10.1007@978 3 642 65567 8

Ergebnisse der Mathematik
und ihrer Grenzgebiete
Band 78
Herausgegeben von P. R. Halmos . P.]. Hilton

R. Remmert· B. Sz6kefaIvi-Nagy
Vnter Mitwirkung von L. V. Ahlfors . R. Baer

F. L. Bauer . A. Dold . ]. L. Doob . S. Eilenberg
K. W. Gruenberg· M. Kneser . G. H. MillIer
M. M. Postnikov . B. Segre . E. Sperner
Geschaftsfiihrender Herausgeber: P.]. Hilton

Janos Bognar
Indefinite Inner
Product Spaces
Springer-Verlag
Berlin Heidelberg New York 1974
Janos Bognar
Mathematical Institute of the Hungarian Academy of Sciences,
1053 Budapest, Hungary
AMS Subject Classification (1970)

Primary 46-02, 46 D 05, 47-02, 47 B 50
Secondary 46 C xx, 47 A xx
ISBN -13: 978-3-642-65569-2 c-ISBN -13:978-3-642-65567-8

001: 10.1007/978-3-642-65567-8
This work is subject to copyright. All rights arc reserved, whether the whole or patt of
the material is concerned, specifically those of translation, reprinting, fe-use of illustra-
tions, broadcasting, reproduction by photocopying machine or similar means, and
storage in data banks. Uneler § 54 of the German Copyright Law ,,,here copies are made
for other than private usc, a fee is payable to the publisher, the amount of the fee to
be determined by agreement with the publisher. © by Springer-Verlag Berlin Heidel-
berg 1974. Library of Congress Catalog Card Number 73-10669.Softcover reprint of
the hardcover 1st edition 1974
Preface
By definition, an indefinite inner product space is a real or complex

vector space together with a symmetric (in the complex case: hermi-
tian) bilinear form prescribed on it so that the corresponding quadratic
form assumes both positive and negative values. The most important
special case arises when a Hilbert space is considered as an orthogonal
direct sum of two subspaces, one equipped with the original inner prod-
uct, and the other with -1 times the original inner product.
The subject first appeared thirty years ago in a paper of Dirac [1]
on quantum field theory (d. also Pauli [lJ). Soon afterwards, Pontrja-
gin [1] gave the first mathematical treatment of an indefinite inner prod-
uct space. Pontrjagin was unaware of the investigations of Dirac and
Pauli; on the other hand, he was inspired by a work of Sobolev [lJ,
unpublished up to 1960, concerning a problem of mechanics.
The attempts of Dirac and Pauli to apply the concept and elemen-
tary properties of indefinite inner product spaces to field theory have
been renewed by several authors. At present it is not easy to judge
which of their results will contribute to the final form of this part of
physics. The following list of references should serve as a guide to the
extensive literature: Bleuler [1], Gupta [lJ, Kallen and Pauli [lJ, Heisen-
berg [lJ-[4J, Bogoljubov, Medvedev and Polivanov [lJ, K.L.Nagy
[lJ-[3], Berezin [lJ, Arons, Han and Sudarshan [1], Lee and Wick [1J.
In particular, the book of K. L. Nagy [2J contains a critical survey of the
literature up to 1966.
The work of Pontrjagin, on the geometry of certain indefinite inner
product spaces and on the spectral theory of their linear operators, has
also been carried on. This line of research led to a number of deep
results connected with the names of M. G. KreIn, r. S. Iohvidov, Ju. P.
Ginzburg, R. S. Phillips, H. Langer, M. A. Nalmark, Ju. L. Smul'jan,
G. Wittstock, and others. The present book is an account of this
second line of investigations.
The main body of the text offers an introduction, with full proofs,
to the subject. Prerequisites do not extend much beyond standard
facts of linear algebra, topology, Hilbert and Banach space theory.
VI Preface
The treatment is, whenever possible, invariant, i.e. independent of

the choice of bases, decompositions and other additional structures.
The presentation proceeds from the general to more and more special
cases. Terminology and notation are partially new; in favour of the
new terms and because of the large variety of the old ones no glossary
is included.
The notes at the end of each chapter contain, besides historical and
other comments relating to the main text, a survey of the literature on
more advanced topics such as the spectral function, groups and algebras
of operators, applications to differential equations and to quasi-definite
functions, etc. For applications in stability theory, however, the reader
is referred to the monograph of Daleckil and KreIn [1J. Applications
to characteristic functions of Hilbert space operators are only touched
upon, since they do not seem to have reached a final form.
The author expresses his sincere gratitude to B. Sz.-Nagy for pro-
posing to write this book, for constant encouragement and helpful
suggestions.
Thanks are also due to the following persons: G. Borg (Royal In-
stitute of Technology, Sweden) and 1. S. Louhivaara (University of
J yvaskyHi, Finland) for invitations to lecture on the subject; H. Langer
and G. Wittstock for certain counter-examples that have been included
in the text; G. Adler, K. MaJyusz and A. Szep for useful discussions.
The author is greatly indebted to his superiors at the Mathematical
Institute of the Hungarian Academy of Sciences for affording time to
accomplish this work.
Finally, the author is deeply obliged to all members of his family
for their patience and understanding with which he could not dispense
during the last several years.
Budapest, October 1973 Janos Bognar

Table of Contents
Chapter 1. Inner Product Spaces without Topology
1. Vector Spaces
2. Inner Products 3
3. Orthogonality 7
4. Isotropic Vectors 9
5. Maximal Non-degenerate Subspaces 11
6. Maximal Semi-definite Subspaces . 12
7. Maximal Neutral Subspaces 14
8. Projections of Vectors on Subspaces 15
9· Ortho-complemented Subspaces 18
10. Dual Pairs of Subspaces . . . 20
11. Fundamental Decompositions 24
Notes to Chapter I . . . . . . . 26
Chapter II. Linear Operators in Inner Product Spaces without Topology
1. Linear Operators in Vector Spaces. 28

2. Isometric Operators . . 31
3. Symmetric Operators . . . . . . 34
4. Cayley Transformations . . . . . 38
5. Principal Vectors of Cayley Transforms 39
6. Pairs of Inner Products: Semi-boundedness 42
7. Pairs of Inner Products: Sign 44
8. Plus-operators . . . . 45
9. Pesonen Operators 48
10. Fundamental Projectors 49
11. Fundamental Symmetries. Angular Operators 52
Notes to Chapter II. . . . . . . . . . . . . . 55
Chapter III. Partial Majorants and Admissible Topologies on Inner

Product Spaces
1. Locally Convex Topologies on Vector Spaces 58

2. Partial Majorants. The Weak Topology . 59
3. Metrizable Partial Majorants . . . . . . 61
4. The Polar of a Normed Partial Majorant . 63
5. Admissible Topologies. . . . . . . . . 65
6. Orthogonal Companions and Admissible Topologies 68
VIII Table of Contents
7. Projections and Admissible Topologies. 70

8. Intrinsic Topology . . . . . . . 71
9. Projections and Intrinsic Topology 72
Notes to Chapter III . . . . . . . . 75
Chapter IV. Majorant Topologies on Inner Product Spaces
1. Majorants . . . . . . . . . . . . . . . 77
2. Majorants and Metrizable Partial Majorants 79
3. Orthonormal Systems . . . . . 81
4. Minimal Majorants . . . . . . 84
5. Majorants and Decomposability. 88
6. Decomposition Majorants 90
7. Invariant Properties of G);+ and G);- • 93
8. Subspaces of Spaces with a Hilbert Majorant . 95
Notes to Chapter IV . . . . . . . . . . . . . 98
Chapter V. The Geometry of Krein Spaces
1. Krein Spaces. . . . . . . 100

2. Krein Spaces as Completions 102
3. Subspaces . . . . . . . . 103
4. Maximal Semi-definite Subspaces 105
5. Uniformly Definite Subspaces. . 107
6. Non-uniformly Definite Subspaces 109
7. Maximal Uniformly Definite Snbspaces 112
8. Regnlar and Singular Subspaces. . . 113
9. Alternating Pairs . . . . . . . . . 114
10. Dissipative Operators in l-Iilbert Space 116
Notes to Chapter V . . . . . . . . . . 118
Chapter VI. Unitary and Seifadjoint Operators in Krein Spaces
1. Preliminaries. . . . . . . 120
2. The Adjoint of an Operator 121
3. Isometric Operators. . . . 124
4. Unitary and Rectangular Isometric Operators. 128
5. Spectral Properties of Unitary Operators 131
6. Selfadjoint Operators . 133
7. Cayley Transformations 136
8. Unitary Dilations 139
Notes to Chapter VI . . . 14S
Chapter VII. Positive Operators and Plus-operators in Krein Spaces
1. Positive Operators . . . . . 147

2. Operators of the Form T* T . 149
3. Uniformly Positive Operators. 151
4. Plus-operators . . . . . . . 153
Table of Contents IX
5. Strict Plus-operators. 154

6. Doubly Strict Plus-operators 158
Notes to Chapter VII . . . . . 161
Chapter VIII. Invariant Semi-definite Subspaces of Linear Operators

in Krein Spaces
1. Fundamentally Reducible Operators . . . . . . . . . . . . . . . . . 163
2. Invariant Positive Subspaces of Plus-operators. . . . . . . . . . . . 165
3. Invariant Semi-definite Subspaces of Unitary and Selfadjoint Operators 168
4. Quadratic Pencils of Operators in Hilbert Space 172
5. Quadr"atic Operator Equations in Hilbert Space . 175
6. Spectral Functions 178
Notes to Chapter VIII 180
Chapter IX. Pontrjagin Spaces and Their Linear Operators

1. The Spaces ilk' Positive Subspaces 184
2. Closed Subspaces . . . . . . . . 186
3. Isometric Operators: Continuity 188
4. Isometric and Symmetric Operators: Number and Length of Jordan Chains 189
5. Proof of Theorem 4.3 . . . . . . . . 192
6. Regular Symmetric Extensions . . . . . 197
7. Invariant Positive Subspaces: Existence. 200
8. Invariant Positive Subspaces: Uniqueness 201
9. Common Invariant Positive Subspaces for Commuting Operators 204
Notes to Chapter IX . . . . . . . . . . . . . . . " . . . . . 207
Bibliography. . 210
Index of Terms. 221
Index of Symbols 224
Chapter I. Inner Product Spaces without Topology
The basic notions connected with general inner product spaces are introduced and
analysed. In particular, non-degenerate spaces (Sections 4- 5), maximal semi-
definite subspaces (Sections 6-7), ortho-complemented subspaces (Section 9), and
decomposable spaces (Section 11) are investigated. Because of their simplicity, the
proofs can alternatively be worked out by the reader. Lemmas 4.4, 5.1, 6.3, 10.1,
10.7 and Corollaries 9.5, 11.9 deserve special attention.
1. Vector Spaces
In this section we fix some conventions (terms and symbols) concerning

vector spaces with no additional structure to be used throughout the
book.
By a vector space we shall always mean a vector space over the
field C of complex numbers.
Two vector spaces (;\;, (;\;' are said to be isomorphic, if there is a one-
to-one mapping T of (;\; onto (;\;' such that T(C<lXl + CX 2X 2 ) = CXI TXI+
+ CX 2 TX2 for every Xl> x 2 E (;\; and CX I , CX2 E C. The mapping T is called
an isomorphism between (;\; and (;\;'.
A subspace of a vector space is a subset which is a vector space itself.
In topological vector spaces eventual closedness requirements will be
cl early pointed out.
The whole space is a subspace of itself. The set consisting of the
zero vector alone is also a subspace, called the zero subspace.
The symbol 0 stands for any of the following objects: the number
zero, the zero vector, and the zero subspace. The empty set is denoted
by fl.
m
lf is a subset of a vector space, then its span, denoted by <m),
is the least subspace containing m. The span <ml, ... , m,,) of several
m
subsets i (j = 1, ... , n) is defined by the equation
<ml> ... , m,,) <\l{l U. . . m,,) .
= u
It will frequently occur that some of the sets mj consist of a single
element Xi' In this case we denote the subset mj by Xi instead of {Xi}'
m
We write, for example, <Xl> 2) rather than {Xl}' 2).< m
2 1. Inner Product Spaces without Topology
If 2j are subspaces (j = 1, ... , n), then obviously
(1.1) (21) ... ,2n>={tXj:

1=1
xj E3 i , i=1, ... ,n}.
Therefore (31) ... , 2,,> can alternatively be denoted by 21 3", + ... +
and called the vector sum of the subspaces 2 j . It is easy to see that for
arbitrary sets mj (j = 1, ... , n) we have
(1.2) <m l , ... , mn>= (m >+ ... + <m,,> .
l
The subspaces 2j (f = 1, ... , n) are said to be linearly independent,

if the representations LXi in ('1.1) are unique or, what is the same, the
relations
n
LXi = 0, XiE3j U=1, ... ,n)
i=1
imply Xl = X2 = ... = x" = 0. Another equivalent condition IS the
following:
(j = 2, ... ,n) .
The vectors vi U = 1, ... , n) are said to be linearly independent,
if none of them is zero and the subspaces (vi> (j = 1, ... , n) are
linearly independent. Equivalently, the vectors VI' . • . , vn are linearly
independent if the relation
= 0.
L
i=1
n
IXjV j = °implies IXI = IX2 = ... = IXn =
An infinite system of elements is said to be linearly independent if

each of its finite subsystems is so.
The maximal number of linearly independent elements of a vector
space Q; is the dimension of Q; (in symbols: dim Q;). Unless otherwise
stated, we allow it to be either 0, or a positive integer, or co. Distinction
between infinite dimensionalities will only be made in topological vector
spaces.
In the case dim Q; <
co any maximal set of linearly independent
elements is called a basis of Q;.
The vector sum (span) of linearly independent subspaces 21 , . . • ,2n
is termed also a direct sum and denoted alternatively by 21 3n" + ... +
If 2 is a subspace of the vector space Q; then, by virtue of Zorn's
lemma, there exists a subspace We c Q; satisfying the relations
(1. 3)
Moreover, given a subspace W1l c Q; such that 2 n WC l = 0, a subspace
W1 ) W1l with the properties (1.3) can be found. We say that W1 is
a complem,entary subspace for 2 (with respect to Q;).
Consider, once again, a subspace 2 of the vector space Q;. In Q;
an equivalence relation rv can be introduced by letting X rv y if
2. Inner Products 3
x - Y E 2. The set of equivalence classes becomes a vector space @/2
(the quotient space of @ with respect to 2) once addition and scalar
multiplication of classes are defined by the corresponding opera-
tions on representatives. The zero vector of (§3/2 is evidently the class
consisting of all elements of 2.
The dimension of the quotient space (§3/2 will be called the
codimension of 2 with respect to (§3, and denoted by codim~ 2 or,
when (§3 is the whole space under consideration, simply by codim 2.
The following well-known statement, cited here with a view to
later application, shows the close connection between complementary
subspaces and quotient spaces.
Lemma 1.1. If 2 is a subspace of the vector space (§3, then every

complementary subspace we tor 2 with respect to (§3 is isomorphic to (§3/2.
A suitable isomorphism is given by Tx = x
(x E we), where E @/2 is the x
equivalence class containing x. D
Corollary 1.2. The dimension of any complementary subspace f01-

a subspace 2 is equal to codim 2. D
It is also possible to define the quotient space we/2 of an arbitrary
subspace we c (§3 with respect to the subspace 2 C (§3. In this case the
equivalence classes are formed as follows: the elements x, y E 9)1 belong
to the same class if x - y E 2. The zero vector of we/2 turns out to be
the class consisting of all elements of Wc n 2. It is clear that we/2 is
isomorphic to a subspace of (§3/2.
2. Inner Products
In our terminology, an inner product on a vector space (§3 is a complex-

valued function (. , .) defined for all pairs x, y E (§3 so that the conditions
(2.1)
(2.2) (y, x) = (x, y)
are fulfilled for everyIXvIX2 E C and Xl' X2 , x, Y E (§3. Relations (2.1) and
(2.2) obviously imply
(y, 1X1X1 + 1X2X2) = (X1(y, Xl) + (X2(y, x 2) .
The number (x, y) is termed the inner product of x and y.
A vector space equipped with an inner product will be called an
inner product space.
4 I. Inner Product Spaces without Topology
Two inner product spaces ~ and ~I with inner products (. , .) and

(. , .)" respectively, are said to be isometrically isomorphic, if a one-to-
one mapping T of ~ onto ~I can be found so that
T (CXlX l + CX X = cx1Txl + CX 2 TX2 ,
2 2)
(TXl' Tx 2 )' = (Xl' x 2)

for every Xl' %2 E ~ and CXl' ();2 E C. The mapping T is called an isometri-
cal isomorphism between ~ and ~/.
In an inner product space ~ the number (x, x) may be termed the
inner square of the element % E~. One verifies easily that
(2-3 ) (x, y) = -
1
(x + y, + y)
X
1
- - (x - y, x - y) +
4 4
+ -i(x + ~y,
. x + ly
.) - -i (
% -
.ly , X -
.)
zy
4 4
for every x, y E~. Thus the inner product is uniquely determined by
the values of the inner square. Relation (2.3) is called the polarization
formula.
Consider an element % of the inner product space (2;. As (x, x) =
= (x, x), there are three possibilities: either (x, x) 0, or (x, x) 0, > <
or (x, x) = o. Correspondingly x is said to be positive, negative, or
neutral. It is clear that the .zero vector is neutral.
The set of all neutral vectors in ~ is called the neutral part of (2;, or
the neutral set of the inner product (. , .), and will be denoted by ~o.
The symbol ~oo will stand for the set obtained from ~o by omitting the
element 0:
~oo = {x E ~ : (x, x) = 0, % =1= O} .
We denote by ~++ (resp. ~--) the set consisting of the zero element
together with all positive (negative) elements of ~, and by ~+ (resp.
~-) the set of all non-negative (non-positive) elements of~. Thus e.g.
~++ = {x E~: (x, x) >0 or x = O},

~+ = {x E~: (x, x) > O} .
If ~ contains positive as well as negative elements, we say that the
inner product is indefinite on ~, or that ~ is an indefinite inner product
space.
Lemma 2.1. Every indefinite inner product space contains non-zero
neutral vectors.
Proof. Let (x, x) >
0, (y, y) <
o. We set z = x + A y, where A is
a (real) solution of the equation
(x, x) + 2 A Re(x, y) + },2(y, y) = 0.
2. Inner Products 5
Then (z, z) = O. On the other hand, z = 0 would imply (x, x)
= 1.412 (y, y), which contradicts our starting point. 0
The inner product is said to be semi-definite on Gf, if it is not indefi-
nite. In this case we speak of a semi-definite inner prodt{ct space.
Lemma 2.2. In a semi-definite inner prodttct space Gf the Schwarz
inequality
I(x, y) 12 < (x, x) (y, y) (x, y E Gf)
holds.
The proof is the same as in Hilbert space. 0
A semi-definite inner product may either be a positive inner product
((x, x) > 0 for every x E Gf), or a negative inner product ((x, x) < 0 for
x E Gf). A neutral inner product ((x, x) = 0 for every x E Gf) is positive
and negative at the same time.
An inner product on Gf is said to be definite, if (x, x) = 0 implies
x = o. In view of Lemma 2.1 a definite inner product is semi-definite.
>
Hence we have either (x, x) 0 for x"* 0 (Positive definite inner prod-
uct), or (x, x) <
0 for x "*
0 (negative definite inner product).
A positive (negative, neutral, definite, positive definite, negative defi-
nite) inner product space is a vector space with an inner product of the
respective kind.
Example 2.3. Let lh e2' ... be real numbers. Denote by Gf the
vector space of sequences {~l' ~2' ... } of complex numbers satisfying
< CXJ. The formula
00
L' leil l~il2

i=l
00
(x, y) = .1' ellii (x= {~l> ~2' ... }EGf, y= {171> 172' .•. }EGf)
i=l
defines an inner product on Gf, which may be indefinite, semi-definite
etc. depending on the numbers ej"
Example 2.4. VVe consider a real-valued function (1 of bounded
variation on the real line R, and the vector space Gf of all complex-
valued functions on R that are measurable and square-summable ",rith
respect to the total variation of (1. The relation
(x, y) = J x(~) y(~) d(1(~) (x,y E Gf)
R
defines an inner product on Gf, which may be indefinite, semi-definite

etc. depending on the function (1.
A subspace ~. of an inner product space Gf is an inner product space
with respect to the restriction of (. , .) to~. If the inner product is
indefinite on ~, we say that 2 is an indefinite subspace of Gf. Similarly
we can speak of semi-definite, positive, negative, neutral, definite, positive

definite, and negative definite subspaces. In particular, the zero subspace
is positive definite as well as negative definite.
Positive definite inner product spaces are well-known objects
("prehilbert spaces", "inner product spaces" in the classical sense).
Negative definite inner product spaces do not possess any new prop-
erties, and semi-definite inner product spaces can be reduced to
definite ones (see Lemma 4.4 and Corollary 5-3 below). For this reason
indefinite inner product spaces will be our main concern. Nevertheless,
in order that the results would be applicable to any subspace, semi-
definite as well as indefinite, we shall formulate them for arbitrary
inner product spaces whenever possible. In such cases it will be under-
stood that the space under consideration may be a subspace of another
inner product space.
Throughout this chapter, when not specified, Q: denotes
an arbitrary inner product space.
Lemma 2.5. The span of a positive (negative, neutral) vector 2S

a positive definite (negative definite, neutral) s~tbspace.
This statement is trivial. 0
Most of the results concerning inner product spaces have two

variants obtained from each other by interchanging the words "pos-
itive" and "negative". Making use of the next definition it is always
sufficient to treat only one variant.
Let Q: be an inner product space. Replacing each value (x, y) of the
inner product by (x, y)' = - (x, y) we obtain an inner product space
Q:'. We shall say that Q:' is the anti-space of Q:.
Lemma 2.6. If the inner product space Q: contains at least one positive
(negative) vector, then every element of Q: is the sum of two positive (nega-
tive) vectors.
Proof. If X,X o E Q:, (xo, x o) > 0, then for sufficiently large positive
numbers (X the quantity
will be positive. Hence the vectors Xl = X +

(XX O, X 2 = - (xxo are
positive and we have x = Xl +

X 2• The other half of the lemma follows
by taking the anti-space of Q:. 0
Coronary 2.7. If the inner product is indefinite on Q:, then none of

the sets ~+, ~-, ~++, ~-- is a subspace. 0
3. Orthogonality 7
3. Orthogonality
If (x,y) = 0 or, equivalently, (y, x) = 0 for a pair of vectors x,y in the

inner product space @, we say that x and yare orthogonal to each other,
and we write x .1 y.
Two sets m, ~ c @ are said to be orthogonal (in symbols: 12{ .1 Q3), if
x.l y for every x E m, y E Q3.
The relation x.l y implies (x +y, x +
y) = (x, x) +
(y, y). Conse-
quently, we have the following result.
Lemma 3.1. If each of the pairwise orthogonal subspaces 21 , . . . ,2"
is positive (negative, neutral, positive definite, negative definite), then so
is the span <2v ... , 2n)· 0
Remark 3.2. Making use of Lemma 2.5 and equation (1.2)
Lemma 3.1 can be extended to orthogonal families of sets 12lv ... , n , m
m
where each i is either a subspace or a single vector.
In connection with Lemma 3.1 we introduce the following defini-
tions and notations.
If 2 is the vector sum (resp. direct sum) of pairwise orthogonal
subspaces 2; (i = 1, ... ,n), we say that 2 is the orthogonal sum
(resp. orthogonal direct sum) of the subspaces 2i and we write 2 =
=21 (+)", (+) 2n (resp. 2 = 21 (+)", (+) 2n)·
We now turn our attention to a concept which is the natural gener-
alization of orthogonal complement in Hilbert space, but has, as a rule,
less favourable properties.
For any set mc @ we put
m.L = {x E @: x .lmL
and say that m.L is the orthogonal companion of m.
It is clear that a) the orthogonal companion of any set is a subspace;
b) m c Q3 implies Q3.L C 12{.L; c) if 21 , 22 are subspaces of @, then
(21 + 22).1.. = 2f n 2f .
Denoting the second orthogonal companion (m.L).L of the set me @ by
12tH we have
(3.1) mcm H .
The case m= mH will be characterized in Chapter III.
The third orthogonal companion S)VH of m
always coincides
with 12l:.L:
(3. 2)
Really, taking the orthogonal companions of both sides of (3.1) we
obtain m.L ) 12l:.L H. On the other hand, replacing of m in (3.1) by m.L
yields m.L c s)FH.
8 I. Inner Product Spaces without Topology
Lemma 3.3. Ij £, we are subspaces ojrg and dim we = m < 00, then
dim (£ n weI) > dim £ - dim we .
Proal. We may assume that dim £ = l < 00, since in the opposite
case the result follows by considering subspaces of £ with increasing
finite dimensions. We may also assume that m > 1. l>
Let el , ... , el and Iv ... , I", be bases of £ and we,
respectively. The
elements of £ n weIare of the form aIel + ... +
aiel, where the
coefficients ai satisfy the equations
I
}; (ej' Ik)ai = 0 (k=1, ... ,m).
j=l
These are m homogeneous linear equations for l unknowns. Conse-

quently, the number of linearly independent solutions is at least
l-m. 0
Corollary 3.4. Ij £, we are subspaces of rg satisfying £ n weI = 0,

then dim £ < dim we . 0
Lemma 3.5. For every subspace we erg we have co dim weI < dim we.
Proal. If co dim weI
> n for some non-negative integer n, then by
Corollary 1.2 a subspace £ ( rg with dim £ = n, £nwe l = 0 can be
found. Applying Corollary 3.4 we obtain the relation dim >n. 0 we
We end this section by introducing one more notion related to
orthogonality.
Let rgv"" rgn be inner product spaces. Consider the set
rgl X··· X rgnofalln-tuples {xv . .. ,xn }, where xi E rgj (j = 1, ... , n).
The conventions
{Xi}~ + {Yi}~ = {xi + Yi}~ (Xi' Yi E rgi) ,
a {xi}~ = {axi}~ (a E C, xi E rgi) ,
n
( {xi }~, {Yi }~) = L' (xi' Yj) (xi' Yi E rgi)
i=l
turn rgl X ... X rgn into an inner product space, called the cartesian
product of the inner product spaces rgl> ... , rgw
It is easy to see that the elements {Xi}~ E rgl X· .. X rgn with
xi = 0 for j =F k form a subspace rg(k}, which is isometrically isomorphic
to rgk (k = 1, ... , n), and rgl X . . . X rgn is the orthogonal direct sum
of these subs paces :
rgl X ... X rgn = rg(l} (+) ... (+) rg(n} .
4. Isotropic Vectors 9
4. Isotropic Vectors
Example 4.1. Let ~ be a two-dimensional vector space with basis ev

e2 • Define an inner product on ~ by the relations (e l , e2 ) = 0, (e v el ) = 1,
(';2' e2 ) = - 1. It is easy to see that for the one-dimensional subspace
+ >
£ = (el e2 we have £.1. = £. This shows that a) a subspace may
have a non-zero intersection with its orthogonal companion; b) even
for a subspace of finite dimension the orthogonal companion need not
contain a complementary subspace.
Example 4.1 motivates the following definitions.
Let £ be a subspace of an inner product space~. The subspace
£ n £.1., to be denoted by £0, is called the isotropic part, and its elements
the isotropic vectors of £. If £0*0, we say that £ is degenerate, or that
the inner product is degenerate on £.
In particular, the whole space ~ is degenerate if the isotropic part
~o = ~.1. is not equal to 0.
The subspace £ of Example 4.1 is a degenerate subspace in a non-
degenerate space. The space ~ of Example 2.3 is degenerate if and
°
only if sf = for at least one j.
The next lemma is a simple consequence of the definition of iso-
tropic part.
Lemma 4.2. If £ = £1( +) ... (+) £", then for the isotropic parts
we have £0 = £~ (+) . . . (+) £~, . 0
Corollary 4.3. The orthogonal direct sum of non-degenerate s2tbspaces
is non-degenerate. 0
As isotropic vectors are orthogonal to themselves, the isotropic part
of any subspace is neutral. For this reason every definite subspace is
non-degenerate.
In Example 4.1 the vector el +
e2 is neutral, but it is not an iso-
tropic vector of ~ (however, it is isotropic for £). The following lemma
asserts that vectors of this kind may only occur in indefinite spaces.
Lemma 4.4. The isotropic part ~o of a semi-definite inner product
space ~ consists of all neutral elements of ~.
Proof. We have already mentioned that an isotropic vector is
always neutral. On the other hand, according to Lemma 2.2 the rela-
tions x E ~, (x, x) = °imply (x, y) = °
for every y E ~. 0
As a consequence, a neutral inner product on a vector space is
necessarily the trivial one:
Corollary 4.5. If ~ is a neutral inner prod'uct space, then (x, y) =°
for every pair x, y E~. 0
Corollary 4.6. If two vectors x, y of an inner product space satisfy

the relations (x, x) = 0, (x, y) =1= 0, then the subspace <x, y) is indef-
inite. 0
Lemma 4.7. If 2 is a neutral subspace, so is 211.

Proof. On account of Corollary 4.5 our statement is equivalent to
the following: 2 c 2.1. implies 21.L c (21.L).1.. But this implication
follows by a repeated application of property b) of the orthogonal
companion. 0
The next example shows that if 2 is positive, 2.L.L need not be
positive.
Example 4.8. Consider the vector space ~ of complex sequences

co
X = {~o, ~l> •.• } with 2.; l~il2 < 00. For y = {l7o, 171> ... } E ~ put
i=O
1 _
+ i=l
_ 00
(x, y) = - ~o 170 1.,' 1 ~i l7i .

2
One verifies easily that this inner product is non-degenerate and in-
definite on~. The subspace 2 c ~ characterized by the relation ~o =
= I, ~ ~i is positive definite. Let y = {l7i}~
i=l 21
E 2.1.. Since the element
with o-th coordinate equal to 1, n-th coordinate equal to 2n , and
°
x(n)
all other coordinates equal to belongs to 2 (n = 1,2, ... ), we have

(n = 1,2, ... ).
< 00 we obtain y =
00
In view of the relation 2-' Il7i l2 0. Thus 2.1. = 0,

2.L.L = ~ . j=O
It is also untrue that 21.L would be non-degenerate whenever 2

is so.
Example 4.9. Let, again, ~ be the vector space of square-summable
numerical sequences x = gi}~' For y = {l7i}~ E ~ put
co
(x, y) = - ~ol7o + 2-' ~il7i'
j=l
Denote by 2 the subspace of ~ consisting of all finite sequences {~i}~

00
,'lith ~o = ~1 = .1' ~i' Then 2.1. is the 1-dimensional subspace spanned

i=2
by the neutral vector Yo = {1, 1,0,0,0, ... }, and
2.L.L = {{~i}~ E ~: ~o = ~1}'
5. Maximal Non-degenerate Subspaces 11
It is easy to see that Yo is an isotropic vector of ,2LL though ,2 is

positive definite, hence non-degenerate.
A possible extension of Lemma 4.7 reads as follows.
Lemma 4.10. If,2 is a degenerate subspace, so is ,2.Ll.
Proof. Owing to the relatiuns (3.1), (3.2) we have:
,2 n,21 C ,2LL n,21 = £LL n,2.Lll . 0
5. Maximal Non-degenerate Subspaces

Non-degenerate spaces (subspaces) have several convenient properties
not shared by degenerate ones. With the aid of the following lemmas
many problems can be reduced to problems concerning non-degenerate
spaces (or subspaces).
Lemma 5.1. Let Q;0 be the isotropic part of the inner product space Q;,
and let Q;l be a complementary subspace for Q;0. Then Q;l is non-degenerate,
and we have Q; = Q;0 (+) Q;l.
Proof. The relation Q;0 ..1 Q;l follows from Q;0 ..1 Q;. Since Q;0 is the
isotropic part of itself, Q;l is non-degenerate by Lemma 4.2. 0
The contents of Lemma 5.1 can be expressed 'in another way if we
consider the quotient space Q;jQ;0. The latter has up to now been merely
a vector space, but it can be equipped with an inner product by setting
(x, x,
y( = (x, y), where y E Q;jQ;0, and x, yare representatives of and x
y, respectively: x,y E Q;, x Ex, Y E y.
Let xl>x2 Ex; Yl>Y2 E y. Then XI -X2 , Yl-Y2 E Q;0. It follows that
(Xl' YI) - (X2' Y2) = (X I -X2, Yl)+ (X2' YI-Y2) = O. Thus the defini-
tion of the function (. , .ris correct. It is easy to see that (. , .( is
really a'1 inner product on the vector space Q;jQ;0. For an inner product
space Q; the quotient space Q;jQ;0 will always be understood to carry
this inner product.
The inner product space Q;j2 can also be defined in the more general
case ,2 C Q;o, but not for an arbitrary subspace ,2 c Q;.
Lemma 5.2. If Q;l is a complementary subspace for the isotropic part
Q;O of Q;, then Q;l is isometrically isomorphic to the quotient space Q;jQ;0.
The isometrical isomorphism is given by the formula Tx = x (x E Q;l),
x
where E Q;jQ;0 is the equivalence class containing x.
Proof. According to Lemma 1.1 the mapping T serves as an iso-
morphism between Q;l and Q;jQ;0. It remains to note that, by defini-
tion, (Tx, Tyr = (x, yt = (x, y). 0
From Lemmas 5.1-5.2 we obtain:
Corollary 5.3. 03/030 is non-degenerate. 0

The complementary subspaces to 03° can also be characterized as
'maximal non-degenerate subspaces of 03, i.e. non-degenerate subspaces
which are not contained in any other non-degenerate subspace of 03.
Theorem 5.4. The set 031 C 03 is a complementary subspace for (to

if and only if 031 is a maximal non-degenerate sttbspace of 03.
Proof. Let 03 = 03° + 031 . By Lemma 5.1 031 is non-degenerate. If
.$3)031, .$3 =-1= 031, then the subspace .$3 n 03 0 is non-zero, and its elements
being orthogonal to the whole space 03, they are orthogonal to.$3; hence
.$3 is degenerate.
Conversely, assume that 031 is a maximal non-degenerate subspace
of 03. Then 031 n 03° = o. Let we be a complementary subspace for
@o + @1, i.e. @ = @o + @1 + we. On account of Lemma 5.1 the
subspace @1 + we is non-degenerate. The maximality of @1 implies
me = o. As a result, 03 = @o + @1. 0
Remark 5.5. From Theorem 5.4 and the existence of complemen-
tary subspaces, or by a direct application of Zorn's lemma, we obtain
that every inner product space contains maximal non-degenerate sub-
spaces. Moreover, every non-degenerate subspace admits a maximal
non-degenerate extension.
6. Maximal Semi-definite Subspaces
A subspace .$3 of the inner product space @ is said to be maximal semi-

definite, if .$3 is semi-definite and .$3 is not contained in any other semi-
definite subspace of @. The definitions of maximal positive, maximal
negative, max'imal neutral, maximal definite, maximal positive definite,
and maximal negative definite subspaces are similar.
By Zorn's lemma every semi-definite (resp. definite etc.) subspace
can be extended to a maximal one. Applying this statement to the
zero subspace we obtain that @ always contains maximal subspaces of
each kind introduced above.
Most of the following lemmas will be used later.
Lemma 6.1. If each of the subspaces .$31,.$32 C 03 is either maximal

positive or maxitnal negative, then (.$31, .$32)0 consists of the neutral elements
of .$31 n .$32 .
Proof. Let x E .$31 n .$32' (x, x) = O. The vector x is a neutral element
of the semi-definite subspace .$31' Hence, by virtue of Lemma 4.4,
x E :£31 n:£3t. In the same way we obtain the relation x E :£3 2 n:£3,t. Thus
+
x E (:£31 +
:£3 2) n (:£31 :£3 2) 1..
Let, conversely, x E (:£31> :£3 2)°. It is evident that x is a neutral
vector orthogonal to :£31 and :£3 2 , Let :£3 1 be maximal positive (maximal
negative). According to Remark 3.2 the span <x,
:£31) is a positive
(resp. negative) extension of :£3 1 , Therefore xE:£31 • The relation xE:£3 2
follows in the same manner. 0
Lemma 6.2. If(§; = :£31 + :£3 2 with a positive subspace:£3 1 and a nega-
tive definite subspace :£3 2 , then :£31 is maximal positive and :£3 2 is maximal
negative definite. For negative :£31 and positive definite :£3 2 a similar
statement holds.
Proof· For any subspace :£3~):£31 (:£3~ *

:£31) we have :£3~n:£32 o.
Thus :£3~ cannot be positive. Analogously, if :£3~ is a subspace which is a
*
proper extension of :£3 2 , then :£3~ has a non-zero intersection with the
positive subspace :£31 , so that it cannot be negative definite.
If :£31 is negative and :£3 2 is positive definite, we consider the anti-
space of (§;. 0
Lemma 6.3. The orthogonal companion of a maximal positive
(maximal negative) s1,f,bspace is negatl:ve (Positive).
Proof. Let:£3 be a maximal positive subspace. Suppose that for
some x E :£31. we have (x, x) >
o. Then, by Remark 3.2, (x,:£3> is a
positive extension of:£3. Moreover, it is a proper extension, since x E :£3
would imply x E :£30, (x, x) = O. This contradicts the maximality of :£3.
If :£3 is maximal negative, we consider the anti-space. 0
As a positive subspace is not necessarily positive definite and a
maximal positive definite subspace is not always maximal positive
(namely it may happen to have positive extensions without having
positive definite extensions), the next lemma neither implies nor is
implied by the preceding one.
Lemma 6.4. The orthogonal companion of a maximal positive definite
(maximal negative definite) s2fbspace is negative (positive).
The proof is similar to that of Lemma 6·3· 0
Remark 6.5. The orthogonal companion of a maximal positive or
maximal positive definite subspace need not be maximal negative. For
instance, the subspace :£3 of Example 4.8 is maximal positive and maxi-
mal positive definite on account of Lemma 6.2, but :£31. = 0 is not
maximal negative.
Finally we mention a result related to Lemma 6.3.
Lemma 6.6. The orthogonal companion of a positive definite, 1naximal

positive (negative definite, maximal negative) subspace is negat1've definite
(Positive definite).
Proof. Let 2 be a positive definite, maximal positive subspace. By
°
Lemma 6.3 21. is negative. If x E 21., (x,x) = 0, x oF then, in view
of Remark 3.2, <x, 2) is a proper positive extension of 2 contrary to
the assumption. 0
7. Maximal Neutral Subspaces
Maximal neutral subspaces do not play such an important role in

further developments as, say, maximal positive ones. Nevertheless,
they have some interesting properties which deserve to be mentioned.
Lemma 7.1. If 2 is a maximal neutral subspace, then 21. is semi-
definite and its isotropic part coincides with 2.
>
Proof· Suppose that Xl> X 2 E 21., (Xl' Xl) 0, (X2' x 2)<0. According
to Lemma 2.1 the two-dimensional subspace <Xl' %2) contains a non-
zero neutral vector Xo' We have Xo ~ 2, since otherwise X 2 would be an
element of the subspace <Xl' ,\2), which is positive by Remark 3.2.
Therefore <xo, ,\2) oF 2. Furthermore, owing to the same Remark 3.2,
<xo, 2) is a neutral subspace. This contradicts the maximality of ,\2.
Consequently, 21. is semi-definite. Replacing X o by any neutral element
of 21. in our reasoning we find that Xo must belong to 2. Now the asser-
tion 21. n2l.l. = 2 follows from Lemma 4.4. 0
Lemmas 7.1 and 4.7 yield necessary conditions in order that a
neutral subspace be maximal neutral. It turns out that both conditions
together are sufficient too.
Theorem 7.2. The neutral subspace 2 C ~ is maximal neutl'al if and

only if ,\21. is semi-definite and ,\2J.J. = ,\2.
Proof. If,\2 is maximal neutral, then ,\21. is semi-definite by Lemma
7.1, whereas the relation 2J.J. = 2 follows from Lemma 4.7. Now let
2 be a neutral subspace with 21. semi-definite and 2J.J. = 2. Let,\21 be
a neutral subspace, 2 1 ) 2. Corollary 4.5 implies 2 ~ ,\21 i.e. ,\21 C ,\21..
Applying Lemma 4.4 to the semi-definite subspace 21. we obtain the
relation 21 ~ 21.. Hence 2 1 c2J.J. = 2 and, consequently, 21 = 2. 0
The following theorem asserts that a maximal neutral subspace is
either maximal positive or maximal negative (or both maximal positive
and maximal negative).
Theorem 7.3. If a maximal neutral subspace £ c @ admits a positive
proper extension, then .53 does not admit any negative proper extension.
Proof. Suppose there exist a positive subspace £1 ( @ and a negative

subspace £2 c @ that contain £ properly. Let Xl E£1> Xl E! £. By Lem-
ma 4.4 we have £1 ~ £. In particular, Xl ~ £. Therefore, in view of
Remark 3.2 and the maximality of £, the vector Xl must be positive.
<
Similarly, if X 2 E£2' x 2 E! £, then x 2 E £1, (x2 , x 2) o. This contradicts
Lemma 7.1. 0
If a maximal neutral subspace £ has neither positive nor negative
proper extensions, then 5] is said to be a hypermaximal neutral subspace.
Theorem 7.4. The subspace £ c @ is hypermaximal neutral if and
only if £1 = £.
Proof. Let £ be hypermaximal neutral. Corollary 4.5 yields
£ C£1. To prove the inclusion in the opposite direction, let X E £1 be
arbitrary. X can be positive, negative, or neutral. In each case the
span (x, £) is semi-definite (see Remark 3.2). Therefore the hyper-
maximality of £ implies x E £.
Conversely, assume that £1 = £. Then £ ~ £, hence £ is neutral.
Denote a semi-definite extension of £ by £1. On account of Lemma 4.4
£ is orthogonal to £1. Thus £1 c £1 = £. 0
The subspace £ appearing in Example 4.1 is hypermaximal neutral.
On the other hand, there are inner product spaces that do not contain
hypermaximal neutral subspaces. For instance, in a one-dimensional
definite inner product space 0 is the only neutral subspace, and it is
not hypermaximal neutral.
8. Projections of Vectors on Subspaces
Let £ be a subspace and x an element of the inner product space @. If

x = y + Z, where y E £ and Z E £1, we say that y is a projection of x
on £.
Neither the existence, nor the uniqueness of the projection of a
vector on a subspace is guaranteed. For instance, in Example 4.1 the
elements of £ have infinitely many projections on £ (namely every
Y E £ is a projection), while the vectors outside £ have none.
The problem of uniqueness can easily be settled.
Lemma 8.1. Two projections of the vector x E@ on the subspace

£ c @ differ in an arbitrary isotropic vector of £.
Proof· Let x = 'Yj + Zj' where Yj E 2, zi E 21 (j = 1,2). Then

)'1-Y2 E 2 and YI-Y2 = Z2 - Zl E 21. Thus )'1-)'2 E 2°.
On the other hand, if x = )'1 + ZI' where Yl E 2, Zl E 2 1 , then for
every 1t E 2° we have x = (Yl + +$&) (ZI-1t), where Yl+Zt E 2,
ZI-1t E 21. 0
Corollary 8.2. If there is an element in @ having exactly one pro-

jection on 2, then 2 -is non-degenerate. If 2 is non-degenerate, every
element of @ has at most one projection on 2. 0
The question concerning the existence of projections is more diffi-
cult. We are going to treat it in a few special cases.
Theorem 8.3. Let 2 be a ne1ttral s1tbspace. The vector x admits a pro-

jection on 2 if and only if x ~ 2. If this condition is satisfied, every
element of 2 is a projection of x on 2.
Proof. By Lemma 8.1 and Corollary 4.5 x has a projection on 2 if
and only if every element of 2, in particular the zero vector, is a pro-
jection of x on 2; i.e., x = 0 + z, where 0 E 2, Z E 21. The proof is
complete. 0
Theorem 8.4. Let 2 be a positive definite s1tbspace of the inner prod$tct

space @. The element x E @ admits a projection on 2 if and only if the
j1tnction T(Y) = (x-y, x-y), considered for y E 52, attains its minim1tm
at some Yo E 2:
(8.1 ) (x-Yo, x-Yo) = min (x-Y, x-y) .

yE,t!
The element Yo def£ned by (8.1) is 1tniq1te and it is the projection of x on

2. For a negative definite 2 a similar condition with max in place of min
holds.
Proof. Let 2 be positive definite. If x = Yo + z, where Yo E 2,
Z E 2 1 , then for every Y E 52, Y =1= Yo, we find:
(x-Y, x-y) = (z+Yo-y, z+Yo-Y) =
= (z, z) + (Yo-Y, Yo-Y):> (z, z) = (x-Yo, x-Yo)'

If, conversely, for some Yo E 52 the condition (8.1) is fulfilled, then for
every y E 52, A E C it follows that
Hence
A (y, x-Yo) + A (x-Yo, y) + IAI2 (y, y);;:; 0 .
Setting j, = ~(x-Yo, Y), where f1, runs through all non-zero real
f1,
numbers, we obtain:
2f1, !(x-Yo, y)!2 + !(x-Yo, y)!2 (y, y) > °

Thus (x-Yo, y) = 0, x-Yo..l 2 .
The case of a negative definite 2 can be reduced to the preceding
one by passing to the anti-space of~. 0
Theorem 8.5. Let the subspace 2 C ~ be the orthogonal direct sum

ot n subspaces:
(8.2) 2 = 21 (+) . . . (+) 2n .
The vector Y E 2 is a profection ot the vector x E ~ on the subspace 2 ij and
only ij
(8·3) y = Y1 + ... + Yn ,
where yjis a projection of x on 2j (j = 1, ... ,n). In particular, x has
a profection on 2 it and only if it has a profection on each 2j (j = 1, ... , n).
Prooj. Suppose that Y is a projection of x on 2. Then x = Y z, +
where yE2, zE21. Let (8.3) be the decomposition of y corresponding
to (8.2). Using the notation
(8.4) Z,:. = I: Yk
k*j
(j = 1, .. .,n)
we have x = Yj+(z+zj), where Yj E 2 j , z+z/ E 27 for every J.

Thus Yj is a projection of x on 2j"
Let, conversely, x = Yj+Zj' where Yj E 2 j' Zj E 2f for every f.
Then with the notations (8.3), (8.4) we obtain x = Y+(Zj-zj), where
Y E 2, Zj-z/ E 2f for every j. Since Zj-zj = X-Y does not depend
on j, it is orthogonal to every 2j and, in view of (8.2), it is orthogonal
to 2. 0
Theorems 8.3 - 8.5 enable us to find the projections of any vector x

on a subspace 2 that can be decomposed into the orthogonal direct sum
of a neutral, a positive definite, and a negative definite subspace. As
we shall see later (Example 11.3), not every 2 belongs to this class.
Nevertheless, it is always possible to represent 2 as the orthogonal
direct sum of a neutral and a non-degenerate subspace (d. Lemma 5.1).
Therefore the problem of findipg projections on 2 can always be
reduced to the case where 2 is non-degenerate.
9. Ortho-complemented Subspaces
In the last section we were looking for projection.,; of different vectors x
on a fixed subspace 3. One may also raise the following question: what
are the conditions on the subspace 3 in order that every x in the space
~ would have at least one projection on 3? Clearly, a necessary and
sufficient condition is <3, 3 1 ) = ~.
If a subspace 3 of the inner product space ~ satisfies the condition
<3,3 1 ) = ~, we shall say that 3 is ortho-complemented. In other words,
the subspace 3 C ~ is ortho-complemented, if it admits an orthogonal
complementary subspace, i.e. a complementary subspace contained in
3 1 . For a non-degenerate 3 this happens if and only if 3 (+) 3 1 =~.
Lemma 9.1. If 3 is ortho-complemented, so is 3 1 .
Proof. Since 3 c 3 11 , the relation <3, 3 1 ) = ~ implies
<3 1 ,3 11 ) =~. D
We remark that 3 1 may be ortho-complemented without 3 being
so. An example is a non-closed subspace in Hilbert space.
The following result is a special case of Theorem 8.5.
Lemma 9.2. Let 3 be the orthogonal direct sum of n subspaces: 3 =
= 3 1 (+) ... (+) 3 n . Then 3 is ortho-cmnplemented if and only if each
3 j (j = 1, ... ,n) is so. D
Lemma 9.3. A neutral subspace 3 C~ is ortho-complemented if and
only if 3 c ~o.
Proof. According to Corollary 4.5 we have 3 C 3 1 . Therefore the
property <3, 3 1 ) = ~ is now equivalent to 3 1 = ~ or, what is the
same, to 3 ~~. D
Theorem 9.4. The following two conditions are necessary and suffi-
cient in order that the subspace 3 C ~ would be ortho-complemented:
a) 3 0 C ~o ;
b) 3jBo is ortho-complemented in ~/~o.
Proof. Let 3 be ortho-complemented in~. Then, in view of Lem-

mas 5.1 and 9.2, also 3° is ortllo-complemented. Thus the relation
a) is a consequence of Lemma 9.}.
To prove b) we first remark that 313° is isometrically isomorphic to
x
the subspace {i c ~/~o consisting of those classes E ~/~o which contain
at least one element of 3. This can be seen as follows. Let xJ2" E 313°.
Let x E 3 be a representative of XJ2o, and let x E ~/~o be the class
containing x. On account of a) the mapping XJ2' --). x is single-valued.
Moreover, it is one-to-one, since 3 n~O = 3°. Finally, it preserves
sums, scalar multiples, and inner products.
9. Ortho-complemented Subspaces 19
Next we show that ,2 is ortho-complemented in ~/~o. Let xE ~/(J;o.

For XEX, XE~ we have a decomposition
(9.1) x= y + z; Y E,2, z 1- ,2 .
Therefore
(9.2)
where YEY, ZEZ.
A Let, conversely, a) and b) be satisfied (the latter in the sense that
,2 is ortho-complemented in ~/~O). Then for any x E @ the respective
x
class E @/~o admits a decomposition of the form (9.2). Choosing a
representative y E ,2 of y and an arbitrary representative Z of z we
obtain relation (9.1). 0
Corollary 9.5. In a non-degenerate space @ every ortho-complemented
subspace is non-degenerate. 0
According to Theorem 9.4 and Corollary 5.3, in the study of ortho-

complemented subspaces it is sufficient to concentrate upon non-
degenerate subspaces of non-degenerate inner product spaces. In many
cases (namely when the subspace under consideration is the orthogonal
direct sum of a positive definite and a negative definite subspace)
Lemma 9.2 makes possible a further reduction: we may restrict our
attention to definite subspaces of non-degenerate spaces. Since most
of the results known for this situation are of a topological nature, they
will be given in subsequent chapters.
Here we mention only some simple facts which can be obtained
without topological tools.
Lemma 9.6. If ,2 is an ortho-complemented subspace of the non-
degenerate space Q;, then ,211 = ,2.
Proof. Our assumptions imply the relation ,2 (+) ,21 = @. Conse-
quently, if ,211 is a proper extension of ,2, then ,211 n,21 oF O. On the
other hand, by Lemma 9.1 and Corollary 9.5 the subspace,21 cannot be
degenerate. 0
It is well known that in Hilbert space the condition ,211 = ,2 is not
only necessary, but also sufficient for a subspace ,2 to be ortho-comple-
mented. In an arbitrary non-degenerate inner product space both of
the conditions ,211 =,2, ,2 n,21 = 0 are necessary (d. Lemma 9.6 and
Corollary 9.5), but, as the following example shows, even together they
are not sufficient for the ortho-complementedness of ,2.
Example 9.7. Consider the space @ of Example 2.3 with Cj = (- 1)i
(f = 1,2, ... ). Thus @ is the vector space of complex numerical
00
sequences x = gv ~2' •.. } satisfying }; l~jl2 < 00, and the inner product
is defined by the relation j=\
00
(x, y) = }; (- 1)1 ~j1jj'

i=1
00
where y = {1]v 1]2' ... }, }; l1] j l2

j=1
< 00. Set
2= {{~i}~ E @: ~2j = 2i ~ 1 ~2i-l (i = 1,2, ... ) }.
The subspace 2 is positive definite. In particular, 2° = O. Further-

more,
°
~1= { {1]'1}oo E\\!-'1]2j=--.-1]2i-1
{c. 2i-1 (._
1- 1,2, ... )} ,~
011_0
-~.
1 21
But 2 is not ortho-complemented. Really, suppose that the element
z = {Cv C2 , ••• }; C2j-l = 0, C2i = ~ (1 = 1, 2, ... )

21
can be written in the form z = x + y,
where x = {~j}~ E 2,
y = {1],.}~ E 21. Then an easy ca1cu1atlOn · . lds '>2j-l
yIe [: 2i - 1
= ---''--.--
00 41 - 1
(i = 1, 2, ...), so that };1~jI2 = 00, contradicting the definition of @.
i=1
Lemma 9.8. Every definite subspace of finite dimension is ortho-

complemented.
Proof. If 2 is an n-dimensional positive definite subspace of the
inner product space @, then we can find a basis ev ... , en for 2 which
is orthonormal: (e j , ek ) = (jjk (i, k = 1, ... , n). Given a vector x E @
n
it is easy to verify that L' (x, ej)ei is the projection of x on 2.
i=1
The case of a negative definite 2 can be reduced to the above one
by considering the anti-space of @. 0
10. Dual Pairs of Subspaces
In Section 5 it was shown that every subspace 2 of an inner product

space @ is the orthogonal direct sum of the isotropic part 2° and a non-
degenerate subspace 21. In certain problems, e.g. those concerning
intrinsic properties of 2 or the restriction of an operator to 2, the iso-
10. Dual Pairs of Subspaces 21
tropic part can be neglected or easily handled, and it remains to con-

sider the non-degenerate component. In other problems, e.g. when a
decomposition of (t corresponding to 2 is required, such a neglection is
unjustified, and one would like to proceed by extending 2° to a non-
degenerate subspace. Sometimes, mainly if dim 2° <
co, this can be
done with the aid of so-called dual companions of 2°.
We will say that the subspaces 2, me ( (t form a dual pair, or that 2
and 9)( are dual companions for each other (in symbols: 2 # me), if
2 n 9)( 1 = 21 n 9)( = 0. In other words, two subspaces of the inner
product space (t form a dual pair if there is no non-zero vector in any of
them that would be orthogonal to the whole of the other.
Every non-degenerate subspace is a dual companion for itself. In
Example 4.1 the neutral subspace 9)( = (el - e2) is a dual companion
for the neutral subspace 2 = (el +e2 ).
Lemma 10.1. If 2 # 9)(, where 2 or 9)( is ne1ttral, then 2n9)( = 0,

+
and 2 9)( is non-degenerate.
Prool. Let 2 # 9)(, where e.g. 9)( (9)(1 (d. Corollary 4.5). Then
2(9)((2n9)(1 =0. Further let x+y (xE2, YE9J~) denote an
isotropic vector of 2+9)(. From the relations x+y E 9)(1, y E 9)(1 we
deduce x E 9)(1. Therefore x E 2(9)(1, that is, x = 0. Thus x+y E .8 1
yields y E 2 1 , Y E 21 n9)(, y = 0. 0
In order to get a converse of this statement we must require that

both of the subspaces 2, 9)( be neutral.
Lemma 10.2. If 2(9)( = ° and (2+9)()0 = 0, where 8, 9)( are

neutral subspaces 01 (t, then 2 # 9)(.
Proal· On account of Corollary 4.5 we have 2 (2 1 , 9)( ( 9JU.
Therefore the elements of 2 n9)(l as well as the elements of 21 n me
are isotropic vectors of 2+9)(. 0
Lemma 10.3. If 2 # 9)(, then dim 2 = dim 9)(.

