Skip to content

Changelog

All notable changes to QuantUI are documented in this file.

The format is based on Keep a Changelog, and this project follows Semantic Versioning.

Unreleased

0.12.0 - 2026-11-12

This release also includes the fixes from a full codebase audit (2026-11-10; PRs #169, #170, #172). The most important ones change computed numbers: NMR ¹³C shifts, wB97X-D energies, and the entropy of high-symmetry molecules. Results from earlier versions in those areas should be recomputed.

Added

  • ECD (electronic circular dichroism) spectra (M-SPECTRA2 SPC2.3). UV-Vis (TD-DFT) runs now also compute rotatory strengths (velocity gauge, plotted; length gauge shown alongside) in 10⁻⁴⁰ cgs, and a new ECD panel beside UV-Vis shows the stick or Gaussian-broadened Δε spectrum (Stephens & Harada convention) with HTML/PNG export and Copy data. Enantiomers give exactly mirror-image spectra; an achiral molecule is labelled as such. Needs a diffuse basis for meaningful Rydberg bands; not available with ECP basis sets (PySCF limitation). Saved with results and batch jobs.
  • NPA charges and bond orders (M-CHEM CHEM.2/CHEM.3). The Populations panel (now Populations & Bond Orders) shows natural population analysis (NPA) charges beside Mulliken, with a Mulliken/NPA toggle for the chart and 3D colours, plus Mayer and Wiberg bond orders in a table and, optionally, as labels on the bonds in 3D. NPA comes from PySCF's natural atomic orbitals (the NAO/NPA step of NBO, not full NBO); Wiberg indices are computed in the NAO basis, as NBO programs report them. NPA charges change far less with basis set than Mulliken charges (water's O: −0.92 to −0.96 across 6-31G…def2-TZVP vs −0.31 to −0.90). Saved with results and batch jobs; for MP2/CCSD they come from the HF reference density.
  • Transition-state guess from a reactant and a product (M-TS TS.4, QST2-style). On the Transition State calc type, pick a Product (a saved Geometry Opt result with the same atoms, or pasted XYZ): QuantUI aligns it to the current molecule, builds an IDPP-interpolated path (7 images by default), computes each image's energy, starts the Sella search from the highest one and plots the path profile in the PES Scan panel. Warns when atoms pass through each other (usually mismatched atom order) or when the path has no barrier. quantui.ts_guess. Batch jobs still start from one geometry.
  • Help topics for Frequency (IR, Raman, thermochemistry), UV-Vis (TD-DFT), NMR, Reorganization Energy, implicit solvent (PCM), GPU acceleration and running on a SLURM cluster. These had no help coverage.
  • The full 3Dmol.js BSD-3-Clause license now ships with the vendored copy (quantui/data/js/3Dmol-LICENSE.txt).

Security

  • Every Voilà launch now needs a per-run token and listens on 127.0.0.1 only (quantui run app, the native launchers and the Apptainer app runscript), as quantui view already did; the JupyterLab launcher no longer disables Jupyter's token. Another user on a shared machine could otherwise drive your session.
  • A results folder is treated as data, not page code. Opening someone else's results (Viewer zip, shared folder) could inject HTML/JavaScript through cached vibrational animations, saved orbital cubes or symmetry labels, and two index files allowed reading files outside the folder. Cached animations are now signed per user; file names, script-embedded JSON and all result text shown in History, Results and error cards are checked or escaped.
  • Batch job folders are created owner-only (0700), and the SLURM partition name is validated like account/QOS before it reaches the batch script.

Fixed

  • NMR chemical shifts were offset by up to 52 ppm (¹³C). The TMS reference table did not match QuantUI's own GIAO shieldings (TMS itself came out at δ(¹³C) = −7.7 ppm at B3LYP/6-31G*, −52 ppm at RHF/STO-3G). The table is now computed by QuantUI on TMS optimized at each level, so TMS gives δ ≈ 0. NMR also refuses open-shell, post-HF and meta-GGA cases up front instead of failing with a cryptic error.
  • wB97X-D now includes its dispersion term. LibXC's functional is only the exchange-correlation part; QuantUI now adds the Chai–Head-Gordon atom-pair dispersion to energies, forces, frequencies and exported scripts (parameters for H–Xe).
  • Thermochemistry uses the right symmetry number for slightly distorted geometries (e.g. benzene σ = 12, was 4; G was 2.7 kJ/mol too low). The point group is shown in the thermo panel.
  • Open-shell HOMO–LUMO gap and orbital diagram use both spins (min LUMO − max HOMO) and draw α and β levels; the gap was α-only.
  • Structure search keeps charge and multiplicity: ions and radicals (sulfate, ammonium, OH⁻, CH₃·, O₂ triplet) no longer load as neutral singlets or fail as "Not found". Search answers from the bundled library first, offline.
  • Calculation runs: one run at a time (a double-click started two); settings are read once at Run and frozen while it runs, so changing the basis mid-run no longer mixes levels of theory; a seed geometry keeps the current charge and multiplicity; a resumed Frequency run never reuses IR or Raman displacements from a different geometry; a failed save, a late Cancel and a failed pre-optimization are reported instead of looking like success.
  • IRC reports the true end minimum when Sella's inner loop gives up (the stored end point could be the previous path point, giving wrong barriers), and mirror-image ends (NH₃ inversion) count as two minima.
  • PES scans: angle scans stay between 0° and 180° and start from linear geometries; angle/dihedral scans rotate the whole moving group instead of one atom.
  • Frequencies: an unconverged displaced SCF no longer corrupts IR or Raman intensities (they are dropped with a message); the mode-following restart from a saddle point moves the stated 0.3 Å.
  • Batch/SLURM: cancelled ("CANCELLED by …"), preempted and lost jobs are recognised instead of holding a job slot forever; a reused request id no longer overwrites a running job; requeued jobs follow the new attempt; ingested results keep their structure; quantui-batch submit --from keeps the source's solvent and refuses an unconverged source; the job monitor can no longer leave Run disabled.
  • Uploads and parsing: planar 3-D MOL/SDF files keep their geometry; charge/multiplicity are taken from a file only when it sets them; non-finite coordinates, impossible spin states and truncated XYZ are rejected; UTF-8 BOM, UTF-16, Fortran D exponents, D/T isotope labels and PDB alternate locations are handled.
  • Many smaller fixes from the same audit: History cards, exports following the run instead of the last-viewed History folder, theme CSS and Dark contrast, 1-based MO numbers everywhere, open-shell TD states no longer labelled S1…Sn, GFN-FF pre-optimization charge, C3v orbital labels, calibration and estimator fixes, launcher robustness.

  • MO symmetry labels no longer depend on the LAPACK build. On a geometry symmetric only to within the detection tolerance, a split degenerate pair (e.g. an NH3 e level) could be labelled "a2" on one platform and "e" on another, because PySCF's symmetrize_space rebuilt each irrep from an arbitrary basis. Each near-degenerate cluster of orbitals (within 5 mHa) is now rotated onto symmetry-pure combinations before labelling, and C3v labels are refused (Cs labels kept) if they would contain more a1, a2 or e levels than the AO basis has SALCs of that irrep. The 90% symmetry-purity check on orbitals now actually applies.

  • Results Viewer (pre-release audit, VA.1–VA.12) — the History + Analysis viewer (desktop installer and browser build, neither of which has PySCF):
  • Frequency results show their saved thermochemistry instead of a red "No module named 'pyscf'" error; T and P are locked, with a note, when nothing can be recomputed. The not-a-minimum warning always shows.
  • The Orbital Isosurface panel draws the saved HOMO and LUMO as before and says other orbitals, density and ESP need the full QuantUI, instead of offering buttons that failed with a "conda install pyscf" error.
  • Mode symmetry labels are saved with each Frequency result, so History and the Viewer show them without PySCF.
  • Switching tabs no longer raises an error; the header says "Viewer"; Help has a Viewer "Getting Started" and only topics that apply; the activity light (always "Idle") is hidden.
  • Browser build: the editable folder box is hidden (the files live in the tab's memory), and Report Issue explains that nothing is kept once the tab closes and links to the issue tracker.
  • Narrow windows: the folder bar and History's Load row wrap instead of pushing buttons off-screen; 3D viewers no longer overflow their panel; the energy-level diagram scrolls sideways instead of being clipped.
  • UV-Vis excited-state table readable on dark themes — alternate rows were near-white (JupyterLab's table striping) under light text.
  • Analysis structure viewer filled for NMR, PES Scan and reorganization results opened from History; it stayed blank when the geometry was saved only at the top level of result.json.
  • Calibration runs save complete results — Frequency entries now include thermochemistry, the max-force check and the saddle-point retry (one payload shared with the app and batch jobs), and Geometry Opt entries keep their Mulliken charges and dipole.
  • Energy-level diagram — HOMO/LUMO labels sit beside the plot, so a gridline at the same energy no longer runs through them.
  • Viewer demo results — the Frequency sample is computed at the optimized geometry (it was at the input geometry: max force 3.2 eV/Å, so its frequencies were not meaningful), and the TD-DFT sample keeps its orbitals.
  • History cards show the point group for every saved result. A single point reloaded from History lost its Point group row (the live card had it): the formatter never read the saved geometry block.
  • Smaller Viewer installers (M-VIEWER VW.8): Windows 338 → 208 MB, Linux 247 → 223 MB, macOS 213 → 185 MB. The Windows build no longer bundles Intel MKL (NumPy now uses OpenBLAS), and pandoc, which Voilà never calls, is left out on every platform.
  • Pasted XYZ coordinates no longer lose their first atom when a count line is followed directly by the atoms with no title line (3, O …, H …, H …): water used to load as H₂, with only a log warning.
  • The tutorial notebooks run again. All five failed: they were written for a removed cluster-submission workflow, needed $USER, or used renamed attributes. They now run in-session in seconds, and every value they quote is the value the code prints (e.g. RHF/6-31G water is −75.984 Ha, not −76.03; STO-3G lies ~667 kcal/mol above cc-pVDZ, not "30–100").
  • Documentation site: the gallery images and the Features → CLI link were broken on the published site.
  • GPU messages, a CLI note and an engine error pointed at a "Status tab" that no longer exists; they now name the System Settings tab.
  • In-app help corrected throughout to match the current UI (labels, Upload File, Compare tab, resume and warm-start details, RHF/UHF behaviour, search order).

