Time averaging smooths waves in every dimension

← All digests

Family 079 · Real and complex analysis
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-27

1. The question

A wave can concentrate sharply when many rays meet. At that instant its spatial profile may be much larger than the initial profile in a norm sensitive to peaks. But a focus is temporary. How much control can we recover by averaging the wave’s size over time?

This is the local-smoothing problem. “Local” refers to a bounded time interval; the spatial region can be all of Euclidean space. The desired gain measures how much less regularity of the initial data is needed for a spacetime estimate than for an estimate at one fixed time.

Family 079 contains a three-dimensional argument and a broader manuscript claiming the conjectured gain in every spatial dimension at least two. The latter also states consequences for reconstructing functions from their Fourier transforms and for the size of sets containing line segments in every direction. The connection is that a localized wave travels along a thin tube, so controlling waves also constrains how such tubes overlap.

2. Background and essential definitions

Waves, norms, and derivative costs

For $x\in\mathbb R^n$, use the Fourier transform with phase $e^{-ix\cdot\xi}$. A half-wave is

$$U(t)f(x)=\frac1{(2\pi)^n} \int_{\mathbb R^n}e^{i(x\cdot\xi-t|\xi|)}\widehat f(\xi)\,d\xi.$$

Here $f$ is the initial profile and $\xi$ its frequency variable. This evolution solves the wave equation $\partial_t^2u-\Delta u=0$, with $\Delta$ the spatial Laplacian. The other sign in the phase gives the second half-wave needed for general initial displacement and velocity.

For finite $p\ge1$, the $L^p$ norm is $(\int |f|^p)^{1/p}$; $L^\infty$ measures the essential supremum. Larger exponents emphasize large values. The spacetime norm integrates over both $x$ and $t$; throughout the main wave estimate the time interval is $I=[1,2]$.

Write $J^s=(1-\Delta)^{s/2}$. In Fourier space this multiplies by $(1+|\xi|^2)^{s/2}$, so $\|J^sf\|_p$ measures $s$ derivatives, including fractional values. At frequencies of size $N$, this costs approximately $N^s$. The estimates start with Schwartz functions—smooth functions whose derivatives decay faster than every power—and extend to the indicated normed spaces by approximation.

The critical exponent

For $p>2$, controlling a half-wave at a fixed nonzero time costs

$$s_{\mathrm{fixed}}(p)=(n-1)\left(\frac12-\frac1p\right)$$

derivatives. The proposed spacetime threshold is

$$s_+(p)=\max\left\{0,\frac{n-1}{2}-\frac np\right\}, \qquad p_c=\frac{2n}{n-1}.$$

Above $p_c$, the difference between the two thresholds is $1/p$. At $p=p_c$, the spacetime threshold reaches zero. The values are $p_c=4$ in two dimensions and $p_c=3$ in three.

An arbitrarily small positive loss means a bound for every $\varepsilon>0$, with a constant allowed to depend on $\varepsilon$. It does not mean a bound with $\varepsilon=0$: even $N^{0.001}$ eventually grows.

3. Development history and previous partial results

Sogge’s 1991 paper formulated local smoothing through the study of wave propagation and maximal averages over circles. It established positive smoothing gains in the plane and connected the question to Fourier summation. Mockenhaupt, Seeger and Sogge developed methods that split waves into angular pieces while retaining information about their overlap.

Wolff’s wave-packet and induction-on-scale arguments obtained sharp gains for large exponents. Bourgain and Demeter’s decoupling theorem subsequently gave the conjectured gain, up to arbitrarily small loss, for

$$p\ge\frac{2(n+1)}{n-1}.$$

This is an important range, but its lower endpoint exceeds $p_c$. In three dimensions it begins at four, leaving the critical exponent three untouched.

Guth, Wang and Zhang’s 2020 cone square-function theorem reached the critical exponent four for waves in two spatial dimensions. The cone lives in three-dimensional spacetime, which explains the dimension appearing in their title. In three spatial dimensions, Gao, Liu, Miao and Xi obtained a Sobolev loss above $1/9$ at $p=3$. Gan, He, Li and Wu’s January 2026 version gives full gain for $p\ge10/3$; interpolation supplies losses above $1/12$ at $p=3$.

The September manuscript in this family claims every positive loss at that critical three-dimensional exponent. The October manuscript gives a different construction in all dimensions, together with a quantitative logarithmic estimate at one frequency scale.

The geometric consequences also have their own history. In particular, Wang and Zahl’s 2025 work already established full Hausdorff dimension for three-dimensional Kakeya sets. The all-dimensional claims here should be understood alongside that earlier result, not as the first resolution of its three-dimensional case. The new manuscript additionally states maximal-operator estimates, which contain more quantitative information than a dimension conclusion alone.

