14
$\begingroup$

The quantum harmonic oscillator relies on two classical objects, the so-called creation and annihilation operator

$$a ^* = x- \partial_x \text{ and }a = x+\partial_x.$$

Fix two numbers $\alpha,\beta \in \mathbb R.$

Can we explicitly compute the spectrum of

$$H = \begin{pmatrix} aa^* + \alpha^2 & \alpha a+ \beta a^* \\ \alpha a^* + \beta a & aa^* +\beta^2 \end{pmatrix}?$$

This looks like an innocent problem, but somehow I find it hard to determine what exactly the spectrum of this operator is.

The operator is acting on a suitable domain of $L^2(\mathbb R) \oplus L^2(\mathbb R).$

$\endgroup$
5
  • 1
    $\begingroup$ What space is $H$ operating on? The direct sum of $L^2$ spaces? $\endgroup$ Commented Aug 15, 2021 at 3:38
  • $\begingroup$ Is not there a solution with any eigenvalue $\lambda$? Goes like ($p_0+p_1x+((1+\alpha^2-\lambda)p_0+(\alpha-\beta)q_1)\frac{x^2}2+...$, $q_0+q_1x+((1+\beta^2-\lambda)q_0+(\beta-\alpha)p_1)\frac{x^2}2+...$) $\endgroup$ Commented Aug 15, 2021 at 6:48
  • 1
    $\begingroup$ @მამუკაჯიბლაძე sorry, can you elaborate? Is your solution in $L^2?$ $\endgroup$
    – Kung Yao
    Commented Aug 15, 2021 at 11:46
  • $\begingroup$ Sorry there was no condition at the time of that comment. The solution has form $\sum_{n\geqslant0}P_n\frac{x^n}{n!}$, $\sum_{n\geqslant0}Q_n\frac{x^n}{n!}$ with $P_n$, $Q_n$ polynomials with integer coefficients of degree $n$ in $\alpha$, $\beta$, of degree $\left[\frac n2\right]$ in $\lambda$ and linear in $p_0$, $q_0$, $p_1$, $q_1$. Unfortunately I do not know how to estimate those integer coefficients but judging by calculations they do not grow quicker than $n!$ so that solution must be nice enough for small $\alpha$ and $\beta$. I'm afraid that's all I can say $\endgroup$ Commented Aug 15, 2021 at 12:07
  • $\begingroup$ this Hamiltonian has a rather special form (in particular it has the special property that the ground state is $\alpha,\beta$ independent), can you inform us on the context in which it appeared? $\endgroup$ Commented Aug 20, 2021 at 12:00

1 Answer 1

14
+500
$\begingroup$

The Hamiltonian $$H=\begin{pmatrix} \alpha^2+a^\ast a&\alpha a+\beta a^\ast\\ \alpha a^\ast+\beta a&\beta^2+a^\ast a \end{pmatrix} $$ is known in the physics literature as the anisotropic Rabi Hamiltonian. (In the most general case there is an additional term $\Delta\sigma_z$.) I give some pointers to the literature in this Physics SE posting. The eigenvalues can be computed from a recursive scheme, but closed-form expressions for the spectrum only exist for either $\alpha=\beta$ or $\alpha\beta=0$.

$\bullet$ Consider first the case $\alpha=\beta$. A unitary transformation $H'=UHU^\ast$ with $U=e^{i\pi\sigma_y/4}$ brings the Hamiltonian to the diagonal form $$H'=\begin{pmatrix} b_+^\ast b_+&0\\ 0&b_-^\ast b_- \end{pmatrix},\;\; b_\pm=a\pm\alpha. $$ The eigenvalues are the integers $N=0,1,2,\ldots$, each twofold degenerate. The corresponding eigenstates $|N,\pm\rangle$ are obtained from the eigenstates $|N\rangle$ of the harmonic oscillator by acting on these with the displacement operator, $$| N,\pm\rangle=e^{\pm\alpha(a-a^\ast)}|N\rangle.$$

$\bullet$ At the other extreme, we can take one of the two parameters $\alpha, \beta$ much smaller than the other. Let me set $\beta=0$, $\alpha\neq 0$. (The spectrum is the same for $\alpha=0$, $\beta\neq 0$.) The Hamiltonian $$H=\begin{pmatrix} \alpha^2+a^\ast a&\alpha a\\ \alpha a^\ast&a^\ast a \end{pmatrix}=(\tfrac{1}{2}\alpha^2+a^\ast a)I+\alpha(\sigma_+a+\sigma_-a^\ast)+\tfrac{1}{2}\alpha^2\sigma_z$$ is known in physics as the Jaynes-Cummings Hamiltonian.

