We examine how the spatial correlation function of galaxies from the CNOC2 Field Galaxy Redshift Survey depends on galaxy color , luminosity and redshift .
The projected correlation function w _ { p } is determined for volume-limited samples of objects with 0.12 \leq z < 0.51 and evolution-compensated R _ { C } -band absolute magnitudes M _ { R } ^ { 0 } < -20 , over the comoving projected separation range 0.04 h ^ { -1 } { Mpc } < r _ { p } < 10 h ^ { -1 } { Mpc } .
Our sample consists of 2937 galaxies which are classified as being either early- or late-type objects according to their spectral energy distribution ( SED ) , determined from UBVR _ { C } I _ { C } photometry .
For simplicity , galaxy SEDs are classified independently of redshift : our classification scheme therefore does not take into account the colour evolution of galaxies .

Objects with SEDs corresponding to early-type galaxies are found to be more strongly clustered by a factor of \sim 3 , and to have a steeper correlation function , than those with late-type SEDs .
Modeling the spatial correlation function , as a function of comoving separation r , as \xi ( r ) = \left ( r / r _ { 0 } \right ) ^ { - \gamma } , we find r _ { 0 } = 5.45 \pm 0.28 h ^ { -1 } { Mpc } and \gamma = 1.91 \pm 0.06 for early-type objects , and r _ { 0 } = 3.95 \pm 0.12 h ^ { -1 } { Mpc } and \gamma = 1.59 \pm 0.08 for late-type objects ( for \Omega _ { M } = 0.2 , \Omega _ { \Lambda } = 0 ) .
While changing the cutoff between early- and late-type SEDs does affect the correlation amplitudes of the two samples , the ratio of the amplitudes remains constant to within 10 % .

The redshift dependence of the correlation function also depends on SED type .
Modeling the redshift dependence of the comoving correlation amplitude r _ { 0 } ^ { \gamma } as r _ { 0 } ^ { \gamma } ( z ) \propto ( 1 + z ) ^ { \gamma - 3 - \epsilon } , we find that early-type objects have \epsilon = -3.9 \pm 1.0 , and late-type objects have \epsilon = -7.7 \pm 1.3 .
Both classes of objects therefore have clustering amplitudes , measured in comoving coordinates , which appear to decrease rapidly with cosmic time .
The excess clustering of galaxies with early-type SEDs , relative to late-type objects , is present at all redshifts in our sample .
In contrast to the early- and late-type SED samples , the combined sample undergoes little apparent evolution , with \epsilon = -2.1 \pm 1.3 , consistent with earlier results .

The apparent increase with redshift of the clustering amplitude in the early- and late-type samples is almost certainly caused by evolution of the galaxies themselves , rather than by evolution of the correlation function .
If galaxy SEDs have evolved significantly since z \sim 0.5 , then our method of classifying SEDs may cause us to overestimate the true evolution of the clustering amplitude for the unevolved counterparts to our early- and late-type samples .
However , if color evolution is to explain the apparent clustering evolution , the color evolution experienced by a galaxy must be correlated with the galaxy correlation function .

We also investigate the luminosity dependence of the correlation function for volume-limited samples with 0.12 \leq z < 0.40 and M _ { R } ^ { 0 } < -19.25 .
We detect a weak luminosity dependence of the correlation amplitude for galaxies with early-type SEDs , d \log \xi / dM _ { R } ^ { 0 } = -0.35 \pm 0.17 , but no significant dependence for late-type objects , d \log \xi / dM _ { R } ^ { 0 } = 0.02 \pm 0.16 .