Uploaded by Ilsa Ingenieros

0.0 (0)

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

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

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

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

Math.X. z o 8. dual group as a one-parameter group of *automorphisms which is trace-scaling. proved the equality S(. for each LE(O. are classifïed exactly by the isomorphism classes of the ergodic groups 8c of *-automorphisms of commutative von Neumann algebras 3. = e -‘T for some faithful normal trace z.(. Prior to Powers’s work only three different kinds of factors of type III..) 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. M is a factor if and only if 8. such as ITPFI and the von Neumann algebra generated by any representation of canonical commulation (or anticommutation) relations on a separable Hilbert space. The work of Araki and Woods shows that there exists only one ITPFI of type III. Takesaki (Acta Math. defined on the Hilbert space L. the crossed product of .(d) and the relation T(d) = 2z110gp(J&)l -l... In that case. 1) (the examples of Powers) as well as for Â = 1. Examples of such factors have been extensively studied by Krieger. with a one-parameter group of trace-scaling *-automorphisms 0. J. to the tenter 3 of A’ is of special importance. In particular. 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 crossed product of a von Neumann algebra . 0 <IL < 1. when J& is an ITPFI. Using the Tomita-Takesaki theory. & is of type III. &! is of type III 1 if M is a factor.. (where 0 <n < 1) and 0 (type III.e. or the set {O.. Approximately fïnite-dimensional factors of type III. if 0. Any such factor is a Krieger’s factor. G of a vo n Neumann algebra . Crossed Products The crossed product U410. had been distinguished.308 J Operator 1160 Algebras conjugacy class of Q. Woods [20] has shown that the classification of ITPFIs of type III. (which are then Powers’s factors) has been demonstrated by Connes [22].M (by *-automorphisms us.(R) induced by the translations of the real line R. < 1. and... while there exist continuously many ITPFIs of type III. Woods [18] in terms of two invariants. i.log A.M) (the set of a11 real t for which the modular automorphism @’ is inner). the set of a11 integral powers of 3. 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 (.Section J) and its application to the following structure theorem for von Neumann algehras of type III. Inst.. i. The restriction 8.36 Banach Algebras H)..&’ (acting on a Hilbert space sj) and a locally compact Abelian group G relative to a continuous action c( of G on . Connes [19] introduced two invariants. the asymptotic ratio set r. In fact the uniqueness of an approximately finite-dimensional factor of type II. if & is aperiodic and not isomorphic to the one-parameter group of *-automorphisms of L. and A is of type III.L.&‘) (the intersection of the spectra of all modular operators) and the T-set T(.). AE&? and i(h). $j) of all . is isomorphic to the original von Neumann algebra . The classification of approximately finite-dimensional factors is almost complete. 1973). Araki and E. 1973) expressed independently a certain class of factors of type III as a kind of crossed product of semifïnite von Neumann algebras with their injective endomorphisms (automorphism in the case of Connes). A structural analysis of factors of type III.e. The S-set S(. 1) (type III. namely the S-set S(.). of the type II. is ergodic. is not smooth.k’ together with the J.).LZ with the group of modular automorphisms @’ is a von Neumann algebra M of with a canonical action 0. A von Neumann algebra on a separable Hilbert space is approximately finitedimensional if and only if it is injective (. a crossed product OFa commutative von Neumann algebra with a single *-automorphism.#) of a factor of type III on a separable Hilbert space is either the set of all nonnegative reals (type III.lr with 0. given independently by Connes [ 193. 0 < 3. modulo inner automorphisms. of factors of type III (now called Powers’s factors). Res. is periodic with period . has been known since the work of von Neumann. (which is then the tensor product of the hyperfïnite factor with a($)) and of type III. which had been constructed by von Neumann in 1938 in terms of an iniïnite tensor product of factors of type 1 (abbreviated as ITPFI). The isomorphism class of d is determined by the isomorphism class of .. (The excluded case does not occur for &X’ of type Il 1. Sci. heG.N) and the p-set p(&‘) of the von Neumann algebra .(G. of 0.!j-valued Lz- .M) = r. 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. Th’z uniqueness of approximately lïnite-dimensional factors of type II. If d is properly infinite.. It is called the hypertïnite factor. along with the same number of factors of type II 1. and Araki (Publ. gE G) is the von Neumann algebra N generated by 1he operators n(A). A systematic classifïcation of ITPFIs was subsequently given by H. This class of von Neumann algebras includes many important examples. Le..

Les C*-algbbres et leurs rep& sentations. In particular.e. Academic Press.. V’i4 is called the natural positive cone. Rings of operators. Math. von Neumann. N.. Takesaki. Topological Algebras Groups and von Neumann Consider a unitary representation g+ Ug of a locally compact Hausdorff group G (. On the equilibrium states in quantum statistical mechanics. Springer. gEAut A. (G. North-Holland. Amer. Stratila and L. Gauthier-Villars. Von Neumann algebras. L. Sot. [ 161 M. homogeneous of degree l/2. J. generated by TX. Fell. [8] S. Hugenholtz. Math.. Examples of groups that are not of type I are known (. and M. A bounded approach to Tomita-Takesaki theory. Tomita’s theory of modular Hilbert algebras and its applications. independent of the weight rp). Gordon & Breach. t($))). Functional Anal. [6] I. Collected works II. Glimm. Natural Positive Cone [l] J.25]. 1979. x E &.437 Unitary Representation El. W. 1967. Takesaki. If this representation is a tfactor representation. 1964. For a C*-algebra &. 1971. Ann. Every normal positive linear functional $ on & has a unique representative <(+) in this cone (i.(B) = p(p)Bp(p)* for BE&” and pee. J3). 215-236. Lectures on von Neumann algebras. Phys. III. Rieffel and A. satisfying the relations U(g)ALJ(g)*=g(A). Trans. W*-algebras. Robinson. 187-221. 1975. Lecture notes in math. Springer. 731. Representations of uniformly hyperfinite algebras and their asso- The closure Vu of the set of vectors A. Sakai [S]). [ 131 M. 5 (1967). and is intrinsic to the von Neumann algebra & (i. Editura Academiei/Abacus Press. A new approach to the Tomita-Takesaki theory of generalized Hilbert algebras. J. Springer. Nakagami and M. 69 (1977). Van Daele.. 1979. Dixmier. [ 121 Y. Lecture notes in math. Springer. M. T. T. Sot. It is a self-dual closed convex cone. [lo] S. [3] 0. P> t(g). U(g)s(cp)= 5((PogP). Math. Springer. A tgroup of type I is a group whose factor. 1961. connected semisimple Lie groups and connected nilpotent Lie groups are of type I (Harish-Chandra. 1968. [ 171 R. G. M. van Daele. 1962). Sakai. 1979. 128.(A)<($). 1953. 1970. Takesaki. 1976. where p(p) is defined by CAPM (9) = (9. Math. n ‘%s reflects certain properties of the von Neumann algebra & for O< c(< l/2 [24. 15 (1974). A *-representation x+ TX gives rise to a von Neumann algebra A. bijective homeomorphism. A C*-algebra is called a C*-algebra of type I if its *-representations are always of type I.27]. Powers. The group of all *automorphisms of & has a natural unitary representation U(g).. and the mapping l is a concave. M. all its representations generate tinjective von Neumann algebras if and only if & is tnuclear [26. Theory of operator algebras I. Winnink. The type of this representation is defined according to the type of &. Kaplansky.(G)) is the algebra of all bounded linear operators on L. Comm. C*-algebras and their automorphism groups. Pergamon. It is known that this class is exactly the class of GCR algebras (. Dixmier..36 Banach Algebras H).q(A) for all positive A in ‘%+. 1979.(G).representations are all of type I.1161 308 Ref.. Proc. Zsido. where the second factor a@. Topological Groups). t+b(A)=(q+. [4] J.423 . 1981.. 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. [S] J. Math. G] @e is isomorphic to & 0 @L. Schwartz. [9] J. An invitation to C*-algebras. C*-algebras and W*-algebras. Pedersen. Bratteli and D. Pacific J. 378393. the type of this representation is defined according to the type of the von Neumann algebra -&’ generated by U. For example. Operator algebras and quantum statistical mechanics. [ 151 A. gE G. Amer. The duality theorem of Takesaki [2t] asserts that [A 0. Haag.(G)). References K.e. C*-Algebras and von Neumann Algebras Let a C*-algebra & be given. Duality for crossed products of von Neumann algebras. Springer. [2] W. 1961.. monotone. Benjamin. [14] R. [7] G. K. [l l] M. Arveson. The canonical action 6( of the dual G on Jlr is defined by 62. Operator on G (relative to the Haar measure) Algebras functions by where t: E L.

for example.3 of the tHamilton-Jacobi equation (. 1 Orbital elements 309 (XX. satellites. Nuclear C*algebras and injectivity: The general case. Math. asteroids. Celestial bodies to which the theory is applied are mainly planets. J.I.. 223 (1976) 19-70. Effros. Q. . Ann. Classification of injective factors. we use the true anomaly u. 25 (1973). Res. 43 (1976) 309-322. Math. [22] A. Connes. Math. Publ. 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. [25] A. J. Connes. can be deter- 1162 B.. Inst. (2) 86 (1967). (2) to determine orbital elements from observed positions of the celestial bodies. 4 (1974): 127-156.. [26] M. Sci. 1333252.. and eclipsing binaries mined by similar methods. Fourier.. which is the mean angular velocity. Duke Math. Indiana Univ. Inst. Modular theory in operator algebras. [29] S. 1). Math. R. The elliptic orbit is fixed by the initial conditions of the motion or the integration constant. Editura Academiei/Abacus Press.. the eccentric anomaly E. Takesaki. [18] H. Math. Woods. 1983. Fast. Stratila. [27] M. J. Krieger. The classification of factors is not smooth. To express the position of the asteroid on the ellipse as a function of time. and (3) to compute their predicted positions utilizing the orbital elements. Kepler’s Orbital Elements Consider. Separable nuclear C*-algebras and injectivity. Kadison and J. The period T of one revolution. [ 191 A. 104 (1976) 733115. although it must be transformed to the true anomaly or to the eccentric anomaly when we compute the coordinates of the aster- . Pacific J. photo- The size and shape of the ellipse are determined by the semimajor axis (half I:he tmajor axis) a and the teccentricity e. e. Canad.. Some properties of modular conjugation operators of von Neumann algebras and a non-commutative Radon-Nikodym theorem with a chain rule. 138-171. 1231 W. and then the position of the asteroid on the orbit is determined by the time t. we adopt as one of the main parameters the perihelion distance 4 = a( 1 -e) instead of the semimajor axis a. Sup. Ann. although orbits of meteors and visual. Connes.. 966 102. Sci. Ringrose. Ann. 24. Ecole Norm. and the mean anomaly M = n(t -to). 50 (1974). measured from the ascending node to the perihelion. (Fig. comets. north Fig. [24] H. [28] R. Acta Math. Choi and E. with p a constant depending on the mass of the asteroid. and artificial satellites in the solar system. 3099354. or mean motion n = 2n/T. A classilication of factors. Academic Press. Caracterisation des espaces vectoriels ordonnes sous-jacents aux algebres de von Neumann. Of these three anomalies the mean anomaly can be derived directly from Kepler’s elements.) The position of the orbital plane is determined by the inclination angle i to the ecliptic and the longitude R of the ascending node. Araki. Choi and E. Ann. metric. 131 (1973) 249-310. i. which is the angular distance of the asteroid from the perihelion. [21] M. Duality for crossed products and the structure of von Neumann algebras of type III.. Fundamentals of the theory of operators I. while the argument w of perihelion.7) Orbit Determination A. is computed by Kepler’s third law u2a3 = p. an asteroid moving on an ellipse with one focus at the sum. Math. and t. of the perihelion passage. Araki and E. shows the direction of the major axis. Woods. General Remarks The purposes of the theory of orbit determination are (1) to study properties of orbits of celestial bodies.55 Celestial Mechanics) and is described by Kepler’s orbital elements a.. On ergodic flows and the isomorphism of factors. Une classification des facteurs de type III. 1981. 1203 E. Ann. V. 26 (1977) 443-446. (4) 6 (1973).309 A Orbit Determination ciated von Neumann rings. Math. w. Effros. (A) 4 (1968) 51-130. (Sometimes.

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

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

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

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

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

Tokyo. Schaefer. ActualitCs Sci. 11 l139.. When a is an upper bound of X and u~X. Functional analysis. Order-preserving maps and integration processes. [ 121 S. semiordering. then sup X is also written 311 (11. then a is called the greatest element (or maximum element) of X. 1959. where A is of the form {i& 1C(n) ). III (1971). Integration. such as the inequality relation between real numbers and the inclusion relation between sets. Studies. 34 (1948). Its dual is the greatest lower bound (or infimum) and is denoted Isy g. {x 1x < a}. 8. I-111. ch. If the ordered set X is the image q(A) of a set A under a +mapping cp. Definitions A subset of an ordered set X of the form {x 1a < x < h} is denoted by (a. its dual is the least element (or minimum element) and is denoted by min X. b satisfying a < h is called a quotient of X and is denoted by b/u. Sci. Elkments de mathkmatique. Bull.X or inf X. [9] M. Acad. Sunouchi. 18 (1942). 1343. and b a successor of a. When an upper bound exists. IV. Izumi. Dunford and J. The dual concept of an upper bound is a lower bound of the subset. an element a is called an upper bound of a subset X if x < (2 for every element x of X. then a is called a predecessor of b. l-9. (ii) the antisymmetric law. 1932.). 50-58. [ 141 E. Stechert. 4477455. Springer. [ 111 N. H. this principle is called the duality principle for ordering. 1974. x < y and y $ x imply x = y. ch. McShane. denoted by < or other symbols. If a universal proposition concerning the ordering is true. 1963). Orihara.311 A Ordering [S] H.c.87~89. and a set of the form (a. When u < b and there is no element lying between a and b. For example. l-4. Princeton Univ. In this manner. or {x 1x > a} is called an interval. Hermann: ch. Bourbaki. [16] N. The Lebesgue integral in abstract spaces. 35 (1949). c is said to lie between a and b. Amer. 535-538. N. M. 1306. Nakamura.518~523. IV. j.539%542. 21~25. Banach. y. In particular. 1244b. 1953. ThCorie des opkrations linkaires. Such an element a (if it exists) is unique and is denoted by max X. H. Stone. 1965. B.12) Ordering A. H. 128la. 45p49. 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. and if the subset has a lower bound. 320-330 (Dover.b. In an ordered set A. Sot. 336-342. while x>y means that x 2 y and x # y. [lo] S. 1965. Matuyama. y of an ordered set A either x < y or y d x must hold. the duals of concepts. 1168 is called a total (or linear) ordering. x<y means that xdy and xfy. 30-32. most of the terminology associated with the inequality of numbers is carried over to general ordering. Z. Ordering The concept of ordering is abstracted from various relations. 1963. Imp. V-X. Ind. 16 (1940). h). ch.543-547. [ 131 S. Wiley. Schwartz. A set bounded both from above and from below is simply said to be bounded. S. ch. M. second edition. Springer. then the ordering < . T. then its dual is also true. and A is called a totally ordered (or linearly ordered) set. conditions. An abstract integral. I (1958). 6. US I-111. the relation between the elements of X.50-52. Proc. X is said to be bounded from above (or bounded above). 17 (1941).53-56. Suppose that we are given a set X = {x. it is also an ordering. Linear operators. 1937. J. h). it is called the least upper bound (or supremum) of X and is denoted by 1. 9. 117a. If for an arbitrary pair of elements x. l-4. Hildebrandt. 5. Yosida. 1964). Ann. or simply order) if the following three laws hold: (i) the reflexive law. More generally. Incidentally. A set X with an ordering between its elements is called an ordered set (partially ordered set or semiordered set). ch. Press. The binary relation > is called the dual ordering of <. order relation. S(c) = {x 1x cc} is called the segment of X determined by c. 7. Nat. Proc. and G. [ 151 T. 1967. Math. x <x. x < y and y < z imply x <z.X or supX. Theory of the Integral. according to the definition of <. Banach.. A subset of an ordered set X is also an ordered set with respect to the same ordering as in X.u. x < y is equivalent to the statement x < y or x = y. and (iii) the transitive law. When u<c<b or b<c<u. Math. is called an ordering (partial ordering.b. Integration in abstract spaces. Saks (ed. 59 (1953). 1969. it is said to be bounded from below (or bounded below). Notes on integration. (Chelsea. Acad.483-490. A pair of elements a. [ 171 K. We sometimes write x < y as y > . second edition. If there is a least element in the set of upper bounds of X. and propositions concerning an ordering are defined by replacing the ordering with its dual. Warsaw. Banach lattices and positive operators.1. and > is the dual of <.

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

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

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

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

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

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

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

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

cp... are equivalent to the conditions U.. .. i=l .b].*[y]=O.y”-l’(b).+.U. uiCYl=“.i=l. and hence for expansion f=a. r are continuous and p(x) > ( [a... aY(4 + PJY’(4= 0. m*. ( tion q.. . Let L*[y] be a formally tadjoint differential operator of L[y].n.f on [a... i = 1. 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. The boundary value problem where G(x. A) such that the conditions L[y] =. .y’j-“(a)+ t j=l Iv. . . 5.. the linear boundary value problem defined by LCYl=f(x).a linear ordinary tor L defined by w. i=1. b]. I) is called the Green’s function of (5). m. VCYI=Yi..(x)#Oforanyxe[a.. . For any value of 1 that is not an eigenvalue. i = 1. b). y.[y]=O..i=l. s Ia 5M5) dt. Given a function f(x) and complex constants yl. Y. The function G(x.390 Spectral Analysis of 01 C. ui” CYI =o.here pk(x) is a complex-valued function of Aclass C”-’ defined on a compact interval a<x<bandp. . i=l .. Then (i) the eigenvalues ing sequence tending to +co. Linear C0nsider. . . Nonlinear Differential Equ: Boundary value problems for 1 ferential equations are very dif sults are obtained only for equ form. h. L natural way operators in the H consisting of functions that are grable in a XX < b. value problem (3) is said to be an adjoint of boundary uYl=o. then the Fourier expansion uniformly to . then the system {cp. and there ex: adjacent zeros of q.. we have J. L.. .} i. b] with weight functi We say that the problem (4) is self-adjoint if L[y] = L*[y] and the conditions Ui[y] =O. The boundary value problet order equation (P(4Y’)’ + M4 + W))Y = 0. i = 1. The boundary value problem containing a parameter 1 LCYI = lY.. A set of m* linear boundary conditions 7JF [y] = 0. m*.. and the gen based on operator theory in H (.. y...m. for example.. A) of (5) and th tion G*(x.ly+J Ui[y] =0 are equivalent to y = c G(x. b) the tParseva1 equality if S is a function of class C” sal 0. . (4) is called a Sturm-Liouville pro1 r are real-valued functions deli and CI....(x) associated with A.. the SI equation Y(X) = a * W. are real constants.(a. .}. <. Y” =f(x. YCYI =O* i=l .by Ui[Y] = f j=l M. .9n. m*. 5. is said to be an adjoint boundary condition of U. When f(x) = 0. ..i=l. . YY@) 6Y’ + L*cYl=o. b]. .bL[y]pdx= fiyL*[y*]dx. .. and any function y* of class C” satisfying Vi* [ y*] =O. If 1= 0 is not an eigenvalue. (5) admits nontrivial solutions only for special values of 1. ! defined in an open interval -0 b < co and pk is of class Cnmk.+a.. n.~. We define a system of linear boundary operators u l. m... i=l. I q. then (5) is equivalent to When the coefficients pO. &II) of L*[y] =ly. yi = 0. otherwise it is called inhomogeneous. . Such values of 1 are called the eigenvalues (or proper values) of (5). l)f(<)d<.m*. . have the relation G(x.1177 315 c ODES (Boundary Differential Equations differential opera- Value ProbIN B. For I function G(x. Consider. . the problem is called homogeneous.(x) a zero ( (iii) the set of eigenfunctions is set on [a.. 5. 0). 5. 5) = G(x.. (2) is a two-point boundary value problem. orthonormal set in the Hilbert consisting of functions that arc grable on (a. there exists a unique function G(x.. fi. if for any function y of class C” satisfying Ui [ y] = 0. . m. zeros in a <x <b. 5.. . Y’) . and the corresponding solutions $0 are called the eigenfunctions (or proper functions) of (5).

II. oj(x). . Suppose that the fi(x. Naimark (Neumark). y(b) = theorem has been proved: Suppose that f(x. . .uj)(D %$x) fj(x. y’“-‘1) the conditions y(u)= b. A. means that the integral curve represented by it passes through the point (a. then a in the (n + I)-dimensional solution of (1) represents a curve in this space called a solution curve (or integral curve). The problem of solving y’“‘=f(x y(uj)=bj. .))~Oforyj=wj(x)andw. . dy. . . Aml:r. .y”) = 0. n. the solution is obtainable by the method of successive approximations. Levitan and I. . Then there exists a solution y(x) such that yj(uj)= bj and gj(x) <y(x) < ai( This theory was applied by M. .) [S] N. General Consider equations Remarks a system of ordinary differential . and that if 0 < 1. (Original in Russian. < 1. A. in general. k #j. Levinson.Y. . The statement that a solution satisfies initial conditions yi(u) = bi. j = 1. and their derivatives aJ/ay. Hartman. then there exists a unique solution of (1) satisfying the conditions yi(u) = bi. i=l. .. Transl. . The problem of finding solutions that satisfy initial conditions is called an initial value problem (or Cauchy problem) If we consider (x. y. 1975.315 Ref. k = 1. u”(X) >f(x. Then (6) admits a solution y(x) such that y(a) = A. 1968. b. .=&. . . y. . for the equation y’“)=f(x. satisfy (x-aj)(D’gj(x)-4(x... Dunford and J. y. A. and w(u)<A<o(a) and o(b)<B<ti(b). y’) is continuous for a < x <b. co(x)<y<~(x). is reduced to Hukuhara’s problem by a suitable change of variables. Coddington and N. Since.) as the coordinates of a point space R”+‘.. . . The problem of finding a solution such that yj(uj) = bj.. If in addition f is an increasing function with respect to y. 1969.). . O’(X)) for a < x < b.. b.Y. The boundary value problem y”’ + 2yy” + 2/l&2 . n. Consider the system of differential equations Y. and satisfy gj(uj) < bj < wj(aj).). all de& nitions and theorems concerning the system (1) can be interpreted as applying to a higherorder equation. Sot. be continuous and right and left differentiable functions and gj(x) < Wj(x) for a <x <b. y(b) = B. y’(u) = b’.. 1967. . .. . k#j. Linear differential operators I. i= l. .. are continuous in a neighborhood Iof a point (a. [Z] P. . M. satisfy (x . 133 E. . where i and k are constants. . called Hukubara’s problem. b. . b. . n. .) are continuous for c(Q x < p and wL(x) < y..~. The following I”& Yt Y’)l c MU +y?.Y. 1963.. y’“-‘)(n)= b’“-‘) constitute initial conditions.3) Ordinary Differential Equations (Initial Value Problems) A.. . 6. .. For example. McGraw-Hill.)) ~0 for yj=pj(x) and o. ODES (Boundary and boundary 1178 Value Problems) [4] M. Addison-Wesley. . y’bbk (x+m). y. reduces to the initial value problem when the uj coincide. S. Y(0) = Y’(O) = 0.y. n. b’. .. Interscience. . ~1.. 1954. . appears in the theory of fluid dynamics. . . (1) >y >y’ . the problem has a solution.. . Then a system of functions (yl(x). . y. . under suitable conditions... [6] B. y. $Q(x). the solution is unique. 1955. we4 o’(x)) and a” <f(x. References [l] E. . Monographs./ n.. Moreover. Cauchy first gave a rigorous proof for the existence and uniqueness of solutions: Iffi.. n. Sargsyan (Sargsjan). Theory of ordinary differential equations. 1964. Introduction to spectral theory: Self-adjoint ordinary differential operators. . b(“-‘) are initial values. Wiley..< b. 316 (X111. Ungar. Math. y’“-1’) j= 1. W(X). T.) conditions y(a) = A. then there exists a uni#que solution satisfying given initial conditions. i= 1.(~)~ yk<<wk(x). b. -co<y’< +co. .< ok(x). n. .. . and the values a. n. . and the values a. initial values. (Original in Russian. .. i=l. Hukuhara to the study of singular points of ordinary differential equations. Hille. and B. L. If f and its derivatives dflay(‘) are continuous. Lectures on ordinary differential equations.ldx=fi(x. j= 1.(x)<Y.n. b. .y. y’.). j = 1.(x:l) is a solu- . 1970.. These conditions are called initial conditions... Linear operators II. of Math. Schwartz. The following result is a generalization of +Perron’s theorem: Let Q~(x). It is known that if I > 0. . ..Y. the solution is unique. and Q(X) <y(x) < W(X) for a <x . y. we can transform a differential equation of higher order into a system of differential equations of the form (1) by introducing new dependent variables.. Ordinary differential equations. .). Suppose that the d are continuous.

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

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

.. x0.(x)} is said to be orthonormal. Proc. yO) is holomorphic with respect to yO. Ein neuer Existenz Beweis fiir die Integrale der Differentialgleichung y’ = f(x. A. if (3) has a solution y = v(t) defined in 0 < t < t. [3] P. 1976. [2] E. v(t)) for 0 < t < t. y).2. Ann. 24 (1943).(n=1. [6] M. is holomorphic at (a. Wiley. Fat. a. Benjamin-Cummings.-Math.(X).(x) with respect to the norm in L*(X). [9] R.} E O(X). we define the inner product (l. then c. Nagumo. .)the References [l] E. Abraham and J. is normalized. If fi (a. b) #O.(x -a)” is a majorant series of 1 c. is holomorphic on C except possibly at x = a. and we write {f. [S] 0...(x -a)“. A). Univ. y. 1) is holomorphic at (a. then we simply say that they are orthogonal. If a solution q(x) is holomorphic on a smooth curve x = x(t). y. I.. b). and satisfying y(0) = b. then the solution of (2).)b.. . x0. Univ. then the solution y =9(x. Okamura.} be a set of linearly independent functions in L*(X). satisfying y(x. Sur la theorie des equations differentielles ordinaires.)/llfn11*. and p(a) = b. We have the following uniqueness theorem: Suppose that f(x. 1) is continuous with respect to (x. y) is holomorphic at (x(t). where fi and fi are holomorphic at (a. If X is a subset of a Euclidean space and p is the tLebesgue measure m. fi(a. b). y. Suppose that f= fi/fi. VII. McGraw-Hill. 1930.(X) if and only if the system is total in L*(X). [8] E.2. yO) of (1) satisfying y(x. then we say that f and g are orthogonal on X with respect to the measure p. satisfying y(0) = b.} E O(X).. we say that they are orthogonal with respect to the weight function q(x). 1955. b). then y =&(t)) is a solution for 0 < t<t. 1958. Kyoto. Differentialgleichungen reeller Funktionen. If {f. Kamke. 1978. If f is continuous with respect to (x. such that ak+a and ~(a.(X-up.u) be a tmeasure space. of Y’ = X’WW)~ Y) (3) tielles ordinaires sans points de Peano. then the solution ~(x. E. [7] M. and x(O) = a. Phys.g)=jxf(x)g(x)dp(x) and the norm Ilf 11 =(Jf)“*. {a. Marsden.) at (a. if any pair of functions in the set are orthogonal. If (cp. ye. Condition necessaire et suffisante remplie par les equations differen- .y) admits a unique solution such that y-b as x-a. Foundations of mechanics.) = y.Wecallthec. x0.fn) = 0 for all n implies q(x) = 0 almost everywhere (a. y) and holomorphic with respect to y. For complexvalued functions A g on X belonging to the tfunction space &(X). Conversely. If(Xg)=O. and if f(x. we say that {f. pt. a. the analytic continuation of a solution continues to be a solution if it does not encounter any singularity off. 1) and holomorphic with respect to (y. ) is said to be an orthogonal system (or orthogonal set). If we can approximate any function f o R arbitrarily closely by a finite linear combination of the f.x. J. and this solution can be expanded into a tPuiseux series: y= 5 C. Coddington and N. 1).(x)} (n = 1. If f(x.. II=0 If f is holomorphic at (a. The orthogonal set {f. (A) 24 (1941). then the series Czl cnfn(x) is called an orthogonal series. 1) is holomorphic with respect to (ye. then the equation y’=f(x. Sci. A). An orthogonal system {f. y) f 0. By a ttheorem of identity.1181 317 A Orthogonal Functions which shows that Z C. b). 5. 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. Hukuhara.). then f is said to be normalized. If IIf II* = 1. vol. and f. Sci. 1964. Mem.. x0.} eO(X).).} is total in R.21) Orthogonal A. Wiley. Ordinary differential equations. Hille.} is complete in L. then ~(x.. [4] H. =(q.. Then q(x) is holomorphic at a. Orthogonal Functions Systems O<t<t. sec. 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. with 0 < t < t. Math. 15 (1945). then { fn} is said to be complete in L*(X). Japan. y = ~(x. Coll. Uber die Lage der Integralkurven gewiihnlicher Differentialgleichungen. Levinson. y. b. Let {f. y) is holomorphic at (a. A set of functions {f. Tokyo.b)=O. Let (X.f.e. Hartman.(a. Perron.) = ye is holomorphic with respect to (x. and there exists a sequence of points on C. Ordinary differential equations in the complex domain. . Akademische Verlag. Theory of ordinary differential equations. Let {f.}. and let R be a subset of L. ye. Suppose that x is a real variable and y is a complex vector. If f(x. b. second edition. and we write { fn} E ON(X) if each f. &). If the series converges to q(x) tin the mean of order 2. A. Sot. then (1) has a solution (p(x) holomorphic on C and y(t)=&(r)) for 317 (X.

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

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

When a driving force term q(t) is present in the right-hand side of (2). . t) 3x2 ’ If E<O in (2) (negative resistance)> the solution (3) increases in amplitude. The theory of vibrations has its origin in the study of mechanical vibrations. resulting i:n an automatic generation of oscillation. which is just the superposition of tb. However. x becomes zero at a fixed interval n/a. C. The difference between the greatest and least values of f(t) (globally or in an interval) is the amplitude. Nonlinear Oscillation u(0. 27c/(~ is called the period. the existence of which has recently been confirmed. Perhaps the simplest case of such an oscillation is represented by the differential equation d2xJdt2+n2x=0. too. a parametron is an electric element utilizing parametrically sustained vibration. are the parametrically sustained vibrations caused by periodic variation of a parameter of the vibrating system. the forced oscillation where the restitutive force is proportional to the displacement from the equilibrium position.g. the solutilsn takes on the additional term B. we mention. k=l i=1. As examples of practical applications of the theory of oscillations. (1) -COSCG s which represents do cp(t)@sin otdt > due to . 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). other than forced oscillations and self-excited . As a limiting case. tunnel diodes). and the extremal values in the intervals decrease to zero in a geometrical progression with the common ratio u = exp( .e fundamental vibration (corresponding to k = 1) and simply harmonic motions of frequencies equal to multiples of the fundamental frequency. General Remarks The solution 1 A. Here the amplitude is A. Actual vibrating systems contain more or less nonlinear elements. Linear Oscillation Periodic solutions of tlinear differential equations have been studied in detail for a long time.. This is called harmonic oscillation or simple barmanic motion. The solution is given by x = A cos(nt + CC). m.vibrations. which give rise to various kinds of oscillations different from those described by the linear theory (. If a resisting force proportional to the velocity is acting.1 4) Oscillations A. where the number of degrees of freedom is infinite. t) = 0. On the other hand.t + a.) is said to be in free harmonic oscillation if the coordinates can be expressed as xi = 2 Ai. such a situation often occurs when the vibrating system has time delay characteristics (. we have the vibration of a string: wzn2 at= i?=u(x. investigations concerning the free oscillation of the earth. in engineering. . In this case. Each of these simple harmonic oscillations is called a normal vibration.). Among sustained vibrations.2 . a=Jn2-Ei (3) n>E. This phenomenon is called self-excited vibration.163 FunctionalDifferential Equations). but its nomenclature has been used also for electric circuits. Exactly periodic oscillations are studied in the theory of tperiodic solutions of differential equations.Z&/C. the equation becomes d2xldt2+2edxJdt+n2x=0.. and its reciprocal the frequency. the period is 2744 n is the circular frequency. (2) is not periodic.318 A Oscillations 1184 318 (XX. either exactly or approximately. whose solution x= AC”‘cos(ot+cc). Besides being caused by some special kinds of circuit elements (e.290 Non- .. x.. t) = u(/. Electric wires and pantographs for use in high-speed electric railways must be designed to prevent unwanted parametrically sustained vibrations.ns/a). The tperiod of a solution f(t) is called the period of the oscillation. and in geophysics.cos(n. and x is the initial phase. sin(knx/l) k is given by a series cos(knnt/l). so that a small disturbance is amplified. A system of m tdegrees of freedom (x.. the prevention of vibrations and the generation of stable sustained oscillations. This phenomenon is called damped oscillation with damping ratio u and logarithmic decrement log u = . A vibration or oscillation is a phenomenon that repeats periodically.

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

.

.

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

.f.ax..p~o=zo... Characteristic Manifolds We consider a partial differential the nth order of two independent x. x..1. The same procedure is applicable to solving other systems of partial differential equations..) that satisfies the given partial dXl>X2.t=azzlay2 when2 is a function of x and y. . Y) asap =Pjk(l). .~.. (The partial differential equation becomes an ordinary differential equation if the number of independent variables is one. .P. Find a (6) and consider the following problem: solution cp(x. . . .. If F vanishes for a system of values x0.> azz azz G’ax. B. Obtaining such a solution for a given partial differential equation is called solving this equation.. .+x”g=o 1 2 ” is equivalent to solving differential equations dx. x2 -dx. . its tpartial derivatives..x..iff. . .. r=aZZlaxZ. and y for x2 when n = 2.~=a2z~axay.. Usually we write pi for az/axi. 2 = @(jr... . Y) ..+ 1 has an integral then we have .. 1 2 ” (4) which is an equation of type (1).‘. x2. and by analogy to the case n = 2.291 Nonlinear Problems). Y)Z = G(x. x.dy”-PP. . x for x1. xi. . x.> Equations Remarks differential equation is a functional a2 X”‘Z. p$(j. Pjk=Pjk(A)t n-l.+P”~+R~=0... .1189 320 B Partial Differential Equations 320 (XIII. . >.‘“’ > Inotherwords.dx”#O (7) dx.k=O. x2..2. j+k<n- 1. y”.. P2. +P”&=R 2 ” hypersurface V(z. the equation &Y)~+%Y)s+ + % C(x.. y”. y: F(~. the integral hypersurface of the equation is z-fp(x. .. where pjr = g$. With this equation.) The torder of the highest derivative appearing in a partial differential equation is called the order of the partial differential equation.~on)=O.~+.P~~..... . x =x(n). and p = &/ax.. differential equation is called a solution of the partial differential equation. A partial differential equation is called nonlinear if it is not linear (.1. . and if the quasilinear partial differential equation (Lagrange’s differential equation) =o that involves a function z of independent variables x1. Y=Y(a)v j.k=0.o. if x.. are functions of n independent variables xi. . . pjk are holomorphic functions of 1 in a neighborhood of 1= 0 and take the respective values x0.dxdy”-‘+ .. then for an arbitrary function a’. R are functions of independent variables x1. For example. P. and if P. Example 1. . General A partial equation l>.k=O..Section C)....y)p p. y. j. we associate a manifold defined by a real parameter 1.. P. and the dependent variable z.1 9) Partial Differential A. f.. Xl the system of ordinary on the curve x=x(n).. .. we define solutions in the same manner.ax. .. . . ..2 . A partial differential equation is called linear if it is a linear relation with respect to z and its partial derivatives. At+ W. (5) is a linear partial differential equation.-. X2. .Y.~+. From this we can obtain a general solution by the method of example 1. ..) = 0. y) of (5) that satisfies aj+kdx. n-l. x. Example 2. Then solving the partial differential equation x. +(-l)“Po.j+k<n)andis tholomorphic in a neighborhood of this system of values.. If PI. . and the independent variables x1. equation variables of y)q + F(x. y = y(L). X. .PI~..) XJ’O. X (2) . . X.-iaren-lindependent tintegrals of (2). . l.x.~+p2~+.. q = &lay.+P2.fn-J is a general solution of (1) (. . The definition of a system of partial differential equations is similar to that of a system of tordinary differential equations. j+k<n-1 For a system of partial differential equations. We call this problem the Cauchy problem for equation (5). pi at 1= 0. A function z = x. . .z. A partial differential equation is quasilinear if it is a linear relation with respect to the highestorder partial derivatives.n. .

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

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

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

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

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

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

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

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

In particular.) with real coefhcients and B is a differential operation in (x. equation (7) is said to be of hyperbolic type (. respectively. For nonlinear equations there are. 11 (1959)) succeeded in obtaining the proof for the exceptlonal cases. y.1 (n is the number of independent variables) is not tcharacteristic for the given equation of order m. 5) are functions of (x. Calderon proved this except for the cases k b 4. The proof of this fact relies on the CauchyKovalevskaya existence theorem. Mizohata. 5. 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 we omit the assumption about the eigenvalues. however. Moreover. Uniqueness of Solutions ik+ 5 qx. Mizohata (J. if an analytic manifold S of dimension n . there is at most one solution of class C’ for the initial value problem. Calderon showed that Carleman’s result can be extended to the case n > 2 (Amer. bpy = 0. Shirota. He proved that if the eigenvalues of the matrix (a. Hormander [4] for an extension to the complex coefftcient case. Hence Petrovskii’s theory guarantees global existence of solutions.... there is only one solution for the original initial value problem with continuous partial derivatives up to the order of the equation. as the hyperplane on which the initial values are assigned and proved the existence of a solution on &r < Ix1 1<Ed. J. 1901). We write u(x. The same method can be applied to nonlinear systems of the first order if . For example.. Math. their derivatives are Holder continuous. and so on. <)= ~.325 Partial Differential Equations of Hyperbolic Type). it is also well posed for the past. are of class C2 and the b. 80 (1958)). 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. If the characteristic equation of (W.()lk-j=O. The method of proof of this theorem is as follows: First. then the Cksolution of the Cauchy problem is unique in a neighborhood of x = a..:hey are special types of quasilinear systems. We assume that the coefficients of 4(x.. Kumanogo discuss the uniqueness theorem for equations of double characteristics or of parabolic type. and we can take T arbitrarily large provided that (10) is satisfied. in general. Then we regard x = E.a. See L..5)+ . and H. In PetrovskiTs theory it is sufficient to assume that the coefftcients in (7) are continuous and bounded. Math.321 F PDEs (Initial 1198 Value Problems) A.(x)5”=h(x.1 coincide with given functions on the manifold S.. t) is a homogeneous polynomial of degree j of 5 = (lr. very few results about the global existence of solutions. because we have a counterexample due to A. This result can be extended to systems of equations under similar assumptions. there is at most one solution whose derivatives of order up to m . G. In this case. . m = 2.327 Partial Differential Equations of Parabolic Type). 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.) are all distinct. PliS (1954) where all coefficients are of class C”. Sot. Construction Expansion of Solutions by Asymptotic Let a(x. n = 2. y. P..u~(x)Da be a linear partial differential operator of order m. If system (7) is of normal form and is well posed for the future. Carleman (1939) about the system: has only distinct roots for 5 ~0. the theorem does not hold in general. if the coefhcients are of class C”. This is the condition obtained by Petrovskii. then the solution of class C’ satisfying the prescribed initial conditions is unique (Holmgren’s uniqueness theorem. . Consider the following linear partial differential equation of the kth order: where C and L are constants independent of 5. then even when some of the eigenvalues are complex. where Pj(x.y. F. D) = &~. An example that is not of a normal form and is well posed is a parabolic equation (. Japan. where a certain topological difficulty arises. then we get a global existence theorem. y) of order at most k . with h I and h’ homogeneous in 5 of degree m and m 1.. T. Therefore. and the coefficients of B are bounded and continuous. j=I If a linear partial differential equation of the first order of normal form has coefficients of class c’“. if the number of independent variables is 2 and the coefftcients are all real. Let K : Q(X) = 0 be a regular .. S. we prove the existence for 1x1<Ed and sufficiently small sI by Picard’s tsuccessive iteration method. S. are continuous. y) of class C’. where the a. with coefficients of class C”.1. then we have a result of T.5)+h’(x.

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

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

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

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

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

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

of L[u] =f(x) we set where f(x. concerning the solutions of the equation Au + cu =0 (c > 0. then the solution u of Au =f(x. the Dirichlet problem for Au = 0 has been studied in detail (. 5) is called a fundamental solution (or elementary solution) of (1) or of L if u(x)= sG E(x. Roughly. and J. a solution u does not attain its local negative minimum in G.<O or if c(x)<0 and f(x) < 0 in (I). y.‘Jaologr (n=2). 5) = 0 except at x = 5.4] for Hopf ‘s maximum principle and Giraud’s theorem). c).Y)’ -& KkYi G ss 5. df(5. A function E(x. 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. Furthermore. 5)= ! 10g(~(xi-&)z)l’2=Iog~. a solution na3. 5)f(5)& provides a solution of (1) for any foci with compact support. n=2.. Let G be a tregular domain in the plane.‘Jaor’-“(n>. if u is a solution of Poisson’s equation Au=f(x). is the tBesse1 function of order v. w.345 Pseudodifferential Operators). y) vanishing on I is given by U(X. Now. u(x) defined by then the function then E(x. concerning the uniqueness of the solution. n=2.[3. n>3.1205 323 C PDEs of Elliptic Type Thus. 5) and put of this equation by where I is the tgamma function. we have a representation of u(xO) by replacing Au by f in the integrals just given. E. because the integral equation in rp as above has a unique solution expressible by Neumann series if the domain G is small enough (. t.3)oro. 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).)) and r= (Caij(~)(xi-&)(xj-~j))1/2 and (a’j) is the inverse matrix of (a. From this it follows that the solution of the Dirichlet problem for (1) is unique if c(x) < 0. If c(x)<0 andf(x). The first and most general is to use pseudodifferential operators (. w.189 Green’s Operator). <)$@)/w” is seen to be a fundamental solution of L. In particular. y) satisfies a tHolder The Dirichlet problem for . c)] =x(x.. A fundamental solution E(x. C. Levi [Z] and is as follows: Let Let G be a bounded domain with boundary I.120 Dirichlet Problem). which uses the geodesic distance between two points x and 5 with respect to the Riemannian metric ~aij(x)dxidxj. n’ = (n . there are three different methods of constructing fundamental solutions. where a(t)=det(@(l. then the solution of the Dirichlet problem is unique. The third method is due to E. if we put (~(Xi-~i)z~-n’*=~2-n. 5)= To obtain r2-n { -lo. Thus Levi’s method enables us to construct a fundamental solution locally by successive approximation.=27FW42) represents a tparticular solution of Au =f(x). Consider now the more general case (1). The Dirichlet Problem n-2 c#.). <) as a function of the variables x satisfies A V(x. Hadamard [ 11. 5) = . constant). where V(x. This is important in applications of elliptic equations to geometry.y(x. Next. 5) is a solution of the equation L[E] = 0 having a singularity at x = 5 of the form -o. If we denote tGreen’s function of A relative to the Dirichlet problem in G by K(x. we obtain the following tintegral equation of Fredholm type in q(x): If we denote the tresolvent ir(x. ?be&> condition. The second is that of J. r/).2)/2. the following relation holds: Writing L[V(x. The maximum principle is one of the most powerful tools available for the treatment of elliptic equations of the second order with real coefficients. This is called the strong maximum principle (.

