tcew / nodal-dg

• 👋 Hi, I’m Tim Warburton (AKA @tcew)
• 👀 I develop and implement GPU accelerated, high-order finite element based PDE solvers
• 🌱 I’m always learning more numerical analysis, PDEs, algorithms, linear algebra, preconditioners, ...

nodal-dg's Issues

Question about "edge noise"

Hi Prof. Warburton!

I am a graduate student at Harvard University working towards my PhD in astrophysics. I have really enjoyed your nodal DG book, and I've been trying to implement the codes in it, using both the book and this repo. I've run into a bit of a snag, however, and I'm not sure what might cause this issue. I thought you might have some insight to share.

I've already checked for obvious bugs in my code, printing out matrices and such and comparing them to the output from yours (using Octave instead of MATLAB for now). So far, everything seems consistent. When I run my code on a simplified version of my target problem, I get a lot of noise at the cell edges, such as the plots I've chosen below. I'm showing both the 2-dimensional view and a slice at y = 0; the variable in both cases is the x-velocity in the Euler equations. You can see that the amplitude of the noise is quite small, but that is a bit misleading -- this particular frame is captured after an absurdly small timestep. In reality, I expect to take timesteps on the order of 5-6 orders of magnitude greater than this, and the noise only grows over time.

This simulation used just 2 cells, and the noise clearly follows the edge of the cell. There should be no shocks in this system, and the solution is expected to be quite smooth. I've pushed filtering to the maximum I can, and it helps a bit, but the noise always comes back through after just a couple of timesteps. I've been debugging for about a week and am really at a loss. Is this a feature inherent to DG-FEM? Is there a workaround? Or a debugging step I could try?

Thanks in advance for your time,
Ellen Price

LINK : fatal error LNK1104: 无法打开文件“D:\Code\nodal-dg-master\nudg++\trunk\Lib\NDGLibd.lib”

When I try to run C++ codes, it turns out like that. I noticed the 'NDGlibd' file is under 'D:\Code\nodal-dg-master\nudg++\trunk\VC8\NDGLib'. Have you guys changed that?

Tks.

Difference in Matlab and C++ Version, which is correct?

These two lines are different, but, as far as I understand, they should not be. Which is the correct version?

nodal-dg/Codes1.1/Codes1D/JacobiGQ.m

Line 8 in b6864ca

if (N==0) x(1)= -(alpha-beta)/(alpha+beta+2); w(1) = 2; return; end;

nodal-dg/nudg++/trunk/Src/Codes1D/JacobiGQ.cpp

Line 26 in b6864ca

if (0==N) { x(1)=(alpha-beta)/(alpha+beta+2.0); w(1)=2.0; return; }

The book (Page 447) includes a listing of the Matlab version, which is different from the Matlab version in this repository but the same as the C++ version.

Trouble making nudg++ on Linux

I was able to get the MATLAB code running, but I can't seem to build the c++ version (I'm guessing it's a c++ version of the MATLAB code?)

I'm trying to build nudg++ on openSUSE Tumbleweed, a rolling relase Linux distribution.

At first glance, I'm getting a bunch of ISO c++ warnings, maybe it has to do with a compiler version mismatch? I'm using gcc 6.2.1 - what compiler is required for this to build correctly?

Please see the output of make Maxwell2D attached here: make.txt

Licensing confusion

Hi! Thank you for sharing this code.

I have a question about its license. In the root of this repository is a LICENSE file containing the MIT license. However, in the Codes1.1 subdirectory, there is a README stating that the code is under a proprietary license.

As-is, I would interpret this to mean that the code in Codes1.1 is under that proprietary license, and that the code in Lebesgue2D and nudg++ (which perhaps you added later, and outside of your contract with the publisher?) is under the MIT license. Is that correct? Or was the code relicensed, and is it all under the MIT license now?

It would be good if you could make explicit who owns the copyright to what, and what license they offer it under.

how to run it