Self-consistent equations and Quantum Diffusion for the Anderson model
Adam Black, Reuben Drogin, Felipe Hernández · 2025
Read Self-consistent equations and Quantum Diffusion for the Anderson model by Adam Black, Reuben Drogin, Felipe Hernández (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-06468.txt.
Introduction
Adam Black, Reuben Drogin, and Felipe Hernández present a 2025 mathematics paper on the Anderson model. The model describes an electron hopping on a lattice with random noise. At low disorder, does the electron spread out or stay stuck? The authors study the resolvent, a tool that probes the system's energies, on lattices of dimension two and up. They derive a diffusive scaling limit at fine energy resolution, with high probability.
From that limit, quantum diffusion follows in a time-averaged sense. The timescale is the longest yet known for this model. Lower bounds on how far eigenfunctions spread also improve. The two-dimensional case is the first diffusion result of its kind. The proof avoids diagram expansions and instead solves self-consistent equations for the resolvent. Physicists gain firmer ground under a fifty-year-old question about disorder and transport.
Catalog record
| CREATOR | Adam Black, Reuben Drogin, Felipe Hernández |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 125 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-06468
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Self-consistent equations and Quantum Diffusion for the Anderson model Adam Black, Reuben Drogin, Felipe Hernández Abstract . We consider the Anderson tight-binding model on ℤ d {\mathbb{Z}}^{d} , d ≥ 2 d\geq 2 , with Gaussian noise and at low disorder λ > 0 \lambda>0 . We derive a diffusive scaling limit for the entries of the resolvent R ( z ) R(z) at imaginary part Im z ∼ λ 2 + κ d \operatorname*{Im}z\sim\lambda^{2+\kappa_{d}} , κ d > 0 \kappa_{d}>0 , with high probability. As consequences, we establish quantum diffusion (in a time-averaged sense) for the Schrödinger propagator at the longest timescale known to date and improve the best available lower bounds on the localization length of eigenfunctions. Our results for d = 2 d=2 are the first quantum diffusion results for the Anderson model on ℤ 2 {\mathbb{Z}}^{2} . The proof avoids the use of diagrammatic expansions and instead proceeds by analyzing certain self-consistent equations for R ( z ) R(z) . This is facilitated by new estimates for ‖ R ( z ) ‖ ℓ p → ℓ q \|R(z)\|_{\ell^{p}\rightarrow\ell^{q}} that control the recollisions. 1. Introduction In this paper, we consider the Anderson tight-binding model (1.1) H=\Delta_{{\mathbb{Z}}^{d}}+\lambda V, where \Delta_{{\mathbb{Z}}^{d}} is the nearest-neighbor Laplacian on {\mathbb{Z}}^{d} (normalized as in ( 2.1 )), the potential V:{\mathbb{Z}}^{d}\to{\mathbb{R}} has independent standard Gaussian values, and \lambda>0 is a coupling constant. The Hamiltonian ( 1.1 ) was originally introduced by Anderson [ And58 ] in order to study the propagation of electrons in a disordered environment, and more generally functions as a simple model for wave transport in random media. For \lambda\gg 1 or d=1 , there is a large body of literature showing that H exhibits Anderson localization , meaning that almost surely it has an orthonormal basis of exponentially decaying eigenfunctions, see [ AW15 ] and the references therein. In contrast, the small coupling regime \lambda\ll 1 is the subject of several outstanding conjectures in