Come out and support OR Ph.D. Students as they practice their talks for the upcoming INFORMS Annual Conference.

Did You Miss It?

Don’t worry. You can Watch it here.
Passcode: #2%kENkQ

Students

• Joshua Grassel
• Regan Richardson
• Hou-An Chen
• Cameron Lisy

Talk Title, Abstract and Biographies

Joshua Grassel

Stress Testing the Numerical Stability of LP and MIP Solvers

Modern solvers for linear programming (LP) and mixed-integer programming (MIP) are indispensable tools in optimization, yet they can yield incorrect results due to roundoff errors from floating-point arithmetic. These errors, such as reporting suboptimal solutions as optimal or misidentifying problems as infeasible, cast doubt on solver reliability. Their impact is particularly acute in domains requiring high numerical precision, such as biochemical modeling, astrophysics simulations, compiler optimization, and mathematical proofs. To mitigate these issues, various strategies have been developed, including presolve scaling, algorithm tuning, and exact arithmetic subroutines. However, current evaluation practices rely heavily on standard problem libraries, which may not expose numerical weaknesses in a consistent or generalizable way. This research proposes a novel framework for generating synthetic LP instances that systematically stress test solvers’ numerical robustness. Inspired by worst-case complexity constructions, the framework allows control over key parameters, such as variable count and coefficient precision, enabling exploration of solver behavior under varying numerical stress. Each generated problem is guaranteed to have an exactly representable input data in floating-point, ensuring that any observed errors originate from solver internals rather than input encoding. This approach provides a systematic method for testing solver reliability and benchmarking numerical stability. The presentation showcases results from applying this framework to a range of commercial and open-source solvers, highlighting both their strengths and limitations when facing numerically challenging LP instances.

Joshua Grassel earned his Bachelor’s degree in Industrial Engineering from California Polytechnic State University in 2021. During his undergraduate studies, he completed three internships—two in manufacturing engineering with defense subcontractors and one in logistics engineering with a wine and spirits bottling and distribution company. Joshua began his PhD in Industrial Engineering at Arizona State University in Fall 2021 and transferred to the Operations Research PhD program at NC State in Fall 2023. His research has spanned topics such as the wisdom of crowds and sustainable solid waste management. Currently, he is focusing on optimization algorithms with a special interest in numerical precision. Outside of academia, Joshua enjoys cooking, playing board games, running, cycling, and exploring the outdoors with his wife and dog.
 
Regan Richardson

Examining the Unintended Impacts of Disruptions to the Illicit Drug Market on Violent Crime and Overdose Death

The use of illicit substances remains an ongoing and widespread part of society, with the illegal drug market estimated to be 1% of the total global trade. The true scale of this market is unknown because it is an illicit underground trade that is poorly understood, dynamic, and complex. While there have been trends towards decriminalization or legalization of marijuana in the United States, the overarching policy and strategy has been consistently been to allocate large amounts of resources to law enforcement agencies to disrupt and dismantle illicit distribution networks. Therefore, the purpose of this project is to utilize national integrated data to develop an understanding of the impact of law enforcement disruptions to the illicit drug supply on both public health (e.g. overdoses) and public safety (e.g. violent crime) outcomes. In the present study, we utilize multivariate granger-causality to investigate the relationships between police drug seizures and violent crime rates, and between police drug seizures and overdose death rates using data from the National Incident Reporting System (NIBRS) and the National Center for Health Statistics Mortality Data from CDC WONDER.

Regan Richardson received her bachelor’s degree in Mathematics with a minor in Data Analytics from Furman University in 2020. She is in her second year of the Operations Research PhD program. Her current area of research is in applications and methodology of causal discovery. Her advisors are Dr. Maria Mayorga and Dr. Osman Ozaltin. In her free time, Regan enjoys reading, watching hockey, and taking ballet classes.
 
Hou-An Chen

Bilevel Formulation for Share Allocation in Food Distribution Auctions

Choice system is a food donation distribution mechanism operated by Feeding America. In Choice, a specialized currency called “shares” is distributed to food banks, and is used to bid on food bundles in auctions. Currently, shares are distributed proportionally to a food bank’s service area population, which often results in smaller food banks receiving fewer shares and being less likely to win auctions. We study the allocation of shares that leads to a better participation, using a multi-follower bilevel model to capture the hierarchical and decentralized nature of this decision process. Feeding America, acting as the leader, seeks to maximize the minimum food bank utilities. The followers consist of multiple food banks that compete for the goods. Each food bank places bids to win allocations and earn utilities, which they seek to maximize. This interdependency between food banks is modelled as a generalized Nash equilibrium problem at the lower level. A single-level reformulation is derived and solved by the cut-and-column generation algorithm. Extensive computational experiments provide managerial insights on the operations of Choice.

Hou-An Chen is a Ph.D. student in the Operations Research Program at North Carolina State University. His research centers on game theory, with a particular focus on applications in food bank operations. In particular, he employs methodologies such as market design and bilevel programming to address complex decision hierarchies and improve resource allocation in humanitarian logistics.
 
Cameron Lisy

Modelling Extreme Weather Risks to Daily US Natural Gas Markets Under Near-term Growth in Electric Power and LNG Export Demand

Recent steady growth in the production of U.S. domestic natural gas has kept up with rising demand in the power sector, and prices have remained stable despite rising geopolitical headwinds from conflict and uncertainty in trade relationships. But this balance of fundamentals may shift in the near term as more LNG export terminals have begun operations and hyperscale datacenter growth encourages the building of more gas turbines by utilities to meet their 24/7 demand for electricity. Simultaneously, natural gas infrastructure vulnerability to extreme winter storms, most recently experienced from weakened polar vortex events in 2019, 2021, 2023, 2024 and 2025 have raised concerns about the potential for disruption at critical times of year when customers and grid operators rely on natural gas deliveries for space heating. To understand the impacts these events may have on the price and availability of natural gas, we develop a daily market clearing model of U.S. production and state-to-state transmission infrastructure.

Cameron is a 6th year PhD student in Operations Research whose research and career interests include fossil fuel market dynamics and how they impact the cost and reliability of the power grid.