Local Integrals of Motion: Quantum Memory and the Failure of Thermalization

In the vast landscape of quantum mechanics, a central tenet is thermalization: the process by which isolated systems evolve to a state of thermal equilibrium, forgetting their initial conditions. This principle underpins statistical mechanics and our understanding of macroscopic matter. However, a fascinating class of systems stubbornly defies this rule, preserving a memory of their past indefinitely. This phenomenon, known as many-body localization (MBL), challenges our foundational assumptions and opens up new frontiers in physics. But what is the underlying mechanism that allows these systems to escape the universal fate of thermalization? The answer lies in a profound concept: the existence of an extensive set of local integrals of motion (LIOMs).

This article provides a comprehensive exploration of Local Integrals of Motion, delving into their structure, consequences, and applications. In the following sections, we will first uncover the fundamental ​​Principles and Mechanisms​​ of LIOMs, exploring how these 'dressed' conserved quantities are constructed and how their existence leads to the breakdown of thermalization and unique dynamical signatures like logarithmic entanglement growth. Subsequently, we will explore the far-reaching ​​Applications and Interdisciplinary Connections​​, revealing how LIOMs serve as the foundation for robust quantum memories and enable the creation of exotic non-equilibrium phases of matter, such as discrete time crystals, connecting their study to fields like condensed matter physics and quantum information.

In physics, we have a deep-seated belief, a foundational principle, that things tend to settle down. A hot cup of coffee cools to room temperature. A shuffled deck of cards becomes random. A complex quantum system, left to its own devices, should forget the intricate details of its birth and relax to a state of thermal equilibrium, remembering only its most basic conserved properties, like its total energy. This process, which we call thermalization, is the bedrock of statistical mechanics. It is the reason we can describe a gas with a trillion trillion particles using just a few numbers like temperature and pressure.

But what if this belief, this cornerstone of our understanding, is not always true? What if there are systems that stubbornly refuse to forget? Systems that hold onto the memory of their initial state not just for a fleeting moment, but forever? This is not a flight of fancy; it is the strange and beautiful reality of ​​many-body localization (MBL)​​. The key to unlocking this mystery lies in a profound generalization of one of physics' most cherished concepts: the conservation law.

We are all familiar with conservation laws. In a closed system, total energy is conserved. If the system has certain symmetries, other quantities are conserved too—momentum is conserved if the system is uniform in space, angular momentum if it is isotropic. These are global quantities, single numbers that describe the system as a whole. They are the bookkeepers of thermalization, defining the final equilibrium state. For a generic, chaotic quantum system, the ​​Eigenstate Thermalization Hypothesis (ETH)​​ tells us that energy is the only local information that matters. Any two states with the same energy density look locally identical, and the system inevitably thermalizes.

An MBL system evades this fate because it possesses a radically different and far more extensive set of conserved quantities. They are not global numbers but a vast collection of local ones, one for nearly every part of the system. We call them ​​Local Integrals of Motion​​, or ​​LIOMs​​. Imagine your system is a long chain of spins. Instead of just one conserved number (the total energy), you have a whole set of "dials," $\{\tau_1, \tau_2, \tau_3, \dots\}$, one for each site along the chain. When you prepare the system in some initial configuration, you set the positions of all these dials. The astonishing property of an MBL system is that, as time evolves, these dials are locked in place. The value of each $\tau_i$ remains perfectly constant: $[\hat{H}, \tau_i] = 0$ for all $i$.

This is a complete game-changer. The system doesn't just remember one number; it remembers an extensive amount of local information, a detailed "fingerprint" of its initial state, which it preserves for all time. It is this local memory that prevents thermalization.

What are these LIOMs, these magical memory keepers? At first glance, one might guess that a LIOM associated with site $i$, say $\tau_i^z$, is simply the spin at that site, $S_i^z$. But this cannot be. In an interacting system, the spin at site $i$ is constantly being jostled and flipped by its neighbors. Its value does not stay constant. The interaction term in the Hamiltonian, something like $J \mathbf{S}_i \cdot \mathbf{S}_{i+1}$, ensures that $[H, S_i^z] \neq 0$.

