MCdevelop - the universal framework for Stochastic Simulations

We present MCdevelop, a universal computer framework for developing and exploiting the wide class of Stochastic Simulations (SS) software. This powerful universal SS software development tool has been derived from a series of scientific projects for precision calculations in high energy physics (HEP), which feature a wide range of functionality in the SS software needed for advanced precision Quantum Field Theory calculations for the past LEP experiments and for the ongoing LHC experiments at CERN, Geneva. MCdevelop is a “spin-off” product of HEP to be exploited in other areas, while it will still serve to develop new SS software for HEP experiments. Typically SS involve independent generation of large sets of random “events”, often requiring considerable CPU power. Since SS jobs usually do not share memory it makes them easy to parallelize. The efficient development, testing and running in parallel SS software requires a convenient framework to develop software source code, deploy and monitor batch jobs, merge and analyse results from multiple parallel jobs, even before the production runs are terminated. Throughout the years of development of stochastic simulations for HEP, a sophisticated framework featuring all the above mentioned functionality has been implemented. MCdevelop represents its latest version, written mostly in C++ (GNU compiler gcc). It uses Autotools to build binaries (optionally managed within the KDevelop 3.5.3 Integrated Development Environment (IDE) ). It uses the open-source ROOT package for histogramming, graphics and the mechanism of persistency for the C++ objects. MCdevelop helps to run multiple parallel jobs on any computer cluster with NQS-type batch system.

💡 Research Summary

The paper introduces MCdevelop, a universal, open‑source framework designed to streamline the development, testing, parallel execution, and result aggregation of stochastic (Monte‑Carlo) simulations. Originating from a series of high‑energy‑physics (HEP) projects such as BHLUMI and EvolFMC, MCdevelop abstracts the common infrastructure needed for large‑scale Monte‑Carlo applications and packages it into a reusable C++ library. The core of the framework consists of two abstract base classes: TMCgen, which defines the interface for event generators, and TRobol, which handles the execution flow and on‑the‑fly analysis (histogram filling, statistical accumulation). Users create concrete subclasses of TMCgen to implement specific probability distributions or physics models, while TRobol subclasses implement custom analysis or monitoring logic.

A central design choice is the reliance on the ROOT data analysis framework for histogramming, graphics, random number generation, and, crucially, object persistency. By storing the random‑number generator state, the full generator object, and analysis modules in ROOT files, MCdevelop enables checkpointing and reproducibility of long‑running batch jobs. This persistency also simplifies the management of independent parallel jobs: each job runs in its own working directory with a unique random seed, yet all can later be merged by ROOT utilities.

The build system uses GNU Autotools (autoconf, automake, libtool) to generate a portable configure script that detects ROOT, sets compiler flags, and builds the shared library libMCdev. Optional integration with the KDevelop 3.5.3 IDE provides a graphical environment for project import, compilation, and debugging (via GDB), reducing the learning curve for new developers.

Parallel execution is facilitated through a set of ROOT‑based shell scripts located in MCdev/farming. These scripts automate the creation of multiple working directories, seed initialization, job submission to an NQS‑type batch system, monitoring, and graceful termination. After the batch run, a merging script combines all ROOT histogram files into a single output file, and a ROOT macro (XPlot) can be used to visualise the accumulated results even before the batch finishes, offering on‑line monitoring capabilities.

MCdevelop also bundles the ROOT Foam package (TFoam), a general‑purpose multidimensional adaptive sampling tool. By inheriting from TMCgen and embedding a TFoam instance, users can automatically generate unit‑weight events from arbitrary high‑dimensional distributions (typically up to ~15 dimensions) without writing custom sampling code. This feature is highlighted as a key advantage for complex integrals or phase‑space generation tasks.

To aid newcomers, the distribution includes two template projects (Template0 and Template). Each contains a minimal MainPr program that drives the generator, a Start.C macro for configuring run parameters, and an XPlot analysis macro. These templates illustrate the typical directory layout (work directory for runtime files, Start.C for initialization) and serve as a starting point for custom applications.

The authors enumerate ten functional requirements that guided the design: universality, minimal dependence on proprietary code, integration of open‑source tools, IDE support, semi‑automatic configuration, robust persistency, transparent input handling, seamless switch between single‑node and farm execution, automated job deployment and result merging, and on‑line monitoring of partial results. MCdevelop satisfies all these criteria, as demonstrated through detailed descriptions of its architecture and workflow.

Limitations are acknowledged. The current farming scripts are tailored to NQS‑type batch systems, so users of SLURM, PBS, HTCondor, or cloud‑based schedulers would need to adapt the scripts. The heavy reliance on ROOT may be a barrier for environments where ROOT is unavailable or where a lighter‑weight solution is preferred. Future work outlined includes extending support to other batch systems, migrating the build system to CMake for broader platform compatibility (including Windows/macOS), providing container images (Docker/Singularity) for reproducible deployments, and developing a plugin architecture for dynamic loading of user‑defined analysis modules.

In summary, MCdevelop offers a cohesive, well‑documented environment that integrates modern C++ development practices, ROOT’s powerful data handling, and automated batch processing. It enables physicists and computational scientists to focus on the scientific aspects of stochastic simulations while abstracting away the repetitive engineering tasks associated with large‑scale Monte‑Carlo production. With the planned extensions, MCdevelop has the potential to become a standard tool not only in HEP but also in finance, biology, and engineering domains where Monte‑Carlo methods are indispensable.