I'm a PhD candidate in Mechanical Engineering at Virginia Tech, in the Laboratory of Transport Phenomena for Advanced Technologies. I defend by February 2027.

nilotpalchakraborty.com: longer write-ups of the work below.

I build the numerical machinery that closed solvers don't provide. Most of my work is inside other people's codes. A ~1,200-line stochastic breakage library in a commercial CFD solver, with special functions, spline inversion, and a per-particle state channel packed into the IEEE-754 mantissa of the diameter word because the API carries no state. An HLLC Riemann solver in an open-source SPH library, written after working out why its surface-tension model failed at high Reynolds number.

My application domain is particle ingestion in gas-turbine compressors: 100M+ cell campaigns on HPC, sponsored by Rolls-Royce and Pratt & Whitney, both of whom took the solver code for internal use. The portable part is the machinery, not the domain.

Found and reported a numerical instability in SPHinXsys. The shipped multiphase surface-tension model breaks down at high Reynolds number: in the square droplet test the fluid particles don't just disorder, they leave the domain. I isolated it with a sequence of experiments and reported it in #378, and #497. The maintainers later traced it to zero-surface-energy modes and fixed it; I'm acknowledged in the resulting paper:

S. Zhang, S.D.N. Lourenço, X. Hu, Multiphase SPH for surface tension: resolving zero-surface-energy modes and achieving high Reynolds number simulations, Computer Methods in Applied Mechanics and Engineering 444 (2025) 118147.

sph-high-re-surface-tension has the diagnostic sequence, the HLLC Riemann solver I wrote while chasing it, and the parameter studies underneath: a screening design over reference velocity and viscosity, a one-factor-at-a-time sweep, a ladder of dissipation-limiter settings, and a 2 × 2 factorial on Poiseuille flow, which is the only case there with a closed-form answer to check against. Video of every run.

Particle redistribution at rotor–stator interfaces in CFX: erosion maps on a compressor rotor came out striped, with 46.8% of blade nodes taking no impacts at all, although particles were injected at random upstream. Eight hypotheses, seven killed by experiment. The eighth held: particles are displaced 4 mm circumferentially crossing the stage interface, landing on the centre-lines of the receiving mesh cells. Isolated by decoupling the particles from the flow entirely (rotor at 10⁻⁶ rpm, fluid forces off), so nothing but the interface could be doing it. The same class of bug as the SPHinXsys one, in a solver whose source I cannot read.

Scaling ANSYS CFX across a cluster: standard advice for this solver is to minimise node count, since every iteration exchanges boundary data across the network. Measured, it went the other way: 3.9× faster on sixteen nodes than one, because the workload is bound by I/O and memory throughput rather than compute. Shown the numbers, the university's research computing director called it "generally the opposite of what I would expect". It also established that licences, not hardware, were the real ceiling, after which the group's allocation was doubled. Independent work, not part of my dissertation.

Lid-driven cavity: incompressible Navier–Stokes on a staggered grid, fractional-step, validated against Ghia et al. (1982). The solver diverged at a time step that sat below both the linear CFL and the viscous limit. The cause is a non-linear CFL condition that restricts higher-order explicit time integrators, and the code now takes its step from that rather than from either textbook criterion.

Iterative solvers for finite-difference systems: the companion study. Gauss-Seidel, SOR and ADI on a manufactured Poisson problem whose exact solution is known, so the error is measured and not inferred from the residual. Observed order of accuracy 2.06; ADI reaches tolerance in about a quarter of the iterations SOR needs, at both grid resolutions. This is the solver behaviour that drives the cavity's cost scaling.

Membrane permeability from molecular structure: can you predict whether a drug-like molecule crosses a lipid membrane from its SMILES string alone, without the free energies and pKa values that normally decide it? Four-person course project; I built the RDKit feature generation (200 molecular descriptors and 1024-bit Morgan fingerprints), and the cross-validated regularisation search and learning-curve diagnostics for the Lasso model. The team's combined Lasso-MLP reached R² = 0.90 from structure alone.

Machine learning projects, spam detection (TF-IDF baselines vs. DistilBERT), and churn prediction, both built around evaluation rather than accuracy: precision–recall, threshold tuning under asymmetric costs, and what the added model complexity actually buys.

Learned surrogates for expensive particle-laden physics on the open NASA Rotor 35 geometry, with the evaluation done properly: held-out conditions, conservation checks on the predicted distributions, and an honest map of where the surrogate stops being trustworthy.

Languages Python · Fortran 90 · C++ · Java · MATLAB · Bash HPC Linux · Slurm · MPI · large-memory nodes · scaling and performance profiling Numerical Special functions from scratch · cubic splines and spline inversion · bit-level IEEE-754 encoding · approximate Riemann solvers (HLLC) · meshfree/SPH · Lagrangian particle tracking · finite volume (RANS/SST) · KD-tree spatial search ML PyTorch · scikit-learn · XGBoost · NumPy/Pandas, surrogate and reduced-order modelling, cross-validation, cost-sensitive evaluation Simulation ANSYS CFX · Fluent · TurboGrid · ICEM CFD · solver-embedded user subroutines

• Summer Cunningham Fellowship, Virginia Tech Graduate School (2026)
• Pratt Fellowship, Virginia Tech College of Engineering (2022)
• Mitacs Globalink Research Fellowship (2019)
• Shastri Research Fellowship, Shastri Indo-Canadian Institute (2019)

📫 nilotpalc@vt.edu · nilotpalchakraborty.com