XMatcher: An Open-Source Framework for X-Ray Diffraction Phase Identification

Bin Cao

REVIEW 3 major objections 5 minor 16 references

XMatcher: An Open-Source Framework for X-Ray Diffraction Phase Identification

T0 review · 3 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read XMatcher makes XRD phase identification a locally runnable, evidence-traceable search-match pipeline.

desk verdict XMatcher is a genuine, openly inspectable XRD search-match tool with honest limitations, but the preprint contains no benchmark or worked example, so its central single-shift alignment assumption remains unvalidated. read the letter →

arxiv 2607.17162 v1 pith:X35C3TSO submitted 2026-07-19 cond-mat.mtrl-sci

Bin Cao This is my paper · ORCID

classification cond-mat.mtrl-sci

keywords X-raydiffractionphaseidentificationsearch-matchmultiphaseanalysisopen-sourcesoftwarenon-negativeleastsquaresinterpretableevidencewholepatterncomparison

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper introduces XMatcher, an open-source, locally run pipeline for identifying crystalline phases from powder X-ray diffraction. It argues that reliable phase assignment needs an unbroken, inspectable chain from measured peaks to candidate structures, rather than a single similarity score. XMatcher builds theoretical peak libraries from crystal-structure records, retrieves candidates under chemical filters and a bounded global angular shift, performs one-to-one peak matching, and reports both supporting and conflicting evidence. Its AutoMix module extends this to mixtures by enumerating candidate phase combinations and fitting non-negative diffraction contributions. If the approach holds, laboratories gain a reproducible, transparent alternative to closed commercial search-match packages.

What carries the argument

The central object is the structured theoretical peak library: each structure record is reduced to its thirty strongest calculated reflections under a chosen radiation and angular window, with positions, relative intensities, reflection families and d-spacings, plus linked full-pattern PDF/CIF records. The matching engine uses a bounded global angular shift per candidate and a one-to-one linear-sum assignment between experimental and theoretical peaks, so no broad peak can support multiple reflections. AutoMix builds a response matrix from these shifted one-to-one assignments and solves a non-negative least squares problem to estimate component contributions. This machinery keeps every step—

What would settle it

Take a well-characterized two-phase pellet with known texture or residual strain, measure its XRD pattern, run XMatcher in AutoMix mode, and compare the ranked phases and NNLS weights against Rietveld-refined phase fractions. If the true majority phase is not ranked first or the fitted weights deviate from refined fractions beyond tolerance, the compact-library-plus-global-shift representation is the point of failure.

Watch

Extended reading notes

Core claim

XMatcher's central claim is that XRD phase identification can be made reproducible and interpretable by structuring it as a staged, locally auditable pipeline. Each candidate phase is returned with assigned experimental-to-theoretical peak pairs, a fitted global shift, coverage statistics, missing theoretical peaks and unexplained experimental peaks; ranking is a composite of angular quality, precision/recall and intensity coverage, not a black-box label. For multiphase samples, AutoMix enumerates bounded combinations of leading candidates and estimates non-negative component weights by least squares, retaining the full theoretical peak list of each component. The result is that a phase assi

Load-bearing premise

The load-bearing premise is that a compact library of the thirty strongest calculated reflections per structure, generated for Cu Kα radiation and corrected only by a single bounded global angular shift, adequately represents real measured powder patterns; if texture, strain, specimen displacement or other sample effects distort positions or intensities beyond that model, the single-phase ranking and AutoMix weights lose physical meaning.

Editorial extensions

If this is right

Researchers can reproduce a phase identification from exported records that include inputs, parameters and peak-level evidence, not just a score.
The one-to-one assignment and explicit residual peaks make counter-evidence visible, so impurity phases and artefacts become prompts for inspection rather than hidden mismatches.
AutoMix's non-negative weights provide a bounded, auditable estimate of relative diffraction contributions for mixtures, with the caveat that they are not quantitative phase fractions.
Because the pipeline is local and open, laboratories can run proprietary or unpublished data without transmitting patterns to remote services and can rebuild libraries with different radiation or peak limits.
The PDF/CIF module offers an independent whole-pattern check that uses the full calculated reflection set, giving a second layer of evidence beyond the truncated retrieval representation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A testable extension is to benchmark XMatcher against known two- and three-phase mixtures with varying texture; if rankings shift substantially with texture, the intensity model would need preferred-orientation corrections.
The design suggests a natural interface to refinement: AutoMix weights could serve as starting values for Rietveld or whole-pattern fitting, since they already carry phase-specific peak attributions.
The same staged, evidence-exposing architecture could be adapted to other diffraction problems, such as neutron or electron diffraction, where proprietary search-match tools dominate.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague — read this if you care about XRD phase identification software. XMatcher is a real piece of open infrastructure: a local library built from MP500 structures, composition filtering, bounded global-shift correction, one-to-one peak assignment with supporting and conflicting peaks exposed, AutoMix multiphase NNLS fitting, and PDF/CIF whole-pattern comparison, all behind a reproducible API and UI. The code and a figshare archive are provided. That puts it ahead of many closed search-match tools and makes it worth serious engagement.

