Cross-correlation of Planck CMB lensing and CFHTLenS galaxy weak lensing maps

We cross-correlate cosmic microwave background (CMB) lensing and galaxy weak lensing maps using the Planck 2013 and 2015 data and the 154 deg(2) Canada-France-Hawaii Telescope Lensing Survey (CFHTLenS). This measurement probes large-scale structure at intermediate redshifts approximate to 0.9, between the high-and low-redshift peaks of the CMB and CFHTLenS lensing kernels, respectively. Using the noise properties of these data sets and standard Planck 2015 Lambda CDM cosmological parameters, we forecast a signal-to-noise ratio approximate to 4.6 for the cross-correlation. We find that the noise level of our actual measurement agrees well with this estimate, but the amplitude of the signal lies well below the theoretical prediction. The best-fit amplitudes of our measured cross-correlations are A(2013) = 0.48 +/- 0.26 and A(2015) = 0.44 +/- 0.22, using the 2013 and 2015 Planck CMB lensing maps, respectively, where A = 1 corresponds to the fiducial Planck 2015 Lambda CDM prediction. Due to the low measured amplitude, the detection significance is moderate (approximate to 2 sigma) and the data are in tension with the theoretical prediction (approximate to 2-2.5 sigma). The tension is reduced somewhat when compared to predictions using WMAP9 parameters, for which we find A(2013) = 0.56 +/- 0.30 and A(2015) = 0.52 +/- 0.26. We consider various systematic effects, finding that photometric redshift uncertainties, contamination by intrinsic alignments, and effects due to the masking of galaxy clusters in the Planck 2015 CMB lensing reconstruction are able to help resolve the tension at a significant level (approximate to 10% each). An overall multiplicative bias in the CFHTLenS shear data could also play a role, which can be tested with existing data. We close with forecasts for measurements of the CMB lensing-galaxy lensing cross-correlation using ongoing and future weak lensing surveys, which will definitively test the significance of the tension in our results with respect to Lambda CDM.