Automated Luminosity Optimization for Future Electron-Positron Colliders

Project Overview

In this project, originally conceived as part of Stanford’s AA 222/CS 361 - Engineering Design Optimization and awarded the best final project award for the 2023-2024 version of the course, we developed a novel optimization framework for maximizing the collision rate, or “luminosity”, at future electron-positron colliders while maintaining controlled background levels.

Technical Approach

In this study, we developed a new luminosity optimization strategy based on the following:

Surrogate Modeling

• Developed a Gaussian Process surrogate model to approximate the beam-beam enhancement factor $H_{D}$, which is a non-analytic contribution to the luminosity.
• Trained the on PIC (Particle-In-Cell) simulation points with careful hyperparameter optimization

Optimization Framework

• Constructed a constrained optimization problem over six beam parameters:
  • Bunch population
  • Horizontal/vertical emittances
  • Horizontal/vertical beta functions
  • Bunch length
• Implemented both single-objective constrained optimization and multi-objective approaches
• Applied bounds based on technical feasibility and background constraints

Key Results

The methodology demonstrated significant improvements across multiple proposed collider designs:

• Achieved luminosity gains of 22-35% for most collider scenarios while maintaing or reducing beam-induced background levels
• Identified optimal operating points within technically feasible parameter ranges
• Discovered a luminosity saturation effect at specific background levels

Impact

The optimization framework provides:

• Automated tool for collider parameter optimization
• Potential for increased physics reach through higher collision rates
• Framework for evaluating design trade-offs between luminosity and backgrounds across different collider proposals (e.g. CLIC, ILC, C$^3$)

Future Developments

We are still working on improving the analysis, focusing on:

• Further validation against full detector simulations
• Integration with detailed beam dynamics studies
• Extension to additional collider scenarios and constraints
• Refinement of the surrogate model with additional training data

Overview

This study presents a comprehensive analysis of beam-beam backgrounds for the Cool Copper Collider (C³), a proposed electron-positron linear collider targeting precision Higgs physics at center-of-mass energies of 250 and 550 GeV. The work validates the compatibility of existing detector concepts with C³ beam parameters and provides a reusable simulation framework for future collider studies.

Paper: JINST 21 (2026) P02024, arXiv:2511.01075

Code: GitHub Repository

Key Findings

Detector Compatibility

• Evaluated the Silicon Detector (SiD) concept across three C³ operational scenarios: baseline, power-efficiency, and high-luminosity modes
• The SiD detector remains compatible with C³ operations without requiring substantial design modifications
• Quantified occupancy levels from incoherent pair production and hadron photoproduction backgrounds in all detector subsystems

Background Characterization

• Full simulation of beam-beam interactions using the GUINEA-PIG PIC simulator
• Detailed modeling of photoproduced hadrons from beam-beam interactions
• Comprehensive tracking of background particles through the SiD detector using the Key4hep software stack
• Assessment of hit densities and timing distributions in the vertex detector, silicon tracker, and calorimeters

Modular Simulation Framework

A key contribution of this work is the development of a modular simulation framework built on the Key4hep ecosystem. This framework provides:

• A versatile toolkit adaptable to different collider proposals and beam parameter configurations
• Standardized analysis methodology for background studies
• Easy integration with various detector geometries and simulation tools
• Reusable components for the broader future collider community

The framework is publicly available on GitHub and can be adapted for background studies at other proposed electron-positron colliders such as ILC, CLIC, or FCC-ee.

Impact

This study provides essential input for:

• Validating detector technologies for C³ conditions
• Guiding detector timing and readout specifications
• Supporting the physics case for C³ as a viable Higgs factory option
• Enabling similar studies at other future collider proposals through the modular framework

Environmental Impact Analysis of Future Particle Colliders

Research Context

In this study we evaluated the environmental impact of next-generation particle physics facilities, focusing on the Cool Copper Collider (C³) proposal, addressing a critical challenge in modern physics: balancing scientific advancement with environmental responsibility.

Key Findings and Impact

The study demonstrated several significant advantages of the C³ design:

• Compact 8-kilometer footprint significantly reduces embodied carbon
• Cut-and-cover construction method offers substantial environmental benefits compared to deep tunneling

We also developed a new metric for comparing different collider proposals based on:

• Physics research potential
• Energy consumption requirements
• Carbon footprint from construction
• Operational environmental impact

Future Implications

This work contributes to the broader discussion of sustainability in big science projects and helps establish environmental impact as a key metric in future accelerator design decisions.

