Feasible quantum engineering of quantum multiphoton superpositions
aa r X i v : . [ qu a n t - ph ] S e p Feasible quantum engineering of quantum multiphotonsuperpositions
Magdalena Stobi´nska
Institute of Theoretical Physics and Astrophysics, University of Gda´nsk, ul. Wita Stwosza57, 80-952 Gda´nsk, PolandInstitute of Physics, Polish Academy of Sciences, Al. Lotnik´ow 32/46, 02-668 Warsaw,Poland
Abstract
We examine an experimental setup implementing a family of quantum non-Gaussian filters. The filters can be applied to an arbitrary two-mode inputstate. We assume realistic photodetection in the filtering process and exploretwo different models of inefficient detection: a beam splitter of a small reflectiv-ity located in front of a perfect detector and a Weierstrass transform applied tothe unperturbed measurement outcomes. We explicitly give an operator whichdescribes the coherent action of the filters in the realistic experimental condi-tions. The filtered states may find applications in quantum metrology, quantumcommunication and other quantum tasks.
1. Introduction
Recent technological advances in the field of quantum optics, such as inte-grated optics schemes, allow unprecedented control of various degrees of freedomof optical quantum systems. Nevertheless, generation of quantum states of lightbeyond the set of squeezed vacuum states (deterministic) and, to some approx-imation, pairs of entangled photons (probabilistic and in a postselective way),still remains challenging. Most of the protocols implementing quantum tech-nologies require however the use of more complex states. They often belongto the class of non-Gaussian quantum states (states with non-Gaussian quasi-probability distribution [1]). Their generation seems possible exploiting theefficient source of quantum light based on parametric down conversion (PDC)and quantum engineering.Quantum engineering implements general quantum operations, often de-scribed by the positive operator-valued measures (POVMs). Since Gaussianquantum superpositions of light are produced directly by the PDC source [2],it is interesting to implement non-Gaussian operations. They will turn the
Email address: [email protected] (Magdalena Stobi´nska)
Preprint submitted to Elsevier February 27, 2018 aussian states into the non-Gaussian multiphoton quantum superpositions ofcertain properties, required for realization of concrete quantum tasks. Suchstates are necessary, for example, for obtaining a quantum speed-up in compu-tation with quantum algorithms [3] and quantum super-resolution in quantumphase estimation using the N00N and NmmN states [4]. They may find also ap-plication in Bell inequality tests performed with homodyne detection, the mosteasy accessible, fast and efficient photodetector at present. These tests can beused to certify quantum devices [5].Up to date, the most often experimentally realized non-Gaussian operationscomprise probabilistic single photon addition [6] or subtraction [7]. They closelyapproximate action of the creation and annihilation operator, respectively. Analternative method of implementing the creation operator, based on repeatedspontaneous parametric down-conversion, was theoretically investigated in [8].These operations can alternate, add or cancel certain components of the initialengineered superposition. Since the probability of the success is very small, theycannot be applied iteratively. Quantum engineering in the form of a quantumfilter can only cancel certain components of the initial superposition. Quantumfiltering was demonstrated for one- and two-photon Fock states [9, 10]. Thefilters were based on the Hong-Ou-Mandel interference [11, 12, 13] and werecapable of blocking single photons over photon pairs. A quantum device capableof filtering out two-mode states of light with mode populations differing bymore than a certain threshold, was proposed in [14]. It is called the modulusof intensity difference filter (MDF) and is based on the multiphoton Hong-Ou-Mandel interference performed in a feed-forward loop. It allows engineering ofthe multiphoton quantum superpositions in a way which is preserving specificsuperpositions. This may turn them useful for Bell test and quantum metrology[15, 16]. Some of the features of this filter has already been experimentallydemonstrated in [17].In this paper we examine the experimental scheme from [14] and show thatin fact it implements a whole family of quantum non-Gaussian filters (the MDFis just one of the possibilities). The filters can be applied to an arbitrary in-put state. We assume realistic photodetection in the filtering process. This isan important step in the analysis of quantum filtering, since lossy detection isdetrimental for the possibility of observation of quantum effects. We model theinefficient detection in two different ways: with a beam splitter of a small reflec-tivity located in front of a perfect detector (the usual way) and a Weierstrasstransform, which implements a Gaussian blur on the unaffected measurementoutcome distribution. We show how the filter acts on an input quantum super-position by computing the photon number distributions and purity of the filteredstates. We also construct the Kraus operator for the filters which reveals their coherent action on an arbitrary input. The filtered states may find applicationsin quantum metrology, quantum communication and other quantum tasks.This paper is organized as follows. Section 2 presents the theoretical descrip-tion of the experimental setup implementing a family of quantum non-Gaussianfilters. In subsection 2.1 we recall the Hong-Ou-Mandel interference and gen-eralize it to the multiphoton case. In subsection 2.2 we show state evolution2 igure 1: Experimental setup implementing a family of non-Gaussian quantum filters. De-scription of the setup is given in the main text. within the feed forward loop. In subsection 2.3 we analyze the family of theoutput states which can result from the filter and we construct its Kraus op-erator describing the action on an arbitrary two-mode input state. In Section3 we discuss two models of realistic photodetection. We also give the Krausoperator describing the filters in presence of inefficient detection. In section 4we present numerical computations demonstrating the action of the filter withrealistic photodetection for two important examples of initial quantum super-positions. The results are given for two population regimes: the few photon andthe mesoscopic population of photons in the initial superpositions. The paperis summarized in conclusions.
2. Experimental setup implementing a family of non-Gaussian filters
In this section we introduce an experimental setup (Fig. 1b) which imple-ments a family of non-Gaussian quantum filters. These filters preserve thesymmetry present in photon number distribution of an input two-mode quan-tum state. They implement a non-Gaussian operation and prepare the inputbeam for further quantum tasks. The principle of work of the filters is based onthe multiphoton Hong-Ou-Mandel (HOM) interference (Fig. 1a) observed in afeed forward loop. The same setup was used before to implement the modulusof intensity filter discussed in [14].
Let us recall the two-photon Hong-Ou-Mandel interference. We analyzethe experimental setup shown in Fig. 2. Two identical photons (one in mode a and the other one in b ) interfering at a balanced (50:50) beam splitter (BS)3 igure 2: Experimental setup for observation of the Hong-Ou-Mandel interference. Behind thebeam splitter are located photon counting detectors. dist( S ( i ) , ∆ ( i ) ) denotes the probabilitydistribution of the total photon number S ( i ) and occupation difference ∆ ( i ) at the output(input) ports of the beam splitter. − − ∆ . . . . . . . . . p S i =2 , ∆ i =0 (∆) − − ∆ . . . . . . . . . p S i =2 , ∆ i =2 (∆) Figure 3: Distribution of the output population difference at the exit ports of a 50:50beam splitter for the interference of 2 photons which enter the beam splitter separately – p S i =2 , ∆ i =0 (∆) or via the same input port – p S i =2 , ∆ i =2 (∆). HOM interference manifestsitself in the double-peaked shape of p S i =2 , ∆ i =0 (∆). always exit together. Behind the beam splitter they are registered by the photoncounting detectors. The only possible measurement outcomes are either K = 0and L = 2 or K = 2 and L = 0. Thus, the probability distribution of the outputpopulation difference between the output ports of the beam splitter denoted by∆ = L − K is p (∆ = ±
2) = 1 / p (∆ = 0) = 0. In this case the eventsof “large” (equal to the total photon number) output difference are more likelythan the events of “small” (zero) output difference. If the two photons enterthe beam splitter through the same input port (e.g. a ) and the other port ( b ) isempty, the most likely is that the photons exit separately, i.e. K = L = 1 and p (∆ = 0) = 1 /
2. We denote the initial total photon number by S i = 2 and theinitial population difference by ∆ i . The probability distributions of the outputpopulation difference p S i =2 , ∆ i (∆) are displayed in Fig. 3 for ∆ i = 0 , | n i enter in mode a and b , the most likely event is that the output populationdifference is large (∆ = L − K ≈ S i = 2 n ) [18]. If a Fock state | n i interfereswith the vacuum state, most likely the photons will split equally between the4utput ports (∆ ≈ | n i and | m i on the beam splitter U BS | n, m i a,b = √ n ! m ! 1 √ n + m n X p =0 m X q =0 (cid:18) np (cid:19)(cid:18) mq (cid:19) (1)( − v − p p ( p + q )!( n + m − p − q )! | p + q, v + w − p − q i c,d . The operator U BS describes the action of the 50:50 beam splitter on two inputmodes a and b ( | n i = a † n √ n ! | i , | m i = b † m √ m ! | i ). In the Heisenberg picture ittransforms the creation operators in the following way U † BS a † U BS = ( c † + d † ) / √ U † BS b † U BS = ( c † − d † ) / √
2, where c and d denote the modes exiting BS. Next, theoutput state (1) is measured by the perfect photon counting detectors locatedbehind the beam splitter. This corresponds to a projection of the state (1) onsome Fock states | K i c , | L i d . The total photon number is conserved by the beamsplitter and equals S i = n + m = K + L = S . The probability distribution of theoutput population difference ∆ conditioned on the initial population difference∆ i = n − m and sum S i reads p S i , ∆ i (∆) = |h K, L |U BS | n, m i| (2)= |h S − ∆2 , S +∆2 |U BS | S i +∆ i , S i − ∆ i i| ,p S i , ∆ i (∆) = ( S i − ∆2 )!( S i +∆2 )!2 S i ( S i − ∆ i )!( S i +∆ i )! (3) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) S i +∆ i X p =0 S i − ∆ i X q =0 δ p + q, S − ∆2 (cid:18) S i +∆ i p (cid:19)(cid:18) S i − ∆ i q (cid:19) ( − q (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) . Fig. 4 shows the probability distribution of the output population difference (3)computed for S i = 200 and two extreme cases of ∆ i = 0 and ∆ i = S i = 200.These figures reveal the essence of the multiphoton HOM interference: for twoequal Fock states interfering on a 50:50 beam splitter the most likely eventis that all photons will exit together (the probability distribution p S i , (∆) isdouble-peaked); for a nonzero Fock state interfering with the vacuum the mostlikely event is that the photons will split equally between the output ports (theprobability p S i ,S i (∆) is given by the single-peaked binomial distribution). Pleasenote, that for any S i , ∆ i the distribution p S i , ∆ i (∆) is symmetric: p S i , ∆ i (∆) = p S i , ∆ i ( − ∆).Moreover, the probability distribution p S i , ∆ i (∆) allows to determine theprobability that the modulus of the output population difference is greater orsmaller than a certain threshold. In Fig. 4 we took the threshold δ th = 30. Theprobability that | ∆ | ≥
30 equals 0 .
905 for ∆ i = 0 and 0 .
04 for ∆ i = 200.5) p ( | ∆ | ≥
30) = 0 . p ( | ∆ | <
30) = 0 . − − −
300 30 100 200 ∆ . . . . . . . p S i =200 , ∆ i =0 (∆) b) p ( | ∆ | ≥
30) = 0 . p ( | ∆ | <
30) = 0 . − − −
300 30 100 200 ∆ . . . . . . . p S i =200 , ∆ i =200 (∆) Figure 4: Distributions of the population difference in the output ports of a 50:50 beamsplitter p S i , ∆ i (∆) after interference of a Fock state | n, m i , with n + m = S i = 200 and withthe initial population difference n − m = ∆ i = 0 (a), ∆ i = 200 (b). The distributions aresymmetric: p S i , ∆ i (∆) = p S i , ∆ i ( − ∆). The vertical dashed lines show the threshold δ th = 30.The probability that | ∆ | ≥
30 is given by p ( | ∆ | ≥ Please note that p S, ∆ i (∆) = p S, ∆ (∆ i ) due to the bi-stochastic nature of thesequantum probabilities [19]. This means that the analysis of the measurementoutcomes of the detectors located behind the BS ( S, ∆) allows to forecast thedistribution of the initial population difference (∆ i ) in the input Fock states.Therefore, the plot of p S, ∆ (∆ i ) is identical to the plot of p S, ∆ i (∆) in Fig. 4.This is one of the two key effects exploited by the setup in Fig. 1 implementinga family of non-Gaussian filters. Let us now describe the action of the setup in Fig. 1b. We assume that theinput state (either mixed ρ i or pure | ψ i i ) is a two-mode quantum state. For con-creteness, we assume the modes to be linear polarizations H – horizontal, V –vertical. The input state impinges on a biased beam splitter with small reflectiv-ity (e.g. 10:90). The reflected beam r is sent to a polarizing beam splitter (PBS),oriented such that it selects the unbiased polarization modes ( a d + , a d − ) with re-spect to the incoming linear polarizations ( a r H , a r V ). In this case, the action ofthe polarizing beam splitter U P BS is the same as U BS ( a d + = ( a r V + a r H ) / √ a d − = ( a r V − a r H ) / √ a r H , a r V , a d + , a d − playthe role of a , b , c , d , respectively. The measurement outcomes of the detectorslocated behind the PBS ( S, ∆) reveal the photon number reflected by the 10:90beam splitter S r = S and allow to forecast the population difference ∆ r beforePBS ( S r and ∆ r play the role of S i and ∆ i in the discussion from Subsec-tion 2.1). The only difference is that now the incoming state impinging on thePBS is not a single Fock state but a superposition of those. We will show belowthat nevertheless the reasoning from Subsection 2.1 still applies. The outcomes S and ∆ parametrize the family of non-Gaussian filters in Fig. 1b because theyallow to choose the shape of the probability distribution of ∆ r . Since the re-flected r and transmitted t beams in the feed-forward loop are correlated, S and ∆ also allow to estimate the distributions for the total photon number andpopulation difference between the polarization modes in the incoming ( S i , ∆ i )and transmitted ( S t , ∆ t ) beams, which are symmetric like those in Fig. 4. Thisis the second important effect exploited by the filter. It is especially pronouncedfor larger photon numbers. For example, in case of the 10:90 beam splitter wemost often obtain ∆ r ≈ . i , ∆ t ≈ . i . Knowing the distributions of S t and ∆ t , the analysis box checks the probability that they fulfill certain desiredcondition C( S t , | ∆ t | ) (e.g. | ∆ t | > S t / | ψ i i = P n,m ξ nm | n, m i . After passing the BS and PBS, the state equals | ψ dt i = X n,m ξ nm n X v =0 m X w =0 c ( n ) v c ( m ) w √ v ! w ! √ v + w v X p =0 w X q =0 (cid:18) vp (cid:19)(cid:18) wq (cid:19) ( − v − p p ( p + q )! ( v + w − p − q )! | p + q, v + w − p − q i d | n − v, m − w i t , (4)where c ( n ) k = q(cid:0) nk (cid:1) r k t n − k , r is the reflectivity of the tapping beam splitter and t = 1 − r . The perfect detectors behind the PBS detect two Fock states | K, L i d | ψ dt i to | ψ t i = d h K, L | ψ dt i , (5)= q K ! L !2 K + L X n,m ˜ ξ nm n X v =0 m X w =0 c ( n ) v c ( m ) w √ v ! w ! δ K + L,v + w " v X p =0 w X q =0 (cid:18) vp (cid:19)(cid:18) wq (cid:19) ( − v − p δ K,p + q | n − v, m − w i t . We note that v + w = S r = S and v − w = ∆ r whereas n − v + m − w = S t and n − v − ( m − w ) = ∆ t . The coefficients ˜ ξ nm are renormalized to ensurenormalization of | ψ t i . We compute the conditional photon number distributionfor the transmitted beam p K,L ( k, l ) = |h k, l | ψ t i| . Here, k denotes the photonnumber in polarization H and l in V . After changing the variables L and K so that they corresponded to the quantities useful for the filtering we obtain p S, ∆ ( S t , ∆ t ) with S t = k + l , ∆ t = k − lp S, ∆ ( S t , ∆ t ) = S (cid:0) S +∆2 (cid:1) ! (cid:0) S − ∆2 (cid:1) ! (cid:16) X n,m ˜ ξ nm n X v =0 m X w =0 c ( n ) v c ( m ) w √ v ! w ! δ S,v + w δ S t +∆ t ,n − v δ S t − ∆ t ,m − wv X p =0 w X q =0 (cid:18) vp (cid:19)(cid:18) wq (cid:19) ( − v − p δ S − ∆2 ,p + q (cid:17) . (6)Plots of the above probability distribution computed for a superposition of Fockstates and presented in Appendix B of [14], are similar to the distributions inFig. 4 computed for a single Fock state. The way the quantum filter in Fig. 1b alters the incoming two-mode quan-tum state depends on the information the incoming state carries about thedistributions of the total photon number and of population difference betweenits polarization modes (dist( S i , ∆ i )).We will consider few generic examples illustrating action of the setup inFig. 1b. Our main tool will be the photon number distributions p S, ∆ ( k, l ) de-picted for the transmitted beam, before the shutter, conditioned on the mea-surement of S photons on the reflected beam and population difference ∆ onthe detectors behind the polarizing beam splitter. We emphasis that these plotswill only visualize qualitatively the action of the setup and will show the photonnumber distributions before the selection performed by the analysis box and theshutter according to a certain condition C( S t , | ∆ t | ).We start with examples of states where the total photon number is knownand well-defined. At first we assume the input state to be a Fock state | ψ i i = | S i − N, N i (7)8ith S i − N ≤ N (∆ i ≤ S t = S i − S . Thus, the only points in the photon number space ( k, l ) of thetransmitted beam, for which p S, ∆ ( k, l ) may be nonzero are those on the line ofconstant photon number S t = k + l , see Fig. 5a. The distribution is unsymmetricwith respect to the line ∆ t = 0 (see the blue curve), independently of the valuesof the measured S and ∆. It indicates that most likely ∆ t ≤
0, revealing theasymmetry in the photon number distribution of the input. From Eq. (5) wenotice that | ψ t i ( ˜ ξ nm = δ n,S i − N δ m,N ) is a superposition state of components | N − v, M − w i . Each component has a fixed photon number S t = S i − S andpopulation ∆ t = ∆ i − ∆ r thus, S t and ∆ t characterize them completely. In thisnew notation, the state | ψ t i equals | ψ t i = | S i − S i ∆ maxr X ∆ r =∆ minr f S, ∆ S i , ∆ i (∆ r ) | ∆ i − ∆ r i , (8) f S, ∆ S i , ∆ i (∆ r ) = vuut S (cid:16) S +∆2 (cid:17) ! (cid:16) S − ∆2 (cid:17) ! (cid:18) S +∆ r (cid:19) ! (cid:18) S − ∆ r (cid:19) ! (9) c (cid:18) S i +∆ i (cid:19) S +∆ r c (cid:18) S i +∆ i (cid:19) S − ∆ r A SS − ∆2 (∆ r ) , (10)where ∆ minr = min {− S, S − S i + ∆ i } , ∆ maxr = max { S, S i − S + ∆ i } , A SS − ∆2 (∆ r )is given by the square brackets in Eq. (5). The filter projects the Fock inputstate coherently on a line S t = k + l with ∆ t = k − l ∈ [∆ i − ∆ minr , ∆ i − ∆ maxr ].The analysis box and the shutter will further select some components from(8) according to a condition C( S t , | ∆ t | ). We conclude that the quantum filterperforms the following operation on the Fock input states | S i +∆ i , S i − ∆ i i ≡| S i , ∆ i i P S, ∆C [ S i , ∆ i ] = | S i − S ih S i | (11) ⊗ ∆ maxr X ∆ r =∆ minr C( | ∆ r | ) f S, ∆ S i , ∆ i (∆ r ) | ∆ i − ∆ r i h ∆ i | . If | ∆ minr | = | ∆ maxr | , the function f S, ∆ S i , ∆ i (∆ r ) is symmetric with respect to theline ∆ t = ∆ i . Note that f S, ∆ S i , ∆ i (∆ r ) > P ∆ r f S, ∆ S i , ∆ i (∆ r ) = 1. The plot of anexemplary f S, ∆ S i , ∆ i (∆ r = ∆ i − ∆ t ) is depicted along the line in Figs. 5a – the bluecurve.As the next example we consider a mixture of two Fock states ρ i = q | S i − N, N ih S i − N, N | (12)+ (1 − q ) | S i − M, M ih S i − M, M | . p S, ∆ ( k, l ) will be a sum of the distributions obtained for S i and S i separately, see Fig. 5b. The filter projects each Fock state coherently on aline, but projections on two different lines are incoherent with respect to eachother P { S j , ∆ j } j =1 , C [ { S i j , ∆ i j } j =1 , ] (13)= q P S , ∆ C [ S i , ∆ i ] + (1 − q ) P S , ∆ C [ S i , ∆ i ] . Let us now consider a superposition state where each term has fixed photonnumber S i and population difference is distributed uniformly | ψ i i = √ S i +1 S i X N =0 | S i − N, N i . (14)Since the photon number S i is known, in this case the nonzero elements of p S, ∆ ( k, l ) must be located on a line as well. However, since the terms with neg-ative and positive ∆ i contribute to this state equally, the probability distributionis symmetric with respect to the line ∆ t = 0. Two generic examples of suchdistributions are depicted in Fig. 5c & d. The distribution shown in Fig. 5c isobtained if behind the PBS, the measured population difference roughly equalsthe sum of the reflected photons ( | ∆ | ≃ | ∆ | ≃ S . Again, the projection of the initial superposition on the line l = S t − k , where S t = S i − S , performed by the filter is coherent. The actionof the setup in Fig. 1 on a superposition with a fixed photon number and dis-tribution of population difference is described by following sum of the operators(11) P S, ∆C [ S i ] = S i X ∆ i = − S i P S, ∆C [ S i , ∆ i ] . (15)Another important example is a state with a uniform distribution of both,the initial population difference and the total photon number | ψ i i = √ S i − S i +1 S i X S i = S i √ S i +1 S i X N =0 | S i − N, N i . (16)The uniform distribution is the worst case scenario with respect to the amountof information it carries about the variable. Now from Eq. (6) we see that | ψ t i is a superposition of the following terms | S i − N − v, N − w i , where S t = S i − S and ∆ t = ∆ i + w − v . We note that S t ∈ [ S i − S, S i − S ] and ∆ t ∈ [ − S t , S t ].Thus, there is more than one S i (and S t ) which contributes to the same ∆ t .While computing photon number distribution for | ψ t i we take projection on suchterms simultaneously, i.e. we add their amplitudes of probability. Therefore,now the projections on different lines of photon number S t are coherent : the10lter projects onto a certain area in space ( k, l ) or ( S t , ∆ t ), see Fig. 6. Weconclude that the quantum filter performs the following operation on a generaltwo-mode input states P S, ∆C = ∞ X S i =0 S i X ∆ i = − S i P S, ∆C [ S i , ∆ i ] . (17)This operator plays the role of the Kraus operator for the filter.Please note that regardless the input state | ψ i i = P S i , ∆ i ξ S i , ∆ i | S i , ∆ i i , thesetup in Fig. 1 preserves the symmetry of the initial state in the photon num-ber space. This follows from the fact that the incoming, the reflected and thetransmitted beams are correlated. The filter convolutes the initial photon num-ber statistics with the beam splitter probability distribution which is symmetricwith respect to the population difference ξ S i , ∆ i → ξ S i , ∆ i · f S, ∆ S i , ∆ i (∆ r = ∆ i − ∆ t ),(see Fig. 4). This results in a “blurred” photon number statistics of the trans-mitted beam with respect to the initial one | ψ t i = P S, ∆C | ψ i i (18)= X S i , ∆ i ξ S i , ∆ i | S i − S i⊗ ∆ maxr X ∆ r =∆ minr C( | ∆ r | ) f S, ∆ S i , ∆ i (∆ r ) | ∆ i − ∆ r i . We would like to comment on the filtering condition C( | ∆ r | ) present in theoperators in Eqs. (11)- (18). It directly results from the condition C( S t , | ∆ t | )(∆ t = ∆ i − ∆ r ). So far, the influence of this condition was not discussed. It is anadditional handle which allows to shape the areas and line of projection shownin Figs. 5-6. The physical implementation of the filter in Fig. 1 allows to imposean arbitrary filtering condition C , symmetric with respect to the line k = l (∆ t = 0), on the sum S t = k + l and the modulus of the difference | ∆ t | = | k − l | .The fact that C is symmetric results from the Hong-Ou-Mandel interference.Fig. 7 depicts exemplary projection areas for various filtering conditions.
3. Quantum filtering in presence of imperfect photodetection
The analysis of the operation performed by the family of non-Gaussian filterson the input quantum state presented in Section 2 has not taken into account anyimperfections in the measurement process. In Appendix D of [14] we considereda simple case of losses in the system, modeled with an additional beam splitterput before the shutter. However, that model still assumed the precise measure-ment of S and ∆ by the perfect photon counting detectors. These parameters,as shown above, are of a great importance in the process of finding an outputphoton number distribution of the transmitted beam p S, ∆ ( S t , ∆ t ). Therefore,11) k − l = ∆ t = kl S t S t b) k − l = ∆ t = kl S t S t S t S t c) k − l = ∆ t = kl S t S t d) k − l = ∆ t = kl S t S t Figure 5: Visualization of the photon number distribution for the transmitted beam before the shutter in Fig. 1b for different inputs. a) – a Fock state. In this case S t = S i − S isfixed, where S denotes the photon number registered by the detectors on the reflected beam.The filter projects the input state on the line coherently. b) – a mixture of two Fock stateswith different photon numbers S i and S i . Here, S t = S i − S . The projections ontwo distinct lines are incoherent with respect to each other. c) & d) – superposition of Fockstates of equal total photon number and uniform distribution of initial population difference, S t = S i − S . The projection is also coherent. k − l = ∆ t = kl S t S t S t S t k − l = ∆ t = kl S t S t S t S t Figure 6: Visualization of the photon number distribution for the transmitted beam before the shutter in Fig. 1b for an input state with a uniform distribution of the initial populationdifference and the total photon number | ψ i i = √ S i − S i +1 P S i S i = S i √ S i +1 P S i N =0 | S i − N, N i . After the projection the coherence is preserved within the colored areas. k − l = ∆ t = kl b) k − l = ∆ t = kl c) k − l = ∆ t = kl d) k − l = ∆ t = kl Figure 7: Exemplary projection areas selected with the quantum filter presented on Fig. 1using filtering condition C( S t , | ∆ t | ) applied by the analysis box and the shutter: a) | ∆ t | > ( aS t ) , b) | ∆ t | > aS t (1 + b sin( cS t )), c) | ∆ t | > ⌊ aS t ⌋ /a , d) | ∆ t | > S t + q b − S t − b . Here,0 < a < b >
13n this section we will discuss the influence of the imperfect photodetection onthe filter output.Detection imperfections in the setup presented in Fig. 1b could be causedfor example by the errors in the photon counting process (e.g. arising in thedetector electronics). They result in lower (e.g. losses) or greater (e.g. darkcounts) number of registered photons than expected. This process is indepen-dent for both detectors placed at the outputs of the polarizing beam splitter(Fig. 1a). As a result, a detector could register a different Fock state | K ′ i than | K i which really leaved the PBS. This is represented by some distribution d K ( K ′ ) giving the probability of registering the state | K ′ i instead of state | K i .Here, P K ′ d K ( K ′ ) = 1. The distribution d K ( K ′ ) represents a detector charac-teristics. It transforms the quantum state before the shutter | ψ t i (Eq. (5)) to amixed state ρ ′ t ρ ′ t = X K ′ ,L ′ d K ( K ′ ) d L ( L ′ ) (cid:12)(cid:12) d h K ′ , L ′ | ψ dt i (cid:12)(cid:12) . (19)The imperfect detectors, which turn the pure output quantum state into a mixedone, may significantly affect the coherent action of the filter. According to theformula (19), the filter will now perform the following operation P C = X S ′ , ∆ ′ d S, ∆ ( S ′ , ∆ ′ ) P S ′ , ∆ ′ C , (20)with d S, ∆ ( S ′ , ∆ ′ ) = d S − ∆2 ( S ′ − ∆ ′ ) d S +∆2 ( S ′ +∆ ′ ). In order to examine its in-fluence on the output state, it is necessary to compute the purity of ρ ′ t : γ =Tr { ( ρ ′ t ) } .In our first model, the detector non-unit efficiency η < η . The detector, instead ofprojecting the incoming beam on the Fock state | K i d , projects on a mixture ofFock states | K ih K | → Tr loss {U BS | K, i} (21)= K X x =0 (cid:18) Kx (cid:19) (1 − η ) x η K − x | K − x ih K − x | , with the binomial distribution d K ( K ′ ) = (cid:18) KK ′ (cid:19) (1 − η ) K − K ′ η K ′ . (22)In the limit of η → d K ( K ′ ) → δ K,K ′ – the Kronecker delta, which gives theresult for perfect detectors. Similar results are obtained for detector measur-ing | L i d . Please note that this photodetection model assumes that K ( L ) isthe maximal possible measurement result. The most probable result is η · K .Therefore, this model includes only losses in the photodetector.14he second model of the imperfect detector is described by a Gaussian dis-tribution of a given standard deviation σ . This corresponds to the Weierstrasstransform (known as the Gaussian blur) applied to the photon number distri-bution measured by the ideal detectors d K ( K ′ ) = 1 √ πσ e − ( K − K ′ ) σ . (23)In the limit of σ → d K ( K ′ ) → δ ( K − K ′ ) – the Dirac delta, which correspondsto the perfect detection. This model assumes that the most probably event isthe detection of the actual photon number K . However, it takes into accountthat the detector, with equal probabilities, can measure higher and lower photonnumbers. The detection of higher photon number K ′ > K may happen due todark counts or cross-talks in the separate channels of photodetectors.
4. Numerical results
In order to illustrate the action of the family of non-Gaussian quantum filters,we performed numerical computations for two input quantum states. Below, wewill present the photon number distributions for the output states before theshutter, which result from two important examples of quantum input statesdiscussed in Section III. The first state is a uniform superposition of Fock statesof a constant photon number S i , given by Eq. (14). The second analyzed stateis a uniform superposition of states (14) with the photon number between S i and S i , given by Eq. (16).Fig. 8 depicts the plots of probability distributions p S, ∆ (∆ t ) computed forthe input state (14) before the shutter, with a constant total number of S i = 200photons, reflectivity of a tapping beam splitter 10% and S = 20 photons reg-istered at the detectors. Left column (plots a), c) and e)) contains the dis-tributions obtained for ∆ = 0, whereas right column (plots b), d) and f)) –∆ = S = 20. Plots a) & b) show the distributions in case of the ideal detectors;plots c) & d) – lossy detectors with efficiency η = 95% (red) and η = 80%(black); plots e) & f) – imprecise detectors with the Gaussian distribution ofthe standard deviation 3 σ = 5 (red) and 3 σ = 20 (black).The probability distributions computed for the above cases allow to predictthe probability of meeting a given condition C( S t , | ∆ t | ). This condition can byarbitrary and chosen in order to optimize performance of a certain quantumtask. For example, let us assume that in some quantum application we need aquantum state with difference of population between the modes larger than 120photons, i.e. C( S t , | ∆ t | ) ≡ {| ∆ t | ≥ } . States filtered according to a condi-tion that the population difference between its two modes is greater than somethreshold value are realization of superpositions of the N00N and NmmN states,which find applications in quantum metrology for enhanced optical phase esti-mation [20]. If we know that the source produces a superposition of uniformlydistributed Fock states (Eq. (14)) of total number of S i = 200 photons, themeasurement result of S and ∆ gives us the information of the probability of15ulfilling C( S t , | ∆ t | ). Here, if detectors were perfect, measurement of S = 20,∆ = 0 would give us a certainty that the condition is met (Fig. 8a), whereas∆ = 20 would inform that is not fulfilled with the probability of 0 .
982 (Fig. 8b).Therefore, when ∆ = 0 the box should open the shutter and close it when∆ = 20. Similar analysis would be performed for all possible values of S and ∆.However, imperfections in the detectors influence the results. In case ofdetector efficiency η modeled by the binomial distribution given by Eq. (22),the probability of | ∆ t | ≥
120 conditioned on ∆ = 0 lowers to 0 .
999 for η = 5%and 0 .
962 for η = 20% (Fig. 8c). In the same time the probability of not fulfillingC( S t , | ∆ t | ) when ∆ = 20 raises to 0 .
988 for η = 5% and even 0 .
998 for η = 20%(Fig. 8d). Similar results are obtained in case of the imperfections modeled withthe Gaussian distribution (Eq. (23)), but here p ( | ∆ t | ≥ .
994 for ∆ = 0,3 σ = 5 and 0 .
962 for 3 σ = 20 (Fig. 8e). Finally, Fig. 8f says that observing∆ = 20 gives us probability 0 .
995 that | ∆ t | for 3 σ = 5 and 1 for 3 σ = 20.Fig. 9 depicts the same collection of probability distributions p S, ∆ (∆ t ) of thestate (14) before the shutter, but computed for a small photon number S i = 6.Reflectivity of a tapping beam splitter remainded 10% and we assumed S = 2photons registered at the detectors. Left column (plots a), c) and e)) containsthe distributions obtained for ∆ = 0, whereas right column (plots b), d) and f))– ∆ = S = 2. Plots a) & b) show the distributions in case of ideal detectors;plots c) & d) – lossy detectors with efficiency η = 95% (red) and η = 80%(black); plots e) & f) – imprecise detectors with the Gaussian distribution ofthe standard deviation 3 σ = 0 .
15 (red) and 3 σ = 0 . | ∆ t | exceeds or is below threshold equal to 4.Figs. 10 and 11 show similar computations performed for input state (16)and two ranges of total photon numbers S i : S i ∈ [80 , S i ∈ [4 , S = 10 photons registered by thedetectors and ∆ ∈ { , } . In the second range, total of S = 2 photons isdetected and the difference between the readouts of the detectors is ∆ ∈ { , } .Plots of the projection areas a) & b) represent the case of ideal photon countingdetectors, plots c) & d) – lossy detectors with efficiency η = 80%; plots e) & f)– imprecise detectors with the Gaussian distribution of the standard deviation3 σ = 10 (for S = 10) and 3 σ = 0 . S = 2).Finally, Figs. 12-13 depict the purity computed for the states (14)-(16), re-spectively, and two models of imperfect detection. The results are presented fortwo cases, in which imperfect detectors are modeled by binomial (black curves)and Gaussian (red curves) distribution. Solid black line represents detector ef-ficiency 95%, dashed – 90% and dot-dashed – 80%. Similarly, red solid linedepicts the purity for standard deviation 3 σ = 5, dashed – 3 σ = 10 and dot-dashed – 3 σ = 20. For certain values of the parameters (very likely), the purityreaches 80%. 16) p ( | ∆ t | ≥ p ( | ∆ t | < − − −
40 40 120 200 ∆ t . . . . . . p , (∆ t ) b) p ( | ∆ t | ≥ . p ( | ∆ t | < . − − −
40 40 120 200 ∆ t . . . . . p , (∆ t ) c) p ( | ∆ t | ≥ . p ( | ∆ t | < . p ( | ∆ t | ≥ . p ( | ∆ t | < . − − −
40 40 120 200 ∆ t . . . . . p , (∆ t ) d) p ( | ∆ t | ≥ . p ( | ∆ t | < . p ( | ∆ t | ≥ . p ( | ∆ t | < . − − −
40 40 120 200 ∆ t . . . . . . p , (∆ t ) e) p ( | ∆ t | ≥ . p ( | ∆ t | < . p ( | ∆ t | ≥ p ( | ∆ t | < − − −
40 40 120 200 ∆ t . . . . p , (∆ t ) f) p ( | ∆ t | ≥ . p ( | ∆ t | < . p ( | ∆ t | ≥ p ( | ∆ t | < − − −
40 40 120 200 ∆ t . . . . . . . p , (∆ t ) Figure 8: The plots of probability distributions p S, ∆ (∆ t ) numerically computed for the inputstate (14) before the shutter, with a constant total number of S i = 200, r = 10%, S =20, ∆ = 0 (left column) and ∆ = 20 (right column). The results were obtained for bothperfect (a & b) and imperfect photodetection, with binomial (c & d) and Gaussian (e & f)distribution d K ( K ′ ) ( d L ( L ′ )), representing the detector characteristics. Black and red curveswere obtained for different values of parameters η and σ . Detailed description is presented inthe main text. The numerical results presented in this Section have shown that the quan-tum filter executed by the schema in Fig. 1b may be implemented in realisticexperimental conditions. It is shown explicitly that the setup preserves its co-herent action on an input state even in presence of inefficient photodetection:the photon number distribution of the output state is not distorted and thepurity of output state is quite high. Although the results were computed for aspecific example of the filtering condition, C( S t , | ∆ t | ) ≡ {| ∆ t | ≥ threshold } , theabove conclusions apply to all possible filtering conditions.17) p ( | ∆ t | ≥
4) = 0 . p ( | ∆ t | <
4) = 0 . − − − ∆ t . . . . p , (∆ t ) b) p ( | ∆ t | ≥
4) = 0 . p ( | ∆ t | <
4) = 0 . − − − ∆ t . . . . . p , (∆ t ) c) p ( | ∆ t | ≥
4) = 0 . p ( | ∆ t | <
4) = 0 . p ( | ∆ t | ≥
4) = 0 . p ( | ∆ t | <
4) = 0 . − − − ∆ t . . . . . p , (∆ t ) d) p ( | ∆ t | ≥
4) = 0 . p ( | ∆ t | <
4) = 0 . p ( | ∆ t | ≥
4) = 0 . p ( | ∆ t | <
4) = 0 . − − − ∆ t . . . . . . p , (∆ t ) e) p ( | ∆ t | ≥
4) = 0 . p ( | ∆ t | <
4) = 0 . p ( | ∆ t | ≥
4) = 0 . p ( | ∆ t | <
4) = 0 . − − − ∆ t . . . . . p , (∆ t ) f) p ( | ∆ t | ≥
4) = 0 . p ( | ∆ t | <
4) = 0 . p ( | ∆ t | ≥
4) = 0 . p ( | ∆ t | <
4) = 0 . − − − ∆ t . . . . . . p , (∆ t ) Figure 9: The plots of probability distributions p S, ∆ (∆ t ) numerically computed for the inputstate (14) with a constant total number of S i = 6, r = 10%, S = 2, ∆ = 0 (left column)and ∆ = 2 (right column). The results were obtained for both perfect (a & b) and imperfectphotodetection, with binomial (c & d) and Gaussian (e & f) distribution d K ( K ′ ) ( d L ( L ′ )),representing the detector characteristics. Black and red curves were obtained for differentvalues of parameters η and σ . Detailed description is presented in the main text. k − l = ∆ t = k l p , ( k, l ) b) k − l = ∆ t = k l p , ( k, l ) c) k − l = ∆ t = k l p , ( k, l ) d) k − l = ∆ t = k l p , ( k, l ) e) k − l = ∆ t = k l p , ( k, l ) f) k − l = ∆ t = k l p , ( k, l ) Figure 10: The plots of probability distributions p S, ∆ (∆ t ) numerically computed for the inputstate (16) with a total number of photons in range S i ∈ [80 , r = 10%, S = 10, ∆ = 0(left column) and ∆ = 10 (right column). The results were obtained for both perfect (a &b) and imperfect photodetection, with binomial (c & d) and Gaussian (e & f) distribution d K ( K ′ ) ( d L ( L ′ )), representing the detector characteristics. Detailed description is presentedin the main text. k − l = ∆ t = k l p , ( k, l ) b) k − l = ∆ t = k l p , ( k, l ) c) k − l = ∆ t = k l p , ( k, l ) d) k − l = ∆ t = k l p , ( k, l ) e) k − l = ∆ t = k l p , ( k, l ) f) k − l = ∆ t = k l p , ( k, l ) Figure 11: The plots of probability distributions p S, ∆ (∆ t ) numerically computed for the inputstate (16) with a total number of photons in range S i ∈ [4 , r = 10%, S = 2, ∆ = 0 (leftcolumn) and ∆ = 2 (right column). The results were obtained for both perfect (a & b) andimperfect photodetection, with binomial (c & d) and Gaussian (e & f) distribution d K ( K ′ )( d L ( L ′ )), representing the detector characteristics. Detailed description is presented in themain text.
100 200 300 400 S i . . . . . . γ b)
100 200 300 400 S i . . . . . . γ Figure 12: The plots of the purity of the state | ψ t i , for the input state (14) with a totalnumber of photons in range S i ∈ [10 , r = 10%, ∆ = 0, numerically computed for S = 20(left figure) and S = r · S i (right figure). Black curves correspond to the imperfect detectionmodeled by a binomial distribution with η = 5% , ,
15% (solid, dashed, dotted), red curves– Gaussian distribution with 3 σ = 5 , ,
20 (solid, dashed, dotted). Detailed description ispresented in the main text. a)
100 200 300 400 S i . . . . . . γ b)
100 200 300 400 S i . . . . . . γ Figure 13: The plots of the purity of the state | ψ t i , for the input state (16) with numberof photons in the range S i = 0 . S i , S i = 1 . S i , where S i ∈ [10 , r = 10%, ∆ = 0,numerically computed for S = 20 (left figure) and S = r · S i (right figure). Black curves corre-spond to the imperfect detection modeled by a binomial distribution with η = 5% , , σ = 5 , ,
20 (solid, dashed,dotted). Detailed description is presented in the main text. . Conclusions In this paper we have examined experimental scheme which implements afamily of quantum non-Gaussian filters. The same setup may apply arbitraryfiltering condition C( S t , | ∆ t | ) which is set by a relation between the total photonnumber S t and the modulus of mode population difference | ∆ t | in the outputstate.Direct applications for some of these filters are already known. It hasbeen shown that filtering according to the condition C( S t , | ∆ t | ) ≡ {| ∆ t | ≥ threshold } allows for generation of states useful for quantum optical phase es-timation [16, 20]. Moreover, this filter helps to increase the distinguishabil-ity of macroscopic qubit in analog detection [14]. It also allows for increasingthe CHSH-Bell inequality violation by a micro-macro singlet state produced bythe phase-covariant quantum cloning [16]. On the other hand, the conditionC( S t , | ∆ t | ) ≡ { S t ≥ threshold } allows to increase the generation efficiency ofthese states [21]. The quantum tasks using the other filtering conditions are notyet known.All filters work for an arbitrary two-mode input state, pure or mixed, with asmall (few photon) or large (mesoscopic) population. We have demonstrated thecoherent action of the filter in presence of realistic photodetection involved inthe filtering process. The imperfect detection was modeled with a beam splitterof a small reflectivity located in front of a perfect detector and a Weierstrasstransform, which implements a Gaussian blur on the unaffected measurementoutcome distribution. We have constructed the operators describing the setupof the filter. We have also presented computations for two exemplary initialquantum superpositions, which reveal the structure of the filtered states in thephoton number space as well as estimate its purity.We believe that the scheme we have discussed will be useful for preparationof the available quantum superpositions for further quantum tasks requiringmore complex quantum states than the Gaussian ones. Acknowledgments
This work is supported by the EU 7FP Marie Curie Career Integration GrantNo. 322150 “QCAT”, NCN grant No. 2012/04/M/ST2/00789, FNP HomingPlus project No. HOMING PLUS/2012-5/12, MNiSW co-financed internationalproject No. 2586/7.PR/2012/2 and EU 7FP project BRISQ2 No. 308803.Computations were carried out at the CI TASK in Gda´nsk and Cyfronet inKrak´ow.