gpt 6 astra pro attempt on Phys-4

Extended States in the Anderson Model

Definitions and mathematical statement

For independent identically distributed real random variables $V_x$, the standard lattice Anderson operator on $\ell^2({\mathbb{Z}}^d)$ has the form \[ (H_\lambda\psi)(x)=-\sum_{|y-x|_1=1}\psi(y)+\lambda V_x\psi(x). \] For a vector $\psi$, its spectral measure is $\mu_\psi(B)=\langle\psi,1_B(H_\lambda)\psi\rangle$; an absolutely continuous spectral component has density with respect to Lebesgue measure. The central target is a nonzero absolutely continuous spectral subspace in a weak-disorder regime in $d\ge3$. The catalogue does not specify a probability law or exact energy interval. Those choices must be fixed in a full model-specific theorem. Its additional language about transport and a mobility edge is not, by definition alone, equivalent to existence of absolutely continuous spectrum.

Formulation and concept references: [S1].

Short English statement

Can weak random disorder in a lattice of dimension at least three be proved to permit an extended-state spectral regime?

Research attempt

Current frontier.

Kirsch distinguishes spectral type, dynamical localization and averaged density-of-states information; these are different assertions. [S1]. Black--Drogin--Hernández establish weak-disorder diffusion for Gaussian disorder down to a stated imaginary-resolvent scale $\eta\asymp\lambda^{2+\kappa_d}$, with $\kappa_d>0$, in their November 2025 version. It does not let $\eta\downarrow0$ at fixed positive $\lambda$. [S3], [E1]. Jakšić--Last--Warzel's September 2026 manuscript proves a Bethe-lattice result and proposes a Poisson--Schur criterion on ${\mathbb{Z}}^d$, explicitly leaving its weak-disorder lower bound unproved there. [S2], [E2]. Becker--Oltman's September 2026 manuscript concerns a hyperbolic-square-lattice product, not ${\mathbb{Z}}^d$ alone. [E3]. Recent statements and their model scopes were checked; they are not treated as independently audited solutions of this entry.

Attack route.

Editorial scope note. The original does not prescribe a single-site law and itself warns that absolutely continuous spectrum is not equivalent, by definition, to transport or a mobility edge. To make a concrete attempt, fix $d\ge3$ and independent $V_x$ uniformly distributed on $[-1,1]$, with $\lambda>0$. This is a nondegenerate random model, not the trivial free case. Results below concern its spectral component only; they do not silently settle the additional transport language or all possible laws.

The selected route is to upgrade resolvent control to a quenched statement about individual spectral measures. A uniform $L^2$ bound on their Poisson transforms is stronger than needed, but it gives a clean sufficient estimate whose limiting steps can be proved. The competing tree-cyclicity route would instead need a lower bound on a matrix Schur complement; independence of forward branches is absent on ${\mathbb{Z}}^d$. [E2], Section 4. I investigate the first route and test whether averaging or small operator norm perturbations could supply its missing estimate.

Attempt.

Write \[ G_x(E+i\eta)=\langle\delta_x,(H_\lambda-E-i\eta)^{-1}\delta_x\rangle, \qquad u_{x,\eta}(E)=\operatorname{Im}G_x(E+i\eta). \] For the probability measure $\mu_x$ associated to $\delta_x$, the spectral theorem gives \begin{equation}\label{wsref-0-72-33-31-3a-70-6f-69-73-73-6f-6e} u_{x,\eta}(E)=\int\frac{\eta}{(t-E)^2+\eta^2}\,d\mu_x(t)\ge0. \end{equation} The absence of $1/\pi$ in this definition will be retained in all limits. Spectral measures and the resolvent framework are those of [S1].

Lemma 31.1 (an all-resolution sufficient estimate). Let $I$ be an open interval. For fixed $\lambda>0$, suppose that for every closed interval $J\Subset I$, \begin{equation}\label{wsref-0-72-33-31-3a-6d-69-73-73-69-6e-67} \sup_{0<\eta<1}\mathbb E\int_J u_{0,\eta}(E)^2\,dE<\infty. \end{equation} Then, almost surely, $H_\lambda$ has no singular spectral component on $I$.

