Skip to content
linaldblinaldblinaldb

Gravitational-Wave Detection

Real strain data from the Gravitational Wave Open Science Center for two physically distinct confirmed detections: GW150914 (the first-ever gravitational-wave detection, a binary black hole merger) and GW170817 (a binary neutron star merger). Real, unmodified 4096 Hz / 32-second strain time series from both LIGO detectors (H1, L1), plus the real GWTC-1-confident event catalog, ingested with zero h5py, csv, scipy, or numpy: linaldb’s own connectors read the raw files directly.

Native ingestion, no Python parsing library

Section titled “Native ingestion, no Python parsing library”
IMPORT DATASET FROM "gwtc1_events.csv" AS gw_events
USE DATASET FROM "GW150914_H1.hdf5" AS h1_150914 FIELDS (strain_Strain)

The HDF5 connector registers a real Vector(131072) tensor per detector per event (32 seconds × 4096 Hz) directly from the raw file.

Real event catalog, classified and ranked in SQL

Section titled “Real event catalog, classified and ranked in SQL”
SELECT event_name,
       CASE WHEN mass_2_source < 3.0 THEN "compact_binary_with_NS" ELSE "BBH" END AS source_class,
       network_matched_filter_snr,
       RANK() OVER (ORDER BY network_matched_filter_snr DESC) AS loudness_rank
FROM gw_events ORDER BY loudness_rank

GW170817 (the neutron-star merger, much closer, at 40 Mpc vs. hundreds for the black hole mergers) tops the real loudness leaderboard at SNR 33.0, ahead of GW150914 at 25.2.

The full frequency-domain DSL, on real physics data

Section titled “The full frequency-domain DSL, on real physics data”
LET h1_noise = PSD h1_150914_strain_Strain WINDOW 131072
LET h1_white = WHITEN h1_150914_strain_Strain WITH h1_noise
LET h1_filt  = BANDPASS h1_white FROM 35.0 TO 350.0 WITH RATE 4096.0

MEAN/STDEV/CORRELATE (the two engine bugs found and fixed via the leukemia notebook) get a real second workout here at 131,072-element scale instead of a 4-element toy vector: STDEV on real H1 strain comes back 2.18e-19, matching the physically expected noise floor (~1e-19 to ~2e-19) exactly.

Reported honestly, both ways

Section titled “Reported honestly, both ways”

A from-scratch, template-free attempt at localizing the real merger (naive per-second energy ranking, and matching one detector’s real whitened merger segment against the other’s) is reported exactly as it came out, not adjusted after the fact: neither approach cleanly landed on the real merger time or the real ~10ms H1-L1 light-travel delay. Real gravitational-wave detection pipelines match against a bank of physically modeled inspiral waveforms, not one noisy real segment against another; that modeling is its own large physics computation, out of scope here, exactly as linal-db-rs/examples/gw_transient_analysis.lnl already disclaims for its own synthetic-template version. Every linaldb primitive involved produced exactly its documented shape and behavior at every step: this is an honest report on the difficulty of the astrophysics, not a shortfall in the engine.

Try it yourself

Section titled “Try it yourself”