Completeness and Nonclassicality of Coherent States for Generalized Oscillator Algebras

The purposes of this work are (1) to show that the appropriate generalizations of the oscillator algebra permit the construction of a wide set of nonlinear coherent states in unified form; and (2) to clarify the likely contradiction between the nonclassical properties of such nonlinear coherent states and the possibility of finding a classical analog for them since they are P-represented by a delta function. In (1) we prove that a class of nonlinear coherent states can be constructed to satisfy a closure relation that is expressed uniquely in terms of the Meijer G-function. This property automatically defines the delta distribution as the P-representation of such states. Then, in principle, there must be a classical analog for them. Among other examples, we construct a family of nonlinear coherent states for a representation of the su(1,1) Lie algebra that is realized as a deformation of the oscillator algebra. In (2), we use a beam splitter to show that the nonlinear coherent states exhibit properties like anti-bunching that prohibit a classical description for them. We also show that these states lack second order coherence. That is, although the P-representation of the nonlinear coherent states is a delta function, they are not full coherent. Therefore, the systems associated with the generalized oscillator algebras cannot be considered `classical' in the context of the quantum theory of optical coherence.


Introduction
The nonclassical properties of light have received a great deal of attention in recent years, mainly in connection with quantum optics [1], quantum information [2], and the principles of quantum mechanics [3]. Among other nonclassical profiles, the related states can be generated by nonlinear processes to have sub-Poissonian statistics [4] or to exhibit squeezing (reduction of the variance) in one quadrature [5]. Some deformations of the algebra generated by the conventional boson operators have been proposed to represent photons with 'unusual properties' [6,7], which have found applications in the photon counting statistics, squeezing and signal-to-quantum noise ratio [8]. The immediate generalizations [9,10] motivated the development of the subject as an important branch of quantum optics [1]. The deformations of the boson algebra include supersymmetric structures [11][12][13][14] for which the so-called polynomial Heisenberg algebras are quite natural [15][16][17][18]. Recently, some non-Hermitian models have been shown to obey the distortions of the boson algebra that arise in the conventional supersymmetric approaches [19][20][21]. In all cases, the deformed oscillator algebras have been used to construct the corresponding generalized (also called nonlinear) coherent states. Most of these states exhibit nonclassical properties that distinguish them from the coherent states of the conventional boson algebra.
In this work we propose a modification of the conventional boson (also called oscillator) algebra that permits the recovering of the majority of the already studied deformed algebras as particular cases. The nonlinear coherent states of these generalized oscillator algebras can be written in the same mathematical form, which facilitates their study. We shall address the discussion to polynomial-like algebras because, as we are going to prove, the closure relation satisfied by the corresponding nonlinear coherent states is expressed uniquely in terms of the Meijer G-function. Accordingly, the P -representation [22,23] of all the nonlinear coherent states we deal with is as singular as the δ-distribution. Then, following [24,25], the fields represented by such states would have a classical analog. However, these states have nonclassical properties that can be exhibited either with the help of a beam splitter [26] (see also [27]) or by showing that their statistics is sub-Poissonian [4]. We face this likely contradiction by interpreting the action of a beam splitter on a given photon state as the equivalent of a double-slit interference experiment in the single-photon regime [28,29] (see also [30]). Then, we use this equivalence to show that the fields represented by the nonlinear coherent states exhibit anti-bunching (the probability of two or more photons arriving together at the same point is zero), so they cannot be modeled in classical form. The conclusion is that the nonlinear coherent states are not full coherent although their P -representation is a delta function. Then, they do not satisfy the notion of classicalness introduced by Glauber in his quantum theory of optical coherence [24].
The paper is organized as follows. In Sec. 2 we generalize the oscillator algebras and give some immediate examples. Sec. 3 deals with the construction of the nonlinear coherent states. We show that the completeness of these states is always possible for the polynomial-like algebras in terms of the Meijer G-function. Some immediate examples are also discussed. In Sec. 4 we show that the P -representation of the nonlinear coherent states is given by the δ-distribution while the photon states are represented by the derivatives of δ. In Sec. 5 we analyze the likely contradiction between the completeness and the nonclassicality of the nonlinear coherent states. Some final remarks are given in Sec. 6. We have added an appendix with some important mathematical expressions that are not required for reading the paper but are necessary to follow the calculations.

