betaFeedback

WorksPhysics and Astronomy

Windowed Catodyne Measurements: Continuum Approximation versus Exact Discrete Photon-Number Treatment

Abstract

We analyze a family of measurements obtained by retaining only a finite window of the photon-sum variable in catodyne detection, where a Schrödinger cat state serves as the local oscillator. In the continuum approximation, the window half-width appears to act as a continuous knob for coherence. However, a treatment on the discrete photon-number lattice shows that, in the strong-LO regime, the coherence conditioned on the difference outcome is not suppressed by the window. The exact parity-sensitive success probability is a staircase function of and , whose plateau structure and magnitude the continuum formulas fail to capture, especially for small and narrow windows. For remote preparation of cat states, an exact calculation with a finitely squeezed resource shows that the remote state keeps a purity close to one when x is resolved, and loses purity mainly when x is not recorded. Our results clarify where the continuum approximation of catodyne detection can be trusted and where the discrete photon-number structure must be used.

Introduction

Quantum-optical measurements are the backbone of modern quantum technologies, from quantum state tomography to continuous-variable quantum computing. Among these, balanced homodyne detection is the most widely used technique for measuring field quadratures [1, 2]. In the standard configuration, a signal mode is mixed with a strong coherent-state local oscillator (LO) at a 50:50 beam splitter, and the photocurrent difference yields a quadrature observable in the limit of large LO amplitude [1, 3]. The corresponding positive operator-valued measure (POVM) is a rank-1 projector onto a quadrature eigenstate ∣xθ⟩|x_\theta\rangle, where θ\theta is the LO phase [4].

The LO, however, need not be a coherent state. Nonclassical LO states can give rise to nonclassical measurements with properties beyond standard homodyne detection [5, 6, 7]. A particularly interesting case is the use of a Schrödinger-cat state ∣β⟩±∣−β⟩|\beta\rangle\pm|-\beta\rangle as the LO, a scheme referred to as “catodyne” [8]. Combes and Lund derived the exact Kraus operators and POVM elements for catodyne and showed that, in the strong-LO limit, the POVM elements contain both diagonal terms and coherence terms of the form ∣xθ⟩⟨−xθ∣|x_\theta\rangle\langle-x_\theta| [8]. They identified two limiting cases. If both the difference and the sum of the photon counts are retained, the POVM has coherence between ∣xθ⟩|x_\theta\rangle and ∣−xθ⟩|-x_\theta\rangle (rank 1). If the sum is integrated out in the continuum approximation, this coherence vanishes and the measurement becomes reflection-symmetric (rank 2). They also noted that the ability to use or ignore the information in the sum variable may provide a means to engineer a measurement [8].

This observation raises a natural question that, to our knowledge, has not been studied quantitatively: what happens between the two limits, when only a finite window of the sum variable is retained? In the continuum approximation, where the sum variable ww is treated as a continuous Gaussian-distributed quantity, the window half-width δ\delta appears to interpolate smoothly between the two limits and to act as a continuous tuning knob for coherence. A related line of work has shown how to tune between photon-number and quadrature measurements by varying the amplitude of a coherent-state LO [9]. Here the LO amplitude is fixed, and the parameter under study is a property of how the photon-counting record is processed.

However, the continuum picture is only an approximation. In an actual experiment the photon counts nn and mm are integers, so that 2∣β∣x=n−m\sqrt{2}|\beta|x=n-m are integers of equal parity. In this work we analyze the windowed catodyne measurement both in the continuum approximation and on the exact discrete photon-number lattice, and we find that the two descriptions differ qualitatively.

The main contributions of this work are threefold:

1-Coherence conditioned on xx is preserved in the strong-LO regime. We show that on the discrete photon-number lattice, windowing the sum variable ww does not reduce the coherence conditioned on the difference outcome xx. The apparent suppression of coherence in the continuum model is an artifact of the continuum approximation. Coherence is lost only when xx is also integrated out or ignored.

2-Parity-sensitive acceptance. We derive exact windowed success probabilities for Fock-state inputs and define a parity-discrimination signal D. We show that D is a staircase function of δ\delta and ∣β∣|\beta|, with magnitudes and plateaus that differ substantially from the continuum prediction.

Limits of the continuum approximation

By comparing exact discrete results with the continuum formulas, we quantify where the continuum approximation breaks down. The breakdown is strongest for small ∣β∣|\beta| and narrow windows, where the window contains only a few photon-number values, as is the case for realistic photon-number-resolving detectors (see § 6.4).

The structure of this paper is as follows. In § 2 we introduce the setup and notation. In § 3 we define the windowed POVM, derive its continuum-approximation expressions, and then present the exact discrete treatment. In Sec. 4 we present numerical results comparing the two. In § 5 we discuss the discrete parity structure. In Sec. 6 we discuss implications for measurement engineering and remote state preparation. We conclude in § 7.

Setup and Notation

We consider a balanced homodyne setup in which a signal mode aa and a LO mode bb are mixed at a 50:50 beam splitter. The output modes are

Balanced homodyne setup

Figure 1. Balanced homodyne setup. A signal state ∣signal⟩|\mathrm{signal}\rangle is mixed with a local oscillator ∣LO⟩|\mathrm{LO}\rangle at a 50:50 beam splitter. The two output ports cc and dd are measured by ideal photon-number-resolving detectors D1\mathrm{D}_1 and D2\mathrm{D}_2. Detector 1 registers nn photons and detector 2 registers mm photons.

c=a+b2,d=a−b2,(1)c = \frac{a+b}{\sqrt{2}}, \qquad d = \frac{a-b}{\sqrt{2}}, \tag*{(1)}

and the two output ports are measured by ideal photon-number-resolving detectors [8]. The joint input state is ρin=ρ⊗∣LO⟩⟨LO∣\rho_{\mathrm{in}}=\rho\otimes|\mathrm{LO}\rangle\langle\mathrm{LO}|, where the signal state is. For catodyn[e], the LO is prepared in a cat state

∣cat±(β)⟩=1N±(∣β⟩±∣−β⟩),(2)|\mathrm{cat}_{\pm}(\beta)\rangle= \frac{1}{\mathcal{N}_{\pm}}(|\beta\rangle\pm|-\beta\rangle), \tag*{(2)}
N±=2(1±e−2∣β∣2).\mathcal{N}_{\pm} = \sqrt{2\left(1\pm e^{-2|\beta|^2}\right)}.

where β=∣β∣eiθ\beta=|\beta|e^{i\theta} is the complex amplitude of the coherent state, ∣β∣|\beta| is the LO amplitude, θ\theta is the LO phase, and N±\mathcal{N}_{\pm} is the normalization constant [8]. The upper sign corresponds to the even cat and the lower sign to the odd cat.

