March 2017 - Nek5000-users - mailman3.cels.anl.gov

turbChannel example
by nek5000-users＠lists.mcs.anl.gov 24 Mar '17

24 Mar '17

Hello, I am trying to reproduce some of the turbChannel results that were published with Nek5000 in the paper that can be found in the following link: https://arxiv.org/pdf/math/0205264.pdf In this paper, two turbulent models are tested: the gradient model and the static Smagorinsky model. The paper also shows results from a coarse DNS for comparison. The implementation of the static Smagorinsky model is easily done when starting from the turbChannel example but I am not sure how to implement the gradient model since it involves a tensor product \grad u \dot \grad u. Can anyone point me to an example that has a similar implementation or explain how to implement the gradient model in Nek5000? Is it correct to assume that the current setup of the turbChannel example in Nek5000 should allow to reproduce the coarse DNS results presented in the paper? What is the relationship between Re_tau and the Reynolds number (param(2)) set in the rea file? Any help will be appreciated. Thanks, Marco

1 3

Copy velocity slice from one boundary to another
by nek5000-users＠lists.mcs.anl.gov 24 Mar '17

24 Mar '17

Hi, Neks! I have two pipes in my simulation. They are placed near each other on one axis with empty space between. One of them with periodic boundary conditions on z direction, another one - inflow -> outflow. How can I copy velocity slice from one of boundary on first pipe with periodic bc to the inflow boundary on second pipe? Regards, Vatslav.

1 1

Dimensionfull passive scalar
by nek5000-users＠lists.mcs.anl.gov 24 Mar '17

24 Mar '17

Hi! I am simulating a flow where thermodynamic processes takes place. Since these thermodynamic processes are rather complicated I decided that developing the set up is best done in dimensionfull variables. Here I ran into unexpected problems since it seems as if the parameters in the .rea file do not work as I thought they would. The temperature solution works well. Putting in the "rhocp" and "conduct" of my fluid into "p007" and "p008" gives the results that I expect. However, I also have two passive scalars. For them I am not sure what the meaning of "rhocp" actually is. >From eqn (1.4) of Ref. [*] I would assume that setting up a passive scalar \phi with diffusivity a = 2.14E-5 m^2/s evolving as D \phi /D t = a Laplacian(\phi) [D/Dt = total derivative] means using "rhocp = 1" and "a = 2.14E-5". However, this seems to not be the case, since under these assumptions the corresponding dimensionless set-up gives totally different results. If my understanding was correct it would only be the ratio of "rhocp" and "conduct" that matters fo a scalar, but I have found that the individual values matter. This leads me to the questions: * What is the meaning of "rhocp" and "conduct" for a scalar? * In which dimensions "rhocp" and "conduct" for a scalar? * Why does the individual values of "rhocp" and "conduct" matter, when they should no according to eqn (1.4) of Ref. [*]? Best, Johan [*]: User Documentation: http://nek5000.github.io/NekDoc/Nek_users.pdf

1 0

Run nek5000 in BlueGene supercalculator (IBM compiler)
by nek5000-users＠lists.mcs.anl.gov 24 Mar '17

24 Mar '17

Hi, I have trouble running Nek in BlueGene in IDRIS, French national supercalculator: the compilation show no error, but the output give NaN before the time loop start. I looked in the internet and I found 1 person ask almost the same question 4 years ago (In my case the compilation was fine), but I can't find the answear. Are there anyone ran Nek5000 in IDRIS can share their experience? Thank you guy,

1 5

Compile optimization flags and GDB
by nek5000-users＠lists.mcs.anl.gov 23 Mar '17

23 Mar '17

Dear all, I found two problems in Nek500 while i try to make some changes to the source codes. 1. When using GDB, I cannot break on some specific lines, it prompts message like missing line number information. I’ve enabled the debuging Flag “G=-g” in bin/makenek 2. I want to use optimization level O0 instead of the default O2 to make gdb debugging easier. So i changed line number 122 in core/makefile.template from "L2=$(G) -O2” to "L2=$(G) -O0”. While compiling with -O2 the result is correct , there is segment fault with -O0. Developing Environment: CENTOS 6.5 GCC-4.9.2 Nek example: eddy_uv Thanks Steven

