February 2018 - Nek5000-users - mailman3.cels.anl.gov

3D v/s. 2D computation
by nek5000-users＠lists.mcs.anl.gov 14 Mar '18

14 Mar '18

Dear All, I am carrying out 3D simulations using NEK5000 in a hexahedral domain. Symmetry boundary conditions have been imposed on the side wall and stress free conditions are prescribed on the top and bottom boundary. At the inlet and outlet, Dirichlet and stress free conditions are used, respectively. I observe that the flow on the symmetry plane is different from that obtained via 2D simulation. Is this a numerical artifact (increased diffusion because of 3D mesh) or is it physical? If I increase the resolution of the mesh in third dimension, do I expect to get the 2D simulation results on the symmetry plane? Any input will be appreciated. Many thanks, NN.

1 2

Regarding base velocity (Mesh deformation)
by nek5000-users＠lists.mcs.anl.gov 28 Feb '18

28 Feb '18

Hi Neks, I had sent you a detailed problem regarding base velocity (mesh motion and mesh deformation) in the morning unfortunately i have not get any replay yet . I think you guys are busy otherwise you guys are very resposive. I appreciate that. Even though I investigated my problem and I think i found where actually the problem is even though I still need some clarifications in that. The problem were in the boundary conditions since there are two type of B.Cs Homogenous (zero on all non- interior boundary faces ) and Inhomogenous( 1 at moving boundary faces). When i was trying on 2D there were no problem with B.Cs. But when i i tried my 3D mesh where a new B.C has come which is called SYMMETRY b.c. I imposed the SYM conditions on the horizontal walls of the fluid domain (not on the object) and also the exterior to the fluid domain on 3rd dimension.(3D means i put a thinkess as size of one element, which makes my foil case into 3D) As a result, for one element near the moving body, I have a m.v condition, SYM conditon on the exterior of the fluid domain in 3rd direction (lets say Z direction, so SYM on both sides) and E on the interior of the fluid domain. So the condition I gave is given below : m1 = -1. y0 = 0.0 nface = 2*ndim do e=1,nelv do f=1,nface c Set Dirichlet for mesh velocity on all boundaries if (cbc(f,e,1).ne.'E ') call facev(msk,e,f,y0,nx1,ny1,nz1) c Inhomogeneous Dirichlet on foil only if (cbc(f,e,1).eq.'mv '.or. cbc(f,e,1).eq.'SYM ') then if((ym1(1,1,f,e).lt.6.0).and.(ym1(1,1,f,e).gt.-6.0)) then !! where my domain +6 to -6 in Y direction and 0 to 0.1 in Z direction and -8 to 8 in X direction call facev(tmp,e,f,m1,nx1,ny1,nz1) endif endif enddo enddo Now I get a better basev (the values ranges between 0 and 1.2 before it was less than zero). But still I am getting a value which is slightly greater than 1.0 which is not good. So could you please tell me what should I do in this case, to have a nice smoothing velocity to all the mesh. Thanks with Regards Sijo GEORGE

1 0

Base velocity problem
by nek5000-users＠lists.mcs.anl.gov 28 Feb '18

28 Feb '18

Hi Neks, I am bit stuck on this occasion for past 2 days, the code below shows a part of the example called ocyl2.usr (from subroutine my_base_meshv(basev,h1,h2,rhs,msk,tmp) ). This code is to generate a base velocity (where basev=1 at wall of object and 0 at the boundary of the domain). I assume that I have a blending coefficient h1.(using Paul's method or volume of elements method). This subroutine working perfectly for the 2D cases. That is I have the base velocity ranges between 0 and 1. But when I tried it on an aero dynamic foil case(its a 3D case where just 1 element in Y direction), the value of base velocity does not even exceed 0. Almsot all values are like -0.0021564... Could you please tell me is there anything that I have to change in this subroutine to make it working. Tha value of m1 and z0 are just explicity giving the base velocity to the elements right? for example on the W or M.V conditions basev becomes 1 and on boundaries of the domain it becomes 0.0 m1 = -1. z0 = 0.0 nface = 2*ndim do e=1,nelv do f=1,nface c Set Dirichlet for mesh velocity on all boundaries if (cbc(f,e,1).ne.'E ') call facev(msk,e,f,z0,nx1,ny1,nz1) c Inhomogeneous Dirichlet on cylinder only if (cbc(f,e,1).eq.'mv ') then call fcaver(xavg,xm1,e,f) if (xavg.gt.-3.0) call facev(tmp,e,f,m1,nx1,ny1,nz1) endif enddo enddo call axhelm (rhs,tmp,h1,h2,1,1) tol = 1.e-4 imsh = 1 ifield = 1 call hmholtz('mshv',basev,rhs,h1,h2,msk,vmult,imsh,tol,200,1) call sub2(basev,tmp,n) c call outpost(basev,basev,basev,basev,basev,' ') c stop Regards Sijo

