Carleson's Problem: Schrödinger Convergence Criteria

• Carleson’s problem is defined as determining the minimal Sobolev regularity needed for Schrödinger evolutions to converge pointwise almost everywhere to the initial data.
• The analysis employs maximal function estimates and summability techniques to link almost-everywhere convergence with strong smoothing and interference behaviors.
• Extensions of the problem cover one-dimensional sharp thresholds, higher-dimensional necessary conditions, and various settings including periodic, non-Euclidean, and perturbed operators.

Carleson’s problem, in the sense used in Euclidean harmonic analysis and dispersive PDE, is the problem of determining the Sobolev regularity of the initial datum $f$ for which the solution of the Schrödinger equation converges pointwise almost everywhere back to $f$ as $t\to 0$. In its standard formulation, one asks for which exponents $s>0$ every $u_0\in H^s(\mathbb R^n)$ satisfies

$$\lim_{t\to 0} e^{it\Delta}u_0(x)=u_0(x) \quad \text{for a.e. }x\in \mathbb R^n.$$

The problem is tightly linked to maximal estimates for the Schrödinger propagator and has developed into a family of related questions involving fractional and generalized dispersive phases, restricted approach regions, periodic and non-Euclidean geometries, and radial or curve-based convergence. The same name also appears in adjacent parts of harmonic analysis and PDE, but the Schrödinger pointwise-convergence problem remains the central modern usage (Lucà et al., 2015, Choi, 2019, Shiraki, 2019).

1. Formulation and maximal-function framework

For the free Schrödinger equation on $\mathbb R^n$,

$i\partial_t u+\Delta u=0,\qquad u(\cdot,0)=u_0,$

the solution can be written as

$$u(x,t)=e^{it\Delta}u_0(x)=\int_{\mathbb R^n} e^{i(x\cdot \xi+t|\xi|^2)}\widehat{u_0}(\xi)\,d\xi.$$

Carleson’s problem asks for the smallest $s$ such that almost-everywhere convergence to the initial datum holds for all $f$0 (Lucà et al., 2015).

The analytic mechanism is the Schrödinger maximal operator

$f$1

A local maximal estimate of the form

$f$2

is sufficient to imply almost-everywhere convergence. Conversely, the a.e. convergence statement implies a weak $f$3 maximal estimate via the Nikishin–Stein maximal principle, and interpolation then yields a strong estimate of the same type. This equivalence makes maximal-function bounds the standard formulation of the problem (Lucà et al., 2015).

A complementary viewpoint treats the question as a summability problem. For perturbed or nonlinear flows, the operators $f$4 are regarded as approximations to the identity as $f$5, analogous to Abel summation for Fourier series, but real-time Schrödinger evolution lies on the boundary of the holomorphic semigroup region, so maximal criteria and smoothing estimates replace semigroup arguments (Choi, 2019).

2. One-dimensional theory and persistence under perturbation

In one spatial dimension, the classical theorem of Carleson gives convergence for $f$6, and Dahlberg–Kenig showed that this threshold is sharp: if $f$7, there exists data in $f$8 for which almost-everywhere convergence fails (Choi, 2019). In the restricted-direction formulation discussed later, the same threshold appears as the special case $f$9, corresponding to vertical approach to the boundary of space-time (Shiraki, 2019).

A later extension shows that the one-dimensional threshold persists for certain perturbed Schrödinger operators

$t\to 0$0

If

$t\to 0$1

then for every $t\to 0$2,

$t\to 0$3

Moreover, in the $t\to 0$4-potential case the threshold remains sharp: if $t\to 0$5, there exists compactly supported $t\to 0$6 for which the perturbed evolution fails to converge on a set of positive measure outside $t\to 0$7 (Choi, 2019).

The same analysis also shows that below the threshold the localized maximal operator remains too large. For every $t\to 0$8 and every $t\to 0$9, a localized strong-type estimate fails, which is consistent with the Dahlberg–Kenig obstruction. The same paper extends the smoothing-based argument to quadratic nonlinear Schrödinger equations with nonlinearities $s>0$0, $s>0$1, and $s>0$2, obtaining a.e. convergence at essentially the same regularity levels, with the mixed term $s>0$3 requiring $s>0$4 (Choi, 2019).

