Thermal Tensor Network Simulations of Lattice Fermions with Fixed Filling

This paper introduces a fixed-particle-number tangent-space tensor renormalization group (tanTRG) algorithm that adaptively tunes the chemical potential during imaginary-time evolution to efficiently and accurately simulate finite-temperature correlated fermion systems, such as the Hubbard model, while overcoming the computational challenges of conventional chemical potential fine-tuning.

The Gist

The Big Picture: The "Crowded Party" Problem

Imagine you are trying to simulate a massive, chaotic party (a quantum system) where thousands of guests (electrons) are interacting, dancing, and bumping into each other. You want to understand how this party behaves when it's hot (high temperature) versus when it cools down.

In physics, this is called studying strongly correlated fermions. The goal is to figure out things like: Do the guests form dance circles? Do they freeze up? Do they start moving in waves?

The problem is that simulating this party is incredibly hard.

• The Old Way (The "Guess and Check" Method): Previously, scientists had to guess how many guests would show up. They would set a "chemical potential" (think of this as the price of entry or the vibe of the party). If they guessed the price wrong, the party would end up with too many or too few people. To fix this, they had to run the simulation, check the headcount, change the price, and run it again. If they wanted to see how the party changed as it cooled down while keeping the number of guests exactly the same, they had to repeat this tedious guessing game hundreds of times. It was slow, expensive, and frustrating.

The New Solution: The "Smart Bouncer"

The authors of this paper (Li, Qu, Chen, Shi, and Li) have invented a new algorithm called Fixed-N tanTRG.

Think of this algorithm as a super-smart bouncer who doesn't just stand at the door; they walk inside the party as it happens.

1. The Magic Trick (Imaginary Time): Instead of simulating time moving forward, they simulate the party "cooling down" in a special mathematical way called "imaginary time." It's like fast-forwarding a video of the party getting colder and more organized.
2. The Adaptive Bouncer: As the party cools, the bouncer constantly checks the headcount. If the number of guests starts to drift away from the target (say, 75% full), the bouncer instantly tweaks the "entry price" (chemical potential) right then and there to push the number back to the target.
3. The Result: They don't have to restart the simulation or guess. They just run the simulation once, and the bouncer keeps the crowd size perfectly steady the whole time.

Why This Matters: The "Stripe" Discovery

To prove their new bouncer works, they tested it on two things:

1. The Practice Run (Free Fermions): They simulated a simple party where guests don't talk to each other (non-interacting). The new method matched the perfect mathematical answer exactly, proving the bouncer is accurate.
2. The Real Challenge (The Hubbard Model): This is a complex party where guests do interact (they repel each other if they try to stand on the same spot). This is the model used to study high-temperature superconductivity (materials that conduct electricity with zero resistance).

What did they find?
They watched a "hole-doped" party (a party where some seats are empty). As the temperature dropped, they saw the guests start to organize themselves into stripes.

• Imagine the guests arranging themselves in alternating rows: Row of people, empty row, row of people, empty row.
• They also saw the "spin" (a quantum property like a tiny magnet) of the guests aligning in a specific pattern.

By keeping the guest count fixed, they could pinpoint exactly when these stripes formed. They identified three specific "temperature milestones":

• The High-Temperature Peak: When the guests stop double-upping on seats.
• The Magnetic Peak: When the guests start aligning their tiny magnets.
• The Stripe Peak: The exact moment the "striped" pattern locks into place.

The Takeaway

Before this paper, studying how these quantum materials behave at specific "filling levels" (how full the material is) was like trying to measure the temperature of a soup while constantly adding or removing water and having to guess how much to add.

This new method is like having a pot with a self-regulating valve. You set the water level you want, and the valve automatically adjusts the flow as the soup cooks, ensuring the level stays perfect the entire time.

This makes it much faster, cheaper, and more reliable for scientists to explore the mysterious world of high-temperature superconductors and other quantum materials, potentially helping us design better electronics and energy technologies in the future.

Technical Summary

1. Problem Statement

Simulating strongly correlated fermion systems at finite temperatures is a central challenge in computational condensed matter physics. While tensor network methods, specifically the Tangent-Space Tensor Renormalization Group (tanTRG), offer an efficient framework for finite-temperature simulations by representing thermal density operators as Matrix Product Operators (MPOs), they face a critical limitation:

• The Grand Canonical Ensemble (GCE) Issue: Standard tanTRG operates in the GCE, where the particle number is not fixed but fluctuates based on the chemical potential ($\mu$).
• The Filling Control Problem: In many physical contexts (e.g., high-$T_c$ superconductivity, topological phases), the filling fraction (particle density) is the intrinsic control parameter, not the chemical potential.
• Computational Bottleneck: To study a system at a specific fixed filling, researchers traditionally must perform a tedious, manual "sweep" of chemical potentials. This requires running multiple independent simulations to find the $\mu$ that yields the target particle number $N_{target}$. This process is computationally expensive, scales poorly with system size, and is particularly difficult in compressible states (metals/superconductors) where particle number is highly sensitive to $\mu$.

2. Methodology: Fixed-$N$ tanTRG Algorithm

The authors propose a Fixed-$N$ tanTRG algorithm that adaptively tunes the chemical potential during the imaginary-time evolution to stabilize the average particle number at a target value.

