I know there are various quantum noise channels, which include the depolarizing channel, the dephasing channel and the bit-flip channel; We can apply them in simulators easily.

However, is there any standard to choose the model we apply?

I mean, why we should (for instance) choose a flip channel over a depolarizing channel and when should we choose a specific channel? Why is the depolarizing channel commonly used?

  • $\begingroup$ I'm not aware of any 'standard' for choosing a channel over another. May depend on your specific scenario. Depolarizing channel is usually chosen as it is sort of the 'worst' that can happen to a travelling qubit. $\endgroup$ Jul 30, 2020 at 2:39
  • $\begingroup$ what do you mean that: 'worst' that can happen to a travelling qubit $\endgroup$ Jul 30, 2020 at 2:41
  • $\begingroup$ Consider what happens in a depolarizing channel. There is a probability of bit-flip, phase-flip and both bit and phase flilp in a depolarizing channel. So it's sort of the worst case scenario. You can find more here: theory.caltech.edu/people/preskill/ph219/chap3_15.pdf $\endgroup$ Jul 30, 2020 at 2:54
  • $\begingroup$ epubs.surrey.ac.uk/849158/1/… in this page also mentioned the worst scenario, but is that neccessary to apply 3/p for X,Y,Z Error? $\endgroup$ Jul 30, 2020 at 3:02
  • $\begingroup$ I mean is that mean: only 3/p for X,Y,Z error called the depolarizing channel(because it told me that in the link i send), if so , why the circumstance when 3/p for X,Y,Z error is the worst case, but not other possibility for X,y,Z error? $\endgroup$ Jul 30, 2020 at 3:04

1 Answer 1


Some thoughts:

A theoretical perspective

From a theoretical perspective, the depolarizing channel is the 'standard' (if there is such a thing) or by some means the most applicable.

Because the Paulis (together with the identity operator) form a basis for $SU(2)$, if a code can correct the $X, Y$ and $Z$ flips on a certain qubit (and it it is able to correctly identify no error (i.e. '$I$-flip') having happened), it can correct all errors on that qubit. A theoretical analysis of a code cares a little less about the relative probabilities of these three flips happening, so we just as well can set them all to $\frac{p}{3}$, thereby obtaining the depolarizing channel.

The depolarizing channel $\Lambda_{\mathrm{depo}}$ can be written as:

$$ \Lambda_{\mathrm{deph}}\left(\begin{bmatrix}a & b \\ b^{*} & d\end{bmatrix}\right) = \begin{bmatrix} (1-\frac{2p}{3})a + \frac{2p}{3}d & (1-\frac{4p}{3}) b \\ (1-\frac{4p}{3})b^{*} & (1-\frac{2p}{3})d + \frac{2p}{3}a\end{bmatrix}. $$ Since $d = 1-a$, we can rewrite this to:

$$ \Lambda_{\mathrm{deph}}\left(\rho_{\mathrm{in}}\right) = (1-\frac{4p}{3})\rho_{\mathrm{in}} + \frac{4p}{3} \frac{I}{2}, $$ which is a convex combination of the input $\rho_{\mathrm{in}}$ and the maximally mixed state $\frac{I}{2}$. Furthermore, if you would equate $p$ to the elapsed time, $p$ would asymptotically go to $\frac{3}{4}$, thereby obtaining the maximally mixed state; therefore the depolarizing channel is in some way the 'worst' noise channel: it destroys both all quantum- (i.e. coherent superpositions) and classical information (there's literally only noise left) in the qubit.

A physical perspective

If you want your simulation to be more true to the physical world, the depolarizing channel is not a very good model, as much as theorists might like it to be. A good first model for noise in qubits is the combination of two channels, the dephasing channel $\Lambda_{\mathrm{deph}}$ and the amplitude damping channel $\Lambda_{\mathrm{amp}}$.

Loosely speaking, the dephasing channel map destroys the coherent phase between the $|0\rangle$ and $|1\rangle$ state:

$$ \Lambda_{\mathrm{deph}}\left(\begin{bmatrix}a & b \\ b^{*} & 1-a\end{bmatrix}\right) = \begin{bmatrix}a & e^{-\frac{t}{T_{2}}} b \\ e^{-\frac{t}{T_{2}}}b^{*} & 1-a\end{bmatrix}, $$

where $T_{2}$ is known as the characteristic qubit dephasing time.

The Kraus operators of $\Lambda_{\mathrm{deph}}$ are $A_{1} = \sqrt{1-p}I$ and $A_{2} = \sqrt{p}Z$, so it is still a Pauli channel, which can help in the analysis or simulation.

$\Lambda_{\mathrm{amp}}$ is a little bit trickier: it simulates the relaxation of the excited (by convention $|1\rangle$) state, and maps it to the $|0\rangle$ state:

$$ \Lambda_{\mathrm{amp}}\left(\begin{bmatrix}a & b \\ b^{*} & 1-a\end{bmatrix}\right) = \begin{bmatrix}a & e^{-\frac{t}{2T_{1}}} b \\ e^{-\frac{t}{2T_{1}}} b^{*} & e^{-\frac{t}{T_{1}}}(1-a)\end{bmatrix}, $$ where $T_{1}$ is known as the qubit relaxation time.

The Kraus operators are also a little trickier: $B_{1} = \begin{bmatrix}1 & 0 \\ 0 & \sqrt{1-p}\end{bmatrix}$ and $B_{2} = \begin{bmatrix}0 & \sqrt{p} \\ 0 & 0\end{bmatrix}$. This means that amplitude damping channel is not a Pauli channel (allthough its Kraus operators can of course be written as linear combinations of the Paulis).

A simulation of a system undergoing both an amplitude damping and dephasing is a good start. There is one big caveat, however: this does not take leakage errors into account whatsoever. Depending on the physical system you are trying to simulate, this might range from either not a very large issue (e.g. for quantum dots) to a gross oversimplification (e.g. for transmon qubits).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.