Supercomputers rely on a "brute force" approach to solve problems, performing billions of calculations very quickly until they arrive at the optimal solution. Quantum computers are emerging that could exponentially speed up this process by performing far more calculations simultaneously, but even they still work on the same principle. But a completely new system has the potential to outperform all of those, using a "magic dust" made of both light and matter as a beacon to highlight the solution directly.
For highly complex systems, like those in biology, finance, physics and space travel, many variations of mathematical formulations need to be sorted through to find the optimal solution – the lowest possible amount of steps that it takes to solve a particular problem.
To illustrate the concept, researchers use the analogy of a hiker trying to find the lowest point of mountainous terrain: exploring on foot in a linear way, the hiker would have to wander the entire area to find the lowest point, never sure if there's an even lower section out of sight, behind another mountain.
Over the years, advances in computer science have constantly sped up that process. The earliest supercomputers were capable of a million calculations per second, while today's fastest supercomputer – the Sunway TaihuLight in China – can perform a staggering 93 quadrillion calculations per second. To continue the analogy, that's like strapping faster and faster rocket skates to that hiker, helping them zip around the terrain and single out the lowest point at blistering speeds.
But researchers from the Universities of Cambridge, Southampton and Cardiff, as well as the Skolkovo Institute of Science and Technology in Russia, are developing a different approach. Rather than constantly speeding up the trial-and-error method, their technique would sample the entire landscape at once and shine a light on the solution immediately.
The idea uses quantum particles called polaritons, which are half light and half matter. In theory, a "magic dust" made of these particles could be sprinkled over the terrain, and when they pool in a section – the lowest point – they would glow there, acting as a beacon for the optimal solution. That was the idea, at least, but implementing it was trickier.
"A few years ago our purely theoretical proposal on how to do this was rejected by three scientific journals," says Natalia Berloff, first author of the paper. "One referee said, 'Who would be crazy enough to try to implement this?!' So we had to do it ourselves, and now we've proved our proposal with experimental data."
To make their magic dust particles, the team shone a laser on layers of atoms like gallium, arsenic, indium and aluminum, causing the electrons in those layers to absorb and emit light of a certain color. That emergent behavior produces the quasiparticles known as polaritons, and since they're 10,000 times lighter than electrons, these polaritons can pack themselves in very close together, forming a strange state of matter called a Bose-Einstein condensate. In that form, the quantum phases of the polaritons sync up, causing them to light up in a way that's detectable.
But that's a practical experiment, while the idea of the hiker's mountainous terrain is purely metaphorical, right? In order to use their proposed technique, the researchers had to find a way to represent the landscape that corresponds to the problem that they want to solve. To do so, they used the XY model, a fundamental kind of optimization problem that's represented on two planes of a graph. This technique is simple enough to model, but versatile enough to be applied to a range of more complicated problems.
The researchers were able to create polaritons on the vertices of an XY graph, and as they condense they pool at a point that corresponds to the absolute minimum of the function they're trying to solve. In the long run, this system could help speed up supercomputers dramatically by removing the "hiker" from the picture entirely.
"We are just at the beginning of exploring the potential of polariton graphs for solving complex problems," says Pavlos Lagoudakis, co-author of the study. "We are currently scaling up our device to hundreds of nodes, while testing its fundamental computational power. The ultimate goal is a microchip quantum simulator operating at ambient conditions."
The research was published in the journal Nature Materials.
Source: University of Cambridge