Our research demonstrates how modern accelerator design can balance scientific ambition with environmental responsibility. The framework developed in this study provides a quantitative basis for evaluating and comparing different collider proposals from a sustainability perspective, while maintaining focus on their primary physics goals.

Searching for New Physics Through Paired Dijet Resonances at the LHC

Project Overview

As an undergraduate student, I contributed to a search for ultra-heavy new particles with the full Run 2 data of the CMS experiment, representing 138 fb⁻¹ of proton-proton collisions at 13 TeV. In that analysis, we looked for new massive particles beyond the Standard Model, in the form of TeV-scale resonances $Y$ decaying to two identical heavy particles $X$, with each one of them decaying hadronically to two jets - collimated sprays of particles that are the hallmark signatures of quarks and gluons. This creates a distinctive four-jet signature: $pp \rightarrow Y \rightarrow XX \rightarrow (jj)(jj)$. A few candidate events were observed in the data, with a maximum local (global) significance of 3.9(1.6)$\sigma$.

Technical Innovations

The analysis employed sophisticated techniques to:

• Distinguish signal from the overwhelming QCD background
• Handle complex four-jet final states
• Develop model-independent search strategies
• Set limits on both resonant and non-resonant production scenarios in low-statistics regimes, where asymptotic appproximations are no longer valid

Results

This search achieved several significant milestones:

• First LHC limits on resonant pair production of dijet resonances via high-mass intermediate states
• Extended previous limits on supersymmetric particle scenarios
• Excluded diquark masses below 7.6 TeV in specific model scenarios
• Placed new constraints on top squark pair production in R-parity-violating supersymmetry

The analysis also revealed some tantalizing potential hints of new physics:

• Two remarkable events were observed with four-jet masses of 8 TeV and average dijet masses of 2 TeV
• These events showed a local significance of 3.9 standard deviations (reduced to 1.6σ when accounting for the look-elsewhere effect)
• Another interesting excess was observed at an average dijet mass of 0.95 TeV, with a local significance of 3.6σ (2.5σ global)

Differentiable and GPU-Accelerated Beam-Beam Simulation

Research Context

Beam-beam interactions set the physics reach of every high-energy lepton collider: the field of each bunch distorts the opposing bunch, and that pinch determines luminosity, beamstrahlung energy spread, and pair-production backgrounds. The luminosity enhancement from this pinch typically has no closed form and must come out of expensive simulations using Particle-In-Cell (PIC) codes.

GUINEA-PIG has been the community standard for over two decades. It is an excellent forward simulator, but it exposes no derivatives and runs on a single CPU core. Collider design therefore proceeds by parameter scans, finite differences, or derivative-free optimizers — none of which scale well into the high-dimensional space of all possible beam configurations. I am approaching this bottleneck along two threads.

GuineaGrad: differentiable beam-beam simulation

GuineaGrad reimplements the deterministic core of GUINEA-PIG in JAX, exposing automatic derivatives with respect to beam parameters. The difficulty lies not as much in the reimplementation but the differentiability, since several processes in the simulation pipeline (scatter-type charge deposition, FFT-based Poisson solver, stochastic sampling for quantum processes etc.) break automatic differentiation in different ways.

Guinea-Pig X (GPX): a CUDA port for throughput

Guinea-Pig X is a CUDA port of GUINEA-PIG. Across reference configurations for several colliders, it delivers roughly an order-of-magnitude end-to-end speedup, while maintaining statistical parity with the legacy code. First results were presented as a poster at the US FCC Workshop 2026. Manuscript in preparation.

Evaluating Detector Design Impact on Jet Flavor Tagging for Future Colliders

Research Context

This study presents a comprehensive evaluation of jet flavor tagging performance for proposed detector concepts at future electron-positron colliders. Jet flavor tagging—the ability to identify the type of quark or gluon that initiated a jet—is crucial for precision measurements of Higgs boson properties, particularly its couplings to bottom, charm, and strange quarks.

