Skip to content
DraconiticResearch
Search Sources API
197,768 studies31 sourcesUpdated 21 min ago
PreprintCC BY-NC-NDChemRxiv

Harnessing Machine Learning and Composite Quantum Chemical Methods for Computing Accurate Bond Dissociation Energies of PFAS

  • Kalpak Ghosh1,2,3
  • Samir Kumar Nayak1,2,4
  • Rohit Batra4,2,5
  • Sharma S. R. K. C. Yamijala1,2,3,4,6
6 affiliations
  1. 1Department of Chemistry
  2. 2Indian Institute of Technology Madras
  3. 3Center for Quantum Information, Communication, and Computing
  4. 4Center for Atomistic Modeling and Materials Design
  5. 5Department of Metallurgical and Materials Engineering
  6. 6Center for Molecular Materials and Functions
Open at sourcePDFDOICite

Abstract

Designing efficient degradation pathways for per- and polyfluoroalkyl substances (PFAS) requires access to accurate bond dissociation energies (BDEs). Composite quantum chemical methods such as G4 provide high accuracy (<1 kcal mol–1) but are limited to small molecules. Density functional theory (DFT) can handle larger systems, but its results vary significantly with the choice of functional and basis set, raising concerns about reliability. Fragment-based methods like the connectivity-based hierarchy (CBH) and its modified form (mCBH) offer G4-level accuracy at DFT cost, but only when using large fragments. Since obtaining G4 energies for large fragments is computationally demanding, these methods are not practical for building large BDE databases. To address this limitation, we introduce a machine learning (ML)-enhanced CBH approach. Here, a regression model is trained on G4-calculated energies of small fragments and is used to predict PFAS BDEs. This strategy achieves near-G4 accuracy without requiring G4 calculations on large fragments. Across diverse PFAS with varying chain lengths and functional groups, our ML-augmented fragmentation-based methods either match or outperform the accuracy of both the mCBH-3 and CBH-3 approaches (both of which require larger fragments), demonstrating that chemical accuracy can be achieved with smaller fragments. By integrating fragment-level quantum data with ML, this hybrid framework provides an efficient pathway for generating reliable BDE datasets and accelerating the discovery of effective PFAS degradation strategies.

Related

  1. Harnessing Machine Learning for Ultrafast Science in Facility Settings: Opportunities and Challenges
    Mousumi Upadhyay Kahaly, Ishraq Md Anjum, Davorin Peceli +2
    arXivPreprintCC BY-NC-ND 4.0Full textApplied physics & instrumentationOptics & photonicsarXiv
  2. Absolute phase recovery from reduced phase-shifted fringe patterns via composite-loss-driven deep learning
    Guorui Zhao, Qiang Liu, Tianhao Wu +4
    Optics ExpressPeer-reviewedLicence not statedOptics & photonicsCrossref
  3. Learning Temporal Patterns in Financial Time Series: A Comparative Study of Quantum LSTM and Quantum Reservoir Computing
    Danyal Maheshwari, Gerhard Hellstern, Martin Zaefferer +2
    arXivPreprintCC BY 4.0Full textQuantum physics & quantum informationarXiv
  4. Multifunctional synaptic learning and neuromorphic computing using crystalline TiOx/NiOx heterojunction-based memory devices
    Dip Manna, Koushik Mondal, Subhajit Karmakar +5
    arXivPreprintarXiv licenceFull textCondensed matterarXiv
  5. Online local learning for generative thermodynamic computing
    Huilin Wang, Weibing Deng
    arXivPreprintarXiv licenceFull textComputational physicsarXiv

Cite

BibTeXRIS
BibTeX
@misc{ghosh2026harnessing,
  title = {{Harnessing Machine Learning and Composite Quantum Chemical Methods for Computing Accurate Bond Dissociation Energies of PFAS}},
  author = {Kalpak Ghosh and Samir Kumar Nayak and Rohit Batra and Sharma S. R. K. C. Yamijala},
  howpublished = {ChemRxiv},
  year = {2026},
  month = sep,
  doi = {10.26434/chemrxiv.15009576/v1},
  url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009576/v1},
  note = {CC BY-NC-ND},
}
RIS
TY  - UNPB
TI  - Harnessing Machine Learning and Composite Quantum Chemical Methods for Computing Accurate Bond Dissociation Energies of PFAS
AU  - Kalpak Ghosh
AU  - Samir Kumar Nayak
AU  - Rohit Batra
AU  - Sharma S. R. K. C. Yamijala
PY  - 2026
DA  - 2026/09/28
PB  - ChemRxiv
DO  - 10.26434/chemrxiv.15009576/v1
UR  - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009576/v1
AB  - Designing efficient degradation pathways for per- and polyfluoroalkyl substances (PFAS) requires access to accurate bond dissociation energies (BDEs). Composite quantum chemical methods such as G4 provide high accuracy (<1 kcal mol–1) but are limited to small molecules. Density functional theory (DFT) can handle larger systems, but its results vary significantly with the choice of functional and basis set, raising concerns about reliability. Fragment-based methods like the connectivity-based hierarchy (CBH) and its modified form (mCBH) offer G4-level accuracy at DFT cost, but only when using large fragments. Since obtaining G4 energies for large fragments is computationally demanding, these methods are not practical for building large BDE databases. To address this limitation, we introduce a machine learning (ML)-enhanced CBH approach. Here, a regression model is trained on G4-calculated energies of small fragments and is used to predict PFAS BDEs. This strategy achieves near-G4 accuracy without requiring G4 calculations on large fragments. Across diverse PFAS with varying chain lengths and functional groups, our ML-augmented fragmentation-based methods either match or outperform the accuracy of both the mCBH-3 and CBH-3 approaches (both of which require larger fragments), demonstrating that chemical accuracy can be achieved with smaller fragments. By integrating fragment-level quantum data with ML, this hybrid framework provides an efficient pathway for generating reliable BDE datasets and accelerating the discovery of effective PFAS degradation strategies.
ER  -