The eigenvalues can be computed exactly, because the Hamiltonian decomposes into an infinite direct product $H_n$ of $2\times 2$-matrix Hamiltonians in the basis $|g,n+1\rangle$ (two-level system in the lower state, $n+1$ quanta excited in the oscillator) and $|e,n\rangle$ (two-level system in upper state, $n$ quanta excited):

$$H_n\begin{pmatrix} |g,n+1\rangle\\ |e,n\rangle \end{pmatrix}= \begin{pmatrix}n+1&\alpha\sqrt{n+1}\\ \alpha\sqrt{n+1}&n+\alpha^2 \end{pmatrix} \begin{pmatrix} |g,n+1\rangle\\ |e,n\rangle \end{pmatrix}.$$ The eigenvalues then follow directly, $$\Omega_{n,\pm}=\tfrac{1}{2}(\alpha^2+1)+n\pm\tfrac{1}{2}\sqrt{(\alpha^2+1)^2+4\alpha^2 n}.$$

This sequence of eigenvalue pairs $\Omega_{n,+},\Omega_{n,-}$ exists for $n\in\{0,1,2,\ldots\}$. In addition, there is an unpaired ground state $|g,0\rangle$ without any excitations, which is annihilated by $H$ so it has eigenvalue $\Omega_{0}=0$ independent of $\alpha$.

$\bullet$ The general case of arbitrary $\alpha,\beta$ does not have a simple closed form expression for the eigenvalues. For $\alpha,\beta\ll 1$ a perturbative solution is given in Appendix C of arXiv:1008.1317.

The ground state does have a simple exact result: the lowest eigenvalue equals zero for any $\alpha,\beta\in\mathbb{R}$. The zero-mode – the eigenstate which is annihilated by $H$ – is a coherent state $|\xi\rangle=e^{\xi(a^\ast-a)}|0\rangle$ for the harmonic oscillator (such that $a|\xi\rangle=\xi|\xi\rangle$). Substitution of $H{p\choose q}\otimes|\xi\rangle=0$ gives the two zero-modes $$\Psi_\pm={{\sqrt\beta}\choose{\mp\sqrt\alpha}}\otimes|\!\pm\!\!\sqrt{\alpha\beta}\rangle.$$ (Thanks to Michal Pacholski for helping me with this.)


Notation: $I$ is the $2\times 2$ unit matrix, $\sigma_x$, $\sigma_y$, $\sigma_z$ are the three Pauli matrices, and $\sigma_\pm=\tfrac{1}{2}(\sigma_x\pm i\sigma_y)$. Also please note that the eigenvalues in the OP are shifted by one unit relative to those given here, because I reordered $aa^\ast\mapsto a^\ast a$.


Since there was an issue with the appearance of spurious "parabolas" in the comparison with the numerics by yarchik (in a now deleted answer), I show here a numerical check of the spectrum as a function of $\alpha$ for $\beta=0$, using a slightly modified version of yarchik's code.

Solid curves are the numerical result, dashed and dotted curve are $\Omega_{n,\pm}$. Note that the ground state is at zero energy for any $\alpha$ when $\beta=0$. The plot below, for $\beta=2\alpha$ shows this also holds for nonzero $\beta$.

$\endgroup$
4
  • 2
    $\begingroup$ Hmm, why can we be sure that the operator has only eigenvalues and no further spectrum? $\endgroup$ Commented Aug 17, 2021 at 14:55
  • 2
    $\begingroup$ there is obviously no continuous spectrum for $\alpha=0$ (two-level system decoupled from the harmonic oscillator); switching on $\alpha$ smoothly perturbs the eigenvalues. $\endgroup$ Commented Aug 17, 2021 at 15:18
  • 1
    $\begingroup$ 20 edits in one week are not few. -- If there is a reason for that number of edits, maybe this would be more obvious if you add some more informative edit summaries than just "added X characters in body"(?) $\endgroup$
    – Stefan Kohl
    Commented Aug 24, 2021 at 9:29
  • 3
    $\begingroup$ Hi Stefan, I spent way too much time on this problem, but the edits are not cosmetic, rather they document progress in an answer to the question posed by the OP. Today's edits give the eigenstate for the ground state solution, which is not in the literature, so I figured it would be worthwhile to document it in case anyone is interested in this Hamiltonian. (And yes, I did feel some obligation to repay the large bounty with as complete an answer as I could provide.) Will be more careful with the "edit summary", good point, thanks for reminding me. $\endgroup$ Commented Aug 24, 2021 at 11:37

You must log in to answer this question.