The Kraus operator for detecting nn photons in mode cc and mm photons in mode dd is obtained by projecting the output state onto the corresponding Fock states [8]

Mn,m[β]=⟨0∣e−∣β∣2/22(n+m)/2n!m!N±(β)×[(a+β)n(a−β)m±(a−β)n(a+β)m].(3)M_{n,m}^{[\beta]}=\langle0|\frac{e^{-|\beta|^2/2}}{2^{(n+m)/2}\sqrt{n!m!}\mathcal{N}_{\pm}(\beta)}\times \left[(a+\beta)^n(a-\beta)^m\pm(a-\beta)^n(a+\beta)^m\right]. \tag*{(3)}

This expression is valid for all β\beta and acts only on the signal mode [8]. The two terms correspond to the two coherent components of the cat state and are added at the amplitude level, which is the origin of the quantum coherence in catodyne.

We introduce the scaled variables

x=n−m2∣β∣,w=n+m2∣β∣.(4)x=\frac{n-m}{\sqrt{2}|\beta|}, \qquad w=\frac{n+m}{\sqrt{2}|\beta|}. \tag*{(4)}

Because nn and mm are integers, 2∣β∣x=n−m\sqrt{2}|\beta|x=n-m and 2∣β∣w=n+m\sqrt{2}|\beta|w=n+m are integers of equal parity, so xx and ww take values on a discrete lattice of spacing 1/(2∣β∣)1/(\sqrt{2}|\beta|). The strong-LO expressions below hold when ∣β∣2|\beta|^2 is much larger than ⟨a†4⟩\langle a^{\dagger4}\rangle [1, 8]; for small ∣β∣|\beta| they are only approximate, and we compare them with the exact Kraus operator of Eq. (3) in Secs. 3.4, 5 and 6.3.2.

In the strong-LO limit, the Kraus operator factorizes as [8]

Mx,w[β]±dw dx≈12ei2∣β∣wθ×[∣xθ⟩eiπ2∣β∣(x−w)/2±⟨−xθ∣eiπ2∣β∣(x+w)/2]×e−(w−∣β∣2)2/2π1/4dw dx.(5)M_{x,w}^{[\beta]\pm}\sqrt{dw\,dx}\approx\frac{1}{\sqrt{2}}e^{i\sqrt{2}|\beta|w\theta}\times \left[|x_\theta\rangle e^{i\pi\sqrt{2}|\beta|(x-w)/2}\pm\langle-x_\theta|e^{i\pi\sqrt{2}|\beta|(x+w)/2}\right]\times \frac{e^{-(w-|\beta|\sqrt{2})^2/2}}{\pi^{1/4}}\sqrt{dw\,dx}. \tag*{(5)}

This factorization shows that the Kraus operator decomposes into a combination of quadrature eigenstates and a Gaussian weight in ww. The corresponding POVM element, obtained by squaring this expression, is

dw dxEx,w[β]≈dw dxe−(w−∣β∣2)2/2π12×[∣xθ⟩⟨xθ∣+∣−xθ⟩⟨−xθ∣±(∣xθ⟩⟨−xθ∣ei2π∣β∣w±∣−xθ⟩⟨xθ∣e−i2π∣β∣w)].(6)dw\,dxE_{x,w}^{[\beta]}\approx dw\,dx\frac{e^{-(w-|\beta|\sqrt{2})^2/2}}{\sqrt{\pi}}\frac{1}{2}\times \left[|x_\theta\rangle\langle x_\theta|+|-x_\theta\rangle\langle-x_\theta|\pm \left(|x_\theta\rangle\langle-x_\theta|e^{i\sqrt{2}\pi|\beta|w}\pm|-x_\theta\rangle\langle x_\theta|e^{-i\sqrt{2}\pi|\beta|w}\right)\right]. \tag*{(6)}

The first two terms are diagonal and describe the classical part of the measurement. The last two terms are coherence terms that couple the quadrature eigenstates and are responsible for the nonclassical character of the measurement [8].

Windowed Sum Variable: Continuum Approximation and Exact Treatment

Definition

We define the partial-resolution POVM by integrating Ex,w∣β∣±E_{x,w}^{|\beta| \pm} over a finite window of ww:

Ex,w(δ,β)±≈∫w0−δw0+δdwEx,w±[β],w0=∣β∣2.(7)E_{x,w}^{(\delta,\beta)\pm} \approx\int_{w_0-\delta}^{w_0+\delta} dw E_{x,w}^{\pm[\beta]}, \qquad w_0 = \frac{|\beta|}{\sqrt{2}}. \tag*{(7)}

The window half-width δ\delta is measured in units of ww; in photon-number units it corresponds to 2∣β∣δ\sqrt{2}|\beta|\delta.

This definition is the central concept of our analysis. In contrast to previous measurements, which either fully discard ww (no coherence) or fully retain it (full coherence), this definition generates a family of measurements parametrized by δ\delta.

Analytical Expressions

The following expressions treat ww as continuous. As shown in Sec. 3.5, this approximation misrepresents the coherence on the true lattice and is retained only as a large-∣β∣|\beta| reference.

