| Stable release | 1.10.0 |
|---|---|
| Repository | github |
| Written in | C++ |
| Type | Optimization solver suite |
| License | MIT |
| Website | ergo-code |
 | |
| Headquarters | Edinburgh |
|---|---|
| Location |
|
Director | Julian Hall |
Key people |
|
| Staff | 5 |
| Website | www |
HiGHS is open-source software to solvelinear programming (LP),mixed-integer programming (MIP), and convexquadratic programming (QP) models.[1]
Written inC++ and published under anMIT license, HiGHS provides programming interfaces toC,Python,Julia,Rust,R,JavaScript,Fortran, andC#. It has no external dependencies. A convenient thin wrapper to Python is available via thehighspyPyPI package. HiGHS is also callable vianuget.
Although generally single-threaded, some solver components can utilize multi-core architectures and, fromVersion 1.10.0, can run its first order LP solver on NVIDIA GPUs. HiGHS is designed to solve large-scale models and exploitsproblem sparsity. Its performance relative to commercial and other open-source software is reviewed periodically using industry-standardbenchmarks.[2]
The termHiGHS may also refer to both the underlying project and the small team leading the software development.
HiGHS is based on solvers written by PhD students from the Optimization and Operational Research Group [3] in the School of Mathematics at theUniversity of Edinburgh. Its origins can be traced back to late 2016, when Ivet Galabova combined her LP presolve with Julian Hall's simplex crash procedure and Huangfu Qi's dual simplex solver to solve a class of industrial LP problems faster than the best open-source solvers at that time. Since then, a C++ API and other language interfaces have been developed, and modelling utilities and other categories of solver have been added.
In early‑2022, theGenX andPyPSA open energy system modelling projects endorsed a funding application for the HiGHS solver in an effort to reduce theircommunity reliance on proprietary libraries.[4] That appeal resulted inCAN$76000 in funding from Invenia Labs, Cambridge, United Kingdom in July 2022.[5]
HiGHS has implementations of the primal and dualrevised simplex method for solving LP problems, based on techniques described by Hall and McKinnon (2005),[6] and Huangfu and Hall (2015, 2018).[7][8] These include the exploitation of hyper-sparsity when solving linear systems in the simplex implementations and, for the dual simplex solver, exploitation of multi-threading. The simplex solver's performance relative to commercial and other open-source software is regularly reported using industry-standard benchmarks.[9]
HiGHS has aninterior point method implementation for solving LP problems, based on techniques described by Schork and Gondzio (2020).[10] It is notable for solving the Newton system iteratively by apreconditioned conjugate gradient method, rather than directly, via anLDL* decomposition. The interior point solver's performance relative to commercial and other open-source software is regularly reported using industry-standard benchmarks.[11]
HiGHS has a branch-and-cut solver for MIP problems. Its performance relative to commercial and other open-source software is regularly reported using industry-standard benchmarks.[12]
HiGHS has an active set solver for convexquadratic programming (QP) problems.
HiGHS can be used as a stand‑alone solver library in bespoke applications, but numerical computing environments, optimization programming packages, and domain‑specific numerical analysis projects are starting to incorporate the software into their systems also.
As powerful open‑source software under active development, HiGHS is increasingly being adopted byapplication software projects that provide support fornumerical analysis. TheSciPy scientific library, for instance, uses HiGHS as its LP solver [13] from release 1.6.0 [14] and the HiGHS MIP solver fordiscrete optimization from release 1.9.0.[15] As well as offering an interface to HiGHS, theJuMP modelling language forJulia [16] also describes the specific use of HiGHS in its user documentation.[17] The MIP solver in theNAG library is based on HiGHS ,[18] and HiGHS is the default LP and MIP solver in the MathWorks Optimization Toolbox .[19]
HiGHS is now also used by some domain‑specific applications, including oneopen energy system modeling environment. The web‑based version of thePyPSA European multi‑sector model deploys the HiGHS solver by default from February 2022.[20][21] The GridCal project developing research‑oriented power systems software added optional support for HiGHS in February 2022.[22]
