Refined regularity at critical points for linear elliptic equations

Hongjie Dong; Jongkeun Choi; Seick Kim

REVIEW 2 major objections 6 minor 7 references

Refined regularity at critical points for linear elliptic equations

T0 review · 2 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read For divergence-form linear elliptic equations with coefficients of Dini mean oscillation, a critical point — where the gradient vanishes — is automatically a point of second-order differentiability, with sharp modulus estimates; the same…

desk verdict Genuinely new refinement of critical-point regularity for linear elliptic equations, with a transparent but load-bearing dependence on the authors' earlier weak-type lemmas. read the letter →

arxiv 2506.08281 v1 pith:HT5HWJ5W submitted 2025-06-09 math.AP

Jongkeun Choi , Hongjie Dong , Seick Kim This is my paper · ORCID

classification math.AP MSC 35B4535B6535J08

keywords linearellipticequationsDinimeanoscillationcriticalpointsgradientdifferentiabilityHessianexistencenon-divergenceweaktype(11)estimatesfreezingcoefficients

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that for linear elliptic equations in divergence form, a critical point — a point where the gradient vanishes — is necessarily a point where the solution is twice differentiable, provided the coefficients and forcing term have Dini mean oscillation. The second derivative exists there, and the gradient is well approximated by the Hessian times the displacement, with an error controlled by moduli of continuity of the coefficients and data. If the principal coefficients are C^$\alpha$, this yields $C^{{2,alpha}}$ behavior at critical points, improving a prior result that only gave nearly Lipschitz regularity of the solution itself. An analogous statement for non-divergence equations says that where the Hessian vanishes, the Hessian is differentiable and the third derivative exists. The point is that degeneracy at a single point upgrades the expected regularity.

What carries the argument

The carrying object is the scale-invariant functional φ(r) = (1/r) inf_{l∈A} ( ∕_{B_r} |Du−l|^{1/2} )^2, where A is the space of affine functions; it measures how far the gradient is from being affine on a ball. The proof splits u = v + w, choosing w to absorb the oscillating-coefficient part of the equation through a constant-coefficient problem and leaving v to solve a constant-coefficient equation, then compares φ at two scales. A weak-type (1,1) estimate for constant-coefficient elliptic equations (Lemma 2.1) controls the fluctuation w by the mean oscillation of A, while interior regularity of v gives the decay φ(κr) ≤ κ^β φ(r) + Cω_coef(r) $r^{{-1}}$‖Du‖_{L^∞(B_r)} plus error terms. Iterating this inequality and freezing coefficients at small balls yields convergence of the sequence S_j that must equal $D^{2}$u(0); the same scheme, one derivative higher, drives the non-divergence theorem.

What would settle it

Check whether Lemma 2.1's weak-type estimate holds for all L^p data with p>1 by testing constant-coefficient operators with data concentrating near the boundary; a failure would undercut the reduction. At the theorem level, a divergence-form example with A in DMO, b in L^p, c and f in DMO, and a solution with Du(x^o)=0 but Du not differentiable at x^o would disprove Theorem 1.1.

Watch

Extended reading notes

Core claim

The central discovery is that vanishing of Du at one point upgrades the generic $C^{1}$ regularity of weak solutions to differentiability of Du at that point, with a quantitative rate. Theorem 1.1 states that if u is a weak solution of div(ADu)+b·Du+cu=f with A in DMO, b in L^p for some p>d, c and f in DMO, and Du(x^o)=0, then Du is differentiable at x^o; the Hessian obeys |$D^{2}$u(x^o)| ≤ C( $r^{{-1}}$ ∫_{B_{2r}(x^o)} |Du| + ϱ_lot(r) + ϱ_f(r) ), and for nearby x, |Du(x)−$D^{2}$u(x^o)(x−x^o)|/|x−x^o| ≤ Cϱ_coef(|x−x^o|)( $r^{{-1}}$∫_{B_{2r}}|Du| + ϱ_lot(r) + ϱ_f(r) ) + Cϱ_lot(|x−x^o|) + Cϱ_f(|x−x^o|). When the coefficients are C^$\alpha$, the relevant moduli are O(t^$\alpha$), so the gradient satisfies |Du(x)−$D^{2}$u(x^o)(x−x^o)| ≲ |x−x^o|^{1+$\alpha$}. Theorem 1.4 is the non-divergence analogue: vanishing Hessian at a point gives differentiability of the Hessian and existence of the third derivative, hence $C^{{3,alpha}}$ behavior at such points when A is C^$\alpha$ and b,c are $C^{{1,alpha}}$.