Performing the integration, we obtain

Ex(δ,β)=Gδ2(∣xθ⟩⟨xθ∣+∣−xθ⟩⟨−xθ∣)E_x^{(\delta,\beta)} = \frac{G_\delta}{2}\left(|x_\theta\rangle\langle x_\theta| + |-x_\theta\rangle\langle-x_\theta|\right)
±12[Iδ(β)∣xθ⟩⟨−xθ∣±Iδ∗(β)∣−xθ⟩⟨xθ∣].(8)\qquad\pm\frac{1}{2}\left[I_\delta(\beta)|x_\theta\rangle\langle-x_\theta| \mathbin{\pm} I_\delta^*(\beta)|-x_\theta\rangle\langle x_\theta|\right]. \tag*{(8)}

where GδG_\delta is the "success probability" (the fraction of the ww distribution that falls inside the window),

Gδ=∫w0−δw0+δe−(w−w0)2πdw=erf⁡(δ),(9)G_\delta= \int_{w_0-\delta}^{w_0+\delta} \frac{e^{-(w-w_0)^2}}{\sqrt{\pi}}dw = \operatorname{erf}(\delta), \tag*{(9)}

and

Iδ=∫w0−δw0+δe−(w−w0)2πei2π∣β∣wdw,(10)I_\delta= \int_{w_0-\delta}^{w_0+\delta} \frac{e^{-(w-w_0)^2}}{\sqrt{\pi}}e^{i\sqrt{2}\pi|\beta|w}dw, \tag*{(10)}

is the "coherence weight" (how much coherence remains in the window). For the symmetric window, IδI_\delta can be simplified by introducing the variable w=w0−yw=w_0-y.

Iδ(β)=2eiπ∣β∣2π∫0δe−y2cos⁡(2π∣β∣y)dy,(11)I_\delta(\beta) = \frac{2e^{i\pi|\beta|^2}}{\sqrt{\pi}}\int_0^\delta e^{-y^2}\cos(\sqrt{2}\pi|\beta|y)dy, \tag*{(11)}

The integral is real; the phase IδI_\delta comes entirely from the global factor eiπ∣β∣2e^{i\pi|\beta|^2}, up to an additional phase of π\pi whenever the integral changes sign.

Normalized coherence factor

To quantify the degree of coherence retained, we define the "normalized coherence factor":

Cδ(β)=∣Iδ(β)∣Gδ(β).(12)C_\delta(\beta) = \frac{|I_\delta(\beta)|}{G_\delta(\beta)}. \tag*{(12)}

This quantity is the ratio of the coherence weight to the success probability. To see why 0≤Gδ(β)≤10 \leq G_\delta(\beta) \leq1, we examine the two weights: GδG_\delta is a real, positive integral of a Gaussian function, while IδI_\delta is a complex integral of the same Gaussian with a phase factor. Since the modulus of an integral of a complex function cannot exceed the integral of its modulus, we have

∣Iδ∣=∣∫w0−δw0+δe−(w−w0)2πei2π∣β∣wdw∣≤∫w0−δw0+δe−(w−w0)2π∣ei2π∣β∣w∣dw=∫w0−δw0+δe−(w−w0)2πdw=Gδ.(13)\begin{aligned} |I_\delta| = \left|\int_{w_0-\delta}^{w_0+\delta} \frac{e^{-(w-w_0)^2}}{\sqrt{\pi}}e^{i\sqrt{2}\pi|\beta|w}dw\right| \\ &\leq\int_{w_0-\delta}^{w_0+\delta} \frac{e^{-(w-w_0)^2}}{\sqrt{\pi}}\left|e^{i\sqrt{2}\pi|\beta|w}\right|dw = \int_{w_0-\delta}^{w_0+\delta} \frac{e^{-(w-w_0)^2}}{\sqrt{\pi}}dw = G_\delta. \tag*{(13)} \end{aligned}

Since ∣eiφ∣=1|e^{i\varphi}|=1 for any real phase φ\varphi, it follows that ∣Iδ∣≤Gδ|I_\delta| \leq G_\delta, and hence 0≤Cδ≤10 \leq C_\delta\leq1. (The bound 0≤Gδ≤10 \leq G_\delta\leq1 follows directly from Gδ=erf⁡(δ)G_\delta=\operatorname{erf}(\delta)). Physical interpretation: "CδC_\delta is the ratio of "retained coherence" to "total success probability." If Cδ=1C_\delta=1, all accepted events have full coherence (a pure state). If Cδ=0C_\delta=0, no coherence remains (a classical mixture).

In the limit of a narrow window (δ→0\delta\to0), we find ∣Iδ∣≈2δπ|I_\delta| \approx\frac{2\delta}{\sqrt{\pi}} and Gδ≈2δπG_\delta\approx\frac{2\delta}{\sqrt{\pi}}, so that Cδ→1C_\delta\to1. In the limit of a wide window (δ→∞\delta\to\infty), the oscillatory factor cos⁡(2π∣β∣y)\cos(\sqrt{2}\pi|\beta|y) causes the integral to cancel, so that ∣Iδ∣→0|I_\delta| \to0 and Cδ→0C_\delta\to0 (up to a residual of order e−π2∣β∣22e^{-\frac{\pi^2|\beta|^2}{2}}).

Within the continuum approximation, CδC_\delta therefore interpolates between 1 and 0. Sec. 3.5 shows that this behavior does not survive on the discrete photon-number lattice.

Parity-Discrimination Signal

The coherence weight IδI_\delta also controls the "parity sensitivity" of the measurement. For a Fock state ∣N⟩|N\rangle, the parity operator P=(−1)a^†a^P=(-1)^{\hat{a}^\dagger\hat{a}} acts as P∣N⟩=(−1)N∣N⟩P|N\rangle=(-1)^N|N\rangle.

