We present deep wide-field K-band observations of six Hyperluminous Infrared Galaxies (HLIRGs) spanning a redshift range 0.44<z<1.55. The sample resides in a wide variety of environments, from the field to Abell 2 clusters, with a mean galaxy-HLIRG clustering amplitude of B_gh=190+/-45. The range in environments, and the mean clustering level, are both greater than those seen in local IR-luminous galaxies, from which we infer that the range of galaxy evolution processes driving IR-luminous galaxy evolution at z>0.5 is greater than locally, and includes mergers between gas-rich spiral galaxies in the field, but also includes encounters in clusters and hierarchical buildup. The similarity in the range of environments and mean clustering amplitude between our sample and QSOs over a similar redshift range is consistent with the interpretation where evolutionary connections between IR-luminous galaxies and QSOs are stronger at z>0.5 than locally, and that, at these redshifts, the processes that drive QSO evolution are similar to those that drive IR-luminous galaxy evolution. From comparison of the HLIRG and QSO host galaxies we further postulate that a larger fraction of IR-luminous galaxies pass through an optical QSO stage at z>0.5 than locally.

We present the K-band Hubble diagrams (K-z relations) of sub-mm-selected galaxies and hyperluminous galaxies (HLIRGs). We report the discovery of a remarkably tight K-z relation of HLIRGs, indistinguishable from that of the most luminous radiogalaxies. Like radiogalaxies, the HLIRG K-z relation at z<~3 is consistent with a passively evolving ~3L* instantaneous starburst starting from a redshift of z~10. In contrast, many sub-mm selected galaxies are >~2 magnitudes fainter, and the population has a much larger dispersion. We argue that dust obscuration and/or a larger mass range may be responsible for this scatter. The galaxies so far proved to be hyperluminous may have been biased towards higher AGN bolometric contributions than sub-mm-selected galaxies due to the 60 micron selection of some, so the location on the K-z relation may be related to the presence of the most massive AGN. Alternatively, a particular host galaxy mass range may be responsible for both extreme star formation and the most massive active nuclei.

Previous modelling has demonstrated that it is difficult to reproduce the SCUBA source counts within the framework of standard hierarchical structure formation models if the sources are assumed to be the high-redshift counterparts of local ultra-luminous infrared galaxies with dust temperatures in the range 40-60 K. Here, we show that the counts are more easily reproduced in a model in which the bulk of the sub-millimetre emission comes from extended, cool (20-25 K) dust in objects with star formation rates of 50-100 solar masses per year. The low temperatures imply typical sizes of about 1 (S_{850}/1mJy})^{1/2} arcsec, a factor two to three larger than those predicted using starburst-like spectral energy distributions. Low dust temperatures also imply a ratio of optical/UV to 850-micron flux which is 30-100 times smaller, for the same optical depth, than expected for objects with a hot, starburst-like SED. This may help explain the small overlap between SCUBA sources and Lyman-break galaxies.