Skip to content
DraconiticResearch
Search Sources API
226,742 studies31 sourcesUpdated 16 min ago
PreprintCC BY 4.0arXiv

Hybrid Machine Learning for Physically Consistent Prediction of Mutual Diffusion Coefficients in Binary Liquid Mixtures

Abstract

Experimental data on mutual diffusion coefficients in liquid mixtures are scarce, and existing prediction methods are unreliable for non-ideal mixtures. Building on HADES, our hybrid model for self-diffusion coefficients, we introduce the Enhanced Vignes Equation (EVE), which augments the Vignes equation with a deep-set model. Its inputs are molecular descriptors derived from SMILES and activity-coefficient information from HANNA, a machine-learning model for activity coefficients. EVE predicts Maxwell--Stefan diffusion coefficients; the corresponding Fickian coefficients follow from the thermodynamic factor, likewise obtained from HANNA. Trained and evaluated on 2106 experimental data points for 123 binary mixtures, EVE outperforms all available semi-empirical models, halving the mean absolute relative error for mixtures without experimental infinite-dilution diffusion data. Combined with HADES, EVE yields the complete set of self- and mutual diffusion coefficients of a binary mixture from SMILES and pure-component viscosities alone. EVE and HADES are available at https://ml-prop.mv.rptu.de/, and the code is fully disclosed.

Related

  1. Grassmannian Density Learning for Physically Consistent Kohn–Sham Electronic Structure Prediction
    Yifei Wang, Pei Liu, Yuheng He +7
    ChemRxivPreprintCC BY-NC-NDFull textMaterials & polymersChemRxiv
  2. Machine learning prediction of DFT-derived electrical conductivity in perovskite and double perovskite materials using physically motivated descriptors
    Fatemeh Mohammad Dezashibi, Farinaz Roshani
    Scientific ReportsPeer-reviewedCC BYFull textMaterials & polymersDOAJ
  3. Physically aligned forest fire risk prediction: A deep learning framework coupling fuel and climate multivariate factors
    Bojie Chen, Anping Zeng, Yu Xie +4
    PLOS OnePeer-reviewedCC BY 4.0Full textEcologyMaterials & polymersPLOS
  4. Composition-Dependent Self-Diffusion Coefficients in Liquid Mixtures from Hybrid Machine Learning
    Jens Wagner, Thomas Specht, Hans Hasse +1
    arXivPreprintCC BY 4.0Full textPhysical chemistryarXiv
  5. JAI: Physically Constrained Prediction of Conformer-Averaged Proton–Proton Coupling Constants in Solution
    Carlos Cobas
    ChemRxivPreprintLicence not statedGeneral chemistryChemRxiv
BibTeX
@misc{wagner2026hybrid,
  title = {{Hybrid Machine Learning for Physically Consistent Prediction of Mutual Diffusion Coefficients in Binary Liquid Mixtures}},
  author = {Jens Wagner and Thomas Specht and Hans Hasse and Fabian Jirasek},
  howpublished = {arXiv},
  year = {2026},
  month = nov,
  doi = {10.48550/arxiv.2611.15235},
  eprint = {2611.15235},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.15235},
  note = {CC BY 4.0},
}
RIS
TY  - UNPB
TI  - Hybrid Machine Learning for Physically Consistent Prediction of Mutual Diffusion Coefficients in Binary Liquid Mixtures
AU  - Jens Wagner
AU  - Thomas Specht
AU  - Hans Hasse
AU  - Fabian Jirasek
PY  - 2026
DA  - 2026/11/11
PB  - arXiv
DO  - 10.48550/arxiv.2611.15235
UR  - https://arxiv.org/abs/2611.15235
AB  - Experimental data on mutual diffusion coefficients in liquid mixtures are scarce, and existing prediction methods are unreliable for non-ideal mixtures. Building on HADES, our hybrid model for self-diffusion coefficients, we introduce the Enhanced Vignes Equation (EVE), which augments the Vignes equation with a deep-set model. Its inputs are molecular descriptors derived from SMILES and activity-coefficient information from HANNA, a machine-learning model for activity coefficients. EVE predicts Maxwell--Stefan diffusion coefficients; the corresponding Fickian coefficients follow from the thermodynamic factor, likewise obtained from HANNA. Trained and evaluated on 2106 experimental data points for 123 binary mixtures, EVE outperforms all available semi-empirical models, halving the mean absolute relative error for mixtures without experimental infinite-dilution diffusion data. Combined with HADES, EVE yields the complete set of self- and mutual diffusion coefficients of a binary mixture from SMILES and pure-component viscosities alone. EVE and HADES are available at https://ml-prop.mv.rptu.de/, and the code is fully disclosed.
ER  -