PACIFIC JOURNAL OF MATHEMATICS

Vol. 52, No. 2, 1974

REPRODUCING KERNELS AND OPERATORS WITH A

CYCLIC VECTOR I
VASHISHTHA N. SINGH

In this paper a study is begun of the complete unitary

invariant ((1 — wT)~1e1 (1 — zT)~1e), first considered by Livsic
in his paper 'On Spectral Resolution of Linear Nonself Ad-
joint Operators' Mat. Sb., 34 (76), 1954, 145-199, of a triple
(T, H, e) where T is a bounded linear operator on a Hubert
space H and e is a cyclic vector for T in H, as a reproducing
kernel. One of the important points is the construction of
a subset of the group algebra of the torus closed under
pointwise addition and convolution. This obviously will
generate a ring called the K-ring. A study of this ring will
be done later.
Several other theorems and constructions are also given.

Introduction* Let T be a bounded linear operator on a Hubert

space H with a topoJlogically cyclic vector e in H. In this paper we
wish to study certain analytic functions associated with the triple
(T, H, e) for the sake of the problem of invariant subspaces of T in
H. (See also [8] and [15].)
The paper is divided into six sections. In § 1 we present some
facts about reproducing kernels with analyticity properties. In § 2
we consider a triple (T, H, e) of the above type. H can then be
represented as a Hubert space of conjugate analytic functions ae[H]
with a reproducing Kernel K. T* on H assumes the form of reverse
shift on the Taylor coefficients of functions in ae[H]. (See also [11]
or [19].) In § 3 we recover (T, H, e) from the reproducing kernel of
ae[H] in two ways. The notion of an analytic function of positive
definite type is introduced and it is shown that only these can arise
as reproducing kernels of ae[H]. These functions are also related
to invariant subspaces (§4). In § 5 a category of triples is constructed
and it is connected to the harmonic analysis of the two-torus via
the analytic functions of positive definite type. Section 6 consists of
some examples and counterexamples about the analytic functions of
positive definite type.
The paper is based on the author's dissertation written under
the guidance of Professor John L. Kelley of the University of
California, Berkeley. The author would like to thank Professor
Kelley for help and advice.

1* Reproducing kernels and analytic functions of positive

type. We start with the definition of a reproducing kernel. Let
567
568 V. N. SINGH

H be a Hubert space of functions on a set X with the inner product

(,). In Theorem 1.2 we present a few facts of a theory of H due
essentially to E. H. Moore and N. Aronszajn. See [1] and [2].

DEFINITION 1.1. A reproducing kernel for H is a complex valued

function K on X x X such that:
(i) For all yeX,KyeH where Ky(x) = ΛΓfo y).
(ii) If g e H then for all yeX9 (gf Ky) = g(y), and
(iii) The linear span of the set {Ky}yex of functions is dense in H.

Note that if K is a reproducing kernel then the map g ι— g(y) =

(g, Ky) is a bounded linear functional on H; i.e., evaluation at a point
is continuous. Also if H is a pre-Hilbert space of funtions on a set
X with continuous evaluation then the completion H oί H can be
realized as a space of functions on X by setting h(x) = (fc, i Q for
ί c e l a n d k f f where iΓ,e 5" is such that (/, Kx) = f(x) for all / e H.

THEOREM 1.2. Lei H be a pre-Hilbert space of functions on a

set X so that the evaluation map f*-*f(x) is continuous for all xe X,
let H be the completion of H. Then the following hold
( i ) There is a unique reproducing kernel K for H.
(ii) If for xeX, εxeH, is such that f(x) = (/, εx) for all fe H
then K(x, y) = (εy, εx).
(iii) // {δjie/ is an orthonormal basis for H, then K(x, y) =

(iv) If H is the set of finite formal sums Σ?=o αA, where xt e X

and als are complex numbers, with an inner product (,) given by
(Σ?=i aiχu ΣΓ=i bjVj) = Σ?=i Σf=i dibjKiyj, χt), the map Σ?=i atxt H-»
Σ ? = I ^<-K*i °/ formal sums to members of H is inner product preserv-
ing and its image is dense in H.
See [1] for a proof.
A reproducing kernel is a function of positive type in the fol-
lowing sense.

DEFINITION 1.3. A complex valued function K defined on X x X

is a function of positive type iff ΣL=i a>i<ϊjK(xs, %d ^ 0 for all complex
numbers al9 α2, , an and all members xlf x2, , xn of X.
Since Σ* atK(x9 xx) = (Σ* a<K.i9 Kx) ^ VK(x, x) VΣ^SMx^x,)
from the Cauchy inequality, a function of positive type K also has
a further property, namely that if ΣL=i Ui<LjK(xs, xd = 0 then
Σa,iK(x, xt) = 0 for all xeX. Thus it is clear from part (iv) of
Theorem 1.2 that a function of positive type is a reproducing kernel.
(A function of positive type K is necessarily selfadjoint i.e., K(x, y) =
K{y, x) for all x, y in X.)
 REPRODUCING KERNELS AND OPERATORS WITH A CYCLIC VECTOR I 569

DEFINITION 1.4. For each s > 0, DS(DS) is the open (closed) disc
of radius s about 0 in the complex plane €. As is the space of
functions defined and conjugate analytic in Ds and As.$ the space of
functions defined on Ds x Ds which are conjugate analytic in the
first variable and analytic in the second variable. & is the space
of polynomials with complex coefficients. If K is a function of positve
type and is in A8m8 for some s then K is an analytic function of
positive type abbreviated a.f.p.t.
Let Ke ASfS. The following theorem gives an alternate construc-
tion of a Hubert space which is equivalent to K being a function of
positive type.

