Application and Performance Assessment of Annealing Methods for Electrostatic-Energy-Based Configuration Search in Mixed Crystals

Tack Saquai, Kenta Hongo, Ryo Maezono, Tom Ichibha · 2026

Read Application and Performance Assessment of Annealing Methods for Electrostatic-Energy-Based Configuration Search in Mixed Crystals by Tack Saquai, Kenta Hongo, Ryo Maezono, Tom Ichibha (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-24557.txt.

Introduction

Tack Saquai and three colleagues pre-screen mixed-crystal configurations with annealing, using Ewald electrostatic energy as the score to minimize. Site occupations become binary variables, mapping energy onto an Ising-type Hamiltonian and turning the search into a combinatorial optimization problem. On a small system, simulated annealing ran about 30 times faster than exhaustive search, and quantum annealing over 100 times faster, both finding all lowest-energy layouts. For medium and large systems, simulated annealing stayed 200 to 300 times faster and robust. Quantum annealing faltered on medium systems through chain breaks.

Catalog record

CREATORTack Saquai, Kenta Hongo, Ryo Maezono, Tom Ichibha
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE58218
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.24557

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Application and Performance Assessment of Annealing Methods for Electrostatic-Energy-Based Configuration Search in Mixed Crystals Tack Saquai, Kenta Hongo, Ryo Maezono, Tom Ichibha Abstract In first-principles design of solid solutions and disordered materials, evaluating the energies of all possible configurations is generally impractical because the number of substitutional-site occupations increases exponentially. In this study, we developed a framework that applies annealing methods to the pre-screening of mixed-crystal configurations using electrostatic energy, specifically Ewald energy, as the objective function, and systematically evaluated its effectiveness. The occupation states of substitutional sites were represented by binary 0/1 variables, and the Ewald energy was mapped onto an Ising-type Hamiltonian. This formulation allowed the search for low-electrostatic-energy configurations to be treated as a combinatorial optimization problem. The formulation was implemented using simulated annealing (SA) and quantum annealing (QA), and their performance was quantitatively assessed by comparison with exhaustive search in terms of speed-up factors and the ability to identify the lowest-energy structures. For the small-scale system CaYAlO 4 , SA achieved a speed-up of approximately 30 times, while QA achieved a speed-up of more than 100 times; both methods successfully captured the lowest-energy configurations without missing any of them. For the medium-scale system β \beta -KSbF 4 and the large-scale Ba-doped SiAlON system, SA achieved speed-ups on the order of 200–300 times while stably identifying the lowest-energy structures. In contrast, although QA was effective for the small-scale system, its speed-up was limited for medium-scale problems, and some low-energy configurations were missed due to chain breaks. These results demonstrate that, for mixed-crystal configuration search based on electrostatic energy, SA is currently the most robust and general-purpose method for rapid pre-screening and is practically applicable even to large-scale problems. The formulation presented in this study can be implemented automatically using publicly available libraries and, when integrated stepwise with first-principles calculations, provides a computational framework that can accelerate candidate-structure generation and property prediction for real materials systems. I Introduction In the materials design of solid solutions and disordered systems, properties are often tuned to a desired range by introducing dopants such as cations, anions, or vacancies 18 . When such systems are treated using first-principles calculations, one must confront the combinatorial number of possible elemental arrangements at the substitutional sites 22 ; 26 . This number can increase exponentially with the number of substitutional sites 26 . For example, in the structural model of Zn 0.5 Fe 2 Cu 0.5 O 4 reported in COD 9012442 10 , as many as 127,400 symmetry-distinct configurations are possible within the unit cell specified therein. In magnetic systems, the number can become even larger because substitutional sites occupied by the same ionic species may still be classified as inequivalent depending on their spin orientations, reaching as many as 1,229,107,765,600 possible configurations 34 . These configurations generally have different energy values 16 . However, in practical first-principles simulations, the configurations that should be considered are usually limited to a subset of patterns with sufficiently low and physically realizable energies 14 . The most straightforward procedure for identifying such configurations would be to perform structural relaxation using first-principles calculations for all possible substitutional-site occupation patterns, compare their energies, identify the most stable structures, and then proceed to property calculations. However, screening hundreds of millions or even trillions of structures by evaluating their energies one by one using first-principles calculations is not practically feasible. This naturally leads to a strategy in which configurations are screened not by using accurate energy values, but by using an alternative descriptor that can be evaluated much more rapidly. One possible descriptor for this purpose is the electrostatic energy 15 . The total energy of a material system is determined by the distribution of many interacting electrons on top of the classical electrostatic interaction energy arising from the ionic arrangement 25 . First-principles calculations are designed to treat the delicate balance associated with the former electronic contribution 29 ; 25 . In ionic crystals, however, the relative stability of possible substitutional-site configurations can often be governed primarily by the latter electrostatic interaction 15 ; 33 . This electrostatic contribution is commonly discussed in terms of the Ewald energy. The Ewald energy can be evaluated much faster than a first-principles calculation. For example, in the present system, its evaluation is approximately 43,000 times faster. Extrapolating this speed-up to all 127,400 configurations, the screening would require approximately 146,000 h using first-principles calculations, whereas it would take only about 3 h when based solely on Ewald-energy evaluations. Software packages that can rapidly estimate Ewald energies are widely available 23 ; 8 , and one might therefore expect that screening can simply be performed using such tools. However, even with this reduced evaluation cost, the computational time becomes prohibitive when realistic mixed-crystal materials are considered. For example, in the case of (Nd 0.7 Ce 0.225 La 0.075 ) 2 Fe 14 B, even if the evaluation time is only about 0.1 s per configuration, the total number of possible configurations reaches 1,229,107,765,600, resulting in an unrealistic computational time of approximately 34,000,000 h, or about 3,880 years 34 . Thus, an exhaustive search over all possible substitutional-site occupation patterns is impractical. Since our ultimate interest lies in substitutional structures with low energies, it is desirable to develop a method that efficiently eliminates high-energy substitutional configurations and selectively extracts low-energy ones. As a method for such selective extraction, we considered the application of annealing methods 2 . To implement an annealing method, it is necessary to formulate a model in which the objective function is expressed in terms of binary variables. This enables the optimization problem to be mapped onto the problem of searching for the most stable spin configuration. As described in § II , the occupation of substitutional sites can be represented by 0/1 binary variables, and the Ewald energy can be expressed as a function of these binary variables, which we refer to as the annealing model. By using this value as the objective function, the problem can be formulated as a combinatorial optimization problem in which one searches for a bit string that gives a lower Ewald energy. In general, combinatorial optimization problems are algorithmically challenging because they are often accompanied by multimodality 19 . Heuristic search methods for addressing such problems have therefore been actively studied. Examples include genetic algorithms 32 ; 1 , particle swarm optimization 13 ; 6 , and Bayesian optimization 39 ; 30 . For an objective function expressed in terms of binary variables, annealing techniques can be applied by mapping the problem onto a spin Hamiltonian. More specifically, the problem can be recast as the search for the spin configuration that realizes the lowest energy. This problem can be solved using methods such as quantum annealing 34 ; 14 and simulated annealing 19 . As described in § III , annealing-based extraction of low-energy substitutional structures does not necessarily guarantee complete coverage of all low-energy candidates. Although a previous study demonstrated that quantum annealing can mitigate this issue 2 , the extent of the speed-up achievable by annealing methods and their ability to recover all relevant low-energy structures without omission have not been quantitatively clarified. The present study therefore aims to evaluate these two aspects systematically.

Read the complete text