JDFTx on HPC

JDFTx is an open-source, plane-wave density-functional theory (DFT) code for electronic-structure calculations, with tools for modelling electrochemical systems and solid–liquid interfaces. Joint density-functional theory couples the electronic system to a density-functional description of the surrounding liquid, while implicit-solvation models such as CANDLE and SaLSA represent solvent effects without explicit solvent molecules. JDFTx overview

For electrochemistry, JDFTx supports constant-potential calculations, allowing the electron number to adjust at a fixed electron chemical potential. It also offers variational energy minimization when conventional self-consistent-field iterations struggle, and truncated Coulomb interactions for molecules, wires, and slabs. JDFTx features

JDFTx runs on CPUs and NVIDIA GPUs through CUDA. The upstream build configuration checked on 13 October 2026 has no HIP backend for AMD GPUs. Upstream build configuration

Note

HIP support on AMD GPUs: Episode 2 shows the ENCCS development port for LUMI’s AMD GPUs, with source excerpts and build instructions.

Before you start

Prerequisites

  • Familiarity with Linux shell commands, environment modules, and Slurm.

  • Access to the target system and an allocation for running calculations.

  • Basic familiarity with JDFTx and electronic-structure calculations.

Short tutorials

Episode 1 covers CPU and CUDA builds on LUMI-C, Dardel GH200, and Leonardo Booster. Episode 2 covers the original HIP port at 9573a3e57 and its build and test procedure on LUMI. Episode 3 is a draft, with the later solver fix, validation results, GPU timings, and CPU profiling.

Credits and licenses

Site infrastructure is adapted from the ENCCS Sphinx lesson template at commit b460ac1af94049533a41abdb03ed4c0cf0177f1e, using the configuration and branding of the companion LAMMPS tutorials.

Template license notices are retained in LICENSE (CC BY-SA 4.0) and LICENSE.code (MIT).