conjugate representation. K is called a splitting field for A. we can assume K = C. In view of M c ML. The number of nonsimilar irreducible representations of G is equal to the number of conjugate classes in G. is a subgroup of the tBrauer group B(K) of K. M” are written pL.) . 8&V) c &‘L(ML) by the natural injections. T (a. If every irreducible representation of A over K is absolutely irreducible. A linear representation over L is said to be realizable in K if it is similar to the scalar extension p” of some linear representation p over K. then the construction of p” from p is called the localization relative to p.If the matrix representation T corresponds to p. If K is an integral domain and p = K . If K is a splitting field for A.1343 362 G Representations some finite group G. is the tfield of quotients of K. A c AL. . For a group G. . For the trivial sub- . sponding to M” relative to the basis (el @ 1. T(u.Section I). any irreducible representation of AL is realizable in K for any field extension L of K. known as the Scbur subgroup of B(K). An irreducible representation p over K is said to be absolutely irreducible if its scalar extension pL to any field extension L is irreducible. every linear representation of G over K is completely reducible (H. A field containing all the nth roots of unity is a splitting field for G (R.. Maschke). a linear representation over K is sometimes called an integral representation..

This implies that u yields an irreducible representation of 5. Z. The following ortbogonality relations hold for irreducible characters x and $ of G: 1344 In the second formula. in Z. i. If the degree of F. Take an algebraic number field R that is a splitting field of G. Hence there corresponds to T a fixed class of irreducible representations of 6. U.* of Fl in U. . is f. to obtain an irreducible representation of 5. Fig. Let k be the number of p-regular classes of G. any two different Young diagrams yield different irreducible characters. appears u. H. over a field of characteristic zero reduces to that over Q.60 Classical Groups). The number of nonsimilar indecomposable components of the regular representation R of G is also equal to k. and we denote these representations by U. .fJ. in which the situation is quite different from the case of ordinary representation. times in R and F. we have the case of modular representation. a fixed irreducible character of 6. x ranges over all the irreducible characters of G.. Z. We can assume that all the coefftcients of Zi are contained in o. As in Fig. an idempotent that is not the sum of two orthogonal nonzero idempotents) of A.. thus obtained may be reducible. that preserves each column and set t = C(sgn r)r. The element u of A is called the Young symmetrizer associated with T. n = 26. over the field Q of rational numbers are absolutely irreducible. and that of U. of F. . Then there exist exactly k nonsimilar absolutely irreducible modular representations F. The modular representations Z. The multiplicities c. Thus there exists a one-to-one correspondence between the Young diagrams and the irreducible characters of 6. In general.L). f’ =2. Moreover.. Z.e. ... is the number of elements in C(a). then u is a primitive idempotent of A except for a numerical factor. Linear Groups Representations of Symmetric I. is u. . the multiplicity of CJin pG coincides with that of p in the restriction a.. it is sufficient to find a tprimitive idempotent (i. . for example.362 H Representations group H = {e}.& = 6. The elements of G whose orders are prime to p are called p-regular.. we obtain another symmetrizer u’ associated with T. If we set u = t.. We put the numerals 1 to n in any order into the n squares of T= T(fi>fz. If p is a divisor of g. that preserves each row and construct an element of A : s = C cr.. fl = 8.=n. Fk. then it is written T= T(f.f~ = 6.. we obtain a modular representation Zi. Replacing every coefficient in Zi by its residue class mod p.. the left ends of which are arranged in a single column. C(a) denotes the conjugate class of G containing a. and let o be the domain of +pintegers of 52.. . we draw a diagram T consisting of n squares arranged in rows of decreasing lengths.e. if it has k rows of lengths fi 2f22 >&>O.>. ..e. we denote by r any permutation of 6. 1 Young diagram. . appears in UK as both its top and bottom component. and any irreducible character is obtained by a suitable Young diagram.fi +fi+ +f. However. The multiplicities di. we obtain the regular representation of G.. 1. conjugate classes of G containing the p-regular elements. Since the group algebra A = Q [S. Hence the representation theory of 6. .. U. Then the residue class field o/p is a finite field of characteristic p and a splitting field of G. Similarly. then UK appears f. of g to H (the Frobenius theorem). . as in Fig. Let p be a prime ideal in R dividing p. Such a diagram T is called a Young diagram. =4. (T. Let Z. are called the decomposition numbers of G. We can number them in such a way that F. i. are called the Cartan invariants of G. . The theory of modular representations of a finite group was developed mainly by Brauer after 1935. Let G be a finite group of order g. The method of determining the character associated with a given diagram was found by Schur and H. s = C f r. . Weyl (. and let K be a splitting field of G of characteristic p # 0. Modular Representations of Finite Groups All irreducible representations of a tsymmetric group 6. Hence we can assume that o/p = K. 1. Such an idempotent can be obtained in the following way. F2.] is semisimple. be the nonsimilar irreducible representations of G in 0. these two irreducible representations associated with LI and u’ are similar. If we put the numerals 1 to n into the n squares of Tin a different order. where K is the field considered at the beginning of this section. .f.. We then denote by 0 any permutation of G. times in R. .. and g. for an irreducible representation c of G and an irreducible representation p of a subgroup H..

. cp. obviously 0 <d < e. then g. (p. The number of blocks of defect e is equal to the number of p-regular classes of defect e. and F.. we say that Zi belongs to B. and if Q # p. . = 0 for any Zi in B.. Let cp?.. we have the following orthogonality relations for the modular characters: In the first sum. Then we may assume that R contains a primitive g’th root of unity 6(~ 0). Since xX=-) = w.. belonging to B.(s). Since (p. Let M be a modular representation of G. A block of The db are called the generalized decomposition numbers of G. and FL belong to the same block if there exists a sequence of indices K. where s. . B. a ranges over all p-regular elements of G. of blocks B.. the converse is also true. (x. Let s be conjugate to an element of D. B. 1 such that cKa#O. Then there exists a p-regular class of defect d containing an element a such that g.. . . = 1). This is an equivalence relation. y. If the irreducible components of Zi belong to B.). Obviously. If we denote the degree of Zi by zi. . then fi. We replace each p by 6’ and obtain an element ((a) of R as the sum of these 6’. (i.. The defect group D of C(a) is called the defect group of B. g’) = 1.. cpi&.~ # 0 if di.(a) of an element a of G. Then d is called the defect of the class C(a). If the order of s is p’.. > y. Let D be any p-Sylow subgroup of the tcentralizer C. and by I]. c.. then ~. If the p-factor of x is not conjugate to any element of the defect group D of B. = y. where a ranges over all ..#O.Xr) = &IX dhC(4 a we have XiW=C ri&r) =I 0 &d(r).xi(a)/zi = g. From these relations we get the following refinement of the orthogonality relations for group characters. Then x. and r is a p-regular element. We say that two elements of G belong to the same section if and only if their p-factors are conjugate in G. of blocks of Cc(s). If p” is the highest power of p that divides all the degrees zi of Zi belonging to E. . the residue class Z(E K) of 6 is a primitive g’th root of unity. .Xj(a)/zj (modp) for all p-regular elements aofG.s~i(a)~i(a-l)=O. . be the absolutely irreducible modular characters of Cc(s). . If F. say s.Xi(a)/zi + 0 (mod p) for any Zi in B.. Zi belongs to a block of defect 0 if and only if x. If Zi belongs to a block of defect d. An arbitrary element x of G can be written uniquely as a product x = sr = r’s. rc Cc(s)9 They are related to the Cartan the fundamental relations The determinant Ic. then the db are algebraic integers of the field of the p’th roots of unity.(D) with defect group D. . We call d = e-u the defect of B. We say that F. Let B.x. called the p-factor of x. &$ B. simpler proofs were given independently by K. Brauer’s original proof of this result was considerably complicated. and let xi be the absolutely irreducible ordinary characters of C. then the power of p dividing zi is pe-d+ki (hi 2 0).. It follows that all the degrees zi of Zi belonging to a block of defect 1 are exactly divisible by p’-‘. and y. Two modular representations have the same irreducible components if and only if their modular characters coincide. Let x. We call 5 the modular character (or Brauer character) of M.(a) = 0 for any element a of G whose order is divisible by p. # 0. is an element whose order is a power of p. hence also exactly one modular representation F.. All the irreducible components of Zi belong to the same block since c. .1345 3621 Representations invariants by defect 0 contains exactly one ordinary representation Zi. . then xi can be considered as the modular character of Zi. c(.. be a block of defect d. Fk are classified into a finite number. a union B. .. We set g = p’g’. then xi(x) = 0 for all Zi in B. g’) = 1. p.. . then it is also the highest power of p dividing all the degrees f.. belongs to a block B. Now Zi and Zj belong to the same block if and only if g.. If Zi and Zj belong to different blocks of G. Moreover... and let (D: l)=pd.~#O . # 0 and di.. c. of F. also belongs to B. and D is called a defect group of C(a). If xi is the ordinary character of Zi. There corresponds to B. each section is the union of conjugate classes of G.. and BP contain no irreducible modular representations in common. be the number of Zi belonging to B. and D is uniquely determined up to conjugacy in G. Denoting by cp.. The number of blocks of G with defect group D is equal to the number of blocks of the tnormalizer N. Iizuka and H. belonging to B. Nagao.e. we say by a stretch of language that the corresponding U. that of U. We have di”.(a)/z. Fz.J of degree k is a power of p. the number of F. for a~ G is an algebraic integer and hence belongs to o. This is obviously an equivalence relation. the modular character of F. . = U. we have Zi = F. The characteristic roots of M(a) for a p-regular element a are powers 2 of 8.. In this manner we define a complex-valued function r on the set of pregular elements of G.

Regarding projective Z [G]-modules . In particular. K*) off is determined by p and is independent of the choice of sections for p.[G]-modules L.) is called a subrepresentation of p. among others. Here K*=K-{0} and K*l. where z is the natural projection of GL( V) onto PGL( V). It provides an integral representation of G as an automorphism group of the R-projective module L. If p and p’ are similar. then C. A mapping (T: G+GL( V) is called a section for (p. then G is called a closed group. The isomorphism problem.. K*) is determined by the characteristic of K. The study of R.e. The +Krull-Schmidt theorem. V) is a projective representation of G such that each p(a) (a~ G) leaves P( V. g M. which can be identified with the quotient group GL(V)/K*l. A homomorphism G+PGL(V) is called a projective representation of G in V or simply a projective representation of G over K. Moreover.. where h. we get a projective representation (pr . The number of genera of G-lattices in V is given by &hp (9 = order G). If 9X(G) = 1. then the order of c. where xi ranges over all the characters of G belonging to a fixed block B. The mapping f is a +2-cocycle of G with values in K*.f: G x G+K* satisfying a(u)g(b) =. When V is absolutely irreducible. holds if R is a complete discrete valuation ring or if R is a discrete valuation ring and K is a splitting field of G. Takahashi).f(u..200 Homological Algebra G. Vi) by restricting the p(a) to P( V.=l. For any R [G]-module L there is an associated family of R. . . The set {f(u. xi(a)xi(h -‘) = 0. and let P(V) be the tprojective space associated with V (. then both the degree of p and the square of the order of cp are divisors of the order of G.. = R. and we have cpol.cIj. The condition for the finiteness of the number of nonisomorphic indecomposable Glattices is known. In this case (p. and any projective representation of G is induced by a K. Let V be a finite-dimensional linear space over a field K. Let p be a prime ideal of R and R.[G]-lattices in V. If an algebraic number field K is specified. Projective Representations of Finite Groups linear representation of G. A G-lattice L is characterized as an R [G]module. i. Glattices).. V) if n(o(u)) = p(u) for each UE G. R[G]-modules L and M need not be isomorphic even when the K [G]-modules K @L and K @ M are isomorphic..) invariant. If K is algebraically closed. C*) = cJ?l(G) is called the multiplier of G. where p ranges over all primes of R. Integral Representations Every complex matrix representation of G is equivalent to a matrix representation in the ring of algebraic integers. the number of R[G]equivalence classes in a genus equals the (ideal) +class number of K (J. Maranda and S. G-lattices L and M in a K [G]-module V are said to be of the same genus if L. cp = c..343 Projective Geometry). denotes the number of R.) c P(V).btc is called the factor set of p with respect to cr. where R is the ring of integers in K. for every p. if p is irreducible. K. the question of whether the isomorphism Z[G] gZ[H] of integral group algebras implies the isomor- . A projective representation is said to be irreducible if there exists no proper subrepresentation of p.. Any section (T defines a mapping . for any ceH’(G. then cp = cp. every K[G]-module V contains G-invariant R-ilattices (briefly. Two projective representations (p. Yamazaki. We can assume that P( V. V. When K is the complex field C. then H’(G. b)o(uh) (a. Let Vi #{O} be a cp0P@)0cp-’ subspace of V. asserting the uniqueness of a direct sum decomposition into indecomposable R[G]-modules. is a divisor of the degree of p (dimension of V). hi G).). IfGisafinite group.e.[G]-equivalence classes of R. If elements a and h of G belong to different sections.362 J Representations the elements belonging to a fixed section S of G. The 2cohomology class c. If (p. K*) there exists an irreducible projective representation p of G over K which belongs to c. b)}a. cp. The set of all projective transformations of P(V) forms the group PGL( V). if p is a projective representation of G over C. which is finitely generated and +torsion free (hence tprojective) as an R-module. A projective representation p has a section o which is a linear representation ofGin Vifandonlyifc. V’) of G are said to be similar if there exists an isomorphism q: PGL(V)+PGL(V’) induced by a suitable isomorphism V-r V’ such that =~‘(a) (uEG). V) and (p’.-representations is intimately related with modular representation theory. for R = Z it reduces to the requirement that the Sylow psubgroup of G be cyclic of order p or p2 for every p 1y.EH’(G. the group ff’(G. The set of Glattices in a fixed K [G]-module V is divided into a finite number of R[G]-isomorphism classes (Jordan-Zassenbaus theorem). In general.. 1346 J. i. 0 L. M. studied the projective representations of finite groups in detail. be the localization of R at p. is the set of all scalar multiples of the identity transformation 1v of V and is the center of CL(V). The tensor product p @ p’ of two projective representations p and p’ can be defined as in the case of linear representations.

Ind. In 1862 he contracted tuberculosis. [3] H. In his inaugural lecture in the same year. Riemann became interested in theoretical physics. Germany. has been answered affirmatively for certain special cases such as tmeta-Abelian groups. Math. Serre. Math. he defined the tRiemann integral and gave the conditions for convergence of trigonometric series. References [l] G. 1948). Brauer and C. Lecture notes in math. [2] B. B. F. In 1857 he rose to assistant professor.1866) was born the son of a minister in Breselenz. In his later years. J. [15] R. 139 (191 l). Springer. Actualites Sci. Yamada. 1972. Representations lineaires des groupes finis. Dedekind and H. Curtis and I. [l] B.. (2) 42 (1941).. 4 (1955). Georg Friedrich Bern hard Georg Friedrich Bernhard Riemann (September 17. Nesbitt. References 363 (Xx1. L. introduced n-dimensional manifolds. Brauer. [ 171 L. Notes of blocks of group characters. Dornhoff. Uber die Darstellung der endlithen Gruppen durch gebrochene lineare Substitutionen. 397. [S] C. R[G] is an R-torder in K[G]. L. second edition. Representation theory of finite groups and associative algebras. R. K-theory of finite groups and orders. Group representation theory. 142. Weber. Weber. Gruppen von linearen Transformationen. formulated the concept of tRiemannian manifolds. Weber. Springer. Elements de mathtmatique. Osima. Springer. Dekker. A. Lectures on modern mathematics.. and defined their curvature. tDirichlet as full professor. Springer. Reiner. II. Gesammelte mathematische Werke und wissenschaftlicher Nachlass. 1970. [7] R. Math. Bull. second edition. 63 (1956). B. Reine Angew. Zur Darstellungstheorie der Gruppen endlicher Ordnung I.1347 363 Ref. van der Waerden. 556-590.1826-July 20. Benjamin. 1974. a considerable portion of the integral representation theory has been extended to more general orders in separable algebras [ 14161. [ 191 T. HuberDyson). The Schur subgroup of the Brauer group. .-P. Brauer. ch. [16] K. J. Schur. Representations of finite groups. G. In his paper of 1857 on tAbelian functions. J.). 1599227. Amer. Algebra II. Sot. II. M. Lecture notes in math. [ 131 W. 1953). 1958. 175188. Reine Angew. his contributions encompassed all aspects of mathematics. he discussed the foundations of geometry. 149. Boerner. Wirtinger (eds. Ann. L. Saaty (ed.. [14] I. Wiley. 72 (1959-1960).40) Riemann. Lecture notes in math. 76 (1970). 1935 (Chelsea. Okayama Univ. Hermann. Algbbre. Schur. He attended the universities of Giittingen and Berlin. Math. he systematized the theory of tAbelian integrals and Abelian functions. Springer. 1971. 1970. 133-175. Feit. Hanover. fifth edition. Interscience.. 406-444. His doctoral thesis (1851) stated the basic theorem on tconformal mapping and became the foundation for the geometric theory of functions. van der Waerden. and in this context. 115. In 1851 he received his doctorate at the University of Gottingen and in 1854 became a lecturer there. A survey of integral representation theory. Math.. Math Z. Bourbaki. 8. It remains for modern mathematics to investigate whether this hypothesis is correct. 25-46. Darstellungen von Gruppen. [S] M. Georg Friedrich Bernhard phism G g H of groups. Teubner. Despite his short life. 1902) (Dover. In his paper presented for the position of lecturer (1854). In his paper of 1858 on the distribution of prime numbers. 85-137. [ 1 l] I. 1962. 132 (1907).). vol. 1261a. [lo] I. He gave lectures on the uses of partial differential equations in physics that were edited and published by H. [4] N. Riemann. Riemann. 155-250. Hermann. Math.).. Uber die Darstellung der symmetrischen und der alternierenden Gruppe durch gebrochene lineare Substitutionen. [6] R. T. 1972. Reiner. 1. Noether and W. 1967. he considered the tRiemann zeta function as a function of a complex variable and stated tliiemann’s hypothesis concerning the distribution of its zeros. W. [IS] J. [ 121 R. 1892 (Nachtrage. 1955. Characters of finite groups. L. 127 (1904) 20-50.. 1963. Untersuchungen tiber die Darstellung der endlichen Gruppen durch gebrochene lineare Substitutionen. and in 1859 succeeded P. [9] I. Springer. and he died at age 40. Reine Angew. W. Lattices over orders I. Schur. influenced by W. Roggenkamp (with V. 1967. On the modular characters of groups. (eds. J. Swan.

We assume that M is connected and of class C”.(p. 2.105 Differentiable Manifolds). it is called the fundamental tensor of M.80 Connections. the covariant differential operator V. a differential form of degree 1 is identified with a tangent vector field. y E M. 1926 (Chelsea. Let (w’).285 Non-Euclidean Geometry). 417 Tensor Calculus). 4~ N.. In particular. then ‘p*g is the Q’irst fundamental form of N. then (M. (For example. X. Then B is an O(n)-subbundle of F of class C”. A normal vector at a point p of a submanifold N of M is well defined as an element of the orthogonal complement of the subspace T. q).+.~~~ on B. In this way we get a one-to-one correspondence between the set of all O(n)-subbundles of F and the set of all Riemannian metrics of M. and we have wj+o{=O. The function d is a tdistance function on M. g). A curve x : [a. and M. Riemannian Connections [2] F. and the topology of M defined by d coincides with the original There exists a unique +afftne connection in the orthogonal frame bundle B whose ‘torsion tensor is zero.$(q)). co) is dctincd so that the value d(p. Using the value g.) A necessary and sufficient condition for a differentiable manifold M of class c’ to have a Riemannian metric is that M be tparacompact. a positive definite inner product g. There exists an essentially unique structure of a Riemannian manifold on (real or complex) telliptic or thyperbolic space (. g). (For a vector field X. = lldc&L)l\. This connection is called the Riemannian connection (or Levi-Civita connection. and y be a +Riemannian metric of class c’-’ on M. The tconnection form of the Riemannian connection is expressed by n2 differential l-forms (~j).~~ together with (Q~) give rise to an absolute parallelism on B (that is. A necessary and sufficient condition for a mapping II/ : N--t M to be an isometry is that d. where a. the length (/x([ is independent of the choice of parameter t. and q is called an isometry.3) Riemannian A. tJ are timmersions of class C”.~~.. then (N. a submanifold and a ‘covering manifold of M have Riemannian manifold structures induced by the natural mappings (.g) or simply M is called a Riemannian manifold (or Riemannian space) of class c* (. Vorlesungen iiber die Entwicklung der Mathematik im 19. to M at p. I(E”) is the tcongruent transformation group. c(= 1. A Euclidean space E” has a Riemannian metric expressed by C&dx’ Q dx’ in terms of an orthogonal coordinate system (xi). If a differentiable manifold M is the product manifold of Riemannian manifolds (M.) If M = E3 and N is a 2-dimensional submanifold of M..)“‘. gl) and (M2. q) = d.80 Connections K). . and d is the distance function of these spaces. I.(X. p.11= IILl\g. (For example. they are linearly independent at . of g at each point PE M. we define the length (IL.($(p). The metric g is a fcovariant tensor field of order 2 and of class c’-‘. h) is said to be isometric to (M.~~~” be the tcanonical l-forms on B. The length ((x(( of such a curve x is defined to be [a” llx’(t)ll dt.364A Riemannian 1348 Manifolds topology of M. is the i&mum of the lengths of curves of class D” joining p and q. and the concepts of tcanonical parameter and orientation of x can be defined (. is introduced on the ttangent vector space T.(L. YE TP.111 Differential Geometry of Curves and Surfaces). ti] such that the restrictions x 1[tieI. 1956). As in a Euclidean space. Utilizing the properties of the space E”. Riemannian Manifolds Metrics Let M be a tdifferentiable manifold of class C (1 < r < o). p.n:g. A function d : M x M + [0.) The covariant differential Vg of the fundamental tensor g vanishes identically.P of L to be the quantity g. +n:g2) is called the Riemannian product of M. b] into finite subintervals [tiei.365 Riemannian Submanifolds). If ‘p*g = h. where x’(t) is the tangent vector of x defined for almost all values oft. b] + M is called piecewise smooth or of class D” if x is continuous and there exists a partition of [a. and M. The set I(M) of all isometries (isometric transformations) of M onto M is a group. Let F be the ttangent n-frame bundle over M and B = B.g2). we can introduce various notions on T. Jahrhundert I. then a Riemannian metric ‘p*g is defined on N by the tpullback process ( 11 L II. Then (~()r$~$. Then (M.(M) be the subset of F consisting of all orthonormal frames with respect to g. called the tangent orthogonal n-frame bundle (or orthogonal frame bundle).(M) with respect to g. Klein. Let V denote the tcovariant differential operator defined by this connection (.(N) of T.. Assume further that ‘p is a diffeomorphism and N has a Riemannian metric h. B.. If there exists an immersion cp of a differentiable manifold N in a Riemannian manifold (M.. given a tangent vector LET. acts on any tensor field T defined on a submanifold having X as a tangent vector field.. are projections from M to M. 364 (Vll. and hence TP can be considered as a fvector space over R with inner product that can be identified with the Euclidean space E” of dimension n = dim M. Springer.. Y).).

