GPU-accelerated broadband analysis of multi-messenger GRB light curves.

High-frequency multi-messenger observations provide a powerful probe of gamma-ray bursts (GRBs), pioneered by {\em BeppoSAX} in gamma-rays and LIGO-Virgo in gravitational waves. Identifying the central engines - magnetars or black holes - also promises to improve on GRBs as probes of cosmological evolution. THESEUS' design is ideally suited to pursue both science objectives. Here, we present a general-purpose {\em graphics processor units} (GPU)-accelerated broadband search algorithm for long duration ascending and descending chirps, post-merger or from core-collapse of massive stars, in electromagnetic and gravitational radiation. It implements butterfly filtering using banks of up to 8 million templates of 1 second duration at over one million correlations per second by heterogeneous computing using a dozen high-end GPUs. We demonstrate its application with the identification of broadband Kolmogorov spectra in long GRBs and the long duration ascending chirp in the merger GW170817. The first shows a noticeable absence of a high frequency bump, otherwise expected from newly formed magnetar central engines. The second illustrates the need for deep searches to identify GRB central engines in descending chirps in gravitational waves, post-merger or from nearby energetic core-collapse supernovae. A future catalogue of THESEUS' GRBs covering a broad range of redshifts may probe the nature of the cosmological vacuum and establish the de Sitter limit as a turning point in cosmological evolution.