introduction
Standard homodyne detection is a key technique for characterizing quantum states of light [1 ] . In this scheme, the unknown state is combined with a high-intensity coherent state, known as the local oscillator (LO) or phase reference, on a balanced beam splitter (BS) at various relative phase settings spanning the full angular range. The output beams are detected by photodiodes, producing electrical signals that are integrated over time and subtracted. In the limit of infinite LO intensity, the result corresponds to a quadrature measurement of the unknown state in the mode matching the LO at the given phase. Here, “mode” typically refers to a normalized spatially localized solution of Maxwell’s equations in free space or a dielectric medium such as an optical fiber. Measuring quadratures across all phases allows for reconstruction of the reduced density matrix of the state in the mode matching the LO [2 , 3 ] .
Standard homodyne detection has some limitations - the most obvious being the need for a high-intensity reference, which can potentially disturb sensitive quantum information in systems such as Rydberg clouds [4 , 5 , 6 , 7 ] , in quantum non-demolition experiments probing atomic ensembles [8 , 9 , 10 ] , and in non-destructive cavity QED experiments [11 , 12 ] . Another obvious limitation is the blindness of the scheme to the modes orthogonal to the LO. These two features of standard homodyne detection are closely related. Even if photodiodes were ideal intensity detectors, in the weak LO regime the final processed signal would no longer correspond to a quadrature measurement, but would contain information about the modes orthogonal to the LO.
Photon-number-resolving (PNR) detectors, also referred to as photon counters, serve a role similar to photodiodes but are designed for situations where, on average, only a few photons are present in the absorbed pulse. [13 , 14 ] Current state-of-the-art PNR detectors, such as transition-edge-sensors (TESs), can distinguish up to sixteen photons in a single pulse [15 ] . Further, a combination of TESs in a multiplexed scheme, acting as an effective PNR detector, can resolve up to 100 photons [16 ] . These advances motivate the exploration of homodyne-like measurement schemes in which photodiodes are replaced by PNR detectors, and the LO is attenuated to contain only a few photons on average.
These schemes, often referred to as “hybrid” or “weak-field homodyne” detectors [13 ] , have been studied both theoretically and experimentally. Early theoretical work showed that a single PNR detector and an unbalanced BS suffice to determine an unknown single-mode state [17 ] . Experimentally, such schemes have been used to measure two-photon phase coherence [18 ] , test Bell inequalities [19 ] , create superpositions of vacuum and single-photon states with well-defined phases [20 ] , perform state discrimination [21 , 22 , 23 ] , and estimate entanglement in bipartite systems [24 ] . Further demonstrations include state reconstruction for single-mode [25 , 26 , 27 , 28 , 29 ] and two-mode squeezed states [30 ] . Other studies have experimentally characterized weak-field homodyne detectors under the assumption that there are no photons in detectable modes other than the mode matching the LO [31 , 32 ] . Interestingly, it was shown that, in principle, an arbitrary multimode state can be determined using one single-photon detector, provided the LO can be prepared in any desired mode [33 ] . Recently, there has also been interest in exploring detection schemes where the LO is not a coherent state, but rather a non-classical state such as a cat state [34 ] .
A common assumption across these studies of weak-field homodyne is that either there is only one, known mode containing photons and this mode matches that of the LO, or the set of modes of interest is known and the LO can be matched to or scanned over them. The former does not account for photons in other modes, while the latter has substantial measurement overhead. For a fixed LO mode, a desirable goal of weak-field homodyne is to not only determine the marginal state of the mode, but also its correlation with photons in other modes, without the need to change or scan the LO mode. Correlations with photons in other modes can yield valuable insights, whereas neglecting them, especially when their content is non-negligible, can lead to significant errors in estimation. Strengthening the power of weak-field homodyne to achieve these goals is one of the key motivations for the present work.
We theoretically investigate what can be learned about an unknown state with an unspecified mode structure by combining it with a LO in a weak coherent state on a BS (not necessarily balanced) and performing photon counting on one or both output ports. We assume that the LO is pulsed, where the LO mode is defined by the pulse shape. The PNR detectors count photons during a time window containing the LO pulse, and they are equally sensitive to all modes that may have photons in this time window. For this work, we assume that the probabilities of all photon number outcomes are learned. This is the infinite data limit for ideal PNR detectors. Perfect PNR detectors are not required in the infinite data limit. For example, inefficiency or loss and added noise can be modeled as classical noise processes following perfect photon counting, and it is often possible to invert the classical noise process given the count probabilities [35 ] .
Our analysis of the infinite data limit establishes what is fundamentally inferrable about the unknown state for the experimental configurations considered. We find that, for a fixed LO mode, the measurement statistics yield significantly more information about the unknown state than is possible with standard homodyne detection. In particular, it is possible to determine the unknown state in the mode matching the LO conditional on each number of photons present in the orthogonal modes, and one may also determine the photon number probabilities in those modes. We find that the measurement statistics from a single PNR detector, collected over a sufficient range of LO magnitudes and phases, are sufficient to extract this information. When two PNR detectors are used, a fixed-magnitude LO is sufficient to extract this information provided the measurement statistics are known for all phase values.
Overview
We begin by summarizing the main theoretical results established in this work. Our analysis considers an unknown multimode input state ρ \rho ρ combined with a weak LO in a fixed mode on an arbitrary lossless BS, followed by detection with one or two noiseless and lossless PNR detectors that do not distinguish between the modes in their respective BS ports (Fig. 1) in the infinite data limit. Throughout the text, such idealized PNR detectors are referred to simply as counters. § 0.3 provides a detailed description of the measurement configuration, relevant physical quantities, and the main mathematical instruments used in deriving the results. Following this setup, we make a key observation in § 0.4 that the measurement statistics exhibit a nontrivial symmetry. Specifically, Prop. 4.1 and Prop. 4.2 together establish that the statistics are invariant under precisely those passive linear transformations (PLTs) on the input state that act trivially on the LO-matched mode (mode 1). Averaging over the symmetry group yields the so-called twirled-state, given in Thm. 4.3 , with Cor. 4.4 showing that there exists a two-mode representative ρ t \rho_t ρ t that is indistinguishable from ρ \rho ρ . We refer to ρ t \rho_t ρ t as the two-mode twirled state corresponding to ρ \rho ρ .
§ 1 then addresses the central question of whether the measurement statistics determine ρ t \rho_t ρ t . The analysis is based on representing the measurement statistics by the moments of the total photon-number operators ( n ^ a ′ (\hat{n}'_a ( n ^ a ′ and n ^ b ′ ) \hat{n}'_b) n ^ b ′ ) measured by the two counters. This in turn imposes a requirement on ρ t \rho_t ρ t : its Husimi function must be determined by its moments. We next show with Prop. 5.2 that, in order to uniquely determine ρ t \rho_t ρ t , the normally ordered moments ⟨ ( a ^ 1 † ) k a ^ 1 ℓ ( a ^ 2 † ) n a ^ 2 n ⟩ \langle(\hat{a}_1^\dagger)^k\hat{a}_1^\ell(\hat{a}_2^\dagger)^n\hat{a}_2^n\rangle ⟨( a ^ 1 † ) k a ^ 1 ℓ ( a ^ 2 † ) n a ^ 2 n ⟩ for k ≥ ℓ k \ge\ell k ≥ ℓ and arbitrary n n n are both sufficient and necessary. Here, a ^ 1 \hat{a}_1 a ^ 1 and a ^ 2 \hat{a}_2 a ^ 2 denote the annihilation operators for modes 1 and 2, respectively. For input states with bounded photon number, Cor. 5.3 establishes that only finitely many of these moments are required. These results rely on the operator-ordering relation of Lem. 5.1 , which establishes a one-to-one correspondence between the Husimi moments of ρ t \rho_t ρ t and the moments ⟨ ( a ^ 1 † ) k a ^ 1 ℓ ( a ^ 2 † ) n a ^ 2 n ⟩ \langle(\hat{a}_1^\dagger)^k\hat{a}_1^\ell(\hat{a}_2^\dagger)^n\hat{a}_2^n\rangle ⟨( a ^ 1 † ) k a ^ 1 ℓ ( a ^ 2 † ) n a ^ 2 n ⟩ .
Within this framework, we obtain detailed results for both the single- and two-counter configurations by asking whether the measurement statistics, known for a given set of probe LO states, determine the moments ⟨ ( a ^ 1 † ) k a ^ 1 ℓ ( a ^ 2 † ) n a ^ 2 n ⟩ \langle(\hat{a}_1^\dagger)^k\hat{a}_1^\ell(\hat{a}_2^\dagger)^n\hat{a}_2^n\rangle ⟨( a ^ 1 † ) k a ^ 1 ℓ ( a ^ 2 † ) n a ^ 2 n ⟩ . For a single counter, Thm. 5.9 shows that access to statistics across a neighborhood of probe amplitudes suffices to recover all required moments, thereby determining ρ t \rho_t ρ t . When the maximum photon number of the input state is bounded by N N N , Thm. 5.7 and Cor. 5.8 relaxes this requirement, showing that ( N + 1 ) 2 (N+1)^2 ( N + 1 ) 2 probe amplitudes are sufficient. If there is no bound on the maximum number of photons, Prop. 5.10 establishes that no finite set of probe amplitudes suffices. Moreover, for fixed-magnitude probes with variable phase, Prop. 5.11 shows that the single-counter scheme is limited to reconstructing states with at most one photon.
The two-counter configuration is strictly more efficient in terms of the number of measurement settings required to obtain the same information. While Prop. 5.13 shows that finitely many probes are still insufficient to determine ρ t \rho_t ρ t without a photon-number bound, Thm. 5.14 demonstrates that collecting measurement statistics by scanning the probe phase at fixed magnitude recovers ρ t \rho_t ρ t in full. The final result, Prop. 5.15, establishes a connection between the two-counter measurement scheme and standard homodyne detection. It shows that the reduced state of the system is fully characterized by the expectations of the normally ordered operators : ( n ^ b ′ − ξ n ^ a ′ ) n : :(\hat{n}'_b-\xi\hat{n}'_a)^n: : ( n ^ b ′ − ξ n ^ a ′ ) n : for all n n n . Here, ξ \xi ξ is a parameter determined by the BS, taking the value ξ = 1 \xi=1 ξ = 1 for a balanced BS. Homodyne detection is recovered in the limit of a strong LO, where ⟨ : ( n ^ b ′ − n ^ a ′ ) n : ⟩ ≈ ⟨ ( n ^ b ′ − n ^ a ′ ) n ⟩ \langle:(\hat{n}'_b-\hat{n}'_a)^n:\rangle\approx\langle(\hat{n}'_b-\hat{n}'_a)^n\rangle ⟨ : ( n ^ b ′ − n ^ a ′ ) n : ⟩ ≈ ⟨( n ^ b ′ − n ^ a ′ ) n ⟩ . In the appendix we provide basic results involving normal and antinormal ordering and PLTs, and we show the Husimi functions of Gaussian states, Fock states and their superpositions and mixtures are determined by their moments.
Preliminaries
Description of the Measurement Configuration
The schematic of the measurement configuration is shown in Figure 1 . The unknown input state ρ \rho ρ and the LO probe state σ \sigma σ are directed toward the two input ports of the BS. The input ρ \rho ρ lives in S S S mutually orthogonal modes, characterized by S S S annihilation operators a ^ 1 , … , a ^ S \hat{a}_1,\ldots,\hat{a}_S a ^ 1 , … , a ^ S . We do not assume that S S S is known, but require that it be finite for our analysis. The BS induces interference between modes in its two input ports that share identical spatio-temporal and polarization characteristics. Thus, there exist S S S modes in the LO input port, characterized by annihilation operators b ^ 1 , … , b ^ S \hat{b}_1,\ldots,\hat{b}_S b ^ 1 , … , b ^ S , where each b ^ i \hat{b}_i b ^ i is matched with a ^ i \hat{a}_i a ^ i .
Figure 1. Diagram of the measurement configuration with both PNR detectors present. The unknown input state ρ \rho ρ and probe state σ \sigma σ are directed towards the inputs of the BS, and the outputs are measured by two PNR detectors, denoted by counter 1 and counter 2. The detectors measure the total-photon-number operators n ^ a ′ \hat{n}'_a n ^ a ′ and n ^ b ′ \hat{n}'_b n ^ b ′ in their respective output paths of the BS. If the measurement configuration uses only one detector, then only counter 1 is present.
We denote the physical Hilbert space of the input state by H a \mathcal{H}_a H a , and of the probe state by H b \mathcal{H}_b H b . We further denote the Hilbert spaces of the output modes of the BS associated with the operators a ^ i ′ \hat{a}'_i a ^ i ′ and b ^ i ′ \hat{b}'_i b ^ i ′ by H a ′ \mathcal{H}_{a'} H a ′ and H b ′ \mathcal{H}_{b'} H b ′ , respectively. We denote the vector of annihilation operators ( a ^ 1 , … , a ^ S ) (\hat{a}_1,\ldots,\hat{a}_S) ( a ^ 1 , … , a ^ S ) by a ^ \hat{\mathbf{a}} a ^ , and the vector of creation operators ( a ^ 1 † , … , a ^ S † ) (\hat{a}_1^\dagger,\ldots,\hat{a}_S^\dagger) ( a ^ 1 † , … , a ^ S † ) by a ^ † \hat{\mathbf{a}}^\dagger a ^ † , and similarly for the operators in the other ports of the BS. We generally interpret vectors written inline as column vectors. We refer to creation and annihilation operators as mode operators. Given any two vectors of mode operators, say a ^ = ( a ^ 1 , … , a ^ S ) \hat{\mathbf{a}}=(\hat{a}_1,\ldots,\hat{a}_S) a ^ = ( a ^ 1 , … , a ^ S ) and b ^ = ( b ^ 1 , … , b ^ S ) \hat{\mathbf{b}}=(\hat{b}_1,\ldots,\hat{b}_S) b ^ = ( b ^ 1 , … , b ^ S ) , the direct sum a ^ ⊕ b ^ \hat{\mathbf{a}}\oplus\hat{\mathbf{b}} a ^ ⊕ b ^ is defined as the concatenated vector ( a ^ 1 , … , a ^ S , b ^ 1 , … , b ^ S ) (\hat{a}_1,\ldots,\hat{a}_S,\hat{b}_1,\ldots,\hat{b}_S) ( a ^ 1 , … , a ^ S , b ^ 1 , … , b ^ S ) .
For any vector of K K K annihilation operators c ^ = ( c ^ 1 , … , c ^ K ) \hat{\mathbf{c}}=(\hat{c}_1,\ldots,\hat{c}_K) c ^ = ( c ^ 1 , … , c ^ K ) with associated Hilbert space H c \mathcal{H}_c H c , we refer to the K K K -dimensional vector space spanned by its components c ^ i \hat{c}_i c ^ i as the mode space of H c \mathcal{H}_c H c . We denote the Fock state with n i n_i n i photons in mode c ^ i \hat{c}_i c ^ i by ∣ n 1 , … , n K ⟩ = ⨂ i = 1 K ∣ n i ⟩ |n_1,\ldots,n_K\rangle=\bigotimes_{i=1}^{K}|n_i\rangle ∣ n 1 , … , n K ⟩ = ⨂ i = 1 K ∣ n i ⟩ . Similarly, a coherent state with complex amplitude α i \alpha_i α i in mode c ^ i \hat{c}_i c ^ i is written as ∣ α ⟩ = ⨂ i = 1 K ∣ α i ⟩ |\boldsymbol{\alpha}\rangle=\bigotimes_{i=1}^{K}|\alpha_i\rangle ∣ α ⟩ = ⨂ i = 1 K ∣ α i ⟩ . We use Greek letters in bra-ket notation to indicate coherent states and Latin letters for Fock states. Throughout this work, the probe state is the coherent state σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⃗ ⟩ ⟨ 0 ⃗ ∣ \sigma=|\gamma\rangle\langle\gamma|\otimes|\vec{0}\rangle\langle\vec{0}| σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⟩ ⟨ 0 ∣ , where ∣ 0 ⃗ ⟩ |\vec{0}\rangle ∣ 0 ⟩ denotes the vacuum state for modes b ^ 2 , … , b ^ S \hat{b}_2,\ldots,\hat{b}_S b ^ 2 , … , b ^ S .
The transformation induced by the BS is described by a 2 × 2 2\times2 2 × 2 unitary matrix B 1 = ( η ζ ζ ′ η ′ ) B_1=\begin{pmatrix}\eta&\zeta\\\zeta'&\eta'\end{pmatrix} B 1 = ( η ζ ′ ζ η ′ ) that, in the Heisenberg picture, combines each pair of matching modes ( a ^ i , b ^ i ) (\hat{a}_i,\hat{b}_i) ( a ^ i , b ^ i ) into two output modes ( a ^ i ′ , b ^ i ′ ) (\hat{a}'_i,\hat{b}'_i) ( a ^ i ′ , b ^ i ′ ) as follows: ( a ^ i ′ , b ^ i ′ ) = B 1 ⋅ ( a ^ i , b ^ i ) (\hat{a}'_i,\hat{b}'_i)=B_1\cdot(\hat{a}_i,\hat{b}_i) ( a ^ i ′ , b ^ i ′ ) = B 1 ⋅ ( a ^ i , b ^ i ) . Here, the central dot denotes the standard product between a matrix and a vector. We assume η ≠ 0 \eta\ne0 η = 0 and ζ ≠ 0 \zeta\ne0 ζ = 0 to exclude trivial BSs. The corresponding unitary operator on the Hilbert space of states is denoted by B \mathcal{B} B , so that a ^ i ′ = B † a ^ i B \hat{a}'_i=\mathcal{B}^{\dagger}\hat{a}_i\mathcal{B} a ^ i ′ = B † a ^ i B and b ^ i ′ = B † b ^ i B \hat{b}'_i=\mathcal{B}^{\dagger}\hat{b}_i\mathcal{B} b ^ i ′ = B † b ^ i B . The action of the BS on a coherent state ∣ α ⃗ ⟩ ∣ β ⃗ ⟩ ∈ H a ⊗ H b |\vec{\alpha}\rangle|\vec{\beta}\rangle\in\mathcal{H}_a\otimes\mathcal{H}_b ∣ α ⟩ ∣ β ⟩ ∈ H a ⊗ H b produces the coherent state ∣ α ⃗ ′ ⟩ ∣ β ⃗ ′ ⟩ ∈ H a ′ ⊗ H b ′ |\vec{\alpha}'\rangle|\vec{\beta}'\rangle\in\mathcal{H}_{a'}\otimes\mathcal{H}_{b'} ∣ α ′ ⟩ ∣ β ′ ⟩ ∈ H a ′ ⊗ H b ′ , where ( α i ′ , β i ′ ) = B 1 ⋅ ( α i , β i ) (\alpha'_i,\beta'_i)=B_1\cdot(\alpha_i,\beta_i) ( α i ′ , β i ′ ) = B 1 ⋅ ( α i , β i ) . The outputs of the BS are directed toward two counters, referred to as counter 1 and counter 2. Counter 1 measures the total-photon-number operator n ^ a ′ = ∑ i = 1 S ( a ^ i ′ ) † a ^ i ′ \hat{n}'_a=\sum_{i=1}^{S}(\hat{a}'_i)^\dagger\hat{a}'_i n ^ a ′ = ∑ i = 1 S ( a ^ i ′ ) † a ^ i ′ , and counter 2 measures n ^ b ′ = ∑ i = 1 S ( b ^ i ′ ) † b ^ i ′ \hat{n}'_b=\sum_{i=1}^{S}(\hat{b}'_i)^\dagger\hat{b}'_i n ^ b ′ = ∑ i = 1 S ( b ^ i ′ ) † b ^ i ′ . If only one of the counters is used, we assume counter 2 is absent. This entails no loss of generality, as omitting counter 1 instead is equivalent to omitting counter 2 up to a change of the BS coefficients.
Unitary transformations that act linearly on the annihilation operators are called passive linear transformations (PLTs). For example, the BS induces a PLT on H a ⊗ H b \mathcal{H}_a\otimes\mathcal{H}_b H a ⊗ H b . Any PLT operator U U U acting on the input state is determined by a unitary matrix U U U acting on vectors of mode operators or coherent state amplitudes. We adopt the convention introduced implicitly for BSs above: For PLTs, the unitary operator acting on the physical Hilbert space is denoted by the calligraphic version of the symbol used for the unitary matrix acting on mode vectors. For example, if U U U is the unitary matrix for a PLT transforming the vector of annihilation operators c ^ \hat{\mathbf{c}} c ^ as c ^ ↦ U ⋅ c ^ \hat{\mathbf{c}}\mapsto U\cdot\hat{\mathbf{c}} c ^ ↦ U ⋅ c ^ , then U U U is the corresponding unitary operator on the Hilbert space H c \mathcal{H}_c H c . In particular, U † c ^ i U = ∑ j U i j c ^ j U^\dagger\hat{c}_iU=\sum_jU_{ij}\hat{c}_j U † c ^ i U = ∑ j U ij c ^ j . We refer to the action of the unitary matrix U U U on c ^ \hat{\mathbf{c}} c ^ as the action or representation of the PLT in the mode space of H c \mathcal{H}_c H c . For instance, the BS operator B \mathcal{B} B is represented in mode-space basis a ^ ⊕ b ^ \hat{\mathbf{a}}\oplus\hat{\mathbf{b}} a ^ ⊕ b ^ by the unitary matrix B ≡ B 1 ⊗ I S B\equiv B_1\otimes I_S B ≡ B 1 ⊗ I S , with I S I_S I S being the S × S S\times S S × S identity. This is because B 1 ⊗ I S B_1\otimes I_S B 1 ⊗ I S transforms each pair ( a ^ i , b ^ i ) (\hat{a}_i,\hat{b}_i) ( a ^ i , b ^ i ) into B 1 ⋅ ( a ^ i , b ^ i ) B_1 \cdot(\hat{a}_i,\hat{b}_i) B 1 ⋅ ( a ^ i , b ^ i ) in this basis.
Operator Ordering and Correspondence with Expressions of Complex Variables
The measurement statistics are the joint photon-counting probabilities p i j = ⟨ Π i ⊗ Π j ⟩ p_{ij} = \langle\Pi_i \otimes\Pi_j \rangle p ij = ⟨ Π i ⊗ Π j ⟩ , where Π i \Pi_i Π i (Π j \Pi_j Π j ) denotes the projector onto the subspace of H a ′ \mathcal{H}_{a'} H a ′ (H b ′ \mathcal{H}_{b'} H b ′ ) containing exactly i i i (j j j ) photons. The operators Π i ⊗ Π j \Pi_i \otimes\Pi_j Π i ⊗ Π j are the measurement operators of the measurement configuration. The expectation defining p i j p_{ij} p ij is with respect to ρ ⊗ σ \rho\otimes\sigma ρ ⊗ σ since the total photon-number operators are given in the Heisenberg picture using the evolved mode operators. From the collection of joint probabilities ( p i j ) i j (p_{ij})_{ij} ( p ij ) ij , one can determine the moments ⟨ ( n ^ a ′ ) k ( n ^ b ′ ) l ⟩ \langle(\hat{n}'_a)^k(\hat{n}'_b)^l\rangle ⟨( n ^ a ′ ) k ( n ^ b ′ ) l ⟩ of the total photon-number operators n ^ a ′ \hat{n}'_a n ^ a ′ and n ^ b ′ \hat{n}'_b n ^ b ′ . For the space of states considered in this work and introduced in Sect. 3.4, these moments are finite, see Lem. 3.1 .
In our analysis, it is convenient to work instead with the normally ordered moments ⟨ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ⟩ \langle:(\hat{n}'_a)^k(\hat{n}'_b)^l:\rangle ⟨ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ⟩ . Here, the double colons “: : :: :: ” denote normal ordering: for any operator P ^ \hat{P} P ^ that is a linear combination of mode operator products, the operator : P ^ : :\hat{P}: : P ^ : is obtained by rearranging each term so that all creation operators appear to the left of annihilation operators. It is crucial to emphasize that normal ordering acts on the algebraic form of an operator expression, rather than the operator itself. Because the canonical commutation relation [ a ^ i , a ^ j † ] = δ i j [\hat{a}_i,\hat{a}_j^\dagger] = \delta_{ij} [ a ^ i , a ^ j † ] = δ ij permits multiple equivalent representations of the same operator, applying normal ordering to these different representations produces distinct operators. In accordance with standard practice, when we write a normally ordered operator : P ^ : :\hat{P}: : P ^ : in closed form, we apply the normal ordering to the fully expanded polynomial of the given expression for P ^ \hat{P} P ^ obtained strictly without invoking any commutation relations. For example, the normally ordered moment ⟨ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ⟩ \langle:(\hat{n}'_a)^k(\hat{n}'_b)^l:\rangle ⟨ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ⟩ is obtained by fully expanding the powers of the number operators, each expressed in the standard form, that is n ^ i ′ = ∑ i = 1 S ( a ^ i ′ ) † a ^ i ′ \hat{n}'_i = \sum_{i=1}^{S}(\hat{a}'_i)^\dagger\hat{a}'_i n ^ i ′ = ∑ i = 1 S ( a ^ i ′ ) † a ^ i ′ , and then multiplying the two expansions and then applying normal ordering.
In addition to normal ordering, we also take advantage of anti-normal ordering. For anti-normal ordering, creation operators are moved to the right of annihilation operators. We denote the anti-normally ordered operator obtained from an operator expression P ^ \hat{P} P ^ by : P ^ : − :\hat{P}:_- : P ^ : − . The comments about normal ordering in the previous paragraph also apply to anti-normal ordering.
Each : ( n ^ a ′ ) k ( n ^ b ′ ) l : :(\hat{n}'_a)^k(\hat{n}'_b)^l: : ( n ^ a ′ ) k ( n ^ b ′ ) l : can be expressed as a finite linear combination of the ordinary moments ( n ^ a ′ ) n ( n ^ b ′ ) m (\hat{n}'_a)^n(\hat{n}'_b)^m ( n ^ a ′ ) n ( n ^ b ′ ) m and vice-versa, see Appendix B. Thus, the ordinary moments of the measurement statistics are finite if and only if the normally-ordered moments ⟨ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ⟩ \langle:(\hat{n}'_a)^k(\hat{n}'_b)^l:\rangle ⟨ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ⟩ are finite.
We use two key properties of normally ordered operators. First, given any K K K mode operators c ^ = ( c ^ 1 , … , c ^ K ) \hat{\mathbf{c}} = (\hat{c}_1,\ldots,\hat{c}_K) c ^ = ( c ^ 1 , … , c ^ K ) , the expectation of a normally-ordered product ( ⨂ i = 1 K ( c ^ i † ) n i ) ( ⨂ j = 1 K c ^ j m j ) \left(\bigotimes_{i=1}^{K}(\hat{c}_i^\dagger)^{n_i}\right)\left(\bigotimes_{j=1}^{K}\hat{c}_j^{m_j}\right) ( ⨂ i = 1 K ( c ^ i † ) n i ) ( ⨂ j = 1 K c ^ j m j ) with a coherent state ∣ δ ⃗ ⟩ \lvert\vec{\delta}\rangle ∣ δ ⟩ , where ∣ δ ⃗ ⟩ \lvert\vec{\delta}\rangle ∣ δ ⟩ is the coherent state in mode c ^ i \hat{c}_i c ^ i , is the monomial ∏ i = 1 K ( δ i ∗ ) n i ∏ j = 1 K δ j m j \prod_{i=1}^{K}(\delta_i^*)^{n_i}\prod_{j=1}^{K}\delta_j^{m_j} ∏ i = 1 K ( δ i ∗ ) n i ∏ j = 1 K δ j m j . Thus, for every normally ordered operator P ^ \hat{P} P ^ we have a corresponding polynomial expression E [ P ^ ] \mathcal{E}[\hat{P}] E [ P ^ ] in the variables δ ⃗ \vec{\delta} δ and δ ⃗ ∗ \vec{\delta}^{*} δ ∗ with values E [ P ^ ] ( δ ⃗ , δ ⃗ ∗ ) = ⟨ δ ⃗ ∣ P ^ ∣ δ ⃗ ⟩ \mathcal{E}[\hat{P}](\vec{\delta},\vec{\delta}^{*}) = \langle\vec{\delta}\lvert\hat{P}\rvert\vec{\delta}\rangle E [ P ^ ] ( δ , δ ∗ ) = ⟨ δ ∣ P ^ ∣ δ ⟩ . We refer to E [ P ^ ] \mathcal{E}[\hat{P}] E [ P ^ ] as the amplitude polynomial of P ^ \hat{P} P ^ . Conversely, given such a polynomial expression we can compute the normally ordered operator P ^ \hat{P} P ^ from which it is obtained. For a polynomial p p p in the variables δ ⃗ , δ ⃗ ∗ \vec{\delta},\vec{\delta}^{*} δ , δ ∗ , we define E − 1 [ p ] \mathcal{E}^{-1}[p] E − 1 [ p ] to be the normally ordered operator P ^ \hat{P} P ^ for which E [ E − 1 [ p ] ] = p \mathcal{E}[\mathcal{E}^{-1}[p]] = p E [ E − 1 [ p ]] = p . The maps P ^ ↦ E [ P ^ ] \hat{P} \mapsto\mathcal{E}[\hat{P}] P ^ ↦ E [ P ^ ] and p ↦ E − 1 [ p ] p \mapsto\mathcal{E}^{-1}[p] p ↦ E − 1 [ p ] are linear and identify the vector space spanned by normally ordered operators with the vector space of polynomials in δ ⃗ \vec{\delta} δ and δ ⃗ ∗ \vec{\delta}^{*} δ ∗ . We refer to this identification of vector spaces as the operator-polynomial correspondence. As operators on the Hilbert space, the vector space spanned by normally ordered operators is the same as the vector space spanned by general products of mode operators and their adjoints.
The second key property is that PLTs map normally ordered operators to normally ordered operators. This follows from the fact that PLTs transform creation operators into linear combinations of creation operators, and annihilation operators into linear combinations of annihilation operators. More generally, PLTs commute with the normal-ordering operation, see Appendix A for a formal proof. For instance, : ( n ^ a ′ ) k ( n ^ b ′ ) l : :(\hat{n}'_a)^k(\hat{n}'_b)^l: : ( n ^ a ′ ) k ( n ^ b ′ ) l : can be expressed as
: ( n ^ a ′ ) k ( n ^ b ′ ) l : = : ( ∑ i = 1 S ( a ^ i ′ ) † a ^ i ′ ) k ( ∑ j = 1 S ( b ^ j ′ ) † b ^ j ′ ) l : = : ( ∑ i = 1 S B † ( a ^ i ) † B B † a ^ i B ) k ( ∑ j = 1 S B † ( b ^ j ) † B B † b ^ j B ) l : = : B † ( ∑ i = 1 S ( a ^ i ) † a ^ i ) k B B † ( ∑ j = 1 S ( b ^ j ) † b ^ j ) l B : = B † : ( ∑ i = 1 S ( a ^ i ) † a ^ i ) k ( ∑ j = 1 S ( b ^ j ) † b ^ j ) l : B = B † : n ^ a k n ^ b l : B (1) \begin{aligned}
:(\hat{n}'_a)^k(\hat{n}'_b)^l:
=
:\left(\sum_{i=1}^{S}(\hat{a}'_i)^\dagger\hat{a}'_i\right)^k
\left(\sum_{j=1}^{S}(\hat{b}'_j)^\dagger\hat{b}'_j\right)^l: \\
&=:\left(\sum_{i=1}^{S}\mathcal{B}^{\dagger}(\hat{a}_{i})^{\dagger}\mathcal{B}\mathcal{B}^{\dagger}\hat{a}_{i}\mathcal{B}\right)^{k}\left(\sum_{j=1}^{S}\mathcal{B}^{\dagger}(\hat{b}_{j})^{\dagger}\mathcal{B}\mathcal{B}^{\dagger}\hat{b}_{j}\mathcal{B}\right)^{l}: \\
&=:\mathcal{B}^{\dagger}\left(\sum_{i=1}^{S}(\hat{a}_{i})^{\dagger}\hat{a}_{i}\right)^{k}\mathcal{B}\mathcal{B}^{\dagger}\left(\sum_{j=1}^{S}(\hat{b}_{j})^{\dagger}\hat{b}_{j}\right)^{l}\mathcal{B}: \\
&=\mathcal{B}^{\dagger}:\left(\sum_{i=1}^{S}(\hat{a}_{i})^{\dagger}\hat{a}_{i}\right)^{k}\left(\sum_{j=1}^{S}(\hat{b}_{j})^{\dagger}\hat{b}_{j}\right)^{l}:\mathcal{B} \\
&=\mathcal{B}^{\dagger}:\hat{n}_{a}^{k}\hat{n}_{b}^{l}:\mathcal{B} \tag*{(1)}
\end{aligned} : ( n ^ a ′ ) k ( n ^ b ′ ) l :=: ( i = 1 ∑ S ( a ^ i ′ ) † a ^ i ′ ) k ( j = 1 ∑ S ( b ^ j ′ ) † b ^ j ′ ) l : =: ( i = 1 ∑ S B † ( a ^ i ) † B B † a ^ i B ) k ( j = 1 ∑ S B † ( b ^ j ) † B B † b ^ j B ) l : =: B † ( i = 1 ∑ S ( a ^ i ) † a ^ i ) k B B † ( j = 1 ∑ S ( b ^ j ) † b ^ j ) l B : = B † : ( i = 1 ∑ S ( a ^ i ) † a ^ i ) k ( j = 1 ∑ S ( b ^ j ) † b ^ j ) l : B = B † : n ^ a k n ^ b l : B (1)
where n ^ a \hat{n}_{a} n ^ a and n ^ b \hat{n}_{b} n ^ b are the total photon-number operators on H a \mathcal{H}_{a} H a and H b \mathcal{H}_{b} H b , respectively. Here, we have used the commutativity of B \mathcal{B} B with the normal-ordering operation in the penultimate step.
