Many-Body Localization in Disordered Spin Chains

From Anderson localization to the many-body problem

Philip Anderson showed in 1958 that a single quantum particle moving on a disordered lattice can become localized: rather than spreading out ballistically or diffusively as it would in a clean, periodic lattice, its wavefunction remains exponentially confined near its starting region forever, a purely wave-interference effect caused by disorder-induced multiple scattering that destructively interferes with any long-range propagation, entirely independent of any particle-particle interaction. For decades this single-particle result was considered largely irrelevant to genuinely interacting many-body systems, where it was widely assumed that even weak interactions between localized particles would inevitably allow them to exchange energy with one another and eventually thermalize, restoring conventional statistical mechanical behavior at long times. The theoretical breakthrough of many-body localization, developed rigorously in perturbative arguments by Basko, Aleiner, and Altshuler in the mid-2000s and substantially clarified and extended by subsequent numerical and analytical work, demonstrated that this assumption is not always correct: for sufficiently strong disorder, an interacting many-body system can remain localized even in the presence of interactions, at any energy density, not merely at low temperature, marking a fundamentally new phase of quantum matter, an interacting generalization of Anderson localization that survives the presence of interparticle interactions rather than being destroyed by them.

Local integrals of motion and emergent integrability

The modern theoretical understanding of the MBL phase centers on the existence of an extensive set of local integrals of motion, often called l-bits, effective conserved quantities that, unlike the bare physical spins of the original disordered chain, are dressed, quasi-local operators, each one built predominantly from a physical spin near a particular site but decorated with an exponentially decaying tail of corrections from more distant sites, reflecting the residual, exponentially weak interactions mediated by the localized single-particle orbitals. Because these l-bits are (approximately) conserved and (approximately) independent of one another, the MBL Hamiltonian can be rewritten, at least approximately and in many rigorously established cases, as a sum of these l-bit operators plus weak, exponentially decaying effective interactions between them, mathematically resembling an integrable model despite the original Hamiltonian containing generic, non-integrable interactions and disorder. This effective integrability is the deep reason MBL systems fail to thermalize: because there are as many independent conserved quantities as there are physical degrees of freedom, the system's dynamics is far more constrained than a generic ergodic system's, and the initial values of all these local conserved quantities, essentially a detailed record of the initial local configuration, are permanently retained rather than being scrambled into a thermal, indistinguishable equilibrium ensemble.

Logarithmic entanglement growth: the defining signature

Perhaps the single most striking and most theoretically celebrated signature distinguishing many-body localization from both Anderson localization and ordinary thermalizing dynamics is the behavior of entanglement entropy growth following a quantum quench starting from an unentangled, or weakly entangled, initial product state. In a thermalizing (ergodic) system, entanglement entropy between two halves of the chain grows ballistically, linearly in time, and eventually saturates at a value proportional to the size of the smaller subsystem, called volume-law entanglement, reflecting the fact that thermalization genuinely spreads quantum correlations and information throughout the entire system. In a simple, non-interacting Anderson-localized system, by contrast, entanglement entropy does not grow appreciably at all after the initial transient, since without interactions there is no mechanism to generate long-range correlations between localized single-particle orbitals. Many-body localization is distinguished from both these limits by an intermediate, remarkably slow logarithmic-in-time growth of entanglement entropy, which nonetheless eventually saturates at a volume-law value, driven physically by the weak, exponentially small in distance, residual interactions between the dressed l-bit degrees of freedom, which allow dephasing-mediated entanglement generation between distant, exponentially weakly coupled l-bits, a process that takes exponentially long for exponentially distant pairs and thus produces the characteristic logarithmic time dependence rather than the linear growth of a thermalizing system. This slow logarithmic entanglement growth, first predicted theoretically and later confirmed in numerical simulations of small disordered spin chains, has become the most widely cited theoretical fingerprint used to identify the MBL phase in numerical and, to a more limited extent, experimental studies.

Disorder strength, the MBL transition, and finite-size caveats

The onset of many-body localization as disorder strength is increased is understood, at least in the simplest models such as the paradigmatic disordered Heisenberg spin chain, as a genuine dynamical phase transition separating a thermalizing (ergodic) phase at weak disorder from a localized (MBL) phase at strong disorder, with the transition itself, and the precise nature of the critical point, remaining an active and genuinely unresolved topic of theoretical debate, since standard renormalization-group and finite-size-scaling tools developed for equilibrium phase transitions apply awkwardly to this fundamentally non-equilibrium, dynamical transition. Numerically, the transition is typically identified through changes in level-statistics measures (Poisson statistics deep in the MBL phase versus Wigner-Dyson random-matrix statistics in the thermal phase), the saturation value and growth rate of entanglement entropy, or the long-time persistence of local memory (such as an initial spin imbalance between even and odd sites), but exact numerical studies are fundamentally restricted to small system sizes, typically a few tens of spins, because the Hilbert space dimension grows exponentially with system size, and a persistent, still-debated theoretical question concerns whether the apparent MBL phase observed in these necessarily finite-size numerical and experimental studies survives, or instead is eventually destabilized by rare, weakly disordered thermal regions (so-called avalanche or 'thermal bubble' instabilities), in the strict thermodynamic limit of an infinitely long chain, a subtlety that has become one of the most actively contested open questions in the field over the past several years.

Experimental realizations in cold atoms and qubit arrays

Because many-body localization is fundamentally a statement about the absence of thermalization in an isolated, closed quantum system, evading essentially any coupling to an external thermal bath, it has proven exceptionally well suited to study using highly controllable, well-isolated synthetic quantum systems rather than conventional solid-state materials, where electron-phonon coupling and other environmental interactions would eventually restore thermalization regardless of disorder strength. Ultracold atoms trapped in optical lattices with programmable, quasi-random disorder potentials provided some of the earliest and most influential experimental evidence for MBL, directly observing the predicted persistence of an initial density imbalance between even and odd lattice sites over remarkably long observation times once disorder exceeded a critical strength, a direct real-time signature of broken ergodicity. Trapped-ion quantum simulators and superconducting qubit processors have since provided complementary and increasingly sophisticated realizations, including direct, qubit-resolved measurements of the logarithmically slow entanglement entropy growth predicted theoretically, made possible by the unique ability of these platforms to measure quantities, like entanglement entropy itself, that are exceedingly difficult to access in conventional solid-state disordered materials. These experimental platforms have not only confirmed core theoretical predictions of MBL but have also enabled discovery of related phenomena such as many-body localization protected by discrete time-translation symmetry breaking (time crystals) and Stark many-body localization, in which a linear potential gradient rather than random disorder can induce qualitatively similar non-thermalizing dynamics.

Frequently asked questions

Try it live

Everything above runs in your browser — open Many-Body Localization in Disordered Spin Chains and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.