Controlled analytic evaluation of the susceptibility-kernel eigenvalue $\lambda(s)$ for the Anderson transition on the Bethe lattice: expansion about $s=0$, retained contributions at $s=1/2$ in the large-connectivity limit, and the weak-disorder constant $\kappa$ in $D_1=1-\kappa W^4$

The Anderson transition on the Bethe lattice (infinite regular tree) can be characterized through the statistics of real-energy cavity propagators: summing the susceptibility along the $K^L$ paths of a tree of depth $L$, the transition occurs when $K\lambda(1/2)=1$, where $\lambda(s)$ is the largest eigenvalue of an integral kernel $K_s$ acting on the distribution of the propagators. Parisi, Pascazio, Pietracaprina, Ros and Scardicchio (J. Phys. A 53, 014003 (2020)) solved this kernel exactly at $s=0$ for Cauchy disorder (finding a gapped spectrum and explicit eigenstates), proved the weak-disorder law $D_1=1-\kappa W^4+o(W^4)$ for the tree fractal dimension with $\kappa$ left only numerically evaluated ($\kappa=(5.0\pm0.6)\times10^{-5}$ at connectivity $K=2$, box disorder), and showed that the forward approximation fails in the large-disorder/large-connectivity limit because it neglects self-energy-induced correlations between consecutive propagators. They explicitly left as future work the systematic perturbative expansion of $\lambda(s)$ about $s=0$, the identification of which contributions of that expansion must be retained at large disorder/connectivity to recover the correct $s=1/2$ asymptotics (equivalently the critical disorder $W_c\sim4K\ln K$ instead of the forward-approximation $2eK\ln K$), and the analytic computation of the constants $c$ and $\kappa$. The problem remains open.

Background

The Anderson model is the simplest description of a quantum particle in a disordered crystal: a tight-binding Hamiltonian with random on-site energies $\epsilon_i$ drawn from a distribution of width $W$ (e.g. uniform on $[-W/2, W/2]$, the 'box' disorder, or a Cauchy/Lorentz distribution). On sufficiently disordered or low-dimensional lattices its eigenfunctions are localized (wavefunction weight concentrated on finitely many sites) rather than extended, with a localization-delocalization transition at a critical disorder $W_c$. The Bethe lattice is the infinite regular tree with branching number $K$ (each site has $K+1$ neighbors); it is the mean-field-like geometry on which the Anderson transition is most tractable, and it is locally indistinguishable from a random regular graph (RRG) - a finite $N$-vertex graph, fixed degree, whose loops all have length of order $\ln N$. The Bethe lattice and RRG also serve as the standard toy geometry for many-body localization (MBL), because the configuration (Fock) space of an interacting disordered system resembles a highly connected graph that is locally tree-like.

The workhorse of the subject is the self-consistent theory of Abou-Chacra, Thouless and Anderson (ACTA, 1973). Removing one link of the tree creates a 'cavity' sub-tree, and the corresponding diagonal resolvent entries ('cavity propagators' or cavity Green functions) satisfy a closed recursion: $G_i=1/(E-\epsilon_i-\sum_{k=1}^{K}G_k)$, where the $G_k$ are independent copies associated with the $K$ branches. In the standard approach one adds a small imaginary part $\mathrm{i}\eta$ to the energy; the modern numerical implementation is 'population dynamics' - iterating the recursion on a large population of propagators to obtain their stationary probability distribution $P(G)$. Parisi et al. (J. Phys. A 53, 014003 (2020)) instead work directly at real energies $E$ (relevant observables such as the local density of states are limits of real-axis quantities), which corresponds to taking $\eta\to0$ before $N\to\infty$, the 'inverted thermodynamic limit' under which the propagators are real random variables. Their central object is the susceptibility along a path of the tree, essentially the product of squared propagators $G_k^2$ along it; the total susceptibility of a site at depth $L$ is the sum of these contributions over the $K^L$ paths. The transition is the divergence of this sum with $L$, and its statistics are controlled by the Mellin transform in $s$ of the distribution of path susceptibilities, whose decay rate is set by the largest eigenvalue $\lambda(s)$ of an explicit integral kernel $K_s$ built from $P(G)$ and the disorder distribution. Because the recursion inverts under $G\to1/G$, the kernel obeys an exact symmetry $\lambda(s)=\lambda(1-s)$, and the transition criterion reduces to $K\lambda(1/2)=1$ - equivalent to the original ACTA stability criterion and to the 'resonant' delocalization criterion of Aizenman and Warzel. The typical single-path decay rate $\theta_{\mathrm{typ}}=-d\lambda/ds$ at $s=0$ governs the linear term of the small-$s$ expansion $\lambda(s)=1-s\theta_{\mathrm{typ}}+O(s^2)$.

