The four-dimensional Anderson model: a case study for critical SPDEs

Technical Summary: The Four-Dimensional Anderson Model: A Case Study for Critical SPDEs

Problem Statement
This paper investigates the weakly coupled elliptic Anderson model on the four-dimensional torus $\mathbb{T}^4$ with spatial white noise. The model is defined by the Schrödinger operator $L_\varepsilon = 1 - \Delta - \lambda_\varepsilon \xi_\varepsilon + C_\varepsilon$, where $\xi_\varepsilon$ is a mollified white noise, $\lambda_\varepsilon = \lambda |\log \varepsilon|^{-1/2}$ is a critical coupling constant, and $C_\varepsilon$ is a renormalization constant. The central challenge lies in the "critical" nature of the equation: unlike subcritical SPDEs where finite-order expansions suffice, critical models require infinite-order expansions and involve infinitely many renormalization terms. Specifically, the authors aim to prove that the Green's function $G_\varepsilon$ of $L_\varepsilon$, when suitably centered and rescaled, converges to a centered Gaussian random field as the regularization parameter $\varepsilon \to 0$.

Methodology
The authors develop a rigorous framework to handle the combinatorial and analytic difficulties inherent in critical SPDEs, where standard perturbative expansions fail due to factorial divergences.

1. Truncated Renormalized Parametrix:
   Instead of attempting to control an infinite series, the authors construct a parametrix $P_\varepsilon$ by truncating the Neumann series expansion at a level $A \sim |\log \varepsilon|$. They define a renormalized operator $L_\varepsilon$ by subtracting a sum of renormalization constants $C_\varepsilon = \sum_{q \le A} C_{2q}$. The error terms $R_\varepsilon$ and $R'_\varepsilon$ are shown to be small in operator norm, allowing the inversion of $L_\varepsilon$ via a Neumann series.

2. Hepp Trees and Multiscale Analysis:
   To estimate the terms in the expansion, the authors employ a multiscale analysis based on Hepp trees. These trees classify the relative scales of distances between points in the integration domain. A key innovation is the adaptation of Hepp trees to handle permutations of indices, distinguishing between "safe" and "unsafe" groups of points based on their scale separation.

3. Combinatorial Estimates and Permutation Summation:
   A major technical hurdle is the summation over permutations $\pi \in S_n$ (where $n \sim |\log \varepsilon|$), which typically introduces a factorial loss $n!$. The authors establish a precise balance between this factorial loss and "logarithmic losses" arising from scale summations.

   • They introduce a classification of permutations based on an index set $W \subseteq B \setminus \{r\}$ (where $B$ are branching nodes of the Hepp tree).
   • They prove that permutations where $W$ is large (contributing to logarithmic losses) are few in number, while those with small $W$ (fewer logarithmic losses) are numerous but controlled by factorial gains.
   • This balance is quantified through sharp estimates involving the structure of the Hepp tree and the automorphism group, preventing the divergence of the series.

4. Renormalization Constants:
   The renormalization constants $C_{2q}$ are constructed algebraically following BPHZ theory but adapted to the linear Anderson model. The authors prove that these constants, which sum up to a diverging number of terms, can be bounded effectively using the same combinatorial machinery developed for the main terms.

Key Results

• Convergence Theorem (Theorem 1.1): For sufficiently small $\lambda > 0$, there exists a choice of renormalization constants $C_\varepsilon$ such that the rescaled difference $\lambda_\varepsilon^{-1}(G_\varepsilon(x, y) - G(x, y))$ converges in law to a centered Gaussian random field $H(x, y)$.
• Explicit Covariance: The limiting field $H$ has a covariance function independent of the cutoff function $\rho$, given explicitly by:
  $$\mathbb{E}(H(x_1, y_1)H(x_2, y_2)) = \frac{2\pi^2}{2\pi^2 - \lambda^2} \int_{\mathbb{T}^4} G(x_1 - z)G(y_1 - z)G(x_2 - z)G(y_2 - z) \, dz$$
• Sharp Bounds: The paper provides sharp high-order bounds for the remainder of the parametrix and the renormalization constants, demonstrating that the series converges despite the critical scaling.

Significance and Claims
The authors position this work as a foundational step toward a general theory for critical SPDEs with weak couplings.

• First Rigorous Result: They claim this is the first convergence result for a critical SPDE where infinitely many renormalization terms are required. Previous results (e.g., [GR26]) established term-by-term convergence but did not sum the infinite series of renormalization constants.
• Critical vs. Subcritical: The paper highlights the distinct difficulty of critical models compared to subcritical ones: the necessity of expanding up to order $|\log \varepsilon|$ and the presence of infinitely many intrinsic renormalization terms.
• Methodological Contribution: The developed machinery, particularly the "truncated renormalized parametrix" and the new estimates for summations over permutations using Hepp trees, is intended to be applicable to more complex nonlinear critical models, such as the $\Phi^4_4$ model and Yang-Mills theory, which are slated for future investigation.
• Comparison to SHE: The authors note a distinction from the 2D Stochastic Heat Equation (SHE), where the heat kernel's one-sided support simplifies the combinatorics (no self-pairing terms). In the 4D Anderson model, the lack of such support necessitates handling infinitely many renormalization terms and complex combinatorial structures involving $n!$ divergences.

The paper does not claim to resolve the phase transition threshold (e.g., whether the limit remains Gaussian for all $\lambda < \sqrt{2\pi}$) but restricts its proof to sufficiently small $\lambda$, noting that extending the result to the full range would require even more precise estimates.