Guidelines for Choosing Between ACA and MLFMM IE Solver Options
The following guidelines have been arrived at by analyzing the features of the MLFMM algorithm and comparing its performance to ACA for several test problems.
- Problem size
In general, the ACA solver will be more efficient than the MLFMM solver in terms of both memory and solution time for problems whose scattering surface(s) has (have) less than 100,000 surface unknowns (~100,000 triangles).
- Aspect Ratio (largest dimension : smallest dimension):
In general, the ACA solver will be more efficient in memory and speed than MLFMM for problems whose geometry has a bounding box with a highly skewed aspect ratio (on the order of 100:1 or higher).
- Design Type
HFSS-IE
Number of Excitations:
For problems having very few excitations (<10), the MLFMM solver will be more efficient than ACA in both speed and memory for problems that satisfy items 1 and 2 above. Since the time required for matrix multiplication is higher in MLFMM, as the number of excitations increases (e.g., mono-static RCS problems), the # of unknowns in the problem needed to realize the increase in speed provided by MLFMM (vs. ACA) will increase (with respect to the limits provided in 1, above). However, the memory savings provided by MLFMM will remain consistent even while the number of excitations increases.
HFSS - FE-BI
For FE-BI surfaces involving few excitations the MLFMM solver will provide similar or slightly better performances in solution time and better performance in memory for problems that satisfy items 1 and 2 above.
For FE-BI surfaces involving a large number of excitations (e.g., several ports, mono-static RCS problems), the MLFMM solver's solution time may be larger than that for ACA but the memory savings will remain.
IE Regions
For IE region problems the ACA solver will nearly always be faster than the MLFMM solver. However, the MLFMM solver will provide memory savings for problems satisfying criteria 1 & 2 above.
Decision Flowchart for Deciding on the IE Solver Options
