A genus obstruction for commuting boundary operators
Henry Zweiman
September 15, 2026
Abstract
Let be a compact connected oriented smooth Riemannian surface with nonempty smooth boundary. We prove that if its Dirichlet-to-Neumann map commutes with the boundary Laplacian, then has genus zero. This answers both cases of Speciel's Open Problem 1.6 and, together with his genus-zero classification, removes the topological restriction from the classification of oriented surfaces with commuting boundary operators. For connected boundary, the finite-dimensional obstruction to harmonic conjugation forces an exact determinant identity on all sufficiently high Fourier planes. The algebra of holomorphic boundary traces then produces a holomorphic map of degree one. For several boundary components, we prove that conformally attaching a Euclidean disk along an arclength identification preserves commutation. The proof separates the finite-period input, the multiplicative trace argument, and the variational elimination of the capped boundary.
Let be a compact connected smooth Riemannian surface with nonempty smooth boundary. Its Dirichlet-to-Neumann map is
where is the outward unit normal and is the nonnegative Laplacian. Write for the nonnegative Laplacian of the induced boundary metric. We consider the exact commutation condition
Speciel [S, Theorem 1.3] classified the surfaces satisfying (1.2) under the additional assumption that the genus is zero or that there are at least three boundary components. His classification contains disks, logarithmic ovals, and flat cylinders, with the equivalence appropriate to two-dimensional Dirichlet-to-Neumann data. The remaining cases, a positive-genus oriented surface with one or two boundary components, are explicitly posed in [S, Open Problem 1.6]. The numbering here refers to the final journal article; the corresponding question is Open Problem 1.12 in its arXiv version.
Theorem 1.1 (Genus obstruction). Let be a compact connected oriented smooth Riemannian surface with nonempty smooth boundary. If (1.2) holds, then has genus zero.
The theorem imposes no curvature condition and no restriction on the number or lengths of the boundary components. In particular, neither of the positive-genus surfaces in [S, Open Problem 1.6] admits the requested metric.
Recall that two surfaces are -isometric if a diffeomorphism identifies their metrics after a conformal change with on the boundary. Both operators in (1.2) are preserved by this equivalence. A logarithmic oval is, up to Euclidean similarities, the image of the closed unit disk under , , using a branch holomorphic near the disk. We use the flat cylinder models of [S].
Corollary 1.2 (Classification without a genus restriction). A compact connected oriented smooth Riemannian surface with nonempty smooth boundary satisfies (1.2) if and only if it is -isometric to a disk, a logarithmic oval, or a flat cylinder.
Proof. Theorem 1.1 reduces the assertion to the genus-zero case of [S, Theorem 1.3], which supplies both the classification and its converse.
The new assertion is the genus obstruction. The models, their commutation property, and their classification in genus zero belong to Speciel. Our proof does not start from a uniformization of a positive-genus surface by a planar domain. Instead, it constructs a degree-one holomorphic map from the boundary operator identity.
When the boundary is connected, let denote positively oriented arclength differentiation. Harmonic conjugation can fail globally because of periods, but these periods lie in a finite-dimensional cohomology space. Consequently, on mean-zero functions, has finite rank. This is a classical feature of the boundary Hilbert transform in the inverse problem for Riemann surfaces; see [B] and the formulation in [BK, Section 2]. We give the needed period argument directly. Under (1.2), the finite-rank defect becomes an exact determinant constraint on every sufficiently high two-dimensional Fourier plane.
The further input is multiplicative. Smooth holomorphic boundary traces form an algebra, characterized by . Condition (1.2) makes that algebra invariant under Fourier-plane projections. Multiplying adjacent high-frequency traces and projecting to frequency one constructs an ellipse-valued boundary parametrization. The exceptional case in which all negative high-frequency coefficients vanish produces a disk-valued parametrization by taking a quotient of two holomorphic functions. In both cases, the argument principle proves that the map has degree one.
For disconnected boundary we attach a Euclidean disk along one boundary component by smooth conformal welding. Two-dimensional Dirichlet energy is unchanged on each piece by conformal rescaling. Minimizing over the seam trace gives a Schur complement of the original Dirichlet-to-Neumann map and the disk operator. The boundary arclength identification makes the latter the square root of the seam Laplacian. This proves that commutation survives capping, including at the constant mode. Iterating reduces to connected boundary without changing genus.
