Projected Entangled Pair States for Lattice Gauge Theories with Dynamical Fermions

This paper demonstrates the feasibility of using gauged Gaussian projected entangled pair states as an ansatz to study a two-dimensional $\mathbb{Z}_2$ lattice gauge theory with dynamical fermions, successfully reproducing exact diagonalization results for small systems and offering a computationally tractable approach for larger systems that avoids the sign problem.

The Gist

The Big Picture: The "Sign Problem" and the New Map

Imagine you are trying to predict the weather for a massive, chaotic city. In physics, this "city" is a Lattice Gauge Theory—a mathematical model used to understand how particles (like electrons) and forces (like magnetism) interact.

For decades, scientists have tried to simulate these cities using a method called Monte Carlo sampling. Think of this like taking a million random photos of the city to guess the average weather. It works great for simple cities. But for some complex cities (specifically those with "dynamical fermions," or moving matter particles), the math gets broken. The "photos" start having negative or imaginary values, making the average impossible to calculate. This is the infamous "Sign Problem." It's like trying to calculate the average temperature of a room where some thermometers read +20°C and others read -20°C, but the math says the -20°C is actually a "negative temperature" that breaks the calculator.

The Solution: This paper introduces a new way to map these cities without taking random photos. Instead, the authors built a specialized blueprint called GGPEPS (Gauged Gaussian Projected Entangled Pair States).

The Blueprint: Building a Lego City with Invisible Rules

To understand GGPEPS, imagine you are building a Lego city, but you have to follow two very strict rules:

1. The Lego pieces must snap together perfectly (this represents the laws of physics).
2. You must obey a secret local rule: Every time you place a red brick, you must place a blue brick next to it in a specific pattern. If you break this pattern, the whole structure collapses. This is Gauge Invariance.

In the past, building these cities was hard because if you tried to build a big one, the number of ways the bricks could snap together became too huge to count.

The Authors' Trick:
Instead of trying to count every possible Lego arrangement, they built a smart, flexible mold.

• Virtual Threads: They imagine invisible threads connecting the Lego bricks. These threads hold the structure together.
• The "Gauging" Step: They weave these threads through the bricks in a very specific way that forces the secret local rule to be obeyed automatically. You don't have to check every brick; the threads ensure the rule is never broken.
• The Gaussian Shape: They shaped the mold so that it naturally favors the most stable, low-energy configurations (the "ground state"). It's like a mold that naturally settles into the most comfortable shape without you having to push it.

The Experiment: Testing the Blueprint

The authors tested this new blueprint on a Z2 Lattice Gauge Theory.

• The Setup: Imagine a 2D grid (like a chessboard). On the squares, you have "matter" (fermions). On the lines connecting the squares, you have "gauge fields" (forces).
• The Test: They tried to find the most stable state (the ground state) of this grid.
• The Comparison:
  • For tiny grids (2x2 and 4x4), they could solve the problem exactly (like solving a small Sudoku puzzle by hand).
  • For larger grids (6x6 and beyond), exact solutions are impossible (the puzzle is too big).

The Result:
Their new blueprint matched the exact solutions perfectly on the small grids. When they scaled it up to the larger grids where they couldn't check the answer, the blueprint still behaved logically and smoothly. It successfully captured the physics of the system, including how the matter and forces interact.

Why This Matters: Avoiding the "Sign Problem"

The most exciting part is that this method doesn't care about the "Sign Problem."

• Old Way (Monte Carlo): Like trying to walk through a foggy forest by taking random steps. If the fog gets too thick (the sign problem), you get lost and can't find the path.
• New Way (GGPEPS): Like having a GPS that draws the path directly based on the terrain's shape. It doesn't need to take random steps; it constructs the path logically.

Because this method avoids the sign problem, it opens the door to studying much more complex theories, including Quantum Chromodynamics (QCD)—the theory that holds the nucleus of an atom together. Currently, we can't simulate QCD with matter on a computer because of the sign problem. This paper is a "proof of concept" that says: "Hey, we found a way to build the map for the simpler version of this city. Now we can use this same map-making technique to tackle the really hard ones."

Summary in a Nutshell

1. The Problem: Simulating certain quantum systems is impossible with current computer methods because the math breaks down (the Sign Problem).
2. The Tool: The authors created a new type of "smart blueprint" (GGPEPS) that builds the system using invisible threads to enforce physical laws automatically.
3. The Test: They built a small model of a quantum city with moving particles. Their blueprint worked perfectly, matching known answers and scaling up to sizes where no one else could get an answer.
4. The Future: This is a major step toward simulating the fundamental forces of the universe on a computer, potentially helping us understand how the universe works at its most basic level, without getting stuck in the math fog.

Analogy Conclusion:
If the universe is a giant, complex video game, current computers are trying to play it by guessing every frame randomly, which crashes the game when the graphics get too complex. This paper introduces a new "engine" that renders the game frame-by-frame using smart rules, allowing us to finally play the high-definition levels that were previously impossible.

Technical Summary

1. Problem Statement

Lattice Gauge Theories (LGTs) are fundamental to understanding the Standard Model and condensed matter physics. However, studying these theories, particularly those involving dynamical fermions, is notoriously difficult using conventional Monte Carlo methods due to the sign problem. The sign problem arises when the probability distribution used in sampling becomes negative or complex, rendering standard importance sampling invalid.

While tensor network methods (like Matrix Product States in 1D) have successfully avoided the sign problem, extending them to higher dimensions (2D and 3D) with dynamical matter remains challenging. Existing approaches often struggle with computational scaling or lack specific tailoring for local gauge symmetries. The authors aim to demonstrate a viable, computationally feasible method for studying 2D lattice gauge theories with dynamical fermions that avoids the sign problem and respects local gauge invariance.

