As quantum computing advances, there are periodic announcements of achieving Quantum Supremacy — a test where Quantum Computers (QC) complete some example algorithm much faster than classical computers. A good Q&A on Quantum Supremacy can be found in Scott Aaronson’s Blog.

One of the publicized results was in 2019. In this case, a quantum computer (Google’s 53-qubit Sycamore chip) built by Google had performed in such a way that the company claimed it would take 10,000 years to reproduce on supercomputing hardware of the day. The specific problem used was simulating the output of a random sequence of gates and qubits in a quantum computer. Though sounding entirely self-referential, the sequences of ones and zeros were derived through the random behavior of the qubits but exhibit a particular kind of random result that researchers can check.

In response, IBM released a paper in which they argue the 250 petabytes of storage on the Summit supercomputer at Oak Ridge could actually store the entire quantum state vector of Google’s Sycamore chip. This configuration would allow the same results to be computed in about 2.5 days by a brute force updating of the entire state vector (all 250 petabytes).

However, adding just a handful of additional qubits would re-establish an insurmountable lead for QC. If Google or someone else upgraded from 53 to 55 qubits, that would be enough to exceed Summit’s 250-petabyte storage capacity. At 60 qubits, you’d need 33 Summits, but who is counting?

In this instance, as the QC walks back to its corner with arms raised in celebration, the classical approach gets up and is ready for another round.

In a 2021 paper, researchers pointed out that Google chose a very specific method of computing the expected behavior of its processor, but there are other ways of doing equivalent computations. Since the published results, several classical options have reported results that perform better. As an example, in their paper, Feng Pan, Keyang Chen, and Pan Zhang described a specific method that allows a GPU-based cluster to produce the same results as the QC run in only 15 hours. The researchers noted that running the problem using a GPU-equipped supercomputer (like Summit) would outperform the Sycamore quantum processor.

I’m Still Standing

In June of this year (2023), IBM published a significant QC result in Nature. This time, instead of creating a special kind of randomness, the researchers used a 127 qubit IBM Eagle processor to calculate what’s known as an Ising model that simulates the behavior of 127 magnetic, quantum-sized particles in a magnetic field. The problem actually has some real-world value, including ferromagnetism, antiferromagnetism, liquid-gas phase transitions, and protein folding. When encoded into 127 qubits, it presents a quantum supremacy of scale rather than speed because even the largest classical computer will not have enough memory to hold the problem.

The IBM team used an interesting approach to mitigate the quantum noise and thus produce a more usable result. The researchers actually introduced more noise and then precisely recorded the effects on each part of the processor’s circuits. Using this data, the researchers could extrapolate what the calculations would have looked like without the noise.

The IBM result seemed like a real gut punch classical computing, but not enough to cause a knock-out. Within 2-weeks of the announcement researchers at the Flatiron Institute’s Center for Computational Quantum Physics rose to the challenge. They have pre-published a paper on their results and report that “By adopting a tensor network approach, we can perform a classical simulation that is significantly more accurate than the results obtained by the quantum device. ” They also mentioned the simulation used “modest computational resources.”

Not to be outdone, a recent pre-print from Tomislav Begušić, Garnet Kin-Lic Chan of California Institute of Technology stated, “Our classical simulations on a single core of a laptop are orders of magnitude faster than the reported wall time of the quantum simulations”

Counter punch. Ouch.

Everybody Wins

There are two somewhat complementary results we can expect as quantum computing advances.

1. QC continues to make strides in achieving workable systems and will eventually succeed (qubits are increasing, and noise is still a big issue)
2. QC will push classical computing to develop new algorithms that benefit many existing users.

Let’s keep up the good fight because every round leaves the market with more superpositioned winners.