Replace Odeint with in-house Runge-Kutta 4th order integrator

[robertodr]

Description

This PR removes the dependency on Odeint by providing code for a Runge-Kutta 4th order integrator.
In addition, the code for handling the radial solutions and some of the internals of the SphericalDiffuse class have been significantly simplified.

How Has This Been Tested?

All unit tests but one pass. green_spherical_diffuse, which checks the value of the Green's function. However, the tests on boundary integral operators and IEF pass without any tweak. It seems to me some loss of accuracy happens with the new integrator, but I cannot see how that could happen given that the Butcher tableau used by Odeint is the same as mine.

Todos

• Developer Interest

• Dropped dependency on Boost.Odeint

Questions

• @ilfreddy please have a look and make sure it's fine to update the reference values for the Green's function tests.

Status

• Ready to go

[MinazoBot]

	1 Warning
⚠️	Consider adding supporting documentation to this change. Documentation sources can be found in the doc directory.

Generated by 🚫 Danger

[ilfreddy]

I don't see that you have included the modifications to the reference values in the test file. If you want me to evaluate whether the difference is significant or not, you should update them. Or you want me to run the tests myself and see?

[robertodr]

The latter. It's probably a good idea if you check this out on stallo and re-run the scans you used to test for the L=0 bug.

[ilfreddy]

I have checked the results against my test set of interfaces. The results are practically identical. Unfortunately the old results were printed with six digits accuracy only and I can therefore not go beyond that, but all numbers match to the last digit available (just some rounding differences).
This is therefore good to go as soon as the reference values are updated.

[ilfreddy]

Maybe we should keep the interface reference calculations (inp/out) somewhere on github, just in case? They are too long to be used as routine tests, but they might turn useful every now and then.

[robertodr]

We had a pcmsolvermeta repository for this kind of stuff when we were on GitLab. I didn't port that repo, but you're welcome to do that.

[codecov]

Codecov Report

Merging #138 into master will increase coverage by 0.24%.
The diff coverage is 89.22%.

@@            Coverage Diff             @@
##           master     #138      +/-   ##
==========================================
+ Coverage   69.83%   70.07%   +0.24%     
==========================================
  Files          92       93       +1     
  Lines        5616     5601      -15     
==========================================
+ Hits         3922     3925       +3     
+ Misses       1694     1676      -18

Impacted Files	Coverage Δ
src/utils/MathUtils.hpp	96.96% <ø> (-1.06%)	⬇️
src/interface/Meddle.cpp	71.13% <ø> (ø)	⬆️
src/green/SphericalDiffuse.hpp	44.44% <ø> (ø)	⬆️
src/utils/Molecule.hpp	100% <100%> (ø)	⬆️
src/utils/Symmetry.cpp	100% <100%> (ø)	⬆️
src/utils/RungeKutta.hpp	100% <100%> (ø)
src/utils/Molecule.cpp	74.46% <66.66%> (ø)	⬆️
src/green/SphericalDiffuse.cpp	80% <69.23%> (-1.46%)	⬇️
src/utils/Symmetry.hpp	76.19% <73.68%> (-23.81%)	⬇️
src/green/InterfacesImpl.hpp	77.27% <83.72%> (-2.34%)	⬇️
... and 4 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 5133636...8b2f3a9. Read the comment docs.