2
$\begingroup$

In his 1963 paper, in which he introduces his formulation of the Glauber-Sudarshan $P$-representation (https://doi.org/10.1103/PhysRev.131.2766), Glauber refers to the convolution of the $P$-representations of two coherent states, $|\alpha_1\rangle$ and $|\alpha_2\rangle$, as giving rise to a "superposed" state, $|\alpha_1+\alpha_2\rangle$. Clearly, this is very different from superposition in the sense of a cat state, which would be $(|\alpha_1\rangle+|\alpha_2\rangle)/\sqrt 2$.

More generally, we can take the convolution of an arbitrary $P$-representation with the $P$-representation of some classical state and obtain the $P$-representation of a new and valid quantum state. In https://doi.org/10.1016/S0375-9601(97)00272-7, the authors discuss "superposition" of an arbitrary state with a thermal state. They say that in the case that one state is thermal, it can be interpreted as the limit of mixing the arbitrary state with a thermal state in a succession of weak beamsplitters.

Firstly, I want to understand what superposition means in this context, which seems very different from the superposition of quantum states, and how it relates to superposition of quantum states.

Secondly, what is the physical intepretation of this convolution in the general case? If I have P-representations $P_1$ and $P_2$, corresponding to quantum states $\rho_1$ and $\rho_2$ respectively, and I calculate the convolution of $P_1$ and $P_2$, $P=P_1\star P_2$, and its corresponding density matrix $\rho$, is there any relationship between $\rho$, $\rho_1$, and $\rho_2$?

$\endgroup$

2 Answers 2

1
$\begingroup$

Not a full answer, but some general considerations that should lead to one when followed through.

Denote with $\chi_\rho(\alpha,1)\equiv \operatorname{tr}(D_1(\alpha)\rho)$ the normally-ordered characteristic function of a state $\rho$, with $D_1(\alpha)\equiv e^{\alpha a^\dagger}e^{-\bar\alpha a}$. The $P$ function can be written as the Fourier transform of this characteristic function: $$P_\rho(\nu) = \int\frac{\mathrm d^2\alpha}{\pi} e^{\nu\bar\alpha-\bar\nu\alpha} \chi_\rho(\alpha,1) \equiv \mathcal F[\chi_\rho(\bullet,1)](\nu),$$ where I used $\mathcal F$ as shorthand for the Fourier transform. Note that with the convention above this operator satisfies $\mathcal F^2=\mathcal F^{-1}=\operatorname{Id}$.

The above generalises to other $s$-ordered quasiprobability distributions, such as Wigner and $Q$, by writing $$W_\rho(\nu,s) = \mathcal F[\chi_\rho(\bullet,s)](\nu),$$ with $P_\rho(\nu)\equiv W_\rho(\nu,1)$, and $\chi_\rho(\alpha,s)\equiv e^{\frac s2|\alpha|^2}\operatorname{tr}(e^{\alpha a^\dagger-\bar\alpha a} \rho)$.

The standard convolution properties of the Fourier transform then tell us that $$W_\rho(\bullet,s)\star W_\sigma(\bullet,s) = \mathcal F[\chi_\rho(\bullet,s)\chi_\sigma(\bullet,s)].$$ In words, the convolution of two quasiprobabilities is the Fourier transform of the pointwise product of the characteristic functions (in any fixed ordering $s$). In particular, $$P_\rho\star P_\sigma = \mathcal F[\chi_\rho(\bullet,1)\chi_\sigma(\bullet,1)].$$ So the question is equivalent to: what is the state (if any exists) whose characteristic function is the pointwise product of the characteristic functions of $\rho$ and $\sigma$?

Example of coherent states

In the case of coherent states, this is easy enough to compute: for coherent states $|\beta\rangle$ and $|\gamma\rangle$ we have $$\chi_\beta(\alpha,1) = e^{\alpha\bar\beta-\bar\alpha\beta}, \qquad \chi_\gamma(\alpha,1) = e^{\alpha\bar\gamma-\bar\alpha\gamma},$$ and thus $$\chi_\beta(\alpha,1)\chi_\gamma(\alpha,1) = \chi_{|\beta+\gamma\rangle}(\alpha,1).$$ In other words, the convolution of the $P$ functions of two coherent states $|\beta\rangle$ and $|\gamma\rangle$ is the $P$ function of the coherent state $|\beta+\gamma\rangle$.

General case

This isn't complete, but observe that given any channel $\Phi$ we can write $$\chi_{\Phi(\rho,\sigma)}(\alpha,s) = e^{\frac s2|\alpha|^2}\operatorname{tr}[D(\alpha) \Phi(\rho\otimes\sigma)] = e^{\frac s2|\alpha|^2}\operatorname{tr}[\Phi^\dagger(D(\alpha)) (\rho\otimes\sigma)],$$ where $\Phi^\dagger$ is the adjoint channel of $\Phi$. This tells you that if you know how to "evolve back" displacement operators, you know the characteristic function of the evolved state. And if you can find a channel $\Phi$ such that $\Phi^\dagger(D(\alpha))\simeq D(\alpha_1)\otimes D(\alpha_2)$, then the resulting characteristic function of $\Phi(\rho\otimes\sigma)$ is indeed a pointwise product of the individual characteristic functions, and thus the $P$-function of $\Phi(\rho\otimes\sigma)$ is the convolution of the individual $P$ functions.

