A Faster Algorithm for the Free Energy in One-Dimensional Quantum Systems

Samuel O. Scalet

2024-02-29

Abstract:We consider the problem of approximating the free energy density of a translation-invariant, one-dimensional quantum spin system with finite range. While the complexity of this problem is nontrivial due to its close connection to problems with known hardness results, a classical subpolynomial-time algorithm has recently been proposed [Fawzi et al., 2022]. Combining several algorithmic techniques previously used for related problems, we propose an algorithm outperforming this result asymptotically and give rigorous bounds on its runtime. Our main techniques are the use of Araki expansionals, known from results on the nonexistence of phase transitions, and a matrix product operator construction. We also review a related approach using the Quantum Belief Propagation [Kuwahara et al., 2018], which in combination with our findings yields an equivalent result.

Quantum Physics,Statistical Mechanics,Mathematical Physics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to efficiently approximate the free - energy density for one - dimensional translation - invariant quantum spin systems in the thermodynamic limit. Specifically, the author proposes a new method that goes beyond the existing sub - polynomial - time algorithms to calculate this physical quantity with a better time complexity.

### Problem Background

The free energy \(F = E - TS\) is an important physical quantity that describes the thermodynamic properties of a system under thermal equilibrium conditions. For one - dimensional quantum many - body systems, although there are already many successful algorithms that can handle free - energy calculations in certain specific cases, these algorithms are usually either heuristic or only applicable to cases without complexity - theory barriers. In particular, in one - dimensional systems, when the temperature is non - zero, there are known efficient algorithms; and in the high - temperature case, lattice systems also have similar efficient algorithms.

### Core Contributions of the Paper

This paper focuses on the thermodynamic - limit case (i.e., an infinitely large system) of one - dimensional translation - invariant systems and proposes a new classical algorithm whose running time is significantly improved compared to previous algorithms. The main technical means include:

1. **Araki Expansion**: Use the Araki expansion to deal with the existence problem of non - phase transitions.
2. **Matrix Product Operator (MPO)**: Calculate the partition function through MPO construction.
3. **Quantum Belief Propagation (QBP)**: Combine the QBP method to further verify the validity of the results.

### Main Results

The main result of the paper is a new classical algorithm. Its input is a Hermitian matrix \(h\in\mathbb{C}^{d^{2}\times d^{2}}\) and a precision parameter \(\epsilon\), and its output is an approximate value \(\tilde{f}\) of the one - dimensional translation - invariant free - energy density, satisfying:
\[|f - \tilde{f}|\leq\epsilon\]

The running time of the algorithm is:
\[ \exp \left( \sqrt{A \log \left( \frac{B}{\epsilon} \right) \log \left( A \log \left( \frac{B}{\epsilon} \right) \right)} \right) \]
where \(A\) and \(B\) are constants that only depend on the inverse temperature and the local dimension.

### Method Overview

1. **Partition Function Increment**: First, relate the global free - energy density to the increment of the partition function.
2. **MPO Construction**: Then use MPO construction to calculate these partition functions separately.
3. **Error Estimation**: Obtain the final result by combining these error estimates.

### Practical Applications

Although this algorithm has strict theoretical guarantees, in practical applications, due to the influence of parameter selection, it may need to be appropriately adjusted according to the computing resources to obtain good approximate results.

### Formula Summary

- Definition of free - energy density:
\[
f(h) = -\lim_{N \to \infty} \frac{1}{\beta N} \log(Z_N) = -\lim_{N \to \infty} \frac{1}{\beta N} \log \left( \text{Tr} \left[ e^{-\beta H[1,N]} \right] \right)
\]
- Error bound of the partition - function ratio:
\[
\left| e^{-\beta f(h)} - \frac{Z_{l + 1}}{Z_l} \right| \leq A e^{-l/\xi}
\]
- Algorithm time complexity:
\[
t \leq n \beta \exp \left( C \sqrt{\beta \log \left( \frac{l}{\epsilon_{\text{MPO}}} \right) \log \left( \beta \log \left( \frac{l}{\epsilon_{\text{MP}}} \right) \right)} \right)
\]