Short answer: Dynamical decoupling suppresses noise on a qubit by applying a sequence of pulses that together act like an identity operation. A 2010 Physical Review Letters paper presents a scheme that the authors say eliminates general decoherence of a qubit to order n using O(n^2) pulses, and a 2021 ACM paper reports that switching it on for every idle qubit on IBM machines does not necessarily help, proposing a software framework that applies it selectively. These are a theory-and-simulation result and a hardware study.
What dynamical decoupling is
The 2021 abstract describes idling errors, which occur when a qubit is idle and not undergoing any operation. To mitigate them, earlier work proposed dynamical decoupling, which reduces stray noise on idle qubits by continuously executing a specific sequence of single-qubit operations that effectively behave as an identity gate. The authors note that existing protocols had mostly been studied for individual qubits, and that their effect at the application level was not yet fully understood.
2010: fewer pulses
The Physical Review Letters abstract, by West, Fong and Lidar, presents what the authors call a near-optimal quantum dynamical decoupling scheme. They say it eliminates general decoherence of a qubit to order n using O(n^2) pulses, which they describe as an exponential decrease in pulses over all previous decoupling methods. Numerical simulations of a qubit coupled to a spin bath demonstrate, in their words, the superior performance of the new pulse sequences. This is a simulation result for a single qubit.
2021: it does not always help
The ACM paper reports experiments showing that naively enabling dynamical decoupling for every idle qubit does not necessarily improve fidelity. It reduces idling error rates for some qubits but raises the overall error rate for others, because the protocol adds operations of its own. The authors also find that idling errors are program-specific, so the set of qubits that benefit changes with each program.
To handle this they propose Adaptive Dynamical Decoupling, or ADAPT, a software framework that estimates how effective the technique is for each qubit combination and applies it only to the subset that gains most. ADAPT uses a decoy circuit, structurally similar to the original program but with a known solution, to pick the sequence that maximizes fidelity. A localized algorithm keeps the search linear in the number of qubits instead of exponential. On IBM quantum machines with 16 to 27 qubits, the authors report that ADAPT improves application fidelity by 1.86 times on average and up to 5.73 times compared with no decoupling, and by 1.2 times compared with decoupling all qubits. These are the authors' own results on the machines they used.
A 2025 paper
A PRX Quantum paper, volume 6, article 010332, is titled "Resource-Efficient Context-Aware Dynamical Decoupling Embedding for Arbitrary Large-Scale Quantum Algorithms". Only the title was readable when this post was written, so its results are not described.
Limits
The two main results differ in kind. The first is a pulse-count result checked in simulation for one qubit, and the second is an empirical study on particular IBM devices with 16 to 27 qubits. The fidelity gains depend on those devices and programs, and neither abstract claims the same gains on other hardware.
FAQ
What is an idling error?
An error that occurs while a qubit is idle and not undergoing any operation, per the 2021 abstract.
What does dynamical decoupling do?
It runs a sequence of single-qubit operations that effectively behave as an identity gate, reducing stray noise on idle qubits.
Can it make things worse?
The 2021 experiments found it can raise error rates for some qubits because of the added operations.
How much did ADAPT improve fidelity?
By 1.86 times on average and up to 5.73 times against no decoupling, as reported by the authors on IBM machines.
What did the 2010 paper claim?
A scheme eliminating decoherence to order n with O(n^2) pulses, shown in numerical simulation.
All field notes