An adaptive inverse-problem framework for one-loop five-gluon BCJ numerators

Technical Summary: An Adaptive Inverse-Problem Framework for One-Loop Five-Gluon BCJ Numerators

Problem Statement
The paper addresses the construction of Bern–Carrasco–Johansson (BCJ) numerators for one-loop five-gluon scattering amplitudes in pure Yang–Mills theory. While BCJ representations organize amplitudes as sums over cubic graphs where kinematic numerators satisfy Jacobi relations mirroring color factors, finding these numerators is an inverse problem with a generally non-unique solution. The challenge lies in reconstructing local, parity-even kinematic numerators from physical data (generalized unitarity cuts) while satisfying algebraic constraints (antisymmetry, Jacobi identities) and locality/power-counting rules. Previous methods often rely on constructive algorithms or specific ansätze that may not capture the full solution space or may lack rigorous validation of the remaining freedom.

Methodology: The Adaptive Inverse-Problem Framework
The authors formulate BCJ numerator construction as a finite, exact inverse problem, $P_\sigma = (A, b; S; T)$, governed by a fixed scientific specification $\Omega$ and an adaptive working specification $\sigma_t$.

1. Fixed Scientific Specification ($\Omega$): This defines the immutable rules of the problem: the theory (pure Yang–Mills), external states, graph conventions, coefficient field (rational functions of the internal spin dimension $D_s$), locality/power-counting rules, the definition of generalized cuts, the observable equivalence map $S$, and independent validation tests $T$.
2. Adaptive Working Problem ($\sigma_t$): At each stage, the working specification compiles to a linear system $Ax = b$.
   • $A$ (Constraint Matrix): Maps candidate numerator coefficients to physical cut data.
   • $b$ (Target Vector): Contains products of on-shell tree amplitudes summed over internal states.
   • $S$ (Observable Map): Classifies solutions by determining if differences between them affect the physical observable (specifically the $R_{12345}$ color-ring projection after LSZ reduction and scaleless integral removal).
   • $T$ (Validation Suite): Holds out independent checks (fresh cuts, integral reductions, helicity benchmarks) to verify the solution.
3. Exact Diagnostics and Adaptation: The framework uses exact linear algebra diagnostics to adapt the problem:
   • Left-Null Obstructions: If $\lambda^T A = 0$ but $\lambda^T b \neq 0$, the current ansatz is insufficient. The vector $\lambda$ identifies the inconsistency, guiding the addition of new basis columns (e.g., repeated-contact terms).
   • Right-Kernel Analysis: If the system is consistent, the right kernel $\ker A$ represents the freedom in the solution. Candidate measurement rows (cuts) are evaluated by their rank gain on this kernel ($\text{rank}(RK)$). Rows with non-zero gain are added to resolve blind directions.
4. Agent-Assisted Construction: The process is designed to be compatible with language-model agents or fixed search policies. An agent receives the exact diagnostic state (ranks, null spaces, residuals) and proposes revisions (adding basis sectors, selecting cut topologies). A compiler and exact evaluator then verify these proposals deterministically.

Key Contributions

1. Formulation as Exact Inverse Problem: The paper establishes a rigorous framework where amplitude representation construction is treated as a finite exact inverse problem, separating the fixed scientific rules from the adaptive computational realization.
2. Two Exact Diagnostics for Adaptation: It introduces a systematic method for refining the ansatz and measurement set using left-null obstructions (to fix consistency) and right-kernel information gain (to maximize measurement efficiency).
3. Complete Reconstruction of the Five-Point Fiber: The authors apply this framework to reconstruct the full 207-dimensional affine solution fiber for the one-loop five-gluon amplitude.
   • Basis Construction: Starting from 17,824 tensor monomials, algebraic reductions yield 10,724 independent structures. Graph symmetries reduce this to 1,127 independent coefficient coordinates. Crucially, the authors identify and add 300 "repeated-contact" directions ($Y_i^2 N_{ii}$) initially omitted in square-free bases, which were necessary for consistency.
   • Cut Design: Maximal, box, and triple cuts provided a rank of 916. A targeted search for informative double-cut rows increased the rank to 920, reducing the kernel dimension from 211 to 207.
4. Observable Classification: The authors prove that the entire 207-dimensional solution fiber maps to a single observable class under the specified $S$ map ($S(\ker A) = 0$). This means all solutions in the fiber are physically equivalent for the chosen readout.
5. Validation and Comparison:
   • The reconstructed solution is validated against independent master-integral reductions and helicity-amplitude benchmarks (all-plus, single-minus, MHV).
   • A published forward-limit numerator [12] is located exactly within the reconstructed fiber, expressed as a specific linear combination of the particular solution and 15 active kernel directions.
   • The paper demonstrates that while the forward-limit and fixed-routing prescriptions yield the same observable class, their raw prescription maps differ by a non-zero coefficient in a specific master integral, highlighting the map-dependence of equality.

Results

• Solution Space: The exact solution is an affine fiber $F_b = x_p + \ker A$ of dimension 207 over $\mathbb{Q}(D_s)$.
• Particular Solution: A reduced-row-echelon (RREF) representative $x_p(D_s) = x(0) + D_s x(1)$ is determined.
• Kernel: 207 rational kernel vectors are computed, representing all deformations invisible to the chosen cuts.
• Observable Equivalence: The map $S$ annihilates all 207 kernel directions, confirming that the cut data uniquely determines the observable amplitude (modulo the specified equivalence).
• Consistency: The system is consistent ($\text{rank} A = \text{rank}[A|b]$), and all independent validation checks (fresh cuts, integral reductions, helicity amplitudes) pass with zero residuals.

Significance and Claims
The paper claims to provide a deterministic, exact, and reproducible framework for constructing BCJ numerators that avoids heuristic guessing. By treating the problem as an adaptive inverse problem, it systematically identifies missing ansatz terms and informative measurements. The authors emphasize that their method:

• Guarantees Completeness: It determines the entire solution fiber compatible with the local ansatz and cut data, rather than finding a single arbitrary solution.
• Ensures Physical Equivalence: It rigorously classifies the residual freedom, proving that the 207-dimensional ambiguity does not affect the physical observable defined by $S$.
• Enables Automation: The framework is designed for agent-assisted construction, providing a common interface for physicists and AI agents to propose and validate revisions.
• Clarifies Prescription Dependence: It explicitly distinguishes between the uniqueness of the numerator modulo the cut/observable map and the equality of different integration prescriptions, showing that the latter requires additional surface or scheme completions.

The work serves as a proof-of-concept for a new paradigm in high-energy physics calculations, combining exact arithmetic, homological algebra concepts (obstructions and deformations), and adaptive computational strategies to solve complex scattering amplitude problems.