Quantum advantage demonstrated using Gaussian boson sampling
An optical circuit has performed a quantum computation called Gaussian boson sampling (GBS) 100 trillion times faster than a supercomputer could, according to researchers in China. This feat was achieved by Jian-Wei Pan and Chao-Yang Lu at the University of Science and Technology of China in Hefei, and colleagues. Although GBS is devised to show that a quantum computation can be done much faster than the same calculation on a conventional computer – a capability called quantum advantage – it may also have specialized practical applications.
Boson sampling is a way of computing the output of a linear optical circuit that has multiple inputs and multiple outputs. Single photons enter the circuit in parallel and encounter optical components such as beam splitters. Due to their bosonic nature, if two photons arrive at a beam splitter at the same time, they will both follow the same path. This property makes it extremely difficult to use a conventional computer to calculate the output of the circuit even for modest numbers of input photons and output channels. While boson sampling is difficult to do and requires state-of-the art quantum optics, it should vastly outperform even the most powerful supercomputers
A boson sampling circuit can be thought of as a matrix that makes a transformation of the input photons. Calculating the output involves working out the “permanent” of the matrix, which is related to the determinant of the matrix but is much more difficult to calculate. Boson sampling determines the permanent by sending groups of single photons into the optical circuit and measuring the output. The number of photons is less than the number of output modes of the circuit – so three photons could be sent into a six-mode circuit, for example.
In this latest research, researchers used a related technique called Gaussian boson sampling (GBS), in which single mode squeezed states of light are used in place of single photons. Instead of determining the permanent of the circuit, Gaussian boson sampling gives a similar quantity that is also extremely difficult to compute using a conventional computer.
According to Pan, GBS offers two important advantages over the single-photon technique. First is that the photon generation rate is much higher for GBS than it is for single-photon boson sampling, and second there are “many proposals for practical applications based on [GBS]”.
The team’s optical circuit has 100 inputs and 100 outputs and comprises 300 beam splitters and 75 mirrors that are arranged in a random manner. The system is fully connected, so a photon at any input port can emerge from any of the output ports.
GBS took about 200 s to make the desired calculation, whereas the team estimate that China’s fastest supercomputer, Sunway TaihuLight, would take 2.5 billion years to do the calculation.
“This experiment is definitely an important milestone for quantum simulations based on linear optical systems,” says Christine Silberhorn at Paderborn University in Germany – who along with colleagues was the first to propose GBS in 2017. She points out that scaling up the system to its 100 × 100 size would have been very challenging.
Ian Walmsley at Imperial College London agrees, adding that the team has made a “heroic effort” at “preparing quantum states that are entirely indistinguishable, and making sure the photons aren’t lost”. However, he points out that use of bulk optics rather than integrated optics could make it difficult to further scale-up the system.
Undaunted, Lu says that the team has made a “considerable improvement of the efficiency of the quantum light sources”, which he says should enable a 144 × 144 version of the experiment. Looking further ahead, he says, “In 2021, we will make the GBS machine more tuneable, more compact and more stable, and look for practical applications.”
Although the current system has no practical application beyond demonstrating quantum advantage, Pan says “we are excited by the potential usefulness of boson sampling as the community has come up with many ideas”.
Walmsley adds, “There are some interesting simulation problems that might benefit from this new scale boson sampler, including modelling of molecular spectra and vibrational dynamics. However, even those require the addition of a deterministic nonlinearity at the single-photon level in order to be able to handle real-world systems accurately.”
The research is described in Science (370 1460).
Hamish Johnston