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

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

and let called Dirichlet’s principle. ss with the given boundary condition satisfies the Euler-Lagrange equation (of type (4)) (pi = au/Zxi) and the boundary condition as well. 0. v0 is the outer normal of unit length at the point x E S. For elements f and g in H”(G).u. Courant. and others to show the existence of solutions for linear self-adjoint elliptic equations.. In general.(~~~. Haar.(x) = . When u satisfies L[u] =f (MEL.. Here the necessity is easily derived from Green’s formula. Furthermore. .. R. then the function that minimizes the integral ^I J= x . cp are continuous. . p > 0. Nikodym.=ug+(p-b)u.) satisfies C(ZZF/(ipii?pj)~i[j>O (and F has some regularity).+. n dx. also 121. x.f.. G. the solutions of the (Ldm= JG c D”f(xP”g(x) dx.. Also..cp). the condition that the solutions must satisfy at the boundary is called the boundary condition.i-a. Then the second and third boundary value problems admit one and only one solution. In general. and others. u.. IY[v. .?I.. Hilbert. and the third boundary value problem (or Robin problem) when /I $0.1. the case of tminima1 surfaces) has been studied by A.g 1 . and at least one of c and p is not identically 0. ... The case where F is a function of p alone (in particular. particularly for F=(1+p:+. and v is the conormal defined by cos(vxi)= -f aijcos(v. 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. F. ax?’ Icrl=a.. j=l i= 1. where b=i=. uijcos(voxj))2)‘~2. account of the boundary condition . .323 F PDEs of Elliptic 1208 Type second boundary value problem are determined uniquely up to additive constants. the coefficients of L and f satisfy Holder conditions.p. .. let H”(G) be the space of functions in L. 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.. lal<m where alnl DE= ax. .-~ f 21 Then if the boundary S of G is of class C’ and . . With respect to this inner product.). @dS=(f. G. and cp and /J’ are continuous [3.) and q(x) is an arbitrary element of H’(G). 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”). Serrin.1 .168 Function Spaces). . 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. . 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. Rado. in order that there exist at least one solution u of the second or third boundary value problem relative to L[u] =..dx. s s av Taking on u: where b.(&I. let M be the tadjoint operator of L. This is also an elliptic partial differential equation in u.. . Green’s formula yields +(c(x)u. H”(G) is a Hilbert space.xj)/a.Z. in boundary value problems...q)+ ua”. If F(x.+p~)1’2inthecaseoftheminima1 surface equation.cos(v..xi)[h. Assume that G is such a domain. The theory of +Hilbert spaces is applicable to the boundary value problems in Section F..(G) whose partial derivatives in the sense of tdistributions up to order m belong to I&(G). was used by D. x. under the assumptions that G is a domain of class C’. (. H. Weyl. When c = 0 and p = 0. . Method of Orthogonal Projection (6) where a=(~~=. T./~X. c GO..L it is necessary and sufficient that fvdxsG cpvdS=O.pr..(x) = b.

hence the L2 a priori estimates play a fundamental role: If u E H”(G). A.l). is a continuous mapping from L.)[u]=GJ + -(Jcp). Variational general boundary value problems have been treated by D. Now. In particular.. by G.. Under certain algebraic conditions (Shapiro-Lopatinskii conditions) on (L. This means that u(x) is a solution of L[u] =f in the sense of distributions. Rellich’s theorem yields that G. 1. XES. Now. I-. the boundary S. where L* is the tadjoint operator of L. Under these estimates the boundary value problem is said to be coercive.189 Green’s Operator). there exists a unique solution u(x)~Hl(G) [12.(G) into H’(G).D)u(x) = cpj(X). 49 (1970)) and others. These problems are formulated as follows: Uul =fM Bj(X.. and mj are the orders of Bj (compare (9) with (2)). 1. Fujiwara and N. So we can apply the tRiesz-Schauder theorem (. a m Thus the problem is reduced to finding U(X)E H’(G) satisfying this equation for all (PE H’(G). {Bj}). Douglis. Shimakura (J.S+ Il”llb (9) where K is a constant determined by (L. which acts on the equation from the left.b(=m/% (8) Conversely. and /?. M. Schechter [17]. If t is large. 15-51) and others. This equation can be regarded as an equation in H’(G). G). if u(x) satisfies J!. aatqav + flu = 0. is a tcompact operator when it is regarded as an operator in L.[u] =f(x) and f(x) belongs to H” on any compact set in G. the boundary value condition is stated as @u/&j =fj(x) (j = 0. u(x) is a genuine solution. it can be shown that if we assume smoothness of the coefficients. and we call such a u(x) a weak solution.t for a large t. Pures Appl. 1954. we have (I+@+ t)G. j=l. denoting its inverse. If we put m=2b. In general.S is the norm in Hk(S). For systems of such equations. .(G) considered above.168 Function Spaces). In this case. If necessary. if Re. then u(x) belongs to H”+” on every compact set in G (Friedrich’s theorem [ 1. the problem is reduced to the displayed equation in L. .(G). Bj. by replacing c(x) by c(x) . Princeton Univ.168 Function Spaces). Browder (Ann. for a solution of this functional equation. . where v is the normal of unit length at the boundary. the theory of interpolation of function spaces are also used [lS] (. (L . if we apply Green’s formula.(G). we have (u. L.13]. Agmon.we is called an elliptic operator ({ # 0).. sBu(pdS= s c>O. b . where the Bj(x. D) are differential operators at the boundary and f and {cpj} are given functions. Such a treatment may be called the method of orthogonal projection. . H. math. In particular. studies 33. and others. . a. we can show that for any f(x)~&(G). we introduce the complex parameter 1 and consider the boundary value problem (L + nr) [u] =L f~ z. Math. Girding studied the Dirichlet problem for strongly elliptic operators [14]. Elliptic Equations of Higher Order The differential operator of order m: . Giirding’s inequality unfix. Press.2 . we can see that u satisfies the boundary condition a au/h + /Iu = 0.iI lIBjullm-mj-(1/2).. following Weyl. z dW>cW. ). since G is bounded and G. E. Thus. if we take (P(X)E 9(G). plays an important role. the problems are treated also in H”(G).51). for an elliptic operator L defined in an open set G. Ilk. In other words.. there are works by F. II . In this treatment.(x)<’ #0 a aujav+pu=O.t1) is a one-to-one mapping from the domain B(L)= {u~H~(G)~a&@v+~u=O} onto L.. (7) where 6 and c are positive constants. we see that u(x)~C~(@ by tSobolev’s theorem [12]. General boundary value problems for elliptic equations of higher order have been considered by S. Next. this boundary condition means that the solutions belong to the closure fib(G) of g(G) in Hb(G) (. and L. then the solution U(X)E H’(G) belongs to H’+‘(G) when ME H”(G) (s = 0.(c). In this case. since the solution u(x) contained in L2(G) (hence also contained in g(L)) satislies the equation and the boundary condition. .. Nirenberg [16]. is called a strongly elliptic operator.. In applications. L*[q]) = (J rp). m is even.1209 323 H PDEs of Elliptic get Type if &=. Thus. Using the notion of function space. when s> n/2. ll”llm~K(llLull +.

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

Schauder. Kyoto Univ. Shimakura (J. [S] J. Baouendi and C. C. Minakshisundaram (Canad. the Dirichlet problem (1) and (12) (with g = 0) has a weak solution UE LP(G) for any f~ LP(G). Differential Geometry.115 Diffusion Processes). Wiss.. Camus (Ann. 4 (1964)). Akad. G. 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 (. (1) on Z2U&. N(T) is no more than O(T”im) if L is of degenerate elliptic type (C. be the set of x E I at which L is not degenerate in the normal direction. i. 10 (1972)). Leipzig. H.. 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. Sulle equazioni lineari totalmente elliptiche alle derivate parziali. Let N( T)( T> 0) be the number of eigenvalues of L smaller than T. Anal. Prentice-Hall.. 1970. Mat. M. J. Scuola Norm. axiaxj (13) s S”-l where a(x. 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. Kyoto Univ.Wesetq=p/(p-l). McKean and I. Numerische Abschltzungen in N. Schauder. +a~. 1 (1967)) treated the case of manifolds and discussed the geometric meaning of this formula. Moser proved that the above inequality is a corollary to a Harnacktype inequality (. and others. L. Math.1211 323 Ref. Ark. [4] M. see also S. where b(x) is defined by b(x)=-i$ v. Circ. v. Palermo. Weinberger. if L is of constant coefftcients. Formula (10) was at first established by H. WI where f and g are given functions. Letl<p<co. Grushin [29]. . Arima (J. Lectures on Cauchy’s problem in linear partial differential equations. F. Press and Oxford Univ. Carleman (Ber. 88 (1936)) studied the behavior of the trace of the Green’s function of zl-L as lzl+ co in the complex plane (. let Z. 9 (1969)) P.. U Z3 satisfying L[u] u=g =f in G.. denote by v(x) = (vl (x). Math. normal vector to G. [2] E.Wealso Put c*(x)=cw~l~+ n abi(x) . E.-Phys. Partial differential equations of elliptic type. Hadamard. 5) is the tprincipal symbol of L and S”-’ is the unit sphere on R”.. Singer (J. 34 (1969)). S. Rend. Protter and H. Yale Univ.. at which b(x)>O. References [l] J. (10) Then the Dirichlet problem for equation (1) is to find a function u(x) defined on G U C. Klasse der Siichs. . 1 (1947)) discussed this formula in connection with the heat equation. 257-282.6) and are independent of the particular choice of (I. M. 1967. Suppose that the coefficients of L and the boundary I of G are smooth enough. Degenerate elliptic equations of type (1) have also been investigated from the probabilistic viewpoint (. T. Math. of the shape of G.327 Partial Differential Equations of Parabolic Type G). C is independent of the boundary condition and. u).(x’)). N.. and Z. H. Maximum principles in differential equations. I. . V. 38 (1934). and =O. 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.. [3] C. Goulaouic. i. Mat. Arch. Math. Also. We have the following existence theorem [28]: If(i)eitherc<OonGorc*<OonGandif (ii) pc + qc* < 0 on G. as T+ 4x. Mizohata and R.. 1923.(x)bi(x)-j$ y}.. The general boundary value problems for degenerate elliptic equations of higher order have been treated by M. 24 (1907).jEl c ” a2aij(X).[27]). Piss. Springer. Levi. 275-317. Umber lineare elliptische Differentialgleichungen zweiter Ordnung. The regularity of solutions is also discussed in [28]. Math. . IV-1 (1974)). At XE I.(x)) the unit outer . it holds that N(T)=CT”‘“+an C =(2x)-” sG dx error term. and hence it is often called Weyl’s formula. 5) -n’m dS. Sup. Weyl’s method is based on the minimax principle [26]. P. Weyl [25] for the case of L = Laplacian. Press. Miranda. respectively. Nordin. 121 (1968)) treated the case of compact manifolds without boundary and obtained the best possible error estimate. S. Hormander (Acta Math. ~0. be the sets ofxsI\Z.O < c(i 1) depend only on (n. Bolley and J. where G is a bounded domain in R” with smooth boundary. Then.also M. Let Z.. PDEs of Elliptic Type where A and tl (A > 0. In general. Z. Rational Mech. [6] J. G’. Vishik and V.

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

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

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

(t. x. + &(x” .xz) = 0 passing through a point p” = (to. u.<) = 0.--au=O.Appendix A.. x) the value of the integral J along the stationary curve from 2I to any point (t. . that is. F.e. English translation. fourth edition. A.2. p) = 0). A typical example the wave equation of hyperbolic equations is a% as a% qU=c7tZ-@-. (Original in Russian. a2u z-c .la~. . III. Petrovskii.)=o.aV/ax)=O 325 (X111. A characteristic curve is also an integral curve. l.to) + . x0. Hilbert.) [3] E.x. Second-Order A tlinear variables Linear Hyperbolic Equations holds.1~=. 1935. then the equation aV/&+H(t.. x. If F is linear in pi.. G. Goursat.a~. a that are functions in (t.) = 0 everywhere on S.. [4] F. Cours d’analyse mathematique. 1954. [2] I. all integral curves coincide with characteristic curves.5)1=c>o. .(t. H(t. 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. . John. Gauthier-Villars. 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. According to the theory of lirst-order par- . . partial differential equation in n + 1 t.5)=1’‘f a&l-i=l (for example..<) is called a characteristic hyperplane of (1) at p” if the direction (A. regularly hyperbolic if these two roots are separated uniformly. quasilinear. that is..(3) Equation (3) is also called the equation of a vibrating string. ... . s. 1967. -a. Lectures on partial differential equations. x=(x. When n = 2. This equation is called the Monge differential equation. Caratheodory. Table 15). 0).) of the second order. 1911.-- au at i=l axi 1 a.-ax. 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). Interscience. Partial differential equations. p) = 0 with coefficients aoi.. Holden-Day.. #O).1. or the equation of sound propagation according as n = 1. (t?. x0) in tx-space and having normal direction (A. In particular.(t. .5)-n.1215 325 A PDEs of Hyperbolic Type this case. Courant and D. u. .ln. [S] C. Calculus of variations and partial differential equations of the first order. <)..j=l aij&tj=O. . x) of G. ~. Springer.X. . In the (n + 1)-dimensional tx-space. 1962. A hyperplane . Conversely. x. . 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. or 3. we obtain Mb i i.. Interscience. by eliminating p between dxJdx. the equation of a vibrating membrane. au which describes the propagation of electric current in a conducting wire with leakage and is called the telegraph equation (. and F = 0 when FP. This equation is called the HamiltonJacobi differential equation or the canonical or eikonal equation. A hypersurface S : s(t. 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. II. The Monge Differential Equation equation (2). st. Teubner. x. (t.<) satisfies the characteristic equation at p” : H(t’. H(t. . and the curve represented by its solution is called an integral curve of the equation. . Variationsrechnung und Partielle Differentialgleichungen Erste Ordnung.fax.x. . i. .. .x. 1953. Another example is References [l] R. . and F(x.25) Partial Differential Equations of Hyperbolic Type A.a2u @-2%=0. = FPi/FP. . 197 1. second edition.5”)#(0..‘=O. Methods of mathematical physics II. <) for any n-tuple of real numbers 5 = (1) is called (51. if we denote by V(t.. 1927. (2) has two distinct real roots 1= I. . .~. n au Consider the partial differential By eliminating p and t between dxi --=Fpp dt :=A PiFpi. ax.l(t .

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

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

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

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

strongly hyperbolic operators have been characterized by K.r)=Ij(~. condition in the case of multiplicity 2 is given by some conditions related to the subprincipal symbol L. a. 5) (P. 5) c a. n a2LN a3. Levi’s condition. Petkov [20]). J. i.<). where L. x. or more precisely.-.I(a/ax)“cp(x)I G C. 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. I.e. E. are real.. if the multiplicity of characteristic roots at (F. 1976..x. even the principal part is not necessarily hyperbolic. J. 1976). for a weakly hyperbolic operator with variable coefficients. j=l. Ya. 2 ( Et+&g$ in the following If: l<l=i.5)-n. namely.. a. 5) = AN + is decomposed LN(t. and to the tPoisson brackets (A. A=A.+/al+qB. (1 <s d N) are the homogeneous lower-order terms of order N -s of L (Ivrii and V.+lal=N rr. N-s.T]xR”=Q 1. x.325 I PDEs of Hyperbolic 1220 Type for a.A’a’lcrl!“}. Mizohata [ 161). the characteristic polynomial L. The Cauchy problem for a weakly hyperbolic system of equations is more complicated.5)l=C>O. 0.x. A.(t.. Cdl.h’-1.1)th order among lower-order terms of L (Mizohata and Y.t))‘i.laO[’ > associated with (17) is not clear in general (references in [20]). s. Hormander.. x. J. Gevrey Classes Classically. Amer.aiacaa~. the functions of tclass s (s > 1) of Gevrey (.58 C”-Functions and QuasiAnalytic Functions G.jik l/z. On the other hand. and A such that ~~p.2 where L. j = 1.. 1977). Anal.X. because of the essential difficulty that the matrix structure Assume that the ij(t. [O. L. Oleinik [ 191 studied the Cauchy problem with nonconstant multiplicities for second-order equations.e. 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.-ii(t. Ohya.QN-1 X)2y way: (‘<ER”\O) fora.. forall (t. Sup. .l.() = 0 be real for any (t.X.. Sci. (1960)).. This class of functions was used efficiently in the studies of the Cauchy problem for tweakly hyperbolic partial differential equations: in j=O. Lax [ 151.+lal=nj c a&(t. M. Chazarain [ 1 S] has studied weakly hyperbolic operators with characteristic roots of arbitrary constant multiplicity. -- i a2L. (“~ER”\O) L. 0.+r]/I]+s(l+q)<.. which corresponds to Levi’s condition in the case of constant multiplicity.7aresatisfied. Moscow Math. Ohya [17]). 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.(t. .+lal+q&+r(fil<p. x). Ann. +C m=.(t. In the case of constant coefficients. 2) is at most p. Kyoto Uniu. it is known that not only regularly hyperbolic operators but also some special classes of not regularly hyperbolic operators are strongly hyperbolic (V. if there exist positive rational numbers q and r (q 2 r) such that 11=o . Yamaguti (Mem. Math.(t. Sot. . Pisa. Scuola Norm. denotes the homogeneous part of (N . of the char- We assume that the multiplicities acteristic roots are constant. the suffrcieni. .-. 1._. Then L is hyperbolic if and only if it satisfies E. Thus. As for operators with variable coefficients. 1977. Kasahara and M. Menikoff.. D.&. i. x.there exist constants C. i=l .I..2. lvrii.x. Math. For higherorder equations.. I”.

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

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

the solution also has singularities for t > 0. Levi concernant des equations hyperboliques. 1968. [7] F. Sci.) [13] K. Interscience. 513(dtj/df)(s) = X.. Anal. Math. Courant and D. (N. Math. Math. x. 1953. Sb. 5 Y). R. 1932.1..S.. WFu(. Guillemin and Sternberg.e. Hadamard. [17] S. [S] J. and all the broken x rays according to the geometric optics starting from~. 1955. Sur la condition d’hyper- . Applications of functional analysis in mathematical physics. 162-189. 1 (1961) 109-127. Linear hyperbolic equations with constant coefficients. Sobolev. (Original in Japanese. 4 (1968) 5 11-526.. Friedrichs.1223 325 Ref. When the initial data have singularities. Interscience. Girding.. [(t))~ T*(R”) 1 Cdxjldt) ts) =tanllatj) -(aW~xi)(s. Schauder. [ 161 S. B. 24 (1935) 213-246. [6] M. 51.x.1=1. Melrose [33] showed the following: Suppose 8 = CR c {x 11 I< R} for some R > 0. Methods of mathematical physics II. Math. Petrovskii. L. Chazarain [18]). E.. 815-870. Sot. . Interscience. (x(O). 1950. Schauder. Math. The theory of partial differential equations.. G. It should be noted that the tpropagation of singularities cannot be derived from the Huygens principle in the wider sense. Duke Math. G.. the solution becomes smooth in R. Appl. 1903. Ludwig and Granoff.i. k =o 91..) [9] J. Assume that uk. Mat. which is a property of hyperbolic equations quite different from the properties of parabolic ones. PDEs of Hyperbolic Type strictly convex obstacles. Kyoto Univ.. Krzyianski and J. 7 (1954). Res. Luneberg [32].(t. 0. have. 6 (1936). 1954. Pure Appl. Pub]. Friedrichs. Mizohata and Y. Math. 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. Sur la condition de E. it is known that the twavefront propagates along the tbicharacteristic strips satisfyingqt--L. Das Anfangswertproblem einer quasilinearen hyperbolischen Differentialgleichung zweiter Ordnung.2 . (Original in Russian. Supersonic flow and shock waves. [lo] 1. K. x. 1965. Then for initial data with singularities in [l] R. Asymptotic solutions of oscillatory initial value problems.. ts.).Vq)=O. 0. Cambridge. Princeton. References M. R. Lectures on partial differential equations.=Rfl{]xl<R}gooutof&inafixed time. D. J. Lecons sur la propagation des ondes et les equations de l’hydrodynamique. Suppose that the multiplicities of characteristic roots of L are constant. Quasilineare Differentialgleichungen zweiter Ordnung vom hyperbolischen Typus. continuous derivatives of sufficiently high order to suffer jump discontinuities across I. Uber das Cauchysche Problem fur Systeme von partiellen Differentialgleichungen.thatis. Fund. Mizohata. Lax..1)-dimensional manifold F. Amer. Hilbert. [ 151 P. gemischte Randwertaufgaben. Math. the reflected wave can be constructed by the superposition of asymptotic solutions of the type (23). (Original in Russian. Acta Math. 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. 1948. Studia Math. [3] J. and across these the partial derivatives of u have jump discontinuities (Courant and Hilbert [ 11). Surveys. 85 (1951) l-62. 2 (44) (1937). 1973. Ohya. Le probleme de Cauchy et les equations aux d&i&es partielles lineaires hyperboliques. 1962. 24 (1957) 627-646. Amer. 14 (1977)). J. Some remarks on the 1 Cauchy problem. Hyperbolic differential equations. Leray. Math. Propagation of Singularities Let L be a hyperbolic operator with C” coetficients and consider the Cauchy problem L[u] = 0 with initial condition (15). For the tacoustic problem. [4] R. Plane waves and spherical means applied to partial differential equations. For more general singularities of initial data... 1952. Hermann. [2] J.. [S] S. Then the solution u has continuous partial derivatives of sufficiently high order everywhere except on the characteristic surfaces of L issuing from F. Courant and K. Symmetric hyperbolic linear differential equations. John. Comm. 1 c141 L . on either side of a sufticiently smooth (n . 1963.. ~(O))E iJkWF(d} (J. Institute for Advanced Study. Hermann. Petrovskii. [ 1l] 1. J. Sot. Inst.) [ 121 S. N . N-s.. for sufficiently large t.. Interscience. t) is contained in {(x(t). The propagation of singularities is more complicated in mixed problems because of the reflections of singularities at the boundary. Lecture notes. Hadamard. 345-392.. Mizohata.

OpCrateurs hyperboliques g caractkristiques de multipliciti: constante.e.27) Partial Differential of Mixed Type A. Ikawa. Le problkme de Cauchy pour les Cquations hyperboliques & caracttristique multiple. 253-279. Fourier. Mathematical theory of optics. 23 (1970). Math. Systkmes IinCaires. Surveys. Japan. aY2 XW)20. 569-586. 2jdme Colloque sur 1’Analyse Fonctionnelle. Equation (1) is of elliptic type if q < c. Math.. 326 (X111. Bronshtein. 112 (1971). Propagation des singularit& pour une classe d’opkrateurs B CaractCristiques multiples et r&solubilit& locale. [22] J. Balaban. If there exist points where the flow is subsonic as well as points where it is supersonic. Sot. 277-298.. 10 (1976) 4. Sakamoto. &-well posedness for hyperbolic mixed problems with constant coeflicients. 19 (1966). 203-223. [23] M. On the mixed problems for the wave equation in an interior domain II. Math. 17 (59) (1945). J. and c is the local speed of sound. B. Mixed problems for hyperbolic equations I. J. (Original in Russian. [lS] J. even when they are linear. On the mixed problem for a hyperbolic equation.) [24] I. Appl. which is a known function of the speed q =: (u’ + u2)1/2 of the flow. 29 (1974). or tparabolic) of the equation depends on the location of the point x. [30] M. 289-370. Kyoto Univ. Singularities and energy decay in acoustical scattering.. B. Equation (2) is hyperbolic for :c > 0 and elliptic for x < 0. Math. u= &p/ax and v = acp/ay are the velocity components. and of hyperbolic type if q > c. (2) which is called Cbaplygin’s differential equation. On the Cauchy problem for weakly hyperbolic equations. D. Acta Math. (1) :IS of mixed type. 63104. Math. CBRM (1964). 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. [28] R. Pure Appl. Pure Appl. 145-206. 14 (1974). [20] V. C. Melrose. the flow is subsonic. Math.. Math. [27] H. (N. 23 (1970). Fourier. Functional Anal. Amer. 24 (1974). Uniform asymptotic expansions at a caustic. Duke Math. J. Ohya. On the diffusion of waves and the lacunas for hyperbolic equations. Cbaplygin’s Differential Equation It is difficult to solve equation (1) directly since it is nonlinear.403-417.. Sakamoto. 1944. and L.. Comm. II.. M. G. 43-59. [26] T. The study of equations of mixNed type has become important with the development of high-speed jet planes. However. Luneberg. Petrovskii. Math. The type (telliptic. Initial boundary value problems for hyperbolic systems. Kyoto Univ. Osaka J. Pure Appl. 93118..). 134 (1973). where cp is the velocity potential. Sot. 105144. Sb. II. 268% 286. the flow is supersonic. is much more difficult and less developed than the study of equations of nonmixed type. Ya. Ludwig. Comm. F. 45 (1979j. The linearized equation takes the form $-Iqx)IZ=O. Russian Math. 1976. Comm. Brown Univ. Ann.326 A PDEs of Mixed 1224 Type bolicitC pour les Cquations g caractCristiques multiples II. Lacunas for hyperbolic differential operators with constant coefficients I.. General Remarks Equations Let A [u(x)] = 0 be a tquasilinear second-order partial differential equation. The parametrix of the Cauchy problem for hyperbolic operators with characteristics of variable multiplicity. 24 (1974). Kreiss.. 0. [29] R. R. the equation is said to be of mixed type. 109-189. l-70. Bott. 40 (1971). Ggrding. 173-202. thyperbolic. i. [31] D. [ 191 0. 124 (1970). 10 (1970). [32] R.) [Zl] Y. J. (Original in Russian.. Ann. Tricomi’s The simplest Differential equation Equation of the form (2) is (3) . Press. Chazarain. hyperboliques non stricts. K. [33] R. 349-373. Inst.S. 83-84.e. we can linearize it by taking q and 0 = arc tan(o/u) as independent variables (the hodograph transformation). If the type varies as the point x moves. [25] M. 215-250. Ivrii and V. Necessary conditions for the correctness of the Cauchy problem for nonstrictly hyperbolic equations. Math. Oleinik. i. J. The study of general equations of mixed type. Mat. 17 (1980). A. Leray and Y. Almost all research so far has been on equation (2) or slight modifications of it. Ohya. Japan.. 1974. Atiyah. 16 (1964). Inst. Mem.. Petkov.

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

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

then it vanishes identically. t)= g(t) and u. If u(x. <x. M.(x). cp. or of a linear combination of u and au/an. vanishes on A^D. and the segment AB.. (15) for sufficiently small h. in some neighborhood of a given curve C nowhere tangent to a characteristic. t) = cpb). (16) where G is the region bounded by the closed curve C. We call (21) Green’s formula for the partial differential equation of parabolic type (3). then the solution to this problem can be expanded as u(x. It corresponds to the problem of heat conduction in an infinite rod. The problem of heat conduction mentioned in Section B corresponds to the particular case ‘pi(t) = constant. lg u.(x) is a normalized function (~~q$(x)dx = 1) that satisfies the boundary conditions d(a) . 1 be such that no characteristic meets either of them in more in the than one point. x. applied to the region G=(ABQP) and the functions $ = 1.bA4 = 0. . According to E. along C. and continuous. is also a mathematical formulation of the problem of heat conduction in a rod [4]. t) that satisfies (8) for t > 0 and the initial condition limtlo u(x. the uniqueness of the solution is established in the following way: Let the curve fi and B^E in Fig. the problem of finding a solution u(x. Consider the equation This problem.. g-hu=O for x=a. 4x9 t) = s(t). this formula is used to establish the uniqueness of the solution of (3) and to derive an integral representation for it. B^E. t) satisfying the conditions (18) and the equation (19) The fourth type of problem is to find a function u(x. au z+H~=O for x=b.e-“n’cp. t) = q(x). t) is continuous closed region (ABED).= ah ---= au ax2 at 0 (10) where cp. of (8) that satisfies the conditions (14’) ljLyu(x.1221 327 D PDEs of Parabolic Type The first consists of determining.(xO.<t<t. 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. h = constant > 0. and the line integral on the right is evaluated in the counterclockwise direction over C. a solution u(x. Here we impose the hypothesis on the curves x = cpl(t) and x= q2(t) that they are nowhere tangent to a characteristic (9). In the second type of problem we are required to find in a region of the form cPl(t)~xGcpz(tX t.(t)=constant. If p(x) is of class C’.. t)= fJ c. Green’s formula (21). t) such that u(x. (13) D. For example. $. t) = h(t). (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. d(b)+HvnW=O s b a dxh(x)dx. posed for equation (lo). a solution u that possesses prescribed values of u and au/an. n=1 CT. (17) in the region a < t <b. t) = h(t) for given functions g(t) and h(t) is a problem of this type. Holmgren.(x. H = constant > 0. Gevrey [3] showed that if such a solution does exist. a solution t>o. As in the case of partial differential equations of elliptic type (. For instance. The third type of problem is to find in a region of the form a<x<b. and satisfies equation (3) in (ABED) except on AB.323 Partial Differential Equations of Elliptic Type). for which condition (15) is automatically fulfilled. c = constant. 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 a similar way.t)=(P(x). + h. to). p ------. 4% =s. we lhave m e-‘*u.. yields 1228 Type where (x.cpw. This idea can also be applied to the solution of parabolic equations with constant coefficients in (n + 1) variables. 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). t) = q(x)./ Q 1 Fig. The Laplace Transform Method Let u(x. s0 t)dt = [C”‘u(x.A(r)u Equations of (I) with f=: 0. For equation (3). Then u is identically equal to K on the segment QP and in the region (ABQP).u(x. we can establish the basic representation formula t) x. we obtain +1 ‘72 emA*u(x. = . by extending Green’s formula suitably. U(x.-x)2/4h PC! provided that lim. various versions of the maximum principle are known for equation (1) [S].)e -(x... t). Once the solution of (26’) has been found. to. t)dt s0 iu(x. 1.t)dx to) PABQ +( ug-+J)dt F. t)u(x. = PASQ s--- u(x.q(x)+ 6’6) 4x. t). where h is a positive number and M is a point with coordinates (x0.. t)/an+h(x. of (3) for t >O and 1>0. the following maximum principle holds: In the region (ABED) of Fig. to) as hJ0. (25) t)dt.x. we are able to prove uniqueness theorems for the four problems stated in Section C for more general linear parabolic equations. be its fLaplace transform with respect to t.. t. We find in view of (3): a*u s = wx. Similarly. ~u(x. A). the desired solution of (3) can be derived by inverting the Laplace transform (25).$ ------. (24) at ) (27) where A(t) is a second-order telliptic operator with parameter t. 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)=f(x. Utilizing integration by parts. t) be a solution (3) and Let we of e-%(x. rj = CJ(x. x. suppose that L[u] 2 0 and that u takes its maximum value K at an interior point M. (16’) derivative in the ) where a/&r is the directional .(x. x.B. + h. t) be a solution cc u(x. Let D be a region (bounded or unbounded) of points x whose boundary is a smooth hypersurface S. t)dx Since the integral in the left-hand side of this equality approaches u(xO. t)=(4n(b-t))-““exp I-&$}. t)= 0 and lirntlo u(x. 1”)= s cl To obtain a representation for solutions proceed as follows: Let u(x. More generally. the following function E (called the Gauss kernel) is a fundamental solution of equation (2): E(a.327 E PDEs of Parabolic cp= u2. such as (1). a) E R” x R” and t < 8. co) equation (27) together with the conditions lpi(x. XES. Formula (23) shows that u(xO. t) U(x. e-%(x. We pose the following initial boundary value problem for (27): Find a function u(x. XED. General Second-Order Parabolic Type Consider the equation be written as F. s-t. t.. t)]fI$ be a particular solution. 4 . which can (23) for solutions u of (3). t). t) that satisfies in D x (0. Applying Green’s formula to the region (PABQMP) and the functions cp= u(x. 1 E.

x.(l). (this regularity assumption can be relaxed). t) = R. Wp(x.. and t-t and is written as U(t. XER”} and /I = a/(2cr+ 2).)v=lv-$ with 1= l/6. when the coefficients of A(t) are infinitely differentiable. (ii) The function u(x. x. t) defined by If the function h in the boundary conditions (16’) and the coefficients of A are independent oft. t > ‘t 2 0) that. where B=sup{ lu(x. 5ED. Then. t) of (27) satisfying (16’) with f(x.. BytheLaplace transform method as described above. t) = 0 is expanded as e-“‘l(l. t) satisfies (27) in D x (0. for any bounded solution u(x. Besides the properties (i) and (ii). J. 7. the inequality (29) holds. < t < s. The Laplace transform is not suitable for solving problems (27) and (16’). any solution u(x.Wpp(4.115 Diffusion Processes).7]: (i) There exists a u = U(& T. GZ. 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). then the fundamental solution U(<.(x. and further SDU(<. . s) such that x ER”. y ER”.x. z. put t. t) is a generalization of the function (24).In particular.$=lv. t) is locally summable and (28) where aij = aji are real-valued functions of class C” and equal to constants outside a fixed compact set of R” for all t 2 t. the constants u and A are positive. w. 1. 1) but independent of the particular choice of (aij). x. m) in the usual sense. then u(x. m-l witha=tfm.oO) X R” x R”. x. y. f= 0. By iterating this procedure m times. then there exist a sequence of teigenfunctions {$p(x. That is.. co) x R” and for any (x. Furthermore. where u = or/@+ 1) with LY (29) and B’= sup lo(x)I. .. z. the inequality Iv(x)-u(y)l<A’B’Ix--yl” (30) j. t) of (27) in the distribution sense is a genuine solution (. As a corollary to this theorem. Thus U(& Z. w) U(z.-I). and h(x. jDUW(~ +A(t)*cp(x. 2) 1 = 1. wz = U([. T. Therefore this theory is of considerable significance from the point of view of the theory of probability (. Nash’s Results Let us consider a parabolic equation is a solution of (27) and satisfies (16’). .t) ) dxdt=O for any function rp(x. t. We put R. t) of class Cz in D x (0. satisfies equation (27) and the homogeneous boundary conditions (16’) with rp= 0. t) 2 0. . t) of (28) in (to. t) depends only on 5. s -cc. y) ER” x R”. t) of (27) and (16’) as the limit of u. Then we obtain a solution u(x. Nash proved that. Instead. x. & t) 2 0. the fundamental solution satisfies UK. as a function of x and t. x. It can be shown that a weak solution of the parabolic equation (27) is a genuine solution.q starting from the initial value cpat t =O. The following results are known [6. (2) The solution u(x. . x. R.(x.125 Distributions and Hyperfunctions).(A)} of measures on the real line for which the following hold: (1) The fundamental solution U(& x. t-z) (t > 7). we have a function u.k where 4cp. t. .. . t) (x.x. if A is tself-adjoint. = k6 fork=0 . t) dx = 1 under some additional assumptions. holds for any (x. The constant in . if u(x. R. dx. t) is expressed in the form U(t. Let m be a large positive integer. G. and t. the theory of l-parameter semigroups of linear operators (. . Suppose that there exists a constant 12 1 such that 1-‘J<[’ ~~aij(x~t)5ilj~al~12 for all (4X. } (A$p + p &Qp= 0) and a sequence {p. co) and vanishing outside a compact subset of D (A(t)* is the adjoint of the partial differential operator A(t) and dx = dx.-. called the fundamental solution of the linear parabolic equation (27) with boundary conditions (16’).2. depending on (n. t) (xES). t) = 2 p=1 m e-“Vp.S)E(t~.-. For t > 0. andofu[9].t) as m-co. t)l 1t > t. In this inequality. we can associate with $ a unique solution v of A(t.378 Semigrbups of Operators and Evolution Equations) can be applied to establish similar fundamental results. SDU(<. UC% w 09x2 t) sCPK)w =D (q(x) is the function given in (16’).1229 327 G PDEs of Parabolic Type outward tconormal direction at (x. t) (z < w < t).(5. . 4&.

is denoted by p(n) and is called the number of partitions of n. Interscience. 17-3 (1962).. . 1953. which is closely related to f(x).+a. [S] M. The igenerating function of p(n) is f(x)=l+ 5 p(n)x”= ft=1 5 (l-X3jl?. D.j(x. 27 (1957).. I. S.(x. Hypoellipticitt ties Cquations paraboliques. Equations differentielles operationnelles et problemes aux limites. t) in the region under consideration and for any <E R”. 105-147. Mizohata. t. Math. Friedman. Let us consider an equation for an unknown function u of (n + 1) independent variables (x.. Euler (1748) obtained the following formula (called the pentagonal number theorem because n(3n . Gevrey.. Linear equations of the second order of parabolic type.t. Prentice-Hall.. 1954. North-Holland. S. References [l] L. Equation (27) is p-parabolic if --A(t) is strongly elliptic of order p. Nash. 4 (1964)). Math. [9] J.j)‘s such that pj + 1aI = pm and that 0 <j < m. Let us denote by {2.6) Partitions of Numbers A partition of a positive integer n is. Parabolic systems. Courant and D.) [ 1 l] S. Prentice-Hall. H. [lo] S. [2] A. John.i 0. 2) (.p . 85 (1957). ( n=1 I for any (x. 1963. where the order of the summands is ignored and repetition is permitted. 9 (1913) 3055471. . The Dedekind eta function. . A. Petrovskii if and only if there exists a positive number 6 such that Rei. Pures Appl. The number of partitions of n.j(X. 1962. Fundamental solutions of parabolic differential equations and boundary value problems. 1967. Hilbert. Kyoto IJniu. Methods of mathematical physics. Arima (J. (32) where Ch. 1961. Russian Math. G. Ann. 10 (1914). Bochner. Japan. Math. The integer p is then seen to be even.Therefore. J. The unit circle 1x1~ 1 is the tnatural boundary of f(x). Ito. 931-954.In (31). J. Protter and H. and 0. which is holomorphic in Ix I< 1. Bers. Oleinik. u. L. [7] A. t)&$=. where a = (a . Springer. [6] S. For example p(5) = 7 since5=4+1=3+2=3+1+1=. We say that the equation (31) is p-parabolic (or of p-parabolic type) in the sense of I. L. J. Bull.p. A’ depends only on (n.1)/2 . France. and F. (33) 328 (V.+. the roots 1 of the equation 1”” +c’ a. Press. [)}r=:=. Eidel’man obtained precise estimates of the fundamental solutions and of their derivatives for p-parabolic equations [lo]. 80 (1958). Studies. Math. Continuity of solutions of parabolic and elliptic equations.) [S] J. A. (Original in Russian. 1964. Bin.j a. Wewritealso1a(=cc. Princeton Univ. Weinberger. C.j is the summation over the (a. an expression of n as the sum of positive integers.. 1969. 1. Amer. p-parabolic equations are known to be thypoelliptic if the coefficients are of class C” [ 111.. H. 1 <k<m. Math. Partial differential equations of parabolic type. Sot. . (Original in Russian 1962.f (31) (a/ax.. Eidel’man. we can define the p-parabolic systems of equations [lo]. F. 15-50. Contributions to the theory of partial differential equations.) and aa/axa=(a/ax.t).321 H PDEs of Parabolic 1230 Type Dif[4] R. Maximum principles in differential equations. D.(x. t) of the type z+c a.!+2+1= 2+1+1+1=1+1+1+1+1. Les equations aux derivees partielles du type paraboliques.143.323 Partial ferential Equations of Elliptic Type L).. n=1 Hence ~(7 + l)= exp(rci/l2)n(z). The heat equation (2) is 2-parabolic in this sense. M.. II. Similarly. Math. S. t)(g)“v= a. [3] M.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. Lions. 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)).i is the summation taken over the (aj) such that pj+lal<pm and O<j<m.. 555102. Kalashnikov.< -bl&‘.. The mixed initial boundary value problems are investigated in detail also by Eidel’man [lo] and by R. Surveys. Partial Differential Parabolic Type Equations of p- Let p and m be given positive integers.

/@ q(r). With regard to the Dedekind sum s(h.. =exp(nis(h. This follows easily from the Jacobi-Biebler equality where W. k) was given by Rademacher in the form By using the ttransformation formula for 9functions. where A and B are suitable constants: (A/n)e’&~(n)<(B/n)e~@J”.r)). b = (hh’ + 1)/k. By making use of a remarkable identity.l/2 otherwise ([ ] is the tGauss symbol).292. The size of p(n) increases rapidly with n. we can infer from the pentagonal number theorem that n( . then the E appearing in the transformation formula of ~(7) is seen to be equal to exp( . c. The theory was improved by Rademacher who expanded p(n) into the . G. in fact P.+A > If we make substitutions a = h’.k)=l..1943). c = k. (h. we can deal with the integral by the tcircle method. Subsequently they obtained Here the symbol ((t)) in the sum denotes the function that is 0 for integral t and t ./I formula [ 121. Hence. ZZO). fl. From this formula we can calculate p(n) successively. It is known that q(r) is a tcusp form of weight -l/2 with respect to the full tmodular group F( 1) [9./$ n(r). Newman (1951). A. L 27ri s rX”+l f(x)dx ’ After.. Iseki (1952). Iseki (1957) gave another proof by using a new method. A direct proof of the transformation formula and the reciprocity law was given by K. k).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. Multiplying both sides of Euler’s formula by the generating function of p(n) and comparing the coefficients. d are integers satisfying ad bc= 1 and c>O. Postnikov [S] succeeded in proving by means of an elementary function-theoretic method. which threw light on recent additive number theory. we have the reciprocity law for Dedekind sums: s(h. J. introduced by Hardy and Ramanujan. for instance p(lO)=42 and p(lOO)= 190. +2. Erdiis (1942) and D. known as the CI.[t] . where p and q are fixed integers. . .. MacMahon obtained in this way the values of p(n) for n up to 200.569. Hardy and Ramanujan thus obtained + O@xp(DJtt)). it follows from the transformation formula that f(x) w exp(w2/6q2(l . hh’= -l(modk). L. we obtain where the contour I is taken inside the unit circle around the origin. W. if a. d) + 7ci(a + d)/l2c). is defined by = 1+ -f q+(Z”+Z-“) “=I (lql< 1.h)= -$+A ( . Hardy and Ramanujan proved the following transformation formula for the generating I ( 1937. 6.nis(c.l/r)= . and d = -h in the Hardy-Ramanujan transformation formula.k)+s(k. C. H. Later S. A. then where E is a 24th root of unity.) is a pentagonal number..=k(3k1)/2 (k=O.k&ev($$$J Xf(exp(y--z)).l/z)= . Siegel (1955) gave a simple proof of the formula q( . letting r+ 1-O in x = r exp(2aip/q). P. Ramanujan (1918) proved the following inequalities. Nevertheless. k)) and the value of s(h. G. p(n) can be represented as an integral: ~(4 -W4~4exp~~J2n/3). 111. where where o. According to Cauchy’s integral formula. Hardy and S.+. The generating function f(x) varies greatly: namely.

