files: unoptimization files: *********************************************************************************************** The methods implicitly all assume that f is twice continuously differentiable. They may fail or converge very slowly if this is not the case. *********************************************************************************************** fmin.f Brent's direct search method, one variable. newuoa.zip Powell's new unconstrained optimization by quadratic approximation (f77) uncmnd.f quasi-Newton method by Schnabel et al (f77) WEDGE.tar.gz Interpolation/trust region method for moderate dimensions(Matlab) praxis.tgz f depends on few variables, the "principal axis method", tries to approximate the curvature of f mctoolbox.zip three direct search methods incl Nelder-Mead, Matlab nelmead.f f depends on few variables, Nelder-Mead simplex-search method (no sound theoretical basis), f77 subplex.tqz f depends on few variables, modification of the Nelder-Mead simplex-search method (no sound theoretical basis), (Matlab version) fminsi modified Nelder-Mead, LGPL-licensed, source, testdriver appspack-5.0.1.tar.gz Asynchronous and Fault Tolerant Parallel Pattern Search (LGLP, C++, serial or PVM or MPI) *********************************************************************************************** f is only Lipschitz continuous or less: n not too large *********************************************************************************************** gradsamp.tar Nonsmooth, nonconvex optimization by gradient sampling, by M. Overton et al; needs one of three QP solvers (Matlab) hanso1_0.zip Hybrid algorithm for nonsmooth, nonconvex optimization using quasi-Newton updating, bundling and gradient sampling (Matlab) OBOE-1.0.tar.bz2 Oracle-Based Optimization Engine for convex problems, uses Proximal-ACCPM interior point method, C++ nda.tar Four different routines for Lipschitz continuous functions and for (small) minimax problems OpenOpt0.36.tar.bz2 Matlab/Octave package, also general constraints lmbm.tar.gz limited-memory bundle method, large-scale, bound constrained version forthcoming *********************************************************************************************** The gradient of f is available or finite difference approximations are applicable : n not too large (= some hundred say) *********************************************************************************************** domin.tar.gz Spellucci's implementation of BFGS, different versions, exact/automatic/numerical gradients, f77/C 500.tgz the Shanno-Phua version of BFGS plus CG 611.tgz trust region BFGS, this is a reliable and fairly efficient code. 739.tgz tensor model quasi-Newton method, may be better for strongly nonquadratic f nmtr-1.2.tar.gz trust-region Newton, requires gradient fsplpak.tar.Z several methods, paper (Matlab) cg_descent-C-1.2.tar.gz A new conjugate gradient method with guaranteed descent (f77/C) cg_descent-F-1.2.tar.gz csminwel.tar.gz BFGS with some robustness against nonsmoothness (Matlab) *********************************************************************************************** f is convex differentiable, n large: *********************************************************************************************** ve08.tgz f is a sum of convex differentiable functions each of which has considerably less variables than n (partially separable problem)(also for bound constraints), special quasi Newton method *********************************************************************************************** n large, f general, but not too wildly behaved : *********************************************************************************************** 630.tgz A combined limited memory qN/cg method (degrades for nonconvex f, nevertheless applicable; for ill-conditioned problem an individual preconditioner must be provided) vmlm.tar.gz A limited memory quasi Newton method. Lbfgsb.2.1.tar.gz (f77) a limited memory quasi Newton method using a new matrix representation; LBFGS-20020202.java.jar directory contains software, drivers and a user's manual. toms778_f90.tar.gz Not to be recommended for ill-conditioned problems. lbfgs.cpp.zip Add your own preconditioner! lbfgs-1.1.zip needs C and Fortran compiler lbfgsb-for-matlab.tar.gz needs C++ and Fortran compiler preqn.1.0.tar.gz preconditioning package *********************************************************************************************** n large, f strongly nonquadratic : *********************************************************************************************** tn.tgz Nash's truncated Newton method based on the Lanczos method, without tnc-1.3.tgz (c) explicit eigenvalue computation. takes directions of negative curvature into account. 702.tgz truncated Newton based on the Lanczos process tron-1.2.tar.gz trust region Newton, preconditioned cg, also for bound constraints pl2.tar.gz truncated Newton method using the Lanczos process with direct computation of a truncated spectral decomposition, uses a so called three directions method from Heinrich, uses directions of negative curvature. for large dimensional problems. *********************************************************************************************** f noisy : *********************************************************************************************** cmaes_c.tar.gz Evolution Strategy with Covariance Matrix Adaptation cmaes_java.tar.gz snobfit_v2.1.tar.gz Stable Noisy Optimization by Branch and FIT (Matlab) MATLAB_IMFIL06.tar.gz implicit filtering subject to box constraints and with multiple minima (Matlab) DIRECTv204.tar.gz the DIRECT algorithm (in Matlab) praxis.tgz often applied to noisy problems although not directly designed for such newuoa.zip Powell's new unconstrained optimization by quadratic approximation (f77) subplex.tgz f depends on few variables, modification of the Nelder-Mead simplex-search method (no sound theoretical basis), (Matlab version) *********************************************************************************************** f expensive to evaluate : *********************************************************************************************** sscode.zip evolutionary method, see here, but claims to use considerably less evaluations, bound constraints (C) spacefiles.zip global optimization through meta-modeling: fitting, cross-validation, prediction, minimization etc (C++) spacemap.zip Optimization using Surrogate Models (Matlab) dace.zip Matlab toolbox to construct a kriging approximation (surrogate) to computer models