Nodal size: a sharp surface bound and higher-dimensional failures

← All digests

Family 350 · Differential geometry
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-26

1. The question

A vibrating membrane has stationary vibration modes. Along the nodal set of a mode, the displacement is zero: these are the lines where sand can gather in a Chladni experiment. On a higher-dimensional manifold, the analogous zero set is usually a hypersurface.

How large can that zero set be as the frequency increases? Yau’s nodal conjecture predicts that its measure is comparable to frequency, with constants depending on the fixed geometry.

Family 350 claims a sharply divided answer in the smooth category. On every smooth closed surface, the predicted upper bound holds. Starting in dimension three, there are fixed smooth metrics with sequences of exact eigenfunctions violating it. In dimensions at least five, the claimed violation even exceeds frequency by a fixed positive power of the eigenvalue.

The phrase fixed metric is central. Choosing a different geometry for every complicated nodal pattern would not contradict Yau’s formulation. Each negative result here claims one metric supporting an entire violating sequence.

2. Background and essential definitions

Eigenvalue and frequency conventions

Let $(M^d,g)$ be a closed connected Riemannian manifold; closed means compact without boundary. Its Laplace–Beltrami operator describes spatial diffusion and vibration. The papers use

$$-\Delta_gu=\lambda u,\qquad \lambda>0,$$

for a nonzero real eigenfunction $u$. Set $k=\sqrt\lambda$. This $k$ is the frequency scale, while $\lambda$ is its square. Confusing them would change every exponent in the conjecture.

The nodal set is

$$Z_u=\{x\in M:u(x)=0\}.$$

Its size is measured by $(d-1)$-dimensional Hausdorff measure $\mathcal H_g^{d-1}$: length on a surface, area in dimension three, and the corresponding hypersurface volume in higher dimensions. Multiplying $u$ by a nonzero constant does not change this set.

The conjectured two-sided estimate is

$$c_g\sqrt\lambda\leq\mathcal H_g^{d-1}(Z_u) \leq C_g\sqrt\lambda.$$

The constants may depend on the manifold and metric, but must work for every nonzero real eigenfunction in every positive eigenspace. The lower and upper bounds are logically separate.

Smooth versus analytic geometry

A smooth metric has derivatives of every order. A real-analytic metric is locally represented by convergent power series, a stronger requirement. Analytic continuation gives additional control unavailable for arbitrary smooth coefficients.

One should also distinguish nodal topology from nodal measure. A zero set can contain a complicated component without violating a total-length or total-area estimate. Conversely, many sign changes in small regions can force large measure without providing a classification of topology.

Growth at the wavelength scale

The natural local length scale is $k^{-1}$. A doubling or growth index compares the size of $u$ on a disk with its size on a larger concentric disk. Large growth indicates strong local oscillation or concentration, but the surface proof does not need every local growth index to be uniformly bounded. It seeks a bound on their average.

3. Development history and previous partial results

Donnelly–Fefferman proved the sharp two-sided estimate for real-analytic manifolds and metrics. This established the predicted frequency scale in the analytic setting, but left the smooth setting open. Donnelly–Fefferman, 1988.

For smooth surfaces, Brüning proved the square-root lower bound. Donnelly–Fefferman and Dong obtained upper bounds of order $\lambda^{3/4}$, and Logunov–Malinnikova improved this to $\lambda^{3/4-\beta}$ for a universal positive $\beta<1/4$. These exponents approach the desired $1/2$ but do not reach it. Surface upper bound, Logunov–Malinnikova.

In all dimensions, Logunov proved the sharp lower bound. He also established a polynomial upper bound in dimensions at least three, replacing earlier nonpolynomial estimates. A polynomial bound with a larger exponent leaves room for the counterexamples claimed here. Lower bound, polynomial upper bound.

Nazarov–Polterovich–Sodin promoted a statistical approach to local growth, and Roy-Fortin made precise its relation to total nodal length on surfaces. M1 supplies the uniform average-growth estimate needed to reach the sharp exponent through that framework. Roy-Fortin, average growth refinement.

Metric flexibility has also been used to prescribe nodal sets. Enciso and Peralta-Salas realized specified separating hypersurfaces by choosing a metric, and related work constructed complicated high-energy nodal components even on fixed analytic spheres and tori. These results concern prescribed topology, not unbounded nodal measure divided by frequency on one smooth metric. The new negative claims require that stronger quantitative feature.