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

Wiley. For generating function (1 + 2)” = x:0 any complex number x. allowing repetition. is the number of permutations of R. no. In particular. Pascal also formulated clear ideas about axioms. and the volume of the figure obtained by rotating it around these lines. (n). This is ais0 the number ofnonnegative integral solutions of &xi = k. . Gallimard. de la Facultt des Lettres de l’Univ. B. In any case. led tLeibniz to discover the fundamental theorem of calculus. the series is conver- [l] E. Permutations and Combinations ical induction. The number of such arrays is (r& = n(n .1) . L’Oeuvre mathematique de Pascal.+.= x(x .+a~~=~&a~~./n! in terms of the Jordan factorial (x). . 21.8) Permutations Combinations and (multinomial theorem). 1958). we have (a.. Oeuvres completes. The same results hold in a tcomplete field with tvaluation. 1960). he did not abandon mathematics. His book Penstes shows his deep involvement with religion. Press. as was noticed by Pascal. . The number of ways of choosing k elements./k!. Pascal. While in Port-Royal. he published Lettres provinciales (1657). Riordan. II. 1981. Combinatory analysis I. however. A subset of R is called a k-subset if it contains exactly k elements. The polynomial (x). Combinatorial identities. . Cambridge Univ. and we have (a + l~)“=~~=~ 0 i a n-kbk (binomial theorem). If we choose k distinct elements of R and arrange them in a row. The study of the methods used by Pascal to obtain these results. Lehrbuch der Combinatorik..1233 330 Ref. (n -k + 1). The 0 X coefficients 0n are defined by the z”. Oeuvres I-XIV. B. [2] J. Teubner. k. A. and with an intuitive idea of limits obtained the formula j: x”dx = amf’/(m + 1).. As a generalization.. Pascal.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. [l] L. [2] P. For integral values x. Bibliothtque de Pleiades. Mem.). As an example of binomial coeflicients with noninteger arguments. 1927 (Chelsea. the barycenter and area of the figure enclosed by a cycloid and straight lines parallel to its base. from a set of n elements is Let there be given a set R of n elements.). we have (-j’2)=(-1)“2-‘“0 References of R is i = (n). (1 + z)” are polynomials. . d’Osaka. second edition. (x -k + 1) in x of degree k is called the Jordan factorial of degree k. = n!.. which were forerunners of differential and integral calculus. [3] J.1) . . [3] KGkiti Hara. and it is verified that x = 0n (x).. where the sum is extended over all nonnegative pi with Z pi = n.. Netto. he determined the area enclosed by a tcycloid and its base. The number of k-subsets (or kcombinations) binomial (-i)k~j=~+[p'). Brunschvicg (ed. 1968. The recursive relation allows us to compute the values of n easily 0k References for small integers n.. He also indicated a way to obtain the sum of the mth powers of the consecutive terms of an arithmetic progression. we have the recursive relation and in general which leads to many identities involving binomial coefficients. Chevalier (ed. 330 (11. in which he carried on a dispute with the Jesuits. 1954. we have a k-array or kpermutation of elements of Q. n factorial. gent for IzI < 1. in particular in a tp-adic number field. MacMahon. 1915-1916 (Chelsea. 19041914. Hachette. In 1658.

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

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

. (1) The essential spectrum (. Japan. ActualitCs Sci.(X. I. Math.) c D(H) and such that the closure of HJ-JH. Conversely.. ser. Palermo. 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. (2) The absolutely continuous spectrum (390 Spectral Analysis of Operators E) is stable under perturbation by the ttrace class. and the study of this stability has been a major topic in perturbation theory (also .9]. Kato. Fat. H. Some problems of resonance can be treated by the technique of dilation analyticity. Sot. J. D. Combes.(H) = a. then CT~(H)=O.(H. M. studying the mode of change under perturbation is not usually a tractable problem. Sh. either by the wave operator approach or by the abstract stationary approach (. Pearson. More generally.. 1966. Comm.331 E Perturbation E. All the results mentioned above are proved by scattering-theoretic methods. 27 (1909)). Math. and 11 lip to be used below . Stijrungstheorie der Spektralzerlegung I-V..[S. (i) can be derived from (ii). Rather.... 600619. Berg (1971) W. Math. HO)~B1 for some [E p(H) fl p(H. if H = HO + K with compact K. of HO which is embedded in the continuous spectrum may diffuse into the continuous spectrum in the presence of a perturbation. and in particular o.(H. Sikonia (1971) J.. 27. An eigenvalue 1. belongs to B. 118 (1942). 117 (1940). T. This phenomenon of spectral concentration is studied. Ann. J.375 Scattering Theory. J. Perturbation theory for linear operators. 361-406. Mat. Pac$c J. Springer. (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. On the perturbation l.. it suffices to assume that R(<. Proc.. Math. H). IC#O. not discussed in this article.heory of closed linear operators.. [4] K. [5] T. Birman. (1963). 356-382. von Neumann. Moreover. Comm. Japan Acad.). In some problems. In this section we discuss only self-adjoint operators and let H=jldE(I). then the absolutely continuous parts of HO and H are tunitarily equivalent. Univ. . U-’ + K (J. On the perturbation of continuous spectra. In fact.) (H. B. H) . Pacific J. 229 (1935)). Voigt (1977). 6 (1951) 145-226.68 Compact and Nuclear Operators. especially for some concrete problems. Pure Appl. has no eigenvalues near Lo but may have a continuous spectrum highly concentrated around /I.) and q(H) are unitarily equivalent for any smooth strictly increasing real function cp (M.375 Scattering Theory). Kato.390 Spectral Analysis of Operators I. 7 (1957). Circ. 116 (1939) 5555570. Akad. then the absolutely continuous parts of cp(H. . H(K). Functional Anal...) (M. 33 (1957)). certain parts of continuous spectra tend to be stable under perturbation. For continuous spectra..) be self-adjoint operators in a Hilbert space X. Friedrichs. Math. On the convergence of the perturbation method.. Namely.. Math. Rend. J.X). C). Kato.. sec.390 Spectral Analysis of Operators E) is stable under compact perturbation. respectively. F. Kuroda.ich is also effective in other problems of spectral analysis (J. 7.. Also . T. Rosenblum.). Phys. (i) If R(<. Sci..(H. then the same conclusion as in (i) holds (D.. 28 (1978)). I. 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. 113 (1’937).). Among generalizations we mention the following two. in relation to resonance poles (or poles of the holomorphic continuation of the resolvent or the scattering matrix). Japan Acad. 462-484. Namely. (2) The detailed structure of continuous spectra is hard to analyze. Perturbation 1236 of Linear Operators of Continuous Spectra the existence and the completeness of the latter implies the unitary equivalence of absolutely continuous parts. . 1 (1948). In such a case. Nauk SSSR.. Voiculescu (1979) have extended these results to normal operators and m-tuples of commutative self-adjoint operators. .R([. [2] T.. Proc. Tokyo. with K E B. if H = HO + K. esp. and D. T. 15 (1965)). if X is separable and if u~(H)=a. Some Other Topics (1) For the perturbation theory for semigroups of operators and evolution equations. (ii) If HO and H act in different Hilbert spaces X0 and X. Perturbation theory for absolutely continuous spectra is closely related to the study of generalized wave operators in scattering theory. [3] T. X) such that JD(H. For B. 22 (1971)). For the topics mentioned in (2) and (3) - c71. Kato. then there exist a unitary operator U and a compact operator K such that H = UH.. 34 (1958)).R([. mat. and if there exists J E B(X. Weyl. 0. Izv. 113 (1937) 677-685. Kato. a technique wh. Rellich. 4 (1952). References [l] F. with I/K lip < E for any p > 1 and E> 0 (S. HO) is compact for some (or equivalently all) [e p(H) fl p(H. Aguilar and J. 3233337. Ind.

Academic Press. In 1882. Lindemann proved that x is a ttranscendental number using Euler’s formula eRi= -1. van Ceulen (1540. Then the relations L. Machin). Ferguson calculated 710 digits of x and found that Shanks’s value was correct only up to the 527th digit.Arc tan l/239 The symbol 7chas been used since W. Euclid gave no statement about the numerical value of 71. Shanks obtained an approximate value of 7cup to 707 decimals. No improvement of his approximation was obtained until 1946 when D.. F. The fact that this ratio is a constant is stated in Euclid’s Elements.. 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. Sot. 195-211. Matunaga computed rc to 50 decimals. . [9] H. Aryabhatta obtained rc+ 3.) 7=1.7. Amer. 1979.OOO decimals has been obtained.230 partial denominators of the regular continued fraction representation of ?I and described how their numerical evidence tallies with theoretical results. Methods of modern mathematical physics I-IV.As an approximate value of n. L. In the 17th and 18th centuries. in 1873 W.>n>l. Fundamental properties of Hamiltonian operators of Schrodinger type. In 5th~century India. Trans.197l.3. The approximate value of K up to 50 decimals is 3. Tanabe. Wiley-Interscience. Brouncker. 1975.(Z. (4/3)4 was used in ancient Egypt. 1975. Equation of evolution. is called Machin’s formula and was often used for calculating an approximate value of n.3. and I. many formulas that represent rc as a sum of infinite series or as a limit have been used to obtain more accurate approximate values.141592653589793238462643383279502884197 16939937510. = 4 Arc tan l/5 .1237 332 Pi (7~) Using this formula. 70 (1951). the Japanese mathematicians T. &L+!.5. L. J. Archimedes obtained 3%~ rc< 35 by calculating L. W. ... 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 .2. F..Appendix B. Linear operators I-III. 1958.5. and an approximate value up to l. Let L. G. F.. The computation of an accurate approximate value of n has been made easier by the recent development of computing machines. 7c 1 12 32 52 -=4 1+2+2+2+.6. . and in 16th-century Europe. Choong et al. 3 has been used from antiquity. In Sth-century China. 2n -L 4 lZn=4z hold. 1978. Euler. Reed and B. obtained the first 21. Since the 17th century.4. Schwartz. (Original in Japanese. (. Lambert used Brouncker’s expression of a in a continued fraction to prove that R is irrational. Pitman. Beckmann [Z] gives a detailed and humorous historical account of the calculation of n from ancient times up to the present computer age. - .. 1979.. Math. Various numerical results obtained by electronic computers are not formally published. These values were obtained by methods similar to those of Archimedes.. F.OOO.4. By utilizing this formula. the initial letter of aspz~srpo~ (perimeter). H. [3]. K.7) Pi (71) (W. Dunford and J. Table 6).6. (J. Tsu Chung-Chih mentioned 2217 as an inaccurate approximate value and 355/l 13 as an accurate approximate value of n. T.14.. [S] T. P..) be the perimeter of a regular n-gon circumscribed about (inscribed in) a circle of radius 1. According to the Rhind Papyrus. some being deposited in the UMT repository of the editorial office of the journal Mathematics of coniputation. which is valid for almost all irrational numbers (e.1416. Seki. however. Adriaen van Roomen obtained nk 355/l 13. obtained by the metrical theory of continued fractions. Simon.1610) calculated rt to 35 decimals. Takebe. In 3rd-century China Liu Hui used x + 3. Leibniz) (J. 83 Continued for Fractions) . Kato. and Y.. using information in Cl].g. [7] M. In 1761. 1963. C41). (J.. Jones (1675-1749) and L. Thus rc can be defined as 332 (V1. Wallis) [6] N. 1972. The following are representations of x known in those days: n 2. Gregory. C.

Jordan domains are simply connected. and let P be a boundary point of D.18) Plane Domains A. Let n 3 2 be an integer and D a plane domain. The thomology group H. Pi (7-r) References [1] 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. Math. If D is a domain on a complex sphere. the subdomain of D separated by qy that contains qy+l. Qn-l such that D (Q. In a domain D.. lie completely in D. (4) Annular domain: r < 1 .. In this article. Let {qv} be a fundamental sequence of cross cuts. (3) Angular domain: c(< arg(z -c) < p. we consider only subsets of the complex plane (or sphere). 25 (1971) 3877392. (6) The complement of D with respect to the complex sphere is a connected (not necessarily arcwise connected) closed set. [2] P. and every 0. The tclosure of such a domain is called a closed plane domain. The boundary element of a cross cuts Q. contains all qv except for a finite number of v. for the domain obtained by removing x = l/(n + l). as above. 1963. The tinterior of a +Jordan curve J in the complex plane is a domain called a Jordan domain. The intersection n 0. second edition.e. This condition determines an equivalence relation. 1). Noordhoff. (5) Slit domain: a z domain obtained by excluding a Jordan arc I from a domain D. Then D is said to be n-ply connected or multiply connected without specifying n. D . where all points on F (except an endpoint lying on the boundary of D) are contained in D.)fromth~zsquare 0 <x < 1. (5) For every Jordan curve C in D.O < y < 1. it can always be continuously deformed to a point. Comp. the Jordan arc I is called the slit of the domain.000 decimals. [3] K. Khinchin. Calculation of rt to 100. there exist n . Golem Press. a tconnected open set) in the tcomplex plane or on the tcomplex sphere is called a plane domain. and q. 1971.. a Jordan arc whose two endpoints lie on the boundary of D is called a cross cut of D. Take an angular domain D’ with vertex at the point P and the initial parts of the two sides of D’ lying in D.. Z) is identical to Z”-’ if and only if the complement of D in the complex sphere has n connected components. . Wrench. 1238 follows: (1) Circular domain: Iz . In this case. B. Dp are the subdomains constructed from q. Beckmann. tending to P such that the line Isegments P. 2): (1) Every q. Two fundamental sequences {q”}. For a domain D. each of the following three conditions is equivalent to the condition that D is tsimply connected: (1) For every cross cut Q of D. A history of pi. W. U U Qnml) is simply connected. Caratheodory [2]. and denote by D.. separates qy-1 and qv+l on D..Q has exactly two tconnected components. (2) Half-plane: Re z > 0.(D. D.c I< R. the boundary points with x = 0. 16 (1962). Let D be a simply connected domain.. Daykin. 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. (2) Every Jordan curve in D is thomotopic to one point. Choong. Here D. Fig.c I< r. [4] A. . Rathbone. Continued fractions. and a plain domain is called simply a domain.. either the interior or the exterior of C is contained in D. 1 Let the domain D be bounded by a smooth Jordan curve. Domains in the Complex Plane A +domain (i. (2) For v-00.. and C. If D is an n-ply connected domain. . {q\} of cross cuts are equivalent if every D. Y. This notion is due to C.. Pz. 76-99.O<y<l/2(n=l. Comp. under which the equivalence class of fundamental sequences of cross cuts is called a boundary element.2. or Im z > 0. contains all qk except for a finite number of v. that is. Math. consists only of the boundary points of D.. Shanks and J. 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. Some typical examples of domains are as . Rational approximations to rc. Boundary Elements 333 (11. For example.332 Ref.O < y < l/2 are all inaccessible (Fig.. (3) The tmonodromy theorem holds in D. E. the sequence qv tends to a point on the boundary.1 suitable mutually disjoint A boundary point P of a domain D is called accessible if there exists a sequence of points P. R.

for all v. Meusnier (1776). Elements of the topology of plane sets of points.77 Conformal Mappings). Caratheodory. Press. then we say that the sequence {GY} converges to K..} has the same domain kernel K. This experiment was performed by the Belgian scientist J. y) having common boundary r. By equating to zero the +lirst variation of the areal functional based on infinitesimal displacement in the normal direction. Domain Kernels P=z’ q=&’ Let {G. for all v. and R. Ann.} be a sequence of domains containing the origin 0. Plateau (1873) to realize minimal surfaces. where H = (R. 73 (1913). u satisfying E = .’ + ture of the surface +radii of principal second differential R. The +Euler-Lagrange differential equation for the functional J[z] is References [1] M. except for the endpoint of F lying on the boundary of the domain. v) by means of parameters u. hence Plateau’s problem is that of determining the minimal surfaces bounded by given closed space curves. The notion of domain kernel is important in considering the limits of a sequence of conformal mappings (. [2] C. If every infinite subsequence of {G. r=a2. 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. with n = n(u. the 334 (X. Since +Beltrami’s form satisfies A. Let its +first fundamental form be dx2 = Edu2 + 2F du dv + G dv2 and its tsecond fundamental form be -drdn=Ldu2+2Mdudv+Ndv2. we put K = (0). A. each point of a slit domain. 1951. i. which is a secondorder tquasilinear partial differential equation of elliptic type and whose geometric interpretation had already been given by M. 323-370. except for a finite number of v. Origin Because of surface tension. C. To formulate the problem more generally. u) the unit normal vector. B. A. we obtain the EulerLagrange equation in the form 2H=(NE-2MF+LG)/(EG-F’)=O.33) Plateau’s Problem A. Then under suitable assumptions on the smoothness of z(x. Newman. or (1 +$)r-22pqs+(l +$)t=O. H. A closed domain is usually considered to be the union of a domain and the set of all its boundary elements. 2 aZ C. 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. Fig.1239 334 B Plateau’s Problem multiply connected domain is defined similarly for each isolated component of the boundary. y) has r as its boundary. The union K of such domains G is called the domain kernel of the sequence {G. determines two distinct boundary elements on each side. Let D be the finite domain bounded by C. a surface of the least area.‘)/2 is the +mean curvaand R. Cambridge Univ. We consider surfaces z = z(x.x = Hn. are the curvature. Various notions of tideal boundary come from considering suitable boundary elements for various purposes (. there exists a domain G such that every closed domain containing the origin and contained in G is contained in G. t = d2zldy2. For example. Math. If a suitable neighborhood of the origin is contained in G.. y). the problem is to minimize the tfunctional J[z]= JJ /mdxdy. s = a2z/axay. If there is no neighborhood of the origin contained in G.e. let a surface be expressed in vector form x = x(u. aZ with the condition that z = z(x. It is a problem of the tcalculus of variations. u.2/ax2.207 Ideal Boundaries).} (Carathtodory). Uber die Begrenzung einfach zusammenhangender Gebiete. the shape of a minimal surface. second edition.

when r is assumed merely to be trectifiable. New Developments Among recent contributions to the study of Plateau’s problem.. However. Rad6 (1930). the existence of a solution was first shown by the limit process by T. CarathCodory’s result on boundary correspondence (. in connection with the +monodromy group concerning a second-order linear ordinary differential equation. Bernshtein (1910) from the viewpoint of a tboundary value problem of the first kind for the elliptic partial differential equation in the previous section. Relation to Conformal Mappings There is a notable relation between Plateau’s problem and conformal mapping when the dimension of the space is 2.e. Based on these facts. He further discussed the general case where r can bound a surface with finite area. Douglas transformed Dirichlet’s functional with harmonic argument functions into a functional whose arguments are boundary functions. Subsequently.275 Minimal Submanifolds). In this formulation. 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. N. (2) Dirichlet’s functional for a scalar function f(u. . R. Haar (1927) dealt with the minimal problem for the functional J[z] by a tdirect method in the calculus of variations. i. Riemann. 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]. Let q(u.lse also. Courant. u). u). Courant. the problem can be easily generalized to an n-dimensional Euclidean space R” (. 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]. by starting from the problem of minimizing Douglas’s functional 1 2n G.u)isdefined byD[f]=~~(f. if the domain of w is the unit disk IwI < 1.77 Conformal Mappings). R.2+f. a minimal surface is defined as a surface with everywhere vanishing mean curvature. n) by D[f] = &.334 c Plateau’s 1240 Problem from the variational problem of minimizing D[f]. F. E.bka2 0 o 4sin2(0-(p)/2 dOdq. and for an n-dimensional vector function f(u. and others treated the generalized case of a fmite number of boundary curves. In general. At present.u) (w=u+iv.Z)dudu. Previously. the following remarkable with components fj(u. The existence of a solution of Plateau’s problem can be discussed by starting . and Plateau’s problem is to determine the minimal surface with a preassigned boundary. C. F.]. The Generalized Case Up to now we have been concerned with Plateau’s problem in the case of a single simple closed curve. C. D[f. by Rad6. Douglas (193 I) succeeded in giving a satisfactory result for the existence problem. Then isothermality is expressed by the condition that the analytic vector ~(w)=x(u. Douglas. J. Garnier (1928) investigated the existence of a solution by the limit process when l. The variational equation of D[f] is Af= 0. Osgood and C. . u) representing a minimal surface is harmonic (the components of this vector are tharmonic functions of u. where +genus and orientability are assigned as the topological structure of the surface to be found.u)+~~(u. respectively. the methods of discussing the existence of solutions of Plateau’s problem can be classified into the following three sorts (represented. Specifically. (3) An analytic vector s(w) is representable in terms of the boundary values of its real part. .i= J-1) satisfies r(w)’ = 0 (Weierstrass). the existence proof of the solution of the former for a Jordan domain implies +Riemann’s mapping theorem together with W. Courant (1937) gave another existence proof by reducing Plateau’s problem to the IDirichlet principle [3].is a simple closed curve with bounded curvature. and others discussed the case where the given space curve r is a polygon. A. u) (j = 1. u) we can prove the existence of solution Plateau’s problem satisfactorily. the vector function that minimizes the Dirichlet integral among functions with fixed boundary values is harmonic. On the other hand. The variational equation of the areal functional becomes H = 0. u) be a harmonic vector conjugate to X(U. of D. On the other hand. the vector x(u. On the other hand. For instance. The existence of a solution has been shown in this c. Namely. F = 0 are chosen. ss 2n 2n h(4 . Weierstrass. and Douglas): (1) The first method is to minimize directly the areal functional jj Jwdu do. by introducing a new functional depending on boundary values instead of the area1 functional. Existence of a Solution The existence of a solution of Plateau’s problem was discussed by S. . Schwarz.

Poincare. [ 1 l] F. Nitsche (1973) proved the uniqueness of solutions for analytic boundaries of ttotal curvature at most 4a. Henri results have emerged. [S] R. 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. 1966. [3] R. J. 1916-1956. La valeur de la science. 1914. teau’s problem. References [l] T. These points are called branch points. In addition. Subsequently. After graduating from the Ecole Polytechnique. Hildebrandt (1969) and others proved that if the boundary is of class Cm+. Springer. Sot. Osserman also gave examples of generalized minimal surfaces in R” (n > 3) with true branch points. Almgren. Anal. The existence and regularity of solutions of Plateau’s problem from this point of view have been discussed by H. Math. Trans. Arch. H. Poincare opened the road to talgebraic topology and made suggestive contributions to the ttheory of relativity and tquantum theory. [2] J. and his popular philosophical works concerning the foundations of natural science and mathematics exhibit a lucid style. algebraic geometry. Springer. Multiple integrals in the calculus of variations. His tuniformization of analytic functions by means of the theory of tautomorphic functions in 1880 is especially notable. [9] J. [4] H. 47-82. Gauthier-Villars. Geometric measure theory. However. 1950. Erg. Springer. Flammarion. Van Nostrand. Courant. . An invitation to varifold geometry. Next. PoincarC. 319-329. his work covered many fields of mathematics. Benjamin. 1975. C. 52 (1973). we mention the question of boundary regularity. Boundary behavior of minimal surfaces. C. W. There are also some recent results on the number of solutions of Plateau’s problem. Rado. conformal mapping and minimal surfaces. Science et hypothese. Reifenberg (1960) and others. Arch. Federer (1969). i. including arithmetic. 33 (1931) 263-321. (2) 91 (1970). For instance. Further developments in connection with Plateau’s problem have emerged in the work of E. In this connection. Gulliver (1973) that for n = 3 the mapping of Douglas’s theorem. 1892-1899. A new uniqueness theorem for minimal surfaces. 550569.. D. Dirichlet’s principle. 1969. Vorlesungen iiber Minimalflachen. A proof of the regularity everywhere of the classical solution to Pla- References [l] H. His paper on the tthree-body problem won the prize offered by the king of Sweden in 1889. 1854-July 17.1899) began a new epoch in celestial mechanics. Plateau’s problem. [3] H. Flammarion. GauthierVillars. The methods he developed in his threevolume Mkanique cdeste (1892. It was then proved by R. Lewy (1951) proved that if the boundary of a minimal surface is analytic. Rational Mech. is free of branch points. Springer. [S] S. is an immersion. 35 (1969). He died in Paris. 1969.39) PoincarQ.e. His achievements center on analysis and applications to theoretical physics and astronomy. He was made a member of the Academic des Sciences in 1887 and of the Academic Francaise in 1908.. 1966. [4] J. He asserted that science is for science’s own sake [4]. Nitsche. Poincare. Solution of the problem of Plateau. On the problem of Plateau. Gulliver also dealt with an analogous problem for surfaces of prescribed mean curvature. C. Morrey. 1903. R. Oeuvres I-XI. Poincare. Rational Mech. 335 (XXl. Henri Henri Poincare (April 29. France. Math. C. H. Les mithodes nouvelles de la mecanique celeste l-111. which is a surface of least area. and others [lo]. Hildebrandt. Nitsche. One of them is connected with the final result of Douglas (1939) on the existence of solution surfaces. Osserman. who sought to minimize the tHausdorff measure among general classes of geometric objects. Anal. Interscience. Ann. 1933. F. B.. Almgren. Fleming. Poincare. [2] H.. A survey of minimal surfaces. Douglas. C.. [7] R.1241 335 Ref. Amer. then the surface is analytic up to the boundary. he taught at the University of Caen in 1879. Osserman (1970) and R.. Federer. not as parametrized manifolds. and topology. Math. spectral theory. [lo] H. Osserman. 1912) was born in Nancy. then at the University of Paris in 1881. J. but as subsets of R”. C. J. m > 1. S. the surface is also of class Cm*” up to the boundary. [6] C.

(x) is the best approximation for .. a function P.(x).. i = 0. k=l (1) such that the inequality If(o)-P.. To prove version (i) of the theorem. Thus we do not always have a unique best-approximation polynomial [S] H. where x0. Best Approximations Let qo(x).c.P. Then x” . but such an approximation is not always unique. . be a sequence of linearly independent continuous functions on a bounded closed domain A in R”. + a..(x)} satisfies this condition..f(x) defined on [a. cosx. b].. + t (a. For example. x. The ..x*. b]. b].(x) is a trigonometric polynomial of the form (I).20) Polynomial A. several mlethods have been developed to find it (. b]. the system of functions {‘pk}k=o.2. .(x)}.. . x + a.. Degrees of Approximation Continuity and Moduli of For a continuous function .(x). . Stone obtained a theorem that generalizes Weierstrass’s theorem to the case of functions of several variables.(x) = 2-(“m1)T. (ui i:< the ordiPut E. We have the following generalization of(i): Let p. B. marion. when the set 4 c R” (n > 2) contains three nonintersecting arcs emanating from a common point.(x).. where T. where ck is indepen- . 336 (X. Then E.cp.T(x) by a linear combination of (cp. then for every E> 0 there exists a polynomial P. Then linear combinations of x0 = 1. A admits no Chebyshev system. where f is a continuous periodic function of period 271 and P. with real coefhcients such that max Ix”--a.P. sin nx} on [0. the modulus of continuity of ktb order is defined by B.-ix”-‘-.(x)=cos(narccosx) is the Chebyshev polynomial of degree n.. . < x.hlf(x)-~‘.. For such an approximation to be unique it is necessary and sufficient that the determinant of the matrix (qk(xi)) (k.(x)=C”. we can apply tFejer’s theorem on +Fourier series. we can assume that the given function f(x) is defined in the segment [0. b]..*(f)<c. (ii) If . To prove (ii).)-P.. x. b]. h. is called a Chebyshev system (or unisolvent system). consider the polynomial P.. . .a.w. Then P.(x) be a linear combination of q.cos k0 + bk sin kO).max. 11.1 -1 GXCl takes its smallest value.. Xpz. .P”(x)1 attains its maximum (these points are called deviation points) and (f(xi)-p.--a. rr] are Chebyshev systems. xnml + ..(x) = a. sets { 1.=.(x)1 C.(x) that is not identical to the function f~C[a. holds for all values of 0.h.(x)l.. C. be a sequence of positive numbers such that limp. .(x) attains inf max I._. Xk(l Ii=000 Then B.>. . For any given n there is a best approximation of Jr(x) by a linear combination of q. H..336 A Polynomial 1242 Approximation Science et mtthode. .f(x) if and only if there are at least n + 2 distinct points xi < .(Xi))(f(Xi+. 1. M. . However. we mention two simple ones.(Xi+ I)<0 (i= 1. . where If(x) . . Of the many direct proofs now available for Weierstrass’s theorem.(x)=if . Consider the Bernshtein polynomial -xp. and conversely.(f. we have Weierstrass’s approximation theorem. b]. Since the best approximation is desired for numerical computation. { 1. In particular. General Remarks Approximation On the existence of polynomial approximations. q.(x) of degree n = n(s) such that the inequality If(x) .(x) converges to f(x) tuniformly.. C161.f(0) is a continuous function of period 2n. . cos nx} on [0. For a Chebyshev system {qk(x)} on [a..300 Numerical Methods).. If {cp. Flamis called the best approximation of . then corresponding to every positive number E there exists a trigonometric polynomial of degree n = PI(E). nary modulus of continuity.+~ of [a..(x). let P. are n + 1 arbitrary distinct points of A (Haar’s condition) (Math.. ‘p.f(x) .._. . ~JX). Poincare. . The second form of Weierstrass’s theorem follows from the first. P”(0) = u.P”(x)1 GE holds throughout the interval [a. which is formulated in the following two forms: (i) Iff(x) is a function that is continuous in the finite interval [a. For any given continuous function f(x). x”} on [a.. 1908. can uniformly approximate each continuous function on [0. l/(n+ 1)) (Jackson’s theorem [I]). Ann.C”xtA for t <(b -a)/k. = co. xp’. 78 (1918)). n) is not zero. rr] and {sin x. 11 with arbitrary precision if and only if C pi’ = --co (Miintz’s theorem). p2.*(f)= info“.(O)1 <F... n + 1) (Chebyshev’s theorem). xi.

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

.(z) on D for R > p (Bernshtein and Walsh [12]).. y) with a pole at infinity. When zp’ is independent of the choice of n.. then there exists a polynomial rr. When .).f(z) n..f’(z1)(z2=z. Lagrange’s Interpolation Formula For each n (n = 0. the limit poly- w(z)=(z-zyy. ).(z)=a. This is called Lagrange’s interpolation polynomial and is given by T. then P.(Z)} is +overconvergent).(z) coincides with the sum of the first n terms of iNewton’s interpolation formula. On the other hand. and so on. The sequence P. when D = [ -1. ..(z-z?$). For real variables also.f‘(z) is holomorphic on D. .( z) of degree n that minimizes the integral ~eu(.n.z”~‘+.(j). y) = log R > 0.2.(z)] < M/R” for some R > 1 (actually {Z.(z)1 < M/R” for ZE D. let z(i”).. Then there exists a polynomial n.(z)] is replaced by methods using a +curvilinear integral or tsurface integral.)(z-z. P. which is the ordinary (real) Chebyshev nomial.-~. In particular. then n.(z) converges to j(z) only at the point 1. On the degree of approximation of polynomials to f‘(z) on a simply connected domain. n + 1). by approximating f(z) = z” by polynomials of degree n .(z)( attains the infimum E. The polynomial n.#z. be a given set of real or complex numbers..(x)=cos(narccosx)/2”-‘. If D is a closed ...+a. I.(z)lPldzl (p>O)..1)st roots of I as zp).(z) is called a Chebyshev polynomial of degree n with respect to the domain D.(z).f(z) is holomorphic in Iz( <p (p> l).P.f(z) is single-valued and holomorphic at every interior point of D. zy).(z-z..(z) is called Newton’s interpolation polynomial and is given by P. Moreover. there exist polynomials P. there exists the largest number p with the following property: . 11. If(z) . Let D be a closed domain in the complex plane with a boundary C that is a rectifiable Jordan curve.. we still obtain similar results.z)If‘(z)n. . Lavrent’ev.)+. be the locus G(x.(z)].(z) of degree n such that max.. there are examples of divergent P.+a. where M is a constant independent of II and z. and f(z) a continuous function on D.)-f(z.(z-z. we can show that there exists a polynomial T.(z) with the (n+ 1)st roots of 1 as nodes converges to j’(z) uniformly in IzI < 1. +a.l 1 I =:IT.n.[17].rc. tion of a continuous function defined in E was given by M.(z) converges to f(z) uniformly on D. .(z-z. and p(D) coincides with the tcapacity and ttranstinite diameter of D [ 151. al=(~(z. Furthermore.))l(z. Similar statements are valid for functions of a real variable. G. Convergence of Newton’s interpolation polynomial is closely connected to convergence of +Dirichlet series.f(z) is single-valued and holomorphic on D.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. For new results and applications of Chebyshev polynomials . E.(z) = z” and P. Generally. there exist no such polynomials P..)...f(z) n.. Then there is a unique polynomial of degree n that coincides with f(z) at each point zp) (k = 1.. =. where ao=S(zl). then P. If D is simply connected and j’(z) is single-valued and holomorphic on D. Suc- exists. if we take f(z) = l/z and the (n $. if .336 G Polynomial 1244 Approximation cessive coefficients of the polynomtal P.(z)1 $ M/R”.(z) of degree n (n = 1.(z) does not always converge to j(z). 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!$:!.) (z. as explained below.(z) of degree n such that min ~~~~z”+a.)+a.(z-z. and let f(z) be an arbitrary function.(z) is unique and is called the best approximation polynomial (in the sense of Chebyshev). If the method of evaluating the degree of approximation using the absolute value I. N. where u(z) is a given positive continuous function on C. P.. H. Let D. W.f‘(z) . Unifying these two results. However. we have T.(z)1 < M/n’R”. ) such that I. a. Sewell [ 141 proved the existence of a constant r such that I. S.. A.f’(z) .). Chebyshev Approximation Let D be a bounded closed subset of the complex plane.. If . For example. 1.(z) can be calculated by +finite differences.). In this case. in this case there exist a number M that does not depend on n and a number R > 1 such that If(z) . Moreover. If R < p. Assuming that S(z) satisfies certain additional conditions.

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

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

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

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). Also. L.)i~asymmetric polynomial of Xi. These potentials are tsuperharmonic in R” and harmonic outside the support of n..a2a. .337 J Polynomials . Because of its particular expression. the functions u(P) given by the integrals u(P) = su(P)= slog PQ dn(Q). X.) (Newton’s formulas). 1959. Moreover. . it is called a symmetric polynomial (or symmetric function) of X. a potential means a function u of n variables xi. can be uniquely expressed as a polynomial ‘p(n.=O(~=n+l.-~. it is called an alternating polynomial (or alternating function). .l. . Conversely.). Usually... . we have .. X. ..+X. van der Waerden.) is invariant under every permutation of Xi. cr.)=(X. a. mz..2r~~.)(X. some authors mean by Newtonian potential the function u(P)=JPQmldp(Q) in R3.)..-X..... then D(! i .-.. Interscience.. X.) is symmetric (alternating). . For Hilbert’s irreducibility theorem. s3 = 0: . = 0 of degree n. seventh edition. Let p(X .].(~.X.X. 1963.2. ch.~. 02. .) on= X.. X. Kurosh. Obviously. X.+2~:+4a. . .X.. the measure p is taken to be a nonnegative +Radon measure with compact tsupport.. For instance.. X.-4rr.. +. Conversely...1)/2 differences between Xi. .) [4] R. a. Lectures on general algebra. are called elementary symmetric polynomials (or elementary symmetric functions).X.u~-4u~u. . . .X” + a. PQ2-“&dQX n>3.gn are equal. . X”-’ + + a. .) be denoted by (. we have a. X.. is identical with the ring I[a.= CXiXj=X.+X. Chelsea.. It can be expressed in terms of coefftcients a. s2 = 0: .... = g. Hence p is an alternating polynomial of Xi. Hermann.+(-l)“~~~“~-lS.-4a.a.. Thus the totality of symmetric polynomials of X.X3+. 1966. a. X.. Let the coefficient of X”-k in the expansion of (X .28) Potential Theory A.) obtained from f(X... Concerning the elementary symmetric polynomials and the s.. .. an of I is called a symmetric (alternating) function of s1i. It is called the simplest alternating polynomial of these variables.n+2 )...... ..) c3u/i?x. Springer. and it is therefore a polynomial in 0i. . X. every symmetric polynomial of X. Diophantine geometry.+X. 1955. x2. This is called the fundamental theorem on symmetric polynomials. . Cours d’algebre.s..-Xx. . If f(X. where s stands for a symmetric polynomial.) is transformed into -..l)krr. Newtonian Potential In dynamics. .. [2] N. Then the polynomial p is invariant under even permutations of Xi. Algebra I..) (X. J. Lang. we have the relations S.a...=CXi=X. for every element cp of the polynomial ring 1248 gives a criterion for the condition tl at some of xi. H owever. 0...-..a. 338 (X. Bourbaki. Elements de mathematique. .a. if n = 2. References [1] B.-i) be the product of n(n. 8.(X -X.) = p2(X.. X. an expression . Hermann.)= 0 are typical examples of potential fu actions. X.-Xx. .cp(o. with elements x. we have a:D = a~a~+18a. . X. . If the characteristic of I is different from 2. cc. Ind.... xn if f(X.S.D = a: -4a. For example. 1962. . gz + 3a. these are symmetric polynomials of X.a..30. . X.r.. ch.). Thenwehavec. .) is called the discriminant of the equation. . ...) by every +odd permutation of X. ICr..X.-ais. If r*i.+. 1102b. x..f(ai. o.. . and gi.ands.=O(v=1. .--27u~u~.. .+~~S.2. (Original in Russian... and p becomes -p under odd permutations. .~~-..)(X. Discriminants The square D(X. n = 2. Godement. .X.).. p is also called the difference product of Xi.f(X. [S] S. . it can be written as . .+(--l)“cT&. X...=a~-4a:~. z. u1 . ..f=ps.) by replacing Xi...) of the simplest alternating polynomial p is a symmetric polynomial.. respectively.). .. second edition. 1970. of the equation. X... D(cc.f(X. .. They are called the logarithmic potential and Newtonian potential. for s. Algebre. Because of this close relation between potentials and harmonic functions. .s.+ (-l)“va. = C Xi’ (v = 1. . . 4. if n = 3. X. .X. .. . Actualites Sci.. (X. u...Y. are the roots of an ialgebraic equation a.X. . X.. .a. G...) gives a field of force in the n-dimensional (n > 2) Euclidean space R” Given a point P in R” and a measure p. an alternating polynomial f is divisible by the simplest alternating polynomial p. such that -grad u = -(fh/dx..X.. .. sometimes potlential .. -X. . [3] A.

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

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

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

[14]. in Q. the balayage principle. The converse is true if @ is positive. Hunt obtained a relation between this principle and the representation of GTf in the form s: P.2 d n <n) in R” and the Yukawa potential with kernel aPQm’ exp( --im) in R3 satisfy the complete maximum principle. ICYis equal to 0 ‘or a Hunt kernel satisfying the complete maxi-mum principle.. of Radon measures of general sign with compact support. then @ satisfies the domination principle or the inverse domination principle under some additional conditions (Kishi). The linear mapping G* of C. into C that is determined by Ifa’Gp = l G*fdp is called a transposed mapping. A Hunt kernel satisfies the domination prinlsiple and the balayage principle to all open sets. For a convolution kernel K satisfying the domination principle. In a special case... The Fourier transform of such a sernigroup has a closed connection with a negativedefinite function [l]. Diffusion Kernels Potential the domination principle (also called Cartan’s maximum principle). p) d @(P. where cp is a continuous exponential function.p)<O(P.}~~~ is sub-Markovian. then the same inequality holds in 0.).20 =E.e. continuous in the wider sense. implies the same inequality in 0 for p E E. Potentials with Distribution Kernels A function f in R” is called slowly increasing in the sense of Deny if there exists a positive . A diffusion kernel G on a locally compact Abelian group induces a convolution kernel K if G is translation-invariant. if @ satisfies the lower envelope principle on every compact set considered as a space. Then c’ satisfies the balayage principle if and only if G* satisfies the domination principle (Choquet and Deny). A kernel @ is said to satisfy the lower envelope principle if given p and v. A. Other Principles A kernel @ is said to satisfy the uniqueness principle if p = v follows from the equality @(P. the domination principle implies Frostman’s maximum principle [ 171. jL) = min(Q(P.(P. into M that is continuous with respect to these two weak topologies is called a diffusion kernel. p). < 1. we now introduce weak topologies in the space C of continuous functions in R and in the class A4.. [12].261 Markov Processes). Similarly. p E M. His result is important in the theory of +stochastic processes (. and a > 0 (Cartan and J.S. and it satisfies the complete maximum principle if and only if {LY. v) + u on S. which asserts that if the inequality O(P. Relations between this principle and some other principles were studied by Kishi [ 141. Convolution Kernels M. C’hoquet and Ohtsuka made a local study. It has a unique decomposition IC= cp.v) on S. v). For the equilibrium and domination principles for nonsymmetric kernels . We can define the balayage principle for G and the domination principle for G* as in the case where kernels are functions. and w2 [S]. + ICY). or.fdt with a tsemigroup P. Ida. Deny. It is called a Hunt kernel when there exists a vaguely continuous such that K = s. Potentials of order CI (n . G. (IC. and ICYis a singular kernel satisfying the domination principle such that ICY* E. p) < @(P. for PEE and an)i I’. 181. A kernel is said to satisfy the complete maximum principle if the inequality @. A positive linear mapping G of M. equivalently. any v. 1950). symmetric. which is valid n.dt and cl0 semigroup {4. Ninomiya and Kishi studied this principle. P. v) is valid on . If Q. we introduce weak topologies in C’. = ICYfor every XE&. This principle implies both Frostman’s maximum principle and the domination principle. X is thin (internally thin) at P if and only if P is an externally (internally) irregular point of X (Cartan [:5]). for . and in the class M of measures of general sign with not necessarily compact support.. Conversely. v)). p)=@(P. and finite outside the diagonal set. Corresponding to inner and outer capacitary mass distributions. we can examine inner and outer balayage mass distributions and inner and outer Gauss variational problems and their coincidences [S. the Dirac measure at the origin. 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. @ is said to satisfy the inverse domination principle if the inequality @(P. The complete maximum principle has been defined and studied for G* (Deny). c:. By means of the bilinear form Sf& (f~ C. While all principles discussed so far are global. @(P.LLLEEand any v implies the same inequality in Q. then @ satisfies the lower envelope principle on every compact set considered as a space. there is a i such that @(P. With respect to the Newtonian potential. satisfies the domination principle and 6 satisfies the continuity principle. 0.338 M 1252 Theory N. the inequality is valid for all relatively compact open sets w. In contrast to the domination principle.