Load-bearing premise

The argument stands on a weak-type (1,1) bound for gradients and Hessians of solutions to constant-coefficient elliptic equations (Lemmas 2.1 and 4.1), which the paper cites to earlier work rather than proves; if that bound is false under the stated L^p hypotheses, the differentiability conclusion does not follow from this proof.

Editorial extensions

If this is right

If A is C^alpha and b is in L^p with p ≥ d/(1−alpha), then at any critical point u enjoys |Du(x)−D^2u(x^o)(x−x^o)| ≲ |x−x^o|^{1+alpha}; the earlier linear result only gave |u(x)−u(x^o)| ≲ |x−x^o|^{1+γ} for every γ<1.
When c is constant the lower-order term ϱ_lot vanishes identically, so the Hessian bound becomes purely coercive: |D^2u(x^o)| ≤ (C/r) ∫_{B_{2r}} |Du| plus the f contribution.
For the non-divergence equation with A in C^alpha and b,c in C^{1,alpha}, every point with D^2u=0 satisfies D^2u(x) = ⟨D^3u(x^o), x−x^o⟩ + O(|x−x^o|^{1+alpha}).
The statements are valid for inhomogeneous equations with f in DMO, so the critical-point regularity is stable when a forcing term with Dini mean oscillation is present.
Because DMO is weaker than Dini continuity, the theorem applies to coefficients whose mean oscillation is integrable over scales even when they are not Dini continuous.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If the weak-type lemma holds under the stated L^p hypotheses, the same iteration should extend to elliptic systems and to equations with VMO-type oscillation, since the proof only uses the pointwise vanishing, local averages, and the weak-type bound.
The method suggests a hierarchy of degeneracies in the linear setting: zeros of u, Du, D^2u, and so on should upgrade to C^{1,alpha}, C^{2,alpha}, C^{3,alpha}, and higher points, respectively; this paper proves the first two rungs of that ladder.
A concrete probe of sharpness would be to take coefficients with mean oscillation ω_A(r) barely failing the Dini condition, for instance ω_A(r) ~ 1/log(1/r), and test whether a solution with a vanishing gradient at a point still has a differentiable gradient there.
The proof's local nature suggests the Hessian bound could feed into nodal-set or free-boundary arguments where solutions are assembled from information at their critical points, giving quantitative control of D^2u in terms of an averaged gradient.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

First thing to know: this paper proves a genuinely stronger statement than anything I've seen before for linear elliptic equations. If Du(x0)=0, then Du is differentiable at x0, with an explicit modulus of continuity that is Dini if the coefficients are DMO and C^alpha if they are C^alpha. That sharpens Teixeira's C^{1,1-} bound. There's also a non-divergence analogue: if the Hessian vanishes, you get differentiability of D^2u. These are new results, not repackaged old ones.

The proof is a Campanato-type iteration. The structure is sensible: first a simple case without lower-order terms, then the general case. The same pattern repeats for the non-divergence theorem. The estimates are quantitative and the dependence on the coefficients' mean oscillation is carefully tracked. I appreciate that they define composite moduli like omega_coef and carry them through rather than hand-waving.

Soft spots: the main one is the reliance on Lemma 2.1 and Lemma 4.1, the weak-type L^1 estimates for constant-coefficient equations. These are proved in the authors' earlier papers, not here. The whole iteration hangs on them: the correction term w is controlled through these lemmas, and if they had hidden hypotheses, both theorems would fall. The stress-test note is right to flag this. But the lemmas are published, the statements in the present text look right, and this kind of dependency is normal in analysis papers. So I'd call it a transparency issue, not a correctness issue. It would help if the paper included a proof sketch or at least an appendix restating the external lemmas with their exact hypotheses. Same for the C^1/C^2 regularity citations.

