XtalOpt Version 14: Variable-Composition Crystal Structure Search for Functional Materials Through Pareto Optimization

Samad Hajinazar, Eva Zurek · 2025

Read XtalOpt Version 14: Variable-Composition Crystal Structure Search for Functional Materials Through Pareto Optimization by Samad Hajinazar, Eva Zurek (2025) 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-2506-17246.txt.

Introduction

Samad Hajinazar and Eva Zurek present version fourteen of XtalOpt in a 2025 paper. The software hunts for the best crystal forms a mix of atoms can take. Past runs held the mix fixed, which left many real materials out of reach.

The new release lets the mix vary as the search runs. It blends quantum methods with fast learned and classic energy models to relax each form. A Pareto method weighs rival goals at once, so no single score hides good trade offs. The paper tells users how to run the code and lists the other gains in this release.

Catalog record

CREATORSamad Hajinazar, Eva Zurek
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-17246

Browse the full library · About the Tower of Babel

Full text

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

XtalOpt Version 14: Variable-Composition Crystal Structure Search for Functional Materials Through Pareto Optimization Samad Hajinazar, Eva Zurek Abstract Version 14 of XtalOpt , evolutionary multi-objective global optimization algorithm for crystal structure prediction, is now available for download from its official website https://xtalopt.github.io . The new version of the code is designed to perform ground state search for novel crystal structures with variable composition by integrating a suite of ab initio methods alongside classical and machine-learning potentials for structural relaxation. The multi-objective search framework has been further enhanced through the introduction of Pareto optimization, enabling efficient discovery of functional materials. Here, we describe the implemented methodologies, provide detailed instructions for their use, and present an overview of additional improvements included in XtalOpt version 14. 1 Introduction Previous versions of XtalOpt [ 1 , 2 ] , available in both command-line interface (CLI) and graphical user interface (GUI) modes, performed evolutionary search on a chemical system specified by its reduced empirical formula, e.g., ⬇ empiricalFormula = Ti1O2 over an optionally determined list of formula units, e.g., ⬇ formulaUnits = 1-4 This resulted in a search over a “fixed chemical composition”. In this search, the initial set of structures are generated from the specified supercells of the given empirical formula. Then, the new structures are produced with the size and composition matching that of the initial list, by applying genetic operations to the selected structures from the parent pool. In this workflow, (i) the selection of the parent structure(s) is based on the fitness of structure, (ii) the fitness of a structure is determined from the normalized relative enthalpies (for a single-objective search) and -possibly- normalized relative value of user-defined objectives (in a multi-objective) search, (ii) the genetic operations include: “crossover” where two parent structure are mixed to generate an offspring with the same composition, and “stripple” and “permustrain” mutations that generate a new structure by applying random distortions to a selected parent structure. In the new version of XtalOpt (version 14), briefly, • Besides the generalized fitness function, the “Pareto optimization” scheme is implemented, • Instead of directly using the enthalpy to measure the structure fitness, the “distance above the convex hull” is used, • The genetic operation “crossover” is generalized and new evolutionary operations are introduced to allow for the generation of structures with new stoichiometries, and • It is possible to conduct the search over multiple compositions, or to generate new structures with compositions that are not listed in the input set of compositions. 2 Input chemical formulas In the new version of XtalOpt , introducing the chemical system and -optionally- desired formula units is different than the previous versions: (i) the “empiricalFormula” and “formulaUnits” pair of flags are removed, and (ii) the chemical system is now entirely specified by the entries of a single input flag “chemicalFormulas”. Generally, an input entry for “chemicalFormulas”: 1. Is interpreted as the “explicit” formula of the simulation unit cell (instead of the reduced empirical formula), and 2. Must include the “full chemical formula” (i.e., combination of element symbols and corresponding atom counts, e.g., “Ti1O4” instead of “TiO4”). That is to say that the input: ⬇ chemicalFormulas = Ti4O8 instructs XtalOpt to search in Ti-O system with “1:2” composition, while all generated unit cells have four Ti atoms and eight O atoms. Various “formula units” of a chemical system can be introduced using a comma-delimited list of full chemical formulas, e.g., ⬇ chemicalFormulas = Ti1O2 , Ti2O4 , Ti3O6 , Ti4O8 , Ti8O16 Besides explicitly listing various formula units in the chemical formulas input, they can also be combined into a hyphen-separated list of formulas in the form of a “ formula1 - formula2 ” entry. In such an entry, both ends must be full chemical formulas of the same composition, while the “formula2” is a proper supercell of the “formula1”. That’s, the number of atoms of each element type in “formula2” are a fixed integer multiple of those in “formula1” (e.g., “A3-A7” does not work, while “A3-A6” is accepted). As an example, the above input can be specified using a hyphen-separated entry as follows: ⬇ chemicalFormulas = Ti1O2 - Ti4O8 , Ti8O16 For comparison, the above input is corresponding to specifying the pair of “empiricalFormula = Ti1O2” and “formulaUnits = 1-4, 8” input flags in the previous versions of code. The “chemicalFormulas” parameter not only determine the initial cell(s) to be generated in the first generation, it also has an important role in specifying what type of the evolutionary search should be conducted; as detailed in the next section. 3 Various types of evolutionary searches Generally, the overall search workflow of the new version of XtalOpt is similar to the previous versions: initial set of structures are generated from the compositions and cell sizes specified in the input, and the genetic operations are applied to the selected parents from the pool to produce offspring. However, now, the details and outcome of the genetic operations are different. That’s, depending on the search type, the size and composition of the offspring produced by genetic operations can be different than that of their parent structure(s) or the input chemical formula list. The search type is determined by the details of the input formulas list and a set of new flags, as follows. 3.1 Fixed-composition search In the simplest case, if all the input formulas are from the same composition, the search is a “fixed-composition” (FC) evolutionary search, where the offspring have their composition match that of their parent structure(s) (i.e., initial list). This is essentially the traditional search performed by previous versions of XtalOpt . 3.2 Multi-composition search In the new version of XtalOpt , on the other hand, the initial list of chemical formulas can also include a combination of different compositions, e.g., ⬇ chemicalFormulas = Ti2O3 , Ti1O2 - Ti4O8 , Ti5O3 , Ti1 O1 - Ti5O5 This input instructs XtalOpt to perform a “multi-composition” (MC) search. This type of search is similar to the FC search in that initial set of structures are generated as usual and subsequent cells generated in the search are forced to have a composition matching that of “one of their parent structure(s)”. Since the initial list now covers various compositions, however, the parent structure(s) can be chosen from different compositions. This extends “desired” motifs found in one composition to other ones, potentially speeding up the search in finding best candidate structures for various compositions. 3.3 Variable-composition search Regardless of whether the input formulas include one or more compositions, if the input flag ⬇ vcSearch = true is specified, after the first generation is produced as usual and according to the input formulas list, genetic operation crossover is now allowed to produce cells with new compositions that are not necessarily match their parent structures’ composition or even those listed among the initial chemical formulas. This is called a “variable-composition” (VC) search: the most general type of evolutionary search conducted by XtalOpt where the entire chemical system of the specified elements is explored to find (meta)stable phases of various compositions. 4 Maximum number of atoms The new version of code allows for generating structures of new composition (hence, various number of atoms in the cell) and random supercells (discussed below).

Read the complete text