.))’ defined on Exp. the differentiable vector bundle over S consisting of all normal vectors at all points of S. of p such that there exists a unique geodesic arcx joining any two points q and r of W.” of I. be the largest U with this property.(x(b)) to x(a)) is regarded as an isomorphism of the inner product space Txe. then t-+1. oi’. Given a tensor field K on M. Y) be an orthonormal basis of a 2-dimensional subspace P of Tp.: U+M of class C” with the following property: There exists a geodesic arcx with the initial tangent vector LE U. of B.. (1. called the normal coordinate mapping.111 Differential Geometry of Curves and Surfaces). Then the inner product K.. then (Exp. is denoted by V. Let N(S) be the normal bundle of a submanifold S of M.” and called the parallel displacement or parallel translation along x. oi). and we get (dJ/)*(@) = coi. The development X of x is the curve in E:. and for a given curve x: [a. (the composite of I.( V. Let U. is an embedding. The image Exp. In this case. In this way the problem of the existence of an isometry from M may be reduced to one of the existence of a diffeomorphism from B preserving absolute parallelism (as well as the order of the basis (w’. is of class C” at t. satisfies the tstructure equation dw’= . then I.1349 364D Riemannian Manifolds every point).80 Connections L). W.. and we get [lx/l = IIxII. Exponential Mapping (178 Geodesics) desic arc. b] of x is a geo- . 417 Tensor Calculus ) of type (1. The rotation part I. to BN.3) on M. defined by x(t) =1&x(t)). (2) Differentiability: If x is of class C” at t.. Let (X. I. then L or Exp.! A oj.. Moreover. length [[XII = IlLll. 279 Morse Theory). Let (@) and (@) be the corresponding set of differential l-forms on the orthogonal frame bundle &. of M. if there exists a diffeomorphism Y: BIM+BN satisfying Y*(P)= oi.). V.w. if Rjkl are the components of R with respect to an orthonormal frame b E B of the tangent vector space T. c Us.(P) of X and R(X. which is denoted by the same symbol I.(L) and points of S. and the focal point of p is called the conjugate point of p.(L) is called the focal point of S on the geodesic s+ Exp. Then S is contained in N(S) as the set of zero vectors at all points of S... is less than n at LE Us. C.) Utilizing the concept of development. that is.. In particular. N({ p}) coincides with the tangent vector space T. If the rank of the Jacobian matrix of Exp. Curvature The set of differential l-forms (w’. and the parallel displacement of E.&‘w: A oj” + Qj. the theory of curves in E” can be used to study curves on M (. 1V. (iii) the restriction Exp.. given as the zero point of the tJacobi field (.K = [dI$(K(x(t)))/dtll=. then x or x( [a. the Riemannian connection on M determines a Wartan connection uniquely with E” = I(E”)/O(n) as tfiber.C. Y). (3) I..h is a diffeomorphism from B = B. b] there exists an isometry Ix. (ii) llLl[ = d(Exp. As a consequence. If Tp is identified with R” (or E”) by means of an orthonormal basis of T.(sL) (0 <s. SL E IQ.* by I. then S has an open neighborhood V.. which is called the Euclidean connection.: Us+ M is determined uniquely by S. $ is uniquely determined if we choose one B. in N(S) satisfying the following three conditions: (i) V.. b] + M of class D” and for t E [a. According to the general theory of alhne connections. There exist a neighborhood U of S in N(S) and a mapping Exp. the tFrenet formula is automatically formulated and proved.(M). we have V. I: is extended to an isomorphism of the ttensor algebra y( Tx& to g( T. Conversely. dwj = . in which case K is said to be parallel. S) for LE V’. is called a convex neighborhood of p. and (4) is called the curvature form of the Riemannian connection of M. Exp. called the curvature tensor.) is a coordinate mapping. every tangent vector space T.(M) is regarded as a Euclidean space E”. translating I. then the differential dt. If S is compact.): (1) If x is a composite of two curves y and z. then K$ = (1/2)x R&ok A 0’ at b. D. In the special case where S consists of only one point p.: E&--+E’& satisfying the following three conditions (we denote Ix. then there exists an isometry II/ : M + N such that dl(/ = Y holds on a connected component B.. is sometimes called the development along x. If there exists an isometry + : M -+ N. Furthermore. r) and contained in W. . and final point Exp..r to T&. = I. if X is a segment....). a necessary and sufficient condition for VK =0 is that I:(K(x(b)))= K(x(a)) for any x. (dtJ)*(@) = o{.178 Geodesics.( V.. The mapping Exp.. where the right--hand member expresses the intimum of the distance between the point Exp. with llxll= d(q. over another Riemannian manifold N with dim N = dim M. b]) is called the geodesic arc (.( VJ contains a neighborhood W. depends on the orientation of x but not on the choice of its parameter.80 Connections.(L). For example. Y) Y is determined by P independently of the choice of the basis (X..) is the tubular neighborhood of S... This form is expressed by a tensor field R (.(L). is called the exponential mapping on S. ‘r*(eij)= w/ and M is torientable. Then Exp. where the i-component of A curve x on M or the image of x is called a geodesic if any subarc x 1[a. by means of which absolute parallelism is given in the orthogonal frame bundle B of M..

. and R. then M is called a space of constant curvature. Y)Z with respect to the basis b of T. the sphere (which is the universal covering Riemannian manifold of a real ielliptic space).).232 Kahler Manifolds).. {j.. were completely classlied by J.412 Symmetric Riemannian Spaces and Real Forms. is identified with E”) independent of the choice of p (. is a subgroup of the orthogonal group O(n) of T. .) containing L and is called the Ricci curvature (or mean curvature) of the direction L at p.(P) for all sections P (2-dimensional subspaces of T. be the set of all closed oriented curves of class D” with initial and final points p and with parameters neglected. is the sign of (ii.. is the mean of K. is the value of the tensor R at p. where b. is a subgroup of I(T. ( . is given by C R&ZjXkY’. A. 413 Symmetric Spaces). C is summation over all pairs of s-tuples satisfying {ii. is a connected component of h and a tcompact Lie group.( V. given by (R. If the Ricci tensor of M is a scalar multiple of the fundamental tensor. positive. The rotation part h.(P) is the +Gaussian curvature of the surface Exp.(P) has a constant value M. A complete. If dim M > 3 and if at every point p of M. . ..). j. E.. YE Tx(. elliptic space.2.(R. For example. If K.(R(Xi. ) is the inner product in T. this scalar is constant. Xi) and R = c.} c { 1. o is the volume element of M. The Ricci tensor (Rij) is defined by R. K. Riemannian metrics of these spaces are uniquely determined by the curvature tensor R up to a constant factor. i. . is defined as follows: For a positive even number s. is a real-valued function of the sdimensional subspaces P of the tangent vector spaces Tp of M...364 E Riemannian 1350 Manifolds classified [224]. in terms of an orthonormal basis (Xi) of T. where R. X.. the integral of a. The compact spaces of positive constant curvature.. and x*1. then M is called a locally symmetric space (. Y) is the endomorphism of the linear space TxcbJ defined in Section D. The mean R of Q(L) for all the unit vectors L at p is called the scalar curvature at p (.l)s/2/(2si2 s!). and simply connected.(X.. called the holonomy group of M. The restriction Ho of this homomorphism to all closed curves homotopic to zero is called the restricted holonomy group.) If M is a +Kahler manifold and P is restricted to a complex plane (invariant under the almost complex structure). or a real thyperbolic space according as K is 0. Then the value Q(L) for a unit vector LE T. then M is called an Einstein space. with respect to gp. If VR = 0.}. then the kth tpontryagin class of M (with real coefficients) vanishes for all k > s/2. complete. E~. then it is biholomorphic to the complex projective space [S.).+o on M is equal to the tEuler-Poincare characteristic. and these properties reflect those of the topological structures of the manifolds (. = K. The set H = {Ix 1xcn. Y))l x: [a. The compact simply connected homogeneous Riemannian manifolds of strictly positive sectional curvature have been in terms of an orthonormal basis (Xi. K. b]-M is of class D”.. If K. Furthermore. and locally symmetric space is a tsymmetric Riemannian space. we have the Gauss-Bonnet formula: If M is an evendimensional compact and oriented Riemannian manifold. The tLie algebra of h. called the restricted homogeneous holonomy group. 61 (. that is. Let Q be the quadratic form on T. Schur). and let 0. In a local sense.80 Connections). (When dim M > 3. . L)L. and X.. and K. is a homomorphism from ap to H.(X.Kc.~. ..) of P.(P) of p and P. = n!/(2”n@(n/2)!).. simply connected. R.) (T..178 Geodesics). or hyperbolic space. Kc. . then K(P) is called the holomorphic sectional curvature... the Riemannian manifolds having the sphere as the universal covering Riemannian manifold. as the solution of the Frankel conjecture. then M. .. then M is isometric to E”.j.. The rotation part h of H. n}..417 Tensor Calculus). or negative. curvature tensors are related to tcharacteristic classes. is spanned by {I. the following holds: If a compact Kahler manifold has strictly positive sectional curvature. The properties of the sectional curvature and the Ricci curvature are closely related to the behavior of geodesics of Riemannian manifolds. where u. of Ho..}. . up to positive constant factors.= --xkRbk. Related to algebraic geometry. Q(L) and R are expressed by Q(L) = Cig.. The curvature tensor R is uniquely determined by the function K. = ( .. A Klhler manifold M of constant holomorphic sectional curvature is locally isometric to a complex Euclidean space. independent of the choice of P. n P) and is called the sectional curvature (or Riemannian curvature) of P.(X. which is given by R(X. K. Holonomy Groups Let p be a fixed point of M. Xj)Xk is as already defined at the beginning of this section. called the homogeneous holonomy group.Q(X. If M is of constant curvature K. of a compact and orientable M is constant for a certain s. Wolf [l].. x(a)=p. is a constant independent of the choice of p (F.(P) is constant. i. In particular.

. If. If M is complete and H (regarded as a transformation group of E”) has a fixed point. there exists a unique parallel tensor field A satisfying A. Conversely.(M) is clearly decomposed into a direct product by the de Rham decomposition of M. If h = {e} (ha = {e}). we get a unique decomposition T. Furthermore. A necessary and sufficient condition for the image of p to be a subbundle of B is that I(M) be transitive. With respect to the linear group h of T. or a real telliptic space (or a sphere). then the image of /I is contained in the h-bundle (.. Y..1351 364 F Riemannian Manifolds If M = E”. the set of all geodesics. r. and the maximum dimension is attained only when M is a space of constant curvature. This equation is called Killing’s differential equation. then I(M) is discrete. In general. then H = (e}. an element of I(M) preserves quantities uniquely determined by the Riemannian metric g. satisfying I$. acts on I&.. P. . if M is compact. a real thyperbolic space.. This decomposition M = l-J MC.80 Connections E). @ . Transformation Groups The group I(M) consisting of all isometries of M with the tcompact-open topology is a tLie transformation group. then M is isometric to E”. 1. i = 1. furthermore. . then h. and a solution X of this equation is called a Killing vector field.9]. Then each connected component of F is a closed ttotally geodesic submanifold of M. are irreducible h-invariant subspaces. then A.). then an isometry of M homotopic to the identity transformation is the identity transformation itself. Any finite rotation group h is the homogeneous holonomy group of some locally flat and compact Riemannian manifold. A necessary and sufficient condition for h to be contained in U(n/2) is that M have a tcomplex structure and the structure of a Kahler manifold. Muto. acts ttransitively on the unit sphere of Tp. > 0) consists of all h-invariant vectors and I$. K. If a tensor field A on M is parallel. . . @ $. of mutually orthogonal subspaces. then M is called irreducible (reducible). = I$. where e is the identity element. If h. Obata. Kuiper.). etc. any element of I(M) commutes with V and the tlaplace-Beltrami operator. then this Lie algebra coincides with that of I(M). then M is complete and is the thomogeneous space of Z(M). is a fixed point of B. i= 1. a homogeneous space M = G/K of a Lie group G by a compact subgroup K has a Riemannian metric invariant under G. It is known that dim I(M) < n(n + 1)/2 if dim M = n. If M is compact and the Ricci tensor is negative definite.. its curvature. The differential dcp of cp~1(M) is a transformation of the orthogonal frame bundle B. called the infinitesimal motion. If M is compact and I(M) is transitive.. if n is even and h. ... extensive work on the structures of M and I(M) has been done by I..). If b is surjective. /? # t( as the identity. that is. such as the Riemannian connection. = T( M. is determined uniquely by M and called the de Rbam decomposition of M [7]. If h or h. Ishihara. n (. For example.. denotes tLie derivation and the & are tcovariant components of X with respect to a natural frame (a/ax.) is invariant under h. E y( T.417 Tensor Calculus). = x(F). In this case h is the direct product of closed subgroups h(. Yano.. Conversely. An element of the Lie algebra of I(M) is regarded as a vector field X on M. r. If the image of fi contains the h-bundle. the sectional curvature is nonpositive. The set of all Killing vector fields is a Lie algebra of finite dimension ( < dim B). . For Riemannian manifolds with large I(M). . cc=O. . 11). The group h acts naturally on the ttensor algebra F(Tp) of Tp.. is irreducible (reducible) on T.... which satisfies the equation L. if A. F. The orthogonal frame bundle B is treducible to the h-bundle. Local flatness is equivalent to M being locally isometric to E”. Egorov. Nagano. T.g = 0...) embeds I(M) as a closed submanifold of B.(dim I’&. Wakakuwa. then h acts transitively on the unit sphere.. and can be regarded as the homogeneous holonomy group of MC. If M is complete. If M is compact and oriented. is the tunitary group U(n/2). N. H.. is irreducible and M is not locally symmetric. is invariant under h. The isotropy subgroup at any point is compact.191 G-Structures).(M) of I(M) preserves any tharmanic differential form. Vjti + Vi tj = 0.. where every h(. then M is a symmetric space. then I. then the mapping fi defined by cp+dq(b. and the differentiable structure of I(M) is thus determined. Mann. For example. it follows from the structure equation that M is of constant curva- ture and equals E”. If b. and others [lo. In particular. where L. . then M is called flat (locally flat) (. where Af denotes the TLefschetz number and . = A. If M is complete and simply connected. H. The classification of possible candidates for such h. L. then the connected component I. then A. so is I(M). If M is compact and f is an isometry of M. N. where I/. let G be any subset of I(M) and F the set of points of M which are left fixed by all the elements of G. S. has been made [8. If I(M) is transitive on M. The fixed point set of a family of isometries has interesting differential geometric properties [lo]. If M is complete and simply connected (hence h = h. M. . @ I$. then M is the Riemannian product of closed submanifolds MC.

namely. S. For example..(M)#I. under the same assumption on the Ricci curvature. unless M is a sphere (n > 2) (K. then f has a fixed point. Hano). if xi = n./?. Tanno. Af= &f. there are compact Riemannian manifolds with constant scalar curvature for which C. Hsiung. Tanaka. A subset of C(M. the eigenvalues of the tlaplace-Beltrami operator A on smooth functionsaregivenbyO<I. Ejiri). They are Lie transformation groups with respect to suitable topologies. a transformation of M which preserves the Riemannian connection.g) is called essential if it cannot be reduced to a subset of I(M. if C. Let P(M) denote the group of all projective transformations of M.(M) is essential if and only if it is not compact. (E. Yamauchi).1). orientable Riemannian manifold admits a lixedpoint-free l-parameter group of isometries. If M is compact.) is an infinitesimal conformal . A transformation preserving the set of all geodesics is called a projective transformation. Tashiro). then A. simply connected. X(F) the tEuler characteristic of the fixed point set F off: As for the existence of fixed points of an isometry. A sphere is characterized by the existence of solutions of certain differential equations on a Riemannian manifold. Ishihara. For example. In general. to e21g.(M) = A. If a compact. A.(M) = I. then its tpontryagin numbers vanish. S. then M is locally isometric to a sphere (Obata. then A(M) = I(M) except when M is a l-dimensional Euclidean space. S.(M) always. N. If M is compact.(M). A transformation preserving the angle between tangent vectors is called a conformal transformation.. In the case of nonpositive curvature.complete. On the other hand. I(M)cA(M)cP(M).) if and only if M is a sphere (Obata. leaves no nonzero vectors. Cartan).(M).1)f as well as a certain system (Ek) of differential equations of degree k + 1 involving the Riemannian metric. If M is complete and irreducible. Let C(M) denote the group of ail conformal transformations. f being any G. Tanno. sufficient conditions for M to be isometric to a sphere have been obtained by S. 3) for any metric g conformal to k b 2. Y.(M) (N. C. then the first eigenvalue 1.. it is known that if M is complete and has a parallel Ricci tensor. Let A(M) denote the group of all affme transformations of M. C. Obata. then A. if g is the standard metric on S”. A. or if dim M is odd and f is orientation reversing. and others. then the connected component C.(M) is essential. Conversely. of A on M satisfies 1. When M is compact. Similarly to the case of P(M). unless M is a space of positive constant sectional curvature (n > 2) (Nagano. Lichnerowicz. simply connected Riemannian manifold with nonpositive sectional curvature has a fixed point (E. Kobayashi. or equivalently which commutes with covariant differentiation V is called an affine transformation. It is a space of positive constant sectional curvature l/r2 (Y = radius) with respect to the natural Riemannian metric as a hypersurface of the Euclidean (n + 1)-space En+‘. The gradient of a solution of (E. the restriction f to S” of a harmonic homogeneous polynomial of degree k satisfies Af= k(n + k . then Af = nfis equivalent to the system of differential equations vjvif+fgji= 0.. then M is conformally diffeomorphic to a sphere or a Euclidean space (n>2) [12-151.(M).. Goldberg and S. Clearly. If M is complete and its restricted homogeneous holonomy group h. If dim M is even and f is orientation preserving. I. If M is complete and has a parallel Ricci tensor. On a unit sphere S” in E”+l. then M is a sphere (Obata).<. the following is basic: Every compact group of isometries of a .. orientable Riemannian manifold M with positive sectional curvature. = k(n + k . Yano. then M is a sphere [14].364 G Riemannian 1352 Manifolds g.. When M is compact and has constant scalar curvature and C. Tashiro). the following are known: Let f be an isometry of a compact. I(M)cC(M)(191 GStructures). C(M) or C.(M) is essential.(M) = I. are the restrictions to S” of harmonic homogeneous polynomial functions F of degree k on En+‘. On a compact Riemannian manifold M. On a Riemannian manifold M.(M) = I. It is known that eigenfunctions f corresponding to L. and has constant scalar curvature. then the connected component P. then P.<&<. If a complete Riemannian manifold M admits a nontrivial solution of (Ek) for some integer smooth function on M.(M). Gallot [17]). however.(M). > Ai = n [16]. is decomposed into a direct product according to the de Rham decomposition of M when M is complete and simply connected (J. In general. if the Ricci curvature of M is not less than that of S”.) A complete Riemannian manifold M (n > 2) admits a nontrivial solution of (E. A conformal transformation remains conformal if the Riemannian metric g is changed conformally.(M) # I. unless M is a sphere (n > 2) (Nagano)..(M)=I. If C. the connected component of A(M).(M). K. Tashiro. Spheres as Riemannian Manifolds A Euclidean n-sphere S” (n > 2) has the properties of a Riemannian manifold.

N. by the tGauss-Bonnet theorem the Gaussian curvature K of M must satisfy the following sign condition in terms of the tEuler characteristic x(M): if x(M) > 0. namely. On a sphere. it has a positive constant sectional curvature [ 143.1353 3641 Riemannian Manifolds has been solved affirmatively transformation and that of (E. it has since been In this paragraph the manifolds under consideration are assumed to be of dimension n 2 3. H. then K changes sign unless it is identically zero. In fact. E. Y. If M carries a metric g with K. For positive scalar curvature. one can say that a smooth function K is the Gaussian curvature of some metric conformally equivalent to the original metric if and only if K satisfies the foregoing sign condition [ 183. any smooth function on a compact manifold M of dimension n > 3 that is negative somewhere is the scalar curvature of some metric on M. of a Riemannian metric g. S. Scalar Curvature On a 2-dimensional Riemannian manifold M. respectively. Marsden. then K is positive somewhere. If M carries metrics gi. Tanno. Then the following are known for a compact manifold M of dimension > 3: If M carries a metric g with K. Ricci Curvature and Einstein Metrics where R is the scalar curvature of g and R a constant which should be the scalar curvature of ~=u4’(“-‘)g (. 3. g2 with Kg1 < 0 and Rgl 2 0. which is a function on M. In particular. Yamabe’s original proof can be pushed to cover a large class of metrics with jM RdM < 0. E. The existence of such a manifold has been shown. then locally it is almost always surjective when M is compact (A.417 . Kobayashi. then there exists a Riemannian metric of positive scalar curvature. denote the sectional curvature and the scalar curvature. any metric of nonnegative scalar curvature on T” must be flat [22]. a Riemannian metric which is conformal to the standard metric and has the same scalar curvature as the standard one is always standard. Let K. 0. On the other hand. If the assignment of the scalar curvature to a Riemannian metric is viewed as a mapping of a space of Riemannian metrics into a space of functions on a manifold M. pointed out that his original proof contains a gap in some cases. then it carries no metric with R > 0. Any smooth function can be the scalar curvature if and only if M admits a metric of constant positive scalar curvature. D. then M carries a Riemannian metric of positive scalar curvature [22]. g) of dimension n > 3. Maeda). ~0. there is a topological obstruction. A compact tspin structure (spin manifold) having nonvanishing ta-genus cannot carry a Riemannian metric of positive scalar curvature. n-2 I. The Bianchi identity (. and the scalar curvature all coincide with the tGaussian curvature. simply connected manifold M of dimension n > 5 is not a spin manifold. Trudinger. if x(M) < 0.183 Global Analysis). If M is compact. This sign condition is also sufficient for a given function K to be the Gaussian curvature of some metric on M. there is a system of differential equations characterizing the complex projective space or the quaternion projective space as a Klhler manifold (Obata. Lafontaine). then both metrics are flat [22]. As the Kahler or quaternion Kahler version of (E2). then K is negative somewhere. however. The Ricci tensor (Rij) is a symmetric tensor field of type (0. More precisely. If a compact. then it carries no metric with R > 0. A torus T” cannot carry a metric of positive scalar curvature. Yamabe [ 191 announced that on every compact Riemannian manifold (M. Furthermore. tensor reduces to the one of solving a system of nonlinear second-order partial differential equations for g. and R. The problem reduces to the following nonlinear partial differential equation on a compact manifold M: &p+m-2) =4n-l*u+Ru. The foregoing results show that there is no topological obstruction to the existence of metrics with negative scalar curvature of a compact manifold of dimension n >. starting with a Riemannian metric with constant Gaussian curvature. if M is not conformally flat and n > 6. on a compact manifold (n > 3) there always exists a Riemannian metric with constant negative scalar curvature [18]. Fischer and J. Furthermore. J. The problem of finding a Riemannian metric g which realizes a given Ricci solved for a wider class: namely. < 0. H. Nevertheless. Blair. if M is a spin manifold and spin tcobordant to M’ with positive scalar curvature. then the problem c201. there exists a strictly positive function u such that the Riemannian metric B = uq(n-2)g has constant scalar curvature. the sectional curvature. if x(M) = 0. the Ricci curvature.2) on a Riemannian manifold. E. or if it is conformally flat and its fundamental group is finite.) is an inlinitesimal projective transformation.

