GENEVA System Framework

• GENEVA System is a multi-domain framework integrating high-precision collider events, branching narrative generation, and event argument extraction for NLP.
• It combines NNLO+PS precision in collider physics with SCET-based resummation, nonlocal subtractions, and modular parton shower matching for robust accuracy.
• Its modular design and rigorous validation across high-energy physics, interactive narrative tools, and NLP benchmarks ensure reproducible, actionable outcomes.

GENEVA System

The term "GENEVA System" encompasses a family of event-generation, benchmarking, and data-processing frameworks that arise in three distinct research domains: perturbative quantum field theory and collider event simulation, computational narrative generation for branching dialogue systems, and large-scale benchmarking for event argument extraction in natural language processing. The most influential and widely cited use is as the fully-differential event generation and resummation framework for high-precision collider physics, developed and maintained by multiple theory collaborations since 2012. In parallel, the GENEVA moniker denotes distinct systems in computational narrative design and NLP benchmarking. Each instance shares a commitment to algorithmic rigor, modular design, and reproducible output, but the technical realization, scope, and validation targets are determined by field-specific constraints and objectives.

1. GENEVA in High-Precision Collider Event Simulation

The GENEVA Monte Carlo framework is a flexible, fully-differential event generation platform designed to combine state-of-the-art fixed-order (FO) perturbative QCD calculations with higher-order logarithmic resummation of jet-resolution variables, interfaced consistently to parton showers and nonperturbative hadronization models. Its central aim is to deliver exclusive hadron-level event samples for collider processes at next-to-next-to-leading order plus parton shower (NNLO+PS) accuracy, with systematically improvable logarithmic control (e.g., resummation up to N³LL) and robust uncertainty quantification (Cridge et al., 2021, Alioli et al., 2013, Marinelli, 2023, Alioli et al., 2019, Alioli et al., 2022, Gavardi et al., 20 May 2025, Alioli et al., 2021, Alioli et al., 15 Apr 2025, Alioli et al., 11 Jul 2025, Alioli et al., 2023, Alioli et al., 2020).

Key architectural modules include:

• A FO matrix-element engine that builds cross sections up to NNLO (for colour singlets, also V+jet and heavy-flavour channels), interfaced to amplitude providers (OpenLoops, MATRIX, etc.).
• Resummation kernel for jet resolution variables (notably N-jettiness T₀, T₁ or p_T), leveraging SCET-based factorization and RG evolution to NNLL′ or N³LL, including careful scale profile prescriptions.
• Analytically defined splitting and mapping functions that allow event-by-event promotion of resummed cumulants to fully differential phase-space distributions at higher multiplicity.
• Modular interface to industry-standard parton showers (Pythia 8, Sherpa, Dire), which is constrained to avoid double-counting with the analytic resummation and preserve exclusive cross sections.
• Operator and data-flow scheduling to enable simultaneous reweighting, scale-variation envelope generation, and power-correction assignments.

The framework achieves a unitary matching between FO and resummed predictions using "add-subtract" master formulas. For example, for 0-jet bins:

$$\frac{d\sigma_0}{d\Phi_0}(T_0^{\mathrm{cut}}) = \frac{d\sigma^{\mathrm{NNLL'}}}{d\Phi_0}(T_0^{\mathrm{cut}}) - \left[ \frac{d\sigma^{\mathrm{NNLL'}}}{d\Phi_0}(T_0^{\mathrm{cut}}) \right]_{\mathrm{NNLO}_0} + \text{FO terms}$$

This structure guarantees the correct logarithmic behavior at $T_0 \ll Q$, exact NNLO normalization for inclusive bins, and positive-weight event generation for reliable shower/hadronization interfacing (Cridge et al., 2021, Alioli et al., 2013, Alioli et al., 2012, Alioli et al., 2023, Alioli et al., 2020).

