Re: [Nektar-users] FW: Restarting a 3D1D case with a 2D converged case
Hi Andrea, Thanks for your suggestions. I was not aware that we could use the modal energy filter as a diagnostic tool. I will try it. I also was not aware that the SVV cutoff frequencies in different directions could be set separately. I have just been applying a single global frequency so far. Could you describe how this is done in the parameters section? Sincerely, On Wed, Jan 30, 2019 at 3:40 AM Amitvikram Dutta < amitvikram.dutta@uwaterloo.ca> wrote:
------------------------------ *From:* nektar-users-bounces@imperial.ac.ukOn Behalf OfCassinelli, Andrea *Sent:* January 29, 2019 1:29:32 PM (UTC-05:00) Eastern Time (US & Canada) *To:* nektar-users
*Subject:* Re: [Nektar-users] Restarting a 3D1D case with a 2D converged case
Hi Amitvikram,
What you described seems to be the correct procedure (initialise the 3DH1D field with well-established 2D field). What kind of initial conditions are you using for the spanwise direction? I would suggest using low-magnitude noise to kick the Fourier modes (i.e. <E VAR="w" VALUE="0.0001*awgn(1.0)" />).
I would also suggest to turn on the modal filter to see if the simulation blows up because energy accumulates in the high frequency modes. Are you using SVV for the homogeneous direction or just in the spectral/hp plane? You can try to lower the cutoff frequency for the spanwise direction and see if you still observe this behaviour.
Hope this helps.
Kind regards,
Andrea
On 29 Jan 2019, at 18:10, Amitvikram Dutta <amitvdutta23@gmail.com> wrote:
Hi all,
I'm studying 3D devlopement of jets and am using a 3D1D setup with Fourier coefficients in the spanwise directions. Instead of starting a 3D1D case from scratch (the mesh contains 32 million DOF in 3D) I was thinking of trying out a few strategies in order to speed up convergence.
I have tried out the steady-state module but it seems to take a rather long time to converge as well.
I was thinking therefore, whether it would be possible to start the case in 2D, and once it reaches convergence, to switch over to a 3D 1D mode using a 2D .chk file as a restart point.
I have tried out a few cases with the procedure described above, but cases always seem to blow up after restarting in the 3D 1D mode. I could explore the parameter space by reducing the time-step or number of expansion modes etc, but I'd like to know if there is a fundamental block to this kind of approach, or indeed if there are better approaches to speeding up convergence.
Sincerely, --
*Amitvikram Dutta*
Graduate Research Assistant
Fluid Mechanics Research Lab
Multi-Physics Interaction Lab
University of Waterloo _______________________________________________ Nektar-users mailing list Nektar-users@imperial.ac.uk https://mailman.ic.ac.uk/mailman/listinfo/nektar-users
--
*Amitvikram Dutta* Graduate Research Assistant Fluid Mechanics Research Lab Multi-Physics Interaction Lab University of Waterloo
Dear Amitvikram, You can indeed set SVV separately for the spectral/hp plane and the Fourier expansion. For instance, If you set: <I PROPERTY="SpectralVanishingViscositySpectralHP" VALUE="DGKernel" /> <I PROPERTY="SpectralVanishingViscosityHomo1D" VALUE="True" /> You can then specify independently the coefficients required for the SVV kernel that you select. In this case, employing the DG Kernel for the spectral/hp plane and the exponential kernel for the spanwise direction, you can then set their parameters independently: <P> SVVDiffCoeff = 1 </P> <P> SVVDiffCoeffHomo1D = 1 </P> <P> SVVCutoffRatioHomo1D = 0.5 </P> (For the DG kernel, no parameters are actually needed although you can modify the diffusion coefficient). For reference to the user guide, you can check section 11.3.1. Kind regards, Andrea On 30 Jan 2019, at 14:58, Amitvikram Dutta <amitvdutta23@gmail.com<mailto:amitvdutta23@gmail.com>> wrote: Hi Andrea, Thanks for your suggestions. I was not aware that we could use the modal energy filter as a diagnostic tool. I will try it. I also was not aware that the SVV cutoff frequencies in different directions could be set separately. I have just been applying a single global frequency so far. Could you describe how this is done in the parameters section? Sincerely, On Wed, Jan 30, 2019 at 3:40 AM Amitvikram Dutta <amitvikram.dutta@uwaterloo.ca<mailto:amitvikram.dutta@uwaterloo.ca>> wrote: ________________________________ From: nektar-users-bounces@imperial.ac.ukOn<mailto:nektar-users-bounces@imperial.ac.ukOn> Behalf OfCassinelli, Andrea Sent: January 29, 2019 1:29:32 PM (UTC-05:00) Eastern Time (US & Canada) To: nektar-users Subject: Re: [Nektar-users] Restarting a 3D1D case with a 2D converged case Hi Amitvikram, What you described seems to be the correct procedure (initialise the 3DH1D field with well-established 2D field). What kind of initial conditions are you using for the spanwise direction? I would suggest using low-magnitude noise to kick the Fourier modes (i.e. <E VAR="w" VALUE="0.0001*awgn(1.0)" />). I would also suggest to turn on the modal filter to see if the simulation blows up because energy accumulates in the high frequency modes. Are you using SVV for the homogeneous direction or just in the spectral/hp plane? You can try to lower the cutoff frequency for the spanwise direction and see if you still observe this behaviour. Hope this helps. Kind regards, Andrea On 29 Jan 2019, at 18:10, Amitvikram Dutta <amitvdutta23@gmail.com<mailto:amitvdutta23@gmail.com>> wrote: Hi all, I'm studying 3D devlopement of jets and am using a 3D1D setup with Fourier coefficients in the spanwise directions. Instead of starting a 3D1D case from scratch (the mesh contains 32 million DOF in 3D) I was thinking of trying out a few strategies in order to speed up convergence. I have tried out the steady-state module but it seems to take a rather long time to converge as well. I was thinking therefore, whether it would be possible to start the case in 2D, and once it reaches convergence, to switch over to a 3D 1D mode using a 2D .chk file as a restart point. I have tried out a few cases with the procedure described above, but cases always seem to blow up after restarting in the 3D 1D mode. I could explore the parameter space by reducing the time-step or number of expansion modes etc, but I'd like to know if there is a fundamental block to this kind of approach, or indeed if there are better approaches to speeding up convergence. Sincerely, -- Amitvikram Dutta Graduate Research Assistant Fluid Mechanics Research Lab Multi-Physics Interaction Lab University of Waterloo _______________________________________________ Nektar-users mailing list Nektar-users@imperial.ac.uk<mailto:Nektar-users@imperial.ac.uk> https://mailman.ic.ac.uk/mailman/listinfo/nektar-users -- Amitvikram Dutta Graduate Research Assistant Fluid Mechanics Research Lab Multi-Physics Interaction Lab University of Waterloo
participants (2)
-
Amitvikram Dutta
-
Cassinelli, Andrea