This is a sequel of this question, and assumes similar notation.
In the previous question I essentially asked if the process of entanglement is just a formal idealization, in which there's some sort of sudden change
$(\sum_i a_i \psi_i,\varphi_0) \mapsto \sum_i a_i \psi_i\otimes\varphi_i$
or if it's realized by unitary evolution. The answer is that it is realized gradually by unitary evolution $U(t)((\sum_ia_i\psi_i)\otimes\varphi_0)$, occurring at all times in the Hilbert space $\mathcal H =\mathcal{H}_A\otimes\mathcal{H}_B$ of the joint system, and at a given time $t=T$ we finally have:
$U(T):\;\sum_ia_i\psi_i\otimes\varphi_0 \mapsto \sum_ia_i\psi_i\otimes\varphi_i$
where $U(T)$ is the time evolution at time $t=T$ corresponding to a Hamiltonian $\hat H\in\mathcal L(\mathcal H)$. Also, the resulting time-evolved states $U(t)((\sum_ia_i\psi_i)\otimes\varphi_0)$ will be non-entangled if and only if the Hamiltonian is of the form $\hat H = \hat H_A\otimes 1 + 1\otimes \hat H_B$.
Now I would like to have a concrete example of this unitary entanglement process.
More specifically, I would like to see a toy model comprising of the following:
- Hilbert spaces $\mathcal H_A$, $\mathcal H_B$, and $\mathcal H =\mathcal H_A\otimes\mathcal H_B$,
- distinct states $\psi_i\in \mathcal H_A$, $\varphi_0$ and $\varphi_i$ in $\mathcal H_B$, and complex numbers $a_i$,
- an explicit Hamiltonian $\hat H\in \mathcal L(\mathcal H)$, of the form $\hat H = \hat H_A\otimes 1 + 1\otimes \hat H_B + H_{\mathrm{int}}\;,$
such that, for a time $T>0$, $U(T)((\sum_ia_i\psi_i)\otimes\varphi_0)=\sum_ia_i\psi_i\otimes\varphi_i$ where $U(t)$ is as above.
If possible, I would also like to have $\langle\psi_i\mid\psi_j\rangle=0$ for $i\neq j$, and $\langle\varphi_i\mid\varphi_j\rangle=0$ for $i\neq j$.
Also: is it possible to do the above literally, or some sort of approximation is needed? (for example the initial state not being literally separable but only approximately so?). Is it possible with the Hamiltonian being a bounded operator (so not having to use rigged Hilbert spaces, generalized eigenvectors and the like)?