Abstract:Ising machines have attracted attention as efficient solvers for combinatorial optimization problems, which are formulated as ground-state (lowest-energy) search problems of the Ising model. Due to the limited bit-width of coefficients on Ising machines, the Ising model must be transformed into a bit-width reduced (BWR) Ising model. According to previous research, the bit-width reduction method, which adds auxiliary spins, ensures that the ground state of the BWR Ising model is theoretically the same as the Ising model before bit-width reduction (original Ising model). However, while the dynamical process is closely related to solution accuracy, how the BWR Ising model progresses towards the ground state remains to be elucidated. Therefore, we compared the dynamical processes of these models using simulated annealing (SA). Our findings reveal significant differences in the dynamical process across models. Analysis from the viewpoint of statistical mechanics found that the BWR Ising model has two characteristic properties: an effective temperature and a slow relaxation. These properties alter the temperature schedule and spin flip probability in the BWR Ising model, leading to differences in the dynamical process. Therefore, to obtain the same dynamical process as the original Ising model, we proposed SA parameters for the BWR Ising model. We demonstrated the proposed SA parameters using a square lattice Ising model, in which all coefficients were set uniformly to the same positive values or randomly. Our experimental evaluations demonstrated that the dynamical process of the BWR and original Ising model became closer.

What problem does this paper attempt to address?

### Main Problem Addressed by the Paper The paper primarily addresses a key issue encountered when using bit-width-restricted Ising machines to solve combinatorial optimization problems: how to ensure the dynamic equivalence between the bit-width-reduced Ising model (BWR Ising model) and the original Ising model. ### Overview of Core Contributions 1. **Analysis of the Dynamic Process of the BWR Ising Model**: - Using the Simulated Annealing (SA) algorithm, the dynamic process differences between the original Ising model and the BWR Ising model were compared. - It was found that the BWR Ising model has two significant characteristics: effective temperature and slow relaxation. These characteristics stem from the entropy effect of auxiliary spins, which do not exist in the original Ising model. 2. **Proposed Improved SA Parameters**: - To make the dynamic process of the BWR Ising model identical to that of the original Ising model, the authors proposed methods to adjust the SA parameters, including modifying the temperature schedule and the number of inner loops. - The adjustment of the temperature schedule takes into account the effective temperature of the BWR Ising model; the adjustment of the number of inner loops is based on the flip probability to achieve dynamic behavior similar to that of the original Ising model. ### Experimental Validation - Experimental evaluation was conducted using a square lattice random Ising model, and the results showed that with the proposed SA parameters, the dynamic process of the BWR Ising model became very close to that of the original Ising model. - The experimental results confirmed the effectiveness of the proposed method and demonstrated that even with larger system sizes, the BWR Ising model can achieve a dynamic process similar to that of the original model. ### Conclusion By deeply analyzing the statistical mechanics characteristics of the BWR Ising model, this paper successfully revealed the dynamic differences between it and the original Ising model and proposed effective solutions to overcome these differences. This provides important theoretical foundations and technical support for efficiently solving combinatorial optimization problems using bit-width-restricted digital Ising machines.

Dynamical process of a bit-width reduced Ising model with simulated annealing

Mathematical Mechanism on Dynamical System Algorithms of the Ising Model

Thermodynamic quantities of two-dimensional Ising models obtained by noisy mean-field annealing and a coherent Ising machine

Annealing Processing Architecture of 28-nm CMOS Chip for Ising Model With 512 Fully Connected Spins

Hybrid Optimization Method Using Simulated-Annealing-Based Ising Machine and Quantum Annealer

E-Spin: A Stochastic Ising Spin Based on Electrically-Controlled MTJ for Constructing Large-Scale Ising Annealing Systems.

ON CONNECTION AMONG QUANTUM-INSPIRED ALGORITHMS OF THE ISING MODEL

A tree search algorithm towards solving Ising formulated combinatorial optimization problems

Bound on annealing performance from stochastic thermodynamics, with application to simulated annealing

Modeling a frustrated Ising square lattice with the D-Wave Quantum Annealer

Simulated Ising Annealing Algorithm with Gaussian Augmented Hamiltonian Monte Carlo

Designing Unit Ising Models for Logic Gate Simulation through Integer Linear Programming

Noise-augmented Chaotic Ising Machines for Combinatorial Optimization and Sampling

Predicting Ising Model Performance on Quantum Annealers

Simulated bifurcation assisted by thermal fluctuation

Critical behavior of Ising model by preparing thermal state on quantum computer

Greedy parameter optimization for diabatic quantum annealing

Resampling schemes in population annealing: Numerical and theoretical results

Simulation of Dynamical Quantum Phase Transition of the 1D Transverse Ising Model with a Double-chain Bose-Hubbard Model

Effectiveness of quantum annealing for continuous-variable optimization

Improving quantum annealing by engineering the coupling to the environment