84
Table 3.3: Performance of reduced model (here kv = feʃ) of HH fork, N = 1501, as
compared with the full model. The first simulation set does not use Branch-Ortho for the
reduced system, while the next four sets use it, leading to far superior accuracy.
|
Speed-up |
% Matched |
% Mismatched |
Γ | |
|
30, no Branch-Ortho |
17.5× |
349 |
Ï9?Ô |
0.484 |
|
10 |
24.1× |
63.4 |
18.7 |
0.707 |
|
15 |
22.5× |
91.5 |
4.5 |
0.933 |
|
20 |
21.2x |
94.3 |
1.5 |
0.963 |
|
30 |
17.5× |
99.6 |
0 |
0.998 |
Table 3.4: Performance of reduced model (here kυ = kf~) of HHA fork, N == 1501, as
compared with the full model.
|
kv |
Speed-up |
% Matched |
% Mismatched |
Γ |
|
20 |
23.5× |
48.27 |
41.42 |
0.515 |
|
25 |
22.0× |
85.87 |
0 |
0.924 |
|
30 |
19.7× |
98.9 |
2.6 |
0.981 |
|
40 |
15.4× |
100 |
0 |
1 |
kept the same, except in this case, we use εj,oca∣ = 10^4 and ε1ζcal = 10-51 and we
use 1000 random stimuli of between 0-250 pA. Note that the tolerances here are
not tuned to give the best performance, but rather have been chosen after just a
little experimentation to give good performance; it is possible that a better choice
may exist. Even though the spatially-varying A-type K+ conductance necessitates a
slightly larger reduced system in order maintain the accuracy seen in the HH fork,
we still observe a similar speed-up, as Table 3.4 indicates.