Fourier extension from positively curved surfaces

← All digests

Family 077 · Real and complex analysis
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-14

1. The question

Suppose we superimpose plane waves whose frequencies lie on a sphere. Each wave extends throughout space, but their phases usually cancel. How much cancellation must occur when the frequency surface is curved?

Fourier restriction theory asks questions of this kind. The geometry of a surface in frequency space controls how strongly the resulting wave can concentrate in physical space. Curvature spreads directions, yet many localized pieces of the wave can still overlap along thin tubes.

Family 077 contains two claimed sharp open-range results in three-dimensional space. One concerns bounded data on the sphere. The other treats data in an $L^p$ space on any compact smooth positively curved surface, including smooth boundary patches. The second paper uses both the first paper’s packet theorem and the Kakeya maximal theorem from family 074.

2. Background and essential definitions

Extension as a superposition of waves

Let $\Sigma\subset\mathbb R^3$ be a smooth surface with area measure $d\sigma$. Its extension operator is

$$E_\Sigma f(x)=\int_\Sigma f(\xi)e^{2\pi i x\cdot\xi}\,d\sigma(\xi).$$

The point $\xi$ is a frequency vector, and $e^{2\pi i x\cdot\xi}$ is its plane wave. The function $f$ specifies the amplitude and phase attached to each frequency. The output is defined on all of physical space, even though the input lives on a two-dimensional surface.

Why is this called restriction theory? A general Fourier transform of an $L^p$ function need not have well-defined values on a surface of zero volume. Restriction estimates make such a trace meaningful in appropriate norms. By duality, these estimates can be expressed as extension bounds. The manuscripts use the extension formulation.

An $L^p$ norm measures $(\int|f|^p)^{1/p}$. The $L^\infty$ norm is the essential supremum. An estimate with bounded input and an estimate with $L^p$ input are different assertions: the latter can admit arbitrarily large amplitudes on small subsets of the surface.

Positive curvature and the number three

Positive curvature here means that the second fundamental form is definite at every point. Locally, after choosing the normal appropriately, the surface is the graph of a function whose Hessian is positive definite. Sphere patches and elliptic paraboloid patches are standard examples. A saddle surface has a different geometry.

For the unit sphere and constant input, direct integration gives

$$E_{S^2}1(x)=\frac{2\sin(2\pi|x|)}{|x|}\qquad(x\ne0).$$

Its amplitude typically decays like $|x|^{-1}$. A shell of radius $r$ has volume proportional to $r^2\,dr$, so the integral of its $p$-th power has the large-radius behavior of $\int r^{2-p}\,dr$. This converges only when $p>3$. Thus the threshold three already appears in the simplest smooth example.

3. Development history and previous partial results

The restriction problem originates in Stein’s work. The Tomas–Stein theorem gives the classical extension estimate from $L^2$ on a positively curved surface to $L^4$ in $\mathbb R^3$. Improving the output exponent toward three requires more detailed control of how oscillatory pieces interact.

Bilinear methods study pairs of separated frequency pieces. Multilinear methods study several pieces whose normal directions span space. The Bennett–Carbery–Tao theorem supplied a central estimate for these transverse interactions. Bourgain and Guth developed methods to turn such information back into estimates for one full extension operator.

Guth’s polynomial partitioning argument reached bounded-data estimates for $p>13/4$. It divides physical space into cells separated by an algebraic surface. Most tubes pass through relatively few cells, while the difficult remaining ones linger near that surface.

Wang’s broom method improved bounded-data restriction for the truncated paraboloid to $p>42/13$. The method exploits the fact that tubes clustered near one region must spread elsewhere, restricting repeated concentration. Wang and Wu subsequently combined refined decoupling with two-ends incidence estimates to obtain diagonal bounds for $p>22/7$ on compact positively curved surfaces.

The input norms in this history matter. A diagonal bound uses the same $p$ on the surface and in space, while a bounded-data theorem uses $L^\infty$ on the surface. The new family claims the full open output range $p>3$, with the second manuscript supplying the general-surface diagonal form.

4. What the new manuscripts claim

Bounded sphere data

The first manuscript states that every $p>3$ has a finite constant $C_p$ such that

$$\|E_{S^2}g\|_{L^p(\mathbb R^3)} \le C_p\|g\|_{L^\infty(S^2)}$$

for every bounded measurable complex-valued $g$. The norm is global in physical space. The constant depends on $p$, not on the input.

The constant-input example above rules out extending this strong estimate to $p\le3$. The stated result is therefore the sharp open range in this formulation.

General positively curved surfaces

The companion states that every compact smooth positively curved $\Sigma\subset\mathbb R^3$, possibly with smooth boundary, satisfies

$$\|E_\Sigma f\|_{L^p(\mathbb R^3)} \le C_{\Sigma,p}\|f\|_{L^p(\Sigma)} \qquad(3<p<\infty).$$

The surface is fixed, and the constant may depend on its geometry. This is not a uniform assertion over all surfaces with arbitrarily degenerating curvature.