Evolution through a beamsplitter

Consider two generic states $\rho\otimes\sigma$ evolving through a balanced beamsplitter. This means in particular that coherent states evolve as $$|\alpha\rangle\otimes|\beta\rangle\to \left\lvert\frac{\alpha+\beta}{\sqrt2}\right\rangle\otimes \left\lvert\frac{\alpha-\beta}{\sqrt2}\right\rangle.$$ Using this we can derive a formal expression for the result of evolving $\rho\otimes\sigma$ through a beamsplitter, via their $P$ representations: $$\rho\otimes\sigma= \int\mathrm d^2\alpha\mathrm d^2\beta P_\rho(\alpha)P_\sigma(\beta) (\mathbb{P}_\alpha\otimes\mathbb{P}_\beta) \\ \to \int\mathrm d^2\alpha\mathrm d^2\beta P_\rho(\alpha)P_\sigma(\beta) (\mathbb{P}_{(\alpha+\beta)/\sqrt2}\otimes\mathbb{P}_{(\alpha-\beta)/\sqrt2}) ,$$ using the shorthand notation $\mathbb{P}_\alpha\equiv |\alpha\rangle\!\langle\alpha|$. If we only look at the marginal state on the first output mode, tracing out the second one, we get $$\rho' = \int\mathrm d^2\alpha\mathrm d^2\beta P_\rho(\alpha)P_\sigma(\beta) \mathbb{P}_{(\alpha+\beta)/\sqrt2} \\ = \int\mathrm d^2\gamma \left( \frac{\mathrm d^2\eta}{4} P_\rho(\tfrac{\gamma+\eta}{\sqrt2})P_\sigma(\tfrac{\gamma-\eta}{\sqrt2}) \right) \mathbb{P}_\gamma \\ =\int\mathrm d^2\gamma \left( \mathrm d^2\theta P_\rho(\sqrt2\gamma-\theta)P_\sigma(\theta) \right) \mathbb{P}_\gamma.$$ In words, we found that the state obtained evolving $\rho\otimes\sigma$ through a symmetric beamsplitter and looking only at the first output mode has $P$ representation equal to the convolution of the $P$ representations of $\rho$ and $\sigma$, modulo a rescaling of the input parameter.

$\endgroup$
0
$\begingroup$

Found a copy of the Glauber paper that is not behind the paywall. It seems that what is called "convolution of the P-representations of two coherent states" by the OP is actually the sequential excitation of states with given P representations within a phase space spanned by displaced coherent states. It then leads to a convolution of the two P representation. Does that correspond to a superposition?

Let's distinguish between the Schroedinger cat state $$ |\text{cat}\rangle = \frac{1}{\sqrt{2}} |\alpha_1\rangle + \frac{1}{\sqrt{2}} |\alpha_2\rangle , $$ and a displaced coherent state $|\alpha_1+\alpha_2\rangle$. The latter can be produced by sending two coherent states through a beam splitter. In effect one may think of this as a combination of two states in terms of the superpositions of the individual photons. Such a process is only possible for a coherent state because all the photons in a coherent state carry exactly the same properties.

The Schroedinger cat state is a true quantum superposition. Each term can represent an arbitrary state (doesn't have to be a coherent state) with multiple photons. The phase space representation of such a state would consist of four terms, two of which can be combine into a oscillatory term. Ons can show this by computing the characteristic function as the trace of the density operator by an adjoint displacement operator and then perform a symplectic Fourier transform (with the necessary Gaussian factor depending of the choice of representation). Since this whole process is linear in the state, it does not seem to lead to a convolution between the individual phase space representations.

$\endgroup$
3
  • $\begingroup$ Yes, my question is why Glauber calls it a superposition in the text when it is very different from a quantum superposition. I suppose the distinction is in calling it a "superposition of fields" rather than of states. But then what does the former mean in general? And is there any link between the concepts? This same terminology of "superposition of fields" seems to have been used in the more general context of convolution of two P-representations (per the second link). So is this a general rule/what does it mean physically in general? $\endgroup$ Commented Jun 11 at 21:34
  • $\begingroup$ In terms of the way people understanding these things today, the scenario described by Glauber in that paper would not be called a superposition at all. In fact, the resulting state should be a tensor product of the two fields produced by the two sources. The reason why it comes out with the superposition of fields is because some of the photons from the two sources would be indistinguishable. $\endgroup$ Commented Jun 12 at 4:00
  • $\begingroup$ OK, so this is more historical terminology than a distinct/currently used concept then? That makes sense, since it is a 60 year old paper. If so, that answers my first question. $\endgroup$ Commented Jun 13 at 1:16

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.