⚛️ phenomenology

Tailored PDFs for new physics searches

This paper demonstrates that new physics signals in high-energy collider data can be inadvertently absorbed into Parton Distribution Functions (PDFs), potentially hiding them from future searches, but shows that both conservative energy cuts and simultaneous PDF-SMEFT fits using the SIMUnet tool can successfully recover these signals.

Original authors: Elie Hammou

Published 2026-07-16

📖 3 min read🧠 Deep dive

CC BY 4.0

Original authors: Elie Hammou

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the universe as a giant, chaotic orchestra playing a song we know by heart: the Standard Model. For decades, physicists have been the conductors, trying to hear every single note perfectly. But lately, they've started listening for something else—a faint, new melody hiding in the high-pitched, screaming notes of the music. This is the hunt for "New Physics," things that don't fit our current musical score. To hear these new notes, scientists need to know the orchestra's baseline perfectly. They need to know exactly how loud the violins and drums usually are so they can spot when a new instrument sneaks in. In the world of particle colliders, these "instruments" are the tiny building blocks inside protons, called quarks and gluons. The map of where these particles live and how they move is called a Parton Distribution Function, or PDF for short. The problem is, the loudest, most energetic notes in the orchestra are the hardest to predict. If a new, mysterious instrument starts playing loudly in those high-energy zones, our current maps might accidentally "learn" that new noise as part of the old song, effectively erasing the evidence of the new instrument before we even realize it's there.

This paper, written by Elie Hammou from the Nikhef Theory Group, tackles this exact musical mix-up. The author sets up a clever simulation—a "closure test"—to see what happens when a fake new particle (a heavy $W'$ boson) is secretly injected into data from the High-Luminosity LHC. The results show a worrying scenario: if scientists try to update their maps (PDFs) using all the data first, assuming only the old Standard Model is playing, the new particle's signal gets swallowed up. The map changes to fit the noise, and when they later try to find the new particle, it's gone, hidden inside the map itself. The paper demonstrates two ways to fix this musical earworm. The first is a "conservative" approach: simply turn down the volume on the highest-energy notes during the map-making process, ignoring anything above a certain energy cutoff (like $Q_{max} = 500$ GeV). This keeps the map clean, allowing the new particle to be spotted later. The second, more sophisticated strategy is to stop making the map and listening for new notes in separate steps. Instead, they use a tool called SIMUnet to do both at once. By fitting the map and the new particle's signal simultaneously, the method prevents the map from absorbing the new physics. The simulations show that both methods successfully recover the hidden signal, proving that we can find the new music without losing the rhythm of the old song.

Technical Summary: Tailored PDFs for New Physics Searches

Problem Statement
In the absence of direct discoveries of Beyond the Standard Model (BSM) physics at the LHC, indirect searches utilizing the Standard Model Effective Field Theory (SMEFT) have become crucial. These searches rely on detecting broad distortions in high-energy tails of collider observables. However, a critical methodological conflict exists between determining Parton Distribution Functions (PDFs) and performing SMEFT fits.

Global PDF fits are traditionally performed under the assumption that the Standard Model (SM) is the correct theory. If unaccounted-for new physics (NP) exists in the high-energy tails of the data used for these fits, the fitting procedure can "absorb" these energy-growing effects into the PDFs, particularly in the large Bjorken- $x$ region where data constraints are weakest. This results in a "BSM-biased" PDF set that mimics the NP signal. Consequently, when these biased PDFs are used as fixed inputs for subsequent SMEFT analyses, the underlying new physics is masked, and the fit incorrectly favors the SM, potentially hiding genuine deviations.

Methodology
The paper investigates this contamination effect using a "closure test" scenario at the High-Luminosity LHC (HL-LHC, $\sqrt{s} = 14$ TeV). Synthetic charged-current high-mass Drell-Yan (HMDY) data is generated using the NNPDF4.06 set as the true underlying PDFs, with an injected heavy, universally-coupled $W'$ boson effect parametrized by the SMEFT oblique coefficient $\hat{W} = 8 \times 10^{-5}$ .

Two distinct strategies are evaluated to mitigate this bias and recover the injected signal:

Conservative Separate Fits: This approach decouples the PDF determination from the SMEFT analysis. Data above a specific energy scale ( $Q_{\text{max}}$ ) is explicitly excluded from the PDF fit to prevent BSM-sensitive regions from biasing the PDFs. The resulting "conservative" PDF set is then used as a fixed input for a standard SMEFT-only fit on the full dataset. The study tests various cut values ( $Q_{\text{max}} = 500$ GeV and $1500$ GeV) to determine the threshold required to remove the distortion.
Simultaneous Fits: This approach abandons the sequential logic entirely. Using the open-source tool SIMUnet, which extends the NNPDF neural-network methodology, the PDFs ( $\theta$ ) and SMEFT Wilson coefficients ( $c$ ) are determined concurrently in a single joint fit ( $T(\theta, c) = \text{PDF}(\theta) \otimes \hat{\sigma}(c)$ ). This method utilizes the entire dataset without assuming the SM ( $c=0$ ) a priori, ensuring that any BSM signal cannot be selectively absorbed into the PDFs without manifesting in the fitted Wilson coefficients.

Key Results

The Bias Effect: When the synthetic data (containing the injected $\hat{W}$ ) is fitted under the SM assumption, the resulting PDFs absorb the signal. A subsequent SMEFT fit using these biased PDFs fails to identify the new physics, appearing compatible with the SM while excluding the true underlying theory at high significance.
Conservative Strategy Performance: Applying a sufficiently aggressive energy cut ( $Q_{\text{max}} = 500$ GeV) successfully removes the BSM-induced distortion from the PDF luminosity. The subsequent SMEFT fit correctly centers on the injected $\hat{W}$ value and excludes the SM point. However, milder cuts (e.g., $1500$ GeV) are insufficient, and the bias persists. A trade-off exists: lowering $Q_{\text{max}}$ inflates PDF uncertainties due to reduced data constraints in the large- $x$ region.
Simultaneous Strategy Performance: The simultaneous fit using SIMUnet successfully recovers the PDF luminosity consistent with the true underlying law, with uncertainties comparable to the conservative approach. Crucially, it correctly excludes the SM point in the $(\hat{W}, \hat{Y})$ plane with high significance. This method was also tested with a second correlated direction ( $\hat{Y}$ ) and in an independent scenario involving a heavy colour-octet affecting the $t\bar{t}$ tail; in all cases, the simultaneous fit recovered the signal, often yielding tighter bounds than the conservative approach.

Significance and Claims
The paper demonstrates that a naive, sequential approach (PDF fit followed by SMEFT fit) risks hiding new physics by allowing the PDFs to absorb BSM signals. It establishes that both conservative separate fits (with appropriate energy cuts) and simultaneous fits are effective, complementary strategies to obtain robust PDFs and SMEFT bounds that are resistant to hidden new physics.

The simultaneous approach offers the advantage of utilizing the full high-energy dataset to constrain PDFs without the need for an arbitrary, a priori energy cut, though it requires specifying the SMEFT directions to be fitted and involves a higher-dimensional parameter space. The conservative approach serves as a useful diagnostic; observing a systematic, energy-dependent drift in results as the cut $Q_{\text{max}}$ is relaxed can signal BSM contamination. The authors conclude that these methods are essential for precision BSM phenomenology to ensure that PDF uncertainties do not obscure genuine deviations from the Standard Model.

Technical Summary: Tailored PDFs for New Physics Searches

More like this