Arxiv:1811.01969V1 [Quant-Ph] 5 Nov 2018 Npooi Unu Esn.W Ee H Edrt De- to Reader Al

Home , Quantum illumination

arXiv:1811.01969v1 [quant-ph] 5 Nov 2018 npooi unu esn.W ee h edrt De- to reader al. the et refer We gen sensing. quantum photonic in such of detection simplicity and features. relative cer- manipulation quantum the exotic is generation, to regime thanks the photonic setting in the best scenario, the this tainly In in improvements states precision. orders-of-magnitude squeezed achieve to and order photons in single such entanglement, characteristics physicsas quantum of important leverage laws can fundamental one exploiting is sensing By of that version exception. anything quantum no The exceed classically. far done cases be can some in that provements on. so and probing readout, biological data clocks, ob- detection, atomic in- target astronomical tasks, microscopes, detection, of servations, number wave a gravitational strat- improves classical cluding any sensing of Quantum performance the egy. exceed which to parameters esti- term able physical of are broad of protocols a discrimination quantum and is those mation This ar- all advanced sensing. encompasses most quantum which the certainly information of One is quantum technologi- eas of promising. the and field fact, many the of are of matter applications a cal As pace. precedented olwn rtcl fqatmsnigta ewl dis- will we the that sensing identify the quantum for to of benchmarks protocols classical us following the allow and structure, will the goals, background general This u uis eew tr ypoiigmathematical estimation providing by parameter start quantum we in and Here ions, background trapped qubits, qubits. spin flux to related areas, photonic lrshmsb mlyn ehd fcanlsimula- channel programmability of methods employing sim- tion to by reduced schemes de- be can pler we they by how allowed fact, and protocols mechanics In (adaptive) quantum powerful most problems. the these scribe of formulation general testing hypothesis quantum hsrve ist rvd uvyo eetadvances recent of survey a provide to aims review This im- offers physics quantum technologies, most with As un- at developing today are technologies Quantum 26,27 tfn Pirandola, Stefano 3 hsi napoc ae ntoso quantum of tools on based approach an is This . eateto hsc n srnm,SaeUiest fNe of University State Astronomy, and Physics of Department 1 optrSineadYr etefrQatmTechnologies, Quantum for Centre York and Science Computer 8 4 eateto hsc,TcnclUiest fDnak Fy Denmark, of University Technical Physics, of Department o noeve fqatmsnigi non- in sensing quantum of overview an for aa-iedtcino an bet,teraoto infor of number readout a the of objects, performance faint the of boost detection to radar-like resources quantum using rmatertcladeprmna on fve.Bsdst steps. Besides next view. potential of and quantu point of experimental problems applications and emerging theoretical review a from then hypothesi we mo and hands, estimation the our parameter in discussing quantum sensing, of quantum problems behind the c tools extremely basic of the resolution on optical the and systems, physical unu esn a eoeamtr n ra ed ti gene is It field. broad and mature a become has sensing Quantum 28–31 6 n eeotto stretching teleportation and eateto ehnclEgneig I,CmrdeM 02 MA Cambridge MIT, Engineering, Mechanical of Department ,2 1, 2 eerhLbrtr fEetois I,CmrdeM 0213 MA Cambridge MIT, Electronics, of Laboratory Research hsa o Bardhan, Roy Bhaskar 16–25 5 aau 7 ihodS ,Trno 5 L,Canada 2L7, M5V Toronto, W, St Richmond 372 Xanadu, dacsi htncQatmSensing Quantum Photonic in Advances rsnigtemost the presenting , 9–15 3 33,34 oisGehring, Tobias and 5–7 1–5 . unu reading quantum experimentally. and theoretically both cuss s hr w ate,syAieadBb rb black a channel. probe quantum Bob, unknown and the formal- Alice containing input-output say box an parties, using two by where prob- ism described the be both can Mathematically, be- lems transformation states. a quantum i.e., channel, tween parameter quantum classical a a in of es- encoded values to the is discriminate and task or estimation the timate re- channel where is discrimination, quantum channel problems of quantum these terms of in discrimi- formulation spectively quantum modern (or A as testing nation). hypothesis known metrology) quantum theory, quantum and (or information estimation which quantum parameter problems quantum two in in central roots are its takes sensing Quantum