The properties described in the previous paragraphs can be used to describe the transformation induced by a PLT on a normally-ordered operator P ^ \hat{P} P ^ by a change in the polynomial E [ P ^ ] \mathcal{E}[\hat{P}] E [ P ^ ] . In particular, consider the action of a PLT operator U \mathcal{U} U on ⨂ i = 1 S ( a ^ i † ) n i ⨂ j = 1 S a ^ j m j \bigotimes_{i=1}^{S}(\hat{a}_{i}^{\dagger})^{n_{i}}\bigotimes_{j=1}^{S}\hat{a}_{j}^{m_{j}} ⨂ i = 1 S ( a ^ i † ) n i ⨂ j = 1 S a ^ j m j :
U † [ ⨂ i = 1 S ( a ^ i † ) n i ⨂ j = 1 S a ^ j m j ] U = ⨂ i = 1 S ( U † a ^ i † U ) n i ⨂ j = 1 S ( U † a ^ j U ) m j = ⨂ i = 1 S ( U ∗ ⋅ a ^ † ) i n i ⨂ j = 1 S ( U ⋅ a ^ ) j m j , (2) \begin{aligned}
\mathcal{U}^{\dagger}\left[\bigotimes_{i=1}^{S}(\hat{a}_{i}^{\dagger})^{n_{i}}\bigotimes_{j=1}^{S}\hat{a}_{j}^{m_{j}}\right]\mathcal{U} \\
&=\bigotimes_{i=1}^{S}(\mathcal{U}^{\dagger}\hat{a}_{i}^{\dagger}\mathcal{U})^{n_{i}}\bigotimes_{j=1}^{S}(\mathcal{U}^{\dagger}\hat{a}_{j}\mathcal{U})^{m_{j}} \\
&=\bigotimes_{i=1}^{S}(\mathcal{U}^{*}\cdot\hat{\mathbf{a}}^{\dagger})_{i}^{n_{i}}\bigotimes_{j=1}^{S}(\mathcal{U}\cdot\hat{\mathbf{a}})_{j}^{m_{j}}, \tag*{(2)}
\end{aligned} U † [ i = 1 ⨂ S ( a ^ i † ) n i j = 1 ⨂ S a ^ j m j ] U = i = 1 ⨂ S ( U † a ^ i † U ) n i j = 1 ⨂ S ( U † a ^ j U ) m j = i = 1 ⨂ S ( U ∗ ⋅ a ^ † ) i n i j = 1 ⨂ S ( U ⋅ a ^ ) j m j , (2)
where ( U ∗ ⋅ a ^ † ) i (\mathcal{U}^{*}\cdot\hat{\mathbf{a}}^{\dagger})_{i} ( U ∗ ⋅ a ^ † ) i is the i i i th element of the vector U ∗ ⋅ a ^ † \mathcal{U}^{*}\cdot\hat{\mathbf{a}}^{\dagger} U ∗ ⋅ a ^ † with U ∗ \mathcal{U}^{*} U ∗ the matrix obtained by conjugating each entry of U \mathcal{U} U , and similarly for ( U ⋅ a ^ ) j (\mathcal{U}\cdot\hat{\mathbf{a}})_{j} ( U ⋅ a ^ ) j . Therefore, the transformation of the corresponding amplitude polynomial is
∏ i = 1 S ( α i ∗ ) n i ∏ j = 1 S α j m j → ∏ i = 1 S ( U ∗ ⋅ α ∗ ) i n i ∏ j = 1 S ( U ⋅ α ) j m j . (3) \prod_{i=1}^{S}(\alpha_{i}^{*})^{n_{i}}\prod_{j=1}^{S}\alpha_{j}^{m_{j}}
\rightarrow
\prod_{i=1}^{S}(\mathcal{U}^{*}\cdot\boldsymbol{\alpha}^{*})_{i}^{n_{i}}\prod_{j=1}^{S}(\mathcal{U}\cdot\boldsymbol{\alpha})_{j}^{m_{j}}. \tag*{(3)} i = 1 ∏ S ( α i ∗ ) n i j = 1 ∏ S α j m j → i = 1 ∏ S ( U ∗ ⋅ α ∗ ) i n i j = 1 ∏ S ( U ⋅ α ) j m j . (3)
We use this correspondence under PLT action to describe Eq. (1) as a transformation of expressions. E [ : ( n ^ a ) k ( n ^ b ) l : ] \mathcal{E}\left[:(\hat{n}_{a})^{k}(\hat{n}_{b})^{l}:\right] E [ : ( n ^ a ) k ( n ^ b ) l : ] is given by
⟨ α ⃗ ′ ∣ ⟨ β ⃗ ′ ∣ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ∣ α ⃗ ′ ⟩ ∣ β ⃗ ′ ⟩ = ∣ α ⃗ ′ ∣ 2 k ∣ β ⃗ ′ ∣ 2 l , (4) \left\langle\vec{\alpha}^{\prime}\right|\left\langle\vec{\beta}^{\prime}\right|:(\hat{n}_{a}^{\prime})^{k}(\hat{n}_{b}^{\prime})^{l}:\left|\vec{\alpha}^{\prime}\right\rangle\left|\vec{\beta}^{\prime}\right\rangle
=|\vec{\alpha}^{\prime}|^{2k}|\vec{\beta}^{\prime}|^{2l}, \tag*{(4)} ⟨ α ′ ∣ ⟨ β ′ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ∣ α ′ ⟩ β ′ ⟩ = ∣ α ′ ∣ 2 k ∣ β ′ ∣ 2 l , (4)
where ∣ α ⃗ ′ ⟩ ∣ β ⃗ ′ ⟩ ∈ H a ′ ⊗ H b ′ \left|\vec{\alpha}^{\prime}\right\rangle\left|\vec{\beta}^{\prime}\right\rangle\in\mathcal{H}_{a^{\prime}}\otimes\mathcal{H}_{b^{\prime}} ∣ α ′ ⟩ β ′ ⟩ ∈ H a ′ ⊗ H b ′ , see Eq. (58) in Appendix B for a derivation. Denoting a coherent state in the inputs of the BS by ∣ α ⃗ ⟩ ∣ β ⃗ ⟩ ∈ H a ⊗ H b \left|\vec{\alpha}\right\rangle\left|\vec{\beta}\right\rangle\in\mathcal{H}_{a}\otimes\mathcal{H}_{b} ∣ α ⟩ β ⟩ ∈ H a ⊗ H b , Eq. (1) implies that B \mathcal{B} B induces the following transformation of the corresponding amplitude polynomials:
∣ α ⃗ ∣ 2 k ∣ β ⃗ ∣ 2 l → ∣ α ⃗ ′ ∣ 2 k ∣ β ⃗ ′ ∣ 2 l , (5) |\vec{\alpha}|^{2k}|\vec{\beta}|^{2l}\rightarrow|\vec{\alpha}^{\prime}|^{2k}|\vec{\beta}^{\prime}|^{2l}, \tag*{(5)} ∣ α ∣ 2 k ∣ β ∣ 2 l → ∣ α ′ ∣ 2 k ∣ β ′ ∣ 2 l , (5)
where, according to Eq. (3) , α ⃗ ′ ⊕ β ⃗ ′ = B ⋅ ( α ⃗ ⊕ β ⃗ ) \vec{\alpha}^{\prime}\oplus\vec{\beta}^{\prime}=B\cdot(\vec{\alpha}\oplus\vec{\beta}) α ′ ⊕ β ′ = B ⋅ ( α ⊕ β ) , or, equivalently, ( α i ′ , β i ′ ) = B 1 ⋅ ( α i , β i ) (\alpha_{i}^{\prime},\beta_{i}^{\prime})=B_{1}\cdot(\alpha_{i},\beta_{i}) ( α i ′ , β i ′ ) = B 1 ⋅ ( α i , β i ) for each i i i .
Generating Functionology
We use operator-valued generating functions, which are formal power series with operators as coefficients of the monomials. As we explain below, the generating functions are chosen so that for the family of states considered in this work and introduced in Sect. 3.4, the expectations of the coefficients can be obtained from the collection of joint photon-counting probabilities ( p i j ) i j (p_{ij})_{ij} ( p ij ) ij , which makes it possible to work with more convenient families of operators than the photon-number projectors.
For a systematic treatment of generating functions and their properties, see Ref. [36 ] . The measurement configuration involving the joint counters can be represented by G ( u , v ) = ∑ k = 0 ∞ ∑ l = 0 ∞ 1 k ! l ! Π k ⊗ Π l u k v l G(u,v)=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\frac{1}{k!l!}\Pi_{k}\otimes\Pi_{l}u^{k}v^{l} G ( u , v ) = ∑ k = 0 ∞ ∑ l = 0 ∞ k ! l ! 1 Π k ⊗ Π l u k v l , which should be interpreted as a formal power series in the variables u , v u,v u , v . The coefficient Π k ⊗ Π l \Pi_{k}\otimes\Pi_{l} Π k ⊗ Π l can be extracted by taking the formal derivative ∂ u k ∂ v l G ( u , v ) ∣ u , v = 0 \partial_{u}^{k}\partial_{v}^{l}G(u,v)|_{u,v=0} ∂ u k ∂ v l G ( u , v ) ∣ u , v = 0 , evaluated at u , v = 0 u,v=0 u , v = 0 to obtain the constant term of the resulting formal power series. The generating function G ( u , v ) G(u,v) G ( u , v ) represents the measurement configuration in the sense that for any given state, the expectations of the operator coefficients are the probabilities of the measurement outcomes. For our analysis, it is more convenient to represent the measurement configuration by the generating function whose coefficients are normally ordered moments:
H ′ ( u , v ) = ∑ k = 0 ∞ ∑ l = 0 ∞ 1 k ! l ! : ( n ^ a ′ ) k ( n ^ b ′ ) l : u k v l . (6) H^{\prime}(u,v)=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\frac{1}{k!l!}:(\hat{n}_{a}^{\prime})^{k}(\hat{n}_{b}^{\prime})^{l}:u^{k}v^{l}. \tag*{(6)} H ′ ( u , v ) = k = 0 ∑ ∞ l = 0 ∑ ∞ k ! l ! 1 : ( n ^ a ′ ) k ( n ^ b ′ ) l : u k v l . (6)
If the measurement configuration involves a single counter, the corresponding generating function is obtained by setting v = 0 v=0 v = 0 in H ′ ( u , v ) H^{\prime}(u,v) H ′ ( u , v ) , resulting in a generating function whose coefficients are the operators : ( n ^ a ′ ) k : :(\hat{n}_{a}^{\prime})^{k}: : ( n ^ a ′ ) k : .
The coefficients of H ′ ( u , v ) H^{\prime}(u,v) H ′ ( u , v ) are unbounded operators. It is therefore necessary to restrict the state space in order for the expectations of the operator coefficients to be finite. The restriction to a dense set of states introduced in Sect. 3.4 suffices for this purpose.
Since the operator-valued coefficients of H ′ ( u , v ) H^{\prime}(u,v) H ′ ( u , v ) are normally ordered, we extend the operator-polynomial correspondence to H ′ ( u , v ) H^{\prime}(u,v) H ′ ( u , v ) by taking the expectation of all of its coefficients in a coherent state ∣ α ⃗ ′ ⟩ ∣ β ⃗ ′ ⟩ ∈ H a ′ ⊗ H b ′ \left|\vec{\alpha}^{\prime}\right\rangle\left|\vec{\beta}^{\prime}\right\rangle\in\mathcal{H}_{a^{\prime}}\otimes\mathcal{H}_{b^{\prime}} ∣ α ′ ⟩ β ′ ⟩ ∈ H a ′ ⊗ H b ′ . Using Eq. (4) , we obtain
H ′ ( u , v ; α ⃗ ′ , β ⃗ ′ ) ≡ ⟨ α ⃗ ′ ∣ ⟨ β ⃗ ′ ∣ H ′ ( u , v ) ∣ α ⃗ ′ ⟩ ∣ β ⃗ ′ ⟩ ≡ ∑ k = 0 ∞ ∑ l = 0 ∞ u k v l k ! l ! ∣ α ⃗ ′ ∣ 2 k ∣ β ⃗ ′ ∣ 2 l = e ∣ α ⃗ ′ ∣ 2 u + ∣ β ⃗ ′ ∣ 2 v . (7) \begin{aligned}
H^{\prime}(u,v;\vec{\alpha}^{\prime},\vec{\beta}^{\prime})
\equiv\left\langle\vec{\alpha}^{\prime}\right|\left\langle\vec{\beta}^{\prime}\right|H^{\prime}(u,v)\left|\vec{\alpha}^{\prime}\right\rangle\left|\vec{\beta}^{\prime}\right\rangle\\
&\equiv\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\frac{u^{k}v^{l}}{k!l!}|\vec{\alpha}^{\prime}|^{2k}|\vec{\beta}^{\prime}|^{2l}
=e^{|\vec{\alpha}^{\prime}|^{2}u+|\vec{\beta}^{\prime}|^{2}v}. \tag*{(7)}
\end{aligned} H ′ ( u , v ; α ′ , β ′ ) ≡ ⟨ α ′ ∣ ⟨ β ′ H ′ ( u , v ) ∣ α ′ ⟩ β ′ ⟩ ≡ k = 0 ∑ ∞ l = 0 ∑ ∞ k ! l ! u k v l ∣ α ′ ∣ 2 k ∣ β ′ ∣ 2 l = e ∣ α ′ ∣ 2 u + ∣ β ′ ∣ 2 v . (7)
H ′ ( u , v ; α ⃗ ′ , β ⃗ ′ ) H^{\prime}(u,v;\vec{\alpha}^{\prime},\vec{\beta}^{\prime}) H ′ ( u , v ; α ′ , β ′ ) is a convergent power series for all values of the variables u u u , v v v , and of α ⃗ ′ \vec{\alpha}^{\prime} α ′ , β ⃗ ′ \vec{\beta}^{\prime} β ′ and their complex conjugates. The operator coefficients of H ′ ( u , v ) H^{\prime}(u,v) H ′ ( u , v ) can be recovered from the function H ′ ( u , v ; α ⃗ ′ , β ⃗ ′ ) H^{\prime}(u,v;\vec{\alpha}^{\prime},\vec{\beta}^{\prime}) H ′ ( u , v ; α ′ , β ′ ) by evaluating the derivative ∂ u k ∂ v l e ∣ α ⃗ ′ ∣ 2 u + ∣ β ⃗ ′ ∣ 2 v ∣ u , v = 0 \partial_u^k\partial_v^l e^{|\vec{\alpha}'|^2u+|\vec{\beta}'|^2v}\big|_{u,v=0} ∂ u k ∂ v l e ∣ α ′ ∣ 2 u + ∣ β ′ ∣ 2 v u , v = 0 as usual, and applying the operator-polynomial correspondence to the resulting polynomial of α ⃗ ′ \vec{\alpha}' α ′ , β ⃗ ′ \vec{\beta}' β ′ and their complex conjugates. We refer to H ′ ( u , v ; α ⃗ ′ , β ⃗ ′ ) H'(u,v;\vec{\alpha}',\vec{\beta}') H ′ ( u , v ; α ′ , β ′ ) as the amplitude generating function of H ′ ( u , v ) H'(u,v) H ′ ( u , v ) , and we maintain this terminology whenever the operator-polynomial correspondence is applied to an operator-valued generating function with coefficients that are polynomials of mode operators and their adjoints.
The main advantage of working with H ′ ( u , v ; α ⃗ ′ , β ⃗ ′ ) H'(u,v;\vec{\alpha}',\vec{\beta}') H ′ ( u , v ; α ′ , β ′ ) lies in how it relates operators on H a ⊗ H b \mathcal{H}_a\otimes\mathcal{H}_b H a ⊗ H b to operators on H a ′ ⊗ H b ′ \mathcal{H}_{a'}\otimes\mathcal{H}_{b'} H a ′ ⊗ H b ′ . Consider the operator-valued generating function
H ( u , v ) = ∑ k = 0 ∞ ∑ l = 0 ∞ 1 k ! l ! : n ^ a k n ^ b l : u k v l , (8) H(u,v)=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\frac{1}{k!l!}:\hat{n}_a^k\hat{n}_b^l:u^kv^l, \tag*{(8)} H ( u , v ) = k = 0 ∑ ∞ l = 0 ∑ ∞ k ! l ! 1 : n ^ a k n ^ b l : u k v l , (8)
and the corresponding amplitude generating function
H ( u , v ; α ⃗ , β ⃗ ) ≡ ⟨ α ⃗ ∣ ⟨ β ⃗ ∣ H ( u , v ) ∣ α ⃗ ⟩ ∣ β ⃗ ⟩ = e ∣ α ⃗ ∣ 2 u + ∣ β ⃗ ∣ 2 v , (9) \begin{aligned}
H(u,v;\vec{\alpha},\vec{\beta})\equiv\langle\vec{\alpha}|\langle\vec{\beta}|H(u,v)|\vec{\alpha}\rangle|\vec{\beta}\rangle\\
&=e^{|\vec{\alpha}|^2u+|\vec{\beta}|^2v}, \tag*{(9)}
\end{aligned} H ( u , v ; α , β ) ≡ ⟨ α ∣ ⟨ β ∣ H ( u , v ) ∣ α ⟩ ∣ β ⟩ = e ∣ α ∣ 2 u + ∣ β ∣ 2 v , (9)
where ∣ α ⃗ ⟩ ∣ β ⃗ ⟩ ∈ H a ⊗ H b |\vec{\alpha}\rangle|\vec{\beta}\rangle\in\mathcal{H}_a\otimes\mathcal{H}_b ∣ α ⟩ ∣ β ⟩ ∈ H a ⊗ H b . Using (1) , H ′ ( u , v ) = B † H ( u , v ) B H'(u,v)=B^\dagger H(u,v)B H ′ ( u , v ) = B † H ( u , v ) B , where the transformation acts independently on each coefficient of the power series. The corresponding polynomials are related by the transformation of the complex variables, ( α i ′ , β i ′ ) = B 1 ⋅ ( α i , β i ) (\alpha'_i,\beta'_i)=B_1\cdot(\alpha_i,\beta_i) ( α i ′ , β i ′ ) = B 1 ⋅ ( α i , β i ) for each i i i , by applying (5) to each coefficient of H ( u , v ; α ⃗ , β ⃗ ) H(u,v;\vec{\alpha},\vec{\beta}) H ( u , v ; α , β ) . Thus, it is possible to express H ′ H' H ′ as a function of α ⃗ \vec{\alpha} α and β ⃗ \vec{\beta} β :
H ′ ( u , v ; α ⃗ ′ , β ⃗ ′ ) = e ∑ i = 1 S ∣ α i ′ ∣ 2 u + ∑ i = 1 S ∣ β i ′ ∣ 2 v = e ∑ i = 1 S ∣ η α i + ζ β i ∣ 2 u + ∑ i = 1 S ∣ ζ ′ α i + η ′ β i ∣ 2 v . (10) \begin{aligned}
H'(u,v;\vec{\alpha}',\vec{\beta}')=e^{\sum_{i=1}^{S}|\alpha'_i|^2u+\sum_{i=1}^{S}|\beta'_i|^2v} \\
&=e^{\sum_{i=1}^{S}|\eta\alpha_i+\zeta\beta_i|^2u+\sum_{i=1}^{S}|\zeta'\alpha_i+\eta'\beta_i|^2v}. \tag*{(10)}
\end{aligned} H ′ ( u , v ; α ′ , β ′ ) = e ∑ i = 1 S ∣ α i ′ ∣ 2 u + ∑ i = 1 S ∣ β i ′ ∣ 2 v = e ∑ i = 1 S ∣ η α i + ζ β i ∣ 2 u + ∑ i = 1 S ∣ ζ ′ α i + η ′ β i ∣ 2 v . (10)
Then, by again applying the operator-polynomial correspondence on H a ⊗ H b \mathcal{H}_a\otimes\mathcal{H}_b H a ⊗ H b to each coefficient of H ′ ( u , v ; α ⃗ ′ , β ⃗ ′ ) H'(u,v;\vec{\alpha}',\vec{\beta}') H ′ ( u , v ; α ′ , β ′ ) , we obtain an expression for H ′ ( u , v ) H'(u,v) H ′ ( u , v ) in terms of the mode operators a ^ i \hat{a}_i a ^ i , a ^ i † \hat{a}_i^\dagger a ^ i † , b ^ i \hat{b}_i b ^ i and b ^ i † \hat{b}_i^\dagger b ^ i † :
H ′ ( u , v ) = ∑ k = 0 ∞ ∑ l = 0 ∞ 1 k ! l ! [ ∑ i = 1 S ( η ∗ a ^ i † + ζ ∗ b ^ i † ) ( η a ^ i + ζ b ^ i ) ] k × [ ∑ j = 1 S ( ζ ′ ∗ a ^ j † + η ′ ∗ b ^ j † ) ( ζ ′ a ^ j + η ′ b ^ j ) ] l : u k v l . (11) H'(u,v)=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\frac{1}{k!l!}\left[\sum_{i=1}^{S}(\eta^*\hat{a}_i^\dagger+\zeta^*\hat{b}_i^\dagger)(\eta\hat{a}_i+\zeta\hat{b}_i)\right]^k
\times\left[\sum_{j=1}^{S}(\zeta'^*\hat{a}_j^\dagger+\eta'^*\hat{b}_j^\dagger)(\zeta'\hat{a}_j+\eta'\hat{b}_j)\right]^l:u^kv^l. \tag*{(11)} H ′ ( u , v ) = k = 0 ∑ ∞ l = 0 ∑ ∞ k ! l ! 1 [ i = 1 ∑ S ( η ∗ a ^ i † + ζ ∗ b ^ i † ) ( η a ^ i + ζ b ^ i ) ] k × [ j = 1 ∑ S ( ζ ′ ∗ a ^ j † + η ′ ∗ b ^ j † ) ( ζ ′ a ^ j + η ′ b ^ j ) ] l : u k v l . (11)
We could have obtained (11) more directly from (6) by expanding the coefficients : ( n ^ a ′ ) k ( n ^ b ′ ) l : :(\hat{n}'_a)^k(\hat{n}'_b)^l: : ( n ^ a ′ ) k ( n ^ b ′ ) l : in terms of the mode operators a ^ i , a ^ i † , b ^ i \hat{a}_i,\hat{a}_i^\dagger,\hat{b}_i a ^ i , a ^ i † , b ^ i and b ^ i † \hat{b}_i^\dagger b ^ i † . However, carrying out the derivation in this way provides a useful exercise in manipulating generating functions.
We are interested in using generating functions to represent the linear spaces of operators on H a \mathcal{H}_a H a whose expectations are determined by measurements with different probe states. For a given probe state σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⃗ ⟩ ⟨ 0 ⃗ ∣ \sigma=|\gamma\rangle\langle\gamma|\otimes|\vec{0}\rangle\langle\vec{0}| σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⟩ ⟨ 0 ∣ , the immediately available expectations are the photon-counting probabilities p k l p_{kl} p k l . These probabilities are the expectations, with respect to the input state ρ \rho ρ , of the operators on H a \mathcal{H}_a H a obtained by contracting the measurement operators Π k ⊗ Π l \Pi_k\otimes\Pi_l Π k ⊗ Π l with σ \sigma σ by taking the partial trace of σ ( Π k ⊗ Π l ) \sigma(\Pi_k\otimes\Pi_l) σ ( Π k ⊗ Π l ) over H b \mathcal{H}_b H b :
E ^ k l ( γ ) ≡ tr b ( ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⃗ ⟩ ⟨ 0 ⃗ ∣ Π k ⊗ Π l ) = ⟨ γ ∣ ⟨ 0 ⃗ ∣ Π k ⊗ Π l ∣ 0 ⃗ ⟩ ∣ γ ⟩ . (12) \begin{aligned}
\hat{E}_{kl}(\gamma)\equiv\operatorname{tr}_b\left(|\gamma\rangle\langle\gamma|\otimes|\vec{0}\rangle\langle\vec{0}|\Pi_k\otimes\Pi_l\right) \\
&=\langle\gamma|\langle\vec{0}|\Pi_k\otimes\Pi_l|\vec{0}\rangle|\gamma\rangle. \tag*{(12)}
\end{aligned} E ^ k l ( γ ) ≡ tr b ( ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⟩ ⟨ 0 ∣ Π k ⊗ Π l ) = ⟨ γ ∣ ⟨ 0 ∣ Π k ⊗ Π l ∣ 0 ⟩ ∣ γ ⟩ . (12)
The photon number probabilities are then given by p k l = tr ( ρ ^ E ^ k l ( γ ) ) p_{kl}=\operatorname{tr}(\hat{\rho}\hat{E}_{kl}(\gamma)) p k l = tr ( ρ ^ E ^ k l ( γ )) . Because the states under consideration have finite photon number moments (Lem. 3.1), the normally ordered moments ⟨ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ⟩ \langle:(\hat{n}'_a)^k(\hat{n}'_b)^l:\rangle ⟨ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ⟩ are linearly determined by the photon number probabilities p k l p_{kl} p k l . These expectations are also obtained as expectations of operators on H a \mathcal{H}_a H a with respect to ρ \rho ρ , where these operators are obtained in the same way as E ^ k l ( γ ) \hat{E}_{kl}(\gamma) E ^ k l ( γ ) by contracting the : ( n ^ a ′ ) k ( n ^ b ′ ) l : :(\hat{n}'_a)^k(\hat{n}'_b)^l: : ( n ^ a ′ ) k ( n ^ b ′ ) l : with σ \sigma σ . By taking advantage of the identity of the coefficients of H ′ ( u , v ) H'(u,v) H ′ ( u , v ) as expressed in (6) with those expressed in (11) , the contraction is obtained as follows:
⟨ γ ∣ ⟨ 0 ⃗ ∣ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ∣ γ ⟩ ∣ 0 ⃗ ⟩ = ⟨ γ ∣ ⟨ 0 ⃗ ∣ [ ∑ i = 1 S ( η ∗ a ^ i † + ζ ∗ b ^ i † ) ( η a ^ i + ζ b ^ i ) ] k [ ∑ j = 1 S ( ζ ′ ∗ a ^ j † + η ′ ∗ b ^ j † ) ( ζ ′ a ^ j + η ′ b ^ j ) ] l ∣ γ ⟩ ∣ 0 ⃗ ⟩ = [ ∑ i = 1 S ( η ∗ a ^ i † + ζ ∗ γ ∗ δ i 1 ) ( η a ^ i + ζ γ δ i 1 ) ] k [ ∑ j = 1 S ( ζ ′ ∗ a ^ j † + η ′ ∗ γ ∗ δ j 1 ) ( ζ ′ a ^ j + η ′ γ δ j 1 ) ] l . (13) \begin{aligned}
\langle\gamma|\langle\vec{0}|:(\hat{n}'_a)^k(\hat{n}'_b)^l:|\gamma\rangle|\vec{0}\rangle\\
&=\langle\gamma|\langle\vec{0}|\left[\sum_{i=1}^{S}(\eta^*\hat{a}_i^\dagger+\zeta^*\hat{b}_i^\dagger)(\eta\hat{a}_i+\zeta\hat{b}_i)\right]^k
\left[\sum_{j=1}^{S}(\zeta'^*\hat{a}_j^\dagger+\eta'^*\hat{b}_j^\dagger)(\zeta'\hat{a}_j+\eta'\hat{b}_j)\right]^l|\gamma\rangle|\vec{0}\rangle\\
&=\left[\sum_{i=1}^{S}(\eta^*\hat{a}_i^\dagger+\zeta^*\gamma^*\delta_{i1})(\eta\hat{a}_i+\zeta\gamma\delta_{i1})\right]^k
\left[\sum_{j=1}^{S}(\zeta'^*\hat{a}_j^\dagger+\eta'^*\gamma^*\delta_{j1})(\zeta'\hat{a}_j+\eta'\gamma\delta_{j1})\right]^l. \tag*{(13)}
\end{aligned} ⟨ γ ∣ ⟨ 0 ∣ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ∣ γ ⟩ ∣ 0 ⟩ = ⟨ γ ∣ ⟨ 0 ∣ [ i = 1 ∑ S ( η ∗ a ^ i † + ζ ∗ b ^ i † ) ( η a ^ i + ζ b ^ i ) ] k [ j = 1 ∑ S ( ζ ′ ∗ a ^ j † + η ′ ∗ b ^ j † ) ( ζ ′ a ^ j + η ′ b ^ j ) ] l ∣ γ ⟩ ∣ 0 ⟩ = [ i = 1 ∑ S ( η ∗ a ^ i † + ζ ∗ γ ∗ δ i 1 ) ( η a ^ i + ζ γ δ i 1 ) ] k [ j = 1 ∑ S ( ζ ′ ∗ a ^ j † + η ′ ∗ γ ∗ δ j 1 ) ( ζ ′ a ^ j + η ′ γ δ j 1 ) ] l . (13)
where δ i j \delta_{ij} δ ij denotes the Kronecker delta. In the last step, we applied the identities ⟨ γ ∣ ⟨ 0 ⃗ ∣ b ^ i † = δ i 1 γ ∗ ⟨ γ ∣ ⟨ 0 ⃗ ∣ \langle\gamma|\langle\vec{0}|\hat{b}_i^\dagger=\delta_{i1}\gamma^*\langle\gamma|\langle\vec{0}| ⟨ γ ∣ ⟨ 0 ∣ b ^ i † = δ i 1 γ ∗ ⟨ γ ∣ ⟨ 0 ∣ and b ^ j ∣ γ ⟩ ∣ 0 ⃗ ⟩ = ∣ γ ⟩ ∣ 0 ⃗ ⟩ δ j 1 γ \hat{b}_j|\gamma\rangle|\vec{0}\rangle=|\gamma\rangle|\vec{0}\rangle\delta_{j1}\gamma b ^ j ∣ γ ⟩ ∣ 0 ⟩ = ∣ γ ⟩ ∣ 0 ⟩ δ j 1 γ . The expectation ⟨ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ⟩ \langle:(\hat{n}'_a)^k(\hat{n}'_b)^l:\rangle ⟨ : ( n ^ a ′ ) k ( n ^ b ′ ) l : ⟩ is then obtained by taking the trace of the final result in (13) with ρ \rho ρ .
For a given γ \gamma γ , C 2 ( γ ) C_2(\gamma) C 2 ( γ ) denotes the set of operators obtained by fully expanding (13) into a sum of monomials and collecting the operator-valued coefficients for all k , l ≥ 0 k,l\ge0 k , l ≥ 0 , meaning that both counters are used. When one detector is used, C 1 ( γ ) C_1(\gamma) C 1 ( γ ) similarly denotes the set obtained with all k ≥ 0 k\ge0 k ≥ 0 and l = 0 l=0 l = 0 . We denote by E [ C i ( γ ) ] \mathcal{E}[C_i(\gamma)] E [ C i ( γ )] the set of amplitude polynomials corresponding to the operators in C i ( γ ) C_i(\gamma) C i ( γ ) , obtained via the operator–polynomial correspondence.
The derivation in Eq. (13) of the operators in C 2 ( γ ) C_2(\gamma) C 2 ( γ ) makes it clear that C 2 ( γ ) C_2(\gamma) C 2 ( γ ) is the collection of coefficients of the operator-valued generating function obtained by contracting H ′ ( u , v ) H'(u,v) H ′ ( u , v ) with σ \sigma σ term-by-term, that is by calculating F γ ( u , v ) ≡ ⟨ α ⃗ ∣ ⟨ γ ∣ ⟨ 0 ⃗ ∣ H ′ ( u , v ) ∣ α ⃗ ⟩ ∣ γ ⟩ ∣ 0 ⃗ ⟩ F_\gamma(u,v) \equiv\langle\vec{\alpha}|\langle\gamma|\langle\vec{0}|H'(u,v)|\vec{\alpha}\rangle|\gamma\rangle|\vec{0}\rangle F γ ( u , v ) ≡ ⟨ α ∣ ⟨ γ ∣ ⟨ 0 ∣ H ′ ( u , v ) ∣ α ⟩ ∣ γ ⟩ ∣ 0 ⟩ . C 1 ( γ ) C_1(\gamma) C 1 ( γ ) is therefore the collection of coefficients obtained by specializing F γ ( u , v ) F_\gamma(u,v) F γ ( u , v ) by setting v = 0 v=0 v = 0 . The generating function F γ ( u , v ) F_\gamma(u,v) F γ ( u , v ) has the following analytic expression:
F γ ( u , v ) ≡ ⟨ α ⃗ ∣ ⟨ γ ∣ ⟨ 0 ⃗ ∣ H ′ ( u , v ) ∣ α ⃗ ⟩ ∣ γ ⟩ ∣ 0 ⃗ ⟩ = ∑ k = 0 ∞ ∑ l = 0 ∞ u k v l k ! l ! ⟨ α ⃗ ∣ : [ ∑ i = 1 S ( η ∗ a ^ i † + ζ ∗ γ ∗ δ i 1 ) ( η a ^ i + ζ γ δ i 1 ) ] k × [ ∑ j = 1 S ( ζ ′ ∗ a ^ j † + η ′ ∗ γ ∗ δ j 1 ) ( ζ ′ a ^ j + η ′ γ δ j 1 ) ] l : ∣ α ⃗ ⟩ = ∑ k = 0 ∞ ∑ l = 0 ∞ u k v l k ! l ! [ ∑ i = 1 S ( η ∗ α i ∗ + ζ ∗ γ ∗ δ i 1 ) ( η α i + ζ γ δ i 1 ) ] k × [ ∑ j = 1 S ( ζ ′ ∗ α j ∗ + η ′ ∗ γ ∗ δ j 1 ) ( ζ ′ α j + η ′ γ δ j 1 ) ] l = e ∑ i = 1 S ∣ η α i + ζ γ δ i 1 ∣ 2 u + ∑ j = 1 S ∣ ζ ′ α j + η ′ γ δ j 1 ∣ 2 v . (14) \begin{aligned}
F_\gamma(u,v) &\equiv\langle\vec{\alpha}|\langle\gamma|\langle\vec{0}|H'(u,v)|\vec{\alpha}\rangle|\gamma\rangle|\vec{0}\rangle\\
&= \sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\frac{u^k v^l}{k!l!}\langle\vec{\alpha}|:\left[\sum_{i=1}^{S}(\eta^*\hat{a}_i^\dagger+\zeta^*\gamma^*\delta_{i1})(\eta\hat{a}_i+\zeta\gamma\delta_{i1})\right]^k \\
&\qquad\times\left[\sum_{j=1}^{S}(\zeta'^*\hat{a}_j^\dagger+\eta'^*\gamma^*\delta_{j1})(\zeta'\hat{a}_j+\eta'\gamma\delta_{j1})\right]^l:|\vec{\alpha}\rangle\\
&= \sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\frac{u^k v^l}{k!l!}\left[\sum_{i=1}^{S}(\eta^*\alpha_i^*+\zeta^*\gamma^*\delta_{i1})(\eta\alpha_i+\zeta\gamma\delta_{i1})\right]^k \\
&\qquad\times\left[\sum_{j=1}^{S}(\zeta'^*\alpha_j^*+\eta'^*\gamma^*\delta_{j1})(\zeta'\alpha_j+\eta'\gamma\delta_{j1})\right]^l \\
&= e^{\sum_{i=1}^{S}|\eta\alpha_i+\zeta\gamma\delta_{i1}|^2u+\sum_{j=1}^{S}|\zeta'\alpha_j+\eta'\gamma\delta_{j1}|^2v}. \tag*{(14)}
\end{aligned} F γ ( u , v ) ≡ ⟨ α ∣ ⟨ γ ∣ ⟨ 0 ∣ H ′ ( u , v ) ∣ α ⟩ ∣ γ ⟩ ∣ 0 ⟩ = k = 0 ∑ ∞ l = 0 ∑ ∞ k ! l ! u k v l ⟨ α ∣ : [ i = 1 ∑ S ( η ∗ a ^ i † + ζ ∗ γ ∗ δ i 1 ) ( η a ^ i + ζ γ δ i 1 ) ] k × [ j = 1 ∑ S ( ζ ′ ∗ a ^ j † + η ′ ∗ γ ∗ δ j 1 ) ( ζ ′ a ^ j + η ′ γ δ j 1 ) ] l : ∣ α ⟩ = k = 0 ∑ ∞ l = 0 ∑ ∞ k ! l ! u k v l [ i = 1 ∑ S ( η ∗ α i ∗ + ζ ∗ γ ∗ δ i 1 ) ( η α i + ζ γ δ i 1 ) ] k × [ j = 1 ∑ S ( ζ ′ ∗ α j ∗ + η ′ ∗ γ ∗ δ j 1 ) ( ζ ′ α j + η ′ γ δ j 1 ) ] l = e ∑ i = 1 S ∣ η α i + ζ γ δ i 1 ∣ 2 u + ∑ j = 1 S ∣ ζ ′ α j + η ′ γ δ j 1 ∣ 2 v . (14)
The operator in Eq. (13) is recovered by evaluating the derivative ∂ u k ∂ v l F γ ( u , v ) ∣ u , v = 0 \partial_u^k\partial_v^lF_\gamma(u,v)|_{u,v=0} ∂ u k ∂ v l F γ ( u , v ) ∣ u , v = 0 and applying the operator-polynomial correspondence on H a \mathcal{H}_a H a . Note that the expression for F γ ( u , v ) F_\gamma(u,v) F γ ( u , v ) could have been obtained directly from Eq. (11) by substituting β ⃗ = ( γ , 0 , … , 0 ) \vec{\beta}=(\gamma,0,\ldots,0) β = ( γ , 0 , … , 0 ) —that is, by treating the arbitrary coherent state on H b \mathcal{H}_b H b used in the operator–expression correspondence as the actual probe state. We have chosen the longer route to avoid potential confusion between the different roles of the variables α ⃗ \vec{\alpha} α and γ \gamma γ in F γ ( u , v ) F_\gamma(u,v) F γ ( u , v ) .
As a final note, we emphasize that while F γ ( u , v ) F_\gamma(u,v) F γ ( u , v ) is an analytic function in the variables u u u , v v v , α ⃗ \vec{\alpha} α , α ⃗ ∗ \vec{\alpha}^* α ∗ , its physical purpose is to encode the operators C 2 ( γ ) C_2(\gamma) C 2 ( γ ) through the operator-amplitude correspondence. We treat u u u and v v v strictly as generating variables and suppress the explicit dependence on α ⃗ \vec{\alpha} α and α ⃗ ∗ \vec{\alpha}^* α ∗ in the notation for F γ ( u , v ) F_\gamma(u,v) F γ ( u , v ) to reflect this physical purpose.
States with Moment-Determinate Husimi Functions
Because our analysis characterizes the measurement statistics by the moments of the measured photon number distributions, these moments must be finite. A sufficient condition for this is that the space of states under consideration has well-defined expectations for all normally ordered products ⨂ j = 1 S ( a ^ j † ) m j ⨂ i = 1 S a ^ i n i \bigotimes_{j=1}^{S}(\hat{a}_j^\dagger)^{m_j}\bigotimes_{i=1}^{S}\hat{a}_i^{n_i} ⨂ j = 1 S ( a ^ j † ) m j ⨂ i = 1 S a ^ i n i . The next lemma establishes the required connections between the photon-number measurement statistics and normally ordered operators.