Intermediate regularity assumptions also matter. Hezari obtained stronger estimates for Gevrey metrics and a specified quasianalytic class. Those classes are more restrictive than arbitrary smooth geometry, so the claimed smooth counterexamples do not erase such results.

4. What the new manuscripts claim

Every smooth surface: M1

For each fixed closed connected smooth Riemannian surface, M1 asserts a constant $C(M,g)$ such that

$$\mathcal H_g^1(Z_u)\leq C(M,g)\sqrt\lambda$$

for every nonzero real positive-eigenvalue eigenfunction. No orientation assumption is imposed. Together with the existing lower bound, this gives the two-sided conjecture on smooth surfaces.

Qualitative violations in dimensions three and four: M2

M2 claims that every neighborhood of the round metric on $S^3$ in the smooth topology contains a metric $g_\infty$ with exact real eigenfunctions $u_j$ satisfying

$$-\Delta_{g_\infty}u_j=\lambda_ju_j, \qquad\lambda_j\longrightarrow\infty,$$

and

$$\frac{\mathcal H_{g_\infty}^2(Z_{u_j})}{\sqrt{\lambda_j}} \longrightarrow\infty.$$

It separately claims a fixed smooth metric on $S^2\times T^2$ with the analogous divergence of nodal three-volume. Here $T^2$ is the two-dimensional torus. These are two dimension-specific constructions, not an assertion that one exactification method works unchanged in both dimensions.

Neither result states a fixed positive power-law excess in dimensions three or four. Unbounded ratios may grow more slowly than every chosen positive power.

A fixed power-law excess in dimension five: M3

M3 claims a fixed smooth metric on $S^4\times S^1$, a fixed number $\epsilon_0>0$, and exact real eigenfunctions with

$$\frac{\mathcal H_g^4(Z_{u_j})} {\lambda_j^{1/2+\epsilon_0}}longrightarrow\infty.$$

This also disproves the proposed upper bound with an arbitrarily small positive exponent loss, for that metric. Taking products with fixed closed manifolds extends the power-law conclusion to every dimension greater than five.

Together, the papers claim the following distinction:

Dimension	Universal smooth upper bound of order $\sqrt\lambda$	Negative example supplied
2	Holds for every fixed closed surface	None
3	Fails for some fixed metrics	$S^3$, arbitrarily close to round
4	Fails for some fixed metrics	$S^2\times T^2$
At least 5	Fails, even with a small fixed power loss	$S^4\times S^1$ and products

Failure for some metrics is not failure for every metric or every eigenfunction. Analytic metrics retain their established sharp bounds, and the lower bound remains intact.

5. Why the advance matters

Importance: 5/5. If correct, the family would give an unexpected dimension threshold for a central conjecture in spectral geometry. The same smooth regularity would permit the sharp universal upper estimate in dimension two and fail to guarantee it in every higher dimension.

The closeness-to-round assertion in dimension three is particularly notable. It says that the exceptional behavior can occur in any prescribed smooth neighborhood of a familiar analytic metric, while still belonging to a single smooth limit metric.

The power-law claim goes further than merely producing an unbounded normalized ratio. It rules out a commonly considered near-sharp relaxation with every positive exponent loss.

These distinctions would reshape the search for upper bounds: dimension, regularity class, and additional geometric assumptions would need to be tracked separately. The rating concerns that stated significance, not proof verification.

6. The main idea

On surfaces, sum average growth rather than worst growth

In local isothermal coordinates, a smooth surface metric is conformal to the Euclidean metric. The eigenfunction equation becomes

$$\Delta u+k^2p(x)u=0,$$

where $p$ is a fixed positive smooth function. Divide a coordinate region into squares of side $s$ comparable to $k^{-1}$.

A local theorem of Roy-Fortin bounds nodal length in such a square by a constant times $s$ multiplied by one plus a logarithmic growth index. There are on the order of $s^{-2}$ squares. If their average growth is bounded, summing gives total length on the order of $s^{-1}$, hence $k$.

M1 extracts logarithmic profiles from weighted local norms. A Carleman estimate, a weighted inequality for differential equations, makes these limiting profiles subharmonic. Its decisive trace control is specific to two dimensions.