In the continuum approximation, the success probability for detecting ∣N⟩|N\rangle in the window is

psucc(N,±)=Gδ±(−1)NRe⁡(Iδ)(14)p_{\mathrm{succ}}(N,\pm)=G_\delta\pm(-1)^N\operatorname{Re}(I_\delta) \tag*{(14)}

because ∫dx∣x⟩⟨−x∣=P\int dx|x\rangle\langle-x|=P. The success probability thus depends on the parity of the input state, a parity dependence that is absent in standard homodyne detection. For the even cat, the continuum parity signal is the difference between even and odd inputs:

Dcont=2Re⁡Iδ.(15)D_{\mathrm{cont}} = 2\operatorname{Re} I_{\delta}. \tag*{(15)}

These expressions treat ww as continuous. On the true photon-number lattice, nn and mm are integers, so the window contains only the integer values s=n+ms=n+m satisfying

∣β∣2−2∣β∣δ≤s≤∣β∣2+2∣β∣δ.(16)|\beta|^2-\sqrt{2}|\beta|\delta\le s \le|\beta|^2+\sqrt{2}|\beta|\delta. \tag*{(16)}

These integers are the points of the window [wo−δ,wo+δ][w_o-\delta,w_o+\delta] of Eq. (7), where w=s/(2∣β∣)w=s/(\sqrt{2}|\beta|) and wo=∣β∣/2w_o=|\beta|/\sqrt{2}. We write WW for this window, so that s∈Ws\in W means that w=s/(2∣β∣)w=s/(\sqrt{2}|\beta|) lies inside it. For a Fock input ∣N⟩|N\rangle, the exact windowed success probability is

P(N,+)=∑s∈w,n+m=se−∣β∣2×∣⟨0∣(a+β)n(a−β)m±(a−β)n(a+β)m∣2∣2(n+m)n!m!N±2(β).(17)P(N,+)=\sum_{\substack{s\in w,n+m=s}}e^{-|\beta|^2}\times\frac{\left|\langle0|(a+\beta)^n(a-\beta)^m\pm(a-\beta)^n(a+\beta)^m|^2\right|}{2^{(n+m)}n!m!\mathcal{N}_{\pm}^2(\beta)}. \tag*{(17)}

where N±2=2(1±e−2∣β∣2)\mathcal{N}_{\pm}^2=2\left(1\pm e^{-2|\beta|^2}\right), and the matrix element is evaluated with Eq. (G4) of Combes and Lund [8]. We define the exact parity signal as

Ddise=P(0,+)−P(1,+).(18)D_{\mathrm{dise}}=P(0,+)-P(1,+). \tag*{(18)}

Both signals can take either sign, but the exact signal is piecewise constant and can be larger in magnitude. For example, for ∣β∣=2|\beta|=2 and δ=0.05\delta=0.05 we find Ddise=0.39D_{\mathrm{dise}}=0.39, whereas the continuum formula gives 0.110.11. For ∣β∣=2.3|\beta|=2.3 and δ=0.1\delta=0.1 we find Ddise=−0.33D_{\mathrm{dise}}=-0.33.

For these two inputs the sums can be done in closed form. For the vacuum and an even cat, only even ss contribute, with P(0,+)=∑s even4e−∣β∣2∣β∣2s/(s!N+)P(0,+)=\sum_{s\ \mathrm{even}}4e^{-|\beta|^2}|\beta|^{2s}/(s!\mathcal{N}_{+}). For a one-photon input, only odd ss contribute, with P(1,+)=∑s odd4e−∣β∣2∣β∣2(s−1)/((s−1)!N+)P(1,+)=\sum_{s\ \mathrm{odd}}4e^{-|\beta|^2}|\beta|^{2(s-1)}/((s-1)!\mathcal{N}_{+}). Both sums run over s∈ws\in w.

Discrete-lattice treatment of the coherence (strong-LO)

We work in the strong-LO regime of Combes and Lund [8], where the Kraus operator for the outcome (x,w)(x,w) is, up to a global phase eiπ2∣β∣we^{i\pi\sqrt{2}|\beta|w} and the Gaussian amplitude g(w)\sqrt{g(w)},

Mx,w∝⟨x∣±eiπ2∣β∣w∣−x⟩.(19)M_{x,w}\propto\langle x|\pm e^{i\pi\sqrt{2}|\beta|w}|-x\rangle. \tag*{(19)}

Because nn and mm are integers, 2∣β∣w=n+m\sqrt{2}|\beta|w=n+m and 2∣β∣x=n−m\sqrt{2}|\beta|x=n-m are integers of equal parity. Hence

eiπ2∣β∣w=(−1)n+m=(−1)n−m=(−1)2∣β∣x.(20)e^{i\pi\sqrt{2}|\beta|w}=(-1)^{n+m}=(-1)^{n-m}=(-1)^{\sqrt{2}|\beta|x}. \tag*{(20)}

The relative sign between the two quadrature eigenstates therefore depends on xx only. It is identical for every value of ww that is compatible with a given xx.

Dcont=2Re⁡Iδ.(21)D_{\mathrm{cont}}=2\operatorname{Re}I_{\delta}. \tag*{(21)}

Proposition. On the discrete lattice, all outcomes (x,w)(x,w) with the same xx have Kraus operators that are proportional to each other. For a window W=[wo−δ,wo+δ]W=[w_o-\delta,w_o+\delta] of the sum variable, the windowed POVM element is

Exw=Gxw2(∣x⟩+σx∣−x⟩)(⟨x∣+σx⟨−x∣),(22)E_x^w=\frac{G_x^w}{2}(|x\rangle+\sigma_x|-x\rangle)(\langle x|+\sigma_x\langle-x|), \tag*{(22)}

where σx=±(−1)2∣β∣x\sigma_x=\pm(-1)^{\sqrt{2}|\beta|x},

where GxwG_x^w is the sum of the Gaussian weights g(w)g(w) over the lattice points of ww compatible with xx. This is a rank-1 projector onto a cat state of definite parity. Consequently the normalized coherence factor is

Cδdisc(x)=∣Ixw∣Gxw=1.(23)C_{\delta}^{\mathrm{disc}}(x)=\frac{|I_x^w|}{G_x^w}=1. \tag*{(23)}

This holds for every window and every δ\delta.