D. [3] M.193 Harmonic Functions and Subharmonic Functions. Sci. 57 (1975). and for any compact set K c fi. Acad.. 22 (1946). Furthermore. the method of projection is applicable. 3 (1935). These results are due to Deny [9]. Math. Beurling. Acad. 139-215. Mat. Dirichlet spaces. [4] M.2)7c”‘*/I(n/2). Choquet and J. Springer. 167197. balayage. J. 27 (1966). I? D is dense in both C. and the complete maximum principle hold [Z]. R. In the case of the Newtonian kernel. [2] A. (2) 68 (1944). For any pure potential. Medd.~~=. Then we call D special and characterize it in terms of a real-valued continuous function on 0 [2]. [8] G. Every ordinary potential of a double layer is a special case of UT. Paris. Foundations of potential theory.S. The notion of Dirichlet space was introduced by A. Suppose that n is a locally compact Abelian group and D is a Dirichlet space such that U. Potentiel d’equilibre et capacite des ensembles avec quelques applications a la theorie des fonctions. Centre de Documentation Universitaire. Paris. Grenoble (N.ull= Ilull for every UED and ysR. then u is called a pure potential. C. Q. in R” for any TE W. Choquet. These A are square integrable. 107-183. Nagoya Math. and D. functions are assumed to be complex-valued. Berg and G. yj are components of P. Sci. For instance. 23 (1963). FT.. References [l] C. Proc. n D. 12-36. Dirichlet Spaces In this section. Kellogg. Given such a distribution K. If in addition p > 0. Nagoya Math. the balayage principle. = 2(n . the family of Newtonian potentials of measures with finite energy is not a Hilbert space (Cartan). T2) = j kFT. Nagoya Math. 221-280.. 244 (1957). Deny. Sur les fondements de la theorie fine du potentiel. and problems of equilibrium. J. On the theory of potentials in locally compact spaces. [S] H. Sem. Theorie g&ale du balayage en potentiel newtonien. R. Choquet. respectively. third edition. Les potentiels d’tnergie linie.. and capacity can be examined.afi/ax. Deny.and IITll*=~. 133137.. aUT/axj =fj is defined a. then the corresponding capacity is the Lebesgue measure. [ 111 B. Let 0 be a locally compact Hausdorff space. Acad. [lS] M. 1707-1710. The family of functions of class C” with. US.C3=~Slfil~dz. [14] M. Sci.) integer 4 such that jf(P)(l + OP2)-4dr(P) < co.. called the K-potential of T.. 1974. 160661609. Suites de potentiels. Ito. and C. the equilibrium principle.1253 338 Ref. Lecture notes in math. Bull. [12] M. Sur les ensembles efliles. 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. 404.e. dr. Cartan. 5 2 0 a Radon measure in 0. 165-187.).. Ann. 45 (1959). Elements de la theorie classique du potentiel. Sci. Since W is complete. 103 (1960). [6] G. Deny. 208215. Acta Math. C. j=l k Q. Acta Math. Caracterisation du principe de domination pour les noyaux de convolution non-born&. Brelot. UT=(n-2) where xi. However. the lower envelope principle. Univ. Springer.60112. [9] J. Nat. cp)= j cpdp holds for every . J. For every TE W the function FK x FT is slowly increasing in the sense of Deny. Kishi. [ 131 0. T= -c. 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. Potential theory on locally compact Abelian groups. Fuglede. if tDirac’s distribution 6 is taken as a distribution kernel. Lunds Univ. [lo] 0. there exists a constant A(K) such that jKIu(dc<A(K)IIuII for every u E D. [7] G. Springer. 1929. PotentiaI Theory cpE C. Noyaux de convolution et balayage sur tout ouvert.compact support is tdense in W. An existence theorem in potential theory. Beurling and J. the space of continuous functions with compact support. 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 . 1975. Maximum principles in the potential theory. Conversely. A function u E D is called a potential if there exists a Radon measure p such that (u.. There exists a distribution U = UT that satisfies FU = FK x FT. Forst. 1965. Kishi. Brelot. A Hilbert space D consisting of locally t-integrable functions is called a Dirichlet space if C. Then f is equal to the potential of some TE W with Newtonian kernel up to an additive constant. where l/c.. 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. 244 (1957). Frostman.u(x) = u(x Y)ED and IIU. 82 (1950). s(xj-Yj)~(Q)PQ-“dz(Q). Potentiels sur un ensemble de capacite nulle. (For axiomatic potential theory .

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

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

21 Analytic Functions of Several Complex Variables.. Oxford Univ. Springer. .370 Rings of Power Series. . If logn. Clarendon Press. and interaction between nearest neighboring sites is taken into consideration.2. [S] N. Tauberian theorems..) is overconvergent.references to 198 Holomorphic Functions. S(&. Univ. Les singularit& des fonctions analytiques represent&es par une strie de Taylor. if the zeros of a subsequence S(n. Actualites Sci. Hermann. Math.2. Ostrowski.z) (k=1. z). Bieberbach. =O(n. but a suitable subsequence S(n. Gauthier-Villars. Mandelbrojt. If this situation occurs. Bourion. (1921).“. then all boundary points of the domain of overconvergence are accumulation points of the zeros of S(n. Abh. Tauberian theorems.. we explain here (I) Ising models.z’n has a lacunary structure if the sequence { 1. Ostrowski. Fortgesetzte Untersuchungen iiber die Abschnitte von Potenzreihen. king Models References For the general theory of power series.. z) (k = 1. 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... This result includes Hadamard’s gap theorem as a special case. S] called this phenomenon overconvergence and proved the following result: By definition. [7] A. in which either a + or spin is put on each site of a crystal lattice. [9] G. A known result is that if the power series C c.-B. Press. (4) random Schrodinger equations.y” (n > 1) are all integers (Eisenstein’s theorem.) and S(n. Sem. [lo] R.2. 1958. R. Hamburg.. cw A power series is completely determined by its coefficients. [6] H. -1)’ be the totality of spin configurations in V. For theorems of Tauberian type. The Taylor series..2. ) converges uniformly. Wiss. ). Akad. Ostrowski [7.. Wiener. the sequence of partial sums S( 1.+. B.. For singularities. Acta Math. Preuss. . . Ind.z” with rational coefficients represents a branch of an talgebraic function. L’ultraconvergence dans les series de Taylor. for formal power series . G. On the other hand. and (5) the +Boltzmann equation. . [l] E. = { +l. 1 (1922). Each ele- .. Ann. Overconvergence If the radius of convergence of . then we can find an integer y such that the c. 1852). 1931. ) has no accumulation point on IzI = 1. where S(n. then in a s&iciently small neighborhood of a point on IzI = 1 where f(z) is holomorphic. (2) 33 (1932). Landau.) may still be convergent. R. [S] A. Analytische Fortsetzung. . [4] S. Bourion [9] gave a unified theory of these results using tsuperharmonic functions.. f(z) = C. Ising [ 11 to explain the phenomena of phase transitions of a ferromagnet. Dienes. 5577565. > E. 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. S(2. Let I’ be a cube in the d-dimensional integer lattice space Zd and X. second edition. (3) percolation processes.z” is 1.339 E Power Series E. (2) Markov statistical mechanics. 1937.} such that A. G.. For overconvergence.f(z) = C a..1 7) Probabilistic Methods Statistical Mechanics A.) has a subsequence {A. z) (k = 1. Table lO. Uber eine Eigenschaft gewisser Potenzreihen mit unendlichvielen verschwinden der Koefhzienten. z) (k = 1. 340 (XVII.. For power series of several variables . Math. 1929.IV.z) (Ostrowski 1256 [2] P. As examples of such methods. z). Conversely. 1955.(l + 0) (0 > 0). The king model was proposed by II. 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. Springer. Jentsch. Pitt. z) = C:=O uvzy. For power series expansions Appendix A. 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. . S. 41 (1918) 253-270. 1932. Darstellung und Begriindung einiger neuerer Ergebnisse der Funktionentheorie. A. Uber vollstandige Gebiete gleichmassiger Konvergenz von Folgen analytischer Funktionen. is naturally divergent for IzI > 1. . l-100. 3277350. z) (k = 1. Also . [3] L. then the power series has a lacunary structure and overconvergence takes place for S(n.Oa.

at present the probabilistic delinition of Gibbs states given by Dobrushin [2] and Lanford and Ruelle [3] is prevalent. We explain this field by looking at a typical class of processes. the free energy is defined by where /l= l/kT (k is the tBoltzmann constant. Then there exists a unique state g{. It is known that any Gibbs state of the Ising model is a reversible stationary state of the stochastic Ising model. (variational principle). there exists a critical value /l. every Gibbs state is Z*homogeneous [S. in which various physical quantities are calculated. This process on X is called a spin-flip model. C. For an initial distribution p we denote by pt the distribution at time t. and sociology are formulated as a class of infinite-dimensional tMarkov processes. Putting + or . and many models occurring A percolation process is a mathematical model which describes the random spread of a fluid .x. for a sufficiently low temperature. t + At] with probability ci(qt)At + o(At). Free energy plays an important role in the study of the ergodic properties of these models. 63. then S-i is a stationary state. and the converse is also valid. 0 < t < cc. -lJzd be the totality of spin configurations and an element of X be denoted by r={~(i)}~. Example 2. Percolation Stochastic Ising models. (0 < 1. h).6-i is a unique stationary state.spin on each site on Zd. which has made rapid progress during the last decade. +l.. and if 1> 1.” is called a Gibbs state on V of the Ising model with parameter (p. the phase transition never occurs. Let X= { +l. (3) In the 2dimensional case. the mean free energy is a nondecreasing functional along the distributions pLt.(i) changes to -v].e. Physically. Markov Statistical Mechanics in physics. T is the tabsolute temperature). Denote by -1 the configuration at which all sites are healthy and by 6-r the unit point mass at -1.. there is a Zd-inhomogeneous Gibbs state for sufficiently large p [7]. (i. there is another stationary state p satisfying p[qEX. < co) such that if A < 1.. On the other hand when an external field is absent (h = 0) and d 2 2. (2) For d > 2.&)= v-a -444 lim 1 VI is well defined. j) is a nearest neighboring pair of sites and h stands for the parameter of an external field. a (limiting) Gibbs state can also be defined for the infinite domain V= Zd by the above mentioned variational principle. q(i)= +l for infinitely Processes many i] = 1. The calculation of b. The most important result is the following: There exists a critical value 1.h(a) exp(-8Wd) = c . In the following we assume h = 0. Let Zd be the d-dimensional integer lattice space. biology.(i) in the time interval [t.e. The flip rate of the contact process is given by 1 ci(T) if if q(i)= q(i)= #{jllj-il=l. The process is described in terms of a collection of nonnegative functions ci(q) defined for iEZd and q E X. In particular. We suppose that a spin configuration 0 has a potential of the type ‘V(a)= 1 Oiiaj+hC CT. i. A stochastic Ising model was proposed by Glauber [8] to describe the random motion in a ferromagnet upon contact with a heat bath. there are at least two Gibbs states. Since for each Zd-homogeneous (i. we mention some known facts in this field. q. Finally.. Contact processes.(d) but it occurs for every fi > b. For the configuration qt at time t. The field of investigation of stationary states and statistical or ergodic properties of these processes is called Markov statistical mechanics. = ki Here + 1 denotes an infected individual and -1 denotes a. gj. p is called a stationary state. a Gibbs state is an equilibrium state. Stochastic Ising models.(2) has been carried out by Onsager [4].healthy one.(d) such that the phase transition does not occur for any b < b..j)cV isV where (i. A tprobability measure on Xv is called a state on V.ew-&424)’ aEXv. -1 and r(j)=+l}=k. D. However. a phase transition occurs. A contact process was introduced by Harris [ 1 l] to investigate the spread of infection.(d).~d (v(i)= +l or -1). j) means that (i. (4) For d 2 3. translationally invariant with respect to Z”) state p the mean free energy 1 . If pLt= p for all t > 0.1257 340 D Probabilistic Methods in Statistical Mechanics ment of XV is denoted by c= {ci}ieV (ci= $1 or -1). Then the flip rate {ci(~)j is defined by the potential of the Ising model. Example 1. For each state p. on V which minimizes the free energy F. It is known that if an external field is present (i. infinite interacting particle systems. let us consider a random motion of spins which evolves while interacting with neighboring spins.e. h # 0) there is only one Gibbs state for any fi. (1) In the l-dimensional case.

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

Rev. Advances in Math.rp. Lecture notes in math. Phys.X)= s SxR' (44 X’ML Y’) . Liggett. [ 1 l] T. Smythe and J. 1972. Molchanov.1259 340 Ref.(t) is the solution of the master equation (1) with u. Higuchi.(x. On the absence of nontranslationally invariant Gibbs states for the two-dimensional Ising model. Time-dependent statistics of the Ising model. on Random Fields. 41-59. Math. 235-241. and u(t) is the solution of the following Boltzmann equation with u(O) = u: &U(f.. based on Kac’s work [22]. Comm. 7 (1971). if {u. Harris. [ 131 T. McKean [24].(O) = u. L. [S] M. (Original in Russian. Probability. Broadbent and J. on S(. 149-164.. Onsager. Springer. = xi + (Xi . H. [17] I..z)z. 598.. 629641. a sequence {u. Moreover. [6] Y. 969-988. Goldseid. 253-264. . The propagation of chaos is the stage corresponding to the tlaw of large numbers. Ann. E. 65 (1944). and L. [4] L. Probabilistic distribution on S2 and x. Crystal statistics I: A twodimensional model with an order-disorder transition. 1968. Lecture notes in math.Z)Z. The stochastic evolution of infinite systems of interacting particles. such a Markov process was constructed by solving a certain tstochastic differential equation [27]. 194-215. more precisely. 17 (1972) 582-600.)u. The next stage is the tcentral limit theorem or fluctuation theory..(x... Appl. [ 121 R. Liggett. (Original in Russian... Phys. The critical probability of bond percolation on the square lattice equals l/2. Dobrushin. Markov processes with a large number of locally interacting components: Existence of a limit process and its ergodicity. A. if there exists a probability density u on R3 such that lim c~l(x. Theory Prob. Glauber.. 198-206. Springer. Dobrushin. Translation invariance and instability of phase coexistence in the two dimensional Ising system.. A. [2] R.} is a u-chaotic sequence. The description of a random field by means of conditional probabilities and conditions of its regularity. Phys. P. Ann. Observables at infinity and states with short range correlations in statistical mechanics. Z. 53 (1957). In the case of the spatially homogeneous Boltzmann equation of Maxwellian molecules without cutoff. a process of this type describes the time evolution of the velocity of a particle interacting with other similar particles.x. Gibbs state describing coexistence of phases for a three-dimensional Ising model. Beitrag zur Theorie des Ferromagnetismus.. Cambridge Philos. [ 161 H. Phys. Interaction of Markov processes.=xj-(xj-xi. Appl. The propagation of chaos was verified by Kac [22].. 197-224.1)-dimensional sphere with center 0 and radius . L. 73 (1980). Holley and T. One-dimensional random Schrodinger operator with purely point spectrum. L. 13 (1968).xi. 1971. Math.} x'=x+(y-x. Wierman. which was also discussed by Kac [23] and McKean [25]. Contact interactions on a lattice.) [lo] F. 671. Kesten. This implies the existence of probability measurevalued solutions of the equation.. [9] R. Probability. McKean [26] introduced a class of tMarkov processes associated with certain nonlinear evolution equations including the Boltzmann equation. 1978. [14] S. S. C. Ruelle. 1 <k < m. M. Aizenman. M.44 x)u(t.. 4 (1963) 294307. North-Holland. Y. Kac’s assertion is that the Boltzmann equation is to be derived from the master equation via the propagation of chaos. Spitzer. then {un(t)} is also u(r)chaotic. y'=y-(y-x.149. 246-290. Ising. 1977.)dx. J. Z)Z. M. Sot. References [l] E. Esztergom. Functional Anal. [7] R./&) for each n > 1. 1977. E. Hammersley.). Phys.. Firstpassage percolation on the square lattice.dx. 5 (1970). “-m s S(. 11 (1977). Lanford III and D. 6 (1978). Pastur.. Math. (Original in Russian. Phys..1)1. Percolation processes I: Crystals and mazes. T. J. Comm. Dobrushin. Let u be a positive constant and S(G) denote the (3n . ibid. l-10.(x)Wdx s for each m > 1 and (pkE C. The survival of contact processes. Problems of Information Transmission. Appl. (Originals in Russian.) [8] R. Comm. Proc. Conf. Proc.. 187-248. 2 (1974). J. 74 (1980)./&.) .. where u. Markov processes with many locally interacting components: The reversible case and some generalizations. 1981.) [3] 0. and others for a considerably wide class of nonlinear equations of Boltzmann type (with cutoff). Given a symmetric probability density u. 83-94. R. Math. Theory Prob. 31 (1925).} is said to have Roltzmann’s property or to be chaotic (or u-chaotic to stress u).JG) . Methods in Statistical Mechanics the uniform x..(R’). [15] R. 117. 13 (1969).= fi k=l R3 cp.

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

i. we have j. Specific Distributions Given an n-dimensional distribution @‘. we denote by yk the coefficient of (i~)~/k! in the TMaclaurin expansion of logrp(z). a point a with @({u}) > 0 is called a discontinuity point of Q. @(al) = 0. Given a l-dimensional distribution @ with fik < +co. and @‘3 is continuous and tsingular. . is also called the characteristic function of X.a.[8]. We call yk the (kth order) semi-invariant of 0. Table 22): the unit distribution with @( (0)) = 1. 1) is ~ called the standard normal distribution). n. . 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) (. t=u or b. and then f(z) uniquely determines @. y. a. z. f(t. .. and x=(x.x))d@(x) (zcR”) does not where Q1 is purely discontinuous. . n).] defined by ai6xi<bi(i=1.~.. We now list some frequently used ldimensional lattice distributions (for explicit data . b) denotes the modified function of [a. u. @ is called a lattice distribution. A complex-valued function cp of z E R” is called tpositive definite if it satisfies the inequality in (i). b. When O(D) = 1. By virtue of the tLebesgue decomposition theorem.%~o. The set D of all discontinuity points of @ is at most countable. The characteristic function is often useful for giving probability distributions explicitly. we have necessarily exist for all n-dimensional distributions but does exist for a number of useful probability distributions @..=a. In particular.W1. up.~z. where 81 denotes the boundary of I..ER.@. b. t+. which can also be written as E(ei@. (For characteristic functions of typical probability distributions .fl dz”’ ..e. @ is called a continuous distribution. .=l. y4=a4-3a:-4alag +12$a. the . . the bypergeometric distribution H(N. i 0. aD.a$ a. 4) with parameters m and q..192 Harmonic Analysis). every probability distribution can be expressed in the form @=u. ... CT*)with mean p and variance g2 (sometimes N(0. b. (iii) ~(0) = 1. is tabsolutely continuous with respect to tLebesgue measure. p) with parameters N... .. If the distribution function of @ is a continuous function.. Equation (2) is called the inversion formula for the characteristic function cp.Appendix A.@. A counterpart to Bochner’s theorem holds for any positive definite sequence as well (THerglotz’s theorem).. the binomial distribution Bin(n.-3a. zCp)of the n-dimensional space R” and any complex numbers a. p) with parameters n and p. . .b). l/2. The characteristic function cp of an ndimensional probability distribution has the following properties: (i) For any points z(l).2 0.342 Probability Theory).b)= t $ Cu. 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. xdz. x. indicator (2) where f(t. p) with parameters n and p. the negative multinomial distribution.z’k’)uja.+u. The following k-dimensional lattice distributions are used frequently: the multinomial distribution M(n.1261 341 D Probability Measures Then the Fourier transform of Q..+u.a. . ..+2a:.. If an n-dimensional interval I = [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.. is called a purely discontinuous distribution. a.+a. Let @ be an absolutely continuous distribution. . Q.) The moment generating function defined by f(z)=lexp( -(z.. b] defined by 1.Appendix A. (ii) For any up.. and p.. .. the multiple hypergeometric distribution.2.+u.. (p= 1. (ii) cp(zCk)) converges to ~(0) as zCk)-+O.< b. . the Poisson distribution P(1) with parameter i... Table 22. y2 = CQ-a: = c?. etc. This function f(x) is called the probability density of @. ... a. For general information about criteria that can be used to decide whether a given function is a characteristic function . The moments and semi-invariants are connected by the relations y1 = CI~.. . .X)) (. and the negative binomial distribution NB(m. then the left-hand side of (2) is equal to Q(I).. the geometric distribution G(p) with parameter p. n) is an interval of continuity for the probability distribution Q. D. . bl. .2. The following l-dimensional distributions are absolutely continuous: the normal distribution (or Gaussian distribution) N(p.. n.-6af.) E R”.. . if D is a lattice.

(2) At every continuity point of the distribution function F(x. Prokhorov in metric spaces [lo].. Let X. normal distribution with mean m and variance u.. Convolution Given any two n-dimensional distributions Or. (3) For every continuity set E of @ (namely.(y). etc. the Dirichlet distribution... )= @(cc...341 E Probability 1262 Measures random function c&j variables. If @. each of the following conditions is necessary and sufftcient.xJ = F(x.. (4) For all open G c R”..2). V. p.. *Qz. )there exists(x3. The metric p.. a l-dimensional distribution function I>O.q)* NB(m.. and OI. q).c2. If q(z) is the characteristic function of the convolution of two probability distributions a.F. lim. Yk= P(R.@‘2)=max(8s.(z) (p*(z).. Therefore...(x)=Sa.(G) 2 O(G). (1) For every continuous function with compact support. Linnik proved a similar fact (the decomposition theorem) for a more general family with reproducing property by using the theory of analytic functions [9].& . Such convergence is called weak convergence in probability theory. @I. .q)... (5) For all closed F c R”.f(x)d~. r(P2.p). * Qz is expressed in the form F.+n. we can use it to QF. where xE is the indicator function of the set E.~. such as the k-dimensional normal distribution N( p.a)* C(P.. ) to converge to @ weakly.rr:)=N(~l+~2. *Y2.d* r(~l.%)= Given F(x). u) =Ql em./I. uk (k = 1. etc.. then we say that ct.. and ‘pz. (D. n). N(~c.. P(i.x. the Z-distribution Z(m. the +X2 distribution x’(n). called the Levy distance. the exponential distribution e(a) with parameter a. there are several k-dimensional absolutely continuous distributions.) (k= 1. V. has a density fr(x).(y). etc.). W../l. lim inf. When R is a topological space.. then @. the beta distribution B(p. For a sequence of ndimensional distributions (I$ (k = 1. @I =O..2) exist such that N(m.. then cp is the product of ‘pr and cp2 q(z) = ‘p.. /I].4.Sn.)=uP. * F2 of @. . has a reproducing property.NB(m./?I . and @)2with characteristic functions ‘p. be tindependent random variables with distributions @.. )* @(x. the distribution function F.. ‘Pk (k= 1.(n:-y)dF. + v2.. Levy [6]). n). &. lim. Qz.. and @. Let N(m.~+t)d~l(~)d~Z(y) is called the composition (or convolution) of @. )suchthat@(cc. ). a) with parameters ..a) be the distribution obtained through translation by a of the Poisson distribution with parameter J. +I. Each of these conditions except (2) is still necessary and sufficient for @)nto converge weakly to @ when @. where p is a metric defined in the following way: Given any n-dimensional distributions @r.& . Convergence of Probability Distributions Nh. F.+P2.pJ and covariance matrix I= (a. Iffor(a. is @I * Qz When Fi is the distribution function of Qi (i= 1. C) with mean vector p=(pIrpz. of The concept of convergence of distributions plays an important role in limit theorems and other fields of probability theory. . o1 + is also useful for similar purposes.. /I) on an interval [x.2)... as Cramer’s theorem and Raikov’s theorem. (6) Iirk. Then the distribution of X. we have Qk= N(m. the n-dimensional distribution @(E) = SR2n~E(... II) be the 1-dimensiona. 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- . )and(cr. i?in(n. .)*C(/*. is called the maximal concentration function Since it satisfies the relation (i= 1.37 Banach Spaces).q)=NB(m...2) for some mk.I?‘) = 0). Furthermore. . the Fdistribution F(m.1))2 dx s cc Caucby distribution C(p. The mean concentration defined by . +X. * Qz has a density f(x) = JR”fi (x y)dF. the t-distribution t(n).. lim. E.f(x)d~.~:+~22).... i. %(F)< W’).2). was introduced by Levy [6] in one dimension and by Yu.F(x . x. the uniform distribution U(a. for every : k.). Some of the distributions listed above have the reproducing property: P(Ir)*P(I.) of @. we consider convergence of probability measures on 0 with respect to the tweak topology introduced in the space of measures on Q (.(x). the kth order semi-invariant of the convolution of two distributions is equal to the sum of their kth order semi-invariants. . .)= P(1.u (+median) and a. c). and let P(I.) (Fk is the distribution function of Q.. Ok(E) = D(E). QJl)=- g2).. lim sup. --mcx<‘x max (F(x+i)-F(x-1))..+m..*F2(I)<QFI(I) study the properties of sums of independent F (P. &(x1. we put ~~~=inf{~(@~i(F)<@~(F~)+ E for every closed F} (F” is the s-neighborhood of F) and define p(@1. )} indexed with parameters cc.. If l-dimensional distributions Qk..p)=Bin(n. Yu.a)=Y. . /I’~. * F2(x) = Ja”F. the gamma distribution T(p. and X. . . Suppose that we are given a family of distributions @ = {@(a. ok). respectively. (F(x + I) .2.. c~. and is denoted by Q.a..).2.. P(%. . These are known. a set such that Q(E .p)* Bin(n..

-+O as k+m. . i = 1. (For infinitely divisible distributions on a homogeneous space . Formula (4) is called L&y’s canonical form.Jz) = q(z). Then $ converges weakly to @ if and only if for every z. 1.(K’) <E for all aeh. This condition is sufficient but not necessary. It is called Kolmogorov’s canonical form. then its characteristic function is given by cp(z)=exp ( imz-iz’ G. . *Q2* . o is a nonnegative constant. and n(dx) is a measure on R” such that n({O})=O and -n(dx) < co. then Qk weakly converges to CD. n(k). (cp. If an n-dimensional distribution @ satisfies the condition slxl . If the probability distributions of the sums X.. . and the value of A(u.Y(E)=(D(IE)(1E={~~l~~ E}) for every set E.. v is a nonnegative constant. For the characteristic function of an infinitely divisible ndimensional distribution. . . Let (pk be the characteristic function of an n-dimensional probability distribution ak.. q. Let Xki. and their convergence [ 14161. If a l-dimensional infinitely divisible distribution @ satisfies JR1 x2 d@(x) < co. .) Let CDand Y be n-dimensional distributions. The method of moment-generating functions is also useful. G(u) is a nondecreasing bounded function with G(-co)=O. Let @ and Y be probability distributions with distribution functions F and G . . . 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 say that @ and Y are equivalent. Hence a tight family Qa (a E A) has a weakly convergent subsequence. We can give a criterion for the convergence of probability measures in terms of their characteristic functions.xjd@(x) (j= 0.1263 341 G Probability Measures tute a complete separable metric space with respect to the Levy distance. (4) where rn~:R”. = CyLkj Xki converge to a probability distribution as k+ co. xjd@. and lim.@(dx) <a. the use of tLaplace transforms as a substitute for Fourier transforms provides a powerful tool. then cp is also the characteristic function of an n-dimensional probability distribution CJand the sequence {@. z) - t P. z&R”. co). *$. @. and @ be l-dimensional distributions. Formula (3) is called Khinchin’s canonical form. z)( 1 + uz)/uz at u = 0 is defined to be . A(u.@. i= 1. Suppose that ak and CDare n-dimensional probability measures with characteristic functions (pk and rp.Eu(s) such that @=@. .J?.2. . )... (5) 5 -a’ An n-dimensional probability distribution @ is called infinitely divisible if for every positive interger k.4’1 cpqzpzq ei(zsx)-l -$A$ n(dx) . then the limit distribution is infinitely divisible. . we say that @‘E U(E). Then CDis an infinitely divisible distribution if and onlyifforeverya>Owecanlind@l. Both normal distributions and Poisson distributions are infinitely divisible.z)=exp(iuz)lizu/( 1 + u’). and K(u) is a nondecreasing bounded function such that K( -co)=O. ~~l/?[lu=cc forBj=S”“~lIxljd~(x)<co. A family Qa (aeA) is tight if and only if it is ttotally bounded with respect to the topology induced by the Levy distance. .) is a positive semidefinite matrix.. Infinitely Divisible Distributions + m (eizu -1-izu)$dK(u) > . If all absolute moments exist.. There are many results on the relation between these functions.. and assume that the distribution of Xki belongs to a(~&.5 Additive Processes. The method of probability generating functions is available for the study of arbitrary probability distribution concentrated on the nonnegative integers [14].2. be independent random variables for every k.(x)=J:. lim. The characteristic function of a l-dimensional infinitely divisible distribution can be written in the form cp(z)=exp ( m + 1 4% z).2. where m is a real constant.} converges weakly to @ [7] (Levy’s continuity theorem). @z. * @‘k (= @tk). probability distributions.1 y’ dG(u) > irz-gz2 -02 T (3) where y is a constant. .z2/2. . > > z=(z. If the sequence { (pk} converges pointwise to a limit function cp and the convergence of (Pi is uniform in some neighborhood of the origin. where s.. . . Ifforsome1>O. n(k). . For any probability distribution concentrated on [0. the canonical form is as follows: q(z) =exp ( + J( R” i(m. . there exists a probability distribution Qk such that @ = Qk * @‘r * . Let @r.

... V. Semistable distributions are another generalization of stable distributions.}. If F satisfies the relation 1 . For the normal distribution. We call @ a stable distribution if for every pair of distributions al.~~Jw)EE.t. and the stable distributions with exponent a = 1 are the Cauchy distributions. All L-distributions (and hence quasistable distributions) are unimodal [ 181. = y2 = 0.. Kolmogorov’s Extension Theorem $(z)=imz-c~zlz(l+i~(z/~zl)cr~(z. N. there exists a i = i(l.< < t.)/b. The characteristic function of a quasistable distribution has the canonical representation d4 = exp $(z)? I. Table 22). A necessary and sufficient condition for a distribution to be quasistable is that its characteristic function q(z) satisfy the relation cp(h.c0 lzla for a symmetric stable distribution. we have y.EkE93(R1). the distributions of the sums B. . and X.)-A.>O. Kolmogorov 1171 called stable distributions also. Furthermore.i2)/b2) = F( (x .. and A.. If for suitably chosen constants A. = /lx/a3 is used as a measure of departure from symmetry of a distribution and is called the coefficient of skewness.2.e. Quasistable distributions are L-distributions.O < n < 2.) be a sequence of independent random variables with identical distribution. -&). 1. b. A quasistable distribution with c(# 1 is obtained from a stable distribution by tran:jlation. A distribution is called semistable if its characteristic function q(z) satisfies the relation tj(qz) = q”$(z) for a positive number q (# l).Appendix A. converge to @ and sup. Let {X. ..)* F( (x . The stable distributions with exponent x= 2 are the normal distributions. 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 distributions of the sums B[‘(Cbl X. the general form was obtained by L&y [6]. we have I/J(. The ratio y. and B.0) <p < F(l../B. A l-dimensional probability distribution D is called an L-distribution if the distribution function F of @ is the convolutioc.<co. {o)~nl n..i)/b).2X2)/). In particular.z)cp(b. converge to a distribution. The Shape of Distributions Let F(x) be a l-dimensional distr bution function. i. We can characterize stable distributions in terms of their characteristic functions q(z) as follows: For every pair 1. the convolution @I * @)2 is equivalent to @. every distribution equivalent to @ is also stable. H. such that (i. i. which implies ~(z)=(-~~+i(z/~z~)cJ~z~~. wherec.‘(& X. where q(z) = exp(+(z)).. @)2equivalent to @. Then the following three statements are equivalent: (1) CDand Y are equivalent. . > 0 and real I. By the delinition we see that all stable distributions are infinitely divisible. of F(x/a) and some other distribution function F.3 is called the = coefficient of excess. it is called symmetric.2 is called a median. Given a probability . + j. which B. If y2 f 0.341 H Probability 1264 Measures (2/n)loglzl (SL= 1).z)= cp(bz)e@ (y = i -2. with identical distribution @ and for any positive numbers i. If @ is stable. the limit distribution is a quasistable distribution (L&y). The quantity cP such that F([.)-A. We can restate this characterization as follows: @ is stable if and only if for every pair of independent random variables X. > 0 and A. where T is an arbitrary index set.(x) for every 0 <a < I @ is an L-distribution if and only if there exists a sequence of independent random variables {X. w(z.} such that for suitably chosen constants B. the ratio 1~~ p3/a4 . the quantity [1. there exists a positive number i.~.z)+$(~“~z). Generalizing stable distributions.(W) denotes the tth coordinateofru. The parameter u is called the exponent of the quasistable distribution. Also in this case. >O. and (3) tj(jLz) = q(z) for every z.) (0 < p < 1) is called the quanfile of order p of F.. IflI < 1. yZ expresses the degree of deviation from the normal distribution. every moment of odd order about the mean (if it exists) is equal to zero. In any ldimensional symmetric distribution.F(m + x) = F(m-xx)..z). there correspond a positive number b and a real number i such that we have the relation F((x--I. where X. We have +(z) = .O<x<2. c > 0.. but quasistable distributions with a = 1 are not. .X. putting q(z)= expll/(z). In the l-dimensional case.a)). has the distribution 0.&) > 0 such that cp(iz) = cp(1. Gnedenko and A. and characteristic functions cp and $. Let F be the distribution function of a l-dimensional distribution @. where m is a real number.(w)6 E. -oo<c..z)= tan(lrr/2) (ctf l).The parameter x is called the exponent (or index) of the stable distribution. and Q(z.<t.. @ is said to be quasistable if to every b. > 0. we can define quasistable distributions. l<k<n. (For the stable distribution with exponent l/2 .z)=+(~. We associate with RT the o-algebra dT generated by the cylinder sets.z)cp(iL. z)= Let R = RT.P(Ix. (2) G(x) = F(ix) for every x. and n = 1.I>~)~o as n+ a for every c > 0.

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

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

The following types of stochastic processes have been investigated extensively: tadditive processes. P(E.. or with probability 1)..E. (n = 1. N) is independent if every subsequence satisfies P(EilflEi2fl. P(0) = 0. p ( UE” ” =CW. occur for all n after some number n. ) are mutually independent or not. Frequently. . which correspond to functions in analysis. respectively.2.} (A E A) of events is independent. . . Z” P(E. In the same way as functions are often defined by differential equations. A family {b. applications of part (ii) are greatly hampered by the requirement of independence.). tmartingales. we say that the events E. (n = 1. N) of events.S. . ) is a sequence of pairwise exclusive events and E is the sum event of E. the correspondence of each event E with the set E of all sample points w satisfying E). tMarkov processes and tMarkov chains. the sets lim sup. . . and tstatistical mechanics. P) is called a probability space. The most important concept in today’s probability theory is that of tstochastic processes. The superior limit event (inferior limit event) is the set of all w for which infinitely many events among E.2. we call P(E) or Pr(s) the probability that the event E occurs. The triple (a. which is concerned with the evaluation of P(lim sup” E. The complementary event (resp. and (ii) if the events E.2. the event E is said to occur almost certainly (almost surely (abbrev. we have (i) whether the events E.). 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. tBrownian motion and tbranching processes are important special stochastic processes. ..} (1 E A) of o-subalgebras of events is said to be independent if for every choice of E. If E is an event and E is the b-measurable set corresponding to E. Given a finite sequence {E.} (n = 1. ) of pairwise disjoint sets in 8. the sum event (resp.(n. Moreover. impossible Given an infinite family {E.)=fiP(E. and P(lim sup. it has made tremendous progress. a b-measurable set itself is frequently called an event. reads as follows: Given a sequence {E. .E.} (LoA) of events. 8.} (n = 1. ) of events. . or sample space (resp.2. we say that the events E.} (n = 1.). each element w of a) is called the basic space.). tstochastic filtering. .e.. The concept of independence of events can be generalized to a family {Bk} (noA) of o-subalgebras of b as follows. E.s. a number of sufficient conditions for depen- .flEi..2. The Borel-Cantelli lemma. there are stochastic processes which can be defined by tstochastic differential equations.IfEflF=@. space of elementary events. (Ieh) are mutually independent or that the family {E. .} (I E A) is independent if every finite subfamily is independent. in particular. sure event) is the complementary set EC (empty set 0. . 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. .thenwesaythatE and F are mutually exclusive or that they are exclusive events. we have 0~ P(E) < 1 for any event E. and lim inf. By the definition of P.2. are called the superior limit event and inferior limit event. tstationary processes.) N) are mutually independent or that the sequence {E. Z. E 23. . the probability of the event E. j=l B. (P2) for every sequence {E.} (n = 1.. .. we have P(E)= f P(E. For a sequence {E. . the family {E. 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. occur. We say that a condition E(W) involving a generic sample point w is an event.) = cc imples that P(lim sup” E.). Therefore P(liminf.e. We assume that an event is always a measurable one. depends on w in general. ) of events. . ~ event.. . intersection or product event) of E.2. whole space Q). . (all events except finitely many E.} (AE A).. is the set U..1267 342 B Probability Theory solid foundations by Kolmogorov. (n = 1.} (n = 1. II=1 This property is called the additivity of probability. For a finite or infinite family {E. for almost all w. sample point or elementary event). 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. a. and tGaussian processes. and P(n) = 1. Probability Spaces Let fi be an tabstract space and 23 be a toalgebra of subsets of R. where n.) is the probability that the events E. . if {E. E. .) is the probability that in& nitely many events among E.E..) = 1. i. If P(E) = 1.. E.2.} (n = 1.) = 0. ) are mutually independent.) occur. .) < co implies that P(lim sup” E. 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. n > (P3) P(a) = 1.2. ..

The independence of X and Y implies that p(X. The Chung-Erdik theorem [9] is quite useful in this connection. .] denotes the a-subalgebras of 23 generated by the sets {w I X.N)}) holds for every choice of 1 -dimensional +Borel sets A.2. E A) of r. the set {w 1X(w) < u} is in 23)..-dimensional distribution Qn of X. the independence of the family {X. in particular. Each element of this class is said to be measurable with respect to the family {X. I i E A]. When X and Y have finite variances.) is the joint random variable of X. E(X Y) := E(X)E( Y). Then each random variable X induces a probability measure @ on (R.}. V(X + Y) = V(X) + V(Y) for mutually independent random variables X and Y.23.} (n E A) of random variables. an arbitrary l-dimensional Bore1 set. Given a family {X. are random variables..-) dimensional Bore1 sets A.-) dimensional random variables.E( Y))) is called the covariance of X and Y. X.) induces its n-dimensional probability distribution (or simply n-dimensional distribution) and its n-dimensional distribution function F(u.(~=~. If X. . . N) (X.2. and that. N) and that the (I-dimensional) distribution @ of X is the joint distribution (or simultaneous distribution) of X. It also follows from the definition that -1 :C p(X. P) be a probability space. Since a random variable X is a 2% measurable function. Random Variables Let (0..dimensional random variables (n = 1. If X is integrable relative to P. N). e)-valued random variable if it is measurable.} (1.. the mapping X =(X1. expectation. with A. Y) d 1 in all cases. if b [X.E(X))’ is integrable. 23) into another imeasurable space (S..(o) :Gun}).2.)-dimensional random variable X=(X.. .. the value E((XE(X))( Y. N) is independent. the smallest o-algebra with respect to which every X. X. we say that the random variables X.2. V(X) = E((X . F(u) = 0.2. The variance is important because of the well-known Chebyshev inequal- .) (n = 1. More generally. but the converse is false in general.2.e. E(X. N).(k. if the relation P({~IX. . . is a monotone nondecreasing and right continuous function such that lim.(w)~a. for every set A of e. On the other hand.. we say that the random variables are mutually independent or that the family is independent if every finite subfamily is independent. denoted also by M(X) or m. Given a finite sequence {Xn} (n = 1. . A). u. The latter definition of independence of random variables is compatible with the previous definition of independence of c-subalgebras of d. Y)=O.} (1eA) and is denoted by 23 [X.(k.-. usually denoted by E(X). the set {w I X((U)E A} belongs to d. (n= 1.} (1 E A)) is defined similarly.X over A is denoted by E(X. If X and Y are two random variables for which E((X E(X))(Y-E(Y))) exists.(WIC A.2.) for l-dimensional Bore1 sets in equation (1). . Given an infinite family {X. dent events to have the same conclusion as (ii) have been discovered... X. X.)=P({wIX.E(X))Z) is called the variance of X.. . ..andom variables. . (A.E A) in the latter sense is equivalent to the independence of the family {b [X.E(Y))z)}“2. V(aX + b) = u2 V(X) for any real numbers u. for every real number a.. then the independence of the family {X. (n = 1.~. “‘> N) ({X.. we can speak of the tintegral of X relative to the measure P on 8. is called the marginal distribution of the /-dimensional distribution @.2. i. X. we say that the I( = Cr=. Let ‘23’ be the o-algebra of all +Borel subsets of the real line R... Similarly. . that is. is measurable is called the (ralgebra generated by {X.{ (3. (n = 1. A random variable is a real-valued function X defined on R that is %-measurable (i. . 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--. . .. . (n = 1.. or expected value of X..} (3.2.} (n = 1.} (A E A) of random variables. k. N). B1) such that The measure @ is called the (l-dimensional) probability distribution of the random variable X or simply the distribution of X.. If the X. F(u) = 1.) from s1 into R” is said to be an n-dimensional (or R”-valued) random variable..e. Q). the k. b. (n = 1. it is enough to take k.(~)EA.] } (1 E A) in the previous sense.. (n= 1. UER. . The standard deviation of X is the nonnegative square root a(X) of the variance. C. . .. X. The function F is called the cumulative distribution function (or simply the distribution function) of the random variable X. N) are k. . N) are mutually inde- It follows that E(aX + b Y) = uE(X) + bE( Y).342 C Probability 1268 Theory pendent or that the sequence {X.. an n-dimensional random variable X =(X1 . N) of random variables. If (X . . X. is called the mean. a mapping X from (Q. .2. lim. . the integral of . The point function F defined by W=P({~IX(4d~})..A)) are k. C5)is called an (S. denoted by a’(X). X.. If the X.

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

P. Springer. de Moivre. Les probabilitks dCnombrables et leur application arithmetiques. J.342 F Probability 1270 Theory [6] and the Hewitt-Savage zero-one law.. [Z] J. 1950. . 1931. The doctrine of chances. 1933. Doob. G. ErdGs. and assume that one of them must occur. 1713. 2 (1923). Kolmogorov’s zero-one law can be extended as follows: Let g. Sot. Circ. the event that xi+ X. 72 (1952). This is called Bayes’s formula. Amer. [6] A. English translation.)P(E~E. the determination of the values of a priori probabilities is sometimes difficult. Differential space. When X is a random variable subject to the distribution with continuous probability density f(x). Vorlesungen au.) = l/n in practical applications.. In practical applications E. E.. Macmillan. Kolmogorov’s zero-one law: Every tail event concerning a sequence of independent random variables has probability 0 or 1. ) be an event concerning a sequence of random variables {X”}. Laplace. L. called the tail oalgebra of { &}. usually represent n unknown hypotheses. P(A) = 0 or 1 for every A E 2. Rend. [ 1 l] J.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.} if for every n. although this has caused a great deal of criticism. If E is another random event. Mat.e. [7] E. E. Franz Deuticke... F. ThCorie analytique des probabilitts. 247-271 [S] N. is the o-algebra generated by X.. be a. A history of the mathematical theory of probability from the time of Pascal to that of Laplace. Let a = cc(X. Zero-One Laws In probability theory there are many theorems claiming that an event with certain properties has probability 0 or 1. 27 (1909). For example. a is called a symmetric event concerning {X”} if occurrence or nonoccurrence of c( is invariant under every finite permutation of X. Todhunter. 1949). . Erg. Then the oalgebra 2 = nk Un. Wiener. . Stochastic processes. [4] P. von Mises. This version is called a regular conditional probability of YE E under 3 or the conditional probability distribution of Y under 5. P( YE E 1X). Wahrscheinlichkeitsrechnung und ihre Anwendung in der Statistik und theoretischen Physik. n= 1. and we often set P(E. 1865 (Chelsea. =0} is {Lx”+l. and P(E I X =x0) is the conditional probability of E relative to X.(E 13).2. 1812. . Kolmogorov (Kolmogoroff ).186. Math. Suppose that the probabilities on the right-hand side of the formula are given. [S] R. occurrence or nonoccurrence of CI depends only on }.k s. is trivial.. 1718. . . Grundbefriffe der Wahrscheinlichkeitsrechnung.. . we have P(E. However.) (P(Ei I E)) is called the a priori (a posteriori) probability. .(E 1X=x) are interpreted similarly. this is written as P. 179. [9] K. fW W IX =x01 . S... I El = P(EJP(E I EJ P(E. Palermo. Gauthier-Villars.. Kolmogorov’s zero-one law is a special case where ‘&. L&y. Bernoulli. bility measure in EEL’ for every weR and that such a version is unique almost surely.)+. 1953. 1937. 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. . L. Math. we mention two famous examples. Wiley. . Hewitt-Savage zero-one law: Every symmetric event concerning a sequence of independent and identically distributed random variables has probability 0 or 1.+P(E. X.. {lim. X.. i. P( YE E 1X=x). Chelsea. . Paris. . On the application of the Borel-Cantelli lemma. For example. Trans.. X. .s dem Gebiete der angewandten Mathematik I. E.)P(E~E. Borel. CI is called a tail event concerning {X. Such theorems are called zero-one laws. sequence of independent o-subalgebras of 23. [lo] P.. Phys.)’ where P(E. Bayes’s Formula Let E. > 0 for infinitely many n’s is a symmetric event. Kolmogorov’s zero-one law .. 131-174. ThCorie de I’addition des variables alkatoires. This is why P(E. 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. and P. for every n. Foundations of the theory of probability. [3] A..(E) X). Math.. a tail event. Here. Ars conjectandi. Bayes’s formula is extended to the following form: References [l] I.) 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. be pairwise exclusive events. Chung and P.

