Dara: Automated multiple-hypothesis phase identification and refinement from powder X-ray diffraction
Yuxing Fei, Matthew J. McDermott, Christopher L. Rom, Shilong Wang, Gerbrand Ceder · 2025
Read Dara: Automated multiple-hypothesis phase identification and refinement from powder X-ray diffraction by Yuxing Fei, Matthew J. McDermott, Christopher L. Rom, Shilong Wang, Gerbrand Ceder (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-2510-19667.txt.
Introduction
Yuxing Fei and four co-authors present Dara, a system that automates the interpretation of powder X-ray diffraction patterns for materials with several crystal phases at once. Reading such patterns is still largely manual and expert-driven, and because diffraction gives only structural information, several phase combinations can fit one pattern, inviting mistakes. Dara searches exhaustively over plausible phase combinations within a chemical space.
Each candidate is checked with a robust Rietveld refinement routine, with database filtering and clustering of structurally similar phases keeping the search manageable. When the evidence stays ambiguous, Dara reports multiple hypotheses for experts or other instruments to decide between. The goal is reliable phase identification that scales to realistic, complex patterns and self-driving laboratories.
Catalog record
| CREATOR | Yuxing Fei, Matthew J. McDermott, Christopher L. Rom, Shilong Wang, Gerbrand Ceder |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 74 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-19667
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Dara: Automated multiple-hypothesis phase identification and refinement from powder X-ray diffraction Yuxing Fei, Matthew J. McDermott, Christopher L. Rom, Shilong Wang, Gerbrand Ceder Abstract Powder X-ray diffraction (XRD) is a foundational technique for characterizing crystalline materials. However, the reliable interpretation of XRD patterns, particularly in multiphase systems, remains a manual and expertise-demanding task. As a characterization method that only provides structural information, multiple reference phases can often be fit to a single pattern, leading to potential misinterpretation when alternative solutions are overlooked. To ease humans’ efforts and address the challenge, we introduce Dara (Data-driven Automated Rietveld Analysis), a framework designed to automate the robust identification and refinement of multiple phases from powder XRD data. Dara performs an exhaustive tree search over all plausible phase combinations within a given chemical space and validates each hypothesis using a robust Rietveld refinement routine (BGMN). Key features include structural database filtering, automatic clustering of isostructural phases during tree expansion, peak-matching-based scoring to identify promising phases for refinement. When ambiguity exists, Dara generates multiple hypothesis which can then be decided between by human experts or with further characteriztion tools. By enhancing the reliability and accuracy of phase identification, Dara enables scalable analysis of realistic complex XRD patterns and provides a foundation for integration into multimodal characterization workflows, moving toward fully self-driving materials discovery. † † affiliation: Authors contribute equally † † affiliation: Department of Materials Science & Engineering, University of California, Berkeley, CA 94720, USA † † affiliation: Materials Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA † † affiliation: Authors contribute equally † † affiliation: Materials Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA † † affiliation: Materials Science Center, National Renewable Energy Laboratory, Golden, CO, 80401, USA † † affiliation: Department of Materials Science & Engineering, University of California, Berkeley, CA 94720, USA † † affiliation: Materials Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA † † affiliation: Department of Materials Science & Engineering, University of California, Berkeley, CA 94720, USA † † affiliation: Materials Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA † † email: [email protected] 1 Introduction Accurate characterization of material structures is essential for understanding synthesis-structure-property relationships in materials. For bulk inorganic materials, powder X-ray diffraction (XRD) has long been a pivotal and widely used technique for determining majority-phase crystal structures Waseda et al. 2011 ; Ward et al. 2017 . Moreover, powder XRD plays a critical role in the discovery of inorganic materials, serving as a key tool for confirming the synthesis of predicted target structures, such as those derived from ab initio methods Narayan et al. 2016 ; Shoemaker et al. 2014 . With appropriate analysis, XRD patterns can provide valuable insights into material properties, including phase fractions, lattice parameters, strains, site occupancies, and more McCusker et al. 1999 . XRD analysis typically begins with the identification of all phases present in the pattern. This