``Consider
everything.
Keep the
good.
Avoid evil
whenever you notice it.''
(1 Thess. 5:21-22)
This file is part of my global optimization web site. It contains sections on
See also the free Matlab toolbox
OPTI, for constructing and solving linear, nonlinear, continuous
and discrete optimization problems
(external free solvers apparently run under Windows only)
LP_SOLVE, Sparse Mixed Integer Linear Programming
in ANSI C (by Michel Berkelaar), a
Matlab Interface
(by Kantor and Chen; get lpmex.tar and lpmex_ext.tar), and an
Online Interface (by Dmitry Golovashkin)
LURUPA, Rigorous Error Bounds in Linear Programming in C++ (by Christian Keil)
Verification software in MATLAB / INTLAB, including linear and convex quadratic programming (by Jiri Rohn)
Rigorous lower and upper bounds in linear programming in C (by Ali Baharev)
GLPK, GNU Linear Programming Kit
ANSI C package for large-scale linear
and mixed integer linear programming,
LIPSOL (Matlab sparse LP by interior points), and the corresponding technical report (ps.gz, 129K)
Sparse Linear Programming
in Fortran77 (by Jacek Gondzio)
uses a higher order primal-dual, predictor-corrector logarithmic
barrier algorithm
PCx, sparse interior-point code for linear programming (Argonne National Laboratory and Northwestern University)
CLP, Simplex based linear program solver from COIN-OR
Linprog, low dimensional linear programming in C (Seidel's algorithm, by Mike Hohmeyer)
MCF Network Simplex Algorithm in C (by Andreas Loebel)
OsiDylp, dynamic linear programming
A Description of the MPS-format, the traditional input format for linear programs, and a program to translate MPS into simple LP format and conversely
MINQ - General Definite and Bound Constrained Indefinite
Quadratic Programming in Matlab
(by Arnold Neumaier)
based on subspace searches and an active set strategy
Newton-KKT Interior-Point Methods for Indefinite Quadratic Programming in Matlab
spsolqp.m interior point QP code in Matlab (by Yinyu Ye)
QPC - Quadratic Programming in C (with Matlab interface)
LOQO Quadratic Programming (sparse interior point code by Bob Vanderbei)
Large Scale Bound Constrained QP Solver in Fortran 77 (tar.gz, 23K; by Rolf Felkel) and in C (tar.gz, 55K)
Quadratic programming codes (links by Gould and Toint)
Quadratic programming form the decision tree page
(gradient based; if no gradients are available, see codes for derivative-free optimization)
CG_DESCENT, conjugate gradient method (in Fortran, by Hager and
Zhang); with a Matlab interface
As of April 2013, this is by far the best sequential first order
unconstrained minimization code publicly available. Works in very large
dimensions with memory of the order of the problem dimension, and
typically in significantly less than 1000 grasdient evaluations (though
sometimes it takes much longer, but competing codes are usually not
better on these problems).
ASA_CG, active set method for bound constrained optimization
(in Fortran, by Hager and Zhang) -
MATLAB interface for ASA_CG
This is the bound-constrained version of an older version of CG_DESCENT.
It is roughly comparable in performance with LBFGS-B on the problems
that the latter solves, but is much more robust and succeeds in many
cases where LBFGS-B stops with an error message.
L-BFGS-B, Limited Memory BFGS Algorithm (for large-scale optimization problems with simple bounds on the variables) - MATLAB interface for L-BFGS-B ( same MATLAB interface at mathworks)
TAO, Toolkit for Advanced Optimization
(for parallel optimization, by Argonne National Laboratory)
The current version of TAO has algorithms for unconstrained and
bound-constrained optimization.
PLIS, Limited-memory BFGS for bound constraints (in Fortran, by Ladislav Luksan)
Optizelle - Unconstrained, preconditioned nonlinear-CG, BFGS, Newton-CG, SR1, trust-region Newton, Barzilai-Borwein (in C++, with Matlab, Octave and Python interfaxce)
SPG, spectral projected gradient method for bound constrained
problems (in Fortran, by Birgin, Martinez, and Raydan)
Accepts more general convex constraints if a cheap projector
is available. There is also another bound-constrained solver GENCAN,
part of the ALGENCAN NLP solver at this site.
