In a sheet-metal lab in Hyderabad, I once spent four years pulling specimens off a wire-cut EDM, stretching them until they split, and plotting where the splitting started. Ten years later I do the same thing on a cluster, with a few hundred thousand atoms and no specimens at all.
That journey — from the machine to the model, by way of a server room in Sydney — is the reason this site exists, and it is the reason it is written the way it is.
Who is writing this
I am Dinesh Varma Mudunuri. I hold a Master of Materials Technology from the University of New South Wales, and I am currently completing a PhD in computational materials science at the National Institute of Technology Warangal, where I have been since January 2024.
My research is on atomistic simulation and on the machine-learned interatomic potentials that are quietly rewriting what atomistic simulation can reach: which of them deserve trust, on which systems, and how you would actually know. Day to day that means density functional theory, molecular dynamics, and the growing family of models trained to imitate the first at the speed of the second. In practice: VASP, LAMMPS, Python, and a queue on an HPC cluster.
I also supervise M.Tech and B.Tech project students, which is where most of my sense of what is badly explained comes from. When three students in a row misunderstand the same thing, it is not the students.
The long way round
Most people who write about computational materials science arrived through physics. I arrived through a workshop floor, then through industry, then through a master’s degree on the other side of the world — and every one of those detours changed what I think is worth explaining.
2013–2017 — the lab. Material engineer at Gokaraju Rangaraju Institute of Engineering and Technology. Nakazima tests on ASS 316L, Inconel 718 and ASS 304 at elevated and superplastic temperatures. Forming limit diagrams, strain and thickness distributions, limiting dome heights. Friction in hot sheet forming, on a machine we designed and built in house. Sheet forming simulation in Dynaform with the LS-DYNA solver. Four years of watching real metal decide, in front of me, whether it would stretch or tear.
2017–2019 — the systems. Systems engineer at Tata Consultancy Services. Not materials at all. But it is where I learned that software is something you stay accountable for, not something you get working once and walk away from.
Sydney — the theory. Master of Materials Technology, UNSW. This is where the intuition I had built by hand acquired a formal structure underneath it — and where I first met the computational side of the subject properly.
2021–2023 — the models. Data science engineer at Regressor, ending as Data Science Engineer II. Machine learning and deep learning in production, where a model that is 95% accurate and confidently wrong on the other 5% becomes somebody’s actual problem on Monday morning. That is why the posts here are so insistent that machine-learned potentials fail quietly.
2024 — back to the atoms. PhD, NIT Warangal.
Why this matters for the writing Having built forming limit diagrams by hand, I cannot write about a stress-strain curve as an abstraction. Having shipped models that were wrong in ways nobody noticed for weeks, I cannot write about a potential’s benchmark score without asking what the error distribution’s tail looks like. Both halves of that show up in every post here.
What Materials Lab is for
Materials science is taught, very often, as a list of facts to memorise: this structure has this packing factor, that treatment gives that hardness. It is far more interesting than that. Almost every fact in the subject sits downstream of a mechanism you can actually picture, and once you have the mechanism the facts stop needing to be memorised at all.
So this site explains one idea at a time, properly: the intuition first, then the real mechanism, then what an engineer does differently because of it. It is written for the student first, with the depth a researcher wants tucked into “Going deeper” boxes you can skip without losing the thread.
Two rules I hold myself to
These are what separate a useful science blog from a content farm, and I would rather be judged on them than on anything else here.
- Every figure is either plotted from cited data or labelled as a schematic. There is no third category. Every diagram on this site is drawn from scratch in code — no borrowed images, no stock photographs of glowing atoms. Where a figure shows real numbers, the source is printed on the figure itself.
- No number without a source. If I cannot verify something, I say so plainly rather than rounding it into a fact. Where a reference is uncertain, it is marked uncertain.
When I get it wrong
I will get things wrong. When that happens I would much rather know. Corrections are made in public, the post carries its updated date, and whoever spotted the error is credited by name unless they would rather not be. The fastest route is the contact page.
Nothing here is peer-reviewed, and none of it should be cited as a primary source. It is written to get you to the primary sources faster.
Where to start
If you are new here, the reading guide sorts everything by what you came for — the fundamentals, the simulation methods, the machine learning, or the engineering disasters. Two new posts go up every week, on Monday and Thursday.
Elsewhere: LinkedIn.