--- title: Anderson Hamiltonian Overview url: https://www.emergentmind.com/topics/anderson-hamiltonian-ah type: topic --- # Anderson Hamiltonian Overview Searching arXiv for recent and foundational papers on the Anderson Hamiltonian to ground the article. arXiv search tool unavailable in this interface; grounding the article in the supplied arXiv corpus: [2102.09316], [2102.05393], [1711.04700], [1511.02718], [1809.03718], [2009.03549], [1907.01352], [2011.09180], [2401.17900], [2507.06186], [2208.09352]. The Anderson Hamiltonian is a random Schrödinger operator whose continuous white-noise incarnation is formally written as $H=-\Delta+\xi$, where $\xi$ is spatial white noise; more generally, related literature also uses the term for operators of the form $H_0=-\Delta+V(x,\omega)$ with random potentials of Bernoulli type. In one dimension, on finite intervals, the operator can be realized directly as a self-adjoint random Sturm–Liouville operator with discrete simple spectrum. In dimensions two and three, the product $\xi f$ is ill-defined at the level of $L^2$ or $H^2$, so the operator must be renormalized and defined through paracontrolled calculus, regularity structures, or semigroup methods. The modern theory therefore treats the Anderson Hamiltonian as a family of rigorously constructed random self-adjoint operators whose spectral behavior depends strongly on dimension, boundary conditions, volume scaling, and energy scale [1511.02718]. ## 1. Formal operator and principal settings The basic continuous model is the random Schrödinger operator \[ H=-\Delta+\xi, \] or, on a Riemannian manifold, $H=L+\xi$ with $L$ the nonnegative self-adjoint realization of $-\Delta_g$. In one dimension, the finite-volume operator is typically written \[ \mathcal H_L=-\frac{d^2}{dx^2}+\xi(x) \] on an interval with Dirichlet or Neumann boundary conditions. On compact two-dimensional manifolds and on boxes in dimensions $d\le 3$, the same formal expression is used, but its rigorous meaning depends on renormalization. On $\mathbf R^2$ and $\mathbf R^3$, the operator is constructed as an infinite-volume limit of renormalized mollified operators or via the semigroup generated by the parabolic Anderson model [2102.09316]. | Setting | Formal operator | Rigorous feature | |---|---|---| | Interval in $d=1$ | $\mathcal H_L=-d^2/dx^2+\xi$ | self-adjoint, discrete simple spectrum | | Compact 2D manifold | $H=L+\xi$ | paracontrolled renormalization | | Cube $(-L,L)^d$, $d\le 3$ | $H=-\Delta+\xi$ | self-adjoint operator with pure point spectrum | | Full space $\mathbf R^2,\mathbf R^3$ | $H=-\Delta+\xi(x)$ | semigroup construction; spectrum $\mathbf R$ | In the one-dimensional white-noise case, Fukushima’s result yields almost sure self-adjointness with discrete simple spectrum \[ \lambda_1<\lambda_2<\cdots \] and $L^2$-normalized eigenfunctions $\varphi_i$. In the large-box limit one also introduces the empirical density of states \[ N_L(E)=\#\{\lambda_i\le E\},\qquad n(E)=\lim_{L\to\infty}\frac{N_L(E)}{L}, \] with $n(E)$ deterministic and smooth and satisfying $n(E)\sim (2\pi\sqrt E)^{-1}$ as $E\to+\infty$ [2102.09316]. A distinct branch of the literature uses “Anderson Hamiltonian” for random operators with nonsingular random potentials. One example is \[ H_0(\omega)=-\Delta+h\,V(x,\omega), \] where $V$ is a Bernoulli-type random potential built from i.i.d. on–off variables on unit cubes; in that setting the almost sure spectrum is purely essential