[benchmark] MD+NNP simulations of H, D, and T on Be

[drobnyjt]

Shermukhamedov et al. have a paper out on relatively low-energy H, D, and T sputtering and reflection using MD with neural-network potentials. This could be a decent target for benchmarking the results of the attractive-repulsive potentials.

https://doi.org/10.1088/1741-4326/ac592a

[drobnyjt]

Data digitized from above:

H on Be

20.004726, 0.0029361406
29.880848, 0.013341782
49.791046, 0.024575423
75.00529,   0.016207783
99.68844,   0.017530797

T on Be

20.013433, 0.0025699555
29.85846,   0.01879205
49.756844, 0.04147379
74.707405, 0.034042966
99.68098, 0.01965117

D on Be

19.9699, 0.005002439
29.601622, 0.021064475
49.76866,   0.03461476
74.71549,   0.030082114
99.954605, 0.013559438

H on Be

9.67742 0.016754618
20.32258    0.09511873
30.0    0.13153034
49.677418   0.102242745
74.83871    0.08403694
100.32258   0.031002639

D on Be

10.32258    0.11649077
20.32258    0.19168866
30.32258    0.19722955
49.677418   0.17269129
75.16129    0.11174142
100.0   0.044459105

T on Be

9.67742 0.19485489
19.67742    0.32150397
30.0    0.33575198
50.322582   0.25105542
75.16129    0.1378628
100.645164  0.044459105

[dcurreli]

Thanks Jon, impressive that in 2022 we still need to digitalize figures...

[drobnyjt]

I think there are ways to do it automatically now, but I'm too stubborn to trust a program to do it right, haha.

[dcurreli]

Every journal should request the data for each published figure, downloadable with the push of a click

[drobnyjt]

Using the RustBCA default options, the BCA appears to perform relatively poorly compared to this MD benchmark. There are two caveats - first, this data is significantly off from both Yamamura and Thomas, so I am not sure how well it compares to experiment in the first place. Second, this is a worst-case-scenario for the default options, since H and Be interact quite strongly with each other, and in fact chemical sputtering of Be by H is expected - the default options are for Kr-C and the individual/planar surface binding model - the next step would be to use the developed NNP potentials with the CPR rootfinder and see how the results compare then.

Reflection

Sputtering

[dcurreli]

Excellent analysis. I was expecting disagreement for hydrogen isotopes on Be for standard, purely-repulsive potentials. Do we have access to an expression of the NNP?

[drobnyjt]

I thought it was in the supplementary material, but I cannot find it now... However, they cite Bjorkas et al. who have an expression for an interaction potential:

https://iopscience.iop.org/article/10.1088/0953-8984/21/44/445002

And there's always a fitted Morse potential; I think I know of a routine for any H-metal interaction that's in an old paper somewhere.

[drobnyjt]

There are also the potentials from the PSI-SciDAC folks as well, if I recall correctly

[dcurreli]

Yes, I am sure we can find it