4. What the new manuscripts claim

Local smoothing and a logarithmic bound

Theorem 1.1 of the October manuscript states that for $n\ge2$, $2\le p<\infty$, and every $\varepsilon>0$,

$$\|U(t)f\|_{L^p(\mathbb R^n\times I)} \le C_{n,p,\varepsilon}\|J^{s_+(p)+\varepsilon}f\|_{L^p(\mathbb R^n)}.$$

The constant is independent of $f$. At $p=2$, zero loss is permitted by the usual energy identity. For $2<p\le p_c$, every positive Sobolev order suffices; above $p_c$, the order must exceed the displayed threshold.

Theorem 1.2 is more quantitative. For data $f_N$ supported in a fixed frequency annulus of size $N\ge2$, it gives

$$\|U(t)f_N\|_{L^{p_c}(\mathbb R^n\times I)} \le C_n[\log(2+N)]^{2/p_c}\|f_N\|_{p_c}.$$

Every fixed power of a logarithm grows more slowly than $N^\varepsilon$. This estimate therefore supplies the small-loss theorem after interpolation and frequency summation. In dimension three it gives logarithmic power $2/3$, while recovering the September paper’s critical $L^3$ claim as a consequence.

Fourier reconstruction and restriction

Bochner–Riesz means reconstruct a function by multiplying its Fourier transform by

$$\left(1-\frac{|\xi|^2}{R^2}\right)_+^\alpha,$$

where $R>0$, $a_+=\max(a,0)$, and $\alpha$ softens the frequency cutoff. Theorem 10.1 claims uniform $L^p$ bounds for the individual means, for $1\le p\le\infty$, whenever

$$\alpha>\max\left\{0,n\left|\frac1p-\frac12\right|-\frac12\right\}.$$

For $2\le p<\infty$, it also bounds the pointwise supremum over all radii at the same strict threshold, and obtains almost-everywhere convergence to the original function. Taking a supremum inside the norm is stronger than bounding every radius separately. Stein’s square function instead measures changes with radius by a squared integral; Theorem 11.1 claims its bound for $\alpha>1/2+s_+(p)$ in this same range of $p$.

Fourier extension assembles waves from frequencies lying on a surface. For the unit sphere, it is

$$Eg(x)=\int_{S^{n-1}}e^{ix\cdot\omega}g(\omega)\,d\sigma(\omega),$$

where $g$ is the amplitude and $\sigma$ surface measure. Corollary 9.3 claims $L^a$-to-$L^b$ bounds throughout the conjectured interior:

$$b>\frac{2n}{n-1},\qquad b>\frac{n+1}{n-1}a',\qquad \frac1a+\frac1{a'}=1,$$

with $1<a\le\infty$ and finite $b$. Companion bounds concern the truncated cone and compact paraboloid. The latter yields local smoothing for Schrödinger waves, whose phase uses $|\xi|^2$ instead of $|\xi|$. In $d$ spatial dimensions, Corollary 12.1 states the exact spacetime Sobolev order $d-2(d+1)/p$ for $p>2(d+1)/d$ on compact time intervals.

Tubes and sets containing every direction

A Kakeya set contains a unit line segment in every direction. Its Hausdorff dimension describes its size through covers at arbitrarily small scales; full dimension does not require positive volume.

Let $K_\delta f(\omega)$ be the largest average of $|f|$ over any unit-length tube of radius $\delta$ pointing in direction $\omega$. Corollary 9.1 states

$$\|K_\delta f\|_{L^n(S^{n-1})} \le C_n[\log(2/\delta)]^{4/p_c}\|f\|_{L^n(\mathbb R^n)}, \qquad 0<\delta\le1.$$

This implies the usual conjectured bound allowing every $\delta^{-\varepsilon}$ loss, and full Hausdorff dimension $n$ for every Euclidean Kakeya set. A projective transformation gives the corresponding Nikodym estimate, in which tubes are selected through observation points rather than indexed by direction.

The September companion also states a wave maximal bound in three dimensions: for $3\le p<\infty$, controlling $\sup_{0<t<1}|U(t)f|$ requires Sobolev order $s>1-2/p$. This pointwise-in-time conclusion has a stronger regularity requirement than spacetime smoothing.

5. Why the advance matters

Importance: 5/5. If correct, the all-dimensional theorem resolves a central conjecture about how oscillatory waves concentrate. Its consequences connect several major questions in harmonic analysis: derivative gains for evolution equations, stability of Fourier reconstruction, restriction to curved frequency surfaces, and overlap of thin tubes.

The logarithmic estimate provides more information than an unspecified small power loss. It quantifies the residual growth at the critical exponent and carries that growth into restriction and maximal estimates. This makes it possible to compare the proposed bound with explicit focusing obstructions.

The geometric implication is especially striking. A statement about averaging an oscillating function over time leads to a constraint on every set containing all line directions, without assuming that the set arose from a particular wave. The method’s significance would lie both in the conjectures resolved and in the proposed conservation-law construction connecting them.