sources. point-like two of separation the on depend imag- quantum quantum resolution in optical advances of recent and most ing the basis the is for quan- estimation the idea of parameter core use at Quantum the by correlations. boosted also is tum targets is faint very and It mote quantum energetic light. illumination low system) extremely of physical using equivalent states by an retrieved (or is memory optical an hl ntedsrmnto problem, discrimination the in while a edseldb sn unu erlgcldetec- metrological quantum using schemes by tion dispelled be may u channel tum a rpr h nu n o a eetteoutput. the detect problem, may estimation Bob the and In input the prepare may a nytk w values, most two the take here where only consider discrimination, may symmetric we binary case, scenario: latter basic prob- the prior For some with abilities. values of number finite and discrete ro rbblt.Ti ste qiaett retrieving to equivalent then is and cost This Bayesian probability. same the prior with hypothesis), ternative unu yohsstsigi ttevr ai of basis very the at is testing hypothesis Quantum hrfr,cnie parameter a consider Therefore, protocols discrimination and Estimation oepitlk ore.Hr efis focus first we Here sources. point-like lose ainfo pia eoiso fragile or memories optical from mation oka eee,GnsoN 45,USA 14454, NY Geneseo Geneseo, at York w esaeo-h-r,w lodsusopen discuss also we state-of-the-art, he 4 ivj 80Knes ygy Denmark Lyngby, Kongens, 2800 sikvej, esn ntepooi eieboth regime photonic the in sensing m nvriyo ok okY1 G,UK 5GH, YO10 York York, of University trcn n eea omltosfor formulations general and recent st etn.Wt hsbscbackground basic this With testing. s hita Weedbrook, Christian 50–67 69–71 E fpatcltss nldn the including tasks, practical of θ n trdi lc o,o hc Alice which of box, black a in stored and al eae ihteie of idea the with related rally hr h aa-iedtcino re- of detection radar-like the where , 35–49 hs cuayadsniiiyd not do sensitivity and accuracy whose hr h nomto trdin stored information the where , 3,USA 139, ,USA 9, 69–83 θ θ 0 nl yohss or hypothesis) (null sacniuu parameter, continuous a is hr Ryeg’ curse” “Rayleigh’s where , 5 n ehLloyd Seth and θ noe naquan- a in encoded θ a nytk a take only may ,6 2, θ 1 (al- θ 2 the classical bit u which is encoded in the parameter θu. We are interested in the “scaling” of the QCRB, i.e., Let us analyze the estimation/discrimination problem on how δθ2 behaves for large number of uses n. Here with an increasing level of complexity. In a basic “block there are two main behaviors to consider. The first one unassisted” protocol, Alice prepares an input state ρ is known as the standard quantum limit (SQL) which 2 −1 which is transmitted through the unknown channel θ is the typical scaling δθ & n achievable in classical E 2 −2 and transformed into the unknown output state θ(ρ) for strategies. The other is the Heisenberg limit δθ & n Bob. Consider this process to be performed n times,E so which can only be achieved in a purely quantum setting. ⊗n ⊗n that Alice sends n copies ρ and Bob receives θ(ρ) These have energy analogues when we consider contin- assuming that the channel is memoryless. To retrieveE θ, uous variable systems5 and parameter estimation with Bob applies an optimal generally-joint measurement to bosonic channels. Assuming a single-use (n = 1) of the his n-copy output state. The type of measurement is dif- comb and an energy-constrained input state with N mean ferent depending on the problem. In channel estimation, number of photons, then δθ2 & N −1 corresponds to the the measurement has a continuous outcome from which SQL, while δθ2 & N −2 is the Heisenberg limit. Bob constructs an unbiased estimator θ˜ of θ, affected by For instance, the Heisenberg limit is achievable for some error variance δθ2 := (θ˜ θ)2 . In channel dis- large n when we estimate the phase factor ϕ of a uni- h − i iϕH crimination, Bob uses a dichotomic measurement which tary Uϕ = e with H being the free Hamiltonian. In provides the bit u with some mean error probability perr. fact, it is sufficient to use the sequential protocol with The most general estimation/discrimination protocol an input pure state 0, 1 sr + 1, 0 sr so that we have the inϕ| i | i is based on unlimited entanglement and adaptive quan- output 0, 1 sr + e 1, 0 sr where the phase coherently tum operations (QOs), which are applied jointly by Alice accumulates.