Loading emulator…

The neural network (~12 MB) is downloaded once and then runs entirely in your browser.

Interactive Explorer for Lu. et al 2026 TARDIS Spectral Emulator

This page runs a neural-network emulator that maps the ejecta parameters (design for Type Ic supernova SN2014L), including time since explosion, bolometric luminosity,outer velocity, density structure, and the integrated masses of 13 elements — onto a full optical + near-infrared synthetic spectrum (3000–24000 Å). A important note here is that these integrated elemental masses are NOT the total ejecta mass, rather it is a choice of representation for the ejecta composition. See the Lu. et al 2026 publication for more details.

The emulator was trained on roughly 700,000 radiative-transfer simulations computed with TARDIS. Here is a visualizer to explore the emulated spectra. You can pin down a baseline model, then watch how the spectrum responds when one parameter at a time is swept across the full range covered by the training grid.

Everything below — the network weights and every spectrum you see — is evaluated locally in your browser with ONNX Runtime Web. No data is sent to a server.

1. Baseline model

Each panel shows the distribution of one parameter in the training grid. Drag the vertical line (or type a value) to move the baseline; the comparison plots below update immediately. The starting point is the median of the training grid.

2. Varying one parameter

Each comparison panel sweeps a single parameter across its full prior range while every other parameter stays at the baseline.

up to 3 panels

Plain text (.txt, .dat, .csv), two columns: rest wavelength in Ångströms (Å), then linear (not log) scale flux in arbitrary units. Separate with spaces or commas. Leading text and column headers are skipped automatically.

Browser only. Never uploaded or saved; cleared on reload.

Mean-matched to each panel’s baseline over the overlap, then vertically offset for comparison. Nonpositive flux is hidden on log axes.