Negative weight mitigation with cell resampling and tests with MCFM

Description

MCFM (Monte Carlo for FeMtobarn processes) is a widely used software package in high-energy physics. It simulates particle collisions, such as those at the Large Hadron Collider (LHC), allowing physicists to compare theoretical predictions with experimental data. It specializes in high-precision predictions Next-to-Next-to-Leading Order and beyond) for a vast array of particle processes.

When physicists calculate predictions for these collisions using higher-order quantum field theory, the mathematics often requires “subtraction schemes” to handle infinities. A side effect of this is that some simulated events are assigned a “negative weight” (effectively a negative probability).

While these negative weights make sense mathematically—they cancel out other positive events to give the correct physical result—they are computationally very expensive. In downstream processing (like simulating how a particle detector responds), a negative event and a positive event must both be fully simulated only to cancel each other out later. This “statistical dilution” means we have to generate and store significantly more data just to achieve a standard level of precision.

A new method called “Cell Resampling” (proposed in arXiv:2109.07851 and arXiv:2303.15246) offers a way to fix this by redistributing these negative weights locally in phase space, effectively removing them without changing the physical prediction.

We plan to implement this method within MCFM. This project is a collaboration between theorists and experimentalists to:

  1. Prove the method works within a major parton-level Monte Carlo generator (MCFM).
  2. Stress-test the method to ensure it is robust enough for use in large-scale experimental workflows.

Once successful, this work will result in a public release of MCFM featuring this efficiency upgrade.

Task ideas

Expected results

Release a modified version of MCFM that successfully incorporates the negative weight mitigation strategy, demonstrating reduced statistical dilution.

Requirements

AI usage policy

AI assistance is allowed for this contribution. The applicant takes full responsibility for all code and results, disclosing AI use for non-routine tasks (algorithm design, architecture, complex problem-solving). Routine tasks (grammar, formatting, style) do not require disclosure.

Mentors

Additional Information

Corresponding Project

Participating Organizations