Where equations
meet intuition
I’m Utsav Purkait, a researcher in computational fluid dynamics working on turbulence modelling and vortex dynamics — and on understanding where our physical models succeed and where they break down. I write about that work here.
Research Assistant at IIT Bombay — studying Reynolds stress anisotropy in turbulent planar jets. Manuscript in preparation for Physics of Fluids.
“When I meet God, I am going to ask him two questions: why relativity? And why turbulence? I really believe he will have an answer for the first.”
— Werner HeisenbergWhat I work on
Reynolds stress anisotropy in turbulent planar jets
Using high-fidelity LES data to show that RANS closure failures in free-shear flows are structural — rooted in the Boussinesq hypothesis itself, not in calibration.
Manuscript in preparation →Tornado-like vortex flow simulator
Designed and simulated a dual-shell simulator replicating Rankine vortex dynamics; physical experiments validated the CFD model to within 8.5%.
Bachelor’s thesis →Data-driven turbulence closure
Exploring symbolic regression and hybrid neural–algebraic models as interpretable routes to RANS closure — machine learning that produces physics.
Projects & writeups →Featured
Data-driven turbulence models often fail when deployed in RANS solvers due to distribution shift. This review examines DARSM, a hybrid model that uses neural networks to calibrate physics-based algebraic equations. By ensuring consistency between training and deployment, DARSM improves generalization across complex flows while highlighting the structural limits…
More articles
Can a Machine Discover a Turbulence Model? The Rise of Symbolic Regression in RANS Closure
There’s a quiet revolution happening at the intersection of machine learning and turbulence modelling, and it doesn’t look like what most people expect. It…
Rortex: finally, a vortex vector that means what it says
Traditional fluid dynamics often confuses vorticity with actual rotation, leading to inaccurate vortex identification. The Rortex method solves this by isolating rigid-body rotation from…
Structural Limits of Linear Eddy-Viscosity Models in the Planar Jet
This article examines the validation of planar turbulent jet simulations, highlighting how key metrics like jet spreading and decay laws align with direct numerical…
When pressure-based solvers silently break conservation
Okay, real talk for a minute. Last month, I almost submitted a paper with results that were mathematically beautiful and physically impossible. My pressure-based solver…
“The Boltzmann Gambit” – When Particle Kinetics Challenges Continuum Mechanics
The “collide-and-stream” algorithm seemed too simple to be revolutionary. In 1998, when Exa Corporation first commercialized the Lattice-Boltzmann Method (LBM), traditional CFD engineers dismissed it…
Time-Step Independence Is Not What You Think It Is: The Stiffness vs. CFL Trap
I was simulating vortex shedding behind a cylinder at Re = 100. My explicit compressible solver gave plausible results at Δt = 1e-5 seconds,…
The Playground
Test your knowledge with topic-specific quizzes, watch curated CFD explainer videos, and challenge yourself across five difficulty categories.
Enter playground →