1. Design of white circularly polarized luminescence with high glum across entire visible regions by dual-loop active learning1
Two active-learning loops guided experiments toward white circularly polarized emitters, balancing colour rendering against polarization across the visible spectrum. The authors report materials meeting both objectives and demonstrate organic light-emitting devices. This connects multi-objective modelling with measured material performance. Our account is limited to the publisher abstract; the reported device results do not establish manufacturing-scale performance.1
2. Towards Universal Calibration for Infrared Spectroscopy with Probabilistic Interference Removal2
The unreviewed VIBES preprint learns background covariance from blank infrared measurements and estimates correction settings through variational inference. Tests on two transmission-measurement types separated analyte absorption from interfering backgrounds, with agreement against independent optical measurements. This could make quantitative spectra less dependent on manual correction choices. Reporting is limited to the repository-deposited abstract; universal calibration remains a proposed direction.2
3. Transferable Graph Neural Network Surrogates for Molecular Dynamics Across Crystal Symmetries3
An unreviewed preprint tests one graph-network formulation for directly propagating atomic positions in aluminium, iron and magnesium, spanning three crystal symmetries. Reported trajectories retain coordination patterns and temperature-dependent displacement behaviour over nanosecond timescales. The comparison tests whether a shared architecture can support different materials. Our access was limited to the deposited abstract; this does not establish transfer without retraining.3
4. Complete Neural Electronic Initialization Accelerates Materials DFT4
An unreviewed study adds augmentation occupancies and spin initialization to neural starting guesses for materials density-functional calculations. Controlled ablations show that omitting these components can remove or reverse speed gains. The complete initializer reduced total calculation time while preserving converged energies on unseen structures. The repository abstract supports a practical acceleration claim, without replacing the underlying electronic-structure calculation.4
5. Machine learning magnetic interactions from neutron powder diffraction data5
An unreviewed study uses machine learning to infer magnetic interaction parameters from powder diffuse-scattering patterns across eight high-symmetry lattices. The reported comparison avoids false minima encountered by nonlinear least-squares refinement. This tests a route from compact scattering fingerprints to material interactions. Our account uses the repository abstract and retains the study’s restriction to isotropic interactions.5
References
Peng Yang; Xiaoyue He; Hongli Zhang; Zeyu Feng; Liyang Wen; Li Wen; Yin Xu; Bo Chen; Mingjun Xiao; Xin Chen; Gang Zou. Design of white circularly polarized luminescence with high glum across entire visible regions by dual-loop active learning. Nature Communications; 2026; Peer-reviewed accepted article, advance online publication. DOI: 10.1038/s41467-026-77409-z. Accessed 2026-09-21T16:12:37.222Z.
Source evidence and access
Abstract: dual-loop experimental optimization and OLED fabrication
AI-assisted interactive experiment–learning evolution strategy to accelerate the discovery of WCPL materials with optimal trade-offs
Original publisher abstract and publication metadata inspected; complete paper not inspected.
Francois Kamper; Ekaterina Krymova; Ann M. Dillner; Andrea Baccarini; Nikunj Dudani; Athanasios Nenes; Guillaume Obozinski; Satoshi Takahama. Towards Universal Calibration for Infrared Spectroscopy with Probabilistic Interference Removal. ChemRxiv; 2026; Preprint v1; not peer reviewed. DOI: 10.26434/chemrxiv.15009188/v1. Accessed 2026-09-21T16:12:37.222Z.
Source evidence and access
Publisher-deposited abstract, https://api.crossref.org/works/10.26434/chemrxiv.15009188/v1
Structured interferences from substrates, matrices, and instrumental artifacts in mid-infrared (IR) spectra often overlap with analyte features and introduce errors
Abstract only: publisher-deposited primary metadata through Crossref; full text not inspected.
Judah Immanuel; Avik Mahata; Aniruddha Maiti. Transferable Graph Neural Network Surrogates for Molecular Dynamics Across Crystal Symmetries. ChemRxiv; 2026; Preprint v1; not peer reviewed. DOI: 10.26434/chemrxiv.15009166/v1. Accessed 2026-09-21T16:12:37.222Z.
Source evidence and access
Publisher-deposited abstract, https://api.crossref.org/works/10.26434/chemrxiv.15009166/v1
We present a transferable graph neural network (GNN) surrogate framework for molecular dynamics (MD) that directly predicts atomic displacements and
Abstract only: publisher-deposited primary metadata through Crossref; full text not inspected.
Felix Ærtebjerg; Jonas Elsborg; Arghya Bhowmik. Complete Neural Electronic Initialization Accelerates Materials DFT. arXiv; 2026; Preprint v1; not peer reviewed. DOI: 10.48550/arXiv.2609.21759. Accessed 2026-09-21T16:12:37.222Z.
Source evidence and access
Abstract and v1 submission history: 18 September 2026 13:30:29 UTC
reduces end-to-end DFT wall time by up to ~25% on unseen structures while preserving converged energies
Original repository abstract and submission history inspected; full paper not inspected.
Adit S. Desai; Yongqiang Cheng; Joseph A. M. Paddison. Machine learning magnetic interactions from neutron powder diffraction data. arXiv; 2026; Preprint v1; not peer reviewed. DOI: 10.48550/arXiv.2609.21970. Accessed 2026-09-21T16:12:37.222Z.
Source evidence and access
Abstract and v1 submission history: 18 September 2026 16:32:40 UTC
a comprehensive survey of isotropic interactions on eight high-symmetry lattices
Original repository abstract and submission history inspected; full paper not inspected.
Publication record
Published 2026-09-21.
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.
- 2026-09-21 · Version fa2945bf · Viewing this version