I have a question about the Cuba package for numerical integration (http://www.feynarts.de/cuba/). I've tried to download and install Cuba
tar xzvf Cuba-4.2.tar.gz
(also tried to statically link using MCFLAGS=-st after having read this answer: http://community.wolfram.com/groups/-/m/t/470248)
and I also tried to download and install the Vegas mathlink executable
gunzip Vegas-Linux.gz (or Vegas-Windows.exe.gz)
chmod 755 Vegas (or Vegas-Windows)
We have tried this on two different Ubuntu systems (with version 11.3) and with Windows (with version 11.1 using MinGW).
On both Ubuntu plateforms, we're managing to get Create link: when we run ./Vegas. But then we have to specify the path to Vegas in the notebook for it to find the executable because somehow the Mathematica Installer cannot load Vegas. Anyhow, Install["/path/Vegas"] seems to work. We then try to run code which used to run on Ubuntu Mathematica 11.1 with Vegas before I updated to Mathematica 11.3 and reinstalled Vegas. This code, for which we have previous good results, now gives either the wrong results, or seems to calculate some form of result but then writes failed on all the integrations.
On Windows we get The code execution cannot proceed because cygwin1.dll AND ml32i4.dll was not found. Reinstalling the program may fix this problem. however, I've added C:\cygwin64\bin to my system path environment variable and restarted Mathematica and am still getting the same result on Windows.
The code execution cannot proceed because cygwin1.dll AND ml32i4.dll was not found. Reinstalling the program may fix this problem.
My research cluster team also tried to install Vegas centrally on the cluster, but it outputs the same thing as it does when I run it in Ubuntu: Install["Vegas"] works fine, but then it just writes failed everywhere.
I then tried to use the NIntegrate mathematica functions, but they don't give anywhere near the same precision, despite trying to find out which would be a good method to use (https://mathematica.stackexchange.com/questions/175480/nintegrate-for-extremely-oscillatory-function/175489?noredirect=1#comment460575_175489).
The Cuba aspect is probably a question for the FeynArts developer though perhaps there are other readers who might have familiarity. As for tuning NIntegrate, it is really difficult to speculate without a (simple as possible) example.
I see, thank you. I'll contact him. The simple example is on the stack-exchange link. Kind regards
Sometimes a complex build system like the one used here is more of a hindrance than help. On a Mac, it doesn't find MathLink automatically (because on a Mac there is no foolproof way to detect it). In general, this would be a pretty easy thing to fix. But in this case, I'd first need to understand how the build system (autotools) works. That would probably take a couple of hours. So instead I just give up.