| i The same| limiti is achievable in the number 12,27,31,32 and Bob . As also discussed in ref. 26, this proof photons N by using the N00N state N, 0 sr + 0,N sr tocol can be represented as a quantum comb86. This in a single use of the block-assisted protocol.| i | i is a quantum circuit board whose slots are filled with We know a simple criterion to establish if channel es- the unknown quantum channel θ. The comb is based timation is limited to the SQL. In particular, Pirandola on a register with an arbitrary numberE of systems, pre- and Lupo27 shed light on the role of teleportation87–89 in pared in some fundamental state ρ. The entire register quantum metrology finding that the property of telepor- 33 undergoes arbitrary QOs before and after channel θ is tation covariance implies the SQL. Recall that a chan- probed, as depicted in Fig. 1. The QOs can alwaysE be nel is tele-covariant if, for any teleportation unitary U assumed to be trace-preserving by adding extra systems (PauliE 1 or displacement operator5), we may write33 and adopting the principle of deferred measurement1. At the output of the comb, the state ρn is detected by an (UρU †)= V (ρ)V †, (3) θ E E optimal quantum measurement whose outcome is classically processed. for a generally different unitary V . Because of this prop- It is clear that the quantum comb encompasses the erty, a quantum channel is teleportation-simulable or n ⊗n Choi-stretchable, i.e., we may write the simulation33,34 previous block protocol with output ρθ = θ(ρ) . It also includes the “block-assisted” protocol whereE the in- (ρ)= (ρ ρE ), (4) put state is bipartite and comprises a signal system s, E T ⊗ sent to probe the channel, and an idler or reference sys- where is the QO of teleportation and ρ is the chan- tem r, which only assists the output measurement. By E nel’s ChoiT matrix ρ := (Φ ), with Φ being a repeating the process n times, the output is given by E sr sr maximally entangled state.E ⊗ I ρn = (ρ )⊗n, where is an identity channel. Fi- θ θ sr Here two observations are in order. First, let us remark nally,E the⊗ comb I can also describeI “sequential” protocols that the simulation in Eq. (4) needs a suitable asymptotic where the input state is transmitted through n instances formulation33,34 for bosonic channels, whose Choi matri- of the channel θ θ. See Fig. 1 for this classification. E ◦···◦E ces are energy-unbounded. In fact, a bosonic channel Performance of channel estimation µ E has Choi matrix ρE := limµ ρE , where the Choi sequence Assume that the quantum comb in Fig. 1 is used and µ µ ρE := (Φsr) is based on the two-mode squeezed optimized for the problem of quantum channel estima- E ⊗ I 5 µ vacuum (TMSV) state Φsr with variance µ = 2N + 1, tion. Then, the ultimate performance is limited by the with N being the mean number of thermal photons in quantum Cramer Rao bound (QCRB) each mode. Second, let us note that a teleportation- 2 1 covariant channel is a specific type of programmable δθ n , (1) channel28–31, whereE the quantum gate array is telepor- ≥ QFI(ρθ ) tation and the program state is the Choi matrix of the where QFI is the quantum Fisher information9 channel. The teleportation simulation has specific ad- n n vantages in terms of achievability of the QCRB26. 8[1 F (ρθ ,ρθ dθ)] QFI(ρn)= − + , (2) By definition, a parametrized quantum channel is θ dθ2 θ jointly teleportation-covariant27 if we may write Eq.E (3) † † with F (ρ, σ) := Trp√σρ√σ being the Bures fidelity be- for any θ, i.e., θ(UρU ) = V θ(ρ)V , so that the out- tween two states ρ and σ. put unitary V Edepends on U butE not on θ. Because of 3

Block protocols Sequential protocols (a) Adaptive protocol (b) Block unassisted (d) Sequential unassisted ⊗ ℇ ℇ ℇ ℇ ℇ ℇ

Quantum Comb (c) Block assisted (e) Sequential assisted ⊗ ℇ ℇ ℇ

FIG. 1: Protocols for quantum estimation and discrimination. (a) Adaptive protocol represented as a quantum comb. An input register with an arbitrary number of systems (wires) is prepared in a fundamental initial state ρ. Each probing of the unknown channel Eθ is performed by inputting a system from the register and storing the output back in the register. Probings n are interleaved by arbitrary QOs performed over the entire register. After n probings, the total output ρθ is subject to a joint quantum measurement. (b) Block unassisted protocol where channel Eθ is probed n times in an identical and independent way. (c) Block assisted protocol where channel Eθ is probed by a signal system coupled to a reference system. (d) Sequential unassisted protocol where the input is transmitted through n consecutive instances of the channel Eθ. (e) Sequential assisted protocol where the input is bipartite and partially transmitted through n consecutive instances of Eθ. this property, we may write the teleportation simulation be exceeded without repeaters. Estimating η is therefore

