Phys. Rev. Lett. (in the press); preprint at

Engineering complex quantum states requires complicated experimental set-ups. Take quantum optics, for example: to create high-dimensional multiparticle entanglement one must carefully arrange numerous optical components in the correct order. Anyone who has seen a crowded optical table will appreciate that it takes skill and experience to design such experiments and the way to optimize them is not obvious. But where humans may fail, machines can help.

Mario Krenn and colleagues created a clever algorithm to help explore the possible ways of generating a desired quantum state using given building blocks such as beamsplitters, half-wave plates or Dove prisms. Starting with a random combination, the resulting quantum state is calculated and analysed. If it satisfies the required criteria, the set-up is then simplified and added to the toolbox — otherwise the search continues.

The algorithm learns from experience by adding the useful solutions to its toolbox, speeding up the search significantly. This automated exploration reveals useful but quite unusual techniques, previously unknown to human experimentalists, hinting at unconventional and rather unintuitive approaches to experiment design.