In order to achieve the primary scientific objectives of this proposal, we need to observe a total of 400 galaxies. With 0.1 square arcminutes per NIRI field and 50 galaxies per square arcminute, we need to observe 80 NIRI fields. We have picked out ~20 stars suitable for AO guiding in each of the four CFHTLS Deep fields for a total of 80 NIRI fields. We will observe two Deep fields in 2005A (the D2 at 10hrs and the D3 at 14hrs). We were awarded a fraction of the requested time in 2004B for the other two deep fields (D1 at 02hrs and D4 and 22hrs). We will receive data for at most 10 guide stars in 2004B and will apply for the remaining time in 2005B. Although we can withstand a reduction in the total sample of about a third, reducing the number of observations by a half would severely compromise the statistical robustness of the studies proposed here. As described in the Scientific Justification, we estimate that half of the galaxies we will observe lie at z>1.5, and on order of 20% of these high redshift galaxies will lie in close physical pairs with physical separations less than about 20/h kpc. These pairs are likely to merge within a short timescale (~ 0.5 Gyr), and therefore are an excellent probe of the instantaneous galaxy merger rate (Patton et al. 2000). We will identify close pairs using a fixed angular separation criterion, taking advantage of the fact that the angular diameter-distance relationship peaks at a redshift of about 1.5. Therefore, all angular pairs with separations less than 3.4 arcseconds must have projected physical separations of at most 20/h kpc, regardless of redshift (assuming that both galaxies lie at the same redshift). Contamination due to projected pairs (different redshifts) will be removed statistically, using a standard two-point correlation function approach. Photometric redshifts will be derived from the ugrizK-band data with typical errors of 10% in (1+z). While these redshifts cannot be used to identify mergers directly, they will reduce the contamination by projected (non-physical) pairs by a factor of ~10. Current tests indicate that the ground based data are useful as close as 4" of the selected guide star. A PSF subtraction algorithm is being developed to enable accurate photometry close to the bright guide stars. It has been tested by adding the images of galaxies from other parts of the CFHTLS fields into the wings of the guide stars. Currently, the magnitudes of I=25 galaxies are recovered to within 0.25 magnitudes (better for brighter galaxies). Further refinements are planned. We applied for time in 2004B. So far, only one of our targets has been observed; we received the data the week before the 2005A proposal deadline. The data has been reduced in a "quick and dirty" manner; there are still un-resolved issues with the flat fielding. Even so, the image shows that we are able to reach K_AB=25 with 20% photometry in an hour, confirming the results of the ITC. Figure 1 shows the K-band image and the corresponding (PSF-subtracted) I-band image. The FITS image is available from http://orca.phys.uvic.ca/~gwyn/MMM/NIRI/index.html Quantitative morphology with gim2d requires knowledge of the PSF across the field. Therefore we are requesting PSF calibration images to be taken on a nearby globular field after each science exposure. These PSF calibration images can also be sandwiched between two science observations, reducing the total time somewhat. Note that even if the PSF behaves oddly (as it probably will with AO) as long it is known, gim2d can operate successfully. Even if the PSF is only vaguely known, gim2d can still do a very good job. For example, there was no PSF calibration data taken with the data we have received so far. Merely by assuming a Gaussian 0.08 arcsecond PSF, gim2d was able to determine fairly accurate models for the galaxy structure as shown in Figure 2. This demonstrates the feasibility of image decomposition with the AO images; it will only get better if we have PSF calibration data. Note that we are not planning to do quantitative morphology to the full depth of the survey. However we still need AO just to be able to acheive the full depth of K_AB=25 in the first place, by concentrating the light over fewer pixels (less read noise) and a smaller patch of sky (less background). We are requesting one hour total integration times. Because the counts are quite shallow, about log_10(N) = 0.15 m_AB(K) + const (Thompson et al, 1999) there is a clear natural depth of exposure. Exposure times of much less than one hour lower efficiency as the setup time (30 minutes per field) costs rise. Increasing the exposure time to 8 hours would increase the depth; the corresponding rise in the number of objects per unit area would be only about 50%. Since it would be possible to observe 5-6 fields in the same 8 hours, the total number of observable objects would decrease by a similar factor. Because this project is closely related to two other public projects (CFHTLS and COSMOS), and because it will take a substantial amount of telescope time, we will release stacked images and catalogs to the CFHTLS community within six months of acquiring the last image.