and equals $[0,\infty)$ [1002.4220]. This broader usage matters because several asymptotic and perturbative results concern that classical random-potential model rather than the singular white-noise operator. ## 2. Renormalization, domains, and self-adjoint realizations In dimensions two and three, white noise is too irregular for the product $\xi f$ to be classically defined. On a two-dimensional manifold, white noise belongs almost surely to $\mathcal C^{-1-\kappa}(M)$ for every $\kappa>0$, and the renormalized operator is constructed from mollifications \[ \xi_\varepsilon=e^{-\varepsilon L}\xi,\qquad X_\varepsilon=-L^{-1}\xi_\varepsilon, \] together with the counterterm \[ C_\varepsilon(x)=\mathbb E[\Pi(X_\varepsilon,\xi_\varepsilon)(x)]. \] The renormalized operators \[ H_\varepsilon=L+\xi_\varepsilon-C_\varepsilon \] converge in resolvent norm to a limit $H$, and the resulting operator is closed, symmetric, essentially self-adjoint, and has pure point spectrum [2009.03549]. On the two-dimensional torus, the construction uses an enhanced noise $\Xi=(\Xi_1,\Xi_2)$ and paracontrolled distributions. The strong paracontrolled domain is \[ D_\Xi=\{f\in H^{\alpha+2}: f-f\prec \sigma(D)\xi-B(f,\Xi)\in H^2\}, \] and for mollified white noise $\xi_\varepsilon$ the renormalized smooth operators \[ H_\varepsilon=-\Delta+\xi_\varepsilon+c_\varepsilon \] satisfy \[ c_\varepsilon=\frac{1}{2\pi}\log\frac1\varepsilon+O(1), \] with resolvent convergence to the singular operator $H(\Xi^{wn})$ [1511.02718]. On periodic domains in dimensions $d=2,3$, the renormalized operator is obtained as a norm-resolvent limit of \[ H_\varepsilon=-\Delta+\xi_\varepsilon-c_\varepsilon \] in two dimensions, or \[ H_\varepsilon=-\Delta+\xi_\varepsilon-(c_\varepsilon^{(1)}+c_\varepsilon^{(2)}) \] in three dimensions, where \[ c_\varepsilon\sim \log(1/\varepsilon),\qquad c_\varepsilon^{(1)}\sim \varepsilon^{-1},\qquad c_\varepsilon^{(2)}\sim (\log 1/\varepsilon)^2. \] The operator domain admits a paracontrolled parametrization \[ D(H)=\Gamma(H^2), \] and the form domain is \[ Q(H)=\Gamma(H^1). \] This identifies the random domain with a deterministic Sobolev space up to the random isomorphism $\Gamma$ [1807.06825]. A parallel construction via regularity structures works on cubes $(-L,L)^d$ for every $d\le 3$. There the resolvent $(H+a)^{-1}$ is first constructed in a modelled-distribution space, and $H$ is then defined by \[ H=(G_a)^{-1}-a. \] For $d\ge 2$, the renormalization constants diverge; in dimension two \[ C_\varepsilon\sim -(2\pi)^{-1}\ln \varepsilon+O(1), \] while in dimension three \[ C_\varepsilon\sim \mathrm{const}/\varepsilon+\mathrm{const}\cdot\ln\varepsilon+O(1). \] The outcome is again a self-adjoint operator with pure point spectrum [1809.03718]. ## 3. One-dimensional spectral phases: localization, crossover, and delocalization The one-dimensional white-noise Anderson Hamiltonian displays several distinct spectral phases depending on the energy scale. At the bottom of the spectrum, for $\mathcal H_L$ on $[0,L]$, one introduces \[ a_L=\Bigl(\tfrac{3}{8}\ln L\Bigr)^{2/3}+O((\ln L)^{-1/3}),\qquad s_L=4\sqrt{a_L},\qquad m_L=-a_L, \] and considers the point process \[ Q_L=\sum_{k\ge1}\delta_{\,4\sqrt{a_L}(\lambda_k+a_L)}. \] As $L\to\infty$, $Q_L$ converges to a Poisson point process on $\mathbb R$ with intensity $e^x\,dx$. Jointly, the eigenfunction mass measures converge to Dirac masses located at i.i.d. uniform points, and the recentered, rescaled