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

Projective Coordinates in P”.. ea’sh pi is called a vertex. q3. then the pencil is determined uniquely by these two. when n = 2.)n(q. (pl U p2) n (ql U q2) are collinear.3). The set C. and the Pzm2 common to them is called the center of C. If the three lines pi (i = 1. dent.3) are concurrent. then its dual proposition is also true (duality principle). 4 92% 9. 3 Fig.343 c Projective 1272 Geometry have no points in common with ttem and project each point of P. is called the dual space of P”. Suppose that we are given spaces P. by regarding the hyperplanes of P” as points of Pi. and then cutting it by P2. and P’ and .U~. Now for arbitrary subspaces P. and we call P. i d j < 4) a complete quadrangle p. then r + s = t + u. In P”. In such cases it is impossible to introduce coordinates. we say that C and L” are in perspective and denote the relation by C X c’. The converse is also true.2. This operation is called projection of Z from PO onto P2. each of which is independent and fies pi # q. This theorem holds for n > 3 generally. Suppose that in a proposition or a figure in P”. When n = 2 and 3. (i = 1. Prier+‘.+P. if a proposition is true.~. . 3). so we assume Desargues’s theorem for n=2. that points satisU qi Here we introduce projective coordinates Consider Desargues’s theorem: Suppose p. 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’. p3 and ql. However. then we call it a projective mapping. The one-to-one correspondence PC-P.~C’. (1 < i < I) such that C. (0 < r < n). gz4. we interchange P’ and Pnmrml (O<r < n) and also interchange contains and is contained (and related terms).Uq. 9. . p2. gi3. Pn--’ that contain the same Ponrml is called the star with center P.). 1 . 2). The proposition or the figure thus obtained is said to be dual to the original one. In projective geometry.(~.c’ (Fig.2. Each pencil of hyperplanes of P”. or more generally. too. respectively. If a one-to-one correspondence P.Uq. 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.p2p3p4. the center of projection (Fig.. By projecting Z from P. If for two fundamental figures Z and c’ there exist a finite number of fundamental figures F..)n(q. it is called a pencil of lines and a pencil of planes. thus obtained is called a perspective mapping. we call these non-Desarguesian geometries. and each gij is called a side. the set Zr of all Prier.mrml. we take Ponrm’ that C. F.~F. g.U~.Zr are dual to each other. The projective space PG obtained by the principle of duality.J is represented as the composite of a finite number of perspective mappings. we call P? P4 P3 AIYi!kL 91 q1 Fig. and Pz. each pencil of hyperplanes of a subspace of an arbitrary dimension in P”. If a pencil of hyperplanes contains two distinct hyperplanes of P”.). we call the figure that consists of these four points and the six straight lines gij = p. P. These mappings are extended to those of fundamental figures. P. This is assured because propositions dual to axioms (I))(IV) hold.. If six points qi (I d i < 6) on a line I are points of intersection of six sides gi2. In this case.. and a fundamental figure Z in the space P. If P’U P” = P’ and P’ f’ P‘= P”. there exist projective geometries for which it does not hold. 2 points(~.. When four points pi (1 < i < 4) in P” lie on the same plane and in general position.thenwesaythatCandZ’are projectively related to each other and denote this by C.4r g34. we can construct a fundamental figure c’ on P2. the operation of constructing P’U P” from P’ and P” is called projecting P” from P’. is called a linear fundamental figure of P” or simply a fundamental figure. q. U pj (1 :. are two sets of in P”. g23 of a complete quadrangle with I. We call the latter the dimension theorem (or intersection theorem) of projective geometry. of all hyperplanes that contain a Porn2 in P” is called a pencil of hyperplanes. onto P$ from P$-rm’. then the three Fig. Under the assumption that P’ and P” do not have points in common..

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

).. These three propositions are mutually equivalent. pm. we denote it by O(P”) and call it the group of projective transformations. respectively. x!) (i = 1. 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 we are given two projective spaces P” and P” that are subspaces of a space PN (n < N). q.. we denote it by K(P”).Uq.P*~PIl> IP. P31 =CP3.)n(p. it naturally induces a mapping Pl+P”. respectively. and I. then collineations are always projective transformations.. Next. pJ on a line in a space. If 7 o 7 is an identity. which we also denote by r.3) and qi (i= 1. Then.3) that lie on I. Suppose that the coeffkient field is commutative. Now. If the coeffkient field is the real number field. the supporting point. pz.)n(. it is necessary and suffkient that all automorphisms of the coeffkient field be inner automorphisms.p2. A projective collineation is also called a projective transformation.< r < n .p3. if we denote the group of tautomorphisms of K by ‘%(K). we can determine homogeneous coordinates in an arbitrary subspace of P” by a frame on it. when a collineation in the wider sense cp: Pn-P is a projective mapping.7~Uq~)are collinear. ql) on an arbitrary line onto K can be uniquely determined so that 0-l o 8.3.P31=:> A cPI>P3. The totality of projective transformations of P” constitutes a tnormal subgroup of a(P”). Suppose that the coefficient field is a commutative field whose characteristic is not 2. then the isomorphism 6 of the Staudt algebra K(q. then the three points (p2 U q3) fl(p3 J q2).P4. if P” = P”. l.2.4). (When Desargues’s theorem holds.+l(s). p3 that are collinear. we call cp a collineation.Uq. we call it a projective collineation in the wider sense. then we have the order of the i = CP2rPli P4. can be expressed as cpp’3 CpP”. . this is not necessarily true.J. If we denote the inhomogeneous coordinates of pi with respect to a general frame by (~0. and I. If we denote a correlation by zO. (2) Given frames 5 and 5’ of P”.)-+K for the Staudt algebra K(p.. and the unit point. we consider the following three propositions: (1) The coefficient field of P” is commutative. where p. (3) Given two distinct lines I. we consider a frame such that pl. If P” = P.p. In classical geometry.P1rP*l =cP4>P3. Suppose that cp: P”+ P” is a collineation in the wider sense and 0 . D. we call 7 an involutive correlation. is a projective mapping. contained in a plane in P” and two sets of three distinct points pi (i= 1.. Then zor is a collineation. then ~(P”)/B(P”)z’%(K)/~(K).2.P4. Hence in order for all collineations to be projective transformations. Specifically. We call proposition (2) the fundamental theorem of projective geometry and proposition (3) the theorem of Pappus. then 3. if we assign an isomorphism Q. and p3 are.1. we call q a correlation. cp(pi) (i = 1. only these cases were studied.Pz. p2. The totality of collineations of P” constitutes a ttransformation group and is called the group of collineations of P”. we call the projective space a real (complex) projective space. If z is a correlation. then (~~Uq.2.:K(p. p2. It is isomorphic to the group of tinner automorphisms J(K) of the is not coefficient field K of P”. The totality of projective transformations that leave invariant a frame 5 of P” constitutes a subgroup B)l. 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”. If the coefficient field is the real (complex) number field. 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. For four collinear points pi (1 < i < 4) in P”. there exists a unique projective transformation Moreover.) In this case. For the complex number field.343 D Projective 1274 Geometry Transformations sending 5 onto 5’.P*>P41=l-A .he origin. Utilizing such isomorphisms.p4].p. however. The former is obtained as a composite of a projective transformation and an automorphism of the coeffkient field.3) are also collinear and vice versa.and(~.. any two projective spaces of the same dimension can be identified with subspaces of a projective space of higher dimension. if we interchange four points. any other correlation is obtained as a composite of z0 and a collineation.2. and if P’z P” in P”. A collineation necessarily a projective transformation. p3 are distinct and p4 # pl..

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

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

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

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

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

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

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

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

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

An interesting application of the global theory of Fourier integral operators appeared in J. Y. T].1~k. we have WF(A.For a small 0 < Tg To the solution S(t.. x. as a family of canonical transformations with a parameter tc [0.S(t.) such that the solution u(t) is found in the form u(t) = E. 5) of Egorov’s theorem satisfies the relation 9(x. 4) of the eikonal equation a. Assume that m = 1. e.. E Sp. E = V. Fujiwara [ 181. 5) .u(x)=(27c-“‘2 s R” exp(iS(x.g. T]. also.. W> Y.n algebra pseudo-differential operators.nere exists an amplitude function e(t. x R. [loI> 1. D. Some suflicient conditions for boundedness have been obtained by Eskin (Math.S+p(t. On the other hand.. UEY. 3 (1976)) used a type of Fourier integral operator in deriving the energy estimates and constructing the fundamental solutions for strict hyperbolic operators.~O)#O at (x0.V. + p(t. Then t.. y. J. 5). Then by (7) the Fourier integral operator A = A. of . the Fourier integral operator has come to be recognized as a powerful tool in the theory of linear partial differential operators. which was originally published in 1965. and that p1 (x. and let a(x. such that p(x. y) = S(x. The works of Egorov. I.. (9) Next consider a hyperbolic operator L = 0. [12. 11 (1970)) applied his theorem and the corollary stated above to the study of hypoellipticity and local solvability for pseudodifferential operators of principal type. ‘Then (X(t. let (x. p = 1. d.(xo. 0.<)~S~. Math. which correspond to Fourier integral operators in the theory of distributions. Chazarain (Inoentiones Math. then 9(x. T] with the initial condition SI. Let us consider a phase function of the form cp(x. 5) be an amplitude (8) Let T be the canonical transformation with the tgenerating function S(x.). S:. 7’1.(t)u. 1 References [l] J.p(t.V. J. . Nirenberg and Treves [ 161 obtained decisive results concerning local solvability for linear partial differential operators of principal type. is defined by A. 5 exists in g” ([0.u)= {(x> 5)= T(y. A. and HGrmander motivated the theory of hyperfunctions developed by M. q). T].(R”) and drp.P(~. 5 = V.) with some To > 0. Using Egorov’s theorem and the same corollary. 4. z(t. x.5’). <)-p(x(x. d</dt = . holds.. 5) = (XQ. K. 1 (1976)).325 Partial Differential Equations of Hyperbolic Type L.x.13. 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).~)I(y. USSR-Sb. 3 (1967)).%. Then one can find a Fourier integral operator A such that the function 9(x. X.. Pseudodifferential 1284 Operators function independent of y of class S. E(t. 0. H6rmander (Acta Math.S)=O on [O. 269-305. q).to)=O. = x.2p(R”). 5) with (x. A calculus of Fourier integral operators in R” was given in Kumano-go [4]. 5)M:c W(5M. Pure Appl.. G. Egorov (Math... q). y. Nirenberg and Treves. By virtue of this research. Hiirmander [9]. and others.. where p.rl)~‘~F(u)}. i. 121 (1968)) introduced the term “Fourier integral operators. 24 (1974)). q)) can be solved by means of the relations y = V.S(t. Sato and gave rise to the concept of tquantized contact transformations. Then for the Fourier integral operator A. To]. = (Y. Comm. J. Hiirmander and J. X. a)) be the bicharacteristic strip defined by +Hamilton’s canonical equation dxJdt = V. 5) is a real-valued ?-function. Sl.. q). 5)) belongs to the class S.” and applied these operators to the derivation of highly accurate asymptotic formulas for spectral functions of elliptic operators. Asada and Fujiwara (Japan. <)E&?’ ([0.5. 5) . Partial DiJff: Eq.. SF. q).. these results were completed by Beals and Fefferman [ 171. The propagation of wave front sets by means of a Fourier integral opera&r is described as follows. ultEO = uo.S(x. 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 (-. Nirenberg.e.y ‘5 in R: x R. 4 (1978)). 5. x.(x”. USSR-. T is defined by y = V$(x.. Thus by means of (9) we have WF(u(O)c {(x. x. II)..14]).345 Ref. The theory of Fourier integral operators has its origin in the asymptotic representation of solutions of the wave equation (. 5)~ B”( [O.) for a real-valued symbol p(t.=. 5) =(X(t. y. Consider the Cauchy problem Lu = 0 on [0. 325 Partial Differential Equations of Hyperbolic Type M). x.. Kumano-go (Comm.15] constructed a general global theory of Fourier integral operators making use of Maslov’s theory [14]. Kohn and L.l.. 18 (1965). x. <). o(R”).. The boundedness of Fourier integral operators in the spaces L*(R”) (or the space H”) has been studied in several cases. homogeneous of degree 1 in r for /<I> 1. y. Math. USSRSb. Duistermaat [9. Eskin (Math. 5) -pl(x.

(2) The Bradley-Terry Model. 128 (1972). and . Pseudo-differential operators and hypoelliptic equations. social attitudes. Math. [ 1 l] Yu. [ 151 J. Math. (Original in Japanese. and mental abilities. B.p as correlation coefficient of Xi and Xj. 69 (1972).. In the following sections we describe the basic statistical procedures used in sensory testing. Oj) is pij. Hyperfunctions and pseudo-differential equations.17) Psychometrics A. 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. Ann. Pure Math. Symp.). P. [ 121 P.=. 25 (1969). [lo] J.. Suppose that the probability that Oi is preferred to Oj for a pair (Oi. J. Proc. Math. Proc. [ 131 D. and Oi is preferred when Xi > Xi. Acta Math. Furthermore. and the examining circumstances must be controlled. P. Comm. The panel of judges must be composed appropriately. Fourier integral operators I. Nauk.. Kumano-go. Let Q(x) be the standardized normal distribution function and pij = @(pi pj). On local solvability of linear partial differential equations. Hormander.. the number who prefer Oi is nij. Proc.=p.=p. Math. J. Acta Math. Duistermaat. [9] L. Fourier integral operators II. Pure Appl. Oj to the senses are random variables Xi. the method of comparing them two at a time in every possible way is called paired comparison. The following are typical mathematical models of this method. O. 10 (1967). we can obtain the estimates pi. [ 171 R. D. 11851187. J. Japan Acad. 235236. 183-269. [14] V. General Remarks Psychometrics is a collection of methods for drawing statistical conclusions from various psychological phenomena which are expressed numerically or quantitatively. Calderon and R. MIT Press. Nirenberg and F. 83 (1966). Asymptotic solutions of oscillatory initial value problems. V. Komatsu (ed. Paired Comparison When there are t objects (treatments or stimuli in some cases) O. Lecture notes. Fourier integral operators. Hiirmander. O. Duistermaat and L.nij. Beals and C.. 96-99. Pseudo-differential operators. Various methods of psychological measurements are applied. Pure Appl. 1973.. 359-443. Amer.. 32 (1979).. Pseudo-differential operators. II. Exact and asymptotic solutions of the Cauchy problem.. On local solvability of linear partial differential equations I. Gauthier-Villars. Xi. I-38. jI6] L.. Pure Appl. Hormander. 473-508. D. Duke Math. [ 181 D. The experimental method in the Bradley-Terry model . [8] L. It consists chiefly of statistical methods to deal with psychological measurements and of theories dealing with mathematical models concerning learning processes. In this comparison it is assumed that the strengths of the stimuli Oi. 13 (1960).1285 346 C Psychometrics [2] L. [3] L. Comm. Of the 12judges who compare this pair. Theorie des perturbations et mtthodes asymptotiques. . Maslov.. 1974. Sot. (2) 97 (1973).fij) and p. 129-209. Comm. Lax and L. Fefferman. 19 (1966). we can test the hypothesis thatp.. Pseudo-differential operators and non-elliptic boundary problems. [7] P. . Lecture notes in math. Nirenberg. 51(1975). 1973. Springer. 473-492. . Fujiwara. (1) Thurstone-Mosteller Model. Pure Appl. Sci. US. Math. Comm. A class of bounded pseudo-differential operators. On stability for difference schemes. 18 (1965) 501-517.459-510. Using p. 138183. Math.. 482498. 23 (1970). Uspekhi Mat. [4] H. C. Pure Appl. Nat.) [S] A. 627-646.. Hormander. On the boundedness of integral transformations with highly oscillatory kernels. 1972.. 287. and the number who prefer Oj is nji = n ./n as estimates of the true pij. Egorov. Sensory Tests A measurement wherein human senses are taken as the gauge is called a sensory test.. Courant Institute. Ann. 79-183. Vaillancourt.. 1981. Acad. Ludwig. 24 (1957). Using pz = @(pi . 128 (1971). [19] H. [6] L. Lax.. 346 (XVlll. The Weyl calculus of pseudo-differential operators. Comm. a sharp form of Girding’s inequality. Treves. On canonical transformations of pseudo-differential operators (in Russian).= n. . Math. Hormander. Hormander. Math. There is no loss of generality in assuming that 2a2( 1 -p) = I and & pi = 0.

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

and the tmaximum likelihood based on the normal model by Lawley and Maxwell. . factor loadings.. . .. . . 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. Example (2). called d-trial path dependent. ) with respective probabilities P. we set Y.2.i.)=~.(Pn) =I&..j.1287 346 G Psychometrics f(P. among which those often used are the MINRES by Harman (1967). if A. In nonlinear models the recursive formula cannot be written as a linear function of P. 6$. and A... and they are frequently represented by recursive formulas for response probabilities.. A are the matrices of factor scores. . k). N. Then the behavior itself is modified by such a sequence of trials... and that an event Ei occurs on the nth trial with Pr(& = Ei) = xi (i = 1. on the nth trial.1) trials (& Ni = n . . PI). . As a result of the nth trial. &. . If the response probability is path independent. the manner of their correspondences depends on each trial..) = aiPn + (1 .Nf(Pl) can be obtained. we set Y.. (3) Estimation of factor scores fi. In a linear model. or A. . . . Lute’s P-model [9]. In this model. . the recursive formula is written as a linear function of P”.. it is called path independent. n). = 1.2. . Other linear models have been proposed in which the response probability is either quasi-independent of path [8] or path dependent [lo]. the explicit formula P.). The response Then it follows that and where hi” is called the communality of thejth variable zj and uf is called the specificity. Furthermore. If both event commutativity and path independence of response probability are satisfied. 7ti = 1).. responses. . (2) Determination of factor loadings.. . . fk(Plb where f.Y..). c?“= I$.i. . Hence. Assume that two mutually exclusive response alternatives A. We can consider the stimulus as a set composed of m elements. . If the formula can be written as f=f(P. (2) Linear Models. .. then the response probability is Gl. the response probability is said to be quasiindependent of path [ 111.. xi=./m. Let the response strengths of A. By making f explicit with respect to n. . The Bush-Mosteller model assumes the response probability to be path independent.(P.).. where F. Example (1).. . on the nth trial be u. otherwise. (3) Nonlinear Models. When the recursive = formula can be expressed as f=f(P. G?~ E. we obtain the recursive formula Pn+l = P. = F(N. . =Ek)=Jj.1). among which X. then we have P. Learning models refer to such processes of behavior modification...) (v = i. &I ... Estes’s stimulus-sampling model [7]. 8. &‘” = Ei. reinforcements. .. . t. & &. and R is the correlation matrix of the Z’S G.) that influence the ensuing behavior. the varimax by Kaiser (1958). . elements correspond to A. elements correspond to A. P.. .XA( 1 . 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. . Then the recursive formula for P. j. gl). Suppose that on the nth trial s ( <m) elements are sampled. The recursive formula is expressed as fi(P.cc&.XA( 1 . In the special case d = 0. For simplicity we write f(P. CF. then &. Example (3).. + {Xx Y.. . . If any two events are commutative. occur on the nth trial (n = 1. Assume that a sequence of trials is done in order to study some given behavior and that on each trial particular events occur (stimuli..2.) elements correspond to A. the condition of event commutativity is satisfied..Y. N.k= 1. and A. test scores. ... Learning Theory (1) General Description. In the recursive formula J two events Ei and Ej (i #j) are said to be commutative if~. . Under the condition of event commutativity. YJ. ... is reinforced.X. the explicit formula can be written P.. .). is of the form Pn+l =f(Pn.. Bush-Mosteller model [6]. and 1 -P.(R) = .f.j. which is the estimation of the aji: A number of methods have been proposed. = 0.. Y. where Ni is the frequency of occurrence of Ei in the first (n ... assume that J n+l = Jn +X. If J.-~. respectively. and assume that P. the response probability is path independent and 8” = (X.. CI?“. =fylfp . and the remaining XL ( = s . each of which corresponds to either response A. t).&“-. the response probability of A. . is expressed as un/(un + ub). respectively (both positive). ..)}/m. etc. . It follows easily from (1) that any orthogonal transformation of the scores does not affect the model.. = J.. Thurstone proposed p=ZR-‘A and Harman F=ZA(A’A)-‘.k(Pn) (ij.-~ = Ej. Z... and oh.: Usually factor scores are estimated after factor loadings have been determined. 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. we write P”=F(n..(P..

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

.

In particular. where (i) x:(n)= (. 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. That is. +iincasem=-1. p’ are prime ideals of o. m’ E 3 (mod 4). (iii) x(p) = 0 if (p) = c ’ in O.e. If m = 5 (mod 8) then (2) = p and N(p) =4. and x is the Kronecker symbol. More precisely. The fl.m)=l.4.et k be a real quadratic field.7.19. 163. *E. in case nr = .1 if (p) = p in o. then x(n)= xz(n)(n/lm’j) for d=2em’. 13 (1966). then x(n) =C#. The Kronecker Symbol Quadratic 347 (V. where d is the discriminant of k.14 Algebraic Number Fields).8. (iii) Let 2/ld.3 (mod 4). The Kronecker symbol for k has the following four properties: (1) x(n) = 0 ii‘ (n. In 1934. The mapping (r: a+x’ is an tautomorphism of the field k.l)(nz-‘)/B for e=3. Heilbronn (1934). =( 1 + fl)/2) field. Any quadratic field k is obtained from Q by adjoining a square root of a square-free integer (i. y) of +Pell’s equation x2 dy’ = + 4 by using continued fractions (. Then p is decomposed in o in the form (p) = p2. B. Linfoot proved that there can be at most one more such d.[ 11). + o~(w. Ed is the positive fundamental unit (> I) of k. There exists a unit E” that is the smallest one among the units (> 1) of k. +Gauss that h(d)-+ cxzas Id I+ co.lf(m/p)=l. k is called a real (imaginary or complex) quadratic field (. for m<O. Siegel (Acta Arith. any talgebraic integer x of k has the unique expression x = u + bw with a. and finally A. d) # 1. bE Q) of k over Q is given by a’ = n -b&. Baker and H.83 Continued Fractions. 59 Class Field Theory. an integer #O. m’s 1 (mod4).1. L. ’ is a ifundamental unit of k. Let (1 + fi)/2 (O= t fi for mz 1 (mod4).3 (mod4). then x(n) = (n/jml). Denote by h(d) the class number of the imaginary quadratic field with discriminant d. We define the symbol x for k as follows: (i) x(p)= 1 if (p)=pp’ (pfp’) in 0. If m = 1 (mod 8). that is. Here p. (ii) x(p)= . for e=3. If m is positive (negative). we define x(l)= 1.d)=1. (Property (4) shows that a quadratic field provides a class field. That is. Michigan Math .m'c3 Then (1. G. and (m/p) is the +Legendre symbol. If (m/p) = -. +Dirichlet ods as follows: hlog+. Any unit E of k can be uniquely expressed in the form F:= + a$ (no Z). w is the number of roots of unity in k. w) is a +minimal basis of k. (4) x(n)= 1 if and only if n= N(a) (modd) for some integral ideal a of k such that (a.then(p)= pp’ in o (p # p’) and N(p) = N(p’) = p. (ii) x:(n) = (. and if m=2.3. IJnits E. where d is the discriminant of k. C.I)(“-‘)‘~ for e = 2. 1 with no square factor) m:k=Q(fi). The fundamental unit ~~=(~+y&)/2 (> 1) can be calculated by finding a minimal positive integral solution (x. C. L. and x(n)=*1 if(n. and (iv) x(n) = ni x(pi)‘i for n = ni p. Prime Ideals The decomposition of a prime number p in the tprincipal order D of k is given as follows: (i) Let p 1d. L. Sta.12) Quadratic Fields A. then (p)=p in o and N(p)=p’.i 3>0. Muthematica.(2)X(m)=X(n)ifm=n (modd). + uo. If (n. b E Z.11. F.43. M. This conjecture was proved by H. the symbol x can also be defined using the +Jacobi symbol as follows: If m 3 1 (mod4). for a table of the fundamental unit of k for m < 100 . and let d be the discriminant of k. then (2) = pp’(p # p’) and N(p) = N(p’) = 2. The +class number P.rk. Ideal Classes Let k be an imaginary quadratic tunitsofkare +l. m E 1 (mod4). The conjugate element of an element a = a + b& (a.= -idi I 1 h=Gldir I I x(v)r h of k was calculated by (1840) by analytical meth- &)logsinF for m>O.347 A 1290 Fields D. The symbol x(n) for nE Z is called the Kronecker symbol for k.) (n*~lm+(n-1)/2 Any tcxtension field of the rational number field Q of degree 2 is called a quadratic field. d) # 1.(d)) = 1. If (n.3 (mod 4).. (ii) Letp#2and(p. Heilbronn and E.d)= 1.. 1 (1935)) proved h(d)=1 holdsfor(d(=3. and (iii) x:(4=(1) (mod 4). It was conjectured from the time of C. H. we define X(-n+=(sgnd)X(n). and _+ 1 in all other cases. For a negative integer -n. Stark independently proved that these nine numbers are the only ones for which h(d)= 1 (Baker.67. for m = 2. (3) x(mn)=X(m)X(n). N means the +norm. General Remarks Let k = Q(&).

. Comp. m’) = 1).5. . (d)) = 1. . the principal genus is 8 = {E. . and the class number formula was obtained by Dirichlet as a formula for binary quadratic forms. Let p1 . (i) H. = 1. Table 4). . the theory of genera for quadratic fields explained in Section F was first developed by Gauss in terms of binary quadratic forms.d=2’m ((2.52. 29 (1975)) that h(d) = 2 holds only for IdI = 15. . and they form an Abelian group of type (2. Let (pi) = pf (i = 1. B. Math. A’B. it follows that n = N(a) (mod d) for an integral ideal a of k is a necessary and sufficient condition for x1(n) ..2. the values ci = Xi(N(a)) (i = 1.) = 1. EJ to be a character system for some genus is that ~~=fland~~~~. The conductor f = n. . . . pt be the set of all prime numbers dividing d.) = .13. and 9 is a subgroup of the ideal class group (r of k. . Math.232. d) = 1. Each coset of G modulo H is called an ideal class in the narrow sense. A’}.403. A*. AB.14 Algebraic Number Fields G. A3. B. .(n)= 1 for neZ with (n. EJ the character system of this genus. the Kronecker symbol x is expressed by x(a)=g $ for(d)= 1. t). .. 14 (1967)). Hilbert [4.andfz=2’for2jd. . it is necessary and sufficient that n = N(a) (modd) for an integer tl of k (where N(a) = a~‘). .. For example.24. Dedekind [2] (. and each genus is uniquely determined by its character system. .. 148.. In order that an ideal class C belong to the principal genus. . we identify xi with XT in the definition of the Kronecker symbol. Each ideal class C such that C” = C is called an ambig class of k.. = p forpId. CL= {E. . for m < 0.59 Class Field Theory).348 Quadratic Forms M).sdis the fundamental unit of k = a(. 21 developed the arithmetic of quadratic fields systematically by introducing the Hilbert norm-residue symbol (. .. ..2).. Hence there are 2’-’ genera of k. (This notion is a special case of the notion of ideal classes in the narrow sense of algebraic number fields. The tconductor of H is said to be the conductor of kJQ or simply the conductor of k..51. There are 2’-’ ambig classes. . it is not yet determined whether there exist infinitely many d with h(d) = 1 (. t) are uniquely determined. From this it follows that there are t . . h= Q(n) 8. That is.2). .20. = 1.1. . Since x1 xz . the p-conductor f.. it is necessary and sufficient that C = C: --d = C: hold for some ideal class C. When m > 0.I. By means of the tHilbert normresidue symbol. For example. . x.123. . for k = we have d= -2’. . A. We call (cl. for pi = 2. . Genera Let G be the group of all (tfractional) ideals of k. > 1). p.p#2. d) # 1. Ed)is uniquely determined for the id&al class C containing a and does not depend on the choice of a. . However. xt is equal to the Kronecker symbol. x. . The set fi of all ideal classes of k such that (El .91.1 tinvariants of the ideal class group 6 of k that are powers of 2.115.35. We call an ideal class in the narrow sense simply an ideal class. ch. N(E.(n)= 1 to hold for ncZ with (n. B2 = E. For n EZ with (n. . . Each ideal a of k with au= a is called an ambig ideal of k.“*. Put &i = xi(N(a)) (i = 1. E. N(E..2. . . =x.’ . the usual classification of ideals and classification of ideals in the narrow sense are identical. where a is an integral ideal with (a. d) = 1.. .5. . . t=3. and B”= B. of k = Q(Ji)isgivenbyf=dform>Oandf= dp. where . W. Baker and Stark proved independently (Ann. we define x.))/(log Jli. f. For each genus... pft(a) by some a~ Q and vi = 0. A3B}. R.Appendix B. G. .) Et)=(l. Each ambig class contains exactly two ambig ideals of the form p .187.(n) as follows: For pi # 2. The theory was then translated into the terms of ideal theory by J. . .(n) = (n/pi). A’B}. Also. A’. .[1.267.235. we define x1(4.d)= 1. each ideal class is divided into two ideal classes in the narrow sense. For real quadratic fields.) 1) is called the principal genus of k. F.88. 0 for (a.40. and 427./$ (E. and a/$ is an Abelian group of type (2. Each coset of (E modulo 43 is called a genus of k.6]). . Then each ambig ideal is uniquely expressed in the form a = p.) In the cases (i) m < 0 and (ii) m > 0.. and let H be the group of all tprincipal ideals (a) of k such that N(a) > 0.. 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].. 1. and the ambig classes are {E. Then (Ed.1291 347 H Quadratic Fields J. A necessary and sufficient condition for (cl. . where A4 = E. (2) 94 (1971). . In order that xl(n)= . t). Norm Residues A quadratic field k is the tclass field over Q for an ideal group H. .. A”= A3. Later the arithmetic of quadratic fields assumed the aspect of a simple example of class field theory (. we have 2-i (log(h(d)log&.

we have A(Q) . . for elements a and b in K* = K . 1 C. of America and Wiley. are uniquely determined by Q (Sylvester’s law of inertia). 1. Number theory. Hirzel. Generally. we call ( -l)n(n-‘)‘22”l Al. Vorlesungen iiber Zahlentheorie. General Remarks A quadratic form Q is a quadratic homogeneous polynomial with coefficients in a +field K.y)=‘xAy. 1894 (Chelsea. Hasse. Any quadratic form Q represents 0 by taking the zero matrix as P. .Q(x) . we call Q a real (complex) quadratic form. W. if a mapping Q: V+K satisfies these two conditions.Ax)=‘xAx.x? over K (ui # 0. Conversely. q) the signature of Q. G.Q(Y) = B(x. When we specify a field K. Then we get a new quadratic form Q’(x’) with the matrix ‘PAP. Hilbert. In particular. Landau. The matrix A = (uik) is the matrix of the quadratic form Q. the discriminant of Q. x = Px’. hence over C two forms of the same dimension are equivalent if and only if they are of the same rank. and (ii) Q(x + Y) . we put Q(x) = Q(x 1. IXVIII (1897). 1950. Then Q represents p if and only if Q’ represents 0. n). 1964. Die Theorie der algebraischen Zahlkcrper.1 5) Quadratic Forms A. 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’. Math.) [S] B. we say that Q represents Q’ over K (resp.. xf 1 x4= lxi+j(p+q=r). 1932.e. Consider a linear substitution xi = Cjm& PijxJ (i. 1967. the equivalence of the forms is equivalence over K. B(x.) [2] P. [6] E. . Vorlesungen iiber Zahlentheorie. Let V be an n-dimensional vector space over K.A..x. [4] D. . Verein. Springer.pandq ..e. if the rank of Q is Y. Using the notation for the tinner product of vectors. written Q(xl. R). denoted by A(Q).. we write a-b if a. . i. x. 1927 (Chelsea.256 Linear Spaces). In particular.” we usually mean the nontrivial representation of 0 over K by Q. (Original in Russian. we have Q(x) = x& uitxixk. 1969).. If Q is nondegenerate and represents 0. II. and the determinant 1 1 is the disA criminant of Q. .{ 0}. If the coefficients cik belong to the field of real (complex) numbers. Jber. Academic Press. (Sometimes. This gives rise to a mapping X-Q(I) of I/ into K. Equivalent forms have the same rank. y) is nondegenerate (i. Deutsch. Shafarevich. Complex Quadratic Forms [l] Z. On the other hand. Borevich and I. if 1A I# 0). Putting uik = ski = cik/2 (i < k). We call (p. Such a mapping satisfies the following two conditions: (i) Q(ax)=a*Q(x) ((1E K). Q(x) = 0 for some nonzero vector :r. B. 4.. = cii (i= 1. 7). then Q must come from a quadratic form (-.x~+~)=Q(x)--px~+~. L. we assume that the coefficients of the quadratic forms and the coordinates of linear transformations are all contained in the field K. by the expression “Q represents 0 over K. Y) is a tsymmetric bilinear form on V. then it represents any element of K*. The arithmetic theory of quadratic forms. then a form of rank r is equivalent to the form CI=l x”. Dirichlet. . Another important special case is that of n = m. Springer. (Gesammelte Abhandlungen I. we consider the quadratic form Q’ defined by Q’(~~. Jones. Two quadratic forms of the same dimension . Real Quadratic Forms Now let K be the field of real numbers R. G.. . [3] H.. herausgegeben und mit Zusgtzen versehen von R. R. then it is equivalent to the form cfL’=. i = 1. 2-‘B(x.A(Q’).. Then the discriminant of Q’ is given by IPI’(AI. 1969).ifIPI#O(IP( is an invertible element of R). fourth edition.y)=(x. x. a. b-’ E (K*)2. .e. If Q is of rank r. 348 (III. Hence. We assume that the tcharacteristic of K is not 2.347 Ref. For a vector x in V whose coordinates are x1. . 1950. H. 175-546.). If each pij belongs to the field K (to a subring R of K that contains the unit element of K). Viewig. instead of 1Al.)= c 1biQkSn cikxjxk. with an n x m matrix P). Dedekind. III. 1966. Given an element p in K*.63-363.) The rank of the matrix A is called the rank of Q. Vorlesungen iiber Zahlentheorie I. a. Chelsea. We say that Q is nondegenerate if B(x. Then if Q is (equivalent to Q’. then Q is equivalent to a form Cf=. This gives rise to an equivalence relation. Math.Here. y) is called the symmetric bilinear form associated with Q. R). Assoc. we can write Q(x) If K is the field of complex numbers C. Quadratic References 1292 Fields =(x. we say that Q is equivalent to Q’ over K (resp. . The problem of representing numbers by a quadratic form (representation problem) is the particular case corresponding to m = 1.

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

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

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

.(n) is a number-theoretic function of n determined by the genus of Q.b.. 4 if D = -4. Siegel.. then the corresponding form is given by Q(x. Given a form Q(x. [6] C.. 38 (1937). W. and 2 otherwise. . x. the meaning of w is ex- References [1] B.bc = 1. and E is a character mod N. He also showed that from this fact. y are rational integers. Jones.4. The correspondence is given in the following manner: If a is an ideal in k with a basis x. teQ). Q. h. E.. II. n=O Moreover. Elements de mathematique. where . [S] E. where N is the absolute +norm. We let w=$/2 if d=O (mod4). run over all the integers. 9. s.. we have F(z. c. When D > 0. xm) be a positive definite form with integral coefficients.ifwetakea. Actualites Sci.:=s+tw. Eichler. be the class number in I:he finer sense of the forms of discriminant D. of the units is known: it is 6 if D = -3.n)#1. Binary Quadratic Coefficients Forms with Integral With respect to the numbers else is known. . If Tm z > 0. . D(Q) = -4A(Q)). where xi.): where (D/n) is the +Kronecker symbol (here D =f2d. In other words. Witt. 1272a. O’Meara.=r>O. and let h. b. [4] 0. put eD =(t. 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.e.x+a. Reine Angew.. If we denote by A(n) the number of integral solutions of the equation Q(x. 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. y) = ax* + bxy + cy*.(n) + O(nk/*). Quadratische Formen und orthogonale Gruppen.*. Using the theory of modular forms. 37 1:1936). (2) 36 (1935). we again have a one-to-one correspondence between the classes of ideals of this order and the classes of quadratic forms [3]. ue) be the smallest positive integral solution of t* -Du’ = 4. Hecks showed that A(n) = A. 212-291 (Gesammelte Abhand- . 11. -cu (t+bu)/2 >. we put D(Q) = b* -4ac and call it the discriminant of Q (i. Uber die analytische Theorie der quadratischen Formen I. Let (t. y) = ax2 + bxy + cy* with discriminant D are given by + (t-bu)P ( au gonal group is 2. Wiley. where ~1. 1959. Then the units of the form Q(x. [2] N. When D(Q) is not a square. L. Let (to. . Math. we consider the classes of positive definite forms). Let d be the +discriminant of k and put D=df*. Q) is a tmodular form with respect to the tcongruence subgroup of level N. we put Fk Q) = c expWiQ(x.)4. Ann.sD. Springer.a. Springer. Q) = t .and(D/n) = (d/n) if (J n) = 1).fi)/2. we have the following transformation formula: = ~(d)(cz + d)kF(z. Introduction to quadratic forms. + u. Theorie der quadratiljchen Formen in beliebigen Kiirpern. if m = 2k. Ind.13 Algebraic Groups P) [lo]. . [3] M. Q is said to be primitive if (a. .=(1+ $)/2 if d = 1 (mod 4). Siegel’s theory in this section can be deduced (. 230263. d are integers such that ad . The arithmetic theory of quadratic forms. Theta Series Let Q(xi. When f= 1. 1963. Then. Bourbaki. For a complex number z. N is a natural number determined by Q.4(n)e2ni"z. We also have M. .t>O (r... u) be an integral solution of +Pell’s equation t* Du* = +4. 31-44. 1952.y). If f > 1. if we consider the ring formed by the elements x +fjiw (x. little Now put m = 2. 1950. c) = 1. Math. in the correspondence for the casef=l. Hermann. 527-606.348 L 1296 Quadratic Forms plained shortly). T. L. the series converges and represents an tentire function of z. F(z. x2.weput(D/n)=Oif(f. c z 0 (mod N). and . Algebre. III. x. x. 176 (1937). These series are called theta series. J. For D < 0.y)= N(a)-‘N(a.f That is. ch. we must replace the ring of integers o by the +order of the tconductor .) = II. the theory concerning Q is closely related to the arithmetic theory of the iquadratic field Q(3) = k. the order w. 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. Then the following formula holds (Dirichlel.

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* ih---= -rfl gAk 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,

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

ii)> %=(:A (The y’ are called Dirac’s quantity y matrices. v=O. H = cu. M. and the Dirac equation is considered as the classical field of electron waves and is second-quantized (-. the total angular momentum J = L + (h/2)a can be conserved when c~is a vector whose components can be given as (: i).8tlk (k = 1. This was predicted in the theory of light spectra. 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. the neutron for example have spin.2. Many particles besides the electron. y’. forces us to give up considering the Dirac equation as an equation of one electron. Relativity requires the equal handling of space and time.ihr x V. The positron theory is introduced. also called the spin. The existence of negative energy states. where yk = . When the speed of an electron is very small. Y. The Klein-Gordon equation can also be considered to be the classical wave equation of matter and can be second-quantized. are given “1>. however. Although the negative energy case is physically undesirable.3). However. Klein calculated the cross section of Compton scattering (the Klein-Nishina formula) by use of the Dirac equation and found good agreement with observations. because u has four components. p + mc’p. If the spin-orbit term that appears in the Pauli approximation is also neglected. y3 and the identity matrix. where c is the speed of light. where g”‘=O for p#v and go’= -gkk=l (k= 1.) angular S =z c is the intrinsic where the energy eigenvalues _+Jm2c4fp2c2. 1. pi mesons (n) for example.3). E are There are four independent eigensolutions u(l). Therefore we say that electrons are particles of spin h/2. these two components become independent of each .2is diagonal and is equal to s(s+ l)h21 (I is the identity matrix). which include the four matrices y”.2. 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. The momen- bounded below and A + B is essentially selfadjoint. This approximation is called the Pauli approximation. The Klein-Gordon equation is obtained by replacing pk by (h/i)a/dx. and h/2 is called the absolute value of the spin. so that (o/c)’ can be neglected. If gamma rays are absorbed to excitl: an electron from a negative energy state into a positive energy state. Thus the yv are found to be 4 x 4 matrices satisfying the commutation relations yUyY+yVy’=2gfiY (p. (The same formula holds without the bdundedness assumption when t E iR. L is not conserved. (k = 1. For the electron s = l/2. states of the electron can be expressed in 1:erms of twocomponent wave functions. Two of them are of positive energy and the other two are of negative energy. Wave functions of free particles are believed to satisfy this equation. We can rewrite the Dirac equation as ih&+h/& = He. The coefficients y” can be so determined that every component of $ also satisfies the KleinGordon equation.2. Any 4 x 4 matrix can be expressed as a linear combination of these sixteen matrices. The Dirac equation (8) (x0 = ct) satisfies these requirements. ay=(:’ spin matrices. and uc4). Nishina and 0. where B=(i ol). The matri Y S2 = Sz + S.) G. To solve this difficulty. IL(=). Sixteen linearly independent matrices are obtained by repeated multiplication of five matrices. The Dirac equation has +plane wave solutions and the ckr called Pauli by ox=c A).S.311 Second Quantization). an electron-position pair must be created. The Dirac Equation The Schradinger equation is not relativistically invariant.’ + . A. P. y2.3). and y” =/I Since H cannot commute with the orbital angular momentum of an electron L = r x p = . 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. thus providing evidence that the Dirac equation is correct. obey this equation. Motions of particles with zero spin. ~~i(i~ :).3) and E by ih8/& in the relativistic identity E2=m2c4+p2c2.2. uc3). it has to be taken into account in order to obtain a mathematically complete set.351G Quantum 1306 Mechanics ber of electrons in the normal state of the vacuum.

