The first strong indication of galaxy dark matter halo substructure was the flux ratio anomalies seen in quadruply imaged radio quasars ("radio quads") Metcalf and Madau (2001), Dalal and Kochanek (2002), Metcalf and Zhao (2002). Smooth mass models of lensing galaxies can easily explain the observed positions of the images, but the predictions of such models of the corresponding fluxes are frequently observed to be strongly violated. Optical and X-ray quasars have such small angular sizes that the observed optical and X-ray flux anomalies can be caused by stars ("microlensing"), which has allowed a measurement of the stellar mass along the lines of sight in lensing galaxies Pooley et al. (2012). But because the quasar radio-emitting region is larger, the observed radio flux anomalies can only be caused by relatively massive objects, with masses of order 106 to 108 M⊙ along the line of sight. After some controversy regarding whether ΛCDM simulations predict enough dark matter substructure to account for the observations, the latest papers concur that the observations are consistent with standard theory, taking into account uncertainty in lens system ellipticity Metcalf and Amara (2012) and intervening objects along the line of sight Xu et al. (2012), Xu et al. (2015). But this analysis is based on a relatively small number of observed systems (Table 2 of Chen et al. (2011) lists the 10 quads that have been observed in the radio or mid-IR), and further observational and theoretical work would be very helpful.
Another gravitational lensing indication of dark matter halo substructure consistent with ΛCDM simulations comes from detailed analysis of galaxy-galaxy lensing Vegetti et al. (2010), Vegetti et al. (2012), Vegetti et al. (2014), although much more such data will need to be analyzed to get strong constraints. Other gravitational lensing observations including time delays can probe the structure of dark matter halos in new ways Keeton and Moustakas (2009). Hezaveh et al. (2013), Hezaveh et al. (2014) show that dark matter substructure can be detected using spatially resolved spectroscopy of gravitationally lensed dusty galaxies observed with ALMA. Nierenberg et al. (2014) demonstrates that subhalos can be detected using strongly lensed narrow-line quasar emission, as originally proposed by Moustakas and Metcalf (2003).
The great thing about gravitational lensing is that it directly measures mass along the line of sight. This can provide important information that is difficult to obtain in other ways. For example, the absence of anomalous skewness in the distribution of high redshift Type 1a supernovae brightnesses compared with low redshift ones implies that massive compact halo objects (MACHOs) in the enormous mass range 10−2 to 1010 M⊙ cannot be the main constituent of dark matter in the universe Metcalf and Silk (2007). The low observed rate of gravitational microlensing of stars in the Large and Small Magellanic clouds by foreground compact objects implies that MACHOs in the mass range between 0.6 × 10−7 and 15 M⊙ cannot be a significant fraction of the dark matter in the halo of the Milky Way Tisserand et al. (2007). Gravitational microlensing could even detect free-floating planets down to 10−8 M⊙, just one percent of the mass of the earth Strigari et al. (2012).
A completely independent way of determining the amount of dark matter halo substructure is to look carefully at the structure of dynamically cold stellar streams. Such streams come from the tidal disruption of small satellite galaxies or globular clusters. In numerical simulations, the streams suffer many tens of impacts from encounters with dark matter substructures of mass 105 to 107 M⊙ during their lifetimes, which create fluctuations in the stream surface density on scales of a few degrees or less. The observed streams contain just such fluctuations Yoon et al. (2011), Carlberg (2012), Carlberg et al. (2012), Carlberg and Grillmair (2013), so they provide strong evidence that the predicted population of subhalos is present in the halos of galaxies like the Milky Way and M31. Comparing additional observations of dynamically cold stellar streams with fully self-consistent simulations will give more detailed information about the substructure population. The Gaia spacecraft's measurements of the positions and motions of vast numbers of Milky Way stars will be helpful in quantifying the nature of dark matter substructure Ngan and Carlberg (2014), Feldmann and Spolyar (2015).