Specifies the factor by which the inverse_parameter (same index)
is varied a bit to determine the sensitivity of the targets
with respect to the parameter.
A relative variation is used: (varied factor = (1+variation)*factor.
On the one hand, it is of importance to use a variation small enough to get a good
approximation of the sensitivity of the targets with respect to the
parameter.
On the other hand the variation should not be such small that we
run into problems with the computer accuracy.
Default, variation is .
See also inverse_target.