A.6 What do I do if I run out of memory?

Like all other electronic structure codes, the current bottleneck for FHI-aims when performing calculations for large systems is memory usage. A number of flags exist to alleviate memory usage, but as many of them have an associated trade-off of performance overhead, poor scaling for small system sizes, or breaking post-processing techniques, they have not been enabled by default. For large systems where memory usage becomes an issue, however, enabling them will be worth it. This list of flags may be found at in Section 3.47, “Large-scale, massively parallel: Memory use, sparsity, communication, etc.”

For hybrid functionals please make sure that you really need the basis set that you are requesting. For example, the tier 2 basis sets specified by tight and really_tight settings for light elements can be far too large. You may be able to get away with “intermediate” settings instead. Especially for hybrid functionals, large numbers of basis functions per atom really increase the time and memory consumption much more drastically than for semilocal DFT.

Please ask regarding intermediate settings in our forum – especially for hybrid functionals or for many-body perturbation theory. It’s worth it.