September 2011 - Nek5000-users - mailman3.cels.anl.gov

Cylindrical mesh
by nek5000-users＠lists.mcs.anl.gov 15 Sep '11

15 Sep '11

Hi Nek's, I've just started to look at the cylindrical meshing capabilities of genbox, and I have one question regarding the boundary conditions. Here is the exemple in Nek Primer just for recall: 2 spatial dimension 1 number of fields # # # Y cYlinder 3 -24 1 nelr,nel_theta,nelz .5 .3 x0,y0 - center of cylinder ccbb descriptors .5 .55 .7 .8 r0 r1 .... r_nelr 0 1 1 v ,W ,E ,E , I am not sure though I correctly understood where are applied the boundary conditions. Indeed we now have five "physical" boundaries (the four edges of the box + the cylinder itself), however we specified only four conditions. Regards, -- Jean-Christophe

1 2

Selective frequency damping
by nek5000-users＠lists.mcs.anl.gov 15 Sep '11

15 Sep '11

Hi Nek's, I've been trying to implement the selective frequency damping (Akervik E., Brandt L., Henningson D.S., Hoepffner J., Marxen O., Schlatter P. 2006. *Steady solutions of the Navier-Stokes equations by selective frequency damping.*Physics of fluids *18*) which enables to compute stationnary solution. Here is what I have written in my .usr file: In userchk: common /vxold/ vxold(size(vx,1),size(vx,2),size(vx,3),size(vx,4)) common /vyold/ vyold(size(vy,1),size(vy,2),size(vy,3),size(vy,4)) common /vzold/ vzold(size(vz,1),size(vz,2),size(vz,3),size(vz,4)) common /qxold/ qxold(size(vx,1),size(vx,2),size(vx,3),size(vx,4)) common /qyold/ qyold(size(vy,1),size(vy,2),size(vy,3),size(vy,4)) common /qzold/ qzold(size(vz,1),size(vz,2),size(vz,3),size(vz,4)) common /for_x/ f_x(size(vx,1),size(vx,2),size(vx,3),size(vx,4)) common /for_y/ f_y(size(vy,1),size(vy,2),size(vy,3),size(vy,4)) common /for_z/ f_z(size(vz,1),size(vz,2),size(vz,3),size(vz,4)) real chi, omega_c integer step_filtr chi = .25 omega_c = .1 step_filtr = 020000 if(istep.EQ.step_filtr.AND.nid.EQ.0) write(*,*) 'Filtrage' if(istep.EQ.0) then vxold = 0.0D+00 vyold = 0.0D+00 vzold = 0.0D+00 endif if(istep.LT.step_filtr) then qxold = 0.0D+00 qyold = 0.0D+00 qzold = 0.0D+00 f_x = 0.0D+00 f_y = 0.0D+00 f_z = 0.0D+00 call opcopy(vxold,vyold,vzold,vx,vy,vz) endif if(istep.GE.step_filtr) then qxold = qxold + dt*omega_c * (vxold - qxold) qyold = qyold + dt*omega_c * (vyold - qyold) qzold = qzold + dt*omega_c * (vzold - qzold) f_x = -chi * (vxold - qxold) f_y = -chi * (vyold - qyold) f_z = -chi * (vzold - qzold) call opcopy(vxxold,vyyold,vzzold,vxold,vyold,vzold) call opcopy(vxold,vyold,vzold,vx,vy,vz) endif and in userf: common /for_x/ f_x(size(vx,1),size(vx,2),size(vx,3),size(vx,4)) common /for_y/ f_y(size(vx,1),size(vx,2),size(vx,3),size(vx,4)) common /for_z/ f_z(size(vx,1),size(vx,2),size(vx,3),size(vx,4)) integer e,f,eg c e = gllel(eg) c Note: this is an acceleration term, NOT a force! c Thus, ffx will subsequently be multiplied by rho(x,t). ffx = f_x(ix,iy,iz,eg) ffy = f_y(ix,iy,iz,eg) ffz = f_z(ix,iy,iz,eg) I have tried to compute with such routines the base flow for a 2D lid-driven cavity flow at Re = 8500, however depending on the number of spectral elements I use, I sometimes get a stationnary solution that looks slightly as the one I'm looking for (but some differences near the lid) and sometimes I get one which is actually not stationnary. I applied the same algorithm for a finite-difference code and encountered no problem, so I was wondering then if I missed something specific to spectral elements or Nek 5000 (such as a mass matrix multplication for instance) or to F77 (which I'm not at ease with, see my definition of the common for instance)? Sincerely yours, -- Jean-Christophe

