Nonsmooth Unconstrained Optimization via Limited Memory Bundle Method
-
The solver is a C++11 wrapper for the FORTRAN solver LMBM, which suits Large-Scale Nonsmooth Unconstrained Nonlinear Optimization.
-
It is a substitution to L-BFGS when nonsmooth optimization is inevitable. For example, nonsmoothness is quite common in various robotics applications where spatially discretized data is often incorporated into optimization problem.
To build the lib, first make sure you have gfortran, g++, make, and CMake installed. This toolchain is often installed by default in many Linux environments. Then, clone this repository and rename it LMBM. Now cd into this directory to build the shared lib for LMBM solver.
cd LMBM/lmbm/
make
Then a shared lib for LMBM is generated. Two files lmbm.so and lmbm.h are all you need for conducting nonsmooth optimization in any C++ application. The header file contains sufficient information about how the lib should be called and tuned.
A simple example is also provided where a nonsmooth function is minimized by LMBM. The CMakeList.txt shows how the lib should be linked by your own projects. Run the following commands to build and execute the example.
cd ..
mkdir build
cd build
cmake ..
make
./lmbm_example
- Haarala, N., Miettinen, K. and Mäkelä, M.M., 2007. Globally convergent limited memory bundle method for large-scale nonsmooth optimization. Mathematical Programming, 109(1), pp.181-205.
- Karmitsa, N., 2007. LMBM–FORTRAN subroutines for Large-Scale nonsmooth minimization: User’s manual’. TUCS Technical Report, (856).
- Bagirov, A., Karmitsa, N. and Mäkelä, M.M., 2014. Introduction to Nonsmooth Optimization: Theory, Practice and Software. Springer.