Features¶
From molecule input to spectra and history replay — everything runs in your local Python kernel.
Paste XYZ, browse an indexed three-tier bundled library (presets + curated + ~1,900 QM9 structures, searchable by name/formula), or run a structure search by name, SMILES, InChI, CID, or CAS — PubChem → NCI CACTUS → an offline bundled-library fallback, so the search still works with no network. Or upload a file (XYZ, MOL/SDF, MOL2, PDB, CIF, Gaussian input/output) and edit the structure: set bond lengths, angles and dihedrals, delete atoms, add hydrogens, change elements, freeze atoms.
py3Dmol-first interactive viewer with a capability-aware backend router. Molecules, optimization trajectories, vibrational modes, and orbital isosurfaces all render inline — and fully offline (3Dmol.js is vendored, never fetched from a CDN). Tunable playback FPS + on-disk cache for instant replay.
RHF, UHF, nine DFT functionals, MP2, CCSD, and CCSD(T) — with eight calculation types: single point, geometry optimization, transition-state search (Sella, with an automatic frequency check), frequencies/thermochemistry, TD-DFT UV-Vis, NMR shielding, 1D PES scans and Marcus reorganization energies. PCM implicit solvation for single points, optimizations, frequencies and TD-DFT.
IR spectrum (stick + Lorentzian-broadened), UV-Vis plot, 1H/13C NMR shifts vs TMS, point groups and an orbital diagram with symmetry labels. Any orbital (α/β), an orbital gallery in linked viewers, density, spin density and ESP-mapped surfaces. Vibration viewer with displacement arrows; thermochemistry at any T and P; ⟨S²⟩ for open-shell results; side-by-side comparison tables.
Every calc auto-saves to a timestamped directory and replays
after a kernel restart. Export structures (XYZ, MOL/SDF, PDB),
orbital data (Molden), trajectories (multi-frame XYZ, ASE
.traj), cube files, spectra
as HTML, full result bundles as .zip,
or any run as a standalone .py script
— each one also downloads to your computer, even from a remote
Voilà/OnDemand session.
Optional NVIDIA GPU offload via
gpu4pyscf
— RHF, UHF, RKS/UKS DFT, and TD-DFT auto-migrate to GPU
when available. Numerical IR-intensity SCFs also offload. Set
QUANTUI_DISABLE_GPU=1 to force
CPU; the result card always shows which device produced the numbers.
Four-tier calibration suite anchors a per-machine time-prediction model with GPU-vs-CPU partitioning, IQR outlier rejection, and variance-aware confidence labels. Pre-run estimates show in the Calculate tab; predicted-vs-actual accuracy accrues automatically in the analytics dashboard.
The quantui CLI inspects the
event log (log tail), probes
GPU availability (gpu check),
and builds a self-contained HTML analytics dashboard
(analytics build --open) with
GPU-vs-CPU speedup tables, method usage, and estimator-accuracy
tracking. See the CLI reference.
quantui-batch submits QuantUI
calculations to SLURM from a cluster login node over SSH, without
starting the image there: presets, chaining from another job's
optimized geometry, reruns with more memory or time, and result
summaries. quantui submit does the
same wherever QuantUI is installed.
Serve the notebook as a polished widget-only UI with
voila. Light/Dark themes,
inline log viewer, and an in-app bug-report form. Equally at
home in a research group or a classroom.
First-class transition-metal support: 14 bundled metal complexes (octahedral / tetrahedral / square-planar) with correct charge and spin, a pre-run guard that catches a metal on an incompatible basis (nudges to def2-SVP / LANL2DZ) and an impossible multiplicity, a spin-state helper that suggests high/low-spin multiplicities from oxidation state + geometry, optional GFN-FF (xtb) pre-optimization for metals, and a viewer that draws the coordination bonds.