Projected Entangled Pair States

In the quantum realm, describing a system of many interacting particles presents a staggering challenge known as the "curse of dimensionality." The information required to specify the state of even a modest number of particles grows exponentially, quickly becoming too vast to store or process. This fundamental barrier limits our ability to understand and predict the behavior of complex quantum materials, from novel magnets to the exotic matter needed for quantum computers. How, then, can we forge a practical language to describe these intricate quantum tapestries?

This article delves into Projected Entangled Pair States (PEPS), a powerful theoretical and computational framework designed specifically to solve this problem for two-dimensional systems. By building a global quantum state from a network of simple, local building blocks called tensors, PEPS provides a description that is both physically motivated and computationally efficient. The reader will embark on a journey through the core ideas behind this revolutionary approach. The first chapter, "Principles and Mechanisms," will deconstruct how PEPS are built, why they naturally capture the essential physics of entanglement, and what computational trade-offs they entail. Following this, the "Applications and Interdisciplinary Connections" chapter will showcase the immense power of PEPS in practice, from simulating real-materials and discovering new phases of matter to providing a universal language for complexity that extends even beyond the quantum world.

Figure 1: The PEPS construction. A tensor (circle) at each lattice site has one physical index 's' (leg pointing up) and several virtual indices (legs in the plane) that connect to neighbors.

Imagine you want to describe a picture. You could, in principle, list the exact color of every single pixel, one by one. For a high-resolution image, this would be an astronomically long and mostly useless list. A much better way would be to describe the objects in the picture—a "red ball," a "blue sky." You've compressed an enormous amount of information into a set of meaningful concepts and their relationships.

Describing a quantum state of many particles, say the electrons in a magnet, faces a similar challenge, but it's even more daunting. The number of "pixels"—the complex amplitudes for every possible configuration of electron spins—grows exponentially with the number of particles. For even a handful of atoms, this number exceeds the number of atoms in the universe. This is the infamous "curse of dimensionality." How can we hope to describe, let alone understand, such a monster? This is where the beautiful idea of ​​Projected Entangled Pair States (PEPS)​​ comes to the rescue.

The PEPS approach says: let's not try to write down the entire list of numbers. Instead, let's build the quantum state from small, local building blocks, just as a wall is built from bricks. These building blocks are mathematical objects called ​​tensors​​.

You can think of a tensor as a multi-dimensional array of numbers. For our purposes, we place one tensor on each site of our lattice of particles, like a quantum "atom" for that location. Each tensor has several "arms" or ​​indices​​. One of these arms is special: it's the ​​physical index​​. This index corresponds to the actual physical state at that site—for a qubit, it might take the value 0 for "spin down" or 1 for "spin up."

The other arms are called ​​virtual indices​​. These are the secret to the whole construction. They don't represent a directly observable property; instead, they act as the "glue" that connects a tensor to its neighbors. The number of possible values each virtual index can take is called the ​​bond dimension​​, denoted by $D$. This number is crucial: it determines how much entanglement, or quantum correlation, can be shared between neighboring sites.

Imagine building with LEGOs. Each brick is a tensor. The color of the brick is its physical state. The studs and tubes on the brick are its virtual indices, allowing it to connect to other bricks. The complete, intricate LEGO model is the global quantum state, built from simple, identical pieces.

We have spent some time getting to know these curious objects called Projected Entangled Pair States, or PEPS. We've seen how they are built, piece by piece, by stitching together local tensors into a grand, entangled tapestry that describes a many-body quantum state. It’s a beautiful construction, to be sure. But the natural, itching question is: so what? What is this machinery good for? What secrets of the universe can this construction unlock?

It turns out that this seemingly abstract idea is one of the most powerful and versatile tools we have for understanding the complex world of many-interacting-particle systems. The journey of its applications will take us from the gritty, practical work of simulating real materials to the farthest frontiers of theoretical physics—designing blueprints for quantum computers and uncovering exotic, new phases of matter. And then, in a final, surprising twist, we will see that this language, born to describe the quantum world, transcends it entirely, offering a new perspective on complexity itself.

Let's start with the most immediate challenge in condensed matter physics: we have a material, perhaps some kind of magnet, whose microscopic particles interact according to a given Hamiltonian, and we want to know what it does. What is its ground state—its state of lowest energy, which it will naturally seek out at low temperatures? For all but the simplest systems, this is a fantastically difficult problem. The number of possible configurations grows exponentially with the number of particles, a situation rightly called the "tyranny of numbers."

This is where PEPS comes in not as a description of a known state, but as a fantastically sophisticated "guess" for an unknown one. Using the powerful variational principle, we can use a PEPS as a trial wavefunction, or an ansatz, and systematically adjust its local tensors to find the configuration that minimizes the energy. It’s like tuning a giant, multi-dimensional radio, twisting the knobs on every tensor, to find the station with the quietest static—the lowest possible energy. The state we find is our best approximation of the true ground state.

For instance, this approach lets us tackle venerable problems like the two-dimensional quantum Ising model, a cornerstone for understanding magnetism and phase transitions. By postulating a PEPS structure, we can calculate the expected energy and then vary the parameters within our tensors to find the minimum, giving us a remarkably accurate picture of the system's ground state properties.