Lemma 3.1. If ρ \rho ρ has finite normally ordered moments, then for all LO amplitudes γ \gamma γ , the photon-number measurement statistics have finite moments. If ρ \rho ρ satisfies the property that the moments of the photon-number measurement statistics are finite for a given LO amplitude γ \gamma γ , then the expectations of the normally ordered total photon number moments of the BS outputs at γ \gamma γ are determined as finite linear combinations of moments of the photon-number measurement statistics at γ \gamma γ .
Proof. For the first part of the lemma, Eq. (13) shows that, for a given γ \gamma γ , the normally ordered moments of the total photon number operators after the BS given by ⟨ : ( n ^ a ) k ( n ^ b ) l : ⟩ \langle:(\hat{n}_a)^k(\hat{n}_b)^l:\rangle ⟨ : ( n ^ a ) k ( n ^ b ) l : ⟩ , are finite linear combinations of normally ordered moment operators of the input states. If the latter have finite expectations, then so do the former. Cor. B.2 in the appendix implies that the ordinary moments are finite linear combinations of normally ordered moments, implying that if the ⟨ : ( n ^ a ) k ( n ^ b ) l : ⟩ \langle:(\hat{n}_a)^k(\hat{n}_b)^l:\rangle ⟨ : ( n ^ a ) k ( n ^ b ) l : ⟩ are finite, so are the ordinary moments, and hence so are the moments of the photon-number measurement statistics.
For the second part, Appendix B shows that the ⟨ : ( n ^ a ) k ( n ^ b ) l : ⟩ \langle:(\hat{n}_a)^k(\hat{n}_b)^l:\rangle ⟨ : ( n ^ a ) k ( n ^ b ) l : ⟩ are finite linear combinations of the ordinary moments. Therefore, if the latter have finite expectations at γ \gamma γ , we can determine the expectations of each of the former at γ \gamma γ .
Since any product of mode operators on H a \mathcal{H}_a H a can be expressed as a finite linear combination of anti-normally ordered products by repeated use of the commutation relations [ a ^ i † , a ^ j ] = δ i j [\hat{a}_i^\dagger,\hat{a}_j]=\delta_{ij} [ a ^ i † , a ^ j ] = δ ij , the requirement that all normally-ordered products possess well-defined expectations can be naturally expressed in the Husimi representation of state space. The Husimi function [37 Chap. 3] of an S S S -mode quantum state ρ \rho ρ is defined as Q ( α ⃗ , α ⃗ ∗ ) = 1 π S ⟨ α ⃗ ∣ ρ ∣ α ⃗ ⟩ Q(\vec{\alpha},\vec{\alpha}^*)=\frac{1}{\pi^S}\langle\vec{\alpha}|\rho|\vec{\alpha}\rangle Q ( α , α ∗ ) = π S 1 ⟨ α ∣ ρ ∣ α ⟩ . The definition extends to arbitrary bounded operators. For states, the Husimi function is non-negative and normalized, so it can be interpreted as a probability measure. Its complex moments—the expectations of monomials in the variables α i \alpha_i α i and α i ∗ \alpha_i^* α i ∗ with respect to the measure Q ( α ⃗ , α ⃗ ∗ ) Q(\vec{\alpha},\vec{\alpha}^*) Q ( α , α ∗ ) —correspond to expectations of anti-normally ordered mode operators for the state ρ \rho ρ . Specifically, for every anti-normally ordered monomial of the form ⨂ i = 1 S a ^ i n i ⨂ j = 1 S ( a ^ j † ) m j \bigotimes_{i=1}^{S}\hat{a}_i^{n_i}\bigotimes_{j=1}^{S}(\hat{a}_j^\dagger)^{m_j} ⨂ i = 1 S a ^ i n i ⨂ j = 1 S ( a ^ j † ) m j , the following identity holds for states ρ \rho ρ whose Husimi functions have well-defined complex moments:
tr ( ρ [ ⨂ i = 1 S a ^ i n i ] [ ⨂ j = 1 S ( a ^ j † ) m j ] ) = ∫ d α ⃗ d α ⃗ ∗ Q ( α ⃗ , α ⃗ ∗ ) ∏ i = 1 S α i n i ( α i ∗ ) m i . (15) \operatorname{tr}\left(\rho\left[\bigotimes_{i=1}^{S}\hat{a}_i^{n_i}\right]\left[\bigotimes_{j=1}^{S}(\hat{a}_j^\dagger)^{m_j}\right]\right)
=
\int d\vec{\alpha}\,d\vec{\alpha}^{\,*}\,Q(\vec{\alpha},\vec{\alpha}^{\,*})\prod_{i=1}^{S}\alpha_i^{n_i}(\alpha_i^*)^{m_i}. \tag*{(15)} tr ( ρ [ i = 1 ⨂ S a ^ i n i ] [ j = 1 ⨂ S ( a ^ j † ) m j ] ) = ∫ d α d α ∗ Q ( α , α ∗ ) i = 1 ∏ S α i n i ( α i ∗ ) m i . (15)
For our analysis we require the stronger condition that the moments of the Husimi function be determinate for the twirled state obtained from the input state as described below in Sect. 4.2. Determinateness is a property that arises in the study of the moment problem [38 ] . The following definition is adapted from Ref. [38 Chap. 15] . A multivariate distribution over complex variables α 1 , … , α S \alpha_1,\ldots,\alpha_S α 1 , … , α S and their complex conjugates is said to have a determinate moment sequence if the distribution is uniquely determined by the expectations of all monomials of the form ∏ i = 1 S α i n i ( α i ∗ ) m i \prod_{i=1}^{S}\alpha_i^{n_i}(\alpha_i^*)^{m_i} ∏ i = 1 S α i n i ( α i ∗ ) m i for all non-negative integers n i , m i n_i,m_i n i , m i . The family of expectations of these monomials is referred to as the moment sequence. Being uniquely determined means that there is no other multivariate distribution with the same moment sequence. The space of states with determinate moments is defined as the states whose Husimi function has a determinate moment sequence, and is denoted by S \mathcal{S} S . We refer to such states as states with a determinate Husimi moment sequence. Gaussian states, states with bounded photon number, and finite mixtures and superpositions of such states satisfy this property, see Appendix C. The appendix also shows that the states of S \mathcal{S} S have photon-number measurement statistics whose moment sequences are determinate for all LO amplitudes γ \gamma γ .
The property of having determinate Husimi moments is preserved under PLTs, as the following lemma shows.
Lemma 3.2. The set of states S \mathcal{S} S is closed under PLTs: if ρ ∈ S \rho\in\mathcal{S} ρ ∈ S and U U U is a PLT on H a \mathcal{H}_a H a , then U ρ U † ∈ S U\rho U^\dagger\in\mathcal{S} U ρ U † ∈ S .
Proof. Let ρ ′ = U ρ U † \rho'=U\rho U^\dagger ρ ′ = U ρ U † . We first show that each antinormally ordered moment of ρ ′ \rho' ρ ′ is a finite linear combination of the antinormally ordered moments of ρ \rho ρ , and vice versa. To express the antinormally ordered moments of ρ ′ \rho' ρ ′ we compute
tr ( U ρ U † [ ⨂ i = 1 S a ^ i n i ] [ ⨂ j = 1 S ( a ^ j † ) m j ] ) = tr ( ρ U † [ ⨂ i = 1 S a ^ i n i ] [ ⨂ j = 1 S ( a ^ j † ) m j ] U ) = tr ( ρ [ ⨂ i = 1 S ( U † a ^ i U ) n i ] [ ⨂ j = 1 S ( U † a ^ j † U ) m j ] ) = tr ( ρ [ ⨂ i = 1 S ( U ∗ ⋅ a ^ ) n i ] [ ⨂ j = 1 S ( U ⋅ a ^ † ) m j ] ) . (16) \begin{aligned}
\operatorname{tr}\left(U\rho U^\dagger\left[\bigotimes_{i=1}^{S}\hat{a}_i^{n_i}\right]\left[\bigotimes_{j=1}^{S}(\hat{a}_j^\dagger)^{m_j}\right]\right)
&=\operatorname{tr}\left(\rho U^\dagger\left[\bigotimes_{i=1}^{S}\hat{a}_i^{n_i}\right]\left[\bigotimes_{j=1}^{S}(\hat{a}_j^\dagger)^{m_j}\right]U\right)\\
&=\operatorname{tr}\left(\rho\left[\bigotimes_{i=1}^{S}(U^\dagger\hat{a}_iU)^{n_i}\right]\left[\bigotimes_{j=1}^{S}(U^\dagger\hat{a}_j^\dagger U)^{m_j}\right]\right)\\
&=\operatorname{tr}\left(\rho\left[\bigotimes_{i=1}^{S}(U^*\cdot\hat{\mathbf a})^{n_i}\right]\left[\bigotimes_{j=1}^{S}(U\cdot\hat{\mathbf a}^\dagger)^{m_j}\right]\right). \tag*{(16)}
\end{aligned} tr ( U ρ U † [ i = 1 ⨂ S a ^ i n i ] [ j = 1 ⨂ S ( a ^ j † ) m j ] ) = tr ( ρ U † [ i = 1 ⨂ S a ^ i n i ] [ j = 1 ⨂ S ( a ^ j † ) m j ] U ) = tr ( ρ [ i = 1 ⨂ S ( U † a ^ i U ) n i ] [ j = 1 ⨂ S ( U † a ^ j † U ) m j ] ) = tr ( ρ [ i = 1 ⨂ S ( U ∗ ⋅ a ^ ) n i ] [ j = 1 ⨂ S ( U ⋅ a ^ † ) m j ] ) . (16)
Expanding the transformation ( U ⋅ a ^ ) i = ∑ k U i k a ^ k (U\cdot\hat{\mathbf a})_i=\sum_k U_{ik}\hat{a}_k ( U ⋅ a ^ ) i = ∑ k U ik a ^ k and pulling the resulting sums outside the trace results in a linear combination of the antinormally ordered moments of ρ \rho ρ . To express the antinormally ordered moments of ρ \rho ρ in terms of those of ρ ′ \rho' ρ ′ , it suffices to perform an identical derivation, where we start with ρ = U † ρ ′ U \rho=U^\dagger\rho'U ρ = U † ρ ′ U and replace U U U with U † U^\dagger U † .
The previous paragraph implies that the Husimi function of ρ ′ \rho' ρ ′ has a well-defined moment sequence. We now show that this moment sequence is determinate. Suppose that μ ′ \mu' μ ′ is a probability distribution that has the same moment sequence. Although μ ′ \mu' μ ′ might not be the Husimi function of a state, we can transform μ ′ \mu' μ ′ to μ \mu μ by the inverse of the PLT in the same way that we transform Husimi functions. The corresponding finite transform of the moment sequence of μ ′ \mu' μ ′ gives the same moment sequence as that of the Husimi function of ρ \rho ρ . Therefore, by moment determinateness, μ \mu μ is the Husimi function of ρ \rho ρ , which in turn implies that μ ′ \mu' μ ′ is the Husimi function of ρ ′ \rho' ρ ′ . By arbitrariness of μ ′ \mu' μ ′ , we conclude that ρ ′ ∈ S \rho'\in\mathcal{S} ρ ′ ∈ S . □ \square □
Symmetries of the Measurement Configuration
Equivalence Class of Input States
Because PLTs preserve total photon number, the measurement statistics do not change if PLTs are applied on each of the output paths of the BS. More specifically, if X a : H a ′ → H a \mathcal{X}_a:\mathcal{H}_{a'}\to\mathcal{H}_a X a : H a ′ → H a and X b : H b ′ → H b \mathcal{X}_b:\mathcal{H}_{b'}\to\mathcal{H}_b X b : H b ′ → H b are PLT operators, then [ n ^ a ′ , X ^ a ] = 0 [\hat{n}'_a,\hat{\mathcal{X}}_a]=0 [ n ^ a ′ , X ^ a ] = 0 and [ n ^ b ′ , X ^ b ] = 0 [\hat{n}'_b,\hat{\mathcal{X}}_b]=0 [ n ^ b ′ , X ^ b ] = 0 . The mode space representation of the PLT X a ⊗ X b \mathcal{X}_a\otimes\mathcal{X}_b X a ⊗ X b in the basis a ^ ′ ⊕ b ^ ′ \hat{\mathbf a}'\oplus\hat{\mathbf b}' a ^ ′ ⊕ b ^ ′ is the unitary matrix X a ⊕ X b X_a\oplus X_b X a ⊕ X b . This is because X a \mathcal{X}_a X a acts by transforming each a ^ i ′ \hat{a}'_i a ^ i ′ into a linear combination of the a ^ i \hat{a}_i a ^ i , and analogously, X b \mathcal{X}_b X b transforms each b ^ i ′ \hat{b}'_i b ^ i ′ into a linear combination of the b ^ i ′ \hat{b}'_i b ^ i ′ .
Let U ( S − 1 ) U(S-1) U ( S − 1 ) denote the group of PLTs on H a \mathcal{H}_a H a that have the form U = I 1 ⊕ W U=I_1\oplus W U = I 1 ⊕ W in the mode-space basis a ^ \hat{\mathbf a} a ^ , where I 1 I_1 I 1 is the 1 × 1 1\times1 1 × 1 identity and W W W is an S − 1 × S − 1 S-1\times S-1 S − 1 × S − 1 unitary matrix. We overload the notation and let U ( S − 1 ) U(S-1) U ( S − 1 ) also denote the corresponding group acting on the Hilbert spaces H b \mathcal{H}_b H b , H a ′ \mathcal{H}_{a'} H a ′ , and H b ′ \mathcal{H}_{b'} H b ′ . We first show with two propositions that a PLT U U U on H a \mathcal{H}_a H a preserves the measurement statistics for all input and probe states if and only if U ∈ U ( S − 1 ) U\in U(S-1) U ∈ U ( S − 1 ) .
Proposition 4.1. Consider the measurement configuration with one or two counters. Consider a PLT U ∈ U ( S − 1 ) U\in U(S-1) U ∈ U ( S − 1 ) acting on H a \mathcal{H}_a H a . Then, ρ \rho ρ and U ρ U † U\rho U^\dagger U ρ U † will have identical measurement statistics when combined with the same probe state.
Proof. It suffices to prove the claim for the configuration with two counters. We first note that we may apply the same U U U on H b \mathcal{H}_b H b . Doing so has no effect on mode b ^ 1 \hat{b}_1 b ^ 1 since U U U has the form U = I 1 ⊕ W U=I_1\oplus W U = I 1 ⊕ W in the mode-space basis b ^ \hat{\mathbf b} b ^ . Applying the corresponding transformation to σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⃗ ⟩ ⟨ 0 ⃗ ∣ \sigma=|\gamma\rangle\langle\gamma|\otimes|\vec{0}\rangle\langle\vec{0}| σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⟩ ⟨ 0 ∣ , gives
I 1 ⊗ W † ( ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⃗ ⟩ ⟨ 0 ⃗ ∣ ) I 1 ⊗ W = ∣ γ ⟩ ⟨ γ ∣ ⊗ W † ∣ 0 ⃗ ⟩ ⟨ 0 ⃗ ∣ W = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⃗ ⟩ ⟨ 0 ⃗ ∣ . (17) I_1\otimes W^\dagger(|\gamma\rangle\langle\gamma|\otimes|\vec{0}\rangle\langle\vec{0}|)I_1\otimes W
=|\gamma\rangle\langle\gamma|\otimes W^\dagger|\vec{0}\rangle\langle\vec{0}|W
=|\gamma\rangle\langle\gamma|\otimes|\vec{0}\rangle\langle\vec{0}|. \tag*{(17)} I 1 ⊗ W † ( ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⟩ ⟨ 0 ∣ ) I 1 ⊗ W = ∣ γ ⟩ ⟨ γ ∣ ⊗ W † ∣ 0 ⟩ ⟨ 0 ∣ W = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⟩ ⟨ 0 ∣. (17)
Consequently,
U ⊗ U ( ρ ⊗ σ ) U † ⊗ U † = ( U ρ U † ) ⊗ σ (18) U\otimes U(\rho\otimes\sigma)U^\dagger\otimes U^\dagger=(U\rho U^\dagger)\otimes\sigma \tag*{(18)} U ⊗ U ( ρ ⊗ σ ) U † ⊗ U † = ( U ρ U † ) ⊗ σ (18)
for all ρ \rho ρ and σ \sigma σ . Hence, to establish the claim of the proposition it suffices to demonstrate that the action of U ⊗ U U \otimes U U ⊗ U on H a ⊗ H b \mathcal{H}_a \otimes\mathcal{H}_b H a ⊗ H b does not affect the measurement statistics.
Working in the mode-space representation, we next establish the fact that B ( U ⊕ U ) = ( U ⊕ U ) B B(U \oplus U) = (U \oplus U)B B ( U ⊕ U ) = ( U ⊕ U ) B , from which it follows that applying U ⊗ U U \otimes U U ⊗ U before the beam splitter is equivalent to applying it after. This fact is established by the following identity:
B ( U ⊕ U ) B † = [ η I S ζ I S ζ ∗ I S η ∗ I S ] [ U 0 0 U ] [ η ∗ I S ζ ∗ I S ζ I S η I S ] = [ ∣ η ∣ 2 U + ∣ ζ ∣ 2 U η ζ ∗ U + ζ η ∗ U η ∗ ζ U + ζ ∗ η U ∣ ζ ∣ 2 U + ∣ η ∣ 2 U ] = U ⊕ U , (19) B(U \oplus U)B^\dagger=
\begin{bmatrix}
\eta I_S & \zeta I_S \\
\zeta^* I_S & \eta^* I_S
\end{bmatrix}
\begin{bmatrix}
U & 0 \\
0 & U
\end{bmatrix}
\begin{bmatrix}
\eta^* I_S & \zeta^* I_S \\
\zeta I_S & \eta I_S
\end{bmatrix}
=
\begin{bmatrix}
|\eta|^2 U + |\zeta|^2 U & \eta\zeta^* U + \zeta\eta^* U \\
\eta^*\zeta U + \zeta^*\eta U & |\zeta|^2 U + |\eta|^2 U
\end{bmatrix}
= U \oplus U, \tag*{(19)} B ( U ⊕ U ) B † = [ η I S ζ ∗ I S ζ I S η ∗ I S ] [ U 0 0 U ] [ η ∗ I S ζ I S ζ ∗ I S η I S ] = [ ∣ η ∣ 2 U + ∣ ζ ∣ 2 U η ∗ ζ U + ζ ∗ η U η ζ ∗ U + ζ η ∗ U ∣ ζ ∣ 2 U + ∣ η ∣ 2 U ] = U ⊕ U , (19)
were we applied normalization and orthogonality of the columns of the BS matrix to obtain the last line.
Next, we show that any PLT on H a \mathcal{H}_a H a that preserves the measurement statistics for all input and probe states must lie in U ( S − 1 ) U(S-1) U ( S − 1 ) . If all measurement statistics are preserved, then ⟨ n ^ a ′ ⟩ \langle\hat{n}'_a \rangle ⟨ n ^ a ′ ⟩ is preserved, so it suffices to prove the proposition for the case where only ⟨ n ^ a ′ ⟩ \langle\hat{n}'_a \rangle ⟨ n ^ a ′ ⟩ is assumed to be preserved.
Proposition 4.2. Consider the measurement configuration with one or two counters. Consider a PLT U U U on H a \mathcal{H}_a H a . Suppose that for every ρ \rho ρ and every probe state σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⃗ ⟩ ⟨ 0 ⃗ ∣ \sigma= |\gamma\rangle\langle\gamma| \otimes|\vec{0}\rangle\langle\vec{0}| σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⟩ ⟨ 0 ∣ , the expectation ⟨ n ^ a ′ ⟩ \langle\hat{n}'_a \rangle ⟨ n ^ a ′ ⟩ is the same for the states ρ \rho ρ and U † ρ U U^\dagger\rho U U † ρ U . Then, U ∈ U ( S − 1 ) U \in U(S-1) U ∈ U ( S − 1 ) .
Proof. The assumption in the proposition statement implies that for all ρ \rho ρ and σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⃗ ⟩ ⟨ 0 ⃗ ∣ \sigma= |\gamma\rangle\langle\gamma| \otimes|\vec{0}\rangle\langle\vec{0}| σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⟩ ⟨ 0 ∣ we have
tr ( ( ρ ⊗ σ ) n ^ a ′ ) = tr ( ( U † ρ U ⊗ σ ) n ^ a ′ ) . (20) \operatorname{tr}\left((\rho\otimes\sigma)\hat{n}'_a\right) = \operatorname{tr}\left((U^\dagger\rho U \otimes\sigma)\hat{n}'_a\right). \tag*{(20)} tr ( ( ρ ⊗ σ ) n ^ a ′ ) = tr ( ( U † ρ U ⊗ σ ) n ^ a ′ ) . (20)
We define N σ = tr b ( σ n ^ a ′ ) N_\sigma= \operatorname{tr}_b(\sigma\hat{n}'_a) N σ = tr b ( σ n ^ a ′ ) , where the index b b b implies a partial trace over H b \mathcal{H}_b H b . Then, the above is equivalent to
tr ( ρ N σ ) = tr ( U † ρ U N σ ) = tr ( ρ U N σ U † ) (21) \operatorname{tr}(\rho N_\sigma) = \operatorname{tr}(U^\dagger\rho U N_\sigma)
= \operatorname{tr}(\rho U N_\sigma U^\dagger) \tag*{(21)} tr ( ρ N σ ) = tr ( U † ρ U N σ ) = tr ( ρ U N σ U † ) (21)
for all ρ \rho ρ and σ \sigma σ . Since ρ \rho ρ is arbitrary, this is equivalent to requiring that U N σ U † = N σ U N_\sigma U^\dagger= N_\sigma U N σ U † = N σ for all σ \sigma σ . We show next that this condition holds if and only if U U U has the form U = I 1 ⊕ W U = I_1 \oplus W U = I 1 ⊕ W .
To evaluate the partial trace defining N σ N_\sigma N σ we use the relations a ^ i ′ = η a ^ i + ζ b ^ i \hat{a}'_i = \eta\hat{a}_i + \zeta\hat{b}_i a ^ i ′ = η a ^ i + ζ b ^ i and a ^ i ′ † = η ∗ a ^ i † + ζ ∗ b ^ i † \hat{a}_i^{\prime\dagger} = \eta^*\hat{a}_i^\dagger+ \zeta^*\hat{b}_i^\dagger a ^ i ′ † = η ∗ a ^ i † + ζ ∗ b ^ i † .
N σ = tr b ( ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⃗ ⟩ ⟨ 0 ⃗ ∣ ∑ i = 1 S ( η ∗ a ^ i † + ζ ∗ b ^ i † ) ( η a ^ i + ζ b ^ i ) ) = ∑ i = 1 S ⟨ γ ∣ ⟨ 0 ⃗ ∣ ( η ∗ a ^ i † + ζ ∗ b ^ i † ) ( η a ^ i + ζ b ^ i ) ∣ γ ⟩ ∣ 0 ⃗ ⟩ = ∑ i = 1 S ∣ η ∣ 2 a ^ i † a ^ i + η ∗ ζ γ ∗ a ^ 1 † + ζ ∗ η γ a ^ 1 + ∣ ζ ∣ 2 ∣ γ ∣ 2 = ∣ η ∣ 2 n ^ a + ∣ ζ ∣ 2 ∣ γ ∣ 2 + η ∗ ζ γ a ^ 1 † + ζ ∗ η γ ∗ a ^ 1 . (22) \begin{aligned}
N_\sigma= \operatorname{tr}_b\left(|\gamma\rangle\langle\gamma| \otimes|\vec{0}\rangle\langle\vec{0}| \sum_{i=1}^{S}(\eta^*\hat{a}_i^\dagger+\zeta^*\hat{b}_i^\dagger)(\eta\hat{a}_i+\zeta\hat{b}_i)\right) \\
&= \sum_{i=1}^{S}\langle\gamma|\langle\vec{0}|(\eta^*\hat{a}_i^\dagger+\zeta^*\hat{b}_i^\dagger)(\eta\hat{a}_i+\zeta\hat{b}_i)|\gamma\rangle|\vec{0}\rangle\\
&= \sum_{i=1}^{S}|\eta|^2\hat{a}_i^\dagger\hat{a}_i+\eta^*\zeta\gamma^*\hat{a}_1^\dagger+\zeta^*\eta\gamma\hat{a}_1+|\zeta|^2|\gamma|^2 \\
&= |\eta|^2\hat{n}_a+|\zeta|^2|\gamma|^2+\eta^*\zeta\gamma\hat{a}_1^\dagger+\zeta^*\eta\gamma^*\hat{a}_1. \tag*{(22)}
\end{aligned} N σ = tr b ( ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⟩ ⟨ 0 ∣ i = 1 ∑ S ( η ∗ a ^ i † + ζ ∗ b ^ i † ) ( η a ^ i + ζ b ^ i ) ) = i = 1 ∑ S ⟨ γ ∣ ⟨ 0 ∣ ( η ∗ a ^ i † + ζ ∗ b ^ i † ) ( η a ^ i + ζ b ^ i ) ∣ γ ⟩ ∣ 0 ⟩ = i = 1 ∑ S ∣ η ∣ 2 a ^ i † a ^ i + η ∗ ζ γ ∗ a ^ 1 † + ζ ∗ η γ a ^ 1 + ∣ ζ ∣ 2 ∣ γ ∣ 2 = ∣ η ∣ 2 n ^ a + ∣ ζ ∣ 2 ∣ γ ∣ 2 + η ∗ ζ γ a ^ 1 † + ζ ∗ η γ ∗ a ^ 1 . (22)
The first two terms on the last line commute with every PLT on H a \mathcal{H}_a H a . The remaining two terms are proportional to a ^ 1 † \hat{a}_1^\dagger a ^ 1 † and a ^ 1 \hat{a}_1 a ^ 1 , where the two coefficients are nonzero for nonzero γ \gamma γ . Since a PLT transforms annihilation operators into linear combinations of annihilation operators (and likewise for creation operators), to commute with N σ N_\sigma N σ , U U U must preserve a ^ 1 \hat{a}_1 a ^ 1 and a ^ 1 † \hat{a}_1^\dagger a ^ 1 † . Consequently, U U U must be of the form U = I 1 ⊕ W U = I_1 \oplus W U = I 1 ⊕ W .
The Twirled State
The set of input states can be partitioned into equivalence classes such that the states within each class are indistinguishable by the measurement configuration for all probe states. In particular, for any given ρ \rho ρ , the states of the form U † ρ U U^\dagger\rho U U † ρ U with U ∈ U ( S − 1 ) U \in U(S-1) U ∈ U ( S − 1 ) belong to the same equivalence class as ρ \rho ρ . Moreover, any convex mixture of these states is also in the same class. Each equivalence class contains a particularly simple representative, obtained by the following integral:
T ( ρ ) ≡ ∫ X ∈ U ( S − 1 ) d μ ( X ) X ρ X † . (23) \mathcal{T}(\rho) \equiv\int_{X\in U(S-1)} d\mu(X)\,X\rho X^\dagger. \tag*{(23)} T ( ρ ) ≡ ∫ X ∈ U ( S − 1 ) d μ ( X ) X ρ X † . (23)
Here, the integration is with respect to the normalized Haar measure μ ( X ) \mu(X) μ ( X ) on U ( S − 1 ) U(S-1) U ( S − 1 ) , which is invariant under left and right multiplication by elements of U ( S − 1 ) U(S-1) U ( S − 1 ) . T ( ρ ) \mathcal{T}(\rho) T ( ρ ) is in the same equivalence class as ρ \rho ρ since it is the uniform mixture of states in the equivalence class of ρ \rho ρ . Next, we evaluate this integral for an arbitrary ρ \rho ρ .
In the theorem below, H a 1 \mathcal{H}_{a_1} H a 1 denotes the subsystem Hilbert space of H a \mathcal{H}_a H a corresponding to the modes a ^ 2 , … , a ^ S \hat{a}_2,\ldots,\hat{a}_S a ^ 2 , … , a ^ S , and H a 1 \mathcal{H}_{a_1} H a 1 denotes the Hilbert space corresponding to mode a ^ 1 \hat{a}_1 a ^ 1 . We use Schur’s lemma together with the representation theory of PLTs on bosonic Fock space for the proof of the theorem. We let H a 1 ( n ) \mathcal{H}_{a_1}^{(n)} H a 1 ( n ) denote the n n n -photon subspace of H a 1 \mathcal{H}_{a_1} H a 1 , with H a 1 ( 0 ) \mathcal{H}_{a_1}^{(0)} H a 1 ( 0 ) being the one-dimensional space spanned by the vacuum state ∣ 0 ⃗ ⟩ |\vec{0}\rangle ∣ 0 ⟩ . Then, H a 1 = ⨁ n = 0 ∞ H a 1 ( n ) \mathcal{H}_{a_1} = \bigoplus_{n=0}^{\infty}\mathcal{H}_{a_1}^{(n)} H a 1 = ⨁ n = 0 ∞ H a 1 ( n ) . H a 1 ( n ) \mathcal{H}_{a_1}^{(n)} H a 1 ( n ) can be identified with the symmetric subspace of ( H a 1 ( 1 ) ) ⊗ n \left(\mathcal{H}_{a_1}^{(1)}\right)^{\otimes n} ( H a 1 ( 1 ) ) ⊗ n and therefore has dimension d ( n ) = ( n + S − 2 n ) d(n)=\binom{n+S-2}{n} d ( n ) = ( n n + S − 2 ) . We denote the unitary group of PLTs on H a 1 \mathcal{H}_{a_1} H a 1 by U F ( S − 1 ) U_F(S-1) U F ( S − 1 ) , so that U ( S − 1 ) = I 1 ⊗ U F ( S − 1 ) U(S-1)=\mathbb{I}_1\otimes U_F(S-1) U ( S − 1 ) = I 1 ⊗ U F ( S − 1 ) . For each photon number n n n , the action of U F ( S − 1 ) U_F(S-1) U F ( S − 1 ) leaves the subspace H a 1 ( n ) \mathcal{H}_{a_1}^{(n)} H a 1 ( n ) invariant and constitutes a distinct, irreducible representation of the group. Irreducibility follows by Schur-Weyl duality for the unitary group, since in the identification with the symmetric subspace of ( H a 1 ( 1 ) ) ⊗ n (\mathcal{H}_{a_1}^{(1)})^{\otimes n} ( H a 1 ( 1 ) ) ⊗ n , U F ( S − 1 ) U_{\mathrm{F}}(S-1) U F ( S − 1 ) acts as the unitary group, identically on each factor. For a given W ∈ U F ( S − 1 ) W \in U_{\mathrm{F}}(S-1) W ∈ U F ( S − 1 ) , we denote its restriction to the n n n -photon subspace by W ( n ) W^{(n)} W ( n ) . Accordingly, with respect to the direct sum decomposition H a 1 = ⨁ n = 0 ∞ H a 1 ( n ) \mathcal{H}_{a_1}=\bigoplus_{n=0}^{\infty}\mathcal{H}_{a_1}^{(n)} H a 1 = ⨁ n = 0 ∞ H a 1 ( n ) , the operator W W W takes the block-diagonal form W = ⨁ n = 0 ∞ W ( n ) W=\bigoplus_{n=0}^{\infty}W^{(n)} W = ⨁ n = 0 ∞ W ( n ) , where W ( 0 ) ≡ I 1 W^{(0)}\equiv I_1 W ( 0 ) ≡ I 1 .
Theorem 4.3. Consider an arbitrary input state ρ \rho ρ . Then T ( ρ ) \mathcal{T}(\rho) T ( ρ ) has the form T ( ρ ) = ∑ n χ n ⊗ 1 d ( n ) Π n S − 1 \mathcal{T}(\rho)=\sum_n\chi_n\otimes\frac{1}{d^{(n)}}\Pi_n^{S-1} T ( ρ ) = ∑ n χ n ⊗ d ( n ) 1 Π n S − 1 , where χ n \chi_n χ n are positive-semidefinite operators on H a 1 \mathcal{H}_{a_1} H a 1 and Π n S − 1 \Pi_n^{S-1} Π n S − 1 is the projector onto the subspace of H a ˉ 1 \mathcal{H}_{\bar{a}_1} H a ˉ 1 containing n n n photons.
Proof. Schur’s lemma states that if a group representation decomposes into mutually inequivalent irreducible representations, its commutant consists precisely of those operators that act as scalar multiples of the identity on each irreducible subspace [39 ] . Because the representations W ( n ) W^{(n)} W ( n ) are irreducible and mutually inequivalent, Schur’s lemma restricts any operator commuting with U F ( S − 1 ) U_{\mathrm{F}}(S-1) U F ( S − 1 ) on H a ˉ 1 \mathcal{H}_{\bar{a}_1} H a ˉ 1 to the form ∑ n = 0 ∞ k n Π n S − 1 \sum_{n=0}^{\infty}k_n\Pi_n^{S-1} ∑ n = 0 ∞ k n Π n S − 1 , where k n k_n k n are scalars. The extension to the total Fock space H a = H a 1 ⊗ H a ˉ 1 \mathcal{H}_a=\mathcal{H}_{a_1}\otimes\mathcal{H}_{\bar{a}_1} H a = H a 1 ⊗ H a ˉ 1 follows from the algebraic structure of tensor-product representations. Since the group acts on the joint space as I 1 ⊗ W I_1\otimes W I 1 ⊗ W , an operator commutes with this total action if and only if it belongs to the tensor product of the respective commutants. The commutant of the trivial action I 1 I_1 I 1 is the set of all bounded operators on H a 1 \mathcal{H}_{a_1} H a 1 , while the commutant of U F ( S − 1 ) U_{\mathrm{F}}(S-1) U F ( S − 1 ) is spanned by the projectors Π n S − 1 \Pi_n^{S-1} Π n S − 1 . Thus, the overall commutant consists precisely of operators taking the form ∑ n = 0 ∞ A n ⊗ Π n S − 1 \sum_{n=0}^{\infty}A_n\otimes\Pi_n^{S-1} ∑ n = 0 ∞ A n ⊗ Π n S − 1 , where A n A_n A n are arbitrary bounded operators on H a 1 \mathcal{H}_{a_1} H a 1 .
