The first use of the Gaussian very fast multipole method (GVFMM) for c
alculating the integral derivatives that arise in the Coulomb terms of
density-functional theory (DFT) energy gradients is reported. Tests o
f the GvFMM gradient algorithm indicate that its accuracy, speed, and
near-linear scaling behavior are similar to the GvFMM molecular energy
algorithm. Specifically, 10(-8) hartree per Bohr accuracy in the ener
gy gradient has been achieved, and the ratio of the computational cost
for the GvFMM energy gradient to the GVFMM energy has been found to b
e lower than the ratio for previous state-of-the-art method.