A model of IBM’s Quantum System Two quantum computer

Marijan Murat/dpa picture alliance / Alamy

Quantum computers will not become useful until researchers perfect methods for correcting errors that these devices naturally make. Quantum error-correction (QEC) techniques have advanced rapidly in recent years, but several studies now suggest that the landscape of errors they must tackle is unexpectedly complex, which poses new challenges for making quantum computers practical.

The idea of QEC dates back to the very inception of quantum computing and the first investigations of how quantum computers could be put to work. At that time, no one had built a quantum computer, so researchers relied on mathematical theory to work out how well errors could be corrected, and which computations this error correction would allow the computers to reliably complete. Now, quantum computers do exist and data about their errors is fast accumulating. Increasingly, researchers are finding there are some types of errors that the earlier investigations had overlooked, and they could threaten quantum computers’ progress.

Quantum computers are made from building blocks called qubits, and qubits make errors because they are sensitive to interactions with their environment. One long-standing assumption about these errors – the Markov assumption – has been that they are memoryless: errors that a qubit makes at different points in time are uncorrelated.

Memory effects are detrimental for the useful operation of quantum computers because they mean that making a qubit repeat the same operation at different times may lead to different results based on what its environment “remembers”, says Lorenza Viola at Dartmouth College in New Hampshire. It is as if multiplying a number by 2 gives one value right now and a different value later. The Markov assumption posits that this happens either never or in a small number of cases, which would allow quantum computers to run longer programs. But troublingly, the assumption seems to be incorrect. Memory effects keep appearing in experiments.

Recently, Ioan Pop at the Karlsruhe Institute of Technology and Stuttgart University in Germany and his colleagues developed a measurement method that uncovered very long-lived defects – spurious quantum states that arise from material imperfections, stray molecules or several other sources – in a superconducting qubit’s environment, each of which could act as a memory. “We are concretely showing how this long-lived environment can break one of the assumptions of error correction,” he says.

Earlier this year, Rami Barends at Forschungszentrum Jülich GmbH in Germany and his colleagues studied a different type of superconducting qubit and found similar violations of the Markov assumption.

The assumption has been challenged at the level of whole quantum computers, too. In 2025, John Fidel Kam at the Agency for Science, Technology and Research in Singapore, Kavan Modi at the Singapore University of Technology and Design, Muhammad Usman at Monash University in Australia and their colleagues analysed a key error-correction experiment with Google’s Willow quantum computer. They found that some of the errors that were difficult to correct or account for matched a theoretical model in which the Markov assumption didn’t hold. The qubits’ environment was non-Markovian.

In another analysis, Gregory Quiroz at Johns Hopkins University in Maryland and his colleagues discovered that a model including non-Markovian effects represented an IBM quantum computer that they accessed through the cloud seven times better than the noise models commonly used in studies of error-correction. IBM and Google did not respond to New Scientist’s request for comment.

Working with colleagues, Viola has found that the parts of a qubit’s environment that can “remember” can additionally act as quantum systems in their own right and “respond” to a qubit. This introduces a new kind of error – another effect that QEC approaches currently do not account for.

This work is preliminary, but Viola says that as quantum computers are rapidly improving, previously irrelevant effects could become increasingly important. Modi thinks it is too early to sound the alarm, but he says that there are currently no guarantees, experimental or theoretical, that the problem will not worsen as quantum computers grow in qubit number.

“There is a chance that error correction does not work because correlated noise [memory effects] has not been accounted for,” he says. “We don’t fully know where it breaks down, partly because complex noise is hard to describe.”

In Usman’s view, the future is rather open because the programs researchers use for quantum error-correction are themselves evolving alongside quantum computing hardware. Until recently, QEC programs that date back to the 1990s were the gold standard, but a whole new class of codes known as QLDPC have now gained traction, and how well they work in the face of correlated errors is not yet clear, he says. “There is a lot more to investigate and study before we make any claim about whether error correction is going to work or not,” he says.

“Error correction codes can adapt to non-Markovian errors. They will have to adapt to that,” says Pop. The measurement technique that he and his colleagues developed recently could play a part here because it may enable other teams to identify memory effects in their devices and devise strategies for avoiding them, says his colleague Nicolas Gosling at the Karlsruhe Institute of Technology. Correlated errors could also be mitigated by adjusting conventional computer algorithms that control quantum computers and read their output, says Kam.

So, there are several ways for tackling this neglected class of errors. The researchers who study them say that they haven’t necessarily uncovered an unsolvable problem – as long as it’s taken seriously. “It was not on the radar of many people until now,” says Pop.