Why are the gradients in an optimization sometimes calculated numerically?
For versions of Jaguar before 6.0, if the basis set contains f functions (as in cc-pVTZ), Jaguar calculates the gradients numerically. Using basis sets that do not contain f functions (such as cc-pVTZ(-f) instead of cc-pVTZ) allows the gradients to be calculated analytically. Jaguar 6.0 has support for analytic gradients for f functions.
