Search

When assisting human users in reinforcement learning (RL), we can represent users as RL agents and study key parameters, called \emph{user traits}, to inform intervention design. We study the relationship between user behaviors (policy classes) and user traits. Given an environment, we introduce an intuitive tool for studying the breakdown of "user types": broad sets of traits that result in the same behavior. We show that seemingly different real-world environments admit the same set of user types and formalize this observation as an equivalence relation defined on environments. By transferring intervention design between environments within the same equivalence class, we can help rapidly personalize interventions.

We study the problem of gravitational lensing by binary galaxies, idealized as two isothermal spheres. In a wide binary, each galaxy possesses individual tangential, nearly astroidal, caustics and roundish radial caustics. As the separation of the binary is made smaller, the caustics undergo a sequence of metamorphoses. The first metamorphosis occurs when the tangential caustics merge to form a single six-cusped caustic, lying interior to the radial caustics. At still smaller separations, the six-cusped caustic undergoes the second metamorphosis and splits into a four-cusped caustic and two three-cusped caustics, which shrink to zero size (an elliptic umbilic catastrophe) before they enlarge again and move away from the origin perpendicular to the binary axis. Finally, a third metamorphosis occurs as the three-cusp caustics join the radial caustics, leaving an inner distorted astroid caustic enclosed by two outer caustics. The maximum number of images possible is 7. Classifying the multiple imaging according to critical isochrones, there are only 8 possibilities: 2 three-image cases, 3 five-image cases, and 3 seven-image cases. When the isothermal spheres are singular, the core images vanish into the central singularity. The number of images may then be 1, 2, 3, 4 or 5, depending on the source location, and the separation and masses of the pair of lensing galaxies. The locations of metamorphoses, and the onset of threefold and fivefold multiple imaging, can be worked out analytically in this case., Comment: MNRAS, in press

Gravitational lenses with anomalous flux ratios are often cited as possible evidence for dark matter satellites predicted by simulations of hierarchical merging in cold dark matter cosmogonies. We show that the fraction of quads with anomalous flux ratios depends primarily on the total mass and spatial extent of the satellites, and the characteristic lengthscale R of their distribution. If R is 100 kpc, then for a moderately elliptical galaxy with a line-of-sight velocity dispersion of 250 km/s, a mass of 3 x 10^9 solar masses in highly-concentrated (Plummer model) satellites is needed for 20% of quadruplets to show anomalous flux ratios, rising to 1.25 x 10^10 solar masses for 50%. Several times these masses are required if the satellites have more extended Hernquist profiles. Compared to a typical elliptical, the flux ratios of quads formed by typical edge-on disc galaxies with maximum discs are significantly less susceptible to changes through substructure -- three times the mass in satellite galaxies is needed to affect 50% of the systems. In many of the lens systems with anomalous flux ratios, there is evidence for visible satellites (e.g., B2045+265 or MG0414+0534). We show that optically identified substructure should not be preponderant among lens systems with anomalies. There are two possible resolutions of this difficulty. First, in some cases, visible substructure may be projected within or close to the Einstein radius and wrongly ascribed as the culprit, whereas dark matter substructure is causing the flux anomaly. Second, bright satellites, in which baryon cooling and condensation has taken place, may have higher central densities than dark satellites, rendering them more efficient at causing flux anomalies., Comment: 10 pages, 14 figures, MNRAS, in press

We study the gravitational lensing effects of spiral galaxies by taking a model of the Milky Way and computing its lensing properties. The model is composed of a spherical Hernquist bulge, a Miyamoto-Nagai disc and an isothermal halo. As a strong lens, a spiral galaxy like the Milky Way can give rise to four different imaging geometries. They are (i) three images on one side of the galaxy centre (`disc triplets'), (ii) three images with one close to the centre (`core triplets'), (iii) five images and (iv) seven images. Neglecting magnification bias, we show that the core triplets, disc triplets and fivefold imaging are roughly equally likely. Even though our models contain edge-on discs, their image multiplicities are not dominated by disc triplets. The halo has a small effect on the caustic structure, the time delays and brightnesses of the images. The Milky Way model has a maximum disc (i.e., the halo is not dynamically important in the inner parts). Strong lensing by nearly edge-on disc galaxies breaks the degeneracy between the relative contribution of the disc and halo to the overall rotation curve. If a spiral galaxy has a sub-maximum disc, then the astroid caustic shrinks dramatically in size, whilst the radial caustic shrinks more modestly. This causes changes in the relative likelihood of the image geometries, specifically (i) core triplets are now 9/2 times more likely than disc triplets, (ii) the cross section for threefold imaging is reduced by a factor of 2/3, whilst (iii) the cross section for fivefold imaging is reduced by 1/2. Although multiple imaging is less likely (the cross sections are smaller), the average total magnification is greater., Comment: MNRAS, in press