OptimizationSNOPT.jl: SNOPT for Optimization.jl

OptimizationSNOPT.jl connects SNOPT, a commercial solver for large, sparse, constrained nonlinear optimization problems, to Optimization.jl. A problem defined as an OptimizationProblem is solved with SNOPT by passing SnoptOptimizer to solve, and the derivatives that SNOPT needs come from any automatic differentiation backend that Optimization.jl supports, so the objective and the constraints can stay ordinary Julia functions.

OptimizationSNOPT.jl solves problems of the form

\[\begin{aligned} \min_{x \in \mathbb{R}^n} \quad & f(x, p) \\ \text{subject to} \quad & l_x \le x \le u_x, \\ & l_c \le c(x, p) \le u_c, \end{aligned}\]

where $f$ is the objective, $c$ is a vector of constraint functions, $p$ holds the parameters, and the bounds can be infinite. SNOPT uses a sequential quadratic programming (SQP) method, so $f$ and $c$ must be smooth.

When to use OptimizationSNOPT.jl

With OptimizationSNOPT.jl, the same problem can be solved with SNOPT and with the other Optimization.jl solvers, and the derivatives come from automatic differentiation. The package is built on SNOPT.jl, which wraps the three SNOPT interfaces snOptA, snOptB and snOptC; use SNOPT.jl directly for linear constraints given separately, finite-difference derivatives estimated by SNOPT, or warm and hot starts.

Installation

Install the package with the Julia package manager:

using Pkg
Pkg.add("OptimizationSNOPT")

Then, before starting Julia, set SNOPTDIR to the directory of the SNOPT library:

export SNOPTDIR=/path/to/snopt/lib

Installation gives the commands for Windows and the setup check.

Contents

Tutorial and installation

The tutorial solves a first problem with bounds and nonlinear constraints. The installation page makes the SNOPT library available to Julia.

• Getting started
• • Solve an unconstrained problem
  • Read the solution
  • Add bounds and constraints
  • Use a sparse Jacobian
  • Next steps
• Installation
• • Requirements
  • Install the package
  • Make the library available
  • Check the setup

How-to guides

Each guide shows how to do one task with OptimizationSNOPT.jl.

• Set solver options
• • Common solve keywords
  • Optimizer fields
  • Other SNOPT options
  • Errors in options
• Monitor and stop a solve
• • Print progress
  • Store a trace
  • Stop on a custom condition
• Solve a problem several times

Solver behaviour and reference

The solver behaviour page explains return codes, SNOPT results and limits. The API reference lists every exported type.

• Solver behaviour
• • Return codes
  • SNOPT results in the solution
  • Derivatives
  • Maximization
  • Integer variables
  • One solve at a time
  • Limits
• API reference
• • Optimizer
  • Traces

Contributing

• Report bugs and request features on the issue tracker.
• Before opening a pull request, run the test suite with julia --project=. -e 'using Pkg; Pkg.test()' from the repository root. The solver tests run only when Julia finds a licensed libsnopt7.
• To build this manual, run julia --project=docs -e 'using Pkg; Pkg.instantiate()' and then julia --project=docs docs/make.jl from the repository root. The build runs the examples when it finds the SNOPT library; SNOPT_DOCTESTS=false skips them. The pages are written to docs/build/.