Of course, "tuning the knobs" on a PEPS is a high art in itself. The space of all possible tensors is vast. Physicists and computer scientists have developed ingenious algorithms for this task. One of the most beautiful methods is inspired by a trick from quantum mechanics itself: imaginary time evolution. If we take an arbitrary starting state and evolve it not in real time $t$, but in imaginary time $\tau = it$, the evolution operator $e^{-H\tau}$ does something magical. Instead of making the state oscillate, it exponentially suppresses components with higher energy. If we evolve for long enough, all excited states "decay" away, leaving behind only the pure, lowest-energy ground state. Tensor network algorithms cleverly simulate this process, applying small imaginary time-step "gates" to the PEPS and using a procedure called the singular value decomposition to keep the complexity of the tensors in check, ensuring the simulation remains tractable. It is through this computational engine that PEPS becomes a practical quantum simulator's toolbox.

Suppose our simulation is complete. We have found the PEPS that describes the ground state of our material. We possess a complete, microscopic description. But a list of all the numbers in the tensors is not physics. We want to know the character of the state. Is it a magnet? Does it conduct electricity? Critically, how do different parts of the system talk to each other? If we poke a spin at one location, how far away does another spin feel the effect?

This property is quantified by the correlation length, $\xi$. It tells us the characteristic distance over which correlations decay. A short correlation length means the system is "short-sighted," with local disturbances quickly forgotten. An infinite correlation length is the signature of a critical point, where the system is perched on the edge of a phase transition and fluctuations are felt across the entire sample. The PEPS framework provides a direct way to calculate such properties. By constructing an object called the transfer matrix—which you can intuitively picture as a machine that "scans" the PEPS from one column to the next, telling us how the information stored in the virtual bonds evolves—we can find the correlation length from its leading eigenvalues. The rate at which the "dominant message" of the scan overtakes the sub-dominant ones directly tells us how quickly the system loses its memory of itself, giving us the correlation length.

This ability to move from a microscopic description to macroscopic, physical observables is what elevates PEPS from a mere representation to a true predictive theory.

So far, we have used PEPS to find the ground state of a given Hamiltonian. But we can, and this is a profound shift in thinking, turn the logic on its head. We can start by designing a local tensor $A$ with some interesting properties, build a global PEPS from it, and then ask: what kind of universe, what Hamiltonian $H$, would have this state as its natural ground state?

This leads to the concept of a parent Hamiltonian. It turns out that for many PEPS, one can construct a local, frustration-free Hamiltonian (a sum of terms acting on small, nearby groups of particles) for which the PEPS is the exact ground state with zero energy. The condition boils down to a simple, local check: does the PEPS tensor, when a local piece of the Hamiltonian acts on it, yield zero? If it does, then the global state is automatically "annihilated" by every term in the Hamiltonian, making it the ground state.

This "inverse" approach is not just a mathematical curiosity; it is a blueprint for discovering and designing exotic phases of matter. Perhaps the most celebrated example is the ​​toric code​​. The PEPS representation of the toric code ground state is astonishingly simple. The local tensor, sitting at the vertices of a square lattice, is just a "parity checker": its value is 1 if an even number of its virtual indices are in the '1' state, and 0 otherwise. That's it. From this ridiculously simple local rule emerges a state of breathtaking complexity. It exhibits topological order, a form of order that cannot be described by any local property, but only by global properties, like winding a string around the underlying torus. Its excitations are not ordinary particles but anyons, which have bizarre statistical properties. And, most importantly for technology, this state is naturally robust to local errors, making it a leading candidate for building a fault-tolerant quantum computer. The fact that its PEPS representation requires the smallest non-trivial virtual bond dimension, $D=2$, reveals the fundamental elegance of the model when viewed through the lens of tensor networks.

The toric code is just the beginning. This constructive power allows us to write down exact PEPS models for a whole menagerie of quantum states, from simple toy models that illustrate how local constraints create global correlations to the sophisticated fixed-point wavefunctions for Symmetry-Protected Topological (SPT) phases. In these advanced applications, the tensors are built directly from the mathematical structures of group theory and cohomology, weaving the symmetries of the system directly into the entanglement fabric of the state. PEPS provides a constructive language that unifies the physics of entanglement with the mathematics of symmetry.

The story has one final, beautiful turn. The structure of a PEPS—a network of local tensors whose contraction gives a global number (the quantum amplitude)—is more general than quantum mechanics itself. It is, in fact, a universal language for describing complex systems where local rules generate global patterns.

Consider a classical system, like the squares on a checkerboard or the grid of a cellular automaton. The state of the system is not described by a wavefunction, but by a probability distribution. For a system at thermal equilibrium, this is the famous Boltzmann distribution, $\mathbb{P}(\text{configuration}) \propto \exp(-E/k_B T)$. We can represent this very probability distribution as a PEPS!. The amplitude of the wavefunction is replaced by the (unnormalized) probability of a classical configuration. The entanglement of the quantum state is replaced by the statistical correlation between classical variables. The local tensor no longer represents a projection from virtual qubits, but instead encodes the local Boltzmann weights or the update rules of the cellular automaton.

This realization is a profound unification. It connects the world of quantum many-body physics to classical statistical mechanics, computer science (where these structures are known as graphical models or Markov random fields), and even machine learning. The same mathematical toolkit used to find the ground state of a quantum magnet can be used to analyze the phase transitions of a classical fluid or to perform inference on a complex probabilistic model.

From a practical tool for simulating materials, to a design language for quantum computers, and finally to a universal framework for complexity, Projected Entangled Pair States reveal a deep truth about our world. They suggest that the structure of many complex systems, both quantum and classical, is fundamentally that of a network of local relationships. By understanding the Lego brick—the local tensor—we gain an unprecedented power to understand, and even to build, the intricate and wonderful castles they form.