Skip to content

Release notes

1.0.0 - 2018-05-22

Added

  • Interface to evaluate the integrand on a phase-space point (MoMEMta::setEvent and MoMEMta::evaluateIntegrand functions).
  • Support for python3 - new build options as a consequence: PYTHON_MIN_VERSION and Boost_PYTHON_VERSION_TAG.

Changed

  • Minimum CMake version is now 3.4.0.

Fixed

  • Build on systems where default compiler standard is C++14 (fixed conflict with ROOT enforcing the use of exactly the same standard as the one it was built with).
  • Problem in some module definitions (BinnedTransferFunctionOnEnergy, BinnedTransferFunctionOnPt, BlockB, BlockD): outputs and attributes did not correspond with module registry.
  • Bug in graph creation, visible only when several instances of MoMEMta were created.
  • LHAPDF could not be found in a CMSSW environment (lhapdf-config not available)

1.0.0 beta 1 - 2017-31-08

Added

  • New Looper module to loop over a set of solutions (see below for more details)
  • Python bindings. To enable, pass -DPYTHON_BINDINGS=ON to cmake.
  • Modules to only evaluate transfer functions (without integrating over anything): BinnedTransferFunctionOnEnergyEvaluator, GaussianTransferFunctionOnEnergyEvaluator
  • Matrix element parameters can now be edited from the configuration file.
  • ConfigurationReader constructor now accepts an optional second argument allowing the definition of variables accessible from the lua configuration.
  • When an exception occurs while constructing a module, try to give useful information to help debugging.
  • Support for CMSSW environment. Python, boost and LHAPDF libraries from CMSSW are automatically used and running make install will automatically register MoMEMta as a new external tool.
  • Support use of other integration algorithms implemented in Cuba (Suave, Divonne, Cuhre)
  • Binned and Gaussian transfer function on Pt
  • DEBUG_TIMING cmake option, to print a summary of each module runtime at the end of the integration.
  • New cuba option to configure multi-core integration, ncores and pcores.
  • declare_input lua function, used to declare a new input.
  • append and copy_and_append lua functions are now built-in MoMEMta.
  • Two new lua functions, add_reco_permutations and add_gen_permutations are available to easily insert a permutator module permutating between the function arguments.
  • LinearCombinator templated module allowing to compute combinations (ie sums, subtractions, ...) of LorentzVectors, numbers, ...
  • Secondary blocks A, B, CD, E
  • Main blocks A, C, E and G (not present in MadWeight)
  • getRandom4Vector function to generate random Lorentz vectors of a specified mass (useful in cases where a particle has to be passed from C++, but integrated over all its components).

Changed

  • The way to handle multiple solutions coming from blocks has changed. A module is no longer responsible for looping over the solutions itself, this role is delegated to the Looper module. As a consequence, most of the module were rewritten to handle this change. See this pull request and this one for a more technical description, and this documentation entry for more details
  • The way input tags corresponding to phase-space points are defined in Lua has changed. It is no longer allowed to specify them explicitly (through cuba::ps_point/i). The function getpspoint() has been changed to add_dimension() and should be used each time an additional dimension is needed for the integration (the function returns an input tag for the corresponding phase-space components). Modules do not declare how many dimensions they need (counting the number of dimensions is done automatically through the add_dimension() function). See this Pull Request for more details.
  • GaussianTransferFunction renamed to GaussianTransferFunctionOnEnergy to be consistent. Also, the width of the Gaussian is computed from the 'gen' energy, not the 'reco' (more correct). The width over which the integration is performed, however, is still computed using the 'reco' energy (no choice there): i.e. integrating over +/- 5 sigma with a resolution of 10% gives an integration range of 'E_gen' in +/- 5*0.1*E_reco.
  • Cuba smoothing option is now enabled by default to mitigate numerical instabilities during the integration. See this for more details.
  • If no integration dimension is requested, no integration is performed and the graph is only evaluated once. As such, no error is provided on the final result.
  • Cuba logging is now handled by MoMEMta logger, at debug level.
  • We no longer use boost-log for logging, but our own implementation heavily inspired by spdlog. As a consequence, boost-log is no longer required to build MoMEMta.
  • Boost is no longer a dependency when using MoMEMta (but it's still a build dependency)
  • MoMEMta::computeWeights now expects a vector of Particle and no longer a vector of LorentzVector. A Particle has a name, a LorentzVector and a type. As a result, configuration files must now declare which inputs are expected.
  • The way the inputs are passed to the blocks is changed (the particles entering the change of variables are set explicitly, the others are put into the branches vector of input tags)
  • Built-in lua version is now v5.3.4
  • Block B, D and F: support massive invisible particles

Fixed

  • Cuba forking mode was broken when building in release mode (with -DCMAKE_RELEASE_TYPE=Release).
  • SLHA card reader (matrix element parameter cards) retrieved from MadGraph was broken.
  • Transfer functions on energy take as lower bound the mass of the given "reco" particle