FAQ

Here are some frequently asked questions I often get:

• What is your underlying lens model in the "Lensing of '69"-approach?

None. The method only employs the general lensing formalism as, e. g. detailed in Schneider, Ehlers & Falco (1992). Using a Taylor expansion of the general lensing potential around a point on the critical curve, no specific lens model is needed. Subtracting the lensing equations of multiple images, i.e. images from the same source, from each other, no source model is required, either.

• How do you account for substructure in your lensing method?

Common lens modelling approaches split the total mass distribution into a smooth large-scale part and small-scale perturbations, so-called substructure. Contrary to that, the observation-based approach employs the full lensing potential and does not split it into parts. In most cases, it is difficult to disentangle a main lens and perturbers because the degeneracies of gravitational lensing have to be broken by additional measurements.

• What's the relevance of the approach given that state-of-the-art models are very precise?

From an epistemological point of view, we can only reject models that are inconsistent with observations, but given several equally consistent models, we do not know which model to prefer over the other(s). From a practical point of view, lens modelling is still run-time intensive and often requires manual adjustments to reach sub-arcsecond precision in the reconstruction of the (multiple) images. The observation-based approach separates the information contained in the observations (i.e. the properties of the multiple images) from model assumptions. Hence, it yields the information all models must have in common. Furthermore, it is based on a simple set of equations that, to its current extent, has a unique optimum solution. Thus, we can quickly extract the lens and source properties contained in the multiple images without further manual interaction.

• Can we now forget about all lens modelling?

No. The observation-based approach is purely data-based and as such only yields information in the vicinity of multiple images. If we want to derive global lens properties, like a lensing (aperture) mass, we have to extrapolate the observation-based local information. For this task, model assumptions are necessary and can be based on the observation-based information.

• Why don't you train a neural network to solve the problem by machine learning?

Neural networks have become a suitable means for many tasks in astrophysics and cosmology. Detecting giant arcs or deriving lens properties from multiple images can certainly be accomplished by neural networks. But they are also subject to degeneracies, e.g. in their architecture, and require a lot of training samples. Most importantly, they cannot provide us with a deeper understanding of the underlying physics (yet?). Hence, they are not an appropriate tool to improve our understanding of the fundamental principles behind gravitational lensing, which is the main goal of my research.

• Why haven't I gotten an answer to my phone calls and instant messages?

Simply because I do not like these ways of communication. Write me an email or invite me for a cup of tea!