I was looking at the paper : https://arxiv.org/abs/2002.11649 and the eigenvalue discussion is not clear to me.

Block-encoding is a general technique to encode a nonunitary matrix on a quantum computer. Let $A \in \mathbb{C}^{N \times N}$ be an $n$-qubit Hermitian matrix. If we can find an $(m+$ $n)$-qubit unitary matrix $U \in \mathbb{C}^{M N \times M N}$ such that $$ U_{A}=\left(\begin{array}{cc} A & \cdot \\ \cdot & \cdot \end{array}\right) $$ holds, i.e., $A$ is the upper-left matrix block of $U_{A}$, then we may get access to $A$ via the unitary matrix $U_{A}$. In particular, $$ A=\left(\left\langle 0^{m}\right| \otimes I_{n}\right) U_{A}\left(\left|0^{m}\right\rangle \otimes I_{n}\right) $$

If $U_{A}$ is Hermitian, it is called a Hermitian block-encoding. In particular, all the eigenvalues of a Hermitian block-encoding $U_{A}$ are $\pm 1$.

It is not clear why all the eigenvalues are $\pm 1$? any guidance? Thanks


1 Answer 1


$U_A$ is defined to be a unitary matrix in the paper and your question. Consider the eigenvalue $\lambda$ of the general unitary matrix $U$ given by $U|\lambda\rangle=\lambda|\lambda\rangle$.

$$U|\lambda\rangle=\lambda|\lambda\rangle$$ $$\langle\lambda| U^{\dagger}=\lambda^*\langle\lambda|$$ $$\langle\lambda| U^{\dagger}U|\lambda\rangle=\lambda\lambda^*\langle\lambda|\lambda\rangle$$ $$\rightarrow |\lambda|^2=1$$

In the paper it also says "If $U_A$ is Hermitian then it is called Hermitian Block Encoding" on the first line of page three. Now obviously we know that the eigenvalues of a Hermitian Matrix are real.

Therefore we know the eigenvalues $\lambda$ of the matrix $U_A$ are real and have magnitude 1. There are only 2 such numbers; $+1$ and $-1$.

  • $\begingroup$ @Betrand Einstein IV , Ah. Thanks. Somehow the simple unitary was slipped out of my mind and was looking at the rightway. Thanks! $\endgroup$ Sep 23, 2021 at 17:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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