---
jupytext:
  text_representation:
    extension: .md
    format_name: myst
    format_version: 0.13
    jupytext_version: 1.11.1
kernelspec:
  display_name: Python 3
  language: python
  name: python3
---

# When should I use CDR?

## Advantages

The main advantage of CDR is that it can be applied without knowing the specific details of the noise model.
Indeed, in CDR, the effects of noise are indirectly _learned_ through the execution of an appropriate set of test circuits.
In this way, the final error mitigation inference tends to be tuned to the used backend.

This self-tuning property is even stronger in the case of _variable-noise-CDR_, i.e., when using the `scale_factors` option in {func}`.execute_with_cdr`.
In this case, the final error mitigated expectation value is obtained as a linear combination of noise-scaled expectation values.
This is similar to [Zero-Noise Extrapolation](zne-5-theory.md) but, in CDR, the coefficients of the linear combination are learned instead of being fixed by the extrapolation model.

## Disadvantages

The main disadvantage of CDR is that the learning process is performed on a suite of test circuits which only _resemble_ the original circuit of interest.
Indeed, test circuits are _near-Clifford approximations_ of the original one.
Only when the approximation is justified, the application of CDR can produce meaningful results.
Increasing the `fraction_non_clifford` option in {func}`.execute_with_cdr` can alleviate this problem to some extent.
Note that, the larger `fraction_non_clifford` is, the larger the classical computation overhead is.

Another relevant aspect to consider is that, to apply CDR in a scalable way, a valid near-Clifford simulator is necessary.
Note that the computation cost of a valid near-Clifford simulator should scale with the number of non-Clifford gates, independently from the circuit depth.
Only in this case, the learning phase of CDR can be applied efficiently.