Open juzikong opened 5 years ago
@juzikong, I think the trend you're seeing in the [Mg/Fe]-[Fe/H] plane for [Fe/H] < -2.5 is largely a consequence of the simulations time resolution (100 Myr/step). The evolution of the system from [Fe/H] = -4 to -2.5 is very rapid and corresponds to only ~2-3 model steps. In the following plot of [Mg/Fe] vs. time in the tutorial, the [Mg/Fe] increases slightly over these first few steps as well to reach its peak pretty quickly after less than a Gyr. Without looking into it further, I'm not entirely sure why the peak isn't at the very first time step. And I don't know if this is a physical prediction of the model or an artifact of their only being a few steps after initializing the model. Perhaps someone else can comment to that effect.
If you run the same simulation with 10, 50, and 100 Myr time steps, you'll notice that they may disagree at very low metallicities—though they should agree at late times (see plot in cell 15 of tutorial #5). In general, if you want to use Chempy to fit abundances at very low metallicities, I would recommend increasing the time resolution to 50 or even 10 Myr.
I hope that helps!
Thanks a lot for your reply!
Exactly. I also thought about the time step influence and as you mentioned. The plot showing the effect of different time steps is nice and reasonable but it won't explain why the peak is not at the very first time step.
I see some other simulations can have a flat relation for low [Fe/H] range IF assuming instantaneous recycling (all star>1 M_sun die instantly after formation) but never a positive slope.
If you don't assume instantaneous recycling, then each new time step should have a lower [Mg/Fe] according to the first principle. In this case, even a flat relation (as in cell 15 of tutorial #5) would be wrong.
I am not asking you to go through the code but maybe add some warning and explanations at least. As the plot can be misleading for people that are not familiar with the issue.
Best regards, Yan
I just realized one possibility. The stellar yield table is often not monotonic and has a large fluctuation. If you didn't smooth it with some spline fit. It could possibly cause a slightly positive slope. This can be easily checked.
If it is indeed the problem, I won't say not smoothing is wrong. But in general, the yield table is very sparse, and not smoothing it leads to unphysical random errors. I personally prefer a smoothed yield table in my simulation.
But then, there is no reason for the relation to be exactly flat as shown in your cell 15 of tutorial #5......So this can't be correct...
Hey, sry for not replying so long. I had my notifications turned off. Still not sure if they are on now...
So I think Nathan is right that the sparse time steps in the beginning might be the problem here. When starting Chempy I had thin disk chemical evolution in mind and did not look into very early times. Nathan is now looking into dwarf galaxies and is solving few of those issues. Chempy is linearly interpolating the yield tables in masses and if I recall correctly in log metallicity. The rise is likely due to the CC-SN feedback of Mg/Fe being metallicity dependent, but could also be due to initialisation problems. In order to circumvent this I would propose to use a finer time grid and discard the first timestep.
Hope that helps. Cheers, Jan
Dear all,
Thanks for the update. It is certainly a complicated task to check code behavior and initialization problems. I have my own chemical evolution code and the initialization is the part I made the most mistakes. And depending on my experience, I do agree with you that after a few more time steps, the initialization becomes less important and the results tend to converge to the correct one as long as the stellar yield table is correct. This is generally due to the feedback effect of the stellar yields.
Dear Jan,
I have one question about your reply though.
The rise is likely due to the CC-SN feedback of Mg/Fe being metallicity dependent,
I am not sure this is a possible reason because lower metallicity stars have higher [Mg/Fe] as far as I know it. Thus it is always the case that [Mg/Fe] should only decrease.
Best regards, Yan Zhiqiang
On Wed, 28 Aug 2019 at 10:06, Jan Rybizki notifications@github.com wrote:
Hey, sry for not replying so long. I had my notifications turned off. Still not sure if they are on now...
So I think Nathan is right that the sparse time steps in the beginning might be the problem here. When starting Chempy I had thin disk chemical evolution in mind and did not look into very early times. Nathan is now looking into dwarf galaxies and is solving few of those issues. Chempy is linearly interpolating the yield tables in masses and if I recall correctly in log metallicity. The rise is likely due to the CC-SN feedback of Mg/Fe being metallicity dependent, but could also be due to initialisation problems. In order to circumvent this I would propose to use a finer time grid and discard the first timestep.
