PQEnalyzer

Plot and monitor PQ energy, box, and optimizer output in a browser or desktop interface.

Install

pip install PQEnalyzer

Quick Start

The desktop GUI remains the default:

pqenalyzer /path/to/simulation.en

Use the browser dashboard explicitly:

pqenalyzer web /path/to/simulation.en

PQEnalyzer detects the input format automatically. Use a format flag only when detection is ambiguous:

pqenalyzer --pq pq-output.en
pqenalyzer --qmcfc qmcfc-output.en
pqenalyzer --box box-output.data
pqenalyzer --opt optimization.data

pqenalyzer gui FILE is equivalent to pqenalyzer FILE.

Input

Energy, box, and optimizer parsing is provided by PQAnalysis.

Energy files need a matching .info file in the same directory. PQ .info rows containing a single parameter column are supported.

Box files contain step x y z alpha beta gamma. PQEnalyzer plots BOX-X, BOX-Y, BOX-Z, ALPHA, BETA, GAMMA, and BOX-VOLUME.

Optimizer plots use the optimization step as the x-axis. They include energy changes, forces, convergence states, and limits. A convergence state of -1 means not converged, 0 means disabled, and 1 means converged. The first row is PQ's initialization snapshot. Final completion status remains in the PQ log.

Web

pqenalyzer web FILE [FILE ...] [--port 8766] [--no-open]

This starts a local-only server (loopback, default 127.0.0.1:8766) and opens the dashboard in your browser. Nothing leaves your machine; use --port when the default is taken and --no-open to print the address without opening a browser.

Cluster access over SSH

Keep the server on the cluster loopback interface and forward it through SSH. On a directly reachable login node, start PQEnalyzer there:

pqenalyzer web --no-open --port 8766 /path/to/simulation.en

Then open the tunnel from your desktop:

ssh -N -L 8766:127.0.0.1:8766 user@login.cluster

Open http://127.0.0.1:8766 locally. If the analysis runs on an allocated compute node behind a login node, start PQEnalyzer on that compute node and use the login node as a jump host:

ssh -N -J user@login.cluster \
  -L 8766:127.0.0.1:8766 user@compute-node

Keep the PQEnalyzer process inside the allocation for as long as the tunnel is needed. The server intentionally rejects non-loopback hosts; SSH provides the authenticated, encrypted path without exposing an unauthenticated HTTP server to the cluster network or VPN.

What you see:

• Dashboard — one card per parameter with a sparkline and the latest value. Click a card (or press Ctrl+K and type a name) to inspect it.
• Series / Histogram — press 1 / 2 to switch the chart language. Drag to zoom, double-click to reset, hover or tap for values. Focus a chart and use arrow keys to inspect values; Home / End jump to the edges. Files are read in input order as one dataset, with one series and one distribution.
• Analysis — open Analysis above the chart to choose a reference, a trend, or autocorrelation. The mean is shown initially. Cumulative and running averages continue across file boundaries; the running window defaults to about 5% of all samples. Autocorrelation correlates this same sequence with itself and switches the chart to lag in steps. Analysis choices and dashboard sort persist per browser. Correlation pairs can cross file boundaries; use continuous segments of the same run for this analysis.
• Axis — physical time is used when it increases throughout the combined sequence. If it restarts or overlaps between files, charts use a one-based sample index so every observation remains visible.
• Live updates — the header badge reads watching while files are watched, stale when they changed on disk (refresh or resume watching to reload), paused when watching is off, and offline if the connection drops. Data stays on screen throughout.

Hover or focus the series or histogram to inspect values and bin counts. The source files stay in their original format; the chart can be saved as a PNG. An MSER marker shows an estimated initial cut on eligible time series. The mean, uncertainty, and other summary statistics still use all samples. The ± value is the standard error of the mean (SEM), adjusted for autocorrelation. Estimates that overflow floating-point arithmetic are shown as unavailable. The cut is not a verdict that the simulation has converged.

Parameters split into observables (energy, temperature, pressure, density, …) and diagnostics (LOOPTIME, atom counts, and any other zero-variance series). Diagnostics stay fully viewable, but they get no drift measurements or MSER cuts, and sort after observables: loop time tracks compute cost per step, not the simulated system, so a step change there vetoes the segment — it never proves equilibration.

Desktop GUI

Run pqenalyzer gui FILE [FILE ...] to use the desktop interface.

Double-click a Live Monitor panel to open its focused plot. Plot settings belong to that focused window, so each window can use different overlays.

Auto-refresh starts with the GUI. If native file watching is unavailable, PQEnalyzer uses polling and shows (polling) in the status line.

Plot windows refit when resized. The Live Monitor also redistributes its grid; press f to fit it to the current screen. Use Plot Size in the sidebar, + and - in a plot window, or Ctrl+0 / Command+0 to restore 100%.

PQEnalyzer remembers the plot size, window dimensions, selected parameter, auto-refresh state, and plot settings. Set PQENALYZER_CONFIG_DIR to override the platform settings directory. Desktop and web use the same flat mono design tokens.

Plot Features

Web uses one ordered dataset for every chart and analysis:

Web autocorrelation is mean-centered and normalized to 1 at lag zero, calculated on all selected files as one sequence.

Multiple Files

pqenalyzer md-01.en md-02.en md-03.en

In the web view, common parameters from the files form one ordered dataset. A parameter found in only some files uses those files in input order. Shared parameters must use the same unit.

In the desktop GUI, Difference plotting requires exactly two files and calculates file 1 - file 2. Points are matched by simulation time, simulation step, or optimization step. PQEnalyzer does not interpolate, extrapolate, or concatenate difference data. Raw series are hidden when Difference is enabled.

Development

pip install -e ".[test]"
python -m pytest -m "not benchmark and not e2e"

For the web frontend, use Node.js 24 and build the bundled files served by pqenalyzer web:

cd PQEnalyzer/web/frontend
npm ci
npm test
npm run build

Commit the generated files in PQEnalyzer/web/static when the frontend changes. The frontend installs the shared flat mono controls from a versioned PQDesign release; no adjacent PQSetup checkout is needed. To update the shared design, follow the design package guide, then update the archive URL and lockfile together. Copy that package's tokens.json to PQEnalyzer/design/tokens.json; CI checks they match. Desktop colors derive from this packaged snapshot. Keep chart and dashboard layout in PQEnalyzer.

Run the end-to-end suite separately:

python -m playwright install chromium
python -m pytest -m e2e

The browser test checks parameter selection, keyboard chart inspection, autocorrelation, histogram bins, PNG download, and stale-file refresh. CI rebuilds the frontend and checks that PQEnalyzer/web/static matches the source. A release tag must point to the current main commit; the tag job repeats the tests, builds the distribution, and installs the wheel before publishing it.