Changes for upcoming VAMPyR update #173
Conversation
and protect isVisibleAtScale and isZeroOnInterval.
Removed separate classes for Hilbert and Lebesgue iterators. Now the type is given as argument to the generic TreeIterator. Reason: easier to understand the TreeIterator name.
Codecov Report
@@ Coverage Diff @@
## master #173 +/- ##
==========================================
+ Coverage 65.59% 65.74% +0.14%
==========================================
Files 173 171 -2
Lines 12401 12425 +24
==========================================
+ Hits 8135 8169 +34
+ Misses 4266 4256 -10
Continue to review full report at Codecov.
|
Semi periodic Gaussians are still available from the function_utils::periodify function (the artist formerly known as make_gaussian_periodic).
...for backward compatibility
src/functions/function_utils.cpp
Outdated
| @@ -47,8 +51,8 @@ namespace mrcpp { | |||
| * integral is conserved with respect to the integration limits. | |||
| * | |||
| */ | |||
| template <int D> std::shared_ptr<GaussExp<D>> function_utils::make_gaussian_periodic(const Gaussian<D> &inp, const std::array<double, D> &period, double nStdDev) { | |||
| auto gauss_exp = std::make_shared<GaussExp<D>>(); | |||
| template <int D> GaussExp<D> function_utils::periodify(const Gaussian<D> &inp, const std::array<double, D> &period, double nStdDev) { | |||
There was a problem hiding this comment.
Add #include "functions/function_utils.h" in api/Gaussians. We need this in mrchem.
There was a problem hiding this comment.
Actually, perhaps we should just re-introduce it as a member function of the Gaussian? I like this syntax better:
auto periodic_exp = gfunc.periodify(period);
|
@bjorgve can you think of any issue with the removal of the periodic Gaussians when it comes to its use in MRChem? |
|
I'm looking into it now. |
|
It didn't break any of my periodic tests. So I believe we are safe. |
|
Great! I would still like to move the operator scale stuff from the MRA into the operators themselves, if possible. Looking into it now. |
Also add kernel precision parameter for ConvolutionOperators and fall back to old precision parameters for non-periodic operators.
|
Ok @bjorgve I think I managed to move the Old way of initializing periodic operator: New way: |
|
This is a nice change, this means we can choose a different reach/scale for the |
There was a 1D special case of BoundingBox::getNode implemented which was not periodified.
|
Ok, I think this is good now 🤞 |
TreeIteratorclass with a parameter for parser type (Hilbert or Lebesgue) instead of the old derived classesHilbertIteratorandLebesgueIterator. The main reason for this is for clarity when using it as the old class names were not very descriptive.Gaussian's property of being "periodic", with aGaussExpmember representing it's periodic images. Now there are just regular non-periodic Gaussians and a free functionperiodify(Gaussian<D>)(formerly known asmake_gaussian_periodic) which returns a semi-periodic version of the Gaussian in the form of aGaussExp. So when working in a periodic setting you should firstperiodifyyour Gaussian and then continue with the resultingGaussExp. Not sure if this retains all the functionality of the old version, as I do not remember the ideas behind the old setup.