mathematical physics ( [ Sim84 , Sim00 ] ). Referred to as the extended states and quantum diffusion conjectures, they predict the existence of delocalized states whose Schrödinger evolution is described by a diffusion process. We list a few forms of these conjectures informally, which are expected to hold for d\geq 3 and sufficiently small \lambda : • Diffusion of the propagator : For fixed \psi\in\ell^{2}({\mathbb{Z}}^{d}) the probability density of the wavefunction |(e^{-itH}\psi)(x)|^{2} satisfies a heat equation under an appropriate scaling limit. In particular, if \psi is localized, then (1.2) \sum_{x\in{\mathbb{Z}}^{d}}|x|^{2}|(e^{-itH}\psi)(x)|^{2}\sim\lambda^{-2}t, implying that e^{-itH}\psi has most of its \ell^{2} mass at the diffusive scale \lambda^{-1}\sqrt{t}. • Diffusion of the resolvent: For E in the interior of the spectrum \sigma(\Delta) , the squared entries of the resolvent R(z):=(H-E-i\eta)^{-1} satisfy an elliptic equation under an appropriate scaling limit as \eta\to 0 . In particular, they scale like the Green’s function of the Laplacian: (1.3) |R(z)_{xy}|^{2}\sim\lambda^{2}|x-y|^{2-d}. • Delocalization of eigenfunctions: Solutions of H\psi=E\psi are delocalized for energies E in the interior of \sigma(\Delta) . On a qualitative level, this means that the spectrum is continuous in the bulk. Quantitatively, this should be interpreted to mean that most eigenfunctions of some truncation of H satisfy a lower bound on their “localization length,” i.e., one of several measures of their spatial extent. The factors of \lambda^{-2} and \lambda^{2} in the above are significant because they are, respectively, the critical time and energy scales past which diffusion may first be observed. Since the propagator and the resolvent are related by a Fourier transform in the time/energy variables, the first two items are thought of as equivalent, although in practice establishing the correspondence is non-trivial. In dimension d=2 , it is instead conjectured that the spectrum of H is pure point with dynamically localized eigenfunctions. Nevertheless, in this dimension, diffusion is still expected up to the timescale e^{\lambda^{-2}} (or dually down to the energy scale \operatorname*{Im}z\sim e^{-\lambda^{-2}} ). Progress on these conjectures, reviewed below, has been limited to a handful of results and has mostly focused on establishing a scaling limit for the propagator in expectation and at timescales just beyond the diffusive threshold of \lambda^{-2} . In this paper, we show that for d\geq 2 the resolvent R(z) is diffusive down to energy scales \operatorname*{Im}z\sim\lambda^{2+\kappa_{d}} , \kappa_{d}>0 , with high probability . As consequences, we establish quantum diffusion (in a time-averaged sense) for the propagator at the longest timescale known to date and improve the best available lower bounds on the localization length of eigenfunctions. Our results for d=2 are the first quantum diffusion results for the Anderson model on {\mathbb{Z}}^{2} . We now state our main results before comparing them more closely to prior work. Below we use (e^{-itH})_{xy} and R(z)_{xy} to refer to the entries of e^{-itH} and R(z) considered as matrices. Note that H is self-adjoint as the sum of a bounded operator and a multiplication operator, so there is no trouble defining its propagator. Theorem 1.1 (Dynamical delocalization) . Let \delta>0 and H be as in ( 1.1 ). There exists c_{\delta},C_{\delta}>0 , independent of \lambda , such that for any C_{\delta}\leq T\leq\lambda^{-2-\kappa_{d}+\delta} we have (1.4) \frac{1}{T}\int_{0}^{T}\sum_{x\in\mathbb{Z}^{d}}|x|^{2}|(e^{-itH})_{0x}|^{2}\geq c_{\delta}\lambda^{-2}T and (1.5) \frac{1}{T}\int_{0}^{T}\sum_{|x|\geq C_{\delta}\lambda^{-1}\sqrt{T}}|(e^{-itH})_{0x}|^{2}\leq\delta, with probability at least 1-C_{\delta}\lambda^{1000} and \kappa_{d}>0 given by \displaystyle\kappa_{d}=\begin{cases}\frac{2}{13},&d=2\\ \frac{2}{9},&d=3\\ \frac{2}{3}-O(d^{-1}),&d\geq 4.