How a Neural Network Hears the Dance of Black Holes ✛ ✦
Far out in space, pulsars spin — neutron stars, ultra-dense remnants discovered by Jocelyn Bell Burnell in 1967. With each rotation, they emit a radio beam, like lighthouses, but more precise than any Earthly clock. The slightest delay in their signal may mean a gravitational wave has passed through space — a ripple from merging black holes, predicted by Einstein. The work of Russell Hulse confirmed this, showing the waves' effect on a binary pulsar. For years, scientists have listened to these rhythms with radio telescopes, but picking out the signal of a single black hole pair from the noise was like hearing a whisper at a rock concert.
Now a neural network trained on the laws of physics handles it in seconds. It gets a hint: the instant positions of the black holes in orbit — like an experienced dancer feeling a partner's rhythm and anticipating their steps. This allows more accurate estimates of the giants' masses and orbital periods. Tests on synthetic data showed accuracy improves and computations speed up from days to seconds. Soon, detectors like LIGO and new radio arrays will continuously listen to the cosmos.
The most precise pulsars change rhythm by 10⁻¹⁵ seconds per second — a gravitational wave shifts the pulses by just a few tens of nanoseconds. But even such tremors the neural network catches, turning the cosmic 'hum' into a clear melody.
The best millisecond pulsars are more stable than atomic clocks: their period changes by less than 10⁻¹⁵ seconds per second. A gravitational wave from a distant black hole shifts pulse arrival times by just a few tens of nanoseconds — a billion times faster than a blink.
The idea of decoding cosmic 'hum' echoes Stanisław Lem's novel 'His Master's Voice', where scientists try to understand the nature of a mysterious signal from the depths of the universe.