Gaia Early Data Release 3: Photometric content and validation

[Context] Gaia Early Data Release 3 (Gaia EDR3) contains astrometry and photometry results for about 1.8 billion sources based on observations collected by the European Space Agency Gaia satellite during the first 34 months of its operational phase. [Aims] In this paper, we focus on the photometric content, describing the input data, the algorithms, the processing, and the validation of the results. Particular attention is given to the quality of the data and to a number of features that users may need to take into account to make the best use of the Gaia EDR3 catalogue. [Methods] The processing broadly followed the same procedure as for Gaia DR2, but with significant improvements in several aspects of the blue and red photometer (BP and RP) preprocessing and in the photometric calibration process. In particular, the treatment of the BP and RP background has been updated to include a better estimation of the local background, and the detection of crowding effects has been used to exclude affected data from the calibrations. The photometric calibration models have also been updated to account for flux loss over the whole magnitude range. Significant improvements in the modelling and calibration of the Gaia point and line spread functions have also helped to reduce a number of instrumental effects that were still present in DR2. Results. Gaia EDR3 contains 1.806 billion sources with G-band photometry and 1.540 billion sources with GBP and GRP photometry. The median uncertainty in the G-band photometry, as measured from the standard deviation of the internally calibrated mean photometry for a given source, is 0.2 mmag at magnitude G = 10-14, 0.8 mmag at G ≈ 17, and 2.6 mmag at G ≈ 19. The significant magnitude term found in the Gaia DR2 photometry is no longer visible, and overall there are no trends larger than 1 mmag mag-1. Using one passband over the whole colour and magnitude range leaves no systematics above the 1% level in magnitude in any of the bands, and a lar