THEOREM 1.5. Let Ke A8t8. Then

( i ) K is an a.f.p.t. if 1/Aπ2 Φ Φ K(z, w)p(z~ί)p(w~ί)(dz/z) x
J cJ c
(dwjw) ^ 0 for all p e & where the path of integration C is a simple
contour in Ds such that its inverse winds around Dιίs once.
(ii) Let (H, (,)) be a pre-Hilbert space of functions analytic in
a neighborhood of Dι}s with the inner product (p, q)κ — 1/47Γ2
J Φ K(z, w)p(w~~1)q(z~1)(dz/z)(dw/w). Then the map which takes a
cJ c
member Ka of H, a a member of Ds, into the rational function
1/(1 — az) is inner-product preserving.

Proof. Consider l/4τr2 Φ Φ K{z, w)p(z~ι)p(w~1)(dz/z)(dwlw) as a

limit of Riemann sums l/4π2 Σ?,i=ι K(ZJ, ^dP^iΊP^T^j+i ~ z$lzϊ) x
(zi+1 — zjzj. Since K is a function of positive type it is clear that 1/47Γ2
1 1
(J) (J) K(z, w)p(z~ )p(w~ )(dz/z)(dw/w) is positive. Let now alf a?, , an
be any complex numbers and let au , an e Ds. Consider the rational
function / given by f{z) = Σ* α€(l - a^Y1. Then Σ?,i=i aiά:3 K(ajf az) =
2 ι 2
l/4τr f f K(z, w)f(z^f(w- ){dz/z)(dw/w) - lim l/4ττ f ί K(z,w)x
J C1 J C2 ί-*00 J C1J C2
ι 1
pι(z~ )pι(w~ )(dz/z)(dw/w) ^ 0 where p% is a sequence of polynoimials
converging to / uniformly on compact subsets of Dιjs and Cl9 C2 are
chosen suitably. Thus (i) is proved the proof of (ii) follows from
the observation that the norm in the space of rational functions of
the function z H^ Σ?=I α *(l ~~ ^^Y1 is Σ?,i=i ^jK(ajf aτ) which is the
same as the norm of Σ?=i ^%Ka. in H.

2. The kernel function of cyclic triple* Consider now a Hubert

space H and a bounded linear operator T on H. We construct certain
functions of positive type associated with T and H. Some facts
from the functional calculus of T are used. (See [12].) If δ is a
570 V. N. SINGH

complex valued function analytic in the neighborhood of the spectrum

σ(T) of T define δ(Γ).as follows.

DEFINITION 2.1. Let C be a contour lying in the domain of

analyticity of b which winds around each point of σ(T) once. Then
ι
b(T) = l/2τrί f b(z)(zl - T)~ dz. It is known that b H* b(T) is a linear
and multiplicative homomorphism of the space of functions analytic
in a neighborhood of σ(T) into the space of bounded linear operators
on H. (See [12], page 199.)
The following result is probably well-known and will be used
later.

THEOREM 2.2. Let T be a bounded linear operator on a Hubert

space H and suppose that there is an analytic function b defined
and nonzero on a connected neighborhood of the spectrum o(T) of
T such that b{T) — 0. Then there is a nonzero polynomial p such
that p(T) = 0.

We leave the proof of this fact to the reader. See for example
[20].
Now we make the following definition.

DEFINITION 2.3. For r > 0. J5r is the space of functions analytic

in a neighborhood of the closed disc Dr with topology given as the
inductive limit topology of the spaces A( U) of the functions analytic
in s neighborhood U of Dr. (See [14], page 219, problem D.)

It is known that Br is a Montel space. (See [14], page 196,

problem F(e).)
For r > 0, define Ar as follows.

DEFINITION 2.4. Ar is the space consisting of functions which

are complex conjugates of the functions in Ar.
Let E be a linear space with a locally convex topology r and let
E' be its dual. Then the strong topology for Er is the topology of
uniform convergence on r-bounded sets.
The following result will be important in the sequel.

LEMMA 2.5. Let r > 0. Then AUr and Br are strong duals for
1
the pairing given by [/, g] — I f(z)g(z~ )dz/z for fe Br and g e Allr
where C is the contour tv->(r + έ)eu, 0 ^ t ^ 2π, for some small ε
depending on f and g.
 REPRODUCING KERNELS AND OPERATORS WITH A CYCLIC VECTOR I 571

Proof. We present an outline of the proof and refer to [14] for

more information. Since Br is a Montel space it is reflexive ([14],
20 F(a)) and hence it suffices to show that the dual of 2?r, with
strong topology is Allrf for the pairing above.
To show this, let for each z e D1/r, hz be the member of Br given
by hz(ζ) = 1/(1 — ζz) for ζ e Dr. For any continuous linear functional
g on Br let %{g) be defined by iQ(g)(z) = g(hz). Then %{g) is a well-
defined function and is a member of A1/r. Moreover, g(f) = l/2πi
fi^ioidX^dz/z = l/2πi[f9 %{g)\, where C is a contour as in the
/ c
statement of the proposition ε being so small that it is contained in
the domain of holomorphy of /. Conversely any member of A1/r
defines a continuous linear functional on Br by the above formula
and thus Aίlr and B'r are algebraically isomorphic.
To complete the proof it remains to show that the topology of
u.c.c. on Allr coincides with the strong topology of B'r. To prove
this first observe that each bounded set of Br is contained in a
bounded set of Άr. for some r' > r ([14], 17G (6) (iii)). Hence it
follows that B'r with strong topology is metrizable ([14], 18.4). Since
each bounded subset of Ar, is uniformly bounded on every compact
subset of Όr,, a Cauchy sequence in the u.c.c. topology of ~A^r is also
a Cauchy sequence in its strong topology. That a Cauchy sequence
in the strong topology of B'r is also a Cauchy sequence in its u.c.c.
topology follows from the observation that for any compact subset
K of Dljr the family of functions {hz}zeκ is bounded in Br and also
the fact that for any h e A1/rf [hz, h] = 2πih(z) where z e K.
Consider now an operator T on a Hubert space H and let e e H.

