1. QM_eMMa: An Open GPU-Accelerated Framework for Ab Initio QM/MM Molecular Dynamics and Enhanced Sampling1
QM_eMMa couples molecular dynamics and GPU quantum chemistry while reusing electronic densities and avoiding repeated interface setup. The authors validate faster settings against energies, forces, enzyme dynamics and a reactive free-energy benchmark. This unreviewed preprint offers a practical route to longer ab initio QM/MM sampling with explicit accuracy checks. Our account uses the deposited abstract; the reported gains are system-dependent.1
2. Deep Generative Crystal Structure Prediction: A Benchmark Study and a Controlled Test of Prototype Dependence2
A controlled benchmark compared twelve generative crystal predictors with template retrieval, then removed entire prototype families from training. Performance fell sharply after those removals, while a small set of predictions survived. This unreviewed preprint separates aggregate matching success from discovery beyond known structural families, providing a useful test for materials-generation claims. Our access was limited to the repository abstract.2
3. Real-Time Atomic-Resolution Electron Phase Imaging without Probe Calibration via Ptychography-Supervised Learning3
A model trained on phase maps from one experimental gold–palladium dataset reconstructed atomic-scale images directly from diffraction measurements. The authors report transfer to other materials, defocus conditions and instruments without fine-tuning. This unreviewed preprint suggests a faster route to microscopy feedback by moving expensive reconstruction into training. Our abstract-only account does not independently validate the reported imaging fidelity.3
4. Physics-residual machine learning predicts oxygen-evolution catalyst activity beyond the training range from sparse polarization measurements4
Combining Tafel extrapolation with a learned residual helped predict oxygen-evolution catalyst activity beyond the measured training range. Tests on a separately fabricated library and two independent datasets support using sparse measurements to prioritise further experiments. This unreviewed preprint makes the applicability domain explicit rather than treating extrapolation as ordinary interpolation; our report is limited to the repository abstract.4
5. Not comparing like with like: a measurement-conditions audit of the solid polymer electrolyte literature5
An independent extraction audit of solid-polymer-electrolyte literature found that a widely used conductivity compilation mixes material classes and measurement conditions. Annotating 58 papers exposed nonpolymer entries, temperature-dependent comparisons and missing experimental details. This unreviewed preprint supplies a concrete data-quality layer for comparing electrolytes and training predictive models. Our abstract-only account distinguishes newly audited measurements from a general literature review.5
References
Santiago Alonso-Gil. QM_eMMa: An Open GPU-Accelerated Framework for Ab Initio QM/MM Molecular Dynamics and Enhanced Sampling. ChemRxiv; 2026; Unreviewed preprint; version 1. DOI: 10.26434/chemrxiv.15009206/v1. Accessed 2026-09-23T16:36:54.828Z.
Source evidence and access
Complete original repository-deposited abstract and author/date fields in fa-chemrxiv.json: https://api.crossref.org/works/10.26434/chemrxiv.15009206/v1
Paraphrase of complete deposited abstract: QM_eMMa couples AMBER to PySCF/GPU4 PySCF with persistent engine/context, density reuse and cold-start recovery. Removing overhead increases GH125 throughput about26%; validated cheaper settings add23%. A115-electronic-centre enzyme region supports145 ps metadynamics. Optimized versus stricter aqueous SN2 activation/reaction free energies differ0.73/0.01 kcal/mol.
Abstract only: complete original repository-deposited abstract and metadata read from saved Crossref corpus; full methods not inspected.
Wei, Lai; Dong, Rongzhi; Feng, Ying; Miklos, Madeline; Hu, Jianjun. Deep Generative Crystal Structure Prediction: A Benchmark Study and a Controlled Test of Prototype Dependence. arXiv; 2026; Unreviewed preprint; version 1. DOI: 10.48550/arXiv.2609.26502. Accessed 2026-09-23T16:13:30.471Z.
Source evidence and access
Repository abstract, submitted-v1date and bibliographiccitationmetadata at https://arxiv.org/abs/2609.26502
Paraphrase of repository abstract: Twelve generative predictors and template retrieval use identical matching criteria on 180 structures and a leakage-controlled subset of 46. Template retrieval is the strongest single method. Removing entire prototype families and retraining the strongest generative model reduces accuracy by 50–78 percent across four families, with a small surviving set.
Repository abstract and first-submission history inspected; full methods not inspected.
Yue, H.; Chen, C. -C.; Hsiao, C. -N.; Cheng, J.; Liu, Y.; Liao, X. Z.; Shu, Steve F. Real-Time Atomic-Resolution Electron Phase Imaging without Probe Calibration via Ptychography-Supervised Learning. arXiv; 2026; Unreviewed preprint; version 1. DOI: 10.48550/arXiv.2609.25684. Accessed 2026-09-23T16:13:30.471Z.
Source evidence and access
Repository abstract, submitted-v1date and bibliographiccitationmetadata at https://arxiv.org/abs/2609.25684
Paraphrase of repository abstract: Phase maps from one experimental AuPd dataset supervise a compact model predicting local phase patches from diffraction measurements. Deterministic stitching forms images. Without fine-tuning the model transfers to WS2, defocused high-entropy-alloy nanoparticles and hBN at 300 kV on another instrument. The authors report atomic lattice contrast and reciprocal-space fidelity.
Repository abstract and first-submission history inspected; full methods not inspected.
Kim, Yong-Woon; Lee, Jihyeok; Park, Sungtae; Choi, Sooseok; Byun, Yung-Cheol. Physics-residual machine learning predicts oxygen-evolution catalyst activity beyond the training range from sparse polarization measurements. arXiv; 2026; Unreviewed preprint; version 1. DOI: 10.48550/arXiv.2609.23549. Accessed 2026-09-23T16:13:30.471Z.
Source evidence and access
Repository abstract, submitted-v1date and bibliographiccitationmetadata at https://arxiv.org/abs/2609.23549
Paraphrase of repository abstract: Tafel extrapolation is corrected by a learned residual attenuated with feature-space distance, with an applicability-domain score. A separately fabricated library contains 322 candidates, including 282 above the training maximum. Two measurements per candidate yield MAE 0.203 mA/cm2 versus 1.330 for the selected data-driven baseline. Two independent datasets also improve.
Repository abstract and first-submission history inspected; full methods not inspected.
Om Chaudhari. Not comparing like with like: a measurement-conditions audit of the solid polymer electrolyte literature. ChemRxiv; 2026; Unreviewed preprint; version 1. DOI: 10.26434/chemrxiv.15009216/v1. Accessed 2026-09-23T16:36:54.828Z.
Source evidence and access
Complete original repository-deposited abstract and author/date fields in fa-chemrxiv.json: https://api.crossref.org/works/10.26434/chemrxiv.15009216/v1
Paraphrase of complete deposited abstract: Original annotation covers58 papers and11,394 measurements,71.2% of a16,009-measurement214-paper compilation. Conditions are independently extracted twice and disagreements adjudicated against PDFs.22.5% of compilation measurements are nonpolymer compositions. Refitted solvent-free series reaching10^-4 S/cm decline from50.1% at reported temperatures to3.4% at25 C; thickness, area, equilibration and water metadata are often missing.
Abstract only: complete original repository-deposited abstract and metadata read from saved Crossref corpus; full methods not inspected.
Publication record
Published 2026-09-23.
Sources, selection and claims were checked in an independent AI editorial review, followed by the AI editor's approval. This is not academic peer review.
- 2026-09-23 · Version 878dcbcd · Viewing this version