Poblano, Matlab toolbox of large-scale algorithms for unconstrained nonlinear optimization problems
TRON, Trust Region Newton Method for large bound-constrained optimization problems (in Fortran)
port/mn[fgh][ b].f for unconstrained and bound constrained optimization, using 0, 1, or 2 coded derivatives, drivers port/ex/nls[ab].f and user manual (ps.gz, 53K)
MINPACK-1 for solving nonlinear systems of
equations and nonlinear least squares problems
copy from netlib
MINPACK-2 for solving large-scale nonlinear systems of equations, nonlinear least squares problems, and unconstrained minimization
toms765, STENMIN for Large, Sparse Unconstrained Optimization Using Tensor Methods (in Fortran)
TNPACK, Truncated Newton Method (large-scale unconstrained minimization, in Fortran)
HIFOO, MATLAB package for multiobjective optimization
MProbe for analyzing the shape of multivariate functions and constrained domains
various Matlab optimization codes (by Tim Kelley)
OptSolve++, C++ library for solving nonlinear systems and unconstrained optimization
MERLIN, an integrated environment designed to solve nonlinear optimization problems with box constraints
(subgradient based; if no subgradients are available, see codes for derivative-free optimization)
HOPSPACK, Hybrid Optimization Parallel Search PACKage, for derivative-free optimization. Functions can be noisy, nonsmooth and nonconvex, linear and nonlinear constraints are supported, and variables may be continuous or integer-valued. The C++ framework enables parallel operation using MPI (for distributed machine architectures) and multithreading (for single machines with multiple processors or cores).
SolvOpt, Shor's minimization method with space dilation for unconstrained and constrained nonsmooth problems (in Fortran90, C and Matlab, by Kuntsevich and Kappel)
OOSuite, containg python code for optimization, among others ralg, a constrained NLP solver for nonsmooth problems, with or without explicit (sub)gradients (in Python, by Dmitrey Kroshko)
LMBM, Limited Memory Bundle Method, for nonconvex, nonsmooth
minimization of bound constrained functions of many variables
(by N. Karmitsa = Haarala)
At present, only the unconstrained version is available.
CMA-ES, a stochastic solver for bound-constrained problems
Subroutines for nondifferentiable optimization with linear constraints (in Fortran, by Ladislav Luksan)
(derivative based; if no derivatives are available, see codes for derivative-free optimization)
Newton Methods for Nonlinear Problems (in Fortran or Matlab)
Newton-like methods for nonlinear systems, least squares problems
and parameter-dependent systems (by Peter Deuflhard)
nleq1.m for square systems
nlscon.m for overdetermined systems (nonlinear least squares)
Tensorlab (in matlab)
contains (maong others) algorithms for nonlinear least squares problems
whose variables are complex numbers, with matrix- or tensor-valued
constraints, including complex automatic differentiation
Google Ceres Solver, Nonlinear Least Squares
Nonlinear Least Squares Problems in Matlab (by Hans Nielsen)
MINPACK for solving nonlinear systems of equations and nonlinear least squares problems
PETSc, Portable, Extensible Toolkit for Scientific Computation with tools for parallel sparse nonlinear systems
CMPSm, Continuation Method for Polynomial Systems
(in Matlab)
for finding all isolated solutions of a system of polynomial equations
MANPAK homotopy package for computations on implicitly defined curves and manifolds (by W.C. Rheinboldt)
CONMAX constrained minimax approximation (234K, by Jerry Taylor)
MATLAB Code for Moré/Garbow/Hillstrom test problems
(gradient based or white box; if no gradients are available, see codes for derivative-free optimization)
CVX, Matlab Software for Disciplined Convex Programming in Matlab, (by Grant, Boyd and Ye)
solnp.m, interior point NLP program (by Yinyu Ye)
Optizelle, unconstrained and constrained optimization, including second-order cone and semidefinite constraints (in C++, with Matlab, Octave and Python interface)
DONLP2, Dense Sequential Quadratic Programming in Fortran77 (by Peter Spellucci) and a Fortran 90 version (by Alan Miller)
IPOPT, Interior Point Optimizer from COIN-OR for nonlinear, nonconvex, constrained optimization problems (in C++; there is also an older Fortran version)
The commercial NLP solver KNITRO now offers a free evaluation version with limited capabilities
OPT++ nonlinear optimization algorithms for expensive problems, written in C++
GALAHAD, library of Fortran 90 packages for solving nonlinear optimization problems. ''At present, the areas covered by the library are unconstrained and bound-constrained optimization, quadratic programming, nonlinear programming, systems of nonlinear equations and inequalities, and nonlinear least squares problems.''
