I am currently reading the papers about the QEC (quantum error-correction) experiments of 2023 and while they are impressive (!), I would like to better understand what are their limitations. My goal is to better grasp the state of the art of the experimental side in QEC.

I am in particular focusing, the Google's one, and the Harvard's one (Bluvstein's paper).

Google experiment:

In Google's paper, they only looked at a quantum memory. They showed that for sufficiently low physical error-rates, increasing code distance of the surface code from $d=3$ to $d=5$ reduced the logical error-rate of the quantum memory.

[A] Am I correct to understand that while this effect occured, the logical error-rate of the quantum memory was still higher than the physical error rates of the gates used.

For instance in Fig 3 of the main text, their error-rate can get close to $1-0.97=3 \%$. Yet, the typical error-rate for their physical gates is always below $3 \%$ (see for instance S4 c) for the error-rate of CZ gates in the supplemental) (and an identity gate, which is the good gate to be compared to, for a physical clock cycle would likely have much lower error-rate than that).

Harvard experiment:

Among other things, in Harvard's experiment they create a logical Bell state between two logical qubit (by implementing a cNOT on an initial state $|+\rangle |0\rangle$. They measure the Bell stabilizers $X_1 X_2$ and $Z_1 Z_2$ and show that when the code distance increases from $d=3$ to $d=5$, the error-rate decreases.

My understanding is that there are two limitations:

[B] Like in Google I am (not sure?) that their logical error rate is lower than the physical ones

[C] They only do a single round of syndrome extraction, which from my understanding of the surface code is not sufficient to guarantee fault-tolerance if a sequence of logical gate is implemented in the algorithm (they would have need to show this improvement even if $d$ rounds of QEC are performed).

My questions in summary:

  1. Do you agree with the limitations I pointed out (specifically, my points [A], [B], [C])
  2. In Harvard paper they insist on the fact they do a "projective measurement" to measure the stabilizers. What do they exactly mean by that? Is it that they measure all the physical qubits in $X$ or $Z$ basis (hence they destroy the logical state).
  3. Are there any other important limitation in these experiments?

1 Answer 1

  1. Yes.
  2. Sounds right. Assembling stabilizers by destructively measuring the state is typically less noisy. They're certainly working on extending to multiple rounds of syndrome extraction. Probably first with a reservoir of ancilla qubits, and then later with continuous reloading.
  3. Of course. For neutral atoms, they're running out of atoms as the experiment goes on. For superconductors, real-time decoding, ensuring that leakage isn't growing, fixing cosmic rays, etc.

Both really good papers getting signatures of QEC working though. Totally personally, I think the rep code stuff down to $10^{-7}$ error rates is also very compelling.

  • $\begingroup$ Thank you very much, very useful. For point 2: do you know where it is explicitly said in Harvard's paper (i.e. the place where they explain that for them "projective measurement" means measuring all the physical data qubits in $X$ or $Z$ basis)? I did not find the place. $\endgroup$ Commented Jan 16 at 9:57

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.