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

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). .. Stationary states can be classified by the absolute value A of the angular momentum of the diatomic system around this axis. 48 (1947)). For example. This invariance is called charge symmetry. isospin can be introduced to describe the two states of nucleans. it is invariant under unitary transformations in the 4-dimensional space spanned by the four internal states of the nucleon: up and down spins. .+JZ+DDJ. The right-hand side of this equation gives all possible states of the atom. Danske Vid.. a point first emphasized by K. The classification of the spectral terms of a polyatomic molecule is related to its symmetry. it is useful to consider the reduction to irreducible parts of the ttensor products of representations of the groups attached to the individual particles. If a Hamiltonian of N nucleons is invariant under any transformation belonging to SU(2). When A > 1. Interscience. Selsk. Medd. antisymmetric. etc. an atom with two electrons carrying the angular momenta J.10DJ2=DJ. and JZ has 25’ + 1 different angular momentum states. which can be zero or a positive integer. depending on the character of the reflections. Phys. If these two atoms are identical. up and down isospins.. Analogous to ordinary spin. Kastler (J. Haag and D.l). are neglected. I. If this Hamiltonian is also independent of spin.) l. the orbital-spin wave function of this state must correspond to the Young diagram [2Nm”.. masses. and p. Segal (Ann Math. Fys. where J’= min(J..+JZ-1+. Friedrichs (Mathematical aspects of the quantum theory offields. In C*-algebra approach. two states (labeled by &) arise.J. and possibly the interaction responsible for the proton-neutron mass difference.. Haag (Mat.hysical examples illustrating this point were given by L. whereas when A =O. To construct such a wave function.-J. where T stands for the total isospin of each state. called isospin invariance. J2) corresponding to the decompositionD. because these two particles have very similar natures except for their charges. This invariance. are classified by the irreducible representations of the group Td (which is generated by adjoining reflection symmetry to the ttetrahedral group T). orbital wave functions with a total spin u/2 should be limited to those corresponding to the tYoung diagram [2Nm”.2. then the eigenstate of these N nucleons is classified by its irreducible representations I). l”].. stationary states of methane molecules CH. Charge Symmetry The proton and the neutron can be considered to be different states of the same particle. When a state of N nucleons has an isospin T= v/2. since the isospin wave function multiplied by the orbital-spin wave function is a totally antisymmetric wave funcl:ion. 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 .+ D. 5 (1964)). As an approximation. the wave function of the total system is constructed by multiplying the wave functions of the individual particles. the corresponding state has twofold degeneracy. 1953). Math. E. K. For example. and the physical relationship among all inequivalent representations of a C*-algebra was first discussed by K. Consider transformations belonging to the tspecial unitary group SU(2) in the 2-dimensional space spanned by the proton state and the neutron state. called the nucleon. 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. holds in the nucleus and elementary particles if electromagnetic and weak interactions.. The up state of the isospin corresponds to the proton and the down state corresponds to the neutron. 0. Therefore the states of N nucleons can be classified by the irreducible representations of the group U(4) (Wigner’s supermultiplet theory). l”] = T(2.351 I Quantum 1308 Mechanics and neutrons may be taken to be invariant under the interchange of protons and neutrons. 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.. described by the set of all transformations that interchange identical atoms. The use of C*-algebras in physics was first advocated by I.. The C*-Algebra Approach J. van Hove (Physica. 29 (1955)). 18 (1952)) and R. In the approximation of many-body problems by means of independent particles. Level structures of a crystal are classified by the irreducible representations of its symmetry groups.

G. such as behind the moon (the behind-the-moon argument).. Wightman.(A). 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. 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. C. (K. 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. to H. product. a. when considered as states on the algebra of observables in the form of expectation functionals ($. From another viewpoint..c>b)isb or q v b.. restrict observables to those leaving the subspace spanned by vectors in each class invariant so that the relevant linear combinations of vectors. 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. and a unit cyclic vector R. meet. called questions... i. n. A nontrivial linear combination (superposition) of these two would then be changed to a vector in a different ray. are physically relevant.. together with their order structure (logical implications) and associated lattice structure (join. In principle.. which constitute a mathematical background for quantum mechanics. e. information on any state cp can be obtained by measurements only up to a neighborhood in the weak topology: 1rp(A. are not immediately discernible from physical observations.(A) T for all As2l. then anycbetweenqvbandb(qvb>. and orthocomplementation as logical sum. such that TX.150 Field Theory) and equilibrium or ground states with broken symmetry. This is called the univalence superselection rule and has been pointed out by A. L. of ‘u by bounded linear operators on H. where weak modularity means cr\(c’vb)=b and bv(b’Ac)=c whenever b < c.402 Statistical Mechanics). A$). where it would form a Boolean lattice. for example.g. and E. S.. A set K. If the 360” rotation is not to produce a physically observable effect.1309 351 L Quantum Mechanics the expectation value of the observable A E 2I when measured in that state) that is linear. while the vacuum vector would be unchanged... Under 360” rotation a vector representing a state of a particle with spin h/2 acquires a factor -1 (. Abundant disjoint states occur for a system of infinitely many degrees of freedom.(A)n.. 11 II= 1 or equivalently cp cp(1) = 1 if 1 &I. 88 (1952)). Rev. in the weak topology contains K. Foundation of Quantum Mechanics Hilbert spaces and the underlying field of complex numbers. The set of all mixtures of vector states ($. all states of K. a(A)+) for any fixed faithful representation 71 of 2l is weakly dense in the set of all states of Cu. are weak limits of states in K. equilibrium states of a infinitely extended system with different temperatures (. However. Conversely any such lattice is a di- . This is called quantum logic in contradistinction to the situation in classical physics. 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. equivalently.e. Two states cp and $ (or rather 7crn and 7tJ are called disjoint if there is no nonzero mapping T from H. then we should either forbid nontrivial superpositions of states of the two classes or. a representation n. such that rp(A) = (Q. positive in the sense q(A*A) > 0 for any A E ‘%!I.) ail<Ei. Wigner (Phys.). In quantum field theory. of states can describe measured information on states at least equally well as another set K. The tGNS construction associates with every state cp a Hilbert space H. superselection sectors explained below (. weakly modular (also called orthomodular) tatomic lattice satisfying the covering law.(A) = 7c. A E 2l. Wick. and are physically relevant if states in K. The lattice of all orthogonal projections (corresponding to all closed subspaces of a Hilbert space) in quantum mechanics is a tcomplete. are actually mixtures (rather than superpositions) of states in two classes and are invariant under the 360” rotation.. all physically relevant information is in the vacuum representation. 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).258 Lorentz Group).. and normalized.i=l. P. physically contains K2) if the closure of K. torthocomplemented. 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. and negation). 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). a point emphasized by Haag and Kastler as a foundation of the algebraic viewpoint in the formulation of quantum theory. in H.

Springer. . 12 (1963). 75 (1980)) has given a characterization allowing direct physical interpretation by replacing P-projection with a measure p and the functions A. M. Dirac. The set of all states (without the normalization condition) of a finite-dimensional. 1979. Pauli. 1974. North-Holland. 1958. the expectation value E(u. formally real. In finite-dimensional cases. and others (Actu Math. 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. 1962. b)-E(u. 1959. Mathematical foundations of quantum mechanics. Fourier. such as real. M. b’)+E(u’. (Original in German. General principles of quantum mechanics. P. Due to some features of quantummechanical measurement not in conformity with common sense. [lo] B. Phys. this type of characterization extends to the “natural” positive cones of vectors (A. Press..1 W. Messiah. ~(a) > 0. all 3 x 3 Hermitian matrices over octanions. Then the following Bell’s inequality holds: IE(u. [ 121 T. Inst. Math. Moscow Math. Emch. It is also possible to characterize the set of all states equipped with the convex structure (mixtures) geometrically.) [4] S. [2] A. (Original in French. Group theory and quantum mechanics. Sot. (Original in German. fourth edition. 1972.351 Ref. Bellissard and B. each of which. In this approach.(l)B. 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 *. representing spins and hence satisfying 1A. Connes. Math. Quantum theory and beyond. [S] H. 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. The theory of groups and quantum mechanics.I < I. 1967. von Neumann. L. by E.... W.e. Dover. then the set of all probability measures on the lattice (i. 24 (1974). von Neumann. Ann. Springer. so that hidden variable theories of this type have been rejected. Cambridge Univ. complex. complex.) [9] E. Ann. F. Wigner. A. . Simon. 1958. J. an additional requirement is needed to restrict the underlying field and its * to be more familiar ones. Araki (Commun. L. [7] B. J. Quantum mechanics I. Shultz. For a situation where a pair of (correlated) particles in states a and b are created and their spins (1 or -1. Clarendon Press. . Tomonaga. Jordan. and E.).1966. North-Holland. In infinite-dimensional cases. 1962. 13 (1965)).. if of dimension (length of longest chain) > 3. Wiley. B. called P-projections. or the so-called spin balls (the set of all normalized states being a ball) linearly generated by the identity and yj (j = 1. Fourier. Quantum 1310 Mechanics notion of filtering corresponding to quantummechanical measurement. Principles of quantum mechanics. 28 (1978)). Algebraic methods in statistical mechanics and quantum field theory. [S] J.) [6] B. b)= A. P. there have arisen hidden variable theories that are deterministic and reproduce the quantum-mechanical prediction.) C3. Weyl. 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. 1!)71. Alfsen. G. van der Waerden. 1980. equipped with a nondegenerate Hermitian form. 1955. of hidden variables 1. Functional integration and quantum mechanics. van der Waerden. Iochum. Dover. Press. Group theory and its application to the quantum mechanics of atomic spectra. b) for the product would be given in a hidden variables theory by E(u. Academic Press. b’)1<2. 1932. or quaternion fields and their usual conjugations *. 1949. 140 (1978). n) satisfying yj o yk = 0 ifj # k and yj2=1. References [l] P.e. Sources of quantum mechanics.(L)dp(L) for a probability rect sum of irreducible ones. Inst. up or down spin) measured at positions distant from each other.. b)+E(u’. II. Bastin (ed. Wigner (Ann. 1961.[ < 1 and 1B. and B. Quantum mechanics I. If that is done. (Original in German. where the relevant Jordan algebras were completely classified earlier by P. Princeton Univ. Vinberg (Trans. 1959. or quaternion field. [ 111 G. 144 (1980)). p( 1) = 1) is exactly the restriction to questions of states p(A)= tr(pA) given by the density matrices p (Gleason’s theorem). This contradicts both quantum-mechanical predictions and experimental results. II. 1928. Academic Press. 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. i.

Beurling and L. (1) (the geometric delintion) Let Q be a curvilinear quadrilateral..lf?l) is bounded.l)/(K + l). K. and The inverse mapping of a K-quasiconformal mapping is also K-quasiconformal. The quasiconformality off is defined as follows. which are not conformally equivalent to each other. he determined the quasiconformal mappings between two given domains. 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:.1 I&l)ldzl. for short a p-conformal mapping. K-quasiconformal mappings are simply said to be quasiconformal. If in the above statements f is not necessarily topological but merely a continuous function satisfying the same requirements. Quasiconformal mappings have less rigidity than conformal mappings. V.1 . Virtanen.-quasiconformal mapping fi. A. Also .l . d’Espagnat. which has led to the precise definition for quasiconformality itself. then f is called a Kquasiconformal mapping of D. The maximal dilatation (1 + ll~llm)/(l-ll~ll.quasiconformal.e. we call it a Kquasiregular function or K-pseudoanalytic function. including the theory of Teichmiiller spaces (. those closest to conformality L. Conceptual foundations of quantum mechanics. A 1-quasiconformal mapping is conformal. and (If. 1976. These applications are explained in Sections C and D. When the ratio K(z)=(lf. a closed Jordan domain with four specified points on the boundary.) A p-conformal function is represented as the composite go h of an analytic function g with a p-conformal mapping h. Principal Properties and Results B. e.references to 150 Field Theory. Let f be a K-quasiconformal mapping of lzJ < 1 onto 1w I< 1. Mori [4]. is K.1 + I. Let f(z) be a continuously differentiable homeomorphism with positive Jacobian between plane domains. The image of an infinitesimal circle (dzl = constant is an inlinitesimal ellipse with major axis of length (If. called the modulus of Q and denoted by mod Q.1311 352 C Quasiconformal Mappings [ 131 B. Bers [S] (.367 Riemann Surfaces). Morrey [6]). Ahlfors (. Then f extends to a homeo- . C. Every quasiconformal mapping is ttotally differentiable a. Teichmiiller. and 386 S-Matrices. almost everywhere in D with the measurable coefficient p.-quasiconformal mapping fi with a K. are due to Ahlfors [3]. (If in addition //pII m <(K . Lavrent’ev [Z]) and especially in the problem of moduli of Riemann surfaces.143 Extremal Length). 375 Scattering Theory.l+ l&1)/( If. Griitzsch (1928) introduced quasiconformal mappings as a generalization of conformal mappings.1 + Ifsl)ldzl and minor axis of length (If. is uniquely determined.e. S. The ratio (> 1) of the sides of I. If modf(Q) < K mod Q for any curvilinear quadrilateral Q in D.. Griitzsch noticed that Picard’s theorem still holds under the weaker condition. and for this reason they have been utilized for the type problem or the classification of open Riemann surfaces (Ahlfors. Consider an orientation-preserving topological mapping f of a domain D on the z( =x + iy)-plane. Quasiconformal mappings have important applications in other fields of mathematics. If K = 1. second edition. we call it a p-conformal function.C.I <K+l K-1lf I I almost everywhere in D with some constant K > 1. then f is conformal. where the (-1.1 5) Quasiconformal A. Kakutani. History Mappings H.416 Teichmiiller Spaces).fMlf.I-. in the theory of tpartial differential equations of elliptic type (M. We cannot speak of the history of quasiconformal mappings without mentioning the discovery of extremal length by A. which dispense with continuous differentiability. Definitions The current definitions of quasiconformality. I. i. and let the interior of Q be mapped conformally onto a rectangular domain I. l)-form p&dz-’ is independent of the choice of the local parameter z.e. The composite mapping fi ofi of a K. that is. its Jacobian is positive a. Benjamin. A. 0.I)6 K a. 377 Second Quantization. 352 (XI. f is called quasiconformal. When the value of K is irrelevant to the problem considered. K. B.e. .g. Y. providing the smallest sup K. Sometimes f is called. if it can be defined. (almost everywhere).. These notions are also defined for mappings between tRiemann surfaces.)does not exceedK. Toki. The K-quasiconformal mapping f satisfies the so-called Beltrami differential equation VI.

Let K(f) denote the maximal dilatation of a quasiconformal mapping f: Suppose that a family 9 = { jJ of quasiconformal mappings is given.I@(z)ldxdy of @ in (1) is finite. If.y’)IO<x’<a’.y)(O<x<a. Given a measurable function p in a simply connected domain D with llpll I. at which @ #O (Teichmiiller [ 121.O).]<l. real analytic..I d C for anyzree points <. (1) morphism of ]z] < 1 onto (WI d 1. Consider again a p-conformal mapping 9 of the unit disk D: ]z] < 1 onto itself which induces a topological autom0rphis. +m p(x) = k 00. t E R. Lehto and Virtanen [7] and Ahlfors [S].0)form @ such that the respective local coordinates z. (b) Mapping Problem. the uniqueness does not hold in general (Strebel [ 131). 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. then the Holder condition holdsfor]z. If some f0 E$ exists such that K(j. Otherwse. The extremal mapping f satisfying equation (1) is sometimes referred to as the Teichmiiller mapping. complex analytic) in t. For further properties and bibliography . . 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.l)/(K + l)]‘o/I@. w of R. As to the Teichmiiller mapping. the Teichmiiller mapping is the unique extremal quasiconformal mapping in 9. continuously differentiable. Ahlfors and Beurling characterized the correspondence between ]z] = 1 and 1WI = 1 induced by f [9]. Let L denote a curve which passes through cc and divides C U {a} into two domains Q. [. (c) Extremal Quasiconformal Mappings.and16isthebest coefficient obtainable independently of K (Mori).?formal mappings are made essential use of.]<l. This shows that any family of Kquasiconformal mappings of ]z] < 1 onto ]w] < 1 is tnormal. Concerning the dependence of p-conformal mapping on p. If {p(t) = ~(z. 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. complex analytic). Then the unique extremal quasiconformal mapping for 9 is the aftine mapping x’=(a’/a)x.(0. there exists a p-conformal mapping of D onto a plane domain A which is unique up to conformal mappings of A [S]. a necessary and sufficient condition is proved for the Beltrami . a classical result concerning the tconformal mapping of surfaces asserts the existence of a solution of Beltrami’s differential equation fi=A!fZ. .m of the boundary ]z] = 1.O<y<h}. 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.. On the other hand. the following theorem holds: Let p(x) be a realvalued monotone increasing continuous function on R such that lim. S and a thomotopy class . Strebel [ 133). Next suppose that we are given 1. Let 9 be the family of all quasiconformal mappings of R onto R’ which map each vertex to a vertex with (O.~0 homeomorphic closed Riemann surfaces 19.O<y’<b’} be a pair of rectangular domains. which have opened up a new way to study +Teichmiiller space. & on L such that [3~6.0. then fr@) is also continuous (continuously differentiable. R’= {(x’. and the space of mappings f@ also has the structure of a Banach space with respect to a suitable norm. When p is real analytic and the derivatives of functions are defined in the usual manner. Then 9 contains a unique extremal quasiconfcmrmal mapping. then f0 is called an extremal quasiconformal mapping in 9 Let R={(x. ]/I[. For the proofs of these important results. More precisely. the extension and reflection of quasico. the uniqueness theorem is as follows: If the norm II@11= jJ. Theorem of quasiconformal reflection (Ahlfors [lo:]). then the extremal quasiconformal mapping in 9 exists but is not always determined uniquely (K.<2. What amounts to the same thing. The space of functions p has the structure of a Banach space with L&-norm. (a) Boundary Correspondences and Extensions. 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. < 1.O). c. furthermore.) attains the infimum of K(f) for all fog.]z. Ahlfors [3]).P of orientation-preserving homeomorphisms of R onto S.f(O) = 0. If we define 9 as the family of all quasiconformal automorphisms f of D satisfyingf(e”)=g(e”). S satisfy the dilferential equation (I?w/Z)/(C~W/~~Z)=[(K . real analytic. Q* such that R U L U a* = C U {m}.352 C Quasiconformal 1312 Mappings continuous (resp. y’=(b’/b)y (Grotzsch Cl]).

Noordhoff. Hamilton [14]. A. [2] M. Mori. and VHidlH [ 181). 114-120. Ahlfors. Amer. Trans. S. and that if in addition the supremum of the tMach number is smaller than 1. 1963. Varational methods for boundary value problems for systems of elliptic equations. [S] L. Amer. Verh. On quasiconformal mappings. Sot. 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 ‘Il(x))n < K. Similar Notions The term quasiconformal was used differently by Lavrent’ev. V. For n > 3 the following properties also still hold: A quasiregular mapping is discrete. then the mapping from the physical plane to the thodograph plane is pseudoanalytic. This function is pseudoanalytic in the sense of Section B on every relatively . 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. 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..367 Riemann Surfaces) of Riemann surfaces are invariant under quasiconformal mappings. Carleson [15]. E. Bers used the term pseudoanalytic to describe a certain function related to linear partial differential equations of elliptic type. Math. as they are under conformal mappings. the operator norm and the determinant off’ are denoted by Ilf’ll and det f ‘.. =i 1 ‘(x + y) I”] holds f almost everywhere in G is called the inner dilatation off and is denoted by K..K. VHisala [ 173) if it is a K-quasiregular homeomorphism. Sachs. J. 3 (1954) 1-58. detf’(x) almost everywhere in f G. Faber mijglichst konforme Abbildungen von schlichten Bereichen.1313 352 Ref. If max(K. The smallest K > 1 for which this inequality is true is called the outer dilatation off and is denoted by K. The theory of quasiconformal mappings was also applied by Lavrent’ev [16] and Bers [2] to partial differential equations. Lavrent’ev. Applications In the earlier stage of development of this theory. When n = 2. Analytic transformations in the theory of functions of several variables are called pseudoconformal by some mathematicians. 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. An orientation-preserving mapping is called K-quasiconformal (J. it is reduced to a quasiconformal mapping if the system is uniformly elliptic. 56-77. 1 for which the inequality det f ‘(x) < K. On quasi-conformality and pseudo-analyticity. This is a generalized definition because the system may not be equivalent to a Beltrami equation. and there is a similar term quasi-analytic.367 Riemann Surfaces).207208.( f ).. Leipzig.. Akad. 84 (1957). particularly to those concerning the behavior of fluids. 84 (1932). The latter is an entirely different notion from the one discussed in this article. and O. Martio. quasiconformal mappings were applied only to the ttype problem of simply connected Riemann surfaces and to the classification of Riemann surfaces of infinite genus (. References [1] H. Quasiconformal Mappings coefficient p of a quasiconformal mapping of 9 to be extremal (R. these definitions agree with those given in Section B. 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). (Original in Russian.. then the smallest K >. D. thenfis said to be K’-quasiregular. Moreover. Reich and Strebel in [ 151). Bers.( f ). Analyse Math. compact subset and has properties similar to those of analytic functions. They utilized the fact that if the density and its reciprocal are bounded in a steady flow of a 2-dimensional tcompressible fluid. On a theorem of Mori and the definition of quasi-conformality. open. [min. Grbtzsch. and is absolutely continuous (0. Rickman. 84 (1957) 78-84. Wiss. this last result can be extended to the extremal quasiconformal mapping between arbitrary Riemann surfaces. totally differentiable a. Trans. respectively. Quasiconformal extension of higher-dimensional half-spaces have been studied by Ahlfors and L. [4] A. 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. F. Furthermore. However.) [3] L.e. Sot. S. Math.(f). If f is quasiregular. 1962. (.(f))<K’. E. Ber.

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

.

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

1. Phys. Irreducible tensorial sets. Theoret.jm) =(-1)j1’j.j. Wiley. 1957. [S] L.j. and fsampling or trandomization techniques in statistics. respectively. The matrix element of this quantity between two irreducible components can be written in the form where CL a parameter to distinguish is multiple components with the same j. Edmonds.. while (a(ill T@IIGL’~‘) depends on the dynamics of the system. -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. Rev. It is a basic tool for the tMonte Carlo method. Group theory and its application to the quantum mechanics of atomic spectra. Irreducible A dynamical quantity qk (q = k. 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. Morita. [TV+’ 0 U’kz’]: 354 (XVl. Random numbers used in practice are pseudorandom numbers (. jk) For an irreducible component of the tensor product of two irreducible tensors. De-Shalit. Tsukada. of the total space (tHilbert space) H = H.)j.~~ U(k2)IIalj. H. Bloch. [7] M. Blatt. R. S..C(. Angular momentum in quantum mechanics. . and I.(j. See [6] and [7] for explicit formulas of Clebsch-Gordan coefficients and [S] for Racah coefficients. That is. qkl =qT.j. Princeton Univ. Pseudorandom Numbers Tables of numbers generated by random mechanisms have been statistically tested and . 1959. U. 64-74. Morita. B.)j. E. k . 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. Phys. C. S. Clebsch-Gordan coefficients for j. T.j.+ij. b] = ab .( -l)q qk U.j. .” has the matrix element (a1C(2jlj2jml(T(k).~kl==(kTq)(kfq+l) CL.. Wigner. P. 1935 (reprinted with corrections. Spectroscopic and group-theoretical methods in physics (Racah memorial volume).-jW(j. 3. theoretically. [3] U. Cohen. M. Mod. Racah. 1951).U(k))la. Fano and G. Theory of atomic spectra. Elementary theory of angular momentum. Press. Suppl.. North-Holland. Some properties of the Racah and associated coefficients.. 26 (1963). and M.“. and components of different CIare assumed to be orthogonal. [4] A. and H. Rose. Cj. . G. 1959.ba. = 512. When Vk) and Uck) operate only on the state vectors in the subspaces H. tsimulation of stochastic phenomena in nature or in society. Rose. it satisfies The 9j-symbol can be written as a weighted sum of the products of the three w’s. and M. Press..j. [6] E. 1959.1317 354 B Random Tensors Numbers C. the 9j-symbol. Sambutsky.j.5) Random Numbers as A.(j. Prog. Cambridge Univ. A. [S] F. General Remarks the matrix (aj. x Hz. and 712.. j. Yamada.k+l... Here [a. Academic Press. Talmi. R.jll[T’kl’@ can be written Uck2)]@lldj.) The last factor. 24 (1952) 249-257.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.)jm) =JGj12+ l)Ch+ 1)(2j13+ 1)(2L4+ 1) CL21. Academic Press.(j.j. In this formula the Clebsch-Gordan coefhcients are determined from group theory.Section B). .)jmlj. [2] M.)(cCj. Uck)) = C. their scalar product ( Tck). 1968. Shortley. E.j. T. J. References [ 1] E. Biedenharn..(j. Condon and G.j.

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

102. [lo] T. the cp-subsequence is defined as &. (ii) For each pair of numbers x. 1}-valued function on the set of (say) finite decimal sequences such that {n. N. there exists one and only one number w E R. 2. 3. Harvard Univ. . all tfour arithmetic operations. H. . . References [l] D. Chaitin. (ii) Order and . &i)=l}. called their product and denoted by xy. let N&x. . en-r)= l} is an infinite set for any d<1. x = y. (xy)z = x(yz) (associative law). Algorithmic information theory. No one has so far been able to prove or disprove the normality of such irrational numbers as rr. . Statist. Church. Based on the notion of collectives by R. . Addison-Wesley. there exists one and only one number w E R. Bk) be the number of occurrences of the block B. Axioms for the Real Numbers ClOl. . then x is said to be normal to base r. Kolmogrov. &. R. For a selection function rp and an infinite sequence 5 1.4. 9. F. infinite sequence <i. [2] E. . . 141-146. . e. Sowey. in the sequence x1. Wiley-Interscience.9 appears in it with a limiting relative frequency of l/10.. for each x. A selection function is a (0. 13. ..<n.)/n+r-k as n-+cc for every k and every B. Bull. tZ. . an infinite decimal sequence is called a @collective if each of the numbers 0. a definition of infinite random sequences has been given by A. ch. . 21 (1977). or x > y.B.. [7] A. . Champernowne [S] constructed a normal number given by the decimal expansion 0. J. y E R. . Subsequences of normal sequences. 602-619. fi. Schmidt [9] proved that the normality to base r implies the normality to base p if and only if logr/logp is rational. 9 (1966). IT-14 (1968).. 121-149. &) of numbers 0.. 89. Niederreiter. PER. Schmidt. The art of computer programming II. for which x + y = y + x (commutative law) and (x + y) + z = x + (y + z) (associative law) hold. Martin-LGf. . With x < y meaning x < y or x = y.}={n. .. For a class Ic/ of selection functions. Logical basis for information theory and probability theory. second edition.1. On normal numbers.“‘. in other words. Information Theory. ye R. 6. . 5. . called their sum and denoted by x + y. there exists one and only one number -XER for which x+(-x) = 0. 10. Lehmer.. Furthermore.obey the usual laws (with the single exception of division by zero).. clearer definition is possible. and the same thing holds for any q-subsequence with (PE $. one and only one of the following three relations holds: x < y..(p(~. Kuipers and H.. . there exists a unique number 1 (unity)ER such that lx = x for every XER (existence of tunity element). (2) Order properties: (i) For each x.xn. r .10) Real Numbers A.. The set R of all real numbers has the following properties: (1) Arithmetical properties: (i) For each pair of numbers x. D. ..1319 355 A Real Numbers [4] A. R is a ttield. = (b.. Mathematical methods in large-scale computing units. . Proc. E. Develop. 16 (1973). [6] P. Almost all real numbers are normal to any r. Rev. Press.. .. 1951.9 be the r-adic expansion of the fractional part of a real number x.1. A real number whose decimal expansion is random in the above sense is normal to base 10. fi. Sot. 1981. for each x # 0 (x ER) there exists one and only one number x-l ER for which xx-i = 1. Math. 47 (1940) 130-135. von Mises. Int. . 8. 1. . W. Israel J. Amer. 12. which makes R ttotally ordered.< z. [9] W. . Also. For the converse. a random sequence is a $collective for the class $ of recursive selection functions. [S] G. [ 2.52. [S] L. Normal Numbers Let x . on Large-Scale Digital Calculating Machinery. On the concept of a random sequence. IEEE Trans. G. 3. Pacific J. 662-664. &) depends only on L) to have the property that the normality implies the {cp}-collectiveness has been obtained in 355 (11.. 10 (1960). however. . Math. 46 (1978). . Furthermore. [3] D. Math. Uniform distribution of sequences. 7.. 1974. . there exists a unique number 0 (zero) such that x + 0 = x for every x E R (existence of tzero element). Knuth. . Church [7]. A second classified bibliography on random number generation and testing.. the relation < obeys the transitive law: x < y and y < z imply x .. By definition. Almost all real numbers are random in their decimal expansions. for which xy = yx (commutative law).<. 661-672. 350359. For any ordered set B. 2nd Symp. .. IBM J. Information and Control. . and (x + y)z = xz + yz (distributive law) hold. 11. 1. Kamae. a necessary and sufftcient condition for a selection function cp (for which (p(l. Also. The definition of random sequences. Res. If Nn(x. where{n. .[x] = Z x. . Owing to properties (i) and (ii).

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

third edition. Herbrand. 1969.-mb. any proper closed subgroup I of R is discrete and isomorphic to the additive group Z of integers. how shall we characterize a number-theoretic function that is effectively computable. Kleene later improved Godel’s definition and developed the theory of general recursive functions [2]. Here. and (po.x. 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. cp(pl)= 1.-*a/b (where b #O. pi be two distinct points on 1..b. Weierstrass. Dieudonne. English translation. 1143c. We employ the letters x. Foundations of modern analysis. 189441927. 1930. situated to the left of pl. Iff(l)=a.+ab.1321 356 A Recursive Functions defined by the notion of convergence (. Then the following problem naturally arises: How shall we define a finitary method? In other words. General topology pt.. Then q(p) is called the coordinate of the point p. Addison-Wesley. C. Springer. Ind. Cantor. Godel [l] considered certain number-theoretic functions. pq = p’q’ (pq and p’q’ are tcongruent)o cp(q) cp(p)=cp(q’)-a. introduced independently and almost simultaneously. References [l] N. Kleene [2] (the definition is given in Section B). 1966. then a. say from left to right.2.-+a + b. x2. for some a>0 we have T={neln~Z}.+a and b. Turing defined the notion of computable functions by introducing the concept of Turing machines. a. Gesammelte mathematische Werke I-III. . The Real Line Let 1 be a Euclidean straight line considered to lie horizontally. 1. 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’.. 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. b. These notions. [S] K. were found to be equivalent. Arithmetic operations in R are all continuous: If a. 1932.87 Convergence).references to 381 Sets. a. z. y. Regarding R as a topological Abelian group (with respect to addition). A function whose tdomain and trange are both the set of natural numbers (0. 4. Academic Press. Gesammelte Abhandlungen l-7. Hilbert (1926) and K. Mayer & Miler. To be precise. [2] G. Post and A. In this article. [3] R. then J g are uniquely determined and are written f(x)=aX. Elements de mathematique III. L. and (iii) for two line segments pq and p’q’ (where p and p’ are to the left of q and q’. a natural extension of the notion of primitive recursive functions. A. Church and Kleene defined I-calculable functions using the l-notation (Church [4]). In particular. 356 (1. Nombres reels. Hermann. Hence such functions are now simply called recursive functions. 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. the quotient group R/Z as a topological group is isomorphic to the rotation group of a circle (l-dimensional ttorus group). Bourbaki. Akademische Verlag. Also . General Remarks E. Elements of R/Z are called real numbers mod 1. we give it by utilizing the idea of introducing schemata.. Xl. ch./b. } is called a number-theoretic function.1932. [4] J. enlarged and corrected printing. instead of giving the definition of recursive functions in the original style (the Herbrand-Godel-Kleene definition).9) Recursive Functions A. 1960. (ii) if p lies to the left of q. Braunschweig.b. with p. g(b)= 1. Let PO. the term natural number is used to mean a nonnegative integer. Topologie gtntrale.. and E. for variables ranging over the natural numbers.+b. Dedekind. Hence R is a ttopological field (regarding the characterization of R as a topological group or a topological field .. and a. . 1. Godel introduced an efficient method of arithmetizing metamathematics based on representing certain linitary procedures in metamathematics by primitive recursive functions. respectively). #O). Furthermore. Gesammelte Abhandlungen. Actualitts Sci. 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). M. following the suggestion given by J. That is. g(x)=log. then q(p) < cp(q). pJ the frame of the line 1.422 Topological Abelian Groups). + b. . called recursive functions by them and now called primitive recursive functions after S. 1960.

Q. . x. 4)(X. Now.)=~Y(lcl(x. if 3yR( y).y)=O).. . . when cp is definable by applying a primitive recursive operation to * I..x. R. otherwise.). &. 4)(X.... .. + (-./ x. (the exponent of pi of the unique factorization of u into prime numbers if a #O. or (III). y)=O) is not necessarily defined for each n-tuple (x 1.411 Symbolic Logic). . XJ of natural numbers. ). . .....:c. 1936) inveslzigated in detail functions that are definable: in general. . . u < b. A. . . a ( b (a divides b)..)=~(x. A function is called primitive recursive if it is definable by a finite series of applications of the operations (IV) and (V) ($. General Recursive Functions Then we call P a primitive recursive predicate if its representing function cp is primitive recursive. tjr and from functions each of which is given by (I). that are valid for pri- . = 2. . .$h. . .x.. it equals z.. .. . For a predicate R(y) on the natural numbers.maO. . and the finite sum ZYcZ is a primitive recursive operation in this sense. .). ..) = xi (1 . . pi (the (i + 1)st prime number.356 B Recursive B.“> x. la-bJ. x. ) / x. . The following operation is also primitive recursive: cp(Y. if there exists such a number y. we always transform it by replacing the predicates contained in it (if any) by corresponding representing functions. . . .x2 where /. min(a.y) is the least y such that y<z and R(x. i<y A function cp is said to be primitive recursive uniformly in $1. .Q. . . . in addition to those used to define the primitive recursive functions. ..x.)=0..x. . a’b. .. otherwise. pyR( y) is undefined.x1.. .. . .. .) if cp takes only 0 and 1 as values and satisfies P(x. and others. Primitive 1322 Functions Recursive Functions five definition Similarly. $. .@(y. $I if it is definable by a finite series of applications of (IV) and (V) starting from $1...x.. . such that its representingfunction is general recursive. Robinson (1947).. a”.1)> =x(y. There are further investigations on primitive recursive functions by R. .1 X.. .) X&l >. of cp from any function II. by (definition.. x2. Q 1.(x. VyY.. . . py($(xl. . a!. Ptter (1935. a=b.)=q (4 a given natural number). .. . ... x. where li/( ) is a constant natural number if n = 1. z) The following p-operator is used to define general recursive functions by extending primitive recursive functions.. The facts... (a). . (1.x* . Whenever we are given a concept or a theorem. x..). Almost all results mentioned in this section were given by Gadel [ 11. Then an operation R is called primitive recursive if the function or the predicate a($. Put cp(x.b). X”)O’p(Xl.) that results from the application of 51 to functions tjl. a primitive recursive function is general recursive. . Generally. The following functions and predicates are examples of primitive recursive ones: a + b. .. $[.R(x. otherwise.> *. (II). .. VXJY(~(X. Consider the following schemata: (I) (11) q(x) = x’ (=x+1).Y)=o) for the definition that satisfies vx. v .) is the representing function of a tpredicate P(x 1. x.... . pyR( y) is the least y such that R(y). we define the relativization (with respect to +.. I+m>O) is primitive recursive in +. the bounded quantifiers Sly...x. C. and the bounded p-operator pyYcz defined :as follows: py.. by k-fold recursions for every positive interger k PI.. Pr(a) (a is a prime number).X. max(a. Given the functions G1.. are already-introduced functions) starting from functions each of which is given by (I). A general recursive predicate is a predicate x. or (III). 0).... 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> .b).1 x. . Note that a function definable by a +double recursion is not necessarily primitive recursive.<i<n). P&er (1934). recursive in $.. . definitions by cases. Thus. .“‘. the following operations are primitive recursive: the finite product n:y<rr the logical connectives 1. including the ones concerning relativization.x.. Then cpis primitive =~y<I. p. y) holds. p1 = 3. Q. (III) (IV 4)(X . y). . .. $I and predicates Q. . . x.). . We say that a function cp(x. $J of the definition of primitive recursive functions as follows: A function is called primitive recursive in $1. .. . (II). .. ...

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

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

He dealt with functions defined.31. Fenstad. Given an admissible ordinal K. v. Machover. Tug& S.. Moldestad. The notions of degrees of unsolvability and recursive enumerability can also be generalized.. there are also investigations by A.. and introduced the notion of admissible sets. S. . w (-1 3tf-2) 2 r>2. w. Levy. Sets or functions are described by tpredicates. . be variables ranging over %>‘32.)(m. . wo3pVf-2N(e a 99. enumeration theorem. Kleene succeeded in establishing a theory of hierarchies that essentially contains classical descriptive set theory as an extreme case [S. as in the case of general recursiveness. and others. set theory. . be variables ranging over the set N of natural numbers. . and others in abstract computation theory [20&23]. V& it is called arithmetical in $i . $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.. a. schemata. In particular. . e.. The fine structures of these properties are currently the objects of intensive research.n>O. and GI. An admissible ordinal K has the closure properties required for the construction of the calculus.1325 356 H Recursive recursive predicates M. .. V<. there are some investigations by Y. C. E.g. not on a segment of ordinal numbers. .. T.0 2. Although research following a similar line had also been done by M.a. etc.. and infinitary logic are closely related. . For brevity.. . N. then /?=f(cc) is computable in fewer than K stages... $r (I > 0). and denote by a a finite list of variables (a. .5>%. which we classify as follows. not all the partial recursive functions of variables of types > 2 can be obtained by applying schemata (I))(VIII) and (IX’). but on a set. An admissible set is a transitive s-model of a certain weak set theory. 10. and whenever a. Mostowskii. Davis. Using this.. = K.. and others. he constructed a model of set theory in ordinal number theory. the set NN of all tnumber-theoretic functions with one argument. Takeuti and A.p. Every arithmetical predicate P(a) is expressible in a form contained ... The least admissible ordinal is w. A. and introduced the notion of admissible ordinals. Kino. In connection with recursive functions of ordinal numbers. respectively.32]. 35. In fact. a. (‘-l. x) = 0) instead of (IX) is partial recursive.. . .. In addition to the abovementioned. and the next is the ordinal wi of Church and Kleene. &. the first nonconstructive ordinal. the normal form theorem. by various equivalent methods. Let a. the first ordinal not expressible as the order type of a A. the equation calculus. . . . sets on which a well-behaved recursion theory can be developed. K-recursiveness can be defined. For example. Further developments have been pursued by J. N Functions are two primitive such that @)(a)woV(‘-13v]‘-2M(e. Subsequently Kripke observed that the assumption that IC is a cardinal is not necessary. yielding the notions of k-degrees and K-recursive enumerability. . y. Recent developments have shown that generalized recursion theory. cc. i. .. Every infinite cardinal is admissible. .q’-‘). .. where 0. consider the case I= 0. a. b. For each infinite cardinal IC there are K+ admissible ordinals of power K.. Recursive Functions of Ordinal Numbers G. Takeuti introduced a notion of primitive recursiveness for functions from a segment of the ordinal numbers to ordinal numbers.g.e.) are also valid for K-recursive functions. is the class of all ordinal numbers. J.’ predicate is admissible (Section H). A. Hierarchib Utilizing the theory of trecursive functions. . they are called simply analytic and arithmetical. . .. . Most of the elementary properties of general recursive functions (e. Let $i.. . . x. . . When 1= 0. . Vx. parametrization theorem. and an ordinal K is admissible if and only if there exists an admissible set A such that A n 0.. .e. However. for every n 2 1. A predicate P(a 1. and definability in both quantifier forms.11/# 2 0) be number-theoretic functions. Every function definable using (IX’) $(a)px($(a. Kripke.m+n>O) with variables of two ttypes is called analytic in $i.3x. ra2. /l< K and /I =f(a) is computable. Early treatments of recursive functions of ordinal numbers dealt only with functions on infinite cardinals. M. respectively. a. a. to tgenera1 recursive predicates in $i.). Takeuti considered functions with a fixed infinite cardinal IC as a domain and a range. it was Kleene who succeeded in bringing the theory to its present form. al. Moschovakis and others [14-271.. 1. H. when P is expressible without function quantifiers 35. and defined ICrecursive functions using schemata similar to the abovementioned (I)-(VI). G. . i.

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

3tl. . the hierarchy theorem. Let a’ be a list of variables of types < t. . Post. 1957). each predicate P(a’) of order r + 1 (r > 0) is expressible in one of the following forms: (4 B(a). Recursive functionals and quantifiers of finite types II. Van Nostrand. 3a’V<‘-‘R(a. H. &. Steklov.81 Constructive Ordinal Numbers) of notations for the constructive ordinal numbers as follows [6]: H. Monatsh.. VC(‘38rVS’-1R(a.. a).(a)o3xT.). Symbolic Logic. Trans. 42 (1954). C. L. 3aacHypA(a. Springer. I+$ (I > 0) (for brevity. . for yo0.(a)a = a. Phys. Math. Inst. Ann.cr’. for a II: predicate P(u. t’-‘). Godel.. Kleene. 150-l 55.2. for some y E 0.x). These concepts and the results mentioned below can be relativized to any given functions or predicates.8r. Markov. predicates that are general recursive in ‘I”. 91 (1959). . we have the enumeration theorem. then Hy and Hz are of the same degree (C. <‘-‘). [lo] S.. t’-‘) 03n’-1V<r-ZR(a’. Amer. Denote by Hyp the set ( c NN) of all hyperarithmetical functions a. 1941. We say a predicate P(d) is of order Y in completely defined functions $i. A. Trans. Akademische Verlag. and symbols of the tpredicate calculus with quantification consisting only of variables of types cr. Math. 1. cr’. Church. otherwise E(a) = 1. Trans. 108 (1963).nyp VxR(u. 38 (1931). Enumerability. for the predicates P(a’+‘) in each form. Sot. l-52. 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). VI-i. but the converse does not hold in general (Addison and Kleene. 112 (1936). [2] S. The calculi of lambdaconversion. 53 (1943). Clarke [30] has published a detailed review of the general theory of hierarchies. . Amer. l’-‘) such that 30’V~r-zP(a’. C. [4] A. Hermes.. A. a. Uber formal unentscheidbare SHtze der Principia Mathematics und verwandter Systeme I. Math. . . Math. a’. Press. y) a 3tlmEHYp VxP(x. The predicates of order 0 in Y are exactly the general recursive ones in Y. [ 1 l] S. Kleene. for any l7: predicate P. . . A predicate that is hyperarithmetical relative to ZT. Peter. 3a’V/I’Zll’-‘R(a. Thus a hierarchy of degrees is uniquely determined by constructive ordinal numbers. predicates (k>O) is of A:+i (Kleene [32]). Spector [34]). 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’. and Y are functions of variables of type 0. D.~(u). Sot. decidability. there is a general recursive predicate R such that P(u). <‘-‘). Trudy Mat. Kleene. 3 (1938). This H.1327 356 Ref.a. [S] S.. Va’3<‘-‘R(a. <‘-‘). A function &z. [7] S. Math. Amer.. H~. . a(x)) (G. . Conversely. [6] E.. a predicate P(d) is of order 1 (of order 2) in Y if and only if P is arithmetical (analytic). Theory of algorithms (in Russian)./?. In fact. denote them by Y) if P is syntactically expressible in terms of variables of finite types. General recursive functions of natural numbers.. computability. = (~1 (no). Kleene. n’-i. When t > 1. 5y~0. where Y. 284-3 16. [9] A. a) is always a n: predicate (Kleene [33]). Amer. Kleene. A function or predicate is said to be hyperarithmetical if it is recursive in H... Sot. <r-2). 1965. Kleene. [lo]. x) (Spector [35]). J. . (. Rekursive Funktionen. C.~Y(u)~H. and it is of a properly higher degree than that of Hz when z<. References [1] K.. For an arithmetical predicate A(a. for 3. 195 1. Introduction to metamathematics. Sot. Let E be an object of type 2 defined by: E(a) = 0 if 3x@(x) =0). Math. 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.~~.. [S] R. . . If lyl=lzl (lyl is the ordinal number represented by y). Recursive Functions the system S. <‘-‘). Recursive functionals and quantifiers of finite types I. Recursive predicates and quantifiers. b). (6) Using these equivalences. When t = r + 1. 106-142. 173-198. C. C. table (c) gives the classification of the predicates of order r + 1 into the hierarchy. LX’. a’. This hierarchy is called the hyperarithmetical hierarchy of degrees of recursive unsolvability. is defined for each y E 0. a.y. 727-742.H~(a. Kreisel. 41-73. Math. Recursively enumerable sets of positive integers and their decision problems. where B is of order r and each R is general recursive. [12] H. 1962). 1952.) is hyperarithmetical if and only if it is general recursive in E (Kleene [lo]). 50 (1944). C. we have Vx 3yP(x. Studies.. Bull. [3] S.. .. Ann. On notation for ordinal numbers. As to tuniformization. Princeton Univ. and the theorem on complete form (Kleene [lo]). there is a primitive recursive predicate R(a’.

