Short answer: Probabilistic error cancellation is a quantum error-mitigation method that estimates noise-free results from a noisy device, at a cost in extra samples that grows exponentially with the number of corrected gates. A 2022 npj Quantum Information paper uses optimization to lower that overhead, a 2025 Quantum paper reduces it using Pauli error propagation, and a 2025 arXiv preprint applies a statistical variance-reduction technique in simulation. These are methods papers, and the savings reported are for the specific cases studied.
What the method does and costs
The 2022 abstract explains that error-mitigation techniques reduce noise on current hardware without fault-tolerant error correction. The quasiprobability method simulates a noise-free quantum computer with a noisy one, but only produces correct expected values of observables. The cost appears as a sampling overhead that scales exponentially in the number of corrected gates. The 2025 Quantum paper adds that methods such as probabilistic error cancellation, zero-noise extrapolation and Clifford data regression can significantly reduce the error in expectation values, but at an exponentially scaling sampling cost.
2022: choosing the decomposition with optimization
The authors present an algorithm based on mathematical optimization that picks the quasiprobability decomposition in a noise-aware way. They say this directly gives a significantly lower base of the sampling overhead than existing approaches. A key element is a robust quasiprobability method that trades off approximation error against sampling overhead using semidefinite programming.
2025: Pauli error propagation
The 2025 Quantum paper presents a method to reduce probabilistic error cancellation overhead through Pauli error propagation combined with classical preprocessing. The authors say it significantly reduces sampling overheads for Clifford circuits, using the well-defined interaction between the Clifford group and Pauli noise. They also show it applies to non-Clifford circuits, with more limited effectiveness, mainly constrained by the number of non-Clifford gates. They give examples of Clifford sub-circuits that occur in practice, such as resource state generation for measurement-based quantum computing.
2025: control variates
The arXiv preprint notes that quasiprobabilistic decompositions underlie both probabilistic error cancellation and circuit cutting, and that their sampling overheads grow exponentially, limiting practical feasibility. The authors adapt the control variates technique from statistics to reduce overhead, calling the method CV4Quantum. In simulation experiments that mimic a realistic probabilistic error cancellation scenario, they report that in more than 50% of the estimations in their study, the number of samples needed for a given precision fell by more than 50%. They say better control variates could lead to stronger reduction. As a preprint, this has not completed peer review in this form.
Limits
The exponential scaling remains in all three approaches. The reductions are specific to Clifford-heavy circuits in one paper, to simulations in another, and to the optimization setting in the third. Error mitigation gives expectation values, not corrected quantum states.
FAQ
What is the sampling overhead?
The extra measurements needed to get a precise result, which grows exponentially with the number of corrected gates, per the 2022 and 2025 abstracts.
Is error mitigation the same as error correction?
No. The 2022 abstract describes mitigation as reducing noise without fault-tolerant error correction, and only expected values are produced.
Which circuits benefit most from Pauli propagation?
Clifford circuits, according to the 2025 Quantum paper.
What did the control variates study find?
In simulation, more than 50% sample reductions in more than 50% of the estimations studied.
Does any of this remove the exponential cost?
No. All three papers aim to reduce it.
All field notes