Opened 5 years ago
Last modified 3 years ago
#5766 assigned defect
Number of homotopy steps out of control
| Reported by: | Francesco Casella | Owned by: | Andreas Heuermann |
|---|---|---|---|
| Priority: | high | Milestone: | |
| Component: | Run-time | Version: | |
| Keywords: | Cc: |
Description
After PR #203, only two homotopy steps are taken by default instead of four. Furthermore, adding -ils=XX to the simulation flags in OMEdit apparently has no effect.
@AnHeuermann, could you please check?
Change History (13)
comment:1 by , 5 years ago
| Owner: | changed from to |
|---|---|
| Status: | new → assigned |
comment:3 by , 5 years ago
Indeed -ils will be read but got overwritten and I wasn't aware of the flag.
Now it should work again and I added a testcase to not destroy this again.
Also I set the default value back to 4.
See PR 665.
comment:4 by , 5 years ago
It's not clear to me what is the meaning of -ils if we have adaptive homotopy. Is it a first tentative value to have -ils steps, and then shorter steps are taken in case of trouble? Is this documented somewhere?
comment:5 by , 5 years ago
I can now control the number of homotopy steps again, thanks!
I only find the way steps are counted a bit weird. If I ask for 4 steps, I get lambda = 0.0, 0.333, 0.667, 1.0. To me, this is three steps, not four. Every step is the advancement of lambda from one data point to the next, not one particular data point. In fact, people often refer to "step size", which of course only makes sense if the step is the interval, not the data point.
Not to mention the famous Neil Armstrong quote, which according to -ils would have been "Two small steps for a men, one giant leap for mankind" :)
Could you please fix that as well? I guess that should be easy :)
Thank you!
comment:6 by , 5 years ago
@AnHeuermann, there is another problem with the homotopy logging, which is well demonstrated by running the test case attached to #5727. That model takes about 25 s to initialize with four (i.e. three) steps.
The problem is, if I am using OMEdit, the output window remains empty and silent for 25 solid seconds, until the end of the process, and only then I get the report on the whole thing in single step.
Windows is known to be lazy with the flushing of buffers, I guess you should add some statement to flush whatever output buffer the LOG_INIT_HOMOTOPY is writing to every time a line is written to it.
Thank you!
comment:8 by , 4 years ago
Is this ticket still unsolved?
commit 47fd12c added a fix and a test for this.
comment:9 by , 4 years ago
@AnHeuermann, the default is now back to four, thanks. I am still quite confused by the fact that the "four steps" are lambda = 0, 0.33, 0.66, 1. To me these are three steps, not four. Would you mind to fix that as well?
Regarding my other questions, see #6033.
comment:12 by , 4 years ago
| Milestone: | 1.17.0 → 1.18.0 |
|---|
Retargeted to 1.18.0 because of 1.17.0 timed release.
Maybe that's because now we have adaptive homotopy. In this case, you should definitely update the documentation, by removing the
-ilsparameter or at least deprecating it