# What is your software optimization solver of preference?

 6 5 Please post an answer that is your favorite solver. If your favorite solver is already listed vote it up. Examples include CPLEX, Gurobi, GLPK, Symphony, lpsolve, Excel Solver, etc... asked 15 Apr '10, 12:55 larrydag 1 ♦ 3.1k●5●10●26 accept rate: 9% You might want to specify type of problem (LP/MILP, NLP, nonconvex global optimum, ...)? (15 Apr '10, 13:14) Paul Rubin ♦ Agreed with Paul that the question is vague. This could become a popularity contest, dominated by vocal supporters of one solver or another (ex: employees at one vendor, or people who have contributed to the open source solver). As a longtime employee in the solver industry, I'm refraining from posting my employer (Gurobi) just to win votes. (31 Jan '11, 18:53) Greg Glockner

 8 CPLEX (which is less than 15 characters, so I'm padding it) answered 15 Apr '10, 13:13 Paul Rubin ♦ 11.9k●4●12 accept rate: 19% 1 Good idea. We might want to give a brief description of the solver. Proprietary/Open Source, Maintainer, Company, Years of Existence and what not. (15 Apr '10, 13:22) larrydag 1 ♦ Makes sense. CPLEX is commercial (but there's a free student version limited to 500 variables / 500 constraints, and the full version is free to academics who register). It's currently owned by IBM (who bought ILOG) (who bought CPLEX Inc.). Years of existence = 20 (25?) plus. (17 Apr '10, 20:08) Paul Rubin ♦
 6 GLPK Fast, easy free and open source(GPL) linear,mixed integer, branch & cut modeling system http://www.gnu.org/software/glpk/ From the site... The GLPK (GNU Linear Programming Kit) package is intended for solving large-scale linear programming (LP), mixed integer programming (MIP), and other related problems. It is a set of routines written in ANSI C and organized in the form of a callable library. GLPK supports the GNU MathProg modeling language, which is a subset of the AMPL language. The GLPK package includes the following main components: * primal and dual simplex methods * primal-dual interior-point method * branch-and-cut method * translator for GNU MathProg * application program interface (API) * stand-alone LP/MIP solver  answered 16 Apr '10, 13:56 larrydag 1 ♦ 3.1k●5●10●26 accept rate: 9%
 5 Used to work with CPLEX a lot now I use Gurobi instead (usually with AMPL). Trying to switch to GLPK but still no luck in converting my AMPL models. For nonlinear models depending on the level of complexity I either use Matlab or MINOS. For portfolio optimization I use either R or Cplex (it is easy to model cone programming in Cplex) answered 15 Apr '10, 18:53 Mark ♦ 3.6k●2●20●48 accept rate: 10%
 4 I've also had good luck with XPRESS-MP (commercial, current owner = FICO, been around a while). answered 17 Apr '10, 20:09 Paul Rubin ♦ 11.9k●4●12 accept rate: 19%
 4 gurobi works really well on big problems. answered 06 Nov '10, 18:53 Jerry Shaw 131●2 accept rate: 0% I'd vote for Gurobi -- it's really fast on my MILPs. The latest version (4.0) just added support for QPs and MIQPs. (07 Nov '10, 18:04) Gilead ♦
 4 is it really true that nobody posted SCIP here? answered 11 Jan '12, 13:11 Marco Luebbecke ♦ 3.2k●1●5●15 accept rate: 15%
 2 I have worked with lpsolve and Mathematica for linear programming, and like both. lpsolve seems to have a more open community, however. answered 15 Apr '10, 13:37 Karsten W. 203●1●1●8 accept rate: 0%
 2 I've found SAS/OR to be very powerful and flexible. Here is a link to the documentation page about the math programming modelling language. http://support.sas.com/documentation/cdl/en/ormpug/59679/HTML/default/optmodel.htm answered 19 Apr '10, 11:13 DC Woods ♦ 4.0k●2●17●43 accept rate: 5%
 2 For large-scale Nonlinear Programs (NLPs), I would have to vote IPOPT (free). I have used almost every single commercial NLP solver available, and IPOPT just consistently outperforms all of them on the kinds of problems I'm solving. http://www.coin-or.org/Ipopt/ Pros: Free Modular support for different sparse-structured linear algebra packages (very important for really large scale problems) Generally fast convergence, and converges even with somewhat poor initial guesses. Ill-conditioned problems are also handled fairly well. Exploits 2nd-order and sparsity information. Has a quasi-newton mode, but automatic differentiation works much better. Filter line search. Faster and more robust than most other NLP solvers for a wider class of problems (see Dolan-More benchmark diagrams on test problems). Works with various modeling languages like AMPL, GAMS; and also with MATLAB, C++, Python (limited), etc. Code is fairly mature, and many organizations are adopting it in-house or even bundling it with their software. Cons: Being an interior-point method, the warm-starting is temperamental -- IPMs don't warm-start well. Does not work well when the no. of constraints exceeds the no. of variables, even if some of the constraints are redundant. GRG methods tend to work better in such situations. The dominating performance bottleneck is the linear algebra, and the default free linear solver MUMPS isn't the fastest out there. There are other high performance linear solvers like PARDISO and MA57 which are free for academic use, but have to be licensed for commercial use. answered 07 Nov '10, 17:56 Gilead ♦ 2.2k●3●11 accept rate: 15%
 toggle preview community wiki

By Email:

Markdown Basics

• *italic* or _italic_
• **bold** or __bold__
• image?![alt text](/path/img.jpg "Title")
• numbered list: 1. Foo 2. Bar
• to add a line break simply add two spaces to where you would like the new line to be.
• basic HTML tags are also supported

Tags:

×153
×42
×39