Linearization of a non-linear function
The log-sum-exp function

admits the gradient at the point given by

Hence can be approximated near
by the linear function

The log-sum-exp function
admits the gradient at the point given by
Hence can be approximated near
by the linear function