Proof. The Kraus operators for fixed xx differ only by a scalar factor g(w)eiφ(x,w)\sqrt{g(w)}e^{i\varphi(x,w)}, since the relative sign σx\sigma_x is the same for all compatible ww. Summing Ex,w=Mx,w†Mx,wE_{x,w}=M_{x,w}^{\dagger}M_{x,w} over w∈Ww\in W therefore multiplies the rank-1 operator (∣x⟩+σx∣−x⟩)(⟨x∣+σx⟨−x∣)/2(|x\rangle+\sigma_x|-x\rangle)(\langle x|+\sigma_x\langle-x|)/2, weight GxwG_x^w. The coherence term carries the same weight as the diagonal terms, so ∣Ixw∣=Gxw|I_x^w|=G_x^w.

Consequences.

1-Windowing ww is a coarse-graining of outcomes whose Kraus operators are proportional. Such a coarse-graining cannot reduce coherence. When xx is resolved, the windowed measurement stays rank-1 for every δ\delta.

2-Coherence is lost only when xx is also coarse-grained over several lattice points, because then (−1)2∣β∣x(-1)^{\sqrt{2}|\beta|x} alternates inside the bin and the coherence terms cancel, or when xx is ignored (as in parity-only observables).

3-In the continuum approximation, the phase eiπ2∣β∣xe^{i\pi\sqrt{2}|\beta|x} is treated as a continuously varying function of δ\delta. The integral over the window then averages it out. This produces the decay of CδcontC_{\delta}^{\mathrm{cont}} and is an artifact of the approximation.

Limits. The argument uses the strong-LO approximation. For finite ∣β∣|\beta| the Kraus operators for different ww at fixed xx are only approximately proportional. § 6.3.2 and Figure 5 quantify this correction for the remote-state purity, using the exact Kraus operators of Eq. (3).

Numerical Results

Coherence Factor CδC_{\delta}

Figure 2 compares the coherence factor of the continuum approximation with the discrete strong-LO result. In the continuum approximation, the coherence factor starts at 1 for δ=0\delta=0 and decays with oscillations. Its first zero occurs near δ≈1/(2∣β∣)\delta\approx1/(\sqrt{2}|\beta|), that is, at δ≈0.35\delta\approx0.35, 0.140.14 and 0.070.07 for ∣β∣=2|\beta|=2, 5 and 10. Larger ∣β∣|\beta| therefore leads to

Figure 5. Linear purity of the remote state as a function of the window half-width δ\delta, for 12 dB of two-mode squeezing, computed with the exact Kraus operators. Top: xx resolved, for ∣β∣=3,5,8,12|\beta| = 3,5,8,12 (with x≈0.5x \approx0.5). Bottom: xx not recorded, for ∣β∣=2,3,5|\beta| = 2,3,5.

faster decay. In contrast, the discrete strong-LO coherence conditioned on xx equals 1 for every δ\delta (Sec. 3.5). The apparent loss of coherence is thus a property of the continuum approximation and not of the windowed measurement itself.

Normalized coherence as a function of the window half-width for $|\beta|=2,5,10$

Figure 2. Normalized coherence as a function of the window half-width δ\delta, for ∣β∣=2,5,10|\beta|=2,5,10 (top to bottom). Blue: continuum approximation. Red: discrete strong-LO result, which equals 1 for every δ\delta when the difference outcome xx is resolved (Sec. 3.5).

Parity Signal DδD_{\delta}

Figure 3 compares the exact parity signal with the continuum prediction. The continuum curve is a smooth oscillation whose first extremum lies near δ≈1/(22∣β∣)\delta\approx1/(2\sqrt{2}|\beta|), i.e. δ≈0.18\delta\approx0.18, 0.07 and 0.035 for ∣β∣=2|\beta|=2, 5 and 10. The exact signal differs from it in three ways.

First, it is piecewise constant. It changes only when an integer value of n+mn+m enters or leaves the window, so the jumps are separated by 1/(2∣β∣)1/(\sqrt{2}|\beta|) in δ\delta (for ∣β∣=5|\beta|=5, at δ≈0.14,0.28,0.42\delta\approx0.14,0.28,0.42).

Second, its sign changes abruptly at these jumps, whereas the continuum curve changes sign smoothly. The signs agree for large ∣β∣|\beta|, but the exact signal can vanish identically where the continuum value is nonzero; for ∣β∣=2.3|\beta|=2.3 it is zero for δ\delta below about 0.09.

Third, its magnitude can be much larger than the continuum prediction. For ∣β∣=2|\beta|=2 and δ=0.05\delta=0.05 the exact signal is 0.39, whereas the continuum value is 0.11; for ∣β∣=2.3|\beta|=2.3 and δ=0.1\delta=0.1 the exact value is −0.33-0.33 versus −0.12-0.12; for ∣β∣=5|\beta|=5 and δ=0.05\delta=0.05 it is −0.16-0.16 versus −0.09-0.09.

For large ∣β∣|\beta| the continuum curve follows the sign changes of the exact signal and can be viewed as a smoothed version of it, while the amplitude remains underestimated.

Discrete Parity Structure

Motivation

The analytical results of § 3 are based on the continuum approximation for ww. In an actual experiment, however, the photon counts nn and mm are integers, so ww takes only discrete values. In this section, we show that this discreteness reveals a parity structure that is not captured by the continuum approximation.

Exact Photon-Number Distribution

For vacuum input and a cat-state LO, the exact joint photon-number distribution is [8]

P(n,m∣β,±,0)=e−∣β∣2n!m!N±2(β)(∣β∣22)n+m[(−1)m−n±1]2,(24)P(n,m|\beta,\pm,0)=\frac{e^{-|\beta|^2}}{n!m!\mathcal{N}_{\pm}^2(\beta)}\left(\frac{|\beta|^2}{2}\right)^{n+m}\left[(-1)^{m-n}\pm1\right]^2, \tag*{(24)}

where N±2(β)=2(1±e−2∣β∣2)\mathcal{N}_{\pm}^2(\beta)=2\left(1\pm e^{-2|\beta|^2}\right).

