A new theoretical study involving EPFL shows how the noise in today's quantum computers limits how much work their circuits can really do, and how this affects training and simulation.
Imagine you're trying to build a very long, complicated chain of dominoes. The aim is that each domino hits the next one perfectly, all the way down the line, producing an amazing result at the end.
A quantum circuit is like a domino chain: A long chain of tiny steps ("operations") that work together to process information together in a powerful way.
Now imagine that every domino is a little wobbly. In the quantum circuit, that wobble is called "noise". It might not look like much-after all, all regular systems are subjected to some kind of "noise"-but noise in quantum circuits can accumulate and build up to a crescendo of problems.
Noise constraints the power of quantum circuits
The question is, if every domino piece is wobbly-if noise is inevitable and eventually destructive-does it still make sense to build complex chains? Quantum circuits lie at the heart of the next generation of technology, like quantum computers, which promise to solve certain problems far beyond the reach of today's machines.
A team of scientists has now carried out a broad theoretical analysis of how noise affects quantum circuits. Their findings show that noise sets a surprisingly tight practical limit on how deep such circuits can be - in other words, how many steps can be applied one after another in a quantum circuit - while also making parts of them easier to simulate on classical computers.
The work was led by together by researchers Armando Angrisani and Yihui Quek at EPFL, Antonio Anna Mele at the Free University of Berlin and Daniel Stilck França at the University of Copenhagen. It is published in Nature Physics.
Focusing on the last few layers
The researchers analyzed large families of quantum circuits made of simple two‑qubit operations, with realistic noise affecting each individual qubit after every step. They treated the problem mathematically, tracking how the influence of each layer propagates through the circuit when noise is present.
The analysis showed that that in most noisy quantum circuits, only the last few steps really matter. Even if a circuit is built to be very deep, noise gradually wipes out the influence of the earlier steps.
In the domino analogy, only the last few pieces matter. This means that if we use a quantum computer to estimate a physical quantity, like the energy or the state of a qubit, the result will be mainly shaped by what happens at the very end of the circuit. Earlier operations "fade from memory" as noise builds up.
Tackling noise from the ground up
The study also found that this explains why these noisy circuits can still be adjusted or "trained" for simple tasks. Changing the circuit's settings can still change the outcome, but only because the final layers remain active. As a result, a deep noisy circuit ends up behaving like a much shallower one.
The study clarifies what near‑term quantum machines can realistically deliver. Simply stacking more layers onto noisy circuits is unlikely to unlock new power for common tasks based on local measurements. Rather, progress depends on better noise control or on carefully engineered designs that exploit specific noise properties.
The work also warns that noisy circuits remain trainable only because noise has already weakened most of their power, and treating real hardware noise as a simple blur can lead to false expectations.
Contributors
- Free University of Berlin
- EPFL
- University of Sorbonne
- University of Chicago
- Fraunhofer Heinrich Hertz Institute
- ENS Lyon
- MIT