3. Higher-dimensional Euclidean theory: necessary conditions and divergence

In higher dimensions, the sharp threshold has historically been harder to determine from below. One counterexample due to Bourgain shows that if

$s>0$5

then there exists $s>0$6 whose Schrödinger evolution diverges on a set of positive Lebesgue measure. A later note gives a different construction that strengthens this to positive fractional Hausdorff measure: for

$s>0$7

and

$s>0$8

there exists $s>0$9 such that

$u_0\in H^s(\mathbb R^n)$0

for all $u_0\in H^s(\mathbb R^n)$1 in a set of positive $u_0\in H^s(\mathbb R^n)$2-dimensional Hausdorff measure (Lucà et al., 2017).

An earlier necessary-condition theorem of Lucà and Rogers sharpened Bourgain’s lower bound for the local maximal estimate in dimensions $u_0\in H^s(\mathbb R^n)$3. If

$u_0\in H^s(\mathbb R^n)$4

holds for every Schwartz function $u_0\in H^s(\mathbb R^n)$5, then necessarily

$u_0\in H^s(\mathbb R^n)$6

Equivalently, if Schrödinger solutions converge a.e. to the initial datum for all $u_0\in H^s(\mathbb R^n)$7, $u_0\in H^s(\mathbb R^n)$8, then $u_0\in H^s(\mathbb R^n)$9 is necessary. Their construction uses data whose evolution interferes with itself periodically in time, then an ergodic lemma to choose a direction along which constructive interference appears densely throughout space-time (Lucà et al., 2015).

These results describe obstructions rather than a full solution. A later comparison result states that for the classical Schrödinger equation, the Euclidean sharp threshold in dimension $$\lim_{t\to 0} e^{it\Delta}u_0(x)=u_0(x) \quad \text{for a.e. }x\in \mathbb R^n.$$0 is $$\lim_{t\to 0} e^{it\Delta}u_0(x)=u_0(x) \quad \text{for a.e. }x\in \mathbb R^n.$$1, up to endpoint issues (Dewan, 1 Jun 2025). This suggests that the higher-dimensional theory is organized by a persistent gap between sufficient estimates, necessary estimates, and endpoint phenomena.

4. Generalized phases, restricted directions, and periodic analogues

A major generalization replaces the vertical approach $$\lim_{t\to 0} e^{it\Delta}u_0(x)=u_0(x) \quad \text{for a.e. }x\in \mathbb R^n.$$2 at fixed $$\lim_{t\to 0} e^{it\Delta}u_0(x)=u_0(x) \quad \text{for a.e. }x\in \mathbb R^n.$$3 by approach along a restricted family of lines. In one dimension, for the generalized dispersive evolution

$$\lim_{t\to 0} e^{it\Delta}u_0(x)=u_0(x) \quad \text{for a.e. }x\in \mathbb R^n.$$4

with $$\lim_{t\to 0} e^{it\Delta}u_0(x)=u_0(x) \quad \text{for a.e. }x\in \mathbb R^n.$$5 satisfying curvature-type assumptions and including the fractional Schrödinger phase $$\lim_{t\to 0} e^{it\Delta}u_0(x)=u_0(x) \quad \text{for a.e. }x\in \mathbb R^n.$$6, $$\lim_{t\to 0} e^{it\Delta}u_0(x)=u_0(x) \quad \text{for a.e. }x\in \mathbb R^n.$$7, one fixes a compact direction set $$\lim_{t\to 0} e^{it\Delta}u_0(x)=u_0(x) \quad \text{for a.e. }x\in \mathbb R^n.$$8 and studies convergence inside

$$\lim_{t\to 0} e^{it\Delta}u_0(x)=u_0(x) \quad \text{for a.e. }x\in \mathbb R^n.$$9

The relevant geometric parameter is the upper Minkowski dimension $\mathbb R^n$0. The maximal operator

$\mathbb R^n$1

satisfies

$\mathbb R^n$2

for every $\mathbb R^n$3 whenever

$\mathbb R^n$4

Consequently,

$\mathbb R^n$5

The case $\mathbb R^n$6 recovers the classical one-dimensional Carleson threshold $\mathbb R^n$7 (Shiraki, 2019).