The equations in Methods are internally consistent and clearly explained. The Limitations section is unusually honest: it explicitly says peak positions can be affected by strain, composition, temperature, and specimen displacement, and that AutoMix NNLS weights are evidence weights, not phase fractions. The dependencies on pymatgen, simxrd, and the author-linked MP500.db are stated, and the library is rebuildable, so I don't count the self-citations against the work.

Now the soft spots. The preprint contains no quantitative validation: no synthetic ground truth, no labeled mixtures, no comparison with HighScore, Match!, or JADE, and no example outputs. A reader cannot tell whether the ranking works or merely looks plausible. The stress-test concern is real. The theoretical library is derived from DFT-relaxed MP structures, whose lattice parameters typically differ from experiment by 1–3%, producing peak shifts that vary nonlinearly with 2θ. A single global δ in Eq. 5 cannot absorb that pattern; high-angle reflections will fall outside the tolerance τ even when low-angle reflections are aligned. The Limitations paragraph concedes this qualitatively, but the paper offers no benchmark showing how often it bites or how robust the downstream ranking is. Since one-to-one assignment and AutoMix response matrices build on those peak matches, this is the central assumption of the tool, and the manuscript does not establish it.

This is fixable. A benchmark section with synthetic patterns subjected to known strain, zero-offset, and intensity distortions would answer most of the concern. The authors might also consider an angle-dependent shift model or a per-peak tolerance bound. None of this requires a change in the architecture; it requires evidence.

Who is this for? Anyone building or evaluating XRD phase-ID pipelines, especially labs that want transparent open tools instead of black-box commercial software. I would bring it to a reading group and would likely cite it as an open-source baseline. Verdict: yes, this deserves a serious referee, conditional on validation results. Don't desk-reject it; send it out and make the authors show the tool works on known cases, including the distortion regimes the paper itself acknowledges.

Referee Report

3 major / 5 minor

Summary. The paper presents XMatcher, an open-source Python/GUI framework for XRD phase identification. The workflow builds a theoretical peak library from Materials Project structures (via the MP500.db collection and pymatgen's XRDCalculator), retrieves candidate phases using optional chemical filters and a bounded global angular shift, performs one-to-one peak assignment with a composite ranking score, and extends to multiphase analysis through the AutoMix module using non-negative least squares. A separate PDF/CIF whole-pattern comparison module is included for profile-level validation. The authors emphasize transparency and interpretability: all parameters, matched/unmatched peaks, shifts, and component contributions are exposed to the user, and the analysis record is exportable. The paper is primarily a software-description manuscript; no quantitative benchmark or ground-truth validation is reported.

Significance. If the claimed functionality is substantiated, XMatcher would be a valuable open alternative to commercial XRD search-match tools, with strengths in local data processing, inspectable algorithms, reproducible analysis export, and explicit treatment of multiphase hypotheses. The open-source release, the separation of a local API from the GUI, the structured evidence outputs, and the PDF/CIF validation module are genuine contributions to research infrastructure. However, the scientific claims of reliable phase identification and useful AutoMix ranking are not yet supported by any quantitative evaluation. The central assumptions—notably the single global-shift alignment model and the use of unvalidated NNLS component weights—need targeted benchmarking before the manuscript can be judged as a rigorous methods contribution.

major comments (3)

