Quantitative assessment of high‑SNR posterior recovery

Determine the quantitative accuracy of posterior recovery in the high signal‑to‑noise ratio (approximately 1000) regime for Dingo and Dingo‑IS when analyzing quasi‑circular, spin‑aligned massive black‑hole binaries with the IMRPhenomXHM waveform and the low‑frequency LISA response, by benchmarking against matched nested‑sampling (Nessai) analyses under the same settings.

Background

The paper evaluates Dingo and Dingo‑IS against Nessai for low and moderate SNR injections using Jensen–Shannon divergences, finding close agreement, especially after importance sampling. However, at very high SNR (~1000), the authors observe qualitative differences, broader support in the raw Dingo posterior, and extremely low importance‑sampling efficiency.

Given these qualitative discrepancies at high SNR, the authors explicitly defer a detailed quantitative assessment of this regime to future work, indicating that a systematic, numerical characterization of performance remains unresolved.

References

As discussed above, posterior recovery in the high-SNR case already shows qualitative differences, and we therefore leave a quantitative assessment of this regime to future work.

— Accurate and efficient simulation-based inference for massive black-hole binaries with LISA  (2603.20431 - Spadaro et al., 20 Mar 2026) in Section Results, Subsection Benchmarking (Sec. 3.2)