-
Speed
Each optimisation was measured as it was made, with simple timing and GPU synchronisation. On a single A100 GPU, CETRA searches a two-year PLATO light curve at 600-second cadence in about 3 minutes, compared with about 2 CPU core-hours for TLS, a roughly 40-fold speed-up. At 25-second cadence TLS needs around 100 CPU core-hours, but CETRA takes a similar time to before, because its expensive periodic search depends on the length of the light curve, not its cadence. A workstation RTX A5000 is only about a fifth slower than the A100, bringing survey-scale searches within reach of groups without dedicated HPC access.
Run time against light curve length for CETRA, BLS and TLS at four cadences. -
Sensitivity
In tests on 20,000 simulated light curves, CETRA and TLS performed similarly at low noise, but at the highest noise level CETRA recovered 21 per cent more signals, and more than twice as many Earth–Sun analogues. These gains follow from the design: formal handling of gaps, a uniform step in transit time and proper weighting by measurement uncertainty.

Signals recovered by CETRA (cyan) and TLS (magenta) at three noise levels.
-
A real-data test
Applied to TESS observations of HD 101581 (TOI-6276), a star known to host two Earth-sized planets and a third candidate, CETRA recovered all three. The candidate, TOI-6276.03, previously reported at a signal-to-noise ratio of 7.9, was detected at 16.0, which may now allow it to be validated as a planet. The full analysis took 2.6 seconds on an RTX A5000, around 90 times faster than TLS on six CPU cores.
The three signals of HD 101581 found by CETRA, phase-folded with the best-fitting transit model. -
Getting started
CETRA is open source and can be installed from PyPI. It needs an NVIDIA GPU and runs on detrended light curves. The GitHub repository includes example Jupyter notebooks, among them the full HD 101581 analysis, and the documentation covers installation and use.
More on performance and results
Watch the final video to see how CETRA performs, and how to start using it.