Home / Reproduce

Reproduce the work

These small examples teach methods from my research. Each tutorial lists the files, commands, and expected output. The status tells you which example was run.

Before you start: the ML-potential tutorials use the open MACE-MP-0 model by default (no account needed). Pass --model uma-s-1p1 to use Meta’s UMA, the model in several of the papers. UMA requires a free Hugging Face token. LAMMPS tutorials need a LAMMPS build; see LAMMPS_SETUP.md.

Example run locallyPython · ML potential

Benchmark a universal ML potential on doped MoS₂

Build doped MoS₂ cells, relax them with an ML potential, compute dopant formation energies, then heat a doped bilayer and watch the dopants move.

~1 min + ~1 min MD (GPU)Colab notebook
From: One ML potential, 25 dopants: where it works and where it fails
Example run locallyPython · ML potential

Stretch MoS₂ until it breaks, at three temperatures

Pull a monolayer MoS₂ sheet apart at 100, 200 and 300 K with an ML potential, and measure flake tilt with a PCA method.

~3 min (GPU)Colab notebook
From: Why MoS₂ space lubricants wear out in the cold
Example run locallyPython · NumPy

How force fields and mixing rules change gold–ion interactions

Recompute the Au–ion Lennard-Jones cross terms for the four ion force fields in the paper’s Table I, and see how far they disagree.

< 2 sColab notebook
From: Why classical models get ion adsorption on gold wrong
Example run locallyPython · ML potential

Diagnose a foundation potential on a molten salt

Melt LiCl–FeCl₂ with an off-the-shelf ML potential and compute the partial radial distribution functions that reveal whether Fe ions cluster.

~2 min (GPU)Colab notebook
From: Fixing a foundation model's iron clustering in molten salt
Example run locallyPython · ML potential

From g(r) to thermodynamics: Kirkwood–Buff integrals

Simulate the LiCl–KCl eutectic melt and compute running Kirkwood–Buff integrals, the bridge from structure to activity coefficients and phase diagrams.

~2 min (GPU)Colab notebook
From: From molten-salt structure to the LiCl–KCl phase diagram
Example run locallyPython · ML potential

Let the model choose its next training data

Run Pt(111)–water MD, measure where two ML potentials disagree, and select the most uncertain frames for DFT labeling.

~1 min (GPU)Colab notebook
From: Uncertainty-driven training data for interface potentials
Example run locallyPython · ML potential

Check, then fine-tune, a foundation potential

Compare model predictions with reference labels in a parity plot (the demo compares two foundation models), then follow the LoRA fine-tuning recipe used for MoS₂ with N₂.

< 1 min (GPU)Colab notebook
From: Fine-tuning cuts a foundation model's error eightfold for MoS₂ + N₂
Example run locallyPython · ML potential

Grow a Ni-doped MoS₂ film atom by atom

Deposit Mo, S and Ni atoms onto a MoS₂ substrate one at a time, then measure Ni clustering and MoS₂-like order.

~7 min (GPU)Colab notebook
From: Intermediate nickel lowers shear stress in sputtered MoS₂
Example run locallyPython · NumPy

Build a temperature plot matrix from literature data

Turn digitized friction, wear-rate and wear-life data into the review’s metric × environment plot matrix.

< 5 sColab notebook
From: MoS₂ lubricant performance across temperatures
Inputs not yet validatedLAMMPS

Pull an Inconel-718 nanowire at 100 K and 800 K

Build a random Ni–Cr–Fe alloy nanowire, stretch it at two temperatures with the Bonny FeNiCr EAM potential, and compare the stress–strain curves.

~10–15 min (4 cores, estimate)Colab notebook
From: How temperature and cooling rate affect Inconel nanostructures
Inputs not yet validatedLAMMPS

Quench liquid tin at two rates, then stretch it

Melt Sn, cool it at 10 and 100 K/ps with a MEAM potential, and compare strength and stiffness of the two solids.

~15 min (4 cores, estimate)Colab notebook
From: Faster cooling makes lead-free solder weaker but tougher
Inputs not yet validatedLAMMPS

Stretch CdTe and CdSe₀.₅Te₀.₅ nanowires

Compare the tensile strength of a zinc-blende CdTe nanowire before and after random Se substitution, with the Cd–Te–Se bond-order potential.

~15 min (4 cores, estimate)Colab notebook
From: Selenium makes CdTe nanowires stronger and stiffer
Inputs not yet validatedLAMMPS

Deposit an alloy layer by layer, then test it

Mimic additive manufacturing: deposit Ni–Cr–Fe atoms in batches, cool at two rates, and run a tensile test.

~20 min (4 cores, estimate)Colab notebook
From: Cooling rate and strength in 3D-printed Inconel
Draft · incomplete inputsLAMMPS

Cycle a gold–silver core–shell nanowire (draft)

Tension–compression cycling of Au-core/Ag-shell and Ag-core/Au-shell nanowires with the paper’s MEAM parameters. Draft: some alloy parameters still need checking against the published tables.

—
From: Gold–silver core–shell nanowires under repeated stretching