The factor [(−1)m−n±1]2\left[(-1)^{m-n}\pm1\right]^2 imposes the "parity selection rule": for the even cat, the probability vanishes unless n+mn+m is even; for the odd cat, it vanishes unless n+mn+m is odd.

Windowed success probability

The success probability for the window [wo−δ,wo+δ][w_{\mathrm{o}}-\delta,w_{\mathrm{o}}+\delta], with wo=∣β∣2w_{\mathrm{o}}=\frac{|\beta|}{\sqrt{2}}, is obtained by summing the exact distribution over all photon-number pairs whose w=n+m2∣β∣w=\frac{n+m}{\sqrt{2}|\beta|} lies inside the window:

Psucc±(β,δ)=∑n,m≥0∣β∣2−2∣β∣δ≤n+m≤∣β∣2+2∣β∣δP(n,m∣β,±,0).(25)P_{\mathrm{succ}}^{\pm}(\beta,\delta)=\sum_{\substack{n,m\geq0\\ |\beta|^2-\sqrt{2}|\beta|\delta\leq n+m\leq|\beta|^2+\sqrt{2}|\beta|\delta}}P(n,m|\beta,\pm,0). \tag*{(25)}

Defining the integer bounds smin⁡=⌈∣β∣2−2∣β∣δ⌉s_{\min}=\left\lceil|\beta|^2-\sqrt{2}|\beta|\delta\right\rceil, and smax⁡=⌊∣β∣2+2∣β∣δ⌋s_{\max}=\left\lfloor|\beta|^2+\sqrt{2}|\beta|\delta\right\rfloor, and using the binomial identity

∑n+m=s1m! n!=2ss!,\sum_{n+m=s}\frac{1}{m!\,n!}=\frac{2^s}{s!},

the inner sum can be evaluated explicitly:

P(β,δ)=∑s=smin⁡smax⁡e−∣β∣2(∣β∣2)sN±2(β)s![(−1)s±1]2.(26)P(\beta,\delta)=\sum_{s=s_{\min}}^{s_{\max}}\frac{e^{-|\beta|^2}(|\beta|^2)^s}{\mathcal{N}_{\pm}^2(\beta)s!}\left[(-1)^s\pm1\right]^2. \tag*{(26)}

Discrete versus Continuum

Figure 4 compares the exact discrete success probability with two continuum expressions: Gδ=erf⁡(δ)G_{\delta}=\operatorname{erf}(\delta), and Gδ+Re⁡IδG_{\delta}+\operatorname{Re}I_{\delta}, which includes the parity term of Eq. (14). For large ∣β∣|\beta| (∣β∣=5,10|\beta|=5,10) the steps are small and both continuum curves follow the exact result on average, with Gδ+Re⁡IδG_{\delta}+\operatorname{Re}I_{\delta} tracking the individual steps more closely.

Success probability for a vacuum input and an even-cat local oscillator as a function of the window half-width

Figure 4. Success probability for a vacuum input and an even-cat local oscillator as a function of the window half-width δ\delta, for ∣β∣=2,5,10|\beta|=2,5,10 (top to bottom). Red: exact discrete result. Blue dashed: continuum approximation Gδ=erf⁡(δ)G_{\delta}=\operatorname{erf}(\delta). Green dotted: continuum approximation including the parity term, Gδ+Re⁡IδG_{\delta}+\operatorname{Re}I_{\delta}. The exact result is a staircase because only even values of n+mn+m have nonzero probability.

For small ∣β∣|\beta| the continuum description fails. For ∣β∣=2|\beta|=2, the exact probability is a single plateau of about 0.39 up to δ≈0.71\delta\approx0.71, followed by a jump to about 0.89; neither continuum curve reproduces this.

For ∣β∣=5|\beta|=5, ∣β∣2=25|\beta|^2=25 is odd, so a vacuum input has zero success probability for windows narrower than δ≈0.14\delta\approx0.14.

This parity effect is absent in Gδ=erf⁡(δ)G_{\delta}=\operatorname{erf}(\delta) and is only partly captured by Gδ+Re⁡IδG_{\delta}+\operatorname{Re}I_{\delta}.

The steps are separated by 2/∣β∣\sqrt{2}/|\beta| in δ\delta, since for the even cat only even values of n+mn+m have nonzero probability.

Discussion

Measurement Engineering with a Windowed Sum Variable

Within the continuum approximation, the window width δ\delta appears to tune the coherence continuously, from a rank-1 projector (full coherence) to a rank-2 reflection-symmetric POVM (no coherence). On the discrete lattice, however, the coherence conditioned on xx is preserved for every window in the strong-LO regime (Sec. 3.5), while the parity-sensitive acceptance is a staircase function of δ\delta and ∣β∣|\beta|. The window on ww can therefore be chosen freely, for example to match the resolution of the detectors, without losing the coherence of the measurement.

The normalized coherence factor CδcontC_{\delta}^{\mathrm{cont}} describes the coherence predicted by the continuum approximation, whereas the discrete strong-LO coherence conditioned on xx equals 1. The parity-discrimination signal DδD_{\delta} quantifies the sensitivity of the success probability to the parity of the input state. Its exact value is a staircase function of δ\delta and ∣β∣|\beta| and can be tuned by choosing the window width and the local-oscillator amplitude (Figure 3).

Figure 3. Parity signal DD as a function of the window half-width δ\delta, for ∣β∣=2,2.3,5,10|\beta|=2,2.3,5,10 (top to bottom). Red: exact discrete signal D=P(0,+)−P(1,+)D=P(0,+)-P(1,+). Blue dashed: continuum approximation D=2Re⁡IδD=2\operatorname{Re} I_{\delta}. The exact signal is piecewise constant and changes only when an integer value of n+mn+m enters or leaves the window.

Comparison with Previous Work

Combes and Lund treated two limiting cases: full retention of the sum variable ww, which gives a rank-1 POVM with coherence, and complete integration over ww in the continuum approximation, which gives a rank-2 reflection-symmetric POVM [8]. They also remarked that the sum-variable information can be used or ignored to engineer a measurement. Our work studies the intermediate case of a finite window and shows, on the discrete lattice in the strong-LO regime, that coherence conditioned on xx is preserved for every window.