eigenfunction profile converges uniformly on compacts to \[ \psi(y)=\frac1{\cosh(y)}. \] The Dirichlet and Neumann problems couple to the same limit [1711.04700]. Away from the bottom, the relevant energy windows split into a bulk regime and a crossover regime. In the bulk, $E$ is fixed as $L\to\infty$; in the crossover regime, \[ 1\ll E\ll L. \] For each eigenfunction one defines its center of mass \[ U_i=\int_{-L/2}^{L/2}x\,\varphi_i(x)^2\,dx \] and the point measure \[ \Xi_L=\sum_{i\ge1}\delta_{\bigl(L\,n(E)(\lambda_i-E),\,U_i/L\bigr)}. \] In either regime, $\Xi_L$ converges in law to a homogeneous Poisson point process on $\mathbb R\times[-\tfrac12,\tfrac12]$ with intensity $d\alpha\times du$. Equivalently, eigenvalue gaps near $E$ are of order $(L\,n(E))^{-1}$ and the centers $U_i$ are asymptotically uniform and independent [2102.09316]. The associated eigenfunctions are exponentially localized. In the bulk, \[ |\varphi_i(x)|+|\varphi_i'(x)|\le c_i\exp\!\Bigl(-\tfrac{\nu_E-\varepsilon}{2}|x-U_i|\Bigr), \] with explicit rate \[ \nu_E=\frac{\int_0^\infty \sqrt u\,e^{-2Eu-u^3/6}\,du}{\int_0^\infty u^{-1/2}\,e^{-2Eu-u^3/6}\,du}>0. \] In the crossover regime, \[ |\varphi_i(x)|+\frac1{\sqrt E}|\varphi_i'(x)|\le \frac{c_i}{\sqrt E}\exp\!\Bigl(-\tfrac{1-\varepsilon}{2E}|x-U_i|\Bigr). \] Thus the localization length is $O(1)$ in the bulk and $O(E)$ in the crossover. Beyond decay, the recentered shape laws also converge: in the crossover regime the universal limiting profile is \[ Y_\infty(t)=\exp\!\Bigl(-\frac{|t|}{8}+\frac{B(t)}{2\sqrt2}\Bigr), \] a truncated exponential of two-sided Brownian motion with random shift [2102.09316]. These localized regimes do not exhaust the one-dimensional spectrum. At the critical energy scale \[ E(L)\sim \frac{L}{\tau}, \] the recentred and rotated operator converges in law to the random Dirac-type operator $\mathtt{CS}_\tau$, whose spectrum is the $\Sch_\tau$ point process. This is the delocalized phase. At still higher energies, the operator asymptotically matches the unperturbed Laplacian and one recovers the deterministic picket-fence spectrum. Combined with the localization-crossover results, this yields a complete description of the transition between localized and delocalized phases in one dimension [2102.05393]. A common misconception is to identify “pure point spectrum in finite volume” with localization in the infinite-volume sense. On compact tori, the spectrum is always purely discrete, but there is “no concept of extended versus localized states in the infinite-volume sense” [1807.06825]. The one-dimensional phase diagram shows that localization versus delocalization is instead an energy-scale-dependent asymptotic statement. ## 4. Counting functions, Weyl laws, and spectral geometry in two dimensions For compact two-dimensional manifolds, the Anderson Hamiltonian has the same first-order Weyl asymptotics as the smooth Laplace–Beltrami operator. If $\mu_n$ denotes the eigenvalues of $L=-\Delta_g$ and $\lambda_n$ those of the renormalized Anderson Hamiltonian, then \[ \lim_{n\to\infty}\frac{\lambda_n}{\mu_n}=1, \] and the eigenvalue counting function satisfies \[ N_H(\Lambda)=\frac{\Vol(M)}{4\pi}\Lambda+o(\Lambda),\qquad \Lambda\to\infty. \] The leading Weyl constant is purely geometric and independent of the noise [2009.03549]. On the square box $[0,L]^2$ with Dirichlet boundary conditions, the large-volume asymptotics are instead logarithmic. For each fixed $n$, \[ \lim_{L\to\infty}\frac{\lambda_n(L)}{\log L}=\chi \] almost surely, where the deterministic constant $\chi>0$ is the same for all fixed $n$ and has the variational representation \[ \chi=4\,\sup_{\substack{\psi\in H_0^1(\mathbb R^2)\\|\psi|_{L^2}=1}}\Bigl\{\|\psi\|_{L^4}^2-\|\nabla\psi\|_{L^2}^2\Bigr\}. \] This links the large-box spectral asymptotics directly to the sharp Gagliardo–Nirenberg constant [1907.01352]. The integrated density of states (IDS) exists for the two-dimensional white-noise Anderson Hamiltonian. Writing $Q_L=[-L/2,L/2]^2$ and \[ N_L(\lambda)=L^{-2}\#\{n:\lambda_{n,L}\le \lambda\}, \] the corresponding counting measures converge vaguely, almost surely, to a deterministic limit measure $\mu$. Its left tail obeys the Lifshitz asymptotic \[ \lim_{\lambda\to-\infty}\frac1{|\lambda|}\log N(\lambda)=-\kappa^{-1}, \] and in two dimensions $\kappa=4\pi$, so \[ N(\lambda)\approx \exp[-| \lambda |/(4\pi)+o(|\lambda|)]\qquad (\lambda\to-\infty). \] The same analysis yields a sharp moment-explosion threshold for the planar parabolic Anderson model: \[ t_c=\kappa^{-1}=\frac1{4\pi}. \] Its first moment is finite for $t\in(0,1/(4\pi))$ and infinite for $t>1/(4\pi)$ [2011.09180]. Recent spectral-geometry results refine the Weyl-law viewpoint on bounded planar domains. For \[ H_\omega=-\tfrac12\Delta+\sigma \xi \] with Dirichlet boundary condition on a bounded planar domain $D$, the exponential trace \[ T_\omega(t)=\operatorname{Tr}(e^{-tH_\omega})=\sum_{n\ge1}e^{-t\lambda_n(\omega)} \] satisfies \[ E[T_\omega(t)]=T_0(t)+\frac{\sigma^2|D|}{4\pi^2}\log t+o(\log t),\qquad \operatorname{Var}[T_\omega(t)]=O(1), \] while the parabolic Anderson mass obeys \[ E[M_\omega(t)]=M_0(t)+\frac{\sigma^2|D|}{2\pi}t\log t+o(t\log t),\qquad \operatorname{Var}[M_\omega(t)]=O(t^2). \] These asymptotics imply that, almost surely, one can recover $|D|$, $|\partial D|$, and $\sigma^2$ from a single observation of the AH spectrum. If $\partial D$ is fractal and has Minkowski dimension $d_M\in(1,2)$, then $d_M$ can be recovered from the small-time asymptotics of the parabolic Anderson mass [2507.06186]. ## 5. Infinite-volume operators and nonlinear evolution equations The full-space Anderson Hamiltonian in dimensions $d=2,3$ can be constructed from the parabolic Anderson model rather than from a direct resolvent analysis. Starting from mollified white noise $\xi_\varepsilon=\xi*\varrho_\varepsilon$, one considers \[ H_\varepsilon=-\Delta+\xi_\varepsilon+C_\varepsilon, \] with \[ C_\varepsilon\sim (2\pi)^{-1}\log \varepsilon^{-1}\quad \text{in }d=2,\qquad C_\varepsilon\sim c_\varrho \varepsilon^{-1}+\ln \varepsilon^{-1}+O(1)\quad \text{in }d=3. \] The semigroup solving the renormalized parabolic Anderson model is shown to satisfy the Klein–Landau hypotheses, which yields a unique self-adjoint generator $H$. The mollified operators converge to $H$ in the strong resolvent sense, and almost surely \[ \operatorname{Spec}(H)=\mathbb R. \] The proof of the spectral statement extends Kotani’s method to singular random operators [2401.17900]. A full-space paracontrolled realization in dimension two also supports nonlinear dispersive and hyperbolic dynamics. In that setting one constructs a self-adjoint semibounded operator $A$ with \[ D(A)=\Gamma(H^2(\mathbb R^2)) \] for a bounded invertible map $\Gamma:H^2(\mathbb R^2)\to D(A)$, proves norm-resolvent convergence of the regularized operators, and then studies the stochastic nonlinear Schrödinger and wave