Proal. The conclusion follows from Corollary 3.4· 0
When studying dual pairs of finite-dimensional subspaces it is
often convenient to introduce bases. In this connection we need the
following definition.
Two systems {ei}~ ((t, {/i}~ ( (t will be said to form a d#al pair,
or to be d1fal companions of each other, if (e i , I,,) = 0ik (j,k = 1, ... ,1'1).
Let the systems {ei}~' {/i}~ form a dual pair. The following asser-
tions can immediately be proved: a) the vectors el , ••• , en are linearly
independent and the vectors 11' ... , In are also linearly independent,
b) the subspaces <e1 , • • • , en>, <II> ... , In) form a dual pair. Thenext
lemma enables us to show that a converse of b) holds too.
Lemma lOA. Let 2 be an n-dimensional (1 < n <
(0) subspace of Ii,
and let we c Ii be a subspace satislying 2 n we.!. = O. There exist two
systems {ei}~' {fi}~ fOY'ming a dual pair such that <e1 , • • . , en) = 2,
<11' ... , I,,) c we.
Proal. Choose a basis {gi}~ of 2, and put e1 = gl' As e1 (f we.!., one
can find an 11 in we with (el> 11) = 1.
Suppose that two dual companions {ei}~' {li}~ (k <
n) have already
been constructed, where <e1 , ••• , ek) = <gl' ... , gk) and Ii E we
(j = 1, ... ,k). Put
k
eH1 = gH1 - 2,' (gH1' Ii) ei .
i=1
Then <e1 , · · • , eH1) = <gl' ... , gH1), (eH1' Ii) = 0 (j = 1, ... ,k).
Furthermore, since eHI (f we.!., there is a vector hH1 E 9)l satisfying the
relation (eH1' hH1 ) = 1. We set
k
IH1 = hk+l - 1; (hHj , ei ) Ii'
i=1
and verify easily that IH1 E 9)1, (/H1' ei ) = 0j,k+1 (j = 1, ... , h+1).
In the n-th step we obtain the desired dual pair. 0
Corollary 10.5. If 2, 9)1 are subspaces of Ii such that 1 < dim 2 =
= dim we< 00 and 2 nwe.!. = 0, then some basis 012 forms a dual pair
'I1:'£th some basis 01 9)1. 0
From property b) of dual pairs of systems it follows that under the

assumptions of Corollary 10.5 we have 2 # 9)e. Therefore in verifying
that two finite-dimensional subspaces 2, we are dual companions one
of the requirements 2 nwe.!. = 0, 2.!. n9)1 = 0 may be replaced by the
equation dim 2 = dim we.
Combining Corollary 10.5 with Lemma 10.3 we obtain a result on
bases of dual pairs of finite-dimensional subspaces. However, this
result can be improved.
Lemma 10.6. Let 2 # we, where 2,we are non-zero subspaces of
finite dimension. Then lor any basis {ei}~ 01 2 some basis {/i}~ 01 9)1
serves as a dual companion.
Proof. Denote by 2k (k = 1, ... , n) the span of all vectors ej with
j:# /,. On account of Lemmas 3.3 and 10.3 we have
dim (wen2t) > 1 (h=1, ... ,n).
10. Dual Pairs of Subspaces 23
"*
Let Ik E wen)31, Ik O. As Ik cannot be orthogonal to the whole of)3,
it is not orthogonal to ek • Hence we may assume that (e k , fk) = 1.
Consequently, (e j , Ik) = dik (j = 1, ... , n), and this is true for every k.
It follows that Ik (k = 1, ... , n) are linearly independent; thus they
form a basis of we. 0
Lemma 10.7. In a non-degenerate inner productspace~ every finite-
dimensional subspace has a dual companion.
Prool. Let)3 be a subspace of @ with dim)3 = n 00. If n = 0, <
then we = 0 is a dual companion for)3. If n >
0, we may apply Lem-
ma 10.4 to )3 and ~. Thus we obtain a basis {ei}~ of )3 and a system
{/i}~ ( ~ which form a dual pair. The span </1> ... , In> will be a dual
companion for)3. 0
The following lemma can be applied in some cases as a substitute
for Lemma 10.1.
Lemma 10.8. If)3 # we, where )3 and we are finite-dimensional
subspaces 01 the inner product space ~, then )3 we 1 = ~. +
Prool. If)3 = 0, then necessarily we = 0, hence the assertion is
true. Otherwise we make use of Corollary 10.5 in order to obtain a
basis {ei}~ of)3 and a basis {fi}~ of we such that (ei,l k ) = dik (J',k = 1,
... ,n). For every x E @ we set
n
x.\! = L (x, Ij) ei '
j=1
and verify immediately that x.\! E)3, x-x.\! E we 1. 0

By way of application we consider a partial improvement of Lem-
ma 3.5.
Theorem 10.9. For every subspace we of a non-degenerate inner prod-
uct space @ we have codim we 1 = dim we.
Prool. If dim we < 00, then the statement follows from Lemmas
10.7,10.8,10.3 and Corollary 1.2. If dimwe = 00, let we! denote an
n-dimensional (n < 00) subspace of we. Applying the part just proved
of our theorem we obtain the relation codim = n. Thereforewet
codim we1 :::: n. This being true for every n, it follows that codim we 1 =
= 00. 0
Theorem 10.10. If )3 is a finite-dimensional subspace of a non-
degenerate inner product space, then )311 = £.
Prool. According to Theorem 10.9 and relation (3.2) we have:
dim )311 = codim )3111 = codim )31 = dim)3. It remains to observe
that )3 C)311 (see (3.1)). 0
11. Fundamental Decompositions
The inner product space (g is said to be decomposable, if it can be

represented as the orthogonal direct sum of a neutral subspace (go, a
positive definite subspace (g+, and a negative definite subspace (g-:
(11.1) (g = (gO(+)(g+(+)@-; @o c 1.l50, @+ C 1.l5++, @- C 1.l5--.
Every decomposition of the type (11.1) is called a fundamental decom-
position of @.
We first justify the use of the symbol @o for the neutral component.
Lemma 11.1. If @o, (g+, @- are subspaces satisfying (11.1), then @{}
coincides with the isotropic part of @.
Proof. This is a consequence of Lemmas 4.2 and 4.4. 0
Corollary 11.2. Every fundamental decomposition of a non-degenerate
inner product space (g is of the form
(11.2) (g=@+(+)@-; @+cl.l5++, @-cl.l5--. 0
Not every inner product space is decomposable.
Example 11.3. Denote by @ the vector space of complex numerical
sequences x = {~i}i=-co that are finite from the left: ~i = 0 for
i < io(x). Define an inner product on @ by the formula
co
(11.3) (x, y) = .1' ~J'-i-l'
i=-oo
where y = {1]i r~oo E @ .
It is easy to see that this inner product is non-degenerate. There-
fore, according to Corollary 11.2, every fundamental decomposition of
@ must be of the form (11.2).
Consider the linear mapping {~i} -»- {~i} of (g to (g, where ~i = ~i
for i < 0, and ~i = 0 for i > O. The restriction of this mapping to
any definite subspace is one-to-one. Really, if x = {~i} -»- 0, then
~-1 = ';-2 = ... = 0 so that, in view of (11.3), (x, x) = O. Thus every
definite subspace of @ is isomorphic to some part of the subspace of @
characterized by the relations ~i = 0 U > 0). In particular, if (11.2)
holds, then (g+ and (g- are isomorphic to certain subs paces oUhe vector
space of finite numerical sequences. On the other hand, (g is by defini-
tion isomorphic to the vector space of all infinite sequences. Therefore
(11.2) leads to a contradiction.
Lemma 11.4. Let 53 be a positive definite (negative definite) subspace

of the inner product space (g. In order that @ have a fundamental decom-
position of the form (11.1) with positive definite component @+ = 53
11. Fundamental Decompositions 25
(negative definite component ~- = ~) the following two conditions are
necessary and sttfficient:
a) ~ is maximal positive definite (maximal negative definite);
b) ~ is ortho-complemented.
Proof. Suppose that ~ admits a fundamental decomposition (11.1)
with positive definite component ~+ =~. Since, owing to Lemma 3.1,
+
the subspace ~O( )~- is negative, Lemma 6.2 guarantees that ~ is
maximal positive definite. Condition b) is a consequence of the rela-
+ .
tion ~.i = ~O( )~-
Let, conversely, ~ be an ortho-complemented maximal positive
definite subspace of~. Then ~ = ~(+ )~.i, since ~ is non-degenerate.
By Lemma 6.4 ~.i is negative. ApplyingLemma 5.'1 to~.i and taking
Lemma 4.4 into account we obtain an orthogonal decomposition
+
~.i = [J"(O( )9](1, where 9](0 is neutral and 9](1 is negative definite. Thus
~!-=~, ~o =9](0, ~-= 9](1 make up a fundamental decomposition of~.
If ~ is negative definite, we consider the anti-space of ~. D
Corollary 11.5. An inner product space is decomposable if and only
if it contains ortho-complemented maximal definite subspaces. D
Remark 11.6. If ~ is non-degenerate and admits a fundamental
decomposition (11.2), then Lemma 6.2 yields not only that ~+ is
maximal positive definite, but also that it is maximal positive. Simi-
larly, ~- is maximal negative in this case.
Lemma 11.4 indicates that the fundamental decomposition need

not be unique. In Example 4.1, for instance, anyone-dimensional
positive definite subspace can play the role of ~+. Making use of this
example one verifies easily that, in fact, the fundamental decomposi-
tion of a decomposable indefinite inner product space is never unique.
Invariant properties of fundamental decompositions will be dis-
cussed in Chapter IV.
The next theorem singles out an important class of decomposable
spaces.
We say that the inner product (. , .) is qttasi-positive (quasi-negative)
on ~, or that ~ is a quasi-positive (quasi-negative) inner product space,
if ~ does not contain any negative definite (positive definite) subspace
of infinite dimension.
Theorem 11.7. Every quasi-positive (quasi-negative) inner product
space is decomposable.
Proof. Let ~ be quasi-positive (in the opposite case we consider the
anti-space of Q;). Let ~ be a maximal negative definite subspace of ~.
Then dim 2 < 00. It remams to apply Lemma 9.8 and Corollary
1'1.5. 0
Corollary 11.8. Every inner product space of finite dimension zs
decomposable. 0
From Corollaries '11.8, 11.2 and Lemmas 9.2, 9.8 we obtain the
following strengthening of Lemma 9.8.
Corollary 11.9. Every finite-dimensional non-degenerate subspace oj
an inner product space is ortho-complemented. 0
Notes to Chapter I
Most of the results collected in this chapter are natural extensions of

classical theorems concerning quadratic forms in finite-dimensional
spaces. During the past decades some of the latter have repeatedly been
reformulated within the framework of the axiomatic theory of vector
spaces, and included in textbooks of linear algebra (see e.g. Greub [1],
Mal'cev [lJ).
The systematic study of the geometry of special indefinite inner
product spaces of infinite dimension began with the works of Pontr-
jagin [lJ, Hestenes [1], Nevanlinna [1J - [4J, and those of 1ohvidov
and KreIn [lJ, [2J. Greub and Hestenes consider inner products on real
vector spaces, but this difference is irrelevant for the questions treated
by them.
The first comprehensive treatments of general inner product
spaces are due to Scheibe [1J and to Ginzburg and 1ohvidov [1]. Our
exposition follows mainly these two papers. The article of Ginzburg
and 1ohvidov contains a survey, also some criticism, of earlier results.
The space of Example 2.3 with ci = -1 (j;:£ x), ci = 1 (j> x)
has been considered by Pontrjagin [1], while a special case of Example
2.4 appears in a note by 1ohvidov [2J. The facts contained in Lemma
2.6 and Corollary 2.7 have been observed by KreIn and Smul'jan [1].
Lemma 4.7 was proved by Scheibe [1]. Example 4.8 belongs to
H. Langer (personal communication).
The contents of Section 5 are taken from the paper of Scheibe [1 J.
A special case of Lemma 6.1 was obtained by Bognar [6].
Lemma 7.1 and Theorem 7.2 are due to Scheibe [1]. Theorem 7.3
was proved by Phillips [3J, and independently by Ginzburg and 1ohvi-
dov [1]. Theorem 7.4 belongs to 1ohvidov (see Ginzburg and 1ohvidov
[1 J).
The results of Section 8 were found by Nevanlinna L4].
Notes to Chapter I 27
Theorem 9.4, though evident, seems to be formulated for the first

time. Example 9.7 is, in fact, merely an illustration to a theorem
characterizing ortho-complemented definite subspaces in a particular
type of inner product spaces (see Theorem V.S.2 below).
Lemma 10.1 (for tvvo neutral subspaces) and Lemma '10.2 were
obtained by Iohvidov and KreIn [1J, and independently by Scheibe [1].
For a special case of Lemma 10.8 see Iohvidov and KreIn [2].
Example 11.3 (with real scalars) is due to G. W. Mackey, and cited
by Savage [1]. Lemma 1 'l.4 appears in the paper of Ginzburg and
Iohvidov [1].
We mention that Mal'cev [1J, Aronszajn [1J, Ginzburg and Iohvidov
[1J permit the inner product to be non-hermitian. Fischer and Gross
['IJ consider bilinear forms on vector spaces over fields different from
Rand C and give references to earlier work.
An extension of the Gram-Schmidt orthogonalization process to
indefinite inner product spaces will be given in Section IV.3.
Chapter n. Linear Operators in Inner Product Spaces
without Topology
In Sections 2- 3 algebraic properties of isometric and symmetric operators of

general inner product spaces are examined. Sections 6-7 constitute the common
foundation for the study of plus-operators (an extension of the class of isometric
operators; Section 8) and Pesonen operators (a subclass of symmetric operators;
Section 9). Sections 4- 5 on Cayley transformations and Sections 10-11 on
fundamental decompositions serve applications in subsequent chapters. Special
attention should be payed to Theorems 2.5, 3.3, 6.2, 9.5, 10.1, and 11. 7.
1. Linear Operators in Vector Spaces
Before considering linear operators in inner product spaces, we collect

some notations and definitions concerning linear operators in vector
spaces.
A linear operator T from a vector space @l to a vector space @2 is a
mapping of a subspace'!J(T) c @1 into @2 such that
T ((Xl Xl + (X2 X 2) = (Xl TXl + (X2 TX2 ((Xl> (X2 E C; Xl' x 2 E '!J(T)) .
The subspace '!J(T) is called the domain of T. We say also that Tis
defined on '!J(T). The set ffi(T) = T '!J(T), evidently a subspace of @2'
is the range of T. If @l = @2 = @, we say that T is a linear operator
in @.
If T is a linear operator, then dim ffi( T) is called the rank of T.
A linear form is a linear operator T with ffi(T) c C.
The identity operator I of the vector space @ is defined by the rela-
tion Ix = X (x E @). If confusion can arise, the underlying space will
be indicated by writing If§; instead of I.
The zero operator 0 of @ is defined by Ox = 0 (x E @), where the
symbol 0 stands for an operator on the left-hand, but for a vector on
the right-hand side. (This will be the fourth meaning of the character
0.)
Let T be a linear operator from @l to @2' The set
SJl:(T) = {x E '!J(T): Tx = O}
is a subspace of @1' termed the null space of T.
1. Linear Operators in Vector Spaces 29
If 5JC(T) = 0, then T is said to be invertible. In this case one can
define a linear operator T-1 with SD(T-1) = ffi(T), setting T-l y = x
if and only if Tx = y. The operator T-1 is called the inverse of T.
If T is an invertible operator in @, whereas SD(T) = ffi(T) = @, we
shall say that T is completely invertible.
Let T be a linear operator in @. If for a A E C the operator T - AI
is not invertible, we say that A is an eigenvalue of T. The subspace
5JC(T - AI) is called the eigenspace, its non-zero elements the eigen-
vectors of T belonging to A.
°
The vector x -::/:= is a principal vector of T belonging to the eigen-
value A, if there is a positive integer r such that x E SD( T') and
(T - AI)' x = °.
The span of all principal vectors of T belonging to A is the principal
s2tbspace S;.(T). Equivalently,
co
S;.(T) = .U 5JC((T - AI)i) .
7=0
In particular, 5JC(T - AI) c S;.(T). If}. is not an eigenvalue of T, we

set S;.(T) = °.
The dimension of S;.(T) will be denoted by m;JT) and called the
algebraic mul#plicity of A as an eigenvalue of T.
An eigenvalueA of T is said to be semi-simple, if S;.(T) = 5JC(T -JlI),
i.e., if every principal vector of T belonging to A is an eigenvector.
If xES;.(T), and p 21 is the least exponent such that (T -H)Px =
= 0, then the vectors
xi = (T - AI)P-l- i x (j = 0, 1, ... ,p - 1)
are linearly independent and belong to S;.(T); in particular, Txo = AXo .
We say that x o, Xl' ••• ,xP_ 1 form a Jordan chain of T to the eigen-
value A. Equivalently, the system {Xi }t- 1 c @ is a Jordan chain of the
operator T corresponding to the eigenvalue A, if Xo -::/:= and °
TXi = AXi + Xi-l (j = 0, 1, ... , p - 1; X-I = 0) .
The positive integer p is called the length of the Jordan chain.
It is well known that the principal subspaces of a linear operator T
are linearly independent. Moreover, if SD(T) = @, ffi(T) c @, and
0< dim @ < 00, then a) @ is the direct sum of the principal subspaces
of T; b) each principal subspace S;.(T) has a basis consisting of Jordan
chains of T to the eigenvalue A; c) the set of the lengths of these Jordan
chains is uniquely determined by the operator T.
A subspace .$3 ( @ is said to be invariant for the linear operator T,
if T(.$3n~(T)) c .$3. We say also that.$3 is an invariant subspace of T.
Principal subs paces are a particular type of invariant subspaces.
30 II. Linear Operators in Inner Product Spaces
The restrict'ion TI2 of the operator T to the subspace 2 has, by

definition, domain 2 n '!l(T), its values there coinciding with those
of T. An eigenvalue of TI2 is also called an el:genvalue of Tin 2.
Two operators T I , T2 defined on the vector space 03 and having range
in 03 are said to commute, if T]T2 = T 2T I . The members of an operator
family are said to commute, if they commute in pairs.
The following two results are well known and will be referred to
later.
Lemma 1.1. If two linear operators commute, then every eigenspace-

of the one is an invaria1tt subspace of the other.
Proof. Denoting the two operators by Tl and T2 we find that the
relation (TJ.-AI)x=O implies (Tl-AI)T2X=T2(TI-JeI)X=0. 0
Lemma 1.2. Let 03 be a vector space with 0 < dim 03 co. If <
T 1, ... , Tn are commuting linear operators on 03, then they have a com-
mon el:genvector.
Proof. Restricting Tv ... , T n - 1 to an eigenspace of Tn and making
use of Lemma 1.1 the problem can be reduced to the case of n - 1
operators. Thus the conclusion follows by induction. 0
Let T be a linear operator in 03. Suppose that the subspace 2 c 03

is invariant for T. On the quotient space '!l(T)/2 one can define a
linear operator Tin the following way: for E '!l(T)/2 letx Tx
= (Txt,
where x is a representative of the class X. We say, T
is the operator
induced by T in 03/2 .
A direct decomposition 03 = 21 + ... +
2" is said to reduce the
linear operator T, if 1) the components 21> ... , 2" are invariant sub-
spaces of T, and 2) '!l(T) = ('!l(T) n2]) + ... +
('V(T) n2 n ). In this
case T is called the direct sum of the linear operators T121' ... , TIB".
Let the vector space 03 be the direct sum of n subspaces:
(1.1 ) 03 = 21 + . . . + 2" .
If T is a linear operator in 03 such that
'V(T) = ('V(T) nBl) + ... + ('V(T) nB,,)
then, with respect to (1.1), T can be represented by an operatur matrix
(Tik );: k=I' The entries Tik are linear operators from 2k to Bj , respec-
tively, defined by the relations
n
(1.2) Yj = E Tikx k (f = 'I, ... , n) ,
k=1
2. Isometric Operators 31
where x = Xl + ... +
xn and Tx = Yl + ... +
Yn are the decom-
positions of the vectors X E '1J(T) and Tx corresponding to (1.1). In
the case x = xk (1.2) reduces to 'Yj = Tikxk , giving a more explicit
definition of T ik .
2. Isometric Operators
Let (;l; be an inner product space with inner product (. ,.). A linear
operator U in Q; is said to be iso111etn:c, if (Ux, Uy) = (x, y) for every
pair x, y E '1J(U).
If (;l; is decomposable, and for some fundamental decomposition
(;l; = (;l;O(+)(;l;+(+)(;l;-; (;l;o c ~o, (;l;+ C ~++, (;l;- c ~--
the linear operator U is the direct sum of an operator in (;l;o, an iso-
metric operator in (;l;+, and an isometric operator in (;l;-, then, as it is
easy to see, U is isometric. From Example 2.3 below it turns out,
however, that even in a decomposable space not every isometric oper-
ator can be obtained in this way.
In contrast to the situation in Hilbert space, an isometric operator
in an inner product space need not be invertible; a trivial example is
the zero operator on a neutral inner product space (;l; O. For an *-
invertible isometric operator U the inverse U-l is obviously isometric.
Lemma 2.1. A n isometric operator U transforms isotropic and no}t-
1·sotropic elements of '1J(U) into isotropic and non-isotropic elell1ents
of m(U), respectively.
Proof. Let x E '1J(U). If (x, y) = 0 for every Y E '1J(U), then
(Ux, Uy) = 0 for every Uy E m(U). If (x, y) *- 0 for some y E '1J(U),
then (Ux, Uy) *- 0 for the corresponding Uy E m(U). 0
Corollary 2.2. If the domain of an isometric operator U is non-
degenerate, then U is invertible. 0
The condition appearing in Corollary 2.2 is not necessary for in-
vertibility; e.g. the restriction of an invertible isometric operator U
to a degenerate subspace 2 C '1J( U) remains to be invertible.
We are going to study the eigenvalues and principal subspaces of
isometric operators.
In Hilbert space, as it is well known, an isometric operator has only
unimodular, semi-simple eigenvalues. (We say that the number ex E C
is unimodular, if lex I = 1.) In an indefinite inner product space this is
not necessarily true, not even in the most "regular" case of an iso-
metric operator that maps a finite-dimensional non-degenerate space
onto itself.
Example 2.3. Consider the two-dimensional vector space ~=<e, f).

Define an inner product (. , .) on ~ by the relations
(e, e) = (I, f) = 0, (e, f) = i ,
and a linear operator U in ~ by
Ue=e, Uf=e+f.
We have (U -I)f =F 0, (U -1)2f = 0; so the eigenvalue v = 1 of U
is not semi-simple. Nevertheless, U is easily verified to be isometric.
Example 2.4. In the inner product space of Example 2.3 we define
a linear operator U by the equations
Ue = -1 e, Uf = 2f.
2
Then U is an isometric operator with eigenvalues VI = 1/2, V2 = 2.
The Hilbert space theorem on the orthogonality of eigenvectors
belonging to different eigenvalues fl, v of an isometric operator remains
valid for the principal vectors of an isometric operator in an arbitrary
inner product space under the assumption that fl and v do not lie sym-
metrically with respect to the unit circle.
Theorem 2.5. Let U be an isometric operator in the inner product
space~. If fl, V are eigenvalues of U such that fl v =F 1, then the principal
subspace ®,.(U) is orthogonal to ®v(U).
Proof. Let
(U-flI), x = 0, (U-vIr Y = O.
We must show that (x, y) = O. For r + s = 0 this is evident. Assum-
ing the validity of the same conclusion for r + s < n, we prove it
for J' + s = n.
Introduce the notations
(U-flI) x = Xl' (U-vI) Y = YI'
As the cases where r = 0 or s = 0 can be neglected, we have
(U -flI),-1 Xl = 0, (U _vI)S-1 YI = 0 .
Hence, by the assumption,
(x, YI) = (Xl> y) = (Xl' YI) = 0.
Making use of these relations we obtain
(x, y) = (Ux, Uy) = (flX + Xl' vy + Yl) = flv (X, y) .
The theorem is proved. D
Coronary 2.6. The principal subspace belonging to a non-unimodular
eigenvalue of an isometric operator is neutral. D
For certain classes of isometric operators it is known that the set of

eigenvalues is symmetric with respect to the unit circle. Here we con-
sider only the simplest case.
Theorem 2.7. Let03 be a finite-dimensional, non-degenerate inner prod-
uct space. Let U be an iso'metric operator in 03 with ~(U) = 03. If v
is an eigenvalue of U, then the number v*=1!1i is also an eigenvalue of U.
Moreover, ~v(U) # ~v.(U) .
*
Proof. Owing to Corollary 2.2, v cannot be 0; thus v* is defined.
*
Denoting the direct sum of all principal subspaces ~J'(U) with f-l v,
f-l v* by 2.(U), we have
+
03 = (~v(U) ev'(U)) 2.(U) . +
Here, according to Theorem 2.5, the last component is orthogonal to
the first and second ones. (Note that the case Ivl = 1, ~v(U) = ~.*(U)
is not excluded.) Applying Lemma 1.4.2 we obtain that the subspace
+
6.(U)
*
ev'(U) is non-degenerate. This proves the assertion for
Ivl = 1, while for Ivl 1 the desired conclusion follows from Corollary
2.6, Lemma 1.10.2, and Lemma 1.10·3· 0
The next theorem complements the above one by showing that the
structure of UI~v(U) and UI~p'(U) is the same.
Theorem 2.8. Let 03 be a finite-dimensional, non-degenerate inner
product space. Let U be an isometric operator in 03 with '1)(U) = 03, and
let v be an et'genvalue of U. If the basis {fi}~ of ~.*(U) is a dual companion
for the basis {ei}~ of ev(U), then the matrices of UI@3v(U) and UI(5•• (U)
relative to {ei}~ and {Ii }~, respectively, are the inverses of the transposed
conjugates of each other.
Proof. The existence of the dual pair {ei}~' {fi}~ follows from
Theorem 2,7 and Lemma 1.10,6, Let
n
Uei = I:" 8ri er , Ufi = 2..' Cf!rifr (f = 1, ... ,n) .
1=1 7=1
Making use of the relations

(Uei , Uf k ) = (e i , f,,) = 0ik (j, k = 1, ... , n)
we obtain
n
2..'8'i "(Prk = 0ik (j, k = 1, ... , n). 0
7=1
The following result is a simple extension of a property of isometric

operators in Hilbert space.
Lemma 2.9. Let U be an isometric operator in the inner product
space 03. If 2 is a subspace of 03 such that U(2 n ~(U)) ) 2, then 21 is
invariant for U.
Proof. Let x E £1 n'!l(U), Y E £. The assumption guarantees

the existence of a YI E £n'!l(U) with UYI = y. Hence (Ux, y) =
= (Ux, UYI) = (x, YI) = o. 0
Finally, we mention a sufficient condition for an eigenvalue of an
isometric operator to be semi-simple.
Lemma 2.10. Let v be an eigenvalue of the isometric operator U. If
the eigenspace m(U - vI) is definite, then the eigenvalue v is unimodular
and semi-simple.
Proof. If v is non-unimodular then, by Corollary 2.6, m(U - vI)
is neutral. Next take Ivl = 1. Suppose that (U - vI)' x = 0,
(lJ - vI),-1 x::/= 0 for some vector x and integer r > 2. Then Xl =
= (U - vI)'-1 X is a non-zero element of m(U - vI), and we have
(xv Xl) = ((U - vI),-1 x, (U - vI),-1 x) =
= - ((U - vI)' x, vU(U - vlY-2 x) = 0 . 0
3. Symmetric Operators
A linear operator A in the inner product space ~ is said to be sym-

metric if (Ax, y) = (x, Ay) for every pair x, Y E '!l(A).
If
~ = ~o (+) ~+ (+) ~-; ~o C ~o, ~+ C ~++, ~- C ~--
is a fundamental decomposition of~, and the linear operator A is the

direct sum of an operator in ~o, a symmetric operator in ~+, and a
symmetric operator in ~-, then A is a symmetric operator in the space
~. Nevertheless, there are symmetric operators in decomposable spaces
which cannot be obtained in this way for any fundamental decomposi-
tion.
Example 3.1. Let {e, /} be a basis of the inner product space ~,
where
(e, e) = (I, f) = 0, (e, f) = 1 .
The linear operator A defined by the relations
Ae = e, Af = e+f
is symmetric, but no fundamental decomposition of ~ reduces A.
Example 3.1 shows also that the eigenvalues of a symmetric opera-
tor in an inner product space need not be semi-simple. The next
example exhibits a symmetric operator with non-real eigenvalues.
3. Symmetric Operators 35
Example 3.2. We define the space and the inner product as in
Example 3.1 and observe that the linear operator specified by the
relations
Ae = ie, Aj = - if
is also symmetric.
The analogy between isometric and symmetric operators known
from Hilbert space theory remains valid in inner product spaces. For
the ~ake of completeness, we are going to formulate the analoga of
the assertions 2.5 -2.8 of the preceding section. The proofs will be
omitted, since the only new point is the replacement of the charac-
teristic property of isometric operators by that of symmetric ones.
Theorem 3.3. Let A be a symmetric operator in the inner product

space @;. If A, fl are eigenvalues oj A such that J, *-
ii, then 5,,(A)is ortho-
gonal to 51'(A). 0
Corollary 3.4. The principal subspace belonging to a non-real eigen-

val-ue of a symmetric operator lS neutral. 0
Theorem 3.5. Let @; be a finite-dimensional, non-degenerate inner

product space. Let A be a symmetric operator in @; with :tJ(A) =@;. I j A
is an eigenvahte 0/ A, then A is also an eigenvalue oj A. lVIoreover,
5;JA) # 5;1.(A). 0
Theorem 3.6. Let @; be a finite-dimensional, non-degenerate inner

product space. Let A be' a symmetric operator in @; with :tJ(A) =@;, and
let A be an eigenvalue oj A. I j the basis {ji}~ oj 5:a(A) is a dual c01npanion
jar the basis {ei}~ oj 5;.(A), then the matrices oj AI5;.(A) and AI5:a(A)
relative to {ej}~ and {fi}~' respectively, are the transposed conjugates of
each other. 0
The next result corresponds to Lemma 2.9.
Theorem 3.7. If A is a symmetric operator in the inner product space
52 is an invariant subspace oj A s14ch that 52 c :tJ(A), then 52 1 is
@;, and
also an invariant subspace of A.
Prooj. For x E 52 1 n:tJ(A), y E 52 we have (Ax, y) = (x, Ay) = o. 0
We also state the analogue of Lemma 2.10; in the proof obvious
changes are only needed.
Lemma 3.8. Let A be an eigenvalue of the symmetric operator A. If
m(A - Jel) is dejinite, then the eigenvalue }. is real and
the eigenspace
semi-simple. 0
II. Linear Operators in Inner Product Spaces
Let, again, @ be an inner product space, and let A be a symmetric

operator in @ with ~(A) = @. Set
(3.1) (x, y)A = (Ax, y) (x, y E @) .
Then (. , ')A is also an inner product on @. Vve shall refer to it as the
A -inner product on @.
When we consider @ as an inner product space with inner product
(. , ')A rather than (. , .), we use the symbol @A' In @A one can intro-
duce all notions defined for inner product spaces. These notions ~hen
considered in the original space @ will be distinguished by a prefix A,
and the corresponding symbols by a sUbscript A.
For instance, a vector x E @ is said to be A-positive, if it is a positive
vector of @A' and we set
~~+ = {x E @: (Ax, x) > 0 or x = O} .
Similarly we can speak of A-positive subspaces, A-orthogonality (to be
denoted by -LA)' A-orthogonal companion of a set In (in symbols: In~),
A-isot1·opic part of a subspace ,13 (in symbols: B~), A-orthogonal sums
and A-orthogonal direct sums (the respective symbols being (+)A and
(+ )A)' A-fundamental decompositions (with components @~, @1, @:4),
quasi-A-posdive spaces, A-isometric and A-symmetric operators etc.
The following simple fact is often useful.
Lemma 3.9. Let A be an everywhere defined symmetric operator in
the 1:nner prOdtlct space @. If the s1tbspaces B,W( ( @ are orthogonal to
each other, 'while one of them is 1:nvariant under A, then ,13 and We are
A -orthogonal to each other.
Proof. The relations B.l We, AB ( ,13 imply AB -L me
i. e.
B-LA We. 0
An important subclass of symmetric operators is that of orthogonal
projectors.
An orthogonal projector in an inner product space @ is a symmetric
operator Pin @ such that ~(P) = @, p2 = P. This definition is justi-
fied by the following theorem.
Theorem 3.10. If P is an orthogonal projector in the inner product
space @, then B1(P) is ortho-complemented, and for every x E @ the vector
Px 1:S a p1·ojection of x on B1(P). If, conversely, ,13 is an ortho-complement-
ed s1tbspace of the non-degenerate inner prod'uct space @, and P B is the
mapping that carries each vector oj @ into its projection on ,13, then P B is
an orthogonal projector with ffi(P B) = ,13 •
Proof. Let P be an orthogonal projector in Q;. Then for every
x, y E @ we have
(x - Px, Py) = (Px - p2X, y) = 0 .
3. Symmetric Operators 37
Let, conversely, ct be non-degenerate, and let £ be an ortho-com-

plemented subspace of ct. Then P E = P is everywhere defined,
single-valued (Corollary 1.8.2), and satisfies the relation (x - Px,
Py) = 0 (x, Y E ct). Thus
(x, Py) = (Px, Py) (x, y E ct) .
Interchanging x and y and taking complex conjugates we obtain:
(3·3) (Px, y) = (Px, Py) (x, y E ct) .

From (3.2) and (3.3) it follows that P is symmetric. Further, the sym-
metry of P along with equation (3.3) yield:
(3.4) (P2X, y) = (Px, Py) = (Px, y) (x, y E ct) .
The space ct being non-degenerate, (3.4) implies p2 = P. 0
To finish with, we mention an interesting property of commuting
orthogonal projectors.
Theorem 3.11. Let PI> P 2 be commuting orthogonal projectors in the
inner product space ct. If ffi(Pl ) and ffi(P 2 ) are positive subspaces, then
so is their span ffi(Pl ) +
ffi(P 2 ) •
Proof. For any Xl E ct the orthogonal decomposition PIXI =

= P 2 P l Xl + (1 -P 2 )PI X l (see Theorem 3.10) yields
(PIXl , PlXl ) = (P 2 P l X I , P 2 P l X l ) + ((1- P 2 )PI X I , (1 - P 2 )PI X l ).
Hence, making use of the relations P I P 2 = P 2 P1 , Sl(P1 ) C q5+, we
obtain:
(3.5)
Similarly, for every X2 E ct we have
(3.6)
Furthermore, a simple calculation shows that
(3.7) (PI X 1 ,P2 x 2 ) = (PI P 2X I , P I P 2X 2) ,
(3. 8) (P2X 2 , PIXI ) = (PI P 2X 2 , P I P 2X I ) •
Adding up the relations (3.5) - (3.8) we find:

(3·9) (PIXI +P 2X 2 , PIXI +P 2X 2 ) >
;;:; (P PI +x2 (X I 2) , P 1 P 2 (X l +x 2 )) •
Since ffi(Pl ) C q5+, the right-hand side of (3.9) is non-negative. 0

4. Cayley Transformations
The analogy between isometric and symmetric operators in an inner

product space finds its explanation in some kind of correspondence
between these two classes of operators. The correspondence in question
is closely related to the Cayley transformation kno'\vn from Hilbert
space theory.
Lemma 4.1. Let A be a sym'll1etric operator in the inner product

space Q:. Let e be a unimodular number, and ( a non-real number which
is not an eigenvalue of A:
-
(4.1 ) lei =1, (#- (,
(4.2) 91(A - U) = O.
Then the operator
(4·3 ) U = e(A - U)(A - (1)-1
is isol11eiric, has domain
(4.4) ~(U) = ffi(A - U) ,
and I' is not an eigenvalue of U:
(4.5) 'i)c(U - el) = 0 .
Let, conversely, U be an iS01netric operator in Q:. Suppose that the
numbers 1', (E C satisfy the conditions (4.1), (4.5). Then the operator
A = ((U - EU)(U - E1)-l
is symmell'ie, satisfies (4.2), and its domain is
( 4.7) ~(A) = ffi(U - el) .
The transformations (4.3) and (4.6) are the inverses 01 each other.
Proof. Consider a symmetric operator A and two numbers e, (

satisfying the conditions (4.1), (4.2). We observe that the definition
(4.3) makes sense and can be written in the equivalent form
(4.8) x = Af - (I, Ux = eAf - E(f,
where /is any element of ~(A), while x is an arbitrary element of ~(U).
Setting -
y = Ag - (g, Uy = eAg - E(g
we obtain:
(4.9) (Ux, Uy) - (x, y) = (' - ()((Af, g) - (j, Ag)) .
Therefore U is isometric. The relations (4.4), (4.5) also follow directly
from (4.8), taking into account (4.1).
In order to find the inverse transformation we solve the equations

(4.8) for f and Af :
1 1 -
(4.10) f = ---.:::.- (Ux - .sx) , AI =---_- (CU x - .sCx) .
.s(C - C) .s(C - C)
Thus (4.6) is the expression for A in terms of U.
Moreover, if U is any isometric operator in Q; with the property
(4.5), then (4.6) or (4.10) define a linear operator A in Q; and, (4.8) and
(4.10) being equivalent, U can be expressed in terms of A by the for-
mula (4.3). Consequently, (4.9) even now holds for any X,YE~(U)
and f,g E ~(A). Therefore the isometry of U implies the symmetry
of A. Relations (4.2), (4.7) follow easily from (4.10). 0
Remark 4.2. If A is a non-symmetric linear operator, then U will
be a non-isometric linear operator, and conversely. The other state-
ments of Lemma 4.1 do not involve the inner product and remain valid
for linear operators A and U in an arbitrary vector space.
If two linear operators A and U in a vector space are related by the
formulas (4-3) and (4.6) (the conditions (4.1), (4.2), (4.5) being satis-
fied), we say that A and U are Cayley transforms of each other. The
mutually inverse transformations (4.3), (4.6) are called Cayley trans-
formations.
Let Q; be an inner product space, and let .s, C be fixed numbers
satisfying the conditions (4.1). Consider the class of all symmetric
operators A in Q; such that Cis not an eigenvalue of A, and the class of
all isometric operators U in Q; such that .s is not an eigenvalue of U.
Lemma 4.1 states that the Cayley transformations (4.3), (4.6) induce
a one-to-one correspondence between these classes. However, a sym-
metric operator A may have several Cayley transforms belonging to
different values of 8 and C. It may also happen that A has no Cayley
transforms at all, since every non-real number is an eigenvalue of A.
Similarly, an isometric operator U admits an infinite number of Cayley
transforms provided there is a unimodular .s which is not an eigenvalue
of U; otherwise it admits none.
In certain situations the existence of a Cayley transform can be
guaranteed. If this is not the case or is difficult to establish, indepen-
dent proofs for the isometric and symmetric versions of a theorem are
justified.
5. Principal Vectors of Cayley Transforms

In this section we examine the behaviour of eigenvalues and principal
subspaces under a Cayley transformation.
Lemma 5.1. Let A be a linear operator in a vector space. Let A be

an eigenvalue 01 A, and 1 a princ£pal vector 01 A belonging to A:
(5.1) 1E ~(A'), (A - AI)'! = 0, (A - AI),-I 1"* °.
Let U be a Cayley transform of A, defined by (4.3) with parameters 8, C
satisfying the conditions (4.1) - (4.2). Set
A-C
(5.2) v = 8 A_ C'
(5·3) x = ex'(A - U)'! ,

where
(5.4)
Then v is an eigenvalue of U, and x is a principal vector of U belonging

to v such that
(5.5) x E ~(U'), (U - vI)' x = 0, (U - vI)'-1 x*"o .
jY[ oreover, we have
(5.6) <x, (U - vI)x, ... , (U - VI)'-I x) =
= <f, (A - AI)!, ... , (A - AI),-I f) .

Proof. First observe that the definition of ex is correct because of
the assumptions (4.1). The definition of x is justified by the inclusion
fE'I:J(Ar) (see (5.1)). Also v is properly defined, since A is, while Cis not
an eigenvalue of A (d. (5.'1) and (4.2)). For the same reason ex"* 0 .
From equations (4.3), (5.2), (5.3), (5.4) for s = 0, 1, 2, ... , r we
obtain:
(U - vI)Sx =
= ex r [8(A - [I) (A - C1)-1 - 8(1. - C)(A - Ct 1 I]'(A - Ut f =
= ex'8 S(A - Ct s [(A - C)(A - [1) - (A - [)(A - U)]S (A - UY-S! =
= ex' lOs (A - C)-S (C - C)S (A - AI)S (A - Uy- s f =
= ex r - s (A - UY-S(A - AI)S f .
Consequently, (5.1) implies (5.5). One also sees that (U - vI)Sx is a
linear combination of the vectors f, Af, A 2 f, ... ,A'-If (s = 0, 1,
... , r - 1). Therefore the left-hand member of the relation (5.6) is
contained in the right-hand member. But, owing to (5.1) and (5.5),
both members have dimension r. Thus (5.6) holds. 0
The counterpart of Lemma 5.1 reads as follows.

Lemma 5.2. Let U be a linear operator in a vector space. Let 'I' be an

eigenvalue of U, and x a principal vector of U belonging to 'I' sttch that the
relations (5.5) are valid. Let A be a Cayley transforln of U, defined by
(4.6) with parameters e, /:; satisfying the conditions (4.1), (4.5). Set
(5.7) }.=/:;Y-/:;8,
'1'-8
(5.8) f = (3r(u - sI)'x,

where
y-e
(5.9) {3=-----.
e (I:; - 1:;)
Then /\ is an eigenvahte of A, and f is a principal vector to /\ satisfying
(H). Moreover, relation (5.6) holds.
The proof proceeds along the same lines as that of Lemma 5.1. 0
Remark 5.3. The transformations (5.2) and (5.7) are the numerical
analogues of the Cayley transformations (4.3) and (4.6), respectively.
Therefore, according to Lemma 4.1 and Remark 4.2, they are inverse
to each other. Making use of, say, (5.2) and (4.3), one verifies that (5.3)
and (5.8) are mutually inverse transformations too.
As a consequence of Lemmas 5.1- 5.2, the finite-dimensional in-
variant subspaces of a linear operator and its Cayley transforms coin-
cide. More precisely, we have:
Lemma 5.4. Ass~tme that the linear operators A and U in the vector
space Q; are related by the Cayley transformations (4.3), (4.6), where
lei = 1, I:; =F- 1:;. Let 53 c Q; be a subspace of finite dimension. Then the
relations 53c~(A), A52 c 53 imply the relations 52(~(U), U52 c 52, and
mce versa.
Proof. Let 53 c ~(A), A52 c 52. Then 52 can be decomposed into a
direct sum
where
52 j = <fi , (A - /\/)fi , ... ,
(A - /\/)';-1 fi ) (j = 1, ... ,10)
with some fjEQ;, }'iEC and positive integer r i satisfying the relations
(A - A/),; Ii = 0, (A - /\/)';-1 Ij =F- 0 .
By Lemma 5.1 we have 52 j c ~(U), U52 j c 52 j •
The second half of the lemma can be proved in a similar way,
making use of Lemma 5.2. 0
Another consequence of Lemmas 5.1- 5.2 is that the principal sub-

spaces of an operator coincide with those of its Cayley transforms.
Lemma 5.5. Let A and U be linear operators in a vector space, related
to each other by the Cayley transformations (4-3) and (4.6), where lei = 1,
C-=1= C· Let.le and 'I' be eigenvalues of A and U, respectively, related by
the formulas (5.2) and (5.7). Then ®;,(A) = ®,,(U).
Proof. Consider a non-zero vector fE6;.(A). For a suitable positive
integer r it satisfies the conditions (5.'1). With the notations and by the
conclusion of Lemma 5.1 we have:
f E (x, (U - 'I'I)x, ... , (U - 'I'Ir- 1 x) c ®,,(U).
Thus ®A(A) c CS,,(U). The complementary inclusion can be obtained
from Lemma 5.2 in a similar way. 0
6. Pairs of Inner Products: Semi-boundedness
In this section and the next one we consider the mutual behaviour of
two inner products (.,.) and (., ')1 defined on the same vector space 0;,
when (. , ')1 satisfies a sign condition on the neutral set of (. , .). The
results have applications in the special case (x, yh = (Tx, Ty)
(x,y E 0;) as well as in the case (X,Y)l = (Ax, y) (x,y E 0;), the operator T
(resp. A) being linear (resp. symmetric) and defined everywhere in 0;.
Throughout the present section we assume that (.,.) is
an indefinite inner product, while (., .h is an arbitrary inner
product on 0:.
Lemma 6.1. If (x, x) = °implies (x, X)1 ~ 0, then
(y, Y)l < (z, Z)l
(y, y) = (z, z)
for every pair y,z E 0; with (y, y) < 0 , (z, z) > 0 .
Proof. If the assertion is false, there exist two elements y, Z E 0;
such that
(y, y) = - 1, (z, z) = 1, - (y, yh (z, Z)1 . >
Setting x = e y + z, where lei = 1, we obtain:
(x, x) = 2 Re {e (y, z)} ,
(x, X)1 < 2 Re {e (y, z)d .
Thus for a suitable e we have (x,x) = 0, (x,xlI < ° in contradiction
with the hypothesis. 0
6. Pairs of Inner Products: Semi-boundedness 43
Theorem 6.2. If (x, x) = °implies (x, X)1 > 0, then
(6.1) (x, X)1 2 [ll(X, x) (x E @) ,
where
(6.2) [11 = inf (z,

(z,z) = 1
Z)1 >- 00.
Proof. Lemma 6.1 and the indefiniteness of (. , .) (required through-

<
out the section) yield - 00 [11 < 00. As to the relation (6.1), for
vectors x with (x, x) >°
it follows from the definition (6.2), for (x, x)
°
= it coincides with the assumption, while for (x, x) it is a conse- <°
quence of Lemma 6.1. 0
Theorem 6.2 can be given the following loose formulation. If the

inner product (. , ')1 is semi-bounded on the neutral set of the indefinite
inner product (. , .), then it is selui-bounded with respect to (. , .) on
the whole space.
Theorem 6.3. If (x, x) = ° implies (x, xh = 0, then for some real

number [11 we have
(x, Y)1 = [ll(X, y) (x,y E @) .
Proof. According to Theorem 6.2,

(x, X)1 > [ll(X, x) , (x, x)~ = -(x, X)1 > [I~(x, x)
for suitable real numbers [11' [I~ and every vector x. By addition we
obtain ([11 + [I~) (x,x) :s ° for
every x. Since (.,.) is indefinite,
[I~ = - Thus (X,X)1 = [ll(X,X) identically in x. It remains to apply
[11'
the polarization formula (1.2·3)· 0
The next result is a modification of Lemma 6.1, and will be needed

later.
Lemma 6.4. If the relations (x, x) = 0, x i=- °imply (x, xlI > 0, then
(y, Y)l < (z, Z)l
(y, y) (z, z)
for every pair y,z E @ with (y, y) < °, (z, z) > °.

The proof can be modelled on that of Lemma 6.1, keeping in luind
that a positive and a negative vector are alwayslineadyindependent. 0
7. Pairs of Inner Products: Sign
The results contained in this section are more or less evident from the
intuitive picture of the neutral set of an inner product as a surface
separating the positive and negative elements of the space. The sym-
bols (. , .) and (. , ')1 will denote two arbitrary inner products on the
vector space @;.
Lemma 7.1. Set, as usually,

(7.1) ~oo..:- {XE@;: (x,x)=O, x-=l-O} ,
and assume that x E ~oo implies (x, X)1 ::/= 0. Then either (x, xh > °for
every x E ~OO, or (x, xh <°
for every x E ~OO.
Proof. Let y, z be two vectors satisfying the conditions

(7.2) (y, y) = (z, z) = °,
(7·3) (y, yh = - 1, (z, Z)1 = 1.
For x = ey + o:z, lei = 1, 0: E H, we find
(x, x) = 20: Re {s(y, z)} ,
(x, X)1 = - 1 + 20: Re {s(y, Z)I} + 0: 2 •
°
Thus for a suitable S and every 0: we have (x, x) = 0, and by a sub-
sequent choice of 0: also (x, xh = can be achieved. Since (7.3) guaran-
tees that x -=I- 0, our result contradicts the assumption. Therefore
equations (7.2)-(7.3) cannot hold simultaneously. 0
Lemma 7.2. Suppose that x E ~oo implies (x, xh > 0. Then either
(y, y) ° °
< implies (y, yh > 0, or (z, z) > implies (z, Z)1 > 0.
Proof. If the conclusion is false, there exist two vectors y, z such
°,
that
(y, y) = - 1 , (y, y)l <
(z, z) = 1 , (z, zh < °.
Setting x = sy + z, where \s\ = 1, we obtain:
x -=I- 0, (x, x) = 2 Re {s(y, z)}, (x, xh < 2 Re {s(y, z)d .
Thus for a suitable s we have x -=I- 0, (x, x) = 0, (x, xh < 0, contrary to
the assumption. 0
Lemma 7.3. Suppose that the relation (x, x) = °

implies (x, xh > 0.
Then either (y,y)<O implies (y,y)1 > 0, or (z,z»O implies (z,zh::::::: 0.
8. Plus-operators 45
The proof is a slight modification of the preceding one. Another

possibility is to notice that the assertion follows immediately from
Theorem 6.2 in the case of an indefinite (. , .), while for semi-definite
(. , .) it is trivial. 0
We finally mention a result where the conclusion rather than the
assumption is concerned ,vith the neutral set of (. , .).
Lemma 7.4. If the inner product (. ,.) is not negative semi-definite
and the relation (z, z) >
0 impl/es (z, Z)1 2: 0, then (x, x) = 0 i'mpl£es
(x, XlI > o. Similarly, if (.,.J is not positive semi-definite and the
relation (y,y) <
0 implies (Y'Y)1 > 0, then (x,x) = 0 implies (x, xlI:::2: O.
Proof. Suppose that some vector x satisfies the equations

(7.4) (x, x) = 0, (x, X)1 = - 1 .
Then choosing an element z with
(z, z) = 1 , Re (x, z) :::2: 0
and setting z" = x + o;z (0; E R), we find:
(z"" z"') = 2 0; Re (x, z) + 0;2 ,
(z"" Z,,.)1 = - 1 + 20; Re (x, zh + 0;2(Z, Z)1 .

Hence for sufficiently small positive values of 0; we obtain (z"" z"J > 0,
(za' z",h < o.
From assumption (7.4) one also derives the existence of a vector y'"
with the properties (y"" Ya) 0, (Ya' Y",)l< o. 0 <
Lemmas 7. '1-7.4 can be formulated in terms of inclusion relations
between the sets ~o, ~+ etc. on the one hand, and ~~, ~t etc. on the
other, the latter sets being defined by the aid of the inner product (. , .h.
8. Plus-operators
In this section we assume that the inner product (.,.) is

indefinite on the space @.
A linear operator T in the indefinite inner product space @ is said
to be a plus-operator, if it is defined everywhere in @ and carries non-
negative vectors into non-negative ones: T~+ c ~+.
By Lemma 7.4 applied to the inner products (x, y) and (x, Y)1 =
= (Tx, Ty) (x, y E @) it is sufficient to require that positive vectors
would be transformed into non-negative ones: T~++ C ~+.
Isometric operators U with 'l;(U) = Q: are examples of plus-

operators.
The most important property of plus-operators reads as follows_
Theorem 8.1. Let T be a plus-operator in Q:. Then
(8.1) (Tx, Tx) > p(T) (x, x) (x E Q:) ,
where
(8.2) ,u(T) = inf (Tx, Tx) .
(,",,")=1
Proof. In Theorem 6.2 we put (x, Y)l = (Tx, Ty) (x,y E Q:). 0
For every plus-operator T the number p(T) defined by (8.2) is
non-negative. If p(1'»O, we say T is a strict plus-operator. In the
opposite case (i.e., when p(T) = 0) T is said to be a non-strict plus-
operator.
Corollary 8.2. A strict plus-operator carries positive vectors into
positive vectors. 0
Corollary 8.3. The range of a non-strict plus-operator is a positive
subspace. 0
The properties of strict and non-strict plus-operators contained in
Corollaries 8.2 and 8.3 are not characteristic.
Example 8.4. Let Q: be the vector space of complex sequences
< 00.
00
x = {~i }~1 with 2,' l~il2 For y = {1Ji} E Q: we define

i=1
+i=2
00
(x, y) = - ~l r;1 2,' ~i r;i .
The linear operator T that maps {~i }~1 into {(Xj ~i-l }~l'
where (Xl = ° and

1- 1
(Xi = -._1_ (j >
1), is an invertible non-strict plus-
operator transforming positive vectors into positive ones.
Example 8.5. With the notations of Example 8.4, let T1 be the
operator that carries {~i}~l E Q: into Wi ~i}~l' where (31 = 0 and
(3i = 1 (j> 1). Then T1 is a strict plus-operator with ffi(T1) c 1,j5+.
In order to make the survey of different possibilities complete we
mention that in an indefinite inner product space the zero operator
is a non-strict plus-operator which does not map every positive vector
into a positive vector, and the identity operator is a strict plus-
operator which does not map every vector into a non-negative vector.
By Corollary 8.3 a sufficient condition for a plus-operator T to be

strict is ffi(T) =~. Hence, making use of Lemma I.2.6, we obtain:
Theorem 8.6. If T is a Plus-operator and ffi(T) ) ~++, then T is a
strict plus-operator. 0
Our next aim is to characterize a subclass of strict plus-operators
(see Theorem 8.9 below).
Lemma 8.7. If T is a plus-operator in ~ as well as in the anti-space
of ~ (i.e. TS,W c ~+, T~- c ~-), then with the notation (8.2) we have:
(8.3) (Tx, Ty) = p,(T) (x, y) (x, Y E ~) •
Proof. By assumption, (x, x) = 0 implies (Tx, Tx) = O. There-
fore we may apply Theorem 6.3 to the inner products (x, y) and
(x, y)1 = (Tx, Ty) (x, y E ~). 0
Corollary 8.8. If T is a plus-operator in ~ as well as in the anti-space
of ~, then either T is a positive multiple of an isometric operator, or
ffi(T) c ~o. 0
Theorem 8.9. Let T be a linear operator defined on the indefinite
inner product space~. Then the following statements a) -c) are equi-
valent.
a) T maps ~+ onto itself in a one-to-one manner.
b) T maps ~++ onto itself in a one-to-one manner.
c) T = I); U, where U is a completely invertible isometric operator
and I); is a positive number.
Proof. First suppose that T satisfies a).
Let Tx = 0 for some XE~. On account of Lemma I.2.6 x is the
sum of two positive vectors xl> x 2 • Then TX1 = T( -x2 ); hence a)
yields Xl = -x2 , i.e. x = O. Thus T is invertible. Furthermore, a)
implies the relation ffi(T) ) ~++. Owing to Lemma I.2.6, ffi(T) =~.
In other words, T is completely invertible.
Let (x, x) < 0 for some x E~. Then (Tx, Tx) < O. Really, ac-
cording to a) a non-negative Tx must be the image of a non-negative
x, and by the invertibility just proved it cannot be the image of any
other element of~. Applying Lemma 7.4 to the inner products (x, y)
and (x, y)l = - (Tx, Ty) (x, Y E~) we obtain that (x, x) ~ 0 implies
(Tx, Tx) < O.
To sum up, T is completely invertible and satisfies the relations
TS,W c ~+, T~- c ~-. Therefore, by Corollary 8.8, T satisfies c).
That b) implies c) can be established quite similarly. The only
difference is the application of Lemma 7.4 for verifying the inclusion
T~W c ~+ rather than T~- c ~-.
That c) impJies a) and b) is obvious. 0
The statements as well as the proofs of Lemma 8.7, Corollary 8.8

and Theorem 8.9 easily generalize to the case where ffi(T) is not con-
tained in Gt. In particular, the following fact will be needed later.
Theorem 8.10. Let Gti (j = 1, 2) be indefinite inner product spaces
with respective ,inner products (. , ')i' Put
~t = E Gt i : (x, xli > o}, ~r = {x E 0: j : (x, xli >° or x = o} .
Let T be a linear operator from ;;D( T) = 0:1 to 0:2 , Then the assertions
a) -c) below are eq'uivalent.
a) Tmaps ~: onto ~: in a one-to-one manner.
b) T maps ~:+ onto ~t+ in a one-to-one manner.
c) T = 0; U, where U is an invertible operator with ffi(U) = 0: 2 such
that (Ux, UY)2 = (x, Y)l (x, Y E 0:1 ), and 0; is a positive number. 0
9. Pesonen Operators
We are going to study a subclass of symmetric operators in 0: whose

members behave very much like symmetric operators in Hilbert space.
Let the inner product space 0: be indefinite. Let A be a symmetric
°
do not hold simultaneously for any x
operator.
*
operator in 0: with SD(A) = 0:. If the relations (x, x) = 0, (Ax, x) =
0, we say A is a Pesonen
Applying Lemmas 7.1-7.2 to the inner products (. , .) and (. , ')1'

where
(9.1) (x, yh = (x, Y)A = (Ax, y) (x, Y E 0:) ,
we obtain the following two results.
Theorem 9.1. If A is a Pesonen operator in 0:, then either (Ax, x) >°

for every x E ~OO, or (Ax, x) <°
for e~lery x E ~OO. 0
Theorem 9.2. If A is a Pesonen operator such that x E ~oo implies

(Ax, x) > °
0, then either (y, y) < implies (Ay, y) 0, or (z, z) > >°
implies (Az, z) 0. 0 >
*
Remark 9.3. It is not an essential restriction on the Pesonen
operator A if we assume that (Ax, x) >°
whenever (x, x) > 0, x 0.
In fact, according to Theorems 9.1-9.2 this can always be achieved
rep~acing, if necessary, the space 0: by its anti-space 0:' and the opera-
tor A by -A.
The next result is a consequence of Theorem 6.2.
Theorem 9.4. If A is a Pesonen operator s~tch that (Ax, x) >0 for
every x E ~OO, then
(Ax, x) > ftA (x, x) (x E @) ,
where
ftA = inf (Ax, x)
(X,-<) =1
>-co. 0
The analogy of Pesonen operators to symmetric operators in Hilbert

space originates from the following fact.
Theorem 9.5. Every invariant subspace of a Pesonen operator is non-
degenerate.
Proof. Let 2 be an invariant subspace of the Pesonen operator A.
Then x E 2° implies both (x, x) = 0 and (Ax, x) = 0; hence x must
be zero. 0
With the help of Lemma 1.4.4 we obtain:
Coronary 9.6. Every invariant semi-definite subspace of a Pesonen
operator is definite. 0
Corollary 9.7. A Pesonen operator has no neutral eigenvector. 0
Corollaries 9.6-9.7 and Lemma 1.2.1 yield:
Coronary 9.8. Every eigenspace of a Pesonen operator is definite. 0
Thus, on account of Lemma 3.8, we have:
Coronary 9.9. Every eigenvalue of a Pesonen operator is real and
semi-simple. 0
From Corollary 9.7, Corollary I.9.8 and Theorem 3.7 it follows that
in a space of finite dimension every Pesonen operator is "diagonaliz-
able" :
Theorem 9.10. If A is a Pesonen operator in the finite-dimensional
inner product space @, then @ has a basis consisting of pairwise orthogonal
eigenvectors of A. 0
10. Fundamental Projectors

lVIaking use of certain linear operators anslng from a fundamental
decomposition, we pursue the study of decomposable spaces begun in
Section I.11. We restrict our attention to non-degenerate spaces, i.e.
(see Corollary I.11.2) to fundamental decompositions with @o = o.
Let
(10.1) @ = @"(+) @-; @+ C ~++, @- C ~--
be a fundamental decomposition of the inner product space~. We

define two orthogonal projectors P+,P- in ~ by the relations
(10.2) P+~ = ~+ , P-~ = ~-
(d. Theorem 3.10). In other words, we set

(x E ~) ,
where
(10.4) x = x+ + x- ;
is the decomposition of the vector x corresponding to (10.1). The opera-
tors P+, P- will be called the fundamental projectors belonging to the
fund amen tal decomposition (10.1).
Theorem 10.1. If 2 is a positive subspace of the decomposable, non-
degenerate inner product space ~, and P+, P- are the fundamental
projectors belonging to a fundamental decomposition (10.1) of ~, then the
restriction P+12 is invertible. Similarly, if 2 is negative, then P-12
is invertible.
Proof. Let 2 be positive. Suppose that for some x E 2 we have
P+x = O. Then x = P-x. Hence x E 2n~- C ~+ n~-- = O.
In the case of a negative 2 a similar reasoning applies to P-I~· 0
The rest of the present section is devoted to quasi-positive and
quasi-negative spaces. Recall that these spaces are decomposable
(Theorem 1.11.7).
Lemma 10.2. If ~ is a quasi-negative, non-degenerate inner product
space, then for any maximal positive definite or maximal positive sub-
space 2e@ and any fundamental decomposition (10.1) we have dim 2 =
= dim ~+. Analogously, if @ is a q~tasi-positive, non-denegerate space,
then for any maximal negative definite or maximal negative subspace 2
and any fundamental decomposition (10.1) we have dim 2 = dim @-.
Proof. Let 2 be a positive (possibly positive definite) subspace of
the quasi-negative, non-degenerate space @. Consider a fundamental
decomposition (10.1) of @ and the corresponding fundamental projec-
tors P+, P-.
If P+2 =F @+ then, @+ being finite-dimensional and positive
definite, there is a non-zero vector YE@+ orthogonal to P+2. Since y
is orthogonal to @- as well, for every xE2 we have (x, y) = (P+x, y) +
+ (P-x, y) = O. Thus y 1. 2. Therefore the proper extension (y,2)
of 2 is positive (d. Remark 1.3.2). Moreover, it is positive definite
provided 2 was so.
Consequently, for a maximal positive definite or maximal positive B
we have P+B=@+. Hence, in view of Theorem 10.'1, B is isomorphic
to @+. 0
Corollary 10.3. In a quasi-negative, non-degenerate space ft'ery
maxitnal positive definite subspace is maximal positive. Ina q'ttasi-
positive, non-degenerate space every maxl:mal negative def£11ite subspace
is maximal negative. 0
Making use of Lemma 1.11.4 we also find:
Corollary 10.4. If
(10.5) @=@t(+)@j; @t c ~++, @j c ~-- (j = 1, 2)
are two fundamental decompositions of the quasi-negative, non-degenerate
inner product space @, then dim @i = dim @t. For a quasi-positive @
we have dim @~ = dim @2" . 0
Lemma 10.2 justifies the following definitions.
Let @ be a quasi-negative, non-degenerate inner product space, and
let B be a maximal positive definite subspace of @. The non-negative
integer dim B will be termed the rank of positivity of the space @ (or of
the inner product (. , .)), and denoted by x+(@). For every fundamental
decomposition (10.1) we have
(10.6) x+ (@) = dim Q;+ .
Next let @ be a quasi-positive, non-degenerate inner product space.

Then the rank of negativity ,c(@) of @ can be defined as the dimension
of a maximal negative definite subspace Bc@. For every fundamental
decomposition (10.1) of @ we have
(10.7)
An extension of these definitions to more general spaces will be
given in Chapter IV.
The following consequence of Corollary 10.3 is often useful.
Lemma 10.5. Let@ be a quasi-negative (quasi-positive), non-degenerate
inner prod1£ct space. Then every s~tbspace Bc@ containing a maximal
positive definite (resp. maximal negative definite) subspace of @ is non-
degenerate.
Proof. Let ffi(cBc@, where @ is quasi-negative, @o = 0, and We is
maximal positive definite in @. If xEBo, then (x, x) = 0 and x 1- ffi(,
so that by Remark I,}.2 the subspace <x,9)() is positive. But We is
maximal positive (Corollary 10,}). Therefore x E ffi(, i.e., x = o. 0
11. Fundamental Symmetries. Angular Operators
Let Q: be a decomposable, non-degenerate inner product space. Con-

sider the fundamental projectors P+,P- belonging to a fundamental
decomposition (10.1) of @. We set
(11.1) J = P+ - P- ,
and say that J is the fundamental symmetry belonging to the funda-
mental decomposition (10.1).
The term "symmetry" for the linear operator J is justified hy the
fact that for every x"Q: the vectors x and Jx li.e symmetrically with
respect to the subspace @+. Namely if
(11.2)
then
(11.3) Jx = x+ - x-.
Since the system (1'1.2)-(11.3) can be solved for x+ and X-, the
fundamental symmetry J uniquely specifies the fundamental decompo-
sition to which it belongs. This entitles us to speak of "the fundamental
decomposition belonging to J".
Making use of relations (11.2)-(11.3) one easily verifies the
following.
Lemma 11.1. A fundamental symmetry J is completely invertible, and
we have
(11.4) J- 1 = J.
Moreover, J is symmetric as well as isometric. 0
Let J be a fundamental symmetry on the decomposable, non-
degenerate inner product space Q:. Since J is symmetric (Lemma 11.1),
the formula
(11.5) (x, ylJ = (lx, y) (x, y E Gt)
defines an inner product on Q:, the so-called J-inner product. (For the
more general concept of A -inner product see Section 3 above.)
With the help of the fundamental decomposition (10.1) belonging
to J, relation (11. 5) can be written in the form
(11.6) (x,y)J = (x+,y+)-(x-,y-) (X,YEGt),
where
(11.7) x=x++x-, y=y++y-; X+,Y+EGt+; r,Y-EGt-.
Note that different fundamental decompositions give rise to differ-
ent J-inner products.
11. Fundamental Symmetries. Angular Operators 53
+
Lemma 11.2. If ~ = ~+( )(2;- is a fundamental decomposition of
the non-degenerate space ~, and I is the corresponding fundamental sym-
metry, then ~+ is I-orthogonal to ~-.
Proof. In (1'1.6) put x- = 0, y+ = 0. 0
Lemma 11.3. Every I-inner product is positive definite.
Proof. From (11.6) we obtain
(x, x)J = (x+, x+) - (x-, x-) (x E ~) .
The positive definiteness of ~+ implies (x+, x+) 20, where equality
°
holds only if x+ = 0. Similarly, the negative definiteness of (2;- yields
- (x-, x-) > with equality holding if and only if x- = 0. 0
Let I be a fundamental symmetry on the non-degenerate inner

product space~. Owing to Lemma 11.3 the formula
(11.8) IlxllJ = + V(x, ;;)~ (x E ~)
defines a norm on~. It will be called the I -norm. More explicitly:
(11.9) IlxllJ = ((x+, x+) - (x-, X-))1/2 (x E~),
where x+ and x- are the components of x with respect to the funda-
mental decomposition (10.1) belonging to I.
Different fundamental decompositions induce different I-norms.
Lemma 11.4. For any I-norm we have the inequality
I(x, y)1 < IlxllJ IlyllJ (x, y E ~) •
Proof. Taking the fundamental decomposition (10.1 1 that belongs
to the I-norm in question and applying the notation (11.7), Lemma
I.2.2, the Cauchy inequality, and the definition (11.9) of the I-norm we
find:
!(x,y)1 = + (x-,y-)I <
I(x+,y+)
< (x+, x+)1/2 (y+, y+)1/2 + I(x-, x-ll1/2l(y-, y-)I* :S
:s;; ((x+, x+) - (x-, x-))* ((y+, y+) - (y-, y-))1/2 = IlxllJ IlylIJ' 0
The next result is an application of the concept of fundamental
symmetry.
Theorem 11.5. In a decomposable, non-degenerate inner product
space ~ every Mtbspace has a dual companion. ll!I oreover, one can choose
the dual c01npanion to be isometrically isomorphic to the s'ltbspace given.
Proof. The assumption implies the existence of a fundamental
symmetry I on~. For any subspace £ ( ~ we set 9)1 = I£. If
x E £n9)1.L then, in particular, x ~ Ix i.e. (x, x)J = 0, and by
Lemma 11.3 x = O. The relation 1))( n21 = 0 follows in a similar way once
we realize that]1))( =]22 = 2 (see (11.4)). Finally, I))( is isometri-
cally isomorphic to 2, since] is an isometric operator (Lemma 11.1). D
By the aid of I-norms an extremely useful description of semi-
definite subspaces c:;tn be given. For this purpose we need one more
definition.
Let 2 be a subspace of the decomposable, non-degenerate inner prod-
uct space @. Consider a fundamental decomposition
@ = @+ (+) @-; @+ C \15++ , @- c \15-- ,
and the corresponding fundamental projectors P+, P-. Suppose that
P+ 12 is invertible. Then every element x E 2 is uniquely specified by
its projection P+x, so that the formula
(11.10) (x E 2)
defines a linear operator ]{+ = K+(2) from @+ to @- with domain
<JJ(K+) = P+2 and range ffi(K+) = P-2. We say, K+ is the angular
operator of the subspace 2 with respect to @+. Similarly, if P-12 is
invertible, then the angular operator K- = K-(2) of 2 with respect
to @- can be defined by the relation
(11.11) K- P-x = P+x (x E 2);
we have <JJ(K-) = P-2, ffi(K-) = P+2 .
Theorem 11.6. Suppose that
@ = @+ (+) @-; @+ C \15++, @- C \15-- .
Let 2 be a subspace of @. If KT(K-) is the angular operator of 2 with
respect to @+(@-), then
(11.12) 2 = {x+ + K+x+; x+ E 2d
(resp.
(11.13)
where 2 1 (22) is a subspace of @+(@-). Conversely, if 2 can be written in
the form (11.12) (resp. (11.13)), 'where 2 1 (52 2) is a subspace of @+(@-) and
I(+(K-) is a linear operator from 21 to @- (from 22 to @+), then K+(K-)
is the angular operator of 2 with respect to @+(@-).
The proof is straightforward. D
Theorem 11.7. S~tppose that
@ = @+ (+) @-; @+ C \15++ , @- c \15-- ,
and denote the corresponding fundamental symmetry by]. A subspace
2c@ is positive if and only if the angular operator K+ of 2 with respect
Notes to Chapter II 55
to Q;+ exists and satisfies the condition
I n particular, positive definite subspaces are characterized by the property