Generalized oscillators
The conventional boson ladder operatorsâ andâ † satisfy the algebraic relation [â,â † ] = 1. If the number operatorn =â †â is considered, then [n,â] = −â and [n,â † ] =â † . The latter expressions define the oscillator (or boson) algebra and show that the action ofâ † on any number eigenvector |n , n = 0, 1, 2, . . ., produces a new eigenvector ofn with eigenvalue n + 1. Similarly,â|n + 1 is proportional to the number eigenvector |n , whileâ|0 = 0. In the sequel we shall modify the oscillator algebra by preserving the number operatorn but changing the ladder operators, now writtenâ E andâ † E , as the set of generators. We say that any system obeying the new algebra is a generalized oscillator.

Deformed oscillator algebras
To define the generalized oscillators let us introduce the pair of ladder operatorŝ where E is a nonnegative function,â † E is the Hermitian conjugate ofâ E , and |n is eigenvector of the number operator with eigenvalue n. The following algebraic relations can be proven The productâ † Eâ E preserves the number of quanta since it is equal to the function E(n). Equivalently,â Eâ † E = E(n + 1). Then As the vacuum state |0 does not contain quanta we shall assume E(0) = 0 in order to haveâ E |0 = 0. The number eigenvectors |n = (a † ) n √ n! |0 , n = 0, 1, 2, . . . , are now expressed as and will be used as the orthonormal basis of the Hilbert space H. The latter consists of all vectors |ψ = +∞ n=0 ψ n |n , ψ n := n|ψ ∈ C, such that

Examples
The following list of examples do not exhaust all the possible generalizations of the oscillator algebras that can be performed with the rules defined in Eqs. (1)-(3). We explicitly mention such cases because either they are connected with approaches already reported, or they give rise to very important results in mathematical physics or they find applications in quantum optics.

f -oscillators
An important class of generalized harmonic oscillators has been already introduced and is nowadays known as the set of f -oscillators [10]. These oscillators are recovered here by making E(n) =nf 2 (n) in (1)-(3), with f a properly chosen real-valued function. In this case the ladder operatorsâ E andâ † E are factorized aŝ The f -oscillators have been associated with the center-of-mass motion of a trapped and bichromatically laser-driven ion [9], and are also related to a 'frequency blue shift' in high intensity photon beams [10].

q-deformed oscillators
It has been shown [10] that the special choice of the f -function reduces the commutator (3) of the f -operators (7) to the rule of the q-deformed oscillators obeyed by the "physics" bosons [6,7] a Eâ † The above approach has important applications in quantum optics as regards the photon counting statistics, squeezing and signal-to-quantum noise ratio [8].

Polynomial-like oscillators
Other relevant class of oscillators is obtained by assuming that E(n) is a real polynomial of degree , The expression is easily achieved and will be useful in the sequel. In this case the commutator relations (2)-(3) define a polynomial Heisenberg algebra of degree − 1. The polynomial algebras are quite natural in the higher order supersymmetric approaches [11][12][13][14], and are usually connected with nonlinearities that arise because the differential order of the operators that intertwine the susy partner Hamiltonians is greater than one [12]. The examples discussed in Secs. 2.2.4, 2.2.5 and 2.2.6, below, are special cases of polynomial algebras.
Assuming that the solutions of (15) are given, the number eigenvectors |n + 2 are now generated from the 1-photon state The above construction is associated with the eigenstates of a series of Hermitian Hamiltonians that share their spectrum with the conventional harmonic oscillator, which is shifted in one unit of energy. That is, the energies are given by E n = n − 1 /2 [15]. Such Hamiltonians are supersymmetric (SUSY) partners of the conventional oscillator for which the supersymmetry is unbroken [11]. A more general treatment includes non-Hermitian Hamiltonians whose eigenvalues are given by E n+1 = n + 1 /2, and E 0 = < 1 /2 [19,20]. That is, the spectrum of such non-Hermitian Hamiltonians includes all the oscillator energies n + 1 /2 plus an additional real eigenvalue which is located below the ground state energy of the conventional oscillator. The E-function for this case can be derived from (10) with α 1 = α 2 = α 3 = β 1 = 1, β 2 = 1 2 − , and β 3 = 3 2 − . Thus, E(n + 2) = (n + 1) n + The non-Hermitian Hamiltonians associated to (18) are constructed with a complexvalued potential [19], so that their study requires a bi-orthogonal structure for the space of states [20,21]. In the appropriate limit, the imaginary part of the potential is cancelled and the model becomes Hermitian, although the generalized algebra defined by (18) is preserved. Then, in such limit, the algebraic structure of the oscillators reported in [15] is recovered by making = − 1 /2 in (18).
Notice that the algebras described above depend on the ground energy of the system. That is, systems with different ground energies will be regulated by different algebras.