Two exactly-solvable limits anchor the problem. (1) Cauchy disorder: the Cauchy distribution is stable under the cavity recursion, and at $s=0$ the paper computes the full spectrum and eigenstates of the kernel explicitly (Bessel/Laguerre-type functions), showing the spectrum is gapped - the gap implies exponentially decaying correlations between consecutive propagators and guarantees that the perturbative expansion of $\lambda(s)$ in $s$ about 0 is convergent. Moreover, as $W\to0$ the stationary distribution becomes Cauchy for any bare disorder, so the Cauchy kernel is the universal weak-disorder reference. (2) Large disorder and large connectivity: Anderson's original 1958 argument gives the asymptotic critical disorder $W_c=4K\ln K$ (box disorder), whereas the 'forward approximation' (FA, also called forward scattering approximation) - dropping the self-energy term $\sum_k G_k$ from the recursion denominators, which decouples consecutive propagators along a path - yields the different asymptotics $W_c=2eK\ln K$. The FA is the basis of many analytic treatments, including much of the MBL literature, so understanding precisely how the neglected self-energy correlations repair it is of broad interest. Parisi et al. show that at large disorder the leading eigenvector of $K_{1/2}$ approaches the FA distribution, and construct an interpolation between the FA and the full problem; what is missing is a systematic expansion that identifies, order by order, which classes of terms accumulate to move the FA answer to the correct one.

The weak-disorder regime connects $\lambda(s)$ to eigenfunction statistics. The information (fractal) dimension $D_1$ quantifies how the effectively occupied number of sites of an eigenfunction scales with system size ($D_1=1$ for ergodic extended states, $D_1<1$ for multifractal ones). Defined through tree (inverted-limit) quantities as in Kravtsov, Altshuler and Ioffe (Ann. Phys. 389, 148 (2018)), $D_1=\Lambda/\ln K$, where the exponent $\Lambda$ can be expressed as an extremum over $s$ of $(\ln K+\mu(s))/(1-s)$ with $\mu=\ln\lambda$. Expanding near $s=1$ and using the conjecture $\theta_{\mathrm{typ}}=\ln K+cW^2$ for disorder distributions falling faster than $1/\epsilon^2$, the paper derives $D_1=1-\kappa W^4+o(W^4)$ with $\kappa=c^2/(2\mu''(0,0)\ln K)$, where $\mu''$ is a second derivative at the point $(W,s)=(0,0)$. This recovers analytically the qualitative $W^4$ flatness found numerically by Kravtsov et al., but the constants $c$ and $\kappa$ are not evaluated: the paper only fits $c$ numerically ($c=(4.1\pm0.2)\times10^{-3}$ for box disorder) and quotes a numerical $\kappa=(5.0\pm0.6)\times10^{-5}$ at $K=2$, noting that $c$ 'can be computed using perturbation theory in $s$ and $W$ around the $W=0$, $s=0$ point' with the appendix eigenstates - a computation it leaves undone. Note the interpretive context, established by Tikhonov and Mirlin (Phys. Rev. B 94, 184203 (2016)) and confirmed by later work: the tree/inverted-limit $D_1<1$ does not mean the delocalized phase of the RRG is non-ergodic; on RRGs ergodicity is restored in the thermodynamic limit, with very strong finite-size crossover effects.

Problem Statement

Within the real-energy cavity (population-dynamics) framework of Parisi et al. (J. Phys. A 53, 014003 (2020)) for the Anderson model on the Bethe lattice of branching number $K$, develop the analytic control of $\lambda(s)$, the largest eigenvalue of the integral kernel $K_s$ that determines the $s$-moments of the distribution of cavity susceptibilities, along the lines the source paper explicitly leaves for future work. The common objective uniting the requested results is a controlled analytic evaluation of $\lambda(s)$ - beyond both the exactly solved $s=0$ Cauchy point and the forward approximation - from which the transition criterion $K\lambda(1/2)=1$, the large-connectivity asymptotics of the critical disorder, and the weak-disorder fractal dimension $D_1$ all follow. Specifically:

(1) Construct the systematic perturbative expansion of $\lambda(s)$ about $s=0$, $\lambda(s)=1-s\theta_{\mathrm{typ}}+O(s^2)$, giving explicit expressions for the coefficients of $s^n$ at general order $n$ and their combinatorial origin. The expansion is expected to be convergent because the spectrum of the $s=0$ kernel is gapped.

