Quantum 3xplorer¶

Quantum 3xplorer (or Q3 in short) ¹ ² is an AI-native quantum simulation engine, written from scratch by AI coding agents under the direction and supervision of domain scientists. Its first implementation is a plane-wave pseudopotential code for Density Functional Theory (DFT) calculations of periodic crystals. The engine is written in C++17, with a Python layer on top, and is distributed under the Apache License 2.0 ³.

Q3 follows the ESSE data standard for its inputs and outputs. It reads its own YAML input format as well as unmodified Quantum ESPRESSO pw.x namelist inputs, and writes structured JSON results together with Quantum ESPRESSO-compatible binary checkpoints.

Capabilities¶

The capabilities of Q3 are grouped below by area. The validation status of each feature against established plane-wave DFT codes is listed in the Q3 documentation ². The subset of these capabilities available through the platform workflows is described in the components page.

Structures and symmetry¶

Q3 accepts the 14 Bravais lattice types as well as explicit lattice vectors. Space-group detection, reduction of the Monkhorst-Pack k-point grid to the irreducible Brillouin zone, and density symmetrization are performed with the spglib library. Gamma-only calculations are supported.

Pseudopotentials and functionals¶

Norm-conserving, ultrasoft, and Projector Augmented-Wave (PAW) pseudopotentials are read in the Unified Pseudopotential Format (UPF), versions 1 and 2. The exchange-correlation functionals are provided by the LibXC library: the Local Density Approximation (LDA) in the Perdew-Zunger form, the Generalized Gradient Approximation (GGA) in the PBE, PBEsol, BLYP, revPBE, and RPBE forms, and any other LDA or GGA combination addressed by its LibXC identifier. Spin-unpolarized and collinear spin-polarized calculations are supported.

Ground-state calculations¶

The self-consistent field solver offers the following features.

Total energy, atomic forces, and stress tensor, for all three pseudopotential types.
Direct (LAPACK) and Davidson diagonalization.
Broyden and Anderson density mixing, with Kerker preconditioning for metallic systems.
Fixed occupations, and Gaussian, Methfessel-Paxton, Marzari-Vanderbilt, and Fermi-Dirac smearing.
DFT+U in the Dudarev formulation, and DFT-D3 dispersion correction with Becke-Johnson damping.
Charged cells, with Makov-Payne and Martyna-Tuckerman corrections for isolated systems.
Non-self-consistent calculations and restart from binary checkpoints.

Property workflows¶

The Python layer of Q3 provides workflows built on top of the engine, with an ASE calculator interface.

Geometry relaxation, variable-cell relaxation, and relaxation at target pressure.
Band structure, density of states, and projected density of states with Löwdin charges.
Equation of state (Birch-Murnaghan) and elastic constants.
Gamma-point phonons, and phonon dispersions and thermodynamics through Phonopy.
Wannier interpolation.
Export of mean-field data for BerkeleyGW, and of volumetric fields (density, potential, wavefunctions) in cube and XSF formats.

Parallelization and hardware¶

Q3 runs in serial with OpenMP threading, and in parallel with the Message Passing Interface (MPI) over k-point pools and band groups. GPU builds are available for NVIDIA (CUDA, including several GPUs across MPI ranks), Apple Silicon (Metal), AMD (HIP), and Intel (SYCL) hardware. Container images are published for the x86-64 and ARM64 architectures.

Supported Versions¶

We support version 2026.10.31-0 of Q3, built with the GNU toolchain and parallelized with MPI.

Models and Methods¶

Q3 calculations on our platform use the Kohn-Sham DFT model, with the following exchange-correlation functionals.

Functional	Type
`pz`	Local Density Approximation (LDA)
`pbe`	Generalized Gradient Approximation (GGA)
`pbesol`	Generalized Gradient Approximation (GGA)

The method combines a plane-wave basis set with norm-conserving, ultrasoft (the default), or PAW pseudopotentials, Gaussian smearing of the electronic occupations, and Davidson diagonalization.

Components ¶

We introduce the executable and the calculation flavors of Q3 implemented on our platform in this page.

Workflows¶

The Total energy - Q3 workflow is available in the Workflows Bank. It fetches the material from the platform, writes it to the input directory, and runs a self-consistent field calculation with the q3_scf flavor, extracting the total energy and the Fermi energy.

Command Line Usage¶

Q3 is installed on the compute clusters as an Apptainer container, exposed through the q3 environment module. Loading the module sets the $EXEC_CMD and $EXEC_CMD_Q3 variables to the container execution command. The MPI launcher is part of the container, so mpirun is called inside it, as shown in the batch script example below.

#!/bin/bash

#PBS -N q3_scf
#PBS -l nodes=1
#PBS -l ppn=4
#PBS -q D
#PBS -j oe
#PBS -l walltime=00:10:00

cd $PBS_O_WORKDIR
module load q3
# $EXEC_CMD is set by the environment module
$EXEC_CMD mpirun -np $PBS_NP q3 q3_scf.yml > q3.out

Results are written to <outdir>/<prefix>_results.json, where outdir and prefix are set in the control section of the input file. The input format is described in the Q3 documentation ². Complete example inputs are listed in the Examples section of the components page.