The%20hep_random%20Library

1
The hep_random Library

• A component of the coming C standard and of LHC
  MathCore

2
The standard random library

• Proposed for inclusion in C standard
• Has made it into TR1
• Engines (random number generators) and
  distributions
• Early package in Boost written by Jens Maurer
• Significant conceptual revamping with heavy input
  from FNAL
• Based on CLHEP Random experience

3
What is hep_random

• A fully compliant implementation of the proposed
  standard random library
• With those tweaks we are relatively sure will be
  accepted by the standards committee
• Suitable as a reference implementation
• The standard does not like to incorporate
  completely groundbreaking art
• Users of hep_random will find transition to
  stdrandom trivial
• Namespace will change

4
What are the advantages in having this in the
standard library

• Avoid any temptation for each physics group to
  roll your own
• Engines can have subtle traps that lead to
  incorrect physics results!
• Weight of the compiler vendors and world of
  developers behind improvements
• We are good
• They are often better
• No issue of portability
• (except for non-compliant compilers)
• Let compiler vendors do the maintenance

5
Nature of the random library

• Concepts mostly the same as for CLHEP
• Engines
• Distributions
• Some features that are clearly remnants of
  Fortran-style programming are gone
• Designed to easily be extended
• New distributions new engines
• Heavily templated for maximal user control

6
In what ways does hep_random differ from
CLHEP/Random?

• They are much more similar than different!
• Better model of expressing sharing/not sharing of
  engines
• Engine used is always explicit in distribution
  invocation
• Slightly different set of engines and
  distributions
• We will be providing an extension that supplies
  all the missing ones, for the HEP community
• A significant new distribution that allows for
  variates from a general user-supplied pdf function

7
Where are we now?

• We have the proposed final wording of the library
  specification in a proposal to J16
• Some final tweaks will appear, but mostly
• We were able to say with honesty that the issues
  one discovers by implementing the standard have
  been explored
• We have our reference implementation just about
  completed
• Some alternate algorithms for some distributions
  have to be tested and vetted or excised
• We are in serious Users Guide writing mode

8
Forward From Here

• Two possible paths
• Submit it to Boost
• Create a Boost-independent version

9
The Boost Path

• Submit hep_random for inclusion in Boost
• Our code already uses Boost
• All the experiments we work with already use
  Boost
• This is the best way of gaining widespread
  visibility
• Which also means people helping to improve the
  library
• Will need to modify code and prepare build files
  for Boost build
• This is the path requiring the least work

10
The Non-Boost Path

• Modify the code, removing Boost dependencies
• Package it suitable for any/all of
• MathCore (thus LHC expts, and ROOT)
• Geant4
• CLHEP (as an underlying implementation)
• Requires more work than submission to Boost
  requires
• Magnitude of the task depends upon how well these
  groups can agree on things, and how much of their
  own styles each group insists upon