then in a neighborhood of p there exists a C”’ (or C”‘) Riemannian metric g such that (Rij) is the Ricci tensor of g [24]. Differential Geometry. then they are isometric [30]. [22] M. Spaces of constant curvature. Math. Sup. Simons. Mori. then the tnullity and +coindex at the critical point are finite [28. (9) 55 (1976). The conjectures on conformal transformations of Riemannian manifolds. Math.364 Ref.. [ 121 J. Math. l-44. L. Math. (2) 111 (1980). 1958. 12 (1979). North-Holland. Trans. no. equations diffkrentielles CaractCristiques de la sphkre. (2) 101 (1975). Wallach. Pures Appl. [2] M. Cl. [ 151 D. Compact KLhler manifolds of positive bisectional curvature. 18 (1972). 70. J. Ann. J. Let uy denote the volume element determined by g.29]. [3] N. . Sot. 247-258. (2) 76 (1962). On a deformation of Riemannian structures on compact manifolds. Math. Scuola Norm. The positivity of the Ricci curvature on a Riemannian manifold puts rather strong restrictions on the topology of the manifold (. Kobayashi. 47-67. 59 (1980). Berard-Bergery. Bull. Though S” with standard Riemannian metric is a typical example of an Einstein space. Yau.. is a functional on . Helv. Pure Appl. [4] L. B. on a complete Riemannian manifold M with nonnegative Ricci curvature. Ecole Norm.X. y) is an Einstein space. nonnegative Ricci curvature and positive Ricci curvature are not too far from each other. i Osaka Math. USSR-Sb. 189-204.. 1972.178 Geodesics). (3) 15 (1961). 285-301.. Math. Publish or Perish. Compact homogeneous Riemannian manifolds with strictly positive curvature. In particular. Math. J. Let &Z. France. Schoen and S. Transformations conformes et quasi conformes des varittt&s riemanniennes compactes (dCmonstration de la conjecture de Lichnerowicz). 593-606. ( c A) denote the space of metrics with constant scalar curvature. there is a point at which the Ricci curvature is positive. 277-295. Wolf. 26 (1952). Math. T. (2) 96 (1972). Compact Riemannian manifolds with essential group of conformorphisms.. Ann. 423-434. [S] M. Ann. The structure Tensor Calculus) must be satisfied. Dunod. 235-267. Harmonic spinors. Groups of conformal transformations of Riemann spaces. There is a symmetric (0. Transformation groups in differential geometry. Comment. Sci.. [7] G. 328-344. [ 131 A.. Berger. [Zl] N. Lelong-Ferrand. If a Riemannian manifold (M. Ann. J. . 6 (1971). An Einstein metric is always real analytic in some coordinate system. if a C” (or Cw) symmetric tensor field (Rij) of type (0. [ 111 K. Amer. Yano. Obata. [ 191 H. References [l] J. Erg.~I. SC. When M is compact. Warner. Obata.. 55 (1976). Sot. [14] M. [lo] S. The integral of the scalar curvature y(g) = sM R. Yau. [18] J. Math. de Rham. Pisa.2)tensor on R” which cannot be the Ricci tensor for any Riemannian metric in a neighborhood of OER”. Equations diffkrentielles non lintaires et problkme de Yamabe concernant la courbure scalaire. Alekseeviskii.2) is invertible at a point p. Ann. Yamabe. Hitchin. Gromov and L. [ 161 A. 645-651. Sur les groupes d’holonomie homogine des varittks B connexion afIine et des variCt&s riemanniennes. Aubin. Colloq. S4h+3 (k > 1) carries an Einstein metric that is not standard [31]. Projective manifolds with ample tangent bundles. A. Belg. Ann. Then if ?? is restricted to &‘. then g is called an Einstein metric on the manifold M. On the transitivity of holonomy systems. However. T. Acad. 12 (1960). Sup. [17] S. Les vari&tCs riemanniennes homogknes simplement connexes g courbure strictment positive. The critical points of $? are Einstein metrics (D. [20] T. Math. Gallot. Lichnerowicz. Advances in Math.. Lawson. Hilbert). Kazdan and F. [S] S. then there exists a complete metric on M with positive Ricci curvature [25&271. If. 1977. V. 317-331. Ann. Ledger and M. Math. 279-330. Siu and S. W. Math. 213-234. 269-296. Riemannian 1354 Manifolds dimension impaire g courbure strictment positive. [6] Y. Existence and conformal deformation of metrics with prescribed Gaussian and scalar curvature.. GkomCtrie des groupes de transformations.1 denotes the space of Riemannian metrics on M with total volume 1. Sur la reductibilitt d’un espace de Riemann. if two simply connected Einstein spaces have neighborhoods on which metrics are isometric... Inventiones Math. 179-246. The classification of simply connected manifold of positive scalar curvature. l-55.. 1957. 83 (1955). MCm. 39. 14 (1974). [23] R.. T. 150 (1970)... However. 5 (1971). Roy. (2) 110 (1979). [9] J. The theory of Lie derivatives and its applications. 21-37. R. J. Les variCtCs riemanniennes homog&nes simplement connexes de . Berger.

4 (1970) 383-424. J. 111 Differential Geometry of Curves and Surfaces. f(M) will be identified with M except where there is danger of confusion.. gent bundle T(M). 9 (1974). Manuscripta Math.. [27] J. [25] T. E. Differential Geometry. Springer. and their twhitney sum F(M) @ F. General Results for Immersibility An n-dimensional real analytic Riemannian manifold can be locally isometrically embedded into any real analytic Riemannian manifold of dimension n(n + 1)/2 (M. 105 Differentiable Manifolds. Ecole Norm. 365 Riemannian Submanifolds. Sup.. The generalization to the C” case is an open question even when the ambient space is Euclidean. 249260. which is called the second fundamental form in the direction of 5.(M). Lecture notes in math.(M) of orthonormal normal frames of M. g) into a Riemannian manifold (fi. the ttan- C. Ehrlich.1355 365 C Riemannian Submanifolds v(M). 191 GStructures. Fundamental Equations Let f: (M. An ndimensional noncompact c’ Riemannian manifold (3 < r < co) can be isometrically embedded into a 2(2n + 1)(3n + 7)-dimensional Euclidean space (Nash (1956). 4133421. Ziller. Introduction If an timmersion (or an tembedding) f of a tRiemannian manifold (M. Sot. B. (3) p < n . Greene (1970)). Muto. Bourguignon. Geometriae Dedicata.. 159-183. [26] P. Kazdan. Metriques riemanniennes et courbure. 365 (VII. Aubin. Existence of metrics with prescribed Ricci curvature: Local theory. 16 (1979). Suppose dim M = n and dim fi = n +p. 109 Differential Geometry. O’Neill(l960)). Ann. E. Let V’ denote the covariant differentiation with respect to the normal con- .1. Then the tbundle F(M) of orthonormal tangent frames of M. [31] J. Ricci curvature and Einstein metrics. 18 (1979).Y. M. An n-dimensional compact C’ Riemannian manifold (3 < r < co) can be isometrically embedded into an (n(3n + 11)/2)-dimensional Euclidean space (J. J. S. On Einstein metrics. K. Amer. 178 Geodesics. respectively. Osaka J. defines a symmetric linear transformation on T. the tangential component of vx Y is equal to Vx Y. and fi is complete and simply connected [2]. Otsuki (1954)). F.(M) are.2) tensor field on M. Cartan (1927)). [30] D. Some regularity theorems in Riemannian geometry. K. respectively. S. 179-207. Sci. Then A. 65 (1981). < 0. DeTurck. Y). P. Tnventiones Math. Math. and O(n) x O(p). (1) Then c is a v(M)-valued symmetric (0. Y)= Q(cr(X. Janet (1926). [29] N. O(p). (2) P and fi is complete and simply connected (C. then f is called an isometric immersion (or embedding) and M is called a Riemannian submanifold of ii?f.1 3) Riemannian Submanifolds A. Tompkins (1939). Koiso. For a normal vector 5 at XE M. and iii an (n + p)-dimensional Riemannian manifold with sectional curvature Ka. and F(M) @ F. B. Q) be an isometric immersion. On the second derivative of the total scalar curvature. Y)=VxY-V. D’Atri and W. the tnormal bundle and their Whitney sum T(M) @ v(M). Chern and N. 838. Let V and V denote the tcovariant differentiations with respect to the tRiemannian connections of M and fi. Mem. 14 (1981). E. g) satisfies the condition f*g = g. These are subbundles of the restriction to M of the bundle F(fi) of orthonormal frames of fi. In this article. of manifolds with positive scalar curvature. l-23. respectively. 521-530.42-63. DeTurck and J. The eigenvalues of A. the bundle F. < KG < 0. Kuiper (1952). KG is constant ( < 0). [28] Y. Differential Geometry. R.X. M is compact. Metric deformations of curvature I: local convex deformations.. The connection on v(M) induced from the Riemannian connection of fi is called the normal connection of M (or off). 1981. 28 (1979). 417 Tensor Calculus. M is compact. put g(A. Then M cannot be isometrically immersed into fi in the following cases: (1) p<n-2 and Knr<Ka (T. H. 5 (1976). For vector fields X and Y on M. [24] D. < n 1. L. g)+(fi. Let M be an n-dimensional Riemannian manifold with tsectional curvature K. F. See also references to 80 Connections. Put 0(X. Nash (1956)). 5). which is called the second fundamental form of M (or off). Naturally reductive metrics and Einstein metrics on compact Lie groups.(M). are called the principal curvatures in the direction of 5. The vector bundles associated with F(M).(M) are tprincipal fiber bundles over M with tstructure groups O(n). Math.

respectively.). Let C@jj.. suppose fi is a ispace form of constant curvature c. by q(A. a(e. (3.Z)Y) (V. ker A..X)ii/llXl12 does not depend on XE T.) Q(R%‘.BCn+p be the iconnection form and the tcurvature form of fi with respect to (eJ.. A4 = dim nSEY.X (4) D.. Then the equations of Gauss. then M is called a totally umbilical submanifold of fi.).. ej) =C wT(ej)e. Let R. A point XE M is called an isotropic point if Ila(X.(n is the connection form of M with respect to (ei)isisn.Z)-(V. sACn+p be a local cross section of F(R) such that its restriction to M gives a local cross section of F(M) @ F.Z) A O(X. Y)Z=CMY. and suppose there is given a p-dimensional +Riemannian vector bundle v(M) over M with curvature tensor R’ and a v(M)-valued symmetric (0. is proportional to the identity transformation for all <E v.I.)” (4. Then . As a particular case. a(q.(M).~)(Y. If every point of M is an isotropic point.BQn+p and (&% QA. V. ) is the fiber metric of v(M). that is.. Then the equations of Gauss. where V’ denotes covariant differentiation with respect to the connection in T(M) @ v(M). which is called the equation of Ricci.). (1) is called the Gauss formula. and (4. Y)Z= R(X. Then the tintegrability condition for (1) and (2) implies d(X.+. Y). Basic Notions Let M be a Riemannian submanifold of iii. x E M is an umbilical point if and only if A.p<n+p be the curvature forms of (w.. the tangential (resp.~)(x.X. Moreover. and Ricci reduce respectively to R(X. that is to say. If every point of M is an umbilical point. the relation vx. and R’ be the icurvature tensors of V. respectively. Y. 4i. then M is called an isotropic submanifold of a. and Ricci are given respectively by -dCA<>A. i?. A. (~9)~ Qign<aGn+p gives the second fundamental form.~)(X. M is a totally geodesic submanifold of fi if and only if every geodesic of M is a geodesic of IGJ. 0. Y)~.jgn and (@.) then A4 can be immersed isometrically into an (n + p)-dimensional complete and simply connected space form M”+“(c) of curvature c in such a way that v(M) is the normal bundle and 0 is the second fundamental form. define A.+I $n. A point XE M is called an umbilical point if rr = g 0 h at x.z) (2’) for vector fields X. Y)Z+ Aoo. (u$). and (2) is called the Weingarten formula. If every point of M is a geodesic point. then M is called a totally geodesic submanifold of fi.2) tensor field g on M. (3.Z)=o. A mapping h: M+v(Mj defined by x* ic:=.(. Formulas (l)-(4) are the fundamental equations for the isometric immersion ..f*o’= 0 for n + 1 <a < . Then (wj).. where ( .lX> Y) holds. Let (e.).275 Minimal Submanifolds).) and (c$). the tangential component of (3) is given by the equation of Gauss and the normal component of (3) is given by the equation of Codazzi. For a tangent vector field X and a normal vector field < on M.z)-(v.(M).f*Gg.).. normal) component of vxt is equal to -A:X (resp..j. h is called the mean curvature vector and 11 h 11is called the mean curvature.(M). Y). such an immersion is unique up to an tisometry of M”+“(c). If they satisfy (3. and put wi=. More precisely.= -A<X+V. Conversely. (3) is called the equation of Gauss and Codazzi. is called the index of relative nullity at XE M. g) be an n-dimensional simply connected Riemannian manifold. Similarly. For a icross section 5 of v(M).Aoo. +(v~~)(y.< (2) holds. and let (w”) be its dual. V... Y)= (a(X. M is called a minimal submanifold of fi if b = 0 (.‘<).pgn+p is the connection form of the normal connection with respect to (eJn+lgoGn+p. Codazzi. <i. (3. for vector fields t and q normal to M. Let (G% QA.W--Y(X.cM. Y. Codazzi.X (3) Put wr = C !I@‘.).f: M+fi.365 D Riemannian 1356 Submanifolds n+p. Then (h$) is the matrix representing the symmetric linear transformation A.%ei = 1 htej. Y)+A. the relation Moreover. and V’.~)=Y(CA<>~JX. It is clear that an umbilical point is an isotropic point. sa. let (M. A point x E M is called a geodesic point if cr = 0 at x. Z tangent to M. ei) is independent of the choice of an orthonormal basis (ei). thatis. (1’) nection. with respect to (e.

(3) n 2 5. These results remain true even if the assumption “the mean curvature is constant” is replaced by the weaker condition “the principal curvatures k. with respect to any volume-preserving variation of a domain D in a Euclidean space. . if f’:M-+fi is another isometric immersion. . Ferus (1970)). then there exists an isometry rp of iii such that f’ = cpof: If f: M-+ti is rigid. Then M is a space form M”(c) with c > E. (2) f is an embedding (A. Chen and T. or a parallel hypersurface of a totally geodesic hypersurface [2]. On the contrary. Chern (1955)). and the index of relative nullity is < n . Calabi (1967). (4) n > 4. then f(M) is locally a hyperplane or a hypersphere. if E=O. The answer is affirmative in the following cases: (1) dim M = 2.) = 0 such that &p/dki > 0. In particular. .1357 365 H Riemannian Submanifolds E. > n. . then . . Y. if E < 0. a sphere has plenty of compact minimal submanifolds. An isometric immersion f: M-+M”+l(c) of an n-dimensional Riemannian manifold into an (n + 1)-dimensional complete and simply connected space form is rigid in each of the following cases: (1) n = 2. an n-dimensional n sphere of curvature can be minimals(s+n-1) ly immersed into a (2s + n . . A generalization of (2) for the case of higher codimension was obtained by C. satisfy a relation dk 1.3 at each point (R. a sphere S”+’ admits many compact hypersurfaces of constant mean curvaturej among which totally umbilical hypersurfaces and the product of two spheres are the only ones with nonnegative sectional curvature (B.” Unlike an open hemisphere.2 (N. There is no compact minimal submanifold in a simply connected Riemannian manifold with nonpositive sectional curvature (O’Neill A manifold of constant mean curvature is a solution to a variational problem. Minimal Submanifolds For general properties of minimal submanifolds . Rigidity An isometric immersion f: M+fi is said to be rigid if it is unique up to an isometry of !i?. If E> 0. (2) The index of relative nullity is < n . > . D. Tanaka. Nomizu (1969)). and the tgenus of M is zero (H.[S]). the mean curvature of M = aD is constant if and only if the volume of M is critical. Among all n-dimensional compact minimal submanifolds of an (n + p)-dimensional unit sphere. A. Itoh 2(n + 1) WW). Cohn-Vossen (1929)). Totally Geodesic and Totally Submanifolds Umbilical A totally geodesic submanifold of a space form is also a space form of the same curvature. Harle (1971)).2 at each point (D. Ejiri (1979)).l)(“. Dolbeault-Lemoine. Sacksteder. .l}- dimensional unit sphere and the immeision ’ is rigid if n = 2 or s < 3 (E. < 0. where M”+l(c) denotes a Euclidean space. do Carmo and N. (1960)).275 Minimal Submanifolds. E. Totally geodesic submanifolds of compact tsymmetric spaces of rank 1 were completely classified by J. M. that is. A nonnegatively or nonpositively curved complete surface of nonzero constant mean curvature in a 3-dimensional spaceform M3(c) is either a sphere or a tclifford torus if c > 0 and is either a sphere or a right circular . 2-l/P > n(n . M is complete. and totally geodesic submanifolds of symmetric spaces of rank 2 2 were studied by Wolf and B. Let f: M +M”‘p(C) be a totally umbilical immersion of an n-dimensional Riemannian manifold M into an (n+p)-dimensional space form. or an open hemisphere according as c = 0. a horosphere. then f(M) is locally a hypersphere. and f(M) is contained in a certain (n + l)dimensional totally geodesic submanifold M”‘l(?) of M”+P(i?). (2) Ricci curvature (3) scalar curvature Simons (1968)). k. G. Alexandrov (1958). Wallach (1971)).1) -n (J. Smyth and K. The interesting question “If the mean curvature of an isometric immersion-f: M-M”+‘(c) of an n-dimensional compact Riemannian manifold M into an (n + l)-dimensional space form M”+‘(c) is constant. is M a sphere?” has not yet been completely solved. Various rigidity conditions have been studied by S. Hopf (1951).[S]).f(M) is locally a geodesic sphere. > k. c = 0. the totally geodesic submanifold is isolated in the sense that it is characterized by each of the following conditions: n (1) sectional curvature > __ (T. and M is compact and of positive curvature (S. and others.~n~~~! . a hyperbolic space. R. Kaneda and N. c > 0. Nagano [4]. c # 0. then every isometry of M can be extended to an isometry of A. or > 0. Wolf (1963).1)~) (C. Beez (1876). For each positive integer s. Submanifolds of Constant Mean Curvature F. and the tscalar curvature of M is constant (# n(n . Allendoerfer’ (1939). H.

R. c < ?. Sk x H’-k.4. and H. (3) If c ~0. Bianchi (1896). (1) If n=2. . If [M] E H. Ferus (1975)). M”(O). M’(0) is complete. (1) If c= 0. Liebmann (1901). - cm (3) If n=2. P. M2(0) is complete. c = ?= 0. then . Takagi. c=F>O. (2) If II = 2. c=O<?. An n-dimensional complete and simply connected complex space form is either P. then M is locally E” or Sk x Hnmk (Cartan). g) into a Kahler manifold (A. Volkovand S. Each of the hypersurfaces above except E2 in (2) and B in (3) is given as an orbit of a certain subgroup of the isometry group of M”“(c). A Kahler submanifold is a minimal submanifold.[Z]). then f(M2(0)) is either a horosphere or a set of points at a fixed distance from a geodesic (J. p= 1. and both M”(c)and M”+P(?) are complete. cylinder if c < 0 (T. Isometric Immersions between Space Forms Let f: M”(c)+M”+“(?) be an isometric immersion of an n-dimensional space form into an (n + p)-dimensional space form.1.f is totally umbilical. Isoparametric hypersurfaces of S”+’ having at most three distinct principal curvatures were completely classified by Cartan. y”) is called a Kiibler immersion. Takeuchi obtained several results for isoparametric hypersurfaces of Sn+i with four or six distinct principal curvatures [7]. then M is locally Sk x E”mk. . p = 1. Takagi. Ozeki and M. If c = 0. or a 3-dimensional group manifold with the metric ds2 =em2rdx2 +e2’dyZ +dt2 (Takahashi (1971)). c>O. p = 1. ?>c>O. or 6 (H. (6) If p = 1. p= 1. If a subgroup of the isometry group of M”+l(c) acts transitively on M. (8) If p<n. and M3(?) is complete and simply connected. A4 has at most two distinct principal curvatures (Cartan). Osserman (196661967) D. the number of distinct principal curvatures of M is 1. Klotz and R.f does not exist (D. Wirtinger (1936)). Isoparametric Hypersurfaces A hypersurface A4 of fi is said to be isoparametric if M is locally defined as the tlevel set of a function f on (an open set of) fi with property dfr\dlldfl12=0 and dfr\d(Q”)=O. Moore (1972)). p = 1. then M is isometric to Sk x E’mk (S. Miinzner (1980)). then vol(M)<vol(M’) holds for any submanifold M’E [M] with equality if and only if M’ is a Kahler submanifold (W. and M3(n is complete and simply connected. A tcomplex submanifold of a +Ktihle’r manifold is a Kahler manifold with respect to the induced Riemannian metric. (4) If n = 2. (5) If n > 3. Z) denotes the +homology class represented by a Kahler submanifold M of A. c>F.. If c > 0. S. M”(c) is complete. . Sasaki. Hsiang. M2(c) is complete. or D. Takahashi. Nagano (1960) Takahashi C91). Pogorelov (1956). K. c = 0 > ?. and c > ?.2. and M”+P(F) is complete and simply connected. Hilbert (1901). Y.365 I Riemannian 1358 Submanifolds then f is cylindrical (A. A complex analytic and isometric immersion of a Kahler manifold (M. then . then M is isometric to E”. and M3(c?) is complete and simply connected.[7]). that is. then there exist infinitely many f (L. then . (2) If c > 0. J. T. Let M be an n-dimensional thomogeneous Riemannian manifold which is isometrically immersed into an (n + 1)-dimensional complete and simply connected space form M”+’ (c). but not true in general (Ozeki and Takeuchi [7]). and others). and if c < 0. or if c > 0 and M has at most three distinct principal curvatures (Cartan). and M3(t) is complete and simply connected. Lawson. B. then f is totally geodesic (D. KIhler Submanifolds J. 3. 1. Cn. Vladimirova. Hartman and L. KChler immersions of complex space forms .(C). Every Kghler submanifold of a complex space form is rigid (Calabi [ 10)).f is totally umbilical (H. H. Homogeneous Hypersurfaces A hypersurface A4 of a complete and simply connected space form A4”+‘(c) is isoparametric if and only if M has locally constant principal curvatures (Cartan). then M is isometric to EZ or else is given as an orbit of a subgroup of the isometry group of M”+‘(c) (W. is complete. (7) Ifpdn-1. M’(C) is complete. Hoffman (1973)). and M”+l(O) is complete and simply connected. Kobayashi (1958). .[2]). A Kahler manifold of constant tholomorphic sectional curvature is called a complex space form. Nirenberg (1959). If c GO. and M is called a Kgihler submanifold of iii. The converse is true if c < 0. there exists no submanifold M’ of fi such that M = 8M’. L. I. then M is isoparametric.+(Ii?. A compact KChler submanifold M of a Kahler manifold @ can never be homologous to 0. then f does not exist (J. c < 0.121).

