[BART-devel] possible bug in ctips

[aj-foster]

The following is an archived message from the BART-devel mailing list, which has now closed.

From: ja***@ph*** (Jasmina Blecic)
Date: Fri, 10 Jul 2015 22:57:44 -0400
Subject: [BART-devel] possible bug in ctips

Hi Pato,

I have attached two plots that show opacities, for two different files
made with the same water databases. I have plotted the opacities using
my code, reading the data my way, that I have already showed you.

Figure FORTRAN_TIPS.png : (run with TLI_fix branch before ctips, from
0.6 - 5.5, 1e-6 thresh)
pyline cfg file ==>
/home/jasmina/BART/BART_WASP43b-old/TLI_fix/pylineread/TLI/pyline_4mol-06-55_TLI_fix.cfg
opacity_file ==>
/home/jasmina/BART/BART_WASP43b-old/TLI_fix//run/opacity_file_HITRAN_4mol-06-55um-100-3000K_TLI_fix.dat'

Figure CTIPS.png: (run with new BART from July 8th, with ctips, from
0.61 - 5.5, 1e-6 thresh)
pyline cfg file
==>/home/jasmina/ast/esp01/BART/BART-8Jul2015/modules/transit/tli-files/pyline_HITRAN2012-4mol-061-55.cfg
opacity_file ==>
/home/jasmina/ast/esp01/BART/BART-8Jul2015/opacity_files/opacity_file_HITRAN2012_4mol-061-55um-100-3000K_1e6_100bar.dat

Figure FORTRAN_TIPS.png: uses fully HITRAN 2008
Figure CTIPS.png: uses also HITRAN 2012 (_affects only _CH4)

Figures differ significantly especially in H2O.
Is there some explanation that I am overseeing?

I did not know who actually converted FORTRAN tips to ctips so I sent it
to bart-devel list as well.

Thanks
Jasmina

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://physics.ucf.edu/pipermail/bart-devel/attachments/20150710/6d976ac1/attachment-0001.html
-------------- next part --------------
A non-text attachment was scrubbed...
Name: FORTRAN_TIPS.png
Type: image/png
Size: 52595 bytes
Desc: not available
URL: http://physics.ucf.edu/pipermail/bart-devel/attachments/20150710/6d976ac1/attachment-0002.png
-------------- next part --------------
A non-text attachment was scrubbed...
Name: CTIPS.png
Type: image/png
Size: 61855 bytes
Desc: not available
URL: http://physics.ucf.edu/pipermail/bart-devel/attachments/20150710/6d976ac1/attachment-0003.png

[aj-foster]

From: pc***@fu*** (Patricio Cubillos)
Date: Sat, 11 Jul 2015 10:58:47 -0400
Subject: [BART-devel] possible bug in ctips

Hi,

This is interesting, I would not expected it in principle. However, making the comparison is not so simple.
There are many commits between the two versions of the code you ran. And there are many steps between the partition-function calculation and the opacity-file calculation. And you used different inputs for the TLI file (however, the latter should not affect other molecules except CH4).

You have a code that reads the TLI files, did you check that the partition-functions values differ?
If they don't, then there must be some change elsewhere in the code.

I did not know who actually converted FORTRAN tips to ctips so I sent it to bart-devel list as well.
I wrote it, AJ tested and plugged it in into Transit. AJ checked that the numbers match the FORTRAN TIPS code down to the machine precision.

Patricio.

[aj-foster]

From: ja***@ph*** (Jasmina Blecic)
Date: Sat, 11 Jul 2015 14:14:20 -0400
Subject: [BART-devel] possible bug in ctips

Hi Pato,

I have attached the code that reads TLI and two plots that show
difference between part functions for H20 isotopes and CO2 isotopes.

The difference is not positive on behalf of CTIPs as you have suggested,
rather fluctuating. It is generally larger fluctuation for higher
temperatures.

I gave you the code to read TLI files and produce the plots. Just copy
and paste the lines into ipython. Part functions array carry zeros for
files repating for the same species. Just first has values.

To run the code you must give the file name, number of files in the
pyline config, and indexes of databases to be skipped (because part
functions repeats for the same species).

Hope, you'll figure it out what is wrong.
AJ and I came up with the FORTRAN rounding, precision issue in making
CIA files. Maybe AJ can explain better (if this is a precision issue at
all).

Best
Jasmina

On 07/11/2015 10:58 AM, Patricio Cubillos wrote:

Hi,

This is interesting, I would not expected it in principle. However, making the comparison is not so simple.
There are many commits between the two versions of the code you ran. And there are many steps between the partition-function calculation and the opacity-file calculation. And you used different inputs for the TLI file (however, the latter should not affect other molecules except CH4).

You have a code that reads the TLI files, did you check that the partition-functions values differ?
If they don't, then there must be some change elsewhere in the code.

I did not know who actually converted FORTRAN tips to ctips so I sent it to bart-devel list as well.
I wrote it, AJ tested and plugged it in into Transit. AJ checked that the numbers match the FORTRAN TIPS code down to the machine precision.

Patricio.

---

BART-devel mailing list
BA***@ph***
https://physics.ucf.edu/mailman/listinfo/bart-devel

-------------- next part --------------
A non-text attachment was scrubbed...
Name: H2O_diff.png
Type: image/png
Size: 119293 bytes
Desc: not available
URL: http://physics.ucf.edu/pipermail/bart-devel/attachments/20150711/c347565e/attachment-0002.png
-------------- next part --------------
A non-text attachment was scrubbed...
Name: CO2_diff.png
Type: image/png
Size: 97851 bytes
Desc: not available
URL: http://physics.ucf.edu/pipermail/bart-devel/attachments/20150711/c347565e/attachment-0003.png
-------------- next part --------------
A non-text attachment was scrubbed...
Name: readTLI_Pato.py
Type: text/x-python
Size: 8525 bytes
Desc: not available
URL: http://physics.ucf.edu/pipermail/bart-devel/attachments/20150711/c347565e/attachment-0001.py

[aj-foster]

From: pc***@fu*** (Patricio Cubillos)
Date: Sat, 11 Jul 2015 14:45:23 -0400
Subject: [BART-devel] possible bug in ctips

Hi,

On Jul 11, 2015, at 2:14 PM, Jasmina Blecic <ja***@ph***> wrote:

I have attached the code that reads TLI and two plots that show difference between part functions for H20 isotopes and CO2 isotopes.
The difference is not positive on behalf of CTIPs as you have suggested, rather fluctuating. It is generally larger fluctuation for higher temperatures.
I didn't say that the partition-function values from CTIPS are larger than TIPS. I was talking about the plots you showed, the opacities.
Hope, you'll figure it out what is wrong.
AJ and I came up with the FORTRAN rounding, precision issue in making CIA files. Maybe AJ can explain better (if this is a precision issue at all).
I re-plotted the figures, this time I plotted: 100*(PF1-PF2)/PF1 to see the difference in percentage.
There you can see that the partition functions differ by less than 0.0005%, and the diff are evenly distributed around zero. So, I doubt that CTIPS is the source of the difference in the opacities.

Figures in:
/home/patricio/ast/esp01/code/tests/python/TLI_PF_diff/CO2_diff.png
/home/patricio/ast/esp01/code/tests/python/TLI_PF_diff/H2O_diff.png

P