Machine Learning the Strong Disorder Renormalization Group Method for Disordered Quantum Spin Chains

A. Ustyuzhanin, J. Vahedi, S. Kettemann · 2026

Read Machine Learning the Strong Disorder Renormalization Group Method for Disordered Quantum Spin Chains by A. Ustyuzhanin, J. Vahedi, S. Kettemann (2026) 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-2603-05164.txt.

Introduction

A. Ustyuzhanin and team train machine learning models to infer the entanglement structure of disordered quantum spin chains from the strong disorder renormalization group method. A graph neural network on the interaction graph learns a bond-ranking rule like the decimation policy, beating a Random Forest older model. It reaches near-perfect pairing accuracy on disorder average and reproduces entropy across subsystem sizes and interaction exponents. Finite-temperature acts come out with no retraining by feeding the zero-temperature flow into a two-stage thermal sampling scheme.

Catalog record

CREATORA. Ustyuzhanin, J. Vahedi, S. Kettemann
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35178
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.05164

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Machine Learning the Strong Disorder Renormalization Group Method for Disordered Quantum Spin Chains A. Ustyuzhanin, J. Vahedi, S. Kettemann Abstract We train machine learning algorithms to infer the entanglement structure of disordered long-range interacting quantum spin chains by learning from the strong disorder renormalization group (SDRG) method. The system consists of S = 1 / 2 S=1/2 spins coupled by antiferromagnetic power-law interactions with decay exponent α \alpha at random positions on a one-dimensional chain. Using SDRG as a physics-informed teacher, we compare a Random Forest classifier as a classical baseline with a graph neural network (GNN) that operates directly on the interaction graph and learns a bond-ranking rule mirroring the SDRG decimation policy. The GNN achieves a disorder-averaged pairing accuracy of r P ≃ 0.94 r_{P}\simeq 0.94 and reproduces the entanglement entropy S ⁡ ( ℓ ) S(\ell) in excellent quantitative agreement with SDRG across all subsystem sizes and interaction exponents. RG-flow heatmaps confirm that the GNN learns the sequential decimation hierarchy rather than merely fitting final-state observables. Finite-temperature entanglement properties are incorporated via the SDRG-X framework through a two-stage strategy—using the zero-temperature GNN to generate the RG flow and sampling thermal occupations from the canonical ensemble—yielding results in agreement with both numerical SDRG-X and analytical predictions without retraining. I Introduction Magnetic properties of a wide range of materials are due to randomly placed quantum spins coupled by long-range interactions anderson58 ; loehneysen ; Kettemann2023 ; Kettemann2024 ; reviewtls ; Yu . Tunable interactions have been realized in cold atom systems grass ; Islam2013 ; Richerme2014 ; Signoles2021 ; Brow2020 . The dynamics of disordered spin ensembles at a diamond surface can be studied with single nitrogen-vacancy centers Lukin2022 ; Davis2022 . Interacting spin models at finite temperature in a disordered landscape were simulated in arrays of superconducting qubits, tracking dynamics both in real space and in Hilbert space Altman2021 ; Lunkin2026 . However, it remains a challenge to model and derive thermodynamic, dynamical, and entanglement properties of such systems. The strong disorder renormalization group (SDRG) method was introduced to study disordered quantum spin chain models bhattlee81 ; bhattlee82 ; fisher94 ; fisher95 ; monthus ; Igloi2018 . Short-range antiferromagnetically coupled disordered spin chains were thereby found to be governed in the ground state by an infinite-randomness fixed point (IRFP), whose ground state is a product of randomly positioned singlet pairs (the random singlet phase). The SDRG method has been applied to other random quantum spin chains, such as the transverse-field Ising model monthus , and it allows one to derive thermodynamic properties monthus as well as dynamical properties including entanglement dynamics Vosk2013 ; Igloi2012 . Recently, we applied SDRG to disordered S=1/2 spin chains with long-range interactions Moure2015 ; Juhasz2014 ; Juhasz2016 ; Moure2018 ; Mohdeb2020 ; Kettemann2025 , and compared with numerical exact diagonalization and long-range extensions of DMRG Mohdeb2020 . The SDRG method was extended to study excited states and finite-temperature properties Pekker2014 ; Huang2014 ; Aram2024 and has recently been applied to study excitations Mohdeb2022 ; Kettemann20252 and entanglement dynamics after global quantum quenches of long-range coupled spin chains Mohdeb2023 . In parallel, machine learning methods have become increasingly important in quantum many-body physics Hou23 ; MehtaReview ; RMPReview . Artificial neural