Classical computers can now simulate noisy quantum circuits as fast as shallow ones — depth no longer gives quantum hardware an edge. A new Nature Physics result is a formal proof, not a guess.
image from grok
Researchers from multiple institutions prove in Nature Physics that uncorrected noise truncates most quantum circuits to effectively logarithmic depth, as early operations exponentially fade from memory in final measurements. The team further demonstrates an efficient classical algorithm that estimates noisy circuit outcomes with constant additive accuracy independent of circuit depth, challenging near-term quantum advantage claims for variational algorithms. Counterintuitively, they also show non-unital noise eliminates barren plateaus, potentially making noisy hardware more trainable than ideal hardware.
When quantum computing advocates describe their roadmap, the story is always the same: more qubits, lower error rates, deeper circuits, eventually computational supremacy. A paper published this month in Nature Physics complicates that story in a way that deserves to be stated plainly: noise cuts your quantum circuit down to size whether you want it to or not.
The finding, from researchers at Free University of Berlin, the University of Copenhagen, EPFL, the University of Chicago, MIT, and ENS Lyon: any uncorrected noise truncates most quantum circuits to effectively logarithmic depth. Earlier operations in a circuit fade from memory. Their influence on final measurement outcomes decays exponentially with their distance from the end. Only the last few layers of a deep noisy circuit actually matter. The rest is, in a meaningful sense, wasted computation. EPFL's press release frames it as a fundamental limit on what current quantum hardware can actually compute.
The authors write in their paper on arXiv: "any noise truncates most quantum circuits to effectively logarithmic depth, in the task of estimating observable expectation values." Observable expectation values are exactly what variational quantum algorithms are designed to compute. Those are the near-term quantum computing applications that were supposed to deliver the first real commercial wins.
The paper does not stop there. Using the effective shallowness that noise imposes, the authors design an efficient classical algorithm that estimates noisy circuit outcomes within constant additive accuracy. The runtime is independent of circuit depth. For one-dimensional architectures it runs in polynomial time. For higher dimensions, quasi-polynomial. Classical computers can simulate noisy quantum circuits without the depth penalty that quantum researchers have long assumed they would incur.
The authors put it this way: "such circuits behave like shallow circuits, which have limited computational power."
The implication is direct. If your quantum circuit is deep enough to be interesting, the noise makes it behave like a shallow one. Shallow quantum circuits do not outperform classical algorithms on hard problems. Quantum Insider covered the result as a challenge to near-term quantum advantage claims.
There is one result in the paper that cuts the other way. Non-unital noise, meaning noise that does not preserve the identity element of quantum mechanics, provably eliminates barren plateaus. Barren plateaus are the failure mode where a quantum circuit's gradient becomes exponentially small and the training stops. The authors prove that non-unital noise prevents this. Counterintuitively, noisy hardware may be more trainable than ideal hardware would suggest.
The paper includes Jens Eisert of Free University of Berlin and Fraunhofer HHI, a researcher who has published extensively on quantum many-body physics and variational quantum algorithms. The combination of the noise truncation result with the depth-independent classical simulation algorithm is new. ScienceDaily reported the result under the headline that quantum computers forget most of their work — an accurate summary if imprecise.
What this means for companies building noisy quantum hardware is uncomfortable but specific. The classical algorithm described in the paper does not break RSA or simulate molecular dynamics faster. It simulates noisy circuits estimating observable expectation values. But that is exactly what variational quantum eigensolvers, quantum approximate optimization algorithms, and most quantum machine learning proposals try to do. Those applications were supposed to be the landing zone for near-term quantum computing. If classical software can do the same job without the hardware, the landing zone shrinks.
The authors are careful not to overclaim. Circuits engineered to exploit noise rather than merely tolerate it might still outperform classical simulation. Designing those circuits is an open problem. The paper is an honest assessment of where things stand: unless you know how to weaponize your noise, your deep circuit is probably just a shallow one with extra steps.
Whether that open problem gets solved, and how fast, is the next question worth watching.
Story entered the newsroom
Research completed — 5 sources registered. Noise truncates typical quantum circuits to O(log n) effective depth for expectation value estimation; classical simulation algorithm runs in time ind
Draft (623 words)
Reporter revised draft (630 words)
Reporter revised draft based on fact-check feedback (585 words)
Reporter revised draft based on fact-check feedback (630 words)
Reporter revised draft based on fact-check feedback (590 words)
Approved for publication
Published (630 words)
Get the best frontier systems analysis delivered weekly. No spam, no fluff.
Quantum Computing · 13h 13m ago · 4 min read
Quantum Computing · 14h 13m ago · 5 min read