We now show that T ( ρ ) \mathcal{T}(\rho) T ( ρ ) commutes with U ( S − 1 ) U(S-1) U ( S − 1 ) . For any U ∈ U ( S − 1 ) U\in U(S-1) U ∈ U ( S − 1 ) ,
U ∫ X ∈ U ( S − 1 ) d μ ( X ) X ρ X † = ∫ X ∈ U ( S − 1 ) d μ ( X ) ( U X ) ρ X † = ∫ Y ∈ U ( S − 1 ) d μ ( Y ) Y ρ ( U † Y ) † = ∫ Y ∈ U ( S − 1 ) d μ ( Y ) Y ρ Y † U , (24) U\int_{X\in U(S-1)}d\mu(X)\,X\rho X^\dagger=\int_{X\in U(S-1)}d\mu(X)\,(UX)\rho X^\dagger
=\int_{Y\in U(S-1)}d\mu(Y)\,Y\rho(U^\dagger Y)^\dagger
=\int_{Y\in U(S-1)}d\mu(Y)\,Y\rho Y^\dagger U, \tag*{(24)} U ∫ X ∈ U ( S − 1 ) d μ ( X ) X ρ X † = ∫ X ∈ U ( S − 1 ) d μ ( X ) ( U X ) ρ X † = ∫ Y ∈ U ( S − 1 ) d μ ( Y ) Y ρ ( U † Y ) † = ∫ Y ∈ U ( S − 1 ) d μ ( Y ) Y ρ Y † U , (24)
where in the second line we performed the variable substitution Y = U X Y=UX Y = U X , so that X = U † Y X=U^\dagger Y X = U † Y , and used the invariance of μ ( X ) \mu(X) μ ( X ) under left multiplication by elements of U ( S − 1 ) U(S-1) U ( S − 1 ) . It follows that T ( ρ ) \mathcal{T}(\rho) T ( ρ ) must be an operator of the form ∑ n = 0 ∞ A n ⊗ Π n S − 1 \sum_{n=0}^{\infty}A_n\otimes\Pi_n^{S-1} ∑ n = 0 ∞ A n ⊗ Π n S − 1 . In order for T ( ρ ) \mathcal{T}(\rho) T ( ρ ) to be a quantum state, the corresponding A n A_n A n must be positive-semidefinite operators. Defining χ n \chi_n χ n such that A n = 1 d ( n ) χ n A_n=\frac{1}{d^{(n)}}\chi_n A n = d ( n ) 1 χ n produces the form of T ( ρ ) \mathcal{T}(\rho) T ( ρ ) in the theorem statement. □
The expression for T ( ρ ) \mathcal{T}(\rho) T ( ρ ) in Theorem 4.3 has a natural interpretation: the uniform averaging over all unitary transformations that respect the symmetries effectively
eliminates all components of the density matrix that are unobservable by the measurement configuration. The form of T ( ρ ) \mathcal{T}(\rho) T ( ρ ) also suggests that the number of modes S S S is unobservable, as long as S > 1 S>1 S > 1 . Thus, there should exist a state in the equivalence class of each ρ \rho ρ that occupies only modes a ^ 1 \hat{a}_1 a ^ 1 and a ^ 2 \hat{a}_2 a ^ 2 . In what follows, ∣ n , 0 ⃗ ⟩ ∈ H a 1 |n,\vec{0}\rangle\in\mathcal{H}_{a_1} ∣ n , 0 ⟩ ∈ H a 1 refers to the Fock state with n n n photons in mode a ^ 2 \hat{a}_2 a ^ 2 , and no photons in the other modes. Furthermore, we define T F \mathcal{T}_{\mathrm{F}} T F as the averaging map for operators on the subsystem H a 1 \mathcal{H}_{a_1} H a 1 . This map is defined by integrating over U F ( S − 1 ) U_{\mathrm{F}}(S-1) U F ( S − 1 ) with respect to its Haar measure:
T F ( ⋅ ) = ∫ X ∈ U F ( S − 1 ) d μ ( X ) X ( ⋅ ) X † . \mathcal{T}_{\mathrm{F}}(\cdot)=\int_{X\in U_{\mathrm{F}}(S-1)}d\mu(X)\,X(\cdot)X^\dagger. T F ( ⋅ ) = ∫ X ∈ U F ( S − 1 ) d μ ( X ) X ( ⋅ ) X † .
Corollary 4.4. Consider an arbitrary input state ρ \rho ρ . The two-mode state ρ t = ∑ n χ n ⊗ ∣ n , 0 ⃗ ⟩ ⟨ n , 0 ⃗ ∣ \rho_t=\sum_n\chi_n\otimes|n,\vec{0}\rangle\langle n,\vec{0}| ρ t = ∑ n χ n ⊗ ∣ n , 0 ⟩ ⟨ n , 0 ∣ , where the χ n \chi_n χ n are the operators in T ( ρ ) = ∑ n χ n ⊗ 1 d ( n ) Π n S − 1 \mathcal{T}(\rho)=\sum_n\chi_n\otimes\frac{1}{d^{(n)}}\Pi_n^{S-1} T ( ρ ) = ∑ n χ n ⊗ d ( n ) 1 Π n S − 1 , is indistinguishable from ρ \rho ρ by the measurement configuration.
Proof. It suffices to show that T ( ρ t ) = T ( ρ ) \mathcal{T}(\rho_t)=\mathcal{T}(\rho) T ( ρ t ) = T ( ρ ) . We first argue that T F ( ∣ n , 0 ⃗ ⟩ ⟨ n , 0 ⃗ ∣ ) = 1 d ( n ) Π n S − 1 \mathcal{T}_{\mathrm{F}}(|n,\vec{0}\rangle\langle n,\vec{0}|)=\frac{1}{d^{(n)}}\Pi_n^{S-1} T F ( ∣ n , 0 ⟩ ⟨ n , 0 ∣ ) = d ( n ) 1 Π n S − 1 . Observe that T F ( ∣ n , 0 ⃗ ⟩ ⟨ n , 0 ⃗ ∣ ) \mathcal{T}_{\mathrm{F}}(|n,\vec{0}\rangle\langle n,\vec{0}|) T F ( ∣ n , 0 ⟩ ⟨ n , 0 ∣ ) is in the commutant of the representation of U F ( S − 1 ) U_{\mathrm{F}}(S-1) U F ( S − 1 ) on H a 1 \mathcal{H}_{a_1} H a 1 , and has total photon number n n n . Therefore T F ( ∣ n , 0 ⃗ ⟩ ⟨ n , 0 ⃗ ∣ ) \mathcal{T}_{\mathrm{F}}(|n,\vec{0}\rangle\langle n,\vec{0}|) T F ( ∣ n , 0 ⟩ ⟨ n , 0 ∣ ) is in the commutant of the irreducible representation of U F ( S − 1 ) U_{\mathrm{F}}(S-1) U F ( S − 1 ) on H a 1 ( n ) \mathcal{H}_{a_1}^{(n)} H a 1 ( n ) . By Schur’s lemma, T F ( ∣ n , 0 ⃗ ⟩ ⟨ n , 0 ⃗ ∣ ) = λ Π n S − 1 \mathcal{T}_{\mathrm{F}}(|n,\vec{0}\rangle\langle n,\vec{0}|)=\lambda\Pi_n^{S-1} T F ( ∣ n , 0 ⟩ ⟨ n , 0 ∣ ) = λ Π n S − 1 for some constant λ \lambda λ . Imposing the trace condition tr ( T F ( ∣ n , 0 ⃗ ⟩ ⟨ n , 0 ⃗ ∣ ) ) = 1 \operatorname{tr}(\mathcal{T}_{\mathrm{F}}(|n,\vec{0}\rangle\langle n,\vec{0}|))=1 tr ( T F ( ∣ n , 0 ⟩ ⟨ n , 0 ∣ )) = 1 gives λ = 1 d ( n ) \lambda=\frac{1}{d^{(n)}} λ = d ( n ) 1 .
Then,
T ( ρ t ) = ∫ X ∈ U F ( S − 1 ) d μ ( X ) X ( ∑ n χ n ⊗ ∣ n , 0 ⃗ ⟩ ⟨ n , 0 ⃗ ∣ ) X † = ∑ n χ n ⊗ ∫ X ∈ U F ( S − 1 ) d μ ( X ) X ∣ n , 0 ⃗ ⟩ ⟨ n , 0 ⃗ ∣ X † = ∑ n χ n ⊗ 1 d ( n ) Π n S − 1 = T ( ρ ) . (25) \mathcal{T}(\rho_t)=\int_{X\in U_{\mathrm{F}}(S-1)}d\mu(X)\,X\left(\sum_n\chi_n\otimes|n,\vec{0}\rangle\langle n,\vec{0}|\right)X^\dagger
=\sum_n\chi_n\otimes\int_{X\in U_{\mathrm{F}}(S-1)}d\mu(X)\,X|n,\vec{0}\rangle\langle n,\vec{0}|X^\dagger
=\sum_n\chi_n\otimes\frac{1}{d^{(n)}}\Pi_n^{S-1}=\mathcal{T}(\rho). \tag*{(25)} T ( ρ t ) = ∫ X ∈ U F ( S − 1 ) d μ ( X ) X ( n ∑ χ n ⊗ ∣ n , 0 ⟩ ⟨ n , 0 ∣ ) X † = n ∑ χ n ⊗ ∫ X ∈ U F ( S − 1 ) d μ ( X ) X ∣ n , 0 ⟩ ⟨ n , 0 ∣ X † = n ∑ χ n ⊗ d ( n ) 1 Π n S − 1 = T ( ρ ) . (25)
□
We refer to ρ t \rho_t ρ t as the “two-mode twirled state” corresponding to ρ \rho ρ . Cor. 4.4 implies that we may, without loss of generality, assume S = 2 S=2 S = 2 in our analysis. Then, ρ t \rho_t ρ t has the form ρ t = ∑ n χ n ⊗ ∣ n ⟩ ⟨ n ∣ \rho_t=\sum_n\chi_n\otimes|n\rangle\langle n| ρ t = ∑ n χ n ⊗ ∣ n ⟩ ⟨ n ∣ . When referring to ρ t \rho_t ρ t in the next section, we mean this explicit form. The trace tr ( χ n ) \operatorname{tr}(\chi_n) tr ( χ n ) is the probability that H a 1 \mathcal{H}_{a_1} H a 1 contains n n n photons, so ρ t \rho_t ρ t assigns a state, χ n / tr ( χ n ) \chi_n/\operatorname{tr}(\chi_n) χ n / tr ( χ n ) , to H a 1 \mathcal{H}_{a_1} H a 1 conditional on each number of photons in H a 1 \mathcal{H}_{a_1} H a 1 .
Measurement Statistics Determine the Twirled State
As argued at the end of the last section, it is sufficient to analyze the case S = 2 S=2 S = 2 as each equivalence class of states has a representative ρ t \rho_t ρ t that is in vacuum in all modes except a ^ 1 \hat{a}_1 a ^ 1 , a ^ 2 \hat{a}_2 a ^ 2 . Thus, to simplify our analysis, we assume S = 2 S=2 S = 2 throughout this section.
Informationally Complete Set of Operators
In this section we explore the conditions under which the measurement configuration can determine ρ t \rho_t ρ t . More specifically, we consider the measurement statistics obtained with one or two counters for different families of probe states. We first identify a minimal informationally complete set of products of mode operators on H a \mathcal{H}_a H a , whose expectations fully determine ρ t \rho_t ρ t for ρ t ∈ S \rho_t \in\mathcal{S} ρ t ∈ S . In what follows, we say that “ρ t \rho_t ρ t cannot be determined in general” by the expectations of a given set of operators, if there exists ρ t ′ \rho'_t ρ t ′ such that ρ t ′ ≠ ρ t \rho'_t \ne\rho_t ρ t ′ = ρ t , but the expectation of every operator in the set is the same on the two states. We need the following lemma.
Lemma 5.1. The anti-normally ordered monomial a ^ 1 ℓ ( a ^ 1 † ) k a ^ 2 n ( a ^ 2 † ) n \hat{a}_1^{\ell}(\hat{a}_1^\dagger)^k\hat{a}_2^n(\hat{a}_2^\dagger)^n a ^ 1 ℓ ( a ^ 1 † ) k a ^ 2 n ( a ^ 2 † ) n , where k ≥ ℓ k \ge\ell k ≥ ℓ , can be expressed as a finite linear combination of normally ordered monomials of the form ( a ^ 1 † ) k ′ a ^ 1 ℓ ′ ( a ^ 2 † ) n ′ a ^ 2 n ′ (\hat{a}_1^\dagger)^{k'}\hat{a}_1^{\ell'}(\hat{a}_2^\dagger)^{n'}\hat{a}_2^{n'} ( a ^ 1 † ) k ′ a ^ 1 ℓ ′ ( a ^ 2 † ) n ′ a ^ 2 n ′ with 0 ≤ k ′ ≤ k 0 \le k' \le k 0 ≤ k ′ ≤ k , 0 ≤ ℓ ′ ≤ ℓ 0 \le\ell' \le\ell 0 ≤ ℓ ′ ≤ ℓ , 0 ≤ n ′ ≤ n 0 \le n' \le n 0 ≤ n ′ ≤ n and k ′ − ℓ ′ = k − ℓ k' - \ell' = k - \ell k ′ − ℓ ′ = k − ℓ . Conversely, any normally ordered monomial ( a ^ 1 † ) k a ^ 1 ℓ ( a ^ 2 † ) n a ^ 2 n (\hat{a}_1^\dagger)^k\hat{a}_1^\ell(\hat{a}_2^\dagger)^n\hat{a}_2^n ( a ^ 1 † ) k a ^ 1 ℓ ( a ^ 2 † ) n a ^ 2 n with k ≥ ℓ k \ge\ell k ≥ ℓ can be expressed as a finite linear combination of anti-normally ordered monomials a ^ 1 ℓ ′ ( a ^ 1 † ) k ′ a ^ 2 n ′ ( a ^ 2 † ) n ′ \hat{a}_1^{\ell'}(\hat{a}_1^\dagger)^{k'}\hat{a}_2^{n'}(\hat{a}_2^\dagger)^{n'} a ^ 1 ℓ ′ ( a ^ 1 † ) k ′ a ^ 2 n ′ ( a ^ 2 † ) n ′ with 0 ≤ k ′ ≤ k 0 \le k' \le k 0 ≤ k ′ ≤ k , 0 ≤ ℓ ′ ≤ ℓ 0 \le\ell' \le\ell 0 ≤ ℓ ′ ≤ ℓ , 0 ≤ n ′ ≤ n 0 \le n' \le n 0 ≤ n ′ ≤ n and k ′ − ℓ ′ = k − ℓ k' - \ell' = k - \ell k ′ − ℓ ′ = k − ℓ .
Proof. To establish the first relation, we first consider a general single-mode anti-normally ordered monomial a ^ l ( a ^ † ) k \hat{a}^l(\hat{a}^{\dagger})^k a ^ l ( a ^ † ) k . Each application of the commutation rule a ^ a ^ † = a ^ † a ^ + 1 \hat{a}\hat{a}^{\dagger}=\hat{a}^{\dagger}\hat{a}+1 a ^ a ^ † = a ^ † a ^ + 1 to a given monomial produces two terms: one in which the exponents of a ^ \hat{a} a ^ and a ^ † \hat{a}^{\dagger} a ^ † are each decreased by one, and another with the same exponents but with one fewer instance of a creation operator appearing to the right of an annihilation operator. Hence, iterating this substitution converts any anti-normally ordered monomial to a finite sum of normally ordered monomials of the form ( a ^ † ) k ′ a ^ l ′ (\hat{a}^{\dagger})^{k'}\hat{a}^{l'} ( a ^ † ) k ′ a ^ l ′ , where in this sum, the exponents are bounded by the original ones, and the difference between creation and annihilation exponents is preserved ( k ′ − l ′ = k − l ) (k' - l' = k - l) ( k ′ − l ′ = k − l ) . Applying this single-mode result independently to each mode in the product a ^ 1 l ( a ^ 1 † ) k a ^ 2 n ( a ^ 2 † ) n \hat{a}_1^l(\hat{a}_1^\dagger)^k\hat{a}_2^n(\hat{a}_2^\dagger)^n a ^ 1 l ( a ^ 1 † ) k a ^ 2 n ( a ^ 2 † ) n yields the first relation. Following the same procedure, the second relation is obtained by repeatedly applying the commutation rule a ^ † a ^ = a ^ a ^ † − 1 \hat{a}^{\dagger}\hat{a}=\hat{a}\hat{a}^{\dagger}-1 a ^ † a ^ = a ^ a ^ † − 1 to the normally ordered monomial ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n (\hat{a}_1^\dagger)^k\hat{a}_1^l(\hat{a}_2^\dagger)^n\hat{a}_2^n ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n .
Proposition 5.2. Consider an arbitrary two-mode twirled state ρ t \rho_t ρ t with ρ t ∈ S \rho_t \in\mathcal{S} ρ t ∈ S . Then, ρ t \rho_t ρ t is determined by the family of expectations r ( k , l , n ) r(k,l,n) r ( k , l , n ) of the normally-ordered operators ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n (\hat{a}_1^\dagger)^k\hat{a}_1^l(\hat{a}_2^\dagger)^n\hat{a}_2^n ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n over non-negative integers k , l , n k,l,n k , l , n satisfying k ≥ l k \ge l k ≥ l with k , l , n k,l,n k , l , n not all zero. No proper subset of this family of expectations suffices for determining ρ t \rho_t ρ t .
Proof. To prove the first claim, assume the expectations r ( k , l , n ) r(k,l,n) r ( k , l , n ) are known for all k ≥ l k \ge l k ≥ l and all n n n . It follows from Lem. 5.1 that for all k ≥ l k \ge l k ≥ l and all n n n , the expectations ⟨ ( a ^ 1 † ) l ( a ^ 1 ) k a ^ 2 n ( a ^ 2 † ) n ⟩ \langle(\hat{a}_1^\dagger)^l(\hat{a}_1)^k\hat{a}_2^n(\hat{a}_2^\dagger)^n\rangle ⟨( a ^ 1 † ) l ( a ^ 1 ) k a ^ 2 n ( a ^ 2 † ) n ⟩ of the anti-normally ordered operators are also known. The expectations for k < l k < l k < l can be determined from the expectations for k ≥ l k \ge l k ≥ l . Specifically, using (15) with S = 2 S=2 S = 2 , for each k < l k < l k < l ,
r ( l , k , n ) = ∫ d α ⃗ d α ⃗ ∗ Q ( α ⃗ , α ⃗ ∗ ) α 1 l ( α 1 ∗ ) k ∣ α 2 ∣ 2 n r(l,k,n)=\int d\vec{\alpha}\,d\vec{\alpha}^{*}\,Q(\vec{\alpha},\vec{\alpha}^{*})\alpha_1^l(\alpha_1^*)^k|\alpha_2|^{2n} r ( l , k , n ) = ∫ d α d α ∗ Q ( α , α ∗ ) α 1 l ( α 1 ∗ ) k ∣ α 2 ∣ 2 n = ( ∫ d α ⃗ d α ⃗ ∗ Q ( α ⃗ , α ⃗ ∗ ) ( α 1 ∗ ) l α 1 k ∣ α 2 ∣ 2 n ) ∗ =\left(\int d\vec{\alpha}\,d\vec{\alpha}^{*}\,Q(\vec{\alpha},\vec{\alpha}^{*})(\alpha_1^*)^l\alpha_1^k|\alpha_2|^{2n}\right)^* = ( ∫ d α d α ∗ Q ( α , α ∗ ) ( α 1 ∗ ) l α 1 k ∣ α 2 ∣ 2 n ) ∗ = ⟨ a ^ 1 k ( a ^ 1 † ) l a ^ 2 n ( a ^ 2 † ) n ⟩ ∗ = r ( k , l , n ) ∗ . (26) =\langle\hat{a}_1^k(\hat{a}_1^\dagger)^l\hat{a}_2^n(\hat{a}_2^\dagger)^n\rangle^*=r(k,l,n)^*. \tag*{(26)} = ⟨ a ^ 1 k ( a ^ 1 † ) l a ^ 2 n ( a ^ 2 † ) n ⟩ ∗ = r ( k , l , n ) ∗ . (26) The collection of moments of the Husimi function of ρ t \rho_t ρ t consists of the expectations of antinormally ordered operators ⟨ a ^ 1 l ( a ^ 1 † ) k a ^ 2 n ( a ^ 2 † ) m ⟩ \langle\hat{a}_1^l(\hat{a}_1^\dagger)^k\hat{a}_2^n(\hat{a}_2^\dagger)^m\rangle ⟨ a ^ 1 l ( a ^ 1 † ) k a ^ 2 n ( a ^ 2 † ) m ⟩ . However, due to the structure of ρ t \rho_t ρ t , these vanish when n ≠ m n \ne m n = m . To see this, observe that
⟨ a ^ 1 l ( a ^ 1 † ) k a ^ 2 n ( a ^ 2 † ) m ⟩ = ∑ i tr ( χ i a ^ 1 l ( a ^ 1 † ) k ) ⟨ i ∣ a ^ 2 n ( a ^ 2 † ) m ∣ i ⟩ . (27) \langle\hat{a}_1^l(\hat{a}_1^\dagger)^k\hat{a}_2^n(\hat{a}_2^\dagger)^m\rangle
=\sum_i \operatorname{tr}\left(\chi_i\hat{a}_1^l(\hat{a}_1^\dagger)^k\right)\langle i|\hat{a}_2^n(\hat{a}_2^\dagger)^m|i\rangle. \tag*{(27)} ⟨ a ^ 1 l ( a ^ 1 † ) k a ^ 2 n ( a ^ 2 † ) m ⟩ = i ∑ tr ( χ i a ^ 1 l ( a ^ 1 † ) k ) ⟨ i ∣ a ^ 2 n ( a ^ 2 † ) m ∣ i ⟩ . (27) But for each ∣ i ⟩ |i\rangle ∣ i ⟩ , ⟨ i ∣ a ^ 2 n ( a ^ 2 † ) m ∣ i ⟩ ∝ ⟨ i + n ∣ i + m ⟩ = 0 \langle i|\hat{a}_2^n(\hat{a}_2^\dagger)^m|i\rangle\propto\langle i+n|i+m\rangle=0 ⟨ i ∣ a ^ 2 n ( a ^ 2 † ) m ∣ i ⟩ ∝ ⟨ i + n ∣ i + m ⟩ = 0 , unless n = m n=m n = m . Therefore, knowing all expectations indicated in the lemma statement implies that the Husimi function is fully determined, and hence so is ρ t \rho_t ρ t .
To prove the second claim, consider the set S N \mathcal{S}_N S N of twirled states with at most N N N photons. By the results of App. C, S N ⊂ S \mathcal{S}_N \subset\mathcal{S} S N ⊂ S . S N \mathcal{S}_N S N consists of the states of the form ρ t = ∑ i = 0 N χ i ∣ i ⟩ ⟨ i ∣ \rho_t=\sum_{i=0}^{N}\chi_i|i\rangle\langle i| ρ t = ∑ i = 0 N χ i ∣ i ⟩ ⟨ i ∣ , where χ i \chi_i χ i has at most N − i N-i N − i photons. Then, for all k ≥ l , n k \ge l,n k ≥ l , n we have
r ( k , l , n ) = ∑ i = 0 N tr ( χ i ( a ^ 1 † ) k a ^ 1 l ) ⟨ i ∣ ( a ^ 2 † ) n a ^ 2 n ∣ i ⟩ . (28) r(k,l,n)=\sum_{i=0}^{N}\operatorname{tr}\left(\chi_i(\hat{a}_1^\dagger)^k\hat{a}_1^l\right)\langle i|(\hat{a}_2^\dagger)^n\hat{a}_2^n|i\rangle. \tag*{(28)} r ( k , l , n ) = i = 0 ∑ N tr ( χ i ( a ^ 1 † ) k a ^ 1 l ) ⟨ i ∣ ( a ^ 2 † ) n a ^ 2 n ∣ i ⟩ . (28) Observe that ⟨ i ∣ ( a ^ 2 † ) n a ^ 2 n ∣ i ⟩ = 0 \langle i|(\hat{a}_2^\dagger)^n\hat{a}_2^n|i\rangle=0 ⟨ i ∣ ( a ^ 2 † ) n a ^ 2 n ∣ i ⟩ = 0 for all i < n i<n i < n , and tr ( χ i ( a ^ 1 † ) k a ^ 1 l ) = 0 \operatorname{tr}\left(\chi_i(\hat{a}_1^\dagger)^k\hat{a}_1^l\right)=0 tr ( χ i ( a ^ 1 † ) k a ^ 1 l ) = 0 whenever k > N − i k>N-i k > N − i . Thus, the expectations r ( k , l , n ) r(k,l,n) r ( k , l , n ) for n + k > N n+k>N n + k > N are zero for all states in S N \mathcal{S}_N S N . We have r ( 0 , 0 , 0 ) = 1 r(0,0,0)=1 r ( 0 , 0 , 0 ) = 1 since ρ t \rho_t ρ t has trace 1. For k = l k=l k = l , r ( k , l , n ) = r ( k , k , n ) r(k,l,n)=r(k,k,n) r ( k , l , n ) = r ( k , k , n ) is the expectation of a Hermitian operator and therefore real. For l < k l<k l < k , r ( k , l , n ) r(k,l,n) r ( k , l , n ) may be complex. We count the number of real values not yet constrained to be 0 or 1 provided by the expectations r ( k , l , n ) r(k,l,n) r ( k , l , n ) . These are provided by the family of expectations r ( k , l , n ) r(k,l,n) r ( k , l , n ) with 0 ≤ l ≤ k ≤ N − n 0\le l\le k\le N-n 0 ≤ l ≤ k ≤ N − n excluding l = k = n = 0 l=k=n=0 l = k = n = 0 . First fix the value of n n n . There are N − n + 1 N-n+1 N − n + 1 real values contributed by r ( k , k , n ) r(k,k,n) r ( k , k , n ) with n + k ≤ N n+k\le N n + k ≤ N and ( N − n ) ( N − n + 1 ) / 2 (N-n)(N-n+1)/2 ( N − n ) ( N − n + 1 ) /2 complex values for
r ( k , l , n ) r(k,l,n) r ( k , l , n ) with 0 ≤ l ≤ k ≤ N − n 0 \le l \le k \le N-n 0 ≤ l ≤ k ≤ N − n . This gives a total of ( N − n + 1 ) + ( N − n ) ( N − n + 1 ) = ( N − n + 1 ) 2 (N-n+1)+(N-n)(N-n+1)=(N-n+1)^2 ( N − n + 1 ) + ( N − n ) ( N − n + 1 ) = ( N − n + 1 ) 2 real values at fixed n n n . Summing over n n n and excluding l = k = n = 0 l=k=n=0 l = k = n = 0 gives
κ r ( N ) = − 1 + ∑ n = 0 N ( N − n + 1 ) 2 . (29) \kappa_r(N)=-1+\sum_{n=0}^{N}(N-n+1)^2. \tag*{(29)} κ r ( N ) = − 1 + n = 0 ∑ N ( N − n + 1 ) 2 . (29)
We compare κ r ( N ) \kappa_r(N) κ r ( N ) to the affine dimension of S N S_N S N , which is the number of independent real values required to specify a generic ρ t \rho_t ρ t in S N S_N S N . The operator χ i \chi_i χ i of ρ t \rho_t ρ t is an arbitrary positive semidefinite Hermitian operator on a Hilbert space of dimension N + 1 − i N+1-i N + 1 − i , and is therefore determined by ( N + 1 − i ) 2 (N+1-i)^2 ( N + 1 − i ) 2 independent real parameters. Consequently, the affine dimension of S N S_N S N is
κ ( N ) = − 1 + ∑ i = 0 N ( N − i + 1 ) 2 , (30) \kappa(N)=-1+\sum_{i=0}^{N}(N-i+1)^2, \tag*{(30)} κ ( N ) = − 1 + i = 0 ∑ N ( N − i + 1 ) 2 , (30)
where the summand − 1 -1 − 1 accounts for the unit trace condition. By comparing the expressions, we find that κ r ( N ) = κ ( N ) \kappa_r(N)=\kappa(N) κ r ( N ) = κ ( N ) .
Consider now a strict subfamily of the family of expectations r ( k , l , n ) r(k,l,n) r ( k , l , n ) with k ≥ l k\ge l k ≥ l . Suppose this subfamily does not contain r ( k 0 , l 0 , n 0 ) r(k_0,l_0,n_0) r ( k 0 , l 0 , n 0 ) with k 0 ≥ l 0 k_0\ge l_0 k 0 ≥ l 0 and k 0 , l 0 , n 0 k_0,l_0,n_0 k 0 , l 0 , n 0 not all zero. Let N = n 0 + k 0 N=n_0+k_0 N = n 0 + k 0 and consider states in S N S_N S N . Because 0 ≤ l 0 ≤ k 0 ≤ N − n 0 0\le l_0\le k_0\le N-n_0 0 ≤ l 0 ≤ k 0 ≤ N − n 0 , the number of real parameters provided by this subfamily from expectations r ( k , l , n ) r(k,l,n) r ( k , l , n ) with 0 ≤ l ≤ k ≤ N − n 0\le l\le k\le N-n 0 ≤ l ≤ k ≤ N − n with k , l , n k,l,n k , l , n not all zero is strictly less than the affine dimension of S N S_N S N . Since expectations r ( k , l , n ) r(k,l,n) r ( k , l , n ) for other triples k , l , n k,l,n k , l , n are constant, either 0 0 0 or 1 1 1 , on S N S_N S N , there are two distinct states in S N S_N S N which have the same values for all expectations in the subfamily, which establishes the second claim. □
We also consider input states with bounded photon number. If ρ \rho ρ contains at most N N N photons, then ρ t \rho_t ρ t does as well, since the map T ( ⋅ ) T(\cdot) T ( ⋅ ) , being mixture of PLT operations, cannot increase the photon number. In particular, by the results of App. C, ρ t ∈ S \rho_t\in S ρ t ∈ S . The proof of Prop. 5.2 implies the following corollary.
Corollary 5.3. As in the proof of Prop. 5.2, let S N S_N S N be the set of two-mode twirled states ρ t \rho_t ρ t with at most N N N photons. Consider the space R \mathcal{R} R of complex vectors r = ( r j ) j ∈ J \mathbf{r}=(r_j)_{j\in\mathcal{J}} r = ( r j ) j ∈ J , where J \mathcal{J} J is the family of triples k , l , n k,l,n k , l , n of non-negative integers where k ≥ l k\ge l k ≥ l and k + n ≤ N k+n\le N k + n ≤ N , with k , l , n k,l,n k , l , n not all zero. Then the linear map r : ρ t ∈ S N ↦ r ( ρ t ) ∈ R r:\rho_t\in S_N\mapsto r(\rho_t)\in\mathcal{R} r : ρ t ∈ S N ↦ r ( ρ t ) ∈ R with r k , l , n ( ρ t ) = r ( k , l , n ) = ⟨ ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n ⟩ ρ t r_{k,l,n}(\rho_t)=r(k,l,n)=\langle(\hat{a}_1^\dagger)^k\hat{a}_1^l(\hat{a}_2^\dagger)^n\hat{a}_2^n\rangle_{\rho_t} r k , l , n ( ρ t ) = r ( k , l , n ) = ⟨( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n ⟩ ρ t is one-to-one and its range is a convex set with non-empty interior.
Proof. The result follows from the first part of Prop. 5.2 and the argument given to establish the second part of that proposition. That the map is one-to-one follows from the observation that the values of r ( k , l , n ) r(k,l,n) r ( k , l , n ) for k , l , n k,l,n k , l , n not in J \mathcal{J} J are constant on S N S_N S N or determined by r ( ρ t ) r(\rho_t) r ( ρ t ) . That the range of the map is convex follows from the convexity of its domain, which the range inherits because the map is linear. That the interior of the range is non-empty follows because the real dimension of R \mathcal{R} R is the same as the affine dimension of S N S_N S N . □
Before proceeding to the next section, we present the generating functions for both the two-counter and single-counter configurations adapted to the S = 2 S=2 S = 2 case. In particular, when two counters are used, (14) becomes
F γ ( u , v ) = e ∣ η α 1 + ζ γ ∣ 2 u + ∣ ζ α 1 + η γ ′ ∣ 2 v + ∣ η α 2 ∣ 2 u + ∣ ζ α 2 ∣ 2 v . (31) F_\gamma(u,v)=e^{|\eta\alpha_1+\zeta\gamma|^2u+|\zeta\alpha_1+\eta\gamma'|^2v+|\eta\alpha_2|^2u+|\zeta\alpha_2|^2v}. \tag*{(31)} F γ ( u , v ) = e ∣ η α 1 + ζ γ ∣ 2 u + ∣ ζ α 1 + η γ ′ ∣ 2 v + ∣ η α 2 ∣ 2 u + ∣ ζ α 2 ∣ 2 v . (31)
If the measurement configuration involves a single counter, we denote the corresponding generating function by F 1 , γ ( u ) F_{1,\gamma}(u) F 1 , γ ( u ) :
F 1 , γ ( u ) = e ∣ η α 1 + ζ γ ∣ 2 u + ∣ η α 2 ∣ 2 u = e ( ∣ η α ⃗ ∣ 2 + η ζ ∗ γ ∗ α 1 + ζ η ∗ γ α 1 ∗ + ∣ ζ γ ∣ 2 ) u . (32) F_{1,\gamma}(u)=e^{|\eta\alpha_1+\zeta\gamma|^2u+|\eta\alpha_2|^2u}
=e^{\left(|\eta\vec{\alpha}|^2+\eta\zeta^*\gamma^*\alpha_1+\zeta\eta^*\gamma\alpha_1^*+|\zeta\gamma|^2\right)u}. \tag*{(32)} F 1 , γ ( u ) = e ∣ η α 1 + ζ γ ∣ 2 u + ∣ η α 2 ∣ 2 u = e ( ∣ η α ∣ 2 + η ζ ∗ γ ∗ α 1 + ζ η ∗ γ α 1 ∗ + ∣ ζ γ ∣ 2 ) u . (32)
We consider this measurement configuration first.
Results With One Counter
In this section we present our results for the measurement configuration with one counter. We first introduce several definitions and notational conventions that will be used throughout the statements and proofs of the subsequent results. For all non-negative integers k , l , n k,l,n k , l , n , we define the following amplitude polynomials:
m ( k , l , n ) = E [ ( a ^ 1 † ) k a ^ 1 l a ^ 2 † n a ^ 2 n ] = ( α 1 ∗ ) k α 1 l ∣ α 2 ∣ 2 n m(k,l,n)=\mathcal{E}\left[(\hat{a}_1^\dagger)^k\hat{a}_1^l\hat{a}_2^{\dagger n}\hat{a}_2^n\right]=(\alpha_1^*)^k\alpha_1^l|\alpha_2|^{2n} m ( k , l , n ) = E [ ( a ^ 1 † ) k a ^ 1 l a ^ 2 † n a ^ 2 n ] = ( α 1 ∗ ) k α 1 l ∣ α 2 ∣ 2 n
m ′ ( k , l , n ) = E [ ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n ] = ( α 1 ∗ ) k α 1 l ∣ α 2 ∣ 2 n . m'(k,l,n)=\mathcal{E}\left[(\hat{a}_1^\dagger)^k\hat{a}_1^l(\hat{a}_2^\dagger)^n\hat{a}_2^n\right]=(\alpha_1^*)^k\alpha_1^l|\alpha_2|^{2n}. m ′ ( k , l , n ) = E [ ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n ] = ( α 1 ∗ ) k α 1 l ∣ α 2 ∣ 2 n .
For each non-negative integer N N N , we define the sets of these expressions restricted by their degrees as M N = { m ( k , l , n ) : n + max ( k , l ) ≤ N } \mathcal{M}_N=\{m(k,l,n):n+\max(k,l)\le N\} M N = { m ( k , l , n ) : n + max ( k , l ) ≤ N } and M N ′ = { m ′ ( k , l , n ) : n + max ( k , l ) ≤ N } \mathcal{M}'_N=\{m'(k,l,n):n+\max(k,l)\le N\} M N ′ = { m ′ ( k , l , n ) : n + max ( k , l ) ≤ N } . For convenience, we define M − 1 = M − 1 ′ = ∅ \mathcal{M}_{-1}=\mathcal{M}'_{-1}=\varnothing M − 1 = M − 1 ′ = ∅ . We have M N ⊂ M N ′ \mathcal{M}_N\subset\mathcal{M}_{N'} M N ⊂ M N ′ and M N ′ ⊂ M N ′ ′ \mathcal{M}'_N\subset\mathcal{M}'_{N'} M N ′ ⊂ M N ′ ′ for N ≤ N ′ N\le N' N ≤ N ′ .
We adopt the notation of expectation brackets ⟨ ⟩ \langle\ \rangle ⟨ ⟩ around an amplitude polynomial P P P to denote the expectation value of the corresponding operator E − 1 [ P ] \mathcal{E}^{-1}[P] E − 1 [ P ] with respect to the twirled state ρ t \rho_t ρ t under consideration. We refer to this expectation simply as the expectation of P P P . For example, ⟨ m ′ ( k , l , n ) ⟩ ≡ tr ( ρ t ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n ) \langle m'(k,l,n)\rangle\equiv\operatorname{tr}\bigl(\rho_t(\hat{a}_1^\dagger)^k\hat{a}_1^l(\hat{a}_2^\dagger)^n\hat{a}_2^n\bigr) ⟨ m ′ ( k , l , n )⟩ ≡ tr ( ρ t ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n ) . In the proof of Prop. 5.2 we analyzed the set of expectations r ( k , l , n ) r(k,l,n) r ( k , l , n ) with 0 ≤ l ≤ k ≤ N − n 0\le l\le k\le N-n 0 ≤ l ≤ k ≤ N − n . Because r ( k , l , n ) = r ( l , k , n ) ∗ r(k,l,n)=r(l,k,n)^* r ( k , l , n ) = r ( l , k , n ) ∗ , the set of expectations r ( k , l , n ) r(k,l,n) r ( k , l , n ) with 0 ≤ max ( k , l ) ≤ N − n 0\le\max(k,l)\le N-n 0 ≤ max ( k , l ) ≤ N − n may be identified with the set of expectations of members of M N \mathcal{M}_N M N . Furthermore, for ρ t ∈ S N \rho_t\in S_N ρ t ∈ S N the expectations for the m ′ ( k , l , n ) m'(k,l,n) m ′ ( k , l , n ) not in M N ′ \mathcal{M}'_N M N ′ are zero and ρ t \rho_t ρ t is determined by the expectations of the members of M N ′ \mathcal{M}'_N M N ′ .