1 0

periodic boundary conditions in mshconvert.py
by nek5000-users＠lists.mcs.anl.gov 28 Feb '18

28 Feb '18

Hello Neks： I‘m trying to use the mshconvert.py to get a periodic bc. I put in convert('test.msh',periodic_dx={(3,4):[0.0,3.0]},bcs={5:'O',6:'v'}). But there is always a mistake: Traceback (most recent call last): File "<ipython-input-93-7345a535cf67>", line 1, in <module> convert('test.msh',periodic_dx={(3,4):[0.0,3.0]},bcs={5:'O',6:'v'}) File "C:/Users/Administrator/Desktop/test/mshconvert.py", line 1477, in convert create_boundary_section(bcs, temperature, passive_scalars, mesh_format) File "C:/Users/Administrator/Desktop/test/mshconvert.py", line 668, in create_boundary_section periodic_cell_face_map[(c, local_face)] KeyError: (1, 2) The .msh file is the most simple one： (0 "GAMBIT to Fluent File") (0 "Dimension:") (2 2) (10 (0 1 8 1 2)) (10 (1 1 8 1 2)( 1.0000000000e+000 0.0000000000e+000 0.0000000000e+000 0.0000000000e+000 0.0000000000e+000 3.0000000000e+000 1.0000000000e+000 3.0000000000e+000 0.0000000000e+000 1.0000000000e+000 0.0000000000e+000 2.0000000000e+000 1.0000000000e+000 1.0000000000e+000 1.0000000000e+000 2.0000000000e+000 )) (0 "Faces:") (13(0 1 a 0)) (13(3 1 1 4 0)( 2 1 2 0 1 )) (13(4 2 2 e 0)( 2 3 4 0 3 )) (13(5 3 5 25 0)( 2 1 7 1 0 2 7 8 2 0 2 8 4 3 0 )) (13(6 6 8 3 0)( 2 2 5 0 1 2 5 6 0 2 2 6 3 0 3 )) (13(8 9 a 2 0)( 2 7 5 1 2 2 8 6 2 3 )) (0 "Cells:") (12 (0 1 3 0)) (12 (2 1 3 1 3)) (0 "Zones:") (45 (2 fluid fluid)()) (45 (3 inlet-vent p2)()) (45 (4 fan p1)()) (45 (5 axis out)()) (45 (6 wall in)()) (45 (8 interior default-interior)()) Thank you! Lily

1 0

Re: [Nek5000-users] Tripping mechanism
by nek5000-users＠lists.mcs.anl.gov 27 Feb '18

27 Feb '18

Hi Philipp, Thank you for your response. It has been very helpful. Best, Juan Diego

1 0

blending coefficient in 3D
by nek5000-users＠lists.mcs.anl.gov 27 Feb '18

27 Feb '18

Hi, In order to make a smooth blending coefficient for mesh motion, I made a code below, it works fine for a 2D case in my computer but it does not work as expected for a 3D case (I run 3D case in server with 1024 pocessors). Could you please tell me is there anything wrong? (The error is the volume that i am getting is some wrong values i think) and is it right way to make Dirichlet B.C on boundaries also calling hmholtz function to make a blending velocity gradient? Because I took that part from the NEK example which is 2D. So do i have to change anything to make working in 3D? subroutine my_base_meshv(basev,h1,h2,rhs,msk,tmp) include 'SIZE' include 'TOTAL' real basev(lx1,ly1,lz1,lelt),tmp(lx1,ly1,lz1,lelt) real h1(1),h2(1),rhs(1),msk(1) real m1,denom,volume(nx1*ny1*nz1*nelv) integer e,f n = nx1*ny1*nz1*nelv nxyz = nx1*ny1*nz1 do e=1,n volume(e) = vlsum(bm1(e,1,1,1),nxyz) enddo a=glmax(volume,n) b=glmin(volume,n) do i=1,n denom = 1/a h1(i) = ((1-(b/a))/(denom*volume(i))) enddo do i=1,n open (unit=10,file="h.dat") write (10,7020) h1(i),volume(i),a,b 7020 format(7f20.8) enddo m1 = -1. z0 = 0. nface = 2*ndim do e=1,nelv do f=1,nface c Set Dirichlet for mesh velocity on all boundaries if (cbc(f,e,1).ne.'E ') call facev(msk,e,f,z0,nx1,ny1,nz1) c Inhomogeneous Dirichlet on cylinder only if (cbc(f,e,1).eq.'mv ') then call fcaver(xavg,xm1,e,f) if (xavg.gt.-1.2) call facev(tmp,e,f,m1,nx1,ny1,nz1) endif enddo enddo call axhelm (rhs,tmp,h1,h2,1,1) tol = 1.e-4 imsh = 1 ifield = 1 call hmholtz('mshv',basev,rhs,h1,h2,msk,vmult,imsh,tol,200,1) call sub2(basev,tmp,n) c call outpost(basev,basev,basev,basev,basev,' ') c stop return end

