Raj Jung Mahat
Ph.D. Candidate, Mechanical Engineering · Georgia Institute of Technology
I build validated predictive models from noisy measurement data, combining physics simulation, Gaussian-process surrogates, dimensionality reduction and Bayesian inference. My dissertation, with Prof. Surya Kalidindi, studies Bayesian constitutive calibration and identifiability under compressed mechanical observation. I also have industry MLOps experience from Multiscale AI.
Research interests
Experience
Graduate Research Assistant
- Bayesian inference of constitutive models from high-throughput small-punch and tensile tests, from finite-element simulation to posterior, applied to nine alloys.
- Gaussian-process emulators on a PCA-compressed response space; transitional and Hamiltonian MCMC for multimodal posteriors.
- A response-dependent discrepancy model in latent space that cut cross-test error from 28% to 13%.
- Likelihood-informed-subspace identifiability analysis used to guide test design; built a DIC-instrumented small-punch rig.
Data Scientist
- MLOps modules for the inference workflow, data workflow and feature-engineering pipeline, with a logging backend.
- An inverse-inference toolkit (surrogates plus MCMC) packaged for reuse across client problems.
- A Cox proportional-hazards model of adhesive-bond durability for a Novelis aluminum surface-treatment study.
Graduate Research Student
- Neural-network inverse models to infer material properties from instrumented indentation.
Education
Ph.D., Mechanical Engineering
M.S., Materials Engineering
B.S., Mechanical Engineering
Publications and manuscripts
- 2026R. J. Mahat. Measure equivalence as a guide for learning SPDE priors from replicated fields. Preprint. doi · source
- in prep.R. J. Mahat, et al., R. Arróyave, I. Karaman, S. R. Kalidindi. Rapid evaluation of elastic–plastic constitutive laws for HTMDEC high-entropy alloys using small punch tests. In preparation.
- in prep.R. J. Mahat, A. Venkatraman, S. R. Kalidindi. Bayesian calibration of Voce hardening parameters from the small punch test: identifiability and structured cross-test discrepancy. In preparation.
- reviewS. Mahmoud, R. J. Mahat, et al., S. R. Kalidindi. Bayesian identification of elastic-plastic properties from high-throughput small punch tests and DIC measurements. Under review.
- ms.S. Mahmoud, A. Venkatraman, R. J. Mahat, et al., S. R. Kalidindi. Amortized posteriors for estimation of material constitutive parameters from multimodal measurements on small punch tests. Manuscript.
- 2026N. Flores, et al. (incl. R. Mahat, S. R. Kalidindi, R. Arróyave). Data-driven insights into composition–property relationships in FCC high entropy alloys. Scripta Materialia 274, 117136. doi
- 2024M. Mulukutla, A. N. Person, S. Voigt, et al. Illustrating an effective workflow for accelerated materials discovery. Integrating Materials and Manufacturing Innovation 13, 453–473. doi
Projects
HTMDEC high-entropy alloys: high-throughput small punch testing and constitutive calibration
- Bayesian framework (simulation surrogate + model-discrepancy correction) inferring Voce elastic–plastic parameters from small punch tests of HTMDEC/BIRDSHOT FCC HEAs (Al–V–Cr–Mn–Fe–Co–Ni–Cu), benchmarked against tensile tests.
- Ongoing: kinematic-hardening study of HEAs in the small punch test.
- SPT, tensile, EBSD and composition data across eight vacuum-arc-melted HEA batches; simulation–experiment mismatch linked to grain size. Related: Scripta Materialia 2026.
Differentiable plasticity FEM in JAX
- Fully differentiable axisymmetric punch-test FEM with J2 plasticity (Voce + Armstrong–Frederick hardening): implicit return map and Newton via implicit differentiation, exact consistent tangent by autodiff, penalty contact, load stepping.
- Exact parameter sensitivities by forward mode, adjoint and a hand-derived direct recursion (agree to 10−13); Fisher-information comparison of what tension and punch tests identify. Notes and flowcharts.
Hamiltonian and Riemannian-manifold MCMC from scratch
- Leapfrog HMC with a convergence study on ODE-defined posteriors; Riemannian-manifold MCMC benchmarked against HMC on strongly correlated posteriors; normalizing flows compared with MCMC on quality and cost.
Latent-space modeling of high-dimensional mechanical test responses
- PCA, graph-Laplacian and diffusion-map embeddings of load–displacement and DIC data; response-dependent corrections in latent space (basis of the 28% → 13% discrepancy result).
Sensitivity and identifiability analysis of nonlinear simulation models
- Jacobian-based and variance-based sensitivities and likelihood-informed subspaces to rank parameters, detect compensating directions and guide experiment and surrogate design.
Data-driven extraction of uniaxial creep properties from indentation using neural networks
- Neural-network inverse models mapping instrumented-indentation response to creep parameters.
Open notes on GitHub
Self-contained write-ups I keep public: each has a readable web page and a GitHub repository with the LaTeX source and figures.
Differentiable plasticity FEM in JAX
A small, fully differentiable punch-test model with J2 plasticity (Voce + Armstrong–Frederick hardening), with detailed flowcharts of how information about the hardening law travels to the measured load: return map, consistent tangent, contact, history, and exact sensitivities (forward mode, adjoint and a hand-written recursion).
When is a Gaussian prior “the same kind of thing”?
Computational notes on the Feldman–Hájek theorem: when two Gaussian measures are equivalent or mutually singular, and what that means for SPDE priors, which changes (a mean shift, a finite-rank correction, a change of lower-order coefficients) stay in the same class, and which force the operator itself to change.
A poem
Ye deewana kaha kuch puch maat apne dil ka haal Tere dadili ki kisse tho rom rom jane naal Puche puch tu jaag pukare Na sake koi na koi tuje nihare Tu kar jo tu soche Jo soche kay tu howe Jane jan, kabi dekh le nazar uthake Pardangiya par par kar udhe hai subhe saam
Don't ask this madman how his heart fares, Every fiber of my soul knows the stories of your grace. You wake and call out, asking the world, But no one can quite capture you, no one can hold your stare. Go ahead and do what's in your mind, Will you become exactly what you dream. My love, just lift your gaze every now and then: Watch the birds of joy, fly every moment through the sky.