The proof of this restricted-direction theorem is also methodologically notable. It extends Cho–Lee–Vargas from the classical phase $\mathbb R^n$8 to a broader dispersive class and from $\mathbb R^n$9 to all $i\partial_t u+\Delta u=0,\qquad u(\cdot,0)=u_0,$0, while avoiding the time localization lemma used earlier. Instead it uses dyadic frequency decomposition, coverings of $i\partial_t u+\Delta u=0,\qquad u(\cdot,0)=u_0,$1 by intervals of length $i\partial_t u+\Delta u=0,\qquad u(\cdot,0)=u_0,$2, a $i\partial_t u+\Delta u=0,\qquad u(\cdot,0)=u_0,$3 argument, van der Corput estimates, and a Hardy–Littlewood–Sobolev-type inequality (Shiraki, 2019).

On the periodic side, the analogue of Carleson’s problem for

$i\partial_t u+\Delta u=0,\qquad u(\cdot,0)=u_0,$4

asks for the critical exponent

$i\partial_t u+\Delta u=0,\qquad u(\cdot,0)=u_0,$5

For non-singular polynomial symbols $i\partial_t u+\Delta u=0,\qquad u(\cdot,0)=u_0,$6 of degree at least $i\partial_t u+\Delta u=0,\qquad u(\cdot,0)=u_0,$7, one has

$i\partial_t u+\Delta u=0,\qquad u(\cdot,0)=u_0,$8

This covers, in particular, $i\partial_t u+\Delta u=0,\qquad u(\cdot,0)=u_0,$9, corresponding to $$u(x,t)=e^{it\Delta}u_0(x)=\int_{\mathbb R^n} e^{i(x\cdot \xi+t|\xi|^2)}\widehat{u_0}(\xi)\,d\xi.$$0, and $$u(x,t)=e^{it\Delta}u_0(x)=\int_{\mathbb R^n} e^{i(x\cdot \xi+t|\xi|^2)}\widehat{u_0}(\xi)\,d\xi.$$1. The argument uses rational space-time points, multidimensional Weyl sums, and Deligne’s theorem on Weil sums to produce uniform counterexamples (Eceizabarrena et al., 2024).

5. Radial and non-Euclidean versions

Recent work has transported Carleson-type convergence from Euclidean space to negatively curved and solvable Lie-group geometries, typically under radiality assumptions. On Damek–Ricci spaces, for the Schrödinger equation with radial initial data, a local maximal estimate of the form

$$u(x,t)=e^{it\Delta}u_0(x)=\int_{\mathbb R^n} e^{i(x\cdot \xi+t|\xi|^2)}\widehat{u_0}(\xi)\,d\xi.$$2

implies

$$u(x,t)=e^{it\Delta}u_0(x)=\int_{\mathbb R^n} e^{i(x\cdot \xi+t|\xi|^2)}\widehat{u_0}(\xi)\,d\xi.$$3

whenever $$u(x,t)=e^{it\Delta}u_0(x)=\int_{\mathbb R^n} e^{i(x\cdot \xi+t|\xi|^2)}\widehat{u_0}(\xi)\,d\xi.$$4 is radial and $$u(x,t)=e^{it\Delta}u_0(x)=\int_{\mathbb R^n} e^{i(x\cdot \xi+t|\xi|^2)}\widehat{u_0}(\xi)\,d\xi.$$5. A counterexample on $$u(x,t)=e^{it\Delta}u_0(x)=\int_{\mathbb R^n} e^{i(x\cdot \xi+t|\xi|^2)}\widehat{u_0}(\xi)\,d\xi.$$6 shows failure below $$u(x,t)=e^{it\Delta}u_0(x)=\int_{\mathbb R^n} e^{i(x\cdot \xi+t|\xi|^2)}\widehat{u_0}(\xi)\,d\xi.$$7, so the threshold is sharp in that model (Dewan, 2024).

This radial Damek–Ricci theory was extended to the fractional Schrödinger, Boussinesq, and Beam equations, for both the Laplace–Beltrami operator $$u(x,t)=e^{it\Delta}u_0(x)=\int_{\mathbb R^n} e^{i(x\cdot \xi+t|\xi|^2)}\widehat{u_0}(\xi)\,d\xi.$$8 and the shifted Laplace–Beltrami operator $$u(x,t)=e^{it\Delta}u_0(x)=\int_{\mathbb R^n} e^{i(x\cdot \xi+t|\xi|^2)}\widehat{u_0}(\xi)\,d\xi.$$9. For the corresponding local maximal functions on balls $s$0, the paper gives a complete characterization: if $s$1, the estimate fails for every $s$2; if $s$3, it holds if and only if $s$4; if $s$5, it holds if and only if $s$6; and if $s$7, it holds for all $s$8. Consequently, pointwise convergence holds for radial $s$9 whenever $f$00 (Dewan, 14 Jan 2025).