Thekkadath et al. showed that one can tune continuously between photon-number and quadrature measurements by varying the amplitude of a coherent-state LO [9]. Our setting differs in two respects: the LO is a cat state of fixed amplitude, and the varied quantity is the width of the window applied to the sum variable rather than the LO amplitude.

Olivares et al. studied the role of the LO intensity in homodyne-like tomography with photon-number-resolving detectors [10]. Our analysis is complementary: it concerns the structure of the measurement operators themselves and the role of the sum variable, rather than the reconstruction of states.

Implications for State Preparation

Windowing of the sum variable has operational consequences for remote state preparation (RSP) with catodyne measurements [8]. We discuss the ideal EPR limit in the strong-LO regime.

Protocol with an ideal EPR State

Consider an ideal EPR state shared between two parties,

∣EPR⟩ab=∫dx∣x⟩a⊗∣x⟩b.(27)|\mathrm{EPR}\rangle_{ab} = \int dx |x\rangle_a \otimes|x\rangle_b . \tag*{(27)}

A catodyne measurement is performed on mode aa with outcome (n,m)(n,m) and we define q=(n−m)/(2∣β∣)q=(n-m)/(\sqrt{2}|\beta|). The conditional state of mode bb is the Schrödinger-cat state in position

∣cq⟩=12(∣q⟩+σ∣−q⟩),σ=±(−1)2∣β∣q(28)|c_q\rangle= \frac{1}{\sqrt{2}}\left(|q\rangle+\sigma|-q\rangle\right), \qquad\sigma=\pm(-1)^{\sqrt{2}|\beta|q} \tag*{(28)}

where ±\pm is the parity of the cat LO. Because n+mn+m and n−mn-m have the same parity, the relative sign σ\sigma is determined by qq alone. It does not depend on the sum variable ww.

Purity of the remote state under windowing

Consider a window WW of the sum variable and suppose qq is resolved. In the strong-LO limit, all outcomes in the window with the same qq give the same relative sign σ\sigma and the same conditional state up to a global phase (Sec. 3.5). Windowing ww then does not mix different states, and the remote state is pure for every window width δ\delta.

More generally, suppose the relative sign is known only through probabilities p+p_+ and p−p_- for σ=+1\sigma=+1 and σ=−1\sigma=-1. The remote state is then ρ=p+∣c+⟩⟨c+∣+p−∣c−⟩⟨c−∣\rho=p_+|c_+\rangle\langle c_+|+p_-|c_-\rangle\langle c_-|. Since the two cat states are orthogonal, its linear purity is

P=p+2−p−2=1+C22,C=∣p+−p−∣.(29)\mathcal{P}=p_+^2-p_-^2=\frac{1+C^2}{2},\qquad C=|p_+-p_-|. \tag*{(29)}

For a windowed measurement with qq resolved, p+=1p_+=1 or p−=1p_-=1, so C=1C=1 and P=1\mathcal{P}=1.

Exact calculation at finite ∣β∣|\beta|. To go beyond the strong-LO limit, we computed the remote state with the exact Kraus operators of Eq. (3). We used a two-mode squeezed vacuum ∝∑kλk∣k⟩∣k⟩\propto\sum_k\lambda^k|k\rangle|k\rangle with λ=tanh⁡r\lambda=\tanh r, which becomes the ideal EPR state for λ→1\lambda\to1. Here λ=0.882\lambda=0.882 for 12 dB and λ=0.6\lambda=0.6 for 6 dB of squeezing. For an outcome (n,m)(n,m) the unnormalized remote state is ∑kλk⟨0∣Mn,m∣k⟩∣k⟩\sum_k\lambda^k\langle0|M_{n,m}|k\rangle|k\rangle. When xx is resolved and only ww is windowed, the remote state is the mixture of the states with the same n−mn-m and with n+mn+m in the window, weighted by their probabilities. We evaluate its purity

P=∑ijpipj∣⟨ψi∣ψj⟩∣2.\mathcal{P}=\sum_{ij}p_i p_j\left|\langle\psi_i|\psi_j\rangle\right|^2.

Figure 5 (top) shows the result for 12 dB of squeezing. With xx resolved, the purity decreases slowly with δ\delta and approaches 1 as ∣β∣|\beta| grows. For δ=1\delta=1 we find P=0.964\mathcal{P}=0.964, 0.970, 0.986, 0.993 for ∣β∣=3,5,8,12|\beta|=3,5,8,12. For 6 dB of squeezing, P≥0.98\mathcal{P}\geq0.98 for all ∣β∣≥2|\beta|\geq2 up to δ=1\delta=1. The window therefore acts as a weak control of purity at finite ∣β∣|\beta|, and not as a switch between a pure and a mixed state. When xx is not recorded (Figure 5, bottom), the purity drops to about 0.15 to 0.55 for ∣β∣=2|\beta|=2 to 5, because states with different qq and parity are mixed.

For the ideal EPR state (λ→1\lambda\to1) the purity does not approach 1 as ∣β∣|\beta| grows at fixed δ\delta (for λ=0.99\lambda=0.99 we find P≈0.97\mathcal{P}\approx0.97 and 0.90 at δ=0.5\delta=0.5 and 1 for ∣β∣=12|\beta|=12). This is expected, since the ideal EPR state has unbounded photon number, for which the strong-LO approximation is not uniformly valid [8]. Finite squeezing is therefore essential for the conclusions above.

Limitations: we assumed ideal photon-number-resolving detectors, a single mode, and a two-mode squeezed vacuum as the resource. The practical consequence is positive: the window on ww can be chosen freely, for instance to match a finite detector resolution in the sum variable, with only a small loss of purity of the prepared cat state, provided xx is resolved. This is relevant for error-correcting codes based on cat states [13, 14].