[Reproducibility and validation (Section 5, also overall)] The manuscript contains no quantitative validation of the central claim. The 'Reproducibility and validation' subsection reports only 'validation tests for retrieval and matching procedures' and programmatic consistency checks; there is no ground-truth benchmark against known experimental patterns, synthetic mixtures, or withheld crystal structures. Consequently, the reader cannot assess whether the single-phase ranking or AutoMix combination selection actually identifies correct phases. I request a benchmark with quantitative metrics: for example, top-k accuracy on simulated patterns from held-out MP500 structures, recovery of known phases in synthetic two- and three-component mixtures, and comparison against at least one commercial or open search-match reference. This is load-bearing because the paper's utility claim rests on the correctness of the matching and ranking pipeline.
[Methods, Eq. (5), Single-phase searching] The single global angular shift δ in Eq. (5) is too rigid for the stated theoretical library. The peaks are calculated from Materials Project structures via pymatgen; DFT-optimized lattice parameters typically differ from experimental values by 1–3%, producing angle-dependent peak shifts (Δ2θ ≈ −2 tanθ · Δd/d). A single constant offset cannot simultaneously align low- and high-angle reflections within a fixed positional tolerance τ, leading to missed assignments, inflated residuals, and potentially incorrect candidate ranking. The Limitations paragraph acknowledges this effect qualitatively, but the paper provides no benchmark demonstrating that the model is adequate for realistic strain, calibration, or specimen-displacement distortions. I ask for a sensitivity test on synthetic patterns with known distortions, and, if the model fails, a revision of the alignment strategy (e.g., an angl
[Methods, AutoMix searching, Eq. (9)] The AutoMix response matrix A is constructed from one-to-one peak assignments made after a candidate-specific global shift. For overlapping reflections from different phases, only one phase receives attribution, even when both contribute substantially. This assignment arbitrariness is baked into A before the NNLS fit in Eq. (9), and the non-negativity constraint cannot repair it. The paper is appropriately cautious that the resulting ρ_l are not quantitative phase fractions, but the identifying function of AutoMix—selecting the correct multiphase combination—is not validated. I request a controlled study on synthetic mixtures with known phase compositions to show that the correct combination is ranked first and that the ρ_l values at least recover the ordering of component strengths. Without such evidence, the AutoMix module remains an unverified heuristic.

minor comments (5)

[Introduction, first paragraph] The phrase 'its practical strength rapid acquisition' appears to be missing a verb or comma ('its practical strength, rapid acquisition,'). Please correct.
[Methods, Eq. (6)] The shift-search procedure is described as using 'a regular grid plus experimentally/theoretically implied pair differences.' The latter is vague; please specify how pair differences are derived and how the grid spacing is chosen. This affects reproducibility.
[Methods, AutoMix searching] The formula P^k_{k=1} \binom{n}{k} is incorrectly rendered in the text; this is a LaTeX typo. The numerical example n=8, p=3 giving 92 combinations is correct.
[References] Reference [1] is the author's own PhD thesis; it would be more standard to cite a primary textbook or review for the opening statement on powder XRD. Also check that the 'pysimxrd' calculator in the PDF comparison section is correctly described by reference [16], which appears to be a conference paper on simulated XRD data rather than a software package documentation.
[Table 1] The row 'NNLS weights' is helpfully explicit about non-quantitativeness. Since this is a frequent source of misuse, consider adding a short paragraph in the AutoMix section that explicitly warns against interpreting the normalized ρ_l values as mass/volume fractions, in addition to the current statement.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: XMatcher is an open-source implementation whose matching and AutoMix operations are defined algorithms, not fitted predictions that reduce to their inputs.

full rationale

The paper's central claim is the provision of an inspectable, reproducible XRD search-and-match platform, not the derivation of a physical law or fitted constant. The main equations are operational definitions: Eq. (5) defines peak-pair eligibility under a bounded global shift, Eq. (6) defines a pair cost, and Eq. (8) defines a ranking score computed from the resulting one-to-one assignment. These quantities are not predictions of anything external to the matching procedure; they are the framework's own scoring components. AutoMix Eq. (9) is a non-negative least-squares fit of component response vectors to a detected-peak intensity vector, and the paper explicitly states that the returned values are 'relative diffraction contributions and are not quantitative mass or volume fractions.' The Limitations paragraph candidly acknowledges that 'Peak positions can be affected by strain, composition, temperature, specimen displacement and instrumental factors that are not fully represented by a single global shift' and that 'Relative intensities can differ substantially because of texture, microabsorption and sample preparation.' This is an honest validity caveat, not a circular step. The self-citations, notably MP500.db [3] and pysimxrd [16], are data and software dependencies rather than load-bearing arguments: the matching mathematics is defined in the paper itself and does not rely on those references for its derivation. No uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled in via citation: the theoretical library is calculated with pymatgen's standard XRDCalculator from crystal structures, and the PDF module uses pysimxrd as a calculator. Thus no load-bearing step reduces, by the paper's own equations or by self-citation, to its inputs.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central workflow rests on standard diffraction physics plus a series of modeling choices about how well calculated peak libraries represent real measurements. The free parameters are all configurable software choices rather than physically meaningful fitted constants, but several of them (score weights, peak detection settings, AutoMix pool size) have no empirical calibration, so they should be treated as assumptions about what makes a good match.

