takafumi-shiraogawa / apdft Goto Github PK

The following values need to be controlled by apdft.conf:

• SCF energy
• CCSD energy and amplitude
• KS-DFT grid
• Integral screening in the direct SCF

I found numerical instability in analytical calculation results of electronic dipole moments by PySCF. For the computational method, I used CCSD/def2-TZVP, and the input is automatically generated by the APDFT code. The target molecule is N2 ("7 0.000000 0.000000 0.000000;7 0.000000 0.000000 1.000000"). The z-component of the dipole moment varied in the calculations. However, electrostatic potential was numerically stable.

I calculated atomic forces of CO with 1.2 angstrom of the bond distance d based on N2 with d = 1.0 angstrom. The resultant atomic forces are too large. This comes from the term with respect to derivatives of the perturbed electron density.

In a vertical APDFT calculation of CO from N2 along with the z direction with the bond length of 1.1 angstrom, I found that the electric dipole moment has non-negligible x and y components. When the bond length of 1.0 angstrom, the same results were found.

Numerical integration for calculations of perturbed electron densities sometimes fails.

Need to consider cross terms for three Cartesian coordinates in QM calculations for numerical differentiation.

Now vertical energy derivatives are strange. In ver_ele_atomic_forces.csv, the sum of the atomic forces is not zero. However, in numerical evaluations, the sum is zero. I guess that ver_contributions_deriv_rho has errors.

In the current implementation, numerical grids are not suitable to estimate properties of target molecules with different atomic coordinates with the reference. This is because grids are large rough for the new coordinate.

The mixed second derivatives of alphas (coefficients of mixed second derivatives of the reference density) should be twice in the code which only calculate coefficients for combinations of two variables. I do not understand how the code accomplishes this.

Intra- and inter-node parallelization of QM calculations should be controlled by inputs in apdft.conf.