SearcharxivSearch

arXiv subjects

Erisa Hasani

Publications and source records attributed to Erisa Hasani.

4 recordsLinked to original sources

The strongly nonlocal Allen-Cahn problem

We study the sharp interface limit of the fractional Allen-Cahn equation $$ \varepsilon \partial_t u^{\varepsilon} = \mathcal{I}^s_n [u^{\varepsilon}] -\frac{1}{\varepsilon ^{2s}} W'(u^\varepsilon) \quad \hbox{in}~(0,\infty)\times\mathbb{R}^n, ~n \geq 2, $$ where $\varepsilon >0$, $\mathcal{I}^s_n=-c_{n,s}(-\Delta )^s$ is the fractional Laplacian of order $2s\in(0,1)$ in $\mathbb{R}^n$, and $W$ is a smooth double-well potential with minima at 0 and 1. We focus on the singular regime $s\in(0,\frac{1}{2})$, corresponding to strongly nonlocal diffusion. For suitably prepared initial data, we prove that the solution $ u^\varepsilon $ converges, as $\varepsilon\to0$, to the minima of $W$ with the interface evolving by fractional mean curvature flow. This establishes the first rigorous convergence result in this regime, complementing and completing previous work for $s\geq \frac{1}{2}$.

math.AP

Generating synthetic data for neural operators

Recent advances in the literature show promising potential of deep learning methods, particularly neural operators, in obtaining numerical solutions to partial differential equations (PDEs) beyond the reach of current numerical solvers. However, existing data-driven approaches often rely on training data produced by numerical PDE solvers (e.g., finite difference or finite element methods). We introduce a "backward" data generation method that avoids solving the PDE numerically: by randomly sampling candidate solutions $u_j$ from the appropriate solution space (e.g., $H_0^1(\Omega)$), we compute the corresponding right-hand side $f_j$ directly from the equation by differentiation. This produces training pairs ${(f_j, u_j)}$ by computing derivatives rather than solving a PDE numerically for each data point, enabling fast, large-scale data generation consisting of exact solutions. Experiments indicate that models trained on this synthetic data generalize well when tested on data produced by standard solvers. While the idea is simple, we hope this method will expand the potential of neural PDE solvers that do not rely on classical numerical solvers to generate their data.

cs.LG

On the critical $p$-Kirchhoff equation

We study a nonlocal elliptic equation of $p$-Kirchhoff type involving the critical Sobolev exponent. First we give sufficient conditions for the (PS) condition to hold. Then we prove some existence and multiplicity results using tools from Morse theory, in particular, the notion of a cohomological local splitting and eigenvalues based on the Fadell-Rabinowitz cohomological index.

math.AP

On the compactness threshold in the critical Kirchhoff equation

We study a class of critical Kirchhoff problems with a general nonlocal term. The main difficulty here is the absence of a closed-form formula for the compactness threshold. First we obtain a variational characterization of this threshold level. Then we prove a series of existence and multiplicity results based on this variational characterization.

math.AP