1 0

Volume of an element
by nek5000-users＠lists.mcs.anl.gov 27 Feb '18

27 Feb '18

Hi Neks, How to calculate volume of an element in a 3D mesh? If i write n = nx1*ny1*nz1*nelv do i=1,n v(i) = volume(1,1,1,i) enddo Can i store the volume of each element into a varibale called v? What happens if i use bm1(1,1,1,i)? Instead of volume(1,1,1,i) Thanks Sijo

1 5

turbChannel: Definition of CFL and write/print files
by nek5000-users＠lists.mcs.anl.gov 27 Feb '18

27 Feb '18

Dear Nek users, I have a couple quick and preliminary questions. Could you please tell me where could I find the definition of CFL; more specifically, the file/subroutine where CFL is related to the time step, etc. The other question is that which file shall I go to add/modify lines to write out some field information in a file (like .dat files or so) or print on sreen? Thank you a lot for the help! Regards, Emily

1 4

CPU, Memory and MPI
by nek5000-users＠lists.mcs.anl.gov 27 Feb '18

27 Feb '18

Hello Neks,I want to know specific information about optimal usage of computing sources.I would like to show you my own environment for running a case.CPU: 3.3 GHz, 4 cores, 8 processors (Intel i7-6700HQ)RAM: 2133 MHz, 8 GB x 1 (DDR4-2133)Case: 1500 elements, N = 61.Based on Nek5000 document, total memory requirement is1500 (points) X 6^3 X 400 (words/point) X 8 (bytes/word) X (1 GB/1024^3 bytes) = 0.97 GBSo, is this calculation correct? so is 8 GB of memory enough?2.In the Nek5000 document, they also mention about "memory per core."Could you tell me the meaning of memory per core?I thought that, in my case, I assigned 2GB / core (If I use MPI with 4 cores. because I also have 8 GB of memory).If there are manual way to assign, how can I assign memory to cores?3.Actually, I also tried to use nekbmpi to use MPI process in my PC.But when I type like belownekbmpi <case name> 8(or 4)my CPU usage of 8 processors suddenly goes up but, there are no progress in simulation.(log file is remained as 0 bytes. when I check 'Task manager' of Windows 10,Nek5000 processes appear but they take just few MB of memory, but usage of CPU peaked up to 100%)I also tried to use nek5000 directly as below./nek5000 -np 8(or 4) > case.out &But I think, after typing this syntax, PC shows just moderate performance (same performance with ./nek5000 ).(log file also shows Number of processors: 1)I already installed MPI in my ubuntu environment in Windows 10.Also I compiled make file with mpif77 and mpicc.How can I use MPI in my PC?4.Finally,If you have idea to estimate time to run your case approximately(e.g. cycle per second), please give me.And I will also appreciate if you show how to calculate the appropriate number of processors to run a case.(no matter if it is based on your environment or my environment) Best Regards,Hyunduk.

1 0

Re: [Nek5000-users] Tripping mechanism
by nek5000-users＠lists.mcs.anl.gov 26 Feb '18

26 Feb '18

Dear Philipp, Thank you for your response. So if I understand correctly, first you generate a random signal as a function of Z and then use a low-pass filter to obtain h^i(z). Thinking about how to implement this in Nek5000, I guess one could impose random numbers at each node (collocation point) and then apply some kind of filter. Is this correct? Also, could you please clarify what you mean by: make sure that you have the same random numbers for the whole spanwise extent? Since h(z) is a function of z, I'm not sure I understand this statement. Thank you very much for you help. Juan Diego Colmenares PhD Candidate Department of Mechanical Engineering University of New Mexico

1 1