  • quantui-batch and quantui submit refuse a method QuantUI does not support instead of warning and queueing it. The worker fails any method outside the supported list, so --method HF (or a typo) used to wait in the queue and then fail; it is now refused before sbatch, with a hint that Hartree-Fock is RHF or UHF. Accepted names are normalised to their usual spelling (cam-b3lyp becomes CAM-B3LYP).

  • quantui jobs no longer shows a cancelled, timed-out or out-of-memory job as running. A killed worker leaves its last progress.json behind, and any Slurm state other than the queued/running ones read as "running". Finished states from sacct now win: cancelled, or failed with the fix (--more-time, --more-memory, the log); a job that completed without a result is failed.
  • A SLURM Reorganization Energy job resumes after a timeout or OOM. The batch worker never handed the run its checkpoint, so quantui-batch rerun redid all two or three geometry optimizations. Each optimization now continues from where it stopped, as in the app.
  • The Help topic "Resuming an interrupted calculation" no longer says Frequency cannot be resumed, and now lists Reorganization Energy.

Changed

  • Plotly figures load plotly.js once per page from the installed package (still offline) instead of embedding ~4.8 MB in every figure and slider redraw; in-app figures are now under 200 kB.
  • Batch jobs match the app: a geometry optimization or pre-optimization is followed by the same required single point (one more SCF per job); PySCF's memory follows the SLURM allocation; failures write a full traceback to live.log; result.json records elapsed time and warnings.
  • NMR stick spectra run high → low in δ, as in the literature.
  • Packaging: py3Dmol capped below 2.6 (matches the vendored 3Dmol.js); rdkit and psutil declared directly; SPDX licence metadata; provenance files shipped in the wheel.

  • Documentation brought in line with the code: CLI reference (quantui view, gpu check --engine, submit requirements and flags, install-launcher options), SLURM batch guide, Apptainer guide (base image, build options; no .sif is published on Releases), README and docs site (GPU coverage, IRC/Raman/reorganization energy, install from PyPI with quantui run app), SECURITY.md (supported versions, subprocess use), and the contributor and agent guides.

0.11.1 - 2026-11-02

Changed

  • Documentation site: an "Open the Viewer" link in the Home sidebar and on the home page, a two-column Features page, and shorter text throughout.

Fixed

  • IRC no longer ends both directions in the same minimum. Sella's iterative estimate of the starting Hessian could carry a spurious mode, so from a transition state converged only to the force threshold the first step went sideways and both directions fell into one valley. The IRC now starts from PySCF's analytic Hessian, as the transition-state search already does. If Sella's inner loop fails, the direction is finished by mass-weighted steepest descent before a short-step BFGS; if both ends still match, the verdict says so rather than reporting a connection.
  • Viewer installer builds use the macos-15 runner (macos-14 is retired).

0.11.0 - 2026-11-01

Added

  • IRC calc type (M-TS TS.3, engine Sella): follows the reaction path downhill from a transition state in both directions, in mass-weighted coordinates, to the two minima it connects. Pick the Transition State result under Start from. The card gives the barrier from each side and a verdict; the PES Scan panel plots energy against the reaction coordinate (click a point to jump the Trajectory panel there); Trajectory plays reactant → TS → product; History replays both. Steps of 0.2 Å·amu½ with Sella's gamma at 0.4 (measured: 35–71 SCFs per direction where the defaults stalled); if Sella's inner loop gives out near a minimum, that direction is finished by minimizing and the card says so. PCM supported. Also a batch job (--calc irc, usually --from a Transition State job; no mid-path resume) in quantui submit, quantui-batch and the app's SLURM path; results prints the barriers and verdict.
  • Full symmetry labels for D4h, D4, C4v and D2d vibrational modes (square-planar, square-pyramidal and allene-type molecules) instead of their D2h/C2v/D2 subgroup labels. B1 and B2 follow the usual convention (C2′ axes, or σv planes, through the ligands), so a square-planar AB4's B1g is the in-plane stretch. Point-group detection for mode labels no longer drops to a lower group at the loose tolerance (an exact square pyramid was labelled C2v).
  • The Viewer draws the HOMO and LUMO (M-VIEWER VW.7): when orbitals are saved (app or batch job), the HOMO and LUMO, each spin for unrestricted results, are also saved as gzipped cubes at the Isosurface panel's default grid (about 0.4 MB each; frontier_cubes/). The desktop and browser Viewer, which have no PySCF, draw those; for any other orbital, density or ESP they say to open the result in the full QuantUI. Set QUANTUI_FRONTIER_CUBES=0 to skip them.
  • quantui view opens the last folder it showed when started without one (desktop shortcuts), instead of the home directory.
  • quantui jobs (M-BATCH2 B2.9): your batch jobs, newest first, with where each one is: waiting, running (with the worker's progress), done or failed. With squeue on PATH it refreshes the job records from the queue first; without it (an OnDemand session) it reads each job folder. --running lists unfinished jobs, --all every job.
  • History lists batch jobs not finished yet (B2.10): on app start and on Refresh, jobs submitted from a terminal show their state and progress from their job folders, since OnDemand has no squeue. A terminal-submitted job's record is marked succeeded when its result is saved to History.
  • Transition State as a batch job (M-TS TS.5): quantui submit, quantui-batch submit --calc transition_state and the app's SLURM path run the same Sella search and frequency check as the window, PCM included. A killed job resumes the search from its last step, never redoes a finished search, and keeps the frequency check's finished displacements. History shows the stationary point found with its search path; results prints the verdict; --from a TS job starts from its saddle point. A post-HF method is refused before the job queues. Both Apptainer images now install the [ts] extra (Sella), which Transition State needs on the cluster in the window as well.
  • Clarity and layout pass (M-UIPOLISH, from a rendered-UI walk-through):
  • Molecules open facing the camera in every viewer (principal axes), so water no longer looks linear and the dipole arrow is visible.
  • The library finds common names ("water", "ammonia", "benzene"); QM9 entries come after the curated ones and duplicates are listed once.
  • Two-column layouts on wide screens: Calculate (molecule | setup + run), Results (card | structure), and the Analysis viewer, Mulliken, Vibrational and Orbital panels; they stack on narrow windows.
  • One-line header with the toolbar, a one-line step indicator, the "Calculation complete" links right under Run, and a shorter, resizable log that hides internal bookkeeping lines.
  • Analysis lists the panels that have data first; plots opened from a collapsed panel fill their width; Raman follows the IR axis direction.
  • UV-Vis shows vacuum-UV bands (the window started at 100 nm) and lists the excited states; NMR groups equivalent nuclei and uses one Atom | σ | δ table.
  • Live and History result cards show the same rows; frequencies are listed with their symmetry labels; formulas read NH3, not H3N.
  • Compare uses checkboxes and renders its chart; History offers recent results as links with the other filters folded away; Help and Report Issue open as side drawers (Esc closes both).
  • Shorter copy throughout, a 12 px minimum text size, clearer section headings, neutral Report Issue / Exit / idle Cancel buttons.
  • Fixed: the System Settings GPU row could stay at "checking…" forever.
  • Vibrational-mode symmetry labels (M-SPECTRA2 SPC2.4) — after a Frequency calculation each normal mode gets its irreducible representation: a Sym. column and a Γvib caption in the Vibrational mode table (NH₃: 2A₁ + 2E), the label in the mode dropdown, and a Symmetry line in the IR and Raman stick hovers. Works for live results and History reloads (computed from the saved displacements). Full-group labels for abelian groups, C3v, D3h, D3, D3d, D6h, Td, Oh and linear molecules (σ/π); other groups (e.g. D4h) show labels in PySCF's computational subgroup, said so in the caption. quantui.symmetry.label_mode_irreps().
  • Diffuse basis sets — 6-31+G*, 6-31++G**, aug-cc-pVDZ and def2-SVPD join the basis menu (each next to the set it extends), for anions, radical anions and weakly bound species. Available with PySCF in the app, quantui submit and quantui-batch; PyFock's menu is unchanged. 6-31+G(d) / 6-31++G(d,p) are accepted as the same sets. The basis card marks them Diffuse, the educational notes explain them and nudge an anion on a non-diffuse basis toward one, and the basis-set help topic has a new Anions need diffuse functions paragraph. SLURM memory factors and the time estimator's basis-function table cover the new sets. config.DIFFUSE_BASIS_SETS / config.is_diffuse_basis() list them.
  • quantui results JOB_DIR — the quantui-batch results summary (convergence, energy, imaginary modes, strongest IR bands, excited states, NMR shifts) from inside the image. Both now also note a saddle-point retry.
  • Site limits for batch jobs — QUANTUI_SLURM_MAX_MEMORY_GB, QUANTUI_SLURM_ESTIMATE_MAX_MEMORY_GB and QUANTUI_SLURM_MAX_CORES set the hard limits and the automatic estimate's own memory ceiling (defaults unchanged: 128 GB, 32 cores). Set them in the image environment before quantui install-launcher so quantui-batch uses the same values.
  • Viewer mode (quantui view [FOLDER]) — a lightweight QuantUI with only the History and Analysis tabs, pointed at a results folder, needing no quantum engine. QuantUIApp(viewer=True, results_dir=...) adds a results-folder bar above the tabs; quantui view starts Voilà on a free 127.0.0.1 port and opens the browser. Packaging for two delivery routes lives in packaging/viewer/ (unsigned conda-constructor desktop installers and a Voici/Pyodide browser build), built by the new viewer-builds workflow.
  • QuantUI Viewer in the docs — new QuantUI Viewer page (/viewer/) with a link to the browser Viewer, which the Pages workflow now builds and publishes at /viewer/app/ on each release and docs change. Publishing a release also attaches the three Viewer installers to it.
  • SLURM GPU and account settings — QUANTUI_SLURM_GRES, QUANTUI_SLURM_QOS and QUANTUI_SLURM_ACCOUNT add #SBATCH --gres, --qos and --account lines to generated batch scripts (app, quantui submit and quantui-batch). A GRES that asks for a GPU also turns on apptainer exec --nv; QUANTUI_SLURM_NV=1/0 overrides that.
  • Batch Geometry Opt results are complete — a SLURM geometry_opt result now records the optimized geometry (final_molecule), the final SCF's dipole moment and Mulliken charges, so History and quantui-batch results show them without a follow-up single point. --from chaining reads the recorded geometry.

