Refactor bonded interaction parameter handling #4161

biermanncarl · 2021-03-16T10:03:46Z

Fixes #3019, partial fix for #3680, closes #4066

Description of changes:

Bonded parameters are now stored inside a boost::variant
Most bond types are now in self-contained header files (dependency reversals compared to former state)
Most bond kernels have become members of the respective structs
Each bond type is represented by its own C++ type (the three tabulated types are now separate, too), which is important because C++ type is now equivalent to bond type
Removed the former bond type identification scheme from the C++ core
Separate bonded IA cutoff from non-bonded IA cutoff
Skip bond loop independently from non-bonded loop

KaiSzuttor · 2021-03-17T11:50:21Z

src/config/config.hpp

@@ -90,6 +90,11 @@
 #define TINY_OIF_ELASTICITY_COEFFICIENT 1e-10
 #endif

+/** Maximal number of soft particles per IBM volcons bond (?) */
+#ifndef IBM_MAX_NUM
+#define IBM_MAX_NUM 1000


It's basically a relocation of a constant from here

This constant had to be moved out to help with dependency inversion, IIRC.

src/core/bonded_interactions/angle_cosine.hpp

src/core/bonded_interactions/angle_cossquare.hpp

src/core/bonded_interactions/bonded_coulomb.hpp

KaiSzuttor · 2021-03-17T11:59:27Z

src/core/bonded_interactions/bonded_interaction_data.cpp

-                          return std::max(max_cut, cutoff(bond.type, bond.p));
-                        });
+  auto const max_cut_bonded = boost::accumulate(
+      bonded_ia_params, -1.,


why do you initialize with -1 and not 0?

I think -1. is the magic value that will be replaced by the more expressive macro BONDED_INACTIVE_CUTOFF in #4066.

Still don't understand this... The minimum value for the maximum cut off is zero and not some magic number.

KaiSzuttor · 2021-03-17T12:00:58Z

src/core/bonded_interactions/bonded_interaction_data.cpp

@@ -99,31 +60,28 @@ double maximal_cutoff_bonded() {
  return (any_dihedrals) ? 2 * max_cut_bonded : max_cut_bonded;
 }

+int bonded_ia_params_num_partners(int bond_id) {


Suggested change

int bonded_ia_params_num_partners(int bond_id) {

int no_of_partners(int bond_id) {

The bonded_ia_params_ prefix is used in this function and the ones above and below because it's acting on the variant called bonded_ia_params. This will become clear once we add the correct doxygen group delimiters. We could even move these functions to src/script_interface.

Function moved to src/script_interface/interactions/bonded.hpp and modified to call a new function with signature number_of_partners(Bonded_ia_parameters const &iaparams) in src/core/bonded_interactions/bonded_interaction_data.cpp.

src/core/bonded_interactions/bonded_interaction_data.hpp

src/core/bonded_interactions/bonded_interactions.dox

src/python/espressomd/interactions.pxd

- store bonded parameters in a boost::variant - make force/energy kernels methods of the bond structs - use boost::serialization for MPI communication Co-authored-by: Jean-Noël Grad <[email protected]>

jngrad

good to go from my side

KaiSzuttor

Please check that all types follow our naming convention (SnakeCase) and adjust the naming of the bond classes (remove the Parameters suffix)

src/core/bonded_interactions/angle_cosine.hpp

src/core/bonded_interactions/angle_harmonic.hpp

src/core/bonded_interactions/bonded_coulomb_sr.hpp

src/core/bonded_interactions/bonded_interaction_data.hpp

src/core/bonded_interactions/bonded_tab.cpp

src/core/bonded_interactions/bonded_tab.hpp

biermanncarl changed the title ~~Refactor bonded interaction parameter handling~~ WIP - Refactor bonded interaction parameter handling Mar 16, 2021

biermanncarl force-pushed the fix-3019 branch from bae6fbc to a788a53 Compare March 16, 2021 14:26

KaiSzuttor reviewed Mar 17, 2021

View reviewed changes

biermanncarl force-pushed the fix-3019 branch 2 times, most recently from 1e4f86f to bcefbc8 Compare March 19, 2021 15:12