Aditya Dendukuri
Aditya Dendukuri

Aditya Dendukuri

PhD Candidate in Computer Science
University of California, Santa Barbara

My research develops numerical methods for simulation and inference in large-scale dynamical systems, with an emphasis on structure-exploiting linear algebra and Markov chains. I am advised by Linda Petzold and am an affiliate at Lawrence Berkeley National Laboratory.

Email GitHub Google Scholar LinkedIn

Research

Numerical simulation

Flux-preserving adaptive finite state projection

An adaptive method for multiscale stochastic reaction networks. It uses probability flux to retain transient pathways that density-based pruning can miss. Accepted for publication in SIAM Journal on Multiscale Modeling & Simulation. Preprint

Multigrid finite state projection

A multilevel finite state projection method for spatially resolved reaction networks. Coarse models guide refinement so that fine resolution is used only where it is needed. Presented at the 2026 Copper Mountain Conference on Multigrid Methods. Paper in preparation for a SISC special issue.

Parameter inference and scientific machine learning

Adjoint Krylov methods for parameter inference

We estimate the coefficients of large linear evolution equations from observations at discrete times. A forward and adjoint Krylov calculation is shared across the operator library, which keeps the gradient calculation manageable when many candidate mechanisms are considered. Under review at SIAM Journal on Scientific Computing. Preprint

Structured Koopman generator learning

We estimate reaction rates from molecular states observed at a fixed time separation, without constructing the full Chemical Master Equation. The method learns the evolution of selected observables and uses reaction-generated functions to reduce projection bias. Under review at SIAM Journal on Applied Dynamical Systems. Preprint

2026 LBNL internship

Research on numerical algorithms for accelerating computations with large Markov chains.

I am also interested in sparse numerical linear algebra, matrix functions, reduced-precision solvers, and adaptive numerical methods.

Past research

Undergraduate research

Real-time fluid and heat simulation for virtual surgery

I worked on a real-time physical simulation for ViRCAST, a virtual trainer for arthroscopic surgery. We used smoothed particle hydrodynamics to model irrigation flow and heat transfer in a single framework, then parallelized the solver with OpenMP for interactive performance. Paper

Master's research

Learning coordinates for thermodynamically small systems

I studied how to identify useful low-dimensional variables for small molecular systems from Monte Carlo data. Diffusion maps, autoencoders, and neural networks revealed a two-dimensional description of three-particle Lennard-Jones clusters and guided the choice of simple geometric coordinates. Paper

Talks and service

SIAM Annual Meeting 2026

Data-Driven Rate Recovery and Sensitivity Analysis for Discrete Stochastic Chemical Reaction Systems. Cleveland, Ohio.

Session chair. CP17: Stochastic Systems.

Colorado Conference for Iterative and Multigrid Methods 2026

Adaptive Multigrid Finite State Projection for the Reaction-Diffusion Master Equation. Boulder, Colorado.

Selected publications and preprints

  1. An Adjoint Krylov Approach to Parameter Inference for Affine Families of Linear Evolution Equations. Aditya Dendukuri, Shivkumar Chandrasekaran, and Linda Petzold. Under review at SIAM Journal on Scientific Computing, 2026. Preprint
  2. Koopman Operator Inference for the Chemical Master Equation. Aditya Dendukuri and Linda Petzold. Under review at SIAM Journal on Applied Dynamical Systems, 2026. Preprint
  3. Flux-Preserving Adaptive Finite State Projection for Multiscale Stochastic Reaction Networks. Aditya Dendukuri, Shivkumar Chandrasekaran, and Linda Petzold. Accepted and forthcoming in SIAM Journal on Multiscale Modeling & Simulation, 2026. arXiv
  4. Preconfigured neuronal firing sequences in human brain organoids. Tjitse van der Molen, Alex Spaeth, Mattia Chini, et al. Nature Neuroscience 29, 123–135, 2026. Journal
  5. Unified Framework for Real-Time Fluid Simulation in Virtual Rotator Cuff Arthroscopic Skill Trainer. Aditya Dendukuri, Mustafa Tunc, Doga Demirel, Sinan Kockara, and Tansel Halic. IEEE BIBE, 2023. Best Paper Award. Paper
  6. Machine Learning to Identify Variables in Thermodynamically Small Systems. David M. Ford, Aditya Dendukuri, Gulce Kalyoncu, Khoa Luu, and Matthew J. Patitz. Computers & Chemical Engineering, 2020. Paper
  7. A hierarchical task analysis of shoulder arthroscopy for a virtual arthroscopic training platform. Doga Demirel, Alexander Yu, Seth Cooper-Baer, Aditya Dendukuri, et al. International Journal of Medical Robotics and Computer Assisted Surgery, 2017.

Software and teaching