H. Hano (1975)). and M is called a totally real submanifold of iii. More holds (Chern and R. (M) and P-l. then each of the following is sufficient for an n-dimensional compact Klhler submanifold to be totally geodesic: (1) holomorphic sectional curvature > l/2 (A. Naito. Cheng (198 1)). and others [ 121. the total curvature of the immersion f is defined as If?Ql 1 voU~“-‘) s v.. Let o and R be the tvolume elements of v. Kuiper (1958)). More generally. and let fi : v1 (M)+Smel be the parallel translation.(C) that is a complete intersection (J. Verstraelen (1984)). As a generalization of the ttotal curvature for a space curve. Takagi.. 1958).(C) satisfies Min.(M) at each point x E M is called a totally real immersion.(C) is 1.g) satisfying JT.. and others.) is the only THermitian symmetric space that can be immersed in Cm (resp. K. (2) sectional curvature > l/8 (A. I.1359 365 0 Riemannian Submanifolds into complex space forms were completely determined by Calabi [lo] and by H. For example. S”-’ the unit sphere centered at 0 E Em. respectively. C” (resp. Sakamot0 [14]). and KHhler immersions of Hermitian symmetric spaces into P. then iqf$f)=/?(M)>2 An isometric immersion of a Riemannian manifold (M. then M is homeomorphic to S” (Chern and Lashof (1958)). Y. S. Total Curvature Let J’: M + E” be an isometric immersion of an n-dimensional compact Riemannian manifold M into a Euclidean space. tExotic spheres do not have minimum immersions (Kuiper (1958)). Nomizu W’W. Q” is the only Einstein-KHhler submanifold of P.(C) into Pntn+3j. 0. A totally geodesic submanifold P. S’ x S’ in P2(C) and an immersion P.(C)is the only Einstein-Kghler hypersurface of a complex space form that is not totally geodesic (B. The result remains true even if “Einstein” is replaced by “parallel Ricci tensor” (Takahashi (1967)). Submanifolds with V’a = 0 in symmetric spaces have been studied by Ferus. S.2(C)) W.(C) were precisely studied by Nakagawa. Totally Real Submanifolds If /l(M) is the least number of critical a tMorse function on M.. If the image of each geodesic of M is contained in a 2dimensional totally geodesic submanifold of M”(c). Yau (1975). M. Takeuchi. Q~={[zil~P~+~(C)ICzf=O} inP.. Tsukada. (4) embedded and scalar curvature > n* (J. Hong (1973). Chern (1967)). Ros (1985)). then f is either a totally geodesic immersion. N. Submanifolds with the above property are closely related to isotropic submanifolds with V’o = 0.+. These results generalize theorems for space curves by W. Integral theorems and pinching problems with respect to various curvatures for compact KBhler submanifolds of P. An isometric immersion f which attains inf(f) is called a minimum immersion or a tight immersion. generally. and others. Then for each 5 E vi(M). Fary (1949).p(x)=O or 2n (K. T. The index of relative nullity p(x) of an ndimensional complete KHhler submanifold M of P. Let vi(M) be the unit normal bundle. Ros and L. A KBhler submanifold of a complete and simply connected complex space form with the same property as above is either a totally geodesic submanifold or a Veronese submanifold (a KHhler immersion of P. Smyth (1967).. Takeuchi. Ogiue (1976). Tanno (1973)... 0.(M) c v. the last case occurs only when c > 0 (S. f?fi = (det A@ holds. 7(f) = 2 if and only if f is an embedding and f(M) is a convex hypersurface of some En+’ in E” (Chern and Lashof (1958)). D.. Fenchel(1929). Milnor (1950).(C). Lashof (1957. S. J. tR- .J as a Klhler submanifold (Nakagawa and Takagi [ 1 I]).(R) in P.(C) give typical +P n(n+Z)(C) defined by [zi]+[zizj] examples of totally real submanifolds... let f: M + M”(c) be an isometric immersion of M into a complete and simply connected space form M”(c). a totally umbilical immersion or a minimal immersion of a compact symmetric space of rank 1 by harmonic functions of degree 2. If 7(f) < 3. J. if the holomorphic sectional curvature of P.(C) have been studied by K. Nakagawa and K. K.. Sakane. Kghler immersions of homogeneous Klhler manifolds into P. J. Little (1976). L.(M) points of ldetA<lo. and others. g) into a KBhler manifold (a. Abe (1973)). H.(C) were determined by Takeuchi (1978). Ogiue. (3) Ricci curvature >n/2 [12]. Besides linear subspaces. Submanifolds with Planar Geodesics A surface in E3 whose geodesics are all plane curves is (a part of) a plane or sphere.+. S.

Y. A. 515-559. TBhoku Math. Takahashi.f= 0. F. Chen [3]). Chern. TBhoku Math. by definition. [lo] E. and S. The +Riemann-Roth theorem (abbreviation: R. 27 (1975). Takagi. Ogiue. Trans. Math. together with the constant . in particular. Ann. theorem will provide a description of dim. Geometry of submanifolds. Takeuchi. A characteristic property of spheres.6) Riemann-Roth A.. [ 121 K. 745-755. (2) 58 (1953).g + 1 + r(D). V.. 22 (1970).bj>O). Springer. [9] T. [ 141 K. The set of meromorphic functions . Homogeneity and some curvature conditions for hypersurfaces. Type Keeping the notation given in Section A. 303-315. R. [S] S. Spivak. It satisfies JM 1 llbll”*13vol(S”). Amer. [6] A. Pura Appl. TBhoku Math. B. J. Kobayashi. R. where r(D) is a nonnegative integer determined by D. J. Riemannian 1360 Submanifolds [ 131 B. [2] M.. D. Math. [7] H. Nomizu. 21 (1969). Springer. Let X be a compact +Riemann surface of tgenus g. Ann. Singer. 44 (1977). Y. Hirzebruch’s Theorem of R. Sakamoto. 405-425. Japan. and let D = C mi Pi be a tdivisor on X. 25-56. IV. If K is the tcanonical divisor of X. Interscience. S. When B is determined by a divisor D of X. forms a finite-dimensional linear space L(D) over C. Foundations of differential geometry II. @(B))z L(D). Isometric imbedding of complex manifolds. We denote by deg D the degree of D. References [l] S. Advances in Math. On locally symmetric Klhler submanifolds in a complex projective space. Totale Absolutkriimmung in Differentialgeometrie und Topologie. where the Qi are the zeros of order ai and the Rj are the poles of order bj. 193 (1974). The R. Nagano. Math. Grothendieck. 1973. Functional Analysis and Related Fields. Minimal submanifolds of a sphere with second fundamental form of constant length. 363-388. 8(B)).f such that (f) + D is positive. and others. B be a +complex line bundle over X.. M. Willmore (1968). The equality holds if and only if f is totally umbilical (T. Totally geodesic submanifolds of symmetric spaces I. theorem) is one of the most significant results in the classical theory of ‘algebraic functions of one variable. On totally real submanifolds. II. Sot. and B(B) be the +sheaf of germs of holomorphic cross sections of B. then r(D) = dim L(K -D) (.. 1969. Atiyah and I.. R. Math. 257-266. . 19(B)) in terms of quantities relating to the properties of X and B.9 Algebraic Curves C. 7-55. J. [ 1 l] H. various theorems of Riemann-Roth type have been obtained. l23.. we have H”(X. 395-410. 366 (Vlll. General Remarks Theorems where S” is the n-dimensional unit sphere. Kobayashi and K..15 Algebraic Surfaces D. Y. doCarmo. 1970. Let X be a compact complex manifold. H’(X. Sot. Ogiue. of an n-dimensional compact Riemannian manifold into a Euclidean space is. Dekker. Differential geometry of Kahler submanifolds. 58 (1962). The divisor D is said to be positive if D # 0 and mi > 0 for all i. [4] B. 28 (1976). theorem asserts that dim L(D) = deg D . 638-667. which is defined to be C mi. 11 Algebraic Functions D). Japan. The total mean curvature of an isometric immersion f: M-E. 45 (1978). M. 73-l 14. Sot. 13 (1974). Planar geodesic immersions. Hence a desirable generalization of the R. 28 (1976). Duke Math. A comprehensive introduction to differential geometry III. 1968. [ 151 D. Chen and K. spaces have minimum immersions and. [3] B. Kodaira. Ryan. F. we put x(X.. On some types of isoparametric hypersurfaces I. Mat. A nonzero tmeromorphic function f on X determines a divisor (f)=CaiQi-CbjRj(ai. theorem for algebraic surfaces . II. J. c’(B)) = C. III. R. Chen and T. Following this idea. Publish or Perish. Nakagawa and R. J.( -1)4 dim H4(X. J. Ferus. Homogeneous hypersurfaces in spaces of constant curvature. J. (For the R. 1975. Hirzebruch. Chen. JM i llbll"* 1. [S] P. Calabi. M. 29 (1977).) Generalizations of this important theorem to the case of higher-dimensional compact tcomplex manifolds were obtained by K. R.. a minimum immersion of a symmetric R-space is a tminimal immersion into a hypersphere (Kobayashi and Takeuchi (1968)). Alexandrov.365 Ref. Ozeki and M.

C. of X and express it formally as the product I-& (1 + yi). .e-?i). if the sequence O-tF-+S+F’+O is exact. the intersection number (FI F2) is defined to be cl(F. A similar construction for the (equivalence classes of the) tcoherent algebraic sheaves instead of the vector bundles yields another Abelian group K’(X). + d. Hirzebruch’s theorem of R. then for complex line bundles F.. (The multiplication CA(F).) T(X. which is the quotient of the free Abelian group generated by the equivalence classes of algebraic vector bundles over X modulo the subgroup generated by the elements of the form F-F’F”. X is a closed subvariety of an open set in a projective space (over an algebraically closed ground field). (aqf)F is coherent.. we consider the formal expression of the Chern class of the vector bundle F as 1 + d. where q is the dimension of the fiber of F. that is. can be expressed as a polynomial in ci. 9’) + x(X. It can be shown that c(F) and ch(F) are determined by the image of F in K(X). O(F)) K is the canonical line bundle of X and cz(X) denotes the value at X of the 2nd Chern class of X. we have x(X.+c. T(X) can be expanded as a formal power series in the yi. its Chern class c(F)=l+c. =(F’)/2-(KF)/2+((K2)+c2(X))/12. i.5] (.9) =x(X. and we have 45 + 4 = c(&(?b CM5 + f7)= cm3 + CW? &(&)=&(&h(q) (<. R. The R. F) is called the Todd characteristic with respect to F. If an tanalytic vector bundle F depends continuously on auxiliary parameters. R.) We define h(F) as before. a compact complex manifold of dimension 2. Atiyah and Singer developed a theory on indices of elliptic differential operators on a compact orientable differentiable manifold and obtained a general result that includes the proof of Hirzebruch’s theorem for an arbitrary compact complex manifold [4. Serre [S]. If F satisfies the conditions for the vanishing theorem of cohomologies. (. F’. y.e.237 K-Theory H). For a vector bundle F. Consider the group K(X). In 1963. . type asserts that x(X.. F = [II] (the line bundle determined by the divisor D). being symmetric in the yi. The following is a description of his idea as reformulated by A. theorem for surfaces is a powerful tool for the study of compact complex surfaces. D.(F)+. . F) to be the value of ch(F) T(X) at the tfundamental cycle X. + c. Consider the formal expression T(X) = I$=. Type 1111.!Sqf)P is the qth tdirect image of 9 under f: (Since f is proper. then x(X.4(X) is the ring of the rational equivalence classes of algebraic cycles on X. = n&t (1 + Sj). + . we can define x(X. If we have a tproper morphism f: Y-X between quasiprojective algebraic varieties Y and X. and thus determines a cohomology class of X. F” are classes of bundles such that there exists an exact sequence O+F’+F+F”+O. and each homogeneous term. Hirzebruch’s formula yields the classical R. 9”). B(F))= T(X.. We call h(F) the tChern character of F (. It can be shown that K(X) is isomorphic to K’(X) by the correspondence F-+0(F) (= the sheaf of germs of regular cross sections of F). the Euler number of X. and K(X) has the structure of a ring with multiplication induced by the tensor product. Grothendieck’s Theorem of R. The Noether formula xx =((K’) + c.(F) is the class of a cycles of codimension i. R. The former is defined by taking the induced vector bundle and the latter by the correspondence F-+X( -l)q(aqf)F. F).) Between Chow rings we have homomor- . Similarly. and c.16 Algebraic Varieties) over a ground field of arbitrary characteristic.. 9) has simple properties in various respects. theorem. c. where where S is a coherent algebraic sheaf on Y and (. 9) using the same formula (replacing O(B) by F). Namely. . For example.. The quantity x(X. Thus the ith Chern class ci is expressed as the ith elementary symmetric function of yi. Let X be a projective algebraic manifold of complex dimension n. R. Lo(F)) remains constant. and F. T(X. where F. T(X) is formal.qsK(X)). R.. if 9 is an arbitrary tcoherent analytic sheaf on X. The R.1361 366D Riemann-Roth Theorems Generally. yi/ (1 . R. Bore1 and J. . we have homomorphismsf!:K(X)-tK(Y) and $:K(Y)+K(X). In particular.(X))/12 follows from the above identity. Addition in K(X) is induced by the twhitney sum of the bundles.) U c1(F2)[X]. We consider the tChern class c = 1 + ci + . 0(F)) Grothendieck took an entirely new point of view in generalizing Hirzebruch’s theorem.-P. and define T(X.(F)(q= the dimension of the fiber) is defined as an element of the tChow ring A(X) with appropriate properties. Theorem for Surfaces If X is a compact complex surface.237 KTheory B). . . theorem for a line bundle F on X is stated as follows: x(X. F) is determined by the term of dimension 2n alone. the formula gives an estimate for dimH’(X. We consider a nonsingular quasiprojective algebraic variety X (. We put ch(F) = &i e% The formal series ch(F) is also an element of the cohomology ring of X whose (v + 1)st term consists of a 2v-dimensional cohomology class. when n = 1. R.

Sot. 423-432.f*:A(X)+A(Y) and. [ 151 J. (2) 100 (1974). then <@~EK’(X) and ~(<@q)= ch(<) (7(q)). This theorem is obtained by taking X to be a single point. I9 (1973). Amer.f*II). Singer. Patodi. the theorem gives Hirzebruch’s theorem for algebraic bundles. Math. The index of elliptic operators. Segre. there exist two values of w satisfying g(z) = w. I 1 I .152..e (0/2+n)i.. This is called Grothendieck’s theorem of R. The geometrical invariants of algebraic loci. theorem for X formulated by P. Math.O < 0 <27-c). Ann. Inst. 127-138. Amer. (2) 71 (1961).(X) such that (I) if <EK(X) and ~EK’(X). M. Math. B. Proc. phisms. 422-433. 279-330. L. Grothendieck. M. defined by taking inverse and direct images of cycles. Math. W. now named after him. Lond. Then g(0) = 0 and g( co) = cc. R. 229-234. [ 121 K. F. [ 131 R.. 45 (1975). R. Note that the naturality of 7 means that for any f: X+ Y and any 4 E K’(X). [9] W.:A(Y)+A(X). lnventiones Math. Fulton and R. R. Ann.237 KTheory). 225. Singer. no. Spectral asymmetry and Riemann geometry. the R.. Topological methods in algebraic geometry. and its inverse function w = g(z) = J. Sot. Rational equivalence on singular varieties. theorem can be expressed as follows: R(Y) is mapped into R(X) by a proper morphism . then f. 367 (XI. and V. Now consider how we should modify the complex z-plane so that we can obtain a single-valued function on the modified surface representing the same relationship . HES. 1966. the theorem asserts that if X and Y are quasiprojective and . France. [2] M. [7] P.f: Y+ X is a proper morphism. If X consists of a single point. G. [S] A. [I41 B. Mat. Atiyah. Math. Bull Sot. Ann. MacPherson. Sot. Chern classes of singular varieties.. With this notation. La th&or&me de Riemann-Roth. London Math. On compact analytic surfaces I. and I. Atiyah and F. Theorem for Singular Varieties Let X be a projective variety over C and let H. W. One of the remarkable results is that an element of R( Y) takes an integral value at the fundamental cycle. using the notions developed by Grothendieck.366 E Riemann-Roth I362 Theorems Singer. F. where w varies in the complex plane. MacPherson. Fulton. Bull. Hirzebruch.. 5 (1973). A. Serre. 45 (1975). Ann. this result covers Hirzebruch’s theorem (. Intersecting cycles on an algebraic variety. III by Atiyah and Singer). z(Io. Math. K. Note that K(X) may not agree with K’(X) when X is singular. Atiyah. Generalizations to +almost complex manifolds and tdifferentiable manifolds were made by Atiyah and Hirzebruch in this latter form [I].(<))T(X). Springer. Todd. H’(X)) denote the singular homology (resp.(ch(t) T( Y)) = ch(f. Kodaira. Real and Complex Singularities. The R. Illusie..f: Y-X. General Remarks References [I] M. 69 (1963). cohomology) group with rational coefficients.-P. Springer. [S] M. Bore1 and J. Berthelot.(X) (resp. (2) 87 (1968). whereas if z # 0. II by Atiyah and Segal. (2) whenever X is nonsingular. Sijthoff & Noordhoff. [4] M. R. F. SGA 6. l-128. Publ. MacPherson. Math.f. 1976. no. m. [IO] W. 484-604 (I by Atiyah and Riemann considered certain surfaces. [3] M. (2) 43 (1937). et al.59). E. HES. 276-281. and R. third edition. obtained by modifying in a suitable manner the domains of definition of multiple-valued ianalytic functions on the complex plane in order to obtain single-valued functions defined on the surfaces. [6] P. Bott. Hirzebruch. Segal. 97-136. Riemann-Roth for singular varieties. Since algebraic and analytic theories of coherent sheaves on a complex projective space are isomorphic. Fulton. On the heat equation and the index theorem. Baum. (4) 35 (1953). By setting z = rei0 (r > 0. Bull. type. and R. 1971. Patodi. Thus. f*7(9)=7(. consider the function z =f(w) = w2. the corresponding two values of w are w1 = ~e(0i2)’ and w2 = J. K. M. Baum. Math. [I I] F. RiemannRoth theorems for differentiable manifolds.) = T(X)(X). Inst. MacPherson [6] says that there exists a unique natural transformation ?:K’(X)+H. Fulton. Bull. Nuovi metodi e risultati nella geometria sulle varieta algebriche. Atiyah and I. V. The subgroup R(Y) of A(Y) given by R(Y) = { ch(Q T(Y) I< E K( Y)} is called the RiemannRoth group of Y. 65 (19. Thkorie des intersections et thkorirme de Riemann-Roth. 101-145. R. 147-167. 179-197. F. R. F.. Atiyah.1 2) Riemann Surfaces A. Sot. Singer. 86 (1958). The index of elliptic operators on compact manifolds. For example. Lecture notes in math. Math. and I. Publ.

1. and 211. Therefore R is a tlocally compact metric space. suppose that for mappings between plane regions we are given a property ‘$3 that is invariant under conformal mappings. and R. tiu.) has the property ‘p for every pair of analytic neighborhoods VJ. with R its base space and (2X) its conformal structure. there are two points zi and z2 in rrr and rc2. Thus such a mapping formal. 1: D+D). h2) in R. Then the function w = & becomes single-valued on R. A pair (R. I+&). U 91u. its basic surface and T its projection. H. R. onto another Riemann surface R.( U. respectively. If there exists a one-to-one conformal mapping of a Riemann surface R.. Often R is called simply a covering surface of R. by definition. For (U. are equivalent.wojea U. are said to be conformally equivalent. It is not possible to define curve lengths on R compatible with the conformal structure. and a continuous open mapping T of R into. tanalytic. tquasiconformal. &. etc.$u2)~21 with I/= U. R is considered to be a real 2-dimensional space with a tconformal connection. Delete the nonnegative real axes from rci and rc2 and patch them crosswise along the slits (Fig. Two such Riemann surfaces are sometimes identified with each other. respectively. and situations at the origin and the point at infinity are as indicated in Fig. Suppose. Two such sets 2X. i. # 0. It is also orientable (. then R..e. and rcz be two copies of the complex plane.. 0 $. that there are two surfaces R and R. with the same coordinate z. Covering Surfaces One of the main themes of the theory of functions is the study of analytic mappings of a Riemann surface R into another Riemann surface R. respectively.72 Complex Manifolds). such that the inverse image of a point in R. 1 Working from the idea illustrated by this example. (A Riemann surface in this sense is sometimes called an abstract Riemann surface..410 Surfaces).1363 367 B Riemann Surfaces between z and w. the theory of covering surfaces of Riemann surfaces. T) a covering surface with R.... The equivalence class (2I) of such Iu is called a conformal structure (analytic structure or complex structure) on R (. Let rc. R. For z#O. n U. tholomorphic. $c) (or sometimes U itself) is called an analytic neighborhood. under T is an tisolated set in R. Weyl and T. (U. 1925).(V). (2l)) consisting of a connected Hausdorff space R and a conformal structure (‘?I) is called a Riemann surface. and wr and w2 are holomorphic functions of zr and z2. (U. A point p0 with p.. and R. of U onto plane regions satisfying the following two conditions: (i) R= UcU. where U.) It is a complex manifold of tcomplex dimension 1. (ii) for each (U. Let w1 and w2 correspond to zi and z2. In the remainder of this article we call R itself a Riemann surface. by condition (ii) it can be deduced that R satisfies the tsecond countability axiom and consequently is a tsurface and admits a tsimplicial decomposition (T. harmonic. It is customary in the theory of functions to call R closed if it is compact and open otherwise.) nd W. The surface R thus obtained is locally homeomorphic to the complex plane except for the origin and the point at infinity.. B. and $c is called a local uniformizing parameter (or simply a local parameter). l/q}. $c) of open sets U in a tconnected tHausdorff space R and topological mappings $. but since angles can be defined. o To I+&. holomorphic. Fig.(V) onto a corresponding one of tic. Rado. co. cp} and {U. (U.: gives an (orientationpreserving) tconformal mapping of each connected component of ti”.. A Riemann sphere is also considered to be a closed Riemann surface whose analytic neighborhoods are given by {U.) into Il/u. More generally.. Moreover.’ of I. onto another Riemann surface R. h. if QII. = . respecT may be contively. The surface R is called the Riemann surface determined by w = $.).&. A mapping T of a Riemann surface R.) is the domain corresponding to {1~~<2]({121>1/2}U{co})underthestereographic projection cp. is said to have the property ‘$3 if the mapping $“. 1). or harmonic on R if f o $6’ is tmeromorphic. The usual definition nowadays is as follows: Let ‘$l be a set of pairs (U. Then T is called an inner transformation in the sense of Stoilow and (R. A function f on a Riemann surface is said to be meromorphic. in general. Rado gave rigorous definitions of Riemann surfaces. or tharmonic in the usual sense on tie(U) for every analytic neighborhood (U. From condition (i) it follows that a Riemann surface R is a real 2-dimensional ttopological manifold. $c) in 2lu(%). (U. also satisfies conditions (i) and (ii). A plane region D is considered an open Riemann surface with the conformal structure ‘u =(D..