1 2

Generating a 2D circular mesh
by nek5000-users＠lists.mcs.anl.gov 13 Sep '11

13 Sep '11

Hi Neks, I'm having a little trouble with my first foray into circular geometries. I tried to construct a 2D annular mesh using genbox with the following .box file: ----------------------------------------------------------------------- an.rea 2 spatial dimension 2 number of fields Y -4 -12 nelr,nel_theta 0 0 x0 y0 ccbb 1 2 1 0 1 1 SYM,SYM, V bc's ! NB: 3 characters each ! f ,f , T bc's ! You must have 2 spaces!! ----------------------------------------------------------------------- genbox seems to work properly, with the exception of the fact that nothing after the boundary data gets copied over from the old .rea file. Nonethess, I copy that chunk over manually and run nek5000, which exits with the following error: WARNINGa: Detected non-right-handed element. Number 1 C1-4: 0.6667E+00 0.8333E+00 -0.2083E+00 -0.4167E-01 Any guidance would be much appreciated. Best, David

1 4

set obj for internal fluid surface
by nek5000-users＠lists.mcs.anl.gov 12 Sep '11

12 Sep '11

All, I'm working with the calc_torque routine, and I want to calculate the torque/drag along a surface within the fluid domain (not on the boundary). Is there a straight-forward way to "set_obj" for such a surface? It's straightforward for external fluid boundaries, as you can just search on certain boundary conditions and grab those element faces. It's not so clear for me how to do the same for some internal surface. Thanks. --Mike

1 3

genmap error: "infinite loop"
by nek5000-users＠lists.mcs.anl.gov 12 Sep '11

12 Sep '11

Dear Nek devs, I get the following error when using genmap to generate the .map file from a mesh involving a large number of elements, in this case 157464 elements. infinite loop not connected 1 2 1 0 I have attached the full log file from genmap along with the .box file used to generate the mesh. How does one fix this error? Thanks, Mani

1 2

possibly starting to use nekton, questions
by nek5000-users＠lists.mcs.anl.gov 08 Sep '11

08 Sep '11

Hi all, I'm considering to start using Nekton. I would like to simulate a spatially evolving 3D boundary layer with a spanwise array of cylinders poking out of the flat wall. The cylinders would also move up and down in the wall normal direction, generating a time-varying disturbance. My goal is to delay bypass transition by dampening this disturbance via feedback control of localized body forces near the wall. Does Nekton currently have these aspects implemented, namely: 1. spatially evolving boundary layer at transitional Re (Re based on displacement thickness of at least ~1000) 2. cylinders independently poking up and down with prescribed height as function of time 3. feedback control with multiple inputs and outputs 4. multiple arbitrary body forces, with independent magnitudes given by the feedback controller or as a function of time 5. ability to calculate the perturbation velocity from laminar for feedback control 6. linear and nonlinear cases (for 1-5) If the control and/or body force parts are not implemented, I'm considering using Nekton to get the wake behind the cylinders and feeding this to a simpler code that can't simulate them directly but would have control and body force aspects. Thanks for your guidance! Brandt

1 1

Non-Newtonian flows: nek options
by nek5000-users＠lists.mcs.anl.gov 08 Sep '11

08 Sep '11

Hi Neks. I'm trying to get Nek to work for a non-Newtonian incompressible fluid. I'm considering a Herschel-Bulkley fluid, whose yield-stress behavior I model with a regularized representation (defined as udiff in subroutine uservp). In this approach, the viscosity gets HUGE (e.g. 1e7) as the strain rate goes to zero. The two main challenges that I face are (i) an excessive number of Helmholtz-solve iterations and (ii) torque-magnitude calculations that are several orders of magnitude greater than I was expecting. Below is my understanding of what I can and should do with nek. Please verify that this is all correct, and please clarify if I have any misunderstanding: 1) I must use the stress formulation: IFSTRS = T 2) The stress formulation only works with PN/PN-2, not PN-PN, e.g. ABORT: Stress formulation in Pn-Pn is not supported 3) The viscous term in the stress formulation is treated implicitly; there's no nu_star option for the stress formulation, where one can treat a portion of the viscous term explicitly. 4) The torque_calc routine should work with this variable viscosity setup. A few more notes: -- I found my velocity system to be difficult to solve, and I had to increase the maximum iterations for the solve: I increased NMXH from 100 to 1000 (or greater) in drive2.f -- Per Paul's advice, I've set 0.000000E-05 p021 DIVERGENCE <--- set these to zero 0.000000E-08 p022 HELMHOLTZ <--- set these to zero 0.00000 p023 NPSCAL 0.100000E-01 p024 TOLREL <--- this is the important one 0.100000E-01 p025 TOLABS Does anyone have any additional tips or insights that might be helpful in solving such a system? Thanks. --Mike

1 3

Explicit Viscosity for Pn-Pn
by nek5000-users＠lists.mcs.anl.gov 02 Sep '11

02 Sep '11

Howdy Neks, I had a question about using variable viscosity with the Pn-Pn formulation. I know I need to have ifuservp and ifexplvis set to TRUE. Looking through the source code, I see that how it works is the viscosity is split into an implicit part (nu_star) and explicit. My question is, how is nu_star set? Doing a grep on the source code, I don't see where it is set anywhere. So, is this something left up to the user? Thank you, -- Josh Camp "All that is necessary for the triumph of evil is that good men do nothing" -- Edmund Burke

1 2