I am stuck on calculating $\mathcal{E}(\rho)=\text{Tr}_b[{U(\rho\otimes\rho_b)U^{\dagger}}]$. For example, in the case when $U$ is the CNOT matrix $$U=\begin{pmatrix} 1 & 0 & 0 & 0\\\ 0 & 1 & 0 & 0\\\ 0 & 0 & 0 & 1\\\ 0 & 0& 1& 0 \end{pmatrix},$$ and $\rho_b=|0\rangle\langle0|$ as is shown in figure $8.4$ on page 359 in Quantum Computation and Quantum Information by Isaac Chuang and Michael Nielsen.

I calculated the whole matrix multiplication inside the partial trace (assuming that $\rho=|\psi\rangle\langle\psi|$). I end up with a 4-by-4 matrix and I cannot find its partial trace. In the textbook it is said that "it is easily seen that" $$\mathcal{E}(\rho) = |0\rangle\langle0|\rho|0\rangle\langle0| + |1\rangle\langle1|\rho|1\rangle\langle1|.$$

I do not understand how to see this easily and find $\mathcal{E}(\rho)$ in general.

For example, when we have this:

$$\text{Tr}_b[{e^{-iH}(\rho\otimes\rho_b)e^{iH}}],$$ with $\rho_b=|0\rangle\langle0|$ and $H$ is of the Ising type $H=J(X^{(0)}X^{(1)}+Y^{(0)}Y^{(1)})+\alpha(Z^{(0)}+Z^{(1)})$ and where $X^{(j)}$, $Y^{(j)}$, and $Z^{(j)}$ are the Pauli $X$, $Y$, and $Z$ operators on qubit $j$ with $j=0$ being the ancillary qubit for the density operator $\rho_b$.


1 Answer 1


The partial trace of a bipartite state $\sigma_{ab}$ of two qubits $a$ and $b$ is $$ \mathrm{tr}_b(\sigma_{ab}) = \langle 0_b|\sigma_{ab}|0_b\rangle+\langle 1_b|\sigma_{ab}|1_b\rangle.\tag1 $$

Substituting $\sigma_{ab}:=U(\rho\otimes\rho_b)U^\dagger$ and $\rho_b=|0\rangle\langle 0|$ into $(1)$, we obtain

$$ \begin{align} \mathcal{E}(\rho)&=\mathrm{tr}_b(U(\rho\otimes\rho_b)U^\dagger)\\ &= \langle 0_b|U(\rho\otimes|0_b\rangle\langle 0_b|)U^\dagger|0_b\rangle+\langle 1_b|U(\rho\otimes|0_b\rangle\langle 0_b|)U^\dagger|1_b\rangle\\ &= \langle 0_b|U|0_b\rangle\rho\langle 0_b|U^\dagger|0_b\rangle+\langle 1_b|U|0_b\rangle\rho\langle 0_b|U^\dagger|1_b\rangle. \end{align}\tag2 $$

CNOT gate

Now, if $U$ is the CNOT gate, then we calculate

$$ \begin{align} \langle 0_b|U|0_b\rangle=|0_a\rangle\langle 0_a|\\ \langle 1_b|U|0_b\rangle=|1_a\rangle\langle 1_a| \end{align}\tag3 $$

so in this case

$$ \mathcal{E}(\rho) = |0_a\rangle\langle 0_a|\rho|0_a\rangle\langle 0_a|+|1_a\rangle\langle 1_a|\rho|1_a\rangle\langle 1_a|\tag4 $$

as expected.

Ising type Hamiltonian

The above procedure works for general $U$. In particular, when $U=e^{-iH}$ we compute $\langle 0_b|e^{-iH}|0_b\rangle$ and $\langle 1_b|e^{-iH}|0_b\rangle$ and substitute into $(2)$.

  • $\begingroup$ Thank you for your help. Please I need some clarifications on your great answer. First, when you write $\langle0_b|\sigma_{ab}|0_b\rangle$, how do you multiply the matrix $\sigma_{ab}$ with the vector $|0_b\rangle$ when their dimensions do not match? In the CNOT gate case, $\sigma_{ab}$ is of size 4-by-4 but $|0_b\rangle$ is of size 2-by-1. Am I wrong? Second, how did you go from line $(2)$ to the next line? In other words, how did you remove the tensor product in line $(2)$? I would appreciate your help. $\endgroup$
    – zdm
    Commented Feb 7, 2022 at 14:46
  • $\begingroup$ Formally, $\langle i_b|U_a\otimes V_b|j_b\rangle=U_a \langle i_b|V_b|j_b\rangle =v_{ij} U_a$ extended by linearity. Alternatively, you can think of a two-qubit operator, such as CNOT or $\sigma_{ab}$, as tensor with four indices: input and output for qubit $a$ and input and output for qubit $b$. To evaluate $\langle\psi_a|\sigma_{ab}|\phi_a\rangle$, we contract the two indices for qubit $a$ and end up with two-index tensor, i.e. a matrix, acting on qubit $b$. If the kets are computational basis then the whole calculation simplifies to "cutting out" a submatrix of $\sigma_{ab}$. $\endgroup$ Commented Feb 7, 2022 at 17:20
  • $\begingroup$ Thanks again. For $U=e^{iH}$, how do we find it? I have so far $H=J\begin{pmatrix}2 & 0\\\ 0 & 2\end{pmatrix}+\alpha\begin{pmatrix}2 & 0\\\ 0 & -2\end{pmatrix}$. I am missing something here. $Z^{(0)}$ is a 2-by-2 matrix but $H$ should be a 4-by-4 matrix right? $\endgroup$
    – zdm
    Commented Feb 7, 2022 at 20:37
  • $\begingroup$ You are right that we cannot add matrices of different sizes. However, there is a convention wherein identity operators are made implicit, so the $Z^{(0)}$ term in the Hamiltonian most likely means $Z^{(0)}\otimes I^{(1)}$ etc. $\endgroup$ Commented Feb 7, 2022 at 20:45

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.