2. Holomorphic traces and the period defect
We use the complex structure determined by the orientation and conformal class of . A holomorphic function on will mean one holomorphic in the interior and smooth up to the boundary. Orient positively, so that is a positively oriented orthonormal frame when is the positive unit tangent. Set on each component. Our Hodge-star convention is in an oriented isothermal coordinate .
For complex-valued traces we extend complex linearly. Green's identity gives
In particular, is self-adjoint and nonnegative on ; its kernel consists of the global constants, since is connected. Smooth elliptic boundary regularity is understood throughout.
Lemma 2.1 (Holomorphic trace criterion). A function is the trace of a holomorphic function on if and only if
Proof. If is holomorphic, then . At the boundary this gives and , proving (2.2).
Conversely, extend harmonically to . By (2.1), (2.2), and Stokes' theorem,
The imaginary term is . Pointwise,
Its integral is zero by (2.3), hence . Thus is holomorphic.
Let be the space of these boundary traces. It contains the constants and is closed under multiplication. Lemma 2.1 identifies it as the kernel of on smooth complex functions. This standard trace-algebra viewpoint is also used in the inverse problem; our use below concerns its additional Fourier invariance under (1.2).
From now through Section 3, assume that is connected. A constant rescaling of preserves (1.2), since and scale with degrees and . Normalize its length to . Positively oriented arclength is then , and .
Let denote the real smooth functions with zero boundary mean. The restriction of to this space has a smooth inverse: the Neumann problem is solvable precisely for mean-zero normal data, and its additive constant is fixed by the boundary mean. Define
Lemma 2.2 (Finite period defect). If has genus and one boundary component, then has rank at most on .
Proof. If is the harmonic extension of , then is closed. Consider the linear period map
This cohomology space has dimension . If , there is a smooth real function with . It is harmonic, and we choose its additive constant so that its boundary trace has mean zero. The boundary Cauchy-Riemann identities are
Consequently and
Thus vanishes on and factors through a vector space of dimension at most . This proves the rank bound. No existence of a harmonic conjugate is asserted outside .
The exact topological rank formula for a related boundary Hilbert transform is stronger than Lemma 2.2; see [BK, Section 2, formulas (12)-(16)] and its attribution to [B, Lemma 1]. Only the elementary upper bound just proved is needed here.
3. Connected boundary forces genus zero
Assume (1.2), and keep the normalization . For , let
These are the positive eigenspaces of . Exact commutation implies that preserves each . Self-adjointness then implies that its orthogonal Fourier projections commute with on smooth functions. The operators , , and also preserve .
Lemma 3.1 (High Fourier planes). There exists such that for every there is a unique , , for which
Proof. In the real orthonormal cosine-sine basis of , write for the positive symmetric matrix of and . The matrix of is . For any positive symmetric two-by-two matrix , direct multiplication gives
It follows that
If the scalar in (3.4) were nonzero for infinitely many , the range of would contain infinitely many independent Fourier planes. This contradicts Lemma 2.2. Hence
for some .
Set . In the complex orthonormal basis proportional to , reality and self-adjointness give
By (3.5), . Its kernel is one-dimensional, because its lower diagonal entry is . Lemma 2.1 identifies this kernel with .
For any nonzero trace in this kernel, strict positivity of (2.1) and (2.2) imply
The trace is nonconstant, so its harmonic extension has strictly positive energy. Thus and . Normalization proves existence and uniqueness in (3.2).
Write for the orthogonal projection onto . Since both and commute with , Lemma 2.1 gives
Proposition 3.2 (A degree-one boundary trace). There is a holomorphic function on whose boundary trace has the form
Proof. First suppose that for some . The product belongs to . Its frequency-one projection is
This is a nonzero holomorphic trace by (3.8). Applying (3.7) at frequency one shows that . Dividing (3.10) by proves (3.9).
Suppose instead that for all . Fix such an . Let be the holomorphic extensions of . The holomorphic function has zero boundary trace, so the maximum principle gives
The quotient is meromorphic in the interior, is smooth near the boundary since is nonzero there, and satisfies wherever it is initially defined. A pole of would give a pole of , which is impossible. Its apparent singularities are therefore removable. Thus is holomorphic on , smooth to the boundary, and has boundary trace . Take and .