The true LIOM is a more subtle and beautiful object. It is a "dressed" version of the physical spin. Think of the LIOM $\tau_i^z$ as being composed of a core, which is indeed the spin $S_i^z$, but it is surrounded by a "cloud" of entanglement with its neighbors. This dressing is precisely tailored to cancel out all the kicks and jostles from the Hamiltonian, ensuring that the total object, $\tau_i^z$, remains perfectly conserved. We can even construct this dressing perturbatively: the LIOM is the bare spin plus small corrections involving operators on adjacent sites, then smaller corrections on sites further away, and so on.

This cloud is not some fuzzy, undefined thing. It has a definite structure. Its influence decays exponentially with distance from the core site $i$. This is the crucial meaning of "local" in LIOMs; they are more precisely ​​quasi-local​​. The spatial extent of this exponential tail is characterized by a fundamental parameter called the ​​localization length​​, $\xi$. In the MBL phase, this length is finite. Strong disorder and weak interactions conspire to keep this dressing cloud tightly bound to its core spin, with a localization length given by a relation like $\xi \approx a / \ln(\Delta E/J)$, where $\Delta E$ is the characteristic disorder strength, $J$ is the interaction strength, and $a$ is the lattice spacing. This formula beautifully illustrates the underlying mechanism: strong disorder (large $\Delta E$) makes the LIOMs more localized (smaller $\xi$), stabilizing the MBL phase.

The existence of an army of LIOMs completely dismantles the standard picture of statistical mechanics.

First, it invalidates the Eigenstate Thermalization Hypothesis. ETH claims that a local measurement on a typical energy eigenstate should depend only on the state's total energy. But in an MBL system, an eigenstate is simultaneously an eigenstate of the Hamiltonian and all the LIOMs. Its properties depend not just on its energy, but on the value of every single LIOM. This means we can easily find two eigenstates with virtually identical energy but with different values for some of the $\tau_i$'s. A local measurement on these two states will yield different results, in stark violation of ETH. The Hilbert space is finely fractured into countless sectors, each labeled by a distinct set of LIOM eigenvalues.

So, if a quenched MBL system doesn't relax to the familiar thermal Gibbs state $\rho \propto \exp(-\beta H)$, what does it relax to? It settles into a state that honors the conservation of every single LIOM. Using the principle of maximum entropy, we can construct the correct statistical description, known as the ​​Generalized Gibbs Ensemble (GGE)​​:

Here, each LIOM $\tau_i$ gets its own Lagrange multiplier $\lambda_i$, which is determined by the initial value of that LIOM in the starting state. The GGE is the mathematical embodiment of the system's memory. While a thermal system only needs one "handle," the temperature $\beta$, to describe it, an MBL system needs a whole control panel of handles, $\{\lambda_i\}$, one for each piece of local information it remembers.

This underlying structure of LIOMs leaves behind spectacular and unmistakable dynamical signatures—footprints in time that we can observe in experiments.

The most famous of these is a strange decoupling of transport and information. Since the LIOMs are localized and their values are fixed, they cannot move through the system. This means that physical quantities like charge, energy, and spin are also locked in place. An MBL system is a perfect insulator with zero DC conductivity. And yet, something does spread: quantum information, in the form of entanglement.

Imagine two distant spins in an initial product state (meaning they are unentangled). They are the cores of two different LIOMs, $\tau_i^z$ and $\tau_j^z$. While the LIOMs themselves don't move, their exponential tails can overlap and interact. This interaction is too weak to destroy the LIOMs, but it is strong enough to cause ​​dephasing​​. The two spins start to precess at slightly different rates, their relative phase evolving in a way that depends on the states of all the other LIOMs in the system. This gradual loss of phase coherence between different parts of the system is what generates entanglement.

