The document presents duality theory for composite geometric programs (CGPs), which include exponential GPs (EGPs) as a special case. An EGP allows some posynomial terms in the objective function to be multiplied by an exponential factor of another posynomial term. The key results are:
1) EGP problems can be formulated as convex programs by a change of variables.
2) The dual problem of an EGP is a posynomial program.
3) Strong duality holds between primal and dual EGP programs, and optimal solutions can be recovered from each other using extremality conditions.
4) Motivating examples like maximum likelihood estimation of Poisson and exponential parameters can be solved as