search

arfc / sneks4me

Simplified Nek Simulations for Moltres Enhancement
BSD 3-Clause "New" or "Revised" License
2 stars 0 forks source link

Add filter to simulations #17

Closed anshchaube closed 5 years ago

anshchaube commented 5 years ago

The turbOutflow example has :

filtering = hpfrt # set to none in case of Smagorinski 
filterWeight = 10
filterCutoffRatio = 0.9

Is this an appropriate filter for our simulations? Are these parameters appropriate?

Resources to consider - Pope; Deville, Fischer, Mund.

dshaver-ANL commented 5 years ago

what are the flow conditions and what polynomial order (lx1) are you using?

anshchaube commented 5 years ago

Re=1000, everything's normalized like it was during the summer for the flat plenum case, lx1 = 6.

dshaver-ANL commented 5 years ago

okay. For a laminar flow the filter isn't really important (everything should be fully resolved). Just for stability though, use the explicit filter:

filtering = explicit filterWeight = 0.02 filterCutoffRatio = 0.65

When you run it, check the logfile after the first time step for the qfilt line. Make sure it's only touching the last 2 modes.

anshchaube commented 5 years ago

I searched for qfilt, didn't find it. I did see this however:

filt amp  .0000  .0000  .0000  .0000  .0050  .0200
filt trn 1.0000 1.0000 1.0000 1.0000  .9950  .9800
dshaver-ANL commented 5 years ago

That's the one!

-------- Original message -------- From: Ansh Chaube notifications@github.com Date: 9/16/19 5:08 PM (GMT-06:00) To: arfc/sneks4me sneks4me@noreply.github.com Cc: "Shaver, Dillon R." dshaver@anl.gov, Comment comment@noreply.github.com Subject: Re: [arfc/sneks4me] Add filter to simulations (#17)

I searched for qfilt, didn't find it. I did see this however:

filt amp .0000 .0000 .0000 .0000 .0050 .0200 filt trn 1.0000 1.0000 1.0000 1.0000 .9950 .9800

— You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://github.com/arfc/sneks4me/issues/17?email_source=notifications&email_token=AE3653E7GRVUUZECKR6OVX3QJ77T3A5CNFSM4IXGJ4W2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD62VIOA#issuecomment-531977272, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AE3653BHTR3RFIIT4GJSZRDQJ77T3ANCNFSM4IXGJ4WQ.

anshchaube commented 5 years ago

Ah!

filt amp  .0000  .0000  .0000  .0000  .0050  .0200

So this shows it's only touching the last two modes? So that I know when I am writing this up, what "modes" are they? Where can I read more about the theory of the filters used by Nek?

And finally, does this mean the filter is working and this issue can be closed?

dshaver-ANL commented 5 years ago

The modes are the polynomial approximation orders. There will be 'lx1' numbers on that line corresponding to the amplitude of the filter. The filter acts by interpolating the solution to each order, then rebuilding it and weighting each mode by (1-amp). By damping the highest order modes, corresponding to the shortest wavelengths, it effectively removes energy from the flow, mimicking a viscous effect and stabilizing the solution.

For a laminar flow, the solution should be well represented on the lower orders and the filter doesn't do anything. However, if the flow is transitional, it will add diffusivity and can prevent it from transitioning to turbulence.

-------- Original message -------- From: Ansh Chaube notifications@github.com Date: 9/16/19 5:11 PM (GMT-06:00) To: arfc/sneks4me sneks4me@noreply.github.com Cc: "Shaver, Dillon R." dshaver@anl.gov, Comment comment@noreply.github.com Subject: Re: [arfc/sneks4me] Add filter to simulations (#17)

Ah!

filt amp .0000 .0000 .0000 .0000 .0050 .0200

So this shows it's only touching the last two modes? So that I know when I am writing this up, what "modes" are they? Where can I read more about the theory of the filters used by Nek?

And finally, does this mean the filter is working and this issue can be closed?

— You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://github.com/arfc/sneks4me/issues/17?email_source=notifications&email_token=AE3653EYYSKQ454CZLICWSTQKAACVA5CNFSM4IXGJ4W2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD62VQ5Q#issuecomment-531978358, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AE3653AQ7W3J4D5O7I3QWULQKAACVANCNFSM4IXGJ4WQ.

anshchaube commented 5 years ago

Thanks! So, we have some turbulence in the plenum. Is the filter going to take us away from the physical solution by increasing dissipation and reducing vortices, thereby giving us the "wrong" answer?

anshchaube commented 5 years ago

Future simulations won't have the filter. See discussion below:

katyhuff commented 5 years ago

Why not?

anshchaube commented 5 years ago

@dshaver-ANL's comment suggests that it's not particularly useful for our case with Re=1000, and it may lead us away from the correct physical answer by adding an artificial diffusivity that unphysically mitigates the very vortices (and any consequent recirculation zones) in the plenum that we're investigating.

dshaver-ANL commented 5 years ago

That's not exactly what I meant. Yes, the filter does add some artificial diffusivity, but it is quite small, and usually only stabilizes the simulation. I would suggest keeping it on.

anshchaube commented 5 years ago

Okay, future simulations will have the filter.