DEFINITION 2.6. A Hubert triple is a triple (T.H.e) where H is

a Hubert space. T a bounded linear operator on H and e a member
of H* A Hubert triple is a cyclic triple if the orbit of e (the linear
span of the set {T*e}T=o is dense in H. If r = || Γ|| the map βe\ Br-*H
is given by βe(b) = b(T)e.
The map βe depends on T, H, and e. Note that βe(Br) is dense
in H iff (Γ, Hf e) is cyclic.
The following is a consequence of Theorem 2.2.

2.7. If(T,H, e) is cyclic then His infinite dimensional

THEOREM
if and only if βe is injective.

Proof If H is finite dimensional then it follows from the Caley-

Hamilton theorem of linear algebra that p(T) = 0 for some pe^
([4], page 320) and thus βe is not injective. Since (Γ, H, e) is cyclic
the converse is the statement of Theorem 2.2.
572 V. JSΓ. SINGH

Now we establish commutativity of certain maps and construct

a reproducing kernel which is an a.f.p.t. The reason for this round
about construction will be clear from Theorem 4.3.

DEFINITION 2.8. Multiplication by z, Mz: Br-+Br is the operator

given by M,f(Q = ζ/(ζ) for ζe Domain (/).
The following lemma is an easy consequence of the functional
calculus.

LEMMA 2.9. βe and Mz are continuous linear maps such that

β.(ϊ) = e and T-βe = βe-Mz.

DEFINITION 2.10. φ: H—>H' is defined by Φ(x)(y) = (y, x) for all

x, yH.
H' is conjugate linearly isomorphic to H via the map φ and
φ.T* = T' φ where T* is the Hubert space adjoint of Ton H and T
is the Banach space dual of T. Also from T-βe = βe-Mz we obtain
βe T = Mϊ-β'e where β'e, Mz, etc. are Banach space duals of βe, Mz,
etc. We know from Lemma 2.5 that B'r can be identified with Alfr
via the map i 0 .

DEFINITION 2.11. The map ae: H~> A1}r is the composition of the
maps φ, β'e, i0 and — as given in the diagram

H •H > Br • Aur •A1/r

where — takes a function into its complex conjugate.

From Lemma 2.9 and the discussion following it we see that there
is a map S*: Aί/r—* Al!r so that ae-T* = S* Λ:e where ae is as above.

THEOREM 2.12. Let (T, H, e) be a Hilbert triple with \\T\\ = r.

Then S*ae = ae T* where ae is such that ae(x)(z) = (x, (1 — zTγγe)
for all z e DUr and all x e H, and S*f(z) = f(z) — f(0)/z. Moreover, ae
is injective iff (T, H, e) is cyclic and in that case} if dim H = oo,
the range of ae is dense in A1/r.

Proof. We show that ae is such that ae{x){z) = (x, (1 — zT)~λe)

for all z e DιSr and S*f(z) = f(z) — f(0)/z, because if this were so then
a.(T*x)(z) = (Γ*α?f (1 - zT)~ιe) = (x, (1 - zT)~ιTe) = S*(a.{x))(z) and
hence jg*-αg = α,-Γ*, Now -0i00βΌφ(x)(z) = β'0φ(x)(hz) = (x)(βe(hz)) =
((1 - zT)-\ x) = (a?, (1 - zTY'e).
Now we prove the rest of the theorem. Since ae(x) - 0 iff (x, (1 —
zT)~ιe) = 0 for all zeDl!r it follows that ae(x) = 0 iff (x, p(T)e) = 0
for all pe&>. If e is cyclic then this holds iff x = 0. The statement
 REPRODUCING KERNELS AND OPERATORS WITH A CYCLIC VECTOR I 573

that if e is cyclic and if dim H = oo then the range of ae is dense in

Aι!r is a mere dualisation of the statement in Theorem 2.2.
Note that in general ae is injective on the orbit of e and zero
on its orthogonal complement. If e is cyclic then we can assign an
inner product ( )«β on ae[H] by (a£x), ae(y))*e = (a?, y). This inner
product makes T* on Hunitarily equivalent to S* on ae[H]. Moreover,
and this is the point of the construction, ae[H] has a reproducing
kernel.

THEOREM 2.13. Let (T,H,e) be cyclic. Then for each zeDllr

the evaluation map ae(x) f-+ ae(x)(z) defined on ae[H] is continuous
ι
and ae(x)(z) = (ae(x), εz) where ez = ae{(l — zT)~ e). Consequently ae[H]
has a reproducing kernel K given explicitly as K(z, w) = ((1 —
l x m m
wT)~ e, (1 - zTY e) = ΣϊU-oίΓ e, T e)z w\

Proof. Since ae(x)(z) = (x, (1 - zT)fty = {ae{x\ ae((l -

for all ae(x) e ae[H]f we see that the evaluation at z is explicitly given
by inner product with a member of ae[H] and is hence continuous.
Furthermore, the particular member of ae[H] corresponding to eval-
uation at z is ae((l — zTY1) and hence εz = ae((l — zTγxe). Now in
view of Theorem 1.2 it follows that ae[H) has a reproducing kernel
given explicitly as K(z, w) = (αβ((l -wTYιe\ ae((l - zTγιe))*9 = ((1 -
1 1 ι
loϊ )- ^ (1 - zTY e).

DEFINITION 2.14. The kernel function of a cyclic triple (I7, £Γ, e)

is the reproducing kernel of ae[H]. If (Γ, ίZ, e) is just a Hubert
triple, not necessarily cyclic, then the kernel function for (T, H, e) is
defined to be the kernel function for (Γo, 0(e), e) where Q(e) is the
closed orbit of e and To is the restriction of T to 0(e).
Two cyclic triples (TΊ, Jϊi, ex) and (T2, H2, e2) are defined to be
unitarily equivalent, (2\, flΊ, ex) — (JΓ2, £Γ2, e2), iff there exists a unitary
map C7: fli ^ H2 with U-T, = T2-U and ^(βO - e2.

Let (T, ίί, e) be a cyclic triple with kernel function K. Consider

the Hubert space ae[H] with the inner product ( )ae obtained from
(Γ, H, e). We can construct ae[H] from Kin the manner of Theorem
1.2, since K is the reproducing kernel for ae[H]. Since ae(e) = Ko
we have the following as a corollary to the preceding.