LOQO Sequential Quadratic Programming (sparse interior point code by Bob Vanderbei)
HQP, Huge Sequential Quadratic Programming for large-scale nonlinear optimization problems with (block-)banded sparsity structure in SIF or Omuses (Rüdiger Franke)
FSQP, Feasible Sequential Quadratic Programming in C and Fortran (for large scale problems)
PSQP, SQP variable metric methods for general nonlinear programming problems (in Fortran, by Ladislav Luksan)
Logilab Software for large-scale linear and convex programming
SOCP, Second-Order Cone Programming in C with Matlab interface
(Lobo, Vandenberghe and Boyd)
minimize f^T x
subject to || A_i x + b_i ||_2 <= c_i^T x + d_i , i=1,...,L
Quadratically Constrained Quadratic Programming (by Henry Wolkowicz)
TOMLAB, a commercial, Matlab based optimization environment for dense nonlinear programming.
NLP++ Toolbox, a commercial C++ class library providing optimization routines for a large variety of mixed-integer nonlinear constrained optimization problems (and simpler variants)
FSQP, Feasible Sequential Quadratic Programming
for constrained optimization and minimax problems (in Fortran77 and C)
Free for academic users
(see also the codes for derivative-free optimization)
SNOBFIT, MATLAB 6 package for the robust and fast solution of
noisy, expensive optimization problems with continuous variables
varying within bound, possibly subject to additional soft constraints
very efficient for up to about 15 variables, but work grows like
O(n^{6}).
HOPSPACK, Hybrid Optimization Parallel Search PACKage, for derivative-free optimization. Functions can be noisy, nonsmooth and nonconvex, linear and nonlinear constraints are supported, and variables may be continuous or integer-valued. The C++ framework enables parallel operation using MPI (for distributed machine architectures) and multithreading (for single machines with multiple processors or cores).
CONDOR in C++ (with Matlab interface)
solves constrained nonlinear programs with noisy, expensive
objective functions (even multiple ones) and cheap constraints
DFO COIN-OR program for bound constrained noisy optimization
download
CMA-ES, a stochastic solver for bound-constrained problems
Asynchronous Parallel Pattern Search (for expensive, noisy objective and linear constraints)
Space Mapping Toolbox for unconstrained optimization with a coarse and a fine model
CLOP Confident Local Optimization for Noisy Black-Box Parameter Tuning (minimizing expected function value, including noisy (0,1)-valued functions)
(see also the codes for noisy smooth optimization)
LINCOA, LINearly Constrained Optimization Algorithm
for up to a few hundred variables
(in Fortran 77, by Mike Powell)
SNOBFIT, MATLAB 6 package for the robust and fast solution of
expensive optimization problems with continuous variables
varying within bound, possibly subject to additional soft constraints.
very efficient for up to about 15 variables, but work grows like
O(n^{6}).
HOPSPACK, Hybrid Optimization Parallel Search PACKage
for derivative-free optimization. Functions can be noisy, nonsmooth
and nonconvex, linear and nonlinear constraints are supported,
and variables may be continuous or integer-valued. The C++ framework
enables parallel operation using MPI (for distributed machine
architectures) and multithreading (for single machines with multiple
processors or cores).