Proof. Fix $J$ and put $\eta_l=2^{-l}$. Fatou's lemma implies \[ \mathbb E\left[\liminf_l\|u_{0,\eta_l}\|_{L^2(J)}^2\right] \le\liminf_l\mathbb E\|u_{0,\eta_l}\|_{L^2(J)}^2<\infty. \] Thus for almost every realization there is a subsequence with bounded $L^2(J)$ norm. It need not be a deterministic subsequence or a subsequence chosen independently of the realization. Weak compactness of $L^2(J)$ gives a further weakly convergent subsequence with limit $h\in L^2(J)$. For $\phi\in C_c(J^\circ)$, Fubini and the Poisson approximate identity give \[ \lim_{\eta\downarrow0}\int\phi(E)u_{0,\eta}(E)\,dE =\pi\int\phi(t)\,d\mu_0(t). \] Indeed, the inner Poisson integral converges uniformly to $\pi\phi(t)$ and is bounded by $\pi\|\phi\|_\infty$, so integration against the finite measure is justified. Weak $L^2$ convergence identifies the same limit as $\int\phi h$. Hence $\mu_0|_{J^\circ}=h(E)dE/\pi$.

Translation invariance of the law gives the same conclusion for each $x$. Intersect the probability-one events for countably many $x\in{\mathbb{Z}}^d$ and a countable exhaustion of $I$ by such intervals. All basis-vector spectral measures are then absolutely continuous on $I$. The singular spectral projection on $I$ annihilates every $\delta_x$, and thus, by density, the whole Hilbert space. This proves the lemma without interchanging an uncontrolled random pointwise limit and expectation.$\square$

Lemma 31.2 (nonvanishing is available without a new randomness estimate). Choose a nonnegative $\phi\in C_c(I)$ such that $a_0=\langle\delta_0,\phi(H_0)\delta_0\rangle>0$. There is $\lambda_1>0$ so that for every $0<\lambda<\lambda_1$ and every realization of bounded $|V_x|\le1$, \begin{equation}\label{wsref-0-72-33-31-3a-6d-61-73-73} \langle\delta_0,\phi(H_\lambda)\delta_0\rangle>a_0/2. \end{equation} Proof. The operator norm of $H_\lambda-H_0$ is at most $\lambda$. For $\lambda\le1$, both spectra lie in $[-2d-1,2d+1]$. Approximate $\phi$ uniformly on this interval by a polynomial $P$. For a fixed polynomial, \[ \|A^j-B^j\|\le j\max(\|A\|,\|B\|)^{j-1}\|A-B\| \] follows from the telescoping identity, with no commutativity required. Consequently $\|P(H_\lambda)-P(H_0)\|\to0$ uniformly in the realization. Choosing the approximation error first, and then $\lambda$, proves (r31:mass).$\square$

Such a $\phi$ exists for every nonempty open $I\subset(-2d,2d)$: Fourier transform diagonalizes $H_0$ as multiplication by $-2\sum_{j=1}^d\cos\theta_j$, and the preimage of a nonempty open energy subinterval has positive torus measure. Thus proving (r31:missing) for a nonempty $I$ and every sufficiently small positive $\lambda$ would establish a nonzero absolutely continuous spectral subspace for this specified model, indeed purity on $I$. The constants may depend on $\lambda$; uniformity in resolution $\eta$ at each fixed $\lambda$ is the indispensable demand.

Attempted shortcut: average density instead of squared density. This shortcut is false. Let $V$ be uniform on $[-1,1]$ and take the one-dimensional operator with spectral measure $\mu=\delta_V$. Every realization is purely atomic, while $\mathbb E\mu$ has density $1/2$ on $[-1,1]$. Moreover \[ \mathbb E\frac{\eta}{(V-E)^2+\eta^2}\le\frac\pi2 \quad\text{for all }E,\eta>0, \] yet \begin{equation}\label{wsref-0-72-33-31-3a-61-74-6f-6d-69-63} \int_{{\mathbb{R}}}\left(\frac{\eta}{(V-E)^2+\eta^2}\right)^2dE =\frac{\pi}{2\eta}. \end{equation} The last identity follows by $E-V=\eta t$ and $\int_{{\mathbb{R}}}(1+t^2)^{-2}dt=\pi/2$. A bounded averaged Poisson density can therefore conceal singular spectrum. This is not a counterexample to the nearest-neighbor lattice conjecture; it disproves the proposed inference from annealed to quenched spectral type.