T can be regarded as an analytic mapping of R onto R. Conversely.. -cc <@<co) such that a+rei8= T(I. p) 2 p.. in R.(U.. w*) (z* = P(t). The universal covering surface exists for every R. is the sphere. A tsimply connected surface R” that is an unramilied unbounded covering surface of R. then p is called a branch point. and if R is an n-sheeted covering surface of R. Historically. by giving R a conformal structure in a natural manner. is isomorphic to the tfundamental group (i. The group of all covering transformations of a universal covering surface of R. If R and R. it can be deduced from the existence theorem of an- C.(t). With respect to a suitable metric on R. n the multiplicity. then R is said to have a relative boundary above the endpoint of C.*.-‘(r. (or the lift of C. set z=F(p)=P(O). is always constant. R. T(p) is called the projection of p. for open Riemann surfaces. is particularly important.. This correspondence generally gives rise to a multiple-valued analytic function w =f(z). where the branch points of R are counted with their multiplicities. Then two meromorphic functions z = F(p) and w = G(p) . w* = Q(t)) in f..~. sharing the initial point with C.. If there exists a topological mapping Ic. n is called the number of sheets of R over R.hL. be either compact surfaces with simplicial decompositions or their closed subregions with boundaries consisting of lsimplexes and vertices such that T preserves simplicial decompositions. T) whose basic surface R. Behnke and K. and a point p in R lying above the initial point of C. w= G(p)=Q(O). In this case. Ahlfors’s theory of covering surfaces. respectively. in contrast. w* = Q(t)) in the wider sense containing po.. Then the number of points on R that lie above each point of R. T) not only from the topological viewpoint but also from the metrical one.. If n > 1. which states that an open Riemann surface is a Stein manifold. T) and the projection Tare said to be unramified.~.. then R is said to have a logarithmic branch point above a.. =(z.. such that $(U)= {z 1lzl< l}.V. if R. This fact had long been known for closedRiemann surfaces. then the covering surface (R.. Let R and R. until the two notions were proved identical.. or R is said to be n-sheeted over R. onto itself such that To S = T is called a covering transformation. and --p and -pO the tEuler characteristics of R and R. w*) (z* = P(t). then its covering surface is a Riemann surface. and p is said to lie above pO. with the projection T: R+R. A topological mapping S of an unramilied unbounded covering surface R of R.367 C Riemann 1364 Surfaces alytic functions proved by H. is a Riemann surface. In particular. Uniformization Suppose that we are given a correspondence between the z-plane and w-plane determined by a tfunction element p. and denote by R.. This has been applied widely in various branches of mathematics. then we have the Riemann-Hurwitz relation: x = nXo . I&. Suppose that R is an unbounded covering surface of R.)=(~II~I<~}. there exist surface coordinates (U. R. ~(P)=O.@IO<r<r. A covering surface without a relative boundary is called unbounded. of R. wz) (zz = P. We use f again to mean the connected component of the set of function elements p = (z*. Suppose that R is a covering surface of the z-sphere R. Here we call the part of the boundary of R whose projection is in the interior of R.. which treats covering surfaces (R. where h > 0 is a constant determined only by R.. If there is no branch point.) starting from p.)=O.~~~ w=(t&o TOI+-‘)(z)=z”. a branch point of multiplicity n of the type defined previously is sometimes called an algebraic branch point. respectively. admits a prolongation along itself starting from p but C. The set of all branch points forms an at most countably infinite set of isolated points. the region that lies above 0 < lz al < r..e.. For a given curve C. a curve C in R with p its initial point satisfying T(C) = Co is called the prolongation along C. $) and (U. w: = QJt)). Then f is a Riemann surface.1 the degree of ramification. let S be the ratio of areas of R and R. We show how to construct a Riemann surface so that the function w =f(z) can be considered a single-valued function on it.. the l-dimensional homotopy group) of Rw In a covering surface (R. and n . the relative boundary.S . any Riemann surface can be regarded as a covering surface of the sphere. say n (< +co).(P. Stein. Then Ahlfors’s principal theorem asserts that max(O.@). L the length of the relative boundary... V being the sum of the degrees of ramification.. respectively. For a point p =(z*. onto {(r. where the analytic neighborhood of each point p is defined to be the set of elements that are direct analytic continuations of p. with the positive integer n independent of the choice of coordinate neighborhoods. are compact surfaces with tEuler characteristic x and x0.) at p and po. R. does not admit a prolongation along itself starting from p. If any proper subarc of C. is said to be a universal covering surface of R. by a Riemann surface mathematicians meant either the abstract Riemann surface or a covering surface of the sphere. including the theory of distribution of values of analytic mappings between Riemann surfaces.

Then R is said to be elliptic. In particular. J. and many detailed reuslts of the type problem obtained in the 1930s are limited mainly to the case where all branch points lie above a finite number of points on the sphere. various new families are created. Oao. f is uniformized by z = F o T(c) and w = Go T(c). or the unit disk Ill< 1. the finite plane ill < co. respectively. and hence f is uniformized by z = F o T(c) and w = Go T(c). From works . an talgebraic function f considered as a Riemann surface is always closed. then (f m.. T) off is simply connected. Suppose that there exist two meromorphic functions z = cp(c) and w = $([) on a region D in the c-plane. A s&Iicient condition for R to be of parabolic type.. Myrberg found an example of a Riemann surface of infinite genus which has a large boundary but belongs to O.” f. is called the type problem for Riemann surfaces.) and w = Q([ . “not merely devices for visualizing the manyvaluedness of analytic functions. then f is uniformized by z = F o T(c) and w = Go T(c) (Schottky’s uniformization). Therefore analytic functions in the wider sense are always uniformizable. are significant results on the type problem. The family of positive functions. The set of all Riemann surfaces R for which X(R) does not contain any function other than constants is denoted by Ox. In general. When l<l< 1.” So the problem naturally arises of how to extend various results in the theory of analytic functions of a complex variable to the theory of analytic mappings between Riemann surfaces. then f is the sphere and is thus uniformized by rational functions z = F(c) and w = G(c). In connection with this. since the universal covering surface (f m.1365 367 E Riemann Surfaces are defined on J and f can be considered as a covering surface of the z-sphere and the wsphere. OaBD. the family of Riemann surfaces on which there are no Green’s functions. then f can be uniformized by z = F(p) and w = G(p). and let z = P(< . the only soil in which the functions grow and thrive. the finite plane.4(R) n B(R) tl D(R). If the function element psO= (z*. and also OG. L. OHD. However. are considered. z and w are tautomorphic functions with respect to the group of linear transformations preserving l<l< 1. if { pF 1[ED} =A then f is said to be uniformized by z = cp([) and w = $(c). and D(R). parabolic.. it is difticult to measure the ramifications of covering surfaces. in D. and that of functions with finite Dirichlet integrals are denoted by P(R).&. Nevanlinna. . Kakutani (using quasiconformal mappings). If its tgenus g =0. then the correspondence w = f(z) determined by the function element pro is said to be locally uniformized on D by z = cp(<) and w = $([). or hyperbolic. For example. T) is conformally equivalent to I(1 < 1 or l[l< co. The family of analytic functions and the family of harmonic functions are denoted by A(R) and H(R). Kobayashi (using the so-called Kobayashi net. The idea behind Myberg’s example is often used to construct examples in classification theory. or the unit disk. f is not conformally equivalent to a plane region. f. Oas. is conformally equivalent to a region D in the c-plane. but rather an essential component of the theory . and a sufficient condition for R to be of hyperbolic type. one usually sets certain restrictions on the properties of the Riemann surfaces.g. it is conformally equivalent to the sphere l[l i co. given by Z. Denote by X(R) the totality of functions on a Riemann surface R with a certain property X. If g > 0. For example. The problem of determining the types of simply connected covering surfaces of the sphere by their structures. P.. f. e. but if an unramified unbounded covering surface (f.[. respectively. Consequently. The Nevanlinna theory of meromorphic functions stimulated this type problem. OHBD. R. w* = Q(t)) belongs to the Riemann surface f. We call f an tanalytic function in the wider sense. Usually OHB. given by S. In general. bolic (Picard’s theorem). which classifies Riemann surfaces by the existence (or nonexistence) of functions with certain properties. such as the distributions of their branch points. E. as pointed out by Weyl. Classification Theory of Riemann Surfaces D. that of bounded functions. if a simply connected covering surface does not cover three points on the sphere. it must be hyper- Riemann surfaces are. MD(R) = . The type problem had by then been extensively generalized to the following classification theory. they are telliptic functions. The Type Problem A simply connected Riemann surface R is conformally equivalent to the sphere. Thus we obtain a single-valued function w = G(p) on the Riemann surface f that can be regarded as a modification of the original function w = f (z). open Riemann surfaces can have infinite genus and are quite complicated. while if l[l< co. and others initiated the classification theory of Riemann surfaces. B(R).) be tLaurent expansions of cp and 1(1at each point &. T) off is conformally equivalent to a region D in the [-plane. So to obtain fruitful results and systematic development. w*) (z* = P(t). In particular. Sario. From these families. If an analytic function fin the wider sense. considered as a Riemann surface. respectively.

and UHDg U. boundaries of Riemann surfaces. Kuramochi. Closed Riemann surfaces are all in 0. and others. Then R. Royden. and he extended the main theorems of Nevanlinna to analytic mappings between arbitrary Riemann surfaces (. There are inclusion relations U. which is utilized to construct harmonic functions on Riemann surfaces with given singularities at their ideal boundaries. H. M.. de Possel. The proof is based on the fol- .. Virtanen. . 2. every open Riemann surface is homeomorphic to a prolongable Riemann surface (S.. and O. are invariant under quasiconformal mappings. A.. Sari0 studied the method of normal operators. Um $ On.. J. as indicated in Fig. Blaschke type. that U.~O. among the classes just mentioned. L. for every g in X(R) with fag>0 such that g-C. Tamura. A nonprolongable Riemann surface is said to be maximal. and others which are special classes of analytic mappings.62 Cluster Sets) by studying the behavior of analytic mappings at ideal boundaries. and others extended various results in the theory of cluster sets (.e.. However... Suppose that a Riemann surface R is conformally equivalent to a proper subregion R’ of another Riemann surface R.O. Closed Riemann surfaces are always maximal. Bochner).. L. Mori. K. G..O. T. and those not in O. Gunning and Narasimhan proved that there exists a holomorphic function on R whose derivative never vanishes. We call a noncompact region 0 which is the complement of a compact subset of a Riemann surface a Heins’s end. where m is the class of positive functions in HD or limits of monotone decreasing sequences of such functions. and r/. it can be seen that there are inclusion relations. it is still an open problem whether O. Analytic Mappings of Riemann Surfaces F.. I. and admitting X-minimal functions on R is denoted by U. M. Heins introduced the notions of Lindelof type.UO. Constantinescu and A.124 Distributions of Values of Functions of a Complex Variable). M. and others.$ O. Sario.. hyperbolic (or of positive boundary). parts with cycles not homologous to 0) accumulate. Kuroda. Pfluger and Royden showed that the classes 0. is said to be a prolongation of R. . Fig. Such a locally homeomorphic mersion analytic mapping is called the imof R. Heins called the minimal number (< co) of generators of the semigroup of the additive class of HP-functions that vanish continuously at the relative boundary of a Heins’s end Q the harmonic dimension of Q. Cornea and others studied Riemann surfaces in U. = O. pathological Apart from the development of the classification theory of Riemann surfaces. M. These are caused by the complexity of ideal surface of the sphere. However.. discovered by Kuramochi. are also said to be parabolic (or of null boundary). In other words. efforts have been made to extend various results in the theory of analytic functions of a complex variable to the case of analytic mappings between Riemann surfaces. Utilizing these notions. Furthermore. R is conformally equivalent to an unramified covering phenomena occur for Riemann surfaces from the viewpoint of function theory in the plane.. the classification problem for subregions was studied in detail by Parreau. M. Open Riemann surfaces in 0. Characterizations of prolongable Riemann surfaces and relationships between the various null classes mentioned in Section E and prolongations were investigated from several viewpoints by R. Kuramochi. TBki. C. Parreau.. and O. is invariant in this sense. Constantinescu and Cornea.. Prolongations As classification of Riemann Surfaces theory shows. a function f is said to be Xminimal if f is positive and contained in X(R) and there exists a constant C.& The family of Riemann surfaces R not belonging to 0.. and in particular by the complexity of the set of ideal boundary points at which handles of Riemann surfaces (i. For Riemann surfaces of finite genus. L. and U. On every open Riemann surface R there exists a nonconstant holomorphic function (Behnke and Stein). Classification theory has a very deep connection with the theory of +ideal boundaries.367 F Riemann Surfaces 1366 of Y.. Its properties were investigated by Z... Sakai. and R is prolongable. and others. The theory of tcapacities on ideal boundaries has also been developed. 0. and others.U O.. Hence it is desirable to find larger Riemann surfaces so that ideal boundary points of original surfaces that are not accumulating points of handles become interior points.. Generally. !$ O. One of the interesting results in classification theory is the fact. There is no inclusion relation between O. Ozawa... 2 From a similar point of view. Several characterizations for parabolic Riemann surfaces are known. A.. but there also exist maximal open Riemann surfaces (Rado).

Using the Dirichlet principle. differential l-forms o=udx+udy and differential 2-forms CI = c dx A dy are defined on them.. there exists the extremal horizontal (vertical) slit mapping F....e. and I. F. A differential is said to be semiexact if its period along every dividing cycle vanishes. A surface of genus 0 is said to be planar (or of planar character or schlichtartig).. etc. Iss’sa obtained a noteworthy result which established that open Riemann surfaces are also determined by their meromorphic function fields [24].(i=1.F. (iii) the total area of the slits and points is zero.. I. Abelian Integrals on Open Riemann Surfaces H.) maps R conformally to a region on C whose boundary consists of horizontal (vertical) slits and possibly points. However. then we have the orthogonal decompositions r.iw is said to be pure.e. or O. SupposethatFi=l/z+ai+ciz+. Koebe proved the following general uniformization theorem: Every planar Riemann surface R can be mapped conformally to the canonical slit regions on the extended complex plane C...) such that (i) Fi (F.)/2 I(*) > 0. then w is called a holomorphic (or analytic) differential. Nehari...105 Differentiable Manifolds). In the case of finite genus g. Virtanen. respectively. where the equality holds if and only if R belongs to O. if f is a holomorphic function. The method of orthogonal decomposition in the theory of Hilbert spaces is the main device to study this space and also its suitable subspaces and to obtain the existence theorem of harmonic and holomorphic differentials with various properties (.. The differential form w is a holomorphic differential if and only if it is closed (i. Y. A simply connected surface is planar. only those Riemann surfaces with small boundaries (i. i *rhm. and operations such as the exterior derivative dw = (au/ax . and others). ones in 0.. and set F. A differential form o satisfying dw = d * w = 0 is The systematic effort to extend the theory of Abelian integrals on closed Riemann surfaces to open Riemann surfaces was initiated by Nevanlinna in 1940. Next..(F. P... . Since coordinate transformations satisfy the Cauchy-Riemann differential equation. i r. Ahlfors proved that a Riemann surface of genus g bounded by m contours can be mapped conformally to an at most (2g + m)sheeted unbounded covering surface of the unit disk. More precisely. stands for the space {w I *w E TX} and I. let r’(T. = {WC r. Differential Forms on Riemann Surfaces Since Riemann surfaces are considered real 2dimensional differentiable manifolds of class C”. in contrast to differentiable manifolds. where *F. The differential o is called exact (or total) if o is written as dF = F. L. The structures of closed Riemann surfaces are determined by the algebraic structures of meromorphic function fields on them. are known as the space of analytic Schottky differentials and the space of differentials of harmonic measures. and F. = r.194 Harmonic Integrals). A pure differential is expressed as w =fdz.) be the Hilbert space of analytic (harmonic) differentials with finite norm. Then s = (cl c. H. Mergelyan’s theorem): Suppose that K is a compact set on R such that R -K has no relatively compact connected component. 1w is tsemiexact}. (ii) F1 and F. later a more general treatment was made possible by the discovery of the notion of semiexact differentials (K. called a harmonic differential. given a point p on R. A l-dimensional cycle C is called a dividing cycle if for any compact set in R there exists a cycle outside the compact set homologous to C. 1w is exact}. dw = 0) and pure. then o is called a meromorphic (or Abelian) differential. have remarkable properties [9]. = *r. Let R be an arbitrary open Riemann surface. the set of all measurable differentials w with IIwllz=jsR wr\*o<coformsatHilbert space with respect to the norm IIw//. Then every continuous function which is holomorphic on the interior of K can be approximated uniformly on K by a holomorphic function on R [22]. At the first stage of the development. have a simple pole at p with residue 1. Both spaces r..2)in terms of local parameter z at p. Kusunoki.1367 367 I Riemann Surfaces lowing deep result (S. and if f is a meromorphic function. For instance. It is known that s (= Ild(Fi . Ahlfors). dx + Fy dy with a globally single-valued function F..)/2 is called the span of R.. It is known too that open Riemann surfaces are determined by their rings of holomorphic functions. and one with * o = . = {o E I. it should be noted that finer orthogonal decompositions into subspaces with specific properties hold for open Riemann surfaces.. the conjugate differential *co= -vdx+udy of w can be defined...) were treated. there also exist the conformal mappings of R onto the parallel slit regions on the (g + 1)-sheeted covering surface of C (Z. Here.&lay) dx A dy and exterior product can be defined (..

and therefore f reduces to a constant by the boundary theorem of Riesz type if R has a large boundary. 267-276. Ahlfors (ed. Sommer. [ 151 L. Subalgebras of functions on a Riemann surface.. and others further generalized the Kusunoki type theorems. 174 (1967). 142 (1979). Spencer. Classification theory of Riemann surfaces. Leqons sur les principes topologiques de la thtorie des fonctions analytiques. Theorie der analytischen Funktionen einer komplexen Veranderlichen. Amer. 1964. Math.. Press. [ 191 L. H.. Trans. 24 (1950). This generalizes the results above and the GunningNarasimhan theorem. Tsuji. Oikawa.9] and the following: [ 143 C. M. Springer. Ideale Rander Riemannscher Flachen. 1913. 1962. . Lectures on Riemann surfaces. [20] L. Riemann’s period relation on R has been studied for various classes of differentials. Royden [28] and B. Shiba. Studies. [22] E. second revised edition. Functionals of finite Riemann surfaces. the analogy to the classical theory is lost completely if no restriction is posed on the differentials on R. Constantinescu and A. Comm. 299 50. Sario. Gunning. M.. V. Contribution to the theory of Riemann surfaces. For the classification of Riemann surfaces . 1368 References [ 1] H. Gunning and R. Heins. Acta Math. Introduction to Riemann surfaces. [9] L. 1960.. C. Ahlfors.} and {q. In this context the following results due to Behnke and Stein [26] are outstanding: (I) There exists an Abelian differential of the first kind on R with infinitely many given periods.). Value distribution theory. 1969.[4. Addison-Wesley. 1966.. New York Univ. Riemann surfaces. Mizumoto. Springer. (2) For two discrete sequences {p. third edition. For analytic mappings of Riemann surfaces . Principal functions.. J. Pacif. but for the case of infinitely many nonvanishing periods. 65 (1935). Yoshida. 1954. no definitive result has been obtained. 1957. Stoilow. [23] M. 1953. Pfluger. there exists a single-valued meromorphic function with zeros at pm and poles at 4. Springer. Helv. English translation. Press. Nakai. Press. H. [2l] R. [3] L. Sot. Lecture notes.6. Sario and M. Math. Addison-Wesley. [lo] M.367 Ref. Riemann surfaces. Independently. It was proved that a meromorphic differential df =du + idu is semiexact canonical if and only if du is (real) distinguished. Hence every meromorphic function .. Sario and K. Behnke and F. [ 181 L. The same is true for the theory of Abelian differentials with infinitely many singularities. Open Riemann surfaces and extremal problems on compact subregions. [4] R. [7] G. The concept of a Riemann surface. Math. Immersion of open Riemann surfaces. Van Nostrand. 1999220. It is further proved that there exists an Abelian differential with prescribed divisor and periods (Kusunoki and Sainouchi). Bers. Potential theory in modern function theory. Y. Princeton Univ. and then u is almost constant on every ideal boundary component of R (in appropriate tcompactification of R). Analytic functions with finite Dirichlet integrals on Riemann surfaces. Rodin also gave generalizations of the Riemann-Roth theorem.f such that df is (exact) canonical exists on any open Riemann surface R with finite genus. [6] A.f for which df is distinguished is almost constant on every ideal boundary component. 1966. Springer.[6. 103-108. 1970. Theorie der Riemannschen Flachen. Bishop.9] and the following: [ 173 B. Math. [ 121 M. Schiffer and D. [ 111 L. 8 (1958). C. Gauthier-Villars. 1959 (Chelsea. [ 161 M. Nevanlinna. Zur Theorie der Uberlagerungsflachen. Teubner. 89 (1958). Algebraic structure and conformal mapping. Math. Noshiro.} of points in R. 1968.Section G). Springer. [S] L. 1975). [2] S. Weyl. Sakai. 1955. Narasimhan.. Acta Math. H. 1957. Princeton Univ. 1963. V. Die Idee der Riemannschen Flache. Van Nostrand. Princeton Univ. [ 131 R. Sario. 1957. Ahlfors and L. On the other hand. Capacity functions. Maruzen. Riemann Surfaces Ahlfors defined the distinguished (complex harmonic) differentials with polar singularities and obtained in terms of them a generalization of Abel’s theorem. Springer. Rodin and L. [S] H. 1938. The Riemann-Roth theorem for a closed Riemann surface can be deduced from that for open Riemann surfaces by considering an open Riemann surface obtained from a closed surface by deleting a point. Press. Kusunoki defined. Whereas by the RiemannRoth theorem above a nonconstant meromorphic function . Springer. Cornea. 157-194. Ann. 100-134. L. and f gives a canonical parallel slit mapping of R (. Princeton Univ. Ahlfors. Ann.the semiexact canonical (meromorphic) differentials and gave in terms of them a formulation of Abel’s theorem and the Riemann-Roth theorem on R [27]. 1953. Sario and K. Uniformisierung.

an element a’ such that u’a = 1 (au’ = 1) is called a left (right) inverse element of a.6. 0). l-34. 34 (1960). (A. The Riemann-Roth theorem. (a+ b)c=ac+ bc (distributive laws). Acad. 235-258. Thus a ring A has the structure of a tleft A-module and a fright A-module. Helv. SS is denoted by S2 (similarly for S3. etc. the unity element is distinct from the zero element.67 Commutative Rings). Math. If ST= 7’S= {0}.. these operations satisfy the following four laws: (i) law of addition). The set of all invertible elements of a unitary ring forms a group under multiplication. For any subsets S and T of a ring A. if a ring has more than one element. Univ. Mem..277 Modules). b. This element is called the inverse element of a and is denoted by a-‘. 33 (1961). The inverse .. and R. On the meromorphic function field of a Stein variety.). [27] Y. [26] H.(x) = ax. Fenn. 416 Teichmiiller Spaces. Hence we often call a unitary ring simply a ring. Theory of Abelian integrals and its applications to conformal mappings. A ring is called a commutative ring if it satisfies (v) ab = ba (a. are defined. and furthermore. KY&o.1369 368 B Rings [24] H.149 Fields). L.67 Commutative Rings). (1) Two toperations. Entwicklung analytischer Functionen auf Riemannschen Fllchen. Nevanlinna. Ann. and (iv) for every pair a. I&a. 430-461.reference to 11 Algebraic Functions. {a} +S({a}S) is denoted by a+S(uS). and idempotent if S2 = S. and a nonzero element a satisfying a2 = a is called an idempotent element. Stein. b of elements of A. A nonzero unitary ring is called a tskew field (or tdivision ring) if every nonzero element is invertible. Ann. In this article we shall discuss associative rings. let S + T(ST) denote the set of elements s + t(st) (s E S. An element a # 0 of a ring A is called a zero divisor if there exists an element b # 0 such that ub = 0 or ba = 0. An element a satisfying a” = 0 for some positive integer n is called a nilpotent element. Certain nonassociative rings are important.) 32 (1959). a skew field that satisfies the commutative law is called a commutative field or simply a field (. then A is a tsemigroup with the identity element 0 with respect to this operation. t E T). Also . Math. Sci. There exists a left (right) inverse element of a if and only if A is generated by a as a left (right) A-module. An element eE A is called a unity element (identity element or unit element) of A if it satisfies ae = ea = a (a E A). Math. then they coincide and are uniquely determined by a. Since the operations L. (iii) u(b+c)=ab+uc. Most of the important examples of rings are unitary. it is unique and is often denoted by 1. Behnke and K. b of elements of A. Such a ring is called a zero ring. b of A to elements a + b and ub of A.9] and the following: [25] R. Math. defined by L. For an element a of a unitary ring A. Ann. the ring A is also an A-A-bimodule (. an example is tLie algebra. Coll. Each element a of a ring A determines operations L. be A) (commutative law for multiplication) (. then 0 is the unity element of the ring.b)+aob by setting aob=a+b-ub. called addition and multiplication (the ring operations). In particular. if we define a new operation (u. (2) 83 (1966). (An algebra is a ring having a tground ring. S4. Quadratisch integrierbare Differentiale auf einer Riemannschen Mannigfaltigkeit. R. (ab)c=u(bc) (associative laws). (2) For arbitrary elements a. 37-51. [28] H.9) Rings A. which send an arbitrary pair of elements a. However. commute for every pair a. A ring with unity element is called a unitary ring. An idempotent element is said to be primitive if it cannot be represented as the sum of two orthogonal idempotent elements. Further Definitions 368 (111. An element that has an inverse element is called an invertible element (regular element or unit). there exists a unique element c of A such that a + c = b.. Thus a ring A is an tAbelian group under addition. Furthermore. 34-46. 1 (1941). Comment. c of A. Sci. Elements a and b of a ring are said to be orthogonal if ub = bu = 0. The identity element of A under addition is called the zero element and is denoted by 0. 120 (1949).) B. In a general ring A. and R.[4. If a ring has only one member (namely. For the generalization of algebraic functions and Abelian integrals . A subset S of a ring is said to be nilpotent if S” = 0 for some positive integer n. Definition A nonempty set A is called a ring if the following conditions are satisfied. If there exist both a right inverse and a left inverse of a. Royden. Kusunoki. It satisfies the equation a0 = Ou = 0 (a E A).(x) = xu (x E A). (ii) a + b = b + a (commutative (a+ b)+c=u+(b+c). If A has such a unity element. 429-433. A commutative unitary ring having more than one element is called an integral domain if it has no zero divisors (. then subsets S and Tare said to be orthogonal.