6. The main idea

Preserve interference while selecting directions

At one frequency scale, the wave is a superposition of plane waves moving in different directions. Their interference creates the peaks that the estimate must control. Squaring the pieces before adding them would discard this information.

The October argument instead continues a finite superposition to carefully chosen complex spacetime points. An imaginary displacement along a light ray multiplies each frequency by a Gaussian weight in its angular separation from that ray. It thus selects nearby directions while retaining the sum and its interference.

Using a symmetric first-order formulation of the wave equation, the paper constructs a positive tensor in an enlarged space of physical and auxiliary variables. Its columns have zero divergence. This is a conservation law: changes inside a region are related to flux through its boundary.

An auxiliary equation turns conservation into an estimate

On slices of the enlarged space, the argument solves a nonlinear boundary-value problem for a potential. Its prescribed determinant is tied to a nonnegative density $\mu$ that detects the squared wave amplitude.

The determinant is formed from a quaternionic matrix of second derivatives. Quaternions extend complex numbers to four real components; the relevant positive Hermitian matrices have real eigenvalues whose product defines their determinant. The role of this machinery is to build an energy sensitive to concentration while permitting integration by parts with the conserved current.

A background potential contributes curvature in one direction. The remaining determinant estimate makes the resulting energy detect the power $\mu^{n/(n-1)}$. Since $\mu$ corresponds to a squared wave amplitude, this produces $|U(t)f|^{2n/(n-1)}$—exactly the critical exponent $p_c$.

The paper corrects this energy so that difficult interior determinant terms cancel after differentiation. The surviving boundary flux is controlled by Gaussian decay. The result is a scalar differential inequality for the energy in physical time and two auxiliary time variables.

Why two logarithms appear

The final comparison moves from a small circle in the two auxiliary variables to a circle of fixed radius, transfers the estimate toward initial time, and returns to a small circle. Radial harmonic functions in two dimensions depend on the logarithm of the radius. The two annular comparisons each contribute a logarithmic factor.

The resulting bound for the $p_c$-th power of the norm has a squared logarithm. Taking its $p_c$-th root gives the power $2/p_c$. Removing an intermediate spatial period and summing frequency scales yields the Euclidean Sobolev estimate.

The September companion follows a separate path: it tracks wave-packet masses across changing scales and sheared coordinate frames. Lower-dimensional wave and incidence estimates constrain an entropy quantity describing packet configurations. Its contradiction argument rules out excessive concentration in three dimensions. The two papers therefore offer different mechanisms for the overlapping three-dimensional conclusion.

7. Scope, limitations, and open questions

The main wave theorem concerns the free Euclidean equation. Variable coefficients and curved manifolds can focus rays differently; the Euclidean critical exponent is not a universal threshold for every wave equation.

At the critical exponent, the manuscript gives an annular lower obstruction of logarithmic power $1/p_c$, compared with its upper power $2/p_c$. Thus a frequency-independent lossless critical bound is excluded by its focusing examples, while the optimal logarithmic power for unrestricted data remains between those values. Additional angular regularity permits sharper statements in Appendix B.

The restriction conclusions use strict inequalities, and the displayed Bochner–Riesz and square-function orders are also strict. They should not be read as every possible endpoint theorem. Likewise, bounded-time smoothing is different from a global-in-time estimate or regularity improvement at each individual instant.

Full Kakeya dimension permits zero volume. The maximal estimate is quantitative, but it does not say that all collections of thin tubes have a volume bounded below independently of their thickness.

The repository’s formalization description reports local smoothing and specified Bochner–Riesz, spherical extension, square-function and Kakeya results. Its scope statement should be consulted for those formalization claims; the exposition here follows the two manuscripts’ mathematical statements.

8. Sources and further reading

In the October manuscript, Theorems 1.1–1.2 state local smoothing and the annular logarithmic bound. Sections 2–7 develop the auxiliary equation and energy argument; Section 8 gives endpoint obstructions; Sections 9–12 derive the applications. The September paper supplies its independent three-dimensional packet argument and maximal consequences.

The primary historical sources linked above trace Sogge’s formulation, Bourgain–Demeter decoupling, the Guth–Wang–Zhang planar theorem, the Gan–He–Li–Wu three-dimensional range, and Wang–Zahl’s Kakeya dimension theorem. Optional background: Sobolev spaces, Fourier restriction, and Kakeya sets.

Source manuscripts

• M1. OpenAI, Local smoothing for the Euclidean wave equation (October 25, 2026).
• M2. OpenAI, Critical local smoothing for the three-dimensional wave equation (September 24, 2026; also retained in the current source revision).

Sources consulted on 27 October 2026 at repository revision 7ef5c0c. Download the manuscript citations.