Comment on "Pushing the frontiers of density functionals by solving the fractional electron problem"

Area of Science:

• Computational chemistry
• Quantum mechanics
• Materials science

Background:

• The DM21 density functional was developed using neural networks and trained on fractional charge (FC) and fractional spin (FS) systems.
• Kirkpatrick et al. claimed DM21 exhibits high accuracy for strongly correlated chemical systems.

Purpose of the Study:

• To critically evaluate the generalization capabilities of the DM21 density functional.
• To reassess the claimed accuracy of DM21 for systems with strong correlation.

Main Methods:

• Analysis of the generalization performance of DM21.
• Revisiting the published results concerning DM21's accuracy on FC and FS systems.

Main Results:

• The ability of DM21 to generalize to strongly correlated systems is not supported by the published evidence.
• The claimed outstanding accuracy of DM21 requires further investigation and validation.

Conclusions:

• The generalization performance of DM21 for strongly correlated systems needs re-evaluation.
• The findings challenge the robustness of DM21's accuracy claims.