Last year Rigetti Computing, a Berkeley, California-based developer of quantum computers, acquired QxBranch, an Australian-U.S. quantum software business. Rigetti is among the first companies to produce the first quantum computers.
The QxBranch team, which had already been working on prototypes with firms such as CBA and UBS, was repositioned to develop applications. This adds the software side to Rigetti’s hardware business.
DigFin caught up with Michael Brett, senior vice president of applications, based in Washington, D.C., and David Garvin, principal researcher of quantum applications, based in Sydney, to understand where the technology stands with regard to financial services.
If you need some more background on what quantum computing is and how it works, see our glossary entry.
What’s the state of play right now?
Brett says there are three big trends.
First, the past two years have seen the number of companies providing access to quantum computers has gone from two (IBM and Rigetti) to many more. “These offerings have gone beyond prototypes,” he said. “The computers do empirical work to study applications, create benchmarks, and investigate new approaches to computing.”
Second, quantum has been transformed by the entry of big cloud vendors. Rigetti for example has a deal with AWS to use its cloud compute time. Microsoft Azure offers a similar service, and other cloud players are expected to follow. This not only opens up a more efficient array of computing, but it changes the market dynamics: “AWS is both an aggregator and a distributor of quantum compute, as well as a provider of software tools to help developers,” Brett said.
Third, performance is improving, although it isn’t good enough yet to offer a commercial, cost-effective advantage over classical computing. “But benchmarking reports suggest we’re getting close,” Brett said.
What has to happen to create meaningful use cases? Brett says the industry needs to achieve three more goals before quantum goes commercial.
First, companies like Rigetti need to make quantum computing better, which means more qubits and lower error rates. An even bigger technical challenge is improving the interaction between quantum and classical compute to solve real problems. In Rigetti’s case, this includes new approaches to get the most from hybrid quantum-classical environments and bringing that into the AWS environment.
Third, companies need to build applications. Banks, oil and gas companies, and chemical companies are likely to be the first major consumers.
What apps matter to banks?
Garvin, who previously headed the quantitative research team at CBA and worked on its early experiments with the tech, says there are three potential applications.
First is optimization of portfolios: how a trader manages a book, how a desk transacts big trades without impacting the market, and how to settle trades with minimal credit or liquidity risk.
Second is using Monte Carlo algorithms to manage and price risks, and to put a valuation on assets. A Monte Carlo simulation attempts to work out probabilities despite the existence of many variables that otherwise make it difficult to make confident decisions. It’s a quant tool to try to measure the likelihood of various outcomes in messy real-world situations. In practical terms, banks use it to price options.
- Read more:
- Quantum computing: as explained by MIT
- CBA embarks on quest for quantum supremacy
- Huishang Bank turns to quantum communication
Banks are big users of Monte Carlo simulations but running these takes a long time because the complexity of it eats up a huge amount of computing resource. Quantum Monte Carlo simulations could be much faster and should in theory enable much faster decisions. “This is still out of reach of our capacity,” Garvin said. “Currently available quantum computers don’t have enough qubits to encode sufficiently complicated problems.”
Third is to improve machine learning. Quantum computing can improve a machine’s ability to learn, the speed at which it can classify data, and generate inputs for trading or other types of algorithms. The upshot: algos at banks should become a lot cleverer and adaptable.
How long will it take to see such apps in the market?
Garvin says nobody knows for certain. Banks are still trying to find the right use case given the existing constraints on the technology.
Some users are exploring “quantum-inspired” algorithms, taking the ideas behind how quantum works and applying it to classically produced algos. Garvin says this is not a stopgap but a new area of research that can help banks realize quantum’s potential faster.
How does this work? Garvin says that a classical computer approaches a Monte Carlo simulation one factor at a time, and tries to work out whether a particular factor is “right”, a match, or not. A quantum computer would address all the factors simultaneously, looking for probabilities or levels of confidence, but not an outright match. Classical computers can’t do this but they can take factors in batches, crunching them in parallel, while looking for matches, not probabilities.
These offerings have gone beyond prototypesMichael Brett, Rigetti
This hybrid approach could yield results. Other banks, however, are taking a longer view, building internal knowledge and developing in-house talent in order to take full advantage as the tech matures.
In January, J.P. Morgan Chase in New York hired Marco Pistoia, formerly IBM’s head of quantum computing algorithms. Goldman Sachs has set up a partnership with another quantum computing company, QCWare, to investigate using the tech to price options via Monte Carlo simulations.
What are the differences between classical and quantum computing that make its deployment so uncertain?
Garvin says some things are similar. The computer languages can be the same. He says quantum apps are usually written in Python, a popular coding language for traditional computers. “But the commands you use are completely different,” he said.
That difference centers around probabilities, which speaks to the nature of quantum mechanics, based on concepts such as entanglement and superpositioning. Quantum computers begin with giving all measurements and equal probability of being true. The process of writing an algo is about adjusting those probabilities, so that at the end, you are left with a high probability of measuring whatever is being examined.
Take the idea of a database search. If you ask Google to search for this article, you input the keywords, and Google will start to query relevant pages one by one, to see if they are a good match, and then it ranks them. But it’s a brute-force, one-after-the-other approach. If a search query has, say, 100 items to sift through, on average, it’s going to take 50 attempts before Google finds the right result.
In quantum computing, every item is measured in parallel. You only need one query, with the algo adjusting the physical hardware (via commands to entangle a qubit and rotate its phase) to fit the logic of the command. If your quantum computer is robust enough, this can happen at lightning speed.
Because today’s quantum technology is still in its infancy, computation times are unpredictable and unreliable. Running an algo is a fiddly business.
The reason is something called decoherence time. It means the computation breaks down because the computer can’t keep its qubits aligned for long enough.
Garvin says banks are making progress in both growing the power of quantum computing while improving decoherence times, so the algos can run for longer.
It’s this still-clunky but maturing nature of the technology that makes quantum computing both potentially revolutionary but also uncertain.