## Work establishes benchmark for beating classical computers, optimizes quantum performance

Quantum computers promise vastly expanded computing power. But they’ve yet to prove beyond a doubt that they’re faster than classical computers for specific tasks. It isn’t even completely clear how best to compare the two. Scientists from Carnegie Mellon University (CMU) and Intel Labs used PSC’s Bridges system to simulate a quantum computer, showing how to optimize a particular approach to quantum computing and benchmarking it successfully against classical computing. The work provides a signpost of both how to compare quantum and classical computers, and how to improve the quantum computer so it can win.

## Why It’s Important

The jury is still out on quantum computing’s real-world relevance. But there’s a lot of promise.

Classical computers, which deal with data in *bits* of ones and zeroes, are incredibly powerful but have their limitations. Today, we’re bumping up against the end of Moore’s Law — not really a law, but an observation that, for a long time, we could double the computing power on a single chip about every two years. That hot streak in computer design gave us smartphones today that are as powerful as PSC’s first supercomputer in 1986. But there are a lot of signs that the era of Moore’s Law is coming to a close. If we want to pack more computing power into the same space, we may need a completely different approach.

Quantum computers take advantage of the weird fuzziness of the quantum realm. Instead of computing in one or zero bits, they can compute in “qubits,” which are both one *and *two at the same time, in varying degrees. Also, qubits can be *entangled*. That is, once coupled, two quantum particles will keep communicating with each other no matter how far apart they travel. That potentially makes quantum computers able to communicate and work across distant parts of the Earth or even farther. These features make quantum computers potentially far more powerful and able to tackle vastly more complex problems than classical computers.

### “This paper is a study on this particular quantum algorithm — QAOA — trying to ask the question ‘What does this algorithm need in terms of quantum computing resources?’ [as well as] design the algorithm such that it could be competitive with classical state-of-the-art computing. That’s the question that everyone is asking in this space: what are quantum computers good for [and] what are they going to be good for in five years and 10 years?”

—Jason Larkin, SEI

The biggest “to do” in quantum-computing research, though, is to demonstrate that a quantum computer can solve a given problem *fundamentally* faster than a classical computer. Google has claimed such “quantum supremacy” in one problem, but scientists at IBM and in China have questioned whether the quantum result was *sufficiently *better. If achieved, quantum supremacy could offer vast improvements in problems like managing data security; simulating real-world quantum chemistry and physics for designing drugs or advanced reactors; solving logistics problems such as airline-flight routing; and applications that scientists haven’t even thought of yet.

Jason Larkin of CMU’s Software Engineering Institute (SEI) and colleagues at CMU and Intel Labs wanted to better define the comparison of quantum and classical computers, as well as to build a simulated quantum computing framework that both allows testing on as many different types of problems as possible and maximally improves an eventual quantum computer’s performance. To do this, they had to “build” a simulated quantum computer. They needed this simulation because current quantum computers are neither large enough nor is their error correction good enough yet to fully test a quantum-versus-classical matchup. For this daunting task they chose the National Science Foundation-funded Bridges advanced research computer at PSC.

**Jason Larkin**

SEI

## How PSC Helped

One reason why scientists are keen on developing practical quantum computers is that some important problems grow exponentially with the data — if you double the data, the problem takes four times as many calculations; triple the data, eight times; quadruple, 16 times. In theory a quantum computer’s complex handling of data and computations — which also grows exponentially, with the number of qubits in the machine — could solve that kind of problem easily. (__Readers can find a catalog of current quantum algorithms and their speed-up over classical computers here__.)

Simulating a quantum computer, though, also takes exponentially ballooning data, as the number of simulated qubits grows. Bridges, with dozens of large- and extreme-memory computing nodes that offered either 3 or 12 terabytes (3,000 or 12,000 gigabytes) of memory (RAM) each, was the perfect fit for simulating a small quantum computer. (By comparison, a typical laptop computer has 16 or 32 gigabytes of RAM.)

### “Simulating quantum computers, at least the way we’re doing it, you have to store this data structure. That’s two to the number of qubits in size, and by the time you get to around 40 qubits you’re well into the terabyte range … Whatever is the max memory on a node, that kind of defines the biggest qubit-size job we can run. So PSC had these very nice nodes that have lots of memory and lots of RAM per node, and so we took advantage of that.”

—Jason Larkin, SEI

Larkin and the team used Bridges to test the quantum approximate optimization algorithm (QAOA), a hybrid quantum-classical set of instructions for running the computer that combines strong error correction — a must, with the fuzzy math of quantum computers — with the ability to run on both classical computers and, at larger scales, on future quantum computers. QAOA gave a strong performance running on Bridges. In particular, it produced approximate solutions to the *maximum cut problem*, useful in both designing electronic circuits and software validation and verification. Judging QAOA’s performance versus classical computing will require more detailed study. The quantum algorithm’s performance relative to the classical approach depended on the size of the problem and number of qubits involved, and the scientists would like to understand that relationship better. But their results tested the QAOA approach successfully, taught the team how to optimize it and set a benchmark for what a real quantum computer will need to do to demonstrate quantum supremacy. __The team reported their results in the journal IOPscience in April 2022__.

Next, the team would like to expand their understanding of QAOA’s performance, giving it more rigorous tests against classical computing. The team are now using Bridges-2, Bridges’ successor system. They’re also working with the team running PSC’s Neocortex experimental artificial intelligence computer, whose huge Cerebras CS-2 “wafer scale” chips may enable even larger quantum-computer simulations.