We present the QSO luminosity function of the completed 2dF-SDSS LRG and QSO ( 2SLAQ ) survey , based on QSOs photometrically selected from Sloan Digital Sky Survey imaging data and then observed spectroscopically using the 2dF instrument on the Anglo-Australian Telescope .
We analyse 10637 QSOs in the redshift range 0.4 < z < 2.6 to a g -band flux limit of 21.85 ( extinction corrected ) and an absolute continuum magnitude of M _ { g } ( z = 2 ) < -21.5 .
This sample covers an area of 191.9 deg ^ { 2 } .

The binned QSO luminosity function agrees with that of the brighter SDSS main QSO sample , but extends \sim 2.5 mags fainter , clearly showing the flattening of the luminosity function towards faint absolute magnitudes .
2SLAQ finds an excess of QSOs compared to the 2dF QSO Redshift Survey at g > 20.0 , as found previously by Richards et al .
( 2005 ) .
The luminosity function is consistent with other previous , much smaller , samples produced to the depth of 2SLAQ .

By combining the 2SLAQ and SDSS QSO samples we produce a QSO luminosity function with an unprecedented combination of precision and dynamic range .
With this we are able to accurately constrain both the bright and faint ends of the QSO LF .
While the overall trends seen in the evolution of the QSO LF appear similar to pure luminosity evolution , the data show very significant departures from such a model .
Most notably we see clear evidence that the number density of faint QSOs peaks at lower redshift than bright QSOs : QSOs with M _ { g } > -23 have space densities which peak at z < 1 , while QSOs at M _ { g } < -26 peak at z > 2 .
By fitting simple luminosity function models in narrow M _ { g } intervals we find that this downsizing is significant at the 99.98 per cent level .

We show that luminosity function models which follow the pure luminosity evolution form [ i.e . M _ { g } ^ { * } \equiv M _ { g } ^ { * } ( z ) ] , but with a redshift–dependent bright end slope and an additional density evolution term , \Phi ^ { * } \equiv \Phi ^ { * } ( z ) , provide a much improved fit to the data .
The bright end slope , \alpha , steepens from \alpha \simeq - 3.0 at z \simeq 0.5 to \alpha = -3.5 at z \simeq 2.5 .
This steepening is significant at the 99.9 per cent level .
We find a decline in \Phi ^ { * } from z \simeq 0.5 to z \simeq 2.5 which is significant at the 94 per cent level .