Section F). Examples (1) Rings of numbers. When condition (ii) for a homomorphism is replaced by (ii’) f(ab) =f(b)f(u) (a. the real number field R. An element a of a unitary ring is quasi-invertible if and only if the element 1 -a is invertible. the commuter of A itself is called the center of A. then the mapping x+axa-’ (x E A) is an automorphism of A. The identity mapping 1. The set &K(M) of tendomorphisms of a +module M over a ring K is in general a noncommutative ring. Homomorphisms A mapping 1‘: A +B of a ring A into a ring B satisfying conditions (i) f(a + h) =f(a) +f(b) and (ii) f(ub)=f(a)f(b) (a.. The smallest subring containing a subset T of a ring A is called the subring generated by T.Section E) of R’. There exists only one homomorphism of any ring onto the zero ring. and f is called an anti-isomorphism. 277 Modules). For unitary rings A and B. then the inverse mapping f-’ : B-rA is also a homomorphism. and 439 Valuations. X. Antiendomorphisms and antiautomorphisms are defined similarly. In particular. If a homomorphism f is tbijective. and the equivalence relation R in A defined by f(uRbof(u)=f(b)) is compatible with the ring operations of A. If a is an invertible element of a unitary ring A.256 Linear Spaces. represented by a and b. If we deal only with unitary rings and unitary homomorphisms. . The set K [X. A composite of homomorphisms is also a homomorphism.]] of tformal power series in indeterminates Xi. a homomorphism f: A+ B is said to be unitary if it maps the unity element of A to the unity element of B. and let I be an interval of R. Thus the ring operations of S are the restrictions of those of A. The set of elements that commute with every element of T forms a subring and is called the commuter (or centralizer) of T. the set C’(I) of functions that are r-times continuously differentiable. A homomorphism of a ring A into itself is called an endomorphism. Then the definition of factor ring implies that c(+ /?(G$) is the equivalence class represented by a + b(ab). bE A). then the image f(A) is a subring of B. . Then the set C’(1) of continuous functions. Every ring A has two trivial factor rings. and the set C”(l) of analytic functions on I are subrings (. E. . 370 Rings of Power Series). X. and in this case f is called an isomorphism. is a family of rings.Section G). In this sense there exists a unique hom- omorphism of the ring Z of rational integers into an arbitrary unitary ring.e.. and Direct Products A subset S of a ring A is called a subring of A if a ring structure is given on S and the canonical injection S+ A is a ring homomorphism..] of polynomials and the set K [ [X. if M is a finite-dimensional +linear space over a field K. then a subring S necessarily contains the unity element of A. X. By a homomorphism. and an isomorphism of A onto itself is called an automorphism of A. of a ring A is an isomorphism. In particular. aRu’ and bRb’ imply (a + b)R(u’ + b’) and (ab) R(a’b’)). . If A has no nontrivial factor rings.. Subrings. with coefficients in a commutative ring K are commutative rings (. called an inner automorphism. The set K’ of functions defined on a set I and taking values in a ring K forms a commutative ring under pointwise addition and multiplication. a homomorphism (isomorphism) of rings is often called a ring homomorphism (ring isomorphism). if an antihomomorphism f is bijective. and the complex number field C are familiar examples (. 284 Noetherian Rings. namely. then A is called a quasisimple ring (. . and an element that has a quasi-inverse element is called a quasiinvertible element (or quasiregular element). the Cartesian . A tquotient set A/R of a ring A by an equivalence relation R is called a factor ring (quotient ring) of A if a ring structure is given on A/R and the canonical surjection A-+ AIR is a ring homomorphism. (5) For other examples . D. A itself and the zero ring 0.368 C Rings element under this operation is called the quasi-inverse element. Factor Rings. If f: A-t i? is a ring homomorphism. a mapping satisfying (i) and (ii’) is called an antihomomorphism. In particular. The ring Z of rational integers.14 Algebraic Number Fields. 257 Local Fields). This is the case if and only if the equivalence relation R is compatible with the ring operations (i. . 1370 C. 67 Commutative Rings. then the inverse mapping f -’ is also an antihomomorphism. Thus / induces an isomorphism A/R df(A) (. More precisely. (4) Endomorphism rings of modules. then &JM) can be identified with a +full matrix ring (.29 Associative Algebras. bs A) is called a homomorphism. If { AijiE. 36 Banach Algebras. (2) Rings of functions. Let LY and /J’ be elements of the factor ring A/R. (3) Rings of expressions.369 Rings of Polynomials. a unitary homomorphism is usually meant. In particular. respectively. the rational number field Q. let K = R.

. . quasisimple. then each Ji is a ring with unity element e.. Thus A is a simple ring if and only if A is a nonzero.. is the direct sum of left ideals and 1= e. The mapping pi:A-+Ai that assigns to each (a. For an ideal J of a ring A. Conversely. J is a ring (however. . . + Ae.e. In particular. Jn are two-sided ideals. if A = J. Any ideal of A is the direct sum of a finite number of simple components of A. A left (right) ideal of a ring A is said to be maximal if it is not equal to A and is properly contained in no left (right) ideal of A other than A. Ideals A subset of a ring A is called a left (right) ideal of A if it is a submodule of the left (right) Amodule of A (. is called a canonical homomorphism. . + . . A ring A is called a left (right) Artinian ring if it is tArtinian as a left (right) A-module (i. Under the operations induced from A.277 Modules).e. and by a natural correspondence.+. are orthogonal idempotent elements. . If A is a semisimple ring. if A satisfies the tmaximal condition for left (right) ideals of A). + J. then 1 -e and e are orthogonal idempotent elements. that is. In other words. left (right) Artinian ring. A ring A is called a left (right) Noetherian ring if it is tNoetherian as a left (right) A-module (i. or A=CZ. . the maximal and minimal conditions for left (right) ideals are independent.. The property of being Artinian or Noetherian is inherited by quotient rings and the direct product of a finite family of rings. unitary. and the equivalence relation defined by J coincides with R. If f: A+B is a ring homomorphism. If it is a field. .. the ring A is isomorphic to the direct product nyZ1 Ji.. given an equivalence relation R that is compatible with the operations of A.1371 368 G Rings to 1.then A=Ae. Quotient rings of a semisimple ring and the direct product of a finite number of semisimple rings are also semisimple. but for unitary rings.. Peirce’s right decomposition is defined similarly. Every module over a semisimple ring is also semisimple. A semisimple ring is called a simple ring if it is nonzero and has no proper ideals except {0}. . C.. A left ideal J of A can be expressed as J = Ae for some idempotent element e if and only if there exists a left ideal J’ such that A = J + J’ is a direct sum decomposition. .. called a simple component of A.Ji. and is denoted by A= @. Conversely. . if A is a quasisimple ring. A is called the direct sum of ideals J. then e. It S is a subring and J is an ideal of a ring A. + A. and f induces an isomorphism A/J+f(A). . Then R is an equivalence relation that is compatible with the operations of A. (e. a left (right) ideal of A is said to be minimal if it is nonzero and properly contains no nonzero left (right) ideal of A. then S + J is a subring of A and S fl J is an ideal of S. in this case A is called a semisimple ring (. J. if e. . If A is commutative. Semisimple Rings The statement that a unitary ring A is tsemisimple as a left A-module is equivalent to the statement that A is semisimple as a right Amodule. if J i. . A semisimple ring is left (right) Artinian and Noetherian.e. the equivalence class of 0 forms an ideal J of A.) its ith component a.. left and right are omitted in these definitions. More generally. Furthermore. + e. This is called Peirce’s left decomposition.... This ring is called the direct product of the family of rings {Ai}i. Each equivalence class is called a residue class modulo J. . it is called a residue (class) field.=‘=. and Ae (eA) is minimal if and only if e is primitive.) + (bi) = (ai + bi) and (ai) =(aibi). then the tkernel off as a homomorphism of additive groups forms an ideal J of A. is the direct sum of left ideals. In this case. . are orthogonal idempotent elements whose sum is equal G. and the quotient ring AIR is denoted by A/J and called the residue (class) ring (or factor ring) modulo J. Akizuki. Ji. A subset of A is called a two-sided ideal or simply an ideal of A if it is a left and right ideal. there exists a unique homomorphismf:B+A such that fi=pioffor each i. the natural homomorphism S+ (S + J)/J induces an isomorphism S/S ll J+ (S + J)/J (isomorphism theorem). . + . For general rings. but not necessarily by subrings. Hopkins). In particular. if A satisfies the tminimal condition for left (right) ideals of A). . J is not necessarily unitary). + . . If e is an idempotent element of a unitary ring A. . Similarly. . For any set of homomorphisms h:B+Ai (iEI). then it has only a finite number of minimal ideals A.. and A = Ae + A( 1 -e) is the direct sum of left ideals.E Ji) is the corresponding decomposition of the unity element. and A is expressible as the direct sum A = A. e. left (right) Artinian rings are necessarily left (right) Noetherian (Y.Section H). a minimal left (right) ideal is a simple left (right) A-module that is contained in a certain simple component. we define a relation R in A by aRboa-bsJ. Two minimal left (right) ideals are iso- . A. a left (right) ideal J of A is an additive subgroup of A such that AJ c J (JA c J). where each Ai is a simple ring. F. Any left (right) ideal of a semisimple ring A is expressible as Ae (eA) for some idempotent element e. product A = niel Ai forms a ring under the componentwise operations (a.

with coefficients in R is called the ring of polynomials (or polynomial ring) in n variables Xi. . If . Let R be a ring. [S] R. In general. are said to be algebraically independent over R. Furthermore. . J. when R and R’ are rings with common unity element and R c R’.] generated by a finite system of algebraically independent elements xi. Jacobson. which is called the nilradical (or simply the radical) and denoted by %(A) (.. 1958. 1964. generated by S over R. then xi. Rings and homology.. [7] N. On the other hand.M # {0}. . Furthermore. 1966. 337 Polynomials.x. Colloq. for any ring A. there exists a largest one. McCoy. a ring A is called a primary ring (completely primary ring) if A/%(A) is a simple ring (skew field). A is called a left (right) primitive ring if it has a tfaithful simple left (right) A-module. On the other hand. %(A) is the largest nilpotent ideal of A. I. .&. . Academic Press. van der Waerden. X. Jans.(M) (Wedderburn’s theorem). [ 111 N. Lang. Fundamental concepts of algebra. The dimension r is also equal to the ilength of A as an A-module. over R and is denoted by R[X. Elements de mathematique. Godement.29 Associative Algebras). algebraically dependent over R. Chevalley. References 1372 [l] B. and R. Math. and M is isomorphic to a minimal left ideal contained in Ai. Holt. Hermann. 1967.] onto R [S] defined by If cp is an isomorphism. 368 Rings). . Rings with minimum condition. Surveys. xn}.(D”) of degree r over the field D”. which is a commutative field. Van Nostrand. 1963. X. In a unitary ring A.67 Commutative Rings. The radical of the residue ring A/%(A) is (0). Actualites Sci. and if A is left (right) Artinian. be variables (letters. Jacobson. . Amer.x. . [3] E. Ind. Cours d’algebre. Macmillan. Then among ideals consisting only of quasi-invertible elements of A. [lo] S. indeterminates. Univ. we have %(A)= %(A). Amer. then for each simple left A-module M there exists a unique i such that A. then gD(M) is isomorphic to the full matrix ring M. The radical ‘%(A) is equal to the intersection of all ideals J such that A/J is a left (right) primitive ring. . Algebra.] (. by R[S]. H. and the condition ‘%(A) = {0} is equivalent to the condition that A is a semisimple ring. When S= {xi. 1144b. [2] N. Algebre. [6] C. Conversely. of Michigan Press. Radicals Let A be a ring. Structure of rings. P. Sot. then the endomorphism ring A =&D(M) of M over D is a simple ring. which is called the radical of A and denoted by %(A). then for a subset S of R’ we denote the subring of R’. Rinehart and Winston. The theory of rings. ch. we mean by a ring a tcommutative ring with +unity element. Pub]. Thus a simple ring is an associative algebra over a commutative field (. If A is a simple ring. Lectures in abstract algebra I. The theory of rings. H.368 H Rings morphic as A-modules if and only if they are contained in the same simple component. there exists a largest one. Among ideals consisting only of nilpotent elements of A. X. 1976). over R. and otherwise. or symbols). Nesbitt. Math. are the simple components of A. then any simple A-module M can be considered as a finite-dimensional linear space over D = gA(M). The center of A = G. which is anti-isomorphic to D. A ring A is called a semiprimitive ring if ‘%(A) is {O}. 1 < i < n. C.. the endomorphism ring D = &A(A4) of a simple A-module M is a (skew) field (Schur’s lemma). if r is the dimension of M over D..x. in a left (right) Artinian ring A. . X. General Remarks In this article. Springer. 284 Noetherian Rings. %(A) coincides with the intersection of all maximal left (right) ideals of A. [S] N. X. . 1261a. and let X. .1 3) Rings of Polynomials A. . %(A) is contained in %(A). . 1951 (Springer. Sot. A primary ring is isomorphic to a full matrix ring over a completely primary ring. . Thus the ring of polynomials in n variables over R may be regarded as a ring R [x. A ring A is called a semiprimary ring if A/%(A) is left (right) Artinian and therefore semisimple. 1964. If M is a finite-dimensional linear space over a (skew) field D. Furthermore. Bourbaki. . . [9] J. Addison-Wesley. 8. . and A is isomorphic to &. 1964. [4] N.67 Commutative Rings). 1956. 1956. Hermann.(M) is isomorphic to the center of D. II. M. Math. 1965. 1943. . Then the set of +polynomials in Xi.. 369 (III. . The nilradical of A/‘%(A) is (0). 1. Algebra. L. Jacobson. 1944. Thrall. then there is a homomorphism cp of the polynomial ring R [X. Artin.

If a. If R is a normal ring. . a. the equality holds if R is Noetherian. If R is a tunique factorization domain (u. Zero points of S are zero points of the tideals generated by S. 2 # 0) is called a zero point of a polynomial f(X. K of K[X] such that (i) K[X] is tintegraloverK[Y]=K[Y. iffb. . for every pair (i. a. then pR [X] is a prime ideal of R[X]. . . . .j). then R is generated (as a ring) by a finite number of elements over its subring R.) of an ndimensional tafhne space R” = {(a. Then there exist elements Yi . . 1~0. Then we call R a graded ring. if the ideal Clpa_i Ri has a finite basis. . . then R[X]/aR[X] 2 (R/a) [Xi. if p is a tprime ideal of R.)= V(a...). }) is integral over I [a-‘. If a ring R is finitely generated over an integral domain I. z. +a. . . . is Noetherian and R is generated by a finite number of elements over R. and the quotient ring R[X]/a is called a homogeneous ring. and propositions similar to (i). XJES. and every prime divisor of a homogeneous ideal is a homogeneous prime ideal. . . then (height of p) + (tdepth of p) = (ttranscendence degree of R over K). f being expressed as Cfi with homogeneous polynomials fi of degree i.sR.d. . A polynomial fcR[X] is a tzero divisor if and only if there is a nonzero member a of R for which af =O. and (iii) of part 1 of this section hold for homogeneous ideals a. (ii) V(a.f. . we may restrict ourselves to the case.. II a.. in order to investigate the set of zero points of S.l)-dimensional tprojective space over R (with a. and an element in R. By the tHilbert basis theorem. then so is R[X]. . Using this theorem..d.2. then R[X] is also a u.. 1.. . then R/m is algebraic over K. and Graded a} is called a zero point of a subset S of K[X] . .?e Ri of its submodules Ri (i = 0. then there exist an element a ( # 0) of I and algebraically independent elements zi. every homogeneous ideal is the intersection of a finite number of homogeneous tprimary ideals.)U V(a. then Krull dim R [X] > n + m.) = 0 for every i (this condition f. but also a tMacaulay ring. Homogeneous Rings Consider the polynomial ring R [X] = R [X. a2. A point (la. . . . a. (ii).) is called algebraic over K (K-rational) if every ai is algebraic over K (is an element of K). A point (a. If m is the tKrul1 dimension of R. then V(a. Ideals.) if. If a is an tideal of R.. Therefore. . zt of R over I such that the king of quotients R. in order to study the sets of zero points.. . . da..X. . .X.. then the ideal is called a homogeneous ideal (or graded ideal).) = 0 for any element I in R. . provided that D contains infinitely many elements). . if m is a maximal ideal of R..)= V(a. we obtain the following important theorems on finitely generated rings. the graded ring R = x Ri is Noetherian if and only if R. a. If R is a field... then R[X] is not only a u. We consider the polynomial ring K [X] = K [Xi. In particular. In a graded ring R = C&Ri. . and a2 have a common tradical.) = V(a. Therefore.).. . see below) In a graded ring R.f..a. The Normalization Theorem C. as a tmodule. ) and that RiRjc Ri.. .where S is an ideal. if R is tNoetherian.. are ideals of K[X]. . .) 1aie Let a be an ideal of theight h in the polynomial ring K [X] = K [Xi. zi..1373 369 D Rings of Polynomials Rings. . Zero Points (1) Zero Points in an AfRne Space. (1) Normalization theorem for finitely generated rings. if an ideal is generated by homogeneous elements. (2) If p is a prime ideal of an integral domain R that is finitely generated over a field K. .d.. then R [X] is also Noetherian. a. K generate a n K [ Y] (normalization theorem for polynomial rings). .)fl l’(a. D.] in n variables over a field K.. . The notion of a graded ring is generalized further as follows: A ring R is graded by an additive semigroup I (containing 0) if R is Cier Ri (direct sum) and if RiRjc Ri+j. .... . . . . In this way we define algebraic zero points and rational zero points. .f.)= V(a. and (iii) if a. (In some literature the term graded ring is used in a wider sense. . . . it is sufficient to consider sets of zero points of homogeneous ideals. A homogeneous ideal of R [X] is an ideal generated by a set of thomogeneous polynomials fA (the degree of fi may depend on 2). and the depth of p coincides with the transcendence degree of R/p over K. (2) Zero Points in a Projective Space. X. When a is a homogeneous ideal. .. assume that a ring R is.]. .). More generally. X.] in n variables over a held K and a field R containing K.Y. zero points of a subset S of K[X] are zero points of the smallest homogeneous ideal containing S. (where S = {a” 1n = 1. a homogeneous element of degree d. Therefore. Therefore.]and(ii) Y1. Thus. then (i) Y(a.)=Oforeveryf(X. an element in R [Xl/a is defined to be a homogeneous element of degree d if it is the class of a homogeneous polynomial of degree d modulo a. i. Denote by V(S) the set of zero points of S. . .1. . In this case. . . B. . . da.@ holds if and only if f&z. X. A point (a. a. .. .2. . some ai#O.] in n variables over a ring R.) of an (n .). . the tdirect sum C.