Lemma 5.4. Every element of M N ∖ M N − 1 \mathcal{M}_N \setminus\mathcal{M}_{N-1} M N ∖ M N − 1 can be expressed as a linear combination of elements in M N ′ ∖ M N − 1 ′ \mathcal{M}'_N \setminus\mathcal{M}'_{N-1} M N ′ ∖ M N − 1 ′ , and conversely, every element of M N ′ ∖ M N − 1 ′ \mathcal{M}'_N \setminus\mathcal{M}'_{N-1} M N ′ ∖ M N − 1 ′ can be expressed as a linear combination of elements in M N ∖ M N − 1 \mathcal{M}_N \setminus\mathcal{M}_{N-1} M N ∖ M N − 1 .
Proof. M N ∖ M N − 1 \mathcal{M}_N \setminus\mathcal{M}_{N-1} M N ∖ M N − 1 consists of the m ( k , l , n ) m(k,l,n) m ( k , l , n ) with max ( k , l ) = N − n \max(k,l)=N-n max ( k , l ) = N − n and similarly for M N ′ ∖ M N − 1 ′ \mathcal{M}'_N \setminus\mathcal{M}'_{N-1} M N ′ ∖ M N − 1 ′ . The result follows directly from expanding the magnitude terms ∣ α ⃗ ∣ 2 n \lvert\vec{\alpha}\rvert^{2n} ∣ α ∣ 2 n via the binomial theorem, utilizing the relation ∣ α ⃗ ∣ 2 = ∣ α 1 ∣ 2 + ∣ α 2 ∣ 2 \lvert\vec{\alpha}\rvert^2=\lvert\alpha_1\rvert^2+\lvert\alpha_2\rvert^2 ∣ α ∣ 2 = ∣ α 1 ∣ 2 + ∣ α 2 ∣ 2 . We can express the elements of M N ′ \mathcal{M}'_N M N ′ in terms of M N \mathcal{M}_N M N as
m ′ ( k , l , n ) = ( α 1 ∗ ) k α 1 l ( ∣ α ⃗ ∣ 2 − ∣ α 1 ∣ 2 ) n = ∑ j = 0 n ( n j ) ( − 1 ) j ( α 1 ∗ ) k α 1 l ∣ α 1 ∣ 2 j ∣ α ⃗ ∣ 2 ( n − j ) = ∑ j = 0 n ( n j ) ( − 1 ) j m ( k + j , l + j , n − j ) . (33) \begin{aligned}
m'(k,l,n)=(\alpha_1^*)^k\alpha_1^l(\lvert\vec{\alpha}\rvert^2-\lvert\alpha_1\rvert^2)^n \\
&=\sum_{j=0}^{n}\binom{n}{j}(-1)^j(\alpha_1^*)^k\alpha_1^l\lvert\alpha_1\rvert^{2j}\lvert\vec{\alpha}\rvert^{2(n-j)} \\
&=\sum_{j=0}^{n}\binom{n}{j}(-1)^jm(k+j,l+j,n-j). \tag*{(33)}
\end{aligned} m ′ ( k , l , n ) = ( α 1 ∗ ) k α 1 l (∣ α ∣ 2 − ∣ α 1 ∣ 2 ) n = j = 0 ∑ n ( j n ) ( − 1 ) j ( α 1 ∗ ) k α 1 l ∣ α 1 ∣ 2 j ∣ α ∣ 2 ( n − j ) = j = 0 ∑ n ( j n ) ( − 1 ) j m ( k + j , l + j , n − j ) . (33)
For the converse, we can express the elements of M N \mathcal{M}_N M N in terms of M N ′ \mathcal{M}'_N M N ′ as
m ( k , l , n ) = ( α 1 ∗ ) k α 1 l ( ∣ α 1 ∣ 2 + ∣ α 2 ∣ 2 ) n = ∑ j = 0 n ( n j ) ( α 1 ∗ ) k α 1 l ∣ α 1 ∣ 2 j ∣ α 2 ∣ 2 ( n − j ) = ∑ j = 0 n ( n j ) m ′ ( k + j , l + j , n − j ) . (34) \begin{aligned}
m(k,l,n)=(\alpha_1^*)^k\alpha_1^l(\lvert\alpha_1\rvert^2+\lvert\alpha_2\rvert^2)^n \\
&=\sum_{j=0}^{n}\binom{n}{j}(\alpha_1^*)^k\alpha_1^l\lvert\alpha_1\rvert^{2j}\lvert\alpha_2\rvert^{2(n-j)} \\
&=\sum_{j=0}^{n}\binom{n}{j}m'(k+j,l+j,n-j). \tag*{(34)}
\end{aligned} m ( k , l , n ) = ( α 1 ∗ ) k α 1 l (∣ α 1 ∣ 2 + ∣ α 2 ∣ 2 ) n = j = 0 ∑ n ( j n ) ( α 1 ∗ ) k α 1 l ∣ α 1 ∣ 2 j ∣ α 2 ∣ 2 ( n − j ) = j = 0 ∑ n ( j n ) m ′ ( k + j , l + j , n − j ) . (34)
The lemma follows because max ( k , l ) + n = max ( k + j , l + j ) + n − j \max(k,l)+n=\max(k+j,l+j)+n-j max ( k , l ) + n = max ( k + j , l + j ) + n − j .
We previously showed that the expectations of m ′ ( k , l , n ) ∉ M N ′ m'(k,l,n)\notin\mathcal{M}'_N m ′ ( k , l , n ) ∈ / M N ′ are zero when ρ t ∈ S N \rho_t\in S_N ρ t ∈ S N . Observe that as a consequence of the above lemma, the expectations of the m ( k , l , n ) ∉ M N m(k,l,n)\notin\mathcal{M}_N m ( k , l , n ) ∈ / M N are also zero when ρ t ∈ S N \rho_t\in S_N ρ t ∈ S N . Cor. 5.3 can be phrased in terms of the expectations of the m ′ ( k , l , n ) m'(k,l,n) m ′ ( k , l , n ) . Since the above lemma implies that there is an invertible linear relationship between M N \mathcal{M}_N M N and M N ′ \mathcal{M}'_N M N ′ , this observation enables replacing m ′ ( k , l , n ) m'(k,l,n) m ′ ( k , l , n ) in the proof of this corollary with m ( k , l , n ) m(k,l,n) m ( k , l , n ) , which yields the following result.
Corollary 5.5. The conclusion of Cor. 5.3 holds if the linear map r ( ρ t ) r(\rho_t) r ( ρ t ) is replaced by the map s : ρ t ↦ ( ⟨ m ( k , l , n ) ⟩ ρ t ) k , l , n : l ≤ k ≤ N − k , ( k , l , n ) ≠ ( 0 , 0 , 0 ) s:\rho_t\mapsto(\langle m(k,l,n)\rangle_{\rho_t})_{k,l,n:l\leq k\leq N-k,(k,l,n)\neq(0,0,0)} s : ρ t ↦ (⟨ m ( k , l , n ) ⟩ ρ t ) k , l , n : l ≤ k ≤ N − k , ( k , l , n ) = ( 0 , 0 , 0 ) .
In the following proofs, we make use of the fact that the product of two generating functions yields another generating function, with coefficients given by finite linear combinations of the products of the original coefficients. In particular, we find it useful to work with the generating function
F 1 , γ ′ ( u ) = e − ∣ ζ γ ∣ 2 u F 1 , γ ( u ) = e ( ∣ η ⃗ ∣ 2 + η ζ γ ∗ α 1 + ζ γ η ∗ α 1 ∗ ) u . (35) \begin{aligned}
F'_{1,\gamma}(u)=e^{-\lvert\zeta_\gamma\rvert^2u}F_{1,\gamma}(u) \\
&=e^{(\lvert\vec{\eta}\rvert^2+\eta\zeta_\gamma^*\alpha_1+\zeta_\gamma\eta^*\alpha_1^*)u}. \tag*{(35)}
\end{aligned} F 1 , γ ′ ( u ) = e − ∣ ζ γ ∣ 2 u F 1 , γ ( u ) = e (∣ η ∣ 2 + η ζ γ ∗ α 1 + ζ γ η ∗ α 1 ∗ ) u . (35)
The coefficient of u n u^n u n in F 1 , γ ′ ( u ) F'_{1,\gamma}(u) F 1 , γ ′ ( u ) is a discrete convolution of the first n + 1 n+1 n + 1 coefficients of F 1 , γ ( u ) F_{1,\gamma}(u) F 1 , γ ( u ) and e − ∣ ζ γ ∣ 2 u e^{-\lvert\zeta_\gamma\rvert^2u} e − ∣ ζ γ ∣ 2 u , as can be seen by writing F 1 , γ ( u ) F_{1,\gamma}(u) F 1 , γ ( u ) and e − ∣ ζ γ ∣ 2 u e^{-\lvert\zeta_\gamma\rvert^2u} e − ∣ ζ γ ∣ 2 u in terms of their explicit formal power series, formally multiplying, and adding terms of equal degree in u u u . The coefficients of F 1 , γ ( u ) F_{1,\gamma}(u) F 1 , γ ( u ) encode the operators’ polynomial correspondence. Because the coefficients of e − ∣ ζ γ ∣ 2 u e^{-\lvert\zeta_\gamma\rvert^2u} e − ∣ ζ γ ∣ 2 u are simply scalar constants proportional to ( − ∣ ζ γ ∣ 2 ) j (-\lvert\zeta_\gamma\rvert^2)^j ( − ∣ ζ γ ∣ 2 ) j , it follows that the operators corresponding to the coefficients of F 1 , γ ′ ( u ) F'_{1,\gamma}(u) F 1 , γ ′ ( u ) are strictly finite linear combinations of the operators in C 1 ( γ ) \mathcal{C}_1(\gamma) C 1 ( γ ) . Conversely, because the exponential factor e − ∣ ζ γ ∣ 2 u e^{-\lvert\zeta_\gamma\rvert^2u} e − ∣ ζ γ ∣ 2 u is invertible, the coefficients of F 1 , γ ( u ) F_{1,\gamma}(u) F 1 , γ ( u ) can similarly be expressed as linear combinations of the coefficients of F 1 , γ ′ ( u ) F'_{1,\gamma}(u) F 1 , γ ′ ( u ) .
This relationship between the coefficients of F 1 , γ ( u ) F_{1,\gamma}(u) F 1 , γ ( u ) and F 1 , γ ′ ( u ) F'_{1,\gamma}(u) F 1 , γ ′ ( u ) motivates a formal definition: we consider two generating functions to be equivalent if their respective sets of coefficients algebraically span the same linear space.
Lemma 5.6. The generating functions F 1 , γ ( u ) F_{1,\gamma}(u) F 1 , γ ( u ) and F 1 , γ ′ ( u ) F'_{1,\gamma}(u) F 1 , γ ′ ( u ) are equivalent.
Proof. This follows from the preceding discussion. □
We begin by assuming the measurement statistics are known for a finite number of probe amplitudes. For each non-negative integer N N N , let P N \mathcal{P}_N P N denote the space of polynomials spanned by all monomials of the form γ k ( γ ∗ ) l \gamma^k(\gamma^*)^l γ k ( γ ∗ ) l , where k , l ≤ N k,l\leq N k , l ≤ N .
Theorem 5.7. Consider a measurement configuration with a single counter. Suppose the input state, and hence the two-mode twirled state ρ t \rho_t ρ t , contains at most N N N photons. Let Γ = { γ 1 , … , γ K } ⊂ C \Gamma=\{\gamma_1,\ldots,\gamma_K\}\subset\mathbb{C} Γ = { γ 1 , … , γ K } ⊂ C be a set satisfying the following condition: if a polynomial p ∈ P N p\in\mathcal{P}_N p ∈ P N satisfies p ( γ i ) = 0 p(\gamma_i)=0 p ( γ i ) = 0 for all γ i ∈ Γ \gamma_i\in\Gamma γ i ∈ Γ , then p = 0 p=0 p = 0 . Then, ρ t \rho_t ρ t is determined by the collection of the measurement statistics for each probe amplitude γ ∈ Γ \gamma\in\Gamma γ ∈ Γ .
Proof. We show that the expectations of all operators ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n (\hat{a}_1^\dagger)^k\hat{a}_1^l(\hat{a}_2^\dagger)^n\hat{a}_2^n ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n with n + max ( k , l ) ≤ N n+\max(k,l)\leq N n + max ( k , l ) ≤ N are determined by the measurement statistics. The theorem statement then follows from Cor. 5.3. Note that the set of E [ ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n ] \mathbb{E}[(\hat{a}_1^\dagger)^k\hat{a}_1^l(\hat{a}_2^\dagger)^n\hat{a}_2^n] E [( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n ] with n + max ( k , l ) ≤ N n+\max(k,l)\leq N n + max ( k , l ) ≤ N is exactly M N ′ \mathcal{M}'_N M N ′ . Lem. 5.4 shows that M N ′ \mathcal{M}'_N M N ′ is in the span of M N \mathcal{M}_N M N , so it suffices to show that the expectations of the operators in M N \mathcal{M}_N M N can be extracted from the generating functions F 1 , γ ′ ( u ) F'_{1,\gamma}(u) F 1 , γ ′ ( u ) for the γ ∈ Γ \gamma\in\Gamma γ ∈ Γ , which we proceed to do. By “extracted” we mean that they can be obtained from the expectations of the coefficients of the generating functions by taking their linear combinations.
The assumption on Γ \Gamma Γ in the theorem statement implies that the map defined by p ∈ P N ↦ ( p ( γ i ) ) i = 1 K p\in\mathcal{P}_N\mapsto(p(\gamma_i))_{i=1}^K p ∈ P N ↦ ( p ( γ i ) ) i = 1 K is linear and injective. We can express this map in the basis of monomials ( γ ∗ ) l γ k ∈ P N (\gamma^*)^l\gamma^k \in\mathcal{P}_N ( γ ∗ ) l γ k ∈ P N as a matrix with independent columns, which implies that the row span of the matrix is complete. In particular, for every basis vector, there is a linear combination of the rows identical to this basis vector. The matrix entry corresponding to monomial p p p and row i i i is p ( γ i ) p(\gamma_i) p ( γ i ) . It follows that for any given monomial p ∈ P N p \in\mathcal{P}_N p ∈ P N , there is a tuple of coefficients ( c i ) i = 1 K (c_i)_{i=1}^K ( c i ) i = 1 K such that for all monomials q ∈ P N q \in\mathcal{P}_N q ∈ P N different from p p p ,
∑ i = 1 K c i q ( γ i ) = 0 , \sum_{i=1}^{K} c_iq(\gamma_i) = 0, i = 1 ∑ K c i q ( γ i ) = 0 ,
∑ i = 1 K c i p ( γ i ) = 1. (36) \sum_{i=1}^{K} c_ip(\gamma_i) = 1. \tag*{(36)} i = 1 ∑ K c i p ( γ i ) = 1. (36)
We define S p ( u ) = ∑ i = 1 K c i F 1 , γ i ′ ( u ) S_p(u) = \sum_{i=1}^{K} c_iF'_{1,\gamma_i}(u) S p ( u ) = ∑ i = 1 K c i F 1 , γ i ′ ( u ) . As a generating function, the coefficients of S p ( u ) S_p(u) S p ( u ) are linear combinations of the coefficients of the F 1 , γ i ′ ( u ) F'_{1,\gamma_i}(u) F 1 , γ i ′ ( u ) . The coefficient of u n u^n u n in S p ( u ) S_p(u) S p ( u ) is given by
( n ! ) coeff u n S p = ∑ i = 1 K c i ( ∣ η α ⃗ ∣ 2 + η ζ i ∗ α 1 + ζ η i ∗ α 1 ∗ ) n = ∑ i = 1 K c i ∑ r + s + t = n ( n r , s , t ) ( η ζ i ∗ α 1 ) r ( γ i ∗ ) r ( ζ η i ∗ α 1 ∗ ) s γ i s ∣ η ∣ 2 t = ∑ r + s ≤ n ( n r , s , n − r − s ) ( ζ η ∗ ) s ( η ζ ∗ ) r ∣ η ∣ 2 ( n − r − s ) ( ∑ i = 1 K c i ( γ i ∗ ) r γ i s ) m ( s , r , n − r − s ) . (37) \begin{aligned}
(n!)\operatorname{coeff}_{u^n} S_p = \sum_{i=1}^{K} c_i\left(|\eta_{\vec{\alpha}}|^2+\eta\zeta_i^*\alpha_1+\zeta\eta_i^*\alpha_1^*\right)^n \\
&= \sum_{i=1}^{K}c_i\sum_{r+s+t=n}\binom{n}{r,s,t}(\eta\zeta_i^*\alpha_1)^r(\gamma_i^*)^r(\zeta\eta_i^*\alpha_1^*)^s\gamma_i^s|\eta|^{2t} \\
&= \sum_{r+s\leq n}\binom{n}{r,s,n-r-s}(\zeta\eta^*)^s(\eta\zeta^*)^r|\eta|^{2(n-r-s)}
\left(\sum_{i=1}^{K}c_i(\gamma_i^*)^r\gamma_i^s\right)m(s,r,n-r-s). \tag*{(37)}
\end{aligned} ( n !) coeff u n S p = i = 1 ∑ K c i ( ∣ η α ∣ 2 + η ζ i ∗ α 1 + ζ η i ∗ α 1 ∗ ) n = i = 1 ∑ K c i r + s + t = n ∑ ( r , s , t n ) ( η ζ i ∗ α 1 ) r ( γ i ∗ ) r ( ζ η i ∗ α 1 ∗ ) s γ i s ∣ η ∣ 2 t = r + s ≤ n ∑ ( r , s , n − r − s n ) ( ζ η ∗ ) s ( η ζ ∗ ) r ∣ η ∣ 2 ( n − r − s ) ( i = 1 ∑ K c i ( γ i ∗ ) r γ i s ) m ( s , r , n − r − s ) . (37)
By construction, when r , s ≤ N r,s\leq N r , s ≤ N , the sum ∑ i = 1 K c i ( γ i ∗ ) r γ i s = 0 \sum_{i=1}^{K}c_i(\gamma_i^*)^r\gamma_i^s=0 ∑ i = 1 K c i ( γ i ∗ ) r γ i s = 0 unless r = l r=l r = l and s = k s=k s = k . Consequently, if n < k + l n<k+l n < k + l , the entire expression vanishes. For n ≥ k + l n\geq k+l n ≥ k + l , we consider the expectation of the coefficient. Because ρ t \rho_t ρ t contains at most N N N photons, the expectations ⟨ m ( s , r , t ) ⟩ \langle m(s,r,t)\rangle ⟨ m ( s , r , t )⟩ strictly vanish whenever the relevant degree exceeds this bound, namely when r + t > N r+t>N r + t > N or s + t > N s+t>N s + t > N . This truncation ensures that the value of ∑ i = 1 K c i ( γ i ∗ ) r γ i s \sum_{i=1}^{K}c_i(\gamma_i^*)^r\gamma_i^s ∑ i = 1 K c i ( γ i ∗ ) r γ i s is irrelevant whenever r > N r>N r > N or s > N s>N s > N , as it multiplies a zero expectation. Taking the expectation on both sides of (37) thus yields
n ! ⟨ coeff u n S p ⟩ = ( n l , k , n − k − l ) × ( ζ η ∗ ) k ( η ζ ∗ ) l ∣ η ∣ 2 ( n − k − l ) ⟨ m ( k , l , n − k − l ) ⟩ n!\langle\operatorname{coeff}_{u^n}S_p\rangle
=
\binom{n}{l,k,n-k-l}
\times(\zeta\eta^*)^k(\eta\zeta^*)^l|\eta|^{2(n-k-l)}
\langle m(k,l,n-k-l)\rangle n ! ⟨ coeff u n S p ⟩ = ( l , k , n − k − l n ) × ( ζ η ∗ ) k ( η ζ ∗ ) l ∣ η ∣ 2 ( n − k − l ) ⟨ m ( k , l , n − k − l )⟩
when n ≥ k + l n\geq k+l n ≥ k + l such that n − l ≤ N n-l\leq N n − l ≤ N and n − k ≤ N n-k\leq N n − k ≤ N , and ⟨ coeff u n S p ⟩ = 0 \langle\operatorname{coeff}_{u^n}S_p\rangle=0 ⟨ coeff u n S p ⟩ = 0 otherwise, so that ⟨ m ( k , l , n − k − l ) ⟩ \langle m(k,l,n-k-l)\rangle ⟨ m ( k , l , n − k − l )⟩ can be determined from the expectation of coeff u n S p \operatorname{coeff}_{u^n}S_p coeff u n S p of S p ( u ) S_p(u) S p ( u ) . Performing this procedure for all monomials γ l ( γ ∗ ) k ∈ P N \gamma^l(\gamma^*)^k\in\mathcal{P}_N γ l ( γ ∗ ) k ∈ P N extracts the expectation of every member of M N \mathcal{M}_N M N , which completes the proof. □
Thm. 5.7 indicates that, for ρ t \rho_t ρ t with at most N N N photons, a minimal number h 1 ( N ) h_1(N) h 1 ( N ) of probe amplitudes exists that are necessary and sufficient to determine ρ t \rho_t ρ t .
Corollary 5.8. Consider a measurement configuration with a single counter. Suppose the input state, and hence the two-mode twirled state ρ t \rho_t ρ t , contains at most N N N photons. Suppose the conditions of Thm. 5.7 are satisfied. Then, there exist ( N + 1 ) 2 (N+1)^2 ( N + 1 ) 2 distinct probe amplitudes that determine ρ t \rho_t ρ t . Moreover, almost every set of ( N + 1 ) 2 (N+1)^2 ( N + 1 ) 2 distinct probe amplitudes possesses this property.
Proof. The claim implies that one can take K = ( N + 1 ) 2 K=(N+1)^2 K = ( N + 1 ) 2 in Thm. 5.7. Namely, there exist ( N + 1 ) 2 (N+1)^2 ( N + 1 ) 2 distinct γ i ∈ C \gamma_i\in\mathbb{C} γ i ∈ C such that the only polynomial p ∈ P N p\in\mathcal{P}_N p ∈ P N that satisfies p ( γ i ) = 0 p(\gamma_i)=0 p ( γ i ) = 0 for all i = 1 , … , ( N + 1 ) 2 i=1,\ldots,(N+1)^2 i = 1 , … , ( N + 1 ) 2 is the zero polynomial. This is a standard result in Lagrange interpolation theory on a real-analytic polynomial space over C \mathbb{C} C , where γ \gamma γ and γ ∗ \gamma^* γ ∗ are treated as independent variables [40 ] . The last claim follows from the fact that the Lagrange interpolation map p ↦ ( p ( γ i ) ) i = 1 K p\mapsto(p(\gamma_i))_{i=1}^K p ↦ ( p ( γ i ) ) i = 1 K fails to be injective only for ( γ i ) i = 1 K (\gamma_i)_{i=1}^K ( γ i ) i = 1 K in a Lebesgue-null subset of C K \mathbb{C}^K C K (see Cor. 5.6.3 in Ref. [40 ] ). □
Cor. 5.8 implies that h 1 ( N ) ≤ ( N + 1 ) 2 h_1(N)\leq(N+1)^2 h 1 ( N ) ≤ ( N + 1 ) 2 since Thm. 5.7 shows only that the stated number of probe amplitudes are sufficient for determining ρ t \rho_t ρ t .
Next, we consider the case where measurement statistics are known for an infinite number of probe amplitudes and we do not assume an upper bound on the number of photons. To that end, for each non-negative integer N N N , we introduce the set M = N = { m ( k , l , n ) : k + l + n = N } \mathcal{M}_{=N}=\{m(k,l,n):k+l+n=N\} M = N = { m ( k , l , n ) : k + l + n = N } . We further denote by P N , ≥ \mathcal{P}_{N,\geq} P N , ≥ the polynomial space spanned by all monomials of the form ( γ ∗ ) l γ k (\gamma^*)^l\gamma^k ( γ ∗ ) l γ k satisfying k + l ≤ N k+l\leq N k + l ≤ N .
Theorem 5.9. Consider a measurement configuration with a single counter. Consider an arbitrary two-mode twirled state ρ t \rho_t ρ t with ρ t ∈ S \rho_t\in\mathcal{S} ρ t ∈ S . If the measurement statistics are known for all probe amplitudes in a set Γ \Gamma Γ of positive Lebesgue measure, then ρ t \rho_t ρ t is uniquely determined.
Proof. We show that the expectations of the operators ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n (\hat{a}_1^\dagger)^k\hat{a}_1^l(\hat{a}_2^\dagger)^n\hat{a}_2^n ( a ^ 1 † ) k a ^ 1 l ( a ^ 2 † ) n a ^ 2 n for all non-negative integers k , l , n k,l,n k , l , n are determined by the measurement statistics. The theorem statement then follows from Prop. 5.2.
Note that ( n ! ) coeff u n S n ( u ) (n!)\operatorname{coeff}_{u^n}S_n(u) ( n !) coeff u n S n ( u ) in (37) depends only on the terms m ( r , s , t ) m(r,s,t) m ( r , s , t ) with r + s + t = n r+s+t=n r + s + t = n , which exactly constitute the set M = n \mathcal{M}_{=n} M = n . Since ⋃ N M N = ⋃ N M = N \bigcup_N\mathcal{M}_N=\bigcup_N\mathcal{M}_{=N} ⋃ N M N = ⋃ N M = N , it suffices to show that for every N N N , M = N \mathcal{M}_{=N} M = N can be extracted from the coefficients of the generating functions F 1 , γ ′ ( u ) F'_{1,\gamma}(u) F 1 , γ ′ ( u ) for probe amplitudes γ \gamma γ in the given neighborhood.
As stated in the proof of Cor. 5.8, the polynomial evaluation map p ∈ P N , ≥ ↦ ( p ( γ i ) ) i = 1 K p\in\mathcal{P}_{N,\geq}\mapsto(p(\gamma_i))_{i=1}^K p ∈ P N , ≥ ↦ ( p ( γ i ) ) i = 1 K , where K K K is a function of N N N satisfying the conditions of Thm. 5.7, fails to be injective only for ( γ i ) i = 1 K (\gamma_i)_{i=1}^K ( γ i ) i = 1 K in a measure-zero subset of C K \mathbb{C}^K C K . Since γ \gamma γ has positive measure, so does Γ K \Gamma^K Γ K for all K K K . Consequently, for any integer N N N , we can always choose a tuple of K ( N ) K(N) K ( N ) probe amplitudes ( γ i ) i = 1 K ( N ) ∈ Γ K ( N ) (\gamma_i)_{i=1}^{K(N)}\in\Gamma^{K(N)} ( γ i ) i = 1 K ( N ) ∈ Γ K ( N ) such that the map p ∈ P N , ≥ ↦ ( p ( γ i ) ) i = 1 K ( N ) p\in\mathcal{P}_{N,\geq}\mapsto(p(\gamma_i))_{i=1}^{K(N)} p ∈ P N , ≥ ↦ ( p ( γ i ) ) i = 1 K ( N ) is injective.
The number of required amplitudes, K ( N ) K(N) K ( N ) , can be chosen to match the dimension of P N , > \mathcal{P}_{N,>} P N , > , which is exactly the number of pairs ( k , l ) (k,l) ( k , l ) satisfying k + l ≤ N k+l \le N k + l ≤ N , namely
K ( N ) = 1 2 ( N + 1 ) ( N + 2 ) . K(N) = \frac{1}{2}(N+1)(N+2). K ( N ) = 2 1 ( N + 1 ) ( N + 2 ) .
With N N N fixed, for each target monomial p = ( γ ∗ ) l γ k p=(\gamma^*)^l\gamma^k p = ( γ ∗ ) l γ k satisfying k + l ≤ N k+l \le N k + l ≤ N , we construct the linear combination S p ( u ) = ∑ i = 1 K ( N ) c i F 1 , γ i ( u ) S_p(u)=\sum_{i=1}^{K(N)}c_iF_{1,\gamma_i}(u) S p ( u ) = ∑ i = 1 K ( N ) c i F 1 , γ i ( u ) as in the proof of Thm. 5.7, choosing the coefficients c i c_i c i such that ∑ i c i q ( γ i ) = δ q p \sum_i c_iq(\gamma_i)=\delta_{qp} ∑ i c i q ( γ i ) = δ q p for q ∈ P N , > q\in\mathcal{P}_{N,>} q ∈ P N , > . Then, the N N N -th order term in the expansion of S p ( u ) S_p(u) S p ( u ) in Eq. (37) , is proportional to m ( k , l , N − k − l ) m(k,l,N-k-l) m ( k , l , N − k − l ) . Because the expansion is strictly restricted to terms m ( r , s , t ) m(r,s,t) m ( r , s , t ) where r + s + t = N r+s+t=N r + s + t = N , no terms of degree higher than N N N appear in the summation, rendering the physical photon-number truncation step from Thm. 5.7 unnecessary. Performing this procedure for each value of N N N extracts all monomials in ⋃ N M = N \bigcup_{N}\mathcal{M}_{=N} ⋃ N M = N , which completes the proof.
In the remainder of this section we establish two no-go results for general states in S \mathcal{S} S . The first establishes that lim N → ∞ h 1 ( N ) \lim_{N\to\infty}h_1(N) lim N → ∞ h 1 ( N ) diverges. Namely, no finite collection of probe amplitudes are sufficient to obtain ρ t \rho_t ρ t in the absence of a bound on total photon number. The second shows that if only one counter is used, measuring at just one probe magnitude is insufficient. The proofs of the results first establish that there are states with bounded photon numbers that cannot be distinguished based on knowledge of the photon-number moment sequences associated with the observed photon number distributions at the set of probe amplitudes under consideration. States with bounded photon numbers have determinate photon-number moment sequences (see Cor. C.5 in Appendix § 1.8 ), which implies that knowledge of the photon-number moment sequence is equivalent to knowledge of the photon-number probability distribution. The no-go results therefore imply that there are states that cannot be distinguished based on the measurement statistics, that is, the observed photon number distributions at the given set of probe amplitudes. A subtle point is that with this proof strategy, the two states that cannot be distinguished based on the photon-number moment sequence must also be indistinguishable by their photon-number distributions to complete the no-go result. To ensure the latter indistinguishability, we need the photon-number moment sequence to be determinate in the sense that there is a unique photon-number probability distribution with this moment sequence. We do not know whether all states in S \mathcal{S} S have determinate photon-number moment sequences.
Proposition 5.10. Consider a measurement configuration with a single counter. Consider an arbitrary two-mode twirled state ρ t \rho_t ρ t with ρ t ∈ S \rho_t\in\mathcal{S} ρ t ∈ S . Suppose the measurement statistics are known for only a finite set of probe amplitudes Γ \Gamma Γ . Then, in general, the two-mode twirled state ρ t \rho_t ρ t is not determined by the collection of measurement statistics for the γ ∈ Γ \gamma\in\Gamma γ ∈ Γ .
Proposition 5.11. Consider a measurement configuration with a single counter. Suppose that, for a fixed probe magnitude R = ∣ γ ∣ R=|\gamma| R = ∣ γ ∣ , the measurement statistics are known for all phase values θ \theta θ of γ = R e i θ \gamma=Re^{i\theta} γ = R e i θ . Then, the two-mode twirled state ρ t \rho_t ρ t is determined by this collection of measurement statistics if and only if the number of photons is bounded by N ≤ 1 N\le1 N ≤ 1 .
Proof. Recall that, for each γ ∈ Γ \gamma\in\Gamma γ ∈ Γ , the coefficients of the generating function F 1 , γ ( u ) F_{1,\gamma}(u) F 1 , γ ( u ) correspond to the operators in C 1 ( γ ) \mathcal{C}_1(\gamma) C 1 ( γ ) . The set of expectations of these operators are strictly equivalent to the set of all normally ordered moments ⟨ : ( n ′ ) k : ⟩ \langle:(n')^k:\rangle ⟨ : ( n ′ ) k : ⟩ , which in turn contain the exact same information as the set of all expectations ⟨ ( n ′ ) k ⟩ \langle(n')^k\rangle ⟨( n ′ ) k ⟩ of photon-number moments, as shown in Appendix B. Furthermore, because F 1 , γ ( u ) F_{1,\gamma}(u) F 1 , γ ( u ) and F 1 , γ ′ ( u ) F'_{1,\gamma}(u) F 1 , γ ′ ( u ) are equivalent generating functions (Lem. 5.6 ), the information accessible from the measurement statistics is completely characterized by the linear span of the coefficients of the F 1 , γ ′ ( u ) F'_{1,\gamma}(u) F 1 , γ ′ ( u ) collected over the probe amplitudes γ ∈ Γ \gamma\in\Gamma γ ∈ Γ .
The coefficient of u j u^j u j in F 1 , γ ′ ( u ) F'_{1,\gamma}(u) F 1 , γ ′ ( u ) is ( ∣ η α ∣ 2 + η ζ ∗ γ ∗ α 1 + η ∗ ζ γ α 1 ∗ ) j / j ! (|\eta\alpha|^2+\eta\zeta^*\gamma^*\alpha_1+\eta^*\zeta\gamma\alpha_1^*)^j/j! ( ∣ η α ∣ 2 + η ζ ∗ γ ∗ α 1 + η ∗ ζ γ α 1 ∗ ) j / j ! , which expands as a linear combination of the monomials m ( k , l , n ) m(k,l,n) m ( k , l , n ) with k + l + n = j k+l+n=j k + l + n = j . Thus, the sets M = j = { m ( k , l , n ) ∣ k + l + n = j } \mathcal{M}_{=j}=\{m(k,l,n)\mid k+l+n=j\} M = j = { m ( k , l , n ) ∣ k + l + n = j } partition the monomials m ( k , l , n ) m(k,l,n) m ( k , l , n ) according to the degree at which they occur in the coefficients of F 1 , γ ′ ( u ) F'_{1,\gamma}(u) F 1 , γ ′ ( u ) . Each probe amplitude yields the expectation of one complex linear combination of monomials in M = j \mathcal{M}_{=j} M = j for each j j j . For K = ∣ Γ ∣ K=|\Gamma| K = ∣Γ∣ probe amplitudes and a given N ′ N' N ′ , this yields at most N ′ K N'K N ′ K such linear combinations of members of M ≤ N ′ = ⋃ 1 ≤ j ≤ N ′ M = j \mathcal{M}_{\le N'}=\bigcup_{1\le j\le N'}\mathcal{M}_{=j} M ≤ N ′ = ⋃ 1 ≤ j ≤ N ′ M = j . We now restrict the twirled state to S N S_N S N . Because the expectations for m ( k , l , n ) m(k,l,n) m ( k , l , n ) not in M N \mathcal{M}_N M N are zero for states in S N S_N S N , they do not contribute to the expectations of linear combinations of monomials. Since M N ⊆ M ≤ 2 N \mathcal{M}_N\subseteq\mathcal{M}_{\le2N} M N ⊆ M ≤ 2 N , the monomials outside M ≤ 2 N \mathcal{M}_{\le2N} M ≤ 2 N do not contribute and it suffices to consider N ′ = 2 N N'=2N N ′ = 2 N . The information available from the measurement statistics is therefore contained in expectations of linear combinations of monomials in M ≤ 2 N \mathcal{M}_{\le2N} M ≤ 2 N , which gives at most 2 N K 2NK 2 N K independent complex or 4 N K 4NK 4 N K independent real values. Since K K K is constant, this grows much more slowly with N N N than the κ ( N ) = Ω ( N 3 ) \kappa(N)=\Omega(N^3) κ ( N ) = Ω ( N 3 ) independent such values required to identify the states of S N S_N S N according to Eq. (30) .
Our final result in this subsection addresses the case where measurement statistics are known only for probe states with a fixed magnitude ∣ γ ∣ |\gamma| ∣ γ ∣ . This scenario is experimentally relevant, since it is typically much easier to vary the phase of the LO and collect data than to change its magnitude.