Distorted SUSY-like oscillators
Another E-function that satisfies the supersymmetric relations (15)-(17) is defined as E(n + 2) = w +n, with w a nonnegative parameter and E(1) = E(0) = 0. In this case the commutator (3) gives Thus, the operatorsâ E andâ † E are the generators of an algebra that imitates the Heisenberg one. For if we concentrate on the subspace spanned by {|n , n ≥ 2} only, then the commutator (19) is completely equivalent to the oscillator one. The same occurs in the subspace spanned by |1 , up to the constant w ≥ 0. For this reason, the algebra defined by (19) is referred to as distorted Heisenberg algebra, and w is called the distortion parameter [16,17].
We would like to remark that, contrary to what happens with the algebra of the previous section, the distorted Heisenberg algebra does not depend on the ground energy of the system. Besides, the results associated to (19) are also easily extended to non-Hermitian Hamiltonians by using the bi-orthogonal approach developed in [20,21].
Other versions of polynomial algebras can be obtained from either the N -fold or the nonlinear supersymmetric models discussed in [13,14] and [12] respectively.

Nonlinear coherent states
Up to a normalization constant, the solutions to the eigenvalue equationâ E |z E = z|z E , with z ∈ C, can be written in the form The vector |z E belongs to H only if the series (20) must be finite. Clearly, not any E and |z| are allowed. In the following we assume that N E (|z|) is finite for z ∈ S ⊆ C, with S defined whenever E has been provided, and say that |z E N is a generalized (nonlinear) coherent state. Concrete realizations will be shown in the examples.
The probability of having n photons associated to the nonlinear coherent state |z E N is given by , and the average photon number is Following [34] we may introduce the E-exponential function (see details in Appendix A) Then, using (4), the coherent state (20) and the normalization constant (21) can be written in simpler form, respectively |z E = e za † E E |0 and N E (z) = (e |z| 2 E ) −1/2 . • For the identity function E = id we have e x id = e x as an appropriate limit, see Eq. (A-4) of the Appendix. Then, the vectors |z id N ≡ |z N represent the conventional coherent states of the harmonic oscillator.

Completeness
As usual, although the nonlinear coherent states |z E N are not mutually orthogonal (hereafter the symbol * denotes complex conjugation), they satisfy a closure relation with dµ E (z) a measure function to be determined. Let us write where Λ E is an additional function to be determined, d 2 z = rdrdθ, and z = re iθ . After integrating over θ, the expression (24) is as follows As the number eigenvectors |n form a complete set, the above equality is achieved when- After the change n → m−1, the integral equation (27) coincides with the Mellin transform [35] of Λ E (x).

Examples
Once the algebras (2)-(3) and the related coherent states (20), with closure relation (24), have been given in general form, it is profitable to analyze concrete realizations in detail. The coherent states for the f -oscillators of Sec. 2.2.1 and those for the q-oscillators of Sec. 2.2.2 have been exahustively studied in [10] and [6,7], respectively. Hence, it is not necessary to revisit them in the present work. Nevertheless, we would like to mention that the former have been used in approaching the Jaynes-Cummings model for some nonlinear Kerr media [36], and that the coherent states of the q-oscillators can be recovered from those reported in [10] as a particular case (see [37,38] for early constructions using other definition of the E-function). Next, we pay attention to the coherent states associated with the polynomial algebras derived in Sec. 2.2.3. The reason is that such algebras are general enough to include a plenty of cases which have a common property. Namely, the measure permitting the resolution of the identity (24) is given in terms of a Meijer G-function.