Technical Approach

We developed a unified analysis framework to enable fair comparisons between detector concepts:

• Detectors evaluated: SiD (ILC), IDEA (FCC-ee), and FCCeeDetWithSiTracking
• Simulation: Delphes fast simulation with advanced modules for particle identification (PID), including time-of-flight measurements and ionization cluster counting
• Algorithm: ParticleNet, a Graph Neural Network architecture adapted for e⁺e⁻ collision environments, trained on Higgsstrahlung events (e⁺e⁻ → ZH)

Key Results

Detector Comparison

• All three detector concepts achieve excellent flavor tagging performance, representing an order of magnitude improvement over current LHC experiments
• IDEA and FCCeeDetWithSiTracking demonstrate superior discrimination, particularly for strange-quark tagging, due to dedicated PID capabilities
• For s-tagging, SiD’s mistag rates are 1.4–2.5× larger than detectors with PID—highlighting the critical role of kaon identification

Detector Parameter Studies

We systematically investigated variations in the SiD detector design:

• Vertex detector geometry: Modest degradation in heavy flavor tagging with increasing radial distance of first layer (10-16 mm range)
• Calorimeter resolution: Flavor tagging performance shows remarkable robustness against ECAL and HCAL energy and spatial resolution variations
• Practical implications: Results suggest potential cost optimization opportunities in calorimeter systems without significantly compromising flavor tagging capabilities

Impact and Future Directions

This work provides the first systematic, apples-to-apples comparison of jet flavor tagging across multiple future collider detector concepts using a consistent framework. The flexible analysis pipeline enables rapid re-evaluation as detector designs evolve, providing valuable feedback for optimization studies. While these fast simulation results offer important insights, validation with full Geant4-based simulation remains essential for confirming the observed trends and informing final detector design decisions for future e⁺e⁻ colliders.

Geometric Signatures of Conceptual Reorganization

Research Context

Scientific knowledge is organized as a network of interconnected concepts, and that organization occasionally undergoes rapid reorganization. Understanding these episodes has traditionally relied on historical analysis and qualitative interpretation, while existing quantitative approaches characterize scientific change through citation networks, bibliometric indicators, or topic evolution. Those methods describe structures external to conceptual organization itself, and inherit the biases of external metadata.

This project asks a complementary question: can conceptual reorganizations be detected directly in the geometry of document embedding spaces? The claim behind it is a shift in how embedding spaces are used — not merely as representations for retrieval or semantic similarity, but as a quantitative scientific observable that can be measured, perturbed, statistically validated, and compared across independent historical case studies.

Method: Counterfactual Ablation

The framework represents scientific documents and natural-language concept descriptions in a common embedding space, then measures the structural role of a concept by removing it. For a candidate concept, all associated documents are removed, the geometry is recomputed, and the resulting perturbation is compared before and after a candidate pivot year. If a concept genuinely reorganized an existing body of knowledge, removing it should perturb the post-emergence geometry far more than the pre-emergence geometry.

Because a geometric signature is easy to produce by accident, the framework is validated against several null hypotheses: removing temporally matched random documents instead of the target, scrambling concept labels while preserving the corpus’s temporal structure, a single-document jackknife, and a look-elsewhere effect correction.

Outlook

The immediate contribution is retrospective: establishing that historically recognized conceptual reorganizations leave measurable geometric signatures, and that embedding spaces can be treated not merely as semantic representations but as scientific objects of study. The longer-term ambition is to be able to detect future conceptual revolutions in science and beyond. However, whether systems that recognize the signatures of past reorganizations can help identify emerging ones remains open; this work is a prerequisite for asking that question, not an answer to it.

With A. Schwartzman (SLAC Physics), C. Chafe (Stanford Music), and T. A. Ryckman (Stanford Philosophy). Preprint: arXiv:2609.14917. Interactive results on the companion website.

Studying High-Energy Higgs Bosons in Association with Vector Bosons

Analysis Overview

During my first research rotations at Stanford, and working with the SLAC ATLAS group, I contributed to the first measurement attempt of highly energetic (“boosted”) Higgs bosons produced alongside $W$ or $Z$ bosons (collectively termed as vector bosons) with the ATLAS experiment at the Large Hadron Collider, focusing on events where all particles in the final state manifest as collimated sprays of hadrons, known as jets.

Technical Approach

The analysis employs several advanced experimental techniques to identify and measure these rare processes:

Particle Reconstruction

• The decay products of both the Higgs and vector bosons are captured within single large-radius jets due to their high momentum
• Sophisticated jet substructure techniques analyze the internal patterns of these jets to distinguish between signal and background processes
• Specialized algorithms identify jets containing b-quarks from Higgs boson decays

Background Estimation

• The dominant background comes from generic quantum chromodynamics (QCD) multijet production
• These background processes are measured directly from collision data rather than relying on simulation
• A likelihood fit to the jet mass distribution is used to separate signal from background

Measurement Strategy

The analysis measures the production rate:

• Inclusively across all kinematic regions
• Differentially in three ranges of Higgs boson transverse momentum:
  • 250-450 GeV
  • 450-650 GeV
  • Above 650 GeV

Results and Future Work

The measurement yields a signal strength relative to the Standard Model prediction of $\mu = 1.4+1.0-0.9 $, which translates to a production cross-section of $3.1 \pm 1.3(\mathrm{stat})^{+1.8}_{-1.4}(\mathrm{syst}) \ \mathrm{pb}$.

