mpLBFGS

mpLBFGS is a header-only C++ library that implements the Polak-Ribière conjugate gradient descent and the limited-memory BFGS algorithm (L-BFGS) for unconstrained optimization. The library uses multiple precision scalars instead of the default double machine precision. This allows to obtain much more accurate approximations of the minimizers, effectively to arbitrary precision.

Build instructions and dependencies

If you have the Eigen and mpfr libraries installed on your system, make builds the wtpframe target by default. Alternatively, to download the dependencies first and then build wtpframe, issue

make all

To build with cmake:

mkdir build && cd build && cmake ..

The library requires Eigen for matrix computations and mpfr for multiple precision floating point scalars, respectively. You will need gmp and mpfr installed on your system in order to use the multiple precision scalars. The optimizers are templatized in the scalar type, so modifying that should be straightforward.

Credits

The code is derived from libLBFGS and LBFGS++ libraries; the original algorithm and implementation goes back to Jorge Nocedal and his collaborators. The multiple precision numbers are provided by the GNU MPFR, wrapped in mpreal class by Pavel Holoborodko.

mpLBFGS is licensed under the GPL license.