The self-citation pattern is noticeable but appropriate: [2] and [3] are exactly where the regularity and weak-type results come from. I don't see gratuitous self-citation. There are a few typos ("and and" in the intro, "omega_dat" in (5.19), a cross-reference to (2.21) that should be (4.21)), but they're cosmetic.

Who this is for: people working on elliptic regularity, critical points, free boundaries, and DMO/borderline coefficients. The simple-case sections make it accessible. I'd take it to a reading group. It deserves a serious referee and likely acceptance after minor revisions.

Referee Report

2 major / 6 minor

Summary. The paper proves pointwise higher regularity for linear elliptic equations at points where the gradient (in divergence form) or the Hessian (in non-divergence form) vanishes. Theorem 1.1 shows that if u is a C^1 weak solution to Lu = div(ADu)+b·Du+cu = f with A in DMO, b in L^p for p>d, c in DMO, f in DMO, and Du(x0)=0, then Du is differentiable at x0, with quantitative estimates in terms of moduli of continuity. Theorem 1.4 establishes the analogous statement for D^2u in the non-divergence case under the assumptions A, b, Db, c, Dc, f, Df in DMO. The proofs follow a Campanato-type iteration: at each scale, u is decomposed into a constant-coefficient solution v and a correction w, and w is controlled via weak-type (1,1) estimates for constant-coefficient elliptic equations (Lemmas 2.1 and 4.1). The paper is organized as a simpler case followed by the general case for each theorem.

Significance. If the results are correct, they refine a theorem of Teixeira (Math. Ann. 358 (2014)) in the linear setting, upgrading the C^{1,1-} regularity at critical points to genuine C^{2,α} when the coefficients are C^α, and providing the analogous C^{3,α} statement in non-divergence form. The proof is well structured and the final estimates are explicit, including sharp moduli of continuity. A notable strength is the clean reduction of the problem to a one-scale Campanato iteration; the main caveat is that the proof relies on external weak-type lemmas, so the self-containedness is limited. The results are significant and likely correct if the cited lemmas hold as stated.

major comments (2)

[Section 2, Eq. (2.2) and Section 4, Eq. (4.2)] The L^{1/2} estimates for Dw (resp. D^2w) are asserted to follow from the weak-type estimate in Lemma 2.1 (resp. Lemma 4.1), but the derivation is not shown. A weak-type (1,1) bound does not by itself imply a strong L^{1/2} norm bound with the stated scaling; an additional interpolation or L^q-embedding argument, or a precise citation to the relevant result in [2] (and [3] for the non-divergence case), is needed. This point is load-bearing, since (2.2) and (4.2) feed directly into the iteration inequalities (2.8) and (4.5).
[Lemmas 2.1 and 4.1] The two central weak-type lemmas are stated without proof and are attributed to [1, Lemma 3.3] and [2, Lemmas 2.2/2.20]. Since [1] is an unpublished arXiv preprint, the paper should either quote the exact result from the published source [2] or include a short proof in an appendix. This is the most fragile point in the argument; if the lemmas were to have hidden hypotheses, the entire iteration would collapse. A precise reference to the published version would remove this concern.

minor comments (6)

[Proof of Lemma 2.35] In the sentence 'By induction, it follows form (2.37) and (2.40)', 'form' should read 'from'.
[Section 4, Eq. (4.21)] The text says 'we obtain from (2.21)' but the reference should be to equation (4.15), not (2.21).
[Section 5, Eq. (5.19) and the estimate before (5.23)] There is a missing '+' sign between two integral terms in the displayed formulas; the current typesetting makes these expressions unreadable.
[Theorems 1.1 and 1.4] The modulus ϱ_lot is stated to depend on ||u||_{L∞(B_r(x0))} in the theorem, but in the proof (after (3.21)) it is defined using ||u||_{L∞(B_{r0})}. Please align the notation.
[Section 2, notation] The notation ∫_{B_r} for the normalized average integral is used throughout the proofs but is not explicitly defined after the introduction. Since the distinction from the ordinary Lebesgue integral is crucial, a short remark at the beginning of Section 2 would help.
[Eq. (3.1) and (5.1)] The factor r^{1-d} in the definition of ω_coef may appear surprising; a brief comment that it is the correct scaling for normalized averages would aid the reader.

