Question about Poisson Accuracy Test

[bewagner1]

It is my understanding that testPoissonAccuracy.m is verifying that the order of accuracy that the laplacian operator achieves is >= (k - 0.5) for k = 2, 4, 6, 8.

However, if you run lines 4 to 51 in their own file and replace lines 48 and 49 with assert(order(i) >= k - 0.5) it fails for k = 8. Why does it still pass?

My gut is telling me that it is something to do with the errors being so low (on the order of 1e-12)

Semi-related to #319, in my C++ implementation the test fails (with similar error values)

[jbrzensk]

The errors are essentially near machine precision once you take the log2 of those numbers. You can make it appear earlier by setting the number of points tested to something bigger, like try [40, 80] and it happens for order 6 and 8. The errors are essentially 0. You could try it with the new addBCScalar functions to see if the error remains. I would guess it stays. Hard to get lower than 10e-13 in MATLAB.

8th order is also a temperamental scheme, only useful if the solutions are smooth, wave-dominated, uniform smooth simple grids. Probably why it is missing from the gradient.cpp file.

[bewagner1]

That's what I figured. I don't think addScalarBC will change it, too. I was mostly wondering how MATLAB considers it as passing even though it doesn't actually pass

[jbrzensk]

I see. Those tests are never run, AFAIK. If you run the test file by hand, it fails.

Running testPoissonAccuracy

================================================================================
Verification failed in testPoissonAccuracy/testforEnergy.
    ----------------
    Test Diagnostic:
    ----------------
    Test failed for k = 8
    ---------------------
    Framework Diagnostic:
    ---------------------
    verifyGreaterThan failed.
    --> The value must be greater than the minimum value.
    
    Actual Value:
      -1.576120623969349
    Minimum Value (Exclusive):
       7.500000000000000
    ------------------
    Stack Information:
    ------------------
    In /home/jbrzensk/github/MOLE/mole/tests/matlab_octave/testPoissonAccuracy.m (testPoissonAccuracy.testforEnergy) at 48
================================================================================
.
Done testPoissonAccuracy
__________

Failure Summary:

     Name                               Failed  Incomplete  Reason(s)
    ================================================================================
     testPoissonAccuracy/testforEnergy    X                 Failed by verification.

[bewagner1]

It doesn't get run when doing make run_matlab_octave_tests?

[jbrzensk]

@bewagner1 I checked it out further, and it seems the CMakeList.txt file runs the tests, but only checks for output from MATLAB, not if the test passes or fails. I tried playing around with it but I am not super familiar with the CMake testing mixed with MATLAB testing.

@cpaolini Can you look into this? I will try more later, but this is lower on my plate.