2. Resolution Variables and SCET-Based Resummation

GENEVA employs careful selection of infrared-safe resolution variables to partition phase space into exclusive jet bins, crucial for the matching of fixed-order and resummed results. The preferred variables and associated resummation schemes include:

• N-jettiness ($T_N$ or $\tau_N$): For $N = 0$, $T_0$ (beam thrust) cleanly separates 0-jet from higher-multiplicity configurations. For $N = 1$, $T_1$ is used for 1-jet bin boundaries. The definition is universal but admits process-dependent recombination schemes and mapping functions (Cridge et al., 2021, Alioli et al., 2013, Alioli et al., 2012).
• Transverse momentum ($q_T$): For colour-singlet systems, GENEVA supports $q_T$ as the 0-jet resolution variable, interfaced to N³LL resummation via SCETlib and the RadISH formalism. This extension enables more efficient matching to $p_T$-ordered parton showers and improves theoretical control on the $q_T$ spectrum (Gavardi et al., 20 May 2025, Alioli et al., 2021).
• SCET factorization: The cross section for a process with hard scale $Q$ factorizes as

$$\frac{d\sigma^{\mathrm{SCET}}}{d\Phi_0\, dT_0} = \sum_{i,j} H_{ij}(\Phi_0, \mu) \int dt_a dt_b\, B_i(t_a, x_a, \mu)B_j(t_b, x_b, \mu) S\bigl(T_0 - (t_a + t_b)/Q, \mu\bigr)$$

where the hard, beam, and soft functions are evolved from their canonical scales to a common scale using RG kernels $U_H$, $U_B$, $U_S$. Resummation up to NNLL′ (and N³LL for $q_T$) is realized by including cusp anomalous dimensions up to three loops and noncusp for two loops, together with full matching corrections (Cridge et al., 2021, Alioli et al., 2022, Gavardi et al., 20 May 2025).

• Matching scheme: The resummed spectrum is subtracted at the relevant FO order from the pure FO cross section to avoid double counting. Event-by-event mappings using normalized splitting probabilities maintain the full singular structure of the matrix elements across all relevant limits (Alioli et al., 2023, Marinelli, 2023).

3. Parton Shower Matching, Subtractions, and Nonperturbative Effects

GENEVA's MC cross sections are unitarily matched to parton showers (Pythia 8, Sherpa, Dire, etc.) under explicit constraints:

• Shower vetoing: Parton showers are vetoed from emitting radiation above the resolution variable cut in the bin of origin (e.g., $T_0^{\text{cut}}$ or $q_T^{\text{cut}}$), guaranteeing that the MC-level 0/1-jet bins are not spoiled and resummed log accuracy is preserved (Cridge et al., 2021, Alioli et al., 2022, Gavardi et al., 20 May 2025, Marinelli, 2023).
• Unitarity and normalization: By construction, the total cross section in each bin remains exactly normalized after showering and hadronization, ensuring NNLO accuracy for inclusive observables (Alioli et al., 2019).
• Nonlocal NNLO subtractions: Recently, GENEVA incorporates a nonlocal subtraction scheme employing expanded resummed cross sections as local counterterms, supplemented by projection-to-Born (P2B) methods that restore fiducial power corrections below numerical IR cutoffs (Alioli et al., 15 Apr 2025).
• Fiducial observables and P2B: For processes with nontrivial generator cuts or acceptance definitions, power corrections induced by mismatch between real emission and Born phase space are recovered, ensuring that agreement with external codes remains at the per-mille level for fiducial cross sections (Alioli et al., 15 Apr 2025).

Hadronization and multiple parton interactions are handled by the underlying PS backend per its default or tuned prescription. GENEVA does not supply an intrinsic nonperturbative model for soft radiation; all such effects arise from the external shower and hadronization modules (Alioli et al., 2013, Alioli et al., 2016).

4. Validation, Phenomenological Reach, and Representative Results

GENEVA's modular structure enables broad coverage of LHC precision processes, including Drell–Yan, vector-boson plus photon, Higgsstrahlung, diphoton, heavy-flavour initiated, and double Higgs production channels. Validation practices include:

