dasslAUTO runs differently on monch versus Windows.

[jdmo]

Hola!

I ran dasslAUTO on a SolverInput.dat file that I was interested in, both on monch and on my Windows laptop machine. However, they integrated to different times -- monch went to 0.19 seconds, while my Windows machine went to 0.21 seconds.

I've uploaded my SolverInput.dat file to my Athena account at:

http://web.mit.edu/jdmo/Public/SolverInput.dat

Thanks!

[gmagoon]

The file is now on github under Downloads: http://github.com/downloads/GreenGroup/RMG-Java/SolverInput.dat

[rwest]

In case you care, SolverOutput.dat on my mac at a277d05a5299684a874d30c1e86b30766eebe0d2 starts

1580
103
0.19823840374238824
2.607683348961955E-7
0.0000032027378365837386

...which is 0.20 seconds (unless you want to go beyond 2 sig fig..)

[gmagoon]

Thanks Richard. Jeff, could you post the exact results you got from Windows vs. Linux?

[jdmo]

SolverOutput.dat on my Windows laptop machine at a277d05a5299684a874d30c1e86b30766eebe0d2 starts

 1523
 103
 0.21628313157279216
 1.2707096380240536E-7
 0.0000029201348570868475

[jdmo]

SolverOutput.dat on monch at a277d05a5299684a874d30c1e86b30766eebe0d2 starts

1718
 103
 0.1983644732746758
 2.6002754339931665E-7
 0.0000032015569586521763

So it seems I rounded incorrectly (but consistently!) when I reported the numbers -- 0.22 versus 0.20 sec.

[rwest]

Greg - what do the numbers mean? I notice that my Mac and Monch both have 0.198.. third and 2.6E-7 fourth, but they have different first numbers.

Jeff - what is your RTOL and ATOL and what happens if you decrease them? It could well be that when solving the amount of each species to the accuracy you requested, the time it takes for a flux to exceed the critical flux could vary anywhere between 0.19 and 0.22 seconds (i.e. this may be expected behaviour.)

[gmagoon]

First=number of ODESolver iterations to time point in the output Second=number of state variables or number of species...not sure off the top of my head which, but it should just be a difference of one Third=time where the concentrations are reported (i.e. when the threshold was first exceeded) Fourth=concentration for species 1

Good point about tolerances and expected behavior, etc., by the way.

[jdmo]

My RTOL is 1e-8, and my ATOL is 1e-16.

I'm testing the behaviour when I adjust them as we speak and will post the results once they're available =]

[rwest]

Jeff, did you get bored of this? I have just found that the RMG jobs when fortran codes are compiled on gfortran vs g95 diverge (see https://gist.github.com/786699 ) and I wonder if it's also in dasslAUTO.

[jdmo]

I haven't looked into this recently, sorry Richard! I'll try to get back to this in a month or so?