Ill
Table 4.1: Top section: Performance of reduced strong-weak model of the MIG fiber as
compared with the full model. Here kw = 10, N = 1001, and xτ is located 500 pm from
the distal end. Middle section: Same as the top section, but now xτ is located 400 pm from
the distal end. Bottom sect ion performance of the POD+DEIM model (here kv = k∣) on
the same fiber with the same inputs.
|
Reduced strong-weak model: |
xτ located 500 pm from distal end | |||
|
ks |
Speed-up |
% Matched |
% Mismatched |
Γ |
|
10 |
3.2× |
98.9 |
15.7 |
0.905 |
|
20 |
2.9× |
98.1 |
2.3 |
0.978 |
|
25 |
2.7× |
98.5 |
________3.4______ |
0.974 |
|
Reduced strong-weak model: |
xτ located 400 pm from distal end | |||
|
Speed-up |
% Matched |
% Mismatched |
Γ | |
|
10 |
3.3× |
98.9 |
22.4 |
0.863 |
|
20 |
3.0× |
97.7 |
3.4 |
0.970 |
|
25 |
2.7× |
98.5 |
2.3 |
0.980 |
POD+DEIM model
|
ky |
Speed-up |
% Matched |
% Mismatched |
Γ |
|
10 |
5.5× |
75.4 |
12.5 |
0.807 |
|
20 |
5.0× |
95.8 |
3.1 |
0.962 |
|
25 |
4.6× |
99.2 |
0.8 |
0.992 |
4.2 Generalizing the Strong-Weak Model to Arbitrary Mor-
phologies
The RSW model can be extended to handle arbitrary branched neurons, with two
main changes that must be considered. First, in partitioning the cell into strong and
weak components, there is a possibility that the branches of the individual compo-
nents will not be ordered to take advantage of the improvements of (Hines, 1984) that
permit efficient Gaussian Elimination. Hence we must reorder the branches locally
and reindex the compartments locally in order to achieve this result. Second, the