ConceptIntermediateQuantum computing fundamentalsLesson 15

Quantum Noise

Quantum noise is any deviation of the actual operations from the ideal ones, modelled by quantum channels such as bit flip, phase flip, depolarizing, amplitude damping, and readout errors.

On a real device, gates and measurements are not carried out perfectly. Quantum noise is modelled by quantum channels, which are linear, trace-preserving, completely positive maps, written in the Kraus representation: $$\mathcal{E}(\rho)=\sum_k K_k\rho K_k^\dagger,\qquad \sum_kK_k^\dagger K_k=I.$$

The most important single-qubit models: the bit flip applies $X$ with probability $p$; the phase flip applies $Z$ with probability $p$, weakening superposition without changing the $Z$ probabilities; the depolarising channel $\mathcal{E}(\rho)=(1-p)\rho+p\,I/2$ shrinks the whole Bloch vector by the factor $1-p$; and amplitude damping represents $T_1$ relaxation towards $|0\rangle$. The $T_1$ and $T_2$ channels result from decoherence with the environment.

Types of error must be distinguished. Incoherent errors are random and turn the state into a mixture. Coherent errors are systematic over-rotations, such as $R_X(\theta+\epsilon)$ instead of $R_X(\theta)$; they keep the state pure but can accumulate with an amplitude that grows linearly with the number of repetitions. Readout errors make the device record 1 while the qubit is in 0, or the reverse; we denote them $p(1|0)$ and $p(0|1)$.

There are two ways to deal with noise. Error mitigation corrects the results statistically without protecting the state, for example by inverting the readout error matrix, usually at the cost of more shots. Quantum error correction encodes a logical qubit in many physical qubits; the threshold theorem states that an arbitrarily long computation with arbitrary precision is possible if the error rate per operation is below a certain threshold, at a polylogarithmic overhead.

Noise limits circuit depth in the NISQ era, and it can flatten the cost landscape of variational algorithms. The ideal simulator contains no noise; the optional noise models provided by the lab are a teaching approximation, not a description of any particular device.

To estimate the effect of noise on a whole circuit, a simple approximation is used: if each gate succeeds with probability $1-\epsilon$ and there are $G$ independent gates, the probability that no error occurs is about $(1-\epsilon)^G$; with $\epsilon=10^{-3}$ and $G=1000$ it is only about $0.37$. This shows why long computations require error correction.

Mathematical formula

$$\mathcal{E}(\rho)=\sum_k K_k\rho K_k^{\dagger},\quad \sum_k K_k^{\dagger}K_k=I$$

Worked example

  1. Readout error: the state $|+\rangle$ theoretically gives $P(0)=0.5$. If $p(1|0)=0.02$ and $p(0|1)=0.05$, the observed probability is $P_{\text{obs}}(0)=0.5(1-0.02)+0.5(0.05)=0.49+0.025=0.515$.
  1. Depolarising with $p=0.1$ on $|0\rangle$: the Bloch vector $(0,0,1)$ becomes $(0,0,0.9)$, so $P(0)=(1+0.9)/2=0.95$.
  1. Phase flip with $p=0.1$ on $|+\rangle$: $r_x=1\to1-2p=0.8$, so $P(+)=0.9$, while $P(0)$ in the $Z$ basis stays at 0.5.

Test your understanding

The quiz needs JavaScript.

Related topics

References

  1. Michael A. Nielsen; Isaac L. Chuang (2010). Quantum Computation and Quantum Information (10th Anniversary Edition). Cambridge University Press doi:10.1017/CBO9780511976667The standard reference for quantum computing; the Crossref record gives 2012 for the Cambridge Core online edition.
  2. John Preskill (2018). Quantum Computing in the NISQ era and beyond. Quantum 2:79 doi:10.22331/q-2018-08-06-79
  3. Qiskit contributors (2025). Qiskit Aer documentation. GitHub Pages qiskit.github.io