PHYSICAL REVIEW A VOLUME 50, NUMBER 5 NOVEMBER 1994 Detection of the density matrix through optical homodyne tomography without filtered back projection G.M. D'Ariano" Dipartimento di Fisica “Alessandro Volta”, Universit degli Studi di Pavia, Via A. Bassi 6, 1-27100 Pavia, Italy ‘and Istituto Nazionale di Fisica Nucleare, Sezione di Pavia, via A. Bassi 6, 1-27100 Pavia, Italy C. Macchiavellot and M. G. A. Parist Dipartimento di Fisica “Alessandro Volta", Universita degli Studi di Pavia, via A. Bessi 6, [-27100 Pavia, Italy (Received 9 March 1994) We propose an alternative method for detecting the density matrix of a radiation field via optical homodyne tomography. The method needs no assumption on the state, and the density matrix is recovered directly from averages on data, The tomographic reconstruction is very fast and provides very reliable statistics on line with detection, Ara test of the method we present numerical results for set of computer-simulated experiments, corresponding to different quantum states of the radiation source. Then a tomographic reconstruction is given for experimental data provided by Smithey et fl [Phys. Rev. Lett. 70, 1244 (1993)) PACS mumber(s): 42.50.Dv, 03.65.Be, 42.65.Ky Ina recent paper Smithey et al {1} developed a method for detecting the quantum mechanical state of the field the so-called optical homodyne tomography. This opened new perspectives in providing complete experi- ‘mental characterization of nonclassical states of radia~ tion, making all information about the quantum source directly detectable (2). ‘A homodyne tomography of a single field mode a con- sists of an ensemble of repeated measurements of the quadratures dy = H(ae~* + ale) for various phases @ relative to the local oscillator of the homodyne detec- tor. Smithey et al. reconstruct the density matrix via the Wigner function W(a,@) of the field, namely, the quasiprobability for symmetrically ordered field opera- tors. In this way their method needs a filtered backpro- jection of the Wigner function, setting the resolution with, which W(a,@) is determined. In some sense the coarse- raining cutoff corresponds to an a priori hypothesis on the detected state, and hence it may ultimately affect the result and the error statistics of the tomographic re- construction (the cutoff should be tuned carefully as a function of the total number of measurements in the to- mographic scanning). ‘The purpose of this paper is to show that there is no need for backprojection filtering 1 (nlaim) = Pam = where Hy (2) is the Hermite polynomial of degree n. Notice that Eq. (1) allows evaluation of the Wigner function only when the probability density p(z,) is “Blectronic address: dariano@py.infn.j¢ 'Blectronie address: macchiavelloOpy infa.it ‘Electronic address: parisOpy.infu.it 1050-2947/94/50(8)/4298(51/$06.00 's the data, whereas the density matrix can be directly re- covered from averages on experimental data. This leads to an improved method of tomographic reconstruction of the density matrix, which greatly speeds up data anal- ysis, and achieves very reliable error statistics on line with detection. Before presenting our method, here in the following we briefly recall the procedure of Ref. [1] for comparison. From the probability distributions p(x, ¢) of the out comes 2 of ay, the Wigner function W(a,@) is recon- structed through the formula [3] & [ani [a [one x exp {in [2 Re (ae*)]} a ‘The density matrix is then obtained using the Fourier transform Wa, (e+2'fe—2) = [Laveewee sine) @) In the number representation the change of basis is ex- pressed by the double integral f de [ote 8O tale tn 2) ale)» (3) SS given in analytical form. In fact, any approximation of the average over p(z, ¢)dzd@ in terms of a finite sum over cxperimental data i meaningless, because the integral ver n doesnot converge for any datum (e,¢)- In order to muke the method analytic, s smoothing procedure is Fequied: in Ref [1 this is accomplished by means of fl tered backprojection techniques, which are standard in tomographic imaging [4 4298 ©1994 The American Physical Societyry 40 60 80 100 Relative frequency 20 0.08 0 0.05 833 FIG. 1. Distribution of the tomographic outcomes for the ‘matrix clement pss (deviations éps,s around the average). ‘The computer-simulated experimental values are obtained for ‘a coherent state with (n) = 3. ‘The histogram contains 5000 experiments (eubensembles of data), each performed with P= 27 scanning phases and 100 measurements for each phase. ‘The Gaussian curve with variance equal to the ex- perimental rms error is superimposed. Let us consider now a completely different procedure. ‘The transformation (1) can be written more generally for any “s-ordering” quasiprobability distribution W, (a, 4) 8) Walesa) = ay fail fae [do 012.0) : Re(ae)}} ® For s < Othe integral over n converges for any (7,4), and thus commutes with the double integral over z,¢. This xem {jr inke SSS Pons = zs [E820 [7 vl2.0) 28 (m— where RD (6) is the fixed matrix DETECTION OF THE DENSITY MATRIX THROUGH OPTICAL ... 299 1.0005, 1 we “4000. 6000 ™@) 0.9995 0.999 1000 2000 3000 Number of experiments FIG. 2. Normalization ofthe density matrix versus number of experiments (subensembles of data) for coherent state with (n) = 8, Bach experiment has been performed asin Fig. 1. The matrix is truncated at n= 20 photons. allows averaging on p(z,¢) in terms of statistical sums ‘over finite sets of experimental data. For s = -1 the ‘quasiprobability W-s(a, a) coincides with the Husimi Q function Q(,4) = (alla) = Wa(a,4) , 6) which itself generates the matrix elements yim in the ‘umber representation according to the formula Deon Onn tO) ((a.a) €*")| ives in Eq. (6), with Q(a,a) given by Eqs. (4) and (5), can be evaluated analytically. After a lengthy calculation one obtains a 9 (d+); + (ate!) : (7 - 124) OE ses SA aya (2) (4 ! : RQ(6) = (2%) Faas (5) (4) ert tin aiei tint nfl) StF) hth y2n+d—23 at 21 +52 1 (cog) 25-5 (cin SY (8) mma aed asamp xt (me pao bat pda +) lm t)- ® In Eq. (8), (2)2 denotes the rest of the divisi (a.,6;2) is the confluent hypergeometric function of z with parameters a6 [5]. For F equally spaced phases ¢, 2/2, fa] is the integer part of z, 6(n) is the Kronecker delta, and e4300 ‘TABLE I. First few detected matrix elements of G.M. D'ARIANO, C. MACCHIAVELLO, AND M. G. A. PARIS squeeaed state with (n) fand one squeesing photon (the matrix is real the theoretical values are given in parentheses), ‘The detected values are obtained for 120 experiments with F ~ 27 scanning phases and 120 measurements each | o 1 2 3 3 a] W276 0.0013] ——o.153830.0012| —O.2BATO0OTS)——W.T396-L0.00I6] o2006+0.0019, —O.1T35L0.0019 (0.5278) (0.1547) (0.2957) | __(0.1393) (0.2015) _ (0.1145) i] 0isssz0.0012] —on4s20.0091, —O.08480.0014) o.0s18-E0.0017| .0s68:E0.0016| 0.0336:20.0015 (0.1547) (0.0454) (0.0867) (0.0409) (0.0591) (0.0336) 2} o20a7s0.0018] —Osasz0.0014] —0-1657=0.0021, o.0790-0.0013] —o.i1230.0016) 0.0860:50.0017 (0.2987) (0.0867) (0.1058) (0.0781) (0.1130) (0.0942) 3] 0:1396:0.0018 | —Oosi8s0.00i7) 0.079940.0013) .a38T-E0.0020 —dOs19=0.001g} —OsIs=0.0NT6 (0.1393), (0.0409) (0.0781) (0.0368) (0.0552) (0.0303) a] W2000z0.0019] —Oasess0.on16| —0.112340.0016) o.0si9:-0.001a] o.07s7-E0.0023) — 0078 (0.2015) (0.0591) (0.1130) (0.0532) (00770) (0.0437) 5] :i1s30.0018] —003500.00151 —0-06600.0077) .0313-0.0016] o.bado-t0.0015] 00271-+0.0034 | (oss) | (0.0396) (0.0042) 0.0503) | (0.0437) (0.0249) average: (edd ra where {)4y denotes averaging over the subsammble of data for ited phase 4 = dy. Equation (9) is particularly suited to on-line data analysis; in fact, apart fom the tum over data, the procedure requires just single sum over m, whereas the hypergeometric functions (a, 3:2) are connected to each other iteratively, and the matrix R()(4,) is stored in the machine before beginning ex- periments. 100 Relative frequency 40 60 20 NTH -2 uh 2 am FIG. 3. Distribution of normalized deviations from the the- oretical values Spam = (Bnm Pr) /Emm for the frst 30 x 80 matrix elements. ‘The quantum state is a coherent fone with (n) = 4. The histogram pertains to 1000 experi- ‘ments (subensembles of data) with F = 27 scanning phases ach, and 200 measurements for each phase. A standardized (unit variance) Gaussian curve is superimposed. wet Ye BERR (eee (mon 0,..,F = 1) the experimental mean value Pani of the matrix element pps is obtained from the following —Hasin +0122) o — We test our method on a set of Monte Carlo simulated experiments, with the aim of evaluating both the cor- rectness of error statistics and the absence of systematic deviations from theoretical values. In order to evaluate the statistical errors, we have pre- liminarily studied the distribution of the tomographic ‘outcomes for each matrix element around its averaged value, checking that it is perfectly Gaussian for every ma- trix element, independent of the kind of considered quan- tum state. We have considered coherent, squeezed, and general quantum superpositions of number states (within this paper F = 27, with at least 100 measurements for each phase; the resolution F = 27 is used for comparison with Ref. (1), but one can get satisfactory results already for F = 12). Due to the normal distribution of deviations Sons the erFOr Em of the matrix element pay can be evaluated as usual, namely, by dividing the ensemble of data into subensembles (here also called “experiments”), and then calculating the rms deviation of the subensem- ble average with respect to the global average. In Fig. 1, we report a typical distribution of outcomes for a fixed ‘matrix element, with the source in a coherent state: the agreement with the Gaussian probability is remarkable ‘As regards the reliability of detection of the whole ‘matrix, two significant tests have been performed: (i) the matrix normalization, and (ji) the x? test. In Fig. 2, a sample of the normalization is given ver sus the number of data subensembles, for a highly ex- cited coherent state (less excited states—coherent, or not—exhibit a more rapid convergence). The distribu- tion of normalized deviations from the theoretical val- ues Apnm = (Brn ~ Prin) /Enm follows a standardized Gaussian curve. A’ sample histogram for the fist 30 x 30 matrix elements is given in Fig, 3, for a coherent state with (n) =4. Notice that about 68% of the deviations