high performance computing on graphics processing units: hgpu.org

In the era of synoptic surveys, the number of known gravitationally lensed quasars is set to increase by over an order of magnitude. These new discoveries will enable a move from single-quasar studies to investigations of statistical samples, presenting new opportunities to test theoretical models for the structure of quasar accretion discs and broad emission line regions (BELRs). As one crucial step in preparing for this influx of new lensed systems, a large-scale exploration of microlensing convergence-shear parameter space is warranted, requiring the computation of O(10^5) high resolution magnification maps. Based on properties of known lensed quasars, and expectations from accretion disc/BELR modelling, we identify regions of convergence-shear parameter space, map sizes, smooth matter fractions, and pixel resolutions that should be covered. We describe how the computationally time-consuming task of producing ~290000 magnification maps with sufficient resolution (10000^2 pixel/map) to probe scales from the inner edge of the accretion disc to the BELR can be achieved in ~400 days on a 100 teraflop/s high performance computing facility, where the processing performance is achieved with graphics processing units. We illustrate a use-case for the parameter survey by investigating the effects of varying the lens macro-model on accretion disc constraints in the lensed quasar Q2237+0305. We find that although all constraints are consistent within their current error bars, models with more densely packed microlenses tend to predict shallower accretion disc radial temperature profiles. With a large parameter survey such as the one described here, such systematics on microlensing measurements could be fully explored.