IIK+x+II J < Ilx+II J
and neutral subspaces by
IIK+x+II J = Ilx+II J
For negative subspaces similar statements, involving K- instead of K+,
are valid.
Proof. If Q is positive, then the angular operator K+(Q) exists by
Theorem 10.1. Moreover, in view of Theorem 11.6 for every x E Q we
have
(x, x) = (x+, x+) + (K+x+, K+x+) = Ilx+IIJ - IIK+x+II}.
Hence, under the assumption that K+ (Q) exists, for x E Q the relation
(x, x)> 0 is equivalent to IIK+x+II J <
Ilx+II J , while (x, x) = 0 is
equivalent to IIK+x+II J = Ilx+II J . 0
Notes to Chapter II
Most results of this chapter were originally obtained as tools for the
study of linear operators in special kinds of inner product spaces, but
it has been evident from the beginning that they remain valid under
more general assumptions. Here we bring them together in order to
make clear their respective ranges as well as to make subsequent proofs
of deeper theorems more transparent.
Isometric and symmetric operators in inner product spaces of
finite dimension have been studied for a long time. An account of
the results appears in the book of Marcev [1].
Symmetric operators in certain quasi-positive inner product spaces
of infinite dimension were introduced by Pontrjagin [1J, in a wider class
of spaces by Ginzburg [2J (d. also Pesonen [1], Langer [1J), and in
arbitrary non-degenerate spaces by Berezin [1J.
More and more general types of isometric operators in indefinite
inner product spaces of infinite dimension have been considered by
KreIn and Rutman [1], Iohvidov [1], [2J, Krein [4J, Iohvidov and
KreIn [1], Ginzburg [1], [2J, and again by Iohvidov [4J, [11].
Corollary 2.2 was noted by Iohvidov and Krein [1J, and Lemma 2.1
by Iohvidov [4]. For the rest of the results of Section 2 see Mal' cev [1 J,
Iohvidov and Krein [1].
For finite-dimensional spaces a more detailed description of the

structure of an isometric operator was given by Mal'cev [1]. Karrer [1J
found essentially the same facts in a more algebraic language.
Theorem 3.10 stated for a smaller class of spaces is due to Pontrja-
gin [1], while Theorem 3.11 belongs to Langer [7], [12]. Other results
of Section 3 have been presented by Mal'cev [1] for finite-dimensional
spaces, and by Pontrjagin [1] and Langer [2] for certain classes of
infinite-dimensional spaces (d. also Iohvidov and Krein [1]). A more
complete structure theorem for symmetric operators of a finite-dimen-
sional space is to be found in the book of Mal'cev [1J.
Simple properties of a skew-symmetric operator defined in a quasi-
positive space were employed by Lax and Phillips [1], [2; Chapter VI]
to the study of a scattering problem.
Cayley transformations in indefinite inner product spaces of finite
dimension have been considered by Mal'cev [1]. In the infinite-dimen-
sional case the first application was made by Iohvidov [1]. In Sec-
tions 4- 5 we follow the general and detailed treatment of Ioh vidov
and Krein [1].
Several results of Sections 2- 5, along with examples of symmetric
and isometric operators having no Cayley transform, are contained in
the expository paper of Azizov and Iohvidov [2].
A special case of the results of Section 6 is due to Kuhne [1], while
in the general case they belong to Krein and Smul'jan [1]. Cf. also
Kuhne [2J.
Lemmas 7.1-7.3 were given by Pesonen [1] (d. also Kuhne [1]).
Lemma 7.4 appears implicitly in the paper of KreIn and Smul'jan [1].
The class of linear operators T with the property that (Tx, Tx) >
> (x, x) whenever (x, x) > 0 was introduced by Krein [4] (d. also
Iohvidov and Krein [1]). Operators T satisfying (Tx, Tx) > (x, x)
for every x in the space have first been investigated by Potapov [1] and
Ginzburg [1], [2]; d. also the Notes to Chapter VII. Both classes are
contained in that of plus-operators.
The characteristic property of plus-operators first appears in a paper
of M. L. Brodskil [1]. It plays an important role in a basic result of
KreIn [7], [8] and in its improvements (see Chapter VIII). The results
of Section 8 belong to Krein and Smul'jan [1].
Most results of Section 9 have been obtained by Pesonen [1], but in
terms of quadratic forms rather than operators. He established Theo-
rem 9.10 and an extension of it to spaces of infinite dimension. Peso-
nen's results were improved and reformulated by KUhne [1]. Inpartic-
ular, the explicit forms and the elementary proofs of our 'propositions
9.5-9.9 belong to him. Theorem 9.10 was subsequently reproved by
Kraljevic [1].
Notes to Chapter II 57
Theorem 9.10 remains valid in real inner product spaces of dimen-
sion greater than 2. A proof given by J. Milnor is included in the book
of Greub [1]. For more elementary proofs see Kraljevic [1J and Wonen-
burger [1].
Theorem 10.1 was stated in this generality by Ginzburg [1]. In
special cases it had earlier been applied by Pontrjagin [1J as well as by
Iohvidov and KreIn [1]. The latter work also contains the rest of the
material of Section 10. It should be noted that Corollary 10.4 is an
extension of Sylvester's law of inertia.
Fundamental symmetries and J-norms have been introduced by
Nevanlinna [1J, [3J. Particular cases of Theorem 11.5 are due to
Iohvidov and KreIn [1J, and to Scheibe [1].
The concept of angular operator as well as Theorems 11.6-11.7
belong to Phillips [2J. An independent and more explicit treatment was
given by Ginzburg [3J (d. also Ginzburg and Iohvidov [1], KreIn [8J).
Chapter HI. Partial Majorants and Admissible
Topologies on Inner Product Spaces
Partial majorants (locally convex topologies ill which the inner product is separately
continuous; Sections 2- 4) and admissible topologies (partial majorants with respect
to which every continuous linear form is an inner multiplication; Sections 5-7),
especially their relation to orthogonal companions and to the existence of projec-
tions are discussed. Sections 8-9 deal with the natural normed topology of definite
subspaces. The list of more important results includes Corollary 2.5 as well as
Theorems 3.3, 5.3, 6.1, 6.5, 7.1, and 9.2.
1. Locally Convex Topologies on Vector Spaces
In this section we list some definitions and facts related to topological

spaces and, especially, topological vector spaces.
A function p defined on a vector space Q; is said to be a semi-norm

on Q; if the following three requirements are fulfilled:
(1.1) P(x) ~ ° (x E Q;) ,
(1.2) P(IXX) = IIXI P(x) (IX E C, x E Q;) ,
(1·3 ) P(x1 +x 2) < P(x1 ) + P(x 2) (Xl' x 2 E Q;) .
If, in addition, P(x) = °

implies X = 0, then P is a norm.
The semi-norm P on the vector space Q; is said to be quadratic if
P(x) = [x, XJ1/2 (x E Q;), where [. , .J is a positive inner product on Q;.
A family {PY}YE r of semi-norms on the vector space Q; defines
a topology r on Q; through the following prescription: the set ® c Q;
is r-open if and only if to every x E 0J there are a finite number of in-
dices Yl' ... , y" E r and a positive number 8 such that the relations
YEQ;, PYj (x - y) <8 (j = 1, ... ,n) imply YE®. The topology r is
locally convex, which means that 1) the vector space operations are
r-continuous, and 2) any r-neighbourhood of a point x E Q; contains
a convex r-neighbourhood of x.
Every locally convex topology is defined by some family of semi-
norms.
2. Partial Majorants. The Weak Topology 59
A topology, on ~ is said to be separated if for any pair of distinct
points Xl' X 2 E ~ there is a ,-neighbourhood U1 of Xl and a ,-neigh-
bourhood U2 of X 2 such that U1 n U2 = O.
The topology defined by the family {Pl'}l'Er of semi-norms is sepa-
rated if and only if the relations Pl'(x) = 0 (y E T) imply X = O.
We shall mainly be concerned with separated locally convex topol-
ogIes.
A separated locally convex topology is metrizable if and only if it
can be defined by a countable family of semi-norms. A locally convex,
metrizable, complete topology is called a Frechet topology.
If the separated topology, is defined by a finite number PI' ... , p"
of semi-norms on ~, then, is also defined by the single semi-norm
P(x)= max Pi(x) (x E ~), which is actually a norm. In this case, is
1~1&n
said to be a normed topology. A complete normed topology may be
termed a Banach topology.
If the topology, can be defined by a quadratic norm, we say, is
a quadratic-normed topology. A quadratic-normed, complete topology
will be called a Hilbert topology.
In contrast to the notion of separated topology, a topology, on
the space ~ is said to be separable, if ~ contains a countable -r-dense
subset.
Let , be a locally convex topology defined by the semi-norms
Pl' (y E T) on the vector space~. The restrictions Pl'I2 of the semi-
norms Pl' (y E T) to a subspace 2 C ~ define a locally convex topology
,12, the topology induced by , on 2.
Let '1> '2
be two topologies on the same space~. vVe say that is 'I
weaker than '2 (or that '2 is stronger than 'I)' and write '1 ::;; '2, if every
'I-open set is '2-open or, equivalently, if every 'I-neighbourhood of
an arbitrary X E ~ contains a '2-neighbourhood of x.
If 'I '2
~ and '2 'I
< 'I at the same time, then and coincide: '2
'1 = '2'
The semi-norm PI defines a weaker topology on ~ than the semi-
norm P2 if and only if Pl(X) ~ aP2(x) for some a >
0 and every x E ~.
If two semi-norms PI' P2 define the same topology, we say PI and pz
are equivalent.
2. Partial Majorants. The Weak Topology
Let ~ be a vector space with an inner product (. , .), and let, be a topol-
ogy on~. We say that, is a partial maforant of the inner product (or:
a partial majorant on ~) if 1), is locally convex, and 2) for any fixed
y E ~ the function fJ!y(x) = (x, y) is ,-continuous on ~.
60 III. Partial Majorants and Admissible Topologies
Since (x, y) = (y, x) (x, y E Ii\;), condition 2) is equivalent to the

T-continuity of 1fJx(Y) = (x, y) for fixed x. Therefore a locally convex
topology T on Ii\; is a partial majorant if and only if the inner product
is separately T-continuous.
Let us mention an important example of a partial majorant.
The weak topology To of the inner product space Ii\; is the locally con-
vex topology defined by the family {PY}YEQ; of semi-norms, where
(2.1) Py(x) = I(x, y)1 (x Eli\;) .
Theorem 2.1. The weak topology To of the inner product space Ii\; is a
partial rnaforant on Ii\;. A locally convex topology T on Ii\; is a partial
rnaforant if and only if T > To .
Proof. Letxo,YEIi\;, 8>0 be given. Therelationl(x,y)-(xo,Y)I<8
holds for every x belonging to the set {x: py(x - xo) < 8}, which
is a weak neighbourhood of Xo (see the notation (2.'1)). Hence To is
a partial majorant. As a consequence, every locally convex topology
T > To is also a partial majorant on Ii\;.
Let T be a partial majorant on Ii\;. Then for any elements xo, Yj E Ii\;
(f = 1, ... ,n) and any number 8> 0 there are T-neighbourhoods Uj
of Xo such that x E Uj implies I(x'Yj) - (xo,YjJI < 8 i.e. PYj(x-xoJ< 8.
n
Since also n Uj is a T-neighbourhood of xo, it follows that every weak
i=!
neighbourhood of Xo contains a T-neighbourhood of XO' D
Corollary 2.2. Every partial maJ'orant of a non-degenerate inner
product is separated. D
Remark 2.3. A partial majorant T on a degenerate inner product
space Ii\; is not necessarily separated. Replacing, however, each of the
semi-norms Py that define T by Py + q, where q is any norm on Ii\;, one
obtains a separated partial majorant Tl > T.
Next we describe some simple properties of partial majorants and
weak topologies involving the concept of orthogonality.
Lemma 2.4. If T is a partial rnajorant on the inner product space Ii\;,
then for any subspace 53 c Ii\; the orthogonal cornpanion 53 1 is T-closed.
Moreover, 53 and its T-closure 53 have the sarne orthogonal cornpanion.
°
Proof. If Xo ~ 531, then (xo, y) =F for some y E 53; thus for every x
in a suitable T-neighbourhood of Xo we have (x, y) =F 0, x ~ 53 1 , In
order to prove the second assertion suppose that the vector x E Ii\; is
-
° -
not orthogonal to 53, i.e., (x, yo) =F for some Yo E 53. Then (x, yJ =F 0
for every Y in aT-neighbourhood U of Yo' Choosing Y E 53 nU it
follows that x ~ 53 1 , D
3. Metrizable Partial Majorants 61
Making use of Lemma 1.9.6 we obtain:

Corollary 2.5. Let T be a partial 111ajorant on the non-degenerate inner
product space~. Then every ortho-complenlented subspace 01 ~ is
T-closed. 0
Further specialization yields:
Corollary 2.6. Let T be a partial majorant on the non-degenerate inner
product space ~, and let
~ = ~+ (+) ~-; ~+ C ~++, (,£;- C ~--
be a f~mdamental decomposition of ~ (d. Corollary 1.11.2). Then ~+ and
~- are "-closed. 0
3. Metrizable Partial Majorants

Theorem 2.1 (and Remark 2.3) establish the existence of (separated)
partial majorants. One wonders if it is always possible to find a metriz-
able partial majorant. The answer turns out to be negative.
Theorem 3.1. II an inner product has a metrizable partial majorant,
then it has a normed partial1l1aiorant too.
Proof. Let T be a metrizable partial majorant on the inner product
space~. Then T is defined by a sequence PI' P2' ... of semi-norms.
Introduce the following notations:
(3.1) Un = {x E~: .i.;Pj(x) < ~J (n =1, 2, ... ) ,

J=1 n f
(3.2) U~ = {x E~: I(x, y)1 < 1 for every y E Un} (n = 1,2, ... ) ,
co
(3·3) ~ = U (Un n U~) ,
n=1
(3.4) ~l=t~;CXjXi: n21, {Xi}~C)ZS' j~1ICXii<1}'

(3·5) P(x) = inf {e > 0: X E e )ZS1} (x E ~) .
We shall prove that p is a semi-norm which defines a partial
+
majorant on~. Replacing p by P q, where q is an arbitrary norm
on ~, the theorem follows.
As T is a partial majorant, and any T-neighbourhood of 0 contains
some of the sets Un' for every x E ~ there exists an index no such that
x E U~o' On the other hand, by virtue of (3.1), for sufficiently small
(J>Owehave(JxEUno" Setting min{1,(J}=(J1 wefind(JlxE)81 i.e.
1
x E-
(Jl
)ZS1' Thus P(x) < 00.
p clearly satisfies (1.1). The property (1.2) follows from the fact
that for a positive number (} the relation x E (}~l is equivalent to
lal x E lal (} ~l or, in view of (3.4), to ax E lal (} ~1' In order to check
(1.3) let xi E (}i~l (i = 1,2). Then ~ xi E ~l and, by the convexity
of ~l> (}i
1
(}l
_ .._ - - Xl + ---
(}2
-1 x2 E ~l'
(}l + (}2 (}l (}l + (}2 (}2
i.e. Xl + X 2 E ((}l + (}2) ~l'

It remains to show that the semi-norm P defines a partial majorant
on Q:.
First let x,Y E~. According to the definition (3.3) of ~, there are
indices m, n such that X E Umnu;,., y E UnnU~. If m < n, then from
the relations X E U~, Y E Un' Un C Umand (3.2) we infer that I(x, y)1 < 1.
In the case m >
n we arrive at the same conclusion by interchanging
the roles of X and y.
Next let x, y be a pair of elements in Q: with P(x) 1, P(y) 1. On < <
account of (3.5) and (3.4) we have x, y E ~l' so that (3.4) yields the
represen tations
m m
X = };a.x.;
i=1 1 1
{xi};n c ~ , !-' lail <
1=1
1,
n n
y = l:,' p"y,,; {y,,}~ c~ , }; IP"I < 1.
"=1 "=1
Applying the result of the preceding paragraph to xi and y" we obtain

(f = 1, ... ,m; k = 1, ... , n) .
Therefore
(3·6) I(x, y)1 < 1 (P(x) < 1, P(y) < 1) .
Finally, if x, yare arbitrary elements of Q;, then replacing X by
_n_ _ 1_ x or nx (n = 1, 2, ... ) according as P(x) -=1= 0 or P(x) = 0,
n + 1 P(x)
and proceeding analogously for y, from (3.6) we conclude that
(3·7) I(x, y) I < P(x) P(y) (x, y E Q;) •

In particular, the topology defined by P is a partial majorant of the
inner product. D
Example 3.2. Let, as in Example I.11.3, Q; be the vector space of

those doubly infinite numerical sequences where only a finite number
4. The Polar of a Normed Partial Majorant 63
of terms with negative index is different from zero, and for x =
= {~i}~oo E~, Y = {1]i}~oo E ~ let
00
(3·8) (x, Y) = L
~i n-i-l .
i=-oo
Suppose that this inner product has a partial majorant defined by
a norm II· II. Then I(x, y)1 < (3(y) Ilxll for a suitable (3(y) and >°
every x, y E~. Setting
y = en = {(Ji,-n}~-oo (n = 1, 2, ... )
we obtain:
I(x, e_n)1 = l~n-ll < (3(e_ n ) Ilxll (x E~; n = 1,2, ... ) .
The
for
choice x = {~j}~oo; ~i = ° for f < 0, ~j = (j + 1) (3(e-i-l)
f > 0, yields n (3(e_ n ) < (3(e_ n ) Ilxll (n = 1,2, ... ), which is
impossible. Thus our inner product has no normed partial majorant;
by Theorem 3.1 it cannot have a metrizable partial majorant either.
As to the uniqueness of metrizable partial majorants, we have the

following:
Theorem 3.3. Let ~ be a non-degenerate inner product space. Suppose
that the Frechet topologies Tv T2 are partial maforants on~. Then Tl = T2.
Proof. Each of Tv T2 is defined by a countable family of semi-norms.
The union of these two families defines a metrizable locally convex
topology •.
Let {x n };'" C~ be a Cauchy sequence with respect to T. Since
(3·9) T>Tj (f=1,2),
{x n } is a Cauchy sequence with respect to Tj as well. Therefore {x,.}
converges to some element Yi E ~ with respect to Tj (f = 1,2). Taking
into account that Tj is a partial majorant, for every Z E ~ we obtain
(xn' z) -+ (Yj' z) (n -+ 00; f = 1,2). Thus (yv z) = (Y2' z) (z E @), i.e.,
YI = Y2' The TI-limit and the T2-limit of {xn} being equal, it is also a
T-limit. Consequently, T is complete, hence a Fnkhet topology.
Applying the closed graph principle to the identity mapping of the
space ~ topologized by T onto the same space topologized bYTj' from
(3.9) we derive T = Ti (f = 1,2). D
4. The Polar of a Normed Partial Majorant
In the following lines we show that every normed partial majorant of

a non-degenerate inner product gives rise to another normed partial
majorant. This fact, and the properties of the correspondence, will be

applied in Chapter IV.
Consider a partial majorant i defined by the single norm II . lion
the non-degenerate inner product space Q;. Put
(4.1) Ilxll' = sup I(x, y)1 (x E Q;) .
Ilyll;;;;l
It is easy to see that II· II' is a norm on Q;. We say, II· II' is the polar
of the norm II· II. The topology defined by II· II' will be denoted by
i' and called the polar of the topology i.
i' only depends on 7: and not on the norm defining i. Really, if
II . III is another norm on Q; such that
(\llxll < IIxl11 :s;; Pllxll (x E Q;) ,
where (x, 1'1 > 0, then
1 1
- sup I(x, y)l:S;; sup I(x, y)1 < - sup I(x, y)1 (x E Q;) •
1'1 Ilyll;;;;l IIYiI,S! ex Ilyll;;;;1
Simultaneously we have proved the following:
Lemma 4.1. Let iv i2 be two nonned partial majorants of a non-
degenerate inner product. If i1 < i 2, then i{ 2': i~ . 0
The polar of a partial majorant is also a partial majorant. Namely

from (4.1) we obtain
(4.2) I(x, y)1 ~ lIyllllxll' (x, y E Q;) .
Thus one can form the normed partial majorants i" = (i')', i'" = (i")'
etc.
Lemma 4.2. If i is a normed partial majorant on the non-degenerate
inner product space Q;, then so is its polar i'. Moreover, we have
(4·3) i" < i ,
(4.4) i/l I === i' .
Proof. ';Ye already know that i' is a normed partial majorant.

Interchanging x and y in (4.2) we find sup I(x, y)1 < IIxll, which
Ilyll';;;;1
proves (4.3). From (4.3), replacing i by i', one obtains i'" ~ i'.
Finally, application of Lemma 4.1 to i1 = i", i2 = i yields i'" 2':
2': i', 0
Lemma 4.3. If a norm Ii· II that d~fines a partial majorant on a nOtl-
degenerate space is quadratic, then so is the polar norm II . II'·
Proof. Let Q; be a vector space with non-degenerate inner product
(. ,.). Consider the quadratic norm IIxll = [x, xJ1/2 (x E Q;), where
5. Admissible Topologies 65
c. ,.J is a positive definite inner product on @. The completion of @
with respect to II . II is a Hilbert space~. If II . II defines a partial
majorant of (. , .), then for any Y E @ the linear form qJy(x) = (x, y)
(x E @) is continuous with respect to II . Ij, hence it admits a unique
continuous extension to~. Consequently, by the Riesz representation
theorem, there exists a y'E@ such that
(x, y) = [x, y'J (x E @) .
The correspondence y ---+- y' is an isomorphism between the vector
space @ and a subspace ,53 ( 0;. Moreover, Ily'll = IlqJyll =
= sup I(x, y)1 = IIYII'· Since II· [I is quadratic on ,53, [I . [I' is
Ilxll;£1
quadratic on @. 0
A partial majorant 'i satisfying the relation 'i' = 'i is said to be
self-polar.
5. Admissible Topologies
A topology 'i on the inner product space @ is said to be admissible,

if 1) 'i is a partial majorant on @, and 2) for any 'i-continuous linear
form qJ defined on @ there is an element Yo E @ such that qJ(x) = (x, Yo)
(x E @).
Theorem 5.1. The weak topology of an inner product space is admis-

sible.
Proof. Consider the weak topology 'io of the inner product space @.
We have already shown in Theorem 2.1 that 'io is a partial majorant
on @. Now let qJ be a weakly continuous linear form on @. Then we
can find elements Yl> ... , Yn E @ and a number 0> 0 such that
(5.1 ) IqJ(x) I <1 whenever [(x, Yi)1 <0 (f=1, ... ,n).
If Yl> ... , Yn are the fixed elements occurring in (5.1) and x runs
over @, the n-tuples {(x, Yl), ... , (x, Y,,)} of complex numbers form
a vector space of dimension not exceeding n. This space admits a basis
(5.2) {(xv Yl)'·· ., (Xl> y,,)}, ... , {(x"" Yl),·· ., (x"" Yn)},
where m < n, and Xl> ... , x'" are suitable elements of @. Thus for
every x E @ there exist numbers IX1> ••• ,IX", satisfying
11t
{(x, Yl)' ... , (x, y,J} = 1.: IXk{(X k , YI)' ... , (Xk' y,,)}
k=1
or, written by coordinates,

(5.3) (x, Yj) = CE'" O(k X" , Yj) (f = 1, ... , n) .

1<=1
Relations (5.1) and (5.3) imply
(5.4) rp(x) = cp(}; O("x,,) .

k=1
Really, setting x' = '"

x - };O("x" we see from (5.3) that (x', Yj) = 0
"=1
(f = 1, ... , n). Hence for every e>O we find (ex', Yj) = 0 (f =1,
... ,n), and therefore, by (5.1), Icp(e x') I 1 (0 9 (0), i.e. < < <
cp(x') = 0 .
As the system (5.2) is linearly independent, the equations
(5.5) "
(x", }; fJjYj) = cp(xk ) (k = '1, ... , m)
j=1
can be solved for fJl' ... ,fJn' We set

n
(5.6) L' fJi Yj = Yo .
j=1
Then (5.4), (5.5), (5.3) and (5.6) yield:
cp(x) = cp (J.;
k=1
O(k X ,,) = ( i'
k=1
O(kXk '
= (x, Yo)· 0
It is clear that the weak topology is separated if and only if the
inner product is non-degenerate. This remains true for every admissible
topology.
Lemma 5.2. I] the inner product space 0; is non-degenerate (degenerate),

then every ad'missible topology oi 0; is separated (non-separated).
Proof. In view of Corollary 2.2 it is sufficient to consider the de-

generate case. So let Xo be a non-zero isotropic vector of 0;. Further-
more, let i be a separated locally convex topology on 0; defined by the
semi-norms Pl' (y EF). Then for some p = Pl'o we have P(xo) =F O.
According to the Hahn-Banach extension principle there is a linear
form (p on 0; such that cp(xo) = P(xo), Icp(x)j < P(x) (x E 0;). In particu-
lar, qJ is i-continuous and (p(xo) =F o. On the other hand, if i is admissi-
ble, then for some Yo E 0; we have cp(x) = (x, Yo) (x E 0;); hence cp(xol =
= (xo, Yo) = O. Contradiction. 0
Concerning the uniqueness of admissible topologies one can prove
the following.
5. Admissible Topologies 67
Theorem 5.3. If i1 is an adlldssible topology defined by a c01mtable

family of semi-no1'1'ns on the inner product space ~, and i is any admissible
topology on ~, then i1 ;::::: i. In particular, no more than one ad1-nissible
topology of ~ is metrizable.
Proof. Let the topologies iv i be defined by the semI-norms

Pi (j = 1, 2, ... ) and qy (y Er), respectively. Assume that i1 is not
stronger than i. Then there exist indices Yl' ... , Ym E and a number r
c >0 such that the set
{x: qr.(x) <c (k = 1, ... , m)}
does not contain any iI-neighbourhood of the origin, in particular, any
of the sets
P(x)
I
<-n1 (j = 1, ... , n)} (n = '1,2, ... ) ,
Consequently, there is a sequence {xn}~ C ~ \vith
max qYk(X,,) > c .

l;?;k;?;m
Since the maximum is taken over a finite set, it means no loss of

generality if we write q(x n ) > c, where q denotes one of the semi-
norms qYk' Setting nXn = y" we obtain:
(5.7) max Pi(Yn)
1 ;?;i;?;n
<1 (11 = '1,2, ... ) ,
(5.8) q(y,.);::::: 11£ (n = 1,2, ... ).

The set 2 = JX: q(x) = o} is a subspace of~. We consider the
quotient space ~ = ~/2 as a vector space with the (proper) llorm
Ilxll=q(x) (XE @, XE x). Let;P denote a linear form on & which is
continuous:
(5.9)
Then the formula
ip(X) = T(x) (x E 0;, x E x)
defines a linear form on 0; such that lip(x) I ;;;; IITII q(x) (x E~). In partic-
ular, ip is i-continuous. As i is admissible, ip(x) = (x, zo) (x E ~) with
a suitable Zo E~. Hence q.' is iI-continuous. Thus for some positive
integer r and positive 0 we have 1q:(x)l<l whenever max Pi (x) <0 or,
equivalently, 1 ;?;j;?;r
(5.10) (x E ~) .
From (5.7) and (5.10) we obtain
l<p(y,,)I < b1 (n > r) .
Therefore the sequence {q3(y,,)}~ is bounded for any fixed (j; belonging
to the conjugate space Q;* of the normed space &. But q3(y,,) can be
regarded as the value of a linear form y" on the element (j; of the Banach
space Q;*. According to (5.9) this linear form is continuous; its norm
is not greater than, and by the Hahn-Banach extension principle not
less than, the value Ily,,11 = q(y,,). Thus (5.8) contradicts the principle
of uniform boundedness. 0
6. Orthogonal Companions and Admissible Topologies
The results to be presented in this and the next section show that ad-
missible topologies are a natural tool in the study of orthogonal com-
panions, the existence of projections etc. It also turns out that often
in these applications the particular choice of the admissible topology
is indifferent.
Theorem 6.1. Let 't be an admissible topology on the inner product
space Q;. Then the i-closure of any s~tbspace 2 c Q; coincides with 211 .
Proof. Denoting the i-closure of 2 by 2 we have 2 C 2 11 , since
2 c 211 by definition and 211 is i-closed by Lemma 2.4.
Now let Xo (£ 2. Then Xo has a i-neighbourhood disjoint from 2.
This neighbourhood contains a set of the form
{x: PYj(x-xo) <8 (j=1, ... ,n)},
where Py" ... , PYn belong to the family of semi-norms defining't, and
8is positive. Introducing the notations
P(x) = max pyj(x) (x E Q;) ,
1;;;;1;;;;"
(6.1) q(X) = inf P(x - x') (x E Q;)

x'E2
it is easy to see that q is a semi-norm and q(xo) ;;::: 8. Therefore the Hahn-
Banach extension principle assures the existence of a linear form <p on
Q; with
<p(xo) = 8 ,
(6.2) l<p(x) I < q(x) (x E Q;) •
Since, by virtue of (6.1), q(x) :S P(x), relatioil (6.2) implies the

i-continuity of <po But i is an admissible topology. Consequently, there
6. Orthogonal Companions and Admissible Topologies 69
exists a Yo E Q: such that

(6.3) <p(x) = (x, Yo) (x E Q:) .
If x E ,2, then (6.1) yields q(x) = O. Hence with the aid of (6.2) and
(6.3) we obtain Yo E ,21. On the other hand, (xo, Yo) = <p(xol = 8 =F 0 .
Thus Xo E! ,211. 0
Remark 6.2. Theorem 6.1 implies that the closure of a subspace with
respect to any admissible topology is the same. Keeping this fact and
Theorem 5.1 in mind, it will be convenient to speak of weakly closed
and weakly dense subs paces instead of "subspaces which are closed
(resp. dense) relative to some (and then every) admissible topology".
Corollary 6.3. ,211 =,2 if and only if the s1fbspace ,2 is weakly
closed. 0
Corollary 6.4. A subspace ,2 of the non-degenerate space Q: satisfies
,21 = 0 if and only if,2 is weakly dense in ct. 0
In order to place Corollary 6.3 (above) and Theorem 6.5 (below)
among previous results, recall that if,2 is a subspace of the non-degener-
ate inner product space Q: such that ,2 + ,21 = Q: , then a) ,2 is non-
degenerate (Corollary I.9.5), b) ,211 = ,2 (Lemma I.9.6), and c) ,2 is
weakly closed (Corollary 2.5 and Theorem 2.1). By Corollary 6.3 con-
ditions b) and c) are equivalent. By Example I.9.7 conditions a) and
b) do not imply ,2 + ,21 = Q:. The next result says that a weakening
of "a) plus b)" is necessary and sufficient for,2 +,21 to fill Q: "approxi-
mately" .
Theorem 6.5. Let Q: be a non-degenerate inner product space, and let ,2
be a subspace of Q:. The s~tbspace ,2 + ,21 is weakly dense in Q: if and
only if ,211 is non-degenerate.
Proof. For any subspace ,2, in view of (I.3.2) and the definition of
orthogonal companion, we have (,2 + ,21)1 = ,21 n ,211 = ,2111 n ,211
= ,211 n (,211) 1. It remains to apply Corollary 6.4. 0
Corollary 6.6. For every subspace ,2 of a definite inner product space
Q: the subspace ,2 + ,21 is weakly dense in Q:. 0
From Theorem 6.5 one can also derive the following extension of
Corollary 6.6.
Lemma 6.7. Suppose that the space Q: as well as the subspace,2 c Q: are
non-degenerate and decomposable:
(6.4) Q: = Q:+( + )Q:-; Q:+ c ~++, Q:- c ~-- ,
(6.5) ,2=,2+(+),2-; ,2+c~++, ,2-c~--.
70 III. Partial MajOl'ants and Admissible Topologies
If the fundamental decompositions (6-4) and (6.5) can be chosen in such

a way that
(6.6)
then 3 + 3 1 is 7£'Cakly dense in @.
Proof- Let x E 3 1 . Then x.l. 3+. Relation (6-4) yields a decom-
position x = x+ +x-, where x+ E @+, x- E @-. Since (6-4) and (6.6)
imply x- .l. £+, we obtain that the vector x+ = x - x- is orthogonal
to 3+. On the other hand, again by (6-4) and (6.6), x+ .l. 3-. As a
result, x+ E 3 1 , and x- = x - x+ E 3 1 . Therefore
3 1 = (3 1 n @+) (+) (£1 n @-) .
In particular, 3 1 fulfils the conditions postulated for 3 (ef. Corollary
1.4.3). Replacing in the above argument £ by 3 1 we obtain:
3 11 = (3 11 n @+) (+) (311 n @-) .
Thus, in view of Corollary 1-4.3, 3 11 is non-degenerate. By Theorem
6.5 3 +
3 1 is weakly dense in @. D
7. Projections and Admissible Topologies
In this section admissible topologies will be utilized for two purposes:

a) to find all vectors admitting a projection on a fixed subspace, and
b) to characterize ortho-complemented subspaces. Recall that problem
a) was solved for decomposable subspaces in Theorems 1.8-3 - 1.8.5,
while b) was not solved at all.
Theorem 7.1. Let £ be a subspace of the inner product space @.
Denote an admissible topology of the subspace £ by T. The element y E @
admits a projection on 3 if and only if the linear form rpy(x) = (x, y)
(x E £) is T-continuous.
Proof. Since T is admissible on £, the linear form rpy (considered on
3) is T-continuous if and only if there exists an element Yl E £ satisfying
rpy(x) = (x, Yl) (x E 3) i.e. Y - Yl .l. 3. D
Theorem 7.2. Let £ be a subspace of the inner product space ~.
Consider the weak topology TO (3) of £ and the topology Tl = To(@)I£
induced on £ by the weak topology To(@) of @. We have To(3) :s
T1 ; equality
holds if and only if £ is ortho-complemented.
Proof. Tl is defined by the family {Py I£}y E Q; of semi-norms,
where Py(x) = I(x, y) I (x E @), while To(£) is defined by the subfamily
{Pyl £}y E .2' Hence To(£) is weaker than Tl .
8. Intrinsic Topology 71
Let B be ortho-complemented. Then for any y E ~ we have

PylB = Py,IB, where Y1 denotes a projection of y on B. SO in this case
'1
'o('£!) and are defined by the same family of semi-norms.
Conversely, suppose that 'orB) = ' 1 ' For any y E ~, according to
Theorem 2.1, the linear form cpy(x) = (x, y) is 'o(~)-continuous on ~.
Therefore fPy is 'o('\3)-continuous on B. Thus, in view of Theorems 7.1
and 5. '1, Y has a projection on B. 0
8. Intrinsic Topology
Let B be a definite subspace of the inner product space~. The relation

(8.1) IxlE = I(x, x) 11/2 (x E B)
defines a quadratic norm on B. We call I . IE the intrinsic norm on B,
and the corresponding normed topology 'int(B) the intrinsic topology
of B.
The adjective "intrinsic" will always refer to the topology 'int(B)
of a definite subspace B. We say, for instance, that the definite sub-
space B is intrinsically complete, if it is complete with respect to the
norm I . IE' In other words, B is intrinsically complete if B or the anti-
space of B is a Hilbert space.
We define the intrinsic completion B of a definite subspace B as the
completion of B with respect to the intrinsic norm I . IE' The Hilbert
- -
space dimension of B (or of the anti-space of B, if B is negative definite)
will be termed the intrinsic dimension of B, and denoted by dimint B.
It is the minimal power of sets m c B such that the span <m> is dense
in B with respect to I . IE'
Consider the following question. Given a vector y E @ and a defi-
nite subspace B c ~, is the linear form rpy(x) = (x, y) (x E B) intrin-
sically continuous on B (i.e., continuous with respect to 'int(B)) or not?
If B has the property that cpy is intrinsically continuous on B for every
y E ~, we say B is regular. Otherwise B is said to be singular.
Obviously, in a finite-dimensional or definite space ~ every definite
subspace is regular. On the other hand, owing to Theorem 1. 5.4 and
Remark 1.5.5, a definite subspace B of an inner product space ~ is
regular if and only if it is regular when considered in a maximal non-
degenerate subspace of ~ containing B. Therefore the existence prob-
lem of singular subspaces is fully settled by the following theorem.
Theorem 8.1. Every infinite-dimensional, non-degenerate, indefinite
inner product space ~ contains a singular subspace.
Proof. Let e1 denote any non-neutral element of~. The subspace
<e >being
1 ortho-complemented and ~ being non-degenerate, there
is a non-neutral e2 E Q: orthogonal to e1 . Then <el , e2 is also ortho- >

complemented, and in its orthogonal companion one can find a non-
neutral ea. Pursuing the process we obtain a sequence {ej};oo c (j; of
pairwise orthogonal, non-neutral elements. In view of the indefi-
niteness of Q: we may also require sign(e2 , e2 ) 0/:- sign (e v el ).
Suppose that {e j } contains an infinite number of positive elements.
Then taking a subsequence and multiplying its members by suitable
positive numbers we get a sequence {gj};oo c Q: which satisfies the
conditions
(gl,gl)=-1, (gj,gj) =1 (j=2,3,···),
(gj' gk) = 0 (jO/:- k; j,k = 1,2, ... ).
00
The set 2 of all finite sums .L'$·g·,

1 1
where;1 = ;2 = .1.'$1' (d. Example
1=1 ,=3
I.4.9), is easily seen to be a positive definite subspace of Q:. However,
the linear form Cf!g. (x) = (x, gl) (x E 2) is not intrinsically continuous
on 2, since for the vectors
1
+ g2 + -
11+2
Xn = gi l,' gj E 2 (n = 1,2, ... )
it j=3
we have
(n -,'>- 00) ,
(n = 1,2, ... ) .
If no more than a finite number of vectors ej is positive, then the
above construction applies to the anti-space of Q:, again providing a
singular subspace in Q:. 0
9. Projections and Intrinsic Topology
The following lemma is related to Theorem 7.1.

Lemma 9.1. If the element Y E (j; admits a projection on the definite
subspace 2, then the linear form Cf!y(x) = (x, y) (x E 2.) is intrinsically
continuous on 52. Thus every ortho-complemented definite subspace is
regular.
Proof. Let Yl be the projection of yon 52. Then, according to Lem-
ma I.2.2, for every x E 52 we have I(x, y)1 = I(x, Yl)1 ;£ IYll2 Ix12' 0
We note that a non-closed subspace of a Hilbert space is not ortho-

complemented (d. Corollary 2.5), though regular. In a definite, but
(intrinsically) non-complete space a subspace 2 can even be weakly
9. Projections and Intrinsic Topology
closed without being ortho-complemented, and .\3 is automatically

regular again. (Let ~ be the vector space of continuous functions f(t),
o < t < 1, with inner product
1 _
(I, g) = J f(t) g(t) dt ,
o
and .\3 the set of those fE~ which vanish for 0 <t< '1/2.) If, however,
a regular subspace .\3 of an inner product space is sequentially To(.\3)-
complete, then it is ortho-complemented. This turns out from the next
result coupled with the well-known fact that for a definite inner prod-
uct space weak sequential completeness is equivalent to (intrinsic)
completeness.
Theorem 9.2. If a definite subspace is regular and intrinsically
complete, then it is ortho-complemented.
Proof. Let.\3 be a positive definite, regular, intrinsically complete
subspace of the inner product space~. Since.\3 is a Hilbert space,
by the Riesz representation theorem Tint(.\3) is an admissible topology
of.\3. It remains to apply Theorem 7.1. If.\3 is negative definite, we
pass to the anti-space of~. 0
In certain cases intrinsic completeness is necessary, too, for ortho-
complementedness.
Theorem 9.3. If ~ is a non-degenerate inner product space such that
the relations
(9.1) lim (xm - Xn ' y) = 0 (y E ~)
mJ~oo
imply the existence of an x E ~ satisfying

(9.2) lim (xn - x, y) = 0 (y E ~) ,
..-..00
then every ortho-complemented definite subspace .\3 c ~ is intrinsically
complete.
Proof· Let {xn}~ c .\3 ,
(9.3) lim IXm - xnb.l = o.
m,n-;.-oo
Then, by Lemma 1.2.2,

(9.4) lim (xm - x n ' y) = 0 (y E .\3) .
tn, n---+OO
But for any yE~ we have (xm - x n ' y) = (xm - x n ' Yo), where Yo is the
projection of y on.\3. Consequently, (9.4) implies (9.1). Thus for some
x E ~ the relation (9.2) holds. Furthermore, in view of Corollary 2.5
and Theorem 2.1, .\3 is weakly closed. Hence x E 2 .
On the other hand, owing to (9-3), there exists an element z belong-

2
ing to the intrinsic completion of 2 such that IXn - zig --+ 0 (n--+oo).
Working in the space 2, relation (9.2) and (x n - Z, y) --+ 0 (y E 2)
imply that the element z - x E fi is orthogonal to 2, a dense subset of
2. Therefore z = x. In particular, z E 2. D
To finish with, we are going to give "pointwise" characterizations
of regular and ortho-complemented definite subspaces, respectively.
These results should be compared with Theorem 1.8.4.
Theorem 9.4. Consider an elentent y and a positive definite subspace 2
of the inner product space @. The linear form tpy(x) = (x, y) (x E 2) is
intrinsically continumts on 2 if and only if
(9.5) inf (y - x, y - x) >- 00 .
Incase 2 is negative definite, the respective cond1:tion reads

(9.6) sup (y - x, y - x) < 00 .
•~E B
Proof. Let I(x, y)1 < ex [xll.l (x E 2), where 2 is a positive definite
subspace. Then for every x E 2 we have:
(y - x, y - x) = (y, y) - (y, x) - (x, y) + (x, x) 2
> (y, y) - 2ex[x[B + Ix!~ .
The last expression is a quadratic function of IX[B which is bounded
from below.
Let, conversely, (y - x, y - x) ~ fJ for a real number fJ and every
vector x belonging to the positive definite subspace 2. This can be
written in the form
(9.7) (x, x) - (y, x) - (x, y) +
y > 0 (x E 2) ,
where y = (y, y) - fJ. Fix the element x and choose a unimodular
number 8 such that 8(X, y) is real. Replacing x in (9.7) by fAX, where
A E R, we find:
A2(X, x) - 2A8(X, y) +
y 2 0 (- 00 A (0) .< <
It follows that 48 2(X, y)2 - 4y(x, x) ::;; 0, i.e., [(x, y)1 < yl/2 [xJ.e.
Since y is independent of x, the intrinsic continuity of tpy(x) (x E 2) is
proved.
If 2 is negative definite, the same argument can be applied to the
anti-space of @. D
Theorem 9.5. Let 2 be a positive definite subspace of the inner product
space @. The element y E @ admits a projection on 2 it and only if the
following conditions are fulfilled:
Notes to Chapter III 75
a) The linear form 97 y (X) = (x, y) (x E 2) is intrinsically continuous
on 2.
b) There exists an Xo E 2 such that (xo, xo) = (xo, y) = 197yl~, where
197,,1£ = sup 197 y (X) I .
Ix!£ ~j
If 2 is negative definite, condition b) should be replaced by (xo, xo) =

= (xo, y) -197yI1·
=
The element Xo appearing in b) is the pr01'ection of yon 2.
Proof. We may confine ourselves to the case where 2 is positive
definite.
If Xo is the projection of y on 2, then obviously (xo, y) = (xo, xo) =
= Ixol~. It remains to prove that 97" is intrinsically continuous on 2,
and 197,,1£ = IXol£. But this follows from the relations
197y(X) 1 = I(x, y) I = I(x, xo) I ::::; Ixol£ Ixl£ (x E 2) ,
97 y (Xo) = (xo, xo) = Ixol~ .
Assume, conversely, that conditions a), b) hold. Owing to a) and
the Riesz representation theorem, there is an element %0 in the intrinsic
completion 3 of 2 such that 97 y(X) = (x,x-o).1l (x E 2); we have
197y lll = 1%01 2 , Making use of these relations, and also of condition b),
we find:
(Xo, Y) = (xo, %0).2 '
(xo, x o) = (xo, y) = IxolE .
Therefore
!x - xolE = (xo, xo) - (xo, xo)n - (xo, xo).1l + IXolE = 0 .
Hence %0 = Xo, i.e. 97 y(X) = (x, xo) (x E 2), which shows that Xo
is the projection of y on 2. D
Notes to Chapter HI
Partial majorants and admissible topologies have been studied for
some time in the duality theory of topological vector spaces; see e.g.
Bourbaki [1J, Kothe [1J, Robertson and Robertson [1], Schaefer [1J.
The application of the general theory to inner product spaces was
initiated, in different directions and independently of each other, by
Aronszajn [1J and Scheibe [1]. The investigations of Aronszajn were
continued by Wittstock [2J, [3J, and those of Scheibe by Ginzburg and
Iohvidov [1].
The mathematical treatment of the intrinsic topology of definite
subspaces, a subject characteristic for inner product spaces, began with
76 III. Partial Majorahts and Admissible Topologies
a short note of Bognar [2J in connection with a paper on quantum field

theory by A. Uhlmann. The investigations were carried on by Iohvidov
[5J, [6J, [SJ and Ginzburg (see Ginzburg and Iohvidov [1J).
Theorem }.1 is a weakened form of a result of Wittstock [2]. The
idea of Example 3.2 belongs to M. L. Brodskil as quoted by Ginzburg
and Iohvidov [1]. Theorem }.} was proved by Scheibe [.JJ (d. also
Ginzburg and Iohvidov [1J).
For a more detailed discussion of polar topologies see Aronszajn [1J
and Wittstock [}].
Theorems 5.1, 5.} and 6.1 appear in the article of Ginzburg and
Iohvidov [1J, and in the books of Bourbaki [1], Kothe [1], Robertson
and Robertson [1J. The three latter references also contain an exten-
sion (in the non-degenerate case) of Theorem 5.1, according to which
a locally convex topology is admissible if and only if it is stronger than
the weak topology and weaker than another particular topology called
Mackey topology (theorem of Mackey and Arens); for a special case see
Theorem IV.S.} below.
Theorem 7.1 was proved by Ginzburg and Iohvidov [1J, while
Theorem 7.2 is essentially due to Scheibe [1J.
Theorem S.'1 has first been proved by Bognar [2]. However, the
idea of the present construction as well as characterizations of singular,
resp. intrinsically complete singular, subspaces belong to Iohvidov
[5J, [6].
Theorem 9.} seems to be new. All other results of Section 9 were
obtained by Iohvidov [5J, [6J (see also Ginzburg and Iohvidov [1J).
In connection with the intrinsic topology of definite subspaces
Noel [1J introduced new topologies on inner product spaces.
Chapter IV. Majorant Topologies
on Inner Product Spaces
Sections 1, 2, and 4 are concerned with the existence and general properties of
majorants (roughly speaking, normed topologies in which the inner prodnct is
jointly continuous). Section 3 deals with orthonormal systems. Sections 5 -7 dis-
cuss the existence and topological properties of fundamental decompositions in
terms of majorants. In Section 8 the study of the projection problem is carried on.
The following results should be specially mentioned: Theorems 3.3, 5.2,6.4, 7.2, 8.6,
Corollaries 6.3,7.4, and Lemma 7.1.
1. Majorants
Let Q; be a vector space with an inner product (. ,.). A topology 7: on

Q; is said to be a 111aiorant of the inner product (or: a majorant on Q;) if '1)
7: is locally convex, and 2) the inner product is (jointly) 7:-continuous.
It is evident that for a majorant 7: the inner square (x, x) (x E ~) is
7:-continuous. The next result says that for a topology 7: defined by a
single semi-norm the 7:-continuity of the inner square is sufficient, too,
in order that 7: be a majorant.
Lemma 1.1. Let p denote a semi-norm on the inner product space Q;.
If
(1.1) I(x, x)1 < 0:(P(X))2 (x E ~) ,
where 0: :2: 0, then

(1.2) I(x, y) I < 20: P(x) pry) (x, Y E Q;) .
Proof. Making use of inequality (1.1), for every x, y E Q; we find:

1 1 .
+
I(x, y) (y, x)1 = 1 - (x
2
+
y, x +
y) - - (x - y, x -
2
<
-
y)j
1 1
~ "2O:(P(x + y))2 + "2O:(P(x - y))2 ~ o:(P(x) + P(y))2 .
Application of this result to eX and y, where 1151 = 1, e(x, y) = I(x, y)\,
yields:
1
I(x, y)1 < "20: (P(x) + P(y))2 (x, y E Q;) .
78 IV. Majorant Topologies on Inner Product Spaces
We finally replacexby (1/P(x))x or nx (n = 1,2, ... ) according as

P(x) 0:/= 0 or P(x) = 0, and do the same for y. 0
Obviously, every majorant is a partial majorant. But a partial
majorant need not be a majorant. Moreover, while the weak topology
is always a partial majorant, there are inner products having no
majorant at all. This will be seen with the help of the following result.
Lemma 1.2. To every majorant there ex£sts a weaker majorant defined
by a single semi-norm.
Proof. Let i be a majorant defined by the semi-norms Py (y EF).
Since the inner product is i-continuous, there are indices YI' ... ,Yn E F
and a number e >
0 such that [(x, y)[ <
1 whenever Py (x), Py (y) e <
(j = 1, ... ,n). The function j j
P(x) = maxpy (x) (x E @)

l;O;j;O;n j
is a semi-norm on @, and for P(x), P(y) < e we have [(x, y)[ < 1. Thus
the topologYil :::; i defined by p is a majorant. 0
Remark 1.3. In Example II1.3.2 we exhibited a non-degenerate
inner product having no normed partial majorant, hence no normed
majorant. From Lemma 1.2 and Corollary II1.2.2 it follows that this
inner product cannot have any kind of majorant.
Consequently, in the case of Example II1.3.2 the weak topology is
not a majorant. This fact is much more general.
Theorem 1.4. The weak topology io of the non-degenerate inner prod-
~tct space @ is a majorant if and only if dim @ oc. <
Proof. Let dim @ <
00. Then, owing to Corollary 1.11.8, @ is
decomposable. Therefore, by Lemma II.11.4, any J-norm defines a
majorant iJ on @. Since a finite-dimensional vector space has only one
separated locally convex topology (the euclidean topology), io coincides
with iJ.
Suppose, conversely, that io is a majorant. Then there exist ele-
ments Zlo . • • , zn E @ and a number e >
0 such that I(x, y) [ 1 when- <
ever [(x, zj)i<e, [(y, zj)[<e (j = 1, ... , n).
Denote the subspace (Zl"'" zn) by 53. Let xE53 1 , YE@. Then
[(mx, Zj)[ = 0 < e (j = 1, ... , n; 111 = 1,2, ... ) ,
and for a suitable a > 0 also
[(ay, Zj)[ < e (j=1, ... ,n).
It follows that
[(mx,ay)[ <1 (m = 1,2, ... ) ,
i.e., (x, y) = 0.
2. Majorants and Metrizable Partial Majorants 79
We have proved that 21 C 0:: 1. Therefore 2ll ) 0:: ll = 0::. On the
other hand, according to Theorem 1.10.10, 2ll = 2. As a result,
0:: = 2. 0
2. Majorants and Metrizable Partial Majorants
A necessary and sufficient condition for the existence of majorants

reads as follows.
Theorem 2.1. A non-degenerate inner product admits a majorant if
and only if it admits a metrizable partial majorant.
Proof. Let < be a majorant of a non-degenerate inner product. Then
Lemma 1.2 and Corollary IIL2.2 assure the existence of a normed
majorant <I; this is, all the more, a metrizable partial majorant.
The implication in the reverse direction has already been proved.
Namely the normed partial majorant supplied by the proof of Theorem
III.}.1 is, in fact, a majorant (d. relation (IIL}.7)). 0
Theorem 2.1 is merely concerned with the existence of majorants.

It cannot be strengthened by saying that every metrizable partial
majorant, itself, is a majorant.
Example 2.2. Let 0:: be the vector space of finite sequences of com-
plex numbers. Denote by ek the sequence whose kth term is 1 and all
others are 0. Introduce two inner products on 0:: with the help of the
relations (e j , ek) = k 0jk' [e j , ekJ = 0jk (j, k = 1,2, ... ). Set \\xll =
= [x, x]1/2 (x E 0::). Then for any pair
m
X = 2.' ()(je j , y = 2,' Ph
j=l k=l
we have
I(x, y)1 = Ik'~: krxk Pk \< m Ilx1111Y11 = y(x) Ily!\ .
On the other hand, II· II does not define a majorant of (., .), Sl11ce
(e k , ek )"-7-oo (k"-7-oo), whereas Ilekll = 1 (k = 1,2, ... ).
If, however, the metrizable partial majorant is complete, this
phenomenon cannot occur.
Theorem 2.3. Every complete metrizable partial majorant is a
majorant.
Proof. Let < be a complete metrizable partial majorant on the inner
product space 0::. Consider two sequences {x,,}';:", {y,,}';:" ( 0:: such that
80 IV. l\Iajorant Topologies on Inner Product Spaces
lim x"
H-+OO
= lim 1'"
n-+OO
= 0 with respect to i. The linear forms Cfy" (n = 1, 2,
... ) defined by <py,,(x) = (x,Y n ) (x E @) are i-continuous, and for
fixed x the numerical sequence {<py,,(X)}~=l is bounded (tends to 0).
Thus, according to the uniform boundedness principle (valid in Fnkhet
spaces), given any neighbourhood U of OEe there exists a i-neigh-
bourhood ~ of OE@ so that x E ~ implies ipy,,(X)EU (n = 1,2, ... ).
In particular, <py,,(x n ) E U if n is sufficiently large. Hence (x"' Yn ) -+ 0
(n-> 00). 0
If a non-complete, but normed partial majorant of a non-degenerate

inner product is known, then the construction of a normed majorant
(d. the proof of Theorem III. 3.1) can be simplified.
Lemma 2.4. Assume that the norm II . II defines a partial majorant on
the non-degenerate ittner product space @. Then the nann
(2.1 ) IIxll1 = max{llxll,llxll'} (x E @)
defines a majorant on @.
Proof. From relation (III.4.2) we obtain I(x, Y)! < IIxl11 111'111
(x, y E @). 0
If II· II is quadratic, the norm (2.1) will not as a rule enjoy the same
property. Nevertheless, in Lemma 2.4 we can replace II . III by another
norm II· 112 that behaves well in this respect too (and is actually
equivalent to II . 111) .
Lemma 2.5. Assume that the norm II . II defines a partial majorant on
the non-degenerate inner product space @. Then the nonn
(2.2) IIxl1 2 = V'lNr+11X1!;Z (x E @)
defines a majorant on @. If II . II is quadratic, so is II . 112 .
Proof. Owing to relations (III.4.2) and (2.2) we have I(x, 1')1 :::::
< Ilxll' Ilyll < Ilxllz 111'112 (x,y
E @). Furthermore, if there is an inner
product [., .J on @ such that IIxl1 2 = [x, xJ (x E @), then by Lemma
III.4·3 we can find an inner product [. , .J' with Ilxll'2 = [x, xJ' (x E @);
the sum [. , .J2 = [. , .J + [. ,
.J' is clearly an inner product having the
desired property Ilxll~ = [x, XJ2 (x E@). 0
The polar topology can also be used for characterizing majorants
among normed partial majorants.
Lemma 2.6. A normed partial majorant i on the non-degenerate inner
prod~tct space @ is a majorant if and only if i' < i.
Proof. Let i be defined by the norm II· II. The relation .' < i
holds if and only if there exists an ex > 0 such that
Ilxll' < exllxll (x E @) .
3. Orthonormal Systems 81
By the definition of \\xll' this i~ equivalent to

I(x, y)1 < GX I\xll (x E a;, j\yll < 1) ,
or
I(x, y)1 < GX Ilx1111Y11 (x, yEa;). 0
3. Orthonormal Systems
Let ey (y E F) be elements of an inner product space a;. If j (e y , ey ') 1 = Oyy'

(y, y' E F), we say the system {eY}YEI' is orthonormal. In other words,
a system is orthonormal if its elements are pairwise orthogonal and
have inner square +1 or -1.
It is easy to see that the elements of an orthonormal system are
linearly independent and their span is non-degenerate.
Lemma 3.1. If x is a non-zero neutral vector in the non-degenerate

inner product space 0;, then there is an orthonormal system {e, f} ( a; such
that x E <e, f).
Proof. Let y E ~, (x, y) *-
O. Then, by Corollary 1.4.6, <x, y) is
indefinite. Let e E <x, y), (e, e) = 1. It follows from Lemma 1.9.8
that <x, y) is spanned by e and a vector f orthogonal to e. Owing to
Remark 1.3.2 fmust be negative. Hence we may require (f,f) =-1. 0
The next result relies on Lemma 3.1 and the idea of the Gram-
Schmidt orthogonalization process.
Theorem 3.2. Let Xl' x 2 , . •. be a sequence of elements in the non-
degenerate inner product space 0;. Then there exists a countable (finite or
infinite) orthonormal system {e i } ( a; whose span contains each of the
vectors Xv x 2, . . . .
Proof. Let Xi, = Yo be the first non-zero element in the sequence
{xi}' If it is non-neutral, we set el = 1(Yo,Yo) 1- 1/2 Yo' In the opposite
case we apply Lemma 3.1 to obtain an orthonormal system {e V e2 }
satisfying Yo E <el , e2 )·
Suppose that for some positive integern we have already constructed
a finite orthonormal system {el , . . . , ed whose span 53 k contains the
vectors xl> ... ,x". Let jn be the first index with Xi" E! 53 k (if such an
index exists). Put
k
Yn = Xi" - 2.,' (e i , ei ) (xin' ei ) ei .
i~1
*"
According to Corollary 1.11.9 and Lemma 1.4.22* is non-degenerate.
If (y", Yn) 0, we define eHI = I(y", Y")1- 1 / 2Y,,. Otherwise we select
an orthonormal system {eHj, eH2} C 2Ji such that y" E (eHj , ek+2)
(d. Lemma 3.1). In both cases we obtain a finite extension of the
orthonormal system {e 1 , . . . , ek } with span containing X,,+1 as well. 0
Let i be a locally convex topology on the inner product space @.
We say that the system {x"}1'EF C @ is complete in @ with respect to i
<
(or: i-complete in @), if the span x")1' EF is i-dense in @.
Theorem 3.3. Suppose that i is a separable Inajorant on the non-
degenerate inner product space @. Then every orthonormal system in @
is countable, and there is a i-complete orthonormal system in @.
Proof. Let i1 < i be a normed majorant on @ (d. Lemma 1.2 and
Corollary III.2.2). Then also i1 is separable. Let 21 be the subspace
spanned by the positive elements of an orthonormal system {e"}"G r.
The topology il121 induced by the separable normed topology i1 is
separable. On the other hand, il121 being a majorant on 21> the in-
trinsic topology iint(21 ) is weaker than i l 121 . As al result, iint(21) is
separable. Therefore, by the theory of Hilbert spaces, the set of positive
elements ey is countable.
Passing to the anti-space of @ we conclude that the set of negative
elements e1' is also countable.
Finally, let Xl' x 2 , ••• be a i-dense sequence in @. According to
Theorem 3.2 there exists a countable orthonormal system {e j } C @
whose span contains each xi" In particular, {e i } is i-complete in @. 0
The next result yields criteria for the validity of a natural generali-
zation to indefinite inner product spaces of the well-known orthogonal
expansion theorems.
Theorem 3.4. Let {ej}j':1 be an orthono1'1nal system in the no1/.-
degenerate inner product space @. The following conditions are equi-
valent:
a) @ admits a fundamental decomposition
(3.'1) @ = @+(+l@-; @+ C ~++, @- C ~--
such that the e/ s with (e j , eil = 1 belong to @+ and form there a iint (@+)-
complete orthonormal system, whereas the e/s with (e i , eil = -1 belong
to @- and form there a iint(@-)-complete orthonormal system.
b) We have
co
(3·2) .2.' (x, ej )12
1 < 00 (x E @)
i~1
and
(3-3) - L I(x, ei )1 2 :::; (x, x):::; l..,' I(x, ei )[2 (x E @) .
(ej,ej)~-1 (ej,ej)~1
3. Orthonormal Systems 83
c) We have the relations (3.2) and
co
(3.4) (x, y) = L (e i , Gi ) (x, ei ) (e i , y) (x,y E ~) •
j=1
d) The function
(3·5) Ilxll = (£I I (X, ej )12)1/2 (x E~)

is a quadratic norm on ~ and it defines a maiorant topology. With respect
to this topology we have
co
0·6) x = 1..' (e j , ej ) (x, ej ) ei (x E ~) •
i=1
Proof. The following notation will be useful:
e:t- = {e j if (e j , ej ) = 1 , e-:- = {e j if (e j , ej ) = - 1 ,
J 0 otherwise; J 0 otherwise.
a) impl£es b). If x = x+ + X-, where X+E~+, X-E@-, then by the
theory of Hilbert spaces
00 00
1.,' I(x, et)12 = L; I(x+, etw = (x+, x+)

j=1 j=1
and
co 00
1..; I(x, eiW = ,L' I(x-, ej)12 = - (x-, r) .

j=1 j=1
Therefore (3.2) and (3-3) are valid.

b) implies c). We consider an x E ~ and put
(3.7) x" =
"
2,' (e j , Gi ) (x, ej ) Gj .
i=1
Then
for i ~ n,
for i> n.
Applying relation (3-3) to the element x-x" we obtain:
00 n co
- 1.: I(x, ejW ~ (x, x) - 2: (e j , ej )l(x,6i )1 2 :'S 2; I(x, ej)12 .
j=,,+1 i=1 j=n+l
In view of (3.2) the extreme members tend to 0 as n -+00. This proves

(3.4) for x = y. Since the expressions on both sides of (3.4) represent
inner products and the corresponding inner squares coincide, it follows
from the polarization formula (I.2.3) that the inner products coincide
themselves.
84 IV. lVIajorant Topologies on Inner Product Spaces
c) implies d). By (3.2), Ilxllisfiniteforevery x. It is also clear

that IIxl1 2 = [x, x] (x E (§;), where
co
().8) [x, y] = 1: (x, ei ) (e i , Y) (x,Y E ~)
i=1
is a positive inner product on~. This inner product is even definite,
because [x, x] = 0 implies (x, ei ) = 0 (j = 1,2, ... ) and, owing to
(3.4), (x, y) = 0 for every y E~. Consequently, II· II is a quadratic
norm. It defines a majorant of (. , .), since in view of (3.4) and the
Cauchy inequality we have
00
I(x, y)1 < 1: I(x, ei)ll(ei , y)1 < Ilxllllyll .

i=1
Furthermore, considering an x E ~ and using the notation (3.7) as well
as the assumption (3.2) we find:
00
Ilx-x,,1I 2 = 1: I(x-x", e,)12 = 1: I(x, ei )l2->- 0 (n ->-(0) .

i=1 i=n+l
d) impries a). Set
(3·9) ~+ = {x: (x, ej) = 0 (j=1,2, ... )},
(3. 10) ~- = {x: (x, et) = 0 (f = 1, 2, ... )} .
Then etE~+, ejE(§;- (j = 1,2, ... ). Moreover, on account of (3.6),
00
(3.11) x = 1: (x, 4) 4 (x E ~+) ,

i=1
(3.12) x = -
i=1
00
.E (x, en ej (x E ~-) .
Since the topology involved in the convergence of the series (3.6), (3.11),
(3.12) is a majorant, inner products can be calculated term by term.
Making use of this remark it is easy to verify (3.1) and the relations
(see (3.5)). In particular, (3.11) and (3.12) hold in the respective

intrinsic topologies too, which proves the completeness properties
required by condition a). D
4. Minimal Majorants
A majorant T* on the inner product space ~ is said to be minimal, if no

majorant T =F T* is weaker than T*.
4. Minimal Majorants 85
Lemma 4.1. Let -r be a normed majorant on the non-degenerate inner
prodttct space Gr. There exists a normed maforant -roo < -r which is self-
polar.
Proof. Let -r be defined by the norm II· II. For some (X >0 we
have I(x, y)1 < (X Ilxllllyll (x,y E Gr). Setting
(4.1) Ilxll l = (X1/2llxll (x E (g)
we obtain:
I(x, y)1 ~ Ilxlll Ilylll (x,y E Gr) .
Let
IIxl12 = V~ (1lxlli + Ilxll~2) (x E Gr) .
It is easy to see that II . 112 is a norm on Gr. Furthermore, from the

relation I(x, y)1 < Ilxll~ Ilylll (see (III.4.2)) and its symmetric counter-
part I(x, y)1 < Ilxlll IIYII~ we derive
1 , 1 ---- - - - -----------.
I(x, y)1 < "2 (11x11111y111 + Ilxlllllyll{) < "2 Vllxlli + Ilxll~2 Vllylli + IlYII~2
i.e.
I(x, y)1 < IIxl1 2 IIyl12 (x,y E Gr) .
Pursuing the process, we define the norms
and verify by recursion that
(4·3) I(x, y)1 < Ilxll" Ilyll" (X,YEGr; 11,=1,2, ... ).
In view of (4.3) we have Ilxll,: ~ Ilxll". Thus, owing to relation

(4.2),
(4.4) Ilxlln+1 < Ilxll" (x E Gr; 11, = 1,2, ... ) .
Therefore the limit

(4.5) Ilxll oo = n-':>-oo
lim Ilxll" (x E (g)
exists.
11·1100 is clearly a semi-norm. But it is a norm as well. Namely
(4.4) implies
(4.6) (x E Gr; 11, = 1,2, ... );
hence, taking the definition (4.2) into account,
Ilxll,,+! 2 V~ Il xll;,2 > V~llxll~ (n = J,2, ... )

1 ,
and, consequently, Ilxlleo > ~ IlxiI I .
V2
From (4.3) and (4.5) it follows that I(x, y)1 ::;:; IlxlloollYltco (x,y E 0:).
In particular, the topology Too defined by II . 1100 is a majorant on 0:.
Moreover, from (4.5), (4A) and (4.1) we obtain the relation tlxltoo ::;
::;a 1i2 11xll (x E 0:). Thus Too ::; T.
It remains to prove that T~ = Teo' By (4A) and (4.5) we have
Ilxll" > Ilxll oo · Therefore
(4.7) Ilxtl;, < Ilxll~ (x E 0:; n = 1,2, ... ) .
In view of (4.6) and (4.7) the limit
(4.8) Ilxll co = lim
n-+OO
Ilxll;, (x E 0:)
exists and satisfies the inequality
(4.9) Ilxii OO
:s Ilxll~ (x E 0:).
On the other hand, relations (IIIA.2), (4.6), (4.8) yield:
I(x, y)1 < Ilxll~ IIYlln < IlxllOOllyll" (x,y E 0:; n = 1,2, ... ) .
Hence I(x, y)1 ~ IlxilCOIlYilco , i.e.,
(4.10) Ilxll~ ~ Ilxll co
(x E 0:) .
According to (4.9), (4.10) and (4.8) we have:
(4.11) Ilxll~ = lim
/[----'?oo
Ilxll;, (x E 0:) .
Passing in (4.2) to the limit with the help of formulas (4.5), (4.11), we
conclude that Ilxll oo = Ilxll~ . 0
Theorem 4.2. A partial majorant l' on the non-degenerate inner prod-

uct space 0: is a 1ninimal maiorant if and only if it is normcd and self-
polar.
Proof. Let l' be a minimal majorant on 0:. According to Lemma 1.2
and Corollary III.2.2 there exists a normed majorant 1'1 < T. Further-
more, by Lemma 4.1 there exists a normed self-polar majorant Too ;;;; 1'1'
The minimality of l' implies Too = 1'1 = T.
Let, conversely, l' be a normed self-polar partial majorant on 0:.
By Lemma 2.6 l' is a majorant. Let 1'1 ::; l' be also a majorant. In
view of Lemma 1.2 and Corollary IlI.2.2 we may assume that 1'1 is
normed. Then Lemma 2.6 and Lemma IIlA.1 yield: 1'1 ~ 1'; ~ 1" = T. 0
4. Minimal Majorants 87
As a consequence of Lemma 1.2, Corollary IlL2.2, Lemma 4.1, and

Theorem 4.2 we obtain the following.
Corollary 4.3. To every majorant of a non-degenerate inner product a
weaker minimal maforant can be to~tnd. 0
If there is a unique minimal majorant on the non-degenerate space