.e. The Hilbert syzygy theorem states that for any finitely generated graded R-module M. Y. is a zero point of q.. ..be the canonical homomorphism with modulus m.) E. (2) Generalization by Serre... .Section C) of a is a zero point of . .. is contained in the set V of zero points of a. it is not necessarily true that V= I$. let f and g be polynomials in x with coefficients inI:f=a... denoting by b the radical of the ideal generated by fi. is not uniquely n .} is called a system of resultants if the radical of the ideal generated by it coincides with a. a... .. b. .) of the ndimensional afline space fi!. and so forth. Then we can find a finitely generated free R-module F and an Rhomomorphism cp of F onto M. .. . such a g (or an equation g = 0) is called a resultant of fi.. . and assume that the field Q containing K is talgebraically closed. . X. P(X) is called the Hilbert polynomial (or characteristic function) of the graded R-module M. .(M..fj). .. For each maximal ideal m of I. + b. X. ..369 E Rings of Polynomials (3) Hilbert’s zero point theorem (Hilbert Nullstellensatz). a. whose definition can be formulated as follows: Let R = k[X... fN be elements of the polynomial ringR=I[X . ..h) by eliminating Xi from the pairs .. if I is a u. = b.. bn-l . Namely. Then eliminate X. then we have V= yo. Algebra. . ... from fi.x”+a. It follows that if M. . have no common factor. Serre [2] as follows: Let R be a Noetherian ring and M a finitely generated R-module. . let I be a field. X. Y.. . Elimination Theory F. N is called the first syzygy of M. “.. . . Let D(h g) be the following determinant of degree m + n: a...] inm+n variables over an integral domain I. the rth syzygy of M is inductively defined as the first syzygy of the (r. Y. .. . an exact sequence of the form where v < n and each F(‘) (0 < i < v) is a finitely generated free graded R-module.be the set of points (ai. .f.. To obtain R(fi. let cp. .] over the field K.. the free R-module erated by ui. vol.fN. The kernel of cp.. (For other methods of elimination see B. . .. = 0.L then some power off is contained in a. 0 “.370 Rings of Power Series 8.. Then N = Ker(cp) is a graded R-module uniquely determined by M up to isomorphism (of graded R-modules). .. Y”]. there exists a polynomial P(X) of degree <n . . Let cp be the graded Rhomomorphism of F onto M defined by cp(uj) =fj. ) Y. such that the system of equations cp.]..:. ... .. . a.) = P(d) for sufficiently large d. then D(A g) is the required resultant R(f. .. ..)E I[ Y. . denotes the homogeneous part of degree d in M.. and genPut F=Ci <jsrn Ruj. the nth syzygy of M is free. a. 0 a. .)(X. If I is finitely generated over a field.. . b. 1374 Then D(A g) = 0 if and only if either ..-P. . then generalized and clarified by Hilbert [3]. form a minimal basis of M over R consisting of homogeneous elements. and let R.l. Let a be an ideal of the polynomial ring K [X] = K [X. If f~ K [X] satisfies the condition that every algebraic zero point (. Let w. However. we introduce m indeterminates u. fi. . . If we wish to write a system of resultants explicitly. .] such that every point of VI. Set deg(uj) = deg(&) (1 d j < m) and supply F with the structure of a graded R-module. To eliminate X. from these resultants. and also in Y.1)st syzygy of M. 0 a.Y. . II..f and g have a common root or a. we have a = b n 1[Y. X. . u. The notion of syzygy was introduced by Sylvester (Phil. Y. and a. u. The syzygy theory was generalized by J. . van der Waerden. a . M admits a free resolution of length <n. be an algebraically closed field containing I/m.d. 0 b. In other words. Syzygy Theory Let fi... we may use Sylvester’s elimination method.. .)=O(i=1.. . The set a of resultants forms an ideal of I[ Y. .g=h. Y. It is obvious that I$. . bn 0 In c II 110 . 0 0 b. Let M be a finitely generated graded R-module.-.e. For a positive integer r. Y. . fN. If fi. then.. . L. For criteria on whether a finitely generated ring over a field is tregular . a..fi.. g. xn-’ + .. 143 (1853)). a. “.. and { gi . and obtain resultants R(fi.. . over R.+a. .g). X.x”+ b. Therefore..1 such that dim..(g) for every m. and b.(f. ..2 .. R is a graded ring in which each Xi (1 Q i < n) is of degree 1 and elements of k are of degree 0). N)hasasolutionin 0:. . . . called a first syzygy of M. Trans...] be a polynomial ring of n variables over a field k. we can proceed as follows: Regard the fi as polynomials in Xi with coefficients in I[X.f.... If every L is homogeneous in X. ... R has the natural gradation (i. X.fN is to obtain g( Y. (1) Classical Case. i. In particular. a. X.x”-‘+..

. i. with coefficients in R. (iv) tintegral domain.). Rings of Formal Power Series (.. X... then R[ [X.X... Suslin (Dokl.). Number Theory. . .x. X..284 Noetherian Rings B). x. power series m x1. Serre. . In particular. i.X~’ . . . + a. . are analytically independent over R. or more generally. Wiley. .+.. . .X. + . Addition and multiplication are defined by (C ad) + (I: bd) = C(ad + bd).. .. Lecture notes in math.1375 370 A Rings of Power Series degree d of the power series. .. If cp is an isomorphism.11 is a u.]cR[[X.EF. . and X 1. Tokyo and Nikko (1955).+. Serre.. .... A formal infinite sum Cdm.. .]].14) Rings of Power Series A. .. .X~~)=~C~~. Symp.xn}.+a.67 Commutative Rings. . Thus 1 . 175-189. Defining cp by cp(Cci .]].. and N.(fo-f)“.. Ann.d. . .~f’ . and ad is called the homogeneous part of . . . .SetX= RCX IZiXiR[CXl. Quillen (Inuertiones Math. . R[ [X. . An element f of the power series ring is tinvertible if and only if its constant term f0 is an invertible element of R.~k. . [4] J. Math. over R and denoted by RCCxl. The homogeneous part a... . with coefficients in R. If iii is a tmaximal ideal of the formal power series ring R[[X. we obtain a ring homomorphism cp: R[ [X. 284 Noetherian Rings. if N.-P. 1975). (v) tregular local ring. ... Akad. 1976)) solved the Serre conjecture by proving that every projective module over a ring of polynomials over a field is free. is a finite sum of these terms such that C ij < d. (ii) tlocal ring. . .-P. Proc.d. The set of such power series in x1.. Assume that R’ is a commutative ring containing R and having a unity element in common with R.. But even if R is a tunique factorization domain (u.X” (a. .. Let Fd be the module of thomogeneous polynomials of degree a’ in X.. if R is a regular semilocal ring. Nagata.. .. [2] J.1] or R{X. and (ii) R is a graded Noetherian ring xdb0 R.$ is also called a . called the power series ring in x 1. x.. multiplicitCs. Uber die Theorie der algebraischen Formen. a’ is an ideal of R’ such that R’ is complete under the a’-adic topology.ER) has a well-defined meaning in R’ (namely.). .. then nt=?iillR is a maximal ideal of R and iii is generated by m and X 1.. Xnll-RCCxl. then X. This element x ci. x.. + aN. . .f. . . 36 (1976)) and A..~. z N2 @ P2 (. (If R is a field. .. . . 473534... .. . .11 or R{x. . ..... In the following special cases..... .. . .>x. .references to 284 Noetherian Rings. (4) Special Cases. Then R[[Xl. . .. an element of k((X)) is expressed uniquely in the form C~. Then an infinite sum Cci . An important result of Serre is that R is a tregular ring of tKrul1 dimension at most n if and only if an nth syzygy of every finitely generated Rmodule is tprojective. an rth syzygy is defined inductively as in (1) of this section. .11 is also of the same kind: (i) YNoetherian ring. Local rings.. 368 Rings) Let R be a commutative ring with unity element 1. are elements of a’. D. x.. 370 (111. called a formal power series or simply power series in n variables X. [3] D. Intern.rcZ). a RCCX. If R is one of the following. . the set of power series forms a commutative ring. . ‘formal power series ring k[ [Xl] in one variable X over a field k is an integral domain whose field of quotients is called the field of (formal) power series (or (formal) power series field) in one variable X over k and is denoted by k((X)). However. References [l] M. where R. with coefficients in R. X”]]. . X..$ (each ij ranges over nonnegative rational integers and c~..f. is a subring of R’. Algirbre locale. .~f’ . 1965. x. x.S. then the power series x ad is identified with the polynomial a. 1962 (Krieger. XJ] need not be a u. Hilbert..i. .X. Sur la dimension homologique des anneaux et des modules noethkriens.. over R and is denoted by R[[X.a. are first syzygies of M. .“-’ .XJl is tcomplete under the X-adic topology (. x.. . . and in this case f-’ = C&f. @ P. Springer. ad +a..277 Modules K). which is called the ring of (formal) power series (or (formal) power series ring)inX. Nauk SSSR (26 Feb. X. of degree zero is called the constant term.}. X. For a positive integer r. is a field and M is a finitely generated graded R-module [4]. X. (iii) tsemilocal ring. we can define the first syzygy of M uniquely up to isomorphism: (i) R is a Noetherian tlocal ring and M is a finitely generated R-module..+ ofelementsa. .d.operations.f. (vi) Noetherian tnormal ring.is =a. . . . 36 (1890). .x$.. Also . .. . then we say that x1. .X$ .XJl. then the infinite sum is defined to be lim. If there is a natural number N such that ad = 0 for every d > N. . (3) Serre Conjecture. .. . then there exist finitely generated tprojective R-modules P1 and PZ such that N..~. .. . .. determined by M.. Alg.. (C Q)(C bJ=C&+j=daiQ BY these. if S.

ErnR[x]. . i.e.]].. Van Nostrand.. and (ii) if . +. If Q is a +prime ideal of K (( X. h 3 h. (iv) if Q is a ‘normal ring. thenK((X .)=Ccil. . 1975).1 0) Robust and Nonparametric Met hods A.Cl]). The converse is also true if K is a tperfect field (if K is not a perfect field. X.-. X.~K((x. .]]. Y.))/a is isomorphic to a ring that is a finite module over K ((X. A formal power seriesf(X..)) in n variables X.). Nagata. a commutative ring R’ with unity element such that R’ is a finite Rmodule is the direct sum of a finite number of Hensel rings._.. h.fi. then.. and tcomplete valuation rings. General Remarks Cll).R[x] tmR[x] = R[x). X. .. If the +rank of (J(. then there are manic polynomials 9.. Interscience. then 2 ci.x.]. .. 1962 (Krieger. . ....l.. X.)). Then for an arbitrary element g of K ((X.f.i. for which the following statements hold: (i) 0 is a ifaithfully flat Q-module. 1960.afl _. In practical applications. is a regular local ring.f. a polynomial in x is called manic if the coefficient of the term of the highest degree is 1) such that f-g. if ui E K and ~(a. 1975.. For . it often happens that the assumptions made for the model. . and &me = Q/mQ.fu =.. When R is a Hensel ring... then the maximal ideal of 0 is nto. For any quasilocal ring Q.. . . R is a tquasilocal ring). . then e is also a local ring..)). a iJacobian matrix J( fi. then the ring R. Weierstrass’s Preparation Theorem. X. 371 (XVIII.. Xnm. R... Let K be a field.X. Important examples of Hensel rings are complete local rings. Springer. . the complex number field. xfl.< A4 for every (il. _. u.. If L‘ is a +trivial valuation. (iii) if R is a Hensel ring that contains Q and has a maximal ideal n. X. . It is called the ring of convergent power series (or convergent power series ring) in n variables over K and is denoted by K((X. . then K ((X. ..Xbnis said to be a convergent power series if there are positive numbers r.. . K = C.f.)) such that .f. element . ...)r... then there is one and only one Qhomomorphism q of 0 into R. ... .)). . The set of convergent power series is a subring of K[ [X.. In this case.. then the Q-homomorphism q is an injection.401 Statistical Inference). It is a regular local ring of +Krull dimension n.. .X.-.f= C c. there exists a unique q E K ((X. and n n Q = m.i.). hc R[x] such that .]]..X. modifying J(. Commutative algebra II.h.. rings of convergent power series.]] is a prime ideal. Hensel Rings A Hensel ring (or Henselian ring) is a commutative ring with unity element satisfying the Robust and nonparametric or distribution-free methods are statistical procedures specifically devised to deal with broad families of probability distributions. Rings of Convergent Power Series 1376 Let K be a field with multiplicative +valuation L’ (for instance. called the Henselization of Q (for details .i. ...f. X. (v) if Q is a local ring.370 B Rings of Power Series B..)). g.~ R. Local rings. and let q be a prime ideal containing Q.. X. .. C.i=Ofor i=O.f.) is defined to be the t x n matrix whose (i. X. . and t. + +.$ has its svrn in the tcompletion of K. and Q is dense in 0. .R[x] +h. and its completion is K [[X. . l. .)))..) modulo q) is equal to the height of Q. there is an invertible element u of K ((X..f.)>. The Jacobian Criterion. tPartia1 derivatives ?/c?Xi are well defined in R. ... the ring of formal power series K [ [X.f.d.)) or K{X.. new edition.)). then QK [ [X.I. References [l] M.K((X.)) such that g ~~EC:Z~X.. X.xgwK((X X. we can have a similar criterion following two conditions: (i) R has only one maximal ideal m (i. By this theorem.. ./Z.. In the theory of statistical inference it is usual to assume that the probability distribution of the population from which the observed values are chosen at random is specified exactly except for a small number of unknown parameters (. (ii) if tn is the maximal ideal of Q.. . are manic polynomials in one variable x (here.. . . X.f. . we see easily that if n is an ideal of theight h of K ((X. . however. .. ..1 and cO.]].. In particular (considering the case where g = XL).. . .Or #O. X...).l-’ + X: (f. X.))=K[[X .fl. For an I. M such that u(ci.~.. X. or the ring of convergent power series K ((X.l rtn . over K. . and let R be the ring of polynomials K [X. .. X.. modulo m. X. [Z] 0.f= gh and g 3 go. .i”X.. go. Let Q be a +prime divisor of the ideal CifiR..) < r.(z) = 1~~1 for ~EC). Zariski and P. j)-entry is <fi/aXj..f. X.. X. . there exists a Hensel ring &... . _. Hence it is a u. Samuel. assume that c. r.

) be the number of xi that are greater than co. and P. Mann and D. The Two-Sample Problem Let F and G be continuous distribution functions of random variables X and Y. In such cases robust and/or nonparametric procedures that do not require exact knowledge of the shape of the distribution and yet prove to be relatively efficient or valid are required. . 1962) provided the initial foundation for robust estimation. 0 < c < n).. On the other hand. Tukey (1960. Savage (1958) showed. . cording as t > 0..(i) = i is frequently used and is called the Wilcoxon signed rank test. the rank test by F. The developments of the theory of robust estimation are reviewed by Huber [4-61 and R. J. Sen [2]. Y. Andrews et al. . x.. J. may not hold for the actual data..) be an independent trandom sample of size n from F(x).)>c. x.. . A test procedure by the following ttest function cp is tuniformly most powerful in some neighborhood of 5.. . .) be the respective sample values. R. (X... . Let F(x) be a tdistribution function of a trandom variable X..(l). x. The 100~ percentile of F is denoted by tp. Two papers by J. M.x.1377 371 c Robust and Nonparametric B. x. . . x.... . . . The term nonparametric or distribution-free is used for problems of testing hypotheses..2 = 5’ for F(x) = l/( 1 + eex). Wilcoxon (1945). . . Consider the problem of testing the hypothesis H: F(x) = G(x) against the alternative hypothesis A.)=c. The procedure with a. G. . A. . .x. Let R: be the rank of IX. .. ..) and (yl. and (x 1. is G(x) = F(x . called the signed rank test. Huber (1964) proposed an estimator of location by generalizing the method of least squares. . the following procedure is proposed. These findings accelerated the studies of nonparametric tests. In the years that followed two important ideas appeared: the concept of asymptotic relative efficiency by E. .. Takeuchi (1971) proposed an adaptive estimate that is asymptotically fully efficient for a wide class of underlying distribution functions. C.. . L. = 5” for the double exponential distribution. Yanagawa (1967). the theoretical foundation for nonparametric tests was first given in the proposals for the permutation test by R. J... x. x. . Although the idea of the sign test appears in the work of J. where cp(x. V.x. B. For testing the thypothesis H: 5.. and let Y(t)= 1. P.) be the corresponding random samples.X. < 0. K.)=c. E. Lehmann (1963) noticed that estimators of location could be derived from nonparametric tests and that these estimators have sometimes much higher efficiency than the sample mean.X. The following procedure rp.. H. 399 Statistical Estimation. Hogg [7].. is true.. . G. .(R+)Y(Xi-to) i=l for some weights a. $idik Cl].)<c. Hodges and E. Hampel(l974) the estimator proposed by Huber has become a core of subsequent studies of robust estimation. (X1. Wolfe [3].(n). y. . .) and ( Y1. . . .: F(x)> G(x) for all x and F(x) f G(x). Chernoff and I. i. . Puri and P. Along with the idea of the influence curve introduced by F.-t”l. the logistic distribution. < 5’ against the talternative hypothesis A : t... A similar study for scale was made by S. Suppose that F(x) is symmetric about x = lliz. F(<. . x. .e). . K. that the asymptotic efficiencies of nonparametric tests are incredibly high.) =p. L. Randles and D. 1 0 when &(x1.) is defined by the equations 1 when cpb. which is the uniformly most powerful rank test in a neighborhood of (I.)= a when &(x1 . Various proposed estimators are compared in the book by D.. F.. and the test based on Ustatistics by H. A frequently used example of such an alternative hypothesis A. . Pitman (1948) and the development of the theory of Ustatistics by W. . Whitney (1947). x.> to. L. : F(x) f G(x) or A.)>c. Box (1953) first coined the term robustness in his sensitivity studies. [S]. respectively. dx. When the alternative hypothesis A. R. Set ux.e.. 0 . 400 Statistical Hypothesis Testing).0 acIx1-5°1. . H.396 Statistic. and (x 1.)<c 1 (0 < a < 1.. Arbuthnot (1710). a. Fisher (1935). Let i(x.‘.. . 0 when i(x. . is also used for testing the hypothesis H: <Ii2 < 5’ against the alternative hypothesis A: <1/Z> 5’: 1 when &(x1. ... in studying the asymptotic distribution of a class of rank statistics.)= a when i(xl . in which he investigated how the standard statistical procedures obtained under certain assumptions are influenced when such assumptions are violated. Hgjek and Z. R. .. The One-Sample Problem Methods especially those about the shape of the distribution. ..5’1 among IX. i(xl . . A.) be an observed sample value. . ..X. and the term robust is mainly used for problems of point estimation (. .)= f: a. we say that the random variable Y is stochastically larger than X and write F > G. This procedure is called the sign test. Hoeffding (1948). Kakeshita and T. W. . Huber [6]. recent progress is summarized in the books by J.. R.

. .. When J is the inverse function of the distribution function ‘of N(0. .) Y. then cp is considered a desirable test. G) > P. i:=l. . among which the KruskalWallis test is a particular one having ii(x.x. . .. Then TN is represented by the integral J. ekzk. Set HN(x)=&F. . v. y.Xkl. for which the following test was proposed by T.x...O<pi<l. n + m) in the arrangement is an xi or yj.(x)+(l -QG. v.x..: F. . > 0. The tmaximal invariant statistic in this case is the rank (R. 1 i=l... Tamura....Y..(x))~Fm(x) J with ek = J.. The following statistic TN is used frequently in nonparametric problems..: F. v’) = 0 otherwise..A)G(x).oowithni=piN.. then the test by TN is called the Fisher-Yates-Terry normal score test. TN is defined by T. y. Then the hypothesis H and the alternative hypothesis A.. Uk) whose coordinates U’ are defined by means of a function *TX ll.. . y....371 D Robust and Nonparametric 1378 Methods ante o2 of TN are given by P= J(H(x)) Wx). If a test function cp(x.. ..400 Statistical Hypothesis Testing C). ..B. = m/N < 1 . are invariant under the group of transformations of the form xf=h(xJ. . l). Testing the hypothesis H : F = G against the alternative hypothesis A : F # G = F(x/a).. The statistic VA’V has asymptotically a tnoncentral chi-square distribution with degrees of freedom = rank C. then PJF.Uf. WhenN=Cni-. Let x1... D.(x) = . . Cj<f $(xi. . CA = I .. respectively. ..(x) = F(x . Let B be the projection matrix corresponding to the eigenspace for the zero eigenvalues of the matrix .(F. . where F.. .“=. . n) yield cp(x..whereUisa+Ustatistic defined by CP(X.Y. The test function is given by cp= lforU>candcp=OforU<c.. I i Lehman& Theorem.x..Q. then cp is unbiased (. .?J..) with ei f B. If cp satisfies the conditions stating that y? > yj (j = 1.xi). If in addition cp is a tsimilar test. CT> 1.m. .). and I = lim 1. Let x be the family of all strictly increasing continuous functions. .. he . and 0 < I.. where p#‘. The k-Sample Problem a with $(x.Xlm.whereU= (y)-‘(.. Several kinds of test represented by a critical region of the form VAV’ > c are proposed.) of the kth order statistic Z.(x) are the iempirical distribution functions based on (x.Y.) of (Y.2. .. . . . u’. and let A be a matrix such that AB = 0.). The ksample problem is concerned with testing the hypothesis H: F.Y...) of (X. k.x..E.v<u’<v’orv’<u< v.(x). The Wilcoxon test (or the Mann-Whitney Utest) is described by a test function cp= 1 when U>candcp=OwhenU<c. .G)= JJ x n dF(xi) jj dG(Yj).and C pi = 1. .... ...X).X. .) J Na2=2(1 -i) W4(l IJJ X<Y x F(x)(l JJ -G(Y)) J’(H(x))J’(H(y))dF(x)dF(y) 1-i +T -F(Y)) X<Y x J’(H(x))J’(H(y))dG(x)dG(y) where J(H) = lim iv-m J. then the test is called the van der Waerden test..Y:. j = 1. .) when the combined sample (X.Y. < 1. A... Yj> Yj. For a given set of N = n + m reals {ek}.. .(x) are equal. then V=(JN(U’-E(U’)).. ..) and (y.. -i A... y. . ..=m-’ C. v’ < ti <v and $(u.*)~cp(x.y)= 1 when xiy and $(x.... YJ is ordered in an ascending order. C) > c( for any F > G. ..I. When ek is the mean E(Z... Several tests have been proposed for this problem.(x) = F(x/ai) with ai$ (T. H(x) = Nx) + (1 . X... .. Chernoff and Savage [9] proved that under some regularity conditions the asymptotic distribution of TN is normal and that the asymptotic mean p and the vari- Let (X. F) < CIand PJF...rXkmxi. xi.(k/N).. R. ....: not all the F..v’)=l whenv<u<v’..k. y. This test is similar and unbiased. S. yn be arranged in order of magnitude.:. When tnk= k/N. Y.xk)=~~ii(x.(x) and G. ni) be a random sample of size ni from the population with a distribution function Fi(x) for each i = 1....) or the rank (S... using quadratic forms of the vector-valued U-statistic U = (U ‘. = Fk(x) against an alternative hypothesis A.y)=O when x > y. .=h(yj) (i= 1.) with $(~a u’. j = 1. F). the statistic TN is equivalent to the +U-statistic in the Wilcoxon test. .)-’ Ci<i.. . n. .(H). . . Set zk = + 1 or 0 when the kth value (k = 1...k. respectively... or A.) satisfies PJF. X. l). is another two-sample problem. in an independent sample of size N from N(0..JN(Uk-E(Uk))) has asymptotically a tmultivariate normal distribution N(0...

exists and is independent of u and lim fi(@.. D. .(x) . t.say) s.(x)<c. consider a two-sample problem on a tlocation parameter.} and {nl} of positive integers satisfying tl < lim p(e.. . F = G is true. where 0 is an unknown location parameter. If. where cp. .(D.. 1 -e-“‘. E.(x)= 1 a when 0 when 1 when Icl. X..) cp.(x) F. As an example. respectively. = 0 and 0..) be an independent random sample from the population with a distribution function F(x .) and p(0) $J. Put 0.. The tpower functions of (P” and $” are denoted by p(0 1cp. d..} and {&} are based on statistics T..) of alternative hypotheses.*(x)<c*.. If the population distribution is normal and the Wilcoxon test is used to test the hypothesis of equality of means.. are called Kolmogorov-Smirnov tests. respectively. . For the same problem.{~~. 1 whereX=(X. the asymptotic relative efficiency of the Kruskal-Wallis test against the F-test is 3/n./$ limP..(x) .E&4). For the hypothesis of equality of means in the k-sample problem.. .*(x) =c*. y) = 1 when x < y and 6(x.(x)= b when t... then e is called Pitman’s asymptotic relative efficiency of { cp.. If there is more than one test procedure for a given testing problem.. provided that m and n tend to co so that N= mn/(m + n)+ co and m/n is constant.} be such that &+0.I cp. Set 0 when t. are frequently used to test the hypothesis F(x) = F. F.0)... x. and (x. the asymptotic relative efficiencies of the Fisher-Yates-Terry normal score test and the van der Waerden test against Student’s test are both unity.(X) and T.)).(x) be the empirical distribution function based on a random sample of size n from Let(X..) = limB(eiI~~. Set Methods as basic functions.... t.(d..). Asymptotic Relative Effkiency of Tests dn=swIF&)--&)I. .)andx=(x.(x). and D. and T”* are asymptotically normal... 4 = sup(F.. Taking account of these facts. Let {cp.F.:)<l. On. (This problem is called testing goodness of fit. then under some conditions e is given by the formula d. Let 0 be a real parameter and {0.(x) be two empirical distribution functions based on samples of sizes m and n from F(x) and G(x). x(2k+l)-‘e((Zk+l)*n*/E)((l-o)/o)r* e-‘*‘2dt.. = sup(l. = k/& (k = constant) for simplicity.. .1379 371 G Robust and Nonparametric F... Kolmogorov-Smirnov Tests lim P. D.(x) If the hypothesis ..=g... then its asymptotic relative efficiency against Student’s test is 3/n...<z/&)=K(z)= ( -l)ke-Zk2rZ.” are used to test the hypothesis F = G.= Then sup (I < F&G F. Suppose that F(x) is continuous and symmetric about the origin.(x)-G. Consider a hypothesis 0=0.}. in the following manner: 1 when q. for any increasing sequences {n. and $” are test functions based on a sample of size n. t:(x) >c*. D. respectively. then one may wish to compare these procedures. tn(x)=c. as i+co.<z/JN)=K(z).). The statistics d..} against {IL.. let F.(x)).=supIF. where 6(x.. we have = L(z). x. s. Ii-Ii P.(x) and G.) be its observed value.(D. = t.G.* = t:(X).(x). y) = 0 otherwise.limn~lni(=e({cp. Interval Estimation Let F.. provided that the sample is distributed normally. If T.}. X. S. d. G.) In a two-sample problem.} and { &} be two sequences of level tl tests.(x)l.... =. <z/. The tests using the statistics d.(x) .(x)>c. Suppose further that the tests {cp.. and a sequence (0.