For a compact paraboloid patch, a known transfer theorem yields a free Schrödinger local-smoothing consequence in two spatial dimensions:

$$\|e^{it\Delta}f\|_{L^p(\mathbb R^2\times[0,1])} \le C_{p,s}\|(I-\Delta)^{s/2}f\|_{L^p(\mathbb R^2)}, \qquad s>2-\frac6p,\quad p>3.$$

Here $\Delta$ is the Laplacian and $e^{it\Delta}$ is the free Schrödinger evolution. The operator $(I-\Delta)^{s/2}$ measures $s$ derivatives in the Bessel-potential sense. The time interval is bounded but space is not. The Sobolev inequality is strict.

5. Why the advance matters

Importance: 5/5. If correct, these estimates reach a central conjectured boundary in three-dimensional Fourier restriction. The second manuscript also moves from one highly symmetric surface to general smooth positively curved surfaces and permits rough $L^p$ data.

Restriction estimates are useful because many oscillatory problems become local versions of the same geometry. A dispersion relation for an evolution equation traces out a surface in frequency-time space; extension from that surface describes a superposition of propagating waves. The stated Schrödinger consequence is one concrete instance.

The family also connects geometric overlap estimates with genuinely complex oscillation. A tube bound counts where packets can be large, while an extension estimate must handle their phases and interference. Passing from one to the other without losing the conjectured exponent is the core significance of the claimed method.

6. The main idea

Resolve a wave into packets

Divide the frequency surface into small caps. Localizing the input within one cap produces a wave packet concentrated near a long thin tube in physical space. Its direction is determined by the surface normal, while another label records its spatial position.

The packets can carry arbitrary complex coefficients. Consequently, a method that works only for equally weighted positive tubes cannot immediately estimate their sum. The first paper formulates a packet theorem that retains this coefficient information through repeated changes of scale.

Its geometric control uses ellipses with semiaxes $L\ge w$. The weight

$$W(E)=L^{1+\kappa}w^{1-\kappa} =Lw\left(\frac Lw\right)^\kappa$$

is ordinary area multiplied by a small eccentricity penalty. The parameter $\kappa>0$ is small. Long narrow concentration is therefore recorded differently from concentration in a round region of the same area.

The paper tracks the joint distribution of direction and position labels. It does not replace that joint distribution by an independent one. Elliptic tests, planar incidence estimates and entropy calculations are used to control how concentration can propagate along packet trajectories.

Combine oscillation with positive tube estimates

The companion applies the packet theorem to general surface charts. A second input, the three-dimensional Kakeya maximal estimate, controls the positive weighted tube density. These inputs serve different purposes: one retains complex coefficients and oscillatory synthesis, while the other bounds geometric concentration.

Grouping packets according to their concentration allows the two estimates to be combined. The paper then obtains a joint bound on sparse collections of spatial balls. This controls the extension on far-separated regions without a constant that worsens with their separation.

An epsilon-removal argument turns local estimates carrying a small power of the spatial scale into a global estimate for every strictly larger exponent. Finally, finitely many charts cover the compact surface. Their Jacobians and the finite-measure inclusion between surface $L^p$ spaces produce the stated diagonal inequality.

The smooth-boundary case is included by extending the local graph smoothly and extending its input by zero. This does not impose an artificial vanishing condition on the original boundary data.

7. Scope, limitations, and open questions

The estimates are for positively curved surfaces in $\mathbb R^3$. Surfaces with zero curvature or mixed curvature are not included. Nor does the family resolve restriction in every ambient dimension.

The lower endpoint $p=3$ is excluded. Strong global boundedness already fails there for constant sphere data. Separate endpoint formulations, such as suitable weak estimates, are different questions.

The companion records how constants depend on its theorem inputs but does not claim a simple polynomial bound as $p$ decreases to three. The finiteness of the constant at each fixed $p>3$ should not be confused with uniform boundedness near the endpoint.

The Schrödinger consequence uses two spatial dimensions and $s>2-6/p$. It is distinct from the three-spatial-dimensional wave local-smoothing result in family 079. The diagonal surface theorem also retains its dependency on the Kakeya maximal manuscript rather than following from bounded sphere data alone for arbitrary surfaces.

8. Sources and further reading

Primary background includes Bennett, Carbery and Tao’s multilinear theorem, Guth’s polynomial partitioning paper, and Bourgain and Demeter’s decoupling theorem. The source manuscripts give the comparisons with Wang’s and Wang–Wu’s later estimates and state the exact packet and Kakeya inputs.

For optional terminology, see restriction of the Fourier transform. The definitions above suffice to read the claimed inequalities without that reference.

Source manuscripts

• M1. OpenAI, Elliptic capacity propagation and Fourier restriction to the sphere (September 24, 2026).
• M2. OpenAI, Diagonal Fourier extension for positively curved surfaces in three dimensions (September 24, 2026).

Sources consulted on 14 October 2026 at repository revision 1581d4c. Download the manuscript citations.