Changed

  • quantui view needs a per-run token. It listened on 127.0.0.1 with no authentication, so any local process or web page could start kernels on its port; it now starts Voila with a random token and opens the URL that carries it. A request without it gets a login page. (Standalone Voila has no XSRF setting, so the disable_check_xsrf flag was never in effect for the viewer; DEC-001's container setting is unchanged.)
  • Frequency results check the geometry — every Frequency run now computes the forces at the input geometry. When the largest force is above the Geometry Opt threshold (0.05 eV/Å) the result card (live and History) and the Analysis thermochemistry box warn that the geometry is not a stationary point, so its frequencies and thermochemistry are not meaningful, and say how to fix it. FreqResult.max_force_ev_per_ang; saved as spectra.ir.max_force_ev_per_ang.
  • IR spectra run 4000 → 400 cm⁻¹ by default (FTIR convention). An Axis dropdown on the IR panel switches back to 400 → 4000; the choice is remembered (viz.ir_axis in user settings). plot_ir_spectrum() gains descending=.
  • Energy-level diagram hides core orbitals by default — occupied orbitals more than 50 eV below the HOMO (e.g. water's O 1s) no longer set the Y range and squeeze the valence levels into a sliver; the subtitle says how many are hidden and a Show core orbitals checkbox restores them. The HOMO–LUMO gap label no longer overlaps the axis labels.
  • Dipole arrow convention — the Mulliken panel says which end the arrow head marks and offers both conventions (− → +, IUPAC/physics, default; or
  • → −, the chemistry crossed arrow), remembered as viz.dipole_arrow.
  • Frequency card — title reads "H2O · B3LYP/6-31G* (RKS)"; H and G are shown with their thermal correction vs E in kJ/mol instead of an absolute kJ/mol value.
  • Less duplication, tidier layout — viewers no longer repeat the molecule facts in a "Molecule Information" box; the launcher notebook drops its duplicate "QuantUI" heading; result tables are compact and left-aligned; the empty Calculate tab no longer reserves a blank viewer block.
  • Esc closes the Help panel (as does its Close button).
  • Clearer controls — no more truncated labels or "?.." buttons; PES atom chips read "O1"; XYZ table headers line up; a Done button closes the molecule editor; the Help panel has a Close button; System Settings groups all settings in one card and says PlotlyMol; the activity indicator is a neutral status chip; History, Compare and the Analysis heading show readable times ("2026-10-25 04:31").
  • Context-aware extras — the spin-state helper appears only for molecules with a transition metal; "Use mode → new Frequency calc" only when there are imaginary modes; Analysis panels that don't apply say what they need in their title; dark themes give 3D viewers the panel colour (also in PNG exports) instead of black.
  • import quantui no longer loads numpy, RDKit, ASE, Plotly, IPython or requests; every public name is resolved when first used. quantui command-line tools and the batch worker start in a fraction of the time, which matters on cluster login nodes (slow shared filesystems, per-user thread limits).

Fixed

  • "Copy path" in the browser viewer no longer raises: Pyodide cannot start the thread that resets the label, so the label just stays.
  • quantui-batch check prints a fix hint only under a failed or warning line. Passing lines showed their hint too ([ok ] image … followed by "ask your instructor: the course image is missing or unreadable"), which read as an error to students.
  • Generate Isosurface works again for molecular orbitals — it failed for every MO with unexpected keyword argument 'esp_range' since the ESP-map release. The orbital gallery was not affected.
  • Edit Structure stays available after a molecule loads — loading or changing a molecule rebuilt the Calculate panel without it.
  • Mulliken viewer opens charge-coloured with the dipole arrow when those boxes are ticked, instead of only after a box was toggled.
  • Sliders are visible on every theme (ipywidgets 8 sliders were unstyled: no visible track, and a white handle on the Light theme), and the selected option of toggle-button groups and History filter chips is now highlighted.
  • The QM pre-optimization checkbox no longer keeps the PES Scan wording ("…before scan") after switching to another calculation type.
  • The Files tab re-lists its folder when opened, so newly saved results show without clicking Refresh.
  • PES Scan: suggested Start/Stop values are rounded (0.001 Å / 0.1°) and the unit label sits after Stop rather than after Points.
  • Time-estimate basis-function counts — the per-element table behind calc_log.count_basis_functions had drifted from PySCF for 75 entries (every element in def2-TZVP, e.g. H 14 instead of 6; Li/Be and Na–Ar in the 6-31G, cc-pVZ and def2-SVP rows; Al–Ar in 3-21G/6-31G). It is now generated from PySCF's AO counts and a test checks every entry. History records logged before this keep their old n_basis, so estimates for def2-TZVP and second-row/heavier molecules settle as new runs are logged.
  • CPU image no longer overrides a caller's QUANTUI_RESULTS_DIR / QUANTUI_FREQ_PARALLEL (M-BATCH2 B2.12). apptainer/quantui.def hard-exported both, so a site module or job script setting either was silently clobbered inside the container. Both are now defaults (${VAR:-default}), matching quantui-gpu.def.

  • History's Refresh button also saves batch jobs that finished since the app started; before, they appeared only after a restart or a Cluster Jobs refresh.

  • The batch resource estimate knows def2-SVP, def2-TZVP and LANL2DZ (they used a generic factor), treats 6-31G(d) / 6-31G(d,p) like 6-31G* / 6-31G**, and adds walltime for TD-DFT runs with more than 10 states and PES scans with more than 10 points.
  • Batch live.log files no longer contain a copy of QuantUI's worker source. PySCF writes the running script into its output at the log level QuantUI uses; in a batch job that script is the worker (~900 lines per calculation). PySCF's version and input summary are still logged.
  • Batch job and folder names write Pople bases the usual way: 6-31G* → 6-31Gd, 6-31G(d,p) → 6-31Gdp, 6-31+G* → 6-31pGd (was 6-31Gx, 6-31Gxdxpx, …).
  • quantui submit and quantui-batch refuse an XYZ file whose first line gives a different atom count from the atoms listed (a cut-off or padded file). The app still only warns, where you see the structure.
  • quantui submit (and Resubmit) syncs job states with the SLURM queue before checking the concurrent-job limit. With no app open, finished jobs stayed "submitted" in the registry and kept counting, so after two submissions every further one was refused.
  • Generated SLURM scripts ask for one task with N CPUs (--ntasks=1 --cpus-per-task=N) instead of N tasks, which is what the single multi-threaded worker uses; a SLURM_CPUS_PER_TASK inherited from the submitting allocation no longer sets the thread count.
  • apptainer exec --nv is no longer used for CPU jobs, and the worker runs the batch image's own python instead of the submitting process's interpreter path, which need not exist in that image.
  • SLURM jobs resume again after a resubmission. Since job folders got one attempt folder per run, the checkpoint and the cached pre-opt geometry were kept inside the attempt folder, which the next attempt never sees, so killed geometry optimizations, PES scans and frequency runs restarted from scratch. Both now live in the job folder.
  • The geometry optimization before a Frequency calculation (app and batch) now converges to 0.01 eV/Å instead of 0.05. Residual forces at 0.05 left soft modes such as methyl rotors wrong, or imaginary (CHEM-3200 Lab 4). Plain Geometry Opt and the pre-opt for other calc types are unchanged.
  • When that pre-optimized Frequency run still finds an imaginary mode below 20i cm⁻¹ (usually a methyl or other rotor stuck at its eclipsed saddle point), QuantUI displaces along the mode, re-optimizes and repeats the frequencies once (app and batch). The result card and result.json (spectra.ir.imaginary_mode_retry) say what was done; a mode still imaginary after the retry is reported, not retried again.
  • Viewer installers: the shortcut opened and closed a console without starting anything. quantui view looked for the voila command on PATH, which a shortcut launch of an unactivated install does not set; it printed "Voilà is not installed or not on PATH" and exited. It now runs Voilà as python -m voila from its own interpreter and puts the install's folders on PATH for the server. The Windows shortcut runs through cmd /c ... || pause, so any future failure stays on screen. Stopping quantui view now stops Voilà too, instead of leaving it serving. The installer CI smoke test starts the viewer the way a shortcut does and fetches a page (it previously only ran quantui view --help).

  • History → "View Results" / "View Analysis" navigated by fixed tab index, so with the Cluster Jobs tab visible they landed one tab early. They now look the tab up by name.

  • Background renders (vibrational animation, isosurface, export) and the activity-light timer fall back to running inline when threads are unavailable instead of leaving the panel stuck.