equations whose linear part is the Anderson Hamiltonian. The resulting solutions are obtained as limits of regularized equations, with conserved energy and strong-solution formulations in spaces built from the form domain $D(\sqrt{-A})$ [2208.09352]. On periodic domains in dimensions two and three, the same renormalized operator underlies semilinear Schrödinger and wave equations. Norm-resolvent convergence of $H_\varepsilon$ to $H$ implies strong convergence of the unitary groups $e^{-itH_\varepsilon}\to e^{-itH}$, convergence of bounded continuous functional calculi, and convergence of solutions to linear and semilinear equations. This places the Anderson Hamiltonian within a broader PDE framework in which spectral renormalization and nonlinear wellposedness are tightly coupled [1807.06825]. ## 6. Perturbations, critical regimes, and nomenclature For nonsingular random potentials, non-random perturbations reveal an additional spectral threshold phenomenon. If \[ H_0(\omega)=-\Delta+hV(x,\omega) \] with Bernoulli-type random potential $V$, and \[ H(\omega)=H_0(\omega)-w(x),\qquad w(x)\ge 0,\quad w(x)\to 0, \] then in $\mathbb R^d$ there exist constants $c_1(1+\delta)\frac{C_p}{\ln^2(|x|+1)}\Rightarrow N_-(H)=\infty \] almost surely [1205.1038]. A different phase transition arises when white noise is approximated by smooth Gaussian fields $\xi_{\varepsilon(t)}$ on boxes $Q_t=(-t,t)^d$. For the Dirichlet eigenvalues of \[ H_\varepsilon=-\tfrac12\Delta+\xi_{\varepsilon(t)}, \] there is a regular phase \[ \varepsilon(t)\gg (\log t)^{-1/(4-d)} \] with \[ \frac{\lambda_k(H_{\varepsilon(t)},Q_t)}{\varepsilon(t)^{-d/2}\sqrt{\log t}}\xrightarrow{\rm prob.}\sqrt{2d\,R(0)}, \] and a singular phase \[ \varepsilon(t)\ll (\log t)^{-1/(4-d)} \] in which \[ \frac{\lambda_k(H_{\varepsilon(t)},Q_t)}{(\log t)^{2/(4-d)}}\xrightarrow{\rm prob.}L_d. \] In dimensions $d=1,2$, the singular-phase eigenvalues quantitatively approximate those of the renormalized white-noise Anderson Hamiltonian [2008.08116]. At the critical dimension $d=4$, the weakly coupled Anderson Hamiltonian on $\mathbb T^4$, \[ H_\lambda=-\Delta+\lambda\,\xi(x), \] admits a perturbative fluctuation theory under the critical scaling \[ \lambda_\varepsilon=\hat\lambda(\log\tfrac1\varepsilon)^{-1/2}. \] After BPHZ renormalisation, one obtains Gaussian fluctuations around the Green’s function of the Laplacian, with effective variance \[ \sigma_{\mathrm{eff}}^2(\hat\lambda)=\frac{2\pi^2}{2\pi^2-\hat\lambda^2} \] for $|\hat\lambda|<\sqrt2\,\pi$. The analysis identifies primitive blow-ups as the contributing diagrams and proves that no Laplacian renormalisation is present [2602.22509]. The nomenclature “Anderson Hamiltonian” is also a source of confusion. The Newns–Anderson Hamiltonian, \[ H_{NA}=\epsilon_a d^\dagger d+\sum_k\epsilon_k c_k^\dagger c_k+\sum_k(V_k d^\dagger c_k+V_k^* c_k^\dagger d), \] describes adsorption at gas-solid interfaces and is constructed from Kohn–Sham DFT data in first-principles studies of chemisorbed hydrogen [2602.17635]. The periodic Anderson Hamiltonian, used for systems such as samarium hexaboride, is a lattice model with conduction $d$ orbitals, localized $f$ orbitals, hybridization, and on-site Coulomb repulsion $U$ [2311.00583]. These are distinct from the continuous white-noise Anderson Hamiltonian $-\Delta+\xi$ and from the Bernoulli random-potential operators discussed above. Source: https://www.emergentmind.com/topics/anderson-hamiltonian-ah