APPSPACK, derivative-free optimization software for solving nonlinear unconstrained, bound-constrained, and linearly-constrained optimization problems, with possibly noisy and expensive objective functions. (superseded by HOPSPACK, according to the developers)
BFO Brute Force Optimizer
trainable derivative-free solver for mixed integer bound-constrained
optimization (in Matlab, by Porcelli and Toint)
NOMAD, derivative-free code for nonlinear programming with expensive functions, allowing for categorical constraints (in MATLAB)
DFL, Derivative-Free Constrained Optimization (in Fortran 90, some in C, by di Pillo et al.), solves mixed-integer, global and local optimization problems, and minimax problems, without using any derivative information
NEWUOA, unconstrained minimization without derivatives for up to 200 variables (in Fortran 77,, by Mike Powell)
BOBYQA, Bound constrained Optimization BY Quadratic Approximation
for high-dimensional smooth functions without explicit deriviatives
(in Fortran 77, by Mike Powell)
NLopt algorithms for derivative-free optimization
Nelder & Mead simplex algorithm with quadratic model fit in Fortran90 (by Alan Miller)
implicit filtering for derivative-free optimization in matlab
OptQuest Solver Engine for Microsoft Excel (commercial)
VSDP, Verified SemiDefinite Programming, mathematically rigorous bounds on the solution of semidefinite programs (in Matlab, by Christian Jannson)
YALMIP, MATLAB toolbox for rapid prototyping of optimization problems (second order cone, non-convex semidefinite programming, mixed integer, multi-parametric and geometric programming)
SOCP, Second-Order Cone Programming
(in C, with Matlab interface, by Lobo, Vandenberghe and Boyd)
minimizes f^T x subject to one or more constraints of the form
||Ax+b||_2 <= c^T x + d
SeDuMi, Matlab toolbox (Jos Sturm)
solves sparse optimization problems over symmetric cones
(linear, convex-quadratic and positive semi-definite constraints;
complex valued entries are allowed)
SDPSOL solver for semidefinite programming
in C, with Matlab interface (Wu and Boyd)
minimize c^Tx - log det G(x)
subject to G(x) pdef, F(x) psd,
where G(x) and F(x) are linear square matrix expressions.
``The current implementation only exploits block-diagonal structure,
hence is not particularly efficient. But it is very easy to use, and
does work quite nicely on small and medium size problems
(say, a hundred or so variables). ''
SDPA, SemiDefinite Programming Algorithm in C++ utilizing the MESCHACH library for dense and sparse numerical linear algebra
SDPT3, Semidefinite Programming in Matlab (Toh Kim-Chuan)
CSDP library for sparse semidefinite programming in C (Brian Borchers)
SP, Software for Semidefinite Programming
in C, with Matlab interface (Vandenberghe and Boyd)
(
Scilab interface)
SDPpack in Matlab for mixed semidefinite-quadratic-linear programs (Alizadeh, Haeberly, Nayakkankuppam and Overton)
SDPHA homogeneous interior-point algorithm for semidefinite programming (Potra, Sheng and Brixius)
CONMAX continuation / sequential linear programming method (in Fortran 77, 240K; by Kaufman, Leeming and Taylor)
Semidefinite Programming (in Matlab, tar.gz; 191K)
Lmitool interface to solve linear matrix inequality problems
DSDP for solving positive semidefinite programs (Benson and Ye)
Cardinality and Weighted Matching based on algorithms of Gabow and Micali/Vazirani
Linear and Bottleneck Assignment in C++, Pascal, Fortran based on algorithms of Jonker and Volgenant
DFL, Derivative-Free Constrained Optimization (in Fortran 90, some in C, by di Pillo et al.), solves mixed-integer, global and local optimization problems, and minimax problems, without using any derivative information
MILES, Matlab package for solving Mixed Integer LEast Squares
problems
for linear problems with bound and integer constraints
CONMAX constrained minimax approximation (234K, by Jerry Taylor)
SIPAMPL, Software for Semi-infinite Programming Problems
JBendge, Bayesian Estimation of Nonlinear Dynamic General Equilibrium
(DGE) Models
solves infinite-time horizon stochastic control problems and
associated parameter estimation problems
(for continuous states and actions)
Murphy Markov Decision Process (MDP) Toolbox for Matlab (for discrete states and actions)
INRA Markov Decision Process (MDP) Toolbox for MATLAB (for discrete states and actions)
RIOTS, Matlab Toolbox for Solving Optimal Control Problems (free demo version only)
Nested Fixed Point Maximum Likelihood Algorithm (for parameter estimation in discrete control processes)
Reinforcement Learning Software and Stuff
MOSQP, MultiObjective Sequential Quadratic Programming
NEOS Guide to Optimization Software
(high quality software, including commercial packages)
In the section Constrained Optimization,
the above site has information about the following packages:
netlib/opt/index (European mirror site)
ELIB list of mathematical programming software
SAL (Scientific Applications on Linux) Optimization Software List
User-Contributed MATLAB Optimization Software (most routines not of high quality)
Transactions on Mathematical Software (TOMS) and TOMS at Netlib
Nonlinear Programming Packages
Interval Methods
Regularization
Protein Folding
Recent Papers and Preprints
my home page
(http://www.mat.univie.ac.at/~neum)
Arnold Neumaier (Arnold.Neumaier@univie.ac.at)