Proof. The proof follows from similar considerations as the proof of Prop. 5.10, allowing us to omit some of the intermediate logical steps. By hypothesis, the expectations of the coefficients of F 1 , γ ′ ( u ) F'_{1,\gamma}(u) F 1 , γ ′ ( u ) are available for all θ ∈ [ 0 , 2 π ) \theta\in[0,2\pi) θ ∈ [ 0 , 2 π ) . We first demonstrate that the individual monomials in M = n \mathcal{M}_{=n} M = n for n ≥ 2 n\ge2 n ≥ 2 cannot be fully extracted from the coeff u n F 1 , γ ′ ( u ) \operatorname{coeff}_{u^n}F'_{1,\gamma}(u) coeff u n F 1 , γ ′ ( u ) . Plugging γ = R e i θ \gamma=Re^{i\theta} γ = R e i θ into the n n n -th coefficient, we obtain
( n ! ) coeff u n F 1 , γ ′ = ( ∣ η α ⃗ ∣ 2 + η ζ ∗ α 1 R e − i θ + η ∗ ζ α 1 ∗ R e i θ ) n = ∑ r + s + t = n ( n r , s , t ) ( η ∗ ζ α 1 ) r R r e − i r θ ( η ζ ∗ α 1 ∗ ) s R s e i s θ ∣ η α ⃗ ∣ 2 t = ∑ r + s + t = n ( n r , s , t ) ∣ η ∣ 2 t ( η ∗ ζ α 1 ) r ( η ζ ∗ α 1 ∗ ) s ∣ α ⃗ ∣ 2 t R s + r e i θ ( s − r ) . (38) \begin{aligned}
(n!)\operatorname{coeff}_{u^n}F'_{1,\gamma}=\left(|\eta\vec{\alpha}|^2+\eta\zeta^*\alpha_1Re^{-i\theta}+\eta^*\zeta\alpha_1^*Re^{i\theta}\right)^n \\
&=\sum_{r+s+t=n}\binom{n}{r,s,t}(\eta^*\zeta\alpha_1)^rR^re^{-ir\theta}(\eta\zeta^*\alpha_1^*)^sR^se^{is\theta}|\eta\vec{\alpha}|^{2t} \\
&=\sum_{r+s+t=n}\binom{n}{r,s,t}|\eta|^{2t}(\eta^*\zeta\alpha_1)^r(\eta\zeta^*\alpha_1^*)^s|\vec{\alpha}|^{2t}R^{s+r}e^{i\theta(s-r)}. \tag*{(38)}
\end{aligned} ( n !) coeff u n F 1 , γ ′ = ( ∣ η α ∣ 2 + η ζ ∗ α 1 R e − i θ + η ∗ ζ α 1 ∗ R e i θ ) n = r + s + t = n ∑ ( r , s , t n ) ( η ∗ ζ α 1 ) r R r e − i r θ ( η ζ ∗ α 1 ∗ ) s R s e i s θ ∣ η α ∣ 2 t = r + s + t = n ∑ ( r , s , t n ) ∣ η ∣ 2 t ( η ∗ ζ α 1 ) r ( η ζ ∗ α 1 ∗ ) s ∣ α ∣ 2 t R s + r e i θ ( s − r ) . (38)
Let us define the scaled monomial
m R ( s , r , t ) = ( n r , s , t ) ∣ η ∣ 2 t R s + r ( η ζ ∗ ) r ( η ∗ ζ ) s ( α 1 ∗ ) s α 1 r ∣ α ⃗ ∣ 2 t , m_R(s,r,t)=\binom{n}{r,s,t}|\eta|^{2t}R^{s+r}(\eta\zeta^*)^r(\eta^*\zeta)^s(\alpha_1^*)^s\alpha_1^r|\vec{\alpha}|^{2t}, m R ( s , r , t ) = ( r , s , t n ) ∣ η ∣ 2 t R s + r ( η ζ ∗ ) r ( η ∗ ζ ) s ( α 1 ∗ ) s α 1 r ∣ α ∣ 2 t ,
which is proportional to m ( s , r , t ) m(s,r,t) m ( s , r , t ) up to a fixed constant factor. The expansion then simplifies to the phase-dependent sum ∑ r + s + t = n e i θ ( s − r ) m R ( s , r , t ) \sum_{r+s+t=n}e^{i\theta(s-r)}m_R(s,r,t) ∑ r + s + t = n e i θ ( s − r ) m R ( s , r , t ) , which are the linear combinations of the m R ( s , r , t ) m_R(s,r,t) m R ( s , r , t ) whose expectations can be determined from the measurement statistics. These linear combinations satisfy that for n n n fixed, the m R ( s + s ′ , s , n − 2 s − s ′ ) m_R(s+s',s,n-2s-s') m R ( s + s ′ , s , n − 2 s − s ′ ) always occur with the same coefficient of e i θ s ′ e^{i\theta s'} e i θ s ′ . For N ≥ 2 N\geq2 N ≥ 2 , consider the set M 2 \mathcal{M}_2 M 2 with respect to states in S 2 \mathcal{S}_2 S 2 . The span of M 2 \mathcal{M}_2 M 2 contains m R ( 0 , 0 , 2 ) m_R(0,0,2) m R ( 0 , 0 , 2 ) and m R ( 1 , 1 , 0 ) m_R(1,1,0) m R ( 1 , 1 , 0 ) . The expectations that can be determined from the measurement statistics are therefore already determined by the subspace of linear combinations of members of M 2 \mathcal{M}_2 M 2 with identical coefficients on m R ( 0 , 0 , 2 ) m_R(0,0,2) m R ( 0 , 0 , 2 ) and m R ( 1 , 1 , 0 ) m_R(1,1,0) m R ( 1 , 1 , 0 ) . Since this is a codimension 1 subspace of the span of M 2 \mathcal{M}_2 M 2 and by Cor. 5.3 and Cor. 5.5, this is insufficient information to identify members of S 2 \mathcal{S}_2 S 2 .
If ρ t \rho_t ρ t has at most one photon, the expectations of monomials in M = j \mathcal{M}_{=j} M = j with j ≥ 2 j\geq2 j ≥ 2 vanish, with the exception of ⟨ α 1 ∗ α 1 ⟩ \langle\alpha_1^*\alpha_1\rangle ⟨ α 1 ∗ α 1 ⟩ . Consequently, excluding the trivial trace condition ⟨ F 1 , γ ′ ( 0 ) ⟩ = 1 \langle F'_{1,\gamma}(0)\rangle=1 ⟨ F 1 , γ ′ ( 0 )⟩ = 1 , only the expectations ⟨ coeff u F 1 , γ ′ ( u ) ⟩ \langle\operatorname{coeff}_{u}F'_{1,\gamma}(u)\rangle ⟨ coeff u F 1 , γ ′ ( u )⟩ and ⟨ coeff u 2 F 1 , γ ′ ( u ) ⟩ \langle\operatorname{coeff}_{u^2}F'_{1,\gamma}(u)\rangle ⟨ coeff u 2 F 1 , γ ′ ( u )⟩ can be nonzero. Because the higher-order monomials in M = 2 \mathcal{M}_{=2} M = 2 vanish, ⟨ coeff u 2 F 1 , γ ′ ( u ) ⟩ \langle\operatorname{coeff}_{u^2}F'_{1,\gamma}(u)\rangle ⟨ coeff u 2 F 1 , γ ′ ( u )⟩ is directly proportional to ⟨ α 1 ∗ α 1 ⟩ \langle\alpha_1^*\alpha_1\rangle ⟨ α 1 ∗ α 1 ⟩ . For the first-order coefficient, we have
coeff u F 1 , γ ′ = ∣ η α ⃗ ∣ 2 + α 1 R e − i θ + α 1 ∗ R e i θ . \operatorname{coeff}_{u}F'_{1,\gamma}=|\eta\vec{\alpha}|^2+\alpha_1Re^{-i\theta}+\alpha_1^*Re^{i\theta}. coeff u F 1 , γ ′ = ∣ η α ∣ 2 + α 1 R e − i θ + α 1 ∗ R e i θ .
Evaluating this expectation at three distinct phases—for example, θ = 0 , π / 2 , π \theta=0,\pi/2,\pi θ = 0 , π /2 , π —produces three linearly independent equations in the unknowns ⟨ ∣ η α ⃗ ∣ 2 ⟩ \langle|\eta\vec{\alpha}|^2\rangle ⟨ ∣ η α ∣ 2 ⟩ , R ⟨ α 1 ⟩ R\langle\alpha_1\rangle R ⟨ α 1 ⟩ , and R ⟨ α 1 ∗ ⟩ R\langle\alpha_1^*\rangle R ⟨ α 1 ∗ ⟩ . Inverting the system we obtain the monomials in M = 1 \mathcal{M}_{=1} M = 1 . Since ∣ α ⃗ ∣ 2 = α 1 ∗ α 1 + α 2 ∗ α 2 |\vec{\alpha}|^2=\alpha_1^*\alpha_1+\alpha_2^*\alpha_2 ∣ α ∣ 2 = α 1 ∗ α 1 + α 2 ∗ α 2 , knowing ⟨ ∣ η α ⃗ ∣ 2 ⟩ \langle|\eta\vec{\alpha}|^2\rangle ⟨ ∣ η α ∣ 2 ⟩ and ⟨ α 1 ∗ α 1 ⟩ \langle\alpha_1^*\alpha_1\rangle ⟨ α 1 ∗ α 1 ⟩ allows us to immediately deduce ⟨ α 2 ∗ α 2 ⟩ \langle\alpha_2^*\alpha_2\rangle ⟨ α 2 ∗ α 2 ⟩ . By Cor. 5.3 , these extracted expectations (⟨ α 1 ⟩ \langle\alpha_1\rangle ⟨ α 1 ⟩ , ⟨ α 1 ∗ ⟩ \langle\alpha_1^*\rangle ⟨ α 1 ∗ ⟩ , ⟨ α 1 ∗ α 1 ⟩ \langle\alpha_1^*\alpha_1\rangle ⟨ α 1 ∗ α 1 ⟩ , and ⟨ α 2 ∗ α 2 ⟩ \langle\alpha_2^*\alpha_2\rangle ⟨ α 2 ∗ α 2 ⟩ ) are completely sufficient to uniquely determine ρ t \rho_t ρ t .
Results With Both Counters
For a measurement configuration with both counters it is convenient to redefine the variables u u u and v v v of the generating function F γ ( u , v ) F_\gamma(u,v) F γ ( u , v ) in Eq. (31) by making the substitutions u → u / ∣ η ∣ 2 u\to u/|\eta|^2 u → u /∣ η ∣ 2 and v → v / ∣ ζ ′ ∣ 2 v\to v/|\zeta'|^2 v → v /∣ ζ ′ ∣ 2 . Note that this variable substitution merely rescales the coefficients of the original generating function by scalar factors, and thus yields an equivalent generating function that encodes the operators in C 2 ( γ ) \mathcal{C}_2(\gamma) C 2 ( γ ) . By an abuse of notation, we retain the symbol F γ ( u , v ) F_\gamma(u,v) F γ ( u , v ) for the rescaled generating function throughout the remainder of this section. Since B 1 B_1 B 1 is unitary, we have the identities ∣ η ∣ 2 = ∣ η ′ ∣ 2 |\eta|^2=|\eta'|^2 ∣ η ∣ 2 = ∣ η ′ ∣ 2 , ∣ ζ ∣ 2 = ∣ ζ ′ ∣ 2 |\zeta|^2=|\zeta'|^2 ∣ ζ ∣ 2 = ∣ ζ ′ ∣ 2 and η ′ / ζ ′ = − η ∗ / ζ ∗ \eta'/\zeta'=-\eta^*/\zeta^* η ′ / ζ ′ = − η ∗ / ζ ∗ , which we use to write the rescaled F γ ( u , v ) F_\gamma(u,v) F γ ( u , v ) as
F γ ( u , v ) = exp ( ∣ α 1 + ζ η γ ∣ 2 u + ∣ α 1 − η ∗ ζ ∗ γ ∣ 2 v + ∣ α 2 ∣ 2 u + ∣ α 2 ∣ 2 v ) . F_\gamma(u,v)=\exp\left(\left|\alpha_1+\frac{\zeta}{\eta}\gamma\right|^2u+\left|\alpha_1-\frac{\eta^*}{\zeta^*}\gamma\right|^2v+|\alpha_2|^2u+|\alpha_2|^2v\right). F γ ( u , v ) = exp ( α 1 + η ζ γ 2 u + α 1 − ζ ∗ η ∗ γ 2 v + ∣ α 2 ∣ 2 u + ∣ α 2 ∣ 2 v ) .
We further denote ξ = η / ζ \xi=\eta/\zeta ξ = η / ζ so that
F γ ( u , v ) = e ∣ α 1 + ξ − 1 γ ∣ 2 u + ∣ α 1 − ξ ∗ γ ∣ 2 v + ∣ α 2 ∣ 2 u + ∣ α 2 ∣ 2 v = exp ( ∣ α 1 ∣ 2 u + ( α 1 ξ − 1 ∗ γ ∗ + α 1 ∗ ξ − 1 γ + ∣ ξ − 1 γ ∣ 2 ) u + ∣ α 1 ∣ 2 v + ( − α 1 ξ γ ∗ − α 1 ∗ ξ ∗ γ + ∣ ξ γ ∣ 2 ) v + ∣ α 2 ∣ 2 u + ∣ α 2 ∣ 2 v ) = exp ( ( α 1 ξ − 1 ∗ γ ∗ + α 1 ∗ ξ − 1 γ + ∣ ξ − 1 γ ∣ 2 ) u + ( − α 1 ξ γ ∗ − α 1 ∗ ξ ∗ γ + ∣ ξ γ ∣ 2 ) v + ∣ α ⃗ ∣ 2 u + ∣ α ⃗ ∣ 2 v ) . (39) \begin{aligned}
F_\gamma(u,v)=e^{|\alpha_1+\xi^{-1}\gamma|^2u+|\alpha_1-\xi^*\gamma|^2v+|\alpha_2|^2u+|\alpha_2|^2v} \\
&=\exp\left(|\alpha_1|^2u+(\alpha_1\xi^{-1*}\gamma^*+\alpha_1^*\xi^{-1}\gamma+|\xi^{-1}\gamma|^2)u+|\alpha_1|^2v+(-\alpha_1\xi\gamma^*-\alpha_1^*\xi^*\gamma+|\xi\gamma|^2)v+|\alpha_2|^2u+|\alpha_2|^2v\right) \\
&=\exp\left((\alpha_1\xi^{-1*}\gamma^*+\alpha_1^*\xi^{-1}\gamma+|\xi^{-1}\gamma|^2)u+(-\alpha_1\xi\gamma^*-\alpha_1^*\xi^*\gamma+|\xi\gamma|^2)v+|\vec{\alpha}|^2u+|\vec{\alpha}|^2v\right). \tag*{(39)}
\end{aligned} F γ ( u , v ) = e ∣ α 1 + ξ − 1 γ ∣ 2 u + ∣ α 1 − ξ ∗ γ ∣ 2 v + ∣ α 2 ∣ 2 u + ∣ α 2 ∣ 2 v = exp ( ∣ α 1 ∣ 2 u + ( α 1 ξ − 1 ∗ γ ∗ + α 1 ∗ ξ − 1 γ + ∣ ξ − 1 γ ∣ 2 ) u + ∣ α 1 ∣ 2 v + ( − α 1 ξ γ ∗ − α 1 ∗ ξ ∗ γ + ∣ ξ γ ∣ 2 ) v + ∣ α 2 ∣ 2 u + ∣ α 2 ∣ 2 v ) = exp ( ( α 1 ξ − 1 ∗ γ ∗ + α 1 ∗ ξ − 1 γ + ∣ ξ − 1 γ ∣ 2 ) u + ( − α 1 ξ γ ∗ − α 1 ∗ ξ ∗ γ + ∣ ξ γ ∣ 2 ) v + ∣ α ∣ 2 u + ∣ α ∣ 2 v ) . (39)
and introduce the generating function
F γ ′ ( u , v ) = e − ∣ ξ − 1 γ ∣ 2 u − ∣ ξ γ ∣ 2 v F γ ( u , v ) = e ( α 1 ξ − 1 ∗ γ ∗ + α 1 ∗ ξ − 1 γ + ∣ α ⃗ ∣ 2 ) u + ( − α 1 ξ γ ∗ − α 1 ∗ ξ ∗ γ + ∣ α ⃗ ∣ 2 ) v , (40) \begin{aligned}
F'_\gamma(u,v)=e^{-|\xi^{-1}\gamma|^2u-|\xi\gamma|^2v}F_\gamma(u,v) \\
&=e^{(\alpha_1\xi^{-1*}\gamma^*+\alpha_1^*\xi^{-1}\gamma+|\vec{\alpha}|^2)u+(-\alpha_1\xi\gamma^*-\alpha_1^*\xi^*\gamma+|\vec{\alpha}|^2)v}, \tag*{(40)}
\end{aligned} F γ ′ ( u , v ) = e − ∣ ξ − 1 γ ∣ 2 u − ∣ ξ γ ∣ 2 v F γ ( u , v ) = e ( α 1 ξ − 1 ∗ γ ∗ + α 1 ∗ ξ − 1 γ + ∣ α ∣ 2 ) u + ( − α 1 ξ γ ∗ − α 1 ∗ ξ ∗ γ + ∣ α ∣ 2 ) v , (40)
which is equivalent to F γ ( u , v ) F_\gamma(u,v) F γ ( u , v ) as the following lemma shows.
Lemma 5.12. The generating functions F γ ( u , v ) F_\gamma(u,v) F γ ( u , v ) and F γ ′ ( u , v ) F'_\gamma(u,v) F γ ′ ( u , v ) are equivalent.
Proof. The argument is identical to the one used for the proof of Lem. 5.6. In particular, the coefficient of u n v m u^nv^m u n v m in F γ ′ ( u , v ) F'_\gamma(u,v) F γ ′ ( u , v ) is a discrete convolution of the coefficients of F γ ( u , v ) F_\gamma(u,v) F γ ( u , v ) and of e − ∣ ξ − 1 γ ∣ 2 u − ∣ ξ γ ∣ 2 v e^{-|\xi^{-1}\gamma|^2u-|\xi\gamma|^2v} e − ∣ ξ − 1 γ ∣ 2 u − ∣ ξ γ ∣ 2 v that go up to degree u n v m u^nv^m u n v m , and due to the multiplicative invertibility of the exponential factor, the same is true in the reverse direction.
Lem. 5.12 allows us to conduct our analysis using F γ ′ ( u , v ) F'_\gamma(u,v) F γ ′ ( u , v ) .
If the measurement statistics from one counter determine ρ t \rho_t ρ t for a given set of probe states, then the statistics from both counters for the same set of probe states do as well, since the former are a marginal distribution of the latter. Consequently, Thms. 5.9 and 5.7 , as well as Cor. 5.8 are true for the configuration with two counters. The relevant question is whether there are interesting cases where the measurement statistics for a restricted set of probe states fails to determine ρ t \rho_t ρ t with one counter but succeeds with two counters. We first show that the no-go result of Prop. 5.10, namely, that a finite number of probe amplitudes does not determine ρ t \rho_t ρ t , also holds for the two-counter configuration.
Proposition 5.13. Consider a measurement configuration with a two counters. Consider an arbitrary two-mode twirled state ρ t \rho_t ρ t with ρ t ∈ S \rho_t \in S ρ t ∈ S . Suppose the measurement statistics are known for only a finite set of probe amplitudes Γ \Gamma Γ . Then, in general, the two-mode twirled state ρ t \rho_t ρ t is not determined by the collection of measurement statistics for the γ ∈ Γ \gamma\in\Gamma γ ∈ Γ .
Proof. The proof closely follows the proof of Prop. 5.10, replacing the generating function F γ ′ ( u ) F'_\gamma(u) F γ ′ ( u ) there with F γ ′ ( u , v ) F'_\gamma(u,v) F γ ′ ( u , v ) here. We refer to that proof for details omitted here.
The coefficient of u i v j u^iv^j u i v j in F γ ′ ( u , v ) F'_\gamma(u,v) F γ ′ ( u , v ) is 1 i ! j ! \frac{1}{i!j!} i ! j ! 1 times
( ∣ α ⃗ ∣ 2 + α 1 ξ − 1 γ ∗ + α 1 ∗ ξ γ ) i ( ∣ α ⃗ ∣ 2 − α 1 ξ γ ∗ − α 1 ∗ ξ − 1 γ ) j , \left(|\vec{\alpha}|^2+\alpha_1\xi^{-1}\gamma^*+\alpha_1^*\xi\gamma\right)^i\left(|\vec{\alpha}|^2-\alpha_1\xi\gamma^*-\alpha_1^*\xi^{-1}\gamma\right)^j, ( ∣ α ∣ 2 + α 1 ξ − 1 γ ∗ + α 1 ∗ ξ γ ) i ( ∣ α ∣ 2 − α 1 ξ γ ∗ − α 1 ∗ ξ − 1 γ ) j ,
which expands as a linear combination of monomials m ( k , l , n ) m(k,l,n) m ( k , l , n ) with k + l + n = i + j k+l+n=i+j k + l + n = i + j . Thus, for each N N N and for a given j ′ j' j ′ , the elements of M j ′ \mathcal{M}_{j'} M j ′ appear only among the coefficients of u i v j u^iv^j u i v j with i + j = j ′ i+j=j' i + j = j ′ , and there are exactly j ′ + 1 j'+1 j ′ + 1 such coefficients. Consequently, for each probe amplitude γ \gamma γ , there are at most j ′ + 1 j'+1 j ′ + 1 complex linear combinations of members of M j ′ \mathcal{M}_{j'} M j ′ whose expectations can be determined from the measurement statistics at this probe amplitude. For K = ∣ Γ ∣ K=|\Gamma| K = ∣Γ∣ probe amplitudes and a given N ′ N' N ′ , this yields at most ( N ′ + 1 ) ( N ′ + 2 ) K / 2 (N'+1)(N'+2)K/2 ( N ′ + 1 ) ( N ′ + 2 ) K /2 such linear combinations of members of M ≤ N ′ = ⋃ 1 ≤ j ′ ≤ N ′ M j ′ \mathcal{M}_{\leq N'}=\bigcup_{1\leq j'\leq N'}\mathcal{M}_{j'} M ≤ N ′ = ⋃ 1 ≤ j ′ ≤ N ′ M j ′ . Following the proof of Prop. 5.10 we restrict the twirled states to S N S_N S N , and because M N ⊆ M ≤ 2 N \mathcal{M}_N\subseteq\mathcal{M}_{\leq2N} M N ⊆ M ≤ 2 N , we find that the number of independent real values obtained for these states is now at most ( 2 N + 1 ) ( 2 N + 2 ) K (2N+1)(2N+2)K ( 2 N + 1 ) ( 2 N + 2 ) K , which is O ( N 2 ) O(N^2) O ( N 2 ) and thus grows much more slowly than the Ω ( N 3 ) \Omega(N^3) Ω ( N 3 ) required to identify the states of S N S_N S N . □
It follows from Prop. 5.13 that for ρ t \rho_t ρ t with at most N N N photons, there is some minimal number of probe amplitude h 2 ( N ) ≤ ( N + 1 ) 2 h_2(N)\leq(N+1)^2 h 2 ( N ) ≤ ( N + 1 ) 2 that are necessary and sufficient to determine ρ t \rho_t ρ t with this measurement configuration. Since there is significantly more information with two counters than with one counter, we expect that h 2 ( N ) < h 1 ( N ) h_2(N)<h_1(N) h 2 ( N ) < h 1 ( N ) . We leave finding h 1 ( N ) h_1(N) h 1 ( N ) and h 2 ( N ) h_2(N) h 2 ( N ) as open problems. The final result in our analysis of the single-counter configuration concerned the case where the magnitude of the probe is restricted to a single value, while allowing its phase to vary freely, as is done in standard homodyne measurements. We now demonstrate that, unlike with the single-counter configuration, ρ t \rho_t ρ t can be determined under these constraints if both counters are used.
Theorem 5.14. Consider a measurement configuration with two counters. Consider an arbitrary two-mode twirled state ρ t \rho_t ρ t with ρ t ∈ S \rho_t\in S ρ t ∈ S . Suppose that, for a fixed probe magnitude R = ∣ γ ∣ R=|\gamma| R = ∣ γ ∣ , the measurement statistics are known for all phase values of γ = R e i θ \gamma=Re^{i\theta} γ = R e i θ . Then, ρ t \rho_t ρ t is determined by this collection of measurement statistics.
Proof. We treat u u u and v v v as real variables in F γ ′ ( u , v ) F'_\gamma(u,v) F γ ′ ( u , v ) , and show that with suitable variable transformations involving u , v u,v u , v , we can extract every monomial m ( k , l , n ) m(k,l,n) m ( k , l , n ) from the F γ ′ ( u , v ) F'_\gamma(u,v) F γ ′ ( u , v ) for all values of the phase θ \theta θ . Note that ξ ∗ \xi^* ξ ∗ and ξ − 1 \xi^{-1} ξ − 1 have the same complex argument, which we denote by ϕ \phi ϕ ; consequently, ξ \xi ξ and ξ − 1 ∗ \xi^{-1*} ξ − 1 ∗ share the argument − ϕ -\phi − ϕ . This allows us to rewrite the exponent of F γ ′ ( u , v ) F'_\gamma(u,v) F γ ′ ( u , v ) as
α 1 γ ∗ ( ξ − 1 ∗ u − ξ v ) + α 1 ∗ γ ( ξ − 1 u − ξ ∗ v ) + ∣ α ⃗ ∣ 2 ( u + v ) = α 1 R e − i θ ( ξ − 1 ∗ u − ξ v ) + α 1 ∗ R e i θ e i 2 ϕ ( ξ − 1 ∗ u − ξ v ) + ∣ α ⃗ ∣ 2 ( u + v ) = ( α 1 e − i θ + α 1 ∗ e i 2 ϕ e i θ ) R ( ξ − 1 ∗ u − ξ v ) + ∣ α ⃗ ∣ 2 ( u + v ) . (41) \begin{aligned}
&\alpha_1\gamma^*(\xi^{-1*}u-\xi v)+\alpha_1^*\gamma(\xi^{-1}u-\xi^*v)+|\vec{\alpha}|^2(u+v)\\
&=\alpha_1Re^{-i\theta}(\xi^{-1*}u-\xi v)+\alpha_1^*Re^{i\theta}e^{i2\phi}(\xi^{-1*}u-\xi v)+|\vec{\alpha}|^2(u+v)\\
&=(\alpha_1e^{-i\theta}+\alpha_1^*e^{i2\phi}e^{i\theta})R(\xi^{-1*}u-\xi v)+|\vec{\alpha}|^2(u+v). \tag*{(41)}
\end{aligned} α 1 γ ∗ ( ξ − 1 ∗ u − ξ v ) + α 1 ∗ γ ( ξ − 1 u − ξ ∗ v ) + ∣ α ∣ 2 ( u + v ) = α 1 R e − i θ ( ξ − 1 ∗ u − ξ v ) + α 1 ∗ R e i θ e i 2 ϕ ( ξ − 1 ∗ u − ξ v ) + ∣ α ∣ 2 ( u + v ) = ( α 1 e − i θ + α 1 ∗ e i 2 ϕ e i θ ) R ( ξ − 1 ∗ u − ξ v ) + ∣ α ∣ 2 ( u + v ) . (41)
We now change variables from ( u , v ) (u,v) ( u , v ) to the independent variables t = u + v t=u+v t = u + v and s = R ( ξ − 1 ∗ u − ξ v ) s=R(\xi^{-1*}u-\xi v) s = R ( ξ − 1 ∗ u − ξ v ) , in terms of which the exponent of F γ ′ F'_\gamma F γ ′ takes the form
∣ α ⃗ ∣ 2 t + ( α 1 e − i θ + α 1 ∗ e i 2 ϕ e i θ ) s . (42) |\vec{\alpha}|^2t+(\alpha_1e^{-i\theta}+\alpha_1^*e^{i2\phi}e^{i\theta})s. \tag*{(42)} ∣ α ∣ 2 t + ( α 1 e − i θ + α 1 ∗ e i 2 ϕ e i θ ) s . (42)
The coefficient of t n s r t^ns^r t n s r is given by
n ! r ! coeff t n s r F ′ = ∣ α ⃗ ∣ 2 n ( α 1 e − i θ + α 1 ∗ e i 2 ϕ e i θ ) r = ∑ j = 0 r ( r j ) ∣ α ⃗ ∣ 2 n ( α 1 ∗ ) j e i 2 ϕ j α 1 r − j e i θ ( 2 j − r ) = e − i θ r ∑ j = 0 r ( r j ) e i 2 ϕ j m ( j , r − j , n ) e i 2 θ j . (43) \begin{aligned}
n!r!\operatorname{coeff}_{t^ns^r}F'=\lvert\vec{\alpha}\rvert^{2n}(\alpha_1e^{-i\theta}+\alpha_1^*e^{i2\phi}e^{i\theta})^r \\
&=\sum_{j=0}^{r}\binom{r}{j}\lvert\vec{\alpha}\rvert^{2n}(\alpha_1^*)^je^{i2\phi j}\alpha_1^{r-j}e^{i\theta(2j-r)} \\
&=e^{-i\theta r}\sum_{j=0}^{r}\binom{r}{j}e^{i2\phi j}m(j,r-j,n)e^{i2\theta j}. \tag*{(43)}
\end{aligned} n ! r ! coeff t n s r F ′ = ∣ α ∣ 2 n ( α 1 e − i θ + α 1 ∗ e i 2 ϕ e i θ ) r = j = 0 ∑ r ( j r ) ∣ α ∣ 2 n ( α 1 ∗ ) j e i 2 ϕ j α 1 r − j e i θ ( 2 j − r ) = e − i θ r j = 0 ∑ r ( j r ) e i 2 ϕ j m ( j , r − j , n ) e i 2 θ j . (43)
Since this expression is available for all values of θ \theta θ , the sum can be inverted to obtain the m ( j , r − j , n ) m(j,r-j,n) m ( j , r − j , n ) using Fourier analysis. For every k = 0 , … , r k=0,\ldots,r k = 0 , … , r , the following integration isolates m ( k , r − k , n ) m(k,r-k,n) m ( k , r − k , n ) :
∫ 0 2 π d θ e − i 2 θ k [ e i θ r n ! r ! coeff t n s r F ′ ] = ∫ 0 2 π d θ e − i 2 θ k ∑ j = 0 r ( r j ) e i 2 ϕ j m ( j , r − j , n ) e i 2 θ j = ∑ j = 0 r ( r j ) e i 2 ϕ j m ( j , r − j , n ) ∫ 0 2 π d θ e − i 2 θ ( k − j ) = 4 π ( r k ) e i 2 ϕ k m ( k , r − k , n ) . (44) \begin{aligned}
\int_0^{2\pi}d\theta e^{-i2\theta k}\left[e^{i\theta r}n!r!\operatorname{coeff}_{t^ns^r}F'\right]
&=\int_0^{2\pi}d\theta e^{-i2\theta k}\sum_{j=0}^{r}\binom{r}{j}e^{i2\phi j}m(j,r-j,n)e^{i2\theta j}\\
&=\sum_{j=0}^{r}\binom{r}{j}e^{i2\phi j}m(j,r-j,n)\int_0^{2\pi}d\theta e^{-i2\theta(k-j)}\\
&=4\pi\binom{r}{k}e^{i2\phi k}m(k,r-k,n). \tag*{(44)}
\end{aligned} ∫ 0 2 π d θ e − i 2 θ k [ e i θ r n ! r ! coeff t n s r F ′ ] = ∫ 0 2 π d θ e − i 2 θ k j = 0 ∑ r ( j r ) e i 2 ϕ j m ( j , r − j , n ) e i 2 θ j = j = 0 ∑ r ( j r ) e i 2 ϕ j m ( j , r − j , n ) ∫ 0 2 π d θ e − i 2 θ ( k − j ) = 4 π ( k r ) e i 2 ϕ k m ( k , r − k , n ) . (44)
Performing this procedure for every pair ( r , n ) (r,n) ( r , n ) with $r+ n = N n = N n = N yields all monomials in M N \mathcal{M}_N M N , and iterating over all N N N recovers every monomial m ( k , l , n ) m(k,l,n) m ( k , l , n ) .
Our final result aims to clarify the connection between this measurement scheme and standard homodyne detection. Let ρ r \rho_r ρ r denote the reduced state in mode a ^ 1 \hat{a}_1 a ^ 1 . In experimental implementations of homodyne detection, the probe amplitude ∣ γ ∣ |\gamma| ∣ γ ∣ is effectively fixed at a large value, and the photodiode signal approximates a measurement of the total photon number. Consequently, the measurement statistics are equivalent to the set of expectations ⟨ ( n ^ a ′ − n ^ b ′ ) r ⟩ \langle(\hat{n}'_a-\hat{n}'_b)^r\rangle ⟨( n ^ a ′ − n ^ b ′ ) r ⟩ for all r ≥ 0 r\geq0 r ≥ 0 , sampled over a sufficiently dense set of probe phases covering the full angular range. Since our scheme does not require R R R to be large, it is interesting to explore whether some form of subtraction between the measurements from the two counters can directly measure ρ r \rho_r ρ r .
Proposition 5.15. Suppose the Husimi function of the two-mode twirled state ρ t \rho_t ρ t has well-defined moments. Suppose the Husimi function of its marginal on mode 1, ρ r \rho_r ρ r , has a determinate moment sequence. Suppose that for a fixed probe magnitude R = ∣ γ ∣ R=|\gamma| R = ∣ γ ∣ , the normally ordered moments ⟨ : ( ∣ η ∣ 2 n ^ b ′ − ∣ ζ ∣ 2 n ^ a ′ ) n : ⟩ \langle:(|\eta|^2\hat{n}'_b-|\zeta|^2\hat{n}'_a)^n:\rangle ⟨ : ( ∣ η ∣ 2 n ^ b ′ − ∣ ζ ∣ 2 n ^ a ′ ) n : ⟩ are known for all integers n ≥ 0 n\geq0 n ≥ 0 and for all phase values θ \theta θ of γ = R e i θ \gamma=Re^{i\theta} γ = R e i θ . Then this set of expectations alone suffices to determine ρ r \rho_r ρ r .
Proof. By Lem. 5.1, the expectations of the antinormally ordered operators a ^ 1 l ( a ^ 1 † ) k \hat{a}_1^l(\hat{a}_1^\dagger)^k a ^ 1 l ( a ^ 1 † ) k for all k ≥ l k\geq l k ≥ l can be expressed in terms of the expectations of the normally ordered operators ( a ^ 1 † ) k a ^ 1 l (\hat{a}_1^\dagger)^k\hat{a}_1^l ( a ^ 1 † ) k a ^ 1 l . Since the collection ⟨ a ^ 1 l ( a ^ 1 † ) k ⟩ \langle\hat{a}_1^l(\hat{a}_1^\dagger)^k\rangle ⟨ a ^ 1 l ( a ^ 1 † ) k ⟩ forms the complete moment sequence of the Husimi function of ρ r \rho_r ρ r , and therefore uniquely determines ρ r \rho_r ρ r , it suffices to extract the expectations ⟨ ( a ^ 1 † ) k a ^ 1 l ⟩ \langle(\hat{a}_1^\dagger)^k\hat{a}_1^l\rangle ⟨( a ^ 1 † ) k a ^ 1 l ⟩ for all k ≥ l k\geq l k ≥ l from the set of expectations that are assumed to be known in the proposition statement.
