Arxiv:Quant-Ph/0606221V2 25 May 2007

arXiv:quant-ph/0606221v2 25 May 2007 pt h esneglmt[,3] [1, limit sensitivities Heisenberg with the estimations to phase up interfer- allowing revolutionize by to ometry potential the has entanglement Mechanics. Quantum not by is imposed this Yet, limit ultimate process. the interferometric the in employed ilsi mode in ticles where h state The ions): trapped Schrödinger cats of called context the (often in 9] 12] 8, 11, realization [7, [10, experimental states the NOON to litera- of directed ef- several been of recently, have Quite body forts 6]. sub large [5, for strategies performances a and shot-noise states inaugurating in quantum new initiated [4], proposing ture were 80s direction early this the along works Pioneering ihsniiiybuddb h tnadqatmlimit quantum standard [2] the (shot-noise) by bounded estimations sensitivity phase allows with Interferom- particles uncorrelated [1]. with bound etry lower re- Cramer-Rao uncertainty the Heisenberg reduced and by lations be by imposed cannot limits provided and the is Mechanics, beyond uncertainty Quantum of of laws source arbitrarily the second be A can principle, reduced. de- in which, poor micro-seismic efficiencies by fluctuations, tection created temperature be activities, can sources geological noise different quite Classical two noise. esti- by limited precise of is The phases waves. as of gravitational cosmic-scales, mation of to detection in forces, the as Casimir in micro-scales, of from measurement spanning the per- domains external the in some by with turbation generated interferometer shifts the phase technologies. of estimate new interaction to and is goal science main basic Its of development the ntels e er,i a eoecerta quantum that clear become has it years, few last the In Introduction. N T | N, stettl(raeae ubro particles of number average) (or total the is | N-NMBCCne n iatmnod iia-Universit` - Fisica di Dipartimento and Center BEC CNR-INFM unu ttscnann pto up containing exper retrieved states probabilities quantum published using limit noise prahs edmntaetepsiiiyt banaphas a obtain to possibility the demonstrate We approaches. iiysaiga ∆ as scaling tivity among − Ψ 0 b π i nefrmtywt ONqatmsae a rvd unbias provide can states quantum NOON with Interferometry N ( contains i ieversa vice ≤ = nefrmtypasacnrlrl in role central a plays Interferometry p θ ∆Θ √ ∆Θ needn esrmnsadoecmsbscdffiute p difficulties basic overcomes and measurements independent ≤ 1 2 π sn N HL | h rtclrqie oa of total a requires protocol The . u htNieitreoercpaesensitivity phase interferometric Shot-Noise Sub N, h state the = atce nmode in particles = 0 ihBrlimin croigrCtStates Schrödinger Cat ions Beryllium with θ N i N + ∼ T 1 1 1 / T e 1 2 iφ . /N , | | 0 0 T N , N , ie ro nweg httetu hs shift phase true the that knowledge prior a given i uaPz´ n uut Smerzi Augusto and Pezzé Luca i ,while ), . N a eylu ions. beryllium 6 = n par- 0 and φ san is (1) (3) (2) N eoeec n lsia os n i)mxmmpriori maximum shift: phase iii) the and about ignorance noise classical incorporating and experimentally results decoherence implemented measurement be the can of which analysis Bayesian with rigorous shifts a phase of estimation strat- uncertainty measurement unambiguous a the i) propose for we egy Letter splitter this In beam 50/50 been 15]. a has [14, through one (3) passed state and of the phase operator in after shifted obtained parity modes when the output discussed) of the of of value generally structure mean not the if multipeak measuring (even the present instance, with is For technology Eq.(4) enhanced Eq.(3). quantum in state typical is Eq.(4) opeepirinrne h 2 The ignorance. prior complete n osqecswe rigt siaeteunknown the estimate to of value trying when consequences determine to ant sufficient not is alone nality θ et efre ihNO ttshvn different a having particles, states of NOON number with The performed prior). ments smaller arbitrary an requires of protocol case the consider also scluae safnto ftettlnme fparticles of process, number the total the in of used function a as calculated is q()i miuu.Indeed, measurement projective ambiguous. the is the on Eq.