Pith's one-line read
This paper proves that one unbalanced energy inequality is enough to force local boundedness of parabolic De Giorgi functions, with quantitative supremum bounds, and that extra $L^s$ integrability is only required in the subcritical range.
desk verdict
Genuine progress on sharp parabolic local boundedness; proof holds together, but the p<N hypothesis is under-advertised and the abstract overclaims the subcritical case.
read the letter →
A machine-rendered reading of the paper's core claim, the
machinery that carries it, and where it could break.
The reading
This paper introduces a parabolic De Giorgi class $\mathcal{PDG}^+$ whose defining feature is a single unbalanced energy inequality, and proves that every member of the class is locally bounded under an explicit condition relating the growth exponents. The condition is $L:=\max(1,\Lambda)\le M$, where $\Lambda$ measures the largest elliptic growth and $M=p|\lambda/p|+(m+1)p/N$ is the threshold at which the parabolic and elliptic terms balance. When the inequality is strict, the proof yields a quantitative bound on the essential supremum; in the limiting case it still yields local boundedness. The interest is that the class contains local weak subsolutions to doubly nonlinear, double-phase/Orlicz-type, and fully anisotropic parabolic operators, so the result removes extra $L^s$ integrability assumptions that earlier work needed, even for the classical $p$-Laplacian in the subcritical cases.
What carries the argument
Three tools carry the argument. The two-sided bounds on the truncation $g(u^m,k^m)$ of Lemma 2.1 turn the energy inequality (1.9) into a clean recursion on level sets, with time term of exponent $1+1/m$ and spatial terms of exponents $p_i$ and $q_i$. The anisotropic Sobolev embedding of Lemma 2.3 then raises the integrability exponent on each superlevel set to $p_*=\alpha p+(1+1/m)p/N$, where $\alpha=(1/p)(1+(p/m)|\lambda/p|)$; this is the step that produces the critical threshold $M$. Finally, the geometric convergence lemma makes the level-set integrals $y_j$ tend to zero once the initial integral is small enough. The strict supercritical case $L<M$ is exactly $p_*>1+L/m$, which makes the recursion super-linear; the limiting case $L=M$ is handled by choosing the starting level through absolute continuity of the integral; and the subcritical case recovers super-linearity from an $L^s$ assumption with $\kappa_s>0$. The paper also defines the auxiliary quantities $H(\theta,\vec\rho)$ and $R(\theta,\vec\rho)$ that enter all quantitative bounds.
What would settle it
One decisive check is to test the borderline case $L=M$ for the classical doubly nonlinear equation with $m=1$: Theorem 1.1 says every finite-energy subsolution in the class is locally bounded, as long as the $p<N$ embedding applies. If an explicit unbounded function can be exhibited that satisfies (1.9) and has finite $\mathcal{V}_{\mathrm{loc}}$ energy, the theorem is false; if, as the paper asserts, the known borderline blow-up examples fail the energy-class membership, the threshold is exactly sharp.
The paper's central claim is that local boundedness is a property of the unbalanced energy estimate (1.9) itself, not of any particular equation or of extra integrability of the solution. More precisely, Theorem 1.1 states that if $u\in\mathcal{PDG}^+(\Omega_T)$ and $L=\max(1,\Lambda)\le M=p|\lambda/p|+(m+1)p/N$, then $u$ is locally bounded; in the strict case the sup is controlled by an $\int u^{m+L}$ average via (1.12). Theorem 1.2 covers the complementary subcritical range $L\ge M$: assuming $u\in L^s_{\mathrm{loc}}$ for some $s$ with $\kappa_s>0$, local boundedness follows with the quantitative estimate (1.15), where $\kappa_s$ is the explicit non-degeneracy number (1.13). The paper also verifies that local weak subsolutions of the doubly nonlinear equation (1.3), the generalized Orlicz/double-phase equation (1.4), and the fully anisotropic equation (1.5) satisfy (1.9), so all these operators fall under the same theorem. The previously known unbounded borderline examples are noted to fail membership in the relevant Sobolev spaces, so they do not contradict the result.
Load-bearing premise
The load-bearing premise is that the anisotropic Sobolev embedding used at the decisive step is valid, and it is stated in the paper only for $p<N$; the main theorems do not list this restriction explicitly, so the conclusion is conditional on that range.
Editorial extensions
If this is right
For the standard doubly nonlinear $p$-Laplacian-type equation, local boundedness of non-negative weak subsolutions follows from the energy class alone whenever $m(p-1)+(m+1)p/N\ge 1$, eliminating the extra $L^s_{\mathrm{loc}}$ hypothesis used in previous borderline results.
For double-phase and Orlicz operators with growth satisfying (4.1), subsolutions are locally bounded under $\max(1,n(q-1))\le m(p-1)+(m_-+1)p/N$, with a quantitative sup bound in the strict case.
For fully anisotropic doubly nonlinear equations, the same conclusion holds under $\max(1,\Lambda)\le p|\lambda/p|+(m+1)p/N$, and Theorem 1.2 shows that in the subcritical range the usual qualitative local-boundedness assumption can be replaced by an $L^s$ assumption with sharp $\kappa_s>0$.
The a priori estimates (1.12) and (1.15) provide explicit control of the essential supremum on interior cylinders in terms of the data, so they can be used as the first step toward continuity, Harnack inequalities, and further regularity for these classes.
Reading between the lines
Editorial extensions of the paper, not claims the author makes directly.
A natural next step, not taken in the paper, would be to replace the $L^s$ hypothesis in Theorem 1.2 by a weaker Lorentz-space assumption; the proof's H\"older and Chebyshev steps suggest the optimal space is governed by the same exponent $\kappa_s$, but this is an extrapolation.
Because $\mathcal{PDG}^+$ is defined by an inequality rather than by an equation, I would expect the same bounds to hold for parabolic quasi-minimizers of the corresponding energies; the paper stops at weak subsolutions.
A testable extension would be to compute the sup bound from (1.12) for explicit self-similar profiles of the doubly nonlinear equation to see how sharp the constant factor is; that check is my suggestion, not something the paper reports.