Supervised and unsupervised learning of the many-body critical phase, phase transitions, and critical exponents in disordered quantum systems

Aamna Ahmed, Nilanjan Roy · 2025

Read Supervised and unsupervised learning of the many-body critical phase, phase transitions, and critical exponents in disordered quantum systems by Aamna Ahmed, Nilanjan Roy (2025) 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-2501-03981.txt.

Introduction

Disordered quantum systems can settle into phases that spread, freeze, or sit strangely in between, and telling them apart is hard. Aamna Ahmed and Nilanjan Roy ask whether neural networks can learn the difference from raw data: the spacings between energy levels and the shapes of quantum states, barely cleaned up.

Trained networks separated the ergodic, localized, and in-between critical phases, and their outputs marked the borders between phases in line with standard methods. Without any training, a simple compression technique also traced the phases, and its entropy acted as a working order parameter. By scaling their network outputs, the authors extracted critical exponents matching known values. The work appears in 2025 and extends machine learning's reach in quantum matter.

Catalog record

CREATORAamna Ahmed, Nilanjan Roy
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE75 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-03981

Browse the full library · About the Tower of Babel

Full text

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

Supervised and unsupervised learning of the many-body critical phase, phase transitions, and critical exponents in disordered quantum systems Aamna Ahmed, Nilanjan Roy Abstract Supervised and unsupervised machine-learning-based frameworks have been implemented in the past to classify various dynamical phases of disordered quantum systems using rather sophisticated (highly preprocessed or difficult to generate) input data. In this paper, we begin by questioning the existence of a nonergodic extended phase, namely, the many-body critical (MBC) phase in finite systems of an interacting quasiperiodic system. We find that this phase can be separately detected from the other phases, such as the many-body ergodic (ME) and many-body localized (MBL) phases, in the model through supervised neural networks made for both binary and multiclass classification tasks, utilizing, rather unpreprocessed, eigenvalue spacings and eigenvector probability densities as input features. Moreover, the output of our trained neural networks can also indicate the critical points separating ME, MBC, and MBL phases, which are consistent with the same obtained from other conventional methods. We also employ unsupervised learning techniques, particularly principal component analysis (PCA) of eigenvector probability densities, to investigate how this framework, without any training, captures the, rather unknown, many-body phases (ME, MBL, and MBC) and single-particle phases (delocalized, localized, and critical) of the interacting and noninteracting systems, respectively. Our findings reveal that PCA entropy serves as an effective indicator (order parameter) for detecting phase transitions in the single-particle systems. Moreover, this method proves applicable to many-body systems when the data undergoes a suitable preprocessing. Interestingly, when it comes to the extraction of critical (correlation length) exponents through a finite-size scaling, we find that for single-particle systems, scaling collapse of neural network outputs is obtained using components of inverse participation ratio (IPR) as input data. Remarkably, we observe identical critical exponents as obtained from the scaling collapse of the IPR directly for different single-particle phase transitions. However, the current finite-size scaling approach to the network output cannot reliably extract critical exponents for the many-body system that needs further investigation. Preprint: APS/123-QED I INTRODUCTION Machine learning techniques have emerged as transformative tools in the study of quantum systems, offering innovative approaches to tackle problems that are computationally intensive for traditional methods [ 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 ] . These techniques excel in capturing complex correlations and high-dimensional features from raw data. Among various strategies, supervised learning has gained prominence in phase classification tasks by utilizing labelled data to learn distinct phase boundaries within the parameter space [ 14 , 15 , 16 , 17 , 18 ] . This approach has demonstrated significant success in identifying phase transitions across a range of systems, from disordered to strongly correlated models. In parallel, unsupervised techniques like the principal component analysis (PCA) have offered a complementary perspective, capable of detecting