0.10.0 - 2026-10-03

Added

  • Upload structure files — new Upload File tab (Calculate → Molecule Input) and Load as molecule on the Files tab: XYZ (last frame), MOL, SDF, MOL2, PDB, CIF, Gaussian .gjf/.com and .log/.out. Charge and multiplicity are read from the file when it records them; 2-D MOL/SDF drawings are embedded in 3-D.
  • Download to your computer — every export (structure files, script, bundle, cubes, PNGs, plots, CSV, animations) now offers a browser download link, and the Files tab has a Download button. Works on remote Voilà/OnDemand sessions; files over 25 MB get a notice instead.
  • Implicit solvent for optimizations, frequencies and TD-DFT — Geometry Opt is solvated at every step; Frequency uses PySCF's PCM Hessian and solvated IR displacements (Raman is skipped in solvent); TD-DFT uses an equilibrium ground state and non-equilibrium excitations with each solvent's own optical dielectric. Also in SLURM batch jobs.
  • Point groups and orbital symmetry labels — the molecule summary and result cards show the point group (0.01 Å tolerance, with a "nearly X" hint); the orbital energy diagram labels orbitals textbook-style (1a₁, 2a₁, 1b₂, 3a₁, 1b₁ for water; σ/π for linear molecules; C₃ᵥ labels for NH₃).
  • Density, spin-density and ESP surfaces — the Isosurface panel draws the electron density, the spin density (α − β) and the electrostatic potential mapped on the density surface (red = negative, blue = positive, with a colour bar), and α/β orbitals for unrestricted results.
  • Vibration viewer — displacement arrows, an amplitude slider and a table of every mode with IR intensity and Raman activity.
  • Thermochemistry at any temperature and pressure — recomputed from the stored frequencies (no new calculation), with U, H, S, G, Cv, Cp and the rotational symmetry number.
  • Edit Structure — set a bond length, angle or dihedral to a value, delete atoms, add a hydrogen, change an element, undo; pick atoms by clicking the viewer. Freeze atoms for a constrained Geometry Opt.
  • Orbital gallery — the Isosurface panel's Orbital gallery button shows the orbitals around the HOMO–LUMO gap (2, 3 or 4 on each side) as small viewers that rotate together, each labelled with its symmetry and energy; follows the α/β choice for unrestricted results.
  • Transition State calc type — searches for the saddle point near the input geometry with Sella (starting from the analytic Hessian, with the same PySCF settings as an optimization: PCM, density fitting, D3, GPU, SCF rescue), then runs a frequency calculation and says plainly whether it found a transition state (exactly one imaginary frequency), a minimum, or a higher-order saddle point. Results open on the Vibrational panel to animate the imaginary mode, plus IR, search trajectory, orbitals and populations. Needs the optional extra: pip install "quantui[ts]". Local runs only for now (not SLURM batch).
  • Populations and Isosurface for Frequency and TD-DFT results — both record the reference SCF's Mulliken charges, dipole and orbitals (both spin channels for UHF/UKS), so the Populations, Isosurface and (TD-DFT) Energies panels fill for them too, locally and from SLURM. Frequency Molden files now include the orbitals.
  • ⟨S²⟩ for open-shell results — single points record and show ⟨S²⟩ next to the ideal S(S+1), flagged when off by more than 10 % (spin contamination). Reorganization-energy results record it at all four points of each channel (s2_* fields in result.json), so a spin-contaminated ion SCF can be filtered out of a λ dataset.
  • Submit from a login node with quantui-batch — a standard-library Python 3.6+ launcher for SSH users. It never starts the image or imports QuantUI on the login node: it writes the same job folder as SlurmBackend.prepare() (constants copied from QuantUI at install time, logic held equal by tests) and calls sbatch; the calculation runs in the image on a compute node. Install once per image, e.g. apptainer exec IMAGE quantui install-launcher /shared/bin. quantui-batch submit mol.xyz --calc frequency --method B3LYP --basis def2-SVP; status, log, rerun (e.g. --mem=64G after running out of memory), cancel, estimate and path follow it. At most QUANTUI_MAX_CONCURRENT_JOBS (default 2) QuantUI jobs may be queued or running per user, counted from the real queue.
  • quantui-batch workflow commands — --preset NAME (shared presets.json beside the launcher, plus per-user presets); --from JOB starts from another job's optimized geometry, waiting on it with an afterok dependency if it is still running (the worker loads the geometry at run time, quantui/backends/batch_chain.py); --queue-rest lines jobs up behind the user's running ones instead of refusing them; a duplicate guard (--again to override); rerun --more-memory / --more-time; results JOB (energy, convergence, imaginary modes, strongest IR bands, excited states, shifts); presets; and check (python, image, Slurm commands, job folders and free space, presets).
  • Every QuantUI batch job is tagged #SBATCH --comment=quantui, so an operator can list all of them with squeue -o "%k".
  • quantui submit accepts .xyz files with --calc, --method, --basis, --charge, --mult, --solvent, --preopt and repeatable --option KEY=VALUE; an impossible charge/multiplicity is refused before anything is queued. The same flags override a request JSON's values.
  • quantui submit --prepare-only writes the job folder (request.json + submit.slurm) without calling sbatch and prints the script path. Its registry record has the new non-active status prepared.

Changed

  • The GPU image builds from the working tree, like the CPU image — quantui-gpu.def copies the same %files allowlist as quantui.def and installs it editable, instead of a pinned PyPI release, so a GPU node runs the code that was just tested. build-gpu.sh records the commit (git describe) in the image's QuantUICommit label and /opt/build-info/quantui-commit.txt; its --version flag is gone — check out a release tag to build that release.
  • Finished batch attempts reach History when the app starts, even where SLURM is unavailable (e.g. an OnDemand session), so jobs submitted from a terminal show up without opening the Cluster Jobs tab.

Fixed

  • PBE-D3 now includes D3 — it needed pyscf.dftd3, which no QuantUI environment installed, so PBE-D3 ran as plain PBE (with a warning). D3 (zero damping) is now applied through PySCF's built-in dispersion (pyscf-dispersion, added to the [pyscf] extra, the conda environment and the CPU image), including in gradients and Hessians.
  • Rotational symmetry number — thermochemistry detected symmetry at PySCF's strict tolerance, so slightly unsymmetric geometries (any optimized water or ammonia) used σ = 1, overstating the entropy by R ln σ. Symmetric molecules' S and G change.
  • quantui submit --depends-on accepts a request id or a SLURM job id and resolves it; it used to write the request id into --dependency=afterok: and validate nothing.
  • "Geometry optimization before the calculation" on SLURM — Single Point and NMR batch jobs ignored the checkbox (the local run honoured it).
  • Mode-displaced Frequency seeds are re-optimized — picking "displace along mode" as the Frequency seed disabled the pre-optimization, so the frequencies were computed at a geometry that is not a stationary point. The checkbox is now ticked for that seed (left on, user can untick).
  • Reorganization Energy shows its optimizer settings — the run used the max-force and max-steps fields while they were hidden, so values left over from Geometry Opt or PES Scan applied unseen.
  • No placeholder Raman spectrum — when Raman activities were not computed (PCM run, or pyscf-properties missing), the Raman panel drew every mode at the same height. It now says why there is no spectrum.
  • Disconnected structures are flagged for every source — the molecule summary notes separate fragments for pasted, library, uploaded, edited and History structures, not only Online Search results.
  • Pasting a multi-frame XYZ says what it is and how to load it, instead of "Line N: not enough values".
  • Export Script follows the calculation — it used to write a gas-phase single point whatever was set up. It now applies the selected PCM solvent and density fitting, adds the Hessian + thermochemistry for Frequency and the excited states (with the solvent's optical dielectric) for TD-DFT, and says plainly when it covers only the SCF (Geometry Opt, NMR, PES Scan, Reorganization Energy).
  • CPU image: NMR and analytical Raman — the image installed QuantUI without the [pyscf] extra, so pyscf-properties was missing. It now installs the extra, and the build fails if pyscf.prop.nmr or pyscf.dispersion cannot be imported.
  • Solvent names are case-insensitive; an unknown solvent is an error. A request with "water" used to run in the gas phase.
  • Export Script writes into the result folder instead of the server's working directory.

  • Solvated batch requests the worker cannot run are refused up front — quantui submit and quantui-batch reject an unknown solvent, and a solvent on a calc type the batch worker runs gas-phase only (e.g. nmr), instead of letting the job queue and then fail. The supported set now lives in one place (batch_input.SOLVENT_CALC_TYPES).

  • SLURM memory estimate for transition-metal complexes — estimate_slurm_resources() used a short element table without Mn, Co, Ni, Mo and most other metals, so a metal counted as 0 electrons and the memory estimate came out low. It now uses the full table.

0.9.0 - 2026-09-26

Added

  • SLURM job folders with per-attempt subfolders (M-JOBDIRS) — each cluster job gets a readable folder (<formula>_<calc>_<method>_<basis>, or the optional Job name on the Calculate tab) that doubles as the SLURM job name. Every run of its submit.slurm creates a new attempt-NN_job<id>/ subfolder, so Resubmit (new, Cluster Jobs tab) and a hand-run sbatch submit.slurm never overwrite an earlier attempt. The job root is configurable in System Settings (or QUANTUI_STAGING_DIR), for clusters with small home quotas.
  • SLURM results marked in History — entries show 🖥 SLURM <job id>·a<attempt> and the result card has a "Ran on" row with the job folder. Hand-run attempts are added to History on the next Cluster Jobs refresh; failed attempts stay in the job folder only.
  • GPU offload for geometry optimization — gpu4pyscf offload now reaches every SCF/gradient step of a geometry optimization, not just single points. A separate, guarded CuPy GPU path for PyFock, GPU provenance (gpu_used / gpu_name) on optimization results, and quantui gpu check --engine pyfock.
  • PyFock backend (M-PYFOCK) — Phase 1: normalized engine dispatch, native-Windows PBE/def2 single points for neutral closed-shell molecules, capability-driven calculation menus, engine provenance in logs and saved results, a pinned pyfock install extra, and a Windows water parity gate against PySCF; unsupported hybrids, ions, open-shell systems, solvent, checkpoints and orbital analysis are rejected before compute. Phase 2: geometry optimization via PyFock's ASE calculator with analytical density-fitted gradients; results keep orbital arrays, Mulliken charges, dipoles, and engine-native cube generation.
  • quantui submit CLI — submit request JSON files to SLURM from a script (with --job-name, --dry-run, resource and mail options); resource estimates account for parallel IR/Raman workers.
  • Checkpoint/resume for batch and Frequency runs — the SLURM batch worker resumes Geometry Optimization, PES Scan and Frequency runs from their checkpoints; Frequency (interactive and batch) banks each finished displacement so a restart skips completed ones.
  • SCF rescue for hard open-shell cases (M-SCF-ROBUST) — one shared convergence-rescue helper at every SCF call site, with provenance lines in the log.
  • RKS/UKS/RHF/UHF label on every result card, live and saved.

Changed

  • SLURM output layout — new submissions use the job-folder layout above instead of ~/.quantui/staging/<request_id>/. Jobs submitted earlier keep their old folders and still list, reconnect and ingest.
  • Unconverged results are no longer reported as converged — CCSD/CCSD(T) amplitudes, TD-DFT roots, frequency Hessians, and optimizer gradients from an unconverged SCF are now flagged instead of shown as a clean result. Saved cards name the step that failed rather than blaming the SCF.
  • Parallel IR/Raman worker pools are sized from the SLURM/cgroup allocation, not the node's full CPU count.