Mem. and G. Generalized recursion theory II. Symposia in Pure Math. Thus regular polygon sometimes means the convex cell bounded by a regular polygon as described above. Amer.. Descriptive set theory. 30 (1965).p. 1980. (When a polygon in a Euclidean plane bounds a tconvex cell. Amer.. 226-252. Hierarchies of numbertheoretic predicates. General recursion theory. Proc. 29 (1964) 161-162. these regular polyhedral angles are all congruent to each other. Kleene. Spector. Gandy. (2) Its vertices are all surrounded alike.. 357 (Vl. Springer. Intinitary logic and admissible sets. C. Sot. 1964. [23] J.. XIII (1971). Japan. and all faces of 3 are congruent to each other. Moschovakis. Hyperarithmetical quantifiers. C. B. Math. Primitive recursive set functions. 143-176. 34 (1969). it is called a regular n-gon. 46 (1958-l 959) 3377357. [20] J. J. Fund. Amer. 575-578. Symbolic Logic. Springer. Fenstad. [29] J. E. E. 1974. Sot. Kripke. J. 46 (19581959) 123-135. Spector. [21] J. Fund. E. Fenstad. 48 (1959-1960). 1980. North-Holland. W. G. [25] Y. Springer. Fenstad. Kleene..). N. J. Trans. That is. J. North-Holland. by the projection of 3 from each vertex of 3. The n vertices of a regular n-gon are obtained by dividing a circle into n equal parts. 0. North-Holland.. 79 (1955) 312-340. Elementary induction on abstract structures. N. 16 (1964). l-31. 1). Kleene. We call the center of these circles the center of the regular n-gon. Sot. 20 (1955). Hinman.179 Geometric Construction). Barwise. Machover. [3 l] S. Addison...2 . Addison. Generalized recursion theory. which is a 2-dimensional cell. Amer. 1978. J. is a regular polygon. Transfinite recursions on admissible ordinals. Math. Moschovakis. 14 (1959) 23-40. Springer. When the number of vertices (which equals the number of sides) is n. Arithmetical predicates and function quantifiers. Symbolic Logic.) 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. Some consequences of the axiom of constructibility. B. Math. Sacks (eds. Barwise. we obtain a regular polyhedral angle. Quantification of numbertheoretic functions. Admissible sets and structures. Computation in higher types. and G. Symbolic Logic. [32] S. Fig. Compositio Math.wherethepi(i=1. Math. On the partial recursive functions of ordinal numbers. E. Recursion theoretic hierarchies. Separation principles in the hierarchies of classical and effective descriptive set theory... 151-163. 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. [22] J. McGraw-Hill. we call it a regular polyhedron: (1) Each face of 3.(m>O). Regular Polyhedra Consider a regular polygon on a plane. 1975. Clarke. 0. 1967. Recursive well-orderings. A. 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.. [ 171 M. Moldestad.).. 1978. 1331 S. [ 151 R. A formalization of the theory of ordinal numbers. Takeuti. Fund. [27] P. E. Rogers. Gandy. 1 When a iconvex polyhedron 5 in E3 satisfies the following two conditions. [ 191 T. 295-317. [28] J. this 2-cell is sometimes called a convex polygon. Symbolic Logic.357 A Regular 1328 Polyhedra [ 13) H.. Tug&. Theory of recursive functions and effective computability. [14] J.. Sot. and take a point on the line perpendicular to the plane at the center of the polygon. North-Holland. Math. [26] J. W. Math. [35] C. R. 1977.. 67 (1961). From (2) we see that the number of edges emanating from each vertex of 3 is . Bull. Math. Hierarchies of predicates of finite types. [34] C. 1974. 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. 61 (1955) 193-213. C. [24] Y.) are different +Fermat numbers (.6) Regular Polyhedra A. [30] D. Bull. Jensen and C. Sacks (eds. R. 313-320. The theory of transfinite recursion. Math. [ 181 G. Sot. [ 161 S. Karp.

Fig. 6 Regular dodecahedron. 3).1329 357 c Regular Polyhedra independent of the vertex. octahedrons (Fig. let a be the length of an edge. 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. 6) (see also see Table 1). The octahedron and hexaTable 1. 5). Fig. 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’ ‘\ \ /’ ’ . The tetrahedron is dual to itself. C. It has been known since Plato’s time that there are only five kinds of regular polyhedra: tetrahedrons (Fig. 2). we have finite subgroups of O(3) called regular polyhedral groups (. 4). 8). 4 Regular icosakedron. I Fig. and dodecahedrons (Fig. we can obtain another one by taking as vertices the centers of all the faces of the given polyhedron (Fig. f3 the magnitude of the dihedral angle at each edge. 7).=tanptanq AR Fig. Corresponding to these polyhedra. 2 Regular tetrahedron. and R and Y the radii of circumscribed and inscribed spheres. respectively.’ . For a regular polyhedron 5. From a given regular polyhedron. 3 Regular octahedron. In a regular polyhedron. cubes or hexahedrons (Fig. we can obtain a regular polyhedron dual to the given one (Fig. 5 Regular hexahedron (or cube). Fig. BR . Regular Polyhedra in 3-Dimensional Number Vertices 4 6 12 8 20 of (see Table 2).15 1 Finite Groups). icosahedrons (Fig.[ @ ‘\ ‘\ \+--.I \ \ \ (1) r=%otEtan2 p 2. hedron are dual to each other. as are the icosahedron and dodecahedron. 8 Fig. We say that the given regular polyhedron and the one obtained in this way are dual to each other. 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 . Fig.

[S] H.6” 90” 109O28’16.70 Complexes).. y be two variables taking their values in X. Regular b: dual to each other Polyhedra in n-Dimensional Regular Polyhedron Euclidean in R” . b) is true or false for each pair (a. [4] H.3. ) (. we define its inverse Then R is the relation R -’ by yRm’xoxRy. i. .2” R/a G/4 a/2 1/a fi(ti 4 20 2/3 138” 11’22. Geometry and the imagination.2) Relations A. . M. 1932. (1) 1330 l/2 l/G lhGi77 2fi Table 3. n > 4. Springer. . Coxeter. to the case n = 2.8” 2fi 4ti +I> r/a ti/12 Values for Eqs. second edition. 358 (11. Numerical of sine 2ti/3 1 2ti/3 2/G e 7OO31’43. Dower. Y be two sets and x. Hilbert and S. y is called a relation or a binary relation if it can be determined whether R(a. y) is sometimes written as xRy.2. Anschaulithe Geometrie. 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 I 120 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. A relation R(x. Lecons de gkomttrie ClCmentaire II. A proposition R(x. but in this article we restrict ourselves to its most ordinary meaning. respectively.411 Symbolic Logic G). Regular Polyhedra Table Number Faces 4 6 8 12 2. y) containing x. Regular polytopes. 1934. S. General Remarks References [ 11 J. Let X. x divides y. Press. third edition.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. Vorlesungen iiber die Theorie der Polyeder. Cambridge Univ. if both X and Y are the set of rational integers. For example. Methuen. [3]E. b) in the Zartesian product X x Y. [2] D. Hadamard. 1952. 1973.e. then the following propositions are relations: x < y.4” 116O33’54. 1948. English translation. M.425-427. b: dual to each other regular polyhedra in E”. Steinitz and H.y is even. x . Chelsea. 1974. In its wider sense the term relation means inary relation (n = 1. 1906.’ Number n-hedron (2n . Coxeter. Springer. .357 Ref. When n = 4 we have 6 kinds of regular polyhedra (Table 3). For n > 5 we have only 3 kinds (Table 4) (. Cohn-Vossen. S. Regular complex polytopes. For a given relation R. Rademacher. Y. Armand Cohn. S.

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

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

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

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

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

it was the invention of printing in the 15th century. iiber Geschichte second edition. If the upper bound of the momentum is limited to a finite cutoff value A. G. E. and strongly urged the study of mathematics. Da Vinci’s contemporary. Stoner). Tartaglia (1506. Teubner. In 1494. [3] G. one of the first printed books on mathematics. especially in Italy. Cambridge Univ. A system of numeration was imported from Arabia to Europe in the 13th century. vols. References [l] M. scholastic theology and philosophy were at their height with the Summu theologiae of Thomas Aquinas (1225‘?-1274). was also born in the 16th century. L. This constitutes a famous episode in the history of mathematics. [2] J. Elements mathematiques. During the same period. 1977. Bourbaki. and F. Cardan (G. W. since the Greeks were able to solve equations only of degrees 1 and 2. life (trans. but in the latter half of the century. the German painter A.32) Renormalization A. McGraw-Hill. 1979. Group theory and general relativity. Galilei (called Galileo) (1564. left manuscripts in which he wrote about mechanics. Hawking and W. The book enjoyed wide popularity. insisted on the importance of experimental methods in science. N. In 1543. 1980. Opera. N. and L. and Florence. the German priest Nicolaus Cusanus (1401. Its content. Cardano (1501~ 1576). Held (ed.360 Ref. influenced by Arabian mathematics. 360 (Xx1.). the German scholar Regiomontanus (143661476) wrote the first systematic treatise on trigonometry independent of astronomy. in the domains of the arts and literature. and perspective. Copernicus (147331543) published his heliocentric theory (1543). Padua. Carmeli. Israel (eds. the convergence of infinite series. Cardano). for example. Tomonaga. General relativity. The book of my 1963. 1900. The Renaissance flourished first in Italy. Galileo studied at the University of Pisa and taught at the Universities of Pisa. General relativity and gravitation I. Renaissance Mathematics 1336 [S] M. an Einstein centenary survey. 1663. Low in order to overcome the difficulty of divergence in field theory. Copernicus studied at the Universities of Bologna. then physical quantities. d’histoire des 1960. practical mathematics (influenced by the Arabians) had become popular in Europe. 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.). Pacioli (1445?% 15 14) published Summa de arithmetica. 361 (Xx. However. J. II. but what is historically more important is the fact that essential progress beyond Greek mathematics was made by mathematicians of this period. Plenum. although publication of the method was against his will. the forms of the numerals became fixed. Diner (1471I1528). S. Gell-Mann. Newer ideas in mathematics and the natural sciences dominated the 17th century. Algebra was subsequently systematized by the French mathematician F. wrote a textbook on perspective. Ferrari (1522-I 565). [9] S. and with the development and acceptance of printing. the all-encompassing genius born in the same century. and some problems of quadrature.. Lyon. the English philosopher Roger Bacon (1214-1294) attacked Aquinian philosophy in his Opus majus. the translation of the Greek texts of Euclid and Archimedes into European languages. the indomitable proponent of this theory. then in other European countries in the 15th and 16th centuries. 10 [4] N. Cardano. the mass WI of an electron. by the time of S. to whom acknowledgment was made. J. The solution was due to Tartaglia.8) Renaissance Mathematics Toward the middle of the 13th century. Schwinger.1642). Viete (1540. Vorlesungen der Mathematik II. and the importation of Arabian science into Europe during the Renaissance that prepared for this development. and Ferrara. [ 101 A. includes practical arithmetic and doubleentry bookkeeping. In the 15th century.1557) G. geometric optics. Press. M. Stevin (1548?? 1620?) it took the definite form of a decimal system. Leonardo da Vinci (145221519). and more advanced mathematics began to be studied in European universities. Dover. By the end of the 15th century. Padua. Cardano published the solution of equations of the third degree in his book Ars magna (1545).1464) discussed infinity. Cantor.1603). Introduction Group The concept of renormalization was introduced by S. Hermann. can be obtained as finite quantities by letting .

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

& /. small-momentum parts inside each cell are integrated or eliminated. N. 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. The correlation length < diverges like t(T)-(TT.. [4] C. we have (i) momentum-space renormalization group theories [6&10] and (ii) real-space renormalization group theories [IO.. If the operator K thus defined has a negative eigenvalue ii.l)/(b . 29A (1969). Jona-Lasinio. Small distance behaviour in field theory and power counting. and consequently the remaining slowly fluctuating parts..P.. E. The renormalization group method can be applied to other many-body problems.. The fundamental idea is to eliminate some degrees of freedom. Renormalization group and critical phenomena. G.fV 1. one can estimate critical exponents [IO. and then to evaluate critical exponents from their asymptotic behavior near the fixed point.361 Ref. )-td’il. Nuovo Cimento. 1300-1312. The multiplicative renormalization group and the critical behavior in d = 4 -E dimensions.R.: i.. The transformation of X is expressed as dX/dl= G [X]. long-wave parts. [Z] M. In most cases. The first few terms are calculated explicitly for specific models. This transformation R. h.. Comm.) The critical exponent of the specific heat defined by C-t P is given by the formula c(= 2 -d/l. La normalisation des constantes dans la thkorie des quanta. . Math. h’~h.(K). I8 (1970). 1 I].. 49-86. Helv. Stueckelberg and A. Z. A generator G is defined by G = lim. 26 (1953). such as the cp4-model.~+~2~2+. In order to find critical exponents from the asymptotic behavior of G near Z*... 322-323.8* is the solution of G[X*] =O. are renormalized. s’?~=R..f(l . That is. = 1. The fixed point K* is given by the solution of K* =fb(K*). There are many different ways of carrying out this idea explicitly. h. [7] K. DiCastro and G. . and the physical quantity Q. Lett. 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.) is found to have the scaling property . such as the Kondo effect [ 121. C. Math.. Petermann. Phys. where d.... R. Phys..... The correlation exponent v defined by 5 -(K-K*)-’ is given by the Wilson formula v = log b/log A. Low.. . Phys.. 19. Rev. = R. This renormalization operation is written as R. Introduction to the theory of quantized fields. 111. Phys. G. References [1] E....h2. V. Rev. By the normalization we obtain the scaling law we have h”lh. Cell-Mann and F. First the momentum space or real space is divided into cells and rapidly fluctuating parts. Acta. )~b-~f(b”lh .. Callan. namely. where v is called the critical exponent of 5. [S] K. to find recursion formulas for interaction parameters... namely. Symanzik.~~~-.... D2 (1970)... = exp(lG). The fixed point A‘*=R. ).59..../dK). +O(R. . Roughly classifying these into two groups.. 499-520.... Phys.. Bogolyubov and D. The common fundamental structure of these renormalization group techniques is explained as follows. is constructed perturbationally.f‘(t. 95 (1954)./A. hj/t’+ .. Phys. Jr. 227-246. The free energy per unit volume f[X] =f(h. is the case where a single interaction parameter K is transformed into a new parameter K’ by K’=f. and critical exponents are usually calculated in power series of E-d.. Rev.% =X0* + CjhjQj. C. 1541-1547. Interscience. Other scaling exponents { qj} can be obtained via the formula ‘pj = ?. ..1).)-” near the critical point T. . particularly in studies of phase transitions and critical phenomena [6&10]. by introducing a field hj conjugate to the relevant operator Qj.=R. By applying the Bore1 sum method to these &-expansions.=.-d. This is called the c-expansion.e. K=<-‘. Renormalization is useful in evaluating these critical exponents systematically.... denotes the critical dimension. Shirkov. Wilson. On the microscopic foundation of scaling laws. corresponding to a positive eigenvalue lLj > 0 becomes relevant.. as ~=~o+~. . . Renormalization group and the Kadanoff scaling picture. = RbXO.. Renormalization 1338 Group By taking Q1 as the energy operator.. 5 (1972). has the (semi) group property R.... R. [6] C.. Quantum electrodynamics at small distances... we study the Hamiltonian . from the eigenvalues of K. A=(df... Lett. 23 (1971). Comm. the corresponding physical quantity Qi becomes irrelevant after repeating the renormalization procedure. hj . The simplest example of R.=R$F. Thus.. Small distance-behaviour analysis and Wilson expansions. I. e’ = h. we consider a Hamiltonian of the form 2 =X* + wQ and expand G[X] as G[X* + wQ] = wKQ + 0(w2)..= t.. G. Similarly. the correlation function C(R) for the distance R behaves like C(R)-Rm(dm2+q)~exp(-~R). Broken scale invariance in scalar field theory.. [3] N. Similarly we have X1 = R.. Phys. 69-74. -TT. DiCastro.

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

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

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

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

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

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

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

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

406-444. Ann. 25-46. Representations lineaires des groupes finis. Uber die Darstellung der symmetrischen und der alternierenden Gruppe durch gebrochene lineare Substitutionen. Springer. Math. Amer. Actualites Sci. . II. 1963. Gesammelte mathematische Werke und wissenschaftlicher Nachlass. L. vol. his contributions encompassed all aspects of mathematics. (eds. Bull. he defined the tRiemann integral and gave the conditions for convergence of trigonometric series. Representation theory of finite groups and associative algebras. Yamada. has been answered affirmatively for certain special cases such as tmeta-Abelian groups. 1967. 4 (1955). 1599227. Untersuchungen tiber die Darstellung der endlichen Gruppen durch gebrochene lineare Substitutionen. Lecture notes in math. [7] R. J. In his later years. F. 1935 (Chelsea. he considered the tRiemann zeta function as a function of a complex variable and stated tliiemann’s hypothesis concerning the distribution of its zeros. Reiner. formulated the concept of tRiemannian manifolds.1826-July 20. Benjamin. Springer. In his paper presented for the position of lecturer (1854). Riemann became interested in theoretical physics. 142. Bourbaki. Hanover. 1972. Wirtinger (eds. Springer. 1971.. [S] C. B. Reine Angew. The Schur subgroup of the Brauer group. In his paper of 1857 on tAbelian functions. G. Math Z. Lecture notes in math. Math. A. Weber.). Uber die Darstellung der endlithen Gruppen durch gebrochene lineare Substitutionen. Wiley.). Dekker. Reine Angew. he systematized the theory of tAbelian integrals and Abelian functions. It remains for modern mathematics to investigate whether this hypothesis is correct. Algebra II. and defined their curvature.-P. 1962. [ 121 R.1866) was born the son of a minister in Breselenz. 1261a. Okayama Univ. Teubner. 132 (1907). Math. tDirichlet as full professor. Ind. Nesbitt. Reiner. [9] I. van der Waerden. L. a considerable portion of the integral representation theory has been extended to more general orders in separable algebras [ 14161. In his inaugural lecture in the same year. 1. In 1862 he contracted tuberculosis. [ 191 T. Riemann.. Notes of blocks of group characters. Algbbre. Gruppen von linearen Transformationen. Georg Friedrich Bernhard phism G g H of groups. Roggenkamp (with V. M. Boerner. 139 (191 l). Georg Friedrich Bern hard Georg Friedrich Bernhard Riemann (September 17. 1953). [16] K. Lecture notes in math. Noether and W. [15] R. 1970. 115. he discussed the foundations of geometry. Group representation theory. (2) 42 (1941). W. [3] H. 155-250. 149. Dornhoff. Springer. Brauer. Schur. 1967. [2] B. second edition. 1892 (Nachtrage. In 1857 he rose to assistant professor. 397. [14] I. [S] M. 1958. 133-175. Interscience. He attended the universities of Giittingen and Berlin. 1955. A survey of integral representation theory. Math. 175188. Weber. [ 1 l] I. Brauer and C. References 363 (Xx1. Dedekind and H. 8. R. Springer.. Schur. K-theory of finite groups and orders. [lo] I. Curtis and I. Hermann. introduced n-dimensional manifolds.). 76 (1970). Springer. Swan. [ 131 W. Elements de mathtmatique.1347 363 Ref. L. 556-590. In his paper of 1858 on the distribution of prime numbers. fifth edition. [ 171 L. 85-137. Lectures on modern mathematics. J. He gave lectures on the uses of partial differential equations in physics that were edited and published by H. [4] N. T. 1902) (Dover. Characters of finite groups. Hermann. Darstellungen von Gruppen. His doctoral thesis (1851) stated the basic theorem on tconformal mapping and became the foundation for the geometric theory of functions. On the modular characters of groups. HuberDyson). 72 (1959-1960). second edition. References [l] G. Riemann.. 63 (1956). and in 1859 succeeded P. [IS] J. Weber. [6] R. and he died at age 40. 1948).. R[G] is an R-torder in K[G]. 1974.. Math. van der Waerden. Brauer. Serre. Representations of finite groups. J. Math. Saaty (ed. 127 (1904) 20-50. [l] B. influenced by W. Sot. 1972. Reine Angew. Osima. In 1851 he received his doctorate at the University of Gottingen and in 1854 became a lecturer there. L. Lattices over orders I. W. Schur.40) Riemann.. J. Zur Darstellungstheorie der Gruppen endlicher Ordnung I. Feit.. ch. II. 1970. Germany. Despite his short life. and in this context. B.

g). As in a Euclidean space. co) is dctincd so that the value d(p. q).. and M. Let (w’). Y).(p. and y be a +Riemannian metric of class c’-’ on M. We assume that M is connected and of class C”. Assume further that ‘p is a diffeomorphism and N has a Riemannian metric h. ti] such that the restrictions x 1[tieI.~~~ on B. B. Springer. a submanifold and a ‘covering manifold of M have Riemannian manifold structures induced by the natural mappings (. and d is the distance function of these spaces. The tconnection form of the Riemannian connection is expressed by n2 differential l-forms (~j).105 Differentiable Manifolds). is introduced on the ttangent vector space T. q) = d. is the i&mum of the lengths of curves of class D” joining p and q. it is called the fundamental tensor of M.(N) of T.. 1956). we can introduce various notions on T.~~. This connection is called the Riemannian connection (or Levi-Civita connection. . If a differentiable manifold M is the product manifold of Riemannian manifolds (M..$(q)). Using the value g.P of L to be the quantity g. and M. Utilizing the properties of the space E”.11= IILl\g. are projections from M to M.+. tJ are timmersions of class C”. and the concepts of tcanonical parameter and orientation of x can be defined (. In this way we get a one-to-one correspondence between the set of all O(n)-subbundles of F and the set of all Riemannian metrics of M.111 Differential Geometry of Curves and Surfaces). I(E”) is the tcongruent transformation group. gl) and (M2.80 Connections. Jahrhundert I. YE TP. If ‘p*g = h.(X. Riemannian Connections [2] F. to M at p. Klein. Then B is an O(n)-subbundle of F of class C”. p.($(p). and q is called an isometry. Then (~()r$~$. 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. p.(M) with respect to g. A necessary and sufficient condition for a mapping II/ : N--t M to be an isometry is that d.~~~” be the tcanonical l-forms on B. = L lldc&L)l\. then (M. X. I.. and we have wj+o{=O. where a. If there exists an immersion cp of a differentiable manifold N in a Riemannian manifold (M. The length ((x(( of such a curve x is defined to be [a” llx’(t)ll dt. y E M. A function d : M x M + [0. then a Riemannian metric ‘p*g is defined on N by the tpullback process ( 11 II. 4~ N. 2. Then (M. b] + M is called piecewise smooth or of class D” if x is continuous and there exists a partition of [a.~~ together with (Q~) give rise to an absolute parallelism on B (that is. they are linearly independent at .285 Non-Euclidean Geometry). A curve x : [a. +n:g2) is called the Riemannian product of M. (For a vector field X. 417 Tensor Calculus). h) is said to be isometric to (M. the covariant differential operator V. we define the length (IL.) The covariant differential Vg of the fundamental tensor g vanishes identically. The metric g is a fcovariant tensor field of order 2 and of class c’-‘. where x’(t) is the tangent vector of x defined for almost all values oft.) A necessary and sufficient condition for a differentiable manifold M of class c’ to have a Riemannian metric is that M be tparacompact. Let F be the ttangent n-frame bundle over M and B = B. the length (/x([ is independent of the choice of parameter t. The set I(M) of all isometries (isometric transformations) of M onto M is a group. (For example. A Euclidean space E” has a Riemannian metric expressed by C&dx’ Q dx’ in terms of an orthogonal coordinate system (xi). g).3) Riemannian A.(L. then ‘p*g is the Q’irst fundamental form of N...g2). Riemannian Manifolds Metrics Let M be a tdifferentiable manifold of class C (1 < r < o). 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.80 Connections K). 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. b] into finite subintervals [tiei.n:g. a differential form of degree 1 is identified with a tangent vector field. Let V denote the tcovariant differential operator defined by this connection (.) If M = E3 and N is a 2-dimensional submanifold of M. In particular.364A Riemannian 1348 Manifolds topology of M.365 Riemannian Submanifolds). acts on any tensor field T defined on a submanifold having X as a tangent vector field. c(= 1. a positive definite inner product g.. 1926 (Chelsea. called the tangent orthogonal n-frame bundle (or orthogonal frame bundle). The function d is a tdistance function on M.(M) be the subset of F consisting of all orthonormal frames with respect to g.g) or simply M is called a Riemannian manifold (or Riemannian space) of class c* (. (For example. Vorlesungen iiber die Entwicklung der Mathematik im 19. given a tangent vector LET.. then (N. of g at each point PE M.). There exists an essentially unique structure of a Riemannian manifold on (real or complex) telliptic or thyperbolic space (.

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

or negative.) of P.) If M is a +Kahler manifold and P is restricted to a complex plane (invariant under the almost complex structure). ) is the inner product in T. Wolf [l]. with respect to gp. curvature tensors are related to tcharacteristic classes. In particular. or hyperbolic space.(R. x(a)=p. then it is biholomorphic to the complex projective space [S. The compact spaces of positive constant curvature. or a real thyperbolic space according as K is 0. Schur).= --xkRbk. = n!/(2”n@(n/2)!). is a homomorphism from ap to H. . complete. R.). the integral of a. Xi) and R = c. and K. . If K. The curvature tensor R is uniquely determined by the function K.}. n P) and is called the sectional curvature (or Riemannian curvature) of P. simply connected. E. then M is called a space of constant curvature. and locally symmetric space is a tsymmetric Riemannian space. For example.2. is defined as follows: For a positive even number s.~. and R. called the restricted homogeneous holonomy group. then M is called a locally symmetric space (. . Holonomy Groups Let p be a fixed point of M. the Riemannian manifolds having the sphere as the universal covering Riemannian manifold.. ( . K. Related to algebraic geometry. .l)s/2/(2si2 s!). YE Tx(. as the solution of the Frankel conjecture. is the value of the tensor R at p. The properties of the sectional curvature and the Ricci curvature are closely related to the behavior of geodesics of Riemannian manifolds. K. is identified with E”) independent of the choice of p (.(R(Xi. . is spanned by {I. X. called the homogeneous holonomy group. Q(L) and R are expressed by Q(L) = Cig.80 Connections). If dim M > 3 and if at every point p of M.).) (T. b]-M is of class D”. of a compact and orientable M is constant for a certain s. and these properties reflect those of the topological structures of the manifolds (. the sphere (which is the universal covering Riemannian manifold of a real ielliptic space).(P) is constant. The restriction Ho of this homomorphism to all closed curves homotopic to zero is called the restricted holonomy group. = K. Y)Z with respect to the basis b of T.. then M is called an Einstein space. In a local sense.(X. . {j. then K(P) is called the holomorphic sectional curvature. in terms of an orthonormal basis (Xi) of T. . is a constant independent of the choice of p (F. is a real-valued function of the sdimensional subspaces P of the tangent vector spaces Tp of M. is a connected component of h and a tcompact Lie group.. The rotation part h of H. Xj)Xk is as already defined at the beginning of this section.178 Geodesics).. is given by C R&ZjXkY’.. Y) is the endomorphism of the linear space TxcbJ defined in Section D.) containing L and is called the Ricci curvature (or mean curvature) of the direction L at p.j.). Riemannian metrics of these spaces are uniquely determined by the curvature tensor R up to a constant factor. where b. were completely classlied by J. = ( . If the Ricci tensor of M is a scalar multiple of the fundamental tensor..364 E Riemannian 1350 Manifolds classified [224]. i. 61 (. of Ho. Furthermore. be the set of all closed oriented curves of class D” with initial and final points p and with parameters neglected.. this scalar is constant.. . i. Y))l x: [a. then M is isometric to E”.( V. If VR = 0.} c { 1. 413 Symmetric Spaces).}.. independent of the choice of P. up to positive constant factors. The rotation part h.Kc.. positive. and simply connected.Q(X. called the holonomy group of M. The tLie algebra of h. A Klhler manifold M of constant holomorphic sectional curvature is locally isometric to a complex Euclidean space. is the mean of K... which is given by R(X... The compact simply connected homogeneous Riemannian manifolds of strictly positive sectional curvature have been in terms of an orthonormal basis (Xi.. is the sign of (ii. and x*1. The set H = {Ix 1xcn. A complete..232 Kahler Manifolds). and let 0. given by (R.. j. n}.. C is summation over all pairs of s-tuples satisfying {ii.(P) for all sections P (2-dimensional subspaces of T.. If M is of constant curvature K.. o is the volume element of M.(X. Let Q be the quadratic form on T. L)L.. E~. The mean R of Q(L) for all the unit vectors L at p is called the scalar curvature at p (. The Ricci tensor (Rij) is defined by R. is a subgroup of I(T.+o on M is equal to the tEuler-Poincare characteristic. then the kth tpontryagin class of M (with real coefficients) vanishes for all k > s/2. K.. then M. Kc. we have the Gauss-Bonnet formula: If M is an evendimensional compact and oriented Riemannian manifold...(P) is the +Gaussian curvature of the surface Exp. A. where u.(P) has a constant value M.417 Tensor Calculus).(P) of p and P. elliptic space. . and X. that is. (When dim M > 3.412 Symmetric Riemannian Spaces and Real Forms. Then the value Q(L) for a unit vector LE T. If K..(X. where R. is a subgroup of the orthogonal group O(n) of T. the following holds: If a compact Kahler manifold has strictly positive sectional curvature.

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

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

A torus T” cannot carry a metric of positive scalar curvature. there exists a strictly positive function u such that the Riemannian metric B = uq(n-2)g has constant scalar curvature. Marsden. g) of dimension n > 3. g2 with Kg1 < 0 and Rgl 2 0. The problem of finding a Riemannian metric g which realizes a given Ricci solved for a wider class: namely. 3. denote the sectional curvature and the scalar curvature. on a compact manifold (n > 3) there always exists a Riemannian metric with constant negative scalar curvature [18]. < 0. and R. n-2 I.183 Global Analysis).2) on a Riemannian manifold. pointed out that his original proof contains a gap in some cases. E. The existence of such a manifold has been shown. any metric of nonnegative scalar curvature on T” must be flat [22]. Blair. D. S. The Ricci tensor (Rij) is a symmetric tensor field of type (0. Then the following are known for a compact manifold M of dimension > 3: If M carries a metric g with K. Furthermore. simply connected manifold M of dimension n > 5 is not a spin manifold. which is a function on M. N. it has since been In this paragraph the manifolds under consideration are assumed to be of dimension n 2 3. The problem reduces to the following nonlinear partial differential equation on a compact manifold M: &p+m-2) =4n-l*u+Ru. then there exists a Riemannian metric of positive scalar curvature. if M is a spin manifold and spin tcobordant to M’ with positive scalar curvature. The Bianchi identity (. If M carries a metric g with K. If a compact. Yamabe [ 191 announced that on every compact Riemannian manifold (M. Kobayashi. of a Riemannian metric g. there is a topological obstruction. the sectional curvature. then M carries a Riemannian metric of positive scalar curvature [22]. J. On a sphere. Fischer and J. In particular. On the other hand. the Ricci curvature. A compact tspin structure (spin manifold) having nonvanishing ta-genus cannot carry a Riemannian metric of positive scalar curvature. Maeda). In fact. then it carries no metric with R > 0. starting with a Riemannian metric with constant Gaussian curvature. 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. Y. For positive scalar curvature. however. then K is negative somewhere. If M carries metrics gi. and the scalar curvature all coincide with the tGaussian curvature. Scalar Curvature On a 2-dimensional Riemannian manifold M. then locally it is almost always surjective when M is compact (A.417 . if x(M) < 0. Any smooth function can be the scalar curvature if and only if M admits a metric of constant positive scalar curvature. then it carries no metric with R > 0. Furthermore. Tanno. then K changes sign unless it is identically zero. E. 0. if M is not conformally flat and n > 6. ~0. 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. if x(M) = 0. Lafontaine). then K is positive somewhere. tensor reduces to the one of solving a system of nonlinear second-order partial differential equations for g. then both metrics are flat [22]. Nevertheless. If M is compact. 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. a Riemannian metric which is conformal to the standard metric and has the same scalar curvature as the standard one is always standard. E. This sign condition is also sufficient for a given function K to be the Gaussian curvature of some metric on M. 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. or if it is conformally flat and its fundamental group is finite. H. there is a system of differential equations characterizing the complex projective space or the quaternion projective space as a Klhler manifold (Obata. Let K. More precisely. 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 (. Trudinger. As the Kahler or quaternion Kahler version of (E2). 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 >. respectively.1353 3641 Riemannian Manifolds has been solved affirmatively transformation and that of (E. H. it has a positive constant sectional curvature [ 143. namely. then the problem c201.) is an inlinitesimal projective transformation. Yamabe’s original proof can be pushed to cover a large class of metrics with jM RdM < 0.

J.178 Geodesics). 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]. Les vari&tCs riemanniennes homogknes simplement connexes g courbure strictment positive. (2) 96 (1972). Transformation groups in differential geometry.. 18 (1972). A. 12 (1960). [ 161 A. [S] M. 14 (1974). Yamabe.2)tensor on R” which cannot be the Ricci tensor for any Riemannian metric in a neighborhood of OER”. Projective manifolds with ample tangent bundles. de Rham. Advances in Math. Berger.. 423-434. Sot. (2) 110 (1979). Colloq. Berard-Bergery. Erg. [3] N. Sur les groupes d’holonomie homogine des varittks B connexion afIine et des variCt&s riemanniennes. 1958. 1957. Existence and conformal deformation of metrics with prescribed Gaussian and scalar curvature. [22] M. 279-330.X. The positivity of the Ricci curvature on a Riemannian manifold puts rather strong restrictions on the topology of the manifold (. Ann. [18] J. [20] T. 12 (1979). J. Transformations conformes et quasi conformes des varittt&s riemanniennes compactes (dCmonstration de la conjecture de Lichnerowicz). 59 (1980). Scuola Norm. 277-295. Obata. However. 593-606. 150 (1970). (2) 111 (1980). Math. North-Holland. [4] L. Comment. Hilbert).1 denotes the space of Riemannian metrics on M with total volume 1. In particular. (2) 76 (1962). Groups of conformal transformations of Riemann spaces. 269-296.29]. Spaces of constant curvature. equations diffkrentielles CaractCristiques de la sphkre. References [l] J. If. Riemannian 1354 Manifolds dimension impaire g courbure strictment positive. J. MCm. Inventiones Math. Ecole Norm. Compact Riemannian manifolds with essential group of conformorphisms. T.. Acad. The theory of Lie derivatives and its applications. [ 131 A. 26 (1952). Ann. 179-246. [9] J. then g is called an Einstein metric on the manifold M. Publish or Perish. B. Equations diffkrentielles non lintaires et problkme de Yamabe concernant la courbure scalaire. The structure Tensor Calculus) must be satisfied. Ann. J. Lichnerowicz. Hitchin.. Math. Simons. then they are isometric [30]. if two simply connected Einstein spaces have neighborhoods on which metrics are isometric. 213-234. 1972.364 Ref. On a deformation of Riemannian structures on compact manifolds. Alekseeviskii.. Sur la reductibilitt d’un espace de Riemann. Ann. then there exists a complete metric on M with positive Ricci curvature [25&271. Lawson. Lelong-Ferrand. The conjectures on conformal transformations of Riemannian manifolds. Ledger and M. (2) 101 (1975). [lo] S. Sci. Pures Appl. L. 645-651. There is a symmetric (0. Trans. Les variCtCs riemanniennes homog&nes simplement connexes de . on a complete Riemannian manifold M with nonnegative Ricci curvature. Math. Ann. [23] R. Sup. SC. Pure Appl.2) is invertible at a point p. [ 111 K. Wolf. The critical points of $? are Einstein metrics (D. Berger. 6 (1971). The classification of simply connected manifold of positive scalar curvature. [ 121 J. Math. USSR-Sb. J. is a functional on . Math..~I. then the tnullity and +coindex at the critical point are finite [28. Let uy denote the volume element determined by g. However. . S4h+3 (k > 1) carries an Einstein metric that is not standard [31]. [S] S. Dunod. Compact KLhler manifolds of positive bisectional curvature. [17] S. 39. Math. y) is an Einstein space. Sot. if a C” (or Cw) symmetric tensor field (Rij) of type (0. The integral of the scalar curvature y(g) = sM R. Amer. Sup. Bull. T.. (9) 55 (1976). Warner. Though S” with standard Riemannian metric is a typical example of an Einstein space. [ 191 H. Kazdan and F. Gallot. T.. 235-267. GkomCtrie des groupes de transformations. i Osaka Math. Math. (3) 15 (1961). V. 83 (1955). Math. If a Riemannian manifold (M. l-44. An Einstein metric is always real analytic in some coordinate system. 70. Gromov and L. W. Let &Z. Wallach. [Zl] N. 1977. 21-37... . nonnegative Ricci curvature and positive Ricci curvature are not too far from each other. Roy.. [7] G.. [14] M. Differential Geometry. On the transitivity of holonomy systems. l-55. 5 (1971). Ann.. Cl. 247-258. R. Math. Helv. Yau. Yano. Harmonic spinors.. Math. there is a point at which the Ricci curvature is positive.. 55 (1976). 317-331. Mori. Kobayashi. [ 151 D. Math. 328-344. no. When M is compact. 189-204. ( c A) denote the space of metrics with constant scalar curvature. Schoen and S. Then if ?? is restricted to &‘. 285-301. Math. [2] M. [6] Y. Siu and S. France. Obata. 47-67. Ann. Compact homogeneous Riemannian manifolds with strictly positive curvature. Pisa.. Belg. Aubin. Yau.

[27] J. 1981. Metriques riemanniennes et courbure. 18 (1979). DeTurck and J. E. 365 Riemannian Submanifolds. 4 (1970) 383-424. < 0.1. On Einstein metrics. Y)=VxY-V. Put 0(X. and O(n) x O(p). Amer.(M) of orthonormal normal frames of M. Osaka J.(M). 5). 838. Koiso. Let V’ denote the covariant differentiation with respect to the normal con- . 14 (1981). and fi is complete and simply connected [2]. 249260. Differential Geometry. In this article.. 178 Geodesics. respectively. g)+(fi. [31] J. Nash (1956)). The vector bundles associated with F(M). Bourguignon. Introduction If an timmersion (or an tembedding) f of a tRiemannian manifold (M. Q) be an isometric immersion. Ann. g) into a Riemannian manifold (fi. The generalization to the C” case is an open question even when the ambient space is Euclidean. f(M) will be identified with M except where there is danger of confusion. Sci. the ttan- C. M is compact. Then the tbundle F(M) of orthonormal tangent frames of M. Then M cannot be isometrically immersed into fi in the following cases: (1) p<n-2 and Knr<Ka (T. 4133421. Kuiper (1952).2) tensor field on M. Manuscripta Math. F. Ziller. 365 (VII. Y). Existence of metrics with prescribed Ricci curvature: Local theory. 5 (1976). Ehrlich. Y)= Q(cr(X. O’Neill(l960)). 109 Differential Geometry. then f is called an isometric immersion (or embedding) and M is called a Riemannian submanifold of ii?f. KG is constant ( < 0). Let V and V denote the tcovariant differentiations with respect to the tRiemannian connections of M and fi. and iii an (n + p)-dimensional Riemannian manifold with sectional curvature Ka. B. The connection on v(M) induced from the Riemannian connection of fi is called the normal connection of M (or off). 16 (1979).1. respectively.1355 365 C Riemannian Submanifolds v(M). [25] T. DeTurck. [29] N. An n-dimensional compact C’ Riemannian manifold (3 < r < co) can be isometrically embedded into an (n(3n + 11)/2)-dimensional Euclidean space (J. < n . The eigenvalues of A. Greene (1970)).1 3) Riemannian Submanifolds A. respectively. Math. the tnormal bundle and their Whitney sum T(M) @ v(M). 179-207. Ecole Norm. Tnventiones Math. Kazdan. Sot. Geometriae Dedicata. 111 Differential Geometry of Curves and Surfaces. E. [24] D. M. Tompkins (1939). Springer. 191 GStructures.. 521-530. 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. J. (2) P and fi is complete and simply connected (C. Math. [26] P. < KG < 0..Y. P. These are subbundles of the restriction to M of the bundle F(fi) of orthonormal frames of fi. and their twhitney sum F(M) @ F. Chern and N. which is called the second fundamental form of M (or off). Muto. Sup. Ricci curvature and Einstein metrics. Aubin.. Suppose dim M = n and dim fi = n +p. Then A. 65 (1981). H. See also references to 80 Connections. [30] D. of manifolds with positive scalar curvature. Metric deformations of curvature I: local convex deformations. 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). 159-183. g) satisfies the condition f*g = g. K. 28 (1979). On the second derivative of the total scalar curvature. and F(M) @ F. the tangential component of vx Y is equal to Vx Y. S. (3) p < n . 105 Differentiable Manifolds. Lecture notes in math. l-23. E.42-63. (1) Then c is a v(M)-valued symmetric (0. S. Differential Geometry.(M) are. Naturally reductive metrics and Einstein metrics on compact Lie groups. O(p). For vector fields X and Y on M. 417 Tensor Calculus.(M) are tprincipal fiber bundles over M with tstructure groups O(n). Let M be an n-dimensional Riemannian manifold with tsectional curvature K. B. Cartan (1927)). Mem. [28] Y. D’Atri and W.. gent bundle T(M).(M). M is compact. L. R. J. Fundamental Equations Let f: (M. Otsuki (1954)). are called the principal curvatures in the direction of 5. For a normal vector 5 at XE M. which is called the second fundamental form in the direction of 5.X. Janet (1926). K. the bundle F. put g(A. F. defines a symmetric linear transformation on T. Some regularity theorems in Riemannian geometry. 9 (1974).