Affine parts of a profile can persist under subdivision, so the argument subtracts them and tracks their slopes separately. The remaining growth improves on average. Excursions of high growth produce witness sets of positive area. These witnesses can lie outside their parent squares, so a separate geometric packing argument controls boundary-collar discrepancies rather than assuming each witness remains inside its original square. The resulting terminal mean-growth bound yields the sharp length estimate.

In higher dimensions, create many stable sign changes

M2 and M3 begin with explicit high-frequency background waves and modify their logarithmic amplitude envelopes in localized regions. Corrugation increases the integrated gradient of the envelope while retaining inequalities needed for localized complex-phase wave packets.

Finite-order phase and transport equations produce quasimodes: functions whose eigenfunction equation has a small error. A Gaussian combination of packets is then used to select a real wave with strong first-derivative control and many strict opposite-sign tests on separated small regions.

These sign tests provide robust lower bounds on nodal measure. If a function has opposite signs on the two ends of a short cylinder, its zero set must separate them. A collection of disjoint cylinders gives an additive lower bound. Strict signs persist under small perturbations, so the construction need not rely solely on delicate pointwise descriptions of the zero set.

Replace approximate waves by exact metric eigenfunctions

Small residual error is not enough: the theorem is about exact eigenfunctions of an ordinary Laplace–Beltrami operator.

For the three-sphere construction, M2 first adjusts a scalar conductivity and an independent density. Those coefficients describe a weighted elliptic equation but need not yet arise from one metric. Division by a positive solution of a semilinear equation imposes the three-dimensional metric identity in the active region, and a transverse tensor repair handles the remaining annulus.

On $S^2\times T^2$, a conformal scalar correction has a coercive linearized equation. A further change tangent to the eigenfunction’s level sets enforces the four-dimensional determinant relation without changing the corrected eigenfunction equation. The separate repairs preserve the chosen sign tests.

M3 instead works first with a positive principal tensor and density on $S^4$, varied independently. After obtaining a fixed limiting weighted operator, it adds a warped circle. Choosing the circle radius to account for the density makes the Laplacian on functions independent of the circle coordinate equal to the weighted base operator. This produces an ordinary five-dimensional metric.

Preserve infinitely many witnesses in one limit

Each construction must protect earlier eigenfunctions while introducing new high-frequency behavior. The papers perturb the operator or metric while preserving a selected eigenfunction and making its eigenvalue simple. Its strict sign score then persists throughout a neighborhood of that metric.

Nested neighborhoods and summably small changes retain every previous witness and converge to one smooth limit. No uniform spectral gap across the whole sequence is assumed; each finite stage has its own tolerance.

For M3, quantitative control is stronger. Repeated corrugations give a multiplicative gain, while derivatives grow at a controlled exponential rate in the number of generations. Taking the number of generations proportional to $\log k$ converts this into a fixed power gain in $k$. Constants and frequency thresholds may deteriorate with the desired accuracy, but the positive gain exponent is kept fixed. That distinction makes the final power-law violation survive the infinite construction.

7. Scope, limitations, and open questions

The negative results concern smooth metrics, not real-analytic ones. They use exact real eigenfunctions on a single closed manifold and metric for each sequence, not approximate modes or a different metric at every frequency.

M2 gives qualitative divergence in dimensions three and four. M3 gives a fixed positive power excess only from dimension five upward. A corresponding power-law construction in dimensions three or four is not claimed by this family.

The known lower bound is not challenged, and a larger polynomial upper bound remains compatible with the examples. Neither theorem prescribes the optimal universal polynomial exponent for all smooth metrics.

Nor do the claims describe a typical metric, a random eigenfunction, or every high-energy mode. They combine a universal positive statement on surfaces with existential negative statements elsewhere.

This digest reports the manuscripts’ proposed average-growth, exactification, and fixed-limit mechanisms without checking their estimates or independently verifying the conjecture resolutions.

8. Sources and further reading

The linked primary papers explain the analytic theorem, smooth lower and polynomial upper bounds, and the average-growth approach. Optional background appears under nodal set, Laplace–Beltrami operator, and Chladni figures.

Source manuscripts

• M1. OpenAI, Sharp nodal length on smooth surfaces (September 23, 2026).
• M2. OpenAI, Smooth counterexamples to Yau’s nodal upper bound in dimensions three and four (September 23, 2026).
• M3. OpenAI, Power-law violations of Yau’s nodal upper bound (September 23, 2026).

Sources consulted on 26 October 2026 at repository revision 1581d4c. Download the manuscript citations.