Because the interaction between LIOMs decays exponentially with distance, this dephasing process is incredibly slow. The time it takes to entangle a spin at site $i$ with a spin at site $j$ scales exponentially with their separation, $t \sim \exp(|i-j|/\xi)$. If we turn this around and ask how far entanglement has spread by time $t$, we find that the distance grows only as the logarithm of time, $d(t) \sim \xi \ln(t)$. Consequently, the amount of entanglement between two halves of the system grows logarithmically with time:

This logarithmic crawl of entanglement is the ultimate fingerprint of MBL. It is a stark contrast to a thermalizing system, where entanglement sprints ahead linearly with time. This logarithmic growth is not just a qualitative feature; its rate, the coefficient $\alpha$, is directly proportional to the LIOM localization length $\xi$, providing a direct window into the microscopic structure of these hidden conserved quantities.

This persistent memory manifests in other ways, too. A local perturbation, like a single flipped spin, does not wash away in a thermal bath. Instead, its memory decays with an agonizingly slow power law in time, $F(t) \sim t^{-\eta}$, where the exponent $\eta$ is once again controlled by the localization length $\xi$. The system simply cannot forget. The principles of MBL, rooted in the existence of these remarkable local integrals of motion, reveal a universe of quantum dynamics far richer and more mysterious than we ever imagined, a world where memory can, against all odds, endure.

After a journey through the fundamental principles of local integrals of motion (LIOMs), you might be wondering: what are they good for? Are they merely a clever theoretical curiosity, an exception to the grand rule of thermalization? The answer, it turns out, is a resounding no. The existence of LIOMs and the phenomenon of many-body localization (MBL) that they underpin is not an ending, but a beginning. It opens the door to a new universe of non-equilibrium physics, with profound implications that ripple across condensed matter physics, quantum information, and even the most fundamental questions about the nature of time.

Let's explore this new territory. We will see that LIOMs are not just about preventing things from happening—like the flow of heat—but about enabling entirely new, dynamic phenomena that are impossible in the familiar world of thermal equilibrium.

The most immediate consequence of a system having a full set of LIOMs is its complete failure to transport conserved quantities like energy or charge. Imagine a bustling crowd where every person is suddenly confined to their own small, invisible room. No matter how much they move within their room, they can never swap places with anyone else. The overall distribution of people becomes frozen. This is the essence of an MBL phase. It is a perfect insulator, not just in its ground state, but at any energy density. All the information about the initial state is locked into the local configuration of the LIOMs, unable to spread.

This profound lack of transport can be characterized by a quantity called the Thouless energy, $E_{\text{Th}}$, which represents the energy scale associated with the time it takes for information to cross the entire system. In a normal, diffusive system, this time scales with the system size squared, so $E_{\text{Th}}$ is small but finite. In an MBL system, the transport time is effectively infinite, causing the Thouless energy to vanish entirely. This collapse of $E_{\text{Th}}$ is the reason why the energy levels of an MBL system lose the "level repulsion" characteristic of chaotic systems and instead follow simple Poisson statistics—they are like uncorrelated random numbers, unaware of each other's existence.

But this "freezing" of information suggests something more exciting than just good insulation. If information is locked in place and protected from thermal scrambling, could we use it to build a new kind of memory? The answer is yes. LIOMs represent a naturally occurring, robust form of quantum memory.

Consider a thought experiment straight out of the foundations of quantum mechanics: a quantum eraser. We send a particle through a double-slit apparatus and place a detector at one slit to find out which path it took. The detector is not just a single atom, but an entire MBL spin chain. The which-path information is recorded by flipping the state of one of its LIOMs. Because the LIOM is a stable, conserved quantity, this information is securely stored. The beauty of this setup is that a LIOM is not a simple, "bare" physical spin. It is a "dressed" operator, a complex collective property of the physical spins in its vicinity. If you, the experimentalist, try to erase the which-path information by measuring a single physical spin, you find that you only have partial access to the information. The visibility of the interference pattern is only partially restored. This is a stunning demonstration of both quantum complementarity and the cryptic, quasi-local nature of LIOMs. They are robust, hidden variables that protect quantum information from local prying.