COROLLARY 2.15. Let (Γ, H, e) be a cyclic triple. Then the kernel

function K for {T, H, e) is a complete unitary invariant for (Γ, H9 e),
i.e., (T*, H, e) is unitarily equivalent via ae to (S*, ae[H], KQ).

The above corollary can also be deduced directly. See for

example [15].
574 V. N. SINGH

3* The triple of an A*F*P*Ί\ In this section we study the

kernel function of a triple more closely. First we mention some
properties of a kernel function. Note that if K is the kernel
function of (T, H, e) where | | Γ | | = r then K can be written as
1 m m n
Σ ^oίϊ ^, T e)z w for z, w e D1/r. The boundedness of T is reflected
in a special property of K. The infinite dimensionality of H is also
reflected in another property of K. We describe these properties.
First the following definitions.

DEFINITION 3.1. The map S?\ AStS —> ASy8 is given by £f(a)(z, w) =
[a(z, w) — a(z, 0) — α(0, w) + α(0, 0)]/(zw) for z, we Ds. A member K
of As,s which is an a.f.p.t. is an analytic function of positive definite
type (abbreviated a.f.p.d.t.) iff there is a positive real r so that
2
r K — £f{K) is also an a.f .p.t. We write ρ(K) for the least such r.
Ke ASfS is a degenerate a.f .p.t. iff K is an f .p.t. and there is a
polynomial p so that I φ K(z, w)p(z~1)p(w~1)(dz/z)(dw/w) = (p, p) = 0,
J cJ c
where C is such that its inverse contains Dιls and is preferably a
circle with center 0.
It follows from Theorem 1.5 that if 6a' is a function of positive
type so is £^{a)9 since Φ Φ £^{a){z, w)p(z~1)p(w~ί)(dz/z)(dw/w) =
φ φ α(^, w)pί(z~1)p1(w~1)(dz/z)(dw/w), where p1 is the polynomial such
J cJ c
that px(z) = zp{z). As shown by the following result the kernel
function of a Hubert triple is always an a.f.p.d.t.
PROPOSITION 3.2. If K is the kernel function for the triple
(T, Hy e) then K is an a.f.p.d.t. and S^{K) is the kernel function
for the triple (Γ, H, Te). Moreover, if (Γ, H, e) is cyclic, then — p{K) =
| | Γ | | and K is nondegenerate iff άimH = oo.
We omit the straightforward proof of this proposition.
Recall that if (T, H, e) is a cyclic triple then we have seen in
§2 (Theorem 2.15) that (Γ*, H, e) is unitarily equivalent via ae to
(S*, ae[H], Ko) and that ae[H] with ( )«β is a space of functions with
continuous evaluations at points, thus the space ae[H] and in fact
the triple (T*, H, e) and hence (Γ, H, e), is determined by the kernel
function K. We want to describe explicitly in two different ways
corresponding to the two Theorems 1.2 and 1.5 of § 1, the construc-
tion of (T, H, e) from K.
First it follows from ae-T* = S*-^ and the fact that a* = at
that ae-T = S-ae where S is the adjoint of S* relative to the inner
product ( )ae of ae[H]. We construct ae[H] in terms of the kernel
K of (T, H, e). We make the construction for an arbitrary a.f .p.t.
 REPRODUCING KERNELS AND OPERATORS WITH A CYCLIC VECTOR I 575

Ke As,s. A formal for S is given in Proposition 3.4. First a definition.

κ
DEFINITION 3.3. For K an a.f .p.t., Ke Ar>r> let H be the com-
pletion of the space of functions which are finite linear combinations
Y^aiKai for at e C and at e Dr with respect to the inner product
k
(Σ* atK*v Σ ; bjKβ.) = Σiti a&Kiβj, a,).

We have the following result.

κ
PROPOSITION 3.4. Let KeAr,r be an a.f.p.t. ThenH cAr. If
κ
K is the kernel function for a cyclic triple (T, H, e) then ae[H] = H
κ κ
and the adjoint S of S*: H —»H is given by S(KW) = Kw — KJw
for w Φ 0 and S(KQ) = dK/dw \w=0.

Proof. The fact that Hκ consists of functions is known from

§ 1. Let fe Hκ and let fn = Σi &?%<$ be a sequence of finite sums
converging to/. Then since | Σ a<iKaJίz) |2 ^ K(z, z) \\ atKai ||2 it follows
that the sequence fn of conjugate analytic functions is a Cauchy
sequences in the topology of uniform convergence on compact sets
of Dr. Hence the limit function / is also conjugate analytic. Thus
Hκ(zAr.
Since, if K is the kernel function for (T, H, e) then K(z, w) =
((1 - wT)~le, (1 - zTY'e), ae[H] = Hκ. Since e is cyclic the adjoint S
of S* is given by
S(Kw)(z) = ((1 -
= (d =^ r r 6 - β (1
_ z T Γ e
\ =
κw - K.
\ w / w
ι
for w Φ 0 and S(K0){z) = (Te, (1 - zT)~ e) = dK/dw Uβθ(.,.

COROLLARY 3.5. If K is the kernal function of the cyclic triple

(Γ, H, e) then (T, H, e)is unitarily equivalent via ae to (S, Hκ, Ko).
Also this,

COROLLARY 3.6. (S, Hκ, Ko) is a cyclic triple with kernel function
K.
Now we make the second construction corresponding to Theorem
1.5 of § 1. First a definition.

DEFINITION 3.7. If Ke AUrΛ!r and K is an a.f .p.t., Hκ is the

completion of the space Br of functions analytic in a neighborhood
of Dr with respect to ( ) κ given by (blf b2) — l/4ττ2 I I K(z, w) x
J cλ J c2
), where bu b2 are members of Br and where
576 V. N. SINGH

Ci is a contour such that its inverse is rectifiable in the complex

plane, contains Dr and has winding number 1 about Dr and is con-
tained in the domain of holomorphy of bt. If multiplication by z, Mz
is bounded on Br relative to ( ) κ then Ml is its bounded extension
to Hκ.

