Mapping quark-level kinematics to hadrons in a new hybrid model of semileptonic $B$ meson decays

This paper presents a novel hybrid model for semileptonic $B$ meson decays that utilizes an optimal transport algorithm to seamlessly merge resonant and non-resonant components, thereby eliminating unphysical features like discontinuities and negative yields found in current simulation methods.

The Gist

The Great Particle Puzzle: Smoothing the Rough Edges of the Universe

Imagine you are trying to predict the outcome of a massive, chaotic game of billiards, but instead of balls, you are dealing with subatomic particles called quarks. Specifically, you are watching a heavy particle called a B meson decay (break apart) into a lighter particle, a lepton (like an electron), and a neutrino.

Physicists have two ways to describe this game:

1. The "Quark" View: A smooth, mathematical calculation that treats the particles like a continuous fluid. This is great for big-picture predictions.
2. The "Hadron" View: The messy, real-world reality where particles clump together into specific, distinct shapes (like a $\pi$ meson or a $\rho$ meson). This is what we actually see in detectors.

The problem? These two views don't always agree, especially when the particles are moving slowly or forming small, distinct clumps. For decades, physicists have tried to glue these two views together using a "hybrid model," but the old glue was messy. It created jagged, unnatural jumps in the data and sometimes even predicted "negative" particles, which is physically impossible.

This paper introduces a new, elegant way to glue these views together using a concept called Optimal Transport.

---

The Analogy: Moving Dirt vs. Cutting and Pasting

To understand the difference between the old method and the new one, let's use an analogy of moving dirt.

The Old Method: The "Bin-by-Bin" Approach

Imagine you have a pile of dirt (the smooth quark prediction) and you want to reshape it to match a specific sculpture (the real-world hadrons).

• The Old Way: You chop the ground into a grid of square boxes. You look at Box #1. If the sculpture needs 10 pounds of dirt there, but your pile only has 5, you magically add 5 pounds. If the sculpture needs 0 but you have 10, you throw 10 pounds away.
• The Problem: This creates a jagged, stepped landscape. If you look at the edge of Box #1, the dirt level might suddenly drop from 10 pounds to 0. In physics, this creates "discontinuities" (jumps) that look fake. Worse, if the sculpture needs more dirt than the pile has in a specific box, the math forces you to have "negative dirt" to balance the books. That's like saying you have -5 apples; it breaks the laws of physics.

The New Method: Optimal Transport

Now, imagine you are a master landscaper. Instead of chopping the ground into boxes, you look at the entire landscape at once.

• The New Way: You ask, "What is the most efficient way to move the dirt from the pile to the sculpture?" You don't just dump dirt in a box; you gently slide it. If you need to move a little dirt from the left side of the pile to the right side of the sculpture, you do it smoothly.
• The Result: The transition is seamless. There are no sudden jumps or cliffs. The dirt flows naturally from one shape to the other. This is Optimal Transport. It minimizes the "work" (or distance) required to turn the smooth prediction into the realistic model.

Why This Matters for Physics

In the world of B meson decays, this "smoothness" is crucial for a few reasons:

1. No More "Negative" Particles: The old method sometimes forced physicists to subtract events, leading to negative numbers in their data. The new method simply moves events around, so you never have negative counts.
2. Smoother Data: Real-world data shouldn't have jagged, artificial jumps just because we decided to draw a grid on our graph. The new method produces smooth curves that look like real physics.
3. Better Precision: Scientists are trying to measure a fundamental constant of the universe called $|V_{ub}|$ (which helps explain why the universe has more matter than antimatter). To measure this with extreme precision, the "glue" between theory and experiment must be perfect. The old jagged glue introduced errors; the new smooth glue reduces those errors significantly.

The "Traffic" Metaphor

Think of the particles as cars on a highway.

• The Old Model: Traffic lights are placed every mile. If a light turns red, cars stop abruptly. If a lane closes, cars are forced to vanish or appear out of nowhere to keep the count right. It's chaotic and unrealistic.
• The New Model: It's like a smart traffic system. Cars gently merge, slow down, or speed up to fill gaps. The flow is continuous. The total number of cars is preserved, but they are distributed in the most logical, efficient way possible.

The Bottom Line

The authors of this paper, Philipp Horak, Robert Kowalewski, and Tommy Martinov, have developed a mathematical tool (using a library called POT for Python) that acts like a "smart mover" for particle data.

Instead of forcing a square peg into a round hole (or a jagged grid into a smooth curve), they use Optimal Transport to gently reshape the theoretical predictions until they perfectly match the experimental reality. This allows physicists to measure the secrets of the universe with greater accuracy and fewer headaches, paving the way for future discoveries at facilities like the Belle II experiment in Japan.

In short: They found a way to make the math of the universe flow as smoothly as the particles themselves.

Technical Summary

1. Problem Statement

Precise determination of the Cabibbo-Kobayashi-Maskawa (CKM) matrix element $|V_{ub}|$ relies on inclusive semileptonic $B \to X_u \ell \nu$ decay measurements. These measurements depend on the Heavy Quark Expansion (HQE), which provides reliable theoretical predictions for quark-level processes integrated over large phase space regions. However, a fundamental challenge arises in mapping these parton-level calculations to physical, color-neutral hadrons:

• Quark-Hadron Duality: While duality holds over large phase space regions, it breaks down at low hadronic invariant masses ($M_X$), where discrete resonances (e.g., $\pi, \rho$) dominate.
• The Hybrid Model Necessity: To model experimental data, physicists must combine exclusive predictions for known low-mass resonances with inclusive HQE predictions for higher-mass systems.
• Limitations of Current Methods: The standard "conventional hybrid model" uses a bin-by-bin reweighting procedure. It sums exclusive and inclusive contributions and scales the inclusive component to preserve the total rate. This approach suffers from two critical unphysical artifacts:
  1. Negative Weights: If the measured exclusive branching fraction in a specific kinematic bin exceeds the inclusive HQE prediction, the reweighting factor becomes negative. This is particularly problematic for certain theoretical models (e.g., BLNP) and leads to unphysical kinematic distributions.
  2. Discontinuities: The bin-by-bin approach creates sharp, unphysical discontinuities at bin boundaries (most notably in the $M_X$ spectrum), which become increasingly problematic as detector resolution improves.

