⚛️ quantum physics

Lowering the temperature of two-dimensional fermionic tensor networks with cluster expansions

This paper extends the cluster expansion method to two-dimensional fermionic systems to construct projected entangled-pair operator approximations of Gibbs states, benchmarking truncation schemes and successfully resolving the phase boundary of a 2D spinless fermion model with attractive interactions.

Original authors: Sander De Meyer, Atsushi Ueda, Yuchi He, Nick Bultinck, Jutho Haegeman

Published 2026-02-26

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Sander De Meyer, Atsushi Ueda, Yuchi He, Nick Bultinck, Jutho Haegeman

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to predict the weather for a massive, complex city. You have a map of every street, every building, and every person. But the city is so huge, and the interactions between people so complicated, that if you try to calculate the weather for every single second, your computer would explode.

This is essentially the problem physicists face when trying to understand quantum materials (like superconductors) at finite temperatures. They want to know how electrons behave when things get hot, but the math is incredibly difficult because the electrons are "fermions" (a type of particle that hates being in the same place as another) and they interact with each other in a way that creates a "sign problem" (a mathematical nightmare that breaks standard simulation tools).

This paper introduces a new, smarter way to do these calculations. Here is the breakdown using everyday analogies:

1. The Old Way: The "Step-by-Step" Walk (Suzuki-Trotter)

Traditionally, to simulate how a system changes over time (or temperature), physicists use a method called Suzuki-Trotter decomposition.

The Analogy: Imagine you want to walk from your house to the moon. The old method says, "Take tiny, tiny steps." You take one step, then another, then another.
The Problem: To get to the moon (low temperature), you need billions of steps. If your steps are too big, you miss the path (inaccuracy). If your steps are too small, you take forever (computationally expensive). Also, every time you take a step, you have to break the symmetry of the city (the math gets messy), which can lead to errors piling up.

2. The New Way: The "Cluster Expansion" (The Smart Map)

The authors propose a new strategy called Cluster Expansion.

The Analogy: Instead of taking tiny steps, imagine you have a "smart map" that looks at groups of houses (clusters) at once. Instead of calculating the weather for one person, you calculate how a whole neighborhood interacts.
How it works: The math is broken down into "clusters" of connected particles. The method says, "Let's calculate the effect of a group of 2 neighbors, then a group of 3, then a group of 4."
The Benefit: This method respects the natural rules of the city (symmetries) and is much more accurate per "step" of time. It allows the physicists to take giant leaps toward the low-temperature state without losing their way. It's like taking a helicopter ride instead of walking step-by-step.

3. The "Compression" Problem (Truncation)

Even with the smart map, the data gets too big to handle. Every time you combine these "neighborhood" calculations, the amount of information (called bond dimension) explodes.

The Analogy: Imagine you are summarizing a 1,000-page book. If you try to keep every single word, your notebook gets too heavy to carry. You have to summarize (compress) the book to fit in your pocket, but you don't want to lose the plot.
The Solution: The paper tests three different ways to summarize this data:
1. Local Truncation: A quick, rough summary. It's fast and usually good enough.
2. Global Truncation: A very careful summary that looks at the whole book. It's accurate but takes a long time.
3. Variational Truncation: A "smart" summary that tries to find the absolute best version of the plot. It's the most accurate but requires the most computing power.

The Finding: The authors discovered that the "Local Truncation" (the quick summary) is actually the best balance. It's fast enough to run on standard supercomputers and accurate enough to give the right answer, whereas the fancy "smart summaries" were too slow to be practical for this specific job.

4. The Result: Solving the "Attractive" Mystery

The team applied this new method to a specific model of spinless fermions (electrons without spin) that attract each other.

Why it matters: In the world of quantum physics, "attractive" electrons are the key to understanding high-temperature superconductivity (materials that conduct electricity with zero resistance). However, standard computer simulations fail miserably here because of the "sign problem."
The Breakthrough: Using their new "Cluster Expansion" + "Local Truncation" method, they successfully mapped out the phase diagram of this material. They found the exact temperature where the material changes from a uniform state to a state where it separates into high-density and low-density regions (like oil and water separating).
The Proof: Their results matched perfectly with other known methods where those methods worked, but they succeeded in areas where other methods failed.

Summary

Think of this paper as inventing a new GPS for quantum physics.

Old GPS: Told you to take tiny, inefficient steps and often got lost in complex traffic (the sign problem).
New GPS (Cluster Expansion): Looks at the traffic patterns of whole neighborhoods, respects the road rules, and gets you to your destination (low temperature) much faster and more accurately.
The Outcome: They used this new GPS to finally see a clear picture of how certain quantum materials behave when they get cold, solving a puzzle that has been stuck for a long time.

This opens the door to simulating even more complex materials (like the famous Hubbard model) that could one day lead to room-temperature superconductors, revolutionizing our energy grid and technology.

1. Problem Statement

Studying two-dimensional (2D) fermionic many-body systems at finite temperatures is a significant challenge in condensed matter physics.

The Sign Problem: Quantum Monte Carlo (QMC) methods, the standard tool for finite-temperature simulations, often fail due to the notorious "sign problem," particularly in doped regimes or systems with attractive interactions.
Limitations of Current Tensor Network (TN) Methods: While TNs (specifically Projected Entangled Pair States/Operators, PEPS/PEPO) offer an alternative, standard approaches rely on the Suzuki-Trotter (ST) decomposition to discretize imaginary time evolution.
- ST decomposition often requires breaking spatial or internal symmetries.
- Higher-order ST decompositions become computationally expensive and complex.
- To reach low temperatures, many small time steps ( $\Delta\beta$ ) are required, leading to the accumulation of both Trotter discretization errors and truncation errors during the contraction of PEPO layers.

