Maxim Yurkin

This [thread](https://devtalk.nvidia.com/default/topic/992502/why-is-cl_device_max_mem_alloc_size-never-larger-than-25-of-cl_device_global_mem_size-only-on-nvidia-/) suggests that Nvidia developers (and others sometimes also) just take the simplest route and specify CL_DEVICE_MAX_MEM_ALLOC_SIZE to be the _minimum_ possible value mandated by the standard. And modern...

A recent relevant paper is http://dx.doi.org/10.1016/j.jqsrt.2015.06.024, which describes a FFT-based approach.

Some approach from the family of [non-equidistant FFTs](https://www-user.tu-chemnitz.de/~potts/nfft/) may potentially help here.

In particular, Intel BLAS should be relevant for #61

I agree. The point of this issue is to actually try to do it and make the switch convenient, e.g. a simple option to make. Anyway, the main work is...

In terms of unstable behavior, trivial renormalization in file `propaesplibreintadda.f` can help. Now it computes very small value and then divides by the dipole volume.

#299 constitutes first working implementation, which should be sufficient for most purposes in ADDA. However, we still want to improve it further, including making it more robust for arbitrary distance...

Related to this is the discussion of optimal threshold for a specific problem. Now we use 1e-5 to be safe, but many tests show that 1e-3 or even 1e-2 can...

Based on the discussion with @alkichigin, we may want to set the default threshold in ADDA to 1e-4 or 1e-3 to save some computational time for all simulations, but that...

Should also be made compatible with rectangular dipoles - see #245