Cowen-Douglas operators and analytic continuation

arXiv:2505.05649v1 [math.FA] 8 May 2025

Pawel Pietrzycki

Abstract. In this paper, we study certain Banach spaces of analytic func-

tions on which a left-invertible multiplication operator acts. It turns out that,
under natural conditions, its left inverse is a Cowen-Douglas operator. We
investigate how the analytic continuations of functions from an invariant sub-
space of this Cowen-Douglas operator relate to the spectrum of its restriction
to that subspace.

1. Introduction
Let B be a Banach space. Denote by B(B) the Banach algebra of all bounded
linear operators on B. Given T ∈ B(B),Wwe write R(T ) and N (T ) for the range
and the kernel of T , respectively. Here stands for the (closed) linear span. By
C we denote the field of complex numbers. The symbols Z and N are reserved for
the sets of integers and positive integers respectively.
In [1], A. Aleman, S. Richter, and W. T. Ross studied a Banach space B of
analytic functions on D which has the following properties:

Mz B ⊂ B,
B ,→ Hol(D), the inclusion is continuous,
1 ∈ B,
Lλ B ⊂ B,
σ(Mz ) = D.

where for λ ∈ D, the operator Lλ : B → B is given by

f (z) − f (λ)
(Lλ f )(z) = , z ∈ D.
z−λ
This example include the classical Banach spaces of holomorphic functions in the
unit disc: the Hardy space, the Bergman space and the Dirichlet space and the
Besov space. They related meromorphic continuations of the functions from an
L-invariant subspace M of B to the spectrum of L|M . In this general setting, they

2020 Mathematics Subject Classification. Primary 47A10, 47B38, 47B48; Secondary 46E22.
Key words and phrases. Cowen-Douglas operators, left-invertible operator, multiplication
operator, analytic continuation, spectrum, invariant subspace.
1
2 P. PIETRZYCKI

proved that σ(L) = D, σ(L|M ) ⊂ D, for M invariant subspace of L and

1
σap (L|M ) ∩ D = σp (L|M ) ∩ D = {a ∈ D : ∈ M}.
1 − az
Moreover, under a regularity condition on B, they proved that
1
σap (L|M ) ∩ T = T \ { : every f ∈ M extends to be analytic
ζ
in a neighborhood of ζ}.
For a connected open subset Ω of C and n ∈ N, the Cowen-Douglas class Bn (Ω)
consists of the operators T ∈ B(B) which satisfy:
(i) Ω ⊆ σ(T ) = {ω ∈ C : T − ω not invertible},
(ii) R(T
W − ω) = B for ω ∈ Ω,
(iii) ω∈Ω ker(T − ω) = B,
(iv) dim ker(T − ω) = n for ω ∈ Ω.
Cowen-Douglas operators have played an important role in operator theory, servic-
ing as a bridge between operator theory and complex geometry. Namely, for an
operator T ∈ Bn (Ω) the mapping ω → (T − ω) gives rise to a Hermitian holomor-
phic vector bundle over Ω. Let us recall this notion. Let Λ be a manifold with a
complex structure and n be a positive integer. A rank n holomorphic vector bundle
over Λ consists of a manifold E with a complex structure together with a holomor-
phic map π from E onto Λ such that each fibre Eλ := π −1 (λ) is isomorphic to Cn
and such that for each λ0 in Λ there exists a neighborhood U of λ0 and holomorphic
functions γ1 ,γ2 ,...,γn from U to E whose values form a basis for Eλ at each λ ∈ Λ.
For an operator T ∈ Bn (Ω) let (ET , π) denote the subbundle of the trivial bundle
Ω × H defined by
ET = {(w, x) ∈ Ω × B : x ∈ N (T − ω)} and π(ω, x) = ω.

2. Main results
Let B be a Banach space of E-valued analytic functions defined on domain
Ω ⊂ C. Let Mz ∈ B(B) be a left-invertible multiplication operator with a left-
inverse L ∈ B(B), where E is a Hilbert space. We assume that Mz satisfies the
following five properties:
• B ,→ Hol(Ω, E) the inclusion map is continuous,
• dim N (L) = n,
• every f ∈ N (L) is analytic in an open neighborhood of Ω̄,
• for every f ∈ B and every λ ∈ Ω, there exist g ∈ B and h ∈ N (L)
f = (Mz − λ)g + h,
• σ(Mz ) = Ω̄.
This class includes the class described in Introduction, as well as certain adjoints
of Cauchy dual operators for left-invertible operators (see the Shimorin analytic
model [8] and its extensions in [4]).
We prove that the inverse Rλ of (I − λL) has the following property: for every
f ∈ M there exists cλ (f ) ∈ N (L) such that
(Mz − λ)Rf = Mz f − λcλ (f ), (2.1)
 COWEN-DOUGLAS OPERATORS AND ANALYTIC CONTINUATION 3

Preliminary research show that this property is equivalent to the following one,
expressed in terms of spectrum
1
λ ∈ σ(Mz ) =⇒ ∈
/ σ(L).
λ
Observe that if we substitute z = λ in (2.1), where λ ∈ Ω, λ ̸= 0 then we get
f (λ) = cλ (f )(λ). This makes it possible to define the following function
f˜: D ∋ λ → cλ (f )(λ) ∈ E, (2.2)
where D = {λ ∈ C : 1
λ / σ(L|M )}. As before f˜(λ) = f (λ), for λ ∈ Ω.
∈
Theorem 2.1. Suppose that the sequence {Mnz Ln } converges to 0 in the strong
operator topology. Then L is a Cowen-Douglas Bn (Ω ′ ), where Ω ′ = { λ1 : λ ∈ /
σ(Mz )}.
Theorem 2.2. Let M be an invariant subspace of L. Then the following con-
ditions hold:
1 1
σap (L|M ) ∩ { : λ ∈
/ σ(Mz )} = σp (L|M ) ∩ { : λ ∈/ σ(Mz )}
λ λ
1
= { : there exist h such that (Mz − λ)−1 h ∈ M}
λ
Theorem 2.3. Let B the following additional condition: If f ∈ B(B) is analytic
in an open neighborhood of a point ξ ∈ ∂σ(T ), λn → ξ and
(Mz − λ)hn = zf − λn cλn (f ) and (Mz − ξ)h = zf − ξcξ (f ) (2.3)
Then hn → ∞. Then
σ(L|M ) ∩ ∂σ(Mz ) = ∂σ(Mz )\{1/ξ ∈ σ(Mz ) : every f ∈ M extends
to be analytic in a neighborhood of ξ}
Note that since L is a (non-invertible) Cowen-Douglas operator the space N (L)
is finite-dimensional. In the case where the operator L is a backward shift the space
N (L) is one-dimensional and consists of constant functions. However, in the case
where L is any Cowen-Douglas operator, the situation is more complicated. Some
examples show that elements of N (L) behave better than a regular function from
space H, that is, they expand beyond the Ω. For example, let T be a left-invertible
operator such that T can be modelled as a multiplication operator on space of
vector-valued holomorphic functionPon annulus. If T ′∗ is a Cowen-Douglas operator
∞
and h ∈ N (L). Note that, since n=0 (PE T ′∗n h)z n = 0, it follows from [4] that
the series for h is as follows:
∞
X 1
Uh (z) = (PE T n h) n .
n=1
z
Therefore this series is absolutely convergent outside some disc.

3. Proofs
Proof of Theorem 2.2. Let λ ∈ C \ σ(Mz ), and take h ∈ N (L). Then
(Mz − λ)(Mz − λ)−1 h = h. Multiplying both sides by L, we obtain (I − λL)(Mz −
λ)−1 h = 0. If (Mz − λ)−1 h ∈ M then it is an eigenvector of L corresponding to
the eigenvalue λ1 . Hence, λ1 ∈ σ(L|M ).
4 P. PIETRZYCKI

Now, suppose λ1 ∈ σap (L|M ) ∩ { λ1 : λ ∈

/ σ(Mz )}. Then, there exists a sequence
{fn } ⊆ M such that ∥fn ∥ = 1 and
lim ∥(I − λL)fn ∥ = 0. (3.1)
n→∞

Note that (Mz L − I)fn ∈ N (L) and the sequence {(Mz L − I)fn } is bounded.
Since N (L) is finite-dimensional, by the Bolzano-Weierstrass theorem, we can,
without loss of generality assume that {(Mz L − I)fn } is convergent to some h ∈
N (L). Using the identity
Mz (I − λL)fn = λ(I − Mz L)fn + (Mz − λ)fn .
and taking the limit as n → ∞, it follows from (3.1) that limn→∞ ∥(Mz − λ)fn −
h∥ = 0. Since λ ∈/ σ(L), we obtain λ1 ∈ σp (L|M ). □

Lemma 3.1. Let M be an invariant subspace of L. Then the following condition

holds:
(i) If λ1 ∈
/ σ(L|M ) then the inverse of (I − λL)|M has the following property:
for every f ∈ M there exist cλ (f ) ∈ N (L) such that
(Mz − λ)Rλ f = Mz f − λcλ (f ), (3.2)
/ σ(L) and f = (Mz − λ)g + h, where h ∈ N (L), then g = (I −
(ii) If λ1 ∈
λL)−1 Lf .
1
Proof. (i) Let λ ∈ C be such that λ ∈
/ σ(L|M ). Then there exists Rλ ∈ B(M)
such that
(I − λL)|M Rλ f = f, f ∈ M.
Multiplying both sides by Mz and rearranging terms yields
Mz f = (Mz − λ)Rλ f + λ(I − Mz L)Rλ f, f ∈ M.
Note that λ(I − Mz L)Rλ f ∈ N (L), since L(I − Mz L) = L − LMz L = 0. There-
fore, we conclude that for every for every f ∈ M there exist cλ (f ) ∈ N (L) such
that (3.2) holds as claimed in part (i).
(ii) Suppose f = (Mz − λ)g + h, where h ∈ N (L). Appying L to both sides
gives Lf = (I − λL)g. As (I − λL) is invertible by assumption, we can solve for g
to obtain g = (I − λL)−1 Lf . □

Theorem 3.2. Let Mz ∈ B(B) be a left-invertible operator and L ∈ B(B) be

its left-inverse. Then the following conditions are equivalent:
(i) for every f ∈ B and every λ ∈ Ω, there exist g ∈ B and h ∈ N (L) such
that
f = (Mz − λ)g + h,
1
(ii) λ ∈ Ω =⇒ λ ∈
/ σ(L).
1
Proof. (ii)⇒(i) Suppose that λ ∈ Ω then λ ∈
/ σ(L), so the operator I − λL
is invertible. Define
g := (I − λL)−1 Lf and h := f − (Mz − λ)(I − λL)−1 Lf.
Since L(Mz − λ) = (I − λL) it implies that h ∈ N (L).
 COWEN-DOUGLAS OPERATORS AND ANALYTIC CONTINUATION 5

(i)⇒(ii) Fix λ ∈ Ω. Define operator Rλ ∈ B(B) by setting Rλ f = g, where

Mz f = (Mz − λ)g + h for some h ∈ N (L). Therefore
Mz f = (Mz − λ)Rλ f + h.
Applying L to both sides yields
f = (I − λL)Rλ f.
which shows that Rλ is injective. To prove surjectivity, take any g ∈ B and define
h := Mz (I − λL)g − (Mz − λ)g = −λ(Mz L − I)g.
Then h ∈ N (L). Putting f := (I − λL)g, we get
zf = (Mz − λ)g + h.
Since (I −λL)Rλ = I and Rλ is bijection it follows that I −λL is invertible. To prove
that Rλ is bounded, suppose that {fn } is a sequence in B such that limn→∞ fn = 0
and limn→∞ Rλ fn = g for some g ∈ B. Then
0 = lim fn = lim (I − λL)Rλ fn = (I − λL)g.
n→∞ n→∞
Since (I − λL) is injective, we get that g = 0. By the Closed Graph Theorem Rλ
is bounded. □
Lemma 3.3. Let T ∈ B(B) be a left-invertible but not invertible operator, and
/ σ(T ), we have λ1 ∈ σ(L).
L ∈ B(B) be its left-inverse. Then for every λ ∈
Proof. Suppose λ ∈ / σ(T ). Then T − λ is invertible, so there exists S ∈ B(H)
such that (T −λ)S = I and S(T −λ) = I. This implies that (I−λL)S = L(T −λ)S =
L. Multiplying both sides on the left by T , we obtain (I − λL)ST = I. On the
other hand, starting from I = S(T − λ), we expand:
I = S(T − λT L + λT L − λ) = ST (I − λL) − λ(I − T L).
Now, if T were invertible, we would have T L = I, and the second term would
vanish. However, since T is not invertible, T L ̸= I, so I ̸= ST (I − λL). This
combined with (I − λL)ST = I shows that λ1 ∈ σ(L). □
Lemma 3.4. Let
_
P= {Mkz h : h ∈ N (L), k ∈ Z+ },
_
S = (Mz − λ)−1 h : λ ∈ C \ σ(Mz ), h ∈ N (L)}.
Then P = S. Moreover, if the sequence {Mnz Ln } converges to 0 in the strong
operator topology, then B = P = S.
Proof. (i) To show S ⊂ P, fix λ ∈ C \ σ(Mz ) such that |λ| > ∥Mz ∥ and
h ∈ N (L). Consider the polynomial
n
X 1
pn = − Mk h.
λk+1 z
k=0
We estimate the norm of the difference:
1
∥pn − (Mz − λ)−1 h∥ = ∥ Mn+1 (Mz − λ)−1 h∥
λn+2 z
1
≤ ∥Mz ∥n+1 ∥(Mz − λ)−1 h∥
|λ|n+2
6 P. PIETRZYCKI

Since |λ| > ∥Mz ∥, we see that limn→∞ pn = (Mz − λ)−1 h. Therefore, S ⊂ P.
Now, we show P ⊂ S. We proceed by induction. First, let us note that the
following limit holds:
Mz
lim (Mz − λ)−1 h = lim ( − 1)−1 h = h, h ∈ N (L).
λ→∞ λ
λ→∞

Therefore N (L) ⊂ S. Now suppose, that for 0 < k < n for some n ∈ N ∪ {0}, then
for λ ∈ C \ σ(Mz ), we have
n
X
Mn+1
z (Mz − λ)−1 h = λk Mzn−k h + λn+1 (Mz − λ)−1 h, h ∈ N (L),
k=0

Furthermore,
∥λMn+1
z (Mz − λ)−1 h − Mn+1
z h∥ ≤ ∥Mn+2
z (Mz − λ)−1 h∥ (3.3)
But since λ → (Mz − λ)−1 h is an analytic function, then the right-hand side of
(3.3) goes to zero as |λ| → ∞. Thus Mn+1z h ∈ S for h ∈ N (L) and so P ⊂ S.
(ii) Fix f ∈ B. Observe that, since L(I − Mz L) = 0, it follows that
hk := (I − Mz L)Lk f ∈ N (L), k ∈ N.
It now suffices to observe that we can write
n−1
X n−1
X
f = Mnz Ln f + Mkz (I − Mz L)Lk f = Mnz Ln f + Mkz hk .
k=0 k=0

Proof of Theorem 2.1. It follows from Theorem 2.2 that Ω ′ ⊂ σ(L).

Let λ1 ∈ Ω, so there exists Rλ ∈ B(B) such that (Mz − λ)Rλ = I. Multyplying
both sides by L, we get (I − λL)Rλ =WI. Thus R(L − ω) = B. Combining
Theorem 2.2 and Lemma 3.4 we see that ω∈Ω ker(L − ω) = B. It is a general fact
that conditions (i) and (ii) imply that dim ker(T −ω) is constant (see [3, p. 189]). □

Proof of Theorem 2.3. Let 1ξ ∈ ∂σ(Mz ) such that 1/ξ ∈ / σ(L|M ). Then
the following function is an analytic extension of f in a neighborhood of ξ
f˜: D ∋ λ → cλ (f )(λ) ∈ E,
where D = {λ ∈ C : λ1 ∈ / σ(L|M )}.
Now suppose that for every f ∈ M extends to be analytic in a neighborhood
of ξ. By the additional condition (2.3), the function g such that (Mz − ξ)g = zf −
ξcξ (f ) is the norm limit of functions gn defined by (Mz −ξ)gn = zf −λn cλn (f ) ∈ M
for some λn ∈ Ω, thus g ∈ M. We show that the inverse Rλ of (I − λL)|M is given
by Rλ f = g, where g is defined by the equation (Mz − ξ)g = zf − ξcξ (f ). We
show that Rλ is bounded. Suppose that {fn } is a sequence in M such that fn → 0
and gn → h, where (Mz − ξ)gn = zfn − ξcξ (fn ). Let c ∈ N (L) be such that the
sequence {cξ (fn )} converges to c. Thus
(Mz − ξ)h = −ξc,
which yields 0 = (Mz − ξ)h + ξc. By uniqueness this can only hold when h = 0.
Thus by the closed graph theorem, Rλ is a continuous operator on M and so
1/ξ ∈ σ(L|M ). □
 COWEN-DOUGLAS OPERATORS AND ANALYTIC CONTINUATION 7

References
[1] A. Aleman, S. Richter, W. T. Ross, Pseudocontinuations and the backward shift, Indiana U.
Math. J. 47 (1998), 223–276.
[2] S. Chavan, J. Stochel, Weakly concave operators, Proceedings of the Royal Society of Edin-
burgh Section A: Mathematics 154.1 (2024): 1–32.
[3] M. J. Cowen, R. G. Douglas. Complex geometry and operator theory. Acta Mathematica 141
(1978): 187–261.
[4] P. Pietrzycki, A Shimorin-type analytic model on an annulus for left-invertible operators and
applications, J. Math. Anal. Appl. 477 (2019), 885–911.
[5] P. Pietrzycki, Generalized multipliers for left-invertible operators and applications, Integr.
Equ. Oper. Theory 92 (2020), 41, 31pp.
[6] P. Pietrzycki, On the Cauchy dual operator and duality for Banach spaces of analytic func-
tions, Studia Mathematica 271 (2023), 121–150.
[7] W.T. Ross, H. S. Shapiro, Generalized analytic continuation, University Lecture Series, (Prov-
idence: American Mathematical Society, RI) Vol. 25 (2002), pp. xiv149.
[8] S. Shimorin, Wold-type decompositions and wandering subspaces for operators close to isome-
tries, J. Reine Angew. Math. 531 (2001), 147–189.

Wydzial Matematyki i Informatyki, Uniwersytet Jagielloński, ul. Lojasiewicza 6,

PL-30348 Kraków
Email address: [email protected]