publications
Papers are listed below by year of submission before they are published, or year of publication.
2025
- Can Language Models Speed Up General-Purpose Numerical Programs?Ori Press, Brandon Amos, Haoyu Zhao, and 21 more authors2025
Despite progress in language model (LM) capabilities, evaluations have thus far focused on models’ performance on tasks that humans have previously solved, including in programming (Jimenez et al., 2024) and mathematics (Glazer et al., 2024). We therefore propose testing models’ ability to design and implement algorithms in an open-ended benchmark: We task LMs with writing code that efficiently solves computationally challenging problems in computer science, physics, and mathematics. Our AlgoTune benchmark consists of 155 coding tasks collected from domain experts and a framework for validating and timing LM-synthesized solution code, which is compared to reference implementations from popular open-source packages. In addition, we develop a baseline LM agent, AlgoTuner, and evaluate its performance across a suite of frontier models. AlgoTuner achieves an average 1.76× speedup against our reference solvers, which use libraries such as SciPy, sk-learn and CVXPY. However, we find that current models fail to discover algorithmic innovations, instead preferring surface-level optimizations. We hope that AlgoTune catalyzes the development of LM agents exhibiting creative problem solving beyond state-of-the-art human performance.
@misc{algotune2025, title = {Can Language Models Speed Up General-Purpose Numerical Programs?}, author = {Press, Ori and Amos, Brandon and Zhao, Haoyu and Wu, Yikai and Ainsworth, Samuel K. and Krupke, Dominik and Kidger, Patrick and Sajed, Touqir and Stellato, Bartolomeo and Park, Jisun and Bosch, Nathanael and Meril, Eli and Steppi, Albert and Zharmagambetov, Arman and Zhang, Fangzhao and P{\'e}rez-Pi{\~n}eiro, David and Mercurio, Alberto and Zhan, Ni and Abramovich, Talor and Lieret, Kilian and Zhang, Hanlin and Huang, Shirley and Bethge, Matthias and Press, Ofir}, year = {2025}, } - An Operator Splitting Method for Large-Scale CVaR-Constrained Quadratic ProgramsEric Luxenberg, David Pérez-Piñeiro, Steven Diamond, and 1 more author2025
We introduce a fast and scalable method for solving quadratic programs with conditional value-at-risk (CVaR) constraints. While these problems can be formulated as standard quadratic programs, the number of variables and constraints grows linearly with the number of scenarios, making general-purpose solvers impractical for large-scale problems. Our method combines operator splitting with a specialized O(mlogm) algorithm for projecting onto CVaR constraints, where m is the number of scenarios. The method alternates between solving a linear system and performing parallel projections: onto CVaR constraints using our specialized algorithm and onto box constraints with a closed-form solution. Numerical examples from several application domains demonstrate that our method outperforms general-purpose solvers by several orders of magnitude on problems with up to millions of scenarios. Our method is implemented in an open-source package called CVQP.
@misc{cvqp2025, title = {An Operator Splitting Method for Large-Scale {CVaR}-Constrained Quadratic Programs}, author = {Luxenberg, Eric and P{\'e}rez-Pi{\~n}eiro, David and Diamond, Steven and Boyd, Stephen}, year = {2025}, eprint = {2504.10814}, archiveprefix = {arXiv}, primaryclass = {math.OC}, }
2023
- Home energy management with dynamic tariffs and tiered peak power chargesDavid Pérez-Piñeiro, Sigurd Skogestad, and Stephen BoydarXiv preprint arXiv:2307.07580, 2023
We consider a simple home energy system consisting of a (net) load, an energy storage device, and a grid connection. We focus on minimizing the cost for grid power that includes a time-varying usage price and a tiered peak power charge that depends on the average of the largest N daily powers over a month. When the loads and prices are known, the optimal operation of the storage device can be found by solving a mixed-integer linear program (MILP). This prescient charging policy is not implementable in practice, but it does give a bound on the best performance possible. We propose a simple model predictive control (MPC) method that relies on simple forecasts of future prices and loads. The MPC problem is also an MILP, but it can be solved directly as a linear program (LP) using simple enumeration of the tiers for the current and next months, and so is fast and reliable. Numerical experiments on real data from a home in Trondheim, Norway, show that the MPC policy achieves a cost that is only 1.7% higher than the prescient performance bound.
