No optical telescope can beat diffraction: an aperture of diameter
D observing wavelength
λ can only resolve two points separated by the Rayleigh angle
θ = 1.22 · λ / D (radians)
That is why Hubble's 2.4 m mirror, JWST's 6.5 m segmented mirror, Chandra's X-ray optics and the Event Horizon Telescope's Earth-sized virtual dish (radio VLBI) all resolve wildly different detail — the two test dots on the canvas are a fixed physical separation, drawn with a blur radius scaled to each instrument's actual θ, so you can watch them cleanly split apart or smear into one blob exactly as the formula predicts.
A single exposure is noisy: read noise and photon shot noise compete with the real signal. Averaging
N aligned exposures (stacking) adds the signal linearly but averages the noise down, since independent noise partly cancels:
SNR(N) = √N · SNR₁
- Instrument swaps the observing wavelength/aperture — infrared pierces the dust that blocks visible light, revealing embedded protostars; X-ray shows the surrounding hot gas; radio VLBI shows the compact core shadow.
- Stacked exposures raises N in the formula above — watch the sensor-noise grain fade and the SNR readout climb.
- Denoise mimics software noise reduction (Wiener/AI denoising) layered on top of stacking.