process involves comparing the experimental pattern with the calculated patterns of structures available in crystal structure databases like the Materials Project (MP) Jain et al. 2013 , Inorganic Crystal Structure Database (ICSD) Zagorac et al. 2019 , Powder Diffraction File (PDF) Kabekkodu et al. 2024 , and Crystallography Open Database (COD) Gražulis et al. 2012 . The task becomes practically challenging when a sample contains multiple phases that cannot be perfectly matched to reference structures. This is common when characterizing the synthesis products of exploratory inorganic reactions or natural minerals, which may exhibit compositional variance or preferred orientation effects. In these cases, accurate interpretation requires meticulous analysis and the expertise of researchers who are intimately familiar with the material system, enabling them to discern its subtle nuances. In recent years, the development of automated and autonomous laboratories for the discovery of inorganic materials has underscored the need to accelerate the characterization process Szymanski et al. 2023 ; Lunt et al. 2024 ; Chen et al. 2024 ; Yotsumoto et al. 2024 . As the throughput of synthesis and characterization continues to increase, human analysis of patterns has become impractical, further emphasizing the importance of automation. The integration of reliable, robust, and accurate powder XRD phase identification algorithms within autonomous laboratories will be crucial to establishing high-quality experimental databases for inorganic materials. In self-driving labs, high-quality characterization of samples at the early stage of synthesis optimization is of particular importance in the AI-driven decision-making algorithms, which may be more challenged by erroneous interpretations than human experts are. Algorithms for XRD phase analysis have captured the interest of researchers for nearly a century. In 1938, Hanawalt Hanawalt et al. 1938 introduced a qualitative method for phase identification based on major diffraction peaks of known phases. Using this approach, researchers can refer to the “Hanawalt Index”, a tabulated collection of peak data, to identify potential phases. If three major peaks align with those of a specific phase, it strongly suggests the presence of that phase in the sample. This method is also known as the “search-match” method, as it always involves searching the reference database and then matching the phase to the diffraction pattern. Following Hanawalt’s manual search, several computer programs are available for automated peak indexing Smith and Gorter 1991 ; Lin et al. 1993 ; Dinnebier and Billinge 2015 , each with carefully fine-tuned strategies to identify phases. With more peaks taken into consideration and user-friendly interfaces, these programs enable researchers to analyze XRD patterns with rigor and ease. This makes them the dominant method for phase analysis due to their straightforward nature and low computational resource requirements. With the growing trend of applying deep learning to XRD analysis, numerous studies have leveraged neural networks (NNs) to automate the interpretation of diffraction patterns. Oviedo et al. Oviedo et al. 2019 used a convolutional neural network (CNN) with simulated and experimental XRD data to classify crystal dimensionality and space group, achieving 93% and 89% accuracy, respectively, by augmenting limited data with physics-informed simulations. Lee et al. tackled multi-phase identification by training a deep CNN on over 1.7 million synthetic mixed XRD patterns (combinations of 170 compounds in the Sr-Li-Al-O quaternary system), enabling near-100% phase identification in complex mixtures and even quantifying phase fractions with 86% accuracy. Maffettone et al. Maffettone et al. 2021 employed an ensemble of CNNs (a “crystallography companion agent”) that outputs probabilistic phase predictions, avoiding combinatorial explosion in training while providing confidence metrics for each identified phase. Szymanski et al. combined physics-informed peak perturbation augmentation with an ensemble CNN and a branching algorithm to iteratively identify phases in mixtures. This probabilistic approach achieved higher accuracy than traditional profile-matching and earlier deep-learning methods on challenging multi-phase samples. More recently, researchers have created a new NN architecture for the XRD phase identification task. For example, Zhang et al. Zhang et al. 2024 introduced a self-attention CNN (CPICANN) trained on around 700k simulated patterns (from 23k structures), which attained 98.5% accuracy on single-phase identification with element information provided and 80% on experimental scans, significantly outperforming conventional XRD software. Beyond deep NNs, other machine-learning approaches have also been explored. For example, Suzuki et al. Suzuki et al. 2020 used an interpretable tree-ensemble model to classify crystal systems and space groups with 90% accuracy, revealing human-understandable diffraction features. Alongside