Quantum circuits promise incredible computing power, but there’s a catch: noise builds up with every step.
Picture setting up an intricate line of dominoes, where each piece must fall perfectly to trigger the next. The goal is a smooth chain reaction that leads to an impressive final result.
Quantum circuits work in a similar way. They are built from many small steps, known as (“operations”), that act together to process information. When everything functions correctly, these steps combine to produce powerful computational outcomes.
Now imagine that every domino in the chain is slightly unstable. In quantum systems, this instability is called “noise.” While small amounts of noise exist in all physical systems, in quantum circuits, it can accumulate over time and create significant disruptions.
Noise Limits Quantum Computing Power
This raises an important question. If every step in a quantum circuit is affected by noise, does building longer and more complex circuits still make sense? Quantum circuits are central to technologies such as quantum computers, which are expected to solve problems beyond the reach of today’s machines.
A new theoretical study has taken a close look at how noise influences these circuits. The results show that noise places a surprisingly strict limit on how deep a quantum circuit can be, meaning how many steps can be applied in sequence. At the same time, noise makes parts of these circuits easier to simulate using classical computers.
The research was led by 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. Their findings were published today (April 2) in Nature Physics.
Why Only the Final Layers Matter
To understand the impact of noise, the researchers examined large groups of quantum circuits built from simple two-qubit operations. They modeled realistic conditions, where each qubit experiences noise after every step, and tracked how the effects of each layer move through the circuit.
Their analysis revealed a striking pattern. In most noisy quantum circuits, only the final steps have a meaningful impact on the outcome. Even if a circuit is designed to be very deep, the influence of earlier steps gradually disappears as noise builds up.
Returning to the domino comparison, it is as if only the last few pieces in the chain actually determine the final result. Earlier pieces still fall, but their contribution fades away.
This has practical consequences. When a quantum computer is used to estimate something like the energy or the state of a qubit, the result is largely determined by the final layers of the circuit. Earlier operations effectively “fade from memory” as noise accumulates.
Why Noisy Circuits Still Work at All
The study also helps explain why noisy quantum circuits can still be adjusted or “trained” for certain tasks. Changing the circuit’s settings can influence the outcome, but mainly because the last layers remain effective while earlier ones lose their impact.
As a result, a deep circuit affected by noise behaves much like a shallower one. Adding more layers does not necessarily improve performance, because most of those extra steps no longer contribute in a meaningful way.
What This Means for Future Quantum Technology
These findings offer a clearer picture of what current and near-term quantum machines can realistically achieve. Simply increasing the number of steps in a circuit is unlikely to unlock new capabilities for many common tasks, especially those involving local measurements.
Instead, progress will depend on reducing noise or designing circuits that can work effectively despite it. The study also highlights a potential pitfall. Noisy circuits may appear trainable, but this is partly because noise has already reduced their overall complexity. Treating noise as a simple, harmless effect could lead to overly optimistic expectations about quantum computing performance.
Reference: “Noise-induced shallow circuits and the absence of barren plateaus” by Antonio Anna Mele, Armando Angrisani, Soumik Ghosh, Sumeet Khatri, Jens Eisert, Daniel Stilck França and Yihui Quek, 2 April 2026, Nature Physics.
DOI: 10.1038/s41567-026-03245-z
Contributors
- Free University of Berlin
- EPFL
- University of Sorbonne
- University of Chicago
- Fraunhofer Heinrich Hertz Institute
- ENS Lyon
- MIT
Never miss a breakthrough: Join the SciTechDaily newsletter.
Follow us on Google and Google News.
2 Comments
Why do we need excessive computing power beyond what we already have? Modern computing power and its use for data analysis already enables us to hunt down and kill designated people remotely and easily. Witness our current pointless wars.
thanks