Table 3 The mean absolute error (MAE) of atomic forces (meV Å−1) of three bulk systems and three molecular systems.
From: Accurate and efficient molecular dynamics based on machine learning and non von Neumann architecture
Method | Computer architecture | Bulk systems | Molecular system | |||||
|---|---|---|---|---|---|---|---|---|
Sb | GeTe | Li-Ge-P-S | Benzene | Aspirin | Naphthalene | |||
NVNMD (this work) | QNN (after quantization) | NvN | 85.1 | 161.0 | 68.1 | 11.9 | 28.8 | 18.8 |
CNN (before quantization) | vN | 79.4 | 161.0 | 65.9 | 6.7 | 24.7 | 10.3 | |
MLMD | DeePMD30 | vN | 64.2 | 168.0 | 79.6 | -- | 19.4 | 13.1 |
SchNet84 | – | – | – | 7.4 | 14.3 | 4.8 | ||
DimeNet85 | – | – | – | 8.1 | 21.6 | 9.3 | ||
sGDML86 | – | – | – | 2.6 | 29.5 | 4.7 | ||
PaiNN87 | – | – | – | – | 16.1 | 3.6 | ||
SpookyNet88 | – | – | – | – | 11.2 | 3.9 | ||
GemNet89 | – | – | – | 6.3 | 9.5 | 2.4 | ||
NewtonNet90 | – | – | – | – | 15.1 | 3.6 | ||
UNiTE91 | – | – | – | – | 6.1 | 1.5 | ||
NequIP92 | – | – | – | – | 8.2 | 1.6 | ||
μF | 64.2 | 168.0 | 79.6 | 6.1 | 15.1 | 4.9 | ||
Difference between QNN (i.e., final results of NVNMD) and μF of MLMD | 20.9 | −7.0 | −11.5 | 5.8 | 13.7 | 13.9 | ||
