Pith. sign in

REVIEW 3 major objections 5 minor 37 references

Hybrid DiffractGPT-Rietveld Refinement Framework for Automated X-ray Diffraction Analysis

T0 review · 3 major / 5 minor · reviewed 2026-07-13 · grok-4.5

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

Pith's one-line read A hybrid database-and-generative pipeline returns valid lattice parameters for most experimental mineral powder patterns and nearly all of a large simulated set, automating end-to-end crystal structure determination from powder XRD.

desk verdict Solid hybrid pipeline and honest lattice benchmarks; the soft spot is framing lattice-return rates as end-to-end structure determination when full StructureMatcher agreement is only ~23%. read the letter →

arxiv 2607.08890 v1 pith:6MKF2EUP submitted 2026-07-09 cond-mat.mtrl-sci

Charles Rhys Campbell , Justin Ely , Jaehyung Lee , Frank M. Abel , Kamal Choudhary This is my paper
classification cond-mat.mtrl-sci
keywords powderX-raydiffractionRietveldrefinementgenerativeAIcrystalstructuredeterminationpatternmatchinglatticeparametersautomatedcrystallographyDiffractGPT
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

Getting from a raw powder X-ray diffraction pattern to a usable crystal structure still usually requires expert phase search, a good starting model, and careful Rietveld refinement. This paper presents AGAPI-XRD, one automated pipeline that first matches the experimental pattern against large structure databases by cosine similarity, then uses DiffractGPT—a generative transformer that proposes full lattice parameters, space group, and coordinates from peak lists and composition—then optionally relaxes the cell with a machine-learned force field and runs classical Rietveld refinement. On a filtered set of 276 experimental mineral patterns, the pipeline returns valid lattice parameters for 79.7 percent of minerals and a candidate structure for 93.8 percent; on a thousand-structure simulated DFT subset it returns valid lattices for 94.8 to 98.1 percent. Pattern matching is most accurate when a known phase sits in the databases, while the generative step extends coverage to complex or missing compositions. The claim is that this hybrid stack makes powder-XRD structure work far more accessible without discarding quantitative refinement when it is still needed.

What carries the argument

AGAPI-XRD: a staged hybrid pipeline that matches experimental powder patterns to database structures by cosine similarity on re-binned intensities, falls back to DiffractGPT (a fine-tuned generative transformer mapping peaks plus formula to lattice, space group, and coordinates), then optionally relaxes with a machine-learned force field and refines with sequential Rietveld parameter release.

What would settle it

On experimental powder patterns whose full crystal structures (including atomic coordinates) are independently known, measure how often the pipeline’s top candidate matches the true structure under a standard structure matcher; if that match rate stays near the reported ~23 percent on the simulated set while lattice-return rates remain high, the end-to-end determination claim does not hold.

Watch

Extended reading notes

Core claim

AGAPI-XRD combines cosine-similarity pattern matching over large structure databases with DiffractGPT generative structure prediction, optional force-field relaxation, and automated Rietveld refinement, and thereby returns valid lattice parameters for 79.7 percent of a 276-mineral experimental powder-XRD benchmark and for 94.8–98.1 percent of a 1,000-structure simulated subset, while identifying a candidate structure for 93.8 percent of the minerals. Database matching supplies the highest lattice accuracy for known phases; the generative model extends structure generation to materials absent from existing databases.

Load-bearing premise

Returning a lattice and a candidate structure file after automated cell rematching is treated as successful crystal structure determination, even when full atomic agreement is rarely checked and refinement often does not improve the lattice.

Editorial extensions

If this is right

  • Known phases already in structure databases can be recovered from powder XRD with high lattice-length skill without manual starting-model selection.
  • Complex or nonstandard compositions that miss database search can still receive a candidate structure from generative prediction, raising overall identification near 94 percent on the mineral set.
  • Automated Rietveld mainly supplies figures of merit and structural consistency rather than large aggregate lattice-accuracy gains when the starting model is already close or wrong in symmetry.
  • A unified public API and web interface let users run the full workflow without local crystallography software setup.
  • Triclinic and other low-symmetry cases remain the weakest regime, so one-dimensional powder data still limit recovery when six independent lattice parameters must be found.

Reading between the lines

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

  • Lattice-return rates and cell rematching can overstate full structure determination when atomic coordinates are not the primary benchmark and structure-match rates on simulated data stay low.
  • Feeding failed refinements back as training signal for the generative model is a natural way to improve triclinic and site-mixed minerals that currently go unmatched.
  • Multiphase mixtures and amorphous fractions are the obvious next laboratory test; the present single-phase design leaves common experimental samples largely unaddressed.
  • Training generative models on broader experimental patterns rather than only simulated DFT XRD could raise true structure-match rates enough for refinement to become consistently helpful.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The manuscript presents AGAPI-XRD, a hybrid pipeline that combines (i) cosine-similarity pattern matching against JARVIS-DFT and COD, (ii) DiffractGPT generative structure prediction from peak lists and composition, (iii) optional ALIGNN-FF relaxation, (iv) automated Rietveld refinement via GSAS-II or BGMN, and (v) optional ALIGNN/SlaKoNet property prediction, exposed as a public API. On a filtered RRUFF experimental mineral set (276 entries with a,b,c≤10 Å; 220 with complete lattice comparisons) the pipeline returns valid lattice parameters for 79.7% of minerals and a candidate structure for 93.8%; pattern matching covers 72.8% and DiffractGPT 81–84%. On a 1,000-structure Alexandria PBE-hull simulated-XRD subset, valid lattice return rates are 94.8–98.1% across six refinement configurations. Method-stratified tables show pattern matching yields low lattice MAE and high skill, while DiffractGPT extends coverage at substantially higher error; StructureMatcher agreement on the Alexandria no-refinement set is 22.9%, and Rietveld does not systematically reduce lattice MAE even on matched structures.

Significance. If the operational claims are framed carefully, the work is a useful systems contribution: it unifies database search, generative inverse design, classical refinement, and ML force-field relaxation into a single accessible API and supplies multi-configuration, multi-metric benchmarks on both experimental minerals and a DFT hull subset. Strengths include transparent reporting of skill scores, JSD, crystal-system and method-stratified MAEs, an honest StructureMatcher analysis (Fig. 6) showing that refinement does not fix generative errors, public code/data links, and an 18-way cell-setting protocol that reduces setting ambiguity. The complementarity result—pattern matching for known phases, DiffractGPT for coverage—is practically relevant for automated XRD workflows even if full atomic-structure recovery remains limited.

major comments (3)
  1. [Abstract / §2 / Conclusion] Abstract, Conclusion, and §2 framing: the central claim of “end-to-end automated crystal structure determination” is stronger than the evidence. RRUFF generally lacks atomic coordinates (Methods), so the 79.7%/93.8% figures are lattice-level identification/return rates after an 18-way cell rematch that minimizes composite relative error against ground truth. On Alexandria, StructureMatcher agreement is only 229/981 (22.9%) at stol=0.5, and Fig. 6 shows neither BGMN nor GSAS-II systematically lowers lattice MAE on that matched subset. The manuscript should restate the claim as automated lattice-level identification plus candidate generation, and report StructureMatcher (or equivalent full-structure) rates as primary success metrics alongside lattice MAE.
  2. [Table 5 / §2] Table 5 and related text: DiffractGPT is presented as the component that “extends structure generation to complex materials,” yet on the 47 RRUFF cases where it supplies the best structure, length MAEs are 1.63–1.82 Å with negative skill scores. The paper should quantify how often these generated structures are crystallographically usable (e.g., correct space-group family, chemically sensible coordination, or refinement convergence to a physically meaningful Rwp) rather than only counting POSCAR emission and post-hoc lattice rematch. Without that, the coverage gain is hard to interpret as successful determination.
  3. [Methods (cell-matching protocol)] Methods cell-matching protocol: selecting the best of 18 cell variants (original/primitive/conventional × 6 axis permutations) by minimizing a composite relative-error score against the reference is a fair diagnostic for setting ambiguity, but it uses ground-truth lattice information that a real user does not have. The main accuracy tables therefore partly measure oracle rematching rather than blind prediction. Please report primary lattice metrics also under a fixed, blind cell convention (or under the single best-scoring candidate without GT-guided permutation), and clarify which numbers in Tables 2–6 use the oracle protocol.
minor comments (5)
  1. [Table 3] Table 3 is dense and transposed; a clearer layout or supplementary long-form tables would help readers compare the twelve configurations.
  2. [§2 / Methods] Peak-finding thresholds (height 0.05, prominence 0.02, min separation 0.5°, top-20 peaks) and the a,b,c≤10 Å filter are free parameters that strongly shape the benchmark; a short sensitivity note would strengthen reproducibility.
  3. [§1 / §2] Mild ecosystem self-use (DiffractGPT trained on JARVIS-DFT; pattern matching also queries JARVIS-DFT) is acknowledged indirectly; a one-sentence explicit statement that RRUFF and Alexandria are external evaluation sets would help.
  4. [Fig. 4] Fig. 4 caption discusses over-/under-prediction of lengths and angles; adding numerical bias (mean signed error) next to MAE would make that discussion quantitative.
  5. [Table 6] Triclinic performance is correctly flagged as weakest (Table 6); a brief note on whether multiphase or preferred-orientation cases were excluded from RRUFF would clarify residual failure modes.

Circularity Check

1 steps flagged · score 1.0 of 10

No circular derivation: empirical pipeline benchmarks on external RRUFF/Alexandria data; only mild non-load-bearing ecosystem self-use of JARVIS-DFT.

  1. self citation load bearing [Introduction; Methods (DiffractGPT); Stage 1 pattern matching]
    "Trained on simulated XRD patterns from the JARVIS-DFT database [24], DiffractGPT performs rapid inverse design... pattern matching against the combined JARVIS-DFT (∼76,000 entries) and COD (∼431,000 entries) structure databases"

    DiffractGPT (same corresponding author) was trained on JARVIS-DFT structure–XRD pairs, and Stage 1 also queries JARVIS-DFT. That is overlapping self-citation of the authors’ ecosystem. It is not load-bearing for the central claims: RRUFF and Alexandria references are external, COD is independent, and reported metrics are comparisons to those external lattices rather than re-statements of the training fit.

full rationale

AGAPI-XRD is an engineering/ML integration paper, not a first-principles derivation. Its load-bearing claims are empirical return rates and lattice MAEs on RRUFF natural minerals and an Alexandria PBE-hull subset, compared to independent experimental or DFT reference lattices. Pattern matching and DiffractGPT are complementary search/generation stages; Rietveld and ALIGNN-FF are optional post-processing. The 18-way cell-setting/axis-permutation protocol is an evaluation alignment against ground truth (standard crystallographic practice), not a fitted parameter renamed as a prediction of the pipeline output. DiffractGPT and ALIGNN-FF are prior work by overlapping authors and were trained on JARVIS-DFT, which is also one of the pattern-matching databases—this is mild ecosystem self-use, not a self-definitional loop or a uniqueness theorem that forces the result. COD and RRUFF are external; Alexandria is a separate DFT corpus. No equation reduces a claimed prediction to its own fitted input by construction. Score 1 for the non-load-bearing self-citation of the JARVIS/DiffractGPT stack; central benchmark content is independent.

Assumptions & free parameters 6 free parameters · 5 assumptions · 1 invented entities

The central performance claims rest on standard crystallographic and ML tooling plus several operational choices (filters, peak thresholds, cell rematch, similarity metric) that are not derived from first principles. No new physical entities are postulated. The main external premises are the kinematic powder pattern model, the correctness of RRUFF/Alexandria reference lattices, and the transferability of a JARVIS-trained generative model to experimental minerals.

free parameters (6)
  • peak-finding thresholds (height 0.05, prominence 0.02, min separation 0.5° 2θ)
    Hand-chosen scipy.find_peaks settings that determine which peaks enter the DiffractGPT prompt; not fitted but control generative input.
  • top-20 peaks retained for DiffractGPT prompt
    Fixed truncation of the peak list; changes the information available to the generative model.
  • cosine-similarity rebinning (0.1° grid, 0–90° 2θ)
    Defines the pattern-matching score S; binning and range are design choices that affect match rates.
  • a,b,c ≤ 10 Å filter (RRUFF → 276; Alexandria → 87,631 then 1,000 subset)
    Post-hoc size cut that defines the reported benchmark populations and can favor easier cells.
  • StructureMatcher stol=0.5 and seed=42 for Alexandria subset
    Controls which structures count as full structural matches and which 1,000 entries are evaluated.
  • 12-term Chebyshev background and sequential Rietveld parameter release order
    Standard but free refinement protocol choices that affect Rwp and refined lattices.
assumptions (5)
  • domain assumption Kinematic diffraction approximation suffices to simulate powder patterns for cosine matching against JARVIS-DFT and COD.
    Stated in Methods for Stage 1 pattern matching; ignores extinction, preferred orientation (unless March–Dollase enabled), and microabsorption.
  • domain assumption Powder XRD is one-dimensional and the pattern-to-structure map is not one-to-one (Patterson ambiguity).
    Cited in Discussion as a fundamental error source; underpins why lattice-level metrics are used.
  • domain assumption OptB88vdW JARVIS-DFT lattices systematically overestimate experimental constants by ~1–3%.
    Invoked to explain residual MAE on pattern-matched structures.
  • domain assumption RRUFF experimental lattice parameters and Alexandria PBE-hull DFT lattices are reliable ground truth for MAE/skill evaluation.
    All accuracy tables rest on these references; RRUFF atomic coordinates are generally unavailable.
  • ad hoc to paper Selecting the best of 18 cell variants (primitive/conventional/original × 6 axis permutations) by composite relative error is a fair accuracy protocol.
    Methods cell-matching protocol; can reduce reported MAE relative to a fixed conventional setting.
invented entities (1)
  • AGAPI-XRD hybrid pipeline / API independent evidence
    purpose: Unify pattern matching, DiffractGPT, ALIGNN-FF, Rietveld, and property models into one automated XRD analysis service.
    The named system is the paper’s product; it is a software composition rather than a new physical entity, with independent evidence via the public API and benchmarks.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hybrid DiffractGPT-Rietveld Refinement Framework for Automated X-ray Diffraction Analysis." pith.science (2026). https://pith.science/paper/6MKF2EUP

@misc{pith2026260708890,
  author       = {Pith},
  title        = {Pith review of: Hybrid DiffractGPT-Rietveld Refinement Framework for Automated X-ray Diffraction Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6MKF2EUP}},
  note         = {Machine review of arXiv:2607.08890}
}
read the original abstract

X-ray diffraction (XRD) is fundamental to structural materials characterization, yet transforming a raw powder pattern into a refined crystal structure still demands considerable domain expertise. We present AGAPI-XRD, a hybrid framework integrating DiffractGPT generative structure prediction, database pattern matching against JARVIS-DFT and COD, and automated Rietveld refinement and ALIGNN-FF relaxation through a unified API at https://atomgpt.org/xrd. First, we used the AGAPI-XRD pipeline to evaluate the crystal structure of a variety of minerals in the RRUFF database that were experimentally characterized using powder x-ray diffraction. Next, we benchmarked the lattice parameter prediction fidelity of the AGAPI-XRD pipeline using a subset of the Alexandria PBE-hull dataset and the subset of RRUFF minerals that have known lattice parameters. AGAPI-XRD returns valid lattice parameters for 79.7\% of the RRUFF benchmark minerals and for 94.8--98.1\% of the Alexandria subset, while identifying a candidate structure for 93.8\% of RRUFF minerals. For this benchmark, pattern matching delivers the highest accuracy for known phases, while DiffractGPT extends structure generation to complex materials absent from existing databases. Together, AGAPI-XRD advances accessible, end-to-end automated crystal structure determination from powder XRD data.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 2 canonical work pages

  1. [1]

    Pearson, ??? (2001)

    Cullity, B.D., Stock, S.R.: Elements of X-Ray Diffraction, 3rd edn. Pearson, ??? (2001)

    2001

  2. [2]

    Dover Publications, ??? (1990)

    Warren, B.E.: X-Ray Diffraction. Dover Publications, ??? (1990)

    1990

  3. [3]

    Rietveld, H.M.: A profile refinement method for nuclear and magnetic structures. J. Appl. Crystallogr.2, 65–71 (1969) 19

    1969

  4. [4]

    (ed.): The Rietveld Method

    Young, R.A. (ed.): The Rietveld Method. Oxford University Press, ??? (1993)

    1993

  5. [5]

    Powder Diffr.34, 352–360 (2019)

    Gates-Rector, S., Blanton, T.: The Powder Diffraction File: a quality materials characterization database. Powder Diffr.34, 352–360 (2019)

    2019

  6. [6]

    Nucleic Acids Res.40, 471–477 (2012)

    Graˇ zulis, S., Chateigner, D., Downs, R.T., Yokochi, A.F.T., Quir´ os, M., Lut- terotti, L., Manakova, E., Butkus, J., Moeck, P., Le Bail, A.: Crystallography Open Database (COD): an open-access collection of crystal structures and platform for world-wide collaboration. Nucleic Acids Res.40, 471–477 (2012)

    2012

  7. [7]

    In: Allen, F.H., Bergerhoff, G., Sievers, R

    Bergerhoff, G., Brown, I.D.: Inorganic Crystal Structure Database. In: Allen, F.H., Bergerhoff, G., Sievers, R. (eds.) Crystallographic Databases, pp. 77–95. International Union of Crystallography, ??? (1987)

    1987

  8. [8]

    Toby, B.H., Von Dreele, R.B.: GSAS-II: the genesis of a modern open-source all purpose crystallography software package. J. Appl. Crystallogr.46, 544–549 (2013)

    2013

Show all 37 references
  1. [9]

    McCusker, L.B., Von Dreele, R.B., Cox, D.E., Lou¨ er, D., Scardi, P.: Rietveld refinement guidelines. J. Appl. Crystallogr.32, 36–50 (1999)

    1999

  2. [10]

    npj Comput

    Choudhary, K., DeCost, B., Chen, C., Jain, A., Tavazza, F., Cohn, R., Park, C.W., Choudhary, A., Agrawal, A., Billinge, S.J.L., Holm, E., Ong, S.P., Wolverton, C.: Recent advances and applications of deep learning methods in materials science. npj Comput. Mater.8, 59 (2022)

    2022

  3. [11]

    Nature559, 547–555 (2018)

    Butler, K.T., Davies, D.W., Cartwright, H., Isayev, O., Walsh, A.: Machine learning for molecular and materials science. Nature559, 547–555 (2018)

    2018

  4. [12]

    npj Comput

    Schmidt, J., Marques, M.R.G., Botti, S., Marques, M.A.L.: Recent advances and applications of machine learning in solid-state materials science. npj Comput. Mater.5, 83 (2019)

    2019

  5. [13]

    npj Comput

    Choudhary, K., DeCost, B.: Atomistic Line Graph Neural Network for improved materials property predictions. npj Comput. Mater.7, 185 (2021)

    2021

  6. [14]

    Digital Discovery2(2), 346–355 (2023)

    Choudhary, K., DeCost, B., Major, L., Butler, K., Thiyagalingam, J., Tavazza, F.: Unified graph neural network force-field for the periodic table: solid state applications. Digital Discovery2(2), 346–355 (2023)

    2023

  7. [15]

    Court, C.J., Yildirim, B., Jain, A., Cole, J.M.: 3-D inorganic crystal structure generation and property prediction via representation learning. J. Chem. Inf. Model.60, 4518–4535 (2020)

    2020

  8. [16]

    ACS Cent

    Kim, S., Noh, J., Gu, G.H., Aspuru-Guzik, A., Jung, Y.: Generative adversarial networks for crystal structure prediction. ACS Cent. Sci.6, 1412–1420 (2020) 20

    2020

  9. [17]

    arXiv:2110.06197 (2022)

    Xie, T., Fu, X., Ganea, O.-E., Barzilay, R., Jaakkola, T.: Crystal diffusion varia- tional autoencoder for periodic material generation. arXiv:2110.06197 (2022)

    2022 arXiv

  10. [18]

    Choudhary, K.: AtomGPT: atomistic generative pretrained transformer for forward and inverse materials design. J. Phys. Chem. Lett.15, 6909–6914 (2024)

    2024

  11. [19]

    Choudhary, K.: DiffractGPT: Atomic structure determination from X-ray diffrac- tion patterns with a generative pre-trained transformer model. J. Phys. Chem. Lett.16, 2110–2116 (2025)

    2025

  12. [20]

    The Journal of Physical Chemistry Letters16(27), 7028–7035 (2025)

    Choudhary, K.: Microscopygpt: generating atomic-structure captions from microscopy images of 2d materials with vision-language transformers. The Journal of Physical Chemistry Letters16(27), 7028–7035 (2025)

    2025

  13. [21]

    IUCrJ4, 486–494 (2017)

    Park, W.B., Chung, J., Jung, J., Sohn, K., Singh, S.P., Pyo, M., Shin, N., Sohn, K.-S.: Classification of crystal structure using a convolutional neural network. IUCrJ4, 486–494 (2017)

    2017

  14. [22]

    Lee, J.-W., Park, W.B., Lee, J.H., Singh, S.P., Sohn, K.-S.: A deep-learning tech- nique for phase identification in multiphase inorganic compounds using synthetic XRD powder patterns. Nat. Commun.11, 86 (2020)

    2020

  15. [23]

    Vecsei, P.M., Choo, K., Chang, J., Neupert, T.: Neural network based classifi- cation of crystal symmetries from x-ray diffraction patterns. Phys. Rev. B99, 245120 (2019)

    2019

  16. [24]

    npj Comput

    Choudhary, K., Garrity, K.F., Reid, A.C.E., DeCost, B., Biacchi, A.J., Hight Walker, A.R., Trautt, Z., Hattrick-Simpers, J., Kusne, A.G., Centrone, A., Davydov, A., Jiang, J., Pachter, R., Cheon, G., Reed, E., Aber, A., Olber, B., Ganguli, S., Tavazza, F.: The joint automated ...

    2020

  17. [25]

    Computational Materials Science259, 114063 (2025)

    Choudhary, K.: The jarvis infrastructure is all you need for materials design. Computational Materials Science259, 114063 (2025)

    2025

  18. [26]

    npj Comput

    Choudhary, K., Lee, J., Ely, J., Zhang, K., Ajith, A., Campbell, C.R.: AGAPI: Agentic graph-augmented physics-informed framework for scientific discovery. npj Comput. Mater. (2025). submitted

    2025

  19. [27]

    CPD Newsl.20, 5–8 (1998)

    Bergmann, J., Friedel, P., Kleeberg, R.: BGMN—a new fundamental parame- ters based Rietveld program for laboratory X-ray sources, its use in quantitative analysis and structure investigations. CPD Newsl.20, 5–8 (1998)

    1998

  20. [28]

    npj Comput

    Choudhary, K.: SlakoNet: Machine-learned tight-binding for materials science. npj Comput. Mater.10, 1–10 (2024) 21

    2024

  21. [29]

    In: Armbruster, T., Danisi, R.M

    Lafuente, B., Downs, R.T., Yang, H., Stone, N.: The power of databases: the RRUFF project. In: Armbruster, T., Danisi, R.M. (eds.) Highlights in Mineralogical Crystallography, pp. 1–30. Walter de Gruyter, ??? (2015)

    2015

  22. [30]

    Scientific Data9, 64 (2022) https://doi.org/10.1038/ s41597-022-01177-w

    Schmidt, J., Wang, H.-C., Cerqueira, T.F.T., Botti, S., Marques, M.A.L.: A new dataset of 415k stable and metastable materials calculated with the PBEsol and SCAN functionals. Scientific Data9, 64 (2022) https://doi.org/10.1038/ s41597-022-01177-w

    2022

  23. [31]

    Nature Methods 17(3), 261–272 (2020) https://doi.org/10.1038/s41592-019-0686-2

    Virtanen, P., Gommers, R., Oliphant, T.E., Haberland, M., Reddy, T., Courna- peau, D., Burovski, E., Peterson, P., Weckesser, W., Bright, J., van der Walt, S.J., Brett, M., Wilson, J., Millman, K.J., Mayorov, N., Nelson, A.R.J., Jones, E., Kern, R., Larson, E., Carey, C.J., Po...

    2020 doi

  24. [32]

    arXiv preprint arXiv:2310.06825 (2023)

    Jiang, A.Q., Sablayrolles, A., Mensch, A., Bamford, C., Chaplot, D.S., de las Casas, D., Bressand, F., Lengyel, G., Lample, G., Saulnier, L., Saux, L.R., Lavril, T., Wang, T.L., Lacroix, T., Tegmark, W.E.S., Jagielski, M., Biderman, S., Four- rier, C., Ouyang, L., Shiebler, D....

    2023 arXiv

  25. [33]

    Zeitschrift f¨ ur Kristallographie81, 285–297 (1932) https://doi.org/ 10.1524/zkri.1932.81.1.285

    March, A.: Mathematische Theorie der Regelung nach der Korngestalt bei affiner Deformation. Zeitschrift f¨ ur Kristallographie81, 285–297 (1932) https://doi.org/ 10.1524/zkri.1932.81.1.285

    1932 doi

  26. [34]

    Journal of Applied Crystallography19, 267–272 (1986) https://doi.org/10.1107/S0021889886089458

    Dollase, W.A.: Correction of intensities for preferred orientation in powder diffrac- tometry: application of the March model. Journal of Applied Crystallography19, 267–272 (1986) https://doi.org/10.1107/S0021889886089458

    1986 doi

  27. [35]

    arXiv preprint arXiv:1802.03426 (2018)

    McInnes, L., Healy, J., Melville, J.: UMAP: Uniform manifold approximation and projection for dimension reduction. arXiv preprint arXiv:1802.03426 (2018)

    2018 arXiv

  28. [36]

    Patterson, A.L.: Ambiguities in the X-ray analysis of crystal structures. Phys. Rev.65, 195–201 (1944)

    1944

  29. [37]

    In: Advances in Neural Information Processing Systems, vol

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., Polosukhin, I.: Attention is all you need. In: Advances in Neural Information Processing Systems, vol. 30 (2017) 22

    2017

Pith tools

Reviewed July 13, 2026 · model on record in the stance chip above.