θ(ρ)= (ρ ρEθ ) for any θ. Replacing this simulation in of paramount importance. The best-known performance Eeach slotT of the⊗ quantum comb in Fig. 1 and “stretching” in estimating the transmissivity η of a pure-loss channel 2 −1 the adaptive protocol, we write the output state as η := η,0 is limited to the SQL δη γN , with a Epre-factorE γ which has been improved over≥ time. ρn = Λ(ρ⊗n) , (5) θ Eθ Monras and Paris90 studied the block unassisted proto- for a global quantum channel Λ. Because the QFI is col with single-mode Gaussian states. They established − monotonic under channels and multiplicative over tensor the limit δη2 ηN 1 achievable by using coherent- n states, also known≥ as shot-noise limit or classical bench- products, one finds that QFI(ρθ ) nQFI(ρEθ ), so that the QCRB must satisfy the SQL ≤ mark. They also found that squeezing may improve the pre-factor and, therefore, beat the shot-noise limit. 2 −1 δθ [nQFI(ρEθ )] , (6) ≥ These results may be achieved by computing the QFI as µ in Eq. (2) and using the formulas for the fidelity between where we implicitly mean QFI(ρEθ ) := limµ QFI(ρEθ ) for 91 a bosonic channel. In other words, the adaptive protocol Gaussian states . The best performance so far has been achieved by using the block protocol with non-Gaussian has been reduced to a block assisted protocol, where n 92 maximally-entangled states Φ⊗n are used to probe the states, including Fock states. In this way, Adesso et al. sr found the improved scaling δη2 η(1 η)N −1. unknown quantum channel θ. ≥ − Because the class of teleportation-covariantE channels is Performance of channel discrimination wide, we have that channel estimation is limited to the Now assume that the quantum comb in Fig. 1 is used and SQL in many situations. For instance, the estimation of optimized for binary discrimination, i.e., for the retrieval the probability parameter p in depolarizing, dephasing or of the parameter θ from a binary alphabet θ0,θ1 where { } erasure channels is limited to27 δp2 p(1 p)n−1. Then, the two possible values have identical Bayesian cost and the estimation of the thermal noise≥n ¯ in− a thermal-loss the same prior probability. This is now a problem of 27 channel η,n¯ with fixed transmissivity η is limited to quantum channel discrimination, where we aim at dis- 2 E −1 δn¯ n¯(¯n + 1)n , which sets the limit for estimating tinguishing between two channels 0 = θ0 and 1 = θ1 , ≥ E E E E excess noise in quantum key distribution. By contrast, or equivalently at retrieving the classical bit u from u. n n E the ultimate estimation limit for the transmissivity η is Let us call ρ0 and ρ1 the two possible output states of not known. Because η,n¯ is not jointly teleportation- the comb. The optimal performance in terms of the min- covariant in η, the reductionE in Eq. (6) does not apply. imum error probability is given by the Helstrom bound16 The optimal estimation of bosonic loss is an open prob- p ( , ) = [1 D(ρn,ρn)]/2, (7) lem with a number of partial results. Solving this prob- err E0 E1 − 0 1 lem is important because loss is the main decoherence where D(ρ, σ) := ρ σ /2 is the trace-distance1. This is effect in quantum optical communications, from fibre- obtained by a suitable|| − dichotomic|| positive-operator val- based to free-space implementations. Recall that the ued measure (POVM), called Helstrom POVM. Equiv- transmissivity η limits the ultimate rate achievable by alently, note that the maximum classical information J point-to-point protocols of quantum and private commu- retrieved from the box is equal to nication according to the Pirandola-Laurenza-Ottaviani- Banchi bound33, so that log (1 η) bits per use cannot J =1 H [p ( , )], (8) − 2 − − 2 err E0 E1 4 where H2 is the binary Shannon entropy. problem of quantum channel discrimination. In the most The error probability greatly simplifies if the two chan- basic description, an optical classical memory can be seen nels 0 and 1 are jointly teleportation-covariant, i.e., we as an array of cells described as microscopic beamsplit- E E † † may write u(UρU )= V u(ρ)V for any u. This allows ters with different reflectivities (see