Coek Overview

Motivation

Coek is a C++ library for expressing and solving optimization problems. Coek enables users to express optimization models in C++, interface directly with optimization solvers, and interact with solutions generated by solvers. The design of Coek was broadly inspired by the object-oriented design in [Gurobi], but Coek is designed to support the expression of mixed-integer nonlinear programs.

Coek is closely developed with the [Poek] library, and the goal of these libraries is to support fast optimization modeling capabilities in both C++ and Python. For example:

• Coek can express large optimization problems in C++ nearly as fast [Gurobi],

• Coek integrates [CppAD] to compute derivatives for nonlinear problems,

• Coek and [Poek] include mutable parameters that allow model constants to be modified in place, and

• Poek can express large optimization problems in Python with modest overhead.

Coek development is being driven by targetted experiments and demonstrations, often with runtime performance as a major driver. Thus, Coek is similar to [Gravity]. Perhaps the biggest difference is that Coek does not focus on vectorized representations. Poek is much simpler than [Pyomo], but it can build models faster.

Although the Coek API has stabilized, it continues to evolve as new capabilities are demonstrated. Consequently, the following documentation does not provide a detailed specification of the Coek API. Rather, we focus on examples that illustrate Coek capabilities.

Capabilities

• Solvers

  • Ipopt - Coek can dynamically load the Ipopt C interface.

  • Gurobi - Coek links with the Gurobi C++ interface

• Automatic Differentiation

  • CppAD and ASL are currently supported

  • There are general hooks in Coek for alternative AD packages

• PyCoek

  • Python wrappers are supported using pybind11 and cppyy

  • The [Poek] library simplifies the use of PyCoek for end-users

• Documentation

  • Sphinx, Doxygen and Breathe are used to generate documentation