I'm currently calculating the dynamics of the density operator for a 3-level system. To do this, I've set up a Hamiltonian as:

\begin{align} H = \left( \begin{array}{ccc} \hbar \delta _q & \hbar \tilde{\Omega }_q & 0 \\ \hbar \Omega _q & 0 & \hbar \Omega _d \\ 0 & \hbar \tilde{\Omega }_d & \hbar \delta _d \\ \end{array} \right) \end{align} For each of the spontaneous decay channels, the lindblad form of the decay is: \begin{align} D_\sigma[\rho]= (\sigma \rho \sigma^{\dagger} - \frac{1}{2}(\sigma^\dagger \sigma \rho + \rho \sigma^\dagger \sigma) \end{align}

Then overall, the time evolution can be cast into the following form:

\begin{align} \dot{\rho}= \frac{i}{\hbar}[\rho, H] + \Gamma_{12} D_{\sigma_{12}}[\rho] + \Gamma_{13} D_{\sigma_{13}}[\rho] + \Gamma_{23} D_{\sigma_{23}}[\rho] = \mathcal{L}[\rho] \end{align} With $\mathcal{L}$ being a linear map acting on the operator $\rho$.

I now try to find the steady state solution of the system, by simply finding the operator $\rho$ that $\mathcal{L}$ computes to 0 - If you want, the kernel of the system. I do so by using a computer algebra system, and perform the calculation symbolically. Now, when I find the 0-space, what usually happens is that the solutions I get have complex diagonal elements, which means that my solution is not hermitian.

Of course, solutions of eigenvalue equations are just solutions up to a complex constant, but by that alone I am not able to cure the non-hermicity in this case.

Which makes me wonder: Should I even expect to get a solution that qualifies as a density matrix? (Self-Adjoint, and trace 1), by finding the kernel of the Liouvillian? I know that the Liouvillian is trace and hermiticity preserving, but does that also play a role in my calculation here, where I didn't pick a density-operator $\rho(0)$ for a start?

  • $\begingroup$ The identity is always in the kernel which, up to renormalization, qualifies as a valid density matrix. $\endgroup$
    – LPZ
    Commented May 6, 2023 at 17:59
  • $\begingroup$ @LPZ this however doesn't tell me wether every linear independent element of the Kernel does, up to renormalization, qualify as a valdi density matrix. $\endgroup$ Commented May 6, 2023 at 19:26
  • $\begingroup$ As you’ve said yourself, there are elements of the kernel that are non valid density matrices even after renormalisation. If I understand correctly, your original question is whether there are density matrices in the kernel. The answer is yes just by this simple example. $\endgroup$
    – LPZ
    Commented May 6, 2023 at 22:29
  • $\begingroup$ @LPZ The identity matrix is not always in the kernel. It is always a left eigenvector, that is, an eigenvector of $\mathcal L^\dagger$, but only a right eigenvector if the channel is unital. $\endgroup$
    – Noiralef
    Commented May 7, 2023 at 0:31
  • $\begingroup$ @Noiralef aren't eigenvectors of $\mathcal{L}$ what I'm interested in, in that case? If I shove all this super-operator terminology asside, it's just a linear map that can have a kernel - no? So why should I be interested in $\mathcal{L}^{\dagger}$? $\endgroup$ Commented May 7, 2023 at 11:12

1 Answer 1


I pondered a bit, and I think I have the answer to the question:

If $\rho$ is part of the Kernel, that is

\begin{align} \mathcal{L}[\rho]= 0 \end{align}

Then \begin{align} \mathcal{L}[\rho]^{\dagger}=0 \\ =(\frac{i}{\hbar}[\rho, H])^{\dagger} - \sum_i \Gamma_i ( \sigma_i \rho \sigma_i^{\dagger} - \frac{1}{2} ( \rho \sigma_i^{\dagger} \sigma + \sigma_i^{\dagger} \sigma \rho))^{\dagger} \\ =(\frac{i}{\hbar}[\rho^{\dagger}, H]) - \sum_i \Gamma_i ( \sigma_i \rho^{\dagger} \sigma_i^{\dagger} - \frac{1}{2} (\sigma_i^{\dagger} \sigma \rho^{\dagger} + \rho^{\dagger} \sigma_i^{\dagger} \sigma)) \\ \end{align} because $H$ is hermitean. Then \begin{align} = \mathcal{L}[\rho^{\dagger}]= 0 \end{align} That means, when $\rho$ is in the kernel, then $\rho^{\dagger}$ is as well, so for every element $\rho$ of the kernel, $\rho + \rho^{\dagger}$ is as well in the kernel, and qualifies (up to normalization of the trace} as a density operator.

  • $\begingroup$ Actually, you’d need to also check positivity of $\rho$ for it to be a density matrix (or else you could have negative probabilities). This is where knowing that the identity is in the kernel comes in handy. By convex combination, you can construct a valid density matrix. $\endgroup$
    – LPZ
    Commented May 6, 2023 at 22:27

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.