Recursive Jigsaw Reconstruction

• Recursive Jigsaw Reconstruction is a framework that recasts complex particle collision events as decay trees, enabling systematic resolution of kinematic and combinatoric ambiguities.
• It employs localized jigsaw rules to optimally assign visible and invisible objects, achieving up to 80–90% correct pairings and improved signal discrimination in LHC analyses.
• The method defines rest-frame observables such as mass and angular estimators, significantly enhancing statistical sensitivity in analyses like toponium reconstruction near the tt̄ threshold.

Recursive Jigsaw Reconstruction (RJR) is a systematic framework for resolving kinematic and combinatoric ambiguities in complex particle physics events, notably those involving invisible particles and multiple indistinguishable final-state objects. By formalizing event topology as a tree of two-body decays and recursively assigning reconstructed objects and missing energy along this tree, RJR enables the definition of rest-frame observables and mass/angle estimators that maximize information extraction and signal discrimination. It is implemented through a set of local, frame-specific algorithms termed "Jigsaw Rules" (JR's), and is realized in full in the RestFrames package and numerous LHC analyses, including recent work targeting toponium at the $\ttbar$ threshold (Desai et al., 27 Jan 2026, Jackson et al., 2017, Desai et al., 26 Jan 2026).

1. Conceptual Basis and Framework

At its core, Recursive Jigsaw Reconstruction abstracts each event as a directed decay tree in which each node corresponds to a decay stage (typically $1\to2$ or $2\to2$). Leaves at the ends of branches are assigned to either visible reconstructed objects (leptons, jets) or invisible particles (neutrinos, LSPs). Each decay node defines a natural rest frame, and the full set of Lorentz boosts mapping the laboratory frame to these rest frames captures all observable and unobservable kinematic degrees of freedom.

Kinematic ambiguities, such as unmeasured invisible momenta or ambiguous associations (e.g., which $b$-jet pairs with which lepton in dileptonic $\ttbar$), are resolved recursively at each node using local prescriptions. These Jigsaw Rules are applied sequentially, exploiting conservation laws, physical constraints (mass shell), and combinatoric minimizations. The final state is then fully reconstructed in "rest frames" with all visible and invisible momenta assigned.

This approach generates a basis of approximately uncorrelated mass and angle estimators that reflect the underlying physics of the hypothesized decay chain (Jackson et al., 2017).

2. Algorithmic Structure and Jigsaw Rules

The RJR procedure consists of several systematic steps:

1. Decay Tree Specification: Define the event topology as a binary (or more general) tree. For dileptonic $\ttbar$, the canonical tree is

$$H \to (t_a)(t_b),\quad t_a\to(b_a)(W_a),\ W_a\to(\ell_a)(\nu_a),\quad t_b\to(b_b)(W_b),\ W_b\to(\ell_b)(\nu_b)$$

where $H$ is the hard system (toponium or continuum), with four visible and two invisible leaves (Desai et al., 27 Jan 2026).

2. Combinatoric Assignment: For ambiguous partitions (e.g., $b$-lepton pairing), the rule is to select the pairing minimizing the sum

$$\chi^2_{\text{pairing}}=M^2(b_i,\ell_i)+M^2(b_j,\ell_j)$$

This prescription achieves correct assignment rates of 80–90% (Desai et al., 27 Jan 2026).

1. Rest Frame Definition: Assign rest frames corresponding to intermediate nodes and perform boosts. For each, the required boosts are computed from assigned four-vectors according to momentum conservation.
2. Invisible Momentum Partition: Parameterize the unknown invisible momenta. For two neutrinos, the partition is encoded by

$$\vec{p}_T^{\,\nu_a} = x \vec{p}_T^{\,\rm miss},\quad \vec{p}_T^{\,\nu_b} = (1-x)\vec{p}_T^{\,\rm miss}$$

with $x\in[0,1]$ (Desai et al., 27 Jan 2026).

1. Jigsaw Rule Selection: Choose local algorithms for resolving unknowns:
   • Equal-top-mass: Enforce $M_{t_a}(x) = M_{t_b}(x)$
   • Equal-W-mass: Enforce $M_{W_a}(x) = M_{W_b}(x)$
   • Minimize sum of squared top masses: $\min_{x} M_{t_a}^2(x) + M_{t_b}^2(x)$
   • Minimize top-mass difference: $\min_{x} |M_{t_a}(x)-M_{t_b}(x)|$ Each yields an equation or minimization for $x$, followed by solving for longitudinal $p_z$ components via mass-shell constraints (Desai et al., 27 Jan 2026, Jackson et al., 2017).
2. Recursive Construction: At each subsequent decay step, the same logic applies—partition, assign, and boost, based solely on previously fixed quantities.

The RestFrames implementation automates tree construction, JR selection, assignment, and all boost operations (Jackson et al., 2017).

3. Applications: Toponium and Beyond

Recent work has focused on reconstructing toponium near the $\ttbar$ threshold at the LHC (Desai et al., 27 Jan 2026, Desai et al., 26 Jan 2026). The RJR approach is employed as follows:

• Event Generation and Selection: Hard events are produced in MadGraph5_aMC@NLO using the NRQCD toponium model, with subsequent showering, hadronization, and object reconstruction (minimum $p_T>25$ GeV, $|\eta|<2.5$ for leptons and $b$-jets).
• Pairing and Tree Construction: $b$-leptons are paired by minimum invariant mass sum, and the decay tree is built according to the hypothesized toponium structure.
• Resolution of Invisible Momenta: All four RestFrames jigsaw rules for partitioning the missing transverse momentum are tested.
• Boosts and Observable Calculation: After all assignments, events are successively boosted into the appropriate rest frames for further calculation of invariant masses and angles.