Polynomial-like oscillators
The explicit form of the E-exponential function (23) for the -polynomial E-function (10)-(11) is given in Eq. (A-2) of the Appendix. Except for some atypical cases, the normalization constant (21) is well defined so that the related coherent states |z E N are in the Hilbert space H. To satisfy the closure relation (24), in this case the Mellin transform (27) is simplified by using the change of variables: Thus, we arrive at the moment problem which is the Mellin-Barnes integral representation [39] of the following Meijer G-function After substituting this last result into (28) we obtain the explicit form of the measure (25) we are looking for.

su(1, 1)-oscillators
For the E-function derived in Sec. 2.2.4, the explicit form of e x E is given in Eq. (A-5) of the Appendix. In particular, using α 1 = α 2 = β 1 = 1 one gets the expression e x E = I 1 (2 √ x)/ √ x, with I ν (z) the modified Bessel function of the first kind [39]. Then, the normalization constant N E (z) = |z|/I 1 (2|z|) is finite for any z ∈ C, and the measure acquires the form with K ν (z) the modified Bessel function of the second kind [39]. Therefore, the generalized SU (1, 1) coherent states form an over-complete set in the Hilbert space H.
The probability P E (n, |z|) of detecting n photons and the average photon number n z E are in this case given by the expressions In Fig. 1(a) we can appreciate that the maximum of the probability P E (n, |z|) is shifted to the right of the |z|-axis as n → ∞. For n = 0, the probability decreases exponentially as |z| → ∞, see Fig. 1(b). The latter is consistent with the behavior of the average photon number n z E since it grows up linearly with |z|, see Fig. 1(c).

SUSY-oscillators
The nonlinear coherent states for the supersymmetric oscillators of Secs. 2.2.5 and 2.2.6 are constructed by adjusting the superpositions (20) to the rule (17) since the vacuum |0 and the 1-photon |1 states are annihilated by both ladder generators,â E andâ † E , just as this has been indicated in Eq. (15). They can be also obtained as a limit case from either the Hermitian approaches reported in [16,17] (important improvements are reported in [18]), or the non-Hermitian ones introduced in [20,21].