free parameters (7)

Positional tolerance τ (Eq. 5)
Configurable tolerance that decides whether an experimental peak and a theoretical reflection can be paired; central to all matching results.
Maximum global shift Δ (Eq. 5)
Configurable bound on the common angular offset searched for every candidate; affects alignment and thus matched peak sets.
Assignment cost weights w_x, w_I (Eq. 6)
User-set weights trading angular error against intensity disagreement; the paper only requires w_x+w_I=1 and gives no default or calibration.
Hybrid score weights (Eq. 8) = 0.45/0.30/0.15/0.10
Weights in H combine quality, coverage, and min(P,R) terms; chosen by hand with no empirical justification, but they determine candidate ranking.
Peak detection settings
Smoothing, baseline estimation, minimum height/prominence/separation, and the number m of retained peaks are configurable and directly change which peaks enter retrieval.
AutoMix pool size/maximum phases and penalties = n=8 default, p max, penalty 0.75(r-1), min contribution 3%
Candidate pool size, maximum number of phases, and the complexity penalty shape which combinations are returned; not validated against known multiphase benchmarks.
Library build choices = 30 strongest reflections, Cu Kα, 10–90° 2θ, d_min 0.5 Å
The distributed MP500_xrd_database.pkl is built with these settings; a different angular window or peak limit would change the library and matching results.

assumptions (5)

standard math Bragg's law relates reflection position to d-spacing (Eq. 1)
Used as the basis for all calculated 2θ positions in the theoretical library.
domain assumption Calculated relative intensities from pymatgen's XRD calculator approximate measured diffraction intensities
The whole matching and NNLS weighting scheme treats calculated intensities as comparable to experimental peak intensities; the Limitations section acknowledges texture, absorption, and other effects can break this.
domain assumption A single global angular shift is sufficient to align a candidate phase to the experimental pattern (Eq. 5)
Calibration-like offsets are modeled, but strain, temperature, and specimen displacement can cause non-uniform shifts; the paper itself lists these as limitations.
domain assumption Multiphase patterns are approximately a linear superposition of normalized single-phase peak intensities (Eq. 9)
AutoMix solves min_w ||Aw-y||^2, which assumes additive, non-negative component contributions to the detected peak vector.
domain assumption The MP500-derived library is sufficiently representative for the intended phase-identification tasks
A phase absent from the database cannot be recovered; the paper acknowledges this in the Limitations section.

how reviews work

0 comments

Cite this review

Pith. "Pith review of XMatcher: An Open-Source Framework for X-Ray Diffraction Phase Identification." pith.science (2026). https://pith.science/paper/X35C3TSO

@misc{pith2026260717162,
  author       = {Pith},
  title        = {Pith review of: XMatcher: An Open-Source Framework for X-Ray Diffraction Phase Identification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/X35C3TSO}},
  note         = {Machine review of arXiv:2607.17162}
}

