Table 1 Performance of ML models compared to other popular water models

From: Machine learning coarse grained models for water

 

ML-BOP

ML-BOP /dih

ML-mW

mW

TIP4P /2005

MB-pol

iAMOEBA

SPC/E

TIP3P

Neighbors3.3 Å, 298 K

9

9

9

10

9

9

9

9

10

resRDF, 298 K

3

4

8

8

7

9

7

6

0

resADF, 298 K

7

7

7

7

6

7

7

6

6

ln D298 K

5

5

0

0

8

9

8

8

0

ρ298K, 1atma

10

10

10

10

9

8

10

9

7

ρmaxa

10

10

10

9

10

7

10

8

2

TMDa

10

10

10

6

10

7

10

5

0

ΔHvap

8

8

9

9

4

8

8

6

10

Tm

10

10

8

10

7

9

7

1

0

TMD - Tm

10

10

7

4

6

8

7

6

4

ΔHmelta

7

7

9

8

6

7

0

0

ΔSmelt

7

7

8

7

7

7

3

0

ρliq at Tma

10

10

10

10

9

7

10

8

6

ρIh at Tma

7

7

8

0

9

9

7

3

3

ΔVmelt

7

7

7

0

8

8

7

4

6

(dp/dT)melt

9

10

9

0

10

9

8

0

Average score

8.1

8.2

8.1

6.1

7.8

8.1

8.2

5.6

3.4

  1. Comparison of the performance of ML models with other popular polarizable8,19 and non-polarizable models26. The numerical scores and tolerance are assigned based on an established system by Vega59. A list of ice and liquid water properties relevant to the capability of ML-BOP models are selected for comparison
  2. aProperties that are included in the training of ML models