(4) and based unknown is estimation incremental small, the phase supposedly interferometry albeit in shift, phase though: problem, a is hr h eeao fteukonpaetranslation phase unknown tor, the of θ generator the where | bepaesiti fteodro 1 of order the of is shift phase able ± ed scetdb h ntr operator unitary the by perturbing The classical created external example. is an simple field, by a induced by shift, This accompanied phase Eq.(2)[13]. often limit is Heisenberg shifts claim the phase unknown at estimate sensitivity can with interferometry Eq.(3) that state believed the widely with is It phase. arbitrary Orthogonality, h nta n gives one initial the Ψ T = π/N stetomd eaienme fprilsopera- particles of number relative two-mode the is N mnal o h raino Schrödinger cat of creation the for imentally 2 = ( ± J ˆ θ z (2 ) hc ol ugs httesals measur- smallest the that suggest would which , i p ( = n θ estvt etn h lsia shot- classical the beating sensitivity e − = 1) + iha rirr ag ubro atce and particles of number large arbitrary an with N atce ueuly distributed (unequally) particles 1 U ˆ ˆ ∼ a iTet,I300Pv,Italy Povo, I-38050 Trento, di a dpaeetmto ihasensi- a with estimation phase ed θ |h π/N | − 1 Ψ Ψ h /N Ψ N N N eeti rvosyproposed previously in resent ˆ p N i | with , b T Ψ needn nefrmti measure- interferometric independent ) | ( = / Ψ N yuigtesaeE.3,wti ii) within Eq.(3), state the using by N N .Tepoeto ftenwstate new the of projection The 2. N T ( θ 1 = | ( N, 2 = ) θ n θ i| ) i 2 0 = isi h interval the in lies , 0 p 2 i ,i rtrahdat reached first is 0, = cos = , − + 4 , ..., , 1 π/N 1. e h − , Ψ i 2 ( 2 π θN N N , . . . , ( 2 Nθ/ ≤ p | silto eidof period oscillation /N + Ψ − φ θ 1 N ) h sensitivity The . ≤ 2) | n ( 0 swl.There well. as θ ihunpleas- with , N , − . ) π U ˆ i θ .Orthogo- 1. btw will we (but i when 0 = ) = / √ e over 2 − i θ J ˆ (4) z = θ , 2

0.4 0.4 iθJˆ A C by applying the operator e− z . The ﬁnal state, ob- 0.3 0.3 tained after a further application of UˆN , is projected over 0.2 0.2 N . Quantum Mechanics provides the probability to | ↑i 0.1 0.1 have N at the output (which will be denoted as a | ↑i 0 0 “yes” result), given the unknown phase shift θ and the 3 1.5 number of particles, N, of the cat state: P (yes N,θ) = B D ˆ | ˆ iθJz ˆ 2 2 Nθ 2 1 N UN e− UN N = cos 2 . Notice that

phase distribution |h↑ | | ↓i| this probability is identical to Eq.(4). The probability 1 0.5 to obtain a “no” result is, obviously, P (no N,θ)=1 | − 0 0 P (yes N,θ). A single interferometric experiment consists −1 −0.5 0 0.5 1 −1−0.5 0 0.5 1 | φ / π φ / π of p independent measurements. We collect a number py of “yes” and pn = p py of “no” results with probability P (p ,p N ,θ), N −being the total number of particles p y n| T T FIG. 1: Phase probability distribution Eq.(6) with p = 1, used in the p measurements. According to the Bayes the- NT = 15 (A), and Eq.(9) with p = 4, NT = 15 (B). In orem [6, 19], we have Pp(φ NT ,py,pn)P (NT ,py,pn) = P (p ,p N , φ)P (φ), where| P (φ) is ﬁxed by the prior both cases, the total number of particles is the same, but the p y n| T distribution of B gives a phase sensitivity at the Heisenberg knowledge and P (NT ,py,pn) provides the normalization 2 k limit. In C we plot the terms cos (2 φ/2) of Eq.(9). The solid of Pp(φ NT ,py,pn), which is a phase probability distri- blue line is for k = 0, the dashed red line for k = 1 and the bution.| We have dot-dashed green line for k = 2. By multiplying these three distributions, as in Eq.(9), all peaks, except the one centered py +pn around the true value of the phase shift, θ = 0, are washed P (φ N ,p ,p ) P (φ N , r ), (5) p | T y n ≈ | j j out to give the phase distribution D (NT = 7 and p = 3). jY=1