read the original abstract

Powder X-ray diffraction (XRD) is widely used for crystalline phase identification, and recent machine learning approaches have demonstrated remarkable capabilities in accelerating diffraction interpretation. However, reliable phase assignment still requires transparent, evidence-based validation, particularly for complex samples where interpretability and expert assessment remain essential. Search-match methods provide a robust and complementary strategy, yet many implementations are proprietary, limiting accessibility and reproducibility. Here, we introduce XMatcher, an open-source, evidence-driven framework that integrates diffraction databases, matching algorithms, and interactive visualization into a portable workflow. XMatcher generates theoretical diffraction libraries from crystal structures, retrieves candidate phases through chemical and diffraction constraints, applies global angular-shift correction and one-to-one peak matching, and reports quantitative agreement metrics together with peak-level evidence. Its AutoMix module extends identification to multiphase patterns by evaluating candidate phase combinations, estimating non-negative diffraction contributions, and visualizing phase-specific peak distributions. Through a local graphical interface, XMatcher enables ranked candidate inspection, interactive pattern comparison, PDF/CIF-based whole-pattern validation, and reproducible analysis export. By exposing both supporting and conflicting evidence rather than relying on a single similarity score, XMatcher provides an interpretable and reproducible platform for crystalline phase identification.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 1 linked inside Pith

[1]

PhD thesis, The Hong Kong University of Science and Technology, 2026

Bin Cao.Physics-Constrained Learning of Crystal Structures and Properties from Powder Diffrac- tion. PhD thesis, The Hong Kong University of Science and Technology, 2026

2026
[2]

Awesome xrd → crystal

awesome-xrd2crystal contributors. Awesome xrd → crystal. https://github.com/Bin-Cao/ awesome-xrd2crystal, 2026. Data set

2026
[3]

Xqueryer: an intelligent crystal structure identifier for powder x-ray diffraction.National Science Review, 12(12):nwaf421, 2025

Bin Cao, Zinan Zheng, Y ang Liu, Longhan Zhang, Lawrence WY Wong, Lu-Tao Weng, Jia Li, Haoxiang Li, and Tong-Yi Zhang. Xqueryer: an intelligent crystal structure identifier for powder x-ray diffraction.National Science Review, 12(12):nwaf421, 2025

2025
[4]

Xdecomposer: Learning prior-free set decomposition for multiphase x-ray diffraction.arXiv preprint arXiv:2605.05866, 2026

Hanyu Gao, Bin Cao, Yunyue Su, Tong-Yi Zhang, and Qiang Liu. Xdecomposer: Learning prior-free set decomposition for multiphase x-ray diffraction.arXiv preprint arXiv:2605.05866, 2026

arXiv 2026
[5]

End-to-end crystal structure prediction from powder x-ray diffraction

Qingsi Lai, Fanjie Xu, Lin Y ao, Zhifeng Gao, Siyuan Liu, Hongshuai Wang, Shuqi Lu, Di He, Liwei Wang, Linfeng Zhang, et al. End-to-end crystal structure prediction from powder x-ray diffraction. Advanced Science, 12(8):2410722, 2025

2025
[6]

Powder diffraction crystal structure determination using generative models

Qi Li, Rui Jiao, Liming Wu, Tiannian Zhu, Wenbing Huang, Shifeng Jin, Y ang Liu, Hongming Weng, and Xiaolong Chen. Powder diffraction crystal structure determination using generative models. Nature Communications, 16(1):7428, 2025

2025
[7]

Ai-driven structure refinement of x-ray diffraction.arXiv preprint arXiv:2602.16372, 2026

Bin Cao, Qian Zhang, Zhenjie Feng, Taolue Zhang, Jiaqiang Huang, Lu-Tao Weng, and Tong-Yi Zhang. Ai-driven structure refinement of x-ray diffraction.arXiv preprint arXiv:2602.16372, 2026

arXiv 2026
[8]

Courier Corporation, 1990

Bertram Eugene Warren.X-ray Diffraction. Courier Corporation, 1990

1990

Show all 16 references

[9]

JADE Pro.https://www.icdd.com/mdi-jade/, 2026

Materials Data Inc. JADE Pro.https://www.icdd.com/mdi-jade/, 2026

2026
[10]

HighScore Plus

Malvern Panalytical. HighScore Plus. https://www.malvernpanalytical.com/en/products/ category/software/x-ray-diffraction-software/highscore, 2026

2026
[11]

Match!: Phase Identification Using Powder Diffraction Data

Crystal Impact GbR. Match!: Phase Identification Using Powder Diffraction Data. https://www. crystalimpact.com/match/, 2026. Accessed: 2026-07-17

2026
[12]

DIFFRAC.EVA

Bruker AXS GmbH. DIFFRAC.EVA. https://www.bruker.com/ en/products-and-solutions/diffractometers-and-x-ray-microscopes/ x-ray-diffractometers/diffrac-suite-software/diffrac-eva.html, 2026

