FAQ

With HALMD, I could do simulations of breath-taking quality and have obtained new scientific insight. How may I thank you?

Please acknowledge the use of HALMD in your publications by citing our article:

P. H. Colberg and F. Höfling, Highly accelerated simulations of glassy dynamics using GPUs: Caveats on limited floating-point precision, Comp. Phys. Comm. 182, 1120 (2011) [Link].
Why does HALMD abort with “[ERROR] overcrowded placeholders in cell/neighbour lists update”?

The GPU backends of HALMD employ fixed-size cell and neighbour lists to accommodate for memory access pattern constraints. By default, a cell or neighbour list is allocated twice as many placeholders as the average expected number of particles. This may not suffice in case of large local density fluctuations, e.g. for unequilibrated systems.

Try lowering the --cell-occupancy value.

Why does HALMD abort with “[ERROR] potential energy diverged”?

An infinite potential energy sum of one or more particles indicates that the integration time-step is too large.

Try lowering the --timestep value.

nvcc fails with ‘cudafe++’ died due to signal 11 (Invalid memory reference)
This is due to a bug in the CUDA compiler, which may be circumvented by including --host-compilation=c in the NVCCFLAGS environment variable passed to cmake, or in CMAKE_CUDA_FLAGS using ccmake.
nvcc fails with error: inline function ‘__signbit’ cannot be declared weak
CUDA 2.3 (or less) is not compatible with GCC 4.4. As a work around install GCC 4.3 and place symlinks in the default CUDA compiler directory, e.g. if the CUDA toolkit is located in /opt/cuda, symlink /opt/cuda/bin/{gcc,g++} to /usr/bin/{gcc,g++}-4.3, respectively. The compiler directory may be overriden with the --compiler-bindir option.

Previous topic

Usage

Next topic

Data files