Daniel Cederberg - Disciplined nonlinear programming

In the first part of this talk, I give a brief overview of disciplined convex programming (DCP), an existing syntax for specifying convex optimization problems. DCP underlies several widely used optimization modeling languages, including CVXPY.

In the second part, I introduce disciplined nonlinear programming (DNLP), a new syntax for specifying nonlinear programming problems. DNLP allows smooth functions to be freely mixed with nonsmooth convex and concave functions, with rules governing how the nonsmooth functions can be used. Problems expressed in DNLP form can be automatically canonicalized to a standard nonlinear programming (NLP) form and passed to a suitable NLP solver.

I conclude by describing the DNLP language in more detail, comparing it with existing NLP modeling languages, and presenting our open-source implementation of DNLP as an extension of CVXPY.

Accompanying draft: https://stanford.edu/~boyd/papers/pdf/dnlp.pdf