One configuration designed by the computer algorithm Melvin, and subsequently implemented in the lab, involved the creation of a high-dimensional GHZ entangled state. [Image: Mehul Malik/University of Vienna]
The various manifestations of “quantum weirdness,” such as entanglement and teleportation, are proverbially counterintuitive—so much so that figuring out optical experiments to implement quantum phenomena poses a tough challenge for human researchers. A group of scientists in Vienna decided to try another approach: create a computer algorithm that can spit out potential designs for quantum experiments (Phys. Rev. Lett., doi: 10.1103/PhysRevLett.116.090405).
The algorithm, christened “Melvin,” has already dreamt up a variety of new configurations for implementing complex quantum states, using standard laboratory optical components such as beamsplitters and nonlinear crystals. And, the researchers say, most of the configurations, while viable, are not approaches they would have thought of themselves—and will prove a challenge to understand.
Rooting around in the optical toolbox
In essence, Melvin approaches its task by taking mathematical descriptions of common laboratory optical components, and shuffling the components around in different hypothetical experimental setups until one of those setups yields the desired goal.
After being given the specs for a quantum phenomenon to implement, the program begins with a random arrangement of the components in the toolbox, and calculates the quantum transformation that would be executed by the setup, to determine whether the arrangement meets the criteria for the desired quantum state. Melvin then reshuffles the components and retests, iterating until it arrives at an arrangement that achieves the targeted quantum transformation. At that point—after a simplification run to cull out any unnecessary components in the setup—it writes a file with the final experimental recipe.
An important step in developing such an algorithm was to put the components of the optical toolbox into a language that the computer could understand. To do so, the team wrote symbolic-algebra rules to summarize the effects of individual optical components such as beamsplitters and half-wave plates, and fed them to Melvin to use as raw material in designing experiments. Even more intriguingly, Melvin can learn from experience; when it develops an experimental configuration that achieves a particular quantum goal, it can store that configuration and add it to its toolbox to use in more complex experiments.
Surprising arrangements
Running for 150 hours on a notebook computer (with Wolfram Mathematica as the programming platform), Melvin was able to develop 51 different experimental setups that, according to the researchers, would produce “genuinely different” kinds of quantum entanglement among more than two particles. One of those is what the team refers to in the paper as the first experimentally realizable scheme for a higher-dimensional Greenberger-Horne-Zeilinger (GHZ) state. These correlated states among more than two photons have the potential to reveal new and interesting physics, but have been tough to produce experimentally in the lab. The computer was also able to devise schemes for creating high-dimensional cyclic transformations that could prove useful in quantum cryptography.
But while Melvin seems to excel at churning out quantum experiments, it’s clear that it is not exactly thinking like a human. Instead, the algorithm tends to offer up surprising schemes for experimental setups that are clearly feasible but, as the authors dryly note, “challenging to understand.” Even so, the researchers find the algorithm’s unconventional approach a feature, not a bug: Melvin’s approach to design, they suggest, “provides some insight into the kind of out-of-the-box thinking that is required for creating such complex quantum states.”