ScalingStacks

5. W 2 , 1 Regularity [04U8]

Original official author HTML, exact retained edition. Historical TeX conversion verdicts remain unchanged. Cited-edition alignment and mathematical self-containment are not assessed.

Complete original source context · Original author HTML

5. W2,1W^{2,1} Regularity

In this section we obtain W2,1W^{2,1} regularity for singular solutions to the Monge-Ampère equation. Furthermore, by examining the examples in the previous section we show that we cannot improve this result to W2,1+ϵW^{2,1+\epsilon} regularity for an ϵ\epsilon depending on λ,Λ\lambda,\Lambda and nn.

The following result of Savin, De Philippis and Figalli (see [DFS]) gives W2,1+ϵW^{2,1+\epsilon} regularity of solutions to λ≤detD2​u≤Λ\lambda\leq\det D^{2}u\leq\Lambda in compactly contained sections:

Theorem 5.1.

Assume that

λ≤detD2​u≤Λ in ​Ω​ and ​Sh​(x)⊂⊂Ω.\lambda\leq\det D^{2}u\leq\Lambda\quad\text{ in }\Omega\text{ and }S_{h}(x)\subset\subset\Omega.

Then u∈W2,1+ϵ​(Sh/2​(x))u\in W^{2,1+\epsilon}(S_{h/2}(x)) for some ϵ\epsilon depending only on λ,Λ\lambda,\Lambda and nn.

W2,1W^{2,1} regularity then follows from our main theorem.

Proof of Theorem 1.2:.

Theorem 5.1 gives local W2,1W^{2,1} regularity on Ω−Σ\Omega-\Sigma. By Theorem 1.1, for any η>0\eta>0 we can cover Σ\Sigma by balls {Bri​(xi)}\{B_{r_{i}}(x_{i})\} such that

∑i=1∞rin−1<η.\sum_{i=1}^{\infty}r_{i}^{n-1}<\eta.

Let A=∪i=1∞Bri(xi)A=\cup_{i=1}^{\infty}B_{r_{i}}(x_{i}). Since uu is a convex function, the second derivatives are controlled by Δ​u\Delta u. It follows that

∫A‖D2​u‖​𝑑x\displaystyle\int_{A}\|D^{2}u\|\,dx ≤∫AΔ​u​𝑑x\displaystyle\leq\int_{A}\Delta u\,dx
≤∑i=1∞∫∂Briuν​𝑑s\displaystyle\leq\sum_{i=1}^{\infty}\int_{\partial B_{r_{i}}}u_{\nu}\,ds
≤C​∑i=1∞rin−1\displaystyle\leq C\sum_{i=1}^{\infty}r_{i}^{n-1}
≤C​η,\displaystyle\leq C\eta,

where CC is the Lipschitz constant of uu. This shows that the second derivatives cannot concentrate on Σ\Sigma. ∎

We now examine the integrability of Δ​u\Delta u for the examples constructed in the previous section. On any ball BrB_{r}, by Hölder’s inequality we have

∫Br(Δ​u)1+ϵ​𝑑x≥c⁡(n)​r−ϵ​n​(∫BrΔ​u​𝑑x)1+ϵ.\int_{B_{r}}(\Delta u)^{1+\epsilon}\,dx\geq c(n)r^{-\epsilon n}\left(\int_{B_{r}}\Delta u\,dx\right)^{1+\epsilon}.

Recall that the subsolutions ww grow like x2β=x21+δ4−3​δx_{2}^{\beta}=x_{2}^{1+\frac{\delta}{4-3\delta}} in the x2x_{2} direction, and that these functions touch uu by below at any x∈Σ=S×{0}×(−1,1)n−2x\in\Sigma=S\times\{0\}\times(-1,1)^{n-2}. It follows that

sup∂Br​(x)(u−u⁡(x))≥rβ\sup_{\partial B_{r}(x)}(u-u(x))\geq r^{\beta}

for any x∈Σx\in\Sigma. Applying convexity,

∫Br​(x)(Δ​u)1+ϵ​𝑑x\displaystyle\int_{B_{r}(x)}(\Delta u)^{1+\epsilon}\,dx ≥c⁡(n)​r−ϵ​n​(∫∂Bruν​𝑑s)1+ϵ\displaystyle\geq c(n)r^{-\epsilon n}\left(\int_{\partial B_{r}}u_{\nu}\,ds\right)^{1+\epsilon}
≥c⁡(n)​r(n+β−2)​(1+ϵ)−ϵ​n\displaystyle\geq c(n)r^{(n+\beta-2)(1+\epsilon)-\epsilon n}
≥c⁡(n)​rn−1−ϵ+(1+ϵ)​δ3.\displaystyle\geq c(n)r^{n-1-\epsilon+(1+\epsilon)\frac{\delta}{3}}.

Fix η\eta small and cover S×{0}×(−1,1)n−2S\times\{0\}\times(-1,1)^{n-2} with balls of radius ri<ηr_{i}<\eta. Take a Vitali subcover {Bri}i=1∞\{B_{r_{i}}\}_{i=1}^{\infty}. It follows that

∫B1(Δ​u)1+ϵ​𝑑x≥c⁡(n)​∑i=1∞rin−1−ϵ+(1+ϵ)​δ3.\int_{B_{1}}(\Delta u)^{1+\epsilon}\,dx\geq c(n)\sum_{i=1}^{\infty}r_{i}^{n-1-\epsilon+(1+\epsilon)\frac{\delta}{3}}.

Taking ϵ=4​δ\epsilon=4\delta above, we conclude that

∫B1(Δ​u)1+ϵ​𝑑x≥c⁡(n)​∑i=1∞rin−1−3​δ,\int_{B_{1}}(\Delta u)^{1+\epsilon}\,dx\geq c(n)\sum_{i=1}^{\infty}r_{i}^{n-1-3\delta},

where the expression on the right goes to ∞\infty as η→0\eta\rightarrow 0 because the Hausdorff dimension of S×{0}×(−1,1)n−2S\times\{0\}\times(-1,1)^{n-2} is n−1−32​δn-1-\frac{3}{2}\delta. Thus, Δ​u\Delta u is not L1+ϵL^{1+\epsilon} for ϵ≥4​δ\epsilon\geq 4\delta.

Remark 5.2.

In future work we intend to present a more precise version of Theorem 1.1 which gives L​log⁡LL\log L regularity of second derivatives of singular solutions to detD2​u=1\det D^{2}u=1.

Original mathematics by the credited authors. Source-backed reader collection; mathematical self-containment is not assessed.