networks have been used for variational representations of many-body wavefunctions (neural quantum states, NQS) Carleo2017 ; Deng2017 ; nomura3 ; choo4 ; choo11 ; Rende2025 ; Viteritti2025 ; roca1 ; sharir2 , for a recent review see Ref. Lange2024 . Beyond variational wavefunctions, machine learning has also been used to learn or accelerate renormalization-group concepts Sheshmani23 ; Mehta ; Fan23 and multiscale transformations, including (i) unsupervised learning of RG transformations and RG equations HouYou2023MLRG , (ii) learning RG-like operator maps targeting a desired property rather than a state ansatz Luo2024OLRG , (iii) learning scale-dependent flow dynamics in functional RG DiSante2022DLfRG , and (iv) learning multiscale, scale-by-scale conditional models with explicit RG structure Marchand2022WCRG ; ZhaoFoglerYou2025RGFlow . Information-theoretic perspectives further support the view of RG as multiscale compression, including results for disordered systems and inhomogeneous geometries Lenggenhager2020OptRG ; Gokmen2021RSMI ; Gokmen2024Compression ; Erdmenger2022RelativeEntropyRG . Machine learning has also been applied directly to disordered quantum matter, including localization physics and disorder-averaged inference Ohtsuki2016 ; Beetar2021MBLGraphs ; Chen2024MobilityEdgesCNN . Recent work has demonstrated that entanglement-related quantities can be inferred by machine learning without a full reconstruction of the many-body wavefunction, including GNN-based prediction of von Neumann entanglement entropy from accessible data Saleh2025GNNEntropy , direct machine-learning-based entanglement detection in noise-relevant settings Huang2025DirectEntanglement and quantum information scrambling Passetti2024DeepLearningScrambling . Representation learning and transfer learning can further reduce the cost of learning entanglement from more accessible observables Schmidt2024TransferEntropy . In this article, we ask whether machine learning models can be trained to infer entanglement properties of disordered long-range interacting quantum spin systems for a given disorder realization by learning from SDRG as a physics-informed teacher. Our aim is not only to reproduce a local decimation criterion, but to approximate the strongly interacting, nonlocal renormalization flow generated by SDRG in long-range coupled systems, where each decimation step reshapes effective couplings across many length scales. Motivated by the success of graph-based policy learning in sequential reduction problems Meirom2022RLContraction , we develop graph-based learning algorithms that take the disorder realization as an irregular weighted interaction graph and predict entanglement properties without explicit diagonalization. Model. — We consider antiferromagnetic long-range XX-coupled chains of N S=1/2 quantum spins, placed at random positions {\bf r}_{i} on a line of length L , defined by H=\sum_{i<j}J_{ij}\left(S_{i}^{x}S_{j}^{x}+S_{i}^{y}S_{j}^{y}\right). (1) We take N even with open boundary conditions. The couplings are antiferromagnetic and decay with a power law in distance r_{ij}=|{\bf r}_{i}-{\bf r}_{j}| with exponent \alpha , J_{ij}=J_{0}\left|({\bf r}_{i}-{\bf r}_{j})/a\right|^{-\alpha}, (2) with J_{0}>0 and |{\bf r}_{i}-{\bf r}_{j}|>a for all i,j , where a\ll L/N is the smallest possible distance. II SDRG-X Method We apply the SDRG-X method, as introduced in Ref. Pekker2014 . Assuming that many body Eigen states of the Hamiltonian Eq. ( 1 ) can be written as tensor products of pair states, one identifies the strongest coupled pair of spins (i,j) for a given random positioning of the N spins on the L lattice sites, yielding with the couplings Eq. ( 2 ) an initial distribution of couplings P(J,\Omega_{0}), where \Omega_{0} is that largest energy scale. We rewrite the Hamiltonian Eq. ( 1 ) as \hat{H}=\hat{H}_{0}+\hat{V}, where \hat{H}_{0} is the Hamiltonian of the most strongly coupled pair of spins (i,j), given by \hat{H}_{0}=J^{x}_{ij}\left(S_{i}^{x}\,S_{j}^{x}+S_{i}^{y}\,S_{j}^{y}\right) and \hat{V} is the part of the Hamiltonian modeling the interaction of spins (i,j) with all other N-2 spins and between them. \hat{H}_{0}, has four eigenstates |s\rangle, s\in\{0,1,2,3\} with eigenenergies E_{s}. Projecting the pair (i,j) on one of these pair states s, one gets the effective Hamiltionian of the remaining N-2 spins \hat{H}_{\rm eff}, to 2nd order in \hat{V}, as given in Ref. Kettemann20252 . Thereby one obtains the renormalized couplings and local fields for all remaining N-2 spins. Even if the form of the Hamiltionian remains unchanged, the renormalized couplings differ from the initial ones, and the distribution function of couplings is changed accordingly to P_{E}(J,\Omega_{0}-d\Omega), with \Omega_{0}-d\Omega the largest energy scale in the reduced system of N-2 spins. E is the total energy of the system. Repeating this procedure until all

Read the complete text