Research paperComputational MLIPComputational DFTComputational MDComputational AimdA Planning-and-Exploring Approach to Extreme-Mechanics Force FieldsPengjie Shi, Zhiping XuarXiv preprint·2023·10.1088/1361-648X/ad5c31·arXiv:2310.19306AbstractWe develop a planning-and-exploring strategy to build a neural network-based force field for fracture (NN-F₃) by combining pre-sampling of strain states with active learning to capture both equilibrium and rare-event non-equilibrium fracture physics. The model is trained on DFT-generated data and validated on h-BN and graphene fracture problems, including twisted bilayer graphene and finite-temperature bond-breaking processes, demonstrating DFT-level accuracy at much larger scales than direct first-principles calculations.Read more
DFT-labeled graphene strain-state and fracture training configurations used for NN-F₃ development.1 propertySimulated Supercell DftCStudied MaterialExpand
DFT-labeled h-BN strain-state and fracture training configurations used for NN-F₃ development.1 propertySimulated Supercell DftBNStudied MaterialExpand
Finite-temperature BN chain/net structures assessed with NN-F₃ and BOMD.2 propertiesSimulatedBNStudied MaterialExpand
Twisted bilayer graphene fracture and crack-interaction models simulated with NN-F₃ plus interlayer LJ interaction.No measurements recordedSimulated Supercell DftCStudied MaterialExpand
Research paperComputational MLIPComputational DFTComputational MDComputational AimdA Planning-and-Exploring Approach to Extreme-Mechanics Force FieldsPengjie Shi, Zhiping XuarXiv preprint·2023·10.1088/1361-648X/ad5c31·arXiv:2310.19306AbstractWe develop a planning-and-exploring strategy to build a neural network-based force field for fracture (NN-F₃) by combining pre-sampling of strain states with active learning to capture both equilibrium and rare-event non-equilibrium fracture physics. The model is trained on DFT-generated data and validated on h-BN and graphene fracture problems, including twisted bilayer graphene and finite-temperature bond-breaking processes, demonstrating DFT-level accuracy at much larger scales than direct first-principles calculations.Read more
DFT-labeled graphene strain-state and fracture training configurations used for NN-F₃ development.1 propertySimulated Supercell DftCStudied MaterialExpand
DFT-labeled h-BN strain-state and fracture training configurations used for NN-F₃ development.1 propertySimulated Supercell DftBNStudied MaterialExpand
Finite-temperature BN chain/net structures assessed with NN-F₃ and BOMD.2 propertiesSimulatedBNStudied MaterialExpand
Twisted bilayer graphene fracture and crack-interaction models simulated with NN-F₃ plus interlayer LJ interaction.No measurements recordedSimulated Supercell DftCStudied MaterialExpand
Research paperComputational MLIPComputational DFTComputational MDComputational AimdA Planning-and-Exploring Approach to Extreme-Mechanics Force FieldsPengjie Shi, Zhiping XuarXiv preprint·2023·10.1088/1361-648X/ad5c31·arXiv:2310.19306AbstractWe develop a planning-and-exploring strategy to build a neural network-based force field for fracture (NN-F₃) by combining pre-sampling of strain states with active learning to capture both equilibrium and rare-event non-equilibrium fracture physics. The model is trained on DFT-generated data and validated on h-BN and graphene fracture problems, including twisted bilayer graphene and finite-temperature bond-breaking processes, demonstrating DFT-level accuracy at much larger scales than direct first-principles calculations.Read more
DFT-labeled graphene strain-state and fracture training configurations used for NN-F₃ development.1 propertySimulated Supercell DftCStudied MaterialExpand
DFT-labeled h-BN strain-state and fracture training configurations used for NN-F₃ development.1 propertySimulated Supercell DftBNStudied MaterialExpand
Finite-temperature BN chain/net structures assessed with NN-F₃ and BOMD.2 propertiesSimulatedBNStudied MaterialExpand
Twisted bilayer graphene fracture and crack-interaction models simulated with NN-F₃ plus interlayer LJ interaction.No measurements recordedSimulated Supercell DftCStudied MaterialExpand
Research paperComputational MLIPComputational DFTComputational MDComputational AimdA Planning-and-Exploring Approach to Extreme-Mechanics Force FieldsPengjie Shi, Zhiping XuarXiv preprint·2023·10.1088/1361-648X/ad5c31·arXiv:2310.19306AbstractWe develop a planning-and-exploring strategy to build a neural network-based force field for fracture (NN-F₃) by combining pre-sampling of strain states with active learning to capture both equilibrium and rare-event non-equilibrium fracture physics. The model is trained on DFT-generated data and validated on h-BN and graphene fracture problems, including twisted bilayer graphene and finite-temperature bond-breaking processes, demonstrating DFT-level accuracy at much larger scales than direct first-principles calculations.Read more
DFT-labeled graphene strain-state and fracture training configurations used for NN-F₃ development.1 propertySimulated Supercell DftCStudied MaterialExpand
DFT-labeled h-BN strain-state and fracture training configurations used for NN-F₃ development.1 propertySimulated Supercell DftBNStudied MaterialExpand
Finite-temperature BN chain/net structures assessed with NN-F₃ and BOMD.2 propertiesSimulatedBNStudied MaterialExpand
Twisted bilayer graphene fracture and crack-interaction models simulated with NN-F₃ plus interlayer LJ interaction.No measurements recordedSimulated Supercell DftCStudied MaterialExpand
