| 1. |
- Fan, Zheyong, et al.
(författare)
-
GPUMD: A package for constructing accurate machine-learned potentials and performing highly efficient atomistic simulations
- 2022
-
Ingår i: Journal of Chemical Physics. - : AIP Publishing. - 1089-7690 .- 0021-9606. ; 157:11
-
Tidskriftsartikel (refereegranskat)abstract
- We present our latest advancements of machine-learned potentials (MLPs) based on the neuroevolution potential (NEP) framework introduced in Fan et al. [Phys. Rev. B 104, 104309 (2021)] and their implementation in the open-source package gpumd. We increase the accuracy of NEP models both by improving the radial functions in the atomic-environment descriptor using a linear combination of Chebyshev basis functions and by extending the angular descriptor with some four-body and five-body contributions as in the atomic cluster expansion approach. We also detail our efficient implementation of the NEP approach in graphics processing units as well as our workflow for the construction of NEP models and demonstrate their application in large-scale atomistic simulations. By comparing to state-of-the-art MLPs, we show that the NEP approach not only achieves above-average accuracy but also is far more computationally efficient. These results demonstrate that the gpumd package is a promising tool for solving challenging problems requiring highly accurate, large-scale atomistic simulations. To enable the construction of MLPs using a minimal training set, we propose an active-learning scheme based on the latent space of a pre-trained NEP model. Finally, we introduce three separate Python packages, viz., gpyumd, calorine, and pynep, that enable the integration of gpumd into Python workflows.
|
|
| 2. |
- Wu, Zhendong, et al.
(författare)
-
Approaching the potential of model-data comparisons of global land carbon storage
- 2019
-
Ingår i: Scientific Reports. - : Springer Science and Business Media LLC. - 2045-2322. ; 9
-
Tidskriftsartikel (refereegranskat)abstract
- Carbon storage dynamics in vegetation and soil are determined by the balance of carbon influx and turnover. Estimates of these opposing fluxes differ markedly among different empirical datasets and models leading to uncertainty and divergent trends. To trace the origin of such discrepancies through time and across major biomes and climatic regions, we used a model-data fusion framework. The framework emulates carbon cycling and its component processes in a global dynamic ecosystem model, LPJ-GUESS, and preserves the model-simulated pools and fluxes in space and time. Thus, it allows us to replace simulated carbon influx and turnover with estimates derived from empirical data, bringing together the strength of the model in representing processes, with the richness of observational data informing the estimations. The resulting vegetation and soil carbon storage and global land carbon fluxes were compared to independent empirical datasets. Results show model-data agreement comparable to, or even better than, the agreement between independent empirical datasets. This suggests that only marginal improvement in land carbon cycle simulations can be gained from comparisons of models with current-generation datasets on vegetation and soil carbon. Consequently, we recommend that model skill should be assessed relative to reference data uncertainty in future model evaluation studies.
|
|