Attempted shortcut: a uniform Neumann expansion. Although $\|\lambda V\|\le\lambda$, the free resolvent norm is at most $1/\eta$. Expanding \[ (H_\lambda-z)^{-1} =(H_0-z)^{-1}\left[I+\lambda V(H_0-z)^{-1}\right]^{-1} \] by its geometric series is guaranteed only when $\lambda/\eta<1$. That regime excludes the required limit $\eta\downarrow0$ with fixed $\lambda$. Averaging may cancel some terms, but a proof of a remainder uniform in $\eta$ is an additional estimate, not supplied by the norm bound. The controlled mesoscopic scale in [E1] does not bridge this gap, and its Gaussian law is not the bounded law chosen above.

Stress test.

The order of limits cannot be changed unnoticed. Every finite-volume spectral measure is atomic. Its positive squared Poisson transform has diagonal contributions of order $\eta^{-1}$, just as in (r31:atomic); therefore one cannot prove the required all-resolution bound at each fixed finite volume and then take volume to infinity. Nor does the small-$\lambda$ continuity in Lemma 31.2 preserve spectral type: it preserves only positive mass for a fixed continuous test function. Lemma 31.1 supplies the separate absolute-continuity step conditionally.

No numerical transport experiment is interpreted as a spectral proof here. The companion script symbolically checks the Poisson-square integral. The probability-one conclusion was derived for each fixed $\lambda$; no uncountable intersection over all disorder strengths is asserted. Finally, existence of an ac subspace, a quantitative conductivity statement, and a sharp mobility edge remain separate goals, as the original editorial formulation already notes.

\Needspace{8\baselineskip}

Outcome.

Outcome: Conditional reduction.

For an explicitly fixed nondegenerate bounded single-site law, an all-resolution second-moment resolvent estimate implies the desired nonzero absolutely continuous spectral regime, with its nonvanishing and almost-sure conclusions proved in detail. The estimate itself is not proved. The averaging and norm-perturbation shortcuts are rejected by an exact model calculation and a parameter-order obstruction. This is a sufficient route for the spectral part of a specified model, not a completion of the underspecified law/transport/mobility-edge package, and the criterion's historical novelty is not asserted.

Sources

• S1 W. Kirsch, "An Invitation to Random Schrodinger Operators", in Random Schrodinger Operators, Panoramas et Syntheses 25, Societe Mathematique de France, 2008; arXiv:0709.3707. https://arxiv.org/abs/0709.3707. Catalogue formulation source.
• S2 A Note on Extended States on the Bethe Lattice. https://arxiv.org/html/2609.06987v1. Catalogue research/status reference; not a proof endorsement.
• S3 Self-consistent equations and quantum diffusion for the Anderson model. https://arxiv.org/abs/2506.06468. Catalogue research/status reference; not a proof endorsement.
• E1 Adam Black, Reuben Drogin and Felipe Hernández, Self-consistent equations and quantum diffusion for the Anderson model, arXiv:2506.06468v2, 6 November 2025, abstract and resolution-scale statement. https://arxiv.org/abs/2506.06468v2. Gaussian disorder and fixed scale are retained; checked 27 September 2026.
• E2 Vojkan Jakšić, Yoram Last and Simone Warzel, A Note on Extended States on the Bethe Lattice, arXiv:2609.06987v1, Sections 1 and 4, Lemma 4.1 and Theorem 4.2. https://arxiv.org/html/2609.06987v1. The lattice criterion is not its unproved weak-disorder estimate. Consulted 27 September 2026.
• E3 Simon Becker and Izak Oltman, Disorder on the hyperbolic square lattice I: Anderson delocalization and absolutely continuous spectrum, arXiv:2609.20798v1, September 2026, abstract and graph hypotheses. https://arxiv.org/abs/2609.20798v1. Recent manuscript on a different graph; checked 27 September 2026, not fully audited here.