Experimental Feasibility

Realizing the windowed measurement requires photon-number-resolving detectors that resolve single photons in both outputs, because the exact windowed quantities depend on the integer value of n+mn + m (Secs. 3.5 and § 5). The window width in photon-number units is 22∣β∣δ2\sqrt{2}|\beta|\delta. For ∣β∣=5|\beta| = 5 and δ=0.1\delta= 0.1 it is about 1.4, so the window contains only one or two integer values of n+mn + m. In this regime the continuum formulas are not valid and the exact discrete expressions must be used.

Current detectors can resolve more than 15 photons with reasonable efficiency [11]. This is at the edge of what is needed for ∣β∣=5|\beta| = 5, where each detector registers on average 12.5 photons. A cat state with ∣β∣=5|\beta| = 5 (mean photon number 25) is also well beyond current optical experiments. A smaller amplitude such as ∣β∣=2|\beta| = 2 (mean photon number 4) is more realistic, and it is also the regime where the discrete and continuum descriptions differ most. In the microwave domain, photon-number resolution inside a cavity is well established [12], and Combes and Lund consider a realization there plausible [8].

Limitations. Our analysis assumes ideal detectors. The parity structure is encoded in outcomes that differ by a single photon, so photon loss and detector inefficiency, which shift nn or mm by one, reduce the visibility of the parity signal and flip the relative sign σ\sigma of the remote cat state [8]. Their effect should be quantified in future work. Furthermore, our analysis is restricted to single-mode fields; an extension to multimode fields is needed for practical applications.

Conclusions

We analyzed catodyne measurements in which only a finite window of the photon-sum variable is retained. Our main results are:

  1. Coherence conditioned on xx is preserved in the strong-LO regime. On the discrete photon-number lattice, windowing the sum variable does not reduce the coherence conditioned on the difference outcome. The apparent decay of coherence in the continuum approximation is an artifact of treating the sum variable as continuous.

  1. Parity-sensitive acceptance. The exact windowed success probability depends on the parity of the input state. The parity signal is a staircase function of the window width, and its magnitude and plateau structure differ substantially from the continuum prediction.

  1. Limits of the continuum approximation. For the success probability, the continuum formulas reproduce the average behavior only for large ∣β∣|\beta|; for small ∣β∣|\beta| and narrow windows, which are the relevant regime for photon-number-resolving detectors, the discrete structure must be used. The amplitude of the parity signal is underestimated by the continuum formula even for ∣β∣=10|\beta| = 10.

For remote state preparation, the conditional cat state keeps a purity close to 1 when xx is resolved (computed exactly for finite squeezing), and loses purity mainly when xx is not recorded. Our analysis assumes ideal detectors and a single mode; detector inefficiency and multimode effects remain to be studied.

References

  1. [1]T. Tyc and B. C. Sanders, Operational formulation of homodyne detection, J. Phys. A: Math. Gen. 37, 7341 (2004).
  2. [2]R. Kumar, E. Barrios, A. MacRae, E. Cairns, E. H. Huntington, and A. I. Lvovsky, Versatile wideband balanced detector for quantum optical homodyne tomography, Opt. Commun. 285, 5259 (2012).DOI
  3. [3]N. Walker, Quantum theory of multiport optical homodyning, J. Mod. Opt. 34, 15 (1987).DOI
  4. [4]S. L. Braunstein and P. van Loock, Quantum information with continuous variables, Rev. Mod. Phys. 77, 513 (2005).DOI
  5. [5]B. C. Sanders, K. S. Lee, and M. S. Kim, Optical homodyne measurements and entangled coherent states, Phys. Rev. A 52, 735 (1995).
  6. [6]G. S. Thekkadath, B. A. Bell, I. A. Walmsley, and A. I. Lvovsky, Engineering Schrödinger cat states with a photonic even-parity detector, Quantum. 4, 239 (2020).
  7. [7]M. S. Kim and B. C. Sanders, Squeezing and antisqueezing in homodyne measurements, Phys. Rev. A 53, 3694 (1996).pubmed.ncbi.nlm.nih.gov/9913321
  8. [8]J. Combes and A. P. Lund, Homodyne measurement with a Schrödinger cat state as a local oscillator, Phys. Rev. A. 106, 063706 (2022).DOI
  9. [9]G. S. Thekkadath, D. S. Phillips, J. F. F. Bulmer, W. R. Clements, A. Eckstein, B. A. Bell, J. Lugani, T. A. W. Wolterink, A. Lita, S. W. Nam, T. Gerrits, C. G. Wade, and I. A. Walmsley, Tuning between photon-number and quadrature measurements with weak-field homodyne detection, Phys. Rev. A.101, 031801(R) (2020).DOI
  10. [10]S. Olivares, A. Allevi, and M. Bondani, On the role of the local oscillator intensity in optical homodyne-like tomography, Phys. Lett. A. 384, 126354 (2020).DOI
  11. [11]L. A. Morais, T. Weinhold, M. P. de Almeida, J. Combes, A. Lita, T. Gerrits, S. W. Nam, A. G. White, and G. Gittel, Precisely determining photon-number in real-time, arXiv: 2012.10158.
  12. [12]D. I. Schuster, A. A. Houck, J. A. Schreier, A. Wallraff, J. M. Gambetta, A. Blais, L. Frunzio, J. Majer, B. Johnson, M. H. Devoret, S. M. Girvin, and R. J. Schoelkopf, Resolving photon number states in a superconducting circuit, Nature. 445, 515 (2007).pubmed.ncbi.nlm.nih.gov/17268464
  13. [13]D. S. Schlegel, F. Minganti, and V. Savona, Quantum error correction using squeezed Schrödinger cat states, Phys. Rev. A. 106, 022431 (2022).DOI
  14. [14]Q. Xu, G. Zheng, Y.-X. Wang, P. Zoller, A. A. Clerk, and L. Jiang, Autonomous quantum error correction and fault-tolerant quantum computation with squeezed cat qubits, arXiv: 2210.13406.arxiv.org/abs/2210.13406

Paper details

vlnc:W_698xafpm
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents