On a conjecture of de Branges

Igor Bereza
a
Department of Mathematics and Computer Science, St. Petersburg State University, 7/9 Universitetskaya
nab., St. Petersburg, 199034, Russia
arXiv:2507.12576v1 [math.FA] 16 Jul 2025

Abstract
We provide a broad class of counterexamples to a conjecture of L. de Branges concerning
the superfluity of the continuity property in the axiomatic description of de Branges spaces.
Keywords: de Branges spaces, reproducing kernel Hilbert spaces
2020 MSC: 46E22, 30H45

1. Introduction
We study the minimality of the following axiomatic description of a class of Hilbert spaces
of entire functions called de Branges spaces (known as de Branges axioms) introduced in [1]:
(H1) Whenever F is in the space and has a non-real zero w, the function F (·)(·− w̄)(·−w)−1
belongs to the space and has the same norm.
(H2) For every complex number w, the linear functional defined on the space by F 7→ F (w)
is continuous.
(H3) Whenever F is in the space, the function F ∗ (z) = F (z̄) belongs to the space and has
the same norm.
The aim of this article is to show that it is possible to transform any de Branges space
from a suitable class into a Hilbert space of entire functions that violates (H2), but satisfies
(H1) and (H3).
One of the most useful properties of de Branges spaces is [2, Theorem 23], given below.
It states that de Branges spaces admit the following constructive definition, which is often
easier to work with once it is established that the axioms are fulfilled.
Definition 1. An entire function E belongs to the Hermite-Biehler class if |E ∗ (z)| < |E(z)|
for all z ∈ C+ .
Definition 2. The de Branges space H(E) associated with a Hermite-Biehler entire func-
tion E is the set of all entire functions F such that
ˆ 2
2 F (t)
||F ||E = dt < +∞
E(t)
R
∗ 2 +
and F/E, F /E belong to H (C ), the Hardy space on the upper half-plane.
Preprint submitted to Journal of Functional Analysis July 18, 2025
Theorem 1 (de Branges). A Hilbert space of entire functions H = ̸ {0} satisfies (H1) –
(H3) iff there exists an entire function E of the Hermite-Biehler class such that H = H(E)
and || · ||H = || · ||E .

We shall not use the constructive definition, it is mentioned only as a part of a brief
historical account. Now we turn our attention to the axiomatic description of de Branges
spaces.
Various aspects of independence of the de Branges axioms have been studied before.
For example, L. de Branges showed that the following result holds. Although his work [2,
Problem 54] includes only the statement, one may find a detailed proof, for instance, in [3].

Statement (de Branges). Let H ̸= {0} be a Hilbert space of entire functions which satisfies
(H1) and (H2). Then there exists a de Branges space H(E) and an entire function U with
U U ∗ ≡ 1 such that F 7→ U F is an isometric transformation of H onto H(E).

We are concerned with independence of the axiom (H2). The subject in question was
considered in the original de Branges’ papers: in [1] L. de Branges showed that Theorem 1
still holds if one weakens the axiom (H2) by requiring the continuity of the point evaluation
functionals only for non-real points. Moreover, at the bottom of page 445 of [4] L. de Branges
posed the following conjecture, which we quote verbatim.

Conjecture (de Branges). The axiom (H2) which appears here is conjectured to be a con-
sequence of (H1).

We shall show that the Conjecture is false using the following class of Hilbert spaces of
entire functions.

Definition 3. Let M be the set of all Borel measures µ supported on infinite discrete (i.e.
having no finite accumulation points) subsets of the real line. We denote by {ti }i∈N an
enumeration of supp µ and by µn the mass of µ at tn .

Definition 4. A Hilbert P space of entire functions H is isometrically isomorphic to ℓ2 (µ),

where µ ∈ M, if ||F ||2H = n µn |F (tn )|2 and the map F →
7 {F (ti )}i∈N is bijective.

Remark 1. For Hilbert spaces of entire functions that are isometrically isomorphic to ℓ2 (µ)
for some µ ∈ M the requirement of equality of norms in the properties (H1) and (H3) is
not needed, since µ is supported on the real line.

Our main result, proven in paragraph 2.2, is as follows.

Theorem 2. For every µ ∈ M and a de Branges space H, isometrically isomorphic to

ℓ2 (µ), there exists a Hilbert space of entire functions H
e such that

• H
e is isometrically isomorphic to ℓ2 (µ);

• H
e satisfies (H1) and (H3), but not (H2);
2
 • H∩H
e is dense in H and H.
e
To refute the Conjecture for the aforementioned class of Hilbert spaces, Theorem 2
requires us to start with a “bootstrap” de Branges space H, which is isometrically isomorphic
to ℓ2 (µ). The following folklore-type result, the proof of which will be given for completeness
in paragraph 2.3, allows us to complete the construction of our counterexamples.
Theorem 3. For every measure µ ∈ M there exists a de Branges space H that is isomet-
rically isomorphic to ℓ2 (µ).

2. Proofs
2.1. Preliminaries
Our construction is an extension of a standard way of producing a discontinuous func-
tional on a Hilbert space: pick a linearly independent sequence {xn } converging to x0 and
set the functional to one on {xn } and to zero on x0 , then extend this partially-defined func-
tional in any way to the entire Hilbert space. In this paragraph we develop the necessary
machinery for extending linear functionals from a linear subset of Hilbert spaces.
Definition 5. A generating set Q of a vector space U is a subset of U such that every
element of U is a finite linear combination of elements of Q.
Definition 6. A Hamel basis of a vector space is a linearly independent generating set.
Theorem 4 (Existence of a Hamel basis). Let P be a linearly independent set contained in
a generating set Q, then there exists a Hamel basis B satisfying P ⊆ B ⊆ Q.
While in the general case the above Theorem would suffice for extending a linear func-
tional, the space H
e must satisfy (H3). For this purpose we introduce an auxiliary definition
of a symmetric linear subset of a Hilbert space and a Lemma which allows us to partition a
Hilbert space into two symmetric linear pieces.
Definition 7. We call a linear subset L ⊆ ℓ2 (µ), where µ ∈ M, symmetric if v̄ ∈ L when
v ∈ L, where v̄ denotes {vi }i∈N .
Lemma 1. For any symmetric linear subset L ⊆ ℓ2 (µ), where µ ∈ M, there exists a sym-
metric linear subset M ⊆ ℓ2 (µ), such that every element of ℓ2 (µ) can be uniquely expressed
as a sum of elements of L and M .
Proof. For a linear subset V ⊆ ℓ2 (µ) let
[
QV = {Re(v), Im(v)},
v∈V

where Re(v) = {Re(vi )}i∈N and Im(v) = {Im(vi )}i∈N . Clearly, V is symmetric if and only if
QV ⊆ V .
If L = {0}, put M = ℓ2 (µ). Otherwise, let c0 ∈ L be a non-zero real-valued sequence,
and let BL be a Hamel basis of L extending {c0 } and consisting of real-valued sequences, by
applying Theorem 4 with P = {c0 } and Q = QL . Applying Theorem 4 again with P = BL
and Q = Qℓ2 (µ) , we can extend BL to Bℓ2 (µ) – a Hamel basis of ℓ2 (µ), consisting of real-valued
sequences. Putting M = span(Bℓ2 (µ) \BL ) concludes the proof.
3
2.2. Construction of the space H
e
Definition 8. For µ ∈ M, w ∈ C\ supp µ and v ∈ ℓ2 (µ) define
Dw (v) = {vi (ti − w)−1 }i∈N .
Proof of Theorem 2. Let W be an isometric isomorphism of H onto ℓ2 (µ). For v ∈ ℓ2 (µ)
and V ⊆ ℓ2 (µ) define
Fv = W −1 (v), FV = Ran(W −1 |V ).
Let A be the operator of multiplication by the independent variable, i.e. Avn = tn vn for
v ∈ ℓ2 (µ), then A is an unbounded self-adjoint operator with dense in ℓ2 (µ) domain L.
By Lemma 1 there exists a symmetric linear subset M ⊆ ℓ2 (µ), with every element
of ℓ2 (µ) uniquely expressible as a sum of elements of L and M . Since A is closed and
unbounded, M ̸= {0}. Therefore, there exist z0 ∈ R\ supp µ and u0 ∈ M with Fu0 (z0 ) ̸= 0.
Finally, for a sequence u ∈ M let
Gu (z) = (z − z0 )FDz0 (u) (z).
Put
He = {Hv = Gu + Fc | v = u + c ∈ ℓ2 (µ), u ∈ M, c ∈ L}.
It is clear that the set H
e with the norm Hv 7→ ||v||ℓ2 (µ) is a Hilbert space of entire functions
that is isometrically isomorphic to ℓ2 (µ). Since L is dense in ℓ2 (µ), we have FL ⊆ H ∩ H e is
a dense subset of H and H. e By Remark 1 we do not need to check the equality of norms in
axioms (H1) and (H3) for the space H. e
He satisfies (H3):
For every v ∈ ℓ2 (µ) we have Hv∗ = G∗u + Fc∗ = Gū + Fc̄ ∈ H, e since L, M are symmetric
and H satisfies (H3).
He satisfies (H1):
If v ∈ ℓ2 (µ) and Hv (w) = 0 for some w ∈ C\R, then
Hv (z)(z − w)−1 = (Gu (z) + Fc (z))(z − w)−1 =
((z − z0 )FDz0 (u) (z) + Fc (z))(z − w)−1 =
FDz0 (u) (z) + ((w − z0 )FDz0 (u) (z) + Fc (z))(z − w)−1 =
FDz0 (u) (z) + F(w−z0 )Dz0 (u)+c (z)(z − w)−1 .
Furthermore, F(w−z0 )Dz0 (u)+c (w) = (w−z0 )FDz0 (u) (w)+Fc (w) = Gu (w)+Fc (w) = Hv (w) = 0
and (w − z0 )Dz0 (u) + c ∈ L as Dz0 (u), c ∈ L.
Applying the property (H1) to F(w−z0 )Dz0 (u)+c ∈ FL ⊆ H we have
F(w−z0 )Dz0 (u)+c (·)(· − w)−1 = FDw ((w−z0 )Dz0 (u)+c) ∈ FL ⊆ H,
e

therefore Hv (·)(· − w)−1 ∈ H.

e
He violates (H2):
Since L is dense ℓ2 (µ) there exists a sequence {cn } ⊆ L converging to u0 in ℓ2 (µ). The
space H satisfies (H2), therefore Fcn (z0 ) → Fu0 (z0 ) ̸= 0, but Gu0 ∈ H
e vanishes at z0 , so
Fcn (z0 ) ↛ Gu0 (z0 ).
4
2.3. Construction of a de Branges space that is isometrically isomorphic to ℓ2 (µ)
Before establishing Theorem 3, we prove a technical Lemma using the following state-
ment.

Theorem 5 (Mittag-Leffler). Let {an } be a discrete subset of the complex plane. For each
natural n let
Kn
X cn,k
pn (z) = .
k=1
(z − an )k
Then there exists a meromorphic function f whose poles are precisely {an } and for every
n ∈ N f − pn has a removable singularity at an . Moreover, if cn,k , an ∈ R for all n, k, then
f may be chosen such that f ∗ = f .

Lemma 2. For every µ ∈ M there exists an entire function S, vanishing precisely on

supp µ, all zeroes of which are simple, satisfying S ∗ = S and
X 1
< +∞.
n∈N
|S ′ (tn )|2 µn

