Comparison between ADIPLS and GYRE

[warrick]

I've been toying around with both ADIPLS and GYRE lately, and recently found that GYRE and ADIPLS seem to give me different results. I noticed this error because I was calculating frequencies for other solar models with GYRE and found that they didn't match observed frequencies (e.g. the BiSON 23 year data) properly. Not as well as the ADIPLS output for Model S, anyway.

I've attached some input files. I took the `FGONG` model for Model S [1] directly from Jørgen Christensen-Dalsgaard's webpage [2]. It didn't want to upload here for some reason, so I leave it to the reader to get it. The `.in` files are ADIPLS input to generate the adiabatic frequencies, plus the one GYRE input file. Here's a side-by-side comparison of the first few frequencies from the output.

Code: Select all

l  n    ADIPLS       GYRE
0  1  258.0090
0  2  404.4811
0  3  535.9398   536.6424
0  4  680.5734   681.4681
0  5  825.3655   826.4537
0  6  972.7470   974.0313
0  7 1118.1517  1119.6280
0  8 1263.5260  1265.1957
0  9 1407.6297  1409.4920
0 10 1548.5240  1550.5783

Now, before anyone starts getting worried, I'd like to point out that I'm not an experienced ADIPLS user, so I might be doing something daft here, like not adding a surface point or an atmosphere or something. I did change the boundary condition in the GYRE input to `JCD` but that didn't help. (I generally haven't found that changing the boundary conditions makes a noticeable difference.) GYRE was compiled on Scientific Linux 6.4 (Carbon) with the MESA SDK (April 2013 edition) and 4 cores. I used a 2nd-order integrator but I found no difference with the fourth-order integrator.

Let me know if any further information is useful, although I'll mostly be away for the next two weeks.

Cheers,
Warrick

[1] http://users-phys.au.dk/jcd/solar_model ... l5bi.d.15c
[2] http://users-phys.au.dk/jcd/solar_models/

[rhtownsend]

I've been toying around with both ADIPLS and GYRE lately, and recently found that GYRE and ADIPLS seem to give me different results. I noticed this error because I was calculating frequencies for other solar models with GYRE and found that they didn't match observed frequencies (e.g. the BiSON 23 year data) properly. Not as well as the ADIPLS output for Model S, anyway.

I've attached some input files. I took the `FGONG` model for Model S [1] directly from Jørgen Christensen-Dalsgaard's webpage [2]. It didn't want to upload here for some reason, so I leave it to the reader to get it. The `.in` files are ADIPLS input to generate the adiabatic frequencies, plus the one GYRE input file. Here's a side-by-side comparison of the first few frequencies from the output.

Code: Select all

l  n    ADIPLS       GYRE
0  1  258.0090
0  2  404.4811
0  3  535.9398   536.6424
0  4  680.5734   681.4681
0  5  825.3655   826.4537
0  6  972.7470   974.0313
0  7 1118.1517  1119.6280
0  8 1263.5260  1265.1957
0  9 1407.6297  1409.4920
0 10 1548.5240  1550.5783

Now, before anyone starts getting worried, I'd like to point out that I'm not an experienced ADIPLS user, so I might be doing something daft here, like not adding a surface point or an atmosphere or something. I did change the boundary condition in the GYRE input to `JCD` but that didn't help. (I generally haven't found that changing the boundary conditions makes a noticeable difference.) GYRE was compiled on Scientific Linux 6.4 (Carbon) with the MESA SDK (April 2013 edition) and 4 cores. I used a 2nd-order integrator but I found no difference with the fourth-order integrator.

Let me know if any further information is useful, although I'll mostly be away for the next two weeks.

Cheers,
Warrick

[1] http://users-phys.au.dk/jcd/solar_model ... l5bi.d.15c
[2] http://users-phys.au.dk/jcd/solar_models/

Interesting stuff! As chance would have it, I've been looking into some differences between ADIPLS and GYRE frequencies for an RGB model that Dennis Stello sent me. It seems that the frequencies depend on how the pulsation equations are formulated -- something that shouldn't matter in the real world, but does matter when one's doing numerical simulations. In a day or two, I'll be making a new release of GYRE which allows the equation formulation to be switched to Jørgen's version; we should see whether that clears up the discrepancies you report above.

cheers,

Rich

[warrick]

Okay, it looks like the errors are actually much smaller. I wasn't normalizing the dimensionless frequencies correctly. While the new version is in the works, the real differences for the original table are (for example):

Code: Select all

l  n    ADIPLS       GYRE
0  1  258.0090
0  2  404.4811
0  3  535.9398   535.9457
0  4  680.5734   680.5834
0  5  825.3655   825.3807
0  6  972.7470   972.7666
0  7 1118.1517  1118.1742
0  8 1263.5260  1263.5528
0  9 1407.6297  1407.6617
0 10 1548.5240  1548.5648

So much smaller, but probably still worrisome. Anyway, I'll wait for the update, just thought I'd own up to my mistake!

W

[rhtownsend]

Okay, it looks like the errors are actually much smaller. I wasn't normalizing the dimensionless frequencies correctly. While the new version is in the works, the real differences for the original table are (for example):

Code: Select all

l  n    ADIPLS       GYRE
0  1  258.0090
0  2  404.4811
0  3  535.9398   535.9457
0  4  680.5734   680.5834
0  5  825.3655   825.3807
0  6  972.7470   972.7666
0  7 1118.1517  1118.1742
0  8 1263.5260  1263.5528
0  9 1407.6297  1407.6617
0 10 1548.5240  1548.5648

So much smaller, but probably still worrisome. Anyway, I'll wait for the update, just thought I'd own up to my mistake!