This measurement opens new possibilities for studying the Higgs boson in previously unexplored kinematic regimes and provides crucial insights into the Higgs mechanism at high energies, complementing existing measurements in other decay channels.

Although the sensitivity of this analysis was not sufficient to measure Higgs production in this phase-space with adequate statistical significance, future improvements in the analysis techniques, as well as additional data collected during Run 3 of the LHC, will enable a more precise determination of the Higgs boson properties at high energies.

Search for Boosted Higgs Bosons in Associated Production with Vector Bosons

Project Overview

This analysis builds upon our previous ATLAS study of highly energetic (“boosted”) Higgs bosons produced alongside $W$ or $Z$ bosons, incorporating significant improvements in both dataset size and analysis techniques. The study focuses on events where all final state particles manifest as collimated sprays of particles called “jets”, offering unique sensitivity to new physics effects at high energies.

Technical Advancements

The analysis features several major improvements:

Enhanced Dataset

• Combination of full Run 2 data with partial Run 3 statistics
• More than double integrated luminosity compared to previous analysis

Advanced Reconstruction

• Implementation of latest b-jet identification algorithms with higher background rejection
• Novel b-jet mass regression techniques for improved mass resolution
• Refined jet substructure methods for better signal-background discrimination

Methodology Improvements

• Sophisticated multivariate analysis techniques for event selection
• Enhanced background estimation methods
• Potential inclusion of vector-boson fusion (VBF) production category
• Implementation of dedicated high-transverse-momentum $p_{T}$ regime optimizations

Theoretical Interpretation

• Extended analysis framework incorporating Effective Field Theory (EFT) operators
• Constraints on Wilson coefficients sensitive to high-momentum Higgs production
• Investigation of possible new physics contributions in the high-energy regime

Physics Goals

The analysis aims to:

• Improve precision of $VH(b\bar{b})$ production cross-section measurement
• Probe potential deviations from Standard Model predictions at high energies by constraining relevant EFT operators through the unique sensitivity of the boosted topology

Impact

This measurement will:

• Enhance our understanding of Higgs boson interactions at high energies
• Provide crucial inputs for constraining possible new physics scenarios
• Demonstrate the capability of advanced reconstruction techniques

The analysis leverages the latest experimental and theoretical developments to maximize sensitivity to both Standard Model measurements and potential new physics effects.

Beam Physics Studies for the Cool Copper Collider

Research Significance

A high-energy electron-positron collider is widely recognized as the crucial next step for detailed studies of the Higgs boson and other fundamental particles. As part of my graduate studies, I worked on analyzing the luminosity and beam-beam interaction characteristics of the Cool Copper Collider (C³), a novel linear collider concept.

This study addresses a fundamental challenge in particle physics: achieving high collision rates while managing beam-induced backgrounds in future electron-positron colliders. The work provides crucial validation of the Cool Copper Collider (C³) design’s capabilities relative to other proposed facilities.

Technical Analysis

The research examines several critical aspects of collider performance:

Beam-Beam Interactions

• Analysis of electromagnetic forces between nanometer-scale particle bunches
• Evaluation of beamstrahlung effects on luminosity distribution
• Quantification of beam-induced background particle production
• Impact on detector design and physics measurement precision

Luminosity Optimization

• Comprehensive parameter space exploration for beam characteristics
• Development of optimization strategies for collision rates
• Assessment of trade-offs between luminosity and background levels
• Comparison with other linear collider proposals

Key Findings

In this study we demonstrated that C³ can:

• Achieve equivalent or superior collision rates compared to alternative proposals
• Maintain controlled levels of beam-induced backgrounds
• Provide robust physics performance across its energy range

Broader Impact

This work serves as an additional Validation of the C³ design concept and lays out a common framework for comparing future collider proposals

Our analysis showed that C³ can achieve the same or higher collision rates compared to other proposals, while maintaining lower beam-induced background fluxes. These findings position C³ as a promising candidate for future particle physics research, leveraging technological advances in accelerator physics developed across various linear collider initiatives.