phase transitions without requiring prior knowledge of the phases [ 19 , 20 , 21 , 22 ] . PCA effectively reduces the dimensionality of large datasets, extracting dominant features that encode critical behaviours, making it a compelling alternative to traditional analytical tools. The study of quantum dynamics of isolated many-body systems has been pivotal in understanding phase transitions in out-of-equilibrium quantum matter, especially in the presence of disorder [ 23 , 24 , 25 ] . These systems in 1D exhibit two well-established phases: the ergodic phase and the many-body localized (MBL) phase. In the ergodic phase, the system adheres to the Eigenstate Thermalization Hypothesis (ETH) [ 26 ] , achieving thermal equilibrium with subsystem properties aligning with statistical mechanics predictions. Conversely, the MBL phase avoids thermalization under strong disorder, retaining memory of its initial state over a long period. Recent studies have highlighted the existence of a third phase, namely, the nonergodic extended (NEE) phase which is an intermediate one between the ergodic and MBL phases [ 27 , 28 , 29 , 30 , 31 ] . It was first proposed in the interacting generalized Aubry-André-Harper (GAAH) model where NEE states were found in a small fraction of the many-body spectrum [ 27 ] . Very recently, a more robust presence of NEE states has been found throughout the spectrum, dubbed as the many-body critical (MBC) phase, in an interacting extended AAH (EAAH) model [ 28 ] . While the MBC phase (NEE states) violates ETH, its entanglement entropy still follows the volume law like the ergodic phase. To distinguish nonergodic phases, methods such as the one-particle density matrix and entanglement spectrum have been widely applied [ 32 , 33 , 34 ] . However, these conventional techniques often entail significant computational overhead. Notably, in this context earlier studies have explored various types of sophisticated input data for machine learning including entanglement spectra, intrinsic dimensions, and others [ 20 , 21 , 1 , 35 , 36 , 12 ] . While conventional measures are effective for well-understood phase transitions, they rely on predefined observables tailored to specific cases, limiting their applicability to more intricate regimes. For example, metrics such as the gap ratio and many-body inverse participation ratio (MIPR) successfully distinguish thermal and MBL phases but face significant challenges in identifying complex phases such as many-body critical (MBC) and other non-ergodic phases (with a lot of spectral gaps or non-monotonic density of states). In these cases, the standard metrics struggle and mere deviations from standard indicators are often interpreted as signatures of criticality rather than definitive phase boundaries (see Appendix A ). These limitations underscore the need for a more flexible and data-driven framework—one that can systematically classify phases without relying on prior assumptions about order parameters or correlation functions. Machine learning provides such a framework, enabling unbiased phase identification and uncovering subtle structures in the spectrum that conventional measures may overlook because of the lack of complex regression analysis unlike the supervised machine learning which performs a complex non-linear regression per se before it generates a stable output [ 12 ] . This advantage is particularly evident in cases where traditional measures, such as level statistics, fail due to spectral complexities (see Appendix C for details). In this work, we introduce a machine learning-based framework that leverages simple yet highly informative input features—eigenvalue spacings and eigenvector probability densities (PDs)—to uncover and analyze phase transitions in the interacting EAAH model mentioned earlier. Building on the findings of Ref. [ 28 ] , which claimed the existence of a many-body critical phase alongside the ergodic and many-body localized phases, we provide a detailed study of searching the MBC phase alongside ergodic and MBL phases, albeit in finite systems, using supervised learning. Fully connected neural networks trained on eigenvalue spacings and eigenvector PDs not only classify these phases with remarkable accuracy but also reveal the tantalizing possibility of distinct sub-phases within the many-body critical regime. Furthermore, our application of PCA to eigenvector PDs highlights its potential as a tool for unsupervised phase identification, with the first principal component capturing dominant phase-transition-related information after suitable preprocessing in the many-body systems. For single-particle systems, we demonstrate the effectiveness of PCA entropy [ 37 ] and its numerical derivative in systematically identifying phase transitions. Another significant outcome is the validation of critical

Read the complete text