Using machine learning to optimise chameleon fifth force experiments

The chameleon is a theorised scalar field that couples to matter and possess a screening mechanism, which weakens observational constraints from experiments performed in regions of higher matter density. One consequence of this screening mechanism is that the force induced by the field is dependent on the shape of the source mass (a property that distinguishes it from gravity). Therefore an optimal shape must exist for which the chameleon force is maximised. Such a shape would allow experiments to improve their sensitivity by simply changing the shape of the source mass. In this work we use a combination of genetic algorithms and the chameleon solving software SELCIE to find shapes that optimise the force at a single point in an idealised experimental environment. We note that the method we used is easily customised, and so could be used to optimise a more realistic experiment involving particle trajectories or the force acting on an extended body. We find the shapes outputted by the genetic algorithm possess common characteristics, such as a preference for smaller source masses, and that the largest fifth forces are produced by small `umbrella'-like shapes with a thickness such that the source is unscreened but the field reaches its minimum inside the source. This remains the optimal shape even as we change the chameleon potential, and the distance from the source, and across a wide range of chameleon parameters. We find that by optimising the shape in this way the fifth force can be increased by $2.45$ times when compared to a sphere, centred at the origin, of the same volume and mass.