Hκ is well-defined despite the arbitrariness of the contours and

r. This is so because of Cauchy's theorem. Also note that Hκ does
not consist of functions in general (see counterexample 1, § 6).

PROPOSITION3.8. Let K be an a.f.p.t. e AUrΛir. Then

s
( i ) Mz on Br has bounded extension M z to Hκ iff K is an
a.f.p.d.t.
(ii) If K is an a.f.p.d.t. then K is the kernel function of

(iii) If K is the kernel function of a cyclic triple (T, H, e) then

(MZA, HKf 1) is unitarily equivalent to (T, H9 e) under the map pv-»
p(T)e for all polynomials p.

Proof.
( i ) K is an a.f.p.d.t. iff there is t ^ 0 so that ?K- S^(K) is of
positive type. However, ί I S^{K)(z, w)a(z~1)a(w~1)(dz/z)(dw/w) =

f J K(z, wXM^iz-'XM^Xw-'JϊdφXdw/w), a e Br. Hence t2K-

is a function of positive definite type iff t ^ ||ikC ||, Br being dense
in Hx. To prove
(ii) note that if λ e Dllr then the function Rλ: z —> (1 — λ^)"1 e Br
and {Rh Rμ) = l/4π2 f f K(z, w)l/((l - Xw'1)^ - μz-ι))(dzlz){dwlw)
J cι J c2
where Cίf C2 are contours specified as in Definition 3.7. It follows
from the Cauchy formula (see [13], page 26) that (Rλ, Rμ) = K(X, μ)
and thus (ii) is proved.
The proof of (iii) is straightforward and is left to the reader.
COROLLARY 3.9. A function K is the kernel function for some
cyclic triple iff K is an a.f.p.d.t.

4* Further properties of an A*F*P*Ί\

4.1. The linear map C": Hκ —> Hκ is given on K'a$ by

DEFINITION
C\Ka)(z) = (1 — az)~ι and the map J: Hκ -— Hκ is given on polynomials
by J(p) ~ l/2πi I Kvp(w~1)dw/w for a simple contour C which lies
in the domain of analyticity of the function w^ Kw and whose
inverse contains ΊTr and has winding number 1 about it.
 REPRODUCING KERNELS AND OPERATORS WITH A CYCLIC VECTOR I 577

To be sure we should refer to vector valued integration but here

we mean that J(p) is the function so that J(p)(z) = l/2πi Φ K(Z, W) X
J c
ι
p{w~~ )dwlw. Clearly J(p) is independent of the contour C so long it
has the properties given in the definition.
We have the following proposition.
PROPOSITION4.2. // (Γ, H, e) is a cyclic triple then ae and C
are unitary and the diagram

is commutative. Moreover, C'~ι = J for any a a.f.p.t. K and C" is

unitary for any a.f.p.t.

Proof. We know that ae is unitary. That C is unitary for any

a.f.p.t. K is a consequence of Theorem 1.5. Since ae[H] = Hκ we
know that the diagram

τ s
l
is commutative. To prove that the rest of the diagram is also
commutative it suffices to prove that C'-S = Mz. C on elements
of the type Ka of Hκ. However, this is true since C'-S(Ka)(z) = C".
Ka
~~K°(z) = ( 1 ~ g g ) " 1 ~ 1 = z(l - az)-1 - M,-C'(K)(z) .
a a
Now we prove that Cf~ι = J for any a.f.p.t. K. To do so it
suffices to consider elements of the form C\Ka) of Hκ. For such an
element

J-C'(Ka) = -1-r ί Kw(l - aw~ι){dwlw) = - ? v f -^-dw = Ka .

2π% J c 2τz% J c w — a

Now let H be the closure of Br under an inner product. Then

we have the

THEOREM 4.3. H = Hκfor some a.f.p.t. Ke Alίr, i/r iff H* = Hκ.

Proof. If H — Hκ then the inclusion of Br in H is continuous

578 V. N. SINGH

where Br is taken with its u.c.c. topology. Hence the dual of H can
κ κ
be identified with H as in Proposition 4.2. Conversely if if* = H
for some a.f.p.t. K in A1}r, 1/r, then the dual can be identified with
H is the same as Hκ c Br via the map C".
The following two theorems give characteristics of an a.f.p.d.t.
supplementing Definition 3.1 and furnishing a connection between an
a.f.p.d.t. and an invariant subspace.

THEOREM 4.4. Let Ke Ar,r be an a.f.p.t. Then K is an a.f.p.d.t.

if and only if Hκ is S*-invariant.

The above theorem is an immediate consequence, of the fact

κ
that the inclusion of H in Ar is continuous, and the closed graph
theorem. We leave the proof to the reader.
Let (Γ, H, e) be a cyclic triple. We prove another proposition
which relates invariant subspaces of T in H to functions of positive
definite type arising from the kernel function of (T, H9 e) and the
projection corresponding to the invariant subspace.

THEOREM 4.5. Let {T, Hy e) be a cyclic triple. Then P is an

orthogonal projection so that P{H) is invariant under T if and only
if the function K' defined by K'(z, w) = ((I - P)(l - wT)'ιef(I - P) x
(1 — zT)~ιe) is an a.f.p.d.t.

We have to show that P(H) is invariant under T if and only if

the operator (I — P)x\-*(I — P)Tx is bounded. We prove this fact
as a consequence of a more general lemma.

LEMMA 4.6. Let (Γ, H, e) be a Hilbert triple and H a first

countable Hausdorff linear topological space, T a continuous linear
operator defined on H and e a member of H. Let H? = {x e H; there
is a sequence {pi{T)e}i—>x and {Pi(f)e}i—*0 where each PieP,i =
1,2, •••}. Then H? is a closed invariant subspace for T.

Proof. Hz is clearly a linear subspace of H invariant under T.

The whole difficulty lies in showing that it is closed. This is also
quite straightforward and we do it as follows. Let {pT(T)e}i be
a sequence converging to xm and let also the {xm}m converge to x.
Also let {pΓ(Γ)β}i converge to 0 for each m. Let B(x, 2~n)(B(x, 2~n))
be the open (closed) ball of radius 2~n with center at x and let {n^
be the sequence for which {xm}m Π B(x, 2~ni) - B(x, 2"Wί"1) Φ 0 . Such
a sequence {n^t —• oo exists since {xm}m —* x. For each nt choose an
xn.e B(x, 2~w0 — B(x, 2~%ί~1). Suppose also that {[/„.}* is a countable
neighborhood basis at 0 for H. For each nt choose a p*f%i) so that
G Un. and p*f%i)(T)e is in B(x, z~ni) - B(x, 2-ni'1). I t is possible
 REPRODUCING KERNELS AND OPERATORS WITH A CYCLIC VECTOR I 579

to do so since {pp(T)e} j—> xn. and {^(ΓjeJy^O for each % by

supposition.
i
Now the sequence {p] {n.){T)e}%i has 0 as a cluster point in H so
there is a proper infinite subsequence of this sequence which tends
to 0. Since {p*fni)(T)e}n{ -+x in that subsequence if we replece T and
e by T and e it tends to x. Thus H? is closed.
The proof of this lemma can perhaps be simplified by construc-
tion of a bounded intertwining operator from H to H whose kernel
coincides with the subspace given by the limits of {Pi(T)e}i as above.
Now we complete the proof of Theorem 4.5 as follows.
Proof of Theorem 4.5. Suppose that P(H) is invariant for
T. We k n o w t h a t \\p \\Bκ, = \\(I - P)p{T)e\\H. Thus \\Mzp\\Hκ, =
| - P)Tp{T)e \\H = || (I - P)TPp{T)e + ( I - P)T(I - P)p(T)e \\H =
P)T(I- P)p(T)e \\H ^ || T(I- P)p(T)e \\H ^ M\\ (I- P)p(T)e\\H =
M\\p\\Hκ,, where Λf is the norm of T, and the suffix indicates the
space in which norm is to be taken. It follows that Mz is bounded
on Hκ, and hence K' is an a.f.p.d.t.
In order to prove the converse we use Lemma 4.6 and put for
H, ϊ, e HKfJ Ml and 1 respectively.
We conclude by characterizing interlacing maps of cyclic triples
to another triple in terms of kernel functions as a preparation for § 5.

THEOREM 4.7. Let (Tlf Hlf ex) be a cyclic triple and let (T2, H2, e2)
be another triple and φ12 the unique map given by ψ12{p{T^e^) = p(T2)e2
for all p e ^ . Then φ12 extends to a unique bounded linear operator
Φiz' Hx —+ H2 iff there exists a real number r so that r2K± — K2 is an
a.f.p.t. where Kx and K2 are the kernel functions of the triples
(Tlf Hu ex) and (T2, H2, e2) respectively.

5* A category of triples and harnomic analysis* Let & be

the class of all triples (T, H, e) so that T is a proper contraction.
Then ^ becomes a category if we define morphism between triples
(Tlf Hu eθ and (T2, H2, e2) to be a bounded linear map φl2\ΊIι'-+Ή2 so
that φ12{e^) = e2 and T2-φ12 = φ^ T^ We discuss some elementary facts
about this category. See [16] for terminology.
Let .Hi and H2 be two Hubert spaces with linear operators 2\
and T2. Then H, φ H2 and Tλ 0 H2 on J?i φ H2 are defined as usual.
(See [17].) It is well-known that if T.'.H.^H, and T2.H2-*HZ are
bounded linear operators then Tλ 0 T2 is also bounded and infact

We summarize some properties of this category ^ in the follow-

ing lemma.
580 V. N. SINGH

LEMMA 5.1. The following hold in the category ^:

( i ) A morphism φl2: (Tlf Hu et) —+ (T2, H2, e2) is an epimorphism
iff ΦuiHi) is dense in H2.
(ii) If (T, H, e) is cyclic then there is at most one morphism
from (T, H, e) to any other object (Tu Hl9 et) of <&.
(iii) (T, H9 e) is cyclic iff every morphism to it is an
epimorphism.
(iv) ^ has a terminal object, namely the triple (0, 0, 0).
( v ) The operation 0 is a sum as well as a product in ^ .
(vi) A morphism φ12: (Tu Hlf βx) —• (T2, H2, e2) is a monomorphism
iffKeτφ12 = {0}.

Proof. We prove only (vi) and leave the proof of the rest of
the proposition to the reader. If φ12: ί?Ί —* H2 is an injection then
clearly φ12: (Tu Hί9 ej —> (T2, H2y e2) is a monomorphism. To prove the
converse let φ12 be a monomorphism. Then we have to show that
Ker (φ12) = {0}. I t is easy to see that Ker (φ12) is a closed subspace of Hx
invariant under 2\. Consider the triple (Tx 0 Tu H, 0 Ker(&2), e1 0 0).
Define the morphisms / and g:(TLφ Tu Hx 0 Ker (φ12), ex φ 0) - *
(Tlf Hu ej by setting / = pι and g = φn + Φn where pt is the projec-
tion on the first coordinate and φn is the identity mapping of (T19 H19 ej
into (ϊ\, Hu e,) and φn: (Tlf Ker (φ12), 0) ~> (2\, Hu 0) is the injection of
Ker (φi2) into Ή.γ. Thus g(x ξ&y) = x + y. We see now that the
morphisms φ12g and φ12f are equal. However f Φ g. Thus φl2 is not
a monomorphism if Ker (φί2) Φ {0}.
The category ^ also admits the usual tensor product operation
(x). iZΊ ® H2 and its ϊ2-completion Hx (g) H2, Tx (x) T2 and its extension
Tiβ§Tt are defined as usual. I t is well-known that if TΊ and T2 are
bounded linear operaters then so is ϊ\ ® T2 and in fact || Tx ® Γ 2 1| ^
l l ϊ ί || || TJI. (See [6].)