Key Technical Components:

• Imaginary-Time Evolution with Feedback: The algorithm integrates the chemical potential $\mu$ directly into the time-dependent variational principle (TDVP) evolution of the MPO. Instead of a static $\mu$, the algorithm treats $\mu$ as a time-dependent variable $\mu(\tau)$.
• Geometric Feedback Mechanism:
  • The evolution is viewed as a gradient descent on the MPO manifold.
  • The authors derive a feedback equation (Eq. 14) to update $\mu$ at each step based on the deviation of the current particle number $\langle N \rangle$ from the target $N_{target}$.
  • The update rule utilizes the Riemannian gradient of the particle number operator ($\nabla N$) and the energy operator ($\nabla E$). Specifically, $\mu$ is adjusted to satisfy:
    $$\mu = \frac{\delta\beta \langle \nabla N, \nabla E \rangle + 4(N_{target} - \langle N \rangle)}{\delta\beta \langle \nabla N, \nabla N \rangle}$$
  • This creates a negative feedback loop that traps the particle number around the target value.
• Correction Step: If the particle number drifts beyond a preset tolerance after a TDVP sweep, a Newton iteration method using Time-Evolving Block Decimation (TEBD) is applied to restore the target filling.
• Thermodynamic Quantities: The method allows for the direct calculation of the constant-particle-number specific heat ($C_N$) and other thermodynamic quantities without needing to differentiate with respect to $\mu$ post-simulation.

3. Key Contributions

1. Algorithmic Innovation: The first implementation of a fixed-filling constraint in thermal tensor network simulations (specifically tanTRG) using an adaptive chemical potential scheme.
2. Efficiency: Eliminates the need for multiple independent simulations with different chemical potentials. The computational overhead is estimated to be only $1/(2K)$ of the baseline TDVP cost (where $K$ is the Krylov space dimension, typically $\sim 10$), making it significantly more efficient than traditional sweeping methods.
3. Theoretical Framework: Provides a rigorous geometric interpretation of the fixed-$N$ constraint within the MPO manifold, linking the chemical potential tuning to the Riemannian metric of the tangent space.
4. Open Source: The authors released the source code (Fixed-N-tanTRG) and data, facilitating reproducibility and further development.

4. Results and Benchmarks

A. Benchmark on Non-Interacting Fermions

• System: Spinless fermion chain (exactly solvable).
• Performance: The algorithm accurately reproduced thermodynamic quantities (Energy $E$, Entropy $S$, Specific Heat $C_N$) and the Equation of State (EoS) across a wide temperature range.
• Accuracy: Results matched exact solutions with errors systematically decreasing as the bond dimension ($D$) increased. For $D=1024$, the dominant error was the controlled deviation from the target filling ($10^{-6}$), confirming the stability of the feedback mechanism.
• Susceptibility: Successfully captured charge susceptibility $\chi_c$, including van Hove singularities at band edges.

B. Application to Square-Lattice Hubbard Model

• System: Hole-doped Hubbard model ($\delta = 1/12$) on a $4 \times 24$ cylinder.
• Validation: At high temperatures ($T \ge 0.5$), results for energy per site matched Determinant Quantum Monte Carlo (DQMC) simulations. At low temperatures, results smoothly converged to DMRG ground-state energies.
• Stripe Formation: The study identified the temperature evolution of charge and spin stripes:
  • Charge Density Wave (CDW): Emerged with a wavelength $\lambda_{CDW} = 6$ (half-filled stripe).
  • Spin Density Wave (SDW): Developed a $\pi$-phase shift locking to charge minima.
• Characteristic Temperature Scales: The authors identified and quantified four distinct temperature scales:
  1. $T_h$ (High-T): Associated with the suppression of double occupancy (proportional to $U$).
  2. $T_m$ (Magnetic): Onset of short-range antiferromagnetic correlations (proportional to $1/U$).
  3. $T^*_{SDW}$: Onset of SDW order.
  4. $T_{stripe}$: The temperature where CDW and SDW patterns lock together (stripe formation).
• Scaling Behavior: $T_m$ showed a clear $1/U$ dependence, while $T_{stripe}$ and $T^*_{SDW}$ showed weaker sensitivity to $U$, suggesting stripe formation is governed by a balance between kinetic energy ($t$) and magnetic energy ($t^2/U$).

5. Significance

• Enabling New Physics: This method removes a major bottleneck in studying finite-temperature phase diagrams of correlated materials where filling is the primary control parameter. It allows researchers to trace the evolution of physical quantities (like specific heat and correlations) at a fixed doping level, which was previously computationally prohibitive.
• Reliability: By establishing the fixed-$N$ tanTRG as a reliable tool, the paper opens the door to more accurate investigations of high-$T_c$ superconductivity, pseudogap phenomena, and topological phases in twisted materials.
• Generalizability: The strategy of adaptively tuning conjugate fields to fix conserved quantities is not limited to tanTRG; it can be extended to other thermal tensor network methods (e.g., PEPS) and other conserved quantities (e.g., magnetization).

In summary, this work provides a robust, efficient, and theoretically sound algorithm for simulating finite-temperature quantum many-body systems at fixed filling, significantly advancing the capability to explore complex emergent phenomena in strongly correlated electron systems.