Several semi-analytic models (SAMs) try to explain how galaxies form, evolve and interact inside the dark matter large-scale structure. These SAMs can be tested by comparing their predictions for galaxy-galaxy-galaxy-lensing (G3L), which is weak gravitational lensing around galaxy pairs, with observations. We evaluate the SAMs by Henriques et al. (2015; H15) and by Lagos et al. (2012; L12), implemented in the Millennium Run, by comparing their predictions for G3L to observations at smaller scales than previous studies and also for pairs of lens galaxies from different populations. We compare the G3L signal predicted by the SAMs to measurements in the overlap of the Galaxy And Mass Assembly survey (GAMA), the Kilo-Degree Survey (KiDS), and the VISTA Kilo-degree Infrared Galaxy survey (VIKING), splitting lens galaxies into two colour and five stellar-mass samples. Using an improved G3L estimator, we measure the three-point correlation of the matter distribution for mixed lens pairs with galaxies from different samples, and unmixed lens pairs with galaxies from the same sample. Predictions by the H15 SAM agree with the observations for all colour-selected and all but one stellar-mass-selected sample with 95% confidence. Deviations occur for lenses with stellar masses below $9.5h^{-2}\mathrm{M}_\odot$ at scales below $0.2h^{-1}\mathrm{Mpc}$. Predictions by the L12 SAM for stellar-mass selected samples and red galaxies are significantly higher than observed, while the predicted signal for blue galaxy pairs is too low. The L12 SAM predicts more pairs of small stellar-mass and red galaxies than the H15 SAM and the observations, as well as fewer pairs of blue galaxies. Likely explanations are different treatments of environmental effects by the SAMs and different models of the initial mass function. We conclude that G3L provides a stringent test for models of galaxy formation and evolution.

We estimate the halo mass function (HMF) by applying the excursion set approach to the non-linear cosmic density field. Thereby, we account for the non-Gaussianity of today's density distribution and constrain the HMF independent of the linear collapse threshold $\delta_{\textrm{crit}}$. We consider a spherical region as a halo, if its density today exceeds the virial overdensity threshold $\Delta$. We model the probability distribution of the non-linear density field by a superposition of a Gaussian and a lognormal distribution, which we constrain with the bispectrum of density fluctuations, predicted by the kinetic field theory description of cosmic structure formation. Two different excursion set approaches are compared. The first treats the density $\delta$ as an uncorrelated random walk of the smoothing scale $R$. The second assumes $\delta(R)$ to be correlated. We find that the resulting HMFs correspond well to the HMF found in numerical simulations if the correlation of $\delta(R)$ is taken into account. Furthermore, the HMF depends only weakly on the choice of the density threshold $\Delta$.

Galaxy-galaxy-galaxy lensing (G3L) is a powerful tool for constraining the three-point correlation between the galaxy and matter distribution and thereby models of galaxy evolution. We propose three improvements to current measurements of G3L: (i) a weighting of lens galaxies according to their redshift difference, (ii) adaptive binning of the three-point correlation function, and (iii) accounting for the effect of lens magnification by the cosmic large-scale structure. Improvement (i) is designed to improve the precision of the G3L measurement, whereas improvements (ii) and (iii) remove biases of the estimator. We further show how the G3L signal can be converted from angular into physical scales. The improvements were tested on simple mock data and simulated data based on the Millennium Run with an implemented semi-analytic galaxy model. Our improvements increase the signal-to-noise ratio by 35 % on average at angular scales between 0.1 arcmin and 10 arcmin and physical scales between $0.02$ and $2 \, h^{-1}\, \textrm{Mpc}$. They also remove the bias of the G3L estimator at angular scales below 1 arcmin, which was originally up to 40 %. The signal due to lens magnification is approximately 10 % of the total signal.