Why Rare Resonances Make Many-Body Localization Stranger Than Anyone's Averages Suggest

A microscopic theory shows typical and average Fock-space correlations in MBL eigenstates behave qualitatively differently — the geometry of exponentially large Hilbert space amplifies rare events into dominant averages.

Imagine you're measuring commute times across a city. Most people drive 30 minutes, but a handful hit 4-hour gridlock detours. If you report the median, the city looks fine. If you report the mean, the city looks broken. Now imagine the city has an exponentially growing number of roads as you add intersections — suddenly those rare 4-hour detours, even if they're one-in-a-million per road, show up on enough roads to dominate the average. That's the core mechanism here: Fock space is the exponentially large city, many-body resonances are the rare gridlock detours, and the paper shows exactly how the geometry of that space makes typical and average correlation functions tell fundamentally different stories. The committed claim: for many-body localised (MBL) eigenstates on Fock space, typical two-point correlations decay as a clean exponential with a well-defined correlation length, while average correlations decay subexponentially with systematic system-size dependence — and this qualitative split is not a finite-size artifact but a structural consequence of hypercubic Fock-space geometry. The authors develop an analytical framework centred on the distribution of Fock-space correlation lengths, showing its variance scales as 1/L (inverse system size), and validate it against exact diagonalisation (ED) of the disordered tilted-field Ising chain. The ladder context matters. MBL has been a contested topic since the original Basko-Aleiner-Altshuler (2006) prediction and the Anderson localisation analogy on Fock space developed by Logan, Welsh, and others. Recent large-scale numerics (Šuntajs et al. 2020, Sels & Polkovnikov 2021) challenged whether MBL survives in the thermodynamic limit, arguing finite-size effects fool ED studies. This paper doesn't resolve that existential fight directly — it works within the MBL phase at strong disorder — but it provides a microscopic theory for why average quantities (which most numerics report) behave so differently from typical ones. That distinction is load-bearing for interpreting every numerical MBL study. Architecturally, this is an analytical + exact diagonalisation hybrid. The theoretical backbone is a probabilistic framework for Fock-space propagators: self-energies, two-point correlation functions, and their distributions parameterised by correlation lengths ξF. The numerics use full ED of the tilted-field Ising chain — meaning exact quantum eigenstates, no approximations, but limited to L ≤ ~18 sites (Fock-space dimension 2^L ≤ ~260,000). The key structural insight is that on a d-dimensional hypercube with d = L, the number of sites at Hamming distance r from any reference grows combinatorially as C(L, r), and even exponentially rare long-ξF tails in the correlation-length distribution get amplified by this combinatorial weight when you compute averages. Integrity is solid within the standard constraints of the subfield. ED is exact — no variational approximations, no DMRG truncation, no stochastic noise — but it hits hard system-size walls. The authors are upfront about this: the 1/L scaling of variance is demonstrated numerically across accessible sizes and supported analytically, but extrapolation to L → ∞ is inherently uncertain. The comparison is self-consistent (theory vs. own numerics) rather than against independent experiments, which is typical for this subfield since no clean experimental MBL system with tuneable Fock-space correlations exists yet. They also derive the singularity spectrum f(α) of MBL eigenstates and the anomalous symmetry violation of quantum multifractality as consequences — providing multiple internal consistency checks rather than a single cherry-picked metric. The milestone question is subtle. This isn't a race to a bigger qubit count; it's a conceptual advance. The next concrete thing to watch is whether the typical-vs-average distinction the paper quantifies can resolve the MBL existence debate. If someone applies this framework to the critical regime (near the putative MBL transition) and shows that average-based numerics systematically overestimate delocalisation, that's a major result. Concretely, extending this analysis from L ~18 to L ~24-30 (via shift-invert ED or tensor-network methods) would test whether the 1/L variance scaling holds or breaks down. The obvious experiment not run: applying this Fock-space correlation framework to the MBL transition region itself, rather than deep in the localised phase. The authors stay at strong disorder where MBL is unambiguous. The honest read is (c) — they're building the theoretical machinery in the regime where it can be validated cleanly, and the transition-region application is the next paper. It's a reasonable strategy, but it means the paper's impact on the big MBL-existence fight is indirect for now.