W

Hi Warrick --

One thing worth checking is that ADIPLS and GYRE are using the same value for the gravitational constant G. You can set this in GYRE using the G parameter of the &coeffs namelist.

cheers,

Rich

[rhtownsend]

Okay, it looks like the errors are actually much smaller. I wasn't normalizing the dimensionless frequencies correctly. While the new version is in the works, the real differences for the original table are (for example):

Code: Select all

l  n    ADIPLS       GYRE
0  1  258.0090
0  2  404.4811
0  3  535.9398   535.9457
0  4  680.5734   680.5834
0  5  825.3655   825.3807
0  6  972.7470   972.7666
0  7 1118.1517  1118.1742
0  8 1263.5260  1263.5528
0  9 1407.6297  1407.6617
0 10 1548.5240  1548.5648

So much smaller, but probably still worrisome. Anyway, I'll wait for the update, just thought I'd own up to my mistake!

W

Hi Warrick --

One thing worth checking is that ADIPLS and GYRE are using the same value for the gravitational constant G. You can set this in GYRE using the G parameter of the &coeffs namelist.

cheers,

Rich

Also, try v2.1, which I've just released.

cheers,

Rich

[warrick]

Rich,

I thought I'd return to this issue, now that we're a few versions down the line and I have a better idea what I'm doing. I've set G to the same values in ADIPLS and GYRE and selected the "JCD" boundary condition, and I'm using adiabatic GYRE 2.2 compiled with MESA SDK (2013/04/08) on Scientific Linux 6.4.

In short, I'm still finding some discrepancies, and they now appear to be frequency dependent. I've attached an input FGONG file as well as ADIPLS inputs and GYRE inputs that produce the table of frequencies. There were a few misidentified frequencies because of the sparse model grid, and ADIPLS covered a larger frequency range (by some arbitrary choice) so the table is just of the frequencies that appear in both outputs. Here's a plot of the frequency differences, in the sense (GYRE - ADIPLS) plotted as a function of the ADIPLS frequencies.



At low frequencies, the results are basically now indistinguishable. The frequency differences are less than 0.001 uHz, which doesn't phase me. But, as you can see, the differences increase to several uHz at higher frequencies. They look to me like some kind of surface term.

I did these calculations without resampling the model grid, so I'm going to see if that closes the gap. It might increase the ADIPLS frequencies, for all I know.

Cheers,
Warrick

PS: I'm hope I'm being helpful and not still making foolish errors!

[rhtownsend]

Hi Warrick --

I can reproduce these discrepancies, so I don't think you're doing anything wrong. I've tried increasing the shooting grid resolution, but that doesn't fix the problem.

Are you doing any post-processing to the ADIPLS frequencies --- e.g., a surface correction?

cheers,

Rich

[warrick]

Rich,

First off, no, I'm not doing any post-processing: this is straight from "set-obs.d 5 ...". But I think I've solved my problem anyway. After a lot of faffing and unnecessary posts (which I deleted, for the sake of everyone's sanity...) I've pinned down the error to the boundary condition. Here's a new inlist and a new plot (both axes in uHz) of the differences between the ADIPLS and GYRE frequencies. There's a tiny frequency dependent error remaining, but it reaches a maximum of 1.3 nHz at 4.5 mHz, so I'm going to take that as agreement. I just had to change the GYRE boundary condition to JCD.

GYRE-ADIPLS.png (3.65 KiB) Viewed 7521 times

Sorry to have sounded a false alarm!

Cheers,
Warrick

[rhtownsend]

Rich,

First off, no, I'm not doing any post-processing: this is straight from "set-obs.d 5 ...". But I think I've solved my problem anyway. After a lot of faffing and unnecessary posts (which I deleted, for the sake of everyone's sanity...) I've pinned down the error to the boundary condition. Here's a new inlist and a new plot (both axes in uHz) of the differences between the ADIPLS and GYRE frequencies. There's a tiny frequency dependent error remaining, but it reaches a maximum of 1.3 nHz at 4.5 mHz, so I'm going to take that as agreement. I just had to change the GYRE boundary condition to JCD.

GYRE-ADIPLS.png

Sorry to have sounded a false alarm!

Cheers,
Warrick

No problem! In fact, I'm not sure it was completely false. From looking through your ADIPLS input file, it looks like ADIPLS was told to use the "zero" BC and not the isothermal atmosphere BC -- so I'm surprised you get good agreement. Maybe I need to check the input file again.

Are you still having problems with radial modes, or did that go away?

[warrick]

With the 'ZERO' boundary condition, I got the small (up to ~3uHz) difference. With the 'JCD' boundary conditions, the differences decrease to less than 1.3 nHz.

From looking through your ADIPLS input file, it looks like ADIPLS was told to use the "zero" BC and not the isothermal atmosphere BC -- so I'm surprised you get good agreement.

Whoa, I'd have to look through as well. I modified the default input generated by Joao Marques's Python modules for CES(T)AM. So not sure what atmosphere it's using...

Are you still having problems with radial modes, or did that go away?

The radial mode problem is gone for now. I'll have to re-establish out how I produced it in the first place! I think it happened if I used variable_type = 'JCD' without increasing the shooting mesh resolution. I'll post again if I can reproduce that bug.