temperature dependent mobilities

[Bernd]

HI Taka,

The performance will improve, if the time step which is needed for the 2/2 interface is smaller than the minimum allowed time step value which you can set in the time input data of the driving file. But he question is wether it will improve a lot. There are parts in the code which are very sensitive to the time stepping, in general all the list operations, which includes the phase-field solver and the list operations.
Using the .TabP output you can check out how much time is consumed in the different parts of MICRESS. If the columns titled "PF Time" and "List Time" make up a considerable amout of time, increasing the time step can help. In such cases, typically also "Diff Time" is increased, because, in addition to the diffusion itself, the solute redistribution, which is also a list operation, is included here.

As an alternative to reducing the mobility, you could also increase the mimimum time step (4th optional parameter in the time imput line in section "Time input data".
Advantage: Mobility is reduced only in those grid cells which would require a smaller time step!
Disadvantage: You do not have control over that...

Bernd

[ilovemicress]

Dear Bernd,

Thanks for your advice.
I will try changing the minimum time step.

According to .TabP output, the large proportion of the wallclock time is used for the TQ time which amazingly increases every 0.1 sec. I think this increase is due to updating diffusion coefficients data. Can the interval for the updating be changed to the larger value so as to reduce the total TQ time? (Does the lower cooling rate allow the larger interval time?) Would you enlighten me about the criteria for it?

Regards,
Taka

[Bernd]

Dear Taka,

when diffusion coefficients are updated from the mobility database using TQ, an Arrhenius description is calculated, which can be seen in the .diff output file. If "local" is selected (see Today's question about diffusivity input), also the slopes for a linear interpolation of the concentration dependencies are obtained.
So, the question how often you have to update these descriptions depends on how close these approximations fit to the exact description in the database, i.e. there is no strict rule for that! But using the .diff output, you can check how much the diffusion coefficients would deviate if you would not update them as often (just by extrapolating the diffusion coefficients from older updates and comparing to the values obtained with newer updates. How big the deviation is allowed to be is up to you...
Please note that including interpolation of the concentration ("local") increases TQ time for calculating the extrapolation! So it makes no sense to use the local extrapolation in order to be able to reduce the update frequency. This feature is important if there are strong concentration differences inside each phase, leading to locally different diffusion coefficients!

Finally, the hard check is to compare results with high and low update frequency: If the results are almost the same, the low update frequencies are justified!

Bernd

[ilovemicress]

Dear Bernd,

Thanks for your advice.
Comparing the results with high and low update frequency, I decided to use 1 sec for the updating interval.

I have a few questions about treatment of diffusion coefficient in MICRESS.
Pease see “Today's question about diffusivity input”.

PS: Can I ignore the huge amount of massage “Warning:Demixing in interface…” when “diff” diffusion flag option is used?

Regards,
Taka

[Bernd]

Hi Taka,

would be nice if MICRESS were able to give you a nice massage!

But kidding aside, the message “Warning:Demixing in interface…” has nothing directly to do with diffusion! It means that MICRESS detected a "demixing" in at least some cells during relinearisation of the thermodynamic description (opposite signs of solidus and solvus slopes in binary cut through linearised phase diagram, possibly leading to a problem during solute redistribution). Normally you should not get a "huge amount" of these messages but only once per relinearisation step. Sometimes you can ignore them, but as soon as you see any numerical problems in the concentration distribution at some interfaces, you should remember the message and do something about the definition of stoichiometric phases (see ’Demixing in interface LIQUID/FCC_A1, component FE’).

Bernd

PS: If you have more trouble with this point, please open a new thread!