Development roadmap

The JuMP developers have compiled this roadmap document to share their plans and goals with the JuMP community. Contributions to roadmap issues are especially invited.

Most of these issues will require changes to both JuMP and MathOptInterface, and are non-trivial in their implementation. They are in no particular order, but represent broad themes that we see as areas in which JuMP could be improved.

• Support nonlinear expressions with vector-valued inputs and outputs. There are a few related components:
  • Representing terms like log(det(X)) as necessary for Convex.jl
  • Automatic differentiation of terms with vector inputs and outputs
  • User-defined functions with vector–as opposed to scalar–inputs, which is particularly useful for optimal control problems
  • User-defined functions with vector outputs, avoiding the need for User-defined operators with vector outputs
• Add support for modeling with SI units. The UnitJuMP.jl extension is a good proof of concept for what this would look like. We want to make units a first-class concept in JuMP. See #1350 for more details.

Completed

• Done #3106 Make nonlinear programming a first-class citizen. There have been many issues and discussions about this: currently nonlinear constraints are handled through a MOI.NLPBlock and have various limitations and restrictions.
  • https://github.com/jump-dev/JuMP.jl/issues/1185
  • https://github.com/jump-dev/JuMP.jl/issues/1198
  • https://github.com/jump-dev/JuMP.jl/issues/2788
  • https://github.com/jump-dev/MathOptInterface.jl/issues/846
  • https://github.com/jump-dev/MathOptInterface.jl/issues/1397
• Done #3385 Add support for coefficient types other than Float64: https://github.com/jump-dev/JuMP.jl/issues/2025 Since the very beginning, JuMP has hard-coded the coefficient type as Float64. This has made it impossible to support solvers which can use other types such as BigFloat or Rational{BigInt}.
• Done #3385 Add support for constraint programming: https://github.com/jump-dev/JuMP.jl/issues/2227 JuMP has a strong focus on linear, conic and nonlinear optimization problems. We want to add better support for constraint programming.
• Done #3176 Add support for multiobjective problems: https://github.com/jump-dev/JuMP.jl/issues/2099 JuMP is restricted to problems with scalar-valued objectives. We want to extend this to vector-valued problems.
• Done #3629 Refactor the internal code of JuMP's macros. The code in src/macros.jl is some of the oldest part of JuMP and is difficult to read, modify, and extend. We should overhaul the internals of JuMP's macros–-without making user-visible breaking changes–-to improve their long-term maintainability.