Fig. 2). Each cell E E us to use the teleportation simulation u(ρ)= (ρ ρEu ) stores an information bit u = 0, 1 in two equiprobable E T ⊗ over the Choi matrix ρEu . We may then stretch the comb reflectivities, the pit-reflectivity η0 (0, 1) and the land- n ⊗n ∈ and write its output as ρu = Λ(ρEu ) for a global chan- reflectivity η1 > η0. This single-cell model is equivalent nel Λ. Because the trace distance is monotonic under to a black-box model read in reflection so that the reflec- ⊗n ⊗n Λ, we have perr [1 D(ρE0 ,ρE1 )]/2. Now note that tivity plays the role of the transmissivity parameter. The this is achievable≥ by an− assisted protocol which exploits readout may also be affected by thermal noise, e.g., due maximally-entangled states at the input, so that27 to stray photons generated by the source. Thus the read-

⊗n ⊗n out corresponds to discriminating between two thermal- perr( 0, 1) = [1 D(ρE0 ,ρE1 )]/2, (9) E E − loss channels, 0 := η0,n¯ and 1 := η1,n¯ , with different reflectivity, η Eand η E, but fixedE thermalE numbern ¯. Other where D = lim D(ρµ⊗n,ρµ⊗n) for bosonic channels. In 0 1 µ E0 E1 decoherence effects may be included35, such as optical finite dimension, Eq. (9) implies that the diamond dis- diffraction, memory effects and inter-bit interference36. tance between two jointly teleportation-covariant chan- We may consider different “transmitters” composed of nels is simply equal to 0 1 ⋄ = ρE0 ρE1 . signal modes probing the cell and reference modes assist- Starting from Eq. (9),||E we−E may|| write|| simple− lower|| and 93 ing detection. The coherent-state transmitter only uses upper bounds using the Fuchs-van de Graaf relations n signal modes in identical coherent states α α ⊗n. and the quantum Chernoff bound (QCB)20. In fact, re- | is h | ⊗n ⊗n More powerfully, we may define a “classical” transmit- call that for any pair of multicopy states ρ0 and ρ1 , ter in the quantum optical sense94,95. This is a block the minimum error probability p = [1 D(ρ⊗n,ρ⊗n)]/2 err − 0 1 assisted protocol employing mixtures of coherent states satisfies the fidelity-based lower bound and the QCB n R d2 α (α) α α , where (α) is a probability dis- tributionP of amplitudes| i h | α, andP α α is a multimode 1 p1 F (ρ ,ρ )2n p − − 0 1 := F (n)(ρ ,ρ ), (10) coherent state with n signal modes| i h | and n reference err ≥ 2 − 0 1 n modes. The optimal classical transmitter has to be com- Q(ρ0,ρ1) s 1−s pared with an Einstein-Podolsky-Rosen (EPR) transmit- perr ,Q := inf Tr(ρ0,ρ1 ). (11) ≤ 2 s ter. This is a block entanglement-assisted protocol where µ⊗n In particular, for multimode Gaussian states, ρ0 and ρ1, we send part of n TMSV states Φsr , so that each signal we know closed formulas for computing the fidelity91 and mode is entangled with a reference or “idler” mode. the QCB22. These inequalities can be extended to the For both the classical and the EPR transmitter, we adaptive error probability of Eq. (9) valid for jointly therefore assume an input 2n-mode state ρsr, which teleportation-covariant channels, so that we may write27 is transformed by the cell into an output state σu := ⊗n ⊗n (ρsr) for the n reflected signal modes and the n Eu ⊗ I (n) Q(ρE0 ,ρE1 ) n kept reference modes. This output is detected by an F (ρE0 ,ρE1 ) perr( 0, 1) , (12) − ≤ E E ≤ 2 optimal Helstrom POVM with some error probability. with asymptotic functionals over bosonic Choi matrices. We then compare the optimal information retrieved by The results in Eqs. (9) and (12) apply to many cases, the classical transmitter Jclass with that retrieved by the including the adaptive discrimination of Pauli channels, EPR transmitter JEPR, quantifying the information gain erasure channels, and the noise parameters in bosonic ∆ := JEPR Jclass. Positive values ∆ > 0 provide quantum advantage.− A fair comparison between these trans- Gaussian channels, such as the thermal photonsn ¯0 and mitters involves fixing the mean number of signal photons n¯1 in thermal-loss channels η,n¯0 and η,n¯1 with the same transmissivity η. Unfortunately,E theE same reduc- probing the cell. Because each cell is probed n times, we tion does not apply to the discrimination of bosonic loss, may consider different types of energy constraints. One type of constraint is “local”, meaning that we fix e.g., η0 and η1 of a thermal-loss channel with fixed noise the mean number of photons N in each probing, so that n¯, because η0,n¯ and η1,n¯ are not jointly teleportation- covariant. AsE a result,E