Fixed

  • Thermochemistry — entropy units no longer corrupt the Gibbs free energy; thermochemistry is saved and shown again when a frequency result is reopened from History.
  • Raman activities — added the missing a₀³→ų polarizability conversion.
  • Functionals — ωB97X-D resolves to the real functional (not wb97x+D3), and the D3 dispersion status is shown instead of silently dropped.
  • ECP systems — ECPs are passed to parallel IR/Raman workers and embedded in exported scripts; charge/spin inference handles ROHF and ECP systems.
  • Exported scripts for MP2/CCSD/CCSD(T) now run the requested method instead of a DFT branch.
  • Solvation — PCM requested for a calc type that cannot use it is rejected instead of silently ignored; solvated Geometry Optimization works in the SLURM batch worker.
  • SLURM ingest keeps computed properties and provenance (dipole, charges, SCF variant, solvent, GPU); reconnecting to a finished SLURM job no longer saves a second copy of its result to History.
  • SLURM scripts quote paths containing spaces, and no longer export an unused QUANTUI_RESULTS_DIR.
  • Checkpoints — PES Scan checkpoint identity is coordinate-aware, and the interactive Frequency run uses its own checkpoint.
  • NMR fallback-reference metadata is saved and shown.
  • Parallel IR/Raman workers honor density fitting and SCF rescue.
  • Batch results carry Mulliken charges and dipoles; UHF Molden export is complete.
  • Spectra — the x-range covers every real mode; exported NPZ files carry the fields the cube helper needs, and cube provenance can't be overwritten.
  • Packaging — the built wheel ships quantui.engines.
  • Theme — Plotly axis and zero lines follow the active palette instead of staying pale blue-gray on dark themes.

0.8.2 - 2026-08-30

Added

  • SLURM batch execution (M-CLUSTER2) — optional cluster backend when QUANTUI_ENABLE_SLURM=1 and sbatch is on PATH: submit from the same UI, monitor on the Cluster Jobs tab, hybrid squeue/sacct polling, submit cooldown, and stale-registry cleanup. All Calculate-tab calc types supported on the batch worker path.
  • QUANTUI_ENABLE_SLURM site gate — SLURM batch stays hidden and undispatchable until an operator sets this environment variable. Student CPU images and JupyterHub profiles can omit it (default off); instructors enable cluster testing with QUANTUI_ENABLE_SLURM=1 without any extra student command.
  • quantui setup and quantui run app — writes ~/QuantUI.ipynb for NCShare/JupyterLab (Render with Voilà). run app detects Apptainer + Jupyter and exits with Voilà instructions instead of opening an unreachable second port.
  • Parallel IR intensity displacements (CPU) — Settings checkbox (compute.freq_parallel) plus QUANTUI_FREQ_PARALLEL env override fans out the 6N finite-difference SCF loop on hosts with ≥4 cores and ≥2 atoms. The CPU teaching SIF sets QUANTUI_FREQ_PARALLEL=1 by default; the checkbox reflects the deployment value and locks while the env var is set.
  • Imaginary-mode perturbation for Frequency reruns seeded from History.
  • Stage-aware frequency cost estimate and per-phase run status labels (NMR, Hessian, post-HF).
  • Preset theme palettes replacing the invert filter (M-THEME THEME.6).
  • Quantum engine registry (quantui/engines/) and ComputeSettings.quantum_engine field (M-PYFOCK PYF.1 groundwork).
  • Static Raman spectrum on the Analysis tab (M-SPECTRA2).

Fixed

  • Mulliken 3D viewer could render blank after geometry persistence and lazy panel activation (#103).
  • apptainer/quantui-gpu.def never installed the xtb extra, so GFN-FF metal-capable classical pre-optimization (DEC-018) was unavailable in every GPU/HPC image — the app correctly reported the honest-failure message ("RDKit could not perceive bonds ... and the GFN-FF (xtb) metal backend is not installed"), but the underlying cause was a container gap, not a missing local install. Fixed by pinning the image's Python to 3.11 via the deadsnakes PPA (Ubuntu 24.04's own repos ship 3.12 only, and xtb's PyPI wheels stop at cp311) and adding xtb to the pip extras. quantui.def (the CPU teaching image) gained the equivalent xtb-python conda-forge package, which it had also been missing despite local-setup/environment.yml already carrying it for local conda installs.

0.8.1 - 2026-08-26

Added

  • PNG export on the geometry-optimization trajectory, vibrational mode, and molecule viewers — the same camera-accurate Save-PNG capture already available on the orbital isosurface and reorganization-energy geometry viewers now covers every 3-D viewer in the app. The molecule viewer (shown on the Calculate, Results, and Analysis tabs) captures only on the py3Dmol backend, since Plotly already provides its own "download plot as PNG" button; each of its three tabs gets an independent capture button so a save from one tab can never be mislabeled with another tab's molecule or overwrite another tab's file.

0.8.0 - 2026-08-25

Added

  • Transition-metal and coordination-complex support. QuantUI now recognizes and handles metal centers throughout the pipeline instead of erroring or silently mishandling them: a pre-run guard catches basis-coverage and charge/multiplicity problems before a run starts (with a one-click "Switch to def2-SVP" fix when that alone resolves it); a GFN-FF (xtb) pre-optimization backend actually relaxes coordination complexes, since RDKit's organic valence model can't; the 3D viewer falls back to py3Dmol instead of showing a "Visualization failed" box; a fetched structure that resolves to a disconnected salt (e.g. cisplatin as separate NH₃/HCl/Pt fragments) is flagged before you compute the wrong geometry; coordination bonds are drawn dashed since 3Dmol.js's own bond perception draws none to a metal; and an oxidation-state → spin-state multiplicity suggestion engine proposes — never auto-sets — physically reasonable multiplicities, naming the strong- vs weak-field ligand ambiguity where one exists.
  • 14 bundled inorganic examples (up from 3), GFN-FF-validated, spanning octahedral (aqua and cyanide, high- and low-spin), tetrahedral, and square-planar geometries — a teaching spread across spin states and charges.
  • Density fitting (RI) as an opt-in SCF speedup — a large win for TD-DFT and larger systems, roughly neutral for small single points, so it defaults off. Shown on/off in the run-header provenance banner and flagged on the result card ("⚡ RI") whenever it was actually applied.
  • Mulliken Populations Analysis panel. Per-atom Mulliken charges as a table and bar chart after Single Point and Geometry Optimization runs, with its own 3D viewer below the energy-level diagram: atoms colored by charge, a center-of-mass dipole arrow, a color-vividness slider, and a magnitude-only fallback when an older History result has ‖μ‖ but no saved vector.
  • NMR stick plot with a ¹H/¹³C nucleus toggle, mirroring the existing IR/UV-Vis panels; UV-Vis gains λ min/max axis-bound inputs.
  • Click-to-measure atom selection on the Analysis-tab viewer — click 2, 3, or 4 atoms to read a bond length, angle, or dihedral, GaussView-style (py3Dmol only for now; the Plotly viewer shows an explanatory message instead of a silent no-op).
  • Interactive XYZ input. An atom-row builder and a cleanup-coordinates accept/reject preview on the Calculate tab. Loading no longer enforces charge/multiplicity parity at parse time — an odd-electron geometry loads with a note instead of an error — and charge/multiplicity now stay owned by Calculation Setup.
  • Suggest charge & multiplicity from geometry — a Calculation Setup button proposes a neutral charge and a parity-compatible multiplicity from the loaded structure, with an explicit Apply step rather than a silent auto-fill.
  • Export provenance. Cube, PNG, and XYZ exports now carry method, basis, and resolution (plus charge/multiplicity for XYZ) as embedded metadata, not just filenames. Reorganization Energy gained its own XYZ export — one file per distinct geometry — and a generalized PNG-capture bridge shared with the isosurface viewer.
  • Reorganization Energy is checkpointed. The run most likely to be interrupted — two geometry optimizations plus four SCF energies under one Calculate-tab run — can now resume mid-leg instead of restarting from scratch.
  • Orbital isosurfaces gained smoothed meshes (no more visible marching-cubes facets) and a wireframe finish toggle.
  • TD-DFT's live status label now surfaces per-root convergence ("TD-DFT root 3 converged @ 16.663 eV") during the excited-state solve, instead of only a generic heartbeat.
  • A representative example gallery in the README and GitHub Pages site, and new documentation for inorganic/coordination-complex support.

Fixed

  • Heavy-element runs on an ECP basis were silently wrong. Cisplatin and similar complexes on LANL2DZ or def2 were built with the basis set but no effective core potential attached, so PySCF ran all-electron against a valence-only basis — converging to a nonphysical energy (positive HOMO) with garbage gradients, which made a geometry optimization look like it was diverging. ECPs are now attached consistently across every calculation type.
  • Frequency time estimates for app runs were systematically inflated. The estimator's fallback cost model wasn't told which pool a run came from, so small-molecule Frequency estimates drew on the calibration pool's fresh-subprocess import overhead instead of the app-run pool.
  • Density fitting's "⚡ RI" badge could silently vanish. TD-DFT, NMR, Frequency, and Geometry Optimization results discarded the density-fitting flag instead of storing it, and results were never persisting it to disk for any calculation type — so even a live single point's badge disappeared on History replay.
  • A warm-start rejection could abort an otherwise-good calculation. A GPU-migrated mean-field rejecting a host density, or an incompatible chkfile density, now logs a warning and retries from a scratch guess instead of failing the whole run.
  • Resuming a resumed-and-interrupted-again optimization under-reported its progress — the BFGS step counter wasn't seeded from the checkpoint, so it restarted counting at 0 instead of continuing from what had already been banked.
  • The resume list's Restore/Discard buttons now log an unexpected exception instead of letting it escape into ipywidgets' click dispatch, matching every other action button.
  • Converged-but-near-degenerate metal runs no longer show a stale "HOMO == LUMO" warning (observed on ferrocene) — the check now looks at the converged gap, not PySCF's initial-guess density, which is always near-degenerate for a bare d-manifold starting guess.
  • Click-to-measure's highlight halo no longer sits buried inside larger atoms — it was a fixed radius; ball-and-stick atoms render at a radius scaled from each element's VDW radius.
  • Analysis-tab polish: the welcome logo's SVG fill token, accordions expanding automatically when they have data, scroll-to-top on tab entry, and a shared seed-geometry dropdown for NMR Shielding.
  • Older History results without saved Mulliken data now say so explicitly, with a magnitude-only note for dipoles recorded before the full vector was saved, instead of a blank or misleading panel.
  • The CPU Apptainer build crashing on a floating miniforge3:latest base image — libmamba choking on a fuzzy version pin. See apptainer/quantui.def.

Changed

  • Three separate calc-type label→key mappings, and three separate implementations of the Hill-formula algorithm, had drifted apart; each is now a single shared helper.
  • python -m quantui.estimator_eval's replay is now linear time instead of quadratic.
  • ~380 scattered hex color literals across the chrome files are now one theme token map.
  • Real type checking restored in CI.