where N = py +pn N and r yes (no) if, in the T j=1 j j ≡ From the experimental point of view, the demonstra- j th measurement,P done with Nj particles, we obtain − tion of the Heisenberg limit Eq.(2) requires the creation a “yes” (“no”) result. Equation(5) contains all the avail- of Schrödinger cat states with a minimum of N = 8 parti- able information needed to estimate θ. We can choose, as cles, which is within the reach of the present state-of-the- estimator Θest, the maximum of the phase distribution, art. Cat states with up to 6 ions [12] and 5 photons [9] and, as uncertainty ∆Θ, the 68%-confidence interval [6], have been recently reported. As far as realistic techno- namely the phase interval containing 68.27% probabil- Θest+∆Θ logical applications are concerned, however, Heisenberg ity given by dφ Pp(φ NT ,py,pn)=0.6827. In Θest ∆Θ | limited interferometry with NOON states Eq.(3) would the following,R for− analytical simplicity, we will consider, likely never overcome the performances of classical in- unless explicitly specified, the case in which the mea- terferometry Eq.(1), where the typical number of parti- surement results are only “yes”: py = p, pn = 0. That cles NT can be several orders of magnitude larger. We happens with certainty at θ = 0. The extension to an ar- therefore extend the previous protocol to reach sub shot- bitrary value of θ, where both “yes” and “no” are acces- p noise sensitivity ∆Θssn/∆Θsn 1/√2 1, which can sible, will be discussed in [20], also including decoherence ∼ − be implemented experimentally with an arbitrarily large effects. number of particles. We address the possibility to reach 1/NT periodicity of the Bayesian phase distribu- over 3 db sub shot-noise in realistic ion and photon ex- tion. Let us first analyze the interferometric experiment periments within the current technology [8, 9, 10, 11, 12], with a state ΨNT of NT particles and a single measure- both in presence of a strong priori constraint and in the ment: p = 1.| Thei phase distribution becomes more general case of a complete prior ignorance. Our 2 results do not only apply to ultra-sensitive interferom- P1(φ NT ,py =1,pn = 0) cos (NT φ/2). (6) etry, but naturally extend to quantum positioning [16], | ∝ clock synchronization [17], frequency standards [14] and This probability contains NT peaks separated by a dis- quantum metrology [1]. tance 2π/NT , see Fig.(1,A). Hence, our best guess about 2π √2 Bayesian analysis with Schrödinger Cats. In the the real phase shift θ is Θest = n with n = NT ± NT following, we discuss the Bayesian phase estimation strat- 0, 1,...,NT 1, where the error √2/NT is the mean egy considering the Schrödinger cat states realized with square fluctuation− around a single peak. In practice, we trapped ions by Wineland and collaborators [10, 11, 12]. do not have any alternative but to choose, as phase esti- ξ+N+1 The state ΨN = ( N + i N )/√2 was cre- mator, one of the N equivalent peaks of the distribution | i | ↓i | ↑i T ated by applying the “nonlinear beam splitter” operator [cf. discussion after Eq.(4)]. In this case, however, the πξ π 2 i Jˆx i Jˆ UˆN = e 2 e 2 x to N spin-down ions N interferometric experiment does not give any substan- ... , with ξ = 0 when N is even, and| ξ =↓i 1 ≡ when | ↓ tial improvement in phase sensitivity. Tautologically, the i1 | ↓iN N is odd [18]. The ΨN state is then shifted in phase phase uncertainty would scale with the total number of by an unknown quantity| iθ (which has to be determined) atoms as 1/N only if we have the a priori knowledge ∼ T 3 that the phase lies in an interval of width 2π/N around p = 4, N = 15, with a prior π θ π. Again, the T T − ≤ ≤ the real value. The NT -peaks structure of Fig.(1,A) does width of the distribution can be simply calculated with not allow the Heisenberg limit, even in the case of arbi- 2 2kφ a Gaussian approximation of each cos 2 term, giv- 2 trary small incremental phase shifts. How is, therefore, φ (2p+1)(2p 1) ing P (φ N ,p = p,p = 0) exp[ − ]. In possible to to select the “right” peak, so to speak, in p T y n 4 3 the limit| of large p we obtain≃ a phase− sensitivity at the order to build an unambiguous phase estimator? Heisenberg limit ∆Θ = √6/N . The numerical calcula- The ∆θ N 3/4 limit. In this section we consider T − tion gives, asymptotically in the number of measurements multiple independent∼ measurements with different num- p, ber of particles. The first measurement is done with a cat state of a single particle, N = 1 (with a prior 2.55 ∆Θ = (10) knowledge of the phase shift in the interval [ π, π]). We N then perform a second measurement with N− = 2, and T we multiply the resulting distribution with the previous for a 68% confidence [21]. We therefore conclude that it one. We repeat this procedure p times, increasing, in is possible to obtain an unbiased phase estimation at the each shot, the number of particles in an arithmetic se- Heisenberg limit, with repeated measurements and com- quence N =1, 2, 3, ..., p. The total number of particles is plete prior ignorance. The trick is to carefully choose the NT = p(p + 1)/2, and the phase distribution is number of particles in each measurement and to multi- p ply the corresponding phase probabilities so to cancel out 2 kφ the extra peaks of the phase distribution, see Figs.(1). Pp φ NT ,py = p,pn =0 cos . (7) | ∝ 2 To clarify this effect, let us consider a phase distribution kY=1 obtained with 2k particles: cos2(2kφ/2). This has max- In the limit of large p, a Gaussian approximation k k k 2 ima in φmax(n)=2πn/2 , with n =0, 1, ..., (2 1). p 2 φ p(p+1)(2p+1) ± ± −k 1 gives k=1 cos (kφ) exp 4 6 , where Conversely, the phase distribution obtained with 2 − p 2 ≈ − k 1 k k=1 Qk = p(p + 1)(2p + 1)/6. For p 1 we have particles has minima in φmin− (n)=2πn/2 with n = ≫ k 1 PN p2/2, and ∆Θ = (9/2)1/4/N 3/4, which is in good 1, 3, ..., (2 − 1). By multiplying the two distribu- T ∼ T ± ± ± − k agreement with the numerical calculation: tions, the maxima φmax(n) superimpose with the min- k 1 ima φmin− (n), for n = 1, 3, .... When taking into 1.44 ± ± k 2 k 3 ∆Θ = . (8) account also the distributions with 2 − , 2 − , ... parti- 3/4 k NT cles, we obtain the cancellation of all peaks φmax(n) with k k n = 1, ..., (2 1) except the central one, φmax(n = 0), This argument can be generalized to the case of any prior which± is enhanced± − and has a width scaling as 1/2k phase knowledge [ π/L,π/L], with L 1. The goal ∼ ∼ − ≥ 1/NT , see Figs.(1, C,D). This shows that the protocol em- is to obtain an unbiased estimate of θ within a region ploying N = 1, 2, 4, ..., 2p particles is optimal and gives 2π/L, the peaks outside this region being wiped out but the Heisenberg limit with the best prefactor. Indeed, let the prior knowledge. As before, the protocol requires p us consider a general sequence N =1, r, r2, ..., rp 1, with ˜ − measurements: the first one with N L particles, the an arbitrary integer r. If r = 1, we recover the shot- ˜ ≈ ˜ second one with 2N, ..., the last one with pN. The to- noise limit Eq.(1), for r = 2, we have, as already dis- ˜ tal number of particles is Np (p + 1)/2. In the limit of cussed, a perfect superposition of maxima and minima. large p, we obtain an unbiased phase estimate with a For r> 2, the phase distribution is characterized by sev- 1.44 1 sensitivity ∆Θ = 3 4 . With an arbitrary value of N˜ 1/4 / eral, strongly weighted, peaks which increase the phase NT the true phase shift θ = 0, the same protocol provides uncertainty. To overcome this problem we must repeat a distribution peaked about6 θ with a sensitivity scaling M > 1 times each interferometric measurement with a as 1/N 3/4 for N 1. This because the cos and NOON state of fixed N before increasing the number of ∼ T T ≫ p 1 k sin distributions corresponding to “yes” and “no” results particles (employing a total NT = M k−=0 r ). For a overlap out of phase and cancel out each other except sufficient large M, it is possible to recoverP the Heisen- in a region around θ. Yet, even if the scaling 1/N 3/4 berg limit Eq.(2), but at the price of a higher prefactor. ∼ T In fact, we have that ∆θ √M/N , as a consequence overcomes the shot-noise Eq.(1), still this is not the fun- ∼ T damental limit. of the statistics of independent measurements. In Fig.(2) 1 we analyzed the sensitivity obtained with different r as a The Heisenberg ∆θ N − limit. Let us now consider p independent measurements∼ done with a geometric function of M: for r = 2 (black circles) the best perfor- p 1 mance is obtained at M = 1, for r = 3 (blue points) at sequence of particles, N = 1, 2, 4, 8, ...2 − . The phase distribution becomes M = 4, for r = 4 (red diamonds) at M = 6 and for r =5 (green squares) at M = 9. p 1 − 2kφ As mentioned before, the analysis can be extended to P (φ N ,p = p,p = 0) cos2 , (9) p | T y n ∝ 2 estimate an arbitrary unknown phase shift θ = 0. How- kY=0 3/4 6 ever, in contrast to the ∆θ NT− case, here we have where the total number of particles employed in the com- to consider multiple repeated∼ measurements in order to plete process is N = 2p 1. Fig.(1,B) shows the case concentrate the probability in a single peak, even for the T − 4

8 2.5 10

7 2 AB 8

6 1.5 6 1 4 Τ 5 Ν ∆Θ 0.5 2 4

phase distribution 0 0 3 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 φ / π φ / π 2 1 2 3 4 5 6 7 8 9 10 M FIG. 3: Phase probability distributions obtained by combin- ing M times (in (A) M = 1, in (B) M = 10) the Bayesian 2 1 FIG. 2: Phase sensitivity obtained with N = 1,r,r , ..., rp− , distributions for N = 1, 2, 3, 4, 5, 6 particles and θ = π/2. where each measurement (using N = rk particles) is repeated The dashed red line is the ideal case, while the solid blue line M times. The best strategy is obtained when r = 2 (black has been obtained with the experimental fidelities, see [25]. circles) and M = 1, due to a perfect superposition of maxima and minima in the Bayesian distributions, see Fig.(1,D). Notice that the optimal value of M, as well as the phase uncertainty, increases with r: here we consider r = 3 (blue points), r = 4 (red diamonds) and r = 5 (green squares). Lines are decrease the fringes contrast CN ,N , which, in the ideal guide to the eye. case, is equal to one. The experimental↑ ↓ Bayesian phase probability distribution associated to a detection of a “yes” or “no” result, P (φ N, yes, no ), is obtained in- exp | { } case r = 2. While the scaling 1/N is preserved [20], verting P ( yes, no N,θ). We are now ready to simulate T { }| Eq.(10) gives a lower bound of∼ phase sensitivity. a realistic phase estimation experiment with Beryllium Sub shot-noise with Beryllium ions. The experi- ions: i) A “yes” or “no” result is chosen with probability mental demonstration of the Heisenberg limit would re- P ( yes, no N,θ) with an unknown, but fixed, value of θ { }| quire the creation of Schrödinger cats having N = 8 par- and for different number of ions, N = 1, 2, 3, 4, 5, 6; ticles. The biggest Schrödinger cat created experimen- ii) We repeat these measurements M times; iii) We cal- tally so far is with N = 6 9Be+ ions. This is still suffi- culate the Bayesian distribution associated with each ciently large to reach a sub shot-noise phase sensitivity “yes”/“no” result and values of N; iv) We multiply all [22]. In the following, we demonstrate, using the fidelities Bayesian distributions obtained in iii). This provides the measured experimentally in [10, 11, 12] with Beryllium final phase probability, from which we retrieve the esti- ions, the possibility to reach a phase sensitivity gain of mated value of the phase shift and its confidence. The to- 0.8 db with respect to shot-noise for a priori π θ π. tal number of particles used in this process is NT = NpM, − ≤ ≤ 6 The protocol is quite similar to the one discussed above with Np = k=1 k = 21. Ideally, with N =1, 2, ..., p, the and requires Bayesian probabilities calculated with dif- phase sensitivityP would scale as ferent number of ions combined with M replica of the measurement process. We remark, however, that, in the analysis, we should now replace the ideal probabil- (9/8)1/4 (9/8)1/4 ity distributions with those retrieved experimentally. In- ∆Θ = = , (11) 3/4 1/2 1/4 1/2 deed, we need to include the effects of noise and deco- Np M Np NT herence present in the experiments which will inevitably decrease the sensitivity of the interferometer. This step can be considered as a “calibration”: once the experimen- where Np = p(p + 1)/2. In Fig.(3) we plot the theo- tal probabilities are retrieved and inverted with Bayes, retical and experimental phase probabilities for M = 1 the interferometer is ready for its use. Many sources and M = 10. Notice that the experimental distribution of experimental imperfections and decoherence conspire for M = 1 is characterized by a large tail. This arises against phase uncertainty enhancement with cat states. from the reduced fringe visibility and it would strongly Laser intensity fluctuations and magnetic field noise have dilute the confidence of the phase estimation. On the been discussed in [12]. Imperfections in the creation of other hand, by multiplying the M = 10 distributions, the state (3) due to non ideal UˆN and decoherence have step ii), we strongly decrease the weight of the tail with the common effect to decrease the oscillation amplitude respect to the central peak, see Fig.(3,B), at the price, of of P (yes N,θ) [24]. A fit of the experimental probabil- course, to increase the shot-noise. It is worth to empha- | ities with Pexp(yes N,θ) = A + CN ,N /2 cos(Nθ) and size, however, that, because of noise and decoherence, at P (no N,θ) = 1| P (yes N,θ)↑ has↓ been given in the end of the day we could, in principle, get an exper- exp | − exp | [10, 11, 12] with the values of A and CN ,N reported imental phase sensitivity even worse than the classical in [25]. The main effect of the experimental↑ ↓ noise is to shot noise. Asymptotically in M, the phase probability 5

gain (db) 0 In this limit, and with ideal fidelities, we would have a −0.5 phase independent gain Gth = 3.18 db with respect to the shot-noise limit ∆Θsn. Such a gain would be compa- −1 rable with the best performances obtained to date with 0 0.2 0.4 0.6 0.8 1 θ / π photons [26]. As a consequence of imperfections and decoherence, the experimental gain is lower than the ideal prediction and depends on the phase shift, see the solid FIG. 4: Gain (db) with respect to the shot-noise limit red line in Fig.(4). The maximum gain is Gexp = 0.75 obtained with the experimental parameters [25]. The blue dB, around the optimal working point θ 0.3. In princi- ∼ dashed line is the case N = 2, the green dot-dashed line ple, an even higher gain can be obtained using states with N = 3, the brown dot-dot-dashed line N = 4, the sky-blue p 1 N =1, 2, 4, 8, ..., 2 − particles: the sensitivity would be dot-dot-dot-dashed line N = 5, and the black dotted line bounded by the ideal value N = 6. Colored arrows indicate the upper bound to the phase prior for each data set. With maximum priori ignorance, we 2.55 2.55 ∆Θ = = , (13) combine M replica of the Bayesian phase distributions with 1/2 1/2 1/2 N = 1, 2, 3, 4, 5, 6, Eq.(12), (solid red line). The horizontal NpM Np N T dotted line gives the shot-noise. p with Np =2 1. So far we have− considered the case of complete prior ignorance. While this can be important for technological ture. We can expect to have, in a few years, the produc- and basic science applications like in gyroscopes or clock tion of robust high-fidelity cat states up to 10 photons synchronizations, it is sometimes possible to confine the or ions, allowing the saturation, at least in∼ principle, of priori within an interval π/L θ π/L, with L > 1. − ≤ ≤ Eq.(2). Major obstacles to these efforts are creation im- In this case an unbiased phase estimation can be obtained perfections and decoherence [24, 27]. On the other hand, with M replica of the interferometric measurement, each Bose Einstein Condensation might offer the possibility with a Schrödinger cat state of a fixed number of parti- to create NOON states with a larger number of particles N˜ L. With ideal distributions, we would obtain ≃ 1 cles [28]. A different, promising strategy to experimen- a phase-independent sensitivity ∆Θ = 1/2 , with N˜ √NT tally reach sub shot-noise sensitivities is to use number- NT = M N˜. The gain with respect to the shot-noise, squeezed states, which have been recently experimentally obtained with the experimental probabilities, is shown in demonstrated [29, 30] with up to few hundred neutral Fig.(4) where the arrows indicate the prior knowledge atoms. In any case, crucial to the success Heisenberg π/N˜ for the various cases. The gain is maximum at limited interferometry, is the realization of highly effi- θ = π/(2N˜) where the experimental probabilities to have cient number counting detectors, which can be probably a “yes” or “no” result are closer to the ideal ones. No- developed in the next generation of experiments [30, 31]. tice that the sensitivity at first increases and eventually Conclusions. Ultra-sensitive interferometry requires decreases with the number of ions. This is caused by the unbiased phase estimation protocols and carefully en- competition between the gain obtained increasing N˜ and gineered maximal quantum correlations among input the loss of visibility due to decoherence, which is higher states. In this paper we have developed a novel for larger cats. With the experimental fidelities measured Bayesian protocol based on multiple measurements with so far, the best scenario is obtained employing cat states Schrödinger cat states of variable number of particles. ˜ 1 of N = 3 particles, which provides a gain up to 1.63 db This achieves the Heisenberg sensitivity ∆Θ NT− with with respect to the shot-noise. Yet, it is clear that, with an arbitrary prior phase uncertainty. This protocol∼ over- a priori π/L θ π/L and with cat states having a comes difficulties present in previous approaches where − ≤ ≤ number of particles N˜ 2kL, with k =1, 2, 3, ..., we can, the estimate was strongly ambiguous, so to require a in principle, further decrease≃ the phase uncertainty than prior knowledge of the true value of the phase in the using cats with a fixed number of particles. restricted interval π/NT θ π/NT . We have also We remark that the total number of particles used in demonstrated the− possibility≤ to≤ obtain sub shot-noise our phase estimation protocol can be arbitrarily large phase sensitivity with the experimental data published in (being proportional to the number of replica M). This [10, 11, 12] on the creation of Beryllium ions Schrödinger is important for realistic technological applications since cat states. Our results do not depend on how the state the number of particles in Schrödinger cat states would Eq.(3) is created, nor on the specific interferometric ap- probably remain relatively small, at least in the next fu- paratus as far as the phase probability distributions have 6 an oscillating pattern with a period depending on the number of particles.

Acknowledgments

We thank John Chiaverini and Jonathan Dowling for useful discussions.

[1] Giovanetti V., Lloyd S. and Maccone L. Phys. Rev. Lett. [16] Rudolph T. and Grover L., Phys. Rev. Lett. 91, 217905 96, 010401 (2006). (2003). [2] Pezzé L., Smerzi A., Khoury G., Hodelin J.F. and [17] Giovanetti V., Lloyd S. and Maccone L., Science 306, Bouwmeester D., quant-ph/0701158. 417 (2001). [3] In this Letter we consider non interacting particles for [18] Mølmer K. and Sørensen A., Phys. Rev. Lett. 82, 1835 which Eq.(2) represents the ultimate limit [1]. Eq.(2) can (1998). be overcome with interacting particles, Luis A. Phys. [19] Zawisky M., Hasegawa Y., Rauch H., Hradil Z., Myska R. Lett. A 329, 8 (2004); Roy S.M. and Braunstein S.L., and Perina J., J. Phys. A: Mat. Gen. 31, 551 (1998); Hel- quant-ph/0607152; Boixo S., Flammia S.T., Caves C.M. strom C.W., Quantum Detection and Estimation Theory Geremia J.M. Phys. Rev. Lett. 98, 090401 (2007). Academic Press, New York (1976), cap.1. [4] Caves C.M., Phys. Rev. D 23, 1693 (1981). [20] PezzéL. and Smerzi A., unpublished. 38% [5] Yurke B., McCall S.L. and Klauder J.R., Phys. Rev. A [21] We have ∆Θ = 1.25/NT for a 38% confidence and 95% 33, 4033 (1986); Holland M.J. and Burnett K., Phys. ∆Θ = 14.79/NT for a 95% confidence. Rev. Lett. 71, 1355 (1993). [22] With two ions, a phase sensitivity ∆Θ ≥ 1/2 has been [6] PezzéL. and Smerzi A., Phys. Rev. A 73, 011801(R) experimentally demonstrated in [23] using an error prop- (2006). agation analysis and prior knowledge −π/2 ≤ θ ≤ 3π/4. [7] In this context, NOON states first appeared in Sanders [23] Meyer V., Rowe M.A., Kielpinski D., Sackett C.A., Itano B.C., Phys. Rev. A. 40, 2417 (1989), and they were in- W.M., Monroe C. and Wineland D.J., Phys. Rev. Lett. dependently introduced in Boto A.N., Kok P., Abrams 86, 5870 (2001). D.S., Braunstein S.L., Williams C.P. and Dowling J.P. [24] Huelga S.F., Macchiavello C., Pellizzari T., Ekert A.K., Phys. Rev. Lett. 63, 063407 (2001). The name “NOON” Plenio M.B. and Cirac J.I., Phys. Rev. Lett. 79, 3865 appeared in a footnote of Lee H., Kok P., Dowling J.P., (1997). J. Mod. Opt. 49, 2325 (2002). [25] The experimental parameters are A = 0.500(5), C1↑,1↓ = [8] Lamas-Linares A., Howell J.C. and Bouwmeester D., Na- 1, C2↑,2↓ = 0.955(8) [10], C3↑,3↓ = 0.84(1) [11], C4↑,4↓ = ture 412, 887 (2001); Walther P., Pan J., Aspelmeyer M., 0.698(3), C5↑,5↓ = 0.527(3) and C6↑,6↓ = 0.419(4) [12]. Ursin R., Gasparoni S. and Zeilinger A., Nature 429, 158 [26] McKanzie K., Shaddock D.A., McClelland D.E., Butch- (2004); Mitchell W.M., Lundeen J.S. and Steinberg A.M., ler B.C. and Lam P.K. Phys. Rev. Lett. 88, 231102 Nature 429, 161 (2004). (2002); and ref.s therein. [9] Zhao A., Chen Y., Zhang A., Yang T., Briegel H.J. and [27] Zurek W.H., Rev. Mod. Phys. 75, 715 (2003); André Pan J., Nature 430, 54 (2004). A., Sørensen A.S. and Lukin M.D., Phys. Rev. Lett. 92, [10] Leibfried D., DeMarco B., Mayes V., Lucas D., Bar- 230801 (2004). rett M.D., Britton J., Itano W.M., Jelenkovic B., Langer [28] Dunningham J.A. and Burnett K., J. Mod. Opt. 48, 1837 C., Rosenband T. and Wineland D.J., Nature 422, 412 (2001); Sørensen A, Duan L.M., Cirac J.I. and Zoller P., (2003). Nature 409, 63 (2001). [11] Leibfried D., Barrett M.D., Schaetz T., Britton J., [29] Orzel C., Tuchman A.K., Fenselau M.L., Yasuda M. and Chiaverini J., Itano W.M., Jost J.D., Langer C. and Kasevich M.A., Science 291, 2386 (2001); Greiner M., Wineland D.J., Science 304, 1476 (2004). Madel O., Esslinger T., Hänsch T.W. and Bloch I., Na- [12] Leibfried D., Knill E., Seidelin S., Britton J., Blakestad ture 415, 39 (2002); Jo G.B., Shin Y., Will S., Pasquini R.B., Chiaverini J., Hume D.B., Itano W.M., Jost J.D., T.A., Saba M., Ketterle W., Pritchard D.E., Vengalat- Langer C., Ozeri R., Reichle R. and Wineland D.J., Na- tore M. and Prentiss M., Phys. Rev. Lett. 98, 030407 ture 438, 639 (2005). (2007); Li W., Tuchman A.K., Chien H. and Kasevich [13] Giovanetti V., Lloyd S. and Maccone L., Science 306, M.A., Phys. Rev. Lett. 98, 040402 (2007). 1330 (2004). [30] Chuu C.S., Schreck F., Meyrath T.P., Hanssen J.L., [14] Bollinger J.J., Itano W.M., Wineland D.J. and Heinzen Price G.N. and Reizen M.G., Phys. Rev. Lett. 95, 260403 D.J., Phys. Rev. A 54, R4649 (1996); Gerry C.C. and (2005). Campos R.A., Phys. Rev. A 68, 025602 (2003). [31] Khoury G., Eisenberg H.S., Fonseca E.J.S., and [15] Notice that, with a prior ∼ 1/N, the error propagation Bouwmeester D., Phys. Rev. Lett. 96, 203601 (2006). formula predicts a sensitivity ∆θ = 1/N [14].