This robustness can be pushed even further. Imagine you want to prepare a system in a specific MBL state by slowly changing its Hamiltonian—a process known as adiabatic evolution. In any real experiment, the process is never perfectly slow, and errors inevitably creep in. Yet, if you monitor the final state of the LIOMs, you discover something remarkable: their expectation values are protected from errors to a higher order than you might expect. The leading-order non-adiabatic corrections vanish!. It's as if the LIOMs have a built-in resistance to being perturbed, making them exceptionally stable targets for quantum state engineering.

The stability provided by LIOMs does more than just preserve the status quo; it creates a platform for entirely new phases of matter to exist—phases that would be unthinkable in systems that obey the ordinary rules of statistical mechanics.

Perhaps the most spectacular example is the ​​discrete time crystal​​. Imagine periodically shaking a quantum system. Just as repeatedly stirring a cup of coffee makes it uniformly hot, periodically driving an interacting quantum system is expected to heat it up to a featureless, infinite-temperature state. This is the fate of nearly all such systems. Any interesting quantum order is quickly melted away by the absorbed energy. But what if the system were unable to thermalize from this driving? This is precisely the situation in an MBL system. The LIOMs provide a rigidity that prevents the system from heating up under a periodic drive. In this strange, un-heatable environment, a new form of order can emerge. If the drive is engineered correctly—for instance, by applying a pulse that flips all the spins approximately every period $T$—the system can begin to oscillate spontaneously, but at a period of $2T$. It remembers how many times it has been "pushed" and responds only on every second push. This subharmonic response signifies the spontaneous breaking of discrete time-translation symmetry. It is a phase of matter that is intrinsically out of equilibrium, a clock that ticks at its own emergent frequency, stabilized against thermal death by the underlying web of LIOMs. This is not a perpetual motion machine, but a new, robust form of non-equilibrium matter.

Of course, this exotic physics doesn't happen in just any system. There are rules. A deep and beautiful connection exists between the symmetries of a system and its ability to host an MBL phase. It turns out that systems with non-Abelian symmetries—like the continuous $\mathrm{SU}(2)$ rotational symmetry of a Heisenberg spin chain—cannot be fully many-body localized. The reason is subtle: the conserved quantities associated with the symmetry (like the total spin components $S^x$, $S^y$, $S^z$) do not commute with each other. If a full set of commuting LIOMs existed, these conserved quantities would have to be functions of them, which would force them to commute—a contradiction! This means the spin degrees of freedom in such a system can never truly localize; they form an internal "bath" that will eventually thermalize the entire system. To find MBL, one must look to systems with simpler, Abelian symmetries (like conservation of particle number or spin projection along a single axis), where this fundamental obstruction does not exist.

So far, our discussion has focused on the strongly disordered realm of MBL. But the concept of LIOMs also provides a powerful language to understand the dynamics of systems that are almost perfect—the nearly integrable systems that are much closer to the idealized models often studied in textbooks.

An integrable model, like the XXZ spin chain, possesses its own set of LIOMs by definition. Now, what happens if we add a small, integrability-breaking perturbation? The system is no longer truly integrable and, given enough time, should thermalize. But the journey to thermal equilibrium is not instantaneous. For a very long time, the system is trapped in a "prethermal" state. This state is not thermal; instead, its local properties are described by a statistical ensemble built from the LIOMs of the original, unperturbed integrable model. The system remembers its integrable roots long after it has forgotten the fine details of its initial state. LIOMs thus provide the framework for a hierarchy of relaxation timescales, bridging the gap between the idealized world of integrable physics and the messy reality of generic, thermalizing systems.

The story of local integrals of motion is a testament to the richness of quantum mechanics. They are the key to understanding how some systems can defy thermalization, providing a new form of stability. This stability is not just a passive shield but an active ingredient that enables robust quantum memories and gives birth to entirely new, dynamic phases of matter like time crystals. The ongoing quest to understand the precise nature of the MBL transition, with its strange activated scaling where time and space are connected exponentially, and to determine if these phenomena can even exist in more than one dimension, marks one of the most exciting frontiers in modern physics. LIOMs have shown us that even in the most disordered corners of the quantum world, a deep and beautiful order can persist.