Proposition 3.3 (Connected-boundary case of Theorem 1.1). If is connected and (1.2) holds, then has genus zero.
Proof. Take from Proposition 3.2. The real-linear map has determinant . Hence (3.9) maps the oriented boundary diffeomorphically onto a positively oriented nondegenerate ellipse . Let be the bounded component of .
For , the argument principle on the bordered Riemann surface gives
There are finitely many zeros in (3.12), since there are none near the boundary. For completeness, the first equality follows by deleting small disjoint coordinate disks around those zeros and applying Stokes' theorem to the closed form on the remaining surface. Each small circle contributes its positive zero multiplicity.
No interior point can map to : by the open mapping theorem its image neighborhood would contain points outside , contradicting the second case of (3.12). Therefore maps the interior bijectively onto , and every local multiplicity is one. It is a biholomorphism there. Its boundary map is also bijective, so is a homeomorphism by compactness. In particular is a topological disk and has genus zero.
4. Conformal capping preserves commutation
We now allow several boundary components. The following construction concerns Dirichlet energy and keeps the line element on every uncapped component exactly fixed.
Lemma 4.1 (Smooth conformal cap). Suppose has at least two boundary components, and designate one of them by . Let be a Euclidean disk of perimeter . Identify with by an orientation-reversing arclength isometry. The resulting topological surface admits a smooth conformal structure and a smooth compatible metric with these properties:
The inclusions of and are conformal on their interiors and smooth with nondegenerate derivatives up to the seam.
The line element of on equals the original line element of .
If has restrictions and to the two pieces, then
The surface is connected, has one fewer boundary component, and has the same genus as .
Proof. Smooth conformal sewing along an orientation-reversing boundary diffeomorphism is standard; a general bordered-surface statement is [T, Theorem 2.33]. The precise local smoothness can also be obtained from disk welding [HT, Theorem 3.1] and disk-annulus welding [HT, Proposition 3.6]. Here is the reduction to that local form.
Choose a smooth annular collar of in . Its conformal uniformization by a round annulus is smooth with nonzero derivative at both boundary circles. In these coordinates the seam identification is a smooth circle diffeomorphism. Welding the disk to the collar produces a disk with conformal inclusions smooth up to a smooth seam. The conformal structure on the open interior of the original collar is unchanged. It therefore glues to the remainder of along an open collar overlap. This supplies the asserted smooth conformal structure on .
Choose any smooth metric compatible with this structure. Its restriction to is conformal to , and its restriction to is conformal to the Euclidean metric. On each component of , prescribe a positive smooth scalar factor so that a conformal rescaling of has the original line element. Extend this factor smoothly and positively to , using boundary collars, and denote the rescaled metric by .
In dimension two, . Applying this on both pieces proves (4.1). Matching traces across the smooth seam are precisely the compatibility condition for two piecewise functions to define an function on . This follows in a smooth seam chart from the usual trace and gluing property for a hypersurface; a jump is the only additional distributional derivative. No matching of normal metric derivatives across the seam is required. Finally, attaching a disk increases Euler characteristic by one and decreases the boundary count by one, so the genus is unchanged.
The arclength condition determines the disk operator transported to the seam:
Indeed, on a disk of radius , the modes have Dirichlet-to-Neumann eigenvalues and boundary Laplace eigenvalues . Reversing orientation leaves both operators unchanged.
Relative to with the original line elements, write the original operator in blocks as
Proposition 4.2 (Capping and commutation). If commutes with , then the Dirichlet-to-Neumann map of in Lemma 4.1 commutes with the unchanged boundary Laplacian .
Proof. Let and denote the minimized Dirichlet-energy forms for prescribed boundary values. Formula (4.1) and the trace gluing property give, for ,
The brackets in (4.4) denote Sobolev duality, agreeing with the energy pairing on smooth real functions; complex forms follow by polarization. Minimization over the interior of each piece first produces its harmonic extension, which justifies both directions of (4.4).
The seam operator
is coercive. In fact, if has boundary trace on , then the Poincare inequality with zero trace on the nonempty boundary portion , followed by the trace theorem, yields
Adding the nonnegative disk energy preserves coercivity. The bounded form of is therefore invertible by the variational existence theorem. This includes constants on : although annihilates constants, the extension with zero trace on supplies the positive term in (4.6).