2. Methodology

The authors propose a framework that replaces the standard ST decomposition with a Cluster Expansion approach for constructing the time-evolution operator, specifically adapted for fermionic systems.

A. Fermionic Tensor Networks

The authors utilize the framework of super vector spaces to handle fermionic statistics. This involves:

Defining homogeneous vectors with parity (even/odd).
Implementing fermionic reordering (isomorphism) to encode anticommutation relations naturally within the tensor network structure.
Using the TensorKit.jl package, which handles these symmetries internally, though specific care is taken with sign factors during inner products and traces.

B. Cluster Expansion of the Evolution Operator

Instead of the standard ST product of 2-site gates, the evolution operator $e^{-\Delta\beta H}$ is expanded based on the maximal size of connected clusters ( $P$ ).

Concept: The Taylor expansion of the exponential is reorganized. Terms are grouped by the size of the largest disjoint cluster they act upon.
Construction: For a nearest-neighbor Hamiltonian, the expansion up to cluster size $P=3$ $P = 3$ (which corresponds to second-order accuracy in $\Delta\beta$ $Δ β$ ) is constructed using local tensors.
- This results in a translation-invariant, size-extensive PEPO.
- Crucially, this method preserves lattice and internal symmetries naturally.
- The bond dimension for the $P=3$ expansion is $D=5$ for both the Ising model and the spinless fermion model, which is comparable to or lower than higher-order ST decompositions for generic models.

C. Temporal Contraction and Truncation Schemes

To reach a target inverse temperature $\beta$ , multiple PEPO layers must be multiplied. Since the bond dimension grows exponentially ( $D^N$ ), intermediate truncation is necessary. The paper benchmarks three strategies:

Local Truncation: Constructs isometries via SVD/QR/LQ decompositions without considering the environment. It minimizes the local tensor difference.
Global Truncation: Uses the Corner Transfer Matrix Renormalization Group (CTMRG) to compute the environment iteratively, optimizing the isometries to minimize the global error.
Variational Truncation: Directly maximizes the fidelity (overlap) between the exact untruncated PEPO and the truncated PEPO using nonlinear optimization (e.g., L-BFGS) with CTMRG environments.

Selection: The authors found that while Variational Truncation offers the highest fidelity, the improvement over Local Truncation is marginal for the bond dimensions used, while the computational cost of the variational method is significantly higher. Therefore, Local Truncation was chosen for the final phase diagram calculations to allow for higher bond dimensions.

3. Key Contributions

Extension to Fermions: Successfully extended the cluster expansion formalism (previously applied to bosonic/spin systems) to 2D fermionic systems, handling the necessary sign factors and anticommutation relations.
Benchmarking Truncation Strategies: Provided a rigorous comparison of local, global, and variational truncation schemes for imaginary-time evolution, establishing that local truncation is the most efficient trade-off for this specific workflow.
Improved Low-Temperature Access: Demonstrated that the cluster expansion allows for larger imaginary time steps ( $\Delta\beta$ ) with lower discretization error compared to ST decomposition, enabling the simulation of significantly lower temperatures without prohibitive computational cost.
Open Source Implementation: Released the ClusterExpansions library, providing tools for constructing cluster expansion PEPOs for various models (Ising, Heisenberg, spinless fermions, t-J, Hubbard) on square and triangular lattices.

4. Results

The methodology was applied to a 2D spinless fermion model with nearest-neighbor attractive interactions ( $V < 0$ ) at half-filling.

Benchmark (Quantum Ising Model):
- Compared against ST decomposition and Simple Update (SU).
- The cluster expansion yielded a more accurate critical temperature ( $T_c$ ) and correlation length peak than ST decomposition, even with the same time step.
- It outperformed SU significantly, which suffered from smeared peaks and shifted transition temperatures.
Truncation Comparison (Free Fermions):
- Variational truncation achieved the highest fidelity, but Local truncation achieved fidelities very close to 1 for low bond dimensions ( $D=2,3$ ).
- Global truncation was found to be sensitive to initial guesses.
Phase Diagram (Attractive Fermions):
- The authors mapped the phase diagram for interaction strengths $V$ ranging from weak to intermediate.
- They identified a clear phase boundary between a homogeneous phase (high $T$ ) and a phase-separated state (low $T$ ).
- Validation: Results matched QMC data where the sign problem was manageable.
- Novelty: In the regime where QMC fails (stronger attractive interactions), the tensor network method provided reliable data, resolving the phase boundary where previous studies could not.

5. Significance

Overcoming the Sign Problem: This work provides a robust, sign-problem-free method to study finite-temperature phases of 2D fermionic systems with attractive interactions, a regime critical for understanding high- $T_c$ superconductivity and other correlated phenomena.
Efficiency: By utilizing cluster expansions, the method achieves higher accuracy per time step than standard ST decomposition, reducing the number of layers needed to reach low temperatures and mitigating error accumulation.
Scalability: The framework is generalizable to other lattices (triangular, honeycomb) and more complex models (t-J, Hubbard), paving the way for accurate simulations of low-temperature physics in strongly correlated electron systems that were previously inaccessible.
Methodological Insight: The paper clarifies that for finite-temperature TN simulations, the "cheaper" local truncation scheme is often sufficient, allowing resources to be allocated to increasing the bond dimension rather than optimizing the truncation environment, which is crucial for capturing critical phenomena.