@article{pineiro2023hem, title = {Home energy management with dynamic tariffs and tiered peak power charges}, author = {P{\'e}rez-Pi{\~n}eiro, David and Skogestad, Sigurd and Boyd, Stephen}, journal = {arXiv preprint arXiv:2307.07580}, year = {2023}, }
2021
- Supervisory control design for balancing supply and demand in a district heating system with thermal energy storageCristina Zotică, David Pérez-Piñeiro, and Sigurd SkogestadComputers & Chemical Engineering, 2021
This paper presents a systematic comparison between three alternatives to design the supervisory control layer of a district heating network composed of a waste heat boiler, an electric boiler, a dump, a hot water storage tank, and a set of consumers. The three alternatives are split range control, controllers with different setpoints, and model predictive control. We evaluate the closed-loop performance in the face of time-varying supply and demand, and constant electricity prices. All alternatives were found to give similar performance. Controllers with different setpoints is the easiest to implement, while model predictive control is the most difficult.
@article{zotica2021supervisory, author = {Zotic{\u{a}}, Cristina and P{\'e}rez-Pi{\~n}eiro, David and Skogestad, Sigurd}, journal = {Computers \& Chemical Engineering}, volume = {149}, pages = {107306}, year = {2021}, publisher = {Elsevier}, }
2020
- Optimal operation and control of a thermal energy storage system: Classical advanced control versus model predictive controlCristina Zotica, David Pérez-Piñeiro, and Sigurd SkogestadIn Computer Aided Chemical Engineering, 2020
The objective of this work is to define the optimal operation and control for a thermal storage system with heat sources and a consumer, which exchange utilities using one hot water thermal energy storage tank. In this work, we compare a decentralized control structure using classical advanced control with PID controllers and logic blocks (split-range control and selectors) and a centralized control structure (model predictive control) to implement optimal operation for a simple thermal energy storage system, which is a multivariable system with constraints. We analyze a varying heat supply profile over a horizon of 24 hours. We show that the supply and demand can be balanced, and we achieve optimal operation by using the energy stored in the tank while minimizing the heat from the market.
@incollection{zotica2020optimal, title = {Optimal operation and control of a thermal energy storage system: Classical advanced control versus model predictive control}, author = {Zotica, Cristina and P{\'e}rez-Pi{\~n}eiro, David and Skogestad, Sigurd}, booktitle = {Computer Aided Chemical Engineering}, volume = {48}, pages = {1507--1512}, year = {2020}, publisher = {Elsevier}, } - Self-optimizing control of a continuous-flow pharmaceutical manufacturing plantDavid Pérez-Piñeiro, Anastasia Nikolakopoulou, Johannes Jäschke, and 1 more authorIFAC-PapersOnLine, 2020
This article considers the real-time optimization under uncertainty of a compact reconfigurable system for on-demand continuous-flow pharmaceutical manufacturing. Self-optimizing control is employed, which optimizes operation in the presence of uncertainty by controlling a carefully chosen combination of measurements to a constant setpoint. The method is applied to a simulated plant based on the physical process. The closed-loop simulations indicate that this simple policy is able to maintain the process operation close to optimality despite disturbances, sensor noise, and parametric model uncertainty.
@article{pineiro2020self, title = {Self-optimizing control of a continuous-flow pharmaceutical manufacturing plant}, author = {P{\'e}rez-Pi{\~n}eiro, David and Nikolakopoulou, Anastasia and J{\"a}schke, Johannes and Braatz, Richard D}, journal = {IFAC-PapersOnLine}, volume = {53}, number = {2}, pages = {11601--11606}, year = {2020}, publisher = {Elsevier}, }