Bright
Bright / ExperiencesJust a minute? Start here ↗

GW150914 · 14 September 2015

A ripple reaches Earth.

Two instruments. Half a second of measured strain. One event that changed what people could observe.

HANFORD · H1LIVINGSTON · L1SAME HALF-SECOND
H1 + L135–350 Hz · whitened display
GWOSC · GW150914-v3 · calibrated 4 kHz R1 strain · CC BY 4.0
01 / 04Two instruments held the same instant.

Hanford and Livingston each recorded calibrated strain around the event.

THE MEASUREMENT

A fraction, measured twice.

The source quantity is calibrated, dimensionless gravitational-wave strain h(t). In Filtered strain mode, values use strain ÷ 10⁻²¹ as a shared display unit. They are not displacement, audio, or a reconstructed waveform.

Bright applies a fourth-order 35–350 Hz Butterworth band-pass with zero-phase filtering to the full 32-second files before cropping this half-second view. The default whitened view first balances the noise spectrum using GWOSC’s archived tutorial recipe; its values are normalized display units, no longer strain. Switch to Filtered strain to see the difference. The ridges map jointly normalized log PSD from 16–400 Hz. Height and time are exaggerated for legibility.

GWOSC whitening tutorial ↗. Bright uses a 35–350 Hz final band instead of the tutorial’s 20–300 Hz. A 32-second noise estimate is a display aid, not a detection, reconstruction, SNR or significance analysis.

4,096 Hzsource sample rate

2,048samples in this focus

32 ssource file duration

What the files carry

Instrumental lines and glitches can appear in detector strain. All seven released data-quality bits are set in this interval. H1’s five injection flags are clear; L1’s continuous-wave hardware-injection flag is active. These display derivatives do not independently validate an astrophysical interpretation.

GWOSC technical details ↗

RELATED RESEARCH · A DIFFERENT DATE

A quieter instrument.
A later chapter.

A 2025 paper reports reinforcement-learning mirror control tested in 2024. It did not enable the 2015 observation shown above.

Read the later control research →
About this Bright experience

The encounter

The event GW150914 was observed on 14 September 2015. This authored encounter uses the official GWOSC version 3, R1 Hanford and Livingston strain recordings. Height and color map derived spectral power or processed detector values; the default whitening display has normalized units, while the separate filtered-strain view keeps physical strain units, not the physical shape of spacetime.

Follow the same time through a waveform and a frequency view. Whitening and filtering make a band of the recording easier to see; it is not a detection pipeline or an estimate of significance.

A later chapter in AI

Deep Loop Shaping used reinforcement learning to improve mirror-control noise in tests at LIGO Livingston. That later work is a separate instrument-control story, published in 2025; it did not detect or produce the 2015 event shown here.

The archived tutorial whitening recipe uses a 32-second noise estimate and normalized display units. It is not an independent detector characterization.

The 35–350 Hz zero-phase Butterworth filter and spectral color scale are authored display choices, not a scientific detection analysis.

Both recordings include instrumental noise. The Livingston source mask reports an active continuous-wave hardware injection; the recording is not described as generally injection-free.

Frequency-view color is normalized log power, not confidence, SNR, distance or detection significance.