\end{cases} See the statement of Theorem 5.2 for the precise values of \kappa_{d} . Next we state our result about the diffusive scaling of the resolvent. For technical reasons, explained in Section 2 , we exclude the following discrete set of energies (in addition to the energies outside of the spectrum of \Delta ) \Sigma_{d}:=({\mathbb{R}}\setminus[-2d,2d])\cup(2d+4{\mathbb{Z}})\cup\{0\}. Theorem 1.2 . Let d\geq 3 , E\not\in\Sigma_{d} , and \kappa<\kappa_{d} . Then there is a constant \beta_{E} and \kappa^{\prime}<\kappa such that the following holds: For any f\in C_{c}^{\infty}({\mathbb{R}}^{d}) and setting \ell=\lambda^{-2-\kappa^{\prime}/2} we have, almost surely, (1.6) \lim_{\lambda\to 0}\lambda^{-2}\ell^{-2}\sum_{x\in{\mathbb{Z}}^{d}}f(x/\ell)|R(E+i\lambda^{2+\kappa})_{0x}|^{2}=c_{d}\beta_{E}\int f(y)|y|^{2-d}\mathop{}\!\mathrm{d}y. The analogous result holds in d=2 provided f=\Div\vec{v} for a vector field \vec{v}\in C_{c}^{\infty}({\mathbb{R}}^{d};{\mathbb{R}}^{d}) and |y|^{2-d} is replaced by -\log(|y|) . The above result establishes an averaged version of the scaling ( 1.3 ). Indeed, rearranging the above equation we obtain \sum_{x\in{\mathbb{Z}}^{d}}f(x/\ell)|R_{0x}|^{2}=\lambda^{2}c_{d}\beta_{E}\int f(y/\ell)|y|^{2-d}\mathop{}\!\mathrm{d}y+o(\lambda^{2}\ell^{2}). To understand the scaling of \ell , note that with \eta:=\operatorname*{Im}z=\lambda^{2+\kappa} , we have \ell\ll\lambda^{-1}\eta^{-1/2} . The scale \lambda^{-1}\eta^{-1/2} is the analogue of the diffusive length scale \lambda^{-1}T^{1/2} for the resolvent under the correspondence T\leftrightarrow\eta^{-1} . Finally, we state our result on the delocalization of eigenfunctions. We consider H_{L} , as in ( 1.1 ), but restricted to the torus {\mathbb{Z}}^{d}_{L}:={\mathbb{Z}}^{d}/(L{\mathbb{Z}}^{d}) with L=\left\lceil\lambda^{-100}\right\rceil (fixed throughout the paper). We take \psi_{j} to be an orthonormal basis 1 1 1 With probability 1 , the eigenvalues are distinct. of eigenfunctions with eigenvalues E_{j} so that H_{L}\psi_{j}=E_{j}\psi_{j} . For any r>0 , we define the set of localized eigenvalues by \displaystyle\mathfrak{L}(r) \displaystyle:=\{E_{j}\mid\text{ there exists }x_{0}\in{\mathbb{Z}}^{d}_{L}\text{ such that }\|\psi_{j}\|_{\ell^{2}(B_{r}(x_{0}))}\geq 1-r^{-4d}\}. Note \mathfrak{L}(r) contains all eigenvalues whose eigenfunction exponentially decays at scale \ll r . Theorem 1.3 (Delocalization bound for most eigenfunctions) . Fix any {\varepsilon},\delta>0 and E\in[-2d,2d] with d(E,\Sigma_{d})\geq 2{\varepsilon} . Then with probability at least 1-\lambda^{1000} , \displaystyle\frac{\Big|\mathfrak{L}(\lambda^{-2-\kappa_{d}/2+\delta})\cap[E-{\varepsilon},E+{\varepsilon}]\Big|}{\Big|\sigma(H)\cap[E-{\varepsilon},E+{\varepsilon}]\Big|}{\,\lesssim\,}_{\epsilon,\delta}\lambda^{c\delta} with \kappa_{d} as in Theorem 1.1 . In fact, one can take {\varepsilon}=\lambda^{c} above for c small enough depending on \delta . Moreover, we deduce Theorem 1.3 from a bound on the number of eigenfunctions localized near