2026
[13]

Integrated x-ray powder diffraction software pdxl.Rigaku J, 26:23–27, 2010

Corporation Rigaku. Integrated x-ray powder diffraction software pdxl.Rigaku J, 26:23–27, 2010

2010
[14]

Commentary: The materials project: A materials genome approach to accelerating materials innovation.APL materials, 1(1), 2013

Anubhav Jain, Shyue Ping Ong, Geoffroy Hautier, Wei Chen, William Davidson Richards, Stephen Dacek, Shreyas Cholia, Dan Gunter, David Skinner, Gerbrand Ceder, et al. Commentary: The materials project: A materials genome approach to accelerating materials innovation.APL materia...

2013
[15]

Python materials genomics (pymatgen): A robust, open-source python library for materials analysis

Shyue Ping Ong, William Davidson Richards, Anubhav Jain, Geoffroy Hautier, Michael Kocher, Shreyas Cholia, Dan Gunter, Vincent L Chevrier, Kristin A Persson, and Gerbrand Ceder. Python materials genomics (pymatgen): A robust, open-source python library for materials analysis. ...

2013
[16]

Simxrd-4m: big simulated x-ray diffraction data and crystal symmetry classification benchmark

Bin Cao, Y ang Liu, Zinan Zheng, Ruifeng Tan, Jia Li, and Tong-yi Zhang. Simxrd-4m: big simulated x-ray diffraction data and crystal symmetry classification benchmark. InInternational Conference on Learning Representations, volume 2025, pages 70721–70745, 2025. 17

2025

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.

[1] [1]

PhD thesis, The Hong Kong University of Science and Technology, 2026

Bin Cao.Physics-Constrained Learning of Crystal Structures and Properties from Powder Diffrac- tion. PhD thesis, The Hong Kong University of Science and Technology, 2026

2026

[2] [2]

Awesome xrd → crystal

awesome-xrd2crystal contributors. Awesome xrd → crystal. https://github.com/Bin-Cao/ awesome-xrd2crystal, 2026. Data set

2026

[3] [3]

Xqueryer: an intelligent crystal structure identifier for powder x-ray diffraction.National Science Review, 12(12):nwaf421, 2025

Bin Cao, Zinan Zheng, Y ang Liu, Longhan Zhang, Lawrence WY Wong, Lu-Tao Weng, Jia Li, Haoxiang Li, and Tong-Yi Zhang. Xqueryer: an intelligent crystal structure identifier for powder x-ray diffraction.National Science Review, 12(12):nwaf421, 2025

2025

[4] [4]

Xdecomposer: Learning prior-free set decomposition for multiphase x-ray diffraction.arXiv preprint arXiv:2605.05866, 2026

Hanyu Gao, Bin Cao, Yunyue Su, Tong-Yi Zhang, and Qiang Liu. Xdecomposer: Learning prior-free set decomposition for multiphase x-ray diffraction.arXiv preprint arXiv:2605.05866, 2026

arXiv 2026

[5] [5]

End-to-end crystal structure prediction from powder x-ray diffraction

Qingsi Lai, Fanjie Xu, Lin Y ao, Zhifeng Gao, Siyuan Liu, Hongshuai Wang, Shuqi Lu, Di He, Liwei Wang, Linfeng Zhang, et al. End-to-end crystal structure prediction from powder x-ray diffraction. Advanced Science, 12(8):2410722, 2025

2025

[6] [6]

Powder diffraction crystal structure determination using generative models

Qi Li, Rui Jiao, Liming Wu, Tiannian Zhu, Wenbing Huang, Shifeng Jin, Y ang Liu, Hongming Weng, and Xiaolong Chen. Powder diffraction crystal structure determination using generative models. Nature Communications, 16(1):7428, 2025

2025

[7] [7]

Ai-driven structure refinement of x-ray diffraction.arXiv preprint arXiv:2602.16372, 2026

Bin Cao, Qian Zhang, Zhenjie Feng, Taolue Zhang, Jiaqiang Huang, Lu-Tao Weng, and Tong-Yi Zhang. Ai-driven structure refinement of x-ray diffraction.arXiv preprint arXiv:2602.16372, 2026

arXiv 2026

[8] [8]

Courier Corporation, 1990

Bertram Eugene Warren.X-ray Diffraction. Courier Corporation, 1990

1990