2. Methodology

The authors employ Gauged Gaussian Projected Entangled Pair States (GGPEPS) as a variational ansatz. This approach combines the efficiency of Gaussian states with the structural requirements of gauge theories.

A. Physical System

• Model: A $Z_2$ lattice gauge theory on a 2D square lattice.
• Degrees of Freedom:
  • Gauge Fields: Located on links, represented by $Z_2$ group elements (Pauli matrices $\sigma_z, \sigma_x$).
  • Matter: A single flavor of staggered fermions located on lattice sites. Staggering is used to reduce the fermion doubling problem, placing matter and anti-matter components on even and odd sublattices respectively.
• Hamiltonian: The system is governed by the Kogut-Susskind Hamiltonian ($H = g_E H_E + g_B H_B + g_I H_I + g_M H_M$), including electric, magnetic, interaction, and mass terms.
• Symmetries: The system must satisfy local $Z_2$ gauge invariance, global $U(1)$ charge conservation (at half-filling), and lattice rotation/translation symmetries.

B. The Ansatz: Gauged Gaussian PEPS

The authors construct the wavefunction $|\psi\rangle$ as a superposition over gauge field configurations $G$:
$$|\psi\rangle = \sum_G \psi_I(G) |\psi_{II}(G)\rangle |G\rangle$$

• $\psi_I(G)$: A complex amplitude depending on the gauge configuration.
• $|\psi_{II}(G)\rangle$: A state of the fermionic matter that depends on the gauge configuration to ensure gauge invariance.

Construction Steps:

1. Particle-Hole Transformation: A transformation is applied to the odd sublattice to make the system translationally invariant and simplify the handling of anti-matter.
2. Virtual Modes: For each site, 16 virtual modes are introduced (4 per link direction). These are divided into two sets:
   • Two copies used for constructing the amplitude $\psi_I(G)$.
   • Two copies plus the physical fermion used for constructing the matter state $|\psi_{II}(G)\rangle$.
3. Gaussian Operators: The virtual and physical modes are coupled via Gaussian operators ($A(x) = \exp(T_{ij} \alpha^\dagger_i \alpha^\dagger_j)$). The matrix $T$ is constrained to respect all symmetries (gauge, $U(1)$, rotation).
4. Gauging: A gauging operator $U_G$ couples the gauge fields on links to the virtual modes. This ensures that the resulting state is an eigenstate of the local gauge generator (satisfying Gauss's law).
5. Projection: Virtual modes on neighboring sites are projected onto maximally entangled states. The virtual modes are then traced out, leaving a physical state with only gauge fields and fermions.

C. Algorithm

• Variational Monte Carlo (VMC): The ground state is found by minimizing the energy expectation value with respect to the parameters in the matrix $T$.
• Efficiency:
  • For a fixed gauge configuration $G$, the state is Gaussian. This allows all observables to be computed efficiently using the covariance matrix formalism (scaling polynomially with system size).
  • The probability distribution $p(G) \propto |\psi_I(G)|^2 \langle \psi_{II}(G) | \psi_{II}(G) \rangle$ is sampled using Monte Carlo.
  • Local updates to the gauge configuration allow for efficient re-calculation of the covariance matrix without recomputing from scratch.

3. Key Contributions

1. Inclusion of Dynamical Fermions: This is the first demonstration of GGPEPS applied to a lattice gauge theory with dynamical fermionic matter (previously limited to pure gauge theories).
2. Gauge Invariance by Construction: The ansatz explicitly enforces local gauge invariance through the gauging procedure, eliminating the need for penalty terms or post-selection.
3. Computational Feasibility in 2D: The method successfully handles systems up to $6 \times 6$ lattices, a regime where exact diagonalization is computationally prohibitive (requiring $>100$ GB of RAM for $4 \times 4$).
4. Avoidance of Sign Problem: By using a tensor network ansatz rather than path-integral Monte Carlo, the method naturally bypasses the sign problem associated with fermionic LGTs.

4. Results

• Validation against Exact Diagonalization:
  • For $2 \times 2$ and $4 \times 4$ lattices, the GGPEPS results for ground state energy and individual energy terms (electric, magnetic, interaction, mass) show excellent agreement with exact diagonalization results.
  • The method correctly captures the physics of the ground state, not just the total energy.
• Scaling to Larger Systems:
  • The authors successfully computed ground state energies and Wilson loops for $6 \times 6$ lattices using Monte Carlo sampling.
  • Wilson Loops: The study of $1 \times 1$ and $2 \times 2$ Wilson loops reveals sharp transitions in their values as coupling constants vary, suggesting phase transitions between confined and deconfined phases (though Wilson loops are not strict order parameters in the presence of dynamical matter).
• Free Fermion Limit: In the limit of vanishing gauge coupling, the method reproduces the expected Dirac-like band structure and ground state energies of free fermions on a torus, validating the ansatz's ability to capture continuum physics.

5. Significance and Future Outlook

• Path to QCD: This work represents a critical step toward simulating more complex theories, such as Quantum Chromodynamics (QCD) in $(3+1)$ dimensions, which suffer severely from the sign problem.
• Generalizability: The framework is designed to be generalizable to other gauge groups (e.g., $SU(2)$, $SU(3)$) and theories with multiple fermion flavors and chemical potentials.
• Efficiency: The combination of Gaussian PEPS (for efficient observable calculation) and Monte Carlo (for summing over gauge configurations) offers a scalable route to studying strongly correlated gauge theories where traditional methods fail.

The authors conclude that while the current $Z_2$ model does not suffer from a sign problem, the successful application of GGPEPS to dynamical fermions proves the viability of the approach for tackling the sign problem in more complex, realistic models. Future work will focus on applying this method to models with multiple fermion flavors and different chemical potentials.