Computational Materials Scientist for Machine-Learned Interatomic Potentials (MD/DFT)

Job Description

About Amphiform:

\n

Amphiform (amphiform.com) is building next-gen energy materials for AI data centres, defence and space: light, abundant and powerful. We achieve that by creating a new type of matter: hybrid materials, built atomic layer by atomic layer, where every atom has a pre-programmed purpose.

\n

We just closed a $5.5+1m pre-seed led by General Catalyst and Main Object, with Thomas Wolf (Hugging Face), Charlie Songhurst, and others.

\n

\n

The role:

\n

We are developing a new class of material: an ultrathin, vapour-deposited hybrid materials with custom properties - and we want to understand the fundamental properties. You will own our first atomistic simulation campaign: predicting conductivity (electron, ion, heat), hydration behaviour, and interfacial transport in a material that has never been simulated, and testing your predictions against experiments that we'll run constantly alongside. The workflow (DFT benchmarking → amorphous structure generation → fine-tuned machine-learned interatomic potentials with active learning → large-scale transport MD, enhanced sampling, and path-integral MD) is scoped, budgeted, and precedented in the recent literature; the material is not. You will be our first dedicated computational hire, working directly with the experimental team.

\n

\n

What you'll do:

\n

\n
• Benchmark DFT functionals against coupled-cluster references for proton-transfer energetics, and own the resulting error budget
\n
• Build amorphous model structures of a hybrid metal-organic network validated against our data
\n
• Fine-tune foundation MLIPs (MACE-class) with active learning; run DFT labelling campaigns
\n
• Run nanosecond-scale transport MD, metadynamics, and PIMD to extract diffusion coefficients, activation energies, isotope effects, and interfacial behaviour
\n
• Turn simulation into decisions: feed results into our materials and process development, and publish the methodology with us
\n
• Define our simulation infrastructure (environments, data management, reproducibility)
\n

\n

\n

What we're looking for:

\n

\n
• PhD in computational chemistry, physics, or materials science
\n
• Hands-on ab-initio MD experience on HPC
\n
• You have trained at least one machine-learned interatomic potential (MACE, NequIP, Allegro, DeepMD, GAP, or similar) and can explain how you built the training set and knew it was converged
\n
• Strong scientific Python
\n

\n

\n

Nice to have:

\n

\n
• Ion transport background
\n
• Amorphous/disordered structure generation
\n
• Path-integral MD (i-PI), GCMC (RASPA), charge-partitioning workflows, or workflow engines (psiflow, Parsl, Snakemake)
\n
• A benchmark culture: DLPNO-CCSD(T), functional selection, error bars you can defend
\n