Table 1 Calculation accuracy comparison between the proposed NVNMD and the established MLMD/CMD.

Method	System	Computer architecture	Computing hardware	\|∆E_i\| (kcal mol⁻¹)	μ_e (kcal mol⁻¹)
NVNMD (this work)	Benzene	NvN	FPGA	0.19	0.21
	Naphthalene		FPGA	0.39
	Aspirin		FPGA	0.32
	Antimony (bulk)		FPGA	0.14
	Germanium telluride (bulk)		FPGA	0.09
	Li-Ge-P-S (bulk)		FPGA	0.14
MLMD	Benzene^26,27,31,58	vN	GPU	0.07/0.07	0.17
	Uracil^26,27,31,58		GPU	0.09/0.11
	Naphthalene^26,27,31,58		GPU	0.12/0.12
	Aspirin^26,27,31,58		GPU	0.27/0.27
	Salicylic acid^26,27,31,58		GPU	0.12/0.12
	Malonaldehyde^26,27,31,58		GPU	0.16/0.16
	Ethanol^26,27,31,58		GPU	0.15/0.15
	Toluene^26,27,31,58		GPU	0.12/0.12
	Water^26,27,31		GPU	0.01
	Glycine proton transfer^65,96		CPU	0.3–0.4
	Acetic acid^65,66		CPU	0.2
	Acetamide^65,66		CPU	0.4
	Acetone^65,66		CPU	0.4
	Ethanol^65,66		CPU	0.2
	Germanium telluride⁵⁶		GPU	0.09
	Li-Ge-P-S⁶²		GPU	0.04
CMD	700 different molecular structures⁹³	vN	GPU/CPU	10 (est.)	≈10
	Small-molecule solvation and protein-ligand binding⁶⁷		GPU/CPU	3.6–6.3
	AMBER intermolecular terms w.r.t. DFT-SAPT⁹⁴		GPU/CPU	10 (est.)
	Nobel gas absorption in metal organic framework⁹⁵		GPU/CPU	8.4-12.6

∆E_i is the RMSE with respect to ab-initio calculations. μ_e is the average value of |∆E_i|. The abbreviation FPGA represents field programmable gate array.

Quick links

Search