A spectral approach to the narrow escape problem in two-dimensional domains

Technical Summary: A Spectral Approach to the Narrow Escape Problem in Two-Dimensional Domains

Problem Statement
This work addresses the "narrow escape problem" in a two-dimensional domain $\Omega \subset \mathbb{R}^2$. The problem involves characterizing the exit event of a Brownian motion $X_t$ starting from the quasi-stationary distribution within $\Omega$. The domain has reflecting boundary conditions on most of its boundary, except for $N$ small, disjoint exit windows $\Gamma^\varepsilon_k$ located near boundary points $x^{(k)}$. The size of these windows is parameterized by $\varepsilon = (\varepsilon_1, \dots, \varepsilon_N)$, and the analysis focuses on the asymptotic regime where $|\varepsilon| \to 0$.

The primary objective is to precisely describe the law of the couple $(\tau, X_\tau)$, where $\tau$ is the first exit time and $X_\tau$ is the exit point. Specifically, the authors aim to derive precise asymptotic expansions for:

1. The mean exit time, $E_{\nu^\varepsilon_0}(\tau) = 1/\lambda^\varepsilon_0$, where $\lambda^\varepsilon_0$ is the principal eigenvalue of the associated Laplacian.
2. The probability of exiting through a specific window $k$, $P_{\nu^\varepsilon_0}(X_\tau \in \Gamma^\varepsilon_k)$.

Methodology
The authors employ a spectral approach, analyzing the principal eigenvalue $\lambda^\varepsilon_0$ and the corresponding eigenfunction $u^\varepsilon_0$ of the operator $L^\varepsilon = -\Delta$ with Dirichlet conditions on the exit windows and Neumann conditions elsewhere.

1. Quasimode Construction: The core technical contribution is the construction of a "quasimode" $\varphi^\varepsilon$, which serves as a high-order approximation of the true eigenfunction $u^\varepsilon_0$. Unlike previous works that often required the approximation to lie within the domain of the operator, this quasimode is constructed to satisfy the eigenvalue equation approximately in $L^2(\Omega)$ and the Neumann conditions exactly, while satisfying the Dirichlet conditions only approximately (up to a high-order error).
2. Asymptotic Expansion: The quasimode is expanded in powers of the parameters $K^\varepsilon_k = -1/\ln(\varepsilon_k/2)$. The expansion takes the form:
   $$\varphi^\varepsilon(x) = 1 + \sum_{\gamma \in \mathbb{N}^N, 1 \le |\gamma| \le M} (K^\varepsilon)^\gamma f^\varepsilon_\gamma(x)$$
   where the coefficients and functions $f^\varepsilon_\gamma$ are determined recursively.
3. Singular and Regular Decomposition: The functions $f^\varepsilon_\gamma$ are constructed using a decomposition into singular parts (involving logarithmic potentials related to the exit windows) and regular parts (solving Poisson equations with specific boundary data). This allows for precise control of the behavior near the small windows.
4. Recursive Determination of Coefficients: The coefficients in the expansion (specifically the parameters $K^\varepsilon_k$ and the coefficients $b_{\gamma, j}$) are determined recursively to ensure that the quasimode vanishes on the exit windows up to a desired order $O(|K^\varepsilon|^M)$. This involves solving linear systems derived from the boundary conditions.
5. Spectral Analysis: Once the quasimode is constructed, the authors use Green's formulas and variational arguments to relate the properties of $\varphi^\varepsilon$ to the true eigenvalue $\lambda^\varepsilon_0$ and the exit probabilities.

Key Contributions and Results

• Precise Expansion of Mean Exit Time: The paper provides a precise asymptotic expansion for the principal eigenvalue $\lambda^\varepsilon_0$ (and thus the mean exit time) in terms of the parameters $K^\varepsilon$.
  $$\lambda^\varepsilon_0 = \tilde{\lambda}^\varepsilon_0 + O(|K^\varepsilon|^{M+1})$$
  where $\tilde{\lambda}^\varepsilon_0$ is a polynomial in $K^\varepsilon$ with explicitly computable coefficients. This improves upon previous results that often provided only leading-order terms or were restricted to specific geometries (like disks).

• Expansion of Exit Probabilities: The authors derive an expansion for the probability of exiting through window $k$, denoted $Y^\varepsilon_k$. They show that these probabilities can be expressed as a series in powers of $K^\varepsilon$:
  $$Y^\varepsilon_{k_0} = \sum_{q=0}^{M-1} \sum_{m=0}^q \frac{Q^{(k_0)}_{M,m,q}(K^\varepsilon)}{p_M(K^\varepsilon)^{m+1}} + O(|K^\varepsilon|^M)$$
  where $p_M(K^\varepsilon)$ is a specific homogeneous polynomial related to the determinant of a matrix arising from the linear system of coefficients. The leading term is shown to be $Y^\varepsilon_{k_0} \approx K^\varepsilon_{k_0} / K^\varepsilon$, recovering the intuitive result that exit probability is proportional to the inverse logarithmic capacity of the window.

• General Domain Applicability: Unlike many existing works that rely on specific geometries (e.g., disks or spheres) or layer potential techniques restricted to those shapes, this approach is formulated for general $C^\infty$ two-dimensional domains.

Significance and Claims
The authors position this work as a counterpart to their previous studies on energetic metastability, aiming to identify the "entropic" equivalent of the celebrated Eyring-Kramers laws.

• Mathematical Rigor: The paper claims to provide the first rigorous mathematical results dealing with the distribution of the exit point (not just the mean exit time) in the narrow escape limit for general domains.
• Beyond Leading Order: A significant novelty is the ability to obtain precise expansions of arbitrary order $M$, rather than just first-order approximations. This allows for a more detailed characterization of the exit event.
• Technical Innovation: The method allows for the use of a quasimode that does not belong to the domain of the operator $L^\varepsilon$ (nor even the domain of the associated quadratic form), which simplifies the construction while maintaining high precision.
• Context: The work is motivated by biological applications (e.g., molecular exit from cells) and the study of entropic metastability, where geometric constraints, rather than energy barriers, dictate the timescale of escape.

The paper does not propose new experimental setups or specific future applications beyond the mathematical characterization of the exit law. It focuses strictly on the analytical derivation of these asymptotic expansions.