Dear CC?,
yes the basis functions are atom centered and hence move with the atoms causing
that the overlap integrals needs to be recomputed whenever the positions are updated
(e.g. geometry optimization, molecular dynamics, etc.). By default in CP2K ab-initio
MD simulations are conducted by means of the Born-Oppenheimer scheme (the
original CPMD method is not implemented), i.e. in every time-step a conventional
electronic structure calculation is conducted.
The de facto standard in CP2K are so-called „molecularly optimized“ basis set (aka
MOLOPT basis), whose Gaussian exponents and contraction coefficients are optimized
based on molecular systems using an objective function that is the sum of the total
energy and the logarithm of the condition number of the overlap matrix. In this way,
contrary to typical augmented basis sets, the overlap matrix is generally well conditioned,
so that there are NO near linear dependencies in the basis, even for periodic systems!