For each value of γ \gamma γ , we introduce the operator-valued generating function
H s , γ ′ ( v ) = ∑ n = 0 ∞ 1 n ! 1 ∣ η ∣ 2 n : ( ∣ η ∣ 2 n ^ b ′ − ∣ ζ ∣ 2 n ^ a ′ ) n : v n , (45) H'_{s,\gamma}(v)=\sum_{n=0}^{\infty}\frac{1}{n!}\frac{1}{|\eta|^{2n}}:(|\eta|^2\hat{n}'_b-|\zeta|^2\hat{n}'_a)^n:v^n, \tag*{(45)} H s , γ ′ ( v ) = n = 0 ∑ ∞ n ! 1 ∣ η ∣ 2 n 1 : ( ∣ η ∣ 2 n ^ b ′ − ∣ ζ ∣ 2 n ^ a ′ ) n : v n , (45)
whose coefficients span the measurement operators whose expectations are assumed known for that γ \gamma γ . H s , γ ′ ( v ) H'_{s,\gamma}(v) H s , γ ′ ( v ) is obtained from the two-counter generating function H ′ ( u , v ) H'(u,v) H ′ ( u , v ) in Eq. 6 by setting u = − ∣ ζ / η ∣ 2 v = − ∣ ξ − 1 ∣ 2 v u=-|\zeta/\eta|^2v=-|\xi^{-1}|^2v u = − ∣ ζ / η ∣ 2 v = − ∣ ξ − 1 ∣ 2 v :
H ′ ( − ∣ ξ − 1 ∣ 2 v , v ) = ∑ k = 0 ∞ ∑ l = 0 ∞ 1 k ! l ! : ( n ^ a ′ ) k ( n ^ b ′ ) l : ( − ∣ ξ − 1 ∣ 2 v ) k v l = ∑ n = 0 ∞ ∑ k = 0 n 1 k ! ( n − k ) ! : ( n ^ a ′ ) k ( n ^ b ′ ) n − k : ( − ∣ ξ − 1 ∣ 2 v ) k v n − k = ∑ n = 0 ∞ 1 n ! : ( n ^ b ′ − ∣ ξ − 1 ∣ 2 n ^ a ′ ) n : v n = ∑ n = 0 ∞ 1 n ! 1 ∣ η ∣ 2 n : ( ∣ η ∣ 2 n ^ b ′ − ∣ ζ ∣ 2 n ^ a ′ ) n : v n = H s , γ ′ ( v ) , (46) \begin{aligned}
H'(-|\xi^{-1}|^2v,v)=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\frac{1}{k!l!}:(\hat{n}'_a)^k(\hat{n}'_b)^l:(-|\xi^{-1}|^2v)^kv^l \\
&=\sum_{n=0}^{\infty}\sum_{k=0}^{n}\frac{1}{k!(n-k)!}:(\hat{n}'_a)^k(\hat{n}'_b)^{n-k}:(-|\xi^{-1}|^2v)^kv^{n-k} \\
&=\sum_{n=0}^{\infty}\frac{1}{n!}:(\hat{n}'_b-|\xi^{-1}|^2\hat{n}'_a)^n:v^n \\
&=\sum_{n=0}^{\infty}\frac{1}{n!}\frac{1}{|\eta|^{2n}}:(|\eta|^2\hat{n}'_b-|\zeta|^2\hat{n}'_a)^n:v^n=H'_{s,\gamma}(v), \tag*{(46)}
\end{aligned} H ′ ( − ∣ ξ − 1 ∣ 2 v , v ) = k = 0 ∑ ∞ l = 0 ∑ ∞ k ! l ! 1 : ( n ^ a ′ ) k ( n ^ b ′ ) l : ( − ∣ ξ − 1 ∣ 2 v ) k v l = n = 0 ∑ ∞ k = 0 ∑ n k ! ( n − k )! 1 : ( n ^ a ′ ) k ( n ^ b ′ ) n − k : ( − ∣ ξ − 1 ∣ 2 v ) k v n − k = n = 0 ∑ ∞ n ! 1 : ( n ^ b ′ − ∣ ξ − 1 ∣ 2 n ^ a ′ ) n : v n = n = 0 ∑ ∞ n ! 1 ∣ η ∣ 2 n 1 : ( ∣ η ∣ 2 n ^ b ′ − ∣ ζ ∣ 2 n ^ a ′ ) n : v n = H s , γ ′ ( v ) , (46)
where we have relabeled the indices in the second line, and then applied the binomial identity.
The generating function for the measurement operators on H a \mathcal{H}_a H a is therefore obtained from F γ ( u , v ) F_\gamma(u,v) F γ ( u , v ) by setting u = − ∣ ξ − 1 ∣ 2 v u=-|\xi^{-1}|^2v u = − ∣ ξ − 1 ∣ 2 v . Working instead with the equivalent generating function F γ ′ ( u , v ) F'_\gamma(u,v) F γ ′ ( u , v ) in (40) (where ∣ η ∣ 2 u → u |\eta|^2u\to u ∣ η ∣ 2 u → u and ∣ ζ ∣ 2 v → v |\zeta|^2v\to v ∣ ζ ∣ 2 v → v ), this substitution becomes simply u = − v u=-v u = − v . Reading off the exponent of F γ ′ ( − v , v ) F'_\gamma(-v,v) F γ ′ ( − v , v ) from (41) with u = − v u=-v u = − v gives
( α 1 e − i θ + α 1 ∗ e i 2 ϕ e i θ ) R ( − ξ − 1 ∗ v − ξ v ) + ∣ α ∣ 2 ( − v + v ) = ( α 1 e − i θ + α 1 ∗ e i 2 ϕ e i θ ) R ( ξ − 1 ∗ + ξ ) ( − v ) . (47) \begin{aligned}
(\alpha_1e^{-i\theta}+\alpha_1^*e^{i2\phi}e^{i\theta})R(-\xi^{-1*}v-\xi v)+|\alpha|^2(-v+v) \\
&=(\alpha_1e^{-i\theta}+\alpha_1^*e^{i2\phi}e^{i\theta})R(\xi^{-1*}+\xi)(-v). \tag*{(47)}
\end{aligned} ( α 1 e − i θ + α 1 ∗ e i 2 ϕ e i θ ) R ( − ξ − 1 ∗ v − ξ v ) + ∣ α ∣ 2 ( − v + v ) = ( α 1 e − i θ + α 1 ∗ e i 2 ϕ e i θ ) R ( ξ − 1 ∗ + ξ ) ( − v ) . (47)
Crucially, the ∣ α ∣ 2 |\alpha|^2 ∣ α ∣ 2 term vanishes, so the exponent carries no dependence on α 2 \alpha_2 α 2 or α 2 ∗ \alpha_2^* α 2 ∗ , as intended. The coefficient of v n v^n v n in F γ ′ ( − v , v ) F'_\gamma(-v,v) F γ ′ ( − v , v ) is therefore
[ − R ( ξ − 1 ∗ + ξ ) ] n n ! ( α 1 e − i θ + α 1 ∗ e i ( θ + 2 ϕ ) ) n . (48) \frac{[-R(\xi^{-1*}+\xi)]^n}{n!}(\alpha_1e^{-i\theta}+\alpha_1^*e^{i(\theta+2\phi)})^n. \tag*{(48)} n ! [ − R ( ξ − 1 ∗ + ξ ) ] n ( α 1 e − i θ + α 1 ∗ e i ( θ + 2 ϕ ) ) n . (48)
As in the proof of Theorem 5.14, the availability of this expression for all θ \theta θ allows one to isolate each monomial ( α 1 ∗ ) k α 1 n − k (\alpha_1^*)^k\alpha_1^{n-k} ( α 1 ∗ ) k α 1 n − k for k = 0 , … , n k=0,\ldots,n k = 0 , … , n by Fourier inversion in θ \theta θ . Repeating for every n n n recovers the full collection of monomials ( α 1 ∗ ) k α 1 l (\alpha_1^*)^k\alpha_1^l ( α 1 ∗ ) k α 1 l , whose expectations ⟨ ( a ^ 1 † ) k a ^ 1 l ⟩ \langle(\hat{a}_1^\dagger)^k\hat{a}_1^l\rangle ⟨( a ^ 1 † ) k a ^ 1 l ⟩ determine ρ r \rho_r ρ r .
If the BS is balanced, as in standard homodyne detection, the expectations ⟨ : ( n ^ b ′ − n ^ a ′ ) n : ⟩ \langle:(\hat{n}'_b-\hat{n}'_a)^n:\rangle ⟨ : ( n ^ b ′ − n ^ a ′ ) n : ⟩ in Prop. 5.15 are the normally ordered versions of the usual homodyne moments ⟨ ( n ^ b ′ − n ^ a ′ ) n ⟩ \langle(\hat{n}'_b-\hat{n}'_a)^n\rangle ⟨( n ^ b ′ − n ^ a ′ ) n ⟩ . Although this establishes a neat connection with homodyne detection, and avoids assuming a high-intensity probe, these normally ordered moments cannot be accessed directly in experiment without measuring the full joint total-photon-number distribution of both counters. In particular, unlike in standard homodyne detection, where subtracting the photodiode signals directly measures the moments ⟨ ( n ^ b ′ − n ^ a ′ ) n ⟩ \langle(\hat{n}'_b-\hat{n}'_a)^n\rangle ⟨( n ^ b ′ − n ^ a ′ ) n ⟩ to a good approximation, there is no straightforward method to measure the normally ordered moments ⟨ : ( n ^ b ′ − n ^ a ′ ) n : ⟩ \langle:(\hat{n}'_b-\hat{n}'_a)^n:\rangle ⟨ : ( n ^ b ′ − n ^ a ′ ) n : ⟩ directly for arbitrary probe amplitudes. However, if the probe has high intensity, the normally ordered moments ⟨ : ( n ^ b ′ − n ^ a ′ ) n : ⟩ \langle:(\hat{n}'_b-\hat{n}'_a)^n:\rangle ⟨ : ( n ^ b ′ − n ^ a ′ ) n : ⟩ admit a significant simplification. Since normally ordered products of n ^ a ′ \hat{n}'_a n ^ a ′ and n ^ b ′ \hat{n}'_b n ^ b ′ differ from their standard counterparts by terms involving lower-order operators, and since each such correction is suppressed by a power of the probe intensity relative to the leading term, the normally ordered and standard moments become asymptotically equal in the high-intensity limit. This is a well-established observation in standard homodyne theory [37 ] . Thus, it could be said that for large enough probe magnitudes, the ⟨ : ( n ^ b ′ − n ^ a ′ ) n : ⟩ \langle:(\hat{n}'_b-\hat{n}'_a)^n:\rangle ⟨ : ( n ^ b ′ − n ^ a ′ ) n : ⟩ can be apapproximately measured by subtracting the outcomes of the counters in the two-counter configuration.
Discussion
Standard homodyne detection, viewed within our framework, appears as a limiting case of the two-counter measurement scheme in the presence of a strong LO. Our work provides a general theoretical comparison of the information content accessible through an idealized weak-field homodyne scheme and that obtained from standard homodyne detection for a fixed LO mode. Existing literature performing this comparison assumes the unknown input state resides only in the LO-matched mode with the other modes in their vacuum states. We relax this assumption entirely, providing an analysis where the orthogonal modes are not necessarily in vacuum. Our analysis demonstrates that employing photon counters together with a weak LO enables the measurement to extract strictly richer information about the unknown input state, while requiring no additional device complexity beyond a BS and the counters. Notably, this advantage arises even in the single-counter configuration, provided that statistics are collected for a sufficiently large set of LO amplitudes with varying intensities.
From a practical perspective, realizing the informational advantage of photon counters requires the characterization of detectors that can resolve photon number sufficiently well for inferrability of the photon number probabilities. With the recent progress in superconducting detectors [15 , 16 , 41 , 42 ] and multiplexed photodetectors [16 , 43 ] , the difficulty of this characterization challenge has been significantly reduced. By applying detector tomography techniques [31 , 44 , 45 , 46 , 47 , 48 ] , it is now feasible to infer photon number probabilities for photon numbers below increasingly high detector saturation limits. Thus, while photon counters are more challenging to implement and characterize, the advantages they offer in information gain, together with technological advances, make them increasingly useful for state characterization with weak probes in continuous-variable experiments.
It is worth highlighting that the sufficiency of the single-counter configuration offers a practical simplification. In standard homodyne detection, the two photodiodes may not perfectly capture the full LO mode, due to imperfect coupling, finite detector area, or spectral mismatch, which can introduce ambiguity in the measured quadrature. Similarly, with a weak-field scheme with PNR detectors, ensuring that both detectors measure the same combination of modes can be technically demanding and complicate the interpretation of the resulting statistics. By contrast, a single-counter configuration reduces this source of ambiguity: although there is still uncertainty about which exact modes are measured, compared to the two-counter configuration the consequences of this uncertainty are straightforward to model and interpret. Thus, the single-counter scheme alleviates a common experimental challenge, provided that statistics are collected over a sufficient set of LO amplitudes.
Most of the results presented here were first developed in a doctoral thesis [49 ] , with several extensions and refinements introduced here. The thesis included analysis of measurement configurations in which the BS acts differently on distinct mode sets, as occurs in polarizing, fiber-optic beam splitters with unequal transmission–reflection ratios for orthogonal polarizations. Accompanying this analysis, a demonstration experiment with such a BS was conducted, while the theoretical results showed that two-counter measurements can reveal significantly more information about the unknown input state than is contained in the two-mode twirled state. The thesis also reported simulations reconstructing the two-mode twirled state via maximum-likelihood estimation for input states containing up to eight photons, with probe states of order one photon in amplitude to ensure resolvability by modern PNR detectors. Fidelities exceeding 99.6% between reconstructed and true two-mode twirled states were obtained with 10 6 10^{6} 1 0 6 samples distributed over fewer than ten LO amplitudes, and remained high even at 50% detector efficiency, indicating that statistical noise and known losses are unlikely to pose significant limitations. A first attempt to experimentally implement these weak-field homodyne schemes is presented in the thesis of D. S. Phillips [50 ] .
The simulations did not consider the effect of mode-dependent loss, which violates the assumptions of the model since the total photon number can no longer be inferred from the detector signal. Future work could investigate the impact of such unaccounted-for losses, which might arise in experimental settings. Additional directions include exploring how the information obtained from this measurement configuration could be enhanced by employing different sets of LO modes. For example, one might consider the situation in which, for the same unknown state, multiple two-mode twirled states have been constructed for multiple, potentially overlapping LO modes, and ask what conclusions can then be drawn about the underlying unknown state.
Acknowledgments
This work includes contributions of the National Institute of Standards and Technology, which are not subject to U.S. copyright. The U.S. Government is authorized to reproduce and distribute reprints for governmental purposes notwithstanding any copyright annotation thereon. The use of trade names and software does not imply endorsement by the US government, nor does it imply these are necessarily the best available for the purpose used here. This document has not been peer reviewed but has been cleared by NIST for release.
Use of AI: Most results were obtained without the use of AI and can be found in the pre-AI thesis [49 ] . When writing this paper, AI was used for researching the relevant literature, finding relevant works to reference and take advantage of in updated proofs and added results, and to improve the language.
References
Appendix
In this appendix, we derive results concerning the use of normal and anti-normal orderings introduced in the main text. We consider an arbitrary collection of S S S orthogonal bosonic modes, denoted by a ^ 1 , … , a ^ S \hat{a}_1,\ldots,\hat{a}_S a ^ 1 , … , a ^ S , with the total number operator defined as n ^ = ∑ i = 1 S a ^ i † a ^ i \hat{n}=\sum_{i=1}^{S}\hat{a}_i^\dagger\hat{a}_i n ^ = ∑ i = 1 S a ^ i † a ^ i .
Normal ordering and PLTs
We begin by establishing that the normal ordering operation commutes with PLTs. To clarify, let P ^ \hat{P} P ^ denote an arbitrary operator that is a linear combination of products of mode operators. We want to show that for any PLT U \mathcal{U} U ,
U † : P ^ : U = : U † P ^ U : (49) \mathcal{U}^{\dagger}:\hat{P}:\mathcal{U}=:\mathcal{U}^{\dagger}\hat{P}\mathcal{U}: \tag*{(49)} U † : P ^ : U =: U † P ^ U : (49)
By the linearity of both PLTs and the normal ordering operation, it suffices to verify this property for a single product of mode operators. Let M ( a ^ † , a ^ ) M(\hat{a}^{\dagger},\hat{a}) M ( a ^ † , a ^ ) denote such a product. Then
U † M ( a ^ † , a ^ ) U = M ( U ∗ ⋅ a ^ † , U ⋅ a ^ ) , (50) \mathcal{U}^{\dagger}M(\hat{a}^{\dagger},\hat{a})\mathcal{U}=M(\mathcal{U}^{*}\cdot\hat{a}^{\dagger},\mathcal{U}\cdot\hat{a}), \tag*{(50)} U † M ( a ^ † , a ^ ) U = M ( U ∗ ⋅ a ^ † , U ⋅ a ^ ) , (50)
which follows by inserting the identity U U † \mathcal{U}\mathcal{U}^{\dagger} U U † between each consecutive pair of mode operators in the product and using U † a ^ i = ( U ∗ ⋅ a ^ ) i \mathcal{U}^{\dagger}\hat{a}_{i}=(\mathcal{U}^{*}\cdot\hat{a})_{i} U † a ^ i = ( U ∗ ⋅ a ^ ) i and U † a ^ i † = ( U ∗ ⋅ a ^ † ) i \mathcal{U}^{\dagger}\hat{a}_{i}^{\dagger}=(\mathcal{U}^{*}\cdot\hat{a}^{\dagger})_{i} U † a ^ i † = ( U ∗ ⋅ a ^ † ) i , where the index i i i denotes the i i i th element of the corresponding vector.
Denote the transformed mode operators ( U ⋅ a ^ ) i (\mathcal{U}\cdot\hat{a})_{i} ( U ⋅ a ^ ) i by b ^ i \hat{b}_{i} b ^ i , so that ( U ∗ ⋅ a ^ † ) i = b ^ i † (\mathcal{U}^{*}\cdot\hat{a}^{\dagger})_{i}=\hat{b}_{i}^{\dagger} ( U ∗ ⋅ a ^ † ) i = b ^ i † . We further define b ^ = ( b ^ 1 , … , b ^ S ) \hat{b}=(\hat{b}_{1},\ldots,\hat{b}_{S}) b ^ = ( b ^ 1 , … , b ^ S ) and b ^ † = ( b ^ 1 † , … , b ^ S † ) \hat{b}^{\dagger}=(\hat{b}_{1}^{\dagger},\ldots,\hat{b}_{S}^{\dagger}) b ^ † = ( b ^ 1 † , … , b ^ S † ) . For a product of mode operators, we have
: M ( a ^ † , a ^ ) : = ∏ i = 1 S ( a ^ i † ) k i ∏ j = 1 S a ^ j l j , (51) :M(\hat{a}^{\dagger},\hat{a}):=\prod_{i=1}^{S}(\hat{a}_{i}^{\dagger})^{k_i}\prod_{j=1}^{S}\hat{a}_{j}^{l_j}, \tag*{(51)} : M ( a ^ † , a ^ ) := i = 1 ∏ S ( a ^ i † ) k i j = 1 ∏ S a ^ j l j , (51)
where k i k_i k i and l j l_j l j denote the number of occurrences of a ^ i † \hat{a}_i^{\dagger} a ^ i † and a ^ j \hat{a}_j a ^ j in M ( a ^ † , a ^ ) M(\hat{a}^{\dagger},\hat{a}) M ( a ^ † , a ^ ) , respectively. Hence,
U † : M ( a ^ † , a ^ ) : U = ∏ i = 1 S ( U ∗ ⋅ a ^ † ) i k i ∏ j = 1 S ( U ⋅ a ^ ) j l j = ∏ i = 1 S ( b ^ i † ) k i ∏ j = 1 S b ^ j l j . (52) \begin{aligned}
U^{\dagger}:M(\hat{a}^{\dagger},\hat{a}):U=\prod_{i=1}^{S}(U^{\ast}\cdot\hat{a}^{\dagger})_{i}^{k_i}\prod_{j=1}^{S}(U\cdot\hat{a})_{j}^{l_j} \\
&=\prod_{i=1}^{S}(\hat{b}_{i}^{\dagger})^{k_i}\prod_{j=1}^{S}\hat{b}_{j}^{l_j}. \tag*{(52)}
\end{aligned} U † : M ( a ^ † , a ^ ) : U = i = 1 ∏ S ( U ∗ ⋅ a ^ † ) i k i j = 1 ∏ S ( U ⋅ a ^ ) j l j = i = 1 ∏ S ( b ^ i † ) k i j = 1 ∏ S b ^ j l j . (52)
where the second line follows using the same technique as in Eq. (50) . On the other hand,
: U † M ( a ^ † , a ^ ) U : = : M ( U ∗ ⋅ a ^ † , U ⋅ a ^ ) : = : M ( b ^ † , b ^ ) : = ∏ i = 1 S ( b ^ i † ) k i ∏ j = 1 S b ^ j l j . (53) \begin{aligned}
:U^{\dagger}M(\hat{a}^{\dagger},\hat{a})U:=:M(U^{\ast}\cdot\hat{a}^{\dagger},U\cdot\hat{a}): \\
&=:M(\hat{b}^{\dagger},\hat{b}):=\prod_{i=1}^{S}(\hat{b}_{i}^{\dagger})^{k_i}\prod_{j=1}^{S}\hat{b}_{j}^{l_j}. \tag*{(53)}
\end{aligned} : U † M ( a ^ † , a ^ ) U :=: M ( U ∗ ⋅ a ^ † , U ⋅ a ^ ) : =: M ( b ^ † , b ^ ) := i = 1 ∏ S ( b ^ i † ) k i j = 1 ∏ S b ^ j l j . (53)
Normally ordered powers of n ^ \hat{n} n ^
Here, we show that the normally ordered powers of n ^ \hat{n} n ^ can be expressed as linear combinations of ordinary powers of n ^ \hat{n} n ^ . We define the falling factorial of the operator n ^ \hat{n} n ^ by
( n ^ ) k = n ^ ( n ^ − 1 ) ⋯ ( n ^ − k + 1 ) , (54) (\hat{n})_k=\hat{n}(\hat{n}-1)\cdots(\hat{n}-k+1), \tag*{(54)} ( n ^ ) k = n ^ ( n ^ − 1 ) ⋯ ( n ^ − k + 1 ) , (54)
where each integer is understood as a scalar multiple of the identity operator on the full multimode Hilbert space.
Lemma B.1. For every nonnegative integer k k k , the normally ordered power of n ^ \hat{n} n ^ is given by
: n ^ k : = ( n ^ ) k . (55) :\hat{n}^{k}:=(\hat{n})_k. \tag*{(55)} : n ^ k := ( n ^ ) k . (55)
Proof. The operator : n ^ k : :\hat{n}^{k}: : n ^ k : is normally ordered, while the operator ( n ^ ) k (\hat{n})_k ( n ^ ) k can be written in normally ordered form by repeated application of the commutation relations a ^ i a ^ i † = 1 + a ^ i † a ^ i \hat{a}_i\hat{a}_i^{\dagger}=1+\hat{a}_i^{\dagger}\hat{a}_i a ^ i a ^ i † = 1 + a ^ i † a ^ i . Hence, the operator–expression correspondence implies that the claim follows if we show that for all α ⃗ \vec{\alpha} α ,
⟨ α ⃗ ∣ : n ^ k : ∣ α ⃗ ⟩ = ∣ α ⃗ ∣ 2 k , (56) \langle\vec{\alpha}|:\hat{n}^{k}:|\vec{\alpha}\rangle=|\vec{\alpha}|^{2k}, \tag*{(56)} ⟨ α ∣ : n ^ k : ∣ α ⟩ = ∣ α ∣ 2 k , (56)
⟨ α ⃗ ∣ ( n ^ ) k ∣ α ⃗ ⟩ = ∣ α ⃗ ∣ 2 k . (57) \langle\vec{\alpha}|(\hat{n})_k|\vec{\alpha}\rangle=|\vec{\alpha}|^{2k}. \tag*{(57)} ⟨ α ∣ ( n ^ ) k ∣ α ⟩ = ∣ α ∣ 2 k . (57)
Let e ⃗ 1 = ( 1 , 0 , … , 0 ) \vec{e}_1=(1,0,\ldots,0) e 1 = ( 1 , 0 , … , 0 ) be the vector with 1 in the first entry and 0’s elsewhere. Given α ⃗ \vec{\alpha} α , define α = ∣ α ⃗ ∣ \alpha=|\vec{\alpha}| α = ∣ α ∣ , and let U U U be a PLT that transforms U ∣ α ⃗ ⟩ = ∣ α e ⃗ 1 ⟩ U|\vec{\alpha}\rangle=|\alpha\vec{e}_1\rangle U ∣ α ⟩ = ∣ α e 1 ⟩ . We use the fact that the number operator commutes with PLTs, together with the fact that normal ordering also commutes with PLTs, in the following:
⟨ α ⃗ ∣ : n ^ k : ∣ α ⃗ ⟩ = ⟨ α e ⃗ 1 ∣ U † : n ^ k : U ∣ α e ⃗ 1 ⟩ = ⟨ α e ⃗ 1 ∣ : U † n ^ k U : ∣ α e ⃗ 1 ⟩ = ⟨ α e ⃗ 1 ∣ : n ^ k : ∣ α e ⃗ 1 ⟩ = ⟨ α e ⃗ 1 ∣ ( a ^ 1 † ) k a ^ 1 k ∣ α e ⃗ 1 ⟩ = α 2 k = ∣ α ⃗ ∣ 2 k . (58) \begin{aligned}
\langle\vec{\alpha}|:\hat{n}^{k}:|\vec{\alpha}\rangle=\langle\alpha\vec{e}_1|U^{\dagger}:\hat{n}^{k}:U|\alpha\vec{e}_1\rangle\\
&=\langle\alpha\vec{e}_1|:U^{\dagger}\hat{n}^{k}U:|\alpha\vec{e}_1\rangle\\
&=\langle\alpha\vec{e}_1|:\hat{n}^{k}:|\alpha\vec{e}_1\rangle\\
&=\langle\alpha\vec{e}_1|(\hat{a}_1^{\dagger})^{k}\hat{a}_1^{k}|\alpha\vec{e}_1\rangle\\
&=\alpha^{2k}=|\vec{\alpha}|^{2k}. \tag*{(58)}
\end{aligned} ⟨ α ∣ : n ^ k : ∣ α ⟩ = ⟨ α e 1 ∣ U † : n ^ k : U ∣ α e 1 ⟩ = ⟨ α e 1 ∣ : U † n ^ k U : ∣ α e 1 ⟩ = ⟨ α e 1 ∣ : n ^ k : ∣ α e 1 ⟩ = ⟨ α e 1 ∣ ( a ^ 1 † ) k a ^ 1 k ∣ α e 1 ⟩ = α 2 k = ∣ α ∣ 2 k . (58)
When replacing : n ^ k : :\hat{n}^{k}: : n ^ k : with ( a ^ 1 † ) k a ^ 1 k (\hat{a}_1^{\dagger})^{k}\hat{a}_1^{k} ( a ^ 1 † ) k a ^ 1 k in the fourth step, we implicitly expanded the expression : n ^ k : :\hat{n}^{k}: : n ^ k : in terms of the commuting mode operators of the different modes and used the fact that all modes except the first are in the vacuum state. Consequently, any term involving at least one mode operator from the other modes vanishes, since it contains an annihilation operator acting on vacuum. Similarly,
⟨ α ⃗ ∣ ( n ^ ) k ∣ α ⃗ ⟩ = ⟨ α e ⃗ 1 ∣ U † ( n ^ ) k U ∣ α e ⃗ 1 ⟩ = ⟨ α e ⃗ 1 ∣ ( n ^ ) k ∣ α e ⃗ 1 ⟩ = ⟨ α e ⃗ 1 ∣ ( a ^ 1 † a ^ 1 ) k ∣ α e ⃗ 1 ⟩ , (59) \begin{aligned}
\langle\vec{\alpha}|(\hat{n})_k|\vec{\alpha}\rangle=\langle\alpha\vec{e}_1|U^{\dagger}(\hat{n})_kU|\alpha\vec{e}_1\rangle\\
&=\langle\alpha\vec{e}_1|(\hat{n})_k|\alpha\vec{e}_1\rangle\\
&=\langle\alpha\vec{e}_1|(\hat{a}_1^{\dagger}\hat{a}_1)_k|\alpha\vec{e}_1\rangle, \tag*{(59)}
\end{aligned} ⟨ α ∣ ( n ^ ) k ∣ α ⟩ = ⟨ α e 1 ∣ U † ( n ^ ) k U ∣ α e 1 ⟩ = ⟨ α e 1 ∣ ( n ^ ) k ∣ α e 1 ⟩ = ⟨ α e 1 ∣ ( a ^ 1 † a ^ 1 ) k ∣ α e 1 ⟩ , (59)
where, as before, in the last line we have implicitly expanded ( n ^ ) k (\hat{n})_k ( n ^ ) k and retained only the terms involving mode 1. Direct computation shows that ( a ^ 1 † a ^ 1 ) k = a ^ 1 † k a ^ 1 k (\hat{a}_1^{\dagger}\hat{a}_1)_k=\hat{a}_1^{\dagger k}\hat{a}_1^k ( a ^ 1 † a ^ 1 ) k = a ^ 1 † k a ^ 1 k :
( a ^ 1 † ) k a ^ 1 k = ∑ m = k ∞ ( m ) ⋯ ( m − k + 1 ) ∣ m ⟩ ⟨ m ∣ = [ ∑ m 1 = 1 ∞ m 1 ∣ m 1 ⟩ ⟨ m 1 ∣ ] [ ∑ m 2 = 1 ∞ ( m 2 − 1 ) ∣ m 2 ⟩ ⟨ m 2 ∣ ] ⋯ \begin{aligned}
(\hat{a}_1^{\dagger})^k\hat{a}_1^k=\sum_{m=k}^{\infty}(m)\cdots(m-k+1)|m\rangle\langle m| \\
&=\left[\sum_{m_1=1}^{\infty}m_1|m_1\rangle\langle m_1|\right]\left[\sum_{m_2=1}^{\infty}(m_2-1)|m_2\rangle\langle m_2|\right]\cdots
\end{aligned} ( a ^ 1 † ) k a ^ 1 k = m = k ∑ ∞ ( m ) ⋯ ( m − k + 1 ) ∣ m ⟩ ⟨ m ∣ = [ m 1 = 1 ∑ ∞ m 1 ∣ m 1 ⟩ ⟨ m 1 ∣ ] [ m 2 = 1 ∑ ∞ ( m 2 − 1 ) ∣ m 2 ⟩ ⟨ m 2 ∣ ] ⋯
⋯ [ ∑ m k = 1 ∞ ( m k − k + 1 ) ∣ m k ⟩ ⟨ m k ∣ ] = ( a ^ 1 † a ^ 1 ) ( a ^ 1 † a ^ 1 − 1 ) ⋯ ( a ^ 1 † a ^ 1 − k + 1 ) = ( a ^ 1 † a ^ 1 ) k , (60) \begin{aligned}
\cdots\left[\sum_{m_k=1}^{\infty}(m_k-k+1)|m_k\rangle\langle m_k|\right] \\
&=(\hat{a}_1^{\dagger}\hat{a}_1)(\hat{a}_1^{\dagger}\hat{a}_1-1)\cdots(\hat{a}_1^{\dagger}\hat{a}_1-k+1)=(\hat{a}_1^{\dagger}\hat{a}_1)_k, \tag*{(60)}
\end{aligned} ⋯ [ m k = 1 ∑ ∞ ( m k − k + 1 ) ∣ m k ⟩ ⟨ m k ∣ ] = ( a ^ 1 † a ^ 1 ) ( a ^ 1 † a ^ 1 − 1 ) ⋯ ( a ^ 1 † a ^ 1 − k + 1 ) = ( a ^ 1 † a ^ 1 ) k , (60)
where the kets and bras correspond to Fock states in mode 1. It then follows that Eq. (59) evaluates to α 2 k \alpha^{2k} α 2 k .
As a direct consequence of Lem. B.1, the ordinary powers of n ^ \hat{n} n ^ can be systematically recovered from the normally ordered powers of n ^ \hat{n} n ^ .
Corollary B.2. For every nonnegative integer m m m , the ordinary power n ^ m \hat{n}^{m} n ^ m can be expressed as a linear combination of the normally ordered powers : n ^ k : :\hat{n}^{k}: : n ^ k : for k ≤ m k\leq m k ≤ m .
Proof. By the definition in Eq. (54) , the falling factorial ( n ^ ) m (\hat{n})_m ( n ^ ) m is a polynomial in n ^ \hat{n} n ^ of degree m m m with a leading term of n ^ m \hat{n}^{m} n ^ m . Rearranging this expression yields n ^ m = ( n ^ ) m + P ^ m − 1 ( n ^ ) \hat{n}^{m}=(\hat{n})_m+\hat{P}_{m-1}(\hat{n}) n ^ m = ( n ^ ) m + P ^ m − 1 ( n ^ ) , where P ^ m − 1 ( n ^ ) \hat{P}_{m-1}(\hat{n}) P ^ m − 1 ( n ^ ) is a polynomial in n ^ \hat{n} n ^ of degree at most m − 1 m-1 m − 1 . By recursively applying this same logic to the lower-degree terms in P ^ m − 1 ( n ^ ) \hat{P}_{m-1}(\hat{n}) P ^ m − 1 ( n ^ ) , the operator n ^ m \hat{n}^{m} n ^ m can be expanded entirely as a linear combination of falling factorials ( n ^ ) k (\hat{n})_k ( n ^ ) k for k ≤ m k\leq m k ≤ m . Substituting ( n ^ ) k = : n ^ k : (\hat{n})_k=:\hat{n}^{k}: ( n ^ ) k =: n ^ k : from Lem. B.1 establishes the operator identity.
Determinate moment sequences
Here we first show that the class of states with Husimi functions that have a determinate moment sequence includes Gaussian states, states with bounded photon number, and their finite mixtures, as well as finite superpositions of pure states of these forms. We then show that for this class of states, the measurement statistics for every probe state γ \gamma γ are well-behaved, in the sense that they are uniquely determined by their photon-number moments.
Proposition C.1. Let ρ \rho ρ be an S S S -mode quantum state that is Gaussian or has bounded photon number. Then its Husimi function Q ( α ⃗ , α ⃗ ∗ ) Q(\vec{\alpha},\vec{\alpha}^{*}) Q ( α , α ∗ ) is bounded by C e − ϵ ∣ α ⃗ ∣ 2 C e^{-\epsilon|\vec{\alpha}|^{2}} C e − ϵ ∣ α ∣ 2 for some state-dependent positive constants C C C and ϵ \epsilon ϵ .
Proof. We characterize the asymptotic tail behavior of Q ( α ⃗ , α ⃗ ∗ ) Q(\vec{\alpha},\vec{\alpha}^{*}) Q ( α , α ∗ ) for the specified classes of states. For a pure state ∣ ψ ⟩ |\psi\rangle ∣ ψ ⟩ with a bounded total photon number N N N , the overlap ⟨ ψ ∣ α ⃗ ⟩ \langle\psi|\vec{\alpha}\rangle ⟨ ψ ∣ α ⟩ is a polynomial in the variables α j \alpha_j α j of maximum degree N N N , weighted by the coherent state normalization factor e − ∣ α ⃗ ∣ 2 / 2 e^{-|\vec{\alpha}|^{2}/2} e − ∣ α ∣ 2 /2 . Consequently, its Husimi function takes the form Q ( α ⃗ , α ⃗ ∗ ) = ∣ P ( α ⃗ ) ∣ 2 e − ∣ α ⃗ ∣ 2 Q(\vec{\alpha},\vec{\alpha}^{*})=|P(\vec{\alpha})|^{2}e^{-|\vec{\alpha}|^{2}} Q ( α , α ∗ ) = ∣ P ( α ) ∣ 2 e − ∣ α ∣ 2 , where P P P is a polynomial. For any S S S -mode Gaussian state, the Husimi function is a Gaussian [2 ] . Consequently states of bounded photon number and Gaussian states both satisfy that their Husimi functions are bounded by a function of the form C e − ϵ ∣ α ⃗ ∣ 2 C e^{-\epsilon|\vec{\alpha}|^{2}} C e − ϵ ∣ α ∣ 2 , where C C C and ϵ \epsilon ϵ are positive and depend on the state.
Let S g S_g S g be the family of positive semidefinite bounded operators whose Husimi function are bounded by C e − ϵ ∣ α ⃗ ∣ 2 C e^{-\epsilon|\vec{\alpha}|^{2}} C e − ϵ ∣ α ∣ 2 for some positive constants C C C and ϵ \epsilon ϵ . We refer to positive semidefinite bounded operators as positive operators. Such operators are non-negative multiples of density operator states.
Proposition C.2. The family S g S_g S g is closed under finite mixture, pure state superpositions, PLTs, squeezing, displacements and transformations of the form ρ ↦ p ρ p † \rho\mapsto p\rho p^\dagger ρ ↦ pρ p † with p p p a polynomial of the mode creation operators.