.. This estimator is called the a-trimmed mean..<S(X. The first method is touse T. where p = 1 + [car] -cm. x.)<0<u. Johns [lo] proved that under some regularity conditions TR is an tasymptotically efficient estimator of 0 for F....x. then the confidence interval of 0 with lOO( 1 .... The second method is to estimate 0 by T. Let J be a real-valued function such that j. converges to T(F)= fhJ(t)F-‘(t)dt in probability..X.+2)+ “’ +PX...) if and only if d. 4(x 1. . for some constant K. .y)% confidence coefficient is given by (L.> x. X.M1/0. If $(t)= --f’(t)/f(t) is chosen for $(x). . where P. .y(x. .) and U..x. are given by L. 9 be a class of distribution functions of the form F(x) = (1 -E)@(x)+EN(x) for a given s (O<. X. V. is any robust estimate of scale. and let I(F) be the tFisher information on 0. then T.~I/(p. and let (x1. < 0 and S(X.(x. Let X(. . x”) for the one-sample nonparametric test for testing the hypothesis H: 0 = 0. then as n+co..1 (... Let J be a real-valued and nondecreasing function such that jhJ(t)dt =0 and R:(O) be the rank of IX... . X.)ln(l . L.. . -8.0. = S(x. x.X.. < <X.s(x. i. and M.-O)<p}. = K&+. is called the M-estimator. and d..-tI)>n}. -. . e. 1 is chosen to be x(t) = t$(t) . . < K&. .. Generally. . or by solving the simultaneous equations (1) and gx(fq)=o n with respect to T. . where n is the expected value of S(X..SetY(t)=l.T..X~. and set ak = lf/Ll.)+x~[. x. X... . .F)isgivenbyp.. the M-estimator converges as n+ 00 to T(F) in probability. > K. . . = M/..-8. fI**=inf{&S(X. where M is the sample median...f’(t)/f(t) and J(t)=$‘(F-‘(t))/l(F). Let CD(x) be the distribution function of the standard normal distribution..2. . In the context of the maximum likelihood estimation. .) under the hypothesis H: 0 = 0. T. For an appropriately given y (O<y < l).forsomegiven constants a. a. There are four methods of constructing robust estimators of 8. When p(t) = t2.) that satisfy P. J((R:(@+n)/(2n+ l))Y(X. n). -. If there exist statistics L.(~)=(Px~[. H(x) be an arbitrary continuous distribution function which is symmetric about the origin. . . = 1.(t)=t’/2.-0 .M-dIjr where M = n(n + 1)/2 and q. .b(x.=a.e. and S. and this estimator is called the Restimator.) d 0 < U. x.-[.-8). x.(x.O according as t > 0. .. such that & a.x..0) =Ci=.. is the statistic for the Wilcoxon signed rank test. . IX.l+. . This interval is distributionfree.)).. and the . Then Chernoff. J(t)dt = 1. Let fI*=sup{t)... . is the tmaximum likelihood estimator of 0 for F and is asymptotically efficient under some regularity conditions.) be an independent random sample from the population with a distribution function F(x . .“.K(t(K2/2 defined for (t] < K. the median of { ]xi.) such that L... <S(x. S tends to be the statistic for the Wilcoxon signed rank test.) or by satisfying Using the statistic S.. . a tconlidence interval of fJ is constructed as follows. the M-estimator defined above is not scale invariant. for all 0... Under quite general conditions.399 Statistical Estimation PI. .. A scale invariant version of the M-estimator is obtained by replacing the defining equation by &$(y)=o. it holds that P(L.+a.. Gastwirth.... respectively.(x r .. Let p be a real-valued (usually convex) function of a real parameter with derivative ti = p’. means the probability under the hypothesis H: 0 = 0.2c0. .{d. J.1. Huber [ 1 l] proved that p minimizing s~p. are ordered values for M averages (xi + xi)/2 (i <j = 1.. and U. it agrees with the least squares estimator. x.X.}=l-y.-Q(. is called the L-estimator. =(Q* + 0**)/2.+. T. X. and V(p.-O( .$.?(xI.T(F))dF(x)=O. L.. T. When S. .).) and U. .. Set $(t) = -.(x...-@cd.)} = 1 --y for all F.. which is defined by jt. .J(t)dt. Suppose that F is a distribution function having an +absolutely continuous density function f: Denote the derivative off by f’. ...g. .. The third method employs nonparametric tests for testing the hypothesis H: 0 = 0 against the alternative hypothesis A : 0 > 0. be ordered values of X.2. where Q is an unknown location parameter. select constants d. H. where S... -O(among(X. Hodges and Lehmann [ 121 first proposed this technique.s<l).) be its observed value. F) be the asymptotic variance of T. . Then an estimator of Q is defined by T.X~. When F(x) is symmetric about the origin and J(t) = t-i.371 H Robust and Nonparametric 1380 Methods by minimizing Cf=r p(xi . in the range of S(x. An example is T.)<d.6745}i. Point Estimation (1) n Let (X.(x... . . .H)..

. Jaeckel [14] made an atrimmed mean adaptive estimator by selecting an a that minimizes the estimated asymptotic variance. Among these. R-estimators of the regression coefficients are obtained by minimizing & a. Dependence for all x is called the influence curve. (1) Spearman’s Rank Correlation.CjQjaij.. X. . . F. be the rank of Xi among X 1.(Ri)Ai. A. by J. n. T) = by T((l-e)F+s6.1381 371 K Robust and Nonparametric Methods R-estimator reduces to the median of n(n+ 1)/2 averages (Xi+Xj)/2 (1 <i<j<n). (2) Kendall’s Rank Correlation. . n”z(T(F.. . A real-valued function IC(x.) be random samples from a population with a bivariate distribution function.. As n-co. F-‘(l--)/(1--a) when x>F-‘(l-a). are identically and independently distributed random errors whose distribution function is given by F(x). .). the influence curve for the cl-trimmed mean IC(x.. The curve was first introduced by Hampel [lS] to study the stability aspect of estimators against a small change of F.) that minimizes x. j=l i = 1. Although the above three methods provide robust estimators. For symmetric F. .0.T(F)) is asymptotically normally distributed with asymptotic variance S(IC(x.). . Y. and selecting the best weights of the L-estimator.) as an estimator for (0. and Ed. for F in T(F). by T.F.)=(n-1)-l. when they are rearranged in an ascending order. Set di = Ri -Si. least approximately. I.(i) = 0.2. . As an example. The idea of the M-estimator is a direct generalization of the method of least squares. of 0 calculated from an empirical distribution function F. the R-estimator defined by the statistic S with J(t)= -f’(F-‘(t))/f(F-l(t)) is asymptotically efficient under some regularity conditions. defined by F(x)+l-F(2T(F)-x) 2 dF(x) = 0. If (Rj-Ri)(Sj-Si)>O. The Regression Consider Statistical Problem problem (403 the linear regression Models D) xi= f ejaij+&.Si) and (Rj. E. where S. Let (Xl. . Take pairs (Ri. F. . If there is no dependence between X and Y.~~lIC(Xi. Ed. and Sj be the rank of yj among Y.) is some monotone function satisfying CyZl a. K. R. defined similarly as Ri. the R-estimator converges in probability to T(F). The Influence Curve Let T(F) be a functional of a distribution function F. Thus it follows that n’j2(T..Zl p(X. Y. Under some conditions.( .)--T(F) & where the Xi are observable variables. and a. where Ai = Xi .(Ri)Ai is asymptotically equivalent to minimizing the properties of which were first studied JureEkov& [ 163. is the distribution function of a point mass 1 at x. . . when F is symmetric about the origin. Ri is the rank of Ai amongA. estimating the elements of the tcovariance matrix of the order statistics by Ustatistics. J.=l-6Zid’/(n3-n)iscalled Spearman’s rank correlation. Y. the aij are given constants. F. . a striking one is the asymptotically fully efficient estimator for a wide class of F proposed by Takeuchi [ 131. T) is given by F-+$/(1 x/(1-2a) -201) when x < F-‘(a). their behavior still depends on F. converge to T(F) in probability as n+ co.T))+0 I in probability.. then E(rs)=O and V(r.Thenr. The last method of constructing robust estimators consists of estimating 0 adaptively by utilizing information on the shape of F. .. The estimator is constructed by using subsamples of size K (K <n) drawn from the original sample. . if the Xi and 3 are independent random variables. and let an estimator T. “. at . A.(X. namely. .. Various quantities are devised to measure the degree of dependence between X and Y. L. . T) defined by IC(x.. it can be proved that as n-+ co.Cjejaij) for some function p such as the one described above..e. to adopt (&. . i. = T(F. . the Qj are regression coefficients to be estimated. which are seldom affected by outlying observations or contamination by gross errors.. F. . when F-‘(cc)<x <F-‘(l-a). set By substituting the empirical distribution function F. we can represent the robust estimators discussed in Section H. T))2dF(x).)-7’(F)-.Sj). . .. It has been proved that minimizing & a.. .

. Statist. J. S. H. Robust estimates of location: Survey and advances. replaced the abacus in the 12th century. Isidorus. A uniformly asymptotically efficient estimator of a location parameter. J. Wolfe. P. L. Hampel..) = 2(2n + 5)/(9n(n .5 (1964). and others. when References [ 1] J. J. otherwise. known for his calendar reform in 46 B. then the usual rsample correlation coefficient is calculated from these scores. Amer. P. Johns. [6] P.1270). Statist.371 Ref. Statist. Assoc. [2] M. Robust statistics. 29 (1958). [S] P.. Huber. [12] J. Math. n.). (3) Rankit Correlation. R. their arithmetic consisted of computation (by means of the abacus). which aroused a demand for land surveying techniques. then E(rk) = 0 and V(r. Hodges and E.1)). also undertook to measure his territory. J. 2922301. z=. the year of the fall of the Visigoths-and then through the Crusades (1096. where 2 runs over all possible pairs chosen from (R 1. and if there is no dependence between X and Y. Academic Press. 1041-1067. Randles and D. 1382 Methods r. [7] R. Introduction to the theory of nonparametric statistics. If there is no dependence between X and Y. Rogers. Andrews. Alcuin. Huber. Statist. Math. Chernoff and I. J. 480-524) wrote his two books on arithmetic and geometry. 34 (1963) 598-611. Ann. Math. 43 (1972). Robust statistics: A review. during this period Boethius (c. Estimates of location based on rank tests. Statist. V.. 38 (1967). 1 (1973) 799-821.= [13] K. Asymptotic normality and efficiency of certain nonparametric test statistics. Puri and P. Lehmann. [15] F. Ann. Ri and &. 1981. V. The former was a summarized translation of Nichomachus’ book. and M. Ann. Huber. Statist. (R. [ 1 l] P. Jureckova. H. Takeuchi. [lo] H. Statist. Huber. asymptotically. 52-72. Statist. Robust regression: Asymptotics. Ann. are replaced by the corresponding normal scores. then E(z) = 0. 372 (Xx1.. the quadrivium was to be studied for the glory of God.... they developed a computational method using duodecimal fractions. Chernoff. as seen in books by Bede Venerabilis. Nonparametric estimate of regression coefftcients. Hampel. [ 141 L. 69 (1974). Books on practical geometry which provided this knowledge were called gromatics (a “groma” was a land surveying instrument).). W. Wiley. Hajek and Z.). 733101. . Ann. Some flexible estimates of location.. F. S. J. W. 42 (1971) 15401552. which constituted the “mathemata” of Plato’s Academy (closed in 529) were treated as the “quadrivium” (the four major subjects) in the Encyclopedia of Martianus Capella. [4] P.. and arithmetic. 1967. Wiley. [S] D. Gastwirth. Music. Math. Arabian science was imported first through Spain-under Moorish influence beginning in 711. Princeton Univ. 66 (1971). Robust estimation of a location parameter. [3] R. Sen.. i= 1. The influence curve and its role in robust estimation. L. Computation with figures. Statist. Amer. Statist.. . and the latter included propositions from the first three books of Euclid’s Elements (without proof) and practical geometry. [ 161 J. and money. R. Adaptive robust procedures: A partial review and some suggestions for future applications and theory. L. Theory of rank tests. Math. 1). [9] H. 3.3)-i. 383-393. ‘pij= -1. 1979. Bickel.3) Roman and Medieval Mathematics The Romans were interested in mathematics for everyday use. Wiley. Assoc. Books on mathematics from this period laid emphasis on the computation of an ecclesiastical calendar and mystical interpretations of integers. conjectures and Monte Carlo.e. A. Tukey. Hispalensis. J. 9722994.log~ 1 -r (TFisher’s z-transformation). This correlation coefficient rR is called rankit correlation. Savage. 42 (1971). 69 (1974). Huber. These stat%tics are used to test the hypothesis of independence. the means of order statistics in an independent sample of size n from N(0. K.. J. and J. Math. After the establishment of the Roman Church in the 5th century. Toward the end of the Western Roman Empire (476) Greek mathematics was studied. R.C. Cassiodorus. originating in India. A. The statistic is called Kendall’s rank correla- tion. Sidak. Jaeckel. geometry. 1328-1338. Ann. Amer.. Nonparametric methods in multivariate analysis. and Maurus from the 7th through 9th centuries. 1971. L. J. 1972. Ann. astronomy.C. Statist. Press. J. weights and measures. Asymptotic distribution of linear combinations of functions of order statistics with application to estimation. . Ann. Julius Caesar (102?%44 B. F. Math. V(z) = (n . Assoc. Robust and Nonparametric ‘pij= 1. Hogg. 909-923. where l-i-r. i. For their monetary system.

Representative books of this period are Liber abaci (1202) and Practica geometrica (1220) by Leonardo da Pisa (also known as Fibonacci. 1193-1280) and Thomas Aquinas (1225?-1274) discussed infinity in a way that went beyond Greek thought and thus helped to lay a basis for the modern philosophy of mathematics. Clagett. 1964. and later in Paris. References [1] M.1383 372 Ref. Introduction to the history of science I. 1927. [2] M. The science of mechanics in the Middle Ages. Italian merchants. introduced fractional exponents and conceived the graphic representation of temperature. Abelard-&human. universities developed from theological seminaries. along with Greek books on geometry and astronomy (such as books by Euclid and Ptolemy). and algebra. Univ. Teubner. 1955. rapidly adopted the new system. 1880. 1323-1382). theologians such as Albertus Magnus (c. who influenced Leonardo da Vinci (1452-1519). [3] M. I. Roman and Medieval Mathematics Arabian books on arithmetic and algebra. showing Indian influence. and Cambridge. a precursor of coordinates and functions. whose occupation necessitated computation. II. II. Oxford. Cantor. c. Sarton. The new methods were not limited to merchants. Clagett. . of Wisconsin Press. commercial arithmetic. Univ. 1892. [4] M. although no remarkable creative contributions were made. were translated into Latin. Clagett. 1959. first in Italian cities such as Bologna and Palermo. Carnegie Institution. [S] G.1250). Mathematics was taught in these universities. From the 1 lth century. However. From Homer to Omar Khayyam. 1170. The former includes the four arithmetic operations. The French bishop Nicole Oresme (c. of Wisconsin Press. Vorlesungen iiber Geschichte der Mathematik. Archimedes in the Middle Ages I. Greek science in antiquity.

.

.

l.. . . The decomposition (2) is not unique.. the tconditional probability distribution of (J. . The probability (J.(v))}>O. X) for given (I. . The Problem of Inference of If the sampling procedure is regular. . Specifically.. J).-i).. Hence.ande~=~(ai-2 a) /(N ... Cr= Z q/N is testimable if and onlyifPr{I~i}>Oforalli..=xlj. . . . The two main mathematical problems of sample surveys are to determine a random sampling procedure and to provide methods whereby statistical inferences can be made concerning 0 (... X) = (5i.. X) is given by positions. . . the index number of the unit in the population is J.-I.. I).Xl. (1) Expression (1) is the fundamental model for the sample survey problem. . where 1 XOW> II = 0 if I. N)} > 0.. and let Y = (Yr .. . J. C.. . if cli is real and the sampling procedure is regular. Note that P(J. if the tjoint distribution of J is independent of 0. v=l. where xs(Xi) is defined as 1 if Xi =: clJi and 0 otherwise.(v). .X)) =p(J. I) = x0(X. the second condition can be replaced by Pr {I 3 (jl(v). . J. we can restrict ourselves to the class of procedures depending only on (I.. This method is mainly applied to demographic statistical surveys and opinion polls. 2. . ..l)=EX(ct-aj)‘/N(N . . for example. and of Xi+r . X) is independent of the parameter 0.j. J. There exists an unbiased estimator of g(0) if and only if there exists a decomposition Pr{I=(j. but it is also applicable to random samples of physical materials. .. .i=l 0 otherwise. .373 A Sample Survey 1386 373 (xvlll. ..(v))}>O. it is called regular. Neyman established a method of ascertaining objectively the reliability of such information. and corresponding to different decomCondition (c) is assumed. J).) be the corresponding X values. it is called a random sampling procedure. The set of all characteristics of the units in the population is designated by 6’= { c(~. .X.. .2. ..X.. different unbiased estimators are derived. .401 Statistical Inference).J. hence X=a.. . . duplication may be allowed. which we regard as a parameter. ..X.LIJ. According to the general theory of sufficient statistics.).}Xs(X. Since xe(Y. Pr{(J. v= 1. nz.X1JXB(XZ) . j} > 0 for all i and j. J. . Y)xs(Y. if n is constant and Pr{J} is symmetric in J. The aggregate concerned is usually called the finite population. Y) is independent of x0(X.) be the set of numbers in (Jr.... Y)> = p(I. . . . . X. c+) is a real parameter whose unbiased estimators are under discussion.X.13) Sample Survey A.. . a.=j.. (2) B. J”) after deleting duplications. . . (but may depend on Xi.}. where N is called its size.. and among J. .) (I. for the case of regular sampling . The sample size n may be a random variable.W=(j I. Y). X) is a tsuflicient statistic. . Suppose that the population consists of N units. .X. Therefore ifweletI=(I. j. which is an element of some set R. Also. Estimation Suppose that g(0) = g(cr... We briefly sketch the mathematical structure of this method without going into detail about technical problems that arise in the practical survey situation.)} =Pr{J. then the joint distribution of I and Y can also be expressed as Pr{(I. Also. Theorem. The set of all possible Q is denoted by 0 c @‘. Moreover. <I. and the observed characteristic is X. The probabilistic scheme for J is called the sampling procedure. Suppose that one unit is chosen and observed according to some procedure.j= otherwise.XhK where ifXi=aJi. X.-. I.. . PrC4=. .... it is called uniform. and hence (J. J) for all 8. Denote the whole sample by (J. Introducing a random mechanism. Y. This formula can be shortened to the form The sample survey is a means of getting statistical information about a certain aggregate from the observation of some but not all of it.... n.1) is estimable if and only if Pr{I 3 i.). Each unit has some characteristic c(. I-I:. where n is the sample size and Xi = c(~. General Remarks Pr{(J.}~e(X. ai is not estimable unless Pr{I= (1. . . It is assumed that the observation is without error. . and if it satisfies the condition (c) Pr {Ji =j} is independent of tlJ. .)Pr{J2=j21J1. < . and the observed part is called the sample.

and lim sup n/N < 1. If we can assume that tli and pi are approximately proportional. it is always possible to construct an unbiased estimator g(0) such that Pr{~(0)=g(8. called strata. the sample variance converges to c.the unknown population mean %by %*=(X/Z) x fi where x is the sample mean of the U’Sand Z the sample mean of the /?‘s. E. . it can be estimated by the same procedure as A (say A). An estimator is also called invariant if its value does not change under any permutation YE G of the numbers of sample units. This procedure is called two-stage sampling. where P. we can estimate -. Asymptotic Confidence Intervals allocation. &* is usually called the ratio estimator.=Pr{J3i}. then the invariant estimator is a function of Y (or X) only. Deming proposed an effective method in practical sample surveys. which are known and assumed to have some relation to tli. Assume that for any 0EO and ycG. as an estimator of the population total A = C fxi. If the size of the ith stratum is Ni. we have ~0~0 and g($)=g(e). we may expect that it has small error if the relation between two variables is close. . the size of the sample chosen from this stratum is n. Tukey’s hint. jN. E. When N is unknown. then for fixed cost. the variance of the estimator is minimized when which is called the condition of optimum D. If some kind of symmetry exists among the population units as well as the sampling procedure and the parameter. . the tsymmetric group).)10=00} = 1 if g(0) is e