1 0

What's change have been made between version 1.0rc1 / SVN r1053 and the current version
by nek5000-users＠lists.mcs.anl.gov 23 Mar '17

23 Mar '17

Hello, I continue the work of another one, he use an old version of Nek (version 1.0rc1 / SVN r1053), and everything work fine. My problem is when I change the version to the newest one from the official website, the solution diverge after 60 time step. I can't figure out why? So if I want to work with the new version, what should be changed? Thank you for your attention,

1 0

Re: [Nek5000-users] Project a solution onto a polynomial space
by nek5000-users＠lists.mcs.anl.gov 23 Mar '17

23 Mar '17

Dear Paul, Thank you for answering so quickly. Yes, the polynomial spaces I'm using are the standard Nek5000 basis functions. To understand the problem a little better: I am conducting DNS of a shear flow and I need to compute higher-order velocity moments. I've implemented some subroutines based on the "avg2" and "avg3" subroutines from Nek, but for higher order moments, i.e. <U^3>, <U^2 V>, etc. The problem is that I need to de-alias the non-linear quantities, i.e. U^3, before doing the averaging. Since the flow is planar, I also take the spanwise average using the "z_average" subroutine. I checked out the "convect.f" file to see how dealiasing is applied by mapping onto the finer grid (lxd), but I'm not quite sure how to project these results back onto the original grid. Thank you, Juan Diego

1 5

Project a solution onto a polynomial space
by nek5000-users＠lists.mcs.anl.gov 22 Mar '17

22 Mar '17

Hello, I need to compute some non-linear quantities as part of a post-processing procedure. Because they are non-linear, I will need to do some kind of dealiasing. So far, I know how to map the velocity field onto a finer grid (higher polynomial order) to compute the non-linear quantities. However, I don't know how to project the resulting field onto the original polynomial space. Note, I want to find the "projection" onto the original space, but not the "interpolation". Thanks, Juan Diego

1 1

tubChannel - neknek script
by nek5000-users＠lists.mcs.anl.gov 21 Mar '17

21 Mar '17

Hello, I am running the neknek script with the turbChannel example. My setup consists of two channels: the first one has periodic boundary conditions, and the second one has inlet/outlet boundary conditions. The numerical solution of first channel is used as an inlet condition for the second channel. The numerical solution seems to behave as expected. My next step is to simulate a short channel with periodic boundary conditions, and a longer one with inlet/outlet boundary conditions. The number of elements in the streamwise direction (x-axis) now differs. This is causing some problems since the number of elements in the x,y and z directions are set in the SIZE file that is unique to the sessions named inside and outside. Note that I need the number of elements in each direction to compute the y+ and u+ values. In the turbChannel example, the number of elements in the x, y and z directions are defined in the usr file (nelx, nely and nelz). This approach could work in my case, I think, assuming that I include a if statement on the session name (inside vs outside). However, I am after something more generic that can be applied to more complex geometries where the number of elements in each direction is not known or difficult to know. Is there a way with Nek5000 to get the number of elements in each direction in a automated way? I have another question that is a little bit off topic: when initializing an array in the subroutines defined in the usr file, should I use nelv or lelv as the number of elements per processors? In the turbChannel example, both are used and it confuses me. nelv is the maximum number of elements per processor, whereas lelv is the actual number of elements in a processor, which leads me to think lelv should be used. Is this correct? Thanks, Marco

1 3

turbChannel example
by nek5000-users＠lists.mcs.anl.gov 21 Mar '17

21 Mar '17

Hello, I am trying to reproduce some of the turbChannel results that were published with Nek5000 in the attached paper. In this paper, two turbulent models are tested: the gradient model and the static Smagorinsky model. The paper also shows results from a coarse DNS for comparison. The implementation of the static Smagorinsky model is easily done when starting from the turbChannel example but I am not sure how to implement the gradient model since it involves a tensor product \grad u \dot \grad u. Can anyone point me to an example that has a similar implementation or explain how to implement the gradient model in Nek5000? Is it correct to assume that the current setup of the turbChannel example in Nek5000 should allow to reproduce the coarse DNS results presented in the attached paper? What is the relationship between Re_tau and the Reynolds number (param(2)) set in the rea file? Any help will be appreciated. Thanks, Marco

1 0