Response to Comment on “Pushing the frontiers of density functionals by solving the fractional electron problem”

Abstract

Gerasimov et al . claim that the ability of DM21 to respect fractional charge (FC) and fractional spin (FS) conditions outside of the training set has not been demonstrated in our paper. This is based on (i) asserting that the training set has a ~50% overlap with our bond-breaking benchmark (BBB) and (ii) questioning the validity and accuracy of our other generalization examples. We disagree with their analysis and believe that the points raised are either incorrect or not relevant to the main conclusions of the paper and to the assessment of general quality of DM21.

Toward the Next Generation of Density Functionals: Escaping the Zero-Sum Game by Using the Exact-Exchange Energy Density

“…This should not be too surprising, as DM21 has been trained with data very close to those used for the plot (see section 5). Discussions on the transferability of DM21 have started 99,100 and will certainly continue. We found that, for example, scLH22t performs better than DM21 for the "multireference" MR16 subset of the W4−11 atomization energy benchmark.…”

Mateverse, the Future Materials Science Computation Platform Based on Metaverse

“…The significance of human intelligence over AI or ML models, to some extent, is the deep understanding of nature. Whether the deep learning model can fully understand the physics on the electron scale remains controversial as a result of the possible improper choice of test data. − In a general optimization scheme, researchers tend to consider a linear combination of all possible mathematical bases. Here, the linear coefficient has two functions.…”