[Paper Review] A C++ interface to QCDNUM

Valerio Bertone, M. Botje|arXiv (Cornell University)|Dec 21, 2017

Particle physics theoretical and experimental studies5 references3 citations

TL;DR

This paper presents a C++ interface to the FORTRAN-based QCDNUM program, enabling seamless integration of QCD parton density evolution into modern C++-based high-energy physics software. By wrapping legacy FORTRAN routines with C++-compatible functions, the interface preserves QCDNUM's high performance while supporting object-oriented programming, dynamic memory management, and type safety, with validated examples showing accurate NNLO evolution and structure function computations.

ABSTRACT

In this document we report on the recent development of a C++ interface to the FORTRAN-based evolution program QCDNUM. A short description of the interface is given with a few basic examples of its usage.

Motivation & Objective

• To bridge the gap between legacy FORTRAN-based QCDNUM and modern C++-based HEP software frameworks.
• To maintain QCDNUM's high computational performance while enabling use in object-oriented C++ applications.
• To provide a portable, standards-compliant C++ interface using autotools and namespace encapsulation.
• To support key QCDNUM functionalities such as NNLO PDF evolution, structure function computation, and dynamic memory access.
• To facilitate integration with contemporary tools like xFitter and Alpos by enabling C++-native calls to QCDNUM routines.

Proposed method

• Wrapping existing FORTRAN subroutines in C++ functions with identical names but lowercase and wrapped in the QCDNUM namespace to avoid naming conflicts.
• Converting implicit FORTRAN typing to explicit C++ variable declarations, including handling of integer and double-precision types.
• Mapping FORTRAN array indexing (1-based, with negative indices for antiquarks) to 0-based C++ arrays using index offset adjustments.
• Implementing a 2D array mapping function k(i,j,n) to convert FORTRAN A(n,m) arrays into 1D C++ arrays for efficient memory access.
• Handling function pointers by passing function arguments as pointers to C++ functions, enabling user-defined PDF input functions in the evolution routine.
• Using autotools to build the library with proper linker and name-mangling support for interoperability between C++ and FORTRAN.

Experimental results

Research questions

• RQ1 How can a high-performance FORTRAN-based QCD evolution program be effectively exposed to modern C++ HEP applications?
• RQ2 What are the key challenges in interfacing legacy FORTRAN code with C++ in terms of type safety, memory management, and naming conventions?
• RQ3 Can a C++ interface preserve the computational speed of QCDNUM while enabling object-oriented programming practices?
• RQ4 How can array indexing differences between FORTRAN and C++ be correctly handled in a robust and maintainable way?
• RQ5 What is the impact of using autotools and namespace encapsulation on the portability and usability of the interface?

Key findings

• The C++ interface successfully enables calling QCDNUM's NNLO PDF evolution and structure function computation routines directly from C++ code with full type safety and modularity.
• The interface maintains QCDNUM's high performance by reusing the original FORTRAN codebase without reimplementation, avoiding performance degradation.
• The use of autotools ensures cross-platform compatibility and proper linking, with the interface available from qcdnum release 17-01/14 onward.
• Example code demonstrates accurate evolution of unpolarized PDFs in the variable-flavor number scheme, yielding a charm sea quark value of 1.8708 at x = 1e-3 and μ² = 1000 GeV².
• The strong coupling constant αs(mZ²) is correctly computed as 0.11807 using the interface, confirming consistency with the underlying FORTRAN implementation.
• The interface supports key features such as grid setup, weight computation, and evolution with LO, NLO, and NNLO order, as well as interpolation of PDFs at arbitrary x and μ².