A further dispersive generalization concerns asymptotically concave phases on Damek–Ricci spaces. For radial data and phases satisfying

$f$01

the maximal estimate

$f$02

holds for every $f$03 provided

$f$04

and fails if $f$05. This yields almost-everywhere convergence for radial $f$06 with $f$07. The author explicitly notes that the endpoint $f$08 remains open, even in the Euclidean case (Dewan, 1 Jun 2025).

The approach-region question also has a non-Euclidean counterpart. On Damek–Ricci spaces, pointwise convergence was proved not only along vertical lines but along curve families $f$09 satisfying Hölder control in time and bilipschitz control in the radial parameter: $f$10 and

$f$11

Under these hypotheses, almost-everywhere convergence still holds for radial $f$12 with $f$13. The same work emphasizes a negative phenomenon: by a counterexample on $f$14, Schrödinger solutions, unlike harmonic functions or solutions of the heat equation, do not admit any natural wide approach region (Dewan, 2024).

On $f$15, the threshold is different. For dispersive equations

$f$16

whose phase satisfies

$f$17

one has the endpoint maximal estimate

$f$18

and therefore

$f$19

whenever $f$20, $f$21. This covers the Schrödinger, fractional Schrödinger with convex phase, Boussinesq, and Beam equations (Dewan, 17 Aug 2025).

Setting	Regularity threshold	Source
$f$22, classical Schrödinger	$f$23 sharp	(Choi, 2019)
Restricted directions in 1D	$f$24	(Shiraki, 2019)
Damek–Ricci, radial Schrödinger	$f$25	(Dewan, 2024)
Damek–Ricci, radial fractional/Boussinesq/Beam	$f$26	(Dewan, 14 Jan 2025)
Damek–Ricci, asymptotically concave phase	$f$27, failure if $f$28	(Dewan, 1 Jun 2025)
$f$29, broad dispersive class	$f$30	(Dewan, 17 Aug 2025)
$f$31, non-singular polynomial symbol	$f$32	(Eceizabarrena et al., 2024)

6. Related but distinct uses of the name

The expression “Carleson problem” also appears in several adjacent areas, but these are technically distinct from the Schrödinger convergence problem.

One branch concerns maximally modulated singular integrals. The one-dimensional Polynomial Carleson operator

$f$33

was shown to satisfy

$f$34

The proof develops higher-order wave-packet analysis and a new tile discretization that eliminates exceptional sets, yielding a direct strong $f$35 bound and the full $f$36 range (Lie, 2011). An even broader abstraction places Carleson operators on doubling metric measure spaces, with axiomatic modulation functions and conditional restricted weak-type $f$37 bounds derived from an assumed $f$38 estimate for a stronger non-tangential truncation (Becker et al., 7 Aug 2025).

A second branch involves Carleson measure conditions in elliptic and parabolic PDE. For elliptic operators $f$39 on Lipschitz domains with small Lipschitz constant, small Carleson norm of the coefficient oscillation yields solvability of the regularity problem with $f$40 boundary data and of the Neumann problem with $f$41 data for all $f$42 (Dindoš et al., 2013). For time-varying parabolic domains, small Carleson norm of the oscillation of the matrix $f$43 and the drift $f$44 yields $f$45 solvability of the Dirichlet problem for all $f$46 (Dindoš et al., 2014). In a different direction, Carleson measure estimates and $f$47-approximation for bounded harmonic functions were characterized without Ahlfors regularity by the existence of a subdomain with uniformly rectifiable boundary containing $f$48 (Garnett, 2020), and Green functions in the half-space were shown to be “almost affine” under weak DKP coefficient oscillation (David et al., 2021).

A third usage concerns an infinite-order differential equation studied by Carleson. For an entire function

$f$49

the equation

$f$50

is analyzed under the ABC conditions on the multiplicity sequence $f$51. Under those hypotheses, the solution space is characterized by Taylor–Dirichlet series with frequencies $f$52, and the same framework yields a biorthogonal family and a solution of an associated moment problem (Zikkos, 2022).

These parallel usages share Carleson’s name because they descend from distinct problems introduced by Lennart Carleson or from techniques built around Carleson measures and Carleson operators. In current analysis, however, the unqualified phrase “Carleson’s problem” most often refers to the almost-everywhere pointwise convergence problem for Schrödinger and related dispersive evolutions.