possible numerical issue about ecos

[adomahidi]

This issue seems to be solved with the upcoming version v1.1.0

[adomahidi]

@jaehyunp If I run this with the new version I get this:

It pcost dcost gap pres dres k/t mu step IR
0 +3.190e+03 -6.116e+04 +1e+05 2e-01 4e+00 1e+00 2e+01 N/A 1 1 -
1 +3.275e+03 -7.470e+03 +2e+04 4e-02 7e-01 7e-01 5e+00 0.8185 1 1 1
2 +3.372e+03 -7.658e+03 +2e+04 4e-02 7e-01 1e+00 5e+00 0.1517 1 1 2
3 +3.378e+03 -7.776e+03 +2e+04 4e-02 7e-01 1e+00 5e+00 0.0074 2 2 2
4 +3.393e+03 -7.605e+03 +2e+04 4e-02 7e-01 1e+00 5e+00 0.0075 2 2 1
5 +3.503e+03 -8.023e+03 +2e+04 4e-02 7e-01 2e+00 5e+00 0.2846 1 2 3
6 +3.635e+03 -7.239e+03 +2e+04 4e-02 7e-01 2e+00 5e+00 0.0817 2 3 1
7 +3.452e+03 -5.039e+03 +2e+04 3e-02 5e-01 2e+00 4e+00 0.4487 1 1 1
8 +3.527e+03 -6.190e+03 +2e+04 4e-02 6e-01 3e+00 4e+00 0.0326 2 2 2
9 +4.370e+03 -1.256e+03 +8e+03 2e-02 3e-01 2e+00 2e+00 0.8442 2 2 1
10 +4.554e+03 -6.663e+02 +5e+03 2e-02 3e-01 3e+00 1e+00 0.5235 2 2 2
11 +4.498e+03 +2.332e+03 +2e+03 8e-03 1e-01 1e+00 6e-01 0.9890 1 2 2
12 +4.746e+03 +3.944e+03 +8e+02 3e-03 5e-02 5e-01 2e-01 0.7198 2 2 1
13 +4.881e+03 +4.233e+03 +6e+02 2e-03 4e-02 4e-01 1e-01 0.5991 2 2 2
14 +5.002e+03 +4.821e+03 +2e+02 7e-04 1e-02 1e-01 4e-02 0.7945 2 2 2
15 +5.058e+03 +5.013e+03 +4e+01 2e-04 3e-03 3e-02 1e-02 0.8904 2 3 2
16 +5.073e+03 +5.064e+03 +9e+00 3e-05 6e-04 6e-03 2e-03 0.8155 2 3 2
17 +5.077e+03 +5.076e+03 +1e+00 5e-06 8e-05 9e-04 3e-04 0.8831 2 3 2
18 +5.077e+03 +5.077e+03 +3e-01 1e-06 2e-05 2e-04 7e-05 0.8291 2 2 2
19 +5.078e+03 +5.077e+03 +7e-02 3e-07 4e-06 5e-05 2e-05 0.8694 1 2 2
20 +5.078e+03 +5.078e+03 +6e-03 2e-08 9e-07 4e-06 2e-06 0.9371 2 2 2
21 +5.078e+03 +5.078e+03 +2e-03 1e-08 3e-07 1e-06 4e-07 0.8498 3 3 3
22 +5.078e+03 +5.078e+03 +4e-04 7e-08 2e-07 3e-07 9e-08 0.9490 2 2 2
23 +5.078e+03 +5.078e+03 +9e-05 4e-08 1e-07 7e-08 2e-08 0.7864 3 3 3
24 +5.078e+03 +5.078e+03 +8e-05 3e-08 1e-07 5e-08 2e-08 0.3844 2 1 1
25 +5.078e+03 +5.078e+03 +1e-05 8e-09 5e-08 8e-09 3e-09 0.9305 3 3 3
26 +5.078e+03 +5.078e+03 +4e-06 3e-09 2e-08 3e-09 1e-09 0.8103 2 1 1
Slacks/multipliers leaving the cone, recovering best iterate (26) and stopping.

Close to OPTIMAL (within feastol=2.0e-08, reltol=8.3e-10, abstol=4.2e-06).
Runtime: 0.053363 seconds.

5077.52993137
optimal_inaccurate

Let me know if this is accurate enough (feastol=2.0e-08, reltol=8.3e-10, abstol=4.2e-06).

[jaehyunp]

Yes, that's accurate enough! Just out of curiosity, why does the solver return optimal_inaccurate?

[adomahidi]

It fails to solve to required accuracy (which is somewhere near 2e-8, I think), but it manages to solve to "reduced" accuracy.

[jaehyunp]

Ah, I see. We only require 5-6 significant digits anyway (5077.53), so it's totally fine.