Hope that helps. Cheers, Jan
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/jan-rybizki/Chempy/issues/12?email_source=notifications&email_token=AE5F3MX5XMDVEZ3RKOQRA4TQGYWW7A5CNFSM4HYKWLB2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD5KIERY#issuecomment-525632071, or mute the thread https://github.com/notifications/unsubscribe-auth/AE5F3MUPW7AXPRJ4D7JXIFTQGYWW7ANCNFSM4HYKWLBQ .
Dear Yan,
if you look at https://github.com/jan-rybizki/Chempy/blob/master/tutorials/2-Nucleosynthetic_yields.ipynb output 16 and 17 the [Mg/Fe] yield of CC-SN yields are shown for Nomoto 2013 and Chieffi+ 2004 respectively. So for Nomoto the lowest metallicity is only about -1,2 dex, therefore the CC-SN yield will be flat below. For Chieffi+ 2004 seems to have a peak at -2 dex. Therefore rising trends before -2 could be due to the CC-SN yields. I am not saying this behaviour is physically right. Just that the yield table provides this kind of data.
Cheers, Jan
Indeed. Interesting.
I guess it can be easily tested whether the low Z part of the table is the cause of an increasing [Mg/Fe].
Also, it would be important to know if there is a physical reason behind this yield behaviour.
Best regards, Yan
On Thu, 29 Aug 2019 at 09:37, Jan Rybizki notifications@github.com wrote:
Dear Yan,
if you look at https://github.com/jan-rybizki/Chempy/blob/master/tutorials/2-Nucleosynthetic_yields.ipynb output 16 and 17 the [Mg/Fe] yield of CC-SN yields are shown for Nomoto 2013 and Chieffi+ 2004 respectively. So for Nomoto the lowest metallicity is only about -1,2 dex, therefore the CC-SN yield will be flat below. For Chieffi+ 2004 seems to have a peak at -2 dex. Therefore rising trends before -2 could be due to the CC-SN yields. I am not saying this behaviour is physically right. Just that the yield table provides this kind of data.
Cheers, Jan
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/jan-rybizki/Chempy/issues/12?email_source=notifications&email_token=AE5F3MWVYJDDQ5WX24EYXG3QG54CRA5CNFSM4HYKWLB2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD5NR5KI#issuecomment-526065321, or mute the thread https://github.com/notifications/unsubscribe-auth/AE5F3MVUQV3LSJ3F6ADAWZLQG54CRANCNFSM4HYKWLBQ .
BTW, if you have publications with these plots (output 16 and 17 the [Mg/Fe] yield of CC-SN yields) I can cite it right now.
Dear Yan,
you can just take the jupyter notebook and let this plot run with a higher time-resolution or play with other parameters. As for the physical reason you would have to ask people who create those yield tables. I would be fine if you point to the github tutorial. But in principle its just a visualisation of the tables Nomoto and Chieffi+ provided. The closest visualisation to that in a publication are fig 1 and 2 of https://arxiv.org/abs/1906.05297 but they are not normalised to Fe.
Cheers, Jan
Looking at your tutorials 7-Acessing Chempy paper 1 abundance tracks.ipynb https://github.com/jan-rybizki/Chempy/blob/master/tutorials/7-Acessing%20Chempy%20paper%201%20abundance%20tracks.ipynb The first plot about the Mg/Fe--Fe/H relation.
How is it possible the [Mg/Fe] is flat (or even increasing) for [Fe/H]~-4 to -2.5?
For this low iron abundance, the age of the galaxy must be very young and the metal enrichment is mostly (if not only) due to CCSN. The more massive star explodes early and gives higher [Mg/Fe] ejecta. Thus the [Mg/Fe] value should reach its peak at the earliest time (lowest [Fe/H]) and only decreases afterward.
Why in your simulation it is the contrary? Could you help me understand? Thanks!