(2) Identify which classes of contributions in this expansion must be retained (or resummed) in the large-disorder/large-connectivity limit so that the expansion reproduces the correct asymptotic behavior of $\lambda(s)$ at $s=1/2$ - the point at which the transition criterion $K\lambda(1/2)=1$ holds - recovering the known critical disorder $W_c$ asymptotics ($W_c=4K\ln K$ plus lower-order corrections for box disorder, rather than the forward-approximation value $2eK\ln K$). The result should give indications on how to properly account for the self-energy corrections - the anti-correlations between consecutive propagators along a path that the forward approximation neglects - in path expansions for the localized phase on treelike and finite-dimensional lattices.

(3) Evaluate analytically the constants $c$ and $\kappa$ appearing in the weak-disorder laws $\theta_{\mathrm{typ}}=\ln K+cW^2$ (for disorder distributions falling faster than $1/\epsilon^2$ at large $\epsilon$) and $D_1=1-\kappa W^4+o(W^4)$, where $\kappa=c^2/(2\mu''(0,0)\ln K)$, and compare with the existing numerical benchmarks (in particular $\kappa=(5.0\pm0.6)\times10^{-5}$ for box disorder at $K=2$).

The three facets are scientifically linked aspects of one objective and are posed jointly. Two broader directions that the source paper also acknowledges - a fully analytic treatment of $\lambda(s)$ for general disorder distributions at arbitrary $s$ together with rigorous control of the saddle-point/large-deviation steps connecting $\lambda(s)$ to susceptibility statistics, and the extension of this integral-equation control to interacting many-body localization problems - remain open research directions; progress on them is relevant context but is not by itself an answer to this problem.