• Comparison to independent FO codes: Partonic predictions at the NNLO level are benchmarked against reference implementations (e.g., Matrix, NNLOJET, SusHi, MCFM), for all key kinematic distributions (Cridge et al., 2021, Alioli et al., 15 Apr 2025, Alioli et al., 2022, Gavardi et al., 20 May 2025).
• Showering and nonperturbative effects: Theoretical uncertainties are estimated via correlated scale variations (FO and resummation regions), nonperturbative shifts are assessed by comparison with various PS recoil schemes, and fiducial observables are compared before and after shower+hadronization (Alioli et al., 2020, Alioli et al., 2022).
• Agreement with data: Geneva-predicted distributions agree with ATLAS and CMS data at 7, 13, and 14 TeV within the expected scale and PDF uncertainties for $p_T$, rapidity, and resolution-variable spectra; any systematic reach of theory bounds is consistently interpreted as arising from missing electroweak corrections or power-suppressed QCD terms (Cridge et al., 2021, Alioli et al., 2020, Alioli et al., 2019).

5. Advancements: Transverse Variables, Subtraction Schemes, and Mass Effects

Recent developments in GENEVA include:

• $q_T$ resummation and transverse-measure jettiness: Full embedding of N³LL $q_T$ resummation for colour-singlets enables more precise matching to $p_T$-ordered showers, improved stability across the $q_T \rightarrow 0$ region, and opens the path to NNLO+PS for heavy-flavour and V+jets final states (Gavardi et al., 20 May 2025, Alioli et al., 2021).
• Nonlocal subtractions and P2B integration: Adoption of nonlocal subtraction schemes utilizing resummation-expanded counterterms, with restoration of power corrections via projection-to-Born, improves numerical stability, maintains dynamical cut consistency, and readies the framework for N³LO corrections (Alioli et al., 15 Apr 2025).
• Inclusion of top-mass effects in di-Higgs channels: For double Higgs production, stepwise improvements have progressed from the heavy-top limit (HTL) to the inclusion of all known finite $m_t$ effects, using amplitude reweighting, LO and NLO exact matrix elements, and re-projection algorithms. This augments the realism of predicted rates and shapes in key BSM-sensitive observables (Alioli et al., 11 Jul 2025, Alioli et al., 2022).
• Splitting-function and timelike-logarithm refinements: Improved splitting functions for phase-space mappings now yield vanishing nonsingular remainders at small $T_0$, and timelike logarithms in hard functions are resummed by running on complex scales, reducing $\pi^2$-driven uncertainties (Alioli et al., 2023).

6. GENEVA in Non-Physics Domains

• Branching Narrative Generation: In (Leandro et al., 2023), GENEVA (also called GRIM) is a two-stage system leveraging GPT-4 for graph-based, dynamically editable branching narrative generation for dialogue-driven RPGs. The pipeline synthesizes story beats into directed acyclic graph (DAG) structures, supports constraint-based DAG editing and designer-in-the-loop workflows, and exposes a domain-specific descriptive language for constraints and narrative arcs. Qualitative evaluation highlights scalability, flexibility, and modifiability, but practical limitations arise from model grounding and prompt engineering constraints.
• Event Argument Extraction (NLP Benchmark): In (Parekh et al., 2022), GENEVA denotes a resource and evaluation protocol for cross-event generalization in event argument extraction. A FrameNet-derived ontology of 115 event types and 220 roles (37% non-entity roles) anchors four generalization-focused test suites (low-resource, few-shot, zero-shot, cross-type transfer). Benchmarks quantify the dropoff in F1 for state-of-the-art models on non-entity roles and highlight the challenge posed by a large, long-tail ontology.

7. Summary Table: GENEVA System Across Domains

Context	Application	Core Function	Characteristic Output
Collider simulations	High-precision QCD event gen.	FO+resummation+parton-shower matching to NNLO+PS	Unweighted events, theory+exp comparison
Narrative generation	RPG narrative graphing	DAG-based story beats generation with LLMs, constraints	JSON/D3.js story DAG, interactive edit
NLP benchmarking	Event argument extraction	Large ontology, multi-suite generalization benchmarks	F1 benchmarks, generalization metrics

Each instantiation of GENEVA provides field-specific algorithmic innovations and validation methods. In collider theory, it establishes a rigorous standard for interfacing fixed-order and all-orders resummed calculations with modern parton showers, enabling the highest fidelity phenomenology for the LHC and future colliders (Cridge et al., 2021, Marinelli, 2023, Gavardi et al., 20 May 2025, Alioli et al., 15 Apr 2025).