Proof. We tackle each claim in the order listed in the proposition statement. If ρ \rho ρ is a mixture of positive operators ρ i \rho_i ρ i where the Husimi function of ρ i \rho_i ρ i is bounded by C i e − ϵ i ∣ α ⃗ ∣ 2 C_i e^{-\epsilon_i|\vec{\alpha}|^{2}} C i e − ϵ i ∣ α ∣ 2 , then ρ \rho ρ is bounded by max i ( C i ) e − min i ( ϵ i ) ∣ α ⃗ ∣ 2 \max_i(C_i)e^{-\min_i(\epsilon_i)|\vec{\alpha}|^{2}} max i ( C i ) e − m i n i ( ϵ i ) ∣ α ∣ 2 .
Suppose that ∣ ψ i ⟩ ⟨ ψ i ∣ |\psi_i\rangle\langle\psi_i| ∣ ψ i ⟩ ⟨ ψ i ∣ for i = 1 , … , k i=1,\ldots,k i = 1 , … , k has Husimi function bounded by C i e − ϵ i ∣ α ⃗ ∣ 2 C_i e^{-\epsilon_i|\vec{\alpha}|^{2}} C i e − ϵ i ∣ α ∣ 2 . Then
1 π S ∣ ⟨ α ⃗ ∣ ψ i ⟩ ⟨ ψ j ∣ α ⃗ ⟩ ∣ = 1 π S ⟨ α ⃗ ∣ ψ i ⟩ ⟨ ψ i ∣ α ⃗ ⟩ 1 π S ⟨ α ⃗ ∣ ψ j ⟩ ⟨ ψ j ∣ α ⃗ ⟩ = Q ψ i ( α ⃗ , α ⃗ ∗ ) Q ψ j ( α ⃗ , α ⃗ ∗ ) ≤ max i ( C i ) e − min i ( ϵ i ) ∣ α ⃗ ∣ 2 (61) \frac{1}{\pi^S}\left|\langle\vec{\alpha}|\psi_i\rangle\langle\psi_j|\vec{\alpha}\rangle\right|=\sqrt{\frac{1}{\pi^S}\langle\vec{\alpha}|\psi_i\rangle\langle\psi_i|\vec{\alpha}\rangle}\sqrt{\frac{1}{\pi^S}\langle\vec{\alpha}|\psi_j\rangle\langle\psi_j|\vec{\alpha}\rangle}
=\sqrt{Q_{\psi_i}(\vec{\alpha},\vec{\alpha}^{*})}\sqrt{Q_{\psi_j}(\vec{\alpha},\vec{\alpha}^{*})}
\leq\max_i(C_i)e^{-\min_i(\epsilon_i)|\vec{\alpha}|^{2}} \tag*{(61)} π S 1 ∣ ⟨ α ∣ ψ i ⟩ ⟨ ψ j ∣ α ⟩ ∣ = π S 1 ⟨ α ∣ ψ i ⟩ ⟨ ψ i ∣ α ⟩ π S 1 ⟨ α ∣ ψ j ⟩ ⟨ ψ j ∣ α ⟩ = Q ψ i ( α , α ∗ ) Q ψ j ( α , α ∗ ) ≤ i max ( C i ) e − m i n i ( ϵ i ) ∣ α ∣ 2 (61)
from which it follows that the Husimi function of ∣ ϕ ⟩ ⟨ ϕ ∣ |\phi\rangle\langle\phi| ∣ ϕ ⟩ ⟨ ϕ ∣ with ∣ ϕ ⟩ |\phi\rangle ∣ ϕ ⟩ the superposition ∣ ϕ ⟩ = ∑ i = 1 k w i ∣ ψ i ⟩ |\phi\rangle=\sum_{i=1}^{k}w_i|\psi_i\rangle ∣ ϕ ⟩ = ∑ i = 1 k w i ∣ ψ i ⟩ is bounded by k 2 max i ( C i ) e − min i ( ϵ i ) ∣ α ⃗ ∣ 2 k^{2}\max_i(C_i)e^{-\min_i(\epsilon_i)|\vec{\alpha}|^{2}} k 2 max i ( C i ) e − m i n i ( ϵ i ) ∣ α ∣ 2 .
Let Q ( α ⃗ , α ⃗ ∗ ) Q(\vec{\alpha},\vec{\alpha}^{*}) Q ( α , α ∗ ) be the Husimi function of ρ ∈ S g \rho\in S_g ρ ∈ S g bounded by C e − ϵ ∣ α ⃗ ∣ 2 C e^{-\epsilon|\vec{\alpha}|^{2}} C e − ϵ ∣ α ∣ 2 for some positive C C C and ϵ \epsilon ϵ . PLTs and squeezing transform the Husimi function according to an invertible linear map L L L to Q ′ ( α ⃗ , α ⃗ ∗ ) = Q ( L α ⃗ , ( L α ⃗ ) ∗ ) ≤ C e − ϵ ∣ L α ⃗ ∣ 2 Q'(\vec{\alpha},\vec{\alpha}^{*})=Q(L\vec{\alpha},(L\vec{\alpha})^{*})\leq C e^{-\epsilon|L\vec{\alpha}|^{2}} Q ′ ( α , α ∗ ) = Q ( L α , ( L α ) ∗ ) ≤ C e − ϵ ∣ L α ∣ 2 [37 ] . Let λ \lambda λ be the minimum singular value of L L L , which is positive [37 ] . Then ∣ L α ⃗ ∣ 2 ≥ λ 2 ∣ α ⃗ ∣ 2 |L\vec{\alpha}|^{2}\geq\lambda^{2}|\vec{\alpha}|^{2} ∣ L α ∣ 2 ≥ λ 2 ∣ α ∣ 2 so Q ′ ( α ⃗ , α ⃗ ∗ ) ≤ C e − ϵ λ 2 ∣ α ⃗ ∣ 2 Q'(\vec{\alpha},\vec{\alpha}^{*})\leq C e^{-\epsilon\lambda^{2}|\vec{\alpha}|^{2}} Q ′ ( α , α ∗ ) ≤ C e − ϵ λ 2 ∣ α ∣ 2 . Since ϵ λ 2 \epsilon\lambda^{2} ϵ λ 2 is positive, the transformed operator is also in S g S_g S g . A displacement by β ⃗ \vec{\beta} β transforms the Husimi function of ρ \rho ρ to
Q ′ ( α ⃗ , α ⃗ ∗ ) = Q ( α ⃗ + β ⃗ , ( α ⃗ + β ⃗ ) ∗ ) ≤ C e − ϵ ∣ α ⃗ + β ⃗ ∣ 2 ≤ C e 2 ϵ ∣ β ⃗ ∣ 2 e − ( ϵ / 2 ) ∣ α ⃗ ∣ 2 . (62) Q'(\vec{\alpha},\vec{\alpha}^{*})=Q(\vec{\alpha}+\vec{\beta},(\vec{\alpha}+\vec{\beta})^{*})
\leq C e^{-\epsilon|\vec{\alpha}+\vec{\beta}|^{2}}
\leq C e^{2\epsilon|\vec{\beta}|^{2}}e^{-(\epsilon/2)|\vec{\alpha}|^{2}}. \tag*{(62)} Q ′ ( α , α ∗ ) = Q ( α + β , ( α + β ) ∗ ) ≤ C e − ϵ ∣ α + β ∣ 2 ≤ C e 2 ϵ ∣ β ∣ 2 e − ( ϵ /2 ) ∣ α ∣ 2 . (62)
The last inequality follows from ∣ α ⃗ + β ⃗ ∣ 2 ≥ ∣ α ⃗ ∣ 2 / 2 − 2 ∣ β ⃗ ∣ 2 |\vec{\alpha}+\vec{\beta}|^{2}\geq|\vec{\alpha}|^{2}/2-2|\vec{\beta}|^{2} ∣ α + β ∣ 2 ≥ ∣ α ∣ 2 /2 − 2∣ β ∣ 2 . To see this, first note that ∣ α ⃗ + β ⃗ ∣ 2 ≥ ∣ α ⃗ ∣ 2 − 2 ∣ α ⃗ ∣ ∣ β ⃗ ∣ |\vec{\alpha}+\vec{\beta}|^{2}\geq|\vec{\alpha}|^{2}-2|\vec{\alpha}||\vec{\beta}| ∣ α + β ∣ 2 ≥ ∣ α ∣ 2 − 2∣ α ∣∣ β ∣ . In general, for x , y ≥ 0 x,y\geq0 x , y ≥ 0 we have − 2 x y ≥ − x 2 / 2 − 2 y 2 -2xy\geq-x^{2}/2-2y^{2} − 2 x y ≥ − x 2 /2 − 2 y 2 . We apply this inequality to the term − 2 ∣ α ⃗ ∣ ∣ β ⃗ ∣ -2|\vec{\alpha}||\vec{\beta}| − 2∣ α ∣∣ β ∣ with x = ∣ α ⃗ ∣ x=|\vec{\alpha}| x = ∣ α ∣ and y = ∣ β ⃗ ∣ y=|\vec{\beta}| y = ∣ β ∣ for ∣ α ⃗ + β ⃗ ∣ 2 ≥ ∣ α ⃗ ∣ 2 − ∣ α ⃗ ∣ 2 / 2 − 2 ∣ β ⃗ ∣ 2 = ∣ α ⃗ ∣ 2 / 2 − 2 ∣ β ⃗ ∣ 2 |\vec{\alpha}+\vec{\beta}|^{2}\geq|\vec{\alpha}|^{2}-|\vec{\alpha}|^{2}/2-2|\vec{\beta}|^{2}=|\vec{\alpha}|^{2}/2-2|\vec{\beta}|^{2} ∣ α + β ∣ 2 ≥ ∣ α ∣ 2 − ∣ α ∣ 2 /2 − 2∣ β ∣ 2 = ∣ α ∣ 2 /2 − 2∣ β ∣ 2 .
For the last claim, it suffices to consider a single creation operator a i † a_i^\dagger a i † . If ρ \rho ρ is transformed to a i † ρ a i a_i^\dagger\rho a_i a i † ρ a i , the Husimi function transforms to ∣ α i ∣ 2 Q ( α ⃗ , α ⃗ ∗ ) |\alpha_i|^{2}Q(\vec{\alpha},\vec{\alpha}^{*}) ∣ α i ∣ 2 Q ( α , α ∗ ) . We have that for positive δ \delta δ , there exists C ( δ ) > 0 C(\delta)>0 C ( δ ) > 0 such that ∣ α i ∣ 2 ≤ C ( δ ) e δ ∣ α ⃗ ∣ 2 |\alpha_i|^{2}\leq C(\delta)e^{\delta|\vec{\alpha}|^{2}} ∣ α i ∣ 2 ≤ C ( δ ) e δ ∣ α ∣ 2 . We may choose δ = ϵ / 2 \delta=\epsilon/2 δ = ϵ /2 and C ′ = C C ( δ ) C'=C C(\delta) C ′ = C C ( δ ) to ensure Q ′ ( α ⃗ , α ⃗ ∗ ) ≤ C ′ e − ( ϵ / 2 ) ∣ α ⃗ ∣ 2 Q'(\vec{\alpha},\vec{\alpha}^{*})\leq C'e^{-(\epsilon/2)|\vec{\alpha}|^{2}} Q ′ ( α , α ∗ ) ≤ C ′ e − ( ϵ /2 ) ∣ α ∣ 2 .
Proposition C.3. We have S g ⊆ S S_g\subseteq S S g ⊆ S .
Proof. Let Q ( α ⃗ , α ⃗ ∗ ) Q(\vec{\alpha},\vec{\alpha}^{*}) Q ( α , α ∗ ) be the Husimi function of a state in S g S_g S g . By definition, there exist positive C C C and ϵ \epsilon ϵ such that Q ( α ⃗ , α ⃗ ∗ ) ≤ C e − ϵ ∣ α ⃗ ∣ 2 Q(\vec{\alpha},\vec{\alpha}^{*})\leq C e^{-\epsilon|\vec{\alpha}|^{2}} Q ( α , α ∗ ) ≤ C e − ϵ ∣ α ∣ 2 . We argue that the moment generating function converges in a neighbourhood of 0, which is a sufficient criterion for the sequence to be determinate [38 Chap. 14,15] . The moment generating function is given by
M ( β ⃗ , β ⃗ ∗ ) = ∫ C S e β ⃗ ⋅ α ⃗ + β ⃗ ∗ ⋅ α ⃗ ∗ Q ( α ⃗ , α ⃗ ∗ ) d α ⃗ d α ⃗ ∗ , (63) M(\vec{\beta},\vec{\beta}^{*})=\int_{\mathbb{C}^{S}}e^{\vec{\beta}\cdot\vec{\alpha}+\vec{\beta}^{*}\cdot\vec{\alpha}^{*}}Q(\vec{\alpha},\vec{\alpha}^{*})\,d\vec{\alpha}\,d\vec{\alpha}^{*}, \tag*{(63)} M ( β , β ∗ ) = ∫ C S e β ⋅ α + β ∗ ⋅ α ∗ Q ( α , α ∗ ) d α d α ∗ , (63)
where β ⃗ ∈ C S \vec{\beta}\in\mathbb{C}^{S} β ∈ C S is an arbitrary complex vector. Applying the upper bound Q ( α ⃗ , α ⃗ ∗ ) ≤ C e − ϵ ∣ α ⃗ ∣ 2 Q(\vec{\alpha},\vec{\alpha}^{*})\leq C e^{-\epsilon|\vec{\alpha}|^{2}} Q ( α , α ∗ ) ≤ C e − ϵ ∣ α ∣ 2 , we see that the integrand is strictly bounded by a displaced multivariate Gaussian. Consequently, the integral converges absolutely for all β ⃗ ∈ C S \vec{\beta}\in\mathbb{C}^{S} β ∈ C S . This global convergence implies that M ( β ⃗ , β ⃗ ∗ ) M(\vec{\beta},\vec{\beta}^{*}) M ( β , β ∗ ) converges everywhere, satisfying the sufficient condition for moment determinacy and completing the proof.
The notation and definitions required for the following proposition are established in Sec. 3.1.
Proposition C.4. Consider a measurement configuration with two counters. Suppose that the input state ρ \rho ρ is in S g S_g S g . Then, for every probe state σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⃗ ⟩ ⟨ 0 ⃗ ∣ \sigma= |\gamma\rangle\langle\gamma| \otimes|\vec{0}\rangle\langle\vec{0}| σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⟩ ⟨ 0 ∣ the joint photon-counting probabilities p i j p_{ij} p ij are uniquely determined by the moments of the total-photon-number operators, ⟨ ( n ^ a ) k ( n ^ b ) l ⟩ \langle(\hat{n}^{a})^{k}(\hat{n}^{b})^{l}\rangle ⟨( n ^ a ) k ( n ^ b ) l ⟩ .
Proof. Let α ⃗ \vec{\alpha} α and β ⃗ \vec{\beta} β be the variables for the Husimi functions of the input and probe modes. By definition, there are positive C C C and ϵ \epsilon ϵ such that the Husimi function of ρ \rho ρ is bounded by C e − ϵ ∣ α ⃗ ∣ 2 Ce^{-\epsilon|\vec{\alpha}|^{2}} C e − ϵ ∣ α ∣ 2 for some positive constants C C C and ϵ \epsilon ϵ . PLTs induce orthogonal coordinate rotations in phase space, which implies that the Gaussian bound applies uniformly over all PLT transformations of ρ \rho ρ . Since the twirled state ρ t \rho_t ρ t is obtained by a Haar integral over such PLT transformations of ρ \rho ρ , ρ t \rho_t ρ t also has a Husimi function bounded by the same Gaussian. Because the probe state is Gaussian, it satisfies a bound of the form C ′ e − ϵ ′ ∣ β ⃗ ∣ 2 C'e^{-\epsilon'|\vec{\beta}|^{2}} C ′ e − ϵ ′ ∣ β ∣ 2 . After twirling, the state on input and probe modes is ρ t ⊗ σ \rho_t\otimes\sigma ρ t ⊗ σ , so its Husimi function factors as the product of the Husimi functions of ρ t \rho_t ρ t and σ \sigma σ . The joint Husimi function is therefore bounded by the Gaussian max ( C , C ′ ) e − min ( ϵ , ϵ ′ ) ( ∣ α ⃗ ∣ 2 + ∣ β ⃗ ∣ 2 ) \max(C,C')e^{-\min(\epsilon,\epsilon')(|\vec{\alpha}|^{2}+|\vec{\beta}|^{2})} max ( C , C ′ ) e − m i n ( ϵ , ϵ ′ ) ( ∣ α ∣ 2 + ∣ β ∣ 2 ) . Let C ′ ′ = max ( C , C ′ ) C''=\max(C,C') C ′′ = max ( C , C ′ ) and ϵ ′ ′ = min ( ϵ , ϵ ′ ) \epsilon''=\min(\epsilon,\epsilon') ϵ ′′ = min ( ϵ , ϵ ′ ) . The coordinate rotation property of PLTs implies that the same bound also applies to the state ρ o u t \rho_{\mathrm{out}} ρ out after the BS that is measured by the two counters.
The measurement statistics p j k = ⟨ Π j ⊗ Π k ⟩ p_{jk}=\langle\Pi_j\otimes\Pi_k\rangle p j k = ⟨ Π j ⊗ Π k ⟩ are obtained by projecting ρ o u t \rho_{\mathrm{out}} ρ out onto the total-photon-number eigenspaces of the two output ports. Crucially, total-photon-number measurements are completely insensitive to local PLTs acting within their respective output spaces. We may therefore independently apply a uniform Haar-measure twirl over U ( S ) U(S) U ( S ) on each port’s output modes without altering the probabilities p j k p_{jk} p j k . The Gaussian bound on the joint Husimi function is again preserved. By the same Schur’s lemma argument utilized in Thm. 4.3, this dual twirling operation maps ρ o u t \rho_{\mathrm{out}} ρ out to a state ρ ~ o u t \widetilde{\rho}_{\mathrm{out}} ρ out that is completely diagonal in the joint total-photon-number basis:
ρ ~ o u t = ∑ j , k = 0 ∞ p j k d ( j ) d ( k ) Π j ⊗ Π k (64) \widetilde{\rho}_{\mathrm{out}}=\sum_{j,k=0}^{\infty}\frac{p_{jk}}{d(j)d(k)}\Pi_j\otimes\Pi_k \tag*{(64)} ρ out = j , k = 0 ∑ ∞ d ( j ) d ( k ) p j k Π j ⊗ Π k (64)
where d ( l ) d(l) d ( l ) is the dimension of the k k k -photon subspace of S S S modes. For this diagonal state, the relationship between the photon-number probabilities and the phase-space variables simplifies significantly. Using the expressions for the probabilities of the total photon number of a multi-mode coherent state [37 ] , we evaluate the Husimi function of ρ ~ o u t \widetilde{\rho}_{\mathrm{out}} ρ out , which yields the explicit power series:
Q ~ o u t ( α ⃗ ′ , β ⃗ ′ ) = 1 π 2 S e − ( ∣ α ⃗ ′ ∣ 2 + ∣ β ⃗ ′ ∣ 2 ) ∑ j , k = 0 ∞ p j k d ( j ) d ( k ) ∣ α ⃗ ′ ∣ 2 j j ! ∣ β ⃗ ′ ∣ 2 k k ! . (65) \widetilde{Q}_{\mathrm{out}}(\vec{\alpha}^{\,\prime},\vec{\beta}^{\,\prime})=\frac{1}{\pi^{2S}}e^{-(|\vec{\alpha}^{\,\prime}|^{2}+|\vec{\beta}^{\,\prime}|^{2})}\sum_{j,k=0}^{\infty}\frac{p_{jk}}{d(j)d(k)}\frac{|\vec{\alpha}^{\,\prime}|^{2j}}{j!}\frac{|\vec{\beta}^{\,\prime}|^{2k}}{k!}. \tag*{(65)} Q out ( α ′ , β ′ ) = π 2 S 1 e − ( ∣ α ′ ∣ 2 + ∣ β ′ ∣ 2 ) j , k = 0 ∑ ∞ d ( j ) d ( k ) p j k j ! ∣ α ′ ∣ 2 j k ! ∣ β ′ ∣ 2 k . (65)
Applying the inherited Gaussian upper bound and multiplying through by π 2 S e ∣ α ⃗ ′ ∣ 2 + ∣ β ⃗ ′ ∣ 2 \pi^{2S}e^{|\vec{\alpha}^{\,\prime}|^{2}+|\vec{\beta}^{\,\prime}|^{2}} π 2 S e ∣ α ′ ∣ 2 + ∣ β ′ ∣ 2 , we obtain the inequality
g ( ∣ α ⃗ ∣ 2 , ∣ β ⃗ ∣ 2 ) = ∑ j , k = 0 ∞ p j k d ( j ) d ( k ) ∣ α ⃗ ∣ 2 j j ! ∣ β ⃗ ∣ 2 k k ! ≤ π 2 S C ′ ′ e ( 1 − ϵ ′ ′ ) ( ∣ α ⃗ ∣ 2 + ∣ β ⃗ ∣ 2 ) . (66) g(|\vec{\alpha}|^{2},|\vec{\beta}|^{2})=\sum_{j,k=0}^{\infty}\frac{p_{jk}}{d(j)d(k)}\frac{|\vec{\alpha}|^{2j}}{j!}\frac{|\vec{\beta}|^{2k}}{k!}
\leq\pi^{2S}C''e^{(1-\epsilon'')(|\vec{\alpha}|^{2}+|\vec{\beta}|^{2})}. \tag*{(66)} g ( ∣ α ∣ 2 , ∣ β ∣ 2 ) = j , k = 0 ∑ ∞ d ( j ) d ( k ) p j k j ! ∣ α ∣ 2 j k ! ∣ β ∣ 2 k ≤ π 2 S C ′′ e ( 1 − ϵ ′′ ) ( ∣ α ∣ 2 + ∣ β ∣ 2 ) . (66)
The moment generating function for p j k p_{jk} p j k is
h ( u , v ) = ∑ j , k = 0 ∞ p j k e u j + v k . (67) h(u,v)=\sum_{j,k=0}^{\infty}p_{jk}e^{uj+vk}. \tag*{(67)} h ( u , v ) = j , k = 0 ∑ ∞ p j k e u j + v k . (67)
The goal is to show that this is finite for some neighborhood of the origin. For this purpose, note that d ( l ) ≤ S l d(l)\leq S^{l} d ( l ) ≤ S l , and since ∫ 0 ∞ d x x l e − x = l ! \int_{0}^{\infty}dx\,x^{l}e^{-x}=l! ∫ 0 ∞ d x x l e − x = l ! , we have
e u l ≤ e u l S l d ( l ) ∫ 0 ∞ d x x l e − x l ! = ∫ 0 ∞ d x ( S e u x ) l d ( l ) l ! e − x . (68) e^{ul}\leq e^{ul}\frac{S^{l}}{d(l)}\frac{\int_{0}^{\infty}dx\,x^{l}e^{-x}}{l!}=\int_{0}^{\infty}dx\,\frac{(Se^{u}x)^{l}}{d(l)l!}e^{-x}. \tag*{(68)} e u l ≤ e u l d ( l ) S l l ! ∫ 0 ∞ d x x l e − x = ∫ 0 ∞ d x d ( l ) l ! ( S e u x ) l e − x . (68)
Applying this inequality to the exponential terms in the series for h ( u , v ) h(u,v) h ( u , v ) gives
h ( u , v ) ≤ ∑ j , k = 0 ∞ p j k ∫ x , y = 0 ∞ d x d y ( S e u x ) j d ( j ) j ! e − x ( S e v y ) k d ( k ) k ! e − y = ∫ x , y = 0 ∞ d x d y e − ( x + y ) g ( S e u x , S e v y ) ≤ ∫ x , y = 0 ∞ d x d y e − ( x + y ) π 2 S C ′ ′ e ( 1 − ϵ ′ ′ ) ( S e u x + S e v y ) . (69) \begin{aligned}
h(u,v)\leq\sum_{j,k=0}^{\infty}p_{jk}\int_{x,y=0}^{\infty}dxdy\frac{(Se^{u}x)^{j}}{d(j)j!}e^{-x}\frac{(Se^{v}y)^{k}}{d(k)k!}e^{-y} \\
&=\int_{x,y=0}^{\infty}dxdy\,e^{-(x+y)}g(Se^{u}x,Se^{v}y) \\
&\leq\int_{x,y=0}^{\infty}dxdy\,e^{-(x+y)}\pi^{2S}C''e^{(1-\epsilon'')(Se^{u}x+Se^{v}y)}. \tag*{(69)}
\end{aligned} h ( u , v ) ≤ j , k = 0 ∑ ∞ p j k ∫ x , y = 0 ∞ d x d y d ( j ) j ! ( S e u x ) j e − x d ( k ) k ! ( S e v y ) k e − y = ∫ x , y = 0 ∞ d x d y e − ( x + y ) g ( S e u x , S e v y ) ≤ ∫ x , y = 0 ∞ d x d y e − ( x + y ) π 2 S C ′′ e ( 1 − ϵ ′′ ) ( S e u x + S e v y ) . (69)
which is finite for S e u < 1 / ( 1 − ϵ ′ ′ ) Se^{u}<1/(1-\epsilon'') S e u < 1/ ( 1 − ϵ ′′ ) and S e v < 1 / ( 1 − ϵ ′ ′ ) Se^{v}<1/(1-\epsilon'') S e v < 1/ ( 1 − ϵ ′′ ) . In the second line, the interchange of the sums over j , k j,k j , k with the integrals over x , y x,y x , y is justified by Tonelli’s theorem [51 Chap. 1] . In particular, each summand in g ( S e u x , S e v y ) g(Se^{u}x,Se^{v}y) g ( S e u x , S e v y ) is a continuous, hence measurable, and non-negative function of ( x , y ) ∈ R 2 (x,y)\in\mathbb{R}^{2} ( x , y ) ∈ R 2 , so the double sum and the double integral may be exchanged, both sides being well defined as elements of [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) . The moment generating function therefore converges on a neighborhood of the origin and satisfies the condition for moment determinateness [38 Chap. 14,15] .
Corollary C.5. Consider a measurement configuration with one counter. Suppose that the input state ρ \rho ρ is in S g S_g S g . Then, for every probe state σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⃗ ⟩ ⟨ 0 ⃗ ∣ \sigma= |\gamma\rangle\langle\gamma| \otimes|\vec{0}\rangle\langle\vec{0}| σ = ∣ γ ⟩ ⟨ γ ∣ ⊗ ∣ 0 ⟩ ⟨ 0 ∣ the photon-counting probabilities p i p_i p i are uniquely determined by the moments of the total-photon-number operator, ⟨ ( n ^ a ) k ⟩ \langle(\hat{n}^{a})^{k}\rangle ⟨( n ^ a ) k ⟩ .
Proof. The moment generating function for one counter is obtained by substituting v = 0 v=0 v = 0 in the generating function h ( u , v ) h(u,v) h ( u , v ) for two counters defined in Eq. (67) . It therefore also converges in a neighborhood of the origin, which implies moment determinateness.
Anti-normally ordered powers of n ^ \hat{n} n ^
Here, we show that anti-normally ordered powers of n ^ \hat{n} n ^ can be expressed as linear combinations of ordinary powers of n ^ \hat{n} n ^ . While not utilized in the main text, the lemma below complements Lem. B.1 for normally-ordered powers of n ^ \hat{n} n ^ and provides an explicit expression.
We first observe that anti-normal ordering, like normal ordering, commutes with PLTs. In particular, for any PLT U U U and any product of mode operators M ( a ^ † , a ^ ) M(\hat{a}^{\dagger},\hat{a}) M ( a ^ † , a ^ ) , one has
U † : M ( a ^ † , a ^ ) : U = : U † M ( a ^ † , a ^ ) U : . (70) U^{\dagger} :M(\hat{a}^{\dagger},\hat{a}): U = :U^{\dagger}M(\hat{a}^{\dagger},\hat{a})U: . \tag*{(70)} U † : M ( a ^ † , a ^ ) : U =: U † M ( a ^ † , a ^ ) U : . (70)
The proof of this fact parallels the one given for normal ordering in Appendix A, and can be adapted directly to the anti-normal case.
Lemma D.1. For every nonnegative integer k k k , the anti-normally ordered power of n ^ \hat{n} n ^ is given by
: n ^ k : = ( n ^ + ( S − 1 ) + k ) k . (71) :\hat{n}^{k}:=(\hat{n}+(S-1)+k)_{k}. \tag*{(71)} : n ^ k := ( n ^ + ( S − 1 ) + k ) k . (71)
Proof. We prove the lemma by induction on the number of modes S S S . The base case S = 1 S=1 S = 1 can be verified directly by calculation:
a 1 k ( a 1 † ) k = ∑ m = 0 ∞ ( m + 1 ) ⋯ ( m + k ) ∣ m ⟩ ⟨ m ∣ = [ ∑ m 1 = 0 ∞ ( m 1 + 1 ) ∣ m 1 ⟩ ⟨ m 1 ∣ ] [ ∑ m 2 = 0 ∞ ( m 2 + 2 ) ∣ m 2 ⟩ ⟨ m 2 ∣ ] ⋯ [ ∑ m k = 0 ∞ ( m k + k ) ∣ m k ⟩ ⟨ m k ∣ ] = ( a 1 † a 1 + 1 ) ( a 1 † a 1 + 2 ) ⋯ ( a 1 † a 1 + k ) = ( a 1 † a 1 + k ) k . (72) \begin{aligned}
a_{1}^{k}(a_{1}^{\dagger})^{k}=\sum_{m=0}^{\infty}(m+1)\cdots(m+k)|m\rangle\langle m| \\
&=\left[\sum_{m_{1}=0}^{\infty}(m_{1}+1)|m_{1}\rangle\langle m_{1}|\right]\left[\sum_{m_{2}=0}^{\infty}(m_{2}+2)|m_{2}\rangle\langle m_{2}|\right]\cdots\left[\sum_{m_{k}=0}^{\infty}(m_{k}+k)|m_{k}\rangle\langle m_{k}|\right] \\
&=(a_{1}^{\dagger}a_{1}+1)(a_{1}^{\dagger}a_{1}+2)\cdots(a_{1}^{\dagger}a_{1}+k)=(a_{1}^{\dagger}a_{1}+k)_{k}. \tag*{(72)}
\end{aligned} a 1 k ( a 1 † ) k = m = 0 ∑ ∞ ( m + 1 ) ⋯ ( m + k ) ∣ m ⟩ ⟨ m ∣ = [ m 1 = 0 ∑ ∞ ( m 1 + 1 ) ∣ m 1 ⟩ ⟨ m 1 ∣ ] [ m 2 = 0 ∑ ∞ ( m 2 + 2 ) ∣ m 2 ⟩ ⟨ m 2 ∣ ] ⋯ [ m k = 0 ∑ ∞ ( m k + k ) ∣ m k ⟩ ⟨ m k ∣ ] = ( a 1 † a 1 + 1 ) ( a 1 † a 1 + 2 ) ⋯ ( a 1 † a 1 + k ) = ( a 1 † a 1 + k ) k . (72)
For the general case S > 1 S>1 S > 1 , let n ^ − \hat{n}_{-} n ^ − denote the total number operator on the first S − 1 S-1 S − 1 modes, and let n ^ S \hat{n}_{S} n ^ S be the number operator on the last mode. Assuming the formula holds for the first S − 1 S-1 S − 1 modes, by commutativity of operators on different modes, we obtain
: n ^ k : = ∑ j = 0 k ( k j ) : n ^ − j : : n ^ S k − j : = ∑ j = 0 k ( k j ) : n ^ − j : : n ^ S k − j : = ∑ j = 0 k ( k j ) ( n ^ − + S − 2 + j ) j ( n ^ S + ( k − j ) ) k − j . (73) \begin{aligned}
:\hat{n}^{k}:=\sum_{j=0}^{k}\binom{k}{j}:\hat{n}_{-}^{j}::\hat{n}_{S}^{k-j}: \\
&=\sum_{j=0}^{k}\binom{k}{j}:\hat{n}_{-}^{j}: :\hat{n}_{S}^{k-j}: \\
&=\sum_{j=0}^{k}\binom{k}{j}(\hat{n}_{-}+S-2+j)_{j}(\hat{n}_{S}+(k-j))_{k-j}. \tag*{(73)}
\end{aligned} : n ^ k := j = 0 ∑ k ( j k ) : n ^ − j :: n ^ S k − j : = j = 0 ∑ k ( j k ) : n ^ − j :: n ^ S k − j : = j = 0 ∑ k ( j k ) ( n ^ − + S − 2 + j ) j ( n ^ S + ( k − j ) ) k − j . (73)
We want to show that the last line above is equal to ( n ^ + S − 1 + k ) k (\hat{n}+S-1+k)_{k} ( n ^ + S − 1 + k ) k . Since each term in the sum contains an equal number of annihilation and creation operators for each mode, : n ^ k : :\hat{n}^{k}: : n ^ k : is diagonal in the Fock basis. Hence, it suffices to verify that for any Fock state ∣ m ⃗ ⟩ |\vec{m}\rangle ∣ m ⟩ , with total photon number m = m 1 + ⋯ + m S m=m_{1}+\cdots+m_{S} m = m 1 + ⋯ + m S , ⟨ m ⃗ ∣ : n ^ k : ∣ m ⃗ ⟩ \langle\vec{m}|:\hat{n}^{k}:|\vec{m}\rangle ⟨ m ∣ : n ^ k : ∣ m ⟩ equals
⟨ m ⃗ ∣ ( n ^ + S − 1 + k ) k ∣ m ⃗ ⟩ = ( m + S − 1 + k ) k . (74) \langle\vec{m}|(\hat{n}+S-1+k)_{k}|\vec{m}\rangle=(m+S-1+k)_{k}. \tag*{(74)} ⟨ m ∣ ( n ^ + S − 1 + k ) k ∣ m ⟩ = ( m + S − 1 + k ) k . (74)
Denoting m 1 + ⋯ + m S − 1 = m ′ m_{1}+\cdots+m_{S-1}=m' m 1 + ⋯ + m S − 1 = m ′ , we obtain
⟨ m ⃗ ∣ : n ^ k : ∣ m ⃗ ⟩ = ∑ j = 0 k ( k j ) ⟨ m ⃗ ∣ ( n ^ − + S − 2 + j ) j ( n ^ S + ( k − j ) ) k − j ∣ m ⃗ ⟩ = ∑ j = 0 k ( k j ) ( m ′ + S − 2 + j ) j ( m S + k − j ) k − j = ∑ j = 0 k ( k j ) j ! ( m ′ + S − 2 + j j ) ( k − j ) ! ( m S + k − j k − j ) = k ! ∑ j = 0 k ( m ′ + S − 2 + j j ) ( m S + k − j k − j ) = k ! ( m ′ + m S + S − 1 + k k ) = ( m + S − 1 + k ) k . (75) \begin{aligned}
\langle\vec{m}|:\hat{n}^{k}:|\vec{m}\rangle\\
&=\sum_{j=0}^{k}\binom{k}{j}\langle\vec{m}|(\hat{n}_{-}+S-2+j)_{j}(\hat{n}_{S}+(k-j))_{k-j}|\vec{m}\rangle\\
&=\sum_{j=0}^{k}\binom{k}{j}(m'+S-2+j)_{j}(m_{S}+k-j)_{k-j} \\
&=\sum_{j=0}^{k}\binom{k}{j}j!\binom{m'+S-2+j}{j}(k-j)!\binom{m_{S}+k-j}{k-j} \\
&=k!\sum_{j=0}^{k}\binom{m'+S-2+j}{j}\binom{m_{S}+k-j}{k-j} \\
&=k!\binom{m'+m_{S}+S-1+k}{k}=(m+S-1+k)_{k}. \tag*{(75)}
\end{aligned} ⟨ m ∣ : n ^ k : ∣ m ⟩ = j = 0 ∑ k ( j k ) ⟨ m ∣ ( n ^ − + S − 2 + j ) j ( n ^ S + ( k − j ) ) k − j ∣ m ⟩ = j = 0 ∑ k ( j k ) ( m ′ + S − 2 + j ) j ( m S + k − j ) k − j = j = 0 ∑ k ( j k ) j ! ( j m ′ + S − 2 + j ) ( k − j )! ( k − j m S + k − j ) = k ! j = 0 ∑ k ( j m ′ + S − 2 + j ) ( k − j m S + k − j ) = k ! ( k m ′ + m S + S − 1 + k ) = ( m + S − 1 + k ) k . (75)
where we have used the identity ( n ) j = j ! ( n j ) (n)_{j}=j!\binom{n}{j} ( n ) j = j ! ( j n ) repeatedly.