0.7.0 - 2026-08-06

Fixed

  • Running the test suite no longer corrupts your time estimates. QuantUI's own tests were writing their runs into ~/.quantui/logs/perf_log.jsonl — the file the runtime estimator learns from. Because those tests use a simulated calculation, each one recorded a fabricated timing: 2 773 records were in the log and roughly four fifths of them were test artifacts, including "water frequency" runs whose recorded times ranged from 0.34 s to 143 s for identical chemistry. This is the reason the pre-run estimate had been unreliable. Only developers running the test suite were affected; the fabricated records are now superseded automatically as real runs accumulate, so no manual cleanup is needed.
  • Calibration now times the calculation, not the process. Each calibration step runs in a fresh subprocess and its stopwatch started before PySCF was even imported, so every calibration record was inflated by startup cost that a real run in an open session never pays. Import time is still recorded, just separately.

Added

  • Checkpoints — an interrupted calculation can pick up where it stopped. Geometry Optimization saves its trajectory and the optimizer's accumulated curvature after every step; PES Scan banks each point as it finishes. If a run is cancelled, crashes, or the machine goes to sleep, the Calculate tab offers to resume it — and says how much is already done ("8 of 20 scan points already computed") rather than just asking. The offer appears only when the calculation you have configured is exactly the interrupted one, geometry included, so a resume can never splice two different runs together.
  • An "Unfinished calculations" list on the History tab. After restarting QuantUI you no longer have to remember what you were running: every interrupted calculation is listed with its molecule, type, level of theory, how much finished and how long ago. Load these settings puts the molecule and all its settings back on the Calculate tab, ready to resume; Discard removes one you are done with. Hidden entirely when nothing is unfinished.
  • Checkpoint activity is recorded in the saved output log. Opening a checkpoint, every save, completion and discarding all appear as [checkpoint] lines in pyscf.log. Resuming writes a banner stating plainly that the log covers only the continuation and that the earlier output lives in the interrupted run's result directory — without it the file would read as a calculation that started from the geometry at the top. A warm start now names the file its initial density came from, since the SCF iteration count is only interpretable if you know what it started from.
  • Help topic: "Resuming an interrupted calculation." Covers how to resume, why the offer disappears if you change a setting, which calculation types can be resumed, and where checkpoints are stored.
  • Warm-started SCF. A converged density from an earlier run of the same molecule, charge, method and basis is reused as the starting guess, which usually cuts several SCF cycles. The geometry does not have to match — a density from a nearby geometry is a good guess, which is what a geometry optimization relies on internally.
  • Calculation records say how they were measured. Runs launched from the app and runs measured by the calibration tool are now labelled as such, and the estimator keeps them apart instead of averaging two populations that measure different things. Records also carry a per-stage time breakdown (SCF, Hessian, excited-state solve, …), which is the groundwork for stage-aware estimates.
  • python -m quantui.estimator_eval — replays your recorded history through the estimator and reports how accurate it would have been, split by calculation type. Reports coverage alongside accuracy, so a model that stays silent can't look good by refusing to answer.

0.6.1 - 2026-08-05

Fixed

  • Reorganization-energy results now survive History. Reloading one from History showed the calculation's headline numbers as missing — λ, the four-point energies and the per-channel breakdown were never written to disk, so there was nothing to redisplay. Both the live and History result cards now render from the same saved data, so anything visible during a run is visible after reloading it. Results saved before this cannot be recovered — λ requires two geometry optimizations and four SCF energies. Those results now say so on the card and tell you to re-run, rather than appearing silently incomplete.
  • 3-D backgrounds follow the theme in the new geometry viewers, matching the fix applied to the other viewers in 0.6.0.

Added

  • A Geometries panel for reorganization energy, on the Analysis tab — which previously showed nothing at all for this calculation type. Step through the distinct geometries behind the four energies, or overlay two of them with arrows marking how each atom moved. The arrows can be scaled ×1–×10 for small relaxations; the structures always show true computed positions.
  • Geometry relaxation on the result card — RMSD and the largest single-atom shift between the neutral and ion geometries, which is what λ physically measures.
  • Compare λ across saved results. Selecting several reorganization-energy runs in the Compare tab now shows a λ table with a column per channel, for screening candidates by how much they reorganize.

Changed

  • The NCShare GPU diagnostic notebook now reports the CPU affinity mask alongside the core count, and states plainly when CPU timings cannot be trusted. A job can be granted 6 cores while still seeing 192; whether that matters depends on how the scheduler constrains it, and the core count alone cannot distinguish the two cases.
  • apptainer/README.md records the first verified H200 run — driver, compute capability, partition, and the measured CPU/GPU crossover.

0.6.0 - 2026-08-04

Added

  • Save orbital isosurfaces as images. A Save PNG button sits under the viewer and captures exactly what you are looking at, camera and all. Options for the filename, the DPI written into the file (72–600; this sets the print size, not the pixel count), and a transparent background for dropping a figure onto a slide — the transparency applies only to the saved file, so the viewer on screen is unchanged.
  • Tunable isosurface resolution — Coarse/Medium/Fine/Very fine grids, remembered between launches.
  • Live isosurface controls: isovalue, opacity and a colour-scheme dropdown (GaussView orange/blue, Avogadro/Jmol red/blue, journal yellow/blue, and more). All three update the existing viewer without recomputing anything, so the orientation you rotated into is preserved.
  • The isovalue is explained in chemical terms. Beside the slider, QuantUI reports what fraction of the orbital's density the surface actually encloses — "encloses 98.4% of the density" — because an amplitude threshold is rarely the question anyone has.
  • A Cancel button for isosurface generation. The underlying PySCF call cannot be interrupted, so the computation finishes in the background, but its result is discarded and the controls come back immediately.
  • The molecule is shown before you generate anything, so the panel is never empty — and you can orient the structure first, since that view carries into the isosurface.

Changed

  • Orbital isosurfaces are py3Dmol only. The Plotly renderer is no longer selectable for them: it downsamples the volume by construction, and it cannot carry the new image export. Plotly remains available for the molecule viewers.
  • Exported animations now match what you saw. Vibrational modes and optimization trajectories both export from py3Dmol — the renderer that drew them on screen — instead of being rebuilt in Plotly. Exported files are also now complete HTML documents declaring UTF-8, so labels no longer risk mojibake when opened from disk.
  • For geometry optimizations, the Orbital Isosurface panel opens by default on the Analysis tab.
  • Isosurface viewers are framed like the other 3-D viewers, and generating a new one dims the current one in place rather than collapsing the panel.

Fixed

  • 3-D backgrounds follow the theme. The isosurface, vibrational and trajectory viewers, and the Analysis-tab molecule viewer, kept their old background when switching Light/Dark until something happened to redraw them.
  • The page no longer jumps when generating an isosurface.
  • Isosurface changes now apply correctly. Adjusting the isovalue, opacity or colours previously layered a new surface on top of the old one, so raising the isovalue appeared to do nothing, lowering opacity did nothing, and switching palettes made the surface progressively brighter.
  • Very large orbital grids are dramatically lighter in the browser — the cube data was being embedded three times per render.

0.5.2 - 2026-08-03

Fixed

  • Dark mode now has visible panel edges. Borders, cards, and dividers were effectively invisible in Dark mode — measured at 1.14:1 against the panel background, well under the 3:1 that WCAG asks of a UI component. Dark mode is a whole-page colour inversion rather than a separate palette, so a light grey border inverts to a near-black one on a near-black page and disappears. The borders are now a mid-tone that stays visible in both modes (3.20:1 light, 4.30:1 dark). Text contrast was measured too and was never the problem — it is 7.24:1 in light mode and slightly better inverted — so text colours are deliberately unchanged.

Added

  • Every 3-D viewer now has a border around it, so it is clear where the view ends and where the page can be scrolled without dragging the structure. This covers the molecule preview, the results and analysis structure views, the geometry-optimization trajectory, the vibrational-mode animation, and the classical pre-optimization preview. The frame hugs the plot itself rather than spanning the window width, and looks the same whichever rendering backend (py3Dmol or plotlymol3d) is active.

0.5.1 - 2026-07-31

Fixed

  • The "Geometry optimization before calculation" checkbox no longer gets stuck disabled. Selecting a seed geometry on Frequency or UV-Vis disables that checkbox (a seed is already an optimised geometry, so re-optimising first would be redundant) — but switching to a different calculation type afterwards used to leave it disabled with no way to re-enable it short of going back and clearing the seed selection.
  • The Geometry Opt seed dropdown now updates when you change molecules while already on that panel. It previously only refreshed when you switched into the Geometry Opt calculation type, so loading a new molecule without leaving the panel could leave it listing matches for the wrong molecule.

Changed

  • Internal cleanup: the Geometry Opt / Frequency / UV-Vis seed-geometry dropdowns, refresh buttons, and status notes are now one shared widget group instead of three near-identical copies. No user-visible change — the same dropdown, filtering, and messages appear on each calculation type as before.

0.5.0 - 2026-07-31

First release published to PyPI — pip install quantui.

Added

  • Seed geometry for Geometry Optimization runs — a run can now start from the final geometry of a previous optimization instead of the current molecule, which makes the standard "optimize at a cheap level of theory, then refine at a higher one" workflow a single dropdown choice. Frequency and UV-Vis already supported this; Geometry Opt now matches them.
  • A "still working" heartbeat in the live log. Long silent phases — the TD-DFT excited-state solve most of all — could leave the output log unchanged for minutes while the calculation was running normally, which reads as a hang. The log now reports … still working — <stage> · <elapsed> whenever it has been quiet for 25 seconds. The saved pyscf.log is unaffected; it stays a faithful record of the calculation's own output.
  • Basis-set names now show their alternate notation. 6-31G* and 6-31G(d) are the same basis set written two ways, and nothing in the UI said so. The basis card now notes the equivalent spelling, and the basis-set help topic explains the star/parenthesis convention, the +/++ diffuse markers and when anions need them, and why Dunning sets (cc-pVDZ) have no star at all.
  • The launcher terminal now prints the QuantUI wordmark instead of two bare lines of text.
  • Reorganization energy (Marcus 4-point) — a new "Reorganization Energy" calculation type that computes the internal reorganization energy λ for hole (cation) and/or electron (anion) charge transfer. It optimizes the neutral and ion geometries and evaluates the four single-point energies of the 4-point scheme (λ = [E_ion(R_neutral) − E_ion(R_ion)] + [E_neutral(R_ion) − E_neutral(R_neutral)]), reporting λ, its λ₁ (ion) and λ₂ (neutral) relaxation components, in eV and kcal/mol. mode="both" shares the neutral optimization across both channels. Open-shell HF ions are automatically promoted to UHF.
  • One-click "Calc. Reorganization Energy" button — sets the calculation type to Reorganization Energy, defaults the channel to both hole + electron, and launches the run in a single click.
  • GPU offload can now be switched off from the UI — Status tab → Settings → "Use GPU when available". The preference persists across launches. This exists because GPU offload is not always faster: quantum-chemistry SCF is double-precision throughout, and consumer/workstation GPUs gate FP64 to roughly 1/32–1/64 of their FP32 rate, so offload on such a card can be slower than a many-core CPU. QUANTUI_DISABLE_GPU=1 still overrides the setting for scripted runs.
  • A warning when a detected GPU is unlikely to help. quantui gpu check and the Status tab now flag consumer-class devices with a note that double precision is weak on them and offload may be slower than CPU, instead of presenting any detected CUDA device as free speed.

