2
$\begingroup$

Background

Transversal gates are considered to be fault-tolerant logical operation as they won't lead to the propagation of local errors. Here are some examples:

  1. CSS code has transversal Pauli and CNOTs gates. Particularly, self-dual CSS code has transversal Clifford gates.

  2. $[[5, 1, 3]]$ has transversal Pauli and Phase-Hadamard $SH$ gate.

Motivation

I am recently interested in QEC codes with non-Clifford transversal gates. Particularly, I find the family of quantum Reed-Muller code that allows the transversal Pauli rotation $Z(\frac{\pi}{2^n})$ gate. Although the allowed rotation angle is discrete, I may argue to myself that it's approximately continuous and for any $Z(\theta)$ gate, we can find a code which allows it to be transversal gate.

As Pauli bases are symmetric, I believe this intuition also holds for $X(\theta)$ and $Y(\theta)$.

Problem

Pauli rotation gates are very special and I wanna generalize this intuition to arbitrary single-qubit unitary $U(\theta)$: For arbitrary single-qubit unitary gate, does there exist a code such that the given gate is transversal?

Thought by far

A common practice is to decompose $U(\theta)$ into

$$ U(\theta) = X(\beta_0)Z(\beta_1)X(\beta_2). $$

Maybe we should find out a code both allows small-angle transversal X and Z rotations? But I highly doubt the presence of such code.

$\endgroup$
1
  • 1
    $\begingroup$ Self-dual CSS codes do not necessarily implement all logical Clifford gates transversally, they are only guaranteed to have transversal Paulis (because stabilizer) transversal CNOT (because CSS) and transversal $ H $ (because self-dual). You need to add the assumption that the Self-dual CSS code is also doubly-even to guarantee transversal phase gate and thus all of transversal Clifford. See quantumcomputing.stackexchange.com/questions/26701/… for an example of a $ [[25,1,5]] $ self-dual CSS code that is not doubly even. $\endgroup$ Mar 4 at 13:41

1 Answer 1

2
$\begingroup$

You might be interested in the following paper, which is part of a sequence that helps to classify the possible forms of transversal gates: https://arxiv.org/abs/1409.8320 (up to some local equivalence). Pretty much, it's the type of diagonal gates you've mentioned + Clifford transformations.

On the other hand, I might cheat a little bit (OK, quite a lot!). Imagine you want to implement a $U=VZ(\theta)V^\dagger$ where $Z(\theta)$ is one of the cases for which you already have a code that implements it. Let $|\tilde 0\rangle_L$ and $|\tilde 1\rangle_L$ be the logical states of this code. Now, note that the choice of basis within a subspace is arbitrary. So, I can just happen to choose to define my logicals instead as $|0\rangle_L=V^\dagger|\tilde 0\rangle_L$. Now if I apply what was $Z(\theta)$, its effective action on my new basis is just $U$. So you're done! (The cheat here is that you no longer have nice descriptions of logical X and Z, which you probably, implicitly, want to keep.)

$\endgroup$
3
  • $\begingroup$ Thanks for mentioning this paper! Indeed, I also came up with the same "cheating" idea. However, I realized the gate $V$ is generally not simultaneously transversal. In another word, it just transferred the non-transversality to another part and I am not really satisfied with this solution. $\endgroup$
    – Yunzhe
    Dec 6, 2023 at 9:18
  • $\begingroup$ I agree, but since you didn't specify that in the question... ;) $\endgroup$
    – DaftWullie
    Dec 6, 2023 at 10:48
  • $\begingroup$ In many ways, for me, I find it desirable to fix having a transversal cNOT (since that's the gate that's the biggest contributor t a fault tolerant threshold, so it needs to be as simple as possible). Once you impose that, you're essentially restricted to CSS codes, and the consequences that entails... $\endgroup$
    – DaftWullie
    Dec 6, 2023 at 13:51

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.