Hilbert spaces of analytic functions
One of the main properties of the coherent states is their ability to form a basis of the Hilbert space H, even when they are nonorthogonal [24,25]. In this form, they can be used to represent not only any vector in H, but also the operators defined to act on H in closed form [22,23]. In the previous section we have shown that the coherent states (20) associated with the generalized oscillator algebras (2)-(3) satisfy the identity resolution (24)- (25). Here we shall go a step further by applying such property in constructing new representation spaces for the states of the new oscillators.
Using the identity resolution (24)-(25), the vectors (5) can be expressed as a superposition of coherent states where dσ E (z * ) = dσ E (z), and the complex series defines the representation of |ψ in the basis |z E N . As we are assuming that E(n) is such that N E (|z|) in (21) is finite, it could be shown that (36) converges for all finite |z|. Concrete realizations depend on the explicit form of the E-function. In particular, for the cases discussed in Sec. 3.2, the ψ(z) are complex-valued functions which are analytic over the whole complex z-plane [17,20]. Indeed, as these functions are holomorphic and are in one-to-one correspondence with the number eigenstates, they are elements of a Hilbert space of entire functions F E named after Fock [40] and Bargmann [41]. In general, from the Schwarz inequality we get |ψ(z)| ≤ N −1 E (z)|| |ψ ||, so that the growth of |ψ(z)| will be bounded from above by the reciprocal of the normalization constant. In such a representation it follows that Thus, the action of the creation operatorâ † E on the space F E is reduced to the multiplication by z. On the other hand, for the annihilation operator one gets Comparing with the derivative of ψ(z) with respect to z, we realize thatâ E is not the canonical conjugate ofâ † E in F E for arbitrary forms of E. Nevertheless, the operatorb E that corresponds to (39) and satisfies [b E ,â † E ] = 1 in F E produces also the linearization of the algebra (1)- (3). Preliminary results on the matter can be found in [18], the detailed construction ofb E for the general case we are dealing with will be reported elsewhere.
Of course, if E = id the above expressions are reduced to those of the conventional oscillator for which the Fock-Bargmann space F id is formed by entire analytic functions of growth ( 1 /2, 2). In this case, the usual boson operatorsâ † id =â † andâ id =â correspond to the multiplication by z and the derivative with respect to z, respectively.
The representation (36) is useful to describe pure states only. A more general and versatile representation is offered by the density operatorρ which includes pure states, Trρ 2 = Trρ, as well as mixed states Trρ 2 < Trρ. Following [22,23], let us writeρ in P -representationρ The main point here is to find the appropriate P -function such that (40) can be interpreted as a 'diagonal' continuous matrix representation ofρ. In doing so, P (z) would play the role of a nonnegative weight function, defined at all points of the complex z-plane.
As indicated earlier, for E = id we recover the conventional coherent states |z id N = |z N of the harmonic oscillator. In such a case the density operators (40) describe the light emitted by a completely chaotic source, a model that includes all known natural light sources [24]. Also in this case the P -function need not have the properties of a probability distribution [23] and does not exist for allρ [22]. However, as the classical probability theory allows for delta function distributions, P can be as singular as δ (2) (z) = δ(Rez)δ(Imz) [22,23]. In the quantum theory of optical coherence [24], if the P -function of a given stateρ doest not possess properties of a classical probability distribution, or it does not exist, such state does not have classical analog. Coming back to our approach, in the simplest case, the density operatorρ z = |z E z E | of any of the coherent states in the sum (40) should be represented by a distribution δ (2) (z − z ). Otherwise, the identity resolution (24)-(25) would be not valid. Thus, as in the conventional case, the function P (z) should have, at most, δ-type singularities.
To investigate the P -representation for other states consider the superposition we may calculate the matrix elements Then, the P -function is obtained from the two-dimensional inverse Fourier transform As an immediate application consider the number eigenstateρ n = |n n|, then P n (z) = E(n)! (n!) 2 Λ(r 2 ) ∂ 2n ∂z n ∂z * n δ (2) (z), n ≥ 0. (44) That is, with exception of the vacuum |0 , the P -representation of the number eigenstates |n is as singular as the derivatives of the δ-distribution. Therefore, according to [24], the fields represented by any of the number eigenstates |n + 1 do not have classical analog. This result is quite natural since the states |n + 1 cannot be described in classical terms, no matter the approach used in their study. In this form, the continuous matrix representation (40) is consistent with the results obtained in terms of the conventional coherent states for the vectors |n + 1 .
On the other hand, the P -representation of the vacuum |0 and the generalized coherent states |z E N is the delta function δ (2) (z − z ), with z = 0 for |0 . The above criterion of classicality would mean that the nonlinear coherent states |z E N are able to represent fields with classical analog, at least at the same level as the vacuum |0 . The latter because the singularity of the delta distribution is integrable, so that P = δ (2) is admissible as a classical probability distribution [24]. Such statement is rather clear for E = id since the conventional coherent states |z id N are indeed as classical as the vacuum |0 . The situation changes if E = id, as we are going to see in the next sections.

Do the completeness of generalized coherent states imply classicality?
We have shown that all the nonlinear coherent states |z E N associated with the -oder polynomials (10)-(11) have a resolution to the identity (24) that is expressed in terms of the Meijer G-function (30). The latter means that the P -representation of all the superpositions (20) is the distribution δ (2) (z − z ), just as this occurs for the vacuum |0 which is P -represented by δ (2) (z). Hence, |z E N is a displaced version of |0 . The question is if such property is a sufficient condition for the vectors |z E N to represent fields with classical analog. We look for an answer to this problem by using the criterion introduced in [26], as well as the parameter introduced by Mandel [4], to identify the possible classicality of such states.