(1) is called the Gauss formula.~)(Y. Basic Notions Let M be a Riemannian submanifold of iii. let (M.z)-(v. with respect to (e.).X (4) D. If they satisfy (3. and V’. Formulas (l)-(4) are the fundamental equations for the isometric immersion . define A.+. Let R.I. If every point of M is an isotropic point.f: M+fi. A. x E M is an umbilical point if and only if A.‘<).< (2) holds. for vector fields t and q normal to M. V. Let (G% QA. Let C@jj. If every point of M is an umbilical point.BCn+p be the iconnection form and the tcurvature form of fi with respect to (eJ. sa. and Ricci are given respectively by -dCA<>A. As a particular case..2) tensor field g on M. a(e.(M).. A point XE M is called an isotropic point if Ila(X.= -A<X+V. A4 = dim nSEY..). (u$). Y)Z+ Aoo.). For a icross section 5 of v(M). A point XE M is called an umbilical point if rr = g 0 h at x.W--Y(X..). ei) is independent of the choice of an orthonormal basis (ei). the relation Moreover.). Z tangent to M. by q(A. Codazzi. It is clear that an umbilical point is an isotropic point. Y)~.+I $n. (3.365 D Riemannian 1356 Submanifolds n+p. Then the equations of Gauss.275 Minimal Submanifolds). and (2) is called the Weingarten formula. and R’ be the icurvature tensors of V.)” (4. suppose fi is a ispace form of constant curvature c. Similarly. is called the index of relative nullity at XE M.(M). Y)Z=CMY.BQn+p and (&% QA.jgn and (@..) and (c$). and put wi=. Then the tintegrability condition for (1) and (2) implies d(X. (3) is called the equation of Gauss and Codazzi. +(v~~)(y.(.. is proportional to the identity transformation for all <E v. ej) =C wT(ej)e. ker A. Then (h$) is the matrix representing the symmetric linear transformation A.Z)=o.Z)Y) (V.j.. that is to say.cM. 4i. such an immersion is unique up to an tisometry of M”+“(c). where ( ... Then the equations of Gauss.~)(X. that is. and let (w”) be its dual. then M is called a totally geodesic submanifold of fi. i?. Y)Z= R(X.) Q(R%‘.Aoo. Codazzi. the tangential (resp. 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. Moreover.Z)-(V..f*Gg.. and (4. respectively.~)(x.z) (2’) for vector fields X. (1’) nection. Y)+A. where V’ denotes covariant differentiation with respect to the connection in T(M) @ v(M). which is called the equation of Ricci. A point x E M is called a geodesic point if cr = 0 at x. 0. ) is the fiber metric of v(M)..X (3) Put wr = C !I@‘. If every point of M is a geodesic point.f*o’= 0 for n + 1 <a < . 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. then M is called a totally umbilical submanifold of fi. h is called the mean curvature vector and 11 11is called the mean h curvature. Conversely. g) be an n-dimensional simply connected Riemannian manifold. Then .Z) A O(X.X)ii/llXl12 does not depend on XE T. More precisely. Then (wj). respectively.(n is the connection form of M with respect to (ei)isisn. M is called a minimal submanifold of fi if b = 0 (. Y).).p<n+p be the curvature forms of (w. (3. then M is called an isotropic submanifold of a.. Y)= (a(X. Y).~)=Y(CA<>~JX. <i. and Ricci reduce respectively to R(X.(M). Let (e. (3. M is a totally geodesic submanifold of fi if and only if every geodesic of M is a geodesic of IGJ.X.lX> Y) holds. V. Y. Y. 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. a(q. A mapping h: M+v(Mj defined by x* ic:=. normal) component of vxt is equal to -A:X (resp. For a tangent vector field X and a normal vector field < on M. the relation vx.pgn+p is the connection form of the normal connection with respect to (eJn+lgoGn+p. thatis.) 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.%ei = 1 htej. (~9)~ Qign<aGn+p gives the second fundamental form.

If E> 0. if E=O. Totally Geodesic and Totally Submanifolds Umbilical A totally geodesic submanifold of a space form is also a space form of the same curvature. c = 0. For each positive integer s. A. 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. with respect to any volume-preserving variation of a domain D in a Euclidean space. A generalization of (2) for the case of higher codimension was obtained by C. Totally geodesic submanifolds of compact tsymmetric spaces of rank 1 were completely classified by J. and the index of relative nullity is < n . then every isometry of M can be extended to an isometry of A. 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 others. Sacksteder. and M is compact and of positive curvature (S. M. then f(M) is locally a hypersphere. Minimal Submanifolds For general properties of minimal submanifolds . Wolf (1963). Hopf (1951). or a parallel hypersurface of a totally geodesic hypersurface [2]. and the tgenus of M is zero (H. Various rigidity conditions have been studied by S. an n-dimensional n sphere of curvature can be minimals(s+n-1) ly immersed into a (2s + n . . 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. Dolbeault-Lemoine. a sphere has plenty of compact minimal submanifolds. Among all n-dimensional compact minimal submanifolds of an (n + p)-dimensional unit sphere. > n. (4) n > 4. Ejiri (1979)). G. (2) f is an embedding (A. .1357 365 H Riemannian Submanifolds E. Cohn-Vossen (1929)).2 (N. In particular. Rigidity An isometric immersion f: M+fi is said to be rigid if it is unique up to an isometry of !i?. Tanaka.~n~~~! . the mean curvature of M = aD is constant if and only if the volume of M is critical. 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. if E < 0. (3) n 2 5. a horosphere.f(M) is locally a geodesic sphere.) = 0 such that &p/dki > 0. H. .l}- dimensional unit sphere and the immeision ’ is rigid if n = 2 or s < 3 (E. D. Chen and T. Wallach (1971)). . Beez (1876). and f(M) is contained in a certain (n + l)dimensional totally geodesic submanifold M”‘l(?) of M”+P(i?). > k.1) -n (J. Alexandrov (1958). (2) Ricci curvature (3) scalar curvature Simons (1968)). do Carmo and N. or an open hemisphere according as c = 0. then f(M) is locally a hyperplane or a hypersphere. that is.” Unlike an open hemisphere. 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 . Smyth and K.[S]). < 0. Then M is a space form M”(c) with c > E. and totally geodesic submanifolds of symmetric spaces of rank 2 2 were studied by Wolf and B. These results remain true even if the assumption “the mean curvature is constant” is replaced by the weaker condition “the principal curvatures k. Calabi (1967). if f’:M-+fi is another isometric immersion. > . . M is complete. 2-l/P > n(n . Submanifolds of Constant Mean Curvature F.1)~) (C. The answer is affirmative in the following cases: (1) dim M = 2. Y.[S]). R. Chern (1955)). and the tscalar curvature of M is constant (# n(n . . (2) The index of relative nullity is < n . 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. c > 0. Kaneda and N. then there exists an isometry rp of iii such that f’ = cpof: If f: M-+ti is rigid.2 at each point (D. Nagano [4]. Nomizu (1969)). a hyperbolic space. On the contrary.3 at each point (R. Allendoerfer’ (1939). Harle (1971)). Itoh 2(n + 1) WW). . then . 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.275 Minimal Submanifolds. satisfy a relation dk 1.l)(“. is M a sphere?” has not yet been completely solved. (1960)). E. c # 0. or > 0. where M”+l(c) denotes a Euclidean space. Ferus (1970)). k.

(C). Pogorelov (1956). Takeuchi obtained several results for isoparametric hypersurfaces of Sn+i with four or six distinct principal curvatures [7].[Z]). M”(O). c=O<?. 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). p= 1. S. (1) If n=2. If c = 0. Volkovand S. KIhler Submanifolds J. that is. Sk x H’-k. ?>c>O. Hoffman (1973)).+(Ii?. A Kahler manifold of constant tholomorphic sectional curvature is called a complex space form. is complete. then M is locally E” or Sk x Hnmk (Cartan). (2) If c > 0. and H. Sasaki. M’(C) is complete. and M”+l(O) is complete and simply connected. . but not true in general (Ozeki and Takeuchi [7]). c < 0.[2]). I.4. Miinzner (1980)). (4) If n = 2. Hilbert (1901). R. or D. 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). and others). Z) denotes the +homology class represented by a Kahler submanifold M of A. Y. there exists no submanifold M’ of fi such that M = 8M’. Vladimirova. K. Hartman and L. and M3(t) is complete and simply connected. If [M] E H.f does not exist (D.. g) into a Kahler manifold (A. 3.365 I Riemannian 1358 Submanifolds then f is cylindrical (A. (2) If II = 2. then . (5) If n > 3. then f does not exist (J.[7]). Nagano (1960) Takahashi C91). or 6 (H. - cm (3) If n=2. Takagi. (7) Ifpdn-1. Wirtinger (1936)). Klotz and R. and c > ?. then there exist infinitely many f (L. c = 0 > ?. c=F>O. Takahashi. and M is called a Kgihler submanifold of iii. KChler immersions of complex space forms .121). M”(c) is complete. Ferus (1975)). c>O. A tcomplex submanifold of a +Ktihle’r manifold is a Kahler manifold with respect to the induced Riemannian metric. A Kahler submanifold is a minimal submanifold. and M”+P(F) is complete and simply connected. c>F. Isoparametric hypersurfaces of S”+’ having at most three distinct principal curvatures were completely classified by Cartan. Hsiang. A complex analytic and isometric immersion of a Kahler manifold (M. then vol(M)<vol(M’) holds for any submanifold M’E [M] with equality if and only if M’ is a Kahler submanifold (W. P. M2(0) is complete. Nirenberg (1959). An n-dimensional complete and simply connected complex space form is either P. 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). p = 1. the number of distinct principal curvatures of M is 1. M’(0) is complete. Every Kghler submanifold of a complex space form is rigid (Calabi [ 10)). (6) If p = 1. and M3(?) is complete and simply connected. then f(M2(0)) is either a horosphere or a set of points at a fixed distance from a geodesic (J. cylinder if c < 0 (T. or if c > 0 and M has at most three distinct principal curvatures (Cartan). then M is isoparametric. c < ?. If c > 0. (8) If p<n. then M is locally Sk x E”mk. c = ?= 0. and M3(n is complete and simply connected. T. The converse is true if c < 0. Bianchi (1896). . L. p = 1. . (3) If c ~0. M2(c) is complete. 1. then M is isometric to EZ or else is given as an orbit of a subgroup of the isometry group of M”+‘(c) (W. Takagi. then f is totally geodesic (D. then . 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. Osserman (196661967) D.2. A compact KChler submanifold M of a Kahler manifold @ can never be homologous to 0. J. H. Liebmann (1901). and if c < 0. Lawson. Moore (1972)). y”) is called a Kiibler immersion. then . .f is totally umbilical.1. Cn. A4 has at most two distinct principal curvatures (Cartan). and M3(c?) is complete and simply connected. then M is isometric to E”. and both M”(c)and M”+P(?) are complete. p = 1. then M is isometric to Sk x E’mk (S. B. or a 3-dimensional group manifold with the metric ds2 =em2rdx2 +e2’dyZ +dt2 (Takahashi (1971)). 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. (1) If c= 0. If a subgroup of the isometry group of M”+l(c) acts transitively on M. p= 1. Kobayashi (1958). If c GO. Ozeki and M.f is totally umbilical (H.

Nomizu W’W. K.. (4) embedded and scalar curvature > n* (J. If 7(f) < 3.1359 365 0 Riemannian Submanifolds into complex space forms were completely determined by Calabi [lo] and by H. The index of relative nullity p(x) of an ndimensional complete KHhler submanifold M of P. C” (resp. The result remains true even if “Einstein” is replaced by “parallel Ricci tensor” (Takahashi (1967)).(C) into Pntn+3j. Fenchel(1929).. 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. Kuiper (1958)).. S’ x S’ in P2(C) and an immersion P. K. S. Ros and L.(R) in P. 1958). Verstraelen (1984)). Fary (1949). tExotic spheres do not have minimum immersions (Kuiper (1958)). respectively. Smyth (1967). Tsukada. Tanno (1973). N. and others. a totally umbilical immersion or a minimal immersion of a compact symmetric space of rank 1 by harmonic functions of degree 2. generally. For example. T.(C) satisfies Min. Abe (1973)). Kghler immersions of homogeneous Klhler manifolds into P.(M) points of ldetA<lo. Y. Nakagawa and K. let f: M + M”(c) be an isometric immersion of M into a complete and simply connected space form M”(c). and others. These results generalize theorems for space curves by W. then iqf$f)=/?(M)>2 An isometric immersion of a Riemannian manifold (M. Takeuchi. Q~={[zil~P~+~(C)ICzf=O} inP. then M is homeomorphic to S” (Chern and Lashof (1958)).. Milnor (1950). g) into a KBhler manifold (a.(C) have been studied by K. Let o and R be the tvolume elements of v. Takeuchi.2(C)) W. A totally geodesic submanifold P.. the last case occurs only when c > 0 (S. the total curvature of the immersion f is defined as If?Ql 1 voU~“-‘) s v. J. tR- .(C) were determined by Takeuchi (1978). S. Chern (1967)).. S. Ogiue (1976).(C) that is a complete intersection (J. L.(C).p(x)=O or 2n (K. Total Curvature Let J’: M + E” be an isometric immersion of an n-dimensional compact Riemannian manifold M into a Euclidean space. Besides linear subspaces. Naito. Lashof (1957.(M) at each point x E M is called a totally real immersion. I.. and M is called a totally real submanifold of iii.(C) give typical +P n(n+Z)(C) defined by [zi]+[zizj] examples of totally real submanifolds. Ogiue. 0. Sakamot0 [14]). More generally. H. 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)).(C) is 1. S. Integral theorems and pinching problems with respect to various curvatures for compact KBhler submanifolds of P. Q” is the only Einstein-KHhler submanifold of P. Yau (1975)... then f is either a totally geodesic immersion.J as a Klhler submanifold (Nakagawa and Takagi [ 1 I]).. More holds (Chern and R. (2) sectional curvature > l/8 (A. If the image of each geodesic of M is contained in a 2dimensional totally geodesic submanifold of M”(c). and others [ 121.(C)is the only Einstein-Kghler hypersurface of a complex space form that is not totally geodesic (B. Takagi. Submanifolds with Planar Geodesics A surface in E3 whose geodesics are all plane curves is (a part of) a plane or sphere.+. As a generalization of the ttotal curvature for a space curve. (3) Ricci curvature >n/2 [12].(C) were precisely studied by Nakagawa..+. M. An isometric immersion f which attains inf(f) is called a minimum immersion or a tight immersion. Submanifolds with V’a = 0 in symmetric spaces have been studied by Ferus. Ros (1985)). H.(M) c v. J. Cheng (198 1)). J. Then for each 5 E vi(M). D.g) satisfying JT. Hano (1975)). if the holomorphic sectional curvature of P.) is the only THermitian symmetric space that can be immersed in Cm (resp. Submanifolds with the above property are closely related to isotropic submanifolds with V’o = 0. (M) and P-l. and let fi : v1 (M)+Smel be the parallel translation. f?fi = (det A@ holds. Sakane. 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. Totally Real Submanifolds If /l(M) is the least number of critical a tMorse function on M. and others. Hong (1973). Little (1976). Let vi(M) be the unit normal bundle. 0. S”-’ the unit sphere centered at 0 E Em.. and KHhler immersions of Hermitian symmetric spaces into P.

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

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

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

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

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

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

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

have remarkable properties [9]. ones in 0.2)in terms of local parameter z at p. . It is known too that open Riemann surfaces are determined by their rings of holomorphic functions. the conjugate differential *co= -vdx+udy of w can be defined.. The differential form w is a holomorphic differential if and only if it is closed (i.. stands for the space {w I *w E TX} and I.. let r’(T. A pure differential is expressed as w =fdz. 1w is tsemiexact}. However. given a point p on R. Virtanen.. Differential Forms on Riemann Surfaces Since Riemann surfaces are considered real 2dimensional differentiable manifolds of class C”. respectively. Ahlfors). Then s = (cl c..) were treated. i *rhm... Iss’sa obtained a noteworthy result which established that open Riemann surfaces are also determined by their meromorphic function fields [24]. where the equality holds if and only if R belongs to O. Nehari. I. then o is called a meromorphic (or Abelian) differential.. The structures of closed Riemann surfaces are determined by the algebraic structures of meromorphic function fields on them. In the case of finite genus g. Since coordinate transformations satisfy the Cauchy-Riemann differential equation. and one with * o = ..iw is said to be pure. The differential o is called exact (or total) if o is written as dF = F. later a more general treatment was made possible by the discovery of the notion of semiexact differentials (K. and if f is a meromorphic function. A differential is said to be semiexact if its period along every dividing cycle vanishes. It is known that s (= Ild(Fi . in contrast to differentiable manifolds.) such that (i) Fi (F. SupposethatFi=l/z+ai+ciz+.) be the Hilbert space of analytic (harmonic) differentials with finite norm. then we have the orthogonal decompositions r. 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.. and set F. = {o E I. have a simple pole at p with residue 1. if f is a holomorphic function.. 1w is exact}.. Mergelyan’s theorem): Suppose that K is a compact set on R such that R -K has no relatively compact connected component.)/2 I(*) > 0. dx + Fy dy with a globally single-valued function F. only those Riemann surfaces with small boundaries (i.. A simply connected surface is planar. and I.(F. 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. Here. the set of all measurable differentials w with IIwllz=jsR wr\*o<coformsatHilbert space with respect to the norm IIw//... More precisely.. etc. L.. Kusunoki. Both spaces 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. 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 (. (iii) the total area of the slits and points is zero.F.e. Next. are known as the space of analytic Schottky differentials and the space of differentials of harmonic measures.105 Differentiable Manifolds). and others).e. there exists the extremal horizontal (vertical) slit mapping F. differential l-forms o=udx+udy and differential 2-forms CI = c dx A dy are defined on them. or O. there also exist the conformal mappings of R onto the parallel slit regions on the (g + 1)-sheeted covering surface of C (Z. = *r.) maps R conformally to a region on C whose boundary consists of horizontal (vertical) slits and possibly points.. At the first stage of the development. 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]. dw = 0) and pure..)/2 is called the span of R. and operations such as the exterior derivative dw = (au/ax . 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 1940.194 Harmonic Integrals). it should be noted that finer orthogonal decompositions into subspaces with specific properties hold for open Riemann surfaces.. F. = r. H.1367 367 I Riemann Surfaces lowing deep result (S. For instance. where *F. = {WC r. (ii) F1 and F. and F. Y.&lay) dx A dy and exterior product can be defined (. then w is called a holomorphic (or analytic) differential. i r.. Abelian Integrals on Open Riemann Surfaces H.. P. A surface of genus 0 is said to be planar (or of planar character or schlichtartig). called a harmonic differential. Let R be an arbitrary open Riemann surface.(i=1. Using the Dirichlet principle.

Bers. Constantinescu and A. Sommer. 1962. Press. and f gives a canonical parallel slit mapping of R (. 1957. Riemann surfaces. Princeton Univ. Pfluger. Riemann’s period relation on R has been studied for various classes of differentials. C. [ 121 M. Leqons sur les principes topologiques de la thtorie des fonctions analytiques. 1953. (2) For two discrete sequences {p. 8 (1958). Sot. Van Nostrand. Y. the analogy to the classical theory is lost completely if no restriction is posed on the differentials on R. [20] L.9] and the following: [ 173 B. Trans. [23] M. 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. The concept of a Riemann surface. Ann. [S] L... Lectures on Riemann surfaces. Classification theory of Riemann surfaces. 1938. C. but for the case of infinitely many nonvanishing periods.f for which df is distinguished is almost constant on every ideal boundary component. Springer. Kusunoki defined. Algebraic structure and conformal mapping. Tsuji. 100-134. Rodin and L. 142 (1979). Whereas by the RiemannRoth theorem above a nonconstant meromorphic function . Open Riemann surfaces and extremal problems on compact subregions. Ahlfors. 1957.f such that df is (exact) canonical exists on any open Riemann surface R with finite genus. Pacif. 1966. 1957. Math. Theorie der analytischen Funktionen einer komplexen Veranderlichen. For the classification of Riemann surfaces . Princeton Univ. Gauthier-Villars. Springer. 1953. 174 (1967).[4. Principal functions. 1954. Addison-Wesley. [ 191 L. Behnke and F.. [9] L. Math.367 Ref. Helv. Princeton Univ. Mizumoto. 1968. V. 65 (1935).9] and the following: [ 143 C. 1975). Acta Math. and therefore f reduces to a constant by the boundary theorem of Riesz type if R has a large boundary. Yoshida.. Press. Math. Math. J. The same is true for the theory of Abelian differentials with infinitely many singularities. Acta Math. Heins. Introduction to Riemann surfaces. [3] L.). [2l] R. Shiba. Sario and K. Hence every meromorphic function . Analytic functions with finite Dirichlet integrals on Riemann surfaces. Spencer. 1963. 1969.. 1970. Comm. Van Nostrand. Sario. V. Functionals of finite Riemann surfaces. [S] H. Springer. Studies. Press. H. [22] E.. Riemann Surfaces Ahlfors defined the distinguished (complex harmonic) differentials with polar singularities and obtained in terms of them a generalization of Abel’s theorem. Lecture notes. Schiffer and D. Capacity functions. Ahlfors. 1999220. Zur Theorie der Uberlagerungsflachen. no definitive result has been obtained. 299 50. Subalgebras of functions on a Riemann surface. Ideale Rander Riemannscher Flachen. 1966. 1964. and others further generalized the Kusunoki type theorems. and then u is almost constant on every ideal boundary component of R (in appropriate tcompactification of R). Addison-Wesley. English translation. Ahlfors (ed. Nakai. 1955. [lo] M. M. Rodin also gave generalizations of the Riemann-Roth theorem. New York Univ. Stoilow. [2] S. 157-194. This generalizes the results above and the GunningNarasimhan theorem. Press. [ 181 L.} and {q.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]. third edition. Potential theory in modern function theory. 267-276.. L. Springer. Royden [28] and B.} of points in R. [ 131 R. Contribution to the theory of Riemann surfaces. Sario and M. there exists a single-valued meromorphic function with zeros at pm and poles at 4.6. 24 (1950). Immersion of open Riemann surfaces. [4] R. For analytic mappings of Riemann surfaces .Section G). Noshiro.[6. [7] G. Gunning. Die Idee der Riemannschen Flache. 1368 References [ 1] H. Riemann surfaces. Uniformisierung. 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. Sario. Weyl. Narasimhan. On the other hand. 1913. Teubner. . Independently. [ 111 L. H.. Nevanlinna. 1959 (Chelsea. Springer. Oikawa. 89 (1958). Sario and K. Bishop. Gunning and R. Ahlfors and L. 1960. Theorie der Riemannschen Flachen. M. Cornea. [ 161 M. Sakai. Amer. It is further proved that there exists an Abelian differential with prescribed divisor and periods (Kusunoki and Sainouchi). 103-108. Ann. [6] A. second revised edition. Springer. [ 151 L. Value distribution theory. It was proved that a meromorphic differential df =du + idu is semiexact canonical if and only if du is (real) distinguished. Springer.. Maruzen. Princeton Univ. Math. H.

) 32 (1959). Math. called addition and multiplication (the ring operations). then they coincide and are uniquely determined by a. {a} +S({a}S) is denoted by a+S(uS). If A has such a unity element. In particular.. defined by L.). the unity element is distinct from the zero element.67 Commutative Rings). b of A to elements a + b and ub of A. then A is a tsemigroup with the identity element 0 with respect to this operation. and furthermore. [28] H. and R. be A) (commutative law for multiplication) (. Math.1369 368 B Rings [24] H. 430-461. the ring A is also an A-A-bimodule (. 33 (1961). The identity element of A under addition is called the zero element and is denoted by 0.9] and the following: [25] R. and (iv) for every pair a. then subsets S and Tare said to be orthogonal. Thus a ring A is an tAbelian group under addition. which send an arbitrary pair of elements a. 429-433. A subset S of a ring is said to be nilpotent if S” = 0 for some positive integer n. Univ. if we define a new operation (u. Also . R. KY&o. The inverse .(x) = ax. Certain nonassociative rings are important. commute for every pair a. Hence we often call a unitary ring simply a ring. An element a satisfying a” = 0 for some positive integer n is called a nilpotent element. If ST= 7’S= {0}. A ring is called a commutative ring if it satisfies (v) ab = ba (a. Mem. However. SS is denoted by S2 (similarly for S3. Ann. It satisfies the equation a0 = Ou = 0 (a E A). Royden.(x) = xu (x E A). are defined. In a general ring A. For the generalization of algebraic functions and Abelian integrals . (1) Two toperations. b. if a ring has more than one element. and a nonzero element a satisfying a2 = a is called an idempotent element. This element is called the inverse element of a and is denoted by a-‘.67 Commutative Rings). (An algebra is a ring having a tground ring. and idempotent if S2 = S.[4.149 Fields). then 0 is the unity element of the ring. A commutative unitary ring having more than one element is called an integral domain if it has no zero divisors (. I&a. [26] H. Most of the important examples of rings are unitary. [27] Y. let S + T(ST) denote the set of elements s + t(st) (s E S. 34-46. If a ring has only one member (namely. Fenn. Since the operations L. Comment. these operations satisfy the following four laws: (i) law of addition). 416 Teichmiiller Spaces. 120 (1949). 34 (1960). Entwicklung analytischer Functionen auf Riemannschen Fllchen. An element that has an inverse element is called an invertible element (regular element or unit). Behnke and K. (2) For arbitrary elements a. 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. a skew field that satisfies the commutative law is called a commutative field or simply a field (. For an element a of a unitary ring A. Each element a of a ring A determines operations L. (A. and R. Math. etc. 1 (1941). A nonzero unitary ring is called a tskew field (or tdivision ring) if every nonzero element is invertible. Definition A nonempty set A is called a ring if the following conditions are satisfied.9) Rings A.b)+aob by setting aob=a+b-ub. Kusunoki. Math. 37-51. On the meromorphic function field of a Stein variety. Elements a and b of a ring are said to be orthogonal if ub = bu = 0. Theory of Abelian integrals and its applications to conformal mappings. an element a’ such that u’a = 1 (au’ = 1) is called a left (right) inverse element of a. Furthermore. b of elements of A. (ii) a + b = b + a (commutative (a+ b)+c=u+(b+c). an example is tLie algebra. (ab)c=u(bc) (associative laws). The Riemann-Roth theorem. Such a ring is called a zero ring. Thus a ring A has the structure of a tleft A-module and a fright A-module.6. (iii) u(b+c)=ab+uc. In this article we shall discuss associative rings. 0). Acad. L. 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). 235-258.) B. 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. c of A. Ann.. Quadratisch integrierbare Differentiale auf einer Riemannschen Mannigfaltigkeit. (2) 83 (1966). l-34. Stein. there exists a unique element c of A such that a + c = b. S4. Coll.277 Modules). Sci. b of elements of A. Sci. An idempotent element is said to be primitive if it cannot be represented as the sum of two orthogonal idempotent elements. t E T). Helv.. A ring with unity element is called a unitary ring. If there exist both a right inverse and a left inverse of a. Ann.reference to 11 Algebraic Functions. it is unique and is often denoted by 1. Nevanlinna. (a+ b)c=ac+ bc (distributive laws).. Further Definitions 368 (111.

bE A). the rational number field Q. For unitary rings A and B.29 Associative Algebras. (4) Endomorphism rings of modules. then the mapping x+axa-’ (x E A) is an automorphism of A. A homomorphism of a ring A into itself is called an endomorphism. X. Factor Rings.14 Algebraic Number Fields. then the inverse mapping f-’ : B-rA is also a homomorphism. namely. . and the equivalence relation R in A defined by f(uRbof(u)=f(b)) is compatible with the ring operations of A.] of polynomials and the set K [ [X. The smallest subring containing a subset T of a ring A is called the subring generated by T. and an element that has a quasi-inverse element is called a quasiinvertible element (or quasiregular element). (5) For other examples . If a homomorphism f is tbijective. Subrings. Then the definition of factor ring implies that c(+ /?(G$) is the equivalence class represented by a + b(ab). if an antihomomorphism f is bijective. The set of elements that commute with every element of T forms a subring and is called the commuter (or centralizer) of T. In particular. E.256 Linear Spaces. 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. X. and f is called an anti-isomorphism. the set C’(I) of functions that are r-times continuously differentiable.368 C Rings element under this operation is called the quasi-inverse element. If we deal only with unitary rings and unitary homomorphisms. let K = R. The ring Z of rational integers. Thus / induces an isomorphism A/R df(A) (. (3) Rings of expressions. a homomorphism f: A+ B is said to be unitary if it maps the unity element of A to the unity element of B. D.]] of tformal power series in indeterminates Xi. with coefficients in a commutative ring K are commutative rings (. If a is an invertible element of a unitary ring A. and in this case f is called an isomorphism. represented by a and b. of a ring A is an isomorphism. respectively.Section G). 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. (2) Rings of functions. A composite of homomorphisms is also a homomorphism. a mapping satisfying (i) and (ii’) is called an antihomomorphism. In particular. Then the set C’(1) of continuous functions. Let LY and /J’ be elements of the factor ring A/R. and an isomorphism of A onto itself is called an automorphism of A. X. then A is called a quasisimple ring (. . the real number field R. If f: A-t i? is a ring homomorphism. Thus the ring operations of S are the restrictions of those of A. then the image f(A) is a subring of B. Examples (1) Rings of numbers. and the set C”(l) of analytic functions on I are subrings (. The set &K(M) of tendomorphisms of a +module M over a ring K is in general a noncommutative ring.. called an inner automorphism. There exists only one homomorphism of any ring onto the zero ring. The set K [X. More precisely. if M is a finite-dimensional +linear space over a field K. Antiendomorphisms and antiautomorphisms are defined similarly.Section F). If { AijiE.e. . bs A) is called a homomorphism. If A has no nontrivial factor rings. then &JM) can be identified with a +full matrix ring (. 277 Modules).. the Cartesian . 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. The identity mapping 1. a homomorphism (isomorphism) of rings is often called a ring homomorphism (ring isomorphism). In particular. 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. 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. aRu’ and bRb’ imply (a + b)R(u’ + b’) and (ab) R(a’b’)). 257 Local Fields).. and let I be an interval of R. In this sense there exists a unique hom- omorphism of the ring Z of rational integers into an arbitrary unitary ring. . 36 Banach Algebras.. . This is the case if and only if the equivalence relation R is compatible with the ring operations (i. 1370 C.369 Rings of Polynomials. and 439 Valuations. Every ring A has two trivial factor rings.Section E) of R’. A itself and the zero ring 0. When condition (ii) for a homomorphism is replaced by (ii’) f(ab) =f(b)f(u) (a. 67 Commutative Rings. is a family of rings. In particular. then a subring S necessarily contains the unity element of A. then the inverse mapping f -’ is also an antihomomorphism. . and the complex number field C are familiar examples (. 370 Rings of Power Series). a unitary homomorphism is usually meant. 284 Noetherian Rings. By a homomorphism.

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

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

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

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

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

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

. the rank test by F. and (x 1. x. a.) be an independent trandom sample of size n from F(x).(R+)Y(Xi-to) i=l for some weights a. For testing the thypothesis H: 5. : F(x) f G(x) or A. When the alternative hypothesis A. . 0 when i(x. . .e. ... R. Hampel(l974) the estimator proposed by Huber has become a core of subsequent studies of robust estimation. which is the uniformly most powerful rank test in a neighborhood of (I. L. In the years that followed two important ideas appeared: the concept of asymptotic relative efficiency by E. x. 400 Statistical Hypothesis Testing). H. i.(i) = i is frequently used and is called the Wilcoxon signed rank test. . The Two-Sample Problem Let F and G be continuous distribution functions of random variables X and Y. H. in which he investigated how the standard statistical procedures obtained under certain assumptions are influenced when such assumptions are violated. Huber (1964) proposed an estimator of location by generalizing the method of least squares. A. we say that the random variable Y is stochastically larger than X and write F > G. . . . F(<.)>c. Chernoff and I. 1 0 when &(x1. . Tukey (1960. . Wolfe [3]. . y. A.. Set ux. Arbuthnot (1710).(n).)<c 1 (0 < a < 1.5’1 among IX. recent progress is summarized in the books by J. i(xl . . J. . < 0. These findings accelerated the studies of nonparametric tests.. Wilcoxon (1945). . . .. Pitman (1948) and the development of the theory of Ustatistics by W. . called the signed rank test. . .) is defined by the equations 1 when cpb. Hodges and E.. . is G(x) = F(x .) be the number of xi that are greater than co.X.) be an observed sample value.. A similar study for scale was made by S. x. x. = 5” for the double exponential distribution. R. and (x 1. x. A frequently used example of such an alternative hypothesis A. [S].. . (X. . .) and (yl. 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..(l).)>c.. respectively. . may not hold for the actual data.. the theoretical foundation for nonparametric tests was first given in the proposals for the permutation test by R.. R. < 5’ against the talternative hypothesis A : t.396 Statistic... . x. Along with the idea of the influence curve introduced by F. and P.e). J. . J.X. Randles and D.x.. 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.: F(x)> G(x) for all x and F(x) f G(x). Huber [6]. G. Whitney (1947).) be the corresponding random samples.x. Two papers by J. G. .1377 371 c Robust and Nonparametric B. L. Hgjek and Z. The following procedure rp.. Let i(x. Andrews et al.2 = 5’ for F(x) = l/( 1 + eex). Takeuchi (1971) proposed an adaptive estimate that is asymptotically fully efficient for a wide class of underlying distribution functions. that the asymptotic efficiencies of nonparametric tests are incredibly high. Consider the problem of testing the hypothesis H: F(x) = G(x) against the alternative hypothesis A. B. is true. Savage (1958) showed. Sen [2]... The term nonparametric or distribution-free is used for problems of testing hypotheses. Let R: be the rank of IX.)=c. .. The procedure with a. (X1. . V. M. . .)= a when &(x1 . .) be the respective sample values. Y. . The 100~ percentile of F is denoted by tp. x. The developments of the theory of robust estimation are reviewed by Huber [4-61 and R. and the test based on Ustatistics by H. P. cording as t > 0. and let Y(t)= 1... and the term robust is mainly used for problems of point estimation (. . L.. where cp(x.)= f: a. F. x. . Hoeffding (1948). This procedure is called the sign test.. 399 Statistical Estimation. ..> to.x. Although the idea of the sign test appears in the work of J. x. W.0 acIx1-5°1.) and ( Y1... . the following procedure is proposed... 1962) provided the initial foundation for robust estimation. E. K. A test procedure by the following ttest function cp is tuniformly most powerful in some neighborhood of 5. dx.)= a when i(xl . . Kakeshita and T.)=c. R. $idik Cl]. the logistic distribution.)<c.X. Puri and P.. C. Let F(x) be a tdistribution function of a trandom variable X.‘. Suppose that F(x) is symmetric about x = lliz. Fisher (1935). K. On the other hand. . in studying the asymptotic distribution of a class of rank statistics.. Mann and D. . .. . .-t”l. Box (1953) first coined the term robustness in his sensitivity studies.. is also used for testing the hypothesis H: <Ii2 < 5’ against the alternative hypothesis A: <1/Z> 5’: 1 when &(x1. The One-Sample Problem Methods especially those about the shape of the distribution. . Yanagawa (1967). . . . Various proposed estimators are compared in the book by D...) =p. 0 . 0 < c < n).. Hogg [7].

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

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

. T. Under quite general conditions. .$. X. F) be the asymptotic variance of T. > K. . . and S.6745}i. ..<S(X. 1 is chosen to be x(t) = t$(t) . . . . S tends to be the statistic for the Wilcoxon signed rank test. ...=a. or by solving the simultaneous equations (1) and gx(fq)=o n with respect to T.) that satisfy P..X. means the probability under the hypothesis H: 0 = 0.-[. T. L..forsomegiven constants a. and set ak = lf/Ll.f’(t)/f(t) and J(t)=$‘(F-‘(t))/l(F). are given by L.. n)... .+2)+ “’ +PX. 4(x 1. for some constant K. where Q is an unknown location parameter. a tconlidence interval of fJ is constructed as follows.)<d. If there exist statistics L. . Hodges and Lehmann [ 121 first proposed this technique. For an appropriately given y (O<y < l).) under the hypothesis H: 0 = 0. which is defined by jt.-8). is called the M-estimator.0) =Ci=. Then Chernoff.) such that L. then T.. x..s(x. e.. = 1.) or by satisfying Using the statistic S..-@cd. Point Estimation (1) n Let (X. J((R:(@+n)/(2n+ l))Y(X. and d. <S(x. J.K(t(K2/2 defined for (t] < K.y)% confidence coefficient is given by (L. .. respectively. the median of { ]xi. Let fI*=sup{t)..-O)<p}.. IX.. = K&+. When S. the M-estimator defined above is not scale invariant. x”) for the one-sample nonparametric test for testing the hypothesis H: 0 = 0. where S. An example is T. is the statistic for the Wilcoxon signed rank test. Let p be a real-valued (usually convex) function of a real parameter with derivative ti = p’.0.O according as t > 0. . then the confidence interval of 0 with lOO( 1 .. .(x. -. Let CD(x) be the distribution function of the standard normal distribution. .)} = 1 --y for all F..+. The first method is touse T.)<0<u..) d 0 < U. . . This estimator is called the a-trimmed mean.. . it agrees with the least squares estimator. . H(x) be an arbitrary continuous distribution function which is symmetric about the origin. . There are four methods of constructing robust estimators of 8.{d.b(x. .}=l-y..J(t)dt.1. V...s<l). .. x.M1/0. and let (x1.. .) and U. where p = 1 + [car] -cm.)+x~[.. is the tmaximum likelihood estimator of 0 for F and is asymptotically efficient under some regularity conditions.) be its observed value. x. X.-tI)>n}..g. Let J be a real-valued function such that j. < <X. X. The third method employs nonparametric tests for testing the hypothesis H: 0 = 0 against the alternative hypothesis A : 0 > 0.1 (. The second method is to estimate 0 by T.e. converges to T(F)= fhJ(t)F-‘(t)dt in probability. J(t)dt = 1... and U..H). and M. where P.(t)=t’/2. ..F)isgivenbyp... Let J be a real-valued and nondecreasing function such that jhJ(t)dt =0 and R:(O) be the rank of IX. -8. in the range of S(x.~I/(p...... -.371 H Robust and Nonparametric 1380 Methods by minimizing Cf=r p(xi . T.-Q(.. When F(x) is symmetric about the origin and J(t) = t-i. < 0 and S(X. fI**=inf{&S(X.+a. and the .2..x.X~.2. L.x.(x r . = S(x. X... . = M/.(x.X.y(x. In the context of the maximum likelihood estimation.. Huber [ 1 l] proved that p minimizing s~p.X~. -O(among(X. . =(Q* + 0**)/2. . < K&.. where M is the sample median.. . it holds that P(L...-O( . is any robust estimate of scale. ..x.?(xI. and this estimator is called the Restimator.) be an independent random sample from the population with a distribution function F(x ..“. . and V(p. Then an estimator of Q is defined by T. If $(t)= --f’(t)/f(t) is chosen for $(x).)).-8.(x.M-dIjr where M = n(n + 1)/2 and q. the M-estimator converges as n+ 00 to T(F) in probability.T. are ordered values for M averages (xi + xi)/2 (i <j = 1.. X.2c0.....) if and only if d. H. A scale invariant version of the M-estimator is obtained by replacing the defining equation by &$(y)=o. 9 be a class of distribution functions of the form F(x) = (1 -E)@(x)+EN(x) for a given s (O<. .. . Johns [lo] proved that under some regularity conditions TR is an tasymptotically efficient estimator of 0 for F.(x. such that & a. .T(F))dF(x)=O. Let X(. . be ordered values of X. Generally.)ln(l . ... then as n+co.(~)=(Px~[.. for all 0. Set $(t) = -. select constants d. Suppose that F is a distribution function having an +absolutely continuous density function f: Denote the derivative off by f’.> x. i. x. . . .X. x. where n is the expected value of S(X.). This interval is distributionfree. and let I(F) be the tFisher information on 0..l+.) and U.. . When p(t) = t2.-0 . Gastwirth..SetY(t)=l. is called the L-estimator. a. .399 Statistical Estimation PI. x.

) that minimizes x. A real-valued function IC(x. (1) Spearman’s Rank Correlation.Cjejaij) for some function p such as the one described above...Thenr.Sj). . The curve was first introduced by Hampel [lS] to study the stability aspect of estimators against a small change of F. E. Set di = Ri -Si. The last method of constructing robust estimators consists of estimating 0 adaptively by utilizing information on the shape of F. estimating the elements of the tcovariance matrix of the order statistics by Ustatistics. The Influence Curve Let T(F) be a functional of a distribution function F. K.(Ri)Ai is asymptotically equivalent to minimizing the properties of which were first studied JureEkov& [ 163.. As n-co. Ed. T))2dF(x).Si) and (Rj.) as an estimator for (0. It has been proved that minimizing & a. defined by F(x)+l-F(2T(F)-x) 2 dF(x) = 0.. . namely. . Y. R-estimators of the regression coefficients are obtained by minimizing & a. F-‘(l--)/(1--a) when x>F-‘(l-a). .T))+0 I in probability..Zl p(X. (2) Kendall’s Rank Correlation. A. the influence curve for the cl-trimmed mean IC(x. Take pairs (Ri.=l-6Zid’/(n3-n)iscalled Spearman’s rank correlation.)=(n-1)-l. by J. Let (Xl. are identically and independently distributed random errors whose distribution function is given by F(x). F. for F in T(F). a striking one is the asymptotically fully efficient estimator for a wide class of F proposed by Takeuchi [ 131.1381 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). where S. n”z(T(F. Jaeckel [14] made an atrimmed mean adaptive estimator by selecting an a that minimizes the estimated asymptotic variance. . the aij are given constants. . T) = by T((l-e)F+s6. Y.CjQjaij. R. .(X. 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. set By substituting the empirical distribution function F. where Ai = Xi . Although the above three methods provide robust estimators. i. when F is symmetric about the origin. and Sj be the rank of yj among Y. when F-‘(cc)<x <F-‘(l-a). n. at . For symmetric F..)--T(F) & where the Xi are observable variables. I. . converge to T(F) in probability as n+ co. ... . If there is no dependence between X and Y. which are seldom affected by outlying observations or contamination by gross errors. . .. . when they are rearranged in an ascending order. it can be proved that as n-+ co.0. . defined similarly as Ri. by T. T) is given by F-+$/(1 x/(1-2a) -201) when x < F-‘(a).2.(Ri)Ai. L. “. X. be the rank of Xi among X 1.(i) = 0. The estimator is constructed by using subsamples of size K (K <n) drawn from the original sample.) is some monotone function satisfying CyZl a. F. T) defined by IC(x. j=l i = 1.( . . Dependence for all x is called the influence curve. . then E(rs)=O and V(r.T(F)) is asymptotically normally distributed with asymptotic variance S(IC(x.. the R-estimator converges in probability to T(F). of 0 calculated from an empirical distribution function F. Y.). and Ed. the Qj are regression coefficients to be estimated. Under some conditions. Thus it follows that n’j