Proof. Pick {sn } – a sequence of non-zero real numbers with n µ−1 −2

P
n |sn | < +∞. Let Π be
∗
the Weierstrass canonical product for the sequence {tn }, then Π = Π is an entire function
whose zero set coincides with supp µ. For each n ∈ N define pn (z) = cn (z − tn )−1 , where
cn = Π′ (tn )−1 log |sn /Π′ (tn )| ∈ R.
By Theorem 5 there exists a meromorphic function f such that f ∗ = f and f having the
residue cn at the point tn for every n ∈ N.
Let T = Πf , then T is an entire function satisfying T ∗ = T and T (tn ) = log |sn /Π′ (tn )|.
∗
Finally, put S = ΠeT , then |S ′ (tn )| = |sn | and S ∗ = Π∗ eT = S.
Proof of Theorem 3. Fix an entire function S satisfying the conclusions of Lemma 2 for the
measure µ. Put ( )
X S(z) 1
H = Xv (z) = vn ′
v ∈ ℓ2 (µ) .
n∈N
z − tn S (tn )

We shall verify that the space H satisfies the de Branges axioms (H1) – (H3), then by The-
orem 1 H is a de Branges space.
H is a Hilbert space of entire functions and is isometrically isomorphic to ℓ2 (µ):
Let K be a compact subset of the complex plane. Since supp µ is a discrete set,
max max |S(z)(z − tn )−1 | ⩽ CK < +∞, therefore
n∈N z∈K
s
X √ 1 X 1
sup |Xv (z)| ⩽ CK |vn | µn √ ⩽ CK ||v||ℓ2 (µ) < +∞.
z∈K
n∈N
|S ′ (tn )| µn n∈N
|S ′ (t )|2 µ
n n

5
By the Weierstrass M-test, the series converges uniformly on K, so Xv is continuous. More-
over, for a triangle γ ⊆ C we have
˛ ˛ X X vn ˛ S(z)
S(z) 1
Xv (z)dz = vn dz = dz = 0,
n∈N
z − tn S ′ (tn ) n∈N
S ′ (tn ) z − tn
γ γ γ

so by Morera’s Theorem Xv is entire.

Notice that Xv (tn ) = vn , so Xv (tn ) = 0 ∀n iff Xv ≡ 0, therefore Xv 7→ ||v||ℓ2 (µ) is a norm on
H, which is isometrically isomorphic to ℓ2 (µ) by definition.
H satisfies (H3):
Direct computation shows that Xv∗ = Xv̄ ∈ H, since v̄ ∈ ℓ2 (µ).
H satisfies (H1):
Suppose that Xv (w) = 0 for some w ∈ C\R. Then S(w) ̸= 0 and

X S(z) 1
XDw (v) (z) = vn ′
=
n∈N
(z − tn )(tn − w) S (tn )
! 0
1 X S(z) 1 X 1 1 Xv (z) S(z)  :
vn ′
− S(z) vn ′
= −  Xv (w).
z − w n∈N z − tn S (tn ) n∈N
w − tn S (tn ) z−w  (z−w)S(w)

H satisfies (H2):
√
If z = tm for some m ∈ N, then |Xv (tm )| ⩽ µm −1 ||v||ℓ2 (µ) . Otherwise,
s
X √ S(z) 1 X 1
|Xv (z)| = vn µn ′
√ ⩽ ||v||ℓ2 (µ) |S(z)| max |z − tn |−1 .
n∈N
z − tn S (tn ) µn n
n∈N
|S ′ (t 2
n )| µn

3. Acknowledgements
The author is deeply grateful to Vladimir Kapustin for stating the initial problem, and to
Roman Romanov for providing helpful suggestions and pointing out the connection between
the author’s undergraduate thesis and the de Branges conjecture.

References
[1] L. de Branges, Some Hilbert Spaces of Entire Functions, Proceedings of the American Mathematical
Society 10 (1959) 840–846. URL: https://www.jstor.org/stable/2033484.
[2] L. de Branges, Hilbert Spaces of Entire Functions, Prentice-Hall, Inc., Englewood Cliffs, N.J., 1968.
[3] R. Romanov, Functional Description of a Class of Quasi-Invariant Determinantal Processes, An-
nales Henri Poincaré (2024). URL: https://doi.org/10.1007/s00023-024-01510-6. doi:10.1007/
s00023-024-01510-6.
[4] L. de Branges, Some Hilbert Spaces of Analytic Functions, Transactions of the American Mathematical
Society 106 (1963) 445–468. URL: https://www.jstor.org/stable/1993754.
6