The unique minimizing trace in (4.4) satisfies
Polarization, or differentiation of the minimum energy with respect to , gives the exact Schur complement
The boundary measure used here is the original measure on , which equals the one for by Lemma 4.1. Thus (4.8) is the geometric Dirichlet-to-Neumann map, with no additional boundary multiplier.
It remains to check commutation without an unjustified manipulation of unbounded operators. The original commutation relation gives the block identities
on smooth functions. The diagonal block is self-adjoint and preserves every eigenspace, so it commutes with its orthogonal spectral projections. The same is true of and hence of the form operator . These projection identities extend to the indicated Sobolev spaces by density. Its inverse also commutes with the projections: apply a projection to and use uniqueness of the solution.
Now let belong to the eigenspace of . By (4.9), belongs to the eigenspace of , or is zero if that eigenvalue is absent there. Equation (4.7) therefore puts in the same eigenspace. Equations (4.8)-(4.9) imply that is in the eigenspace of . This also holds for by the invertibility already proved. Hence preserves every eigenspace. Smooth Fourier expansion on the finite union of boundary circles, together with continuity of both operators on smooth functions, proves their commutation.
Proof of Theorem 1.1. If the boundary is connected, use Proposition 3.3. Otherwise cap one boundary component using Lemma 4.1 and Proposition 4.2. The resulting surface has the same genus, one fewer boundary component, and commuting boundary operators. Repeat until the boundary is connected. Proposition 3.3 makes the final genus zero, hence the original genus was zero.
5. Scope and further questions
The proof uses exact commutation twice: first to reduce the finite-period defect to scalar defects on Fourier planes, and then to preserve the holomorphic-trace algebra under projection. A small commutator does not imply either exact determinant identities or exact multiplicative Fourier relations. A quantitative genus obstruction would require additional geometric control and is not a consequence of this argument. The stability results in [S2] concern other settings and do not supply that conclusion for arbitrary bordered surfaces.
Proposition 4.2 may be useful separately when studying boundary symmetries compatible with conformal sewing. Its hypotheses specify the metric on the uncapped boundary and the arclength identification at the seam; omitting either can change the operators in (4.8). The present theorem concerns orientable surfaces with smooth metrics and smooth boundary. We make no classification claim for nonorientable surfaces or for rough boundaries.
Research and preparation statement. This manuscript was developed and written with AI assistance, including its proof development, source comparison, and internal audit. It has not received independent expert review. The accompanying source and review records distinguish prior results, the proposed contribution, and the limits of the priority search. No numerical computation is used as evidence for the theorem.
References
[B] M. I. Belishev, The Calderon problem for two-dimensional manifolds by the BC-method, SIAM Journal on Mathematical Analysis 35 (2003), no. 1, 172-182. DOI: 10.1137/S0036141002413919.
[BK] M. I. Belishev and D. V. Korikov, On stability of determination of Riemann surface from its DN-map, arXiv:2112.14816v2, 2022. Section 2, especially formulas (12)-(16).
[HT] A. G. Henriques and J. E. Tener, The Segal-Neretin semigroup of annuli, arXiv:2410.05929v1, 2024. Theorem 3.1 and Proposition 3.6.
[S] R. Speciel, Surfaces with commuting boundary Laplacian and Dirichlet-to-Neumann map, Journal of Spectral Theory 16 (2026), 1281-1292. DOI: 10.4171/JST/617. Final journal numbering is used throughout.
[S2] R. Speciel, Stability for commutativity properties of the Dirichlet-to-Neumann map, International Mathematics Research Notices 2026 (2026), no. 14, rnag155. DOI: 10.1093/imrn/rnag155. Preprint titled Stability Estimates for Commutativity Properties of the Dirichlet-to-Neumann Map, arXiv:2510.08822v1.
[T] J. E. Tener, Construction of the unitary free fermion Segal CFT, Communications in Mathematical Physics 355 (2017), 463-518. DOI: 10.1007/s00220-017-2959-x. The sewing statement used here is Theorem 2.33 in the inspected author preprint, arXiv:1608.02095v1, 2016.