The problem arises from trying to construct an error envelope over predicted model data that is to be compared in a likelihood-maximisation routine with observed data.
Th