2. Methodology: Optimal Transport Hybrid

The authors propose a novel method based on Optimal Transport (OT) theory to replace the bin-by-bin reweighting. Instead of treating kinematic bins independently, the method treats the problem as a global redistribution of probability mass.

• Core Concept: The goal is to find an optimal mapping that transforms the source distribution (inclusive HQE prediction) into a target distribution (sum of known exclusive resonances) while minimizing the "cost" of moving events in kinematic space.
• Mathematical Framework:
  • The problem is formulated using the Wasserstein distance (specifically the Earth-Mover's Distance, $W_1$), which measures the minimal "work" required to transform one distribution into another.
  • The phase space is discretized using light-cone variables $P_+$ and $P_-$ (related to $M_X$ and $q^2$).
  • A transport plan ($T_{ij}$) is computed numerically (using the Python Optimal Transport library) to determine how much probability mass moves from source bin $i$ to target bin $j$.
• Implementation Details:
  • Source ($\mu$): Inclusive DFN (De Fazio-Neubert) or BLNP model predictions.
  • Target ($\nu$): Sum of exclusive resonant decays ($\pi, \rho, \omega, \eta, \eta'$).
  • Sink Node: A sink node is introduced to absorb source mass that exceeds the total target mass, ensuring total normalization is conserved.
  • Cost Metric: The Euclidean distance between bin centers in the $(P_+, P_-)$ plane is used as the cost function.
  • Regularization: The authors also explored entropic regularization (Sinkhorn algorithm) to introduce smoothness, though they found the exact solution (network simplex) superior for preserving moments.

3. Key Contributions

1. Algorithmic Innovation: Introduction of an Optimal Transport framework to the construction of hybrid models for $B \to X_u \ell \nu$ decays, moving away from local bin-by-bin reweighting to a global, continuous redistribution.
2. Resolution of Negative Weights: The OT method naturally handles cases where exclusive contributions exceed inclusive predictions locally without generating negative weights, enabling the use of previously problematic theoretical models like BLNP.
3. Elimination of Discontinuities: By operating on a continuous (or finely binned) phase space, the method eliminates the artificial jumps in kinematic spectra seen in conventional hybrids.
4. Improved Moment Conservation: The method is designed to minimize the displacement of events, thereby better preserving the spectral moments (mean, variance, skewness, kurtosis) of the original HQE prediction.
5. Open Source Tools: The authors provide the simulation samples on Zenodo and the implementation code on GitHub, facilitating immediate adoption by the experimental community.

4. Results

The authors compared the new Optimal Transport (OT) hybrid against the conventional bin-by-bin hybrid using both DFN and BLNP theoretical inputs for $B^+$ and $B^0$ decays.

• Kinematic Spectra:
  • $M_X$ Distribution: The OT hybrid produces a smooth spectrum without the large discontinuity at $M_X \approx 1.4$ GeV observed in the conventional method.
  • $q^2$ and $E_\ell$: Discontinuities at bin boundaries (e.g., $q^2 \approx 14$ GeV$^2$) are eliminated.
  • $\cos \theta_\ell$: The lepton helicity angle distribution, which was not an input to the reweighting, is better preserved in the OT model, indicating a more physically consistent global mapping.
• Moment Conservation:
  • The OT hybrid significantly outperforms the conventional method in preserving the spectral moments of $M_X^2$, $q^2$, and $E_\ell$.
  • Quantitative Improvement: The mean relative error in moment conservation across variables and orders dropped from 4.1% (conventional) to 1.0% (OT).
  • Specific improvements include reducing the error in $\langle M_X^2 \rangle$ from 7.0% to 0.7% and $\langle E_\ell \rangle$ from 2.2% to 0.2%.
• BLNP Compatibility: The OT method successfully constructs a physical distribution using the BLNP model, whereas the conventional method fails due to a high prevalence of negative weights.
• Strangeness Suppression: The OT hybrid better preserves the fraction of events containing kaons (approx. 10.0%) compared to the conventional hybrid (approx. 8.5%), which artificially depletes strange quark production.

5. Significance

This work addresses a critical bottleneck in flavor physics precision measurements:

• Future-Proofing for Belle II: With the Belle II experiment accumulating high-statistics datasets and improving detector resolution, the unphysical artifacts of the bin-by-bin method are becoming a limiting factor for $|V_{ub}|$ precision. The OT hybrid provides a theoretically motivated solution that scales with these improvements.
• Theoretical Consistency: By better preserving HQE moments and avoiding unphysical negative weights, the method ensures that experimental measurements are more directly comparable to Standard Model predictions.
• Generalizability: While demonstrated for $B \to X_u \ell \nu$, the framework is general and can be applied to other processes requiring the merging of resonant and non-resonant contributions, such as $B \to X_s \ell^+ \ell^-$, $B \to X_s \nu \bar{\nu}$, and $B \to X_s \gamma$.

In conclusion, the paper presents a robust, mathematically rigorous alternative to the 25-year-old standard for hybrid modeling, offering smoother spectra, better theoretical consistency, and the ability to utilize a wider range of theoretical inputs.