Equivalent methodologies have been deployed for searches in compressed SUSY scenarios, electroweakino production, and other new-physics signatures involving multiple invisible particles (Santoni, 2017, Collaboration, 2018).

4. Kinematic and Angular Observables

A major strength of RJR is the ability to define observables in physically relevant rest frames. For toponium, new angular discriminants have been constructed (Desai et al., 27 Jan 2026):

• Modified Helicity ($\cos\theta_{nHel}$): Computed as the dot product of lepton unit-vectors after boosting to the parent top frames, with one $p_z$ coordinate sign-flipped:

$\cos\theta_{nHel} = \hat{p}_a \cdot \hat{p}_b$

with $\hat{p}_a$, $\hat{p}_b$ constructed from the lepton momenta as prescribed.

• Azimuthal Opening ($\Delta\phi(t,\bar t)$): The absolute difference in azimuthal angle between the two reconstructed top quarks.

Distributions in the $(\cos\theta_{nHel},\Delta\phi)$ plane exhibit enhanced discrimination between signal (pseudoscalar toponium) and Standard Model backgrounds, resulting in significance gains of up to 16% over existing methods (Desai et al., 27 Jan 2026, Desai et al., 26 Jan 2026).

5. Performance Metrics and Validation

Performance of RJR-based reconstructions is quantified by:

• Statistical Significance: Defined as $\mathcal S = S/\sqrt{S+B}$, where $S$ and $B$ are signal and background yields after full event selection and binning in angular observables.
  • For toponium, replacement of standard angular variables with the new rest-frame observables increases significance from $\sim11.9\sigma$ to $\sim13.0\sigma$ after standard cuts, and to $14.1\sigma$ for the most optimized bin (Desai et al., 27 Jan 2026).
• Resolution: The reconstructed $M_{t\bar t}$ mass peak has a resolution of $\sim10\ \rm GeV$ using the equal-top-mass jigsaw rule, outperforming standard methods.
• Background Separation: The near-orthogonality of RJR observables and their low cross-correlation allow for robust multivariate analysis or simple cut-based selections with minimal loss in performance (Jackson et al., 2017).

Validation steps include detailed cutflows, sideband modeling, and comparison to truth-level generator observables (Desai et al., 27 Jan 2026, Collaboration, 2018).

6. Practical Implementation and Systematics

Replication or extension of RJR analyses for LHC Run 3 requires:

• Event Generation: MadGraph5_aMC@NLO with NRQCD models, Pythia8 for showering, FastJet for jet clustering, with appropriate PDF sets.
• Object Selection: Two high-quality oppositely-charged leptons, $\geq2\ b$-tagged jets, and kinematic selection cuts such as $E_T^{\rm miss}>40$ GeV, $M_{t\bar t}<550$ GeV.
• Software: The RestFrames package (v1.3+) is used for decay tree definition and application of jigsaw rules.
• Systematic Uncertainties: Key sources include jet energy scale/resolution, lepton and $b$-tagging efficiencies, and theoretical uncertainties in modeling of the signal-line shape, parton distributions, and MET resolution (Desai et al., 27 Jan 2026).

A summary of practical ingredients required for reproducibility is shown below:

Ingredient	Description	Example/Setting
Event generator	MadGraph5_aMC@NLO, NRQCD toponium model	(Fuks et al., 2024)
Shower/hadronization	Pythia 8.3
Jet clustering	anti-$k_T$, $R=0.4$ via FastJet
Object selection	$p_T>25$ GeV, $|\eta|<2.5$, $\geq$2 $b$-jets, 2 OS leptons
Software	RestFrames v1.3+, MadAnalysis5, ROOT

Systematic issues such as real $b$-tag efficiency ($\sim70$–80% in data), PDF/scale variations, and modeling of the NRQCD Green’s function are recognized as critical for future LHC real-data analyses (Desai et al., 27 Jan 2026).

7. Extensions and Broader Context

RJR is applicable far beyond toponium. It provides a generalizable algorithmic infrastructure for any hadron-collider process involving multiple invisible objects, combinatoric ambiguities, and the need for rest-frame–specific kinematic variables. Demonstrations have included:

• Compressed SUSY final states, exploiting initial-state recoil (Jackson et al., 2016)
• Electroweakino production and decay in multilepton plus MET events (Santoni, 2017, Collaboration, 2018, Resseguie, 2019)
• Complex topologies involving multiple decay chains and invisible objects; resolution of jet-to-parent and lepton-to-parent assignments via recursive combinatoric minimization (Jackson et al., 2017)

Limitations include the increased algebraic complexity for many-invisible systems, dependence on good detector resolution, and possible sensitivity to assignment errors in combinatorics. However, RJR observables have shown robustness and superior background rejection properties compared to traditional global variables.

In summary, Recursive Jigsaw Reconstruction realizes a modular, frame-centric, and algorithmic approach to event kinematics that systematically addresses both invisible momentum and object-assignment ambiguities. Its efficacy in reconstructing heavy fermion bound states, new-physics cascades, and compressed-mass scenarios establishes it as a central tool for current and future collider analyses (Desai et al., 27 Jan 2026, Desai et al., 26 Jan 2026, Jackson et al., 2017).