Circularity Check

0 steps flagged · score 1.0 of 10

No by-construction circularity; the paper's Campanato iteration is independent content, though several load-bearing lemmas are cited from the authors' own prior papers.

full rationale

The main theorems (1.1 and 1.4) are proved by a Campanato-type iteration on the quantities φ(r) and Φ(r), in which the differentiability of Du (resp. D^2u) at the critical point is obtained by showing the best affine approximants S_j converge to a limit S and that ||Du−Sx||_{L∞(B_r)}/r→0. The only substantially imported ingredients are the weak-type estimates in Lemma 2.1 and Lemma 4.1, whose proofs are delegated to the authors' earlier papers [1, Lemma 3.3] and [2, Lemmas 2.2 and 2.20], together with the C^1/C^2 regularity results from [2] and [3]. These are external published lemmas with stated hypotheses (constant-coefficient, or DMO-coefficient equations) that do not include the target conclusion; they are tools, not restatements of the theorem. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported to forbid alternatives, and no ansatz is smuggled in via citation. The self-citations are frequent and some are load-bearing, but the central claim—existence of the second (resp. third) derivative at the critical point with the explicit modulus estimates—is established by the paper's own iteration, not reduced to those citations by construction. Accordingly, no circular step can be exhibited with a specific equation reducing an output to an input.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

No data fitting occurs; the constants β, κ, r0 are proof devices, not tunable physical parameters. The argument relies on standard elliptic estimates, Poincaré's inequality, and several prior regularity theorems, mostly by the same authors, which are cited but not reproved. No new entities are introduced.

assumptions (6)

domain assumption Weak-type (1,1) gradient estimate for constant-coefficient elliptic equations (Lemma 2.1)
Used in Section 2 to control the oscillation of Dw in (2.2); proof cited to [1, Lemma 3.3] and [2, Lemma 2.2].
domain assumption C^1_loc regularity for divergence-form equations with DMO coefficients
Invoked in Section 3 to justify u∈C^1_loc and in Lemma 2.35 bounds; cited to [3, Theorem 1.3].
domain assumption C^2_loc regularity for non-divergence equations with DMO coefficients
Invoked in Section 5; cited to [3, Theorem 1.5].
standard math Interior regularity estimates for constant-coefficient elliptic equations
Used in (2.3) and (4.3) to bound higher derivatives of the constant-coefficient part v.
domain assumption DMO condition implies uniform continuity and a modulus of continuity
Used throughout to define rho-coefficients; cited to the appendix of [4].
standard math Poincaré inequality on balls
Used in Sections 3 and 5 to bound ∫|cu - c̄ū| and ∫|f - f̄ - Df·x|.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Refined regularity at critical points for linear elliptic equations." pith.science (2026). https://pith.science/paper/HT5HWJ5W

@misc{pith2026250608281,
  author       = {Pith},
  title        = {Pith review of: Refined regularity at critical points for linear elliptic equations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HT5HWJ5W}},
  note         = {Machine review of arXiv:2506.08281}
}

abstract

We investigate the regularity of solutions to linear elliptic equations in both divergence and non-divergence forms, particularly when the principal coefficients have Dini mean oscillation. We show that if a solution $u$ to a divergence-form equation satisfies $Du(x^o)=0$ at a point, then the second derivative $D^2u(x^o)$ exists and satisfies sharp continuity estimates. As a consequence, we obtain ``$C^{2,\alpha}$ regularity'' at critical points when the coefficients of $L$ are $C^\alpha$. This result refines a theorem of Teixeira (Math. Ann. 358 (2014), no. 1--2, 241--256) in the linear setting, where both linear and nonlinear equations were considered. We also establish an analogous result for equations in non-divergence form.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

