Trouble with recreating published data

[YoffeG]

Hello Prof. Agol,

I'm currently attempting to recreate the adopted parameters of KOI-738 in Hutter et al., 2016 (they also used TTVFaster) using transit data from the Holczer Catalogue. The fit isn't good and only approaches some similarity (otherwise nearly linear - no TTVs) when I raise j_max to 10, however it does not seem the adequate way. I express the Planetary masses as advised in TTVFast (mass ((earth_mass/solar mass) stellar mass). Additionally, the output data is absolutely insensitive to changing some parameters (such as defying co-planarity by changing the longitude of periastron of one of the planets, or both).

Would greatly appreciate any advice you may have that would focus my troubleshooting!

Thanks in advance,

Gidi Yoffe

[ericagol]

Sorry, I'm not able to help out right away. Maybe a plot of the residuals would help?

Eric Agol Astronomy Professor University of Washington

On Aug 18, 2017, at 6:45 PM, Gideon (Gidi) Yoffe notifications@github.com wrote:

Hello Prof. Agol,

I'm currently attempting to recreate the adopted parameters of KOI-738 in Hutter et al., 2016 (they also used TTVFaster) using transit data from the Holczer Catalogue. The fit isn't good and only approaches some similarity (otherwise nearly linear - no TTVs) when I raise j_max to 10, however it does not seem the adequate way. I express the Planetary masses as advised in TTVFast (mass ((earth_mass/solar mass) stellar mass). Additionally, the output data is absolutely insensitive to changing some parameters (such as defying co-planarity by changing the longitude of periastron of one of the planets, or both).

Would greatly appreciate any advice you may have that would focus my troubleshooting!

Thanks in advance,

Gidi Yoffe

― You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

[YoffeG]

Thanks for your quick response!

If by residuals you mean the O-C plot. Attached are two plots:

• This is a residual plot with j_max = 0 (Black dots are model data)

• This is a residual plot with j_max = 10 (Black dots are model data)

Thank you very much again

[ericagol]

Very strange - what version are you using? (C, Python or Julia)

Eric Agol Astronomy Professor University of Washington

On Aug 18, 2017, at 10:11 PM, Gideon (Gidi) Yoffe notifications@github.com wrote:

Thanks for your quick response!

If by residuals you mean the O-C plot. Attached are two plots

one with j_max = 10, and the other with j_max = 0 (the linear one)

Thank you very much again

― You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[ericagol]

I looked back at the paper, and this is not one of the systems we modeled with TTVFaster (only Kepler-26 & Kepler-177). The others were modeled with N-body integrations.

It is possible that resonant terms or higher order terms need to be included to model this system (we didn't investigate this). I can recommend TTVFast if you want a package that does so.

Cheers,

Eric Agol Astronomy Professor University of Washington

On Aug 18, 2017, at 10:11 PM, Gideon (Gidi) Yoffe notifications@github.com wrote:

Thanks for your quick response!

If by residuals you mean the O-C plot. Attached are two plots

one with j_max = 10, and the other with j_max = 0 (the linear one)

Thank you very much again

― You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[YoffeG]

Thanks again for your response,

I use both TTVFast and TTVFaster on Python.

At first, I tried to run an n-dimensional grid-search of parameters (mainly the mean anomalies of both planets, seeing I was unsure of my calculation of them) with TTVFast attempting to replicate the result presented in Hutter et al., 2016, using the parameters they presented for the Holczer Catalogue data just as I did with TTVFaster. I ran the grid-search both in the e/omega and ecosOmega/esinOmega planes and was able to receive a relatively good fit only with parameters within 2sigma range (orbital periods were several sigmas away) - but once again, not with the adopted values. In the TTVFast case, I assumed that due to coplanarity, longitudes and arguments of periastron of each planet would be equal and ran a 2D grid-search to solve for the minimal chi-square mean anomalies.

Attached are two plots of model data for Kepler-29 produced with TTVFast:

• Parameters within 2sigma range (except for orbital periods that were farther off):

• Best-fit of a 2D grid-search solving for mean-anomalies of both planets with all other input parameters being the adopted values for the system from Hutter et al., 2016 for KOI-738 using data from the Holczer Catalogue:

For both of the above results, coplanarity was assumed and the only parameters that were altered were masses, periods, ecosOmega, esinOmega (they were converted to eccentricity and omega using Pythagorean theorem and atan2 function, respectively) and the mean anomalies.

Thank you very much again for advising me on this matter,

Gidi

[ericagol]

Sorry for the delay...

There are some possible issues that might help resolve the discrepancy between TTVFast parameters & Jontof-Hutter et al's parameters: 1). The starting time of the integration may be different (the orbital elements osculate with time); 2). The coordinate system used may be different (for instance, the longitudes may be referred to the Sky plane or the transit plane; for the planet or the star).

The parameters that should be very similar are the mass ratios as these are independent of coordinates (unless one uses Jacobi, but that difference should be slight).

I can't recall the details of what choices Jontof-Hutter used, but I don't think he used TTVFast, so it is likely that the choices are different. The other possibility is that there may be a bug or typo in that paper and/or there is a degenerate set of parameters that fit the data equally well, which would also be interesting! Another possible source of discrepancy is how one handles outliers - J-F et al used a heavy-tailed t-distribution for some of the cases.

I would suggest getting in touch with Daniel J-F about this to track down these possibilities (he is now at the University of the Pacific).

Eric Agol Astronomy Professor University of Washington

On Aug 18, 2017, at 11:00 PM, Gideon (Gidi) Yoffe notifications@github.com wrote:

Thanks again for your response,

I use both TTVFast and TTVFaster on Python.

At first, I tried to run an n-dimensional grid-search of parameters (mainly the mean anomalies of both planets, seeing I was unsure of my calculation of them) with TTVFast attempting to replicate the result presented in Hutter et al., 2016, using the parameters they presented for the Holczer Catalogue data just as I did with TTVFaster. I ran the grid-search both in the e/omega and ecosOmega/esinOmega planes and was able to receive a relatively good fit only with parameters within 2sigma range (orbital periods were several sigmas away) - but once again, not with the adopted values. In the TTVFast case, I assumed that due to coplanarity, longitudes and arguments of periastron of each planet would be equal and ran a 2D grid-search to solve for the minimal chi-square mean anomalies.

Attached are two plots of model data for Kepler-29 produced with TTVFast:

Parameters within 2sigma range (except for orbital periods that were farther off:

Best-fit of a 2D grid-search solving for mean-anomalies of both planets with all other input parameters being the adopted values for the system from Hutter et al., 2016 for KOI-738 using data from the Holczer Catalogue:

― You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[YoffeG]

Prof. Agol,

Thanks a lot for your reply. I've sent an e-mail to your University of Washington address with further details.

Gidi