(;); then, according to Corollary 4.3, it is the weakest maforant on (;); in the
sense that it is weaker than any majorant. If there are several minimal
majorants, then a weakest majorant does not exist.
Example 4.4. Let (;); be the vector space of complex numerical
sequences x
= gi}~ satisfying the relations
;; l~il2
i=1
< 00, i=1
i (2f)2 1~2i12 < 00, .i
1=1
(21')2
1
1~2i _11 2 < 00.
For X,Y E (;);, Y = {1Ji }';"', put
00
(x, y) =
j=1
.1' (~2i-l "iJ2i + ~2i ii2j-l) .
Each of the norms
defines a majorant of this inner product, as one verifies most easily with
the help of Lemma 1.1.
Suppose II . 113 is a norm on (;); that defines a
weaker topology than any of II . 111 and II . 112' For the elements
xn = {
1
;- ()j,2n-l
1}00
+ ,/-()j,2n. '
fn yn 1=1
Yn =
_
{2Vn()j,2n-l + ,I1}00
a ()i,2n .
2 yn 1=1
(1$-* 00). On the other hand, (xn+Yn' xn+Yn)

°
we have Ilxnlll -* 0, IIYnl12 -* (n -> 00). Therefore Ilxn Ynlla -*
4 (1$ = 1,2, ... ). >
+ °
Thus the topology defined by II· 113 is not a majorant. It follows that
our inner product has no weakest majorant.
In Chapter III we introduced admissible topologies, a subclass. of
partial majorants. Below we consider the special case where the admis-
sible topology is even a majorant. (That this case is really a special one
turns out from Theorem 1.4 and Theorem IlLS.1.)
Theorem 4.5. If the admissible topology l' on the non-degenerate inner
product space (;); is a maforant, then a) l' is a minimalmaforant, b) l' is a
Banach topology, c) l' is the only admissible maforant on (;);, and d) l' is
stronger than any admissible topology on (;);.
Proof. a) Let i1 < i be a majorant. Making use of Lemma 1.2 and

Corollary IlL2.2 we find a normed majorant T2 ::;; i 1 . The majorant i2
being weaker than the admissible topology i, it is necessarily admissible.
Therefore, according to Theorem IlLS.3, T2 > T. As a result, i1 = T.
b) It follows from a) and Theorem 4.2 that i is defined by a norm
II· II· SO Q; is a normed space, and its conjugate space Q;* is a Banach
space. The admissibility of i implies that the mapping y ->- rpy' where
Ty(X) = (x, y) (x E Q;), is an isomorphism between Q; and Q;*. Since
IITyl1 = Ilyilt and, by a) and Theorem 4.2, II· II' is equivalent to II· II,
this isomorphism is continuous in both directions. Thus, along with
Q;*, also Q; must be complete.
Assertions c) - d) follow from b) and Theorem IlLS.}. 0
5. Majorants and Decomposability
Every fundamental decomposition of a decomposable non-degenerate

inner product space Q; gives rise to a quadratic norm, the corresponding
J-norm. In view of Lemma I1.11.4 the topologyTJ defined by this norm
is a majorant on @; it will be called the decomposition majorant belonging
to the fundamental decomposition in question.
A space having a quadratic-normed majorant need not be decom-
posable.
Example 5.1. Let Q; denote the vector space of doubly infinite
numerical sequences of the form x = {~i }):-oo, where ~i = (j:S °
< 00.
00
:S io(x)) and 2; l~il2 (The first condition says that all but a
i~1
finite number of terms with negative subscript vanish.) For x, y E Q;,

y= {1')i}':'oo, define
00
(x, y) = 2: ~i 17-i-1 .
i~-oo
One verifies in just the same way as in Example 1.11.3 that the n011-
degenerate inner product space Q; so obtained is non-decomposable.
Nevertheless, Q; admits a majorant defined by the quadratic norm
00 )1/2 (x a;) .
11xII = (
i~~ool~iI2 E
Next let a; be an inner product space with a quadratic-normed and

complete majorant: l(x, y)1 ~ (\ 11xlillyll (x, y E Q;), where 0; 0, >
IIxl1 2 = [x, x] (x E a;), and a; is a Hilbert space with respect to the posi-
5. Majorants and Decomposability 89
tive definite inner product [. ,.J. Then there exists a linear operator G
in @; such that
(5.1) (x, y) = [Gx, yJ (x, y E @;) .
G is called the Gram operator of (. , .) with respect to [. , .J. We have
(5.2) IIGxl1 < (Xllxll (x E @;)
and
(5·3 ) [Gx, yJ = [x, GyJ (x, y E @;) .
Theorem 5.2. If the inner product space @; admits a Hilbert majorant

T, then @; is decomposable. Moreover, the fundamental decomposition of
@; can be chosen so that its three components as well as the direct sum of
any two of them are T-closed.
Proof. Consider a positive definite inner product [. , .J such that
the norm IIxll=[x, x]1/2 (x E @;) definesT. Denote the Gram operator
of the original inner product (.,.) with respect to [., .J by G. On
account of (5.2) and (5.3) G has a right-continuous spectral function
{E(A)}l:_oo satisfying G( -0: -0) = 0, . G«(X) = I ,
(5.4) Gx = f'" A dE(A)x (x E @;) ,

-/X-O
where the integrals exist in the T-topology. It is easy to see that the
subspaces
@;- = E( -O)~ , @;o = (E(O) - E( -O))@;, @;+ = (I -E(O))@;
form a fundamental decomposition of @;; moreover, each of the sub-
spaces 'la, @;+, @;-, @;o +
@;+, @;o @;-, @;++ +
@;- is T-c1osed. 0
Remark 5.3. For a non-degenerate @;the second half of Theorem 5.2
is weaker than Corollary III.2.6.
In the rest of this section we show that the existence of a Banach

majorant is not sufficient for decomposability.
Lemma 5.4. If Tl is a decomposition majorant on the non-degenerate
inner product space @; and T is a Banach maforant on @;, then Tl < T.
Proof. Let Tl belong to the decomposition
(5.5) @; = @;+(+)'l-; @;+ c ~++, @;- c ~--.
Denote the corresponding fundamental projectors by P+, P-. From

Corollary III.2.6 and the closed graph principle it follows that P+, P-
are T-continuous. For J = P+ - P- and any norm II· II defining T we
have IIxll} = (Jx, x) < 0: IIJxII IIxll:S;; (Xl IIxll 2 (x E @;). 0
Theorem 5.5. If r is an admissible majorant on a decomposable, non-

degenerate inner prodttct space, then ris a Hilbert majorant.
Proof. According to Theorem 4.5 r is both a Banach majorant
and a minimal majorant. Therefore, in view of Lemma 5.4, every
decomposition majorant coincides with r. Henceris quadratic-normed.D
Example 5.6. Let @ be a reflexive Banach space with norm II . II not
*
equivalent to any quadratic norm. (We may choose e.g. @ = lP;
1 < p< co, P 2, since it contains closed subspaces without closed
complementary subspaces, as first proved by Murray [1].) Consider the
Banach space @ X @* of all pairs {x, r}, where x belongs to @ and r
is a continuous linear form on @, linear operations and norm being
defined by the relations
(Xl {Xl' rl} + (X2{X2, r2} = {(XlXl + 0;2X 2, C'l(Pl + (X2r2}
((Xl> (X2 E C; X l ,X2 E 0:; (Pl>r2 E @*) ,
(5.6) II{x, <p}11 = (11x11 2 + 11<p112)1f2 (XE0:, <pE0:*).
Introduce the non-degenerate inner product
(5.7) ({X, r}, {y, '1fJ}) = 1p(X) + r(y) (X,y E @; r, 'If' E 0:*).
Obviously, the norm (5.6) defines a majorant of the inner product (5.7).
We are going to prove that this majorant is an admissible topology.
Let (JJ denote a linear form on @ X 0:* which is continuous relative
to the norm (5.6). The restriction of (JJ to 0: X 0 can be looked upon as
a continuous linear form on the Banach space @. Thus for a suitable
<Po E @* we have (JJ({x, O}) = <Po (x) (x E @). Similarly, making use of
the reflexivity of 0:, we find an Xo E@ with (JJ({O, r}) = <p(xo) (<pE@*).
As a result,
({x, <p} E @x@*).
Now suppose that @ X @* is decomposable. Then, in view of
Theorem 5.5 and what we have just seen, the norm (5.6) is equivalent
to a quadratic norm on 0: X @*. The restriction of the latter to @ X 0
yields a quadratic norm on @ equivalent to the original norm. This
contradicts the choice of @.
6. Decomposition Majorants
In this section we consider decomposable, non-degenerate inner product

spaces.
Theorem 6.1. Every decomposition majorant is a minimal majorant.
6. Decomposition JliIajorants 91
Proof. By Theorem 4.2 it is sufficient to prove that every J-norm

is equivalent to its polar. But the latter property follows from Lemma
II.11.1 and the theory of Hilbert spaces:
il:rll~ = sup I(x, y)1 = sup I(x, J,Z)I = sup [(x, 'z)]1 = [Ixll]. 0
IlyllJ;i;l IlzllJ;£;l ilzllJ;i;l
In certain cases decomposition majorants are not only minimal,

but also weakest majorants.
Theorem 6.2. S~(ppose that the inner product space Q; has a fttndamen-
tal decomposition ot the torm
(6.'1 )
where one 0/ the components Q;+, Q;- is intrinsically cOJnplete. Then there
is only one minimal majorant on Q;.
Proof. Denote the fundamental projectors belonging to (6.1) by
P+, P-. Consider the fundamental symmetry J = P+ - P-, the
corresponding decomposition majorant T], and a minimal majorant
Ton Q;. Suppose that, say, Q;+ is intrinsically complete. The proof will
advance through the following stages: 1) TIQ;+ < T]IQ;+, 2) P+ is
T-continuous, 3) T] < T.
Owing to Theorem 4.2, T is defined by a norm II . II, and for some
(X> 0 we have
(6.2) Ilx+11 < (Xllx+II' = (X sup I(x+, y)1 (x' E Q;+) .
Ilyll ;i;1
In order to establish the T] -continuity of the right-hand expression,
we are going to examine the linear forms Cf v (YEQ;, Ilyll~ 1) defined
on Q;+ by the relation .
Cfy(X+) = (x+,)I) (X+Ecr+) .
Since TJ is a majorant (d. Lemma 11.'l1.4), each of the forms ff'y
is continuous on the Hilbert space Q;+ :
Moreover, as T is also a majorant, at every fixed ;t:+ the values ipy(;t:+)

(Ilyll ;::::; 1) constitute a bounded set:
(y E cr, Ilyll~1).
Thus, according to the principle of uniform boundedness, for a suitable
y> 0 we have
I(x+, )1)1 ~ yllx+ll] (x+ E Q;+, Y E cr, Ilyll:S; 1) .
Consequently, from (6.2),
(6·3) Ilxtll ::s <5llx+ll] (X+Ecr+) ,
where <5 = cxy.
Now let XE@ be arbitrary. Setting P+x = x+, with the help of
(6.3) we find:
Ilx+112 < 0211x+IIJ = 02(X+, x+) = 02(X+, x) ~ {J02 Ilx+llllxll .
Hence Ilx+11 < {J021Ixll, so that
Ilxlli = (lx, x) < {J IIIxllllxl1 = {J 112x+-xllllxll ~
< {J(2{J(j2 + 1) Ilx11 2 •
Therefore the majorant i J is weaker than the minimal majorant i. As
a result, i J = i. 0
From Theorems 6.1 -6.2 we obtain:
Corollary 6.3. If@ has a fundallwntal decomposition of the form (6.1),
where one of the subspaces @+, @- is intrinsically complete, then there is
only one decomposition majorant on @. 0
There are also other cases where the decomposition majorants

coincide.
Theorem 6.4. If the non-degenerate inner product (. , .) has a Hilbert
majorant, then (. , .) has only one decomposition maforant.
Proof. Let i be a Hilbert majorant on our non-degenerate space @.
By Theorem 5.2 @ is decomposable. Consider two fundamental sym-
metries 11,12 on @, and denote the respective decomposition majorants
by iv i 2 •
From Corollary III.2.6 and the closed graph principle it follows that
every fundamental projector of @ is i-continuous. In particular, the
operators Iv 12 and T = II 12 are i-continuous. Moreover, owing to
Lemma II.11.i, T is II-symmetric:
(Tx, Y)J, = (lI Tx , y) = (l2 X, y) =
= (x, I2Y) = (x, IITy) = (x, Ty)" (x,y E @) .
Finally, since i is a majorant and II is i-continuous, the II-norm is
i-continuous:
(6.4) (y E @) ,
where II . II is a norm defining i, and ex is a suitable positive number.
Let us show that the facts enumerated so far imply the iI-continuity
of T.
In view of Lemma 1.2.2 for every vector x and non-negative inte-
ger n we have
IlyznxllJ, = (yznx, YZ"X);,2 = (T2n-l-Ix, X);,2 ~ IIT2,,+l xll ;: Il x ll;,2.
Hence, by iteration,
IITxllJ, < IIYZ"xIIJ~" Ilxll}~2--n (n = 0, 1,2, ... ) .
7. Invariant Properties of 0:+ and 0:- 9')
Therefore, applying the inequality (6.4) to y = p"x,

IITxllI, ~ ();2- n IIT!lllxW- n Ilxll);-2-".
For n -7 <Xl this yields:
(6.5) IITxllI < IITI! IlxllI, (x E 0:) .
Consequently, T is T 1-continuous indeed.
With the help of (6.5) we find:
Ilxllj, = (]2 X, x) = (]ITx, x) = (Tx, x)I,:S IITxllI, Ilxl!/t:S
< IITllllxllj,.
Thus T2 < T 1. The relation Tl < T2 can be proved similarly. D
7. Invariant Properties of 0:+ and 0:-
In the previous section we have seen that in some instances the de-
composition majorant does not depend on how the fundamental de-
composition is chosen. In order to prove other invariant properties
of fundamental decompositions, we need the following improvement of
Theorem II.i0.1.
Lemma 7.1. Suppose that the inner product space 0: admits a funda-
mental decomposition of the form
(7.1) 0:=0:+(+)0:-; 0:+c~++, @-c~--.
Consider the corresponding fundamental projectors P+, P-, f'undamental

symmetry J = P+ - P-, and decomposition majorant T I . If ~ c @
is a positive subspace, then the linear operator P+ 12 and its inverse are
Trcontinuous. Similarly, if 2 is negative, then P-12 and its inverse are
T [continuous.
Proof. For every x E ~ Lemma II.i '1.2 yields IIP+xII I < IlxiII. If
2 is positive, then for x E 2 we also have
Ilxllj = IIP+xllj + liP-xii} = 211 P +x llj - (x, x) :S 21I P +x llj.
Analogously, for x in a negative subspace 2 we find
Ilxll} = 21IP-xll} + (x, x) < 21IP-xllj. D
Now, making use of the conclusion of Corollary 6.,), we can establish

the invariant character of the condition appearing in the same corollary
(and in Theorem 6.2).
Theorem 7.2. Let
(7.1 )
and
(7.2) (2(-) ( ~--
be two fundamental decompositions of the inner product space ij. If

(2(+)is intrinsically complete, so is (2~. Similarly, it (2(-) is intrinsically
cornplete, so is @-.
Proof. Denote the fundamental projectors belonging to (7.1) by
P+ and P-. Let (2(+) be intrinsically complete. Then (2;(+) is complete
with respect to the decomposition majorant corresponding to (7.2).
In view of Corollary 6.3, (2(+) is complete with respect to the decom-
position majorant corresponding to (7.1) as well. On account of
Lemma 7.1 the latter property is shared by the subspace P+ij(+). In
other words, P+(2(+) is intrinsically complete.
Suppose that P+(2(+) =1= (2+. As the orthogonal decomposition
theorem known from Hilbert space theory makes use of the complete-
ness of the subspace, but not of the completeness of the whole space,
we can find an element Xo =1= 0 in (2+ such that Xo ~p+(2(+). Then
Xo ~(2(+) and, owing to Remark 1.3.2, <xo, (2(+) is a positive definite
proper extension of @(+). This contradicts Lemma 1.11.4. 0
Lemma 7.3. Suppose that the inner product space (2 admits a funda-
mental decomposition of the form (7.1). Let 3 be a Positl:ve definite sub-
space of @. Then dimillt 3 ~ dimint (2+. Similarly, if 3 ( (2 is negative
definite, then dimint 3 ~ dimint @-.
Proof. Consider the fundamental symmetry] and the decomposi-
tion majorant i ] belonging to (7.1). For any subspace we (@ denote
by dim] 9)( the irdimension of 9)(, i.e., the minimal power of sets
liT ( 9)( whose span <21) is iT-dense in we. Let 3 be positive definite.
Since i ] is a majorant, ij~t (3) i]13; hence dimint 3 ~ dim] 3.
Further, by Lemma 7.1, dim] 3 :::;; dim] @I. Finally, dim] (2+ =
= dimint (2+ by definition. 0
Corollary 7.4. If (7.1), (7.2) are two fl£ndamental decompositions of
(2, then
(7·3 ) dimint (2+ = dimil1t (2( +) , dimint (2- = dim;nt @(-). 0

Remark 7.5. If the inner product space@is allowed to be degenerate,
and (cf. Lemma 1.1 '1.1)
(7.4) (2 = (20 (+) (2+ (+) (2-; (2+ ( ~+~, (2- ( ~--,
(7.5) (2 = (20 (+) (2(+) (+) (2H; (2(+) (~++, (2(-) (1:j3--
are two fundamental decompositions, then in view of Corollary 1.5.3
and Lemma 1.5.2 the subspaces (2+ (+) (2- and (21+) (+) (2H are
8. Subspaces of Spaces with a Hilbert Majorant 95
non-degenerate and isometrically isomorphic to each other. It follovvs

that in this case the relations (7.3) still hold.
On the basis of Corollary 7.4 and Remark 7.5 the rank of positivity
and the rank of negativity, introduced in Section II.i0 for non-degener-
ate quasi-negative resp. quasi-positive spaces, can be defined for
every decomposable space. Namely, if (7.4) is a fundamental decom-
position of the inner product space ~, the intrinsic dimension of ~+
(~-) will be called the ranh 0/ positivity (rank of negativity) of ~, and
denoted by x+(~) (resp. x-(~)). The cardinal number
x(~) = min fx+(~) , x-let)}
will be termed the rank 0/ indefiniteness of ~.
8. Subspaces of Spaces with a Hilbert Majorant
Let the norm II . II define a Hilbert majorant T of the inner product (. , .)

on the space~. Then a) Ilxll = [x, xJ1/2 (x E ~), where Co, .J is a positive
definite inner product on ~ ; b) ~ is complete with respect to II . II;
c) there is an ex> 0 such that
(8.1 ) I(x, y)1 ::::; C\ Ilx1111Y11 (x,y E ~) 0
Let G denote the Gram operator of(. , .) with respect to [. , .J:

(8.2) (x, y) = [Gx, yJ (x.y E ~) •
Consider a T-closed subspace 53 c et and the corresponding r. , .J-orthog-
onal proiector Q£ characterized by the properties
(8·3 )
Set
(8.4)
Obviously, G)3 is an operator in 53 satisfying the equation
(8.5) (x,y E 53) .
Therefore G)3 is the Gram operator of (. , .) with respect to C. , oJ when
both inner products are restricted to 53.
Now the following result can be added to our list of projection
theorems (see Chapters I and III).
Lemma 8.1. Let ~ be an inner product space with a Hilbert majorant
T, and let 53 be a T-closed subspace of~. The element x E ~ admits a
profection on 53 if and only if, with the notations introduced above,
Q)3G x E ffi(G£).
Proof. Let Xo be a projection of x on 53. Then for every y E 53 we

have (x, y) = (xo, y) i.E'. [Gx, yJ = [GS]xo, y] or, equivalently, [Q£Gx, yJ=
= [GS]xo ,yJ. Hence Q£Gx = GS]xo. The reasoning applies in the
reverse direction too. 0
Corollary 8.2. 53 is ortho-complementedlf and only if ffi(Q£G) =
= ffi(G£). 0
Another criterion for ortho-complementedness in a space with a
Hilbert majorant can be formulated by the aid of a new topology.
Let @ be a vector space with inner product (. ,.). Let, at the same
time, @ be a Hilbert space with respect to the positive definite inner
product [.,.J. Suppose that the norm Ilxll=[x, x]l/2 (x E @) defines
a majorant of (. ,.). Set
( 8.6) P(x) = IIGxl1 (x E @) ,
where G is the Gram operator of (. , .) relative to [. , .J, The semi-norm
p defines a topology TM, called the Machey topology of the inner product
space @.
From Theorem 8.3 below and Theorem IlLS.3 it follows that the
Mackey topology does not depend on the choice of [. , .J.
Theorem 8.3. The Mackey topology is adrnissible.
Proof. Since I(x, y)1 = I[Gx, yJI :::::; IIGxllllyll, the topology 1'M
is a partial majorant. Let the linear form ffJ be 1'M-continuous on the
space @:
IffJ(x) I < yllGxl1 (x E @) .
For Gx = 0 we have ffJ(x) = O. Thus we can define a linear form (fj on
ffi (G) setting
(fj (Gx) = ffJ(x) (x E @) .
:s
Then I(fj(Gx) I = IffJ(x) I yllGxl! (x E @). Therefore (fj can be ex-
tended to a continuous linear form on the Hilbert space~. Conse-
quently,
(fj(z) = [z, Yo] (z E ffi(G)) ,
where Yo is a suitable element of @. Hence ffJ(x) = (fj(Gx) = [Gx, YoJ =
=(x, Yo) (XE@). 0
Making use of Theorem IlLS.} we obtain:
Corollary 8.4. The lYIackey topology is stronger than any other
admissible topology. 0
Theorem 8.5. Let @ be an inner product space with a Hilbert majorant
T.The 1'-closed subspace 53 c @ is ortho-complemented l:f and only if its
Mackey topology 1'M(53) coincides with the induced topology 1'1 = Tl\l(@)I53.
8. Subspaces of Spaces with a Hilbert Majoraut 97
Proof. Let 3 be ortho-complemented. According to Theorem 8.}

'TMCI3) is admissible on 3. In view of Theorem III. 5.3 it is sufficient to
show that 'T1 has the same property. With the notations constantly
nsed in this section I(x, y)1 = I[Gx, y]1 ~ IIGxllllyll (x, y E 3), hence
'T1 is a partial majorant on 3. Consider a 'TI-continuous linear form rp
on 3: 197 (x) 1 2:: rllGxll (x E 3). By the Hahn-Banach extension prin-
ciple rp admits a'TM(@)-continuous linear extension defined on @. Thus,
owing to Theorem 8.}, there exists an y E @ such that rp(x) = (x, y)
(x E 3). Denoting a projection of y on 3 by Yo , we have (p(x) = (x, Yo)
(x E 3).
Let, conversely, iI = iM(3). For any fixed y E @ the linear form
ipy(x) = (x, y) (x E 3) is 'TI-continuous, hence iM(3)-continuous. On
account of Theorems 8.3 and IIL7.1, 3 is ortho-complemented. 0
Let us give a sufficient condition of ortho-complementedness in

spaces with a Hilbert majorant 'T. By Corollary II1.2.5, Theorem 5.2,
Lemma 1.9.2 and Lemma 1.9.3 we may restrict our attention to 'T-closed
definite subspaces.
Theorem 8.6. Let @ be an inner product space having a Hilbert

majorant r. If the definite subspace 3 c @ is 'T-closed and intrinsically
complete, then it is ortho-complemented.
Proof. The intrinsic topologYiint(3) as well as the induced topology

'T13 are Hilbert topologies on 3. The first being weaker than the second
one, the closed graph principle implies that they coincide. Therefore,
on account of the Riesz representation theorem, il3 is admissible on 3.
rt remains to apply Theorem II1.7.1. 0
The following modification of Lemma 1. 5.1 is contained in Theo-

rem 5.2. We treat it separately in order to stress its more elementary
nature and facilitate future reference.
Lemma 8.7. Let@ be an inner product space with a Hilbert majorant i,

and let 3 be a i-closed subspace of @. Then the isotropic part 3° of 3 is
i-closed. Moreover, 3 is the orthogonal direct sum of 3° and a 'T-closed
non-degenerate subspace 3 1 •
Proof. 3.1. is 'T-closed by Lemma II1.2.4. Hence 3° = 3 n 3.1. is

'T-closed. According to the orthogonal decomposition theorem of
Hilbert spaces, 3° has a i-closed complementary subspace 3 1 in 3.
Clearly, 3 1 is non-degenerate and orthogonal to 3° (d. Lemma 1. 5.1). 0
98 IV. Majorant Topologies on Inncr Product Spaces
Notes to Chapter IV
Similarly to Chapter III, a considerable part of the present chapter is

closely related to the duality theory of topological vector spaces (see
e.g. Bourbaki [1J, Kothe Robertson and Robertson [1], Schaefer [1]).
Theorem 2.1 belongs to Wittstock [2J. Conditions for the existence
of majorants have also been studied by Jarchow [1J, [2J, who showed
that in a class of structures wider than that of locally convex topologies
every inner product has a "majorant".
For the present proof of Theorem 2-3 see Ginzburg and Iohvidov [1].
Lemmas 2.4-2.6 and related results can be found in the papers of
Aronszajn [1J and \Vittstock [3J.
Theorem 3.2 was proved by Savage [1]. For spaces with a separable
Hilbert majorant Theorem 3.3 is due to Nevanlinna [3J, and for those
with an arbitrary separable majorant to Jelinek and Virsik [1]. Theo-
rem 3.4 was obtained by Nevanlinna [3J and reformulated by Ginzburg
and Iohvidov [1].
Lemma 4.1 is a result of Aronszajn '11]' Variants of its proof have
been given by Witt stock [3], [5]. Theorem 4.2 and Corollary 4.3 belong
to Witt stock [3]. Example 4.4 and Theorem 4.5 are borrowed from the
paper of Jelinek and Virsik [1J (for Theorem 4.5 d. also Jarchow [1J).
An analogue of Example 5.1 appears in a short note of Ov6il1ni-
kov ['1] as a modification of Example 1.11.3. The first formulations of
Theorem 5.2 were given by Hestenes [1J and, independently, Nevan-
linna [3]. The proof only requires a part of the spectral theorem that
can easily be reduced (see Lemma V1.8-3 below) to the more elementary
subject of positive square roots in Hilbert space. Example 5.6 is due
to \Vittstock (personal communication).
A special case of Theorem 6.1 was obtained by N evanlinna [3 J, [4J;
the general case is due to Wittstock [2]. Criteria for the existence of
a weakest majorant, also Theorem 6.2 in a stronger form, are contained
in the works of Wittstock [1J, [3J, [5]. A special case of Corollary 6.3
was earlier established by Iohvidov and KreIn [1].
In Theorem 6.4 the existence of a Banach majorant would suffice.
The original proof belongs to Wittstock [2J, and the present one to
Bognar [7]. In the text of the proof we reproduce an argument of
Reid ['1] rather than refer to the result we need (the latter has first
been obtained by Krein [2J). An example showing that Theorem 6.4
cannot be extended to spaces witb a Frechet majorant was given by
Wittstock [2J.
Lemma 7.1 goes back to Ginzburg
Notes to Chapter IV 99
Theorem 7.2 is due to WiHstock [4]. For special cases see Iohvidov
and KreIn [1], Scheibe [1], Ginzburg [3], Bognar and Kramli [1]. All of
these papers contain special cases of Corollary 7.4 too.
The first steps towards Lemma 8.1 were made by Nevanlinna [4J
and Louhivaara [4J. The lemma itself belongs to Browder [1] (d. also
Louhivaara [5J). Related facts for non-symmetric bilinear forms have
been obtained by Littman [-1J and Browder [2J (d. also Louhivaara [6J),
and in a special case by \iVittstock [1J. An account of all these results
appears in the survey article of Louhivaara [7J. The most complete
study of the problem, however, is due to Ginzburg [4J (d. also Ginzburg
and Iohvidov [1J).
On the initiation of Nevanlinna [1J the theory of projections in an
inner product space with a Hilbert majorant has been utilized for
extending the variational method to general elliptic or even non-
elliptic boundary value problems (Louhivaara [1], [2J, [3J, Browder [1J,
[2J, Littman [I], Hildebrandt [1J; for a summary d. also Dolph [1J and
Louhivaara [7J).
For the definition and properties of the Mackey topology when the
existence of a Hilbert majorant is not assumed see e.g. Robertson and
Robertson [1J.
Theorem 8.5 was announced by Ginzburg [4]. The present proof of
this theorem as well as that of Theorem 8.3 is taken from the paper of
Ginzburg and Iohvidov ['1], where the case of a possibly non-symmetric
bilinear form with a reflexive Banach majorant is considered. Theo-
rem 8.6 appears in the same paper.
Non-symmetric bilinear forms with a Banach or Hilbert majorant
were also investigated by ] alava [2J.
Iohvidov [10J made a study of non-closed maximal definite sub-
spaces in an inner product space having a Hilbert majorant.
Chapter V. The Geometry of Krein Spaces
Krein spaces, the most important type of inner product spaces, can roughly be
characterized as non-degenerate, decomposable, complete spaces. Sections 1-2 deal
with their global properties. In Sections 4, 7, 9, 10 maximal semi-definite subspaces
of Krein spaces are investigated, while Sections 3, 5, 6, 8 are essentially concerned
with ortho-complemented subspaces. Theorems 1.3, 3.4, 3.5, 4.2, 5.3,5.7,6.3,7.1
and 9.1 give an idea of the contents of the chapter.
1. Krein Spaces
If an inner product space s;;, admits a fundamental decomposition of the

form
(1.1) s;;,+ c ~++ ,
where the subspaces s;;,+ and s;;,- are intrinsically complete, then we shall
say that s;;, is a Krein space.
Lemma 1.'11.1 and Theorem IV.J.2 yield:
Theorem 1.1. A Krein space .S) is non-degenerate and dec01nposable.
Every fu-ndamental decomposition of s;;, looks lil?e (1.1), and the compo-
nents s;;,+, Sd- are intrinsically complete. 0
Loosely speaking, a Krein space is a non-degenerate, decomposable,
complete inner product space.
According to our definition, the class of Krein spaces includes Hil-
bert spaces (s;;,-= 0) as well as anti-spaces of Hilbert spaces (s;;,+ = 0).
Furthermore, on account of Corollary 1.11.8 and Lemma I.'I1.1, every
finite-dimensional non-degenerate inner product space is a Krein space.
A connection between Hilbert spaces and arbitrary Krein spaces is

given by the following consequence of Lemmas II.11. 3, II.11.2, and
TheorEm 1.'1.
Corollary 1.2. The deco111,posable, non-degenerate inner product space
@ is a Krein space If and only if, for every fundamental symmetry I, the
J-inner product tu.rns @ into a Hilbert space. 0
1. Krein Spaces 10'1
The rest of this book is devoted to Krein spaces and their linear
operators. The letter Sj, when used without explanation, will
always denote a Krein space.
Theorem 1.3. A vector space Q; with an inner product (. , .) is a Krein

space if and only if 1) (. , .) has a Hilbert maforant T and 2) the Gram
operator of (. , .) with respect to any positive deft:nite inner product [. , .J
such that the norm
(1.2) Ilxll = [x, x]1/2 (x E Q;)
deFnes T 1:S completely invertible.
Proof. Suppose that
(1·3) ~=Q;"(+)Q;-; Q;+cl,lS++, Q;-cl,lS--,
where Q;+ and Q;- are intrinsically complete. Denote the fundamental
symmetry belonging to (1.3) by]. The decomposition majorant T J
defined by II· IIJ is a I-Iilbert majorant on Q;. Since, in view of
LemmaIL1L1, (x,y) = (j2x,y) = (]x'Y)J' the Gram operator of (.,.)
with respect to (. , ')J is ], a completely invertible operator.
Owing to Theorem IlL3.3 (., .) has only one Hilbert majorant.
Let [. ,.J j (f = 1,2) be two positive definite inner products on Q; such
that the norms Ilxlli = [x, X]J/2 (x E Q;) define the Hilbert majorant of
(. , .). Since II· 111 is equivalent to II· 112, there exist linear operators G12 ,
G21 on Q; satisfying the relations
[x, yJ2 = [G 12x, Y]l' [x, yJl = [G 21 X, yJ2 (x,y E Q;) .
We have G12 G21 = G21 G12 = I; in particular, G12 and G21 are completely
invertible. Now if Gi denotes the Gram operator of (. , .) relative to
[. , .J i , i.e.
(x,y E Q;; f = 1,2) ,
then G2 = G21 G1 , G1 = G12G2 ; hence G2 is completely invertible if and
only if G1 is so.
Let, conversely, (x,y) = [Gx,yJ (X,YE G¥), where the norm 11·11
given by (1.2) defines a I-Iilbert majorant T and the T-continuous linear
operator G is completely invertible. Then ~ is non-degenerate and
decomposable (d. Theorem IV.5.2). Thus, according to Lemma 1.11.1,
Q; has a fundamental decomposition of the form (L3). On account of
Corollary 1.2 it is sufficient to show that the corresponding decomposi-
tion majorant T J coincides with T.
The complete invertibility of G and the closed graph principle
imply TM = T. Therefore, by Theorem IV.8.3, the majorant T is an
admissible topology on 0:. Hence, in view of Theorem IVA.S, T is
a minimal majorant. On the other hand, Lemma IV.5A yields T J ;:;; T.
Consequently, T J = T. 0
102 V. The Geometry of Krein Spaces
From the well-known fact that a Hilbert space is completely char-

acterized by its dimension one readily obtains the classification of
Krein spaces by means of their ranks of positivity and negativity.
Theorem 1.4. Two Krein spaces SjI' Sj2 are isometrically isomorphic
if and only 1:j X+(Sj1) = X+(Sj2) and x-(Sj}) = X-(Sj2)' 0
Owing to Corollary IV.6.3 all J-norms on a Krein space Sj are

equivalent. They will be called natural norms on .\j, and the Hilbert
majorant they define, the strong topology of Sj.
In view of Theorem 1.3 the strong topology of ~~ is equal to the
Mackey topology. Therefore it may be denoted by T!,l(Sj).
As a corollary of Theorem IY.8-3 and Lemma 11.11.4 we have:
Theorem 1.5. The strong topology ojSj is an admissible majorant. 0
In a Krein space Sj all notions involving some kind of topology such
as convergent, closed, dense, continuous, homeomorphic, (infinite) dimen-
sion etc. will be understood, unless otherwise stated, to refer to the
strong topology Tlvl(Sj). The closure of a set 2kSj will be denoted by 21.
One more definition. Let 2 + Wc = 2 1 , where 2, Wl and 21 are
closed subspaces of Sj. In this case the cardinal number dimWl is
called the codimension of 2 with respect to 21' and denoted by codims,2.
On account of the closed graph principle, the definition is correct.
Furthermore, according to Corollary 1.1.2, it is compatible with the
definition of codimension we gave in Section 1.1.
2. Krein Spaces as Completions
It is well known that every positive definite inner product space Q; can
be embedded in a Hilbert space Sj as a dense subspace, and that this
procedure is essentially unique. The corresponding result for more
general inner product spaces reads as follows.
Theorem 2.1. Let
(2.1 ) 0;+ C \13+1 , 0;- c \13--
be a f~mdatnental decomposition of the decomposable, non-degenerate inner
prodnct space~. Denote the intrinsic completion of 0;+ and ~- by Sj+ and
Sj-, respectively. Then the cartesian prodnct Sj = Sj+ X Sj- is a Krein
space, and 0; l:S isometrically isomorphic to a dense subspace of Sj.
Let
(2.2) Q;j C \13++ , ~i c \13-- (j = 1,2)
3. Subspaces 103
be two /lMtdamental decompos1:tions of Q;, with corresponding decomposi-

tion majorants if Denote the intrinsic completion of by S";)r If Q;r
i1 = i 2, and only then, there eX1:sts an isometrical isontorphism between
S)l = S); X S);:- and ~)2 = S); X s;,);- which leaves the elements of Q;
unchanged.
Proof. That S) is a Krein space and Q; can be identified with a
dense subspace of S) is clear. Let i l = i 2. If Xl E S";)V there exists
a sequence {:V(1!l}~~l C Q; converging to Xl in the strong topology
iM(s;,1 1 ). Then {y(n l } is a Cauchy sequence relative to iM(S";)I)IQ; =
= 7:1 = '2 = 'M(S)2)1Q;, so that it has a 7:M(S)2)-limit X 2 E S)2' The
mapping T uniquely defined by the relation TX1 = x 2 is an isomor-
phism between S)1 and S)2 which leaves the elements of Q; unchanged.
Since iM(S)l) and'M(S";)2) are majorants, T preserves inner products
too.
Suppose, conversely, that there exists an isometrical isomorphism
between .pI and S)2 which leaves the elements of Q; unchanged. Then
'2
we may regard 7:1 and as induced on Q; by two decomposition major-
ants of the same Krein space. Consequently, '1 = '2' D
Remark 2.2. Comparing Theorem 2.1 with Corollary IV.6.3 and

Theorem IV.6.4, we obtain sufficient conditions on Q; in order that its
"completion" S) be essentially unique.
3. Subspaces
Corollary 1.2 and Theorem IV.5.2 yield:

Theorem 3.1. If 53 is a closed subspace of the Krein space S";), then 53
admits a fundamental decomposi#on whose c01nponents, as well as the
direct sum of any t'liiO of thern, are closed. D
On the other hand, a non-closed subspace of S) need not be de-
composable.
Example 3.2. The non-decomposable space Q; of Example IV.5.1
can be embedded in a Krein space S) as follows. Let S) consist of all
<
co
doubly infinite numerical sequences {~i}~-oo satisfying 2-' l~il2
< 00. Set i~-co
00
({~i}' {1]j}) = E ~i1]-j-l'

j=-oo
The relations
co
[{~j},{1]i}J = E ~i 1]i'
1=-00
define a Hilbert majorant of (. , .), and the respective Gram operator

G is completely invertible: G{~i}~-CXl = {Li-d~-CXl' By Theorem
1.3 Sj is a Krein space.
A closed, non-degenerate subspace of a Krein space, though de-
composable, need not be a Krein space.
Example 3.3. In Example I.9.7 @ is a Krein space and 2 is a closed
positive definite subspace of @. The elements
x(n) = {~~n)}~1 (n = 1,2, ... )
defined by the relations
/:(n) -
"21- 1 -
(?J'
-
- 1) -1/2 (f < n) ,
~~i) = 2i(2j - 1)-3/2 (i < n) ,
t(n) - 0 (j> 2n)
"i -
belong to 2 and we have
Ix(n) - x(nt)l.\l --? 0 (m, n --? (0) .
If x(n) had an intrinsic limit x(O) E 2, then the coordinates of the latter
would be
/:(0.)
"21-1
= (21' _ 1) -1/2 , (j=1,2, ... ),
so that J3
i=1
1~~O)12 = 00. Hence 2 is not intrinsically complete.
According to the next theorem, a closed subspace of Sj is a Krein
space if and only if it is ortho-complemented.
Theorem 3.4. A subspace 2 of the Krein space Sj is ortho-complement-
ed if and only if 1) 2 is closed and 2) 2 is a Krein space.
Proof. If 2 is ortho-complemented, then it is closed by Corollary
III.2.S. On account of Corollary IV.8.2 and Theorem 1.3, a closed
subspace 2 c Sj is ortho-complemented if and only if
(3.1) ffi(G.\l) = 2,
where G.\l is an operator in the subspace 2 determined by the relation
(x, y) = (G.\lx, y) J (x,y E 2) with a fixed fundamental symmetry J of Sj.
Since G.\l is J-symmetric, (3.1) is equivalent to the complete invert-
ibility of G.\l, i.e. (see Theorem 1.3), to 2 being a Krein space. D
Let us give another characterization of ortho-complemented sub-
spaces in Sj.
Theorem 3.5. A subspace 2 of the Krein space Sj is ortho-complement-
ed if and only if 1) 2 is closed, 2) 2 is non-degenerate, and 3) for any
fundamental decomposition
(3.2) 2=2'(+)2-; WcI,15H, 2-c':+5--
of £ a fundamental decomposition
(3·3) S)- c ~--
of S) can be found so that

(3.4) £+ CS)+, £- C,S)- .
Proof. Let £ be ortho-complemented. Then £ and £1 are closed
and non-degenerate (d. Lemma 1.9.1, Corollary 1.9.5, and Corollary
IIL2.5). In view of Corollary 1.2, Theorem IV.S.2 and Lemma 1.11.1
£ and £1 admit fundamental decompositions (3.2) and
£1 = £;(+)£~; £t c ~++, £~ c ~--,
respectively. The subspaces
(3·5) S)+=£+(+)£;, Sj-=£-(+)£~
satisfy (3.3) and (3.4)·
Suppose, conversely, that £ is closed and non-degenerate, and that
for suitable decompositions (3.2), (3.3) the inclusions (3.4) hold. By
Corollary III.2.6 £+ is closed in £. Consequently, £+ is closed in Sj. In
particular, £+ is ortho-complemented in the Hilbert space Sj+. Analo-
gously, £- is ortho-complemented in Sj-. Thus for certain subspaces
£;, £~ we have ().5). The subspace £1 = £;(+)£;:- is orthogonal to £
and satisfies the relation £ (+) £1 = Sj. 0
With a view to later application we mention the following conse-
quence of Theorem 1.5 and Theorem III. 6.1 :
Theorem 3.6. If £ is a subspace of Sj, then £11 = £. 0
Corollary 3.7. If £ is a closed s16bspace of .'I'd, then the isotropic part
of £1 coincides with that of £. 0
4. Maximal Semi-definite Subspaces

The first three results presented below establish conditions (one nec-
essary, one necessary and sufficient, and one sufficient) on a positive
(negative) subspace of Sj for being maximal positive (maximal negative).
Theorem 4.1. Every maximal positive or maximal negative s'Vtbspace
of a Krein space is closed.
Proof. Since the strong topology is a majorant, the closure of a
positive (negative) subspace is positive (negative). 0
Theorem 4.2. Let
(4.1 ) sy c ~c + ,
106 Y. The Geometry of E:l'ein Spaces
be a fltndamental decomposition of the Krein space Sj. Denote the cor-

responding fundamental projectors by P+, P-. The positive (negative)
s~tbspace 53 c Sj is maximal positive (maximal negative) if and only ij
P+53 = Sj+ (resp. P-53 = Sj-).
Proof. Let 53 be positive and P+53 = Sj+. Then, owing to Theo-
rem II.i 0.1, 53 is maximal positive.
Suppose, conversely, that 53 is maximal positive. By Theorem 4.1
and Lemma IV.7.1 p-t 53 is closed. The subspace Sj+ being intrinsically
complete (d. Theorem 1.1), so is P+53. If p"53"#- Sj+ then, according
to the orthogonal decomposition theorem of Hilbert spaces, there
exists a vector x satisfying the relations x E S)+, x~P+53, x"#- 0; in
view of Remark 1.3.2, <53, x) is a proper positive extension of 53. Con-
tradiction. 0
Lemma 4.3. Every closed, maximal positive definite (maxilnalnegative

definite) subspace of Sj is maximal positive (111aximal negative).
Proof. Let 53 c &d be closed and positive definite. Following the
notations and proof of Theorem 4.2 we find that in tbe case P+53 = &d+
the subspace 53 is maximal positive, while in the opposite case 53 admits
a positive definite proper extension. 0
Lemma 1.6.3 says that the orthogonal companion of a maximal

positive (maximal negative) subspace is negative (resp. positive). In
Krein spaces we can prove more.
Theorem 4.4. If 53 is a tnaximal positive (tnaximal negative) subspace
of &d, then 53 1 is maximal negative (maximal positive).
Proof. Let 53 be maximal positive. On account of Lemmas 1.6.3
and III.2,4 53 1 is negative and closed. Consider a fundamental de-
composition (4.1) with corresponding projectors P+, P-. By Lemma
IV.7.1 and Theorem 1.1 p-53 J· is an intrinsically complete subspace of
the intrinsically complete space &d-.
Suppose that 53 1 is not maximal negative. Then, according to
Theorem 4.2, P-53 1 "#- SJ-. Consequently, there exists an XE Sr satis-
fying the relations x~P-531, x"#- o. In particular, (x, x) <
0 and
x E 53 11 . But, in view of Theorems 3.6 and4.1, 8 11 coincides with the
positive subspace 53. Contradiction. 0
In Krein spaces also a converse of Lemma 1.6.3 holds.

Lemma 4.5. If 53 is a positive (negative) subspace with negative
(Positive) orthogonal companion in Sj, then the closure of 53 is maximal
positive (maximal negative).
5. Uniformly Definite Subspaces 107
Proof. Suppose that 2 is positive, but its closure is not maximal

positive. Consider a fundamental decomposition (4.1) and the cor-
responding projectors P+, P-. On account of Theorem 4.2 and Lem-
ma IV.J.1 there exists an x E ~+ with the properties P+2, x O.
In particular, x is a positive element of 2.1... 0
We end this section by introducing a quantity which measures the

non-maximality of positive subspaces. It will next appear in Chap-
ter VII.
Let 2 be a closed positive subspace of the Krein space ~. Let
2 1 ) 2 be a maximal positive subspace of ~ (d. Section 1.6). The co-
dimension of 2 relative to 21 will be called the plus-defect of 2 and
denoted by d+ (£).
From the next result it follows that the cardinal number d+(2)
does not depend on the choice of the maximal positive subspace 21 ) 2.
Lemma 4.6. If 2 C 21 c~, where 21 is maximal positive, then for any
fundamental decMnposition (4.1) and respective fundamental profectors
P+, P- we have
codim's12 = codim.p+ P+2 .
Proof. Let ill( be a closed complementary subspace to 2 in 2 1 ,

Then, by Theorem 4.2 and Lemma IV.J.1, P+ill( is a closed comple-
mentary subspace to P+2 in ~+, and dimP+ill( = dim ill(. 0
5. Uniformly Definite Subspaces

Consider a semi-definite subspace 2 in the Krein space ~. According
to Theorems 3.4 and 1.1 2 is ortho-complemented if and only if it is
closed, definite, and intrinsically complete. On the other hand, the
intrinsic topology Tint(2) of a definite 2 being weaker than the topology
T:M(~)12 induced on 2 by the strong topology of ~), Corollary 1.2 and
the closed graph principle imply that a closed definite subspace 2 is
intrinsically complete if and only if Tint(B) = TM( ~) 1£. This motivates
the following definition.
A (closed or non-closed) subspace 2 of the Krein space ~ is said to be
uniformly positive (~tniformly negative), if 1) 2 is positive definite
(negative definite) and 2) Tint(2) = TM(~)12.
In other words, 2 c ~ is uniformly positive if
(5.1) (x, x) > (Xllxll] (x E B) ,
and uniformly negative if
(5.2) (x, x) S - (Xllxllj (x E 2) ,
108 V. The Geometry of Krein SpacE's
where II . IIJ is any fixed natural norm and ex is a positive number

depending on 52.
Uniformly positive and uniformly negative subspaces may be
termed uniformly definde.
The above arguments and definitions lead to the following two
results.
Lemma 5.1. A closed definite subspace of Sj is intrinsically complete

if and only if ilis 1mifor1l11y definite. 0
Theorem 5.2. A semi-definite subspace of Sj is ortho-complemented if
and only if it is closed and 'uniformly definite. 0
For a possibly indefinite subspace 52 c Sj we have:
Theorem 5.3. A subspace 52 of the Krein space Sj is ortllo-comple-

mented if and only if 52 is the orthogonal direct sum of a closed uniformly
positive and a closed uniformly negative subspace.
Proof. By Theorem 5.2 and Lemma I.9.2 the condition is sufficient.
To prove necessity, suppose that 52 is ortho-complemented. Then, on
account of Theorem 3.4, 52 is the orthogonal direct sum of a positive
definite and a negative definite subspace. Application of Lemma I.9.2
and Theorem 5.2 yields the desired conclusion. 0
Having made clear the basic importance of uniformly definite sub-

spaces, we are going to present further characterizations of them,
valuable partly from the intuitive, partly from the technical point of
view.
Lemma 5.4. A subspace of Sj is uniformly positive (uniformly nega-
tive) if and only If its closure is so.
Proof. Employ the description (5.'1) - (5 .2) of uniformly definite
subspaces. 0
From Lemma 5.4 and Theorem 5.2 one obtains:
Coronary 5.5. A semi-definite subspace of Sj is uniformly definite i]
and only if its closure is ortho-comple11lented. 0
Lemma 5.4 combined with Corollary 5.5 and Theorem 3.5 yields:
Theorem 5.6. The subspace .53 c Sj is 'Uniformly positive (uniformly

negative) if and it Sj admits a fundamental decomposition
only
(5·3) Sj = Sj+( +)~;r; ,~+ c 1.l3++ ,
such that 52 c &:,+ (2 c Sj-). 0

6. Non-uniformly Definite Subspaees 109
In contrast to Theorem 5.6, the followi11g characterization of uni-

f01:mly definite subspaces is useful in cases where a fundamental
decomposition of 5) is fixed in advance.
Theorem 5.7. Consider a fundamental decomposition (5.3) of the
Krein space ~"). Denote the corresponding funda1nental symnletry by ].
A subspace S3 C 5) is uniformly positH'e if and only if the ang~tlar
operator K+(S3) oj S3 with respect to ~)+ exists and satisfies the inequality
<
liKe (S!) IIJ 1. Analogously, .3 is uniformly negative if and only it
K-(S3), the angular operator oj S3 relat~'ve to 5)-, exists and satisfies the
condition IIK-(S3)II J 1. <
Proof. \iV e restrict ourselves to the positive case. Let S3 be uni-
formly positive. Then, according to Theorem H.1 '1.7, K+ (3) exists.
Moreover, for
(5 A)
relation (5.1) yields
(5.5) Ilx+ll} - Ilx-ll} > (X (1Ix+llj + Ilx-II})
i.e.
(5.6)
where (X > O. Thus

IIK+(3)II J :::;: 1 __ ~ < 1.
-1+(X
Let, conversely, .3 be a subspace such that K+(3) exists and

IIK+(3)II J = Q < 1. .. b
Then for the posItlve num er (X =
i-I}
--- - an d
+
'1 I}
any vectors (5A) the relation (5.6) holds. From (5.6) we obtain (5.5)
and, finally, (5.1). 0
6. Non-uniformly Definite Subspaces

By Theorem 5.6 every non-zero Krein space contains uniformly definite
non-zero subspaces. Next we investigate the existence of definite sub-
spaces which are not uniformly definite (non-uniformly definite sub-
spaces for short).
Lemma 6.1. Let 5) be a Krein space with rank of positivity 71:+(5)) and
rank of negativity %-(5)). In order that every positive definite subspace
3 c 5) would have a pos1:tive definite closure it is necessary and sufjicient
that either 71:+(5)) <00 or %-(5)) = o. Similarly, the closure of every
negative definite 3 C S) is negat1:ve definite ij and only 'if at least one of the
conditions w(5)) < 00, x+(5)) = 0 is satisfied.
110 v. The Geometry of Krein Spaces
Proof. Owing to Theorem II.1 0.1 the conditions are sufficient. Next
suppose that x+(5)) = 00, x-(5)) >
O. Let {e j };'" and {II}' respectively,
be orthonormal systems contained in the components 5)+ and 5)- of
a fundamental decomposition of 5). Denote by 2 the set of all elements
+ fMl'
00
of the form 2.Or.lj where o<'j ;:j= 0 only for a finite number of
1 00
indices and, moreover, 1:o<'j = 0<.1 = {Jl (d. also Example 1.4.9 and the
2
proof of Theorem III.S.1). Then £, is a positive definite subspace, but
the vectors
(n = 1,2, ... )
converge to the neutral non-zero vector e1 + fl . 0

From Lemmas 5.4 and 6.1 we obtain:
Corollary 6.2. Every indefinite Krein space of infinite dimension
contains non-closed, non-uniformly definite subspaces. 0
As to the existence of closed, non-uniformly definite subspaces, we

have the following.
Theorem 6.3. Let S) be a Krein space with rank of indefiniteness
x(5)). In order that every closed, positive definite (negative definite) sub-
space of 5) would be uniformly positive (uniformly negative), it is
necessary and.sufficient that x(5)) 00. <
Proof. Consider a fundamental decomposition
(6.1) 5)=5)+(+)5)-; 5)+c~++, 5)-c~--
and a closed, positive definite subspace 2 C 5). On account of Theo-

rem II .11. 7 we have the inequality
(6.2) IIK+x+IIJ <1 (x+E~(K+), Ilx+II J < 1) ,
where K+ is the angular operator of 2 with respect to 5)+ and] is
the fundamental symmetry belonging to (6.1). Moreover, owing to
Lemma IV.7.1, ~(K+) is closed.
Let x(5))< 00. Then either the domain ~(K+) c 5)+ or the range
ffi(K+) C 5)- of K+ has finite dimension. In the first case (6.2) implies
that
(6·3)
where y < 1. In the second case the same conclusion can be drawn by
splitting ~(K+) into the orthogonal direct sum of m(K+) and its finite-
dimensional orthogonal companion in ~(K+). Therefore, in view of
Theorem 5.7, 2 is uniformly positive.
6. Non-uniformly Definite Subspaces 11-1
Now let x(Sj) = 00. Choose two orthonormal systems {e j };:" C Sj+,
{If};:" C Sj-. Define a linear operator K+ in the following way:
~(K+) = {2.;IX.e. :
i=t 1 1
K + ( 2.,' IX ·e· = 2,.' -._7_ IX .f.

00 ) 00 •
i=t 1 1 i=t 7 +1 1 1
Then '1)(K+) is closed, (6.2) is fulfilled, but (6.3) does not hold for any
y < 1. On account of Lemma IV.7.1, Theorem II.H.7 and Theo-
rem 5.7, respectively, the subspace
£ = {x~ + K+x+: x+E;D(K+)}
is closed, positive definite, but not uniformly positive.
The case of negative definite subspaces can be treated by passing
to the anti-space of Sj. D
Remark 6.4. For uniformly definite subspaces, according to their

definition and Lemma 5.1, the properties of being closed or intrinsically
complete imply each other. However, by the same lemma, a non-
uniformly definite subspace cannot have more than one of these prop-
erties. The existence of closed, non-uniformly definite subspaces has
just been discussed in Theorem 6.3. For intrinsically complete, non-
uniformly definite subspaces the existence problem is settled by the
next example.
Example 6.5. Consider a Krein spaceSj with x+(Sj) = 00, O<x-(Sj)<
< 00. Let £ be a positive definite subspace of Sj such that the closure
£1 of £ is degenerate (d. Lemmas 6.1 and I.4.4). Applying Remark I.5.5
and Theorem I.S.4 to the inner product space £1 we find a positive
definite subspace £2 ) £ with the property
£2+£~=£1'
where £~ stands for the isotropic part of £1' On the other hand, by
Lemmas IV.S.7 and I. 4.4, there exists a closed positive definite sub-
space 23 such that .
23+2~=£1'
In view of Theorem 6.3 23 is uniformly positive, hence intrinsically
complete. But, according to Lemma I.5.2, £2 is isometrically isomorphic
to 23, In particular, £2 is intrinsically complete. At the same time,
having a degenerate closure £1> the subspace £2 is not uniformly positive-
(d. Lemma 5.4).
7. Maximal Uniformly Definite Subspaces

A uniformly positive (uniformly negative) subspace of the Krein
space S) is said to be maximal uniformly positive (maximal uniformly
negative) if it is not contained in any other uniformly positive (1.mi-
formly negative) subspace of S).
From Theorem 5.6 and Lemma 1.11,4 we obtain:
Theorem 7.1. The subspace £ c S) is maximal uniformly positive
(maximal uniformly negative) if and only if S) has a fundamental de-
composition
(7.1) S) = S)+(+)S)-; S)+ C ~++ , S)- c ~--
such that £=S)+ (£=S)-). 0
Theorem 7.1 and Remark 1.11.6 yield:
Corollary 7.2. Every maximal uniformly positive (maximal uniformly
negative) subspace of S) is maximal positive (maximal negative). 0
Remark 7.3. Corollary 7.2 should be compared with Lemma 6.1
which, in view of the existence of maximal definite subspaces con-
taining a given definite subspace, implies that in "most" Krein spaces
a maximal positive definite subspace £ need not be maximal positive.
(For a closed £, however, see Lemma 4.3.)
The following consequence of Theorem 7.1 should also be mentioned.
Corollary 7.4. If £ is a maximal uniformly positive (maximal uni-
formly negative) subspace of S), then £1 is maximal uniformly negative
(maximal uniformly positive). 0
The next two results are concerned with complementary subspaces.
Lemma 7.5. Suppose that the subspace £ c S) contains a maximal
uniformly positive (maximal uniformly negative) subspace. Then £ is
ortho-complemented and £1 is uniformly negat£ve (uniformly positive).
Proof. Let £1 C £ be maximal uniformly positive. On account of
Corollary 7,4 53t is uniformly negative. Therefore the subspace £lC £t
is uniformly negative. Moreover, owing to Lemma III.2,4, 53 1 is closed.
By Theorem 5.2 £1 is ortho-complemented. Hence, in view of Lemma
1.9.1, also £11 is ortho-complemented. It remains to observe that
£11 = £ (d. Theorem 3.6). 0
Lemma 7.6. Let £1 be a maximal uniformly positive and £2 a maximal
negative subspace of the Krein space S). Then £1+532 = S). The same
conclusion holds if £1 is maximal uniformly negative and 532 is maximal
positive.
8. Regular and Singular Subspaces 113
Proof. According to Theorem 7.1 Sj has a fundamental decom-
position (7.1) such that 521 = Sj+. On the other hand, Theorem 4.2
yields P-52 2 = Sj-, where P- is the fundamental projector with
?R{P-) = Sj-. Thus
Sj+ =~1 C 521 + 52 2 ,
Sj- = P-~2 C P+~2 + ~2 C Sj+ +~2 D
8. Regular and Singular Subspaces
In Section IlLS definite subspaces were divided into two classes:

those of regular and singular subspaces. In a Krein space these classes
can easily be characterized.
'vVe shall make use of the following lemma.
Lemma 8.1. Let 52 be a subspace with positive dej£nite clOS1tre in the
Krein space Sj. Sttppose that 52 admits a decolnposition 52 = 521 +
52 2 ,
where dim521 = 1 and 52 2 1:S uniformly positive. Then 52 1:S uniformly
positive.
Proof. According to Lemma 5.4 the closure 52 2 of ,$32 is uniformly
- -
positive. Hence for 521 C ,$32 i.e. ,$3. C ,$32 the assertion £o11O\vs at once.
Next consider the case ,$31 n,$32 = O. Set
52' =,$31 + 32 , ,$3; = (~2)ln 52' .
In view of Theorem 5.2 we have
~' = 52; (+) £2 .
Corollary 1.1.2 and the required positive definiteness of 52, respectively,
yield that 52~ is i-dimensional and positive definite. Being the orthog-
onal direct sum of two ortho-complemented subspaces, 52' is ortho-
complemented (d. Lemma 1.9.2). Thus, by Theorem 5.2, 52' is uni-
formly positive, and therefore so is the subspace 52 C 52'. D
Theorem 8.2. A definite subspace of the Krein space Sj is regular if

and only if it is uniformly definite.
Proof. Let 52 C Sj be uniformly definite. Since, for every y E Sj,

the linear form rpy{x) = (x, y) (x E,$3) is continuous relative to TM(~)) 152,
it is continuous relative to T;nt(,$3). So 52 is regular.
Suppose, conversely, that the subspace 52 c Sj is non-uniformly def-
inite, say, positive definite. We shall distinguish between two cases.
"*
1. £ contains a neutral non-zero vector XO' In this case, taking an
element Y E 5) with (xo, y) 0 and a sequence {x,,};x' c £ with x" -;> Xo
(n -;> 00), we find:
lim
1t-'?-00
Ix"ls = 0, lim (xw y)
n-)-oo
"* 0.
Hence £ is singular.
2. £ is positive definite. In this case we proceed as follows.
We consider a fundamental decomposition
(8.1)
with respective fundamental projectors P+, P- and fundamental sym-
metryJ = P+ - P-. Instead of P"'x, as usual, we shall write X+.
The subspace £ being non-uniformly definite, there exists an
Xl E £ such that (Xl' Xl) < Ilxlli~. By Theorem II.10.1 we may require
(xt,xt) = 1.
Set
(8.2) ~2 = {x E ~: x+ ~ xt} .
Then £1 +
£2 = £. Thus, on account of Lemma 8.1, £2 is non-uni-
formly definite. Consequently, there exists an X 2 E £2 satisfying the
conditions (x 2 , x2 ) < ~ Ilx211~, (x~~, xi) = 1. In view of (8.2) we also
have (xt, x;I-) = O.
Pursuing the process, we construct a sequence {xn};x' c £ such that
(m,n = 1,2, ... ) .
Consider the elements

nx" = y" (n = 1,2, ... ) ,
The positive definiteness of £ implies Ilxnll~ < 21Ix;;II~ .. Therefore

2
(Yn,Yn)<- (n=1,2, ... ).
n
On the other hand, (Y n , x+) = 1 for every n. So £ is singular. 0
9. Alternating Pairs
An ordered pair {£v £2} of subspaces of the Krein space ,'9 will be
called an alternating pair provided £1 is positive, £2 is negative, and
9. Alternating Pairs 115
)31 ~ 3 2 , If, in addition, )31 is maximal positive and 3 2 is maximal
negative, we say that {3 v 3d is an alternating maximal pair.
By an alternating extension of the alternating pair l31 , 32 } we mean
an alternating pair {3~, 3~} such that 31 c )3~, 32 C 3~.
Theorem 9.1. Every alternating pair in the Krein space S) can be

extended to an alternahng maximal pair.
(9.1) S) = S)+(+).\j-; S)+ C ~++, ~)- C ~-- ,
the corresponding fundamental projectors P+, P-, and the fundamen-

tal symmetry I = P+ - P- .
Let 31 C ~ +, 3 2 C 1{3 -, 3 1 ~ 3 2 , Since the inner product is con-
tinuous on J), we may assume that 3 1 and 3 2 are closed. Then, accord-
ing to Lemma IV./.1, P+3 l and P-3 2 are also closed. The orthogonal
(and I-orthogonal) projectors with range P+ 3 1 resp. P-)32 (d. Theorem
5.2 and Theorem IIj.'IO) will be denoted by PI and P 2 •
In view of Theorem II.11./ the angular operators K+(3 1 ), K-(3 2 )
exist and their I-norms do not exceed 1. For technical reasons we set
(9.2)
Then K2 is defined throughout S) and we have
(9·3) IIKlll] ~ 1, IIK211J ~ 1 .
Moreover, the assumption 3 1 ~ 52 2 yields
or, equivalently,
(9.4) P 2K l x = Kl;J x (x E P 3l )
7 ,
where K[2*J is the Hilbert space adjoint of K2 specified by the relation

(9.5) (]{~*Jx, y)] = (x, K 2 y)] (x, YES)) .
On account of Theorem 4.4 it is sufficient to find a maximal positive
subspace 3(~) containing 31 and orthogonal to 3 2 , By Theorem II.H./
it will do if we construct the angular operator K = K-t (3~O)), i.e., a
linear operator E which satisfies the conditions Kl c]{, '1:l(K) = S)'
(d. Theorem 4.2), ffi(]{) c S)-, IIKII] ~ 'I and, in analogy with (9.4),
(9.6) P2 K x = ]{l~J x (x E S)+) .
Since (9.6) can be regarded as a definition of the first term in
the representation ]{ = P 2 E +
(1 - P 2 ) E, it remains to choose
(1 - P 2 ) K to be an extension of (1 - P 2 ) Kl with domain S)+ and
range contained in (1 - P 2 ) .Sj- such that
(9./) 11(1 - P2)]{xll~ ~ I!xll~ - 11]{~*lxl!~ (x E S)+) .
Consider the inner product

