One-sided fft; Current issues in consistency between one-sided and two-sided fft

[katgonzalez17]

Looks like the Nusselt number does change more significantly at larger grid sizes. I re-looked at my test script and noticed inconsistencies at 10^15, that (maybe?) these errors are amplified over multiple time steps, particularly for array sizes. Not sure what's going on, and am open to any and all suggestions for what to check out!

Attached is a photo of the Nusselt number over different time steps with a 1728 by 1377 grid.

[dsondak]

A few questions / suggestions:

• It sounds like you checked your test script and found some discrepancies. What do you mean by 10^15? Is this the number of points?
• Which test script is this? Is it the script that is simulating the $N_y$ loop? Or is it your original one?
• I'd break this down a bit further. Compare results after the very first loop. Compare intermediate results within the loop. You can still calculate $Nu$ as a representative scalar quantity. It won't have any meaning, but that's okay. You can also make sure the 1-sided and 2-sided ffts are the giving the same results within the loop.

It's possible that there is a subtle bug. It is also possible that some errors are accumulating naturally. Let's try to isolate the culprit.

[katgonzalez17]

• Sorry, I meant 10^-15, so inconsistencies after 15 places behind the decimal point. The number of points that I used was 2^15, and I checked the last few array entries comparing before forward + backward transformation and after.
• This is my original one where I allocate an array, populate it with sin values, and then do a forward and backward transformation.
• I can write out the Ny loop first and look at Nusselt number between one and two sided transformations!

[dsondak]

Discrepancies of $10^{-15}$ are okay. But try to track down why they are accumulating.