UNIVERSITY PARK, Pa. — Romit Maulik, assistant professor in the Penn State College of Information Sciences and Technology (IST), was granted a three-year, $360,000 Early Career Program Award from the Army Research Office (ARO) to improve predictive modeling of complex dynamical systems for applications such as weather and climate modeling, engineering design and anomaly detection. While a system can be inherently dynamic in nature — where its measurable properties change with time — a dynamical system is one where these changes and the way they occur may be documented and analyzed mathematically, through differential equations.

The award recognizes “early career scientists and engineers who show exceptional ability and promise for conducting basic research,” according to the Army Research Laboratory (ARL), which oversees the ARO. The ARO Early Career Program Award supports scientists and engineers who have held their doctoral degree for fewer than five years at the time of application.

Maulik holds a doctoral degree in mechanical and aerospace engineering from Oklahoma State University and joined Penn State 2023. He leads the Interdisciplinary Scientific Computing Lab (ISCL), where his group performs research at the intersection of data science, applied mathematics and high-performing computing with the goal of solving grand challenge problems in computational science. Maulik is also a co-hire of the Institute for Computational and Data Sciences and Penn State.

Penn State News spoke with Maulik about the new grant and the work it will support.

Q: What do you want to understand or solve through this project?

Maulik: Our research projects aim to solve problems in predictive modeling for dynamical systems arising in geophysical, nuclear fusion and engineering applications. For example, we are very interested in reducing costs associated with weather and climate modeling, and we also wish to accelerate simulations of Tokamak reactors, which are experimental machines designed harvest energy from reactions in confined plasma. To do this, we must develop novel tools that can leverage our prior knowledge of the physics governing these applications and augment them with artificial intelligence (AI) algorithms to significantly improve accuracy and speed of predictions without sacrificing reliability and interpretability.

This award will help us pursue connections between closure models and data assimilation. Closure models are application-specific corrections built into physics-based forecasting techniques to enhance their accuracy. Data assimilation refers to a class of techniques that improves the accuracy of dynamical systems forecasts in the presence of real-time observations — for example, small fractions of the global weather are actually measurable and available for improving a simulation. In this project, we will see how real-time data can be used to identify corrections to the governing laws used in a simulation, so that one can discover novel scientific relationships. Importantly, the connection between the two will be established by deep learning algorithms.

Specifically, we are excited by how diverse sources of data — multimodality — can be used to develop on-the-fly closures to physics-based models in an interpretable and non-invasive manner. Our developed algorithms will balance the strengths of physics-based methods that are trustworthy and well-understood with the expressive quality of machine learning algorithms.

Q: How will advances in this area impact society?

Maulik: We are now in an era where tremendous amounts of measurement data for these systems are being collected from different platforms. In this project, we will try to predict such dynamical systems more accurately and more efficiently by leveraging this measurement data in real time. Applications of the methods we develop will range from more effective weather forecasts to real-time analysis of engineered systems for digital twins, which are computational models of complex real-world systems that can be simulated in real-time and used for forecasting, planning and anomaly detection.

Q: How will undergraduate and/or graduate students contribute to this research?

Maulik: This research will support the training of one doctoral student who will work closely with the rest of our group to develop, document and disseminate the proposed algorithms. This represents an excellent opportunity to work closely with the Department of Defense, not just during the course of the project but also beyond and into their professional career.

Q: This award recognizes researchers who engage in scientific leadership, education or community outreach. How does your team hope to fulfill those responsibilities?

Maulik: This grant represents a tremendous opportunity to make significant advances in the use of multimodal data and model fusion for Army-relevant applications. There are several open questions as to how data-driven models can be utilized for complex dynamical systems without violating physical laws and performance guarantees — this source of support will be instrumental in addressing several such questions. In addition, multimodal data is destined to be an intrinsic component of computing going forward, and this project will also guide curriculum development for students at Penn State with an emphasis on challenges faced by the U.S. Department of Defense.

Other members of the ISCL team include Shivam Barwey from the Mathematics and Computer Science Division at Argonne National Laboratory; Dibyajyoti Chakraborty, Haiwen Guan and Hojin Kim, all doctoral students in the College of IST; and James Henry and Xuyang Li, both postdoctoral scholars at Penn State.