(x, Y)2 = (x, Y) + (K~*Jx , K~*Jy) (x, Y E S)+) .
By (9.2), (9.3) and (9.5) we have
(x, X)2 = Ilxll} - IIK~*l xii} > 0 (x E S)+) .
In terms of the semi-norm P2(X) = (x, X)~/2 (x E S)+) condition (9.7) takes
the form
(9.8) (x E S)+) .
For x E P+53 l the respective inequality
(9.9) 11(1 - P 2)Kl xll] < P2(X)
is guaranteed by (9.4) and (9.3).
Put
S)t = {XE S)+: P2(X)=O}.
On the quotient vector space s)+/~);t the positive inner product (. , ')2
induces (d. Section 1.5, especially Corollary 1.5.3) a positive definite
inner product (. ,.);. With respect to the corresponding norm II· II;
the space ~)+ / S);~ can be completed to a Hilbert space Q;+.
In view of (9.9) the formula
Tlx = (1 - P 2 )Kl x (x E X, x E P+53 l )
uniquely defines a linear operator 1; on a subspace of Q;+ ; this operator
has its values in (1 - P 2 ) S)- and, owing to (9.9), is contractive:
IITlxll]::;; Ilxll; (x E ;])(T1 )) •
We extend Tl to the whole of the Hilbert space Q;+ as a contractive

operator T with values in (1 - P 2 ) ~)-. Finally, we set
(1 - P 2 )Kx = Tx (x E S)+) .
Then (9.8) and all other relevant conditions will be fulfilled. 0
10. Dissipative Operators in Hilbert Space
The purpose of this section is to give an idea of how one applies alter-
na ting pairs.
Let SdO be a I-Iilbert space with inner product [., .Jo and norm
II . 110' Let T l , T2 be dissipative operators in S)o; by this we mean that
Tv T2 are linear operators with dense domains ;])(Tl ), ;])(T2) satisfying
(10.1) [Tjxj' xj]o + [xi' Tjxi]o ~ 0 (xi E '1J(Tj); j = '1,2).
Suppose also that
(10.2) [1;X1' x2Jo = [Xl' T 2x 2]o (Xl E '1J(T l ), X2 E '1J(T2)) •
10. Dissipative Operators in Hilbert Space 117
The question is whether there exist operators T~O) ) T l , T(~)) T2 in &)0

which are maximal dissipative (i.e., dissipative operators having no
dissipative proper extensions) and satisfy the analogue of (10.2):
In order to solve the problem we consider the cartesian product

&) &)0 X&)O, i.e. the Hilbert space of all pairs {x, y} (x,), E &)0) with
=
inner product
(10.4) [{Xl' yd, {X2' )'2}J = [Xl' x2Jo + [Yl' )'2Jo
and norm
(10.5) II{x, y}11 = (1Ixll~ + IIYII~)1/2.
We introduce an indefinite inner product on &) setting
(10.6) ({XV )'l}' {X2,Y2}) = [X1'Y2Jo + [Yl'X 2]0'
The Gram operator G of (.,.) with respect to [., .J exists, namely

G{x, y} = {y, x}. Moreover, G is continuous relative to II· Ii, and
completely invertible. Thus, according to Theorem 1·3, &) equipped
with the inner product (10.6) is a Krein space.
From (10.1) it follows that the graph
®( - T 1 ) = {{Xl' - Tx 1 }: Xl E SD(T1 )}
is a positive subspace of &). Similarly, the graph
®(T2) = {{X2' T 2 x2 }: X2 E SD(T2 )}
is a negative subspace of &). Further, by (10.2), 0.1 (- T1UGSJ(T2)
(in the sense of (10.6)). So {®(-T1 ), ®(T2)} is an alternating pair
in &).
Let {,~~, ~~} be an alternating extension of {®( - T 1 ), ®( 1'2 )}.
Suppose that 2~ contains an element of the form {O,),}. By (10.6)
such an element is neutral. Therefore, on account of Lemma I.4.4,
{O, )'} ~2~. In particular,
({O, Y}, {Xl' -T1x1 }) = 0
i.e. [y, x1Jo = 0 for every Xl E SD(Tl)' Hence)' = 0 .
Consequently, 2~ is the graph of an operator - T~. The positivity
of 2~ implies that T~ is dissipative. In the same manner one verifies
that 2~ is the graph of a dissipative operator T~. Also the properties
T1 C T~, T2 C T~,
(10.7)
follow immediately.
Conversely, if T~, T; are dissipative extensions of Tl and T 2 ,

respectively, such that ('iO.7) holds, then {®(-T~), ®(T~)} is an
alternating extension of {®(-Tl), ®(T2 )}.
We are led to the following result. If {S2(~), S2~)} is an alternating
maximal pair extending {®(-Tl)' ®(T2 )}, then the operators T~O), T~O)
defined by the relations ®( - T~)) = S2~O), ®(TkO)) = S2~) yield a solu-
tion of our problem, and every solution can be obtained in this way.
Thus the existence of a solution is guaranteed by Theorem 9.1.
Notes to Chapter V
The objects called here Krein spaces for the first time occurred implic-
itly in papers of Nevanlinna [4J and Pesonen [1]. Formally they were
introduced by Ginzburg ['I] in 1957 and investigated thereupon by
several authors including Iohvidov, Langer,Phillips, and Scheibe.
Nonetheless, the new term seems to be justified by the fact that the
theory of these spaces grew out, for the most part, of the school of
M. G. KreIn, and has largely been influenced by KreIn's personality
and contributions.
The present definition of Krein spaces was formulated by Scheibe
[1]. Most authors start from a Hilbert space and introduce the indefi-
nite inner product as a derived concept. In the latter treatment the
indefinite rather than the definite inner product is called J-inner
product.
The last assertion of Theorem '1. -, was proved by Iohvidov and
Krein [1J in a special case. In the general case it is due to Scheibe [1].
It follows also from independent investigations of Ginzburg [3].
Theorem 1.3 has first been obtained by Nevanlinna [4J and in-
dependently reproved by Langer [2J, Phillips [3], and Scheibe [1].
Langer also noticed that a non-degenerate, decomposable space is a
Krein space if and only if it is weakly sequentially complete (d. the
assumption of Theorem III.9.3).
That Krein spaces are completions of decomposable non-degenerate
spaces was observed in a special case by Iohvidov and KreIn [1J, and
in the general case by Ginzburg and Iohvidov ~1]. The uniqueness of
the completion seems to be considered in Theorem 2.1 for the first
time.
Theorem 3.4 was obtained by Scheibe [1J and Ginzburg [3J, inde-
pendently of each other. For a special case of Theorem 3.5 see Iohvidov
and KreIn [1].
Theorem 4.2, Theorem 4.4 and Lemma 4.5 belong to Phillips [1], [2].
The first two of these results have also been proved by Ginzburg [3].
Notes to Chapter V 119
Instead of using angular operators, Langer [1], [2] described maxi-

mal positive subspaces .2 C 5) by means of skew projectors that project
5) onto .2 along 5)- (d. also Ginzburg and Iohvidov [1]).
Iohvidov [10] gave a characterization of non-closed maximal defi-
nite subspaces of 5).
The concept of plus-defect was introduced by KreIn and Smul'jan
[1].
Let ~ be a Banach space, and let P+, P- be continuous projectors
in m- such that P+ + P- = I. The space m- equipped with the func-
tion J(x) = IIP+xI1 2 - IIP-xIl 2 (x E~) is called a Banach space with a
J-metric. J-semi-definite subspaces of ~ have been studied by Ioh-
vidov [14].
Uniformly definite subspaces of a KreIn space were introduced and
analysed by Ginzburg [3], [4].
Lemma 6.1 is essentially contained in a paper of Iohvidov and
KreIn [2]. Theorem 6.3 belongs to Ginzburg [3]. Example 6.5 has its
origin in a construction proposed by Iohvidov [6]; for related results d.
also Iohvidov [8].
A special case of Lemma 7.5 is due to Iohvidov and KreIn [1]; in
the general case it was formulated by Smul'jan [3]. Lemma 7.6 appears
in a paper of Krein and Smul'jan [3].
Theorem 8.2 was proved by Ginzburg (see Ginzburg and Iohvidov
[1]). The present proof is based on his idea, but avoids the spectral
theorem. Another method is proposed in the lecture notes of KreIn [8].
Alternating pairs were extensively studied by Phillips [1J, [2], [3].
The present proof of Theorem 9.1 belongs to Langer (unpublished
lecture notes; d. also Langer [5]). In the original proof Phillips [3]
unites the two angular operators K+(.2 1 ), K-(.2 2) into a single operator.
Both proofs make use of a classical argument of KreIn.
The material of Section 10 is borrowed from Phillips [3]. The
existence of the maximal extensions in question has independently
been obtained by Sz.-Nagy and Foia~ [1; Proposition IV,4.2] without
using an indefinite inner product.
Phillips [1], [2] applied his theory of alternating pairs to differen-
tial equations; for a short summary d. also N almark and Ismagilov
[1]. Further developments in this direction are to be found in the
papers of Cordes [1], Olubummo and Phillips [1], Phillips and Sarason
[1], Crandall and Phillips [1], Daleckil and Fadeeva [1J; the latter work
also offers an insight into the main idea of these applications.
The survey of Ginzburg and Iohvidov [1] as well as the lecture notes
of KreIn [8] contain additional facts from the geometry of Krein
spaces.
Chapter VI. Unitary and Selfadjoint Operators in
Krein Spaces
The study of linear operators in Krein spaces begins with this chapter. Main topics
include criteria for the continuity of isometric operators (Section 3) as well as basic
properties and the location of spectra of unitary and selfadjoint operators (Sec-
tions 4-7). In Section 8 it is proved that every continuous linear operator of a Hil-
bert space has a unitary dilation in a Krein space. Theorems 3.5,3.10,5.1,5.5.6.1,
7.5,8.6 and Lemma 4.7 can be mentioned as representative results.
1. Preliminaries
In Chapter II linear operators of general inner product spaces have

been treated. The rest of our book is devoted to linear operators of
Krein spaces.
Since every ,Krein space, when considered with a J-inner product,
is a Hilbert space, we shall need some definitions and facts from the
theory of Hilbert space operators. A few of them are listed below.
If Q;1> Q;2 are Hilbert spaces, then the class of all continuous linear
operators T with '!l(T) = Q;1> ffi(T) C Q;2 will be denoted by 5.8(Q;1> Q;2)'
In the special case Q;1 = Q;2 = Q; we use the symbol 5.8(Q;) or 5.8 for short.
5.8(Q;1> Q;2) is a Banach space with respect to the usual operator norm
IITII = sup II Txll·
11",[1=1
Besides the Banach topology, we shall have to do with another
separated locally convex topology of 5.8(Q;1> Q;2)' the weak operator
topology i wo (Q;l> Q;2) defined by the semi-norms p""y (x E Q;1> Y E Q;2)'
where
p""y(T) = I(Tx, y)1
Recall that a subset ts: of a topological space is said to be compact,
if from any system of open sets whose union contains ts: one can select
a finite sUbsystem whose union also contains ts:.
Let ~1>~2 be Hilbert spaces. We say that the operator TElB(Q;VQ;2)
is compact, if for every bounded set Il( C Q;1 the closure Til( of
its image is a compact set in Q;2' Equivalently, the operator T is com-
pact if for every bounded sequence {x,,}~ C 0: 1 the sequence {Txn}~ c 0:2
contains a convergent subsequence.
Linear combinations of compact operators are compact. If 0: j
(j = 1, 2,3,4) are Hilbert spaces, if 1'j E ~(0:i' 0: i + 1 ) (j = 1,2,3), and
jf 1'2 is compact, then 1'11'2 and 1'21'3 are compact.
The linear operator l' from the I-lilbert space 0:1 to the Hilbert
space G!2 is said to be closed, if the relations {x,J;x'c:D(T), xn-)ox,
Tx" -)0 y imply x E :D(T), Tx = y. This property is equivalent to the
closedness of the graph {{x, Tx}: x E '1)( T)} of l' in the Hilbert space
G!1 X G!2·
A linear operator l' mayor may not have closed extensions. In
the first case there exists a least closed extension, denoted by T and
called the closure of 1'.
Let l' be a closed linear operator in the Hilbert space G!. With
respect to T, the following classification of complex numbers A is
customary. If T - AI is invertible and its range coincides with G!,
then ). is said to belong to the resolvent set e(1') of T. If T - AI is
invertible and its range is a dense proper subspace of 0:, then }, is said
to lie in the continuous spectrum actT) of 1'. If l' - AI is invertible
and its range is not dense in G!, then A is assigned to the residual spec-
trum ar(T) of 1'. Finally, in case l' - AI is not invertible, A is said to
belong to the point spectrum ap(T) of 1'.
The set
a(T) = ac(T) u ar(T) u ap(T) ,
complementary to (2(T) in C, is called the spectrU1n of T.
The points belonging to e(T) are called reg%lar points of T. By the
closed graph principle, }. E (2(T) if and only if (1' - A1)-1 E ~.
If A E ac(T), then (T - }.It J is discontinuous on its domain.
Obviously, ap(T) consists of the eigenvalues of 1'.
An important subset of ap ( 1') is the set of "normal" eigenvalues.
The number A is said to be a normal eigenvalue of the closed linear opera-
tor T in the H.ilbert space 0:, if 1) 0< <
dim @)}.(T) 00; 2) @);JT) has a
complementary subspace ,12;. in 0: which is closed, invariant for T, and
such that). is a regular point of 1'\,12;. .
2. The Adjoint of an Operator

Let l' be a linear operator with dense domain in the Krein space Sj.
We define the adjoint T* of l' as follows. The vector y E Sj belongs to
'1)( 1'*) if and only if there exists a y* E Sj such that
(2.1) (1'x, y) = (x, y*) (x E '1)(T)) .
In this case y* is unique; we set T*y = y*.
122 VI. Unitary and Selfadjoint Operators in Krein Spaces
Evidently, 0 always belongs to ~(T*), and T* is a linear operator.

Moreover, T* satisfies the relation
(2.2) (Tx, y) = (x, T*y) (x E ~(T), Y E ~(T*))
and has largest possible domain.

The adjoint operator in 5) can be expressed through the adjoint in
a Hilbert space.
Lemma 2.1. Let T be a densely defined linear operator in the Krein
space .S). For some fundamental symmetry j E 5.8(~)), denote the j-adioint
of T by T[*l; thus T[*l is the operator satisfying
(2.3) (Tx, zlJ = (x, T[*lzlJ (x E ~(T), z E ~(T[*l))
and having largest possible domain. Then
(2.4)
Proof. Relation (2-3) with the help of Lemma 1I.11.1 yields:
(Tx, y) = (x, jT[*l Jy) (x E ~(T), Y E ~(]T[*l])).
Therefore jT[*lj c T*. Similarly, starting from equation (2.2) we
derive jT* j c T[*l i.e. T* C jT[*lj. 0
Numerous properties of adjoint operators in a Krein space formally
duplicate those known from Hilbert space theory. Below we list some
of them in a more or less logical order, but somewhat arbitrary group-
ing. The proofs can be obtained either by imitating the Hilbert space
argument, or by applying the Hilbert space result itself combined with
Lemma 2.1.
Theorem 2.2. a) 0* = 0; b) 1* = I; c) T* is closed (provided it
exists); d) TE 5.8(5)) implies T* E 5.8(.S)). 0
Theorem 2.3. If the adjoints in question exist, we have: a) (o.:T)* =
= E C); b) (T1 + T 2 )* ) Tf + T:; c) (T1T 2 )* ) T~T~ .
(iT* (0.: 0
Theorem 2.4. If T* and (T-1)* exist, then (T*)-l exists and is equal
to (T-1)*. 0
Theorem 2.5. The linear operator T with dense dmnain in the Krein
space 5) admits closed extensions if and only it T* * exists; in this case
T** is the closure of T. 0
Theorem 2.6. If the adjoints in question exist, we have: a) T1 C T2
implies T~) T~; b )invariance relative to T of the subspace ~ c 5)
implies invariance of ~l relative to T* provided ~n~(T) is dense in~;
c) 'iJ((T*) = ffi(T)l; d) ffi(T*) = SJ/:(T)l provided T is closed. 0
Theorem 2.7. Let T be a densely defined, closed operator in the Krein

space~. Then
-
a) A E e( T) implies ~ E e(T*);
b) A E rYc(T) implies ~ E rYc(T*);
c) A E rYr(T) i1'nplies ~ E rYp(T*);
d) A E rYp(T) implies A E rYp(T*) U rYr(T*). o
Theorem 2.8. If T is a linear operator in ~ such that 'Il(T) =
= 'tI(T*) =~, then T is continuous. 0
The following result is also known in the Hilbert space case. Since,
however, it is less standard, we present the proof, adapted to the Krein
space situation.
Theorem 2.9. Let T be a closed linear operator with dense domain in
the Krein space~. If ffi( T) is closed, so is ffi( T*).
Proof. Let zJJJC(T). Then the relation
cp(Tx) = (x, z) (x E 'tI(T))
uniquely defines a linear form cp on ffi(T).
Since T is closed, so is SJC(T). Let 53 be a closed complementary
subspace to SJC(T) (say, the J-orthogonal companion of SJC(T), where J
is a fundamental symmetry on SJ). For u E ffi(T) let T(-l)U denote the
unique element x E 53 satisfying Tx = u. By the closed graph prin-
ciple, T(-lj is continuous. On the other hand, cp(u) = (T(-1)u, z)
(u E ffi(T)). Thus cp is continuous.
Extending cp to all of ~ as a continuous linear form and applying
Theorem V.1.5 we find an element y E ~ such that cp(Tx) = (Tx, y)
(x E 'tI(T)), i.e.,
(Tx, y) = (x, z) (x E 'tI(T)) .
Hence z = T*y, z E ffi(T*), SJC(T)l c ffi(T*). Now the assertion
follows from Theorem 2.6 d). D
The next lemma exhibits a specific feature of indefinite Krein

spaces as compared to Hilbert spaces.
Lemma 2.10. Let T be a linear operator with dense domain in the
Krein space~. The equation SJC(T*T) = SJC(T) holds if and only if
ffi(T) is non-degenerate.
Proof. For x E 'tI(T) the relations Tx E 'tI(T*), T*Tx = 0 are
equivalent to (Tx, Ty) = 0 (y E 'tI(T)), i.e., to Tx being an isotropic
vector of ffi(T). D
3. Isometric Operators
Let U be an isometric operator in the Krein space S). Though U pre-
serves intrinsic norms, it need not preserve natural norms; what is
more, U can be discontinuous (with respect to the strong topology).
This will turn out from the criteria of continuity given below mainly
in terms of ~(U) and lH(U).
Remark 3.1. When looking for sufficient conditions of continuity
in terms of ~(U) and lH(U), we may restrict ourselves to the case
where ~(U) and ffi(U) are non-degenerate. To see this, we distinguish
between two possibilities.
a) ffi(U) is degenerate. We neglect the trivial case dim~(U)<
< 00, but assume that U is continuous. Choosing a non-zero isotropic
vector of lH(U), say Ue, and a discontinuous linear form rp on ~(U)
with rp(e)*- -1, the formula U1x = Ux +
rp(x) Ue (x E ~(U)) de-
fines a discontinuous isometric operator UI having the same domain
and range as U. Consequently, in this case ~(U) and lH(U) do not
decide on the continuity of U.
b) ~(U) is degenerate, but ffi(U) is non-degenerate. Then, on
account of Lemma II.2.1, U~o = 0, where ~o denotes the isotropic
part of ~(U). Thus the problem of continuity can be split into two
pieces (see Theorem 1.5.4): one concerning the restriction of U to a
maximal non-degenerate subspace ~l of ~(U) (by Lemma I1.2.1 the
range of this restriction is also non-degenerate), and another concerning
the dependence of XO and Xl on the vector XO +
Xl (XO E ~o, Xl E ~l).
The main body of the present section is based on the following

simple fact.
Lemma 3.2. Let U be an isometric operator in the Krein space SJ. If
U is conhn'u,ous and ~(U) is uniform.!y positive (tmiformly negative),
then ffi(U) is uniformly positive (zmifonnly negative). On the other hand,
if lH(U) is u1u:jormly definite, then U is continuous.
Proof. If ~(U) is uniformly positive and non-zero, while U is
continuous, then for a fundamental symmetry] and suitable positive
number ex we have
(Ux, Ux) = (x, x) ~ex Ilxlli ~ IldnjllUxllJ (x E ~(U));
hence lH(U) is uniformly positive. A similar reasoning applies when

~(U) is uniformly negative. If, in turn, lH(U) is uniformly definite,
then (d. Lemma 11.'11.4)
liUxllJ ;;::; f3 I(Ux, Ux)1 = f3 I(x, x)1 :S f3 I!xllj (x E ~(U))
for some f3 > 0; so U is continuous. 0
Making use of Corollary 11.2.2 we obtain:

Coronary 3.3. If U is an isometric operator with umjormly definite
domain in ~), then U-1is continuous. 0
If x(S)) = CXJ, then 'with the help of Lemma 3.2 we can construct a
discontinuous isometric operator that is defined throughout S).
Example 3.4. Let x(,SJ) = CXJ. Consider a fundamental decom-
position
Decompose S)+ and S)- as follows:

,SJ+ = ,SJ; (+) S):, dimSJi count ably infinite,
~)- = s);(+),SJ;-, dimS); = 1.
Example V.6.S shows that the Krein space,SJI = ,SJi (+) ~); contains
a positive definite, intrinsically complete, non-uniformly definite sub-
space 2. The intrinsic dimension of .2 must be infinite and, by Lemma
IV.7.3, not greater than that of,SJt. Consequently, there exists an
isometric operator ui with '1)(U~)= S)i, ffi(Ui)=.\3. Let U- be an
isometric operator that maps ,SJ- onto S):;-. For
x = xi + xt + x-; x~ E ,SJ~, xt E ,SJt, x- E ,SJ-
set
Ux = uixt + xt + U-x- .
Then U is an isometric operator with '1)(U) = ,SJ. Applying Lemma 3·2
to ui we see that U is discontinuous.
The next result extends the necessary condition and a weakening of
the sufficient condition contained in Lemma 3.2 to the case of indefinite
domains.
Theorem 3.5. Let U be an isometric operator in the Krein space ,SJ.
Assume that '1)( U)is the orthogonal direct stun of a um:formly positive
s·ubspace '1)+ and a uniformly negative subspace ';tJ -. Then U will be
contimwtts if and only it U'1)+ and U';tJ- are tlnifonnly definite.
Proof. Lemma V.SA, Theorem V.S.3 and Theorem V.3.5 yield a
fundamental decomposition
(3·1) S)=SY(+),SJ-; ,SJ+clf5++, ,SJ-clf5--
such that '1)+ C S)+, '1)- c ,p-. Denote the corresponding fundamental
projectors and fundamental symmetry by P+, P- and ], respectively.
If U'lY, U'1)- are uniformly definite then, according to Lemma 3.2
and Lemma II. '11.2,
IIUxll J ~ IIUP'xII J + I!UP-xll] ~ fJ11IP+xll] + fJ211 P-x11 J ::;;:
~ (fJI + fJ2) IIxl IJ
for suitable fJI> fJ2 >

0 and every x E ~(U); hence U is continuous. The
reverse implication directly follows from Lemma 3·2. 0
Corollary 3.6. If U is a continuous isometric operator in S), and
~(U) is the orthogonal direct sum of a $tniformly positive and a uniformly
negative subspace, then U-l is continuotts. 0
In Theorem 3.5 the uniform definiteness of the components of two
+
fundamental decompositions ~(U) = ~+( )~-, ffi(U) = ffi+( )ffi- +
plays a part. It is essential that these decompositions are required to
be "coherent": ffi± = U~±.
Example 3.7. Let x(S)) = CD. Consider a fundamental decom-
position (3.1) of S). Let {ei}~ C S)+, {M~ c S)- be orthonormal sys-
tems. Denote by ~+(~-) the span of all vectors ei (Ii). Furthermore,
let ~1(~2) be the span of the vectors ei + --;-.J~-fi (resp . fi + ~-ei)'
J+1 J+1
j = 1,2,. .. . Then ~l' ~2 are orthogonal, non-uniformly definite
subspaces satisfying
(3. 2)
As each of ~I> ~2' ~+, ~- is the span of a countably infinite set of
vectors, there exists an isometric operator U defined on ~+ (+) ~
such that
(3·3 ) U~- = ~2.
For this operator U both ~(U) and ffi(U) can be decomposed into the
orthogonal direct sum of the uniformly definite subs paces ~+, ~-.
Nevertheless, from (3.3) and Theorem 3.5 it follows that U is discon-
tinuous. On the other hand, if V is the identity mapping on ~+ (+) ~
then V is continuous and isometric; one of the fundamental decom-
positions of ffi(V) still consists of the non-uniformly definite subspaces
~l' ~2·
This dependence on the choice of the fundamental decompositions
ceases when ~(U) is closed.
Theorem 3.8. An isometric operator U with ortho-complernented
domain ~(U) in the Krein space .\iis continuous if and only if ffi( U)
is ortho-complemented.
Proof. By Theorem V.5-3 ~(U) can be written in the form
(3.4) ~(U) = ~"( +)~-; ~+ c ~++, ~- c ~-- ,
where ~+ and ~- are closed and uniformly definite. Obviously,
(3· 5)
If U is continuous, then Theorem 3.5 implies that UCJy and U'1J-

are uniformly definite. They are closed as well, since a uniformly
definite subspace is closed if and only if it is intrinsically complete
(d. Remark V.6.4), and the latter property is preserved by an iso-
metric operator. Applying Theorem V.5.3 once more, we obtain that
ffi(U) is ortho-complemented.
Let, conversely, ffi(U) be ortho-complemented. Then, on account of
(3.5) and Lemma 1.9.2, U'1J+ and U'1J- are also ortho-complemented.
Therefore the continuity of U follows from Theorem V.5.2 and The-
orem 3·5. 0
If requiring only that the closure of '1J(U) be ortho-complemented,
which is a weaker restriction on '1J(U) than the hypothesis of Theorem
3.5, some conclusions can still be drawn.
Lemma 3.9. If U is a continuous isometric operator in Sj such that
the closure oj '1J(U) is ortho-complemented, then the clos~tre ot ffi(U) is also
ortho-complemented and U-I is continuous.
Proof. U can be extended to a continuous isometric operator UI de-
--- ---
fined on '1J(U). Clearly, ffi(U)c ffi(UI ) c ffi(U). By Theorem 3·8 ffi(UI )
is ortho-complemented. Hence, on account of Theorem V.3.4, ffi(UI )
is closed. In particular, ffi(UI ) = ffi(U). Furthermore, according to
Corollary 3.6 and Theorem V.5.3, U-;-' is continuous. 0
It should be noted that the ortho-complementedness of '1J(U) and
ffi(U) do not imply the continuity of the isometric operator U (d.
Example 3.7).
The following rather strong result differs in nature from the preced-
ing ones. It contains one half of Theorem 3.8, but not of Theorem 3.5.
Theorem 3.10. Let U be an isometric operator in the Krein space .~.
If '1J(U) is closed as well as non-degenerate and the closure ot ffi(U) is non-
degenerate, then U is continuous.
Proof. Owing to the closed graph principle we only need to prove
that U is a closed operator.
Let {x,,}~ c '1J(U), x" -> x, Ux" -'.>- y (n -;"(0). Then
(Ux",z)-'.>-(y,z) (ZESj).
On the other hand, making use of the invertibility of U (Corollary
II.2.2) and the relation x E '1J(U), we find:
(Uxn' z) = (x"' U-IZ) -'.>- (x, U-1z) = (Ux, z) (z E ffi(U)).
Consequently, (Ux - y, z) = 0 for every z in ffi(U). But Ux - Y
o
--- ---
belongs to ffi(U), and ffi(U) is non-degenerate. Hence Ux = y.
Theorem 3.10 and Lemma II.2.1 yield:

Corollary 3.11. Let U be an isometric operator in 5). If '1:J(U) , ffi(U)
are closed and '1:J(U) is non-degenerate, then U 1:S continuous. D
4. Unitary and Rectangular Isometric Operators

An isometric operator U in the Krein space 5) is said to be unitary
provided '1:J(U) = ffi(U) = S).
Theorem 3.8 yields:
Theorem 4.1. Every unitary operator in 5) is continuous. D
Making use of Corollary 11.2.2 the following characterization of
unitary operators can easily be established.
Theorem 4.2. The linear operator U in the Krein space 5) is 1mitary
if and only if it is completely invertible and U-1 = U*. D
Let U be an isometric operator in the Krein space 5). If '1:J(U) and

ffi(U) are ortho-complemented, we say U is rectangular. In case U is
rectangular and has no rectangular isometric proper extensions we say
that U is maximal rectangular.
Unitary operators are a special kind of maximal rectangular iso-
metric operators. Every rectangular isometric operator is continuous
(Theorem 3.8) and invertible (Corollary II.2.2).
Theorem 4.3. Let U be a rectangular isometric operator in the Krein

space 5). Set '1:J = ';t(U), ffi = ffi(U). The operator U is maximal
rectangular isometric if and only if
(4.1 )
and
(4.2)
Proof. Lemma 1.9.1 and Theorem V-3,4 assure that the values %
appearing in the statement make sense.
Let U1 be a rectangular isometric proper extension of U. Since '1:J
is ortho-complemented, we have '1:J(U1 ) = '1:J( +)'1:J1, where 5\)1 =
= '1:Jl (l '1:J(U1 ). Since '1)(U1 ) is ortho-complemented, so is '1)1 (d.
Lemma 1.9.2). Moreover, '1:J1 =t O. The relation 5\)1 '1) implies
U1'1:J I ~ U1'1); hence U1'1:J1 (ffil. As a result, the isometric operator
U1 maps the ortho-complemented non-zero subspace '1:J1 ( '1)1 into ffil.
Owing to Corollaries 1.9.5 and 1.4.5 '1:J1 contains either a positive or a
negative vector; ffi 1 must contain a vector of the same kind.
4. Unitary and Rectangular Isometric Operators 129
If, conversely, g)l contains a positive or negative vector x and ffil

contains a vector y of the same sign, then it is easy to construct an
isometric extension U 1 of U such that g)(U1 ) = g)(+)(x), ffi(U1 ) =
= ffi( + )(y). In view of Lemma I.9.2 this U1 will be rectangular. 0
In contrast to the theorem above, a rectangular isometric U satis-

fying (4.1) and (4.2) may have isometric proper extensions U1 which
are non-rectangular. This situation occurs when g)l is indefinite,
~U = 0, and U 1 is defined to be zero on a neutral non-zero subspace
of g)l.
Theorem 4.4. Let U be a rectangular ismnetric operator in 5). Then U

admits maximal rectangular isometric extensions. U admits a unitary
extension if and only if
(4·3)
where g) = '£(U), ffi = ffi(U).
Proof. According to Lemma 1.9.1, Theorem V.3.4 and Theorem
V.1.1 the subspaces g)l, ffi 1 admit fundamental decompositions of the
form
~l = ~t(+)g)~; ~t C \,p++, g)i- C \,p--,
ffil = mt(+ )ffi~; ffit c \,p++, m~ c \,p-- ,
where g)t, g)~, mi, ffi~ are intrinsically complete. By the theory of
Hilbert spaces there exists an isometric operator ut from ~t into mt
such that
(4.4) either ~(Ut) = g)t or m(ut) = mt .
Similarly, there exists an isometric operator U~ from g)~ into m~ such
that
(4.5) either '1:( Uj) = '1:;- or m( Uj) = ffij .
We define a linear operator U o by the relations
'1)(Uo) = g)(U) (+) '1:(Ut) (+) g)(Ull,
Uo(x -+ xi -+ xn = Ux -+ utxi -+ Ujx1
(x E '£(U), E '1)(UtJ, Xl E ~(Ul))'
Obviously, U o is an isometric extension of U. Moreover, (4.4) implies
that '1:( Ui) is intrinsically complete, hence ortho-complemented in
'1:1. Similarly, (4.5) implies that g)(Uj) is ortho-complemented in g)j.
Thus ~(Uo) and in much the same way ffi(U o) are ortho-complemented
in 5). Therefore Theorem 4.3 may be applied: owing to (4.4) and (4.5)
U o is maximal rectangular isometric.
If (4.}) holds, then both of the equations under (4.4) and both of
the equations under (4.5) can be satisfied. In that case Uo will be

unitary. If, conversely, U admits a unitary extension Uo , then Uo~1 =
= ffi1; therefore (4.3) is a consequence of Theorem V.1.4. 0
In what follows we establish a few elementary results concerning
invariant subspaces of unitary operators.
Lemma 4.5. Let U be a unitary operator in the Krein space Sj, and
let 5} be a subspace of Sj. If U5} c 5}, then U-I5}1 c 5}1.
Proof. Since U is completely invertible (Theorem 4.2), U5} c 5}
yields ,53 c U-lS!. Now Lemma 11.2.9 can be applied. 0
Taking into account that the relation U5} = S! is equivalent to the

invariance of S! under both of the unitary operators U, U-l (and that
for S!1 a similar statement holds), we obtain:
Corollary 4.6. Let U be a unitary operator in Sj, and let S! be a subspace
of ~). If US! = S!, then U5}1 = S!1. 0
\iV e are going to show that for a maximal positive or maximal
negative subspace S! the condition US! c S! already implies US! = S!.
Lemma 4.7. Let U be a unitary operator in the Krein space Sj. If S!
is a maximal positive (resp. maximal negative) su,bspace of Sj, so is US!.
Proof. By Theorem 4.2 the relation (Ux, y) = 0 is equivalent to
(x, U-ly) = O. Thus for any subspace S! we have (US!)1 = US!l. If
S! is maximal positive then, in particular, S! is a closed positive sub-
space (Theorem VA.1), while 5}1 is negative (Lemma 1.6.3). Since U is
isometric and U-l E 58(Sj) (Theorems 4.'1-4.2), it follows that US! is a
closed positive subspace with negative orthogonal companion. There-
fore, according to Lemma VA. 5, US! is maximal positive. 0
Corollary 4.8. Let U be a wnitary operator in Sj. If the rnaximal

posi#ve or 1naximal negative subspace S! c Sj is invariant for U, then
U5} = S!. 0
When the invariant positive or negative subspace is not maximal,
we can state something less.
Lemma 4.9. Let U be a unitary operator inSj. If the positive (negative)
subspace 5} C ~) is invariant for U, then there exists a positive (negative)
subspace S!1 ) S! Sttc/z that US!1 = S!1 .
Proof· Denote by S!1 the span of the subspaces U-iS! (i = 0,1,2,
... ). Then the relations S!1 ) S!, U5}l = S!1 are obvious. Furthermore,
jf x = ,x:" U-ixi , where xi
1=0
E S! for every j, then unx E S!. Since (x, x) =
= (un:'c, unx), this proves the assertion concerning the sign of S!1' 0
5. Spectral Properties of Unitary Operators 131
5. Spectral Properties of Unitary Operators
In this section we shall repeatedly use the notation

1
v* = -=- (v E C, v*" 0)
v
introduced in Section n.2.
Theorem 5.1. If U is a unitary operator in the Krein space Sj and

v is a non-zero number, then the following implications are valid:
a) v E e(U) implies v* e(U);
E
b) v E CTc(U) implies v* CTc(U);
E
c) v E CTr(U) implies V* E CTp(U);
d) v E CTp(U) implies v* E CTp(U) UCTr(U) .
Proof. In view of Theorem 4.1 we may apply Theorem 2.7. It
remains to observe that, owing to Theorem 4.2,
(5.1) u* - vI = -vU*(U - v*I) (v *" 0) ,
where the operator U * = U-l is unitary, hence a homeomorphism. 0
Corollary 5.2. The residual specf1·um of a unitary operator in Sj does
not intersect the unit circle Ivl = 1. 0
For a unitary operator the point l' = 0 is always regular (d. The-
orem 4.2). Consequently, Theorem 5.1 a) can be reformulated as fol-
lows.
Corollary 5.3. The spectrum of a unitary operator in Sj lies symmetri-
cally with respect to the unit circle. 0
Remark 5.4. A unitary operator in ,S) may really have non-unimodu-

lar numbers in its spectrum. For the continuous and residual spectra
this will be seen from Example 6.4 and Theorems 7.1, 7.3. For the
point spectrum d. Example 11.2.4.
If a unitary operator has a non-empty residual spectrum then, by

Theorem 5.1 c), its point spectrum cannot be symmetric relative to the
unit circle. The set of normal eigenvalues, however, is always sym-
metric, as shown by the following generalization of Theorems n.2.7-
-II.2.8.
Theorem 5.5. Let U be a unitary operator in the Krein space .S). If
v is a normal eigenvalue of U, so is v*, and the principal subspaces
Eiv(U), Eiv'(U) form a dual pair; 11wreover, the 1natrices oj UIEi.(U) and
UIEi v'( U) with respect to any dual pair of bases are the inverses of the
transposed conjugates of each other.
Proal. According to the assumption we have

(S.2)
where <5 = <5.(U), 0 < dim <5 = m < 00, 53 is closed, U53 C 53, and
l' E e(UI53).
Set
(5.3) <5* = 53 1, 53* = <5 1 .
Then <5*n53* = (53 + <5)1 = ~1 = O. Moreover, by Theorem 1.10.9
codim 53 * = dim <5 = m .
On the other hand, relation (5.3), Theorems 1.10.9, III.6.1 and V.1.5
along with relation (S .2) yield
dim <5* = codim 53 11 = codim 53 = dim <5 =m.
Therefore, owing to Lemma 1.1.1,
(S .4) ~ = <5* + 53*.
On account of (S.3) and Lemma II1.2.4 the subspace 53* is closed.
That <5 and 5* form a dual pair is also easy to verify: <5n(5*)1 =
= en53 11 = 5n53 = 0, 5*n<51 = 5*n53* = O.
From the definitions of <5 and m, recalling basic facts of linear
algebra, we obtain:
(S.S) (U - v1)n 5 = 0 (n 2: m) .
At the same time, v E e(UI53) implies
(5.6) (U - v1t 53 = 53 (n = 1,2, ... ) .
Comraring (5.5) and (5.6) we find that
(5.7) 53 = ffi((U - v1t) (n > m) .
Hence, on account of relation (5.3) and Theorems 2.6 c), 2.3,4.1,
5* = 97((U* - v1)n) (n ~ m) ,
or, by (5.1) and TheorEm 4.2,
(5.8) <5* = SJl((U - v* l)n) (n > m).
Thus €* = E •• (U).
Since <5 is a finite-dimensional principal subspace of U and the
operator U is invertible, it follows that U<5 = 5. Therefore, in view
of (5-3) and Corollary 4.6,
(5.9) U53* = 53* .
Consider the restriction (U - v* 1)153*. According to (5.8) and
(5.4) it is invertible. Moreover, its range 53* is contained in 53* (see
(5.9)). Let us prove that 53* fills 53*.
6. Selfadjoint Operators 133
Obviously, (U - v*l)"'B* c B*. By (5.8) and (5.4) this is equi-
valent to ffi((U - v*l),ll) C B* or, by (5.1), (5.9) and Theorem 4.2, to
m((u* - vI)"') c S3*. But, as a consequence of (5.7) and Theorem
2.9, ffi((U':' - vI)"') is closed. Therefore, owing to Theorem 2.6 d),
97.:((U - vI)"') 1 C S3,:,. Recalling the definition of <5 and making use
of relations (5.5), (5.3) we obtain B* c S3* .
Consequently, v* E g(U[S3*).
The assertion concerning the matrices of GI(5 and UI(5* follows
from Theorem II.2.8 applied to the space a; = (5 + 15*, which is non-
degenerate by Corollary II.2.6 and Lemma 1.10.1 in case Ivl oF 1, and
by a;=<5, @;#(5 in the opposite case. 0
Corollary 5.6. Ij v is a unimodular normal eigenvalue oj the unitary

operator U in ,Q), then the principal subspace I5 v ( U) is non-degenerate. 0
6. Selfadjoint Operators
The linear operator A in the Krein space ,'{:> is said to be selfadjoint if
:t'l(A) is dense in ,Q) and A * = A.
In particular, every selfadjoint operator is symmetric and closed
(ef. Theorem 2.2 c). On the other hand, if A is symmetric and :t'l(A)=,Q),
then A is selfadjoint.
From Theorem 2.Jwe obtain the following analogue of Theorem 5.1.
Theorem 6.1. Ij A is a seljadjoint operator in the Krein space ,Q) and
A E C,then
a) }, E etA) implies },Ee(A);
b) A E O'c(A) inzplies }, E O'c(A) ;
-
c) A E O',(A) implies A E O'p(A) ;

d) AEO'p(A) implies }, E O'p(A)uO'r(A) . o
Coronary 6.2. The resid~tal spectrum oj a seljadjoint operator in ,'{:>
does not intersect the real axis. 0
Corollary 6.3. The spectrum oj a selfadjoint operator in ,Q) is sym-

metric with respect to the real axis. 0
That the point spectrum of a selfadjoint operator in a Krein space

may contain non-real points is shown by Example II.3.2. The corre-
sponding fact for the residual and continuous spectra can be established
as follows.
Example 6.4. Let,Q)o be a Hilbert space of countably infinite dimen-
sion. Denote the inner product by [., .Jo. Consider the cartesian
product &)=&)n X&)o, i.e., the Hilbert space of all ordered pairs {x,y}
(x,y E Sjo) with inner product
[{Xl' yd, {X2' Y2}] = [Xl' x2Jo + [Y1' Y2JO·
The new inner product
({xv Y1}, {X2' Y2}) = [xv :>'2JO + [Y1' x2Jo
turns &) into an indefinite Krein space (see Section V.i0). For any
To E )S(&)o) with adjoint T~*l the formula A{x, y} = {Tox, T~*ly}
defines a selfadjoint operator on the Krein space &).
Let A E C, A =F ).. If we fix To by the equations Toe j = Aej ei+l +
(j = 1,2, ... ), where {e j };", is a complete orthonormal system in the
Hilbert space &)0' then A E ar(ToL }. E e(To). Owing to Theorem 2.7 a)
applied to the Hilbert space &)0' the latter inclusion is equivalent to
}. E e( T~*l). Thus A E ar(A). On the other hand, if Toe j = J.e j + ~ ej
J
(f = 1,2, ... ), then}, E ac (1'o) and A E e(T~*l), so that A E ac(A).
The next result is an analogue of Theorem 5.5 and, at the same time,
an extension of Theorems 11.3.5-II.3.6. A further extension, relating
to certain discontinuous selfadjoint operators, will be given in Sec-
tion 7.
Theorem 6.5. Let A be a continuous selfadjoint operator in the Krein
space &). If A is a normal eigenvalue of A, so is A, and the principal sub-
spaces El,.(A), El;:(A) form a dual pair; moreover, the matrices of AIEl).(A)
and A IEl;:(A) with respect to any dual pair of bases are the transposed
conjugates of each other.
The proof essentially duplicates that of Theorem 5.5. One has to
replace U, '1'* and 1'1'1 = 1 by A, A and 1m}, = 0, respectively. As a
counterpart of VEl = El the inclusion AEl c El can only be stated;
then Theorem 11.3.7 yields A.B* c .B*, which suffices in the rest of the
argument. At the end of the proof the references Theorem 11.2.8 and
Corollary 11.2.6 are to be replaced by Theorem II.}.6 and Corollary
II·3.4· 0
Corollary 6.6. If A is a real, normal eigenvalue of the selfadjoint
operator A E )S(&)), then El;.(A) is non-degenerate. 0
The following lemma, actually a Hilbert space result, serves as a
point of departure when analysing selfadjoint operators with the help
of fundamental decompositions (see Chapters VIII-IX).
Lemma 6.7. Let A be a symmetric operator in the Krein space &).
Consider a fundamental decomposition
(6.1) &) = &)+(+)&)-; &)+ C ~++, &)- c ~--
6. Selfadjoint Operators 135
and the corresponding f1t11damental projectors P+, P-. Sttppose that
5)+ ( SD(A). Suppose also that either A is closed or ;:.-:+(5)) is finite. Set
An = P+AI5)+, A12 = P+AI5)-,
(6.2)
A21 = P-AI5)+, A22 = P-AI5)-.
Then:
a) SD(An) =SD(A21) = 5)+, SD(A12l=SD(A22)=SD(A)n5)-;
b) An and A22 are symmetric;
c) An, A12 and A21 are contin1~ous;
d) in order that A22 be densely defined, closed, or selfadjoint in 5)-
it -is necessary and sufficient that A have the respective property in 5).
Proof. a) and b) are clear.

Let us prove c). Owing to a) and b) the operator An is selfadjoint
on 5)+; therefore, by the closed graph principle, it is continuous.
Further, if A is closed, then A 21 , as difference of the closed operator
AI5)+ and the continuous operator An, is closed; being defined
throughout 5)+, it is continuous. On the other hand, if ;:.-:+(5)) co <
then A21 is defined on a finite-dimensional space, hence continuous
again. Finally, for x- E SD(Ad, y+ E 5)+, J = P+-P- the symmetry
of A, Lemma II.11.4 and the continuity of A21 yield
I(A 12 X-, y+)1 = !(x-, A 21 y+)1 :::; Ilx-ll] IIA2111] IIY+II];
setting y+ = A 12X- we obtain
(6.3)
As to the proof of d), the assertion concerning dense domains imme-
diately follows from a). That A22 is closed if and only if A is so can be
established with the aid of a) and c). It remains to consider the prop-
erty of selfadjointness.
Let A be selfadjoint. Then, in particular, SD(A 12 ) is dense in 5)-.
This circumstance, the symmetry of A and the continuity of A12 (see c))
imply
(6.4)
Suppose now that for a certain y- E 5)- there exists a y; E 5)- with
(6.5) (A 22 X-, y-) = (x-, y;) (x- E SD(A 22 )) .
Leaning on (6.2)' (6.4) and (6.5) we derive
(Ax, y-) = (A21X+' y-) + (A 22X-, y-) = (x+, A 12y-) +
+ (x-, Y;) = (x, A12 y- + Y;) (x E SD(A)) ,
where x± = P±x. Hence, A being selfadjoint, Y- E SD(A) I.e.
y- E SD(A 22 ).
VI. Unitary and Selfadjoint Operators in Krein Spaces
Conversely, let A22 be selfadjoint in S)-. Suppose that for some

fixed YES) there exists a y* E Sj with
(x E ~(A)) .
Setting y± = P±y we find:
(A 22X-, y-) = (A 22 X-, y) = (Ax-, y) - (A12X-, y) =
= (x-, y* -A 21 y+) (x- E ~(A22)) .
Therefore y- E ~(A22)' i.e., y E ~(A). 0
7. Cayley Transformations
Applied to operators in a Krein space S), the Cayley transformations

defined in Section 11.4 prove to be a source of the analogies between
unitary and selfadjoint operators of S).
Theorem 7.1. Let A be a selfadjoint operator in the Krein space S).

Let e,' be complex numbers such that lel=1, ,0:/= " CE Q(A). Then the
Cayley transform
(7.1) U = e(A - U)(A - ,1)-1
is tmitary and UJe have e E! Gp(U).
Proof. It was shown in Lemma 11.4.1 (for the more general case
'E! Gp(A)) that U is isometric and e is not an eigenvalue of U. Since,
according to (7.1),
(7.2) ~(U) = 31(A - '1) , 31(U) = 31(A - ;1) ,
from the assumption' E Q(A) and its consequence CE Q(A) (see Theo-
rem 6.1 a)) we obtain ~(U) = 31(U) = S). 0
Theorem 7.2. Let U be a unitary operator in the Krein space Sj.

Let e,' be complex members such thut \e\=1, ,0:/= C, e E! Gp(U). Then the
Cayley transform
A = (CU - el;1)(U - e1)-l
is selfadjoint and we have 'E Q(A).
Proof. By Lemma 11.4.1 the operator A is symmetric and Cis not
an eigenvalue of A; moreover, (7.1) and (7.3) define mutually inverse
transformations. The latter circumstance implies that the relations
(7.2) still hold. In particular, 31(A - '1) = S). Hence' E Q(A). It
remains to show that A * exists and ~(A *) c ~(A).
7. Cayley Transformations 1:37
From (7.3) we obtain:

(7.4) ;t)(A) = ffi( U - El) .
Since 8 E[ (Jp( U) by assumption and 8 E[ (Jr( U) by Corollary 5.2, it £o11O\vs
that ;t)(A) is dense in SJ. Thus A * is defined.
Let g E ;t)(A *). Denoting A *g by g* we have
(Af, g) = (I, g*) (! E ;t)(A)) .
In view of (7.4) and (7.3) this can be written in the form
((CU - 8[I)x, g) = ((U - cI)x, g*) (x E SJ) .
Taking into account that U is unitary we find:
(Ux, Eg - eCUg - g* + eUg*) = 0 (XESJ)·
Hence
(g - sCUg - g* + sUg* = 0,
1.e.,
1
g = ------=(U - cI)(g* - Cg) .
8(C - C)
Therefore, in view of (7.4), g E ;t)(A). 0
Theorem 7.3. Let A be a closed linear operator in SJ. Let e, ( be

complex nmnbers such that \e\=1, C=I=(, (E[(Jp(A). Then the Cayley
transform
(7.5) U = 8(A - CI)(A - C1)-1
is closed. Moreover, if A =I=- Cis a number belonging to (Jp(A), (Jr(A), (JeCA)
or erA), then the nmnber
(7.6) ')J = 8
A-C
.. _ - -
A -C
belongs to (Jp(U), (Jf(U), O"c(U) or e(U), respectively. It, in particular, A is
a normal eigenvalue of A, then')J is a normal eigenvalue 0/ U.
Proof. We have
(7.7) U = cI + 8(C - C)(A - CI)-1.
As inverses, multiples and translates of closed operators are closed,
U is closed.
Let A E C, A =I=- (, and let ')J be the number furnished by (7.6). If
), E O"p(A) then, owing to Lemma II.S.i, ')J E O"p(U). If, on the other hand,
}, E[ O"p(A), then Lemma II.S.2 applied to the inverses of the transform a-
VI. Unitary and Selfadjoint Operators in Krein Spaces
tions (7.5)-(7.6) (see Lemma 11.4.1 and Remark II.5.3) yields v E! O'p(U).
Moreover, from (7.5) and (7.6) we derive
U - vI = [S(A - eI) - S ~:---~(A - CI)J(A - CI)-l =

A-"
= S C- C(A - U)(A - Cltl .
A-C
Hence ffi(U - vI) = ffi(A - AI).
Finally, let}, be a normal eigenvalue of A. Then S) = EiJ(A) 53, +
where dim EiJ(A) < 00, while 53 is closed, invariant for A and such
that A E e(AI53). This decomposition reduces A, since @5J(A) c ;D(A);
therefore, on account of (7.5), it reduces U as well. Applying one of the
assertions just established to A 153, a legitimate procedure because the
proof only involved the Banach (and not the Krein) space structure
of S), we obtain that v E e(UI53). On the other hand, Lemma II.5.5
yields ®,,(A) = Ei.(U). Thus v is a normal eigenvalue of U. 0
Theorem 7.4. Let U be a closed linear operator in S). Let 13, Cbe corn-
*
plex mtmbers such that 1131 = 1, C C, 13 E! O'p(U). Then the Cayley
transform
A (CU - sCI)(U - eI)-l
*
(7.8) =
is closed. Moreover, if v 13 is a number belonging to O'p(U), O'r(U), O'c(U)

or e(U), then the number
(7.9) A = Cv - sC
v-s
belongs to O'p(A), O'r(A), O'e(A) or e(A), respectively. 1 f, in particular, v is
a normal eigenvalue of U, then}, is a normal eigenvalue of A.
Proof. We have
(7.10) A = C1 + s(C - C)(U - eft l •
Thus A is closed.
The assertion concerning O'p(U), O'r(U), O'e(U) and e(U) follows from
the similar statement of Theorem 7.3, since (7.5) and (7.8) as well as
* *
(7.6) and (7.9) are mutually inverse transformations, while s E! O'p(U)
and v 13 are equivalent to CE! O'p(A) and A C, respectively (d.
Lemma 11.4.1 and Remark II.5.3).
For a normal eigenvalue v of U we argue just as in the proof of
Theorem 7.3, replacing A, A, U, v by U, 'V, A, A, respectively, and re-
ferring to (7.8) instead of (7.5). 0
Now we are able to extend Theorem 6.5 to a large class of discon-

tinuous selfadjoint operators.
Theorem 7.5. Let A be a selfadjoint operator in Sj. Suppose that

a(A) *- C. I I }, is a normal eigenvalue of A, so is A, and we have ItlA(A):j:j:
:j:j:1tl;:(A); moreover, the matrices of A\Itl;.(A) and A\Itl;:(A) with
respect to any dHal pair 01 bases are the transposed confHgates ot each
other.
Proal. Since e(A) is open and non-empty, it contains a non-real
point C. By Theorem 7.1 the operator U = (A - U)(A - C1)-1 is
unitary. According to Theorems 2.2 c) and 7.3 the number v =}, - C

Je -C
is a normal eigenvalue of U. On account of Theorem 5.5 also v* is
.
a normal eIgenvalue of U, whereas Itlv(U):j:j:ltlv'(U). But v* = =---=-
Je-t
.
_ Je-C
Expressing A through U and Je through v* (d. Lemma II.4.1 and Re-
mark II. 5.3), with the help of Theorem 7.4 we obtain that Ais a normal
eigenvalue of A. Moreover, 1tl,,(A) and 1tl;.(A) form a dual pair, since
they coincide with Itlv(U) and Itlv'(U), respectively (Lemma II.5.5).
To get the desired conclusion for the matrices of A \1tl.(A) and
A\Itl;.(A), we apply Theorem II,J.6 to the space @ = 1tl,,(A) +
1tl;:(A),
whose non-degeneracy is a consequence of Corollary II.3.4 and
Lemma I.10.1 in the case A *- A, and of the relations @ = 1tl;.(A),
1tl,,(A):j:j:ItlA(A) in the case Je = A. 0
8. Unitary Dilations
Consider a Hilbert space Sjo and a linear operator T E m-(Sjo). Suppose

there exists a Krein space Sj containing an ortho-complemented sub-
space Sj(O) that is isometrically isomorphic to Sjo, and a unitary operator
U E m-(Sj) satisfying the relations
(8.1) ynx = PU"x, (T*)n x = Pu-nx
(x E Sj(O); n = 1,2, ... ) ,
where P is the orthogonal projector from 'l)(P) = Sj onto ffi(P) = Sj(O).
This situation will be expressed by saying that U is a ~tnitary dilation
of T on the Krein space Sj ) ,~o. If the span of the subspaces un Sj(O)
(n = 0, +1, ±2, ... )
is dense in Sj, we say the unitary dilation U is
minimal.
One sees immediately that in the case where indefinite inner prod-
ucts are not admitted T can have a unitary dilation only if T is
a contraction (i.e., its norm does not exceed 1). This condition is
kno-INn to be sufficient as well. It turns out that in the general case
every T E 18(~o) has a unitary dilation. Before proving this, however,
we need some preparation.
Let ~o be a Hilbert space. Recall that a positive operator B E 18(~o)
has a unique positive square root B1f2 E 18(~o); moreover, B1/2 is the
strong limit of a sequence of polynomials of B (see e.g. Riesz and
Sz.-Nagy [1; Section 104J). In particular, we have:
Lemma 8.1. Let ~o be a Hilbert space. If two positive operators
Bv B2 E 18(~o) and an operator 5 E 18(~o) fulfil the condition
SB 1 = B 2S, then SBi/ 2 = B~/2S. 0
The following special case of Lemma 2.'10 is also related to square
roots.
Lemma 8.2. Let ~o be a Hilbert space. If A E )8(.5)0) is selfadjoint,
then m(A2) = 91(A). 0
For a selfadjoint operator A E )8(~o) (~o a Hilbert space) it is

usual to set
(8.2) . IAI = (A2)1/2, sgnA = E;; - E"A '
where E;; and E"A are the orthogonal projectors onto the subspaces
(8.3) ffi(E;;) = ~1 = m(A)l nm(A - IA I)
and
(8.4) lH(E"A) = ~"A = m(A)-Lnm(A +
IAI),
respectively.
The following two results, though seldom mentioned explicitly, are
known. We state and prove them in order to make the discussions
involving sgnA possibly elementary and self-contained.
Lemma 8.3. The s~tbspaces defined by (8.3) and (8.4) satisfy the
relations ~;; ~ ~A' ~1 +
~A = 91(A)-L.
Proof. As a consequence of Lemma 8.1 the operator A commutes
with IAI. Therefore, in view of Theorem 2.6 d), A~1 c ~1. Moreover,
again by Theorem 2.6 d), A~1 is dense in ~1, since 91(AI.5)1) = o.
But for XE~;;, YE~A (denoting the inner product on ~o by (. ")0) we
have:
(Ax, Ay)o = (1/2 (A + IAI)x ,1/2 (A - IAI)Y)o =
= 1/4 ((A2 - IAI2)X, Y)o = O.
Thus A~;; ~ A~A and, taking closures, S); 1 S'j:1.
In order to establish the second half of the statement consider an

element Z E Sjo and put
x = 1/2 (A + IA I)z , y = 1/2 (A - IAl)z.
Then Az = x +y, where (A - IAl)x = 0 and (A +
IAl)y = O.
Making use of the relations ffi(A)d.Q(A) 1, ffi(IAl)d.lC(IAl)l, 9l(IAI) =
= 9l(lA 12) = 9l(A 2) = 9l(A ) (d. Theorem 2.6 d) and Lemma 8.2) we
obtain that x E Sj;L y E SjA'. Consequently, ffi(A) c Sjl+SjA'. It re-
+
mains to observe that Sj::t SjA' is an orthogonal direct sum of closed
subspaces in a Hilbert space, hence closed, while ffi(A) is dense in
9l(A)l. 0
Lemma 8.4. Let Sjo be a Hilbert space. Let S,AI,A2 E ~(Sjo), A! = AI>
A~ = A 2. Denote by Pi (j = 1,2) the orthogonal projector with range
9l(Aj)' If SAl = A 2S, then SPI = P 2 S.
Proof. SAl = A 2 S implies that S9l(A l ) c 9l(A 2 ). Hence
(8.5) SPl = P 2SP1 •
Similarly, AlS* = S* A2 yields S*9l(A 2 ) c 9l(Al) i.e.
(8.6)
Taking adjoints in (8.6) and comparing with (8.5) we obtain

SP} = P 2 S. 0
It is via the next assertion that Lemmas 8.1-8.4 contribute to the
proof of the main result of this section.
Lemma 8.5. Let Sjo be a Hilbert space. For TE ~(Sjo) put

(8.7) QT = II - T*TI 1 / 2 , QT* = II - TT*ll/2,
(8.8) JT = sgn(I - T*T) , fy. = sgn(I - TT*)
(cf. the definitions (8.2)). Then the following relations hold true:
(8.9) TQT = QT,T , T*QT* = QTT* ,
(8.10) TJT = fyoT, T*JT* = JTT*,
(8.11 ) QyJT = JTQT, QT*J T* = J T*QT* ,
(8.12) QilT=I-T*1', Q'}..J T* = I - TT* ,
(8.13) QTJi = QT' QT*J}. = QT* .
Proof· Set
(8.14) AT = 1- 1'*T, AT* = 1- TT*.
Denote by FI, Fy, Fj, Ft., F T*, F'i. the orthogonal projectors satis-
fying the conditions
(8.15) ffi(F~) = m(AT)l n m(AT =+= IATIl,
(8.16) ffi(F~) = m(A T) ,

(8.17) ffi(F~) = m(AT.)l n IJc(Ar- + IAr-1) ,
(8.18) ffi(Fj.) = m(Ap) .
Now, in view of (8.2) - (8.4), the definitions (8.7) -(8.8) can be written
as
(8.19) QT = IATI1/2 , QT* = IA T.1 1!2 ,
(8.20) ]T = Fi - Fy , ] T* = Ft. - FT' .

Lemma 8.2 yields m(IArl)=m(A T), so that m(AT)cm(AT=+=IATI).
Thus the orthogonal· projectors FI F ~, Fy +
F~ have the +
ranges
(8.21 )
respectively. In the same way, for the orthogonal projectors Ff,. + F~*
we have
(8.22)
Observe that in each of the items (8.9)-(8.13) it is sufficient to
verify the first equation, since applying it to the operator T* we obtain
the second one.
After these preliminaries, let us prove (8.9). The definitions (8.14)
yield TAT = ApT. Hence we deduce the relation TQT = QT*T
with the help of (8.19), (8.2) and a repeated application of Lemma 8.1-
Again, owing to (8.14) and Lemma 8.1,
T(AT + IATI) = (Ap + lAp\) T .
Therefore, in view of (8.21), (8.22) and Lemma 8.4,
T(F~ + F;) = (F~. + F~.)T .
Hence, by the definition (8.20), T] T = Jr. T .
Further, making use of the formulas (8.19), (8.2) and (twice) of
Lemma 8.1, we obtain that QTAT = ATQT' Applying Lemma 8.1
once more, we find:
QT(A T + IATI) = (AT + IAT\)QT .
The same steps as in the preceding paragraph lead to the conclusion
QrIT = ]TQT'
To prove (8.12) we start from the relations
IATI Ft = ATFt, IATI Fy. = -ATFy.
(see (8.15)). Along with (8.16) and Lemma 8.3 they imply:
IATI (F:j: - Fr) = AT(F:j: + Fr) = AT(I - F~) = AT .
Hence, in view of (8.19), (8.20) and (8.14), Qif T = 1 - T*T.
Finally, from (8.20), (8.15), (8.16) and Lemma 8.3 we infer
Ji. = (F:j: - Fr)2 = F:j: + Fr = I - F~
so that, according to (8.'19) and Lemma 8.2,
QTJi. = IATI1/2(I - FJ) = IATII/2 = QT' 0
We are now coming back to the dilation problem.

Theorem 8.6. Let.5)o be a Hilbert space, and let T E ~(.5)0). Then T
has at least one minimal unitary dilation on a sttitable Krein space
.5)).5)0'
Proof. We shall use the notations (8.7)-(8.8). The inner product
and the norm on.5)o will be denoted by (. , ')0 and II . 110' respectively.
Consider the space
~) = X.5)i'
i=~oo
the cartesian product of the Hilbert spaces 'S)i (f = 0, ±1, ±2, ... ),
where
f
~~{
.5)0 if = 0,
(8.23 ) e 1 ffi(QT) if f> 0,
ffi(QT*) if j < 0.
In other words, .5) is the Hilbert space of all sequences X= {xi}~-oo
< ex);
00
with Xi E .5)i (j = 0, +1, +2, ... ) and 2: Ilxill~ the

i=~oo
(positive definite) inner product on oS) is given by the formula
(8.24) [x, yJ = }; (Xi' Y)o (X,YE.5)) .

i=-oo
Define a linear operator J on .5) setting
if j = 0,
I
1
Xo
(8.25) (JX)j = J TXj if f> 0,
J T'Xi if j < 0.
By the aid of Lemma 8.5 it is easy to see that ffi(J) c .5) and J- 1 = J.
Moreover, J is continuous and selfadjoint. Thus, according to
Theorem V.i.}, the inner product

(8.26) (x, y) = [lx, yJ (X,y E .5;,»)
turns .S) into a Krein space.

The subspace
( 8.27) .5;,)(0) = {x E .5;,): X I =o for ficO}
is isometrically isomorphic to .5;,)0, and the complementary subspace

{x E .5;,): Xo = O} is orthogonal to .5;,)(0) with respect to both of the
inner products (8.24), (8.26).
Next define a linear operator U on .5;,) by the formula
QT*X_l + Txo if f = 0,
(8.28) (Ux)j = \ - ]TT*Lj + QTXO if f = '1 ,
Xj_l otherwise.
Leaning on Lemma 8.5 one verifies the relations ffi(U) c ,5),

(Ux, Uy) = (x, y) (x,y E .5;,»). Moreover, with the help of the same
lemma it can be checked that U-l is defined on all of .5;,); namely
T*xo + ITQTxl if f= 0,
(8.29) (U-1X)j = 1 ]T*Qr'XO- jT*TX1 if j = -1 ,
Xi+l otherwise
for every x E.5;,). Hence U is unitary with respect to (. , .).
Let x E Sj(O) (see (8,27)). It follows by induction relative to n that
in this case
if j = 0,
(8.}0) if 1 ~ j :S n,
otherwise
and
if j = 0,
(8-31 ) if -n::;; j < -1 ,

otherwise
(n = 1,2, ... ). Considering j = 0 we obtain the dilation property (8.1)
of U.
Notes to Chapter VI 145
Let x,y,z E 5)(0)' Yo = Tx o , Zo = P'xo ' Then (8·30) and (8.3 1)

yield:
if f= n,
otherwise,
if f = -n,
otherwise.
But, according to (8.23), ffi(QT) is dense in 5)", and the sJIbspace
ffi(J T*Qp) ) m(J T*QT*j p) = m(QT*)
(d. Lemma 8.5) is dense in 5)-n (n = 1,2, ... ). Thus the span of the
sets UtII5)(O) (m = 0, +1, +2, ... ) is dense in each of the subspaces
5)(y) = {XE5): xj=Oforf#r} (r=0,±1,±2, ... )
and, consequently, in S). 0
Notes to Chapter VI
Linear operators in quasi-positive Krein spaces have first been studied

by Pontrjagin [1]. The investigation of operators in arbitrary Krein
spaces was begun by Ginzburg [1J, [2J.
Properties of the adjoint in inner product spaces with a Hilbert
majorant are discussed in the expository paper of Azizov and Ioh-
vidov [2]. Lemma 2.10 (for T E )8(5))) was formulated by KreIn and
Smul'jan [2].
The propositions of Section 3, some of them even for inner product
spaces with a Hilbert majOrant, belong to Iohvidov [4J, [11]; d. also
Azizov and Iohvidov [2]. For analogous questions concerning Banach
spaces with a j-metric see Iohvidov [14J, Iohvidov and Senderov [1J.
Theorems 4-3 and 4.4 are due to KreIn and Smul'jan [2]. Lemma 4.9
appears in a paper of Langer [5J, and more explicitly in unpublished
lecture notes by the same author.
A class of "almost unitary" operators in 5) has been investigated by
Kuzel' [1].
The results of Section 5 are natural counterparts for those of
Section 6, obtained by Langer [1J, [2]. In contrast to our geometrical
treatment of normal eigenvalues, Langer established Theorem 6.5,
and its extension to spectral sets of a possibly discontinuous selfadjoint
operator, with the help of general spectral theory (integration of the
resolvent).
Let us note that our method of proving Theorem 6.5 applies to
discontinuous selfadjoint operators A as well, but in this case the aux-
iliary relation ((A - AIt) * = (A - iI)" is far from being triviaL

If we take the latter equation for granted, Theorem 7.5 becomes super-
fluous.
Examples of selfadjoint operators in ~ whose residual or continuous
spectrum contains non-real points have first been published by Azizov
[1], while the construction presented here as Example 6.4 is due to
Langer (personal communication).
Azizov and Iohvidov [2J found that symmetry properties of spectra
do not as a rule remain valid in non-degenerate inner product spaces.
with a Hilbert majorant. Noel [2J generalized Langer's results to
operators T E 5.8(~) such that T* is a function of T. Senderov [1J
examined the spectral radius and principal subspaces of J-isometric
operators in Banach spaces with a J-metric.
Hess [2J studied polynomials of certain selfadjoint operators in ~.
Berezin [1J defined selfadjoint operators in arbitrary non-degenerate
inner product spaces and described selfadjoint extensions of symmetric
operators.
The assertions of Theorems 7.1-7.4 in the case of a quasi-negative
~ largely belong to Iohvidov and Krein [1]. For an exposition in the
case of non-degenerate inner product spaces with a Hilbert majorant
see Azizov and Iohvidov [2]. In the second half of the proof of Theo-
rem 7.2 we follow Riesz and Sz.-Nagy [1J.
Theorem 8.6 is due to Davis [1], who admitted discontinuous opera-
tors and discussed unicity of the dilation too. For the definite case see
e.g. Sz.-Nagy and Foia~ [1]. Davis [2] extended his result to semi-
groups. An application to characteristic operator functions was given
by Davis and Foias [1].
J alava [2J carried over several results of the present chapter to
Banach spaces with a continuous, but possibly non-symmetric bilinear
form.
Chapter VII. Positive Operators and Plus-operators
in Krein Spaces
Positive operators in a Krein space (Sections 1 and 3) have turned out (see Chap-
ter VIII, especially Section 6 and the Notes) to be the most fruitful generalizations
of positive operators in Hilbert space. In Section 2 an alternative extension of the
latter concept is considered. Sections 4- 6 resume the study of plus-operators begun
in Section II.S. A unifying feature of the chapter is the repeated application of the
Schwarz inequality. Theorems 1.3, 3.1, 5.2, 5.4, 5.5 and 6.1 deserve particular
attention.
1. Positive Operators
A symmetric operator A in the Krein space ~ is said to be positive

if (Ax, x) > 0 for every x E 'tJ(A).
We begin with two elementary but substantial facts.

Lemma 1.1. Let A be a densely defined positive operator in~. If an
element Xo E 'tJ(A) satisfies (Axo, x o) = 0, then Axo = O.
Proof. Applying Lemma 1.4.4 to the space @ = 'tJ(A) and the
inner product (x, Y)A = (Ax, y) (x,y E @) we obtain that (Axo, y) =
=--= 0 (y E 'tJ(A)). Thus Axo-L~, Axo = o. 0
Theorem 1.2. Let A be a densely defined positive operator in the
Krein space~. If A is a positive (negative) eigenvalue of A, then the
corresponding eigenspace is positive definite (negative definite).
Proof. Let Axo = AXo , where A >0 and Xo =I- O. Then Axo =I- 0,
so that by Lemma L1 (Axo, x o) -;- O. Hence (xo, x o) = 1/A (Axo, xo) >
> O. For}, <
0 the reasoning is similar. 0
The spectrum of a selfadjoint operator A E ?S(~) need not be con-

tained in the realline (d. Examples 11.3.2 and V1.6A). For a positive A,
however, the situation is different.
Theorem 1.3. The spectrum of a positive operator A E ?S(~) is real.
148 VII. Positive Operators and Plus-operators in Krein Spaces
Proof. Suppose that A E actA) uap(A). Let J be a fundamental

symmetry on.p. Then there exists a sequence {x,,}~ ( Sj with the
propertiEs
(1.1) Ilx"II J = 1 (n = 1,2, ... ) ,
(1.2) IIAx" - AX"IIJ -+ 0 (n ->-(0) .
Making use of Lemma II.11A we find:
(1·3) (n-;..oo) .
Since !(x"' X,.) I ~ Ilx"ll} = ·1 for every n, we may assume that
(1.4) (x", x,,) -;"0; ,
a real number. Then (1-3) yields:

(1.5) (Ax",x,,)-;"AIX (n-;..oo).
According to (1.5), IX >
0 implies A 2: 0, whereas IX <0 implies
), :s:: o.
Next let IX = 0, i.e.,
(1.6) (Ax" ' x,,) ->- 0 (n-;.. (X)).
Applying Lemma 1.2.2 to the A-inner product, making use of Lemma
II .11.4 and recalling (1.1) we obtain:
IIAx"ll} = (Ax", J Ax,,) = (x"' J AXn)A :s::
< (xn' xn)~2 (JAxn , J Ax,,)~2 =
= (Ax", X,,)1 /2 (AJAx", JAxn) 1/2 :S:: (Ax", x,y1211AIIJ2.
Thus, in view of (1.6), IIAxnllJ -;.. 0 (n -;"(0). Hence, on account of
(1.2) and (1.1), }, = O.
Consequently, ac(A) and ap(A) are real. From Theorem V1.6."\ c)
we conclude that a.(A) is empty. D
Under the assumption and with the notations of Theorem 11.9.4
A - fk A I is a positive operator. Therefore, by Theorem 1.3 and Theo-
rem 11.9.1, the following supplement to Corollary 11.9.9 holds:
Corollary 1.4. The spectrum of a Pesonen operator A E ~(.S)) is real. D
The spectrum of a discontinuous positive selfadjoint operator need
not be real.
Example 1.5. Let the space Sj = Sjo X Sjo and the inner products
[. , .J, (. , .)
be defined as in Example VI.6.4. Let Bo be a discontinuous
positive selfadjoint operator in the Hilbert space .S)o' For x E '1:J(Bo),
y E.S)o set A {x, y} = {a, Box}. The operator A is positive and selfad-
joint with respect to (. ,.). On the other hand, choosing an Xo l! '1:J(Bo)
the element {xo, o} does not belong to ffi(A - }el) for any Jc. Hence
alA) = c.
2. Operators of the Form T*T 149
If .iQ is a Hilbert space (,c(S)) = 0), then each of the follmving

conditions is known to be necessary and sufficient in order that a self-
adjoint operator A E ~(~~) be positive: a) the spectrum of A is non-
negative; b) there exists a selfadjoint operator E E $1'(.iQ) such that
A = E2; c) there exists a linear operator T E ~(S)) such that
A = T* T. On the other hand, if ,$ is an indefinite Krein space then I
is not a positive operator on .iQ, though it satisfies the above conditions.
Thus in the general case a), b) and c) are not sufficient for the positivity
of A. It will turn out from Examples 1.6 and 2.3 below that they are
not necessary either.
Example 1.6. Let dimS) <00, x(S)) >
o. Any fundamental sym-
metry J on S) is positive. Suppose that J = E2 for a selfadjoint opera-
tor E. Since -1 is an eigenvalue of J, making use of the Jordan
normal form it follows that i or - i is an eigenvalue of E, i.e., Ex = ix
or Ex = -ix for some x =F O. Corollary II.3.4 yields (x, x) = O.
But Jx = E 2x = -x and the definition of J imply (x, x) O. <
Thus J has no selfadjoint square root.
2. Operators of the Form T*T
Let S) be a Krein space and A E ~(S)) a selfadjoint operator. The

vector space S) equipped with the A-inner product
(2.1) (x, Y)A = (Ax, y) (x,y E S))
can be regarded as an inner product space S)A' Since S)A admits the
strong topology of S) as a majorant, S)A is decomposable (d. Theorem
IV.S.2). The rank of positivity (negativity) of S)A will be called the
rank of A-positi'l;ity (A-negativity) of S) and denoted by x1(S)) (x;(S))).
Thus ~;1(S)) = X+(S)A)' x;(S)) = x-(S)A)'
Theorem 2.1. Consider a Krein space S) and a selfadjoint operator
A E ~(S)). There exists a 11:near operator T E ~(S)) satisfying the equation
(2.2) A = T*T
if and only if
(2·3 )
and
(2.4) x;(S)) < x-(S)) .
Proof. vVe first note that for T E ~(S)) (2.2) is equivalent to the
relation
(2.5) (x, Y)A = (Tx, Ty) (x,y E S)) .
In other words, T must be an isometric linear mapping of S)A into S).
Let us fix a fundamental and an A-fundamental decomposition

of Sj:
(2.6) Sj = Sjf(+)Sj-,
(2.7) SjA = Sj~ (+)A Sjl (+lA Sj::!,
the superscript indicating, as usual, the sign of the respective inner
square on the subspace in question.
Suppose that (2.5) is satisfied by some T. Then the range of
TISjl is a positive definite subspace ~)l+cSj with dimintl)l+ = xl(Sj).
But, according to Lemma IV.7.3, dimintffi+ < x+(Sj). Thus (2.3) holds.
The inequality (2.4) can be proved in a similar way.
Suppose, conversely, that (2.}) and (2.4) are valid. Owing to (2.3)
and the intrinsic completeness of Sj+ there exists an isometric isomor-
phism T+ of Sjl c SjA onto a subspace of Sj+ C Sj. Similarly, (2.4) im-
plies the existence of an isometric isomorphism T _ carrying Sj::! into
a subspace of Sj-. If for
.0 E ,,0
xA "dA' XI E Sjl '
we set
Tx = T +xl + Lx::! '
then (2.5) will be satisfied. Moreover, for the fundamental symmetry]
that corresponds to (2.6) we have
IITxllj = (T ... xl, T+xl) - (Lx::!, Lx::!) =
= (Axl, xI) - (Ax::! ' x::!) < IIAII] (1lxlllj + Ilx::!llj) .

Therefore if we additionally require from the decomposition (2.7) that
xI , x::! continuously depend on x (by Theorem IV. 5.2 and the closed
graph principle this can be achieved), then T E l8(Sj). 0
The following result is a special case of Theorem 2.1.
Theorem 2.2. Let Sj be a Krein space with dimSj = 00, u+ (Sj) =
= u-(Sj). Then /or every sel/adjoint operator A E l8(Sj) there exists a l1:near
operator T E l8(Sj) such that A = T*T.
Proof. In view of the assumption, x+(Sj) = x-(Sj) = x+(Sj) +
+ x-(Sj) = dim Sj. Therefore, by Theorem 2.1, it is sufficient to
prove that xl(Sj) < dimSj and x::!(Sj) s dimSj for every selfadjoint
A E l8(Sj). But this follows from the (strong) continuity of (. , ')A;
namely the intrinsic A-dimension of any A-definite subspace 2 c Sj
is less than or equal to the (strong) dimension of 2. 0
In contrast to Theorem 2.2, there are numerous cases where (2.3)
or (2.4) is not satisfied.
3. Uniformly Positive Operators 151
Example 2.3. Let 5) be a Krein space with rv+(5)) dim 5), and let] <
denote a fundamental symmetry on 5). Then rvt(5)) rv+(5)), so that >
by Theorem 2.1 ] cannot be written in the form T*T.
3. Uniformly Positive Operators

Consider a Krein space 5) and a selfadjoint operator A E )5(5)). We
shall say that A is uniformly positive, if for a fundamental symmetry]
and a suitable positive number (X it satisfies the inequality
(3·1 ) (Ax, x) > (Xllxll} (x E 5)) .
In other words, A is uniformly positive if the function
(3.2) PA(X) = (Ax, x)1/2 (x E 5))
is a norm defining the strong topology TM(S)),
Every uniformly positive operator on an indefinite 5) is a Pesonen
operator (ef. Section 11.9).
The next result should be compared with the following special case
of Theorem 11.9.4:
>
(x, x) > 0 (x *
Let 5) be indefinite, A E )5(5)), A'~ = A, and (Ax, x)
0). Then A - PAl, where
0 for
(3-3) flA = inf (Ax, x) ,

(x,x)=l
is a positive operator.
Theorem 3.1. Let 5) be an indefinite Krein space, A E )5(5)), and
A * = A. Suppose that for a fundamental symmetry ] and suitable
positive number (X the ineq'uality
(3.4) (Ax, x) > (Xllxll} (x E ~+)
holds. Then there exists an co> 0 such that for every 0 < c < Co the
operator A-(PA-c)l (cj. (3.3)) is uniformly positive.
Proof· From (3.3) and (3.4), writing PA = P and applying Lemma
II.11.4, we obtain:
P = inf ~A::~~t > inf jAx, ::). > (X > 0 .
(x,x»o (x, x) -(x,x) >0 Ilxll} -
Further, by Theorem 11.9.4,
(3·5) (Ax, x) > fl(x, x) (x E 5)) .
For x E ~+, 0 < c < P relations (3.5) and (3.4) yield:
(Ax, x) - (p - c)(x, x) > (Ax, x) ( 1 P -

- -----;;:-
c) 2
(Xc ll }.
-;llx
Consequently, if we prove the existence of a number 0> 0 satis-

fying
(3.6) (Ax, x) - p(x, xl ::2:: ollxll}
our theorem will be established with 8 0 = min{,u, o}.
Assume that (3.6) is not fulfilled by any positive o. Then there
exists a sequence {xn}~ C oS) satisfying the relations
(3.7) (xn' x,,) < 0, l!x"ll; = 1 (n = 1,2, ... ),
(3.8) (Ax", x,,) - p(x", x,,) -+ 0 (n -+ co) .
On the other hand, in view of (3-3), (3.7) and the estimate I(x", x,,)1 <
< Ilx"II~, we can find a sequence {y,,}~ c oS) such that
(3·9) (y", y,,) = -(x"' x,,), Re(xn , y,,) = 0 (n = 1,2, ... ),
(3·10) (Ay n , Yn) - p(Y", Yn) -> 0 (n-;..cx)).
Consider the neutral vectors z" = x" - y" (n = 1,2, ... ). Accord-
ing to condition (3.4) we have:
Ilx" - y"ll; = Ilz"ll] < o,-1/2(Az" , zn)1/2 = ex- 1/ 2 ((A - pI)z", Z,,)1/2.
Hence, applying the triangle inequality to the positive semi-definite
inner product (. ")A-pI (see (3.5)) and making use of relations (3.8),
(3.10) we obtain:
(3.11) Ilx,,-ynll]-+O (n-+co).
But (3.4) is valid for x = Yn too:
(3.12) (Ay", Yn) > ex (n = 1,2, ... ) .

IIYnll~ =
From (3.12), (3.11) and (3.7) we conclude that
(Ax", x,,) > -ex2 (n> no),
contrary to the assumption (3.7)-(3.8). 0

In Chapter VIn we shall make use of the following simple fact.
Lemma 3.2. Every uniformly positive operator A E m(oS)) is com-
pletely invertible.
Proof. Relation (3.1) and Lemma II.11.4 yield IIAxll]llxll; >
> ex \\xIIJ i.e. IIAxll; > exllxll; with an ex '-> 0 for every x E oS). As in the
Hilbert space case, Theorem VI.2.6 d) leads to the desired conclu-
SIOn. 0
4. Plus-operators
In this section we assume that x(~) > o.

We are going to study plus-operators T E m-(~), i.e., continuous
linear operators T defined on the indefinite Krein space ~, the value
(4.1) f1o(T) = inf (Tx, Tx)
(x,x)=l
being non-negative (d. Section II.8). Recall that for every operator of
this kind
(4.2) (Tx, Tx) > f1o(T) (x, x) (x E ~)
(Theorem II.8.1).
According to relation (4.2) the operator
(4·3) A = T*T - f1o(T)I
is positive. Thus from Theorem 1.3 we obtain:
Lemma 4.1. If T E ){l(~) is a plus-operator, then a(T*T) is real. 0
In the case where both T and T* are plus-operators we can prove
more.
Theorem 4.2. If the adjoint of the plus-operator T E ){l(~) is also a
fJlus-operator, then a( T* T) is non-negative.
Proof. We use the notations (4.1), (4.3). By relation (4.2) the
operator A is positive. Let AE ap(A) u actA). Repeating the proof of
Theorem 1.3 we find a sequence {xn}~ C ~ such that
(4.4) Ilx"ll, = 1 (n = 1,2, ... ), IIAx" - Ax"llr-+- 0 (n --'>(0) ,
t4.5) (xn' x,,) -+ ex, (Ax", x,,) -)- },ex (n -+(0) ,
where J is a fixed fundamental symmetry and ex E R. According to the
same proof ex >0 implies}, > 0, while ex <
0 yields A o. In par- ::s
ticular, a,(A) = €I (d. Theorem VI.6.1 c)). It remains to show that
}. > -f1o(T) even in the case ex o. <
Let
(4.6) ex < 0 , }. < -f1o(T) .
From (4.3)-(4.5) we obtain:
(T*Tx", T*Txn ) = (Axn' Ax" - }.x,,)+ (Ax", AXn) +
+ 2f1o(T)(Ax", xn ) + f10 (T)(xn' xn ) -)-
2 ex(}. + f1o(T))2.
Therefore, in view of (4.6),
(4.7) lim (T*Tx" , T*Txn )
n-..oo
<0.
154 VII. Positive Operators and Plus-operators in Iu-em Spaces
On the other hand, from (4-3) and (4.5)

(Tx.. , Tx,,) = (Ax.. , x .. ) + f.l(T)(x", x,,) -+ cx(}. + f1(T)) ,
so that (d. (4.6))
(4.8) lim (Tx", Tx,,)
" ...... 00
>0.
Furthermore, Theorem n.8.1 applied to T* (which is a plus-operator by
assumption) yields
(T*Tx", T*Tx,,) > f.l(T*)(Tx", Txn ) (n = 1,2, ... ),
where
f.l(T*) = inf (T*x, T*x) > 0 .
(x,x)=l
Hence, on account of (4.8),

(T*Tx.. , T*Tx n ) > 0 (n > no) .
This, however, contradicts (4.7). 0
In connection with Theorem 4.2 it should be noted that the ad-
joint of a plus-operator need not be a plus-operator.
+
Example 4.3. Let ~ = ~+( )~-, ~+ C q5++, ~- C q5--, dim~+
count ably infinite, dim ~- = 1. Let {ei}~ and {j} be complete
orthonormal systems in ~+ and ~-, respectively. Define a linear
operator T E ~(~) by the relations Te j = ei+l (j = 1,2, ... ), Tf = e1.
Then T is a plus-operator, but T* carries the positive e1 into the nega-
tive -f.
5. Strict Plus-operators
We still assume that x(~) > O.

According to the definition given in Section n.8, the plus-opera-
tor T is said to be strict if the value f1(T) (see (4.1)) is positive.
Lemma 5.1. Let T E ~(~) be a strict plus-operator. If E is a Positive
(Positive definite, uniformly positive) subspace of ~, so is TE.
Proof. For a positive or positive definite E the assertion directly
follows from (4.2). Next let E be uniformly positive:
(x, x) > cxllxllj (x E E)
for a fundamental symmetry J and positive number 0.:. Then, again
by (4.2),
(Tx, Tx) > f.l(T)(x, x) > f.l(T)cxllxllj > ~i~~~ !ITxll;

for every x E E. 0
5. Strict Plus-operators 155
Theorem 5.2. If T E )B(S)) is a strict plus-operator and J is a j2tnda-

mental sY111Jnetry on ,~, then there exists a ntl.lnber 0 0 such that >
(5.1) (x E 1.l3+) .
Proof. The operator A = T*T - ,uI, where It = It(T) > 0, is
positive (d. (4.1) and (4.2)). Let e E 1.l3+, Ilell]=1, XES). Lemma I.2.2
applied to the A-inner product (. , .).4 yields:
I(Ax, e)12 :::; (Ax, x)(Ae, e) :s;;: [(Tx, Tx) - p(x, x)](Te, Te) :::;
2 (IITII J + p) Ilxll} IITell}.
Setting x = J Ae we obtain:
IIAell} < (IITIIJ + pl IITellJ .
On the other hand,
IIAell] ~ fA,llell] -- IIT*Tell] ~ p - IIT*II] IITell].
A.s a result,
I.e.,
p
IITell] ~ (t[ril}+ ,;)1/2 +liT;"I I] . D
Corollary 5.3. If T E )B(S)) is a strict plus-operator and £ is a closed
pos1:tive subspace of S), then T£ is closed. D
Theorem 5.4. If T E )B(S)) is a strict plus-operator, then the null
space ill (T) is uniformly negative.
Proof. The assumption ensures the existence of a number p > 0
such that the operator A = T*T - pI is positive (d. (4.'1) and (4.2)).
Let us fix a fundamental symmetry J and an element XES). Setting
JAx = y, with the help of Lemma 1.2.2 (applied to the A-inner
product) and Lemma II.ltA we obtain:
IIAxllj = (Ax, Ax)] = (Ax, y)2 < (Ax, x)(Ay, y) 2
< (Ax, x) IIAII] Ilyll} = (Ax, x) IIAII] IIAxIIJ.
Consequently,
(5.2) IIAxll] < IIA IIJ (Ax, x) (x S)) .
* *
E
If xE ill(T), x 0, then Ax = -px (in particular, A 0), so that

from (5.2)
(x, xl < - ~ IlxllJ . D
- IIAIIJ
The following theorem yields a classification of strict plus-opera-
tors T E )B(S)).
Theorem 5.5. Let T E ~(5)) be a strict plus-operator. Then for every

maximal positive subspace 2 c 5) the plus-defect d+(T2) is the same.
Proof. By Theorem V.4.1 and Corollary 5.3 d+(T2) exists.

Consider a fundamental decomposition
(5·3)
the corresponding fundamental projectors P+, P-, and the fundamen-
tal symmetry] = P+ - P-. Put x+ = P+x, x- = P-x for every
x E 5). Let
be the operator matrix of T with respect to (5.3), the indices 1,2

referring to the components 5)+ and 5)-, respectively. Then
(5.4) (x E 5)) .
Define
(5.5)
In view of Theorems I1.11.7 and V.4.2 a subspace 2 c 5) is maximal
positive if and only if it can be written in the form
(5.6)
where K E sr. From (5.4) and (5.6) we obtain:
(5.7)
Thus we have to prove (d. Lemma V.4.6) that
(5.8) codim.p+(Tll +T 12K)5)+ = codim.I)+(Tl l + T K')5)+

12
(K, K' E sr) .
sr
Since, however, is convex, it is sufficient to prove (5.8) for 11K' - KII]
smaller than some e >
O. We also may assume that T12 *- O.
Theorem 5.2 and the relation 2 C ~+ yield IITxllJ > ollxll] (x E 2),
where the number 0>
0 is independent of 2. Making use of the
inequality (x, x) > 0 once more, we derive:
(x E 2) .
Therefore, on account of (5.7),

(5.9)
5. Strict Plus-operators 157
Set B = Tn + T12K and define a linear operator B(-I) on Sj+ by

the relations
B(-l)y = B-ly (ydR(B)) ,
B(-1)y = 0 (YE Sj+, Y1- l ffi(B)).
For every K' E st' we have
(5.10) Tn + T12K' = [I + T12(K' - K)BHl]B .
But (5.9) implies IIBH)II] < V;. Consequently, if
(5.11) 11K' - KIIJ < o

-;=---
V2 IITdl l
then IIT12 (K' - K)BH)II] < 1, so that in (5.10) B is multiplied by
a completely invertible operator of the space Sj+. In other words, under
the assumption (5.11) there exists a homeomorphic isomorphism be-
tween Sj+ and Sj+ which maps (Tn +
T 12 K)Sj+ onto (Tn T 12 K')Sj+. +
This proves (5.8) for K, K' satisfying (5.11). 0
Theorem 5.5 justifies the following definition.

Let T E 55-(Sj) be a strict plus-operator. The cardinal number
d+(T£), where £ is any maximal positive subspace of the Krein space Sj,
will be called the plus-defect of the operator T and denoted by d+(T).
Below we present two simple summation rules for plus-defects.
Theorem 5.6. If T E 55-(Sj) is a strict plus-operator and £ is a closed

positive subspace of Sj, then d+(T£) = d+(T) + d+(£).
Proof. Choosing two maximal positive subspaces £1,W(: c Sj such
that £ c £1' T£ c W(: we find:
codimWIT£ = codimW)T£1 + codimTl21T£ .
But the last term is equal to codimE,£, since by Theorem 5.2 T is a
homeomorphic isomorphism between £1 and T£l' 0
Theorem 5.7. If T 1 ,T2 E Q3-(Sj) are strict plus-operators, so is T 1 T 2 •

Moreover, d+(T1 T 2l = d+(T1 ) + d"(T2)'
Proof. From the relations
(T2X, T2X) > /h2 >0 if (x, x) = 1 ,
(TIT2X, T1T2X) > ttl> 0 if (T2X, T2X) = 1
it follows that
(T1T2X, T1T2X);;::: /hl/h2 >0 if (x, x) = 1 .
Hence 1~T2 is a strict plus-operator. Let 3 be a maximal positive

subspace of &). Then, owing to Theorem V.4.1 and Corollary 5·3, T 23
is closed. Consequently, we may apply Theorem 5.6:
d+(T1 T 2l = d~(TIT2Bl = d+(T1 ) + d+(T2B) =
= d+(T1 ) + d+(T2)' D
6. Doubly Strict Plus-operators
A linear operator l' E 58( Sj) is called a doubly strict plus-operator prov-
ided both l' and T* are strict plus-operators. In particular, the
underlying Krein space is required to be indefinite (d. Sec-
tions 4- 5 and II.S).
Example 4.} exhibits a strict plus-operator ,"vhose adjoint is not
even a plus-operator.
The follovving result characterizes doubly strict plus-operators
among strict plus-operators.
Theorem 6.1. The strict pl'Us-operator T E 58(&)) is doubly strict if and

only if d+(1') = O.
Proof. Assume that the value
fJ(T*) = inf (T*x, T*x)
(x,x)=l
is positive. Let 3 c Sj be a maximal positive subspace. Then T2 is

positive and closed (d. Theorem V.4.1 and Corollary 5.}). For y ~ T2
we have T*y ~ 3; hence by Lemma I.6,} (T*y, T*y) ~ 0, so that
in view of Theorem II.S.l
1
(y, y) ~ fJ(T*) (T*y, T*y) ~ 0 .
Thus, on account of Lemma V.4.S, T3 is maximal positive, i.e.,

d+(T) = O.
Let, conversely, d+(1') = o. Suppose that (1'*x, 1'*x) 0 for <
some x E &). Then, owing to Theorems V.5.6 and V.7.1, there exists a
maximal uniformly positive subspace 3 which is orthogonal to T*x.
The relation T*x ~ 3 implies x ~ 1'3. On the other hand, in view
of CorollaryV.7.2, the assumption d+(1') = 0 and Lemma 5.1, the sub-
space T3 is maximal uniformly positive. Thus Lemma I.6.6 yields
(x, x) < O. Hence T* is a plus-operator.
Let us prove that the plus-operator T* is strict. According to
Corollary II.S.} we may confine ourselves to the case where ffi(T*) c q5+
i.e. (see Lemma II.11.4)
(6.1) ffi(T*) c q5+ .
6. Doubly Strict Plus-operators 159
Since m(T) is closed and, by Theorem 5.4, uniformly negative, from

Theorems VL2.6 d) and V.5.2 we obtain:
(6.2) m(T) (+) ffi(T*) = S).
Thus, in view of (6.1) and Lemma 1.6.2, the subspace ffi(T*) is maximal
positive; being ortho-complemented (see (6.2)), Theorem V.5.2 assures
that it is uniformly positive as well.
On account of (6.2) we have:
(6·3) ffi(T) = TS) = Tffi(T*) .
Therefore, owing to the assumption d+(T) = 0 and Lemma 5.1, ffi(T)

is maximal positive and uniformly positive. In particular, ffi(T) is
closed (Theorem V.4.1) and, by Theorem VL2.9, so is ffi(T*).
Since ffi(T) is maximal uniformly positive, Theorems V.7.1 and
VL2.6 c) yield the decomposition
(6.4) ffi( T) (+) m( T*) = S) ,
where m(T*) is maximal uniformly negative. Thus every XES) can be
written in the form
x = x+ + X-; x+Effi(T), x-Em(T*) .
Relying on the uniform positiveness of ffi(T*) (see (6.1)-(6.2)) we
find
(T*x, T*x) = (T*x+, T*x+) > exIIT*x+II],
where J is a fundamental symmetry on S) and ex >
0 is independent
of x. On the other hand, the continuous operator T* Iffi( T) is invertible
(see (6.4)) and has closed range ffi(T*); hence, according to the closed
graph principle and Lemma n.11.4,
IIT*x+IIJ > ,8llx+ll] 2 ,8(x+, x+) > ,8(x, x)
with ,8 > 0 independent of x. As a result,
(T*x, T*x) > ex,8(x, x) (x E S)) .
So T* is a strict plus-operator. 0
From Lemma n.10.2 and Theorems 5.2, 6.1 we derive:
Corollary 6.2. Ina quasi-negative Krein space S) every strict phts-

operator T E )8(S)) is doubly strict. 0
The next result should be compared with Theorem n.S.9.

Theorem 6.3. Let u(Sj) >

O. For an operator T E )8(Sj) the following
assertions a) ~c) are equivalent:
a) TS,W = ~+;
b) T~++ = ~++;
c) SJC(T) is ~tnitormly negative, and TISJC(T)l is a positive number
times an invertible isometric operator with range Sj.
If a), b) or c) holds, then T is a doubly strict plus-operator.
Proof. Suppose that a) is valid. Then, according to Theorem

11.8.6, T is a strict plus-operator. Hence, by Theorem 5.4, SJC(T) is
uniformly negative. In particular, SJC(T) is ortho-complemented (d.
Theorem V.5.2).
Introduce the notations
'" nm+
Ddt ..p ~
- m+
~l'
Obviously, Tl is invertible. We claim that Tl ~t = ~+. The inclusion

Tl~t C ~+ is clear. On the other hand, y E ~+ implies y = Tx for
some x E ~+; decomposing x as
(6.5)
we find the relations
(6.6) Tx = T 1 x1 ,
(6.7) (Xl' Xl) = (X, X) ~ (xo, Xo) > (X, x) > 0 .

Furthermore, Sj1 is indefinite. Namely Sj1 C ~- along with Lemma 1.3.1
+
yields Sj = Sj1( )SJC(T) c ~-, contrary to the assumption x(Sj) O. >
If, in turn, Sj1 C ~+, then ffi( T) = TSj1 C ~+ or, making use of a) once
more, ffi(T) = ~+, which is impossible by Corollary 1.2.7. Applying
Theorem II.8.i0 to Tl we obtain c).
That b) implies c) can be established similarly.
Let c) be satisfied. Then Tl~t = ~+, so that T~+ ) ~+. On the

other hand, for x E ~+ we have (d. (6.5) ~ (6.7)) Tx E Tl~t = ~+.
Thus a) holds. The assertion b) can be deduced from c) in just the same
Inanner.
As already mentioned, every operator T with property a) is a strict

plus-operator. Moreover, from c) it follows that T carries every fun-
damental decomposition of SJC( T) 1 into a fundamental decomposition
of Sj. Hence d+(T) = 0 and, consequently, Tis doubly strict (d. Theo-
rem 6"1). 0
Notes to Chapter VII 161
Notes to Chapter VII
Weighted integral operators which are positive in an indefinite space

Sj have been examined, without explicit use of the indefinite inner
product, by Kre'ln ['1J in 1940. In this special case Theorems 1.2-1.3
are due to him. The general concept of positive operator in a finite-
dimensional Sj belongs to Potapov [1J, and in an arbitrary Krein space
to Ginzburg [2J, who proved Theorem 1.3. Using more efficient tools,
Krein and Smul'jan [2J showed that the eigenspaces appearing in
Theorem 1.2 are uniformly definite. Example 1.5 was proposed by
Langer (personal communication).
Other possible extensions of the concept of positive operator in
Hilbert space, for the special case x+(Sj) < 00, have been investigated
by Bognar [1], [3J - [6]. Theorem 2.1 for a finite-dimensional Sj
belongs to Potapov [1J, and for arbitrary Sj to Bognar and Kramli [1J.
Ginzburg [2J proved that every selfadjoint A E QJ(Sj) with positive
spectrum admits one and only one square root of the same type. An-
other sufficient condition for the existence of a selfadjoint square root
was given by Jalava [3].
Theorem 3.1 is due to Kuhne [1].
Sections 4-6 are mainly based on two papers of KreIn and Smul'jan
[1], [2J. Theorems 5.6-5.7 are borrowed from Smul'jan [2]. Special
cases of Theorem 4.2 have been obtained by Potapov [1J and Ginzburg
[2J; our argument follows the latter author. The present proof of
Theorem 5.4 is due to Iohvidov [14J. Our reasoning in the second half
of the proof of Theorem 6.1 (d. Bognar [9J) seems to have methodical
advantage over the original version; however, we miss the corollary
fl(T*) = fl(T). For dimSj < 00 Corollary 6.2 belongs to Potapov [1].
According to Theorem 6.1, a doubly strict plus-operator T carries
every maximal positive subspace 2 into a maximal positive subspace
T2. Introducing a fundamental decomposition (5.3) of Sj, the corre-
sponding matrix (Til )i,I=1,2 of T, and the angular operators K=K+(2),
([JT(K) = K+(T2), one easily verifies that Tn + T12K is completely
invertible on Sj+ and
Thus ([JT is a linear fractional transformation of the" operator unit ball"

S'e(Sj+,Sj-) to itself. The geometric properties of ([JT for doubly strict
plus-operators as well as for another class of operators T have been
studied in a paper of KreIn and Smul'jan [3J and in subsequent works of
Smul'jan [5J - [7]. In the same context SmuI'jan [4J considered operators
of the form T*T, where T is a doubly strict plus-operator. For a unitary
T Krein and Smul'jan [3] found the general form of the matrix (Til)
(in the special case x(Sj) < co see Iohvidov and Krein [1J). Helton
[2J examined the iterates of CPT'
Conditions for the continuity or closedness of plus-operators in
Banach spaces with a J-metric have been given by Iohvidov [14J.
Operators T E )8(,~) with the property that (Tx, Tx) ~ (x, x)
(resp. (Tx, Tx) :S (x, x)) for every x may be called augmenting (d1'-
minishing) operators.
Evidently, an operator is augmenting in Sj if and only if it is
diminishing in the anti-space of Sj. Furthermore, an operator T E )8(Sj)
is a strict plus-operator if and only if it is a non-zero multiple of an
augmenting operator. Hence many results can equally well be establish-
ed for augmenting (or diminishing) operators and strict plus-operators,
as displayed by the history of the subject. This is, however, not always
the case; for instance, the divisibility conditions obtained by Smul'jan
[1J, [2J are better suited to augmenting operators.
It can be shown that the formula
C = T(D) = (P- - P+D)(P+ - P-D)-l
defines a one-to-one correspondence between a subclass of diminishing
operators DE )8(Sj) and a subclass of operators C E )8(Sj) having the
property llc\l; ~ 1, where J = P+ - P-. The linear fractional trans-
formation r (which has nothing to do with CPT above) has been inves-
tigated by Potapov ['I], Ginzburg [2J, and Iohvidov ['12J, [15].
Analytic functions whose values are diminishing operators in Sj
have first been studied by Potapov ['1J (finite-dimensional Sj) and Ginz-
burg [1J (arbitrary .~). For an idea of how these functions intervene in
the theory of characteristic operator functions see e.g. V. M. Brodskii
['1]. Concerning operator valued analytic functions d. also SmuI'jan
[8l .
Chapter VIII. Invariant Semi-definite Subspaces of
Linear Operators in Krein Spaces
The existence of invariant maximal uniformly definite (Section 1) resp. maximal

semi-definite (Sections 2- 3) subspaces is treated. In Sections 4- 5 an application
to quadratic pencils of Hilbert space operators is given. Section 6 contains the state-
ment of some results concerning the spectral function of a positive or positizable
selfadjoint operator in a Krein space. Theorem 2.1 is the culminating point of the
book. Theorems 1.2, 1.4, 3.1, 3.2 and 5.5 should also be mentioned.
1. Fundamentally Reducible Operators

Let T be a linear operator in the Krein space .s;). If some fundamental
decomposition of &) reduces T, we say T is fundamentally redttcible.
The study of a fundamentally reducible operator is equivalent to
the treatment of two Hilbert space operators.
For simplicity, we shall consider continuous operators only.
Lemma 1.1. If a fundamental decomposition oj &) reduces the operator
T E 58( &)), then the corresponding flmdamental profectors P+, P- and
fundamental symmetry J = P+ - P- commute with T.
Proof. The assumption yields TP+ = P+TP+, TP- = P-TP-.
From the latter equation, SUbstituting P- = 1 - P+, we obtain
P+ T = P+ T P+. Thus P+ T = T P+. It follows that P- and] also
commute with T. 0
Theorem 1.2. The operator T E 58(&)) is jitndamentalZy reducible if

and only if there exists a uniformly positive operator co11t1mding with T.
Proof. If T is fundamentally reducible then, according to Lemma
1.1, there exists a fundamental symmetry .1 that commutes with T. But
] is uniformly positive. Thus the condition is necessary.
Suppose, conversely, that AT = T A, where the operator A E 58(&))
is uniformly positive. Replacing the inner product (. , .) by the A -inner
product (., ')A' the Krein space &) turns into a Hilbert space &)A
whose strong topology coincides with TM(&))' Obviously, A IS a
selfadjoint operator on &)A' In addition, by Lemma VII.3.2, A is
completely invertible: ?JC(A) = 0, ffi(A) = &) = 's;)A .
164 VIII. Invariant Semi-definite Subspaces of Linear Operators
Working in the space SjA' set

(1.1 )
where IAIA denotes the unique A-positive square root of ./12. In view
of Lemma VI.8-3 we have:
( 1.2)
(1-3 ) (x, y)A = 0
Moreover, since the subspaces ('1.1) are invariant under A (d. Lemma
VI.8.1), the relations AISj± = +IAIAISj± and 9c(A) = 0 with the
aid of Lemma VII.1.1 yield:
(1.4) (Ax+, X+)A >0 (x+ E Sj+, x+ =F 0) ,
(1.5) (Ax-, X-)A <0 (x- E Sj-, x- =F 0) .
Returning to the original inner product (. , .) and taking into account
that ASj = Sj implies ASj± = &)±, from (1.3) - (1.5) we obtain:
&)+ 1- &)-, &)+ C ~++, &)- c ~-- .
In other words, (1.2) is a fundamental decomposition of Sj. Finally,
by the assumption and Lemma VI.8.1, this decomposition reduces T. 0
Theorem 1.2 and Theorem VII.3.1 yield:

Corollary 1.3. Let x(&)) >
0 and TE 5B(Sj). If there exists a selfad-
joint operator A E 5B(Sj) commuting with T and satisfying the inequality
(Ax, x) > (X Ilxll} with a fundamental symmetry] and positive number (X
for every x E ~+, then T is fundamentally reducible. 0
Next we apply Theorem 1.2 to unitary operators of &). Loosely

speaking, our result says that a unitary operator similar to a Hilbert
space unitary operator is fundamentally reducible.
Theorem 1.4. The unitary operator U on the Krein space &) is funda-
mentally reducible 1"/ and only if there exist a completely invertible operator
T and a fundamental symmetry I on Sj such that TUT- 1 is I-unitary
(i.e., unitary relative to the I-inner product).
Proof. If U is fundamentally reducible then, by Lemma 1.1, a
certain fundamental symmetry I commutes with U. For this I we
have:
(lUx, Uy) = (U]x, Uy) = (lx, y) (x,y E &)) .
Thus, choosing T = I, the operator TUT- 1 is I-unitary.
Assume, conversely, that the operator
V= TUr 1
2. Invariant Positive Subspaces of Plus-operators 165
is J-unitary, where J denotes a fundamental symmetry and T is com-

pletely invertible. Set
A = T*JT.
Then A E )5-(&)) and, in vie\v of Theorem V1.2.3 c), A':' = A. Moreover,
A is uniformly positive, since T- t is continuous and (Ax, x) = II TxllJ .
Finally, applying Theorem V1.4.2 and Lemma V1.2.1 we obtain:
AU = T*JTU = UU*T*JTU = UT*V*JVT =
= UT*JJV*JVT = UT*JT = UA.
Therefore (see Theorem 1.2) U is fundamentally reducible. 0
2. Invariant Positive Subspaces of Plus-operators
Fundamental reducibility, at least for operators defined throughout &),

means the existence of two, mutually orthogonal, invariant subspaces:
one maximal uniformly positive and one maximal uniformly negative
(ef. Theorem V.7.1).
A weaker, yet useful property can be obtained by omitting the
words "uniformly". In the next section we shall verify this weaker
property for an important subclass of unitary and selfadjoint opera-
tors. The proof will be based on a related result concerning plus-
operators that we are going to treat just now.
Theorem 2.1. Consider a Krein space&) and a plus-operator T E )5-(&)).

Suppose that for some fttndamental decomposition
(2.1) &)=&)+(+)&)-; &)+c~++, &)-c~--
and corresponding fundamental projectors P+, P- the operator P+ T p-

is compact. Then every closed positive subspace 2 c &) with the prope1·ty
T2 = 2 can be extended to a maximal positive subspace 21 c &) such that
T21 c 21 .
Proof. Let 21 be any maximal positive subspace containing 2,
and let 9J(1 be any maximal positive subspace containing T21 (ef.
Section 1.6). Then, in view of Theorem VA.i,
9J(1 = 9J(1 ) T21 ) T2 = 2.
Denoting the angular operators relative to &)+ of 2, 21 and mC l
(ef. Theorems II.i1.7 and VA.2) by K, K1 and NIl, respectively, the
identity operator of the space &)+ by II, the orthogonal projector of
the space &)+ onto P+2 (ef. Lemma IV.7.1) by P, and setting
(2.2) K11(Il - P)&)+ = X, MII(Il - P)&)+ = Y,
we have:
(2·3 ) Kl = KP + X(11 - P) ,
In particular,
(2.4)
and
(2.5) IIKP + X(11 - P)II, < 1, IIKP + Y(11 - P)II, :::; 1 ,
where J = P+ - P- is the fundamental symmetry belonging to the
decomposition (2.1), and Sj+, Sj- are considered as Hilbert spaces
relative to the J-inner product.
Let x+ E S)+. Then, according to Theorems II.11.6 and V.4.2,
x+ + K 1x+ E 21 , On the other hand, if we denote the operator matrix
of T with respect to (2.1) by (T jZ )i,I=I,2' where Til E )8(SjI' Sjj)'
Sjl = Sj+, Sj2 = Sj-, then T(x+ +
K 1x+) can be written in the form
(Tnx+ + T I2 K 1x+) + (T21 X+ + T 22K 1x+) .
Now the assumption T21 C weI yields
T 21x+ + T 22 K 1x+ = Ml(Tnx+ + T I2 K 1x+)
or, x+ being arbitrary,
T21 + T22Kl = M 1(Tn +T 12K 1 ) •
Replacing K1 and 1\11 by their expressions (2.3) we obtain:

+ T 22 (KP + X(11 - P)) -
T21
(2.6)
- (KP + Y(11 - P)){Tn + T I2 (KP + X(11 - pm = o.
Consequently, in order that the maximal positive subspaces 210 9)11
meet the requirements
(2.7)
it is necessary that the operators X, Y defined by (2.2) satisfy the
relations (2.4)-(2.6).
It is easy to see, performing the steps in reverse order, that this
condition is sufficient as well. More exactly, if X and Y obey the
relations (2.4)-(2.6), then they define, via (2.3), maximal positive
subspaces 21> weI
satisfying (2.7).
In addition, 21 is invariant under T if and only if weI
appearing in
(2.7) can be chosen equal to 21> i.e., if in (2.4) - (2.6) the choice Y = X
is possible.
Consider the space
2. Invariant Positive Subspaces of Plus-operators 167
'with the weak operator topology

Two = Two ((II - P)~+, ~-) .
Put
Il{ = {X E )8-]: IIKP + X(II - P)II, < 1},
and define a set-valued mapping <J> on III by the formula
<J>(X) = {Y E 21: (2.6) holds for X, Y} (X E Ill) .
The set Il{ is Two-closed. In fact, if
+ X(II - P)II, > 1 ,
IIKP
then for suitable vectors t E ~+, g E Sj- and a number 8 > 0 we have
I( {KP + X(II - P)} t, g)1 > Iltll,llgll, + 8;
thus every operator X' E )8-1 in the Two-neighbourhood of X character-
ized by the inequality
I((X' - X)(II - P)f, g)1 <8
also lies outside Ill. On the other hand, III is bounded:
(2.8) IIXII, = IIX(I1 - P)ll j < 1 + IIKPII, < 2 (X E Ill) .
Therefore (see e.g. Dunford and Schwartz [1; Exercise VI.9.6]) 21 is
Two-compact. Besides, 21 is easily seen to be convex.
For fixed X the set W(X) is obviously convex and, by the argument
leading to equation (2.6), non-empty.
With the aim of proving the Two-closedness of the set
(2.9) {{X,Y}: X E Ill, Y E W(X)} C )8-1X)8-1'
we will show that the left-hand side of (2.6) is a jointly continuous
function of the variables X, Y E III in the sense of the weak operator
topologies of the domain space )8-1 and the range space )8-(~+,.\>-),
respectively (a Two-continuous function for short).
The only term demanding a closer attention is
Since, by assumption, T12 = P+ T P-I~- is a compact operator, for

every 0 >0 there exists an operator Ro E )8-(~-, ~+) of finite rank
such that IIT12 - Roll] <
0 (see e.g. Riesz and Sz.-Nagy [1; Section
85]). Making use of (2.8) we find:
(X,Y E Ill) .
Hence it is sufficient to prove the Two-continuity of P(X, Y; R),
where R E )8-(~-, ~+) is a fixed operator of finite rank. In order to
establish the latter property of P, write

n
Rx = L(x, vj)zi
j=l
then
n
(P(X, Y; R)f, g) = I (X(II - P)f, vj)(Y(Il - P)zi' g)
j=l
a polynomial of manifestly Two-continuous numerical expressions.

To sum up, 2r is a non-empty compact convex set in the separated
locally convex ~l> the mapping ({J is defined on 2r, its values are non-
empty convex subsets of 2r, and the "graph" (2.9) of ({J is closed. In
these circumstances a theorem independently obtained by Fan U] and
Glicksberg [1J (d. also Berge [1; Sections IX.5 and VI.1J) ensures the
existence of a point Xo E 2r such that Xo E ({J(Xo). Thus for Y = X = Xo
the conditions (2.4) - (2.6) are fulfilled. 0
Remark 2.2. In Theorem 2.1 the choice 2 = 0 is always possible.
3. Invariant Semi-definite Subspaces of Unitary

and Selfadjoint Operators
The invariant subspace theorems, presented below in a weakened form,

rank among the most celebrated achievements of the theory of Krein
spaces.
Theorem 3.1. Let U be a unitary operator in the Krein space ~.
Suppose that for a fundamental decomposition
(3.1) ~ = ~+(+)~-; ~+ C SfS++, ~- C SfS--
and corresponding fundamental projectors P+, P- the operator P+ U p-
is compact. Let 0'0 denote any set of non-unimodular eigenvalues of U
such that v E 0'0 implies v* EE 0'0 •
Then U has an invariant maximal positive subspace 21 and an in-
variant 11'laximal negative subspace 2 2 , each containing all principal S1f,b-
spaces 6 v (U) with l' E 0'0' For every 21 and 22 of this kind and every nor-
mal eigenvalue v E 0'0 zve have the relations 6 •• (U) n21 = 6 v.(U) n22 = o.
One can choose the subspaces 21> 22 to be the orthogonal companions ot
each other.
Proof. Denote the span of the subspaces 6,,(U) (v E 0'0) by 2 0 ,
According to Corollary II.2.6, Theorem II.2.5 and Lemma 1.3.1, 20 is
a neutral subspace. On the other hand, 20 is invariant for U. Moreover,
3. Unitary and Selfadjoint Operators
since the relations 11 E (Jp(U), (U - 1II)'x = 0 imply v =1= 0, x =

= ifv (Ux - Xl)' where Xl = (U - v1)x, (U - 1I1)'-I X1 = 0, by recur-
sion we obtain U®v(U) = ®v(U) (v E ao) i.e. U53 0 = 530 , Applying
now Theorem 2.1 to the operator T = U and the subspace 53 = 53 0 ,
we find a maximal positive subspace 531 which is invariant for U and
contains 530 •
The subspace 53 2 = 53t is maximal negative (Theorem VAA) and
invariant for U (Corollaries V1.4.8 and V1.4.6). Since 530 is a neutral
subspace of the positive subspace 53 1 , Lemma 1.4.4 assures that 530 c 53 2 ,
Theorems V.}.6 and VA.'! yield the relation 53; = 531 .
Let v E ao be a normal eigenvalue of U. As we have just mentioned,
530 ~531' In particular, X E 531 implies x~®v(U). Thus, by Theorem
V1.5.S, X cannot belong to ®•• (U) unless X = O. For 532 the reasoning
is similar. 0
Theorem 3.2. Let A be a selfadjoint operator in the Krein space S).
S#ppose that S)+ C ~(A), where
(3.2) S) = S)+(+)S)-; S)+ C ~++, S)- c ~--
is a s#itable fundamental decomposition of .~.
Then A has at least one non-real regular point.
Further, denoting the ft.tndamental profectors onto S)+ resp. S)- by P+
and P-, setting 11 = IIS)+, 12 = IIS)- and introducing the operator
matrix A = (Ailkl~I,2 by the formulas
An = P+ AIS)+ , AI2 = P+ AIS)- ,
A2I = P-AIS)+ , A22 = P-AIS)-,
slIppose also that for some Co E e(A 22 ) (cf. Theorem V1.2.2 c) and Lemma
V1.6.7 d)) the operator AdA22 - CoI2)-1 1:S compact. Fina0y, let 0'0 be
any set of no'n-real eigenvallIes of A sncll that It E ao implies It ft a o .
Then A has an invariant maximal posit£ve sttbspace 531 c ~(A) and
an invariant maximal negative snbspace 53 2 , each containing all principal
subspaces ®). (A) with It E 0'0' For every 531 and 532 of this kind and every
normal eigenvaltte A. E 0'0 we have the relations ®X(A) n 531 = ®;:(A) n532 = O.
One can choose the sttbspaces 531> 53 2 to be the orthogonal companions of
each other.
Proof. Once a non-real CE e(A) is found, we shall try, with the
help of the Cayley transformation, to reduce the rest of the statement
to Theorem }.1.
Reasoning for the moment heuristically, we have
A _ Ci = (An-Cil
°
0)
A 22 -Ciz
(II
B21
BIZ)
I2 '
where ( E C and
(3·3)
Thus, still formally,
(3.4) (A - (1)-1 =
= ((II-oTI)-l 0 )(
(1 2- TZ)-I
II
-B2I
-B12 ((A
I2) ll -UI)-1
0 (A 22 -U z l '
0)
t
where
(3.5)
-
Let ( i= (. The spectrum of a selfadjoint operator in Hilbert space
being real, from (3-3) and Lemma VI.6.7 we conclude that B2I E
E )S(&)+ ,&)-), whereas BIZ admits a continuous closure BI2 E )S(&)-,&)+).
Since ffi(B 2I ) C 'V(A 22 ) = 'V(A I2 ) = 'V(B I2 ), with the aid of (3.5) it
follows that TI E )S(&)+), while T2 admits a continuous closure
T2 E )S(&)-).
But, in order to ascribe an exact meaning to equation (3.4), we need
more. The elementary identity
IIAnx - (xii} = IIAnx - Re ( . xiiI + (1m C)211xlll (x E &)+) ,
where] = P+ - P-, yields:
(3·6) (C i= C) .
The estimate
(3·7) (C i= C)
can be verified similarly. Now if
(3.8)
-
then by (3.5), (3.3), (3.6) and (3·7) we have IITIII! <1 , IIT211J < 1 ,
so that
As a result, for every Csatisfying condition (3.8) the right-hand side

of (3.4) is a wen-defined linear operator S(C), which is continuous on its
domain. Moreover, an easy calculation shows that
S(C)(A - U)x = x (x E 'V(A)) .
Consequently, Ceither belongs to e(A) or to ar(A). A similar statement
holds for C. Thus, by Theorem VI.6.1 c), CE e(A). In particular, equa-
tion (3.4) is correct.
3. Unitary and Selfadjoint Operators 171
We fix Cin compliance with (3.8) and build the Cayley transform
- -
(3.9) U = (A - U)(A - U)-l = 1+ (C - C)(A - (1)-l .
According to Theorem V1.7.1 U is unitary. Further, from (3A) and

(3.3) we obtain
(3·10) Un = ptUI&)+ = II + (C - C)(I1 - T 1)-1(A ll - U 1)-1 ,
(3. 11 ) U 12 = P+UI&)- =
= - (C - C)(I1 - T 1J-1(A n - U1)-1 A 12 (A 22 - U 2)-1 ;
the matrix elements U 21 , U 22 will not be used in the sequel. The assump-
tion concerning the compactness of AdA22 - CoI 2)-1 for a certain 1;0'
along with the well-known equation
Rco - Rc = (1;0 - C)Rc.Rc
where RJl = (A22 - flI 2J-1 (fl E e(A 22 )), imply the compactness of
A12(A22 - CI2)-1. Hence U 12 is compact.
Owing to Theorem 3.1 there exists a maximal positive subspace
E1 c &) such that UE 1 eEl and @3v(U) eEl (oj! E a~), where
a~ = {~ == f: A E ao} .
Obviously, (U - I)E1 C B1 . Let us prove the stronger fact

(3.12)
In view of Theorem VA.2 the equation (3.12) is equivalent to
P+(U - I)B1 = &)+. Expressing here the left-hand side by the aid of
the angular operator K = K+(B 1l E st'(&)+, &)-) (d. Theorems 11.11.7,
V.4.2 and II.11.6), we find
P+(U - I)B1 = (Un - II + U12K)&)+
or, making use of (3.10) and (3.11),
(3.13) P+(U - I)B1 =
= (1; - 1;)(11 - T 1l-1(A n - U 1)-1[I1 - A12(A22 -U2)-lK]&)+.
But SD(T1) = SD(An) = &)+ and, according to (3.7)-(3.8),

IIAI2(A22 - U 2) -1 KII, < ·1,
so that the range of each multiplier of &)+ in (3.13) coincides with &)+.
This proves (3.12).
Solving equation (3.9) for A, we obtain (d. Lemma 11.4.1):
A = (1;U - U)(U - It1 .
Hence, leaning on the construction of £1 as well as on relation (3.12)

and Lemma II.5.5, we derive the properties £1 C ~(A), A£l c £1 ,
6,,(A) C £1 (J, E 0'0)'
Set £t = £2' Then £2 is maximal negative (Theorem V.4.4), in-
variant for A (Theorem 11.3.7), and its orthogonal companion is £1
(Theorems V.3.6 and V.4.1). lVIoreover, by Corollary II.3.4 each
6,1 (A ) (A E 0'0) is a neutral subspace of the positive subspace £1; thus, in
view of Lemma 1.4.4,
(3·14)
i.e. (5).(A) c £2 (A E 0'0)'
Finally, let A E 0'0 be a normal eigenvalue of A. Then, owing to
Theorem V1.7.5, 6,,(A) # C5J.(A). Comparing with 0.14) yields
6;;(A)n£1 = O. Theproofoftherelation C5;;(A)n£2 = 0 issimilar. 0
4. Quadratic Pencils of Operators in Hilbert Space
Consider the differential equation

d2v I dv
(4.1) -,B-+Cv=O
dt 2 dt '
where t is a real variable, v = v(t) is a function with values in a Hilbert
space SJo, the operator BE 1S(SJo) is selfadjoint, and the operator
C E 1S(SJo) is positive and invertible. The inner product on SJo will be
denoted by (. , ')0'
A solution of the equation (4.1) is called an elementary solution if it
can be written in the form
P-l ti
(4.2) v (t) = eJ.ot}; --;- x p - 1 -i '
)=0 J!
where 1.0 E C, P> 1, {Xi}~-l C SJo, Xo *- o.
A tool in the study of equation (4.1) is the family of operators
(4·3) L(A) = AN + AB + C V' E C) .
A family of this type is called a quadratic pencil.
The number 1.0 E C is said to be an eigenvalue of the pencil L
provided m(L(Ao)) *- O. The set O'p(L) of all eigenvalues of L is called
the point spectrum of L.
We say that the system {xi }~-1 C SJo, where p ~ 1 and Xo *- 0, is a
] ordan chain of the pencil (4.3) to the value 1.0 E C, if it satisfies the
4. Quadratic Pencils of Operators in Hilbert Space 173
conditions
(4.4) {A~1 + AoB + C)xi + (2Ao1 + B)Xi-l + Xi-2 = °
(j = 0,1, ... ,P-1; X-I = X- 2 = 0) .
Setting j = °we see that necessarily),o E O'p{L).
The usefulness of the pencil L in treating the differential equation
(4.1) is partly due to the following circumstance.
Lemma 4.1. A function v(t) of the form (4.2) is an elementary
solution of the equation (4.1) ifand only if x o, Xv ••• , xP_ 1 is a Jordan
chain to the eigenvalueAo of the quadratic pencil (4.3).
The proof is straightforward. 0
Having indicated the connection between the equation (4.1) and

the pencil (4.3), we are going to show that, in turn, the pencil (4.3) is
related to a selfadjoint operator of a suitable Krein space.
Let
(4.5)
where 5)~ stands for the anti-space of 5)0. In other words, 5) is the vector
space of ordered pairs of the form a: = {x+, x-}, where x+, x- run
through the elements of 5)0; the inner product of a: and y = {y+, y-}
is given by the formula
(a:, y) = (x+, y+)o - (x-, y-)o .
Obviously, the subspaces
5)+ = 5)0 X °= {{x+, o}:
(4.6)
5)- = ° X 5)~ = {{a, x-}:
induce a fundamental decomposition of 5):
(4.7) 5) = 5)+(+)5)-; 5)+c $++,
In particular, 5) is a Krein space.
Consider the operator A E ~(5)) whose matrix with respect to (4.7)
looks as follows:
(4.8) A = (0-C 1/2
-B
C1/2) •
One immediately verifies that A is selfadjoint in 5).
Lemma 4.2. If x o, Xl' .•. , x P_ 1 is a Jordan chain of the quadratic

pencil L to the eigenvalue Ao , then the vectors
(4.9) Yi = {C 1~xi' AOXj + xi-d (i = 0,1, ... ,P-1; X-I = °
)
constitzde a Jordan chain of the operator ~4. to the same Ao. Conversely, If
the vectors Yj = {yj, yf} (f = 0,1, ... ,P-1) form a Jordan chain of
A to the eigenvalue Ao , then }'o is non-zero, and the vectors x o, Xl> . . . , x P_ 1
successively obtal'nable from the formula
(4.10) (j = 0,1, ... , P-1; X_I = 0)
constitute a Jordan chain of L to the same }'o' The transformations (4.9)

and (4.10) aI'e mutually inverse.
Proof. From (4.4), (4.9) and (4.8) one easily derives the equations
(4.11) AYj = AOYj + Yj-l (j = 0,1, ... ,P-1; Y-l = 0) .
Since, by assumption, C1/2 is invertible, Xo 0 implies Yo *-
O. Moreover, *-
if {Yi'~-I is defined by (4.9), then {Xj}~-1 can be recovered with the
aid of (4.10).
Suppose, conversely, that (4.11) holds and Yo O. Setting j = 0*-
I/2
and making use of the invertibility of C we obtain that Ao O. If *-
{Xj}~-I denotes the system furnished by (4.10), then the relations (4.9)
and (4.4) can be verified by simultaneous recursion for j. Finally,
(4.9) applied to the case j = 0 yields Xo o. D *-
Corollary 4.3. O'p(L) = O'p(A). D
We say that the number Ao E C belongs to the resolvent set e(L) or to
the spectrum O'(L) of the quadratic pencil L according as L(Ao) is com-
pletely invertible or not. Obviously, O'p(L) C O'(L).
In addition to Corollary 4-3 we have:

Lemma 4.4. O'(L) = O'(A).
Proof. If A does not belong to the set O'p(L) = O'p(A) (d. Corollary
*-
4.3) and A 0, then solving the equation (A - AI)y = g we find:
- ~I + ~CI/2L(}.rIC1/2 -C!/2L(Ar l )
AIr
•
(A - l = ( A A
L(A)-IC1/2 -}.L(Ar l
Therefore (A - AIr
l is defined throughout SJ if and only if L(J,r1 is
defined throughout SJo .
In the case A = 0 we obtain:
5. Quadratic Operator Equations in Hilbert Space -175
Hence :tJ(A-l)=S'j if and only if ~(C-!/2)=S'jO. But the latter equa-

tion is easily seen to be equivalent to ~(C-l)=S'jO' i.e., to the com-
plete invertibility of L(O). 0
Lemmas 4.2 and 4.4 serve as a point of departure for utilizing Krein
space theory in the study of the quadratic pencil (4.3).
5. Quadratic Operator Equations in Hilbert Space
We pursue our investigations concerning the quadratic pencil

(5.1 ) L(A) =},.2] +AB +C (A E C)
of operators in the Hilbert space S'jo with inner product (. , .)0' still
assuming that B E )8-(S'jo) is selfadjoint, while C E )8-(S'jo) is positive and
invertible.
Writing
(5.2) L(Z) = Z2 + BZ + C
we shall exhibit a relationship between the pencil (5.1) and the roots of
the operator equation L(Z) = 0 .
Let us see first a few general properties of the equation L(Z) = O.
Lemma 5.1. If L(Zl) = 0 for some Zl E )8-(S'jo), then setting Z2 =
= - B - zt we have L(Z2) = o. Moreover,
(5.3) L(A) = (AI - Z~)(AI - Zl) (A E C) .
The proof is simple and left to the reader. o
Lemma 5.2. Let Zo E )8-(S.)o) satisfy the eqltation
(5.4) Z~ + BZo + C = 0.
Suppose there exists a set ero of non-real eigenvalues of Zo with the proper-
ties: a) }, E ero implies A E! ero; b) the span of the principal subspaces ®;.(Zo)
(A E ero) is dense in S'jo. Then
(5.5) Zo + Z: = -B, zt Zo = C.
Proof. We define
(5.6) Bl = - Zo - zt , C1 = ztzo ,
and verify immediately that
(5.7) Z~ + BIZO + C 1 = O.
From (5.4) we obtain
Z~ - zt 2 = zt B - BZo ,
whereas (5.7) yields
Thus for
D = Bl - B
we have Z~'kD = DZo , so that Zo is symmetric relative to the D-inner
product
(x, Y)D = (Dx, Y)o (x,y E Sjo) .
Therefore, in view of assumption a) and Theorem II.3.3, the D-inner
product is identically zero on the span of the subspaces 6,.(Zo) (A E 0"0)'
Hence, by assumption: b) and the continuity of the D-inner product,
(x, Y)D = ° (x,y E Sjo). The positive definite inner product (. , ')0 being
non-degenerate, D must be zero. Thus Bl = B. The relation C1 = C
now easily follows from (5.4) and (5.7). D
The next result connects the quadratic equation L(Z) = with the °
operator r1 defined on the Krein space Sj (d. (4.5) - (4.8)).
Lemma 5.3. Let ,S3 be a subspace of the Krein space Sj = SjoxSj~,
where Sj~ stands for the anti-space of Sjo, Suppose that ,S3 has an angular
operator K = K+(,S3) E 18(Sj+, SJ-) with respect to Sj+ (see (4.6) - (4.7)).
The s2tbspace ,S3 is invariant for .11 (cf. (4.8)) 1/ and only 1:f the operator
Zo = KC1f2 satisfies the eq2£ation L(Zo) = 0.
Proof. Sj+ and Sj- being isometrically isomorphic to Sjo resp. its
anti-space, we may regard K as an element of 18(Sjo) and write
,S3 = {{x+, Kx+}: x+ E Sjo} .
Therefore, in view of (4.8),
A,S3 = {{C 1[2Kx+, - C1/2x+ - BKx+}: x+ E Sjo} .
Consequently, the relation A.53 C .53 holds if and only if
(5.8)
Since all operators involved are continuous andm(C1/2) = 9((C1/2) 1 = Sjo ,
(5.8) is equivalent to
KC 1[2KC 1[2 y+ + BKC1/2y+ + Cy+ = 0
and this was to be proved. D
As to the interdependence between the quadratic pencil L and the
roots of the operator equation L(Z) = 0, the simplest fact is the follow-
ing.
Lemma 5.4. If Zo E 18(Sjo) satisfies the equation L(Zo) = 0, then every
Jordan chain of Zo is a Jordan chain of the quadratic pencil L to the
same eigenvalue. In particular, O"p(Zo) C O"p(L).
s. Quadratic Operator Equations in Hilbert Space 177
Proof. Let {xi}e- I be a Jordan chain of Zo to AD· Expressing AOxi

and A~xi from the equations
Z~X, = ;,~xi + 2Aoxj_1 + Xi--2
(f = 0,1, ... , p-1; X-I = x- 2 = 0) ,
respectively, we obtain:
(A~I + l'oB + C)Xj + (2/'° 1 + B)xi_1 + Xi-2 = L(Zo) xi = 0. 0
We are now coming to the main result of this section. It can be
looked upon as a partial converse of Lemma 5.4.
Theorem 5.5. Consider the quadratic pencil

LV,) =}N +AB +C (A E C)
of operators in the Hilbert space &)0' and the qt~adratic operator ftmction
L (Z) = Z2 + BZ + C
Besides the conditions B E ~(&)o) selfadfoint, C E ~(&)o) positive and
invertible, i'mposed on Band C so far, assume that C is a compact opera-
tor. Let eTo be a set of non-real eigenvalues of the pmcil L with A E eTo
implying}. El eTo .
Then there exists an operator Zo E ~(&)o) such that a) L(Zo) = 0,
b) Z;;Zo < C, and c) any Jordan chain of L to an eigenvalue .1.0 E eTo is a
Jordan chain of Zo to the same Ao .
If the span of all Jordan chains of L corresponding to points of eTo
is dense in &)0' then requirement c) completely determines Zo; in this case
we have Zo + zt = -B, ztzo
= C.
Proof. We shall make use of the selfadjoint operator A acting in
the Krein space &) (see the relations (4.5) - (4.8)). Owing to Corollary
4.3, eTo is a set of non-real eigenvalues of A such that A E eTo implies
~ El eTo · F mther, denoting the norm on &)0 by II . 110 we have: II C1/2xll~ :::;:;
< IICxll o Ilxll o (x E &)0)' Therefore if {xi}~ c&)o is bounded and {Cxi}~
is convergent then {C1/2Xj}~ is convergent. Thus, along with C, also
CI/2 is compact.
Consequently, Theorem 3.2 applies to A and eTo: there exists a
maximal positive subspace 2 c &) that is invariant under A and con-
tains all principal subspaces ®;.(A) V, E eTo).
In view of Theorems II.H.7 and V.4.2 2 has an angular operator
K E ~(&)+, &)-) with respect to &)+ (d. the definitions (4.6)), andidenti-
fying K with an element of ~(&)o) we have IIKllo < 1. For Zo = KC 1/2
Lemma 5.3 yields L(Zo) = 0, whereas the inequality Z:Zo < C follows
from IIZoxii o :::;:; IIC1/2xll o (x E &)0)'
Let {Xj }~-1 be a Jordan chain of the pencil L corresponding to an

eigenvalue 1'0 E a o . Then, on account of Lemma 4.2, the vectors
(f = 0,1, ... ,P-1; X-I = 0)
form a Jordan chain of A to the same 11.0 ' Therefore, by the construction
of 53, each Yi is contained in 53. Recalling the definition of K we conclude
that
+
KC1f2xj = l'oX j Xj_l (f = 0,1, ... ,P-1; X-I = 0) .
Hence {Xi}~-l is a Jordan chain of Zo = KC 1/2 to the eigenvalue 11.0 '
If the span of all Jordan chains of L corresponding to points of ao
is dense in Sjo, then condition c) completely determines the values of
Zo on this dense subset and, by continuity, on the whole space. The
rest of the statement follows from Lemma 5.2. 0
6. Spectral Functions
The present-day theory of spectral functions in Krein space makes

extensive use of analytic methods. Therefore we will just try to give
the ideas by stating some results and neglecting the proofs.
Let Sj be a Krein space and let A E ~(Sj) be a positive operator. There
exists one and only one function E that is defined on the set of all non-
zero real numbers A and satisfies the following conditions:
1) E(A) is an orthogonal projector in .'0;
2) E(Al)E(A2) = E(min{l'l' A2 });
<
3) E(A)Sj is uniformly negative if A 0 ,
(I - E(A))Sj is uniformly positive if A> 0;
4) E(A) = 0 if I, is suflic1;ently stnaU,
E (A) = I if A is sufficiently large;
5) E(A - 0) = E(A);
6) AE(A) = E(A)A;
7) a(AJE(A)Sj) C (-oo,AJ ,
a(AJ(I - E(A))Sj) C [11.,(0) .
(Conditions 1), 2), 5), 6), 7) are required to hold for aU admissible, i.e.
non-zero real, values of A, Al and 1'2 .)
E is called the spectral function of the positive operator A. Taking
.c(Sj) = 0 we see that this definition is consistent with Hilbert space
terminology.
According to conditions 1), 6) and 2), the values of the function
E(A)Sj (A E R; A i= 0) are ortho-complemented invariant subspaces
of A, and the function itself is monotone non-decreasing.
6. Spectral Functions 179
From 1)-3) it follows that the strong limits E (A-O), E (}.+O)

exist for every A -j- O. Thus condition 5) means just a nonning of E.
From 1), 2), 6), 7) we easily obtain:
8) A real number }, belongs to a(A) if and only if E(?'2) -E(Al) 0 *"
whenever Al A A2 .< <
Two further properties of the spectral function:
9) n
(E(A 2 ) - E(A1))S) = ®o(A)
},,<O<}.,
(the principal subspace of A corresponding to A = 0) ;
+ J A dEV,) , where
<Xl
10) A = 5 5 is a positive operator such that

-co
52 = 0, 5(E(A2) - EV'l)) = 0 if Al <A2 <
0 or 0 <AI <}'2' and the
integral exists in the strong topology as an improper integral with
singular point A = o.
A few consequences of 9) and 10):
< A2 < 0 or 0 < }'1 < A2;
)':l
11) A (E(?'2) - E(A1 )) = JA dEV,) if }'l

00 Al
12) An = I A"dE(A) (n = 2,3, ... );
-00
13) AI®o(A) = 51®o(A);

14) A2®o(A) = O.
On account of 14) the subspace ®o(A) is closed, and the Jordan
chains of A to the value A = 0 consist of no more than two elements.
It can also be proved that
15) ®o(A) is ortho-complemented if and only if both of the strong
limits E(-O), E(+O) exist; in this case ®o(A) = (E(+O) - E(-O))S)
and 5 = A(E(+O) - E(-O)).
A real A is said to be a critical point of the spectral function E if
the subspace (E(A2) - E(A 1)) S) is indefinite for every ,11' A2 satisfying
Al <A <A2·
In view of 3), the only possible critical point of the spectral function
of a positive operator is O. Also, by property 8), in order that ,1= 0
actually be a critical point of E it is necessary that 0 E a(A).
Suppose 0 is not a critical point of E. Then for some Al 0 and <
A2> 0 the subspace (E(A2) - E(?'l)) S) is semi-definite, say positive.
Being ortho-complemented, it is uniformly positive. It follows that
E( -0) and E( +0) exist. Moreover, 3) yields E( -0) = E(A1)' Further,
owing to 15), 14) and the uniqueness of the positive square root in
Hilbert space, ®o(A) is ortho-complemented and 5 = o.
If the point A = 0 is non-critical, we agree to extend the function E
setting E(O) = E( -0). This convention preserves the properties 1), 2),
5), 6), 7).
Let Sj still denote a Krein space. We say the selfadjoint operator

A E )8-(,Sj) is positz:zable, if there exists a non-constant, real polynomial rp
such that rp(A) is a positive operator; rp will be called a positizing
polynomial for A.
Positive operators A E )8-(,Sj) are positizable by the polynomial
qe(},) = k
If &) is a Hilbert space, then every selfadjoint A E )8-(,Sj) is positiz-
able by the polynomial rp(A) = },2.
The non-real spectrum of a positizable operator A can be proved
"*
to consist of a finite number of normal eigenvalnes. It follows that the
span {5 of the principal subspaces (5,JA) (1m A 0) is ortho-comple-
mented and A restricted to lSi is a positizable operator with real
spectrum.
Let A E )8-(&)) be a positizable operator with real spectnml. There
exists a unique function E defined in all b'ut a finite mtmber of points
of the real line and satisfying conditions 1)-2) and 4)-7) above, as
well as the following ones:
3 a) the critical points of E form a S11bset of the finite set {it E a(A) :
rp(A) = O}, where rp is a positizing polynomial for A;
3b) E(A) is undefined if and only zf A is a critical point of E;
3 c) zf A is a critical point of E, then for every closed interval [}'v A2J
lying in a sufficiently slnall right-hand or left-hand neighbMtrhood of A
the subspace (E(A2) - E(A 1 ))&) is definite.
E is called the spectral function of the positizable operator A.
Property 8) and the analogue of 9) remain valid in the positizable
case. Instead of 10) we have:
co
10a) rp(A) = S + Jrp(A) dE(A) ,
-00
where S is a positive operator such that S2 = 0, S(E(A2l - E(A 1 )) = 0

if the closed interval [Al> A2J does not contain critical points, and the
integral exists in the strong topology as an improper integral with
singularities at the critical points.
Notes to Chapter VIII

The first achievement in the field of fundamentally' reducible operators
belongs to Pesonen [1]. He proved an equivalent of Corollary 1.3 for
T = A in a separable Krein space. Using a simpler method, Kuhne [1J
extended Pesonen's theorem to selfadjoint operators T commuting
with A in a possibly non-separable &). Hess [1J clarified the meaning
of these results and obtained the corresponding fact for closed opera-
tors T. In the same context, J alava [1J considered closed operators T 1 ,
T2 that are A-adjoints of each other for some uniformly positive opera-
tor A E Q3-(S)). In essence, our treatment follows Hess.
Theorem 1.4 coupled with a theorem of Sz.-Nagy ['I] says that the
unitary operator U in the Krein space S) is fundamentally reducible if
and only if the sequence {II unll! }:=1 is bounded. This proposition,
and its extension to commutative groups of unitary operators, have
been proved by Phillips [3J (d. also Krein [6J, [8J, Daleckil and
KreIn [1J). The connection between Theorems -1.2 and 1.4 was pointed
out by Hess [lJ.
A related type of theorem give conditions on an operator T for the
existence of a reducing decomposition S) = S)l + S)2 where S)1 and S)2
are maximal uniformly definite (not necessarily orthogonal) subspaces
with the additional property that O'(TIS)l) is disjoint from 0'(TIS)2)'
These results and their applications to stability theory are largely
due to Krein. A detailed account and a bibliography of the subject
can be found in the book of Daleckil and KreIn ['I]. Cf. also Krein [6J,
[8J for earlier versions as well as Marksjo and Textorius [1J, Langer [9J,
Svarcman [1J for special topics.
Theorems stating the existence of invariant maximal positive (semi-
definite) subspaces for various kinds of operators in S) (d. Sections
2-3) have been proved, in chronological order, by the following authors:
Pontrjagin [1J (x(S)) < 00, selfadjoint operators, analytic method),
KreIn (in the paper of KreIn and Rutman [1J; x(.S)) = 1, unitary opera-
tors, using the theory of cones in Banach spaces), Iohvidov [lJ (x(S)) <
< 00, unitary operators, from Pontrjagin's result by means of the
Cayley transformations), KreIn [4J (x(S)) < 00, strict plus-operators,
by means of the Brouwer fixed point principle; in a paper of Iohvidov
and KreIn [1J reproved by the aid of the Schauder fixed point principle),
Langer [1J-[3] (x(S)) = 00, separable S), selfadjoint operators A with
compact P+ A P-, by means of the Tihonov fixed point principle),
Krein [7J (arbitrary ~; unitary, selfadjoint and certain plus-operators T
with compact P+TP-), Iohvidov [9J (arbitrary S), arbitrary plus-
operators T E Q3-(.S)) with compact P+TP-, by means of a "fixed point
theorem for set-valued mappings"), Langer [5J, [12J (arbitrary S),
positizable operators, by means of the spectral function). For in-
variant subspaces of selfadjoint, quasi-nilpotent operators d. Lario-
nov [4J.
Quite recently Masuda [lJ has attacked the invariant subspace
problem with the help of an ad hoc iteration procedure rather than
a fixed point theorem. In this way he obtained a condition ensuring
the existence of a unique invariant maximal positive subspace for
a unitary operator U on S) in terms of the numerical ranges of
Ull + U12K and U 22 - KU12 , where K runs through st(S)+, S)-).
In contrast to positizability, as pointed out by Noel [2], the com-

pactness condition appearing in Theorems 2.1, 3.1 and 3.2 is not in-
variant against the choice of the fundamental decomposition. We have
still decided in favour of these theorems because they automatically
furnish most of the results known for the case ,,(&)) <CXJ (see Chap-
ter IX), and because their proof, as compared to that available for
positizable operators, does not make use of the analytic machinery of
spectral functions.
The present form of Theorem 2.1 was obtained by Langer [5] and
Wittstock [4], independently of each other. In the proof we follow the
unpublished lecture notes of Langer.
If U is unitary in &) and P+ U P- is compact, then the perturbation
theory of Hilbert space operators yields that every non-unimodular
point of O"(U) is a normal eigenvalue. Taking the set 0"0 of Theorem 3.1
to be maximal, it follows that the non-unimodular spectra of UI21
and UI2 2 coincide with 0"0' Moreover, it can be proved that if one
merely requires the latter property (still with maximal 0"0) from the
invariant maximal positive 21 and maximal negative 2 2 , then they
always contain the principal subspaces ®.(U) (11 E 0"0)' Similar state-
ments hold for selfadjoint operators. Theorems 3.1-3.2 have origi-
nally been proved in this form; see Langer [2], [3] and KreIn [7], [8].
In particular, the compactness condition appearing in Theorem 3.2
and the present proof of this theorem belong to KreIn [7J.
For certain commuting families of unitary or selfadjoint operators
the existence of a common invariant maximal positive subspace was
proved by Nalmark [1], Langer [5], [12], and Helton [1] (d. also
Larionov [1], Helton [2J). The simplest case will be treated in Chap-
ter IX.
A strengthening of the invariant subspace theorems asserts that,
for certain unitary and selfadjoint operators or commuting families of
them, every alternating pair of invariant subspaces can be extended
to an alternating maximal pair of invariant subspaces. Results of this
kind have been obtained by Phillips [3], Langer [5], [12], and Lario-
nov [2J, [3] (d. also Langer [11J).
For invariant subspaces of operators in Banach spaces with a
J-metric see Bonsall [1J, M.L. Brodskil [1], Iohvidov[9], Langer [10], and
Hackevic [1]. For related theorems concerning locally convex spaces
d. Fan [2]-[5J.
The material of Sections 4-5, partly in a stronger form, belongs to
KreIn and Langer [2], [3] (d. also the lecture notes of Kreln [8]). From
the many other facts proved by KreIn and Langer [3] we mention the
following.
If the quadratic pencil L satisfies the conditions of Theorem 5.5

and, in addition, L is strongly damped, i.e.,
(Ex, x) > 2 V(Cx, x)(x, x) (x E Sj; x 01= 0) ,
then a full converse of Lemma 5.4 holds; namely the equation L(Z) = 0
has two roots Zl' Z2 such that every principal vector of L is an eigen-
vector either of Zl or of Z2'
Langer ([5J resp. [8J) extended the latter result to the case of
a non-compact C and to pencils where the coefficient of },2 is different
from I. For related investigations see Kuhne [2J, Larionov [5J, and
Eisenfeld [1].
Kuhne [1J proved the completeness of the system of eigenvectors of
a power-compact Pesonen operator in Krein space. A special case due
to KreIn [1J and an extension by Harazov [1J were obtained without
explicit use of the indefinite inner product. Phillips [4J has found the
minimax characterization of the eigenvalues of a compact, strictly
positive operator in Sj (for the finite-dimensional case d. also Svarc-
man [1]); Lax and Phillips C3J applied this result to a scattering
problem. For other questions concerning compact operators see Ioh-
vidov [7J, Wendland [1].
The theory of the spectral function for selfadjoint operators in
a Krein space Sj with ;Y,(Sj) <co is due to KreIn and Langer [1]. In
Langer's dissertation this was generalized to positizable selfadjoint
and unitary operators of an arbitrary Krein space, and in a paper of
Langer [6J another extension to a class of continuous selfadjoint opera-
tors A with compact P+AP- was given. Jonas [1J proved the exist-
ence of a spectral function for unitary operators satisfying a some-
what complicated condition weaker than positizability.
A complete account of the spectral theory of positizable operators
has not appeared in print. For a short treatment of continuous positive
operators see KreIn and Smul'jan [2J. A formulation of the main results
for continuous positizable selfadjoint operators can be found in two
papers of Langer [8J, [12]. For an application of the theory see Lan-
ger [15].
Making use of the spectral function, KreIn and Smul'jan [2J estab-
lished a polar decomposition for a class of strict plus-operators.
Special cases have earlier been obtained by Potapov [1J and Ginz-
burg [2J.
In connection with the spectral theory of Krein spaces d. also
Langer [4J, [13J and NaImark [12].
Chapter IX. Pontrjagin Spaces
and Their Linear Operators
Those properties of Pontrjagin spaces (Krein spaces with a finite rank of indefinite-
ness) and of their linear operators are considered which are not available in arbi-
trary Krein spaces or can there only be established by more delicate methods. In
particular, Sections 4- 5 give estimates for the number of non-unimodular (resp.
non-real) or non-semi-simple eigenvalnes of an isometric (resp. symmetric) operator,
while in Sections 7-9 the investigation of invariant maximal semi-definite sub-
spaces is carried on. Theorems 1.4, 2.2, 2.5, 3.2,4.3, 7.3, 8.2, 9.3, 9.4 and Corollary
8.5 are worth special mentioning.
1. The Spaces IJ,. . Positive Subspaces
Krein spaces with a finite rank of indefiniteness are called Pontrjagin

spaces.
In particular, every Pontrjagin space is a Krein space, every Hilbert
space is a Pontrjagin space, and every finite-dimensional non-degen-
erate inner product space is also a Pontrjagin space.
A Krein space with a finite rank of positivity k (k = 0,1,2, ... )
will be denoted by Ilk' Thus y,,+(Ilkl = k <
00, whereas y,,-(Ilk ) is
arbitrary.
Obviously, every II" is a Pontrjagin space and every Pontrjagin
space is either a Ilk or the anti-space of a Ilk' so that in the study of
Pontrjagin spaces we may restrict our attention to spaces of the
type Ilk'
Leaning on the foregoing chapters it is not difficult to establish the
main features of the geometry of Ilk'
From Theorem II.10.1 we obtain:
Lemma 1.1. If 53 is a positive s~tbspace of Ilk' then dim53 ~ k. 0
Furthermore, Theorems II.10.1 and V.4.2 yield:
Lemma 1.2. The positive subspace 53 c II k is maximal positive if and
only if dim53 = k. 0
Hence with the aid of Lemmas 1.11.4 and 1.9.8 we derive:
1. The Spaces Ilk' Positive Subspaces 185
Lemma 1.3. If 2 -is a k-d-inzensional positive definite subspace of Ilk'

then there exists a fundamental decomposition
(1.1) Ilk = Il+(+)Il-; II+cl.lS++, Il- c 1.lS--
such that Il+ = 2. 0
In Theorem VrrI.3.2 among other things we have required that
a certain dense subspace of Sj would contain one component of a funda-
mental decomposition. In a Pontrjagin space this condition becomes
superfluous.
Theorem 1.4. Every dense s2tbspace of Ilk contains a k-dirnensional
positive definite subspace.
Proof. The case k = 0 is trivial. So let k ~ 1, and let 2 be a dense
subspace of Ilk' Consider a fundamental decomposition (1.1) and the
corresponding fundamental symmetry]. Fix an orthonormal basis
{e i }:=1 of II+. Choose a positive number c and a system {li}~ C 2
such that
(j = 1, ... , k) .
Let
k k k
X = l,' iXiei ' y = 2,' iX/I ; l,'liXil2 = 1 .
j=1 i=1 i=1
Then
k
Ily - xllJ :S L' liXilllfi - eill J < kc,
i=1 .
k
IlxllJ = 1, IlyllJ :S 2.' liXilllfillJ
j=1
< k(1 + c) ,
and therefore
I(y, y) - 11 = I(y, y) - (x, x)1 :S I()', ),-x)1 + I()'-x, x)1 <
:::;; 11)'!11 II)' - xiiI + IlxllJ II)' - xiiI < (1 + k + kc)kc .
Thus for c < _1_ we have ()" y) > O.
6k 2
Next let
k k
z = 2,.' P/i; LIPi l2 = p2 > 0 .
1=1 i=1
Applying the previous conclusion to )' = ~z we find that (z, z) > O.

Consequently, the vectors f1> ... , fk are linearly independent and their
span is a positive definite subspace of 2. 0
186 IX. Pontrjagin Spaces and Their Linear Operators
2. Closed Subspaces
The following agreeable property of Ilk has already been proved as a
part of Theorem V.6.3.
Lemma 2.1. Every closed definite subspace of Ilk is uniformly def-
inite. 0
The next result, a consequence of Lemma 2.1, plays a decisive role
in the theory of Pontrjagin spaces.
Theorem 2.2. Every closed, non-degenerate subspace 53 c Ilk is ortho-
complemented.
Proof. By Theorem V.3.1 and Lemma 1.11.153 is the orthogonal
direct sum of a closed positive definite and a closed negative definite
subspace. It remains to apply Lemma 2.1 and Theorem V.5·3· 0
From Theorem 2.2 with the help of Theorem V.3.4 and Lemma 1.1
we obtain:
Corollary 2.3. Every closed, non-degenerate subspace of Ilk is a space
of type Ilk' for some k' < k. 0
Example V.6.S shows that a type Ilk' subspace of II", where h' k,<
need not be closed. Nevertheless, for k' = k the following converse
of Corollary 2.3 holds.
Theorem 2.4. If Il(,,) c Ilk are two Pontrjagin spaces with the same
finite rank of positivity k, then Il(k) is closed in Ilk'
Proof. Let
Il(k) = Il(+) (+) IlH; Il(+) C ~++, IlH C ~--
be a fundamental decomposition of Il(k)' In view of Lemma 1.3 the
subspace Il(+) gives rise to a fundamental decomposition of Ilk as well:
II" = Il(+)(+)Il-.
SinceIlH is an intrinsically complete subspace of the negative definite,
intrinsically complete space Il-, the theory of Hilbert spaces assures
that Il(-) is ortho-complemented in Il-. Consequently, Il(k) is ortho-
complemented in II" . Thus by Corollary III.2.S Il(,,) is closed in II" . 0
For possibly degenerate subspaces we have the following substitute

of Theorem 2.2.
Theorem 2.5. Let 53 be a closed subspace of Ilk' Setting$3n53 1 = 53°,
the space Ilk can be decomposed into a direct sttm ot the form
(2.1 )
2. Closed Subspaces 187
where
(2.2)
(2·3) £2 = £2 , £O( + )£2 = £1 ,
(2.4) £a # £0.
If £1 and £2 are prescribed subspaces satisfying (2.2) and (2.3), then apart
from the conditiol'fs (2.4) al1.d
(2.5) £3 c (£1(+)~)1
the subspace £3 may be chosen arbitrarily. On the other hand, for any
fixed £a satisfying (2.4) one and only one choice of £1 and £2 is possible:
(2.6) £1 = £n£;, £2 = £In£; .
Proof. In order to prove the existence of £3 for £1 and £2 given, let

£1' £2 be two subspaces of Ilk satisfying the conditions (2.2)-{2·3)·
(That (2.2) and (2.3) can at all be fulfilled follows from Lemma IV.8.7,
Lemma II1.2.4 and Corollary V.3.7.) In view of Theorem 1.5.4 £1 and
£2 are non-degenerate, hence, by Theorem 2.2, ortho-complemented.
But the relations £1 C £, £2 C £1 imply £1 J.. £2' Therefore £1 (+ ) £2
exists and, owing to Lemma 1.9.2, is ortho-complemented.
Lemma1.9.1 and Corollary 1.9.5 ensure that the subspace (.Bt (+ )~) 1
is non-degenerate, and relations (2.2)-(2.3) show that it contains £0.
Moreover, applying Lemma 1.1 to the neutral subspace £0 we obtain:
(2.7) dim £0 < k.
Thus, according to Lemma 1.10.7, £0 has a dual companion £a in the
space (.Bt (+ ) £2) 1 .
On account of Lemma 1.10.1, relation (2.7), Lemma 1.10.3 and
Corollary 1.11.9 the direct sum £0+£3 exists and is ortho-complement-
ed. By Lemma 1.9.2 this implies the ortho-complementedness of the
subspace
E we 1. Then, in particular, x E (£0 (+ ).Bt) 1 = £1 and

Let x
x (£0(+)£2)1 = £11 = £ (d. Theorem V.3.6), so that x E £n£1 =
E
= £0. On the other hand, x E £;. Consequently, x E £on£;. Making
use of (2.4) we obtain x = O. Thus (2.1) holds.
In order to prove the existence of £1 and £2 for fixed £3' let £3 denote
any dual companion of £0 in Ilk' (That such an £a can actually be found
follows from (2.7) and Lemma 1.10.7.) Set
£1 = £n£;.
Then Lemma III.2.4, the definition of 533 and the relation 5301-53, re-
spectively, yield 531 =53v 2°n21 =0 and 2°-121 , Moreover, 2°+21 =
= 2, since 2° + 2; =ilk (ef. Lemma 1.10.8) and 2° ( 5]. Thus (2.2)
is valid. In a similar \vay we obtain that the subspace 5]2 = 53 1 n 2t
satisfies (2.3). Relation (2.1) is now a consequence of that part of the
theorem we have already established.
Finally, in order to see the uniqueness of 21 and 2 2 , let 5:s be fixed,
23 =1=1= 2°, and let 21> 22 be subs paces satisfying (2.1), (2.2) and
(2.3)· Then necessarily 21 ( 2n2;, 22 c 21n2t. As, however,
2 n2:l- and 21 n2; themselves meet all requirements involving 21 and
22 (d. the preceding paragraph), their proper subspaces cannot. This
proves (2.6). 0
Besides using decompositions of the form (2.1), closed degenerate

subspaces of ilk can be treated by passing to quotient spaces.
Theorem 2.6. Ij 2 is a closed subspace oj ilk' then 2/53° is a space ilk'
for some k' :::; k.
Proof. Lemma IV.8.J supplies a closed, non-degenerate subspace 21
such that 2 = 2°+21. By Corollary 2.3 531 is of type ilk' for some
k' :::; k. But, according to Lemma 1.5.2, 21 is isometrically isomorphic
to 53/2°. 0
3. Isometric Operators: Continuity
Criteria for the continuity of isometric operators in a Krein space have

been discussed in Section V1.3. The following sufficient condition is
specific to Pontrjagin spaces (ef. Example V1.3.J).
---.!.heorem 3.1. Let U be an isometric operator in ilk' Ij the closures
SD(U) and m(U) are non-degenerate, then U and U-l are continuous.
Proof. Since the inner product is continuous, along with SD(U) also
SD(U) is non-degenerate. Moreover, SD(U) is quasi-negative (ef. Lem-
ma 1.1). Let SD+ be a maximal positive definite subspace of SD(U).
Then, owing to Lemma 1.11.4, Lemma I.9.8 and Corollary 1.11.2,
there exists a subspace SD- such that
(3.1) SD(U)=SD+(+)SD-; SD+cl,jS++, SD-cl,jS--.
Here SD+ is finite-dimensional and definite, hence uniformly definite.
Let us show that SD-, too, is uniformly definite.
By Corollary 2.} the closure SD(U) is a space of type ilk" where
k' :::; k. The subspace' SD(U) is dense in this space, since iM(ilk,) =
4. Isometric and Symmetric Operators 189
= TM(IIk)!IIk , (d. Theorem 2.2 and Theorem V.3.S). Thus, according

to Theorem 1.4 and Lemma II.10.2, dim'Il+ = k'. Therefore Lemma 1.3
guarantees the existence of a subspace 'Il(-) c 1,:J.5-- such that
(3.2)
The subspace 'Il(U) being ortho-complemented (Theorem 2.2), so is
'IlH (Lemma 1.9.2). Consequently, 'IlH is uniformly definite (Theo-
rem V.S.2). But (3.1) and (3.2) imply 'Il- c 'Il H . As a result, ;tl- is
uniformly definite.
From (3.1) and the isometric property of U we obtain:
ffi(U) = U'Il+ (+) U'Il-; U'Il+ C 1,:J.5++, U'Il- c 1,:J.5-- .
Making use of the non-degeneracy of ffi( U) and repeating the argument
just applied to 'Il(U), it follows that U'Il+ and U'Il- are uniformly
definite. Now Theorem V1.3.5 and Corollary II.2.2 yield the desired
conclusion. 0
Theorem 3.2. Let U be an isometric operator in II" with 'Il(U) = Ilk'

Then U and U-l are contimtous.
Proof. The image UII+ of a k-dimensional positive definite sub-
space II+ C Ilk is k-dimensional (Corollary II.2.2) and positive definite.
Thus ffi(U) and, all the more, ffi(U) contain a maximal positive definite
subspace of II k' It remains to apply LemmaII.10.S and Theorem 3·1. 0
4. Isometric and Symmetric Operators: Number and Length

of Jordan Chains
As indicated in Chapter II (d. Examples II.2.3-II.2.4), an isometric

operator in an indefinite inner product space may have non-unimodular
or non-semi-simple eigenvalues too. It turns out, however, that in a
Pontrjagin space the number of these eigenvalues and the lengths of
the respective Jordan chains do not exceed certain limits.
Before stating the general result, let us mention an easy special case
where non-unimodular eigenvalues are only considered.
Theorem 4.1. Let U be an isometric operator in Ilk' Let VI' ••• , V"
be a set of non-unimodular eigenvalues of U such that v; =/:- VI (j,l =
= -I, ... ,n). Then
(4.1) "
}; dimiS"j(U) < k .
i~1
Proof· By Theorem II.2.S and Lemma 1.3.1 the subspace iSv'(U)+

+ ... +is,,,(U) is neutral. Thus (4.1) follows from Lemma 1.1. 0
Corollary 4.2. An isometric operator in Ilk has no more than k dgen-

values inside and no more than k eigenvalues outside the unit circle. 0
The next theorem is the main result of this section. Roughly

speaking, it yields a bound for the total length of all Jordan chains
of an isometric operator in II", excepting chains of length 1 to uni-
modular eigenvalues.
Theorem 4.3. Let U be an isometric operator in Ilk .
a) If s is a unimodular, non-semi-simple eigenvalne of U, then
(4.2) codim@i,(u) m(U - sI) < 00.
In particular, the sum
00
(4·3) m(l::) = L'codimw(u_sI)2j m(U - sI)2i-l

i~l
is finite.
b) If 1::1> ••• ,eN are unimodular, non-semi-simple eigenvalues and
VI' . . . ,vn are non-unimodular eigenvalues of U such that v7 =F VI
(j,l = 1, ... ,n), then
N n
(4.4) 1.," m(l:: j )
i~l
+ LmYj :::;
i~1
k ,
where mYj = dim (5Vj(U) (j = 1, ... ,n) .

c) If the operator U is closed and I:: is a unimodular, non-semi-simple
eigenvalue of U, then the principal subspace (5,(U) can be decomposed
into an orthogonal direct sttm of the form
(4.5) (5e(U) = (5' (+) (5",
where (5' is non-zero, finite-dimensional, and invariant for U, while (5ft is
closed, negative definite, and contained in the eigenspace m(U - sI).
Selecting a basis of (5' that consists of Jordan chains of U to the eigen-
valtte s and discarding chains of length 1, the lengths d i (j = 1, ... , 1')
of the remaining chains depend on U and I:: only; in addition,
(4.6) t [diJ2
J~l
= m(e) ,
where the brackets [ ] stand for "integral part".

We postpone the proof to Section 5. 0
Besides Theorem 4.1 and Corollary 4.2 the following special cases
of Theorem 4.3 should be mentioned.
Since, on account of (4.6), for a unimodular, non-semi-simple eigen-
value s we have m(s) ~ 1, while for a non-unimodular, non-semi-
simple eigenvalue V obviously mv ~ 2, relation (4.4) yields:
4. Isometric and Symmetric Operators 191
Corollary 4.4. The number of non-semi-simple eigenvalues of an iso-

metric operator in Ilk does not exceed k. 0
Since in case SJC(U _cI)2k+ 1

would contain at least k +
*
SJC(U - cI)2k+2 the sum (4.3)
1 non-zero terms, from (4.4) we deduce:
Corollary 4.5. The length of a Jordan chain corresponding to a uni-
modular eigenvalue of an isometric operator in Ilk cannot be greater than
2k + 1. 0
Below we formulate the counterparts, involving symmetric opera-

tors, of Theorem 4.'1 and Corollaries 4.2, 4.4, 4.5. The analogue of
Theorem 4.3 holds as well. The proofs can be accomplished either by
imitating the argument applied in the isometric case, or by reduction
to the isometric case via a Cayley transformation (d. Sections 11.4- 5
and Theorem V1.7.3).
*
Theorem 4.6. Let A be a symmetric operator inIlk • Let AI, ... , A"
be a set of non-real eigenvalues of A such that Ai I,t (j,l = 1, ... , 11) .
Then
n
17 dim 6;,(A) ~ k . 0
i=l
Corollary 4.7. A symmetric operator in Ilk has no 1110re than k eigen-
>
valtfes in the half-plane 1m A 0 and no more than!? eigenvalues in the
<
half-plane 1m A o. 0
Theorem 4.8. The nlnnber of non-semi-simple eigenvalues of a sym-
metric operator in Ilk does not exceed k 0
Theorem 4.9. The length of a Jordan chain corresponding to a real
+
eigenvalue of a SY1111Jtet1'1:C operator in Ilk cannot be greater than 2!? 1. 0
Remark 4.10. All of the estimates appearing in the present section

are sharp. To see this, say, for symmetric operators, recall Example
II.3.2, where a symmetric operator with one pair of non-real eigen-
values A, A in III was constructed; taking the cartesian product of !?
* *
copies of the space and defining in the fth copy (i = 1, ... , !?) a sym-
metric operator with non-real eigenvalues Ai' I'i (Ai Al for iI), one
obtains a symmetric operator with non-real eigenvalues AI, AI> ... , Ak ,
Ak in a space Ilk' Starting from Example 11.3.1, a similar process leads
to a symmetric operator with k real, non-semi-simple eigenvalues in Ilk'
Further, choosing two k-dimensional neutral subspaces ~\ # 22 in
a 2!?-dimensional space Ilk (d. Theorem 11.1'1.5), defining Al in 21
to be a linear operator with a Jordan chain of length k to a non-real
eigenvalue A, and defining a linear operator A2 in 22 so that the matrices
of Al and A2 relative to dual bases of .\31 and .\3 2 (Lemma 1.10.6) would
be the transposed conjugates of each other (Lemma 1.10.1), we easily
verify that the direct sum A = Al +
A2 is symmetric on Ilk and has
a Jordan chain of length k to the non-real eigenvalue A. Finally, the
existence of a symmetric operator having a Jordan chain of length
2k + 1 to a real eigenvalue in Ilk will be exhibited by the next
example.
Example 4.11. Let Ilk (k >
0) be the orthogonal direct sum of a
k-dimensional positive definite subspace Il+ and a (k 1 )-dimensional +
negative definite subspace n-.
Let {ej}~ and {If }~+1 be orthonormal
bases of Il+ and Il-, respectively. Set
1.
(I if s_ l' S
) V2 j T
I
ej ) 1 k ,
+1,
lV~-
gj = tk+l if j = k
U'w; - c,w-;) jf /, + 2;;; i < 2k +1.

Then
if j +
Z = 2k +2,
otherwise.
The linear operator A defined by the equations
Agj = gj-l (j = 1, ... , 2k + 1; go = 0)
turns out to be symmetric.
5. Proof of Theorem 4.3
a) The subspaces
(5.1) I](j = I]((U - eI)j (j = 0,1, ... )
obviously satisfy the conditions
(5.2) (j=0,1, ... ),
00
(5.}) U I](j = ®e(U) .

j=o
It is also easy to see that
(5.4) I](j = I](i+l implies I](i+l = 1](j+2 .
Hence part a) of our theorem holds if and only if

codim'J/j+1 I](j < CXJ (j = 1,2, ... )
5. Proof of Theorem 4.3 193
and
SJ'li = SJ'li+ 1 (f > jO) •
Let j > 1, and let @ C SJ'li+l be a subspace linearly independent of
SJ'li' Employing the identity
(5.5) (v, (U - sl)w) = -8((U - e1)v ,Uw) (lei = 1; v,w E ~(U)) ,
which is a simple consequence of the isometric property of U, for
x E SJ'l1 and y E @ we find:
(x, (U - e1)iy) = -e((U -e1) x, U(U - e1)i- 1 y) = 0,
Thus (U - e1)i@ 1.. SJ'l1' On the other hand, the assumption @ C SJ'li+l
yields (U - e1)i@ c SJ'l1' So (U - e1)i@ is orthogonal to itself and
°
therefore, by Lemma 1.1, its dimension does not exceed the value k.
Since, however, the condition @nSJ'li = implies that (U - e1)il@
is invertible, we have dim@ < k. As a result,
(5.6) codim!ni+l SJ'l1 < k (j = 1,2, ... ) .
Let t > 1 be an index such that SJ'lt =1= SJ'lt+l' Then, according to
(5.4), SJ'li =1= SJ'li+l (f = 0,1, ... ,t). Thus, in view of (5.2), each
SJ'li (f < t) has a non-zero complementary subspace in, SJ'li+l:
(5.7) SJ'li+l = SJ'lj + @i (j = 0,1, ... , t) ,
(5.8) @j =1= ° (j = 0,1, ... , t) .
Put
(5.9) (j = 1,2, ... , [t ~ 1]).
From the relations @2i-l
obtain:
C SJ'l2i' @2i-1 n~i-l = ° (see (5.7)) we
(5.10) £1 c SJ'l1' £inSJ'li-1 = ° (j = 1,2,. ", [t ~ 1]).

Therefore, in view of (5.2), the direct sum
(5.11) WIt = £1 + ... + £t' (tl = [t ~ 1])

exists. Obviously,
(5.12)
Applying equation (5.5) j times and recalling (5.7), for x E @2i-l'
Y E@2q-l (1 < j:::;:; q < [t ~ 1])we derive:
((U - e1)ix, (U - e1)qy) =
= (_ e)i((U - sl)2ix, Ui(U - e1)q-iy ) = °.
Hence ~i 1- ~q (1,. q = 1,2, ... , [t +1])

~ and, consequently, met
is a neutral subspace:
(5.13)
In particular, by Lemma '1.1,
(5.14) dim mCt :S k .
On the other hand, since @2i-1 n?Jei c @2i-l n ?Je2i - 1 = 0 (d.
(5.7)), the restriction of (U - 8I)i to @2i-l is invertible. Therefore,
owing to (5.9),
(5.15) dim 2j = dim @2i-l (r. = '\ ,2, ... , + 1]) '
[t-2
so that (5.8) and (5.H) yield
(5. '16) dim 9)(t ;;:::: [t : 1] .

From (5.14) and (5.16) we obtain the inequality t:S 2k. Thus, on
account of the definition of t,
(5.17) ?Je j = ?Je i + 1 (f = 2k + 1, 2k + 2, ... ) .
Relations (5.6) and (5.17) imply (4.2).
b) Pursuing the investigation of a single unimodular, non-semi-
simple eigenvalue 8 for a while, and adhering to the notation introduced
above, we set (d. (5.17), (5.'11))
(5.18) to = max{t: ?Jet *- ?Jet + 1} ,
(5.19) m1(8) = meto .

Then, in accordance with (5.12) and (5.13),
(5.20) 9)((8) C €J.(U), 9)((8) 1- 9)((8) .
Moreover, taking into account relations (5.19), (5.11), (5.15) and (5.7)
as well as the special choice (5.18) of to we find:
t~ t~
dim 9)((8) = L dim 2j = L dim @2i-l =
i~l i~1
t~ . co
= 2: codim!JI2' ?JeZj - 1 = L' codim9/2i ?Je2j - 1 •

j~1 J J~l
Consequently (see (4.3)),
(5.21) dim 9)((8) = m(8) .
Passing to the situation described in part b) of the statement, we
consider the direct sum
5. Proof of Theorem 4.3 195
whose existence follows from the fact that the principal subspaces of an
operator are linearly independent. By Theorem 11.2.5 the components
of IDC are pairwise orthogonal. The same theorem assures that each
(5Vj(U) is orthogonal to itself. Finally, (5.20) yields IDC(si) ..1 IDC(sj)
(j = 1, ... , N). As a result, IDC is a neutral subspace. Hence, in view
of Lemma 1.1,
(5.23) dim IDC < k .
Relations (5.21)-(5.23) imply (4.4).
c) For a closed U the eigenspace m1 = m(U - el) is closed.
m
Therefore, owing to Theorem V.3.1, 1 has a fundamental decompo-
sition of the form
(5.24) m1 = m~ (+) mt (+ ) m1 ;
m~ c 1.)30 , mt c 1.)3++ , m1 C 1.)3-- ,
where m1 is closed.
Choose
(5.25)
Then (5"c (5.(U) and, by Theorem 2.2, (5" is ortho-coinplemented in
il". Consequently, there exists a subspace (5' satisfying
(5.(U) = (5' (+) (5".
As S is a non-semi-simple eigenvalue, (5' cannot be zero. Moreover,
S being non-zero and (5" being contained in the eigenspace m1 =
= m(U - el), we have U(5" = (5", so that Lemma II.2.9 and the
invariance of (5.(U) yield U(5' c (5'.
Set
(5.26) U' = UI(5', mi = m(U' - el)i (j = 0,1, ... ) .
Applying assertion a), already established, to the operator U' we find:
(5.27)
On the other hand, the relations m~ c mI , m; n m1 C (5' n (5" = 0
imply dimm~ < codim\ll,m1 ; hence, in view of (5.24), Lemma 1.3.1
and Lemma 1.1,
(5.28) dim m~ < k.
From '(5.27) and (5.28) it follows that (5' is finite-dimensional.
Since (5" is ortho-complemented whereas (5" c 1 C i (j = 1,2, m m
... ) and, according to (5.26), m;
= min(5' (j = 0,1, ... ), we have
(5.29) (j = 1,2, ... ) .

In particular, the definitions (5.18) and

(5·30) to = max{t: ~; ~;+d *
are equivalent.
Let @;
(f < to) denote a complementary subspace to ~j in ~i+t:
(5·31) ~j+l = ~j + @j (f = 0,1, ... ,to)·
Selecting first @;. and then proceeding to smaller and smaller indices
it can be achieved that
(5.3 2) (U - eI)@; ( @;-1 (f = 1, . . . , to) .
Really, from @j c ~j+l and @j n~; = °
we deduce (U - e1)@j ( ~j
and (U - e1)@j n ~-1 = 0, respectively.
Formulas (4.5), (5.26), (5.30) and (5.31) yield:
6' = 6.(U') = ~;.+1 = @~ @~ + + ... +
@;.,
Therefore we can build up a basis of 6' from bases of the @j .
Let ej,1>"" ej,gj be a basis of @j, where 1 < f < to' Then,
owing to (5.32), the vectors
(5·33) ej_I,1 = (U - el)ej,I' ... , ej_l,gj ~ (U - eI)ej,gj
belong to @f-l' Moreover, they are linearly independent, since by
(5.31) the restriction of U - e1 to @i is invertible:
~(U - e1) n@i = ~; n@j ( ~fn@; = 0.
Consequently, the system (5.33) can be extended to a basis of @f-l'
Starting from a basis of @;. and applying the above procedure suc-
cessively to times, we obtain a basis of 6' which consists of Jordan
chains of U to the eigenvalue e. Simultaneously it turns out that the
number Pj of Jordan chains of a given length f can be expressed as
(5·34) Pj = dim @j-l - dim @i
(f = 1, ... ,to + 1; @;.+1 = 0) .
But, in view of (5.31) and (5.29),
(5·35) dim @j = codimlni+l ~i (f = 1, ... , to) .
Therefore the values P2' Pa, ... ,Pt.+l depend neither on the choice
of the decomposition (4.5) nor on the selection of the Jordan basis
of 6'.
The proof of equation (4.6) is now within reach. In fact, (4.3) and
(5.35) along with (5.18) imply that
t~
(5.3 6) m(e) = L dim @~i-l
i=1
6. Regular Symmetric Extensions 197
Further, from (5.34) we derive
dim ®: =
1,+1
2; p.
i=s+1 J
(s = 0,1, ... ,to).
In particular,
dim ®; = P2 + P3 + P4 + P5 + Ps + P7 + ... + Pto+1 ,

dim ~j; = P4 + P5 + Ps + P7 + ... + Pt,+1 ,
dim ®; = Ps + P7 + ... + Pt.+1 ,
Putting these expressions into (5.36) we finally obtain the relation
which coincides with (4.6). 0
6. Regular Symmetric Extensions
Let A be a symmetric operator with dense domain in the Pontrjagin

space II". Let II~ be a Pontrjagin space containing Ilk and having the
same rank of positivity x+(II~) = x+(IIk) = k <
00. Let Al be a sym-
metric operator in II~ satisfying the conditions SD(A) ( SD(A I ), Alx =
= Ax (x E SD(A)). Then Al is called a regular symmetric extension
of A.
In other words, a regular symmetric extension is a symmetric
extension in a possibly larger space of type Ilk, where k is unaltered.
Let, again, A be symmetric and densely defined in II" . Owing to
Theorem 1.4 and Lemma 1.3 there exists a fundamental decomposition
such that
(6.2) II+ c SD(A) .
Denote the corresponding fundamental projectors by P+, P-. With
respect to (6.1) the operator A can be represented by a matrix
(6·3)
where
An = P+ AIIJ+ , Al2 = P+AIII-,
(6.4)
A2l = P-AIII+ , A22 = P-AIII-.
A few properties of Ail (j,l = 1,2) have been enumerated in Lemma
VI.6.7.
Consider a spacelIo (i.e., a negative definite, intrinsically complete
space) containinglI-, and a symmetric extensionA;2 of A22 iniio . Set
(6.5)
where P n - is the orthogonal projector from lIo onto 1I-. It is easy to
see that the matrix operator
(6.6) (An A' =

A21
acting in the space II; = 1I+ xlIo , is a regular symmetric extension
of A. Conversely, every regular symmetric extension can be obtained
by selecting lIo and A;2 in a suitable way. Moreover, according to
Lemma VI.6.7, A' is selfadjoint in 1I~ if and only if A;2 is selfadjoint
in lIo .
Thus the problem of finding regular sYmmetric extensions of
operators in II" is equivalent to that of finding symmetric extensions
(within or beyond the space) of Hilbert space operators.
Next we examine non-real eigenvalues of regular symmetric exten-
SIOns.
Theorem 6.1. Consider a densely defined sym·metric operator A in II"
and a non-real number I,. The following conditions are equivalent:
a) ffi(A _AI)l n 1,)30"* 0;
b) A has a symmetric extension Ao in IIk such that I, is an eigenvalue
of Ao;
c) A has a regular symmetric extension Al such that I. 1:S an eigenvalue
of Al and one of the corresponding eigenvectors belongs to II" .
Proof· Let Xo ~ffi(A - AI), (xo, x o) = 0, xo "* O. If Xo E SD(A),
then I. is an eigenvalue of A itself, since
(6.7) (Axo - I.X o , y) = (xo, (A - AI)y) = 0 (y E SD(A)) .
If Xo (i SD(A), we define Ao as the linear extension of A to <SD(A), xo >
satisfying Aoxo = ho, and establish that Ao is symmetric:
(Aoxo , xo) - (xo, Aoxo) = I.(x o , x o) - I.(xo, x o) = 0 ,
(Aoxo, y) - (xo, AoY) = - (xo, (A - illy) = 0 (y E SD(A)) .
Thus a) implies b). Further, b) obviously implies c) withAl = Ao.
6. Regular Symmetric Extensions 199
Finally, suppose that c) holds. Then for some non-zero element

Xl Eilk we have Alx1 = AX!. Corollary II.3.4 yields (Xl' Xl) = O. On
the other hand,
(Xl' (A - iI)y) = (AIXI - AXI , y) = 0 (y E ~(A)) .
Hence a) is valid. 0
Theorem 6.2. Consider a densely defined symmetric operator A in ilk
and a non-real number A. The following conditions are equivalent:
a) ffi(A - AI) 1 n ~++ =1= 0;
b) A has a regular symmetric extension A2 such that A is an eigen-
value of A2 and one of the corresponding eigenvectors does not belong
to ilk'
Proof. Let xo..l ffi(A - AI), (xo, xo) >
O. Then Xo E! ~(A), since
otherwise the equation Axo = AXo (see (6.7)) coupled with Corollary
11.3.4 would imply (xo, xo) = O. Set il; = <ilk, u o where >,
and define A2 as the linear extension of A to <~(A), xo+uo satisfying >

+
A 2(x O u o) = A(Xo +
u o). On account of the relations
(A2(XO + uo), Xo + uo) - (xo + U o, A2(XO + uo)) =
= (A -A)(Xo + U o , Xo + uo) = 0,
(A2(XO + uo), y) - (xo + U o , A 2y) = -(xo + U o , (A -iI)y) =
= -(xo, (A - lilY) = 0 (y E ~(A))

the operator A2 is symmetric.
Conversely, let b) be satisfied. Then there exists an X2 E! ilk with
A 2x2 = AX2 • According to Theorems 2.4 and 2.2 x 2 has a projection Xo
on ilk . Since, by Corollary 11.3.4, X2 is neutral and, in view of Lemma
1.1 and Remark 1.3.2, X 2 - Xo is negative, Xo must be positive. More-
over,
(Xo:, (A - lilY) = (X2' (A - ~I)y) = (A2X2 - AX2 , y) = 0
(y E ~(A)). 0
Theorems 6.1-6.2 yield:
Corollary 6.3. Let A be a densely defined symmetric operator in ilk ,
and let A be a non-real number. Unless the orthogonal companion of
ffi(A -AI) is negative definite, A is an eigenvalue of some regular sym-
metric extension of A. 0
7. Invariant Positive Subspaces: Existence
Since every continuous operator of finite rank 15 compact, from

Theorem VIIIj.1 and Lemma -1.2 we obtain:
Theorem 7.1. Let U be a unitary operator in Ilk' Then U has a

k-dimensional invariant positive subspace 21 which contains the principal
subspaces ISv(U) (Ivl >
1). If v is a normal eigenvalue with Ivl >
1, then
ISv.(U) n21 = O.
Similar statements hold for Ivl < 1. D
The analogue, involving selfadjoint operators, of this result reads

as follows.
Theorem 7.2. Let A be a selfadjoint operator in Ilk' Then e(A)

contains at least one non-real point. Moreover, A has a k-dimensional
invariant positive subspace 21 such that 21 c ;ti(A) and IS}, (A ) c 21
(1m ;t> 0). If;t is a normal eigenvalue with 1m;t > 0, then
1S;.(A)n21 = O.
Similar statements hold for 1m A O. <
Proof. Making use of Theorem 1.4 and Lemma 1.3 we can find a
fundamental decomposition (6.1) with the property Il+ c ;ti(A). Intro-
ducing the matrix elements (6.4), from Lemma VL6.7 it follows
that A12 is continuous on ;ti(A 12 ) = ;ti(A 22 ), while e(A 22 ) is non-empty.
Thus for (0 E e(A 22 ) the operator A12(A22 - (01)-1 is continuous;
having a finite rank, it is compact. Now we apply Theorem VIIL3.2
and Lemma 1.2. D
Since positizability plays a basic role in the spectral theory of

Krein spaces (d. Section VIII.6), the next result is particularly im-
portant.
Theorem 7.3. Every continuous selfadjoint operator in Ilk is positiz-

able.
Proof. Let AE)S(Ilk), A = A *. According to Theorem 7.2 A has
a k-dimensional invariant positive subspace 21 , As it is well known
from linear algebra, there exists a non-constant polynomial rp (e. g. the
characteristic polynomial or the minimal polynomial of A \21 in case
k > 0, and rp(;t) = }, if k = 0) such that rp(A \2 1) = O. Let (j5 be the
polynomial whose coefficients are complex conjugate to those of rp.
Then
(<p(A)x, y) = (x, rp(A)y) = 0
i. e., ffiCil(A))lJ3 1 . HEnce, by Lm:ma 1.2 and Lemma 1.6.3,

(ip(A)x, ip(A)x) ~ 0 (x E Ilk) .
Therefore A is positizable by the real, non-constant polynomial-C(ip. 0
The following consequence of Theorem 7.1 will be needed when
proving the existence of common invariant subspaces for commuting
systems of unitary operators in Ilk.
Lemma 7.4. Let U be a unitary operator in Ilk. Let we
be a closed
subspace of Ilk such that Uff)1 = ffJ1. Unless ffJ1 is negative definite,
ffJ1 contains a non-zero positive subspace which is invariant for u.
Proof. If 9)1 is degenerate, then its isotropic part ffJ10 = we
n ffJ11 is
non-zero, neutral and, by Corollary V1.4.6, invariant for U. If 9)1 is
non-degenerate and not negative definite then, in view of Corollary
2.3, ffJ1 is a space of type II", , where 0 < k':S k; thus we may apply
Theorem 7.1 to the restriction UIffJ1· 0
8. Invariant Positive Subspaces: Uniqueness
We are going to examine the question whether the k-dimensional in-

variant positive subspace of a selfadjoint operator in Ilk is unique or
not.
Beginning the study with the case of a Pesonen operator, we first
need the following lemma, interesting on its own right.
Lemma 8.1. Let A E'iS(II,,) be a Pesonen operator such that

(Ax, x) >
0 (x E 1,l300). Then the value
. (Ax, x)
(8.1 ) flA = mf - -- ...
(x, x)
(x,x»O
is /t"nite and it is an eigenvalue of A ; at the corresponding eigenvectors x,

and only there, the infimum (8.1) is attained. 1l10reover, the eigenspace
91(A - flAIl is contained in every k-dimensional invariant positive
definite subspace of A.
Proof. Owing to Theorem 7.2 and Corollary 11.9.6 A has a k-dimen-
sional invariant positive definite subspace 2 1 . By Lemma 1.3 there
exists a fundamental decomposition
(8.2) IIk=II+(+)II-; II+c~++, II-cl,l3--
such that II+ = 21 , Since a Pesonen operator is always understood to
act in an indefinite space, II+ and II- cannot be zero.
Let (x, x)> 0, x <! II+. Then for the decomposition x = x+ x- +

(x± E II±) we have x+ =1= 0, x- =1= 0. Since Lemma 11.6.4 applied to the
inner product (. , .h = (. , ')A yields
(Ax-, x-) / (Ax+, x+)
- - - _.... <-...----'
(x-, x-) (x+, x+)
making use of Lemma II.3.9 we obtain:
(Ax+, x+)
(Ax, x) = (Ax+, x+) + (Ax-, x-) > (x+, x+)
[(x+, x+) + (x-, x-)] .
Consequently,
(Ax, x) (Ax+, x+)
-----
(x, x)
> ---,----:--
(x+, x')
(XE ~++, X <!II+) .
Thus the infimum (8.1) can only be attained on the finite-dimen-

sional, invariant, positive definite subspace II+. There, however, it is
really attained, namely at the eigenvectors of A III+ corresponding
to the eigenvalue flA' as stated by a well-known theorem of linear
algebra and easily verified with the help of the eigenvector expansion
of AIII+.
Let Xo be any eigenvector of A to the eigenvalue flA' We have just
seen that A has at least one positive eigenvector to flA; therefore, on
account of Corollary 11.9.8, Xo is positive. Moreover,
(Axo, xo) (flAXO, x o)
(8.4) -_.- =._-.-...-. = flA .
(xo, x o) (xo, x o)
Relations (8.4), (8.1) and (8.3) imply Xo E II+. 0
Theorem 8.2. Let A be a Pesonen operator inII". Then the k-dinun-

sional invariant positive subspace of A is unique.
Proof. Since A and -A have the same invariant subspaces, by
Theorem 11.9.1 we may assume that (Ax, x) > for every x E ~oo. °
Let ,3 be a k-dimensional invariant positive subspace of A (d.
Theorem 7.2). According to Corollary 11.9.6 and Lemma 8.1 ,3 is
definite and contains the non-zero subspace 91(A - flAIl.
If 91 (A - flAIl =1= B, then theorthogonalcompanionof~(A-flAI)
°
is a space of type IIk , with < k' < k. Furthermore,
,3 = 91(A - flAIl (+),3',
where,3' = ,3nIIk' , dim,3' = k'. Also, byTheoremII.3.7, AIIk, (IIk ,
and A,3' ( ,3'. Setting A' = A IIIk , , introducing flA' in conformity
with (8.1) and applying Lemma 8.1 again, we obtain that flA' is an
eigenvalue of A' and 91 (A ' - flA,I) ( ,3' .
Next, supposing that ?JC(A' - f.tA,I) ic 2', we consider the space

ilk" = (?JC(A - f.tAI) (+) ?JC(A' - f.tA,I))~,
the operator A" = Alilk " , and the invariant subspace 2" = 2nilk ,,;
etc. After no more than k steps the process breaks off. Thus 2 is the
direct sum of a certain number of subspaces uniquely defined by A. D
Owing to Theorem 7.2 and Theorem VI.7.5, the number of
k-dimensional invariant positive subspaces for a selfadjoint operator
having a non-real normal eigenvalue inilk is not less than two. Below
we give a non-trivial example of a selfadjoint operator with real spec-
trum and several k-dimensional invariant positive subspaces in ilk'
Example 8.3. Let ilk (k >
0) be the orthogonal direct sum of a
k-dimensional positive definite subspace il+ and a k-dimensional
negative definite subspaceil-. Let {ei}~ and {fi}~ be orthonormal bases
of il+ and il-, respectively. Set
jf 1~j~k,
if k + 1 <j'~ 2k.
Then
if j +
I = 2k +1,
otherwise.
The linear operator A defined by the relations
(jic1,k+1);
is selfadjoint, has the only eigenvalue 0, and each of the subspaces
<gl' g2, ... , gk>' <gv g2, ... , gk-l, gk+l> '
<gv g2, ... , gk-2, gk+l' gk+2>, ... , <gk+l' gk+2' ... , g2k>
is neutral and invariant under A.
Although the k-dimensional invariant positive subspace need not
be unique, certain parameters connected with it are uniquely deter-
mined.
Lemma 8.4. Let A be a selfadjoint operator in ilk' Let 21 c ;tJ(A) and
B c ;tJ(A) be a k-dimensional and an arbitrary invariant positive sub-
space of A, respectively. Then for every A E R we have
dim i5,,(A \2) <: dim i5 A(A \2 1) .
In particular, every real eigenvahte of A \2 is an eigenvalue of A \21 too.

Proof. The subspace ~1' of dimension k, is the direct sum of the

principal subspaces of A I~l' Replacing the component @S;.,(A I~l)'
where ,10 is real, by @SA, (A I~), the direct sum remains positive because
of Theorem II.).) and Lemma 1.3.1; thus, on account of Lemma 1.1,
its dimension cannot increase. (If ,10 (£ <1p(A I~l), then the "compo-
nent" @3,l.(A I~l) is understood to be zero; d. Section ILL) 0
Corollary 8.5. Let A be a se~fadjoint operator in Jlk • Then the real
eigenvalues Al < < ... <
,12 A, of A in the k-dimensional invariant
positive subspace ~1 c ~(A) and the algebraic multiplicities m,l,(AI~I)'
... , m",(A 1.21) do not depend on the special choice of .21 ' 0
Remark 8.6. The conclusions of the items 8.3 -8.5 carryover to

unitary operators in a natural way. Really, the proof of Lemma 8,4 can
be repeated almost unaltered (substituting "unimodular" for "real"),
while in the case of Example 8.3 it is easier to apply a Cayley trans-
formation and refer to Lemma II.S,4.
9. Common Invariant Positive Subspaces

for Commuting Operators
Lemma 9.1. Let U1> ••• , Un be commuting unitary operators in Ilk'

where k > o. Then Ilk has a non-zero positive subspace .2 such that
U i .2 c.2 (j = 1, ... , n).
Proof. We proceed by induction. Owing to Theorem 7.1 the result
holds for n = 1. Assume that it holds for n.
Let Ul , ••• , Un +! be commuting unitary operators in Ilk (k> 0).
On account of the induction hypothesis there exists a non-zero posi-
tive subspace .2 c Ilk such that Uj~ (.2 (f = 1, ... , n). Since, by
Lemma 1.1, dim.2 < k, Lemma 11.1.2 applies to the operators Ull.~,
... ,Un l.2. Consequently, for a non-zero Xo E .2 and suitable numbers
Ai we have
(9.1) (j=1, ... ,n).
Set
n
(9.2) W1 =
i
=!
n 'Jc:(U.1 -,1.1).
1
Lemma II.1.i yields U,,+!Wc c W1. On the other hand, u;;t! also
commutes with Ui . (Really, if A and B are commuting, completely
invertible operators, then AB-l = B-lBAB-l = B-lABB-l =
= B-lA.) Hence u;;t!WC c WC. In addition, WC is closed and contains
9. Common Invariant Positive Subspaces 205
the non-negative vector X o -=/= O. Therefore, according to Lemma 7.4,

W1 has a non-zero positive subspace 2' which is invariant under U n + 1 •
In view of (9.2) the subspace 2' c me is invariant for u1 , . . . , Un as
well. 0
Lemma 9.2. Let U1 , • . . , Un be commuting unitary operators in Ilk'
Then Ilk has a k-dimensional positive subspace 21 such that U/31 c 21
(j = 1, ... , n).
Proof. The case k = 0 is trivial, so let k> o. Then Lemma 9.'!

provides a non-zero positive subspace 2 satisfying the relations
(9·3) (j=1, ... ,n).
Suppose that
(9.4) dim 2 < k.
Since U i is invertible (Theorem VI.4.2) and dim2 00, in (9.3) inclu- <
sion can be replaced by equality. Therefore, on account of Corollary
VI.4.6,
(9.5) Uj 2 1 = 21 , Ui 2° = 2° (j = 1, ... .' n) ,
where 2° = 2n21.
From Theorem 2.6 with the help of Lemma III.2.4, relation (9.4)
and Corollary V.3.7 it follows that for some k' < k the quotient space
2 1/2° is a space of type [h,. Moreover, owing to (9.4), Lemma 1.2 and
Lemma V.4.S, k' cannot be zero.
~ view of (9.5) the operators Uf induce commuting unitary opera-
tors U i on 2 1/2°. Consequently, by Lemma 9.1, there exists a non-
zero positive subspace We c 2 1 /2° with the property
A A A
(9.6) UJWl c Wl (j = 1, ... ,n) .

Lemma 1.1 assures that dim~ ~ k'. Let xl> ... ,
Xs be a basis ot
fuc, and let Xi E 21 be a representative of the equivalence class Xl
(j -'- 1, ... ,s). Then for the subspace
Wl = <Xl' ... ,xs >c 21
we have
(9.7) (j = 1, ... , n) .
Further, since Wl is positive, non-zero and linearly independent of 2°,
the subspace 2 + Wl is positive (d. Lemma I,}.1) and different from 2-
The invariant positive subspace 2 satisfying (9.4) could thus be
extended to an invariant positive subspace 2+Wl with dim(2+9)() >
> dim 2. Iteration, at most k-dim2 times, of the process yields a
k-dimensional invariant positive subspace 21 as required. 0
Theorem 9.3. Let U be a family of commuting unitary operators in Ilk .

Then Ilk has a k-dimenst"onal positive subspace 21 S1lCh that U2 1 ( 21
(U E U) .
(9.8) Ilk = Il+ (+ )Il-; Il+ c 1,l5++, Il- ( I,l5 -- ,
and denote the corresponding fundamental symmetry by J. Let
(U j1 kl=1,2 be the matrix of a U E U with respect to (9.8). Consider
the space )81 = 'iE;(Il+, Il-) and its 'weak operator topology Two' Set
(9.9)
The assertions a) - c) below can be obtained by specializing the
proof of Theorem VIII.2.1 to the case 2 = 0, .~ = Ilk , T = U .
a) The formula
(9.10) 21 = {x+ +
K1X+: x+ E Il+}
defines a one-to-one correspondence between the class of all k-dimen-
sional invariant positive subspaces 21 of U and the set S't'u of all opera-
tors K1 satisfying the conditions
(9.11)
(9.12) U21 + U22K 1 - K1(Ull + U12K 1) = 0 .
b) Sl' is Two-compact.
c) The left-hand side of (9.12) is a Two-continuous function of the
variable K1 E Sl'.
According to a), the conclusion of our theorem holds if and only if
(9.13) n Sl'u
UEU
"* £}.
But, on account of a) and c), every Sl'u is contained in Sl' and closed
with respect to the induced topology Twol ~1'. Therefore, by assertion
b), instead of (9.13) it is sufficient to show that the intersection of any
finite subfamily of the family {Sl'u }UEU is non-void. In other words
(see a)), it is sufficient to prove the theorem for finite familiesU. This,
however, has been previously done in Lemma 9.2. 0
Theorem 9.4. Let III be a family of commuting continuous selfadjoint

operators in Ilk' Then Ilk has a k-dimensional positive subspace 21 such
that A21 c 21 (A E Ill).
Proof. Since the invariant subspaces of A coincide with those of
i,A (J. E C; i, "*
0), we may assume that
(9.14) IIAII] ~ 1 (A E Ill) ,
where J is a fixed fundamental symmetry.
Notes to Chapter IX 207
Let' E C, 1'1 >

1, ,*- ,. -
On account of (9.H) we have' E g(A) for
every A E Ill. Thus, according to Theorem V1.7.1, the Cayley trans-
forms
(9.15) UA = (A - '1)(A - U)-l = 1 + (' - ,)(A - '1)-]
(A E ill)
exist and are unitary. In addition, the operators U A (A E Ill) are easily
seen to commute. Therefore Theorem 9.3 guarantees the existence of
a k-dimensional positive subspace f!l c Ilk with the property
(9.16) UA'~1 c Bl (A E Ill) .
Let us prove that B1 is invariant under the operators A as well.
Solving (9.15) for A we find (d. Lemma 11.4.1):
(9.17) A = (,U A - U)(U A - I)-I (A E ~() .
Furthermore, the relations (U A - I)Bl C Bl (see (9.16)), dimB 1 < 00,

SJI.(U A-I) = 0 yield
(9.'18) (U A - I)B1 = B1 .
From (9.16) - (9.18) we obtain the desired inclusions ABI C~l
(A E Ill). 0
Notes to Chapter IX
Sequence spaces of type Ilk were introduced by Pontrjagin [1]. Theo-

rem 1.4, Theorem 2.2 and the "symmetric version" of Theorem 4.3
belong to this author.
The axiomatic definition of Ilk is due to Iohvidov and KreIn [1J,
[2J, who established Theorems 2.5, 3.1, 4.3 and Example 4.11. Another
treatment of Theorem 2.5 was given by Bognar [1]. For Theorem 3.2
and further discussion of the continuity of isometric operators d.
Iohvidov [11].
The results of Section 6 were obtained by Krein and Langer [4 J
in the context of a detailed study of defect subspaces and generalized
resolvents. For generalized resolvents in Ilk d. also KreIn and Langer
[5J, Langer :14J, Sorjonen [1J.
Making use of analytic methods, Theorem 7.1 can be significantly
improved (d. also the Notes to Chapter VIII). Namely, if U is unitary
in ilk, then from Hilbert space perturbation theory it follows that the
non-unimodular points of O'(U) are normal eigenvalues. Therefore 1)
the non-unimodular points of 0'( U) form at most k pairs, the members
of each pair lying symmetrically with respect to the unit circle; 2) if
208 IX. Pontrjagin Spaces and Their Linear Operator
the k-dimensional invariant positive subspace 21 contains all principa

subspaces 6 v (U) with Iv\> <
1, then for Iv\ 1 necEssarily 21 n 6 v (U) = C
These facts have further implications.
Another way of obtaining the above improvements of Theorem 7.'
is to prove first, again by perturbation theoretic methods, the existenci
of two k-dimensional invariant positive subspaces 21 and 2~ such tha·
all non-unimodular eigenvalues of Ul21 (resp. UI2~) are locatE(
outside (inside) the unit circle; everything else follows then by rela
tively simple geometric arguments. Cf. especially Iohvidov and Kre'll
[1].
Similar comments can be made in connection with Theorem 7.2.
For historical remarks on invariant subspace theorems see thl
Notes to Chapter VIII.
Azizov and Iohvidov [2J showed that Theorem 7.2 does not exten(
to non-degenerate quasi-negative or quasi-positive spaces ("non
complete Pontrjagin spaces"), not even in case they admit a Hilbe'r·
majorant.
Iohvidov and Kre'ln [2J (d. also Iohvidov [2J) proved an analogUi
of Theorem 7-3 for unitary operators and gave a characterization of thl
"positizing functions" involved. Their results were carried over t(
(possibly discontinuous) selfadjoint operators by Lo [1]. For an appli
cation to differential operators see Langer [15].
In Lemma 7.4 even the existence of a maximal positive subspace oj
we invariant under U could be stated (see Iohvidov and Kre'ln [2J).
Lemma 8.1 and Theorem 8.2 belong to Kuhne [1]. In the rest 0
Section 8 we follow Iohvidov and Kre'ln [2].
For further properties of spectra and invariant subspaces cf
Iohvidov [2J, Iohvidov and Krein [1J, [2].
The results of Section 9 are due to Naimark [1J (d. also Helton [2J)
Extensions to Krein spaces have been given by Langer [5J, [12J anc
Helton [1].
Much of the theory of indefinite inner product spaces, especially 0:
Pontrjagin spaces, was modelled in a finite-dimensional non-degenerate
space and presented as a series of problems by Glazman and Ljubic
[1; Chapter X].
Bognar [1J investigated square roots of selfadjoint operators in Ilk
His results were carried over to more general functions by Pe'lsahovic
[1]. Operators of the form T*T in Ilk have been considered by Holevc
[lJ as well as by Bognar and Kramli [1J. For other special questiom
concerning seHad ioint operators in Ilk d. Bognar [3J - [6J, Lo [2].
Shah [1] and independently, N aimark [12J generalized Stone' E
theorem to one-parameter unitary groups in Ilk.
Notes to Chapter IX 209
Representations by unitary groups and symmetric algebras in Ilk

as well as reductions and canonical models of the latter have been stud-
ied by Na'lmark [2J-[11J, Ismagilov [1J-[4], Langer [7J, Liberzon
and Sul'man [1J, Loginov [1]-[3], Sul'man [1]. Cf. also the survey
articles NaImark [13], NaImark and Ismagilov [1].
Bognar [8] gave a characterization of the mapping T -+ T*
(T E 'i8(Ilk )) regarded as an operation in the algebra of all continuous
linear operators on a Banach space.
Iohvidov [2J, Azizov and Iohvidov [1J, Azizov [2J found conditions
in order that the principal vectors of a compact selfadjoint or compact
"dissipative" operator in Ilk form a complete system.
The theory of Pontrjagin spaces has been applied to obtain integ-
ral representations, extensions and other properties of sequences,
matrices, functions and kernels with a finite rank of indefiniteness in
the following papers: KreIn [3], [5J, Iohvidov [3J, [13J, Iohvidovand
KreIn [2], DaleckiI [1], Shah [1J, V. 1. Gorbacuk [1]-[7J, V.1.
Gorbacuk and M. L. Gorbacuk [1].
Applications of Pontrjagin spaces to concrete problems of mechanics
have been given by S. G. KreIn and Moiseev [1J, Sobolev [1], M. L.
Gorbacuk, Slepcova and Temcenko [1J, to ortho§onal polynomials by
M. G. KreIn [9], and to numerical methods by Gerisch and Gerisch [1J.
Bibliography
Arons, M. E., Han, M. Y., Sudarshan, E. C. G.: [1J Finite quantum electrodynamics:
a field theory using an indefinite metric. Phys. Rev. (2) 137, B 1085-B 1104
(1965).
Aronszajn, N.: [1] Quadratic forms on vector spaces. In: Proc. Internat. Sympos.
Linear Spaces, pp. 29-87. Jerusalem and Oxford: Jerusalem Academic Press
and Pergamon 1961.
Azizov, T. J a.: [1] The spectra of certain operator classes in Hilbert space. Mat.
Zametki 9,303-310 (1971) [Russian].
[2] Invariant subspaces and criteria of completeness for the system of root
vectors of J-dissipative operators in the Pontrjagin space II,.. Dokl. Akad.
Nauk SSSR 200,1015-1017 (1971) [Russian].
Azizov, T. J a., Iohvidov, 1. S.: [1J A criterion, in order to form a complete system
or a basis, for the root vectors of a completely continuous J-selfadjoint operator
in the Pontrjagin space II,.. Mat. Issled. 6, no. 1, 158-161 (1971) [Russian].
[2] Linear operators in Hilbert spaces with a G-metric. Uspehi Mat. Nauk 26,
no. 4, 43-92 (1971) [Russian].
Berezin, F. A.: [1] On the Lee model. Mat. Sb. 60, 425-446 (1963) [RussianJ.
Berge, C.: [1] Espaces topologiques: fonctions multivoques, Paris: Dunod 1959.
Bleuler, K.: [1] Eine neue Methode zur Behandlung der longitudinalen und skalaren
Photonen. Helvetica Phys. Acta 23,567-586 (1950).
Bognar, J. (= Bognar, Ja.): [1] On the existence of square roots of an operator
which is self-adjoint with respect to an indefinite metric. Magyar Tud. Akad.
Mat. Kutat6 Int. K6zl. 6, 351-363 (1961) [Russian].
[2] On a discontinuity property of the inner product in spaces with indefinite
metric. Uspehi Mat. Nauk 17, no. 1, 157-159 (1962) [Russian].
[3] Non-negativity properties of operators in spaces with indefinite metric.
Ann. Acad. Sci. Fenn. Ser. A I, no. 336/10 (1963).
[4J Certain relations among the non-negativity properties of operators in spaces
with an indefinite metric. Magyar Tud. Akad. Mat. Kutat6 Int. K6zl. 8,
201-212 (1963) [Russian].
[5] Certain relations among the non-negativity properties of operators in spaces
with an indefinite metric. II. Studia Sci. Math. Hungar. 1, 97-102 (1966)
[Russian].
[6] Certain relations among the non-negativity properties of operators in spaces
with an indefinite metric. III. Studia Sci. Math. Hungar. 1,419-426 (1966)
[Russian].
[7] On decomposition majorants of an indefinite metric. Math. Z. 101, 65-67
(1967).
Bibliography 211
Bognar, J.: [8J Involution as operator conjugation. In: Colloquia Math. Soc.
Janos Bolyai, Vol. 5, Hilbert space operators and operator algebras, pp. 53 - 64.
Amsterdam/London: North-Holland 1972.
[9J A remark on doubly strict plus-operators. Mat. Issled. (to appear) [Russian}.
Bognar, J., Kramli, A.: [1] Operators of the form C*C in indefinite inner product
spaces. Acta Sci. Math. (Szeged) 29,19-29 (1968).
Bogoljubov, N. N., Medvedev, B. V., Polivanov, M. K.: [1] On the question of an
indefinite metric in quantum field theory. Naucnye Doklady VySSel Skoly,
Fiz.-Mat. Nauki (1958), no. 2,137-142 (1958) [RussianJ.
Bonsall, F. F.: [1] Indefinitely isometric linear operators in a reflexive Banach
space. Quart. J. Math. Oxford Ser. (2) 6, 179-187 (1955).
Bourbaki, N.: [1] Elements de mathematique. XV, XVIII, XIX. Espaces vec-
toriels topologiques. Actualites Sci. Ind., nos. 1189, 1229, 1230. Paris: Hermann
1953 and 1955.
Brodskil, M. L.: [1] On properties of operators mapping the non-negative part of
a space with indefinite metric into itself. Uspehi Mat. Nauk 14, no. 1, 147-152
(1959) [RussianJ.
Brodskil, V. M.: [1] Operator colligations and their characteristic functions. Dokl.
Akad. Nauk SSSR 198,16-19 (1971) [Russian].
Browder, F. E.: [1] A remark on the Dirichlet problem for non-elliptic self-adjoint
partial differential operators. Rend. Circ. Mat. Palermo (2) 6,249-253 (1957).
- [2J On the Dirichlet problem for linear non-elliptic partial differential equations.
II. Rend. Circ. Mat. Palermo (2) 7,303-308 (1958).
Cordes, H. 0.: [lJ On maximal first order partial differential operators. Amer. J.
Math. 82, 63-91 (1960).
Crandall, M. G., Phillips, R. S.: [1] On the extension problem for dissipative
operators. J. Functional Analysis 2,147-176 (1968).
Daleckil, J. u. L.: [lJ Differentiation of non-hermitian matrix functions depending
on a parameter. Izv. Vyss. Ucebn. Zaved. Matematika 2, 52-64 (1962)
[Russian].
Daleckil, J u. L., Fadeeva, E. A. : [lJ Hyperbolic equations with operator coeffi-
cients, and ultra-parabolic systems. Ukrain.Mat. Z. 24,92-95 (1972) [RussianJ.
Daleckil, Ju. L., KreIn, M. G.: [lJ The stability of the solutions of differential
equations in a Banach space, Moscow: Nauka 1970 [RussianJ.
Davis, Ch.: [lJ J-unitary dilation of a general operator. Acta Sci. Math. (Szeged)
31,75-86 (1970).
- [2J Dilation of uniformly continuous semi-groups. Rev. Roumaine Math.
Pures Appl. 15, 975-983 (1970).
Davis, Ch., Foia~, C.: [lJ Operators with bounded characteristic function and their
J-unitary dilation. Acta Sci. Math. (Szeged) 32,127-139 (1971).
Dirac, P. A. M.: [lJ The physical interpretation of quantum mechanics. Proc. Roy.
Soc. London Ser. A 180,1-40 (1942).
Dolph, C. L.: [1] Recent developments in some non-self-adjoint problems of
mathematical physics. Bull. Amer. Math. Soc. 67,1-69 (1961).
Dunford, N., Schwartz, J .T.: [lJ Linear operators. I. General theory, New York/
London: Interscience 1958.
Eisenfeld, J.: [1] On symmetrization and roots of quadratic eigenvalue problems.
J. Functional Analysis 9, 410-422 (1972).
212 Bibliography
Fan, K.: [1J Fixed-point and minimax theorems in locally convex topological linear
spaces. Proc. Nat. Acad. Sci. U.S.A. 38,121-126 (1952).
[2J Invariant subspaces of certain linear operators. Bull. Amer. Math. Soc. 69.
773-777 (1963).
[3J Invariant cross-sections and invariant linear subspaces. Israel J. Math. 2,
19-26 (1964).
[4J Invariant subspaces for a semi group of linear operators. Indag. Math. 27,
447-451 (1965).
[5J Applications of a theorem concerning sets with convex sections. Math.
Ann. 163, 189-203 (1966).
Fischer, H. R., Gross, H.: [1] Quadratic forms and linear topologies. I. Math. Ann.
157,296-325 (1964).
Gerisch, A. (= Geris, A.), Gerisch. vI'. (= Geris, V.): [1J Pontrjagin's space and
convergence of the Bubnov-Galerkin method. Dohl. Akad. Nauk SSSR 193.
1218-1221 (1970) [Russian].
Ginzburg, Ju. P.: [1] On J-contractive operator functions. Dokl. Akad. Nauk
SSSR 117, 171-173 (1957) [Russian].
[2] On J-contractive operators in Hilbert space. Odess. Gos. Ped. Inst. Nauen.
Zap. Fiz.-Mat. Fak. 22, no. 1, 13-20 (1958) [Russian].
[3] Subspaces of a Hilbert space with indefinite metric. Odess. Ped. lnst.
Nauen. Zap. Kaf. Mat. Fiz. Estestv. 25. no. 2,3-9 (1961) [Rnssian].
[4] Projections in a Hilbert space with bilinear metric. Dok!. Akad. Nauk
SSSR 139.775-778 (1961) [Russian].
Ginzburg. Ju. P., Iohvidov.1. S.: [1] A study of the geometry of infinite-dimensional
spaces with bilinear metric. Uspehi Mat. Nauk 17, no.4. 3- 56 (1962) [Russian].
Glazman.1. M .• Ljubie. Ju. 1.: [1] Finite-dimensional linear analysis. Moscow:
Nauka 1969 [Russian].
Glicksberg,1. L.: [1] A further generalization of the Kakutani fixed point theorcm.
with application to Nash equilibrium points. Proc. ArneI'. Math. Soc. 3.170-174
(1952).
Gorbaeuk. M. L.. Slepcova. G. P., Temeenko, M. E.: [1] Stability of motion of a
rigid body suspended on a string and filled with fluid. Ukrain. Mat. Z. 20.
586- 602 (1968) [Russian].
Gorbaeuk. V. 1. (= Pljuseeva, V.!.): [1] The integral representation of hermitian-
indefinite matrices with % negative squares. Ukrain. Mat. Z. 14. 30- 39 (1962)
[Russian].
[2J The integral representation of continuous hermitian-indefinite kernels.
Dohl. Akad. Nauk SSSR 145. 534-537 (1962) [Russian].
[3J The integral representation of hermitian-indefinite kernels (the case of several
variables). Ukrain. Mat. Z. 16.232-236 (1964) [Russian].
[4J The integral representation of hermitian-indefinite kernels. Ukrain. Mat. Z.
17. no. 3.43-58 (1965) [Russian].
[5] On the uniqueness of the representation of hermitian-indefinite functions
and sequences. Ukrain. Mat. Z. 18. no. 2,107-113 (1966) [Russian].
[6] Extensions of a real hermitian-indefinite function with one negative square.
Ukrain. Mat. Z. 19. no. 4.119-125 (1967) [Russian].
[7J Self-adjoint extensions of some Hermitian operators in a space with in-
definite metric. In: Colloquia Math. Soc. Janos Bolyai. Vo1.5. Hilbert space
operators and operator algebras. pp.265-269. Amsterdam/London: North-
Holland 1972.
Bibliography 213
Gorbacuk, V.1., Gorbacuk, M. L.: [lJ Representation of the vacuum-mean of field

operators in a space with an indefinite metric. Ukrain. Mat. Z. 18, no. 6,
108-111 (1966) [RussianJ.
Greub, W. H.: [1] Linear algebra, 2nd edition, New York and Berlin/G6ttingen/
Heidelberg: Academic Press and Springer 1963.
Gupta, S. N.: [1] Theory of longitudinal photons in quantum electrodynamics.
Proc. Phys. Soc. Sect. A 63,681-691 (1950).
Hackevic, V. A.: [1J Invariant subspaces for certain classes of linear operators in
normed spaces with an indefinite metric. Mat. Issled. 6, no. 3, 133-147 (1971)
[Russian].
Harazov, D. F.: [1] Symmetrizable operators that do not satisfy the conditions of
positive-definiteness, and their applications. Studia Math. 34, 241- 252 (1970)
[Russian].
Heisenberg, W.: [lJ Erweiterungen des Hilbert-Raums in der Quantentheorie del'
Wellenfelder. Z. Physik 144,1-8 (1956).
[2] Hilbert space II and the "ghost" states of Pauli and Kallen. Nuovo Cimento
(10) 4, supplemento, 743-747 (1956).
[3] Lee model and quantisation of non linear field equations. Nuclear Phys. 4,
532-563 (1957).
[4J Introduction to the unified field theory of elementary particles, London/
New York/Sydney: Interscience 1966.
Helton, J. W.: [1] Unitary operators on a space with an indefinite inner product.
J. Fnnctional Analysis 6,412-440 (1970).
- [2] Operators unitary in an indefinite metric and linear fractional transforma-
tions. Acta Sci. Math. (Szeged) 32, 261-266 (1971).
Hess, P.: [1] Zur Theorie der linearen Operatoren eines J-Raumes. Operatoren die
von kanonischen Zerlegungen reduziert werden. Math. Z. 106, 88- 96 (1968).
- [2] LJber Polynome J-symmetrischer Operatoren in J-Raumen. Math. Z. 114,
271-277 (1970).
Hestenes, M. R.: [1] Applications of the theory of quadratic forms in Hilbert space
to the calculus of variations. Pacific J. Math. 1, 525- 581 (1951).
Hildebrandt, S.: [1] Rand- und Eigenwertaufgabcn bei stark elliptischen Systemen
linearer Differentialgleichungen. Math. Ann. 148, 411 - 429 (1962).
Holevo, A. S.: [1] Generalization to spaces with an indefinite metric of a theorem
of von Neumann on the operator 1'*1'. Azerbalc1zan. Gos. Univ. Ucen. Zap.
Ser. Fiz.-Mat. Nauk (1965), no.2, 45-48 (1965) [Russian].
Iohvidov, 1. S.: [lJ Unitary operators in a space with an indefinite metric. Zap.
:iVIat. Otd. Fiz.-Mat. Fair. i Har'kov. Mat. Obsc. (4) 21, 79-86 (1949) [Russian].
[2] On the spectra of hermitian and unitary operators in a space with indefinite
metric. Dokl. Akad. Nauk SSSR 71,225-228 (1950) [Russian].
[3] On the theory of indefinite Toeplitz forms. Dokl. Akad. Nauk SSSR 101,
213-216 (1955) [Russian].
[4J Boundedness of J-isometric operators. Uspehi Mat. Nauk 16, no. 4, 167--I70
(1961 ) [Russian].
[5] Regular and projection-complete linear manifolds in spaces with a general
hermitian bilinear metric. Dokl. Akad. Nauk SSSR 139, 791-794 (1961)
[Russian].
[6] Singnlar linear manifolds in spaces with an arbitrary hermitian bilinear
metric. Uspehi Mat. Nauk 17, no. 4,127-133 (1962) [Russian].
214 Bibliography
Iohvidov, 1. S.: [7] Operators with completely continuous iterations. Dokl. Akad.
[8J Singular linear manifolds in the spaceII,.. Ukrain. Mat. Z. 16, 300-308
(1964) [RussianJ.
[9J On a lemma of Ky Fan generalizing the fixed-point principle of A. N.
Tihonov. Dokl. Akad. Nauk SSSR 159,501-504 (1964) [RussianJ.
[10J On maximal definite linear manifolds in a Hilbert space with a G-metric.
Ukrain. Mat. Z. 17, no. 4, 22-28 (1965) [Russian].
[llJ G-isometric and I-semiunitary operators in Hilbert space. Uspehi Mat.
Nauk20,no.3, 175-181 (1965) [Russian].
[12J Linear fractional transformations of I-contractive operators. Akad. Nauk
Armjan. SSR Dokl. 42,3-8 (1966) [Russian].
[13J Unitary extensions of isometric operators in the Pontrjagin space III and
continuations in the ~1 class of finite sequences of the class ~l;n. Dokl. Akad.
[14J Banach spaces with a I-metric and certain classes of linear operators in
these spaces. Bul. Akad. Stiince RSS Moldoven. 1, 60 - 80 (1968) [RussianJ.
[15J On a class of linear fractional operator transformations. Voronez. Gos.
Univ. Trudy Sem. Funkcional. Anal. 18-44 (1970) [RussianJ.
Iohvidov,1. S., Kreln, M. G.: [1J Spectral theory of operators in spaces with an
indefinite metric. I. Trudy Moskov. Mat. Obsc. 5,367-432 (1956) and 6, 486
(1957) [Russian].
[2J Spectral theory of operators in spaces with an indefinite metric. II. Trudy
Moskov. Mat. Obsc. 8, 413-496 (1959) and 1-5,452-454 (1966) [RussianJ.
Iohvidov, 1. S., Senderov, V. A.: [1] The bounded ness of I -semiunitary operators in
Banach spaces with a I-metric. Mat. Issled. 5, no. 4, 166-170 (1970)
[Russian].
Ismagilov, R. S.: [1J Description of the unitary representations of the Lorentz

group in a space with indefinite metric. Dok!. Akad. Nauk SSSR 158, 268-270
(1964 ) [Russian].
[2J Unitary representations of the Lorentz group in a space with indefinite
metric. Izv. Akad. Nauk SSSR SeT. Mat. 30, 497- 522 (1966) [Russian].
[3J Irreducible representations of the discrete group 5 L (2, P) that are unitary
with respect to an indefinite metric. 1zv. Akad. Nauk SSSR Ser. Mat. 30,
923 - 950 (1 966) [Russian].
[4J Rings of operators in a space with an indefinite metric. Dok!. Akad. Nauk
SSSR 171, 269-271 (1966) [RussianJ.
Jalava, V.: [lJ On spectral decompositions of operators in I-space. Ann. Acad.

Sci. Fenn. Ser. A I, no. 446 (1969).
[2J On operators in a linear space with a non-degenerate sesquilinear form.
Univ. JyvaskyUi. Dept. Math., Report 5 (1969).
[3J On the square root of a self-adjoint operator in I-space. Ann. Acad. Sci.
Fenn. Ser. A I, no. 468 (1970).
Jarchow, H.: [lJ Stetigkeit hermitescher Formen. Ann. Acad. Sci. Fenn. Ser. A I,
no. 441 (1969).
- [2J Topologisch stetige hermitesche Formen. Math. Z. 113, 326- 334 (1970).
Jelinek, J., Virsik, J.: [1] Pseudo-unitary spaces. Casopis Pest. Mat. 91, 18-33
(1966).
Bibliography 215
Jonas, P.: [1] Eine Bedingung fUr die Existenz einer EigenspektraUunktion fur ge-
wisse Automorphismenlokalkonvexer Raume. Math. Nachr. 45, 143-160(1970).
Kallen, G., Pauli, W.: [IJ On the mathematical structure of T. D. Lee's model of a
renormalizablc field theory. Danske Vid. Selsk. lVIat.-Fys. lVIedd. 30, no. 7
(1955).
Karrer, G.: [IJ Spektraltheorie der Automorphismen Hermite'scher Formen. Ann.
Acad. Sci. Fenn. Ser. A I, no. 237 (1957).
Kothe, G.: [1] Topologische lineare Raume, Vol. 1, Berlin/Gottingen/Heidelberg:
Springer 1960.
Kraljevic, H.: [1] Simultaneous diagonalisation of two symmetric bilinear func-
tionals. Glasnik Mat. Ser. III 1, 57-63 (1966).
KreIn, lVI. G.: [IJ On weighted integral equations the distribution functions of which
are not monotonic. In: Memorial volume dedicated to D. A. Grave, pp. 88-103.
Moscow/Leningrad: Gostehizdat 1940 [Russian].
[2J Completely continuous linear operators in function spaces with two norms.
Akad. Nauk Ukrain. RSR. Zbirnik Prac' Inst. Mat. no. 9, 104-129 (1947)
[UkrainianJ .
[3J Helices in the infinite-dimensional Lobacevskil space. Uspehi Mat. Nauk 3,
no.3, 158-160 (1948) [Russian].
[4J An application of the fixed-point principle in the theory of linear transforma-
tions of spaces with an indefinite metric. Uspehi Mat. Nauk 5, no. 2,180-190
(1950) [Russian].
[5J Integral representation of a continuous hermitian,indefinite function with a
finite number of negative squares. Dokl. Akad. N auk SSSR 125, 31 - 34 (1959)
[Russian].
[6J Lectures on the theory of the stability of solutions of differential equations
in a Banach space. Kiev: Akad. Nauk Ukrain. SSR Inst. lVIat. 1964 [RussianJ.
[7J A new application of the fixed-point principle in the theory of operators in a
space with indefinite metric. Dokl. Akad. Nauk SSSR 154, 1023-1026 (1964)
[Russian].
[8J Introduction to the geometry of indefinite I-spaces and to the. theory of
operators in those spaces. In: Second mathematical summer school, Part I,
pp. 15-92. Kiev: Naukova Dumka 1965 [Russian].
[9J Distribution of roots of polynomials orthogonal on the unit circle with respect
to a sign-alternating weight. Teor. Funkcil Funkcional. Anal. i Prilozen. no. 2,
131-137 (1966) [Russian].
KreIn, M. G., Langer, H. (= Langer, G. K.) : [1] On the spectral function of a self-
adjoint operator in a space with indefinite metric. Dokl. Akad. Nauk SSSR 152,
39-42 (1963) [Russian].
[2J On the theory of quadratic pencils of self-adjoint operators. DokL Akad.
Nauk SSSR 154,1258-1261 (1964) [RussianJ.
[3J Certain mathematical principles of the linear theory of damped vibrations
of continua. In: Applications of the theory of functions in continuum mechanics,
Vol. II, pp.233-322. iYIoscow: Nauka 1965 [Russian].
[4J The defect subspaces and generalized rcsolvents of a hermitian operator in
the spaccII,.. FunkcionaL Anal. i Prilozen. 5, no.2, 59-71 and no. 3, 54-69
(1971) [Russian].
[5J Dber die verallgemeinerten Resolventen und die characteristische Funktion
eines iSOluetrischen Operators im Raume II". In: Colloquia Math. Soc. Janos
Bolyai, VoL 5, Hilbert space operators and operator algebras, pp. 353-399.
Amsterdam/London : North-Holland 1972.
216 Bibliography
Krem, M. G., Rutman, M. A.: [1] Linear operators leaving invariant a cone in a
Banach space. Uspehi Mat. Nauk 3, no. 1, 3-95 (1948) [Russian].
Krei:n, M. G., Smul'jan, Ju. L.: [1] Plus-operators in a space with an indefinite
metric. Mat. Issled. 1, no. 1, 131-161 (1966) [Russian].
[2] i-polar representations of plus-operators. Mat. Issled. 1, no. 2, 172-210
(1966) [Russian].
[3] Linear fractional transformations with operator coefficients. Mat. Issled. 2,
no. 3, 64-96 (1967) [Russian].
Krei:n, S. G., Moiseev, N.N.: [1] On oscillations of a vessel containing a liquid with
a free surface. Prikl. Mat. Meh. 21, 169-174 (1957) [Russian].
Kuhne, R.: [1) Uber eine Klasse ]-selbstadjungierter Operatoren. Math. Ann. 154,
56-69 (1964).
- [2] Minimaxprinzipe fur stark gedampfte Scharen. Acta Sci. Math. (Szeged) 29,
39-68 (1968).
Kuzel', A. V.: [1] The spectral analysis of quasi-unitary operators in a space with
indefinite metric. Teor. Funkcii Funkcional. Anal. i Prilozen. no. 4, 3-27
(1967) [Russian).
Langer, H. (= Langer, G. K.): [1] On i-hermitian operators. Dokl. Akad. Nauk
SSSR 134,263-266 (1960) [Russian].
[2] Zur Spektraltheoric ]-sclbstadjungierter Operatoren. Math. Ann. 146,
60-85 (1962).
[3J Eine Verallgemeinerung eines Satzes von L. S. Pontrjagin. Math. Ann. 152,
434-436 (1963).
[4] Eine Erweiterung der Spurformel der St6rungstheorie. Math. Nachr. 30,
123-135 (1965).
[5] Invariant subspaces of linear operators acting in a space with indefinite
metric. Dokl. Akad. Nauk SSSR 169,12-15 (1966) [Russian].
[6] Spektralfunktionen einer Klasse ]-selbstadjungierter Operatoren. Math.
Nachr. 33, 107-120 (1967).
[7] Uber einen Satz von M. A. Neumark. Math. Ann. 175,303-314 (1968).
[8] Uber stark gedampfte Scharen im Hilbertraum. J. Math. Mech. 17, 685 -705
(1968).
[9] Uber die schwache Stabilitat linearer Differentialgleichungen mit periodi-
schen Koeffizienten. Math. Scand. 22, 203-208 (1968).
[10] A remark on invariant subspaces of linear operators in Banach spaces with
an indefinite metric. Mat. Issled. 4, no. 1, 27 - 34 (1969) [Russian].
[11] Maximal dual pairs of invariant subspaces of i-selfadjoint operators. Mat.
Zametki 7,443-447 (1970) [Russian].
[12] Invariante Teilraume definisierbarer ]-selbstadjungierter Operatoren.
Ann. Acad. Sci. Fenn. Ser. A I, no. 475 (1971).
[13] Verallgemeinerte Resolvcnten eines ]-nichtnegativen Opcrators mit end-
lichem Defekt. J. Functional Analysis 8, 287-320 {1971}.
[14] Generalized co-resolvents of a n-isometric operator with unequal defect
numbers. Funkcional. Anal. i Prilozen. 5, no. 4,73-75 (1971) [Russian].
[15] Zur Spektraltheorie verallgemeinerter gew6hnlicher Difierentialoperatoren
zweiter Ordnung mit einer nichtmonotonen Gewichtsfunktion. Univ. Jyvaskyla
Dept. Math., Report 14 (1972).
Larionov, E. A.: [1] A commutative family of operators in a space with indefinite
metric. Mat. Zametki 1, 589- 594 (1967) [Russian].
Bibliography 217
Larionov, E. A.: [2] The extension of dual subspaces. Dokl. Akad. Nauk SSSR
176,515-517 (1967) [RussianJ.
[3J The extension of dual subspaces invariant under an algebra. Mat. Zametki 3,
253-260 (1968) [Russian].
[4J Nilpotent I-selfadjoint operators. Dohl. Akad. Nauk SSSR 183, 768-771
(1968) [Russian].
[5J Selfadjoint quadratic pencils. Izv. Akad. Nauk SSSR Ser. Mat. 33,138-154
(1969) [RussianJ.
Lax, P. D., Phillips, R. S.: [1] The acoustic equation with an indefinite energy form
and the Schr6dinger equation. J. Functional Analysis 1, 37- 83 (1967).
[2J Scattering theory, New York/London: Academic Press 1967.
[3J Decaying modes for the wave equation in the exterior of an obstacle. Comm.
Pure Appl. Math. 22,737-787 (1969).
Lee, T. D., Wick, G. C.: [1] Negative metric and the unitarity of the S-matrix.
Nuclear Phys. B 9, 209-243 (1969).
Liberzon, V. 1., Surman, V. S.: [1] Operator-irreducible symmetric operator alge-
bras in the Pontrjagin space IJI. Izv. Akad. Nauk SSSR Ser. Mat. 35,
1159-1170 (1971) [Russian].
Littman, 'V.: [1] Remarks on the Dirichlet problem for general linear partial dif-
ferential equations. Comm. Pure Appl. Math. 11, 145-151 (1958).
Lo, c.-y.: [1J A class of polynomials in self-adjoint operators in spaces with an
indefinite metric. Canad. J. Math. 20, 673-678 (1968).
- [2J On polynomials in self-adjoint operators in spaces with an indefinite metric.
Trans. Amer. Math. Soc. 134,297-304 (1968).
Loginov, A. 1.: [1J Semidegenerate algebras in a Pontrjagin space. Mat. Zametki 6,
73-80 (1969) [Russian].
[2J Commutative symmetric operator algebras in a Pontrjagin space. Izv. Akad.
Nauk SSSR Ser. Mat. 33,549-569 (1969) [Russian].
[3J Complete commutative symmetric operator algebras in the Pontrjagin
space III' Mat. Sb. 84,575-582 (1971) [RussianJ.
Louhivaara, 1. S.: [1] Uber das erste Randwertproblem fur die Differentialgleichung
u xx + U yy + q U + f = O. Ann. Acad. Sci. Fenn. Ser. A I, no. 183 (1955).
[2J Uber das zweite und dritte Randwertproblem fur die Differentialgleichung
u xx + U yy + q U + f = O. Ann. Acad. Sci. Fenn. Ser. A I, no. 203 (1955).
[3J Uber das Dirichletsche Problem fur die selbstadjungierten linearen partiellen
Differentialgleichungen zweiter Ordnung. Rend. Circ. Mat. Palermo (2) 5,
260-274 (1956).
[4J Bemerkung zur Theorie der Nevanlinnaschen Raume. Ann. Acad. Sci.
Fenn. Ser. A I, no. 232 (1956).
[5J Zur Theorie der Unterraume in linearen Raumen mit indefiniter Metrik.
Ann. Acad. Sci. Fenn. Ser. A I, no. 252 (1958).
[6J Uber verschiedene Metriken in linearen Raumen. Ann. Acad. Sci. Fenn.
Ser. A I, no. 282 (1960).
[7J Uber die neuere Entwicklung der Theorie der linearen Raume mit indefiniten
Bilinearformen. In: Festband 70. Geburtstag R. Nevanlinna, pp.66-81-
Berlin/Heidelberg/New York: Springer 1966.
Mal'cev, A. 1.: [1] Foundations of linear algebra, San Francisco/London: Freeman
1963.
Marksj6, B., Textorius, B.: [1] On the stability of linear differential equations in
spaces with an indefinite metric. Math. Scand. 20, 177 -192 (1967).
218 Bibliography
Masuda, K.: [lJ On the existence of invariant subspaces in spaces with indefinite
metric. Proc. Amer. Math. Soc. 32, 440-444 (1972).
Mnrray, F. J.: [lJ On complementary manifolds and projections in spaces Lp and lp.
Trans. Amer.Math. Soc. 41,138-152 (1937).
Nagy, K. L.: [n Indefinite metric in quantum field theory. Nuovo Cimento (10) 17,
supplemento, 92-131 (1960).
[2J State vector spaces with indefinite metric in quantum field theory, Grollin-
gen and Budapest: Noordhoff and Akademiai Kiad6 1966.
[3J Complex poles, cuts, indefinite metric and unitarity. Acta Phys. Acad. Sci.
Hungar. 29, 251-265 (1970).
Narmark, M. A.: [1] On commuting unitary operators in spaces with indefinite
metric. Acta Sci. Math. (Szeged) 24, 177-189 (1963).
[2J Unitary representations of solvablc groups in spaces with indefinite metric.
Izv. Akad. Nauk SSSR Ser. Mat. 27, 1181-1185 (1963) [Russian].
[3J Commutative algebras of operators in the space Ill' Rev. Roumaine Math.
Pures App!. 9, 499-528 (1964) [Russian].
[4J Unitary representations of the Lorentz group in spaces with indefinite
metric. Mat. Sb. 65,198-211 (1964) [RussianJ.
[5J On unitary group representations in spaces with indefinite metric. Acta
Sci. Math. (Szeged) 26,201-209 (1965).
[6J Kommutativc symmetrische Operatorenalgebren in Pontryaginschen Ran-
menilk . Math. Ann. 162, 147-171 (1965) and 170, 166 (1967).
[7J On the structure of the unitary representations of locally compact groups in
the space Ill' Izv. Akad. Nauk SSSR Ser. Mat. 29, 689-700 (1965) [RussianJ.
[8J Conditions for the unitary equivalence of commutative symmetric algebras
in the Pontrjagin space Ilk' Trudy Moskov. Mat. Obse. 15,383-399 ('1966)
[Russian].
[9J Structure of unitary representations of locally compact groups and sym-
metric representations of algebras in the Pontrjagin space II". Izv. Akad. Nauk
SSSR Ser.Mat. 30,1111-1132 (1966) [Russian].
[10J Degenerate operator algebras in the Pontrjagin space Ilk' Izv. Akad.
Nauk SSSR Ser. Mat. 30,1229-1256 (1966) [Russian].
[llJ Representations of commutative symmetric Banach algebras and commu-
tative topological groups in the space II". Dokl. Akad. Nauk SSSR 170, 271-274
(1966) [Russian J.
[12J Analog of Stone's theorem for a space with an indefinite metric. Dokl.
Akad. Nauk SSSR 170,1259-'1261 (1966) [Russian].
[13J On unitary group representations and symmetric algebra representations
in spaces with indefinite metric. In: Proceedings of the symposium in analysis,
pp.145-156. Kingston, Ontario: Queen's University 1967.
Nalmark, M. A., Ismagilov, R. S.: [lJ Representations of groups and algebras in
a space with indefinite metric. In: Itogi N auki, Mathematical analysis 1968,
pp. 73-105. Moscow: Akad. Nauk SSSIUnst. Nauen. Informacii 1969 [Russian].
Nevanlinna, R.: [1] Erweiterung der Theorie des Hilbertschen Raumes. Comm.
Sem. Math. Univ. Lund, Tome Supplementaire, 160-168 (1952).
[2J Dber metrische lineare Raume. II. Bilinoarformen und Stetigkeit. Ann.
[3J Dber metrische Iineare Raume. III. Theorie dor Orthogonalsysteme. Ann.
[4J Dber metrische lineare RauIne. IV. Zur Theorie der Unterraume. Ann.
Bibliography 219
Noel, G.: [1] Topologies sur un vectoriel hermitien non degenere. c. R. Acad. Sci.
Paris 257, 2785-2787 (1963).
- [2] Operateurs fortement (f)-normaux dans un espace de type (f). Acad. Roy.
Belg. Bull. Cl. Sci. (5) 51, 570- 585 (1965).
Olubummo, A., Phillips, R. S.: [1] Dissipative ordinary differential operators.
J. Math. Mech. 14, 929-949 (1965).
Ovcinnikov, V. 1.: [1] The decomposability of spaces with indefinite metric. Mat.
Issled. 3, no. 4,175-177 (1968) [Russian].
Pauli, W.: [1] On Dirac's new method of field quantization. Rev. Modern Phys. 15,
175-207 (1943).
Pelsahovic, E. E.: [1] Sufficient conditions for the existence of a solution of the
equation A = fiB) for selfadjoint operators in a space with indefinite metric.
Vestnik Moskov. Univ. Ser.I Mat. Meh. 21, no. 4, 47-53 (1966) [Russian].
Pesonen, E.: [1] Dber die Spektraldarstellung quadratischer Formen in linearen
Raumen mit indefiniter Metrik. Ann. Acad. Sci. Fenn. Ser. A I, no. 227 (1956).
Phillips, R. S.: [1] Dissipative operators and parabolic partial differential equations.
Comm. Pure Appl. Math. 12,249-276 (1959).
[2] Dissipative operators and hyperbolic systems of partial differential equa-
tions. Trans. Amer. Math. Soc. 90, 193-254 (1959).
[3] The extension of dual subspaces invariant under an algebra. In: Proc.
Internat. Sympos. Linear Spaces, pp. 366- 398. Jerusalem and Oxford: J erusa-
lem Academic Press and Pergamon 1961.
[4] A minimax characterization for the eigenvalues of a positive symmetric
operator in a space with an indefinite metric. J. Fac. Sci. Univ. Tokyo Sect.
I A Math. 17, 51- 59 (1970).
Phillips, R. S., Sarason, L.: [1] Singular symmetric positive first order differential
operators. J. Math. Mech. 15,235-271 (1966).
Pontrjagin, L. S.: [1] Hermitian operators in spaces with indefinite metric. Izv.
Akad. Nauk SSSR Ser. Mat. 8, 243-280 (1944) [Russian].
Potapov, V. P.: [1] The multiplicative structure of J-contractive matrix functions.
Trudy Moskov. Mat. Obsc. 4, 125-236 (1955) [Russian].
Reid, W. T.: [1] Symmetrizable completely continuous linear transformations in
Hilbert space. Duke Math. J. 18,41-56 (1951).
Riesz, F., Sz.-Nagy, B.: [1] Le90ns d'analyse fonctionnelle, 4 e edition, Paris and
Budapest: Gauthier-Villars and Akademiai Kiad6 1965.
Robertson, A. P., Robertson, W.: [1] Topological vector spaces, New York: Cam-
bridge University Press 1964.
Savage, L. J.: [1] The application of vectorial methods to metric geometry. Duke
Math. J. 13, 521-528 (1946).
Schaefer, H. H.: [1] Topological vector spaces, New York and London: Macmillan
and Collier-Macmillan 1966.
Scheibe, E.: [1] Dber hermitische Formen in topologischen Vektorraumen.1. Ann.
Senderov, V. A.: [1] Operators that are absolute indefinitely bounded from below in
spaces with indefinite metric. Mat. Zametki 10, 301-305 (1971) [Russian].
Shah Tao-shing: [1] On conditionally positive-definite generalized functions. Sci.
Sinica 11, 1147-1168 (1962).
220 Bibliography
Smul'jan, Ju. L.: [1] Contractive operators in a finite-dimensional space with

indefinite metric. Uspehi Mat. Nauk 18, no. 6, 225-230 (1963) [Russian].
[2] Division in the class of I-expansive operators. Mat. Sb. 74,516- 525 (1967)
[Russian].
[3J I-expansive operators in I-spaces. Ukrain. Mat. Z. 20, 352-362 (1968)
[RussianJ.
[4J I-majorizing and modular operators in I-spaces. Mat. Issled. 3, no. \,
198-214 (1968) [Russian].
[5J Linear fractional transformations with operator coefficients, and operator
balls. Mat. Sb. 77,335-353 (1968) [Russian].
[6J Linear fractional transformations of the upper half plane of operators. Izv.
Vyss. Ucebn. Zaved. Matematika (1969), no. 1, 97-105 (1969) [Russian].
[7] Linear fractional transformations in a space with involution. Izv. Vyss.
Ucebn. Zaved. Matematika (1969), no. 2, 117 -126 (1969) [Russian].
[8] A certain class of holomorphic operator-valued functions. Mat. Zametki 5,
351- 359 (1969) [RussianJ.
Sobolev, S. L.: [1] The motion of a symmetric top containing a cavity filled with a
liquid. Z. Prikl. Meh. i Tehn. Fiz. (1960), no. 3, 20-55 (1960) [Russian].
Sorjonen, P.: [1] Verallgemeinerte Resolventen eines symmetrischen Operators im
Pontrjaginraum. Univ. JyvaskyHi. Dept. Math., Report 15 (1972).
Sul'man, V. S.: [1] Operator algebras in the space III with indefinite metric. Dokl.
Abd. Nauk SSSR 201,44-47 (1971) [Russian].
Svarcman, P. A.: [1] Inequalities for the eigenvalues of I-hermitian and I-unitary
operators. I. Mat. Issled. 4, no. 4,33-45 (1969) [Russian].
Sz.-Nagy, B.: [lJ On uniformly bounded linear transformations in Hilbert space.
Acta Sci. Math. (Szeged) 11, 152-157 (1947).
Sz.-Nagy, B., FOial?, C.: [1] Harmonic analysis of operators on Hilbert space,
Amsterdam/London and New York and Budapest: North-Holland and American
Elsevier and Akademiiti Kiad6 1970.
~Wendland, W.: [lJ Die Fredholmsche Alternative fUr Operatoren, die bezuglich
eines bilinearen Funktionals adjungiert sind. Math. Z. 101, 61 - 64 (1967).
Wittstock, G.: [lJ Uber koerzive indefinite Metriken. Ann. Acad. Sci. Fenn. Ser.
AI, no. 347 (1964).
[2J Uber Zerlegungsmajoranten indefiniter Metriken. Math. Z. 91, 421-430
(1966).
[3J Uber Majoranten indefiniter Bilinearformen. Ann. Acad. Sci. Fenn. Ser. A I,
no. 381 (1966).
[4J Uber invariante Teilraume zu positiven Transformationen in Raumen mit
indefiniterMetrik. Math. Ann. 172, 167-175 (1967).
[5J Uber indefinit symmetrisierbare lineare Abbildungen. Math. Z. 111,
131-144 (1969).
Wonenburger, M. J.: [1] Simultaneous diagonalization of symmetric bilinear forms.
J. Math. Mech. 15, 617-622 (1966).
Index of Terms
A -fundamental decomposition 36 convergent 102

A-inner product 36 critical point 179
A-isometric operator 36
A-isotropic part 36 decomposable space 24
A-orthogonal 36 decomposition majorant 88
companion 36 definite inner product 5
- direct sum 36 - inner product space 5
- sum 36 - subspace 6
A-positive subspace 36 degenerate inner product 9
- vector 36 - inner product space 9
A-symmetric operator 36 - subspace 9
adjoint operator 121 dense 102
admissible topology 65 dimension 2, 102
algebraic multiplicity 29 diminishing operator 162
alternating extension 115 direct sum (of operators) 30
- maximal pair 115 - sum (of subspaces) 2
- pair 114 dissipative operator 116
angular operator 54 domain 28
anti-space 6 doubly strict plus-operator 158
augmenting operator 162 dual companion (for a subspace) 21
companion (for a system) 21
Banach space with a J-metric 119
pair (of subspaces) 21
- topology 59
pair (of systems) 21
basis 2
cartesian product 8 eigenspace 29

Cayley transform 39 eigenvalue (of a quadratic pencil) 172
- transformation 39 - (of an operator) 29
closed (set) 102 - in a subspace 30
- operator 121 eigenvector 29
closure (of a set) 102 elementary solution 172
- (of an operator) 121 equivalent semi-norms 59
co dimension 3, 102
commuting operators 30 Frechet topology 59
compact operator 120 fundamental decomposition 24
- set 120 - projector 50
complementary subspace 2 - symmetry 52
complete system 82 fundamentally reducible operator 163
completely invertible operator 29
continuous 102 Gram operator 89
- spectrum 121 graph 121
222 Index of Terms
Hilbert topology 59 maximal definite subspace 12

homeomorphic 102 dissipative operator 117
hypermaximal neutral 5U bspace 15 negative definite subspace 'l2
negative subspace 12
identity operator 28 neutral subspace 12
indefinite inner product 4 non-degenerate subspace 12
- inner product space 4 positive definite subspace 'l2
- subspace 5 positive subspace 12
induced operator 30 rectangular isometric operator 128
- topology 59 semi-definite subspace 12
inner product 3 uniformly negative subspace 112
- product space 3 uniformly positive subspace 112
- square 4 minimalmajorant 84
intrinsic completion 71 - unitary dilation 139
dimension 71
- norm 71 natural norm 102
- topology 71 negative definite inner product 5
intrinsically complete 71 definite inner product space 5
invariant subspace 29 definite subspace 6
inverse operator 29 inner product 5
invertible operator 29 inner product space
isometric operator 31 subspace 6
isometrical isomorphism 4 vector 4
isometrically isomorphic 4 neutral inner product 5
isomorphic 1 inner product space 5
isomorphism 1 part 4
isotropic part 9 set 4
- vector 9 subspace 6
vector 4
I-adjoint 122 non-strict plus-operator 46
I-inner product 52 non-uniformly definite subspace 109
I-isometric operator, normal eigenvalue 121
see A-isometric operator normed topology 59
I-norm 53 null space (of an operator) 28
I -orthogonal,
see A -orthogonal opera tor matrix 30
- companion, ortho-complemented 18
see A -orthogonal companion orthogonal 7
I-symmetric operator, companion 7
see A-symmetric operator direct sum 7
Jordan chain (of a quadratic pencil) 172 projector 36
- chain (of an operator) 29 sum 7
orthonormal system 81
Krein space 100 partial majorant 59
length (of a Jordan chain) 29 Pesonen operator 48
linear form 28 plus-defect (of a subspace) 107
- operator 28 - (of an operator) 157
linearly independent subspaces 2 plus-operator 45
- independent vectors 2 point spectrum (of a quadratic pencil)
locally convex topology 58 172
- spectrum (of an operator) 121
Mackey topology 96 polar (of a norm) 64
majorant 77 - (of a topology) 64
Index of Terms 223
polarization formula 4 second orthogonal companion 7

Pontrjagin space 184 selfadjoint operator 133
positive definite inner product self-polar topology 65
definite inner product space semi-definite inner product
definite subspace 6 - inner product space
inner product 5 - subspace 6
inner product space 5 semi-norm 58
operator 147 semi-simple eigenvalue 29
subspace 6 separable topology 59
vector 4 separated topology 59
positizabJe operator 180 singular subspace 71
positizing polynomial 180 span 1
principal subspace 29 spectral function 178, 180
- vector 29 spectrum (of a quadratic pencil) 174
proj ection 1 5 - (of an operator) 121
strict plus-operator 46
quadratic pencil 172
strong topology 102
- semi-norm 58
stronger (topology) 59
quadratic-normed topology 59
strongly damped 183
quasi-A-positive inner product 36
subspace 1
- inner product space 36
symmetric operator 34
quasi-negative inner product 25
- inner product space 25
topology defined by semi-norms 58
quasi-positive inner product 25
- inner product space 25 uniformly definite subspace 108
quotient space 3, 11 negative subspace 107
range 28 positive operator 151
rank (of an operator) 28 positive subspace 107
of A-negativity 149 unimodular number 31
of A-positivity 149 unitary dilation 139
of indefiniteness 95 - operator 128
of negativity 51, 95
of positivity 51, 95 vector space
rectangular isometric operator 128 - sum 2
rcducing direct decomposition 30
regular point (of an operator) 121 weak operator topology 120
- subspace 71 - topology 60
- symmetric extension 197 weaker (topology) 59
residual spectrum 121 weakest majorant 87
resolvent set (of a quadratic pcncil) 174
- set (of an operator) 121 zero operator 28
restriction (of an operator) 30 - subspace 1
Index of Symbols
IAI 140 ~~+ 36

<~{>, <WI' ... '~(n> 1 R 5
9I.L, 2{ J.l 7 lR(T) 28
\if 102 6,l(T) 29
W-2t 36 sgn A 140
91.1587 T-l 29
58({;\;1' Q;2)' 58 ({;\;) , 58 120 r 121
C 1 T* 121
codim(ij'£, codim£ 3, 102 TI£ 30
d+(£) 107 Ilxll' 64
d+(T) 157 IlxllJ 53
stJ(T) 28 Ixl2 71
dima: 2, 102 (x, Y) 3
dimint£ 71 (x, Y)A 36
(;1; 3,6 (x, Y)J 52
Q;0 9 x.1y7
{;\;+, (;1;- 24 x .1 .JY 36
(;1;A 36 u((;1;) 95
Q;A 36 u+({;\;), u-((;1;) 51,95
a:;, (;1;A 36 u:'1,(Sj), uA(Sj) 149
(;1;/(;1;0 11 !-,(T) 46
{;\;/£ 3, 11 /-1'.1 151
(;1;I X ... x (;1;n 8 v* 33, 131
Sj 100, 101 Ilk 184
I(ij', I 28 Q(L) 174
J 52 e(T) 121
J{+(£), I{-(£) 54 a(L) 174
~(Sj+, Sj-), ~ 156 arT) 121
£0 9 ac(T), ap(T), ar(T) 121
£136 ap(L) 172
£1+"'+£n 2 7:', 7:" 64
£1+"'+£,,2 7:1 ~ 7:2 59
£1(+)"'(+)£» 7 7:1£ 59
£d+)'" (+) £n 7 7:0((;1;) 60
£1 (+).1'" (+).1 £n 36 7:int(£) 71
£1 ( + )...1 ••• (+)...1 53,. 36 7:J((;1;) 88
£ =# we 21 7:M({;\;) 96
mJJT) 29 7:M(Sj) 102
9C(T) 28 7:WO({;\;I' a:2) 120
P+, P- 50 o 1,28
~o, ~oo 4 o1
~+, ~-, ~++, ~-- 4
721/17/73 V/12/6
Ergebnisse der Mathematik und ihrer Grenzgebiete
1. Bachmann: Transfinite ZaWen. DM 48,-

2. Miranda: Partial Differential Equations of Elliptic Type. DM 58,-
4. Samuel: Methodes d' Algebre Abstraite en Geometrie Algebrique. DM 34,-
5. Dieudonne: La Geometrie des Groupes Classiques. DM 42,-
7. Ostmann: Additive ZaWentheorie. 1. Teil: Allgemeine Untersuchungen.
DM 42,-
8. \Vittich: Neuere Untersuchungen tiber eindeutige analytische Funktionen.
DM 36,-
11. Ostmann: Additive Zahlentheorie. 2. Teil: Spezielle Zahlenmengen. DM 34,-
13. Segre: Some Properties of Differentiable Varieties and Transformations.
DM 46,-
14. Coxeter/Moser: Generators and Relations for Discrete Groups 3rd edition.
DM 42,-
15. Zeller/Beckmann: Theorie del' Limitierungsverfahren. DM 64,-
16. Cesari: Asymptotic Behavior and Stability Problems in Ordinary Differential
Equations. DM 54,-
17. Severi: II teorema di Riemann-Roch per curve, superficic e varieta questioni
collegate. DM 30,-
18. Jenkins: Univalent Functions and Conformal Mapping. DM 37,-
19. Boas/Buck: Polynomial Expansions of Analytic Functions. DM 24,-
20. Bruck: A Survey of Binary Systems. DM 46,-
21. Day: Normed Linear Spaces. 3rd edition. DM 42,-
23. Bergmann: Integral Operators in the Theory of Linear Partial Differential
Equations. DM 40,-
25. Sikorski: Boolean Algebras. DM 42,-
26. Ktinzi: Quasikonforme Abbildungen. DM 43,-
27. Schatten: Norm Ideals of Completely Continuous Operators. DM 30,-
28. Beckenbach/Bellmann: Inequalities. DM 38,-
29. vVolfowitz: Coding Theorems of Information Theory. DM 30,-
30. Constantinescu/Cornea: Ideale Randel' Riemannscher Flachen. DM 75,-
31. Conner/Floyd: Differentiable Periodic Maps. DM 34,-
32. Mumford: Geometric Invariant Theory. DM 24,-
33. Gabriel/Zisman: Calculus of Fractions and Homotopy Theory. DM 42,-
34. Putnam: Commutation Properties of Hilbert Space Operators and Related
Topics. DM 31,-
35. Neumann: Varieties of Groups. DM 51,-
36. Boas: Integrability Theorems for Trigonometric Transforms. DM 20,-
37. Sz.-Nagy: Spektraldarstellung linearer Transformationen des Hilbertschen
Raumes. DM 24,-
38. Seligman: Modular Lie Algebras. DlVI 43,-
39. Deuring: AIgebrcn. DM 30,-
40. Schutte: Vollstandige Systeme modaler und intuitionistischer Logik. DlVI 30,-
41. Smullyan: First-Order Logic. DM 36,-
42. Dembowski: Finite Geometries. DM 68,-
43. Linnik: Ergodic Properties of Algebraic Fields. DM 44,-
44. Krull: Idealtheorie. DM 34,-
45. Nachbin: Topology on Spaces of Holomorphic Mappings. DM 18,-
46. A. Ionescu Tulcea/C. Ionescu Tulcea: Topics in the Theory of Lifting.
DM 36,-
47. Hayes/Pauc: Derivation and Martingales. DM 48,-
48. Kahane: Series de Fourier absolument convergentes. DM 44,-
49. Behnke/Thullen: Theorie der Funktionen mehrerer komplexer Veranderlichen.
DM 48,-
50. Wilf: Finite Sections of Some Classical Inequalities. DM 28,-
51. Ramis: Sous-ensembles analytiques d'une variete banachique complexe.
DM 36,-
52. Busemann: Recent Synthetic Differential Geometry. DM 32,-
53. Walter: Differential and Integral Inequalities. DM 74,-
54. Monna: Analyse non-archimedienne. DM 38,-
55. Alfsen: Compact Convex Sets and Boundary Integrals. DM 46,-
56. Greco/Salmon: Topics in m-Adic Topologies. DM 24,-
57. Lopez de Medrano: Involutions on Manifolds. DM 36,-
58. Sakai: C*-Algebras and W*-Algebras. DM 68,-
59· Zariski: Algebraic Surfaces. DM 54,-
60. Robinson: Finiteness Conditions and Generalized Soluble Groups, Part 1.
DM 48,-
61. Robinson: Finiteness Conditions and Generalized Soluble Groups, Part 2.
DM 64,-
62. Hakim: Topos anneles et schemas relatifs. DM 48,-
63. Browder: Surgery on Simply-Connected Manifolds. DM 42,-
64. Pietsch: Nuclear Locally Convex Spaces. DM 48,-
65. Dellacherie: Capacites et processus stochastiques. DM 44,-
66. Raghunathan: Discrete Subgroups of Lie Groups. DM 56,-
67. Rourke/Sanderson: Introduction to Piecewise-Linear Topology. DM 42,-
68. Kobayashi: Transformation Groups in Differential Geometry. DM 52,-
69. Tougeron: Ideaux de fonctions differentiables. DM 69,-
70. Gihman/Skorohod: Stochastic Differential Equations. DM 88,-
71. MilnorjHusemoller: Symmetric Bilinear Forms. DM 42,-
72. Fossum: The Divisor Class Group of a Krull Domain. DM 44,-
73. Springer: Jordan Algebras and Algebraic Groups. DM 48,-
74. Wehrfritz: Infinite Linear Groups. DM 59,-
75. Radjavi/Rosenthal: Invariant Subspaces. DM 50,-
76. Bognar: Indefinite Inner Product Spaces. DM 48,-
77. Skorohod: Integration in Hilbert Space. In preparation
78. Bonsall/Duncan: Complete Normed Algebras. DM 68,-

10.1007@978 3 642 65567 8

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

10.1007@978 3 642 65567 8

Uploaded by

Copyright:

Available Formats

Ergebnisse der Mathematik

und ihrer Grenzgebiete

Herausgegeben von P. R. Halmos . P.]. Hilton

Vnter Mitwirkung von L. V. Ahlfors . R. Baer

Geschaftsfiihrender Herausgeber: P.]. Hilton

AMS Subject Classification (1970)

ISBN -13: 978-3-642-65569-2 c-ISBN -13:978-3-642-65567-8

By definition, an indefinite inner product space is a real or complex

The treatment is, whenever possible, invariant, i.e. independent of

Budapest, October 1973 Janos Bognar

Chapter 1. Inner Product Spaces without Topology

Chapter II. Linear Operators in Inner Product Spaces without Topology

1. Linear Operators in Vector Spaces. 28

Chapter III. Partial Majorants and Admissible Topologies on Inner

1. Locally Convex Topologies on Vector Spaces 58

7. Projections and Admissible Topologies. 70

Chapter IV. Majorant Topologies on Inner Product Spaces

Chapter V. The Geometry of Krein Spaces

1. Krein Spaces. . . . . . . 100

Chapter VI. Unitary and Seifadjoint Operators in Krein Spaces

Chapter VII. Positive Operators and Plus-operators in Krein Spaces

1. Positive Operators . . . . . 147

5. Strict Plus-operators. 154

Chapter VIII. Invariant Semi-definite Subspaces of Linear Operators

Chapter IX. Pontrjagin Spaces and Their Linear Operators

In this section we fix some conventions (terms and symbols) concerning

If 2j are subspaces (j = 1, ... , n), then obviously

(1.1) (21) ... ,2n>={tXj:

The subspaces 2j (f = 1, ... , n) are said to be linearly independent,

An infinite system of elements is said to be linearly independent if

Lemma 1.1. If 2 is a subspace of the vector space (§3, then every

Corollary 1.2. The dimension of any complementary subspace f01-

In our terminology, an inner product on a vector space (§3 is a complex-

(2.2) (y, x) = (x, y)

Two inner product spaces ~ and ~I with inner products (. , .) and

(TXl' Tx 2 )' = (Xl' x 2)

~++ = {x E~: (x, x) >0 or x = O},

L' leil l~il2

defines an inner product on Gf, which may be indefinite, semi-definite

we can speak of semi-definite, positive, negative, neutral, definite, positive

Lemma 2.5. The span of a positive (negative, neutral) vector 2S

Most of the results concerning inner product spaces have two

will be positive. Hence the vectors Xl = X +

positive and we have x = Xl +

Coronary 2.7. If the inner product is indefinite on Q:, then none of

If (x,y) = 0 or, equivalently, (y, x) = 0 for a pair of vectors x,y in the

These are m homogeneous linear equations for l unknowns. Conse-

Corollary 3.4. Ij £, we are subspaces of rg satisfying £ n weI = 0,

Example 4.1. Let ~ be a two-dimensional vector space with basis ev

Corollary 4.6. If two vectors x, y of an inner product space satisfy

Lemma 4.7. If 2 is a neutral subspace, so is 211.

Example 4.8. Consider the vector space ~ of complex sequences

(x, y) = - ~o 170 1.,' 1 ~i l7i .

all other coordinates equal to belongs to 2 (n = 1,2, ... ), we have

In view of the relation 2-' Il7i l2 0. Thus 2.1. = 0,

It is also untrue that 21.L would be non-degenerate whenever 2

Denote by 2 the subspace of ~ consisting of all finite sequences {~i}~

,'lith ~o = ~1 = .1' ~i' Then 2.1. is the 1-dimensional subspace spanned

It is easy to see that Yo is an isotropic vector of ,2LL though ,2 is

5. Maximal Non-degenerate Subspaces

Corollary 5.3. 03/030 is non-degenerate. 0

Theorem 5.4. The set 031 C 03 is a complementary subspace for (to

6. Maximal Semi-definite Subspaces