Optimize calculation of reflected Green's tensor

[GoogleCodeExporter]

Currently, somnec.c is used to compute a table of Sommerfeld integrals. There 
are two problems.

1) It is used as black box, without sufficient understanding of its accuracy. 
For some applications, the accuracy is probably not sufficient, for others - 
some of it can be sacrificed to get computational speed. This should be studied 
in more details (see comments in somnec.c). 

2) It takes considerable time, currently approximately equivalent to 200 
iterations. This can be prohibitive, e.g. for biological applications (large 
grid, but small number of iterations). So studying other alternatives 
(described in the beginning of issue 101) is still very relevant. However, the 
issue is not easy (preliminary work showed that it is not that easy to make a 
general "black box" routine with at least any automatic convergence control).

Alternative approach is to use 2D interpolating table to calculate Sommerfeld 
integrals - but that adds another free parameter, effecting the final accuracy.

Also, there is a connected (more specific) issue 175.

#101, #175

Original issue reported on code.google.com by yurkin on 25 Sep 2013 at 8:28

[myurkin]

#175

[myurkin]

Concerning the optimization and accuracy, there are a lot of ideas in Schmehl's thesis (1994) - https://www.researchgate.net/publication/260426740_The_Coupled-Dipole_Method_for_Light_Scattering_from_Particles_on_Plane_Surfaces . It includes building a table and doing 2D interpolation on it.

The major question is what is the proper way to control accuracy. Constant relative error (as it seems to be now) may be not appropriate, since at large distances it is difficult to obtain and not really needed (since the values are then added to a much larger image-dipole term).

Probably, the best way would be to make an accuracy parameters (currently defines MAXH and CRIT in somnec.c) into function arguments to be adjusted at each call. Then a smaller CRIT can be used at larger distances.

This may help with problems as in #215