LEMMA 5.2. The operation (x) is a product in &.

We leave the proof of this statement to the reader. It is easy
to check that (x) is not a sum in <g^ (see counterexample 2, § 6).
Unfortunately we do not know an abstract characterization of
the category <&.

DEFINITION 5.3. An atom in a category is an object such that

every morphism from it is either zero or a monomorphism.

Theorem 5.4. Let (T, H, e)e ^ be a cyclic triple then T has

no proper invariant subspaces in H iff (T, H9 e) is an atom in ^ .

Proof. If T has no proper invariant subspaces then it is clear

from (vi) Lemma 5.1 that (Γ, H, e) is an atom. If T has a proper
 REPRODUCING KERNELS AND OPERATORS WITH A CYCLIC VECTOR I 581

invariant subspace let it be given by the range of an orthogonal

projection P. Let (Γ, H, e) be written as (Γ, {p(T)e \ pe &*}, e) con-
sider also the triple (MΪ{&}P, 1) where the inner product on {^}P is
given by (p, q) = ((I - P)p(T)e, (I - P)q(T)e). Then (JIC WU 1)
is a cyclic triple (Theorem 4.5). Now the desired morphism
A
Φ*. (Γ, {p(T)e\pe^}, e) -> (M β, {^}P, 1) is given by φι2{p{T)e) = p.
It is not clear whether the category & also has an initial object.
2
However, if we adjoin to ^ the triple (Mz, H\ 1) where H is the
Hardy space of analytic functions defined on D1 whose boundary
values are square integrable on the unit circle, and obtain another
category &" then W certainly has an initial object; namely the
2
triple (Mz, H , 1). This will be given as a corollary in the latter
part of this paper.
We now make the following definition and show that the kernel
function map k defined as below connects ^ and the harmonic
analysis of the two-torus.

DEFINITION 5.5. The map k named the kernel function map is the
map which assigns to a member (Ty H, e) of c^ its kernel function.

Now we prove the following properties of k related to the

operations 0 and (x) of ^

THEOREM 5.6. If Kγ and K2 are the kernel functions for (Tίf ΐZΊ, ej
and (Γ2, H2, e2) respectively, then the kernel functions for (Tx 0 T2, Hx 0
Hi, &i 0 e2) and (ϊ\ (§) T2j H^ ® H2, eι (x) e2) are Kγ + K2 and Kx * K2 respec-
tively where K^K2 (z, w) = Γ* I"" K(ze~ίθf we-^)K2{eid, e**) (dθ/2π)(df/2π)
Jo Jo
and where dθj2π and dψ/2π denote the normalized Haar measure oj
the circle.
Proof. Let Cιm,n and C2m>n be the coefficients of zmwn in a Taylor
expansion of the kernel functions of the triples ( T Ί 0 T2, H1
and (2\ (g) T2, Hx (x) H2, e1 (x) e2) respectively.
Then,
Ci,n = ((T, 0 T2y(ei 0 e2\ {Tx 0 T2r{eλ 0 e2))Hι
= (2\* 0 T2\e, 0 e2), T? 0 T^e, ®e2))HlΘff2
= ( I X , T?eι)Hl + (ΓaΛe2, T?e2)H2
and
cι,n = ((T, (§ τ2γ(ei (g) β2), (Γ, (g) τ2yn(ei ® β
= (ΓΓ (g) Γ ^ (x) β 2 ), T Γ (x) Γ 2 m (βi ® β2))
582 V. N. SINGH

1 m n
and hence Σm,%=0 C mtnz~ w = K^z, w) + K2(z, w). We also observe
that since 2\ and T2 sue proper contractions Kγ and K2e A8iS for some
s > 1 and hence
2 m n m imθ n in
Σ C m,nz~ w = Γ* P Σ (ΓΓβi, T?e1)z e w e- +
m,n=0 Jo JO Jra,w=0

x Σ (T?e2, T2me2)e-imθein*dθ/2πdf/2π
m,n=0

- P Γ K(ze~iθ, w
jo Jo

DEFINITION 5.7. The equivalence relation ^ defined on ^ is

such that (Γ, H, e) ~ (Tu Hl9 ej iff there is a unitary map φ: 0(e)~>
7
0(βO so that φ(e) = ^ and ^ Γ = ϊ ^^.

Now let us consider the category ^ with the equivalence relation

~ and denote this new object by |£. Then the operations 0 and (x>
are defined on %_ and we have the following corollary.

THEOREM 5.8. The class % is ίsomorphic to a subset of the group

algebra of the torus and so it is a cancellative abelian semigroup
under the operation 0 . It also has no divisors of zero under the
operation (x).

We remark that the set of a.f.p.t.'s is closed under pointwise

multiplication. However, the set of a.f.p.d.t.'s is not closed under
pointwise multiplication. (See counterexample 3.)

6* Some examples and counterexamples*

EXAMPLES.
1. Consider (Mz, H\ 1). Then {Mz, H\ 1) is a cyclic triple and
its kernel function K is given by

K(z, w) = Σ (Λ zm)zmwn = (1 - zw)'1.

m,n=0

If q is an inner function then k(Mz, H2, 1) = k(Mz, H\ q).

2. The triple (F, U [0,1], 1), where V is the Volterra operator
on the Hubert space of square integrable real-valued functions on
the closed interval [0, 1] with Lebesgue measure, is also a cyclic triple.
Its kernel function K is given by

K(z, w) = Σ (Vn(l), Vm(l))z~mwn

0
I
β
β ((2+
+ w)dx
w)x = t-
o Z
Z +W
 REPRODUCING KERNELS AND OPERATORS WITH A CYCLIC VECTOR I 583

3. The kernel function of the triple (Ma, C, 1) is the function

ι
— az)~\l — aw)~ . This is a special case of an example to be
given in the latter part of this paper.