Beam splitter criterion
Consider a single photon |1 entering a 50 : 50 beam splitter The subindex 'H' ('V ') stands for the horizontal (vertical) channel of the beam splitter, see Fig. 2. As a vacuum |0 enters the other input port of the beam splitter, the entire input state is the product |1 ⊗ |0 ≡ |1, 0 . Hereafter the ket at the left (right) in the tensor products |· ⊗ |· stands for 'horizontal' ('vertical') signal with respect to the beam splitter shown in Fig. 2. The output is the Bell state |β = 1 √ 2 (|1, 0 + i|0, 1 ) which, as it is well known, encodes nonclassical correlations [2,3]. Such state is distinguished from the classical correlation ρ clas = 1 2 (|1 1| ⊗ |0 0| + |0 0| ⊗ |1 1|) because the off-diagonal elements of its density operator |β β| are associated with transitions |0 ↔ |1 , occurring in both channels, that are invariant under a change of basis. That is, the off-diagonal elements |β β|−ρ clas that are different from zero produce entanglement. In general, when the state |n, 0 enters the beam splitter, one gets the well known binomial distribution of bi-partite photon states where a global phase has been dropped. The straightforward calculation shows that the off-diagonal elements of the density operator BS|n, 0 n, 0|(BS) † are different from zero, so that the pure state (46) encodes nonclassical correlations. These off-diagonal elements are such that measuring the number of photons at the horizontal output port of the beam splitter is affected by the result of detecting photons at the vertical port and vice versa. This last result has motivated the conjecture that the entangled output state from a beam splitter requires nonclassicality in the input state [26]. Assuming that the conjecture can be proved (see, e.g. [27]), this would be used as a criterion for nonclassicality. For if the off-diagonal terms of the output state |ψ out ψ out | are nontrivial, then the input state |ψ in ψ in | is nonclassical in at least one of its two channels [26]. .
Here, B(a, b) stands for the Euler beta function [39] which, as far as we know, cannot be expressed as B(a, b) = f (a)g(b) for any functions f and g. Therefore, the probability (47) cannot be factorized as the product of two independent distributions, one for each output port of the beam splitter. This property is concomitant to the impossibility of writing the bi-partite photon states (46) as the product of any state |φ H of the horizontal channel with a state |φ V of the vertical one, that is BS|n, On the other hand, a measure of the nonclassicality of states is given by the Mandel parameter which indeed indicates the degree to which the statistics of a given field is sub-Poissonian [4]. For a field represented by the number eigenvector |n + 1 one gets Q = −1, so that the field is sub-Poissonian (−1 ≤ Q < 0). Classical fields like those represented by either a coherent state or a vacuum are Poissonian (Q = 0), or even super-Poissonian (Q > 0) if they correspond to thermal light. Now, let us writen tot =n H +n V for the total number operator associated to either the input or the output ports of the beam splitter. Heren H =n ⊗ I andn V = I ⊗n are the number operators for the horizontal and vertical channels respectively. We may prove the expression (∆n tot ) 2 = (∆n H ) 2 + (∆n V ) 2 , meaning that the variance of any signal involved with a beam splitter is the result of adding the variances of the horizontal and vertical channels. A simple calculation shows that (46) is such that n H = n V = n 2 , Thus, the horizontal and vertical signals of (46) are sub-Poissonian (nonclassical).
For the present case, the conjecture indicated in [26], hereafter the (Knight) Kconjecture, is trivially verified since the number eigenstate |n injected into the beam splitter to produce the superposition (46) is clearly nonclassical if n = 0.

Classical signals
Although the above discussion is true for the number eigenvectors |n + 1 in any of the input channels, the result cannot be generalized for any superposition of such states. For instance, the input state |z, 0 , with |z id N = |z N a conventional coherent state of the harmonic oscillator, produces the separable (classical) signal The latter confirms that the coherent state |z N is classical. The separability of the state BS|z, 0 can be also studied in terms of the probability P(n, m, |z|) = | n, m|BS|z, 0 | 2 of detecting n photons in the horizontal channel and m photons in the vertical one. The result is shown in Fig. 3 for n = 16 in the input signal. In general, the distribution spreads out on the nm-plane while its center is shifted along the line m = n as |z| → ∞, see Fig. 4.
In this case the probability P(n, m, |z|) can be expressed as the product of two Poisson distributions with mean value |z| 2 / √ 2, one for each output port, The latter means that measuring the number of photons at the horizontal output port does not depend on the result of detecting photons at the vertical output port of the beam splitter. This result enforces the notion of classicality associated to the conventional coherent states of light. Now, it can be shown that the state (49) gives (∆n ) 2 out = n out , with = H, V . Then Q H,out = Q V,out = 0. The latter is consistent with the separability of P(n, m, |z|) indicated above since the same Poisson distribution (50) determines the photon-detection for the two output ports of the beam splitter.
Summarizing the properties of the state (49), generated when the classical signal |z id N enters a 50:50 beam splitter, we have (1.C) The average occupation number P(n, m, |z|) can be factorized as the product of two independent Poisson distributions, one for each output port.
(2.C) The variances of the input and output signals are equal.
(3.C) The variances of the horizontal and vertical output signals coincide and are equal to one half the variance of the input signal.
The point (1.C) and the product in Eq. (49) are concomitant since we can take one of them as a given property to verify the other one, and vice versa. Here, we would like to emphasize that the points (2.C) and (3.C) would serve as a criterion to investigate the possibility of factorizing the state BS|ψ, 0 , equivalently P(n, m, |z|), when |ψ is an arbitrary superposition of photon states. As the factorization of BS|ψ, 0 means no quantum correlations, the properties (2.C) and (3.C) may imply the classicality of |ψ . As we can see, also in this case the K-conjecture is trivially verified since the conventional coherent state |z id N injected into the beam splitter to produce the vector (49) is classical for any |z|.

Nonclassical signals associated with generalized oscillators
The above analysis can be extended to any state of the form (20) that enters the beam splitter in the horizontal port, together with a vacuum |0 in the vertical channel. As we have seen, the separability of the distribution P E (n, m, |z|) plays an important role in the identification of classicality. In the sequel we are going to pay special attention to the SU (1, 1) coherent states derived in Sec. 3.2.2. The other coherent states mentioned in Sec. 3 lead to similar conclusions.
The behavior of this last distribution is shown in Fig. 5 for the average photon number n z E = 16 in the input signal, the latter value has been chosen for comparison with the result of Fig. 3. The squeezing of P E (n, m, |z|) along the line n = m is notable.
Concerning the global profile, as in the previous case, the distribution (51) spreads out in the nm-plane while its center is shifted along the line m = n as |z| → ∞, see Fig. 6. As it occurred for the probability (47), the presence of the beta function in (51) prohibits the factorization of P E (n, m, |z|) as the product of two independent probability distributions if n and m are both different from zero. The latter means that detecting photons at the vertical output port of the beam splitter affects the counting of photons in the horizontal output port.
In turn, the variances (∆n E,tot ) 2 and (∆n E,tot ) 2 out of the input |z E , 0 and output BS|z E , 0 signals are not the same. Indeed, the latter is larger than the former for practically any |z| = 0, see Fig. 7(a). However, the horizontal and vertical output variances coincide and they are shorter than the variance of |z E , 0 , see Fig. 7(b). Such squeezing is specially clear along the line n = m, as this has been indicated above. See Fig. 8 for a detailed comparison.
The above results indicate that the SU (1, 1) coherent sates (33) are nonclassical although they can be interpreted as displaced versions of the vacuum |0 . Indeed, the direct calculation shows that the Mandel parameter associated to |z E N is such that − 1 2 ≤ Q ≤ 0 for all |z| ≥ 0, and Q = 0 for |z| = 0 only. Moreover, Q → − 1 2 as |z| → ∞, see the blue curve depicted in Fig. 9. That is, the statistics associated with the state (33) is sub-Poissonian, so that it is nonclassical. On the other hand, the Mandel parameter for the horizontal and vertical channels of the output signal in the interferometer gives a result that is equal to one-half the result of the input signal, see the dashed-red curve in Fig. 9.
The properties of the state that is generated when the nonclassical signal |z E N enters a 50:50 beam splitter are as follows (1.N) The average occupation number P E (n, m, |z|) cannot be factorized as the product of two independent probability distributions.   Then, the K-conjecture is also verified for the nonlinear coherent states defined in (33). Namely, as the output state BS|z E , 0 is non-separable, the state |z E N in the horizontal input port of the beam splitter is nonclassical. The same conclusion is obtained for any of the nonlinear coherent states (20) that can be constructed with the generalized oscillator algebras (1)-(3). However, we have shown that these states have a P -representation that is proportional to δ (2) (z − z ), so that they can be classified as displaced versions of the vacuum |0 . Is there any contradiction between the non-separability of the states BS|z E , 0 and the P -representation of |z E N ?

Refinement of the criterion
To clarify the results of the previous sections let us emphasize that the states BS|n, 0 defined in (46) are nothing but a class of generalized coherent states [42] (see also [43]) associated with the su(2) Lie algebra realized in terms of two oscillators, i.e., in the  so-called Schwinger representation [44], A simple inspection shows that making ξ = e i π 2 in this last expression gives the state (46). The states |ξ also have a resolution to the identity and are represented by a P -function that is proportional to the δ-distribution. However, they are nonclassical, as this has been discussed in Sec. 5.1.
In many ways, the action of the beam splitter on the incoming state |n, 0 is equivalent to a double-slit interference experiment in the single-photon regime, see e.g. [28,29]. Indeed, since the detector and electronic instrumentation dead-time limitations make difficult the direct measurement of anti-bunching in a double-slit experiment [29], the output intensities are instead measured by using a 50:50 beam splitter and two detectors [28]. These and other photon correlation experiments have their origin in the Hanbury-Brown and Twiss experiments [30], the results of which have shown the importance of distinguishing between the first two orders of coherence. Following Glauber [24], given the normalized form of the correlation functions the necessary condition of coherence is that |g (j) | = 1. The first-order coherence is obtained if j = 1. This coincides with the definition of coherence used in optics previous to the theory of Glauber. On the other hand, a field characterized by n-th order coherence corresponds to j ≤ n. The full coherence implies that |g (n) | = 1 holds for all n. Light beams from ordinary sources can be made optimally first-order coherent but they lack second-order-coherence [24]. In turn, the conventional coherent states are coherent to all orders. Next, we follow [24] and adopt the factorization of G (n) producing g (n) = 1 in (53) as the definition of the n-th order coherent fields.
The straightforward calculation shows that the horizontal and vertical channels of the nonlinear coherent states (46), equivalently (52), lack second-order-coherence since g (2) = 1 − 1 n < 1 for n = 0. This means that the photons in such ports are anticorrelated, so that they cannot be created or annihilated simultaneously. The result g (2) = 1 is due to the fact that G (2) cannot be factorized as the product of two G (1) -functions. The statement is equivalent to the impossibility of factorizing either the vector BS|n, 0 or the probability (47). Notice however that g (2) → 1 for n → ∞. That is, for a large number of photons, the state (46) can be associated with the results of an interference experiment involving a classical field.
This last function satisfies g (2) < 1 for all finite |z|, and goes to one for |z| → ∞, see Fig. 10. As |z| >> 1 implies a large average photon number, see Fig. 1(c), the latter result means that the coherent states (33) allow for bunching of photons only at the limit n z E → ∞. Again, the result g (2) = 1 is associated with the fact that neither the function G (2) nor the probability (51) nor the state BS|z E , 0 can be factorized.
As we have seen, the nonlinear coherent states |z E N are not full coherent although their P -representation is a delta function, so that they cannot be considered 'classical' in the sense established by Glauber [24]. Besides, the impossibility of factorizing either the second order correlation function G (2) , or the probability of detecting n and m photons in the horizontal and vertical output channels, or the output state BS|z E , n associated with a 50:50 beam splitter, means that the fields represented by |z E N are nonclassical.

Concluding remarks
We have shown that the nonlinear coherent states associated to a series of generalized oscillator algebras can be written in the same mathematical form. If such algebras are polynomial the related coherent states satisfy a closure relation that is uniquely expressed in terms of the Meijer G-function. We have also shown that, although the P -representation of these states is as singular as the delta function, they have no classical analog. The latter is due to the fact that such states are not full coherent in the sense established by Glauber in his quantum theory of optical coherence [24]. Then, a field represented by any nonlinear coherent state which is P -represented by a delta function will have classical analog whenever such state is full coherent.
As a byproduct of this work, we have shown that the criterion of separability introduced in [26] for distinguishing nonclassicality of states can be refined by considering also the separability of either the second order correlation function G (2) , or the probability of detecting n and m photons in the horizontal and vertical output ports of a 50:50 beam splitter.