Changed

  • The live calculation log no longer snaps back to the bottom while a run is in progress, so you can scroll up and read earlier output mid-calculation. It still follows new output automatically when you are at the bottom. The log is now a container QuantUI owns and appends to, rather than a widget that rebuilt itself on every line.
  • No pre-optimization preview when the geometry barely moves. Previously a relaxation that changed essentially nothing still opened an animation pane and a Keep/Revert choice between two effectively identical structures. Below 0.05 Å RMS displacement QuantUI now just reports the number and proceeds with your geometry.
  • All launcher scripts moved into a launchers/ folder to declutter the repo root. Behaviour is unchanged — each launcher now resolves the repo root as its parent directory. launchers/launch-app.bat still finds quantui.sif next to itself (student download) or in the repo root (dev build). Docs and shortcuts that referenced the old top-level paths were updated.
  • Apptainer container now builds on condaforge/miniforge3 with the mamba (libmamba) solver. The previous continuumio/miniconda3 base used conda's classic solver, which could hang for hours resolving the PySCF + RDKit + JupyterLab dependency set. The build now installs from conda-forge only (--override-channels, strict channel priority), which resolves in minutes and avoids the base image's implicit defaults channel. No change to the shipped package set or runtime behavior.

Fixed

  • The live log and run header render in a fixed-width font again. A system-font rule was applying to the log, which garbled the ASCII wordmark and pushed the run header's Label : value columns out of alignment.
  • quantui gpu check no longer reports a broken CUDA install as "gpu4pyscf not installed". ModuleNotFoundError is a subclass of ImportError, so catching the latter conflated "the package is absent" with "the package is present but its CUDA libraries are missing" — the second case was reported as the first, sending users back to an install step they had already completed. The two are now distinguished, the underlying exception is included in the message (e.g. the missing libnvJitLink.so), and both cases are logged. The reason string now comes from the detection probe itself rather than being re-derived by the CLI, so the message can no longer contradict what the run dispatcher decided.
  • Exit is now a two-stage confirmation, so one click can no longer tear down an HPC allocation. Exit shuts the server down by sending SIGTERM to the parent process. On a laptop that parent is just Voilà, but on a cluster interactive session (e.g. NCShare) that parent process is the job — so a single stray click destroyed the entire allocation. The first click now only arms the shutdown: Exit re-labels to "Confirm shutdown" and an inline warning plus a Cancel button appear. Only a second click runs the shutdown, which is otherwise unchanged. An armed Exit auto-reverts after 15 seconds so a stray click doesn't leave the button primed.

0.4.1 - 2026-07-16

Bug-fix and hardening release from a full repository audit. No new features and no breaking changes — it fixes correctness bugs across orbital isosurfaces, Molden export, the XYZ parser, NMR references, PES scans, and IR intensities, plus a range of edge-case, packaging, and hygiene issues found by reading the codebase end to end.

Added

  • Single-atom molecules (e.g. a lone He or Ne atom) are now accepted as XYZ input and run end to end — atomic calculations are legitimate targets.

Changed

  • Python 3.9 is now covered by CI. The minimum supported version was previously claimed but untested; adding it surfaced and fixed real geometry-optimization and frequency failures on the older ASE that resolves for Python 3.9.
  • Faster quantui CLI startup — the command-line tool no longer imports the full notebook/GUI stack (ipywidgets, IPython, the app module) just to tail a log.
  • No more logging hijack — importing quantui no longer reconfigures the root logger, so it won't override or duplicate logging in a host application or notebook.
  • Faster time estimates over long sessions — the performance log is cached and re-read only when it changes, instead of being fully parsed on every UI update.

Fixed

  • Orbital isosurfaces for charged and open-shell molecules — the isosurface viewer previously errored for every odd-electron system (radicals, and ions such as H₃O⁺, NH₄⁺, OH⁻); charge and spin are now carried through to the cube generator.
  • Molden export from frequency calculations — the exported geometry was inflated (Bohr coordinates read as Ångström) and internally inconsistent, so it rendered wrong in Avogadro/IQmol; frequency geometries are now stored and written in the correct units. This also fixes history replay of frequency orbital isosurfaces.
  • Export Script and method notes for wB97X-D — "Export Script" failed with a "method not supported" error and the educational notes silently disappeared for wB97X-D; mixed-case method names are now matched correctly.
  • 2D structure images from XYZ input — this path never worked (it always returned nothing) and now renders.
  • XYZ files with a blank or comment (#) title line no longer silently drop the first atom — a very common file layout was losing an atom with only a warning.
  • NMR reference provenance — chemical shifts no longer silently fall back to the B3LYP/6-31G* TMS constants for other method/basis combinations; the reference that was actually applied is now recorded, and the lookup is case-insensitive.
  • IR intensities for open-shell / UHF-singlet frequency runs no longer abort the whole IR-intensity step.
  • PES scans — a failed scan point no longer records a bogus geometry frame or poisons the energy/barrier statistics with NaN, and angle/dihedral scans work with current ASE releases.
  • UHF calculations now report a dipole moment and Mulliken charges (these were silently skipped).
  • Clearer errors for unsupported input — elements outside the supported H–Kr range, and post-Hartree–Fock methods (MP2 / CCSD / CCSD(T)) requested for calculation types that don't support them, now fail with an explicit message instead of a misleading one or an uninformative crash.
  • Windows exports no longer fail when the basis-set name contains * (e.g. 6-31G*) — export filenames are sanitized.
  • Method-notes panel renders bold text correctly instead of leaking literal ** markers.
  • Robustness — the event log is no longer subject to lost entries under concurrent writes; a thumbnail-save failure no longer aborts saving a result; PubChem availability checks respect the configured throttle and timeout; and results carrying NumPy scalar values now save correctly.

0.4.0 - 2026-06-18

Interactive-visualization and offline-readiness release. Adds molecular-orbital isosurfaces and an interactive pre-optimization preview, reworks the 3D viewers to preserve camera orientation across frames, makes all 3D rendering work offline, and substantially speeds up startup.

Added

  • Molecular-orbital isosurfaces (py3Dmol) — interactive HOMO / LUMO / MO isosurface viewer rendered with py3Dmol; the downsampled Plotly path remains a fallback.
  • Interactive classical pre-optimization — a Preview button relaxes the geometry with a fast bonded force field (RDKit MMFF94 → UFF) and animates the relaxation in place; Keep this geometry adopts it as the active structure or Revert discards it. Stepper controls (play/pause, prev/next, scrub slider, and an input ⇄ relaxed flip) let you compare geometries, and the captured trajectory is sampled at even RMSD spacing for smooth playback.
  • Cancel button — stop a running calculation cooperatively at the next SCF cycle / optimization step.
  • Live vibrational-animation framerate — the Vib fps setting updates the running animation immediately.

Changed

  • Single persistent 3D viewers — the trajectory and vibrational-mode viewers now load all frames into one py3Dmol viewer and switch frames/modes client-side, so the camera (rotation/zoom) is preserved across steps and modes and there is no per-frame rebuild or flicker.
  • Pre-optimization is Preview-only — the silent "classical pre-optimize" checkbox is gone; pre-optimization happens only through the transparent Preview → Keep/Revert flow, so nothing relaxes the geometry invisibly. (The separate QM "geometry optimization before calculation" option is unchanged.)
  • Offline-first 3D rendering — 3Dmol.js is vendored and loaded per-view from a local data: URI instead of a CDN, so every 3D view works with no network (the build fails if the vendored asset is missing). Native launchers tolerate offline pip install.
  • Faster startup — GPU detection and History/Compare population are deferred off the synchronous construction path, so the UI paints in ~1 s instead of ~15 s; the GPU status badge and dropdowns fill in shortly after.
  • Clear of the live calculation log is disabled while a calculation runs.

Fixed

  • GPU-offloaded result extraction — HOMO–LUMO gap, dipole moment, and Mulliken charges are now reported for GPU runs. CuPy arrays are copied to host before extraction, and Mulliken falls back to the CPU object (gpu4pyscf does not implement population analysis on the GPU).
  • Vibrational animation glitchiness — stacked animation loops (a new loop started on every mode switch) made playback jittery and too fast and ignored the framerate setting; exactly one loop now runs.
  • Cancel status no longer sticks on "Cancelling…" after a calculation is cancelled.
  • Stale run status — "Pre-optimized geometry accepted." is cleared when a new molecule is loaded or a preview is reverted.
  • Structure provenance is reported and the input viewer is persisted across reloads.

0.3.0 - 2026-06-11

Structure-sourcing release. Repairs the external-database structure search and replaces the 20-entry inline molecule list with an indexed, three-tier bundled library, alongside visualization and result-card fixes.

Added

  • External structure search — resolve molecules by name, PubChem CID, InChI, InChIKey, SMILES, or CAS number. Input is routed by type: SMILES/InChI resolve locally via RDKit with no network, while names and identifiers query PubChem through a hardened client (URL-encoding, request throttling, bounded retry on server throttling). NCI CACTUS acts as a fallback resolver, and an offline bundled-library fallback keeps the search usable without a network connection.
  • Disambiguation pick-list — an ambiguous query (e.g. a name with several PubChem matches) presents a selectable list instead of silently choosing the first hit.
  • Three-tier bundled molecule library — 20 presets, 156 curated named molecules, and ~1,900 bulk QM9 structures (CC0), held in an indexed, lazily-loaded package-data store. Bulk entries are reachable via search.
  • Library browse/search tab — category filter plus name/formula search, replacing the flat preset dropdown.

