- get rid of the
create_result
function - add serialization functions for the
_result
classes that are needed to exchange and accumulate the results from different MPI processors - add functionality to capture non-finite numbers of the integrand and report how many numbers of
the generator need to be discarded to get there: save a copy of the generator the integrator
obtains, print out the state (with
operator<<
) and the number of discards needed - save the ten (n?) largest weights?
- simplify the handling of integrands by requiring that they must accept a
hep::projector<T>
? - remove the integrand classes; they complicate the use of the integration routines
- add under-/overflow bins
- write new integrator using multi channel and VEGAS for each channel; implement different
strategies:
- Multi channel run, then refine using VEGAS for the n largest channels
- Multi channel and VEGAS for all channels, adapting the weights and grids each iteration
- write new integrator using the FOAM algorithm?
- add
non_zero_calls
andfinite_calls
for each channel? - add possibility for the controlled removal of channels?
- add beta parameter to dampen the weight adjustment for better convergence
- get rid of the global configuration option, make it an argument to the vegas function?