References

1. G. Parisi, S. Pascazio, F. Pietracaprina, V. Ros, and A. Scardicchio, 'Anderson transition on the Bethe lattice: an approach with real energies', J. Phys. A: Math. Theor. 53, 014003 (2020). DOI: 10.1088/1751-8121/ab56e8, https://doi.org/10.1088/1751-8121/ab56e8 (preprint arXiv:1812.03531, https://arxiv.org/abs/1812.03531) [source paper]
2. R. Abou-Chacra, D. J. Thouless, and P. W. Anderson, 'A selfconsistent theory of localization', J. Phys. C: Solid State Phys. 6, 1734 (1973). DOI: 10.1088/0022-3719/6/10/009, https://doi.org/10.1088/0022-3719/6/10/009
3. P. W. Anderson, 'Absence of diffusion in certain random lattices', Phys. Rev. 109, 1492 (1958). DOI: 10.1103/PhysRev.109.1492, https://doi.org/10.1103/PhysRev.109.1492
4. V. E. Kravtsov, B. L. Altshuler, and L. B. Ioffe, 'Non-ergodic delocalized phase in Anderson model on Bethe lattice and regular graph', Ann. Phys. (NY) 389, 148 (2018). DOI: 10.1016/j.aop.2017.12.009, https://doi.org/10.1016/j.aop.2017.12.009 [reference [44] of the source paper: origin of the D1 definition via the inverted thermodynamic limit]
5. K. S. Tikhonov and A. D. Mirlin, 'Fractality of wave functions on a Cayley tree: Difference between tree and locally treelike graph without boundary', Phys. Rev. B 94, 184203 (2016). DOI: 10.1103/PhysRevB.94.184203, https://doi.org/10.1103/PhysRevB.94.184203
6. P. A. Nosov, I. M. Khaymovich, A. Kudlis, and V. E. Kravtsov, 'Statistics of Green's functions on a disordered Cayley tree and the validity of forward scattering approximation', SciPost Phys. 12, 048 (2022). DOI: 10.21468/SciPostPhys.12.2.048, https://doi.org/10.21468/SciPostPhys.12.2.048 (preprint arXiv:2108.10326, https://arxiv.org/abs/2108.10326)
7. G. Biroli, A. K. Hartmann, and M. Tarzia, 'Critical behavior of the Anderson model on the Bethe lattice via a large-deviation approach', Phys. Rev. B 105, 094202 (2022). DOI: 10.1103/PhysRevB.105.094202, https://doi.org/10.1103/PhysRevB.105.094202 (preprint arXiv:2110.01524, https://arxiv.org/abs/2110.01524)
8. T. Rizzo and M. Tarzia, 'Localized phase of the Anderson model on the Bethe lattice', Phys. Rev. B 110, 184210 (2024). DOI: 10.1103/PhysRevB.110.184210, https://doi.org/10.1103/PhysRevB.110.184210 (preprint arXiv:2406.18748, https://arxiv.org/abs/2406.18748)
9. A. Aggarwal and P. Lopatto, 'Mobility Edge for the Anderson Model on the Bethe Lattice', arXiv:2503.08949 (2025), https://arxiv.org/abs/2503.08949
10. K. S. Tikhonov and A. D. Mirlin, 'From Anderson localization on random regular graphs to many-body localization', Ann. Phys. (NY) 435, 168525 (2021). DOI: 10.1016/j.aop.2021.168525, https://doi.org/10.1016/j.aop.2021.168525 (preprint arXiv:2102.05930, https://arxiv.org/abs/2102.05930)
11. G. Biroli and M. Tarzia, 'Delocalization and ergodicity of the Anderson model on Bethe lattices', arXiv:1810.07545 (2018), https://arxiv.org/abs/1810.07545 [preprint; reference [67] of the source paper]
12. I. García-Mata, J. Martin, O. Giraud, B. Georgeot, R. Dubertrand, and G. Lemarié, 'Critical properties of the Anderson transition on random graphs: two-parameter scaling theory, Kosterlitz-Thouless type flow, and many-body localization', Phys. Rev. B 106, 214202 (2022). DOI: 10.1103/PhysRevB.106.214202, https://doi.org/10.1103/PhysRevB.106.214202 (preprint arXiv:2209.04337, https://arxiv.org/abs/2209.04337)
13. C. Vanoni, B. L. Altshuler, V. E. Kravtsov, and A. Scardicchio, 'Renormalization group analysis of the Anderson model on random regular graphs', PNAS 121, e2401955121 (2024). DOI: 10.1073/pnas.2401955121, https://doi.org/10.1073/pnas.2401955121 (preprint arXiv:2306.14965, https://arxiv.org/abs/2306.14965)
14. M. Baroni, G. Garcia Lorenzana, T. Rizzo, and M. Tarzia, 'Corrections to the Bethe lattice solution of Anderson localization', Phys. Rev. B 109, 174216 (2024). DOI: 10.1103/PhysRevB.109.174216, https://doi.org/10.1103/PhysRevB.109.174216 (preprint arXiv:2304.10365, https://arxiv.org/abs/2304.10365)
15. G. Biroli, A. K. Hartmann, and M. Tarzia, 'Large-deviation analysis of rare resonances for the many-body localization transition', Phys. Rev. B 110, 014205 (2024). DOI: 10.1103/PhysRevB.110.014205, https://doi.org/10.1103/PhysRevB.110.014205 (preprint arXiv:2312.14873, https://arxiv.org/abs/2312.14873)
16. D. Prado Bandeira and M. Tarzia, 'Anderson localization on Husimi trees and its implications for many-body localization', Phys. Rev. B 113, 144207 (2026). DOI: 10.1103/q3bz-97v6, https://doi.org/10.1103/q3bz-97v6 (preprint arXiv:2601.04155, https://arxiv.org/abs/2601.04155)
17. A. Östlin, I. Titvinidze, M. Kollar, D. Jones, and L. Chioncel, 'Analytical results of Typical Medium Theory for the Bethe lattice with Cauchy disorder', arXiv:2608.03770 (2026), https://arxiv.org/abs/2608.03770
18. S. Pascazio, A. Scardicchio, and M. Tarzia, 'Anderson Localization on the Bethe Lattice', in 'Spin Glass Theory and Far Beyond' (World Scientific, 2023), pp. 335-352. DOI: 10.1142/9789811273926_0017, https://doi.org/10.1142/9789811273926_0017
19. K. S. Tikhonov and A. D. Mirlin, 'Critical behavior at the localization transition on random regular graphs', Phys. Rev. B 99, 214202 (2019). DOI: 10.1103/PhysRevB.99.214202, https://doi.org/10.1103/PhysRevB.99.214202