Reduced order models for parameterized PDEs|| Neural network chemical kinetics || May 8, 2026

Reduced order models for parameterized PDEs|| Neural network chemical kinetics || May 8, 2026

by CRUNCH Group: Home of Math + Machine Learning + X

Formal & Physical Sciences Mathematics PBWApplied mathematicsPBWHMathematical modelling
🎙 CRUNCH Group: Home of Math + Machine Learning + X 👥 4K 📅 May 8, 2026 ⏱ 136 min 👁 311 📄 seminar 🧭 2026-08-15
Available in: English (current) Français

Keywords

ROMDMDDeepONetchemical kineticsdetonation

Summary

This seminar features two talks. The first, by Prof. Zhen Gao, focuses on extending Dynamic Mode Decomposition (DMD) to parameterized PDEs. He introduces KNN-DMD, which uses k-nearest neighbors to handle parameter variations, and then addresses high-dimensional challenges with Tensor Train (TT) decomposition, leading to TDMD and TDMD-GPR. Finally, he proposes TDMD-DeepONet, integrating DeepONet for parameter mapping and DMD for temporal evolution, with an enhanced version using initial conditions. Numerical examples include heat equation, Burgers’ equation, and Navier-Stokes equations, demonstrating improved extrapolation and efficiency. The second talk, by Josh Ethan Lipman and Amitesh Sivaraman Jayaraman, discusses the use of physics-constrained neural networks for chemical kinetics in gas-phase detonation simulations. They highlight the computational challenges of simulating detonations with detailed kinetic models and propose neural networks to accelerate chemical state prediction, enabling simulations of larger hydrocarbon fuels. The talk covers the development of kinetic models like FFCM-2 and HyChem, and the integration of neural networks to overcome stiffness and complexity.

158 words

Critical Evaluation

Value of the Information & Strength of the Argument

The first talk provides a clear and systematic development of ROM methods, with rigorous error analysis and numerical validation. The argumentation is solid, showing progressive improvements from KNN-DMD to TDMD-DeepONet, each addressing specific limitations. The second talk presents a compelling case for using neural networks to accelerate chemical kinetics in detonation simulations, backed by the need for high-resolution simulations and the limitations of traditional methods. Both talks are well-structured and supported by numerical experiments.

Scientific Rigor, Source Quality, Title Accuracy

The seminar is scientifically rigorous, with detailed methodological descriptions and error analyses. The speakers cite relevant literature (e.g., Koopman operator, DeepONet) and present original research. The title accurately reflects the content. No external sources are provided in the description, but the talks themselves reference established methods and models.

137 words

Title / Content Match

The title accurately reflects the two main topics: reduced order models for parameterized PDEs and neural network chemical kinetics.

Quality & Reliability

8/10

The seminar presents two research talks with clear methodological descriptions, numerical experiments, and error analyses. The speakers are affiliated with reputable institutions (Ocean University of China, Stanford University). The content is technical and appears scientifically sound, though not peer-reviewed in this format.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Contribution & Novelties

The seminar presents novel contributions: KNN-DMD extends DMD to parameterized problems, TDMD-GPR and TDMD-DeepONet address high-dimensional challenges, and the neural network approach for chemical kinetics enables more efficient detonation simulations. These methods show improved extrapolation and efficiency over existing techniques.

Pour aller plus loin :

  • Dynamic mode decomposition — Foundational method for data-driven ROM.
  • DeepONet — Operator learning framework used in the talk.
  • Tensor train decomposition — Dimensionality reduction technique for high-dimensional data.
  • Koopman operator — Theoretical basis for linearizing nonlinear dynamics.

82 words

Radar Profile

The radar profile shows high scores in technical level and information quality, with slightly lower scores in quantity and reliability, reflecting the seminar's depth and specificity.

Reliability 8/10