Changed

  • The molecule preset list moved from an inline config.py dictionary into the indexed library store; config.MOLECULE_LIBRARY is preserved as a backward-compatible accessor.
  • Frequency / UV-Vis seed-geometry dropdown entries are now labeled as optimized geometries so their source is explicit.

Fixed

  • Orbital isosurface could exhaust browser memory — the full volumetric grid was serialized into the Plotly figure. The rendered surface is now downsampled to a bounded point count (the saved cube file remains full-resolution).
  • CCSD / MP2 result cards now show the HF reference and correlation-energy breakdown (plus the (T) triples correction for CCSD(T)); these fields are also persisted with saved results.
  • The live calculation log no longer jumps to the top while streaming output.
  • Structures fetched as 2D records are re-embedded in 3D, and salt/counterion fragments are separated so bond perception does not misread them.

0.2.0 - 2026-05-22

First substantial release after v0.1.0. The codebase moved from a single monolithic app.py to a modular package, added six PySCF-backed calculation types end-to-end, introduced a results-persistence layer with history replay, and shipped a complete visualization stack (3D viewer with selectable backend, trajectory animation, IR/UV-Vis/PES plots, orbital isosurfaces, vibrational mode animation with caching). UI runs as a Voilà app suitable for classroom deployment.

Added

Calculations

  • Geometry optimization (optimizer.py) — ASE-BFGS driver around a custom PySCF calculator; per-step trajectory persisted.
  • Vibrational frequency analysis (freq_calc.py) — Hessian via pyscf.hessian, ZPVE, thermochemistry (H/S/G at 298 K), IR intensities via pyscf.prop.infrared or a numerical-derivative fallback for compatibility across PySCF versions.
  • TD-DFT UV-Vis (tddft_calc.py) — excitation energies, oscillator strengths, wavelengths; full spectrum plot in the Analysis tab.
  • NMR shielding (nmr_calc.py) — GIAO shielding via pyscf.nmr (core preferred over pyscf-properties to dodge a known upstream bug); ¹H/¹³C chemical shifts relative to TMS.
  • 1D PES scan (pes_scan.py) — bond / angle / dihedral; energy profile + per-step geometry animation.
  • PCM implicit solvent — Water, Ethanol, THF, DMSO, Acetonitrile via a single checkbox in the Calculate tab.
  • MP2 post-HF method support.

Analysis & visualization

  • Analysis tab with 8 always-in-DOM panels (Energies, Trajectory, Vibrational, IR Spectrum, PES Scan, Isosurface, UV-Vis, NMR) wired through a _PANEL_REGISTRY so live runs and history replay share one code path.
  • IR spectrum chart (ir_plot.py) — stick plot + Lorentzian-broadened curve; broadening toggle and FWHM slider.
  • UV-Vis spectrum plot — Plotly chart with wavelength/energy axes.
  • Orbital visualization (orbital_visualization.py) — energy-level diagram (matplotlib → Plotly HTML) and cube-file isosurface viewer with HOMO-1/HOMO/LUMO/LUMO+1 toggle.
  • Trajectory animation — atomic Output-children swap to avoid Voilà-deferred-display blank frames; py3Dmol-only render path with prev/next arrow navigation.
  • Vibrational mode animation — py3Dmol multi-frame XYZ renderer with amplitude scaling, prev/next mode nav, and dropdown skipping near-zero modes.
  • 3D visualization backend router (viz_backend_router.py) — pure function that picks py3Dmol or plotlymol3d per VizTask based on user preference and runtime availability; immutable Decision carries chosen backend, fallback, and reason.
  • Lifecycle telemetry — _viz_render_event context manager emits viz_render_start / viz_render_done / viz_render_error JSONL events with backend, task, elapsed_ms, and extras at every render dispatch.
  • Side-by-side Compare tab — pick any two saved calculations and view a diff table.

Persistence & logging

  • Results storage (results_storage.py) — every run is saved to a timestamped directory containing result.json (schema v2, additive-only), pyscf.log, optional trajectory.json / orbitals.npz / thumbnail.png.
  • History tab — browse and replay saved calculations after a kernel restart; replay path is identical to live-run analysis activation.
  • Performance log (calc_log.py) — perf_log.jsonl per converged run + event_log.jsonl for startup/calc/error events; 7-day auto-prune.
  • Time estimator — 4-strategy priority chain (N_basis-normalised → cross-method electron-count) populates "Estimated time" before each run.
  • Benchmark suite (benchmarks.py) — one-click calibration suite to populate the time-estimator history with real machine data.
  • Issue tracker (issue_tracker.py) — in-app bug-report UI writing to a local issues.db.
  • Persistent user settings (user_settings.py) — stored at ~/.quantui/settings.json (override via QUANTUI_SETTINGS_PATH). Schema is section-based for additive growth, with atomic writes and graceful fallback to defaults on corruption.
  • Vibrational-animation disk cache (vib_cache.py) — per-result-dir vib_frames/ of pre-rendered py3Dmol HTML keyed by (mode, n_frames, amplitude, renderer, fps). Mode switches on repeat visits and history replay are instant.
  • Vib FPS user preference — viz.vib_framerate_fps exposed as an IntSlider in the Status tab (clamped 1–120, default 10); included in the vib cache key so changing FPS invalidates cleanly.

UI

  • Modular UI package — app.py (orchestration) plus app_analysis.py, app_builders.py, app_exports.py, app_formatters.py, app_history.py, app_runflow.py, app_visualization.py.
  • Seven-tab layout — Calculate, Results, Analysis, History, Compare, Log, Status — with a floating Help overlay (not a tab).
  • Light / Dark theme selector — dark by default on startup.
  • Status tab — environment info, performance-history accordion (two-step reset), default-3D-backend toggle, vib-FPS slider.
  • Files tab + activity indicator for browsing saved results.
  • Plot export UI — save IR, UV-Vis, PES, orbital diagram plots as HTML.
  • Scroll guard for the run output area to keep long PySCF logs from jumping the page.
  • Welcome header, completion banner, structured log header/footer.
  • Compare-tab Copy-path button (replaced a broken Open-folder action).
  • Result directory label + log accordion showing inline pyscf.log.
  • Structure exports — XYZ, MOL/SDF, PDB, plus a standalone runnable .py script export.

Tooling & dev

  • Test suite grew from a handful to 1004 passed / 97 skipped (Windows quantui-win env baseline; the 97 skips are PySCF-gated Linux-only tests).
  • New analysis-history end-to-end tests for every calc type (test_sp_analysis_history.py, test_geo_opt_analysis_history.py, test_freq_analysis_history.py, test_tddft_analysis_history.py, test_nmr_analysis_history.py, test_pes_scan_analysis_history.py).
  • test_code_quality.py enforces:
  • No include_plotlyjs="cdn" anywhere (fails silently in offline Voilà).
  • No bare except: pass blocks.
  • test_viz_backend_router.py + test_viz_backend_sync.py — full task × preference × availability matrix and Calculate/Analysis toggle sync.
  • test_vib_cache.py, test_vib_py3dmol_render.py, test_viz_render_telemetry.py — vib animation + telemetry coverage.
  • _layout(...) helper sanitises widgets.Layout kwargs to eliminate a 4808 → 13 traitlets warning regression.
  • _safe_cb wrapper around every .observe() callback so exceptions surface in the Log tab instead of disappearing into the Voilà kernel console.
  • Kernel io_loop is cached at startup; thread-spawned callbacks are queued onto the main thread to avoid RuntimeError: no current event loop.
  • Native launchers: launch-native.bat (Windows / WSL) and launch-native.command (macOS / Linux) — double-clickable, port 8867, stamp-based editable-install skip, browser auto-open. README documents pinning each to the Start menu / Dock as a real app.
  • Native JupyterLab launcher (launch-native-jupyter.bat) and Apptainer launcher improvements.

Docs

  • .github/copilot-instructions.md — canonical AI-assistant context (now the single source of truth for any AI assistant working on this repo).
  • CLAUDE.md — Claude-specific session/workflow context (git-ignored).
  • Site favicons (ICO + SVG) for the GitHub Pages docs site.

Changed

  • Visualization is py3Dmol-first. plotlymol3d remains an optional fallback for non-trajectory tasks; trajectory rendering is hard-wired to py3Dmol to avoid Plotly/RequireJS flicker.
  • Plotly figures are rendered via plotly.io.to_html(..., include_plotlyjs="require") inside widgets.HTML, not display(fig), so threaded renders work and offline Voilà loads correctly.
  • pyscf is now an optional extra (pip install quantui[pyscf]); the package imports cleanly on Windows with PySCF unavailable.
  • Repo renamed from QuantUI-local to QuantUI.

Fixed

  • Trajectory accordion blank on first expand — switched traj_output from Output to VBox and use atomic children-swap so deferred widget-display is no longer a blank-frame risk.
  • Vib mode races on rapid switching — render-token guard (_vib_render_token) causes stale background renders to bail rather than overwriting newer output.
  • Camera state lost on mode switch — JS hook caches the active $3Dmol.GLViewer state across atomic HTML swaps; reset only on a genuinely new frequency result.
  • PySCF API drift — robust handling for v2 NMR / thermo API and the pyscf.prop.infrared rename; both Infrared.kernel() and the older IR API are supported.
  • Result-dir name collisions — timestamps now include microseconds; same formula + method + basis no longer overwrite each other.
  • IR x-axis — corrected wavenumber axis on the IR Plotly figure.
  • Plotly figures invisible after accordion show — figures are re-rendered on accordion expand to handle RequireJS / display-deferral edge cases.

Removed

  • visualization.py (PlotlyMol fallback) — replaced by the router-backed visualization_py3dmol.py path.
  • All SLURM-era infrastructure already removed during the downstream port: job_manager.py, storage.py, slurm_errors.py, SLURM config templates.

0.1.0 - 2026

Initial public scaffolding of the QuantUI package: quantui package with molecule.py, pubchem.py, config.py, visualization_py3dmol.py, calculator.py, basic notebook launcher, Apptainer container definition, MIT license, and project metadata.