7 extracted references · 7 canonical work pages

[2]

Dong, Hongjie; Kim, Seick.On C 1, C2, and weak type-(1,1)estimates for linear elliptic operators. Comm. Partial Differential Equations42(2017), no. 3, 417–435

work page 2017
[3]

Ann.370(2018), no

Dong, Hongjie; Escauriaza, Luis; Kim, Seick.On C 1, C2, and weak type-(1,1)estimates for linear elliptic operators: part II.Math. Ann.370(2018), no. 1-2, 447–489

work page 2018
[1]

Regularity of elliptic equations in double divergence form and applications to Green's function estimates

Choi, Jongkeun; Dong, Hongjie; Kim, Dong-ha; Kim, Seick.Regularity of elliptic equations in double divergence form and applications to Green’s function estimates. arXiv:2401.06621 [math.AP]

work page Pith review arXiv
[4]

Potential Anal.52(2022), no

Hwang, Sukjung; Kim, Seick.Green’s function for second order elliptic equations in non-divergence form. Potential Anal.52(2022), no. 1, 27–39

work page 2022
[5]

Kim, Seick; Lee, Sungjin.Estimates for Greens functions of elliptic equations in non-divergence form with continuous coefficients. Ann. Appl. Math.37(2021), no. 2, 111–130

work page 2021
[6]

Leit ˜ao, Raimundo; Pimentel, Edgard A.; Santos, Makson S.Geometric regularity for elliptic equations in double-divergence form. Anal. PDE13(2020), no. 4, 1129–1144

work page 2020
[7]

Teixeira, Eduardo V .Regularity for quasilinear equations on degenerate singular sets. Math. Ann.358 (2014), no. 1-2, 241–256. (J. Choi) Department ofMathematicsEducation, PusanNationalUniversity, Busan, 46241, Republic ofKorea Email address:jongkeun choi@pusan.ac.kr (H. Dong) Division ofAppliedMathematics, BrownUniversity, 182 GeorgeStreet, Providence, R...

work page 2014

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [2]

Dong, Hongjie; Kim, Seick.On C 1, C2, and weak type-(1,1)estimates for linear elliptic operators. Comm. Partial Differential Equations42(2017), no. 3, 417–435

work page 2017

[2] [3]

Ann.370(2018), no

Dong, Hongjie; Escauriaza, Luis; Kim, Seick.On C 1, C2, and weak type-(1,1)estimates for linear elliptic operators: part II.Math. Ann.370(2018), no. 1-2, 447–489

work page 2018

[3] [1]

Regularity of elliptic equations in double divergence form and applications to Green's function estimates

Choi, Jongkeun; Dong, Hongjie; Kim, Dong-ha; Kim, Seick.Regularity of elliptic equations in double divergence form and applications to Green’s function estimates. arXiv:2401.06621 [math.AP]

work page Pith review arXiv

[4] [4]

Potential Anal.52(2022), no

Hwang, Sukjung; Kim, Seick.Green’s function for second order elliptic equations in non-divergence form. Potential Anal.52(2022), no. 1, 27–39

work page 2022

[5] [5]

Kim, Seick; Lee, Sungjin.Estimates for Greens functions of elliptic equations in non-divergence form with continuous coefficients. Ann. Appl. Math.37(2021), no. 2, 111–130

work page 2021

[6] [6]

Leit ˜ao, Raimundo; Pimentel, Edgard A.; Santos, Makson S.Geometric regularity for elliptic equations in double-divergence form. Anal. PDE13(2020), no. 4, 1129–1144

work page 2020

[7] [7]

Teixeira, Eduardo V .Regularity for quasilinear equations on degenerate singular sets. Math. Ann.358 (2014), no. 1-2, 241–256. (J. Choi) Department ofMathematicsEducation, PusanNationalUniversity, Busan, 46241, Republic ofKorea Email address:jongkeun choi@pusan.ac.kr (H. Dong) Division ofAppliedMathematics, BrownUniversity, 182 GeorgeStreet, Providence, R...

work page 2014