Installing BigDFT on a 64-bit AMD Opteron Cluster running CentOS 5 Linux

BigDFT is a Density Functional Theory (DFT) parallel electronic structure code using a wavelet basis set. Wavelets form a real space basis set distributed on an adaptive mesh. Surfaces and isolated systems can be simulated with the proper boundary conditions and GTH or HGH pseudopotentials are used to remove the core electrons.

To install, download the source tarball to a sensible location (e.g., /usr/local/src/BIGDFT) and extract (tar xvf bigdft-1.4.0.tar.gz). The installation process is a fairly typical automake process, but with some minor differences. Apart from a usual requirement to differentiate from source and installation directories, this particular installation is designed to operate with CUDA GPUs and without MPI. Thus the configure script takes the following form:

./configure --prefix=/usr/local/bigdft/1.4.0-cuda --with-blas --with-lapack --with-cuda-path=/usr/local/cuda/3.1/cuda --enable-cuda-gpu --disable-mpi

After this is complete the installation with make, make check, and make install. If you are not installing on a GPU machine (but are using the code on a GPU machine, e.g., a cluster with a GPU node) the make check will fail. It is still worth running this as far as it will can to ensure system sanity and Cntrl-C'ing out of it.

Create a module for environment variables the usual way; cd /usr/local/src/Modules/modulefiles/bigdft, ln -s .base 1.4.0-cuda) and modify .version for the default version. Note that the paths in .base are slightly different due to the -cuda suffix.