Counterexamples.
1. This counterexample shows that in general Hκ is not a space
of functions. We claim that the following holds.
Let (Mz, HKi 1) be a canonical triple with
2{ en+1) n n
K{z, W)=ΣA e ~ zw .

Let M(X) denote the space of functions (measurable or otherwise)

for a Borel field (X, A, μ) finite valued except for a set of measure
zero, and let it be given the topology of pointwise convergence.
Then the identity map i of the polynomials contained in H to the
polynomials contained in M(X) if injective is not continuous. This is
seen as follows.
First the space M(X) is complete. Second the sequence of func-
tions {fn}n where fn(z) — een~xzn is a complete orthonormal sequence in
Hκ. Hence the sequence of polynomials {p^neN given by pn{z) =
Σ?=o 2~Vί~V converges to a member h of Hκ. Now let X be any
subset of the complex plane β. Since i is injective XΦ{0}. It
suffices to show that the sequence pn(z)n does not converge for any
z e 0, z Φ 0. Let z = rei0. Then | pn(z) - pn^(z) | = e %~\r/2)n =
e% 1+wlogr/2
e ~ , which tends to oo if r Φ 0. Hence the sequence {pn(z)}n
is divergent unless z = 0.

2. If (ϊ7, H, e) is a cyclic triple then the subspace {p(T)e \p

can be given the structure of an abelian ring if we define
p(T)e q(T)e = pq(T)e. However, this operation is not continuous
and does not extend to all of H. Take for (Γ, H9 e) the triple
(Mz, H2, 1). Consider the convergent sequence {pn}n given by pn(z) =
Σm=i zm/ms/δ. The limit is of course in H2. However, a simple calcu-
lation shows that p\ does not converge in H2. We leave the calcu-
lation to the reader. (To see how it relates to (x) not being a sum
in ΐ f take for j\: (T, Hy e)->{T® T, H® H,e®e) the map f(T)e h->
/(Γ)e(x)eand for j2: (Γ, H, e)->(Γ® Γ, H(g)H, e (x) β) the map flr(Γ)e->
e (x) α(T)e and let (T3, i/"3, βs) be the triple (Γ, fl", e) with 013 and ^28 as
the identity map (f{T)e\-»f(T)e) then there is no map from ^:(Γ(§)
Γ, £Γ(g) ίί, e 0 e) -> (ϊ7, iJ, e) so that ^o^ = φιz and ^o^2 = φ2Z.)

3. We show now that every a.f.p.t. is not an a.f .p.d.t. as well

as the pointwise product of two a.f.p.d.t.'s is not necessarily an
a.f.p.d.t. The kernel function of the triple (1, C, 1) is the function
584 V. N. SINGH

1
K(z, w) — (1 — z)' ^ — w)~\ This is an a.f .p.d.t. Its square the
2 2
function Kx(z, w) = (1 — z)~ (l — w)~ is obviously an a.f.p.t. But it
is not an a.f.p.d.t. as Hκ is one dimensional and the a.f.p.d.t. cor-
ι
responding to a triple (Ma, 0, β) is the function β\l — az)~ (l — aw)~\

REFERENCES

1. N. Aronszajn, Theory of reproducing kernels, Trans. Amer. Math. Soc, 6 8 (1950),

337-404.
2. S. Bergman, The Kernel Function and Conformed Mapping, American Mathematical
Society Surveys Number V, New York, 1950.
3. A. Beurling, On two problems concerning linear transformations in Hilbert space,
Acta. Math., 8 1 (1949), 239-255.
4. G. Birkhofϊ and S. MacLane, A Survey of Modern Algebra, Revised edition,
Macmillan Co., New York, 1963.
5. S. Bochner and W. T. Martin, Several Complex Variables, Fourth printing, Princeton
University Press, 1964.
6. J. Dixmier, Les Algebres dy Operateurs danse Le Espace Hilbertian, (Algebres de Von
Neumann), Gauthier-Villaries, 55 Quai des Grandes Augustins, 1957.
7. N. Dunford and J. Schwartz, Linear Operators, Part II, Interscience, Wiley and
Sons, New York, Third printing, March, 1967.
8. I . C . Gohberg and M. G. Krein, Theory and Applications of Volterra Operators in
Hilbert Space, Translation from the Russian by A. Feinstein, Amer. Math. Soc, Provi-
dence, R. I., 1970.
9. P. R. Halmos, Finite Dimensional Vector Spaces, Second edition, Van Nostrand,
Princeton, N. J., 1958.
10. , Introduction to Hilbert Space, Second edition, Van Nostrand, Princeton,
N. J., 1957.
11. H. Helson, Lectures on Invariant Subspaces, Academic Press, New York, 1964.
12. E. Hille and R. S. Phillips, Functional Analysis and Semi-groups, Revised edition,
Amer. Math. Soc. Colloq. Publ., Vol. 31, Amer. Math. Soc, Providence, R. I., 1957.
13. L. Hormander, An Introduction to Complex Analysis in Several Variables, Van
Nostrand, Princeton, N. J., 1967.
14. J. L. Kelley and I. Namioka, Linear Topological Spaces, Van Nostrand, Princeton,
N. J., 1963.
15. M. S. Livsic, On Spectral Resolution of Linear Nonself Adjoint Operators, Mat.
Sb., 34 (76), (1954), 145-199. English translation Amer. Math. Soc. Translations (2), 5
(1957), 67-114.
16. S. MacLane, Homology, Academic Press, New York, 1963.
17. M. A. Naimark, Normed Rings, P. Noordhoff N. V., Groningen, Netherlands, 1964.
18. R. R. Phelps, Lectures on Choquets Theorem, Van Nostrand Mathematical Studies
7, Van Nostrand, Princeton, N. J., 1966.
19. G. C. Rota, On models for linear operators, Comm. Pure Appl. Math., 13 (1960),
469-472.
20. V. N. Singh, Reproducing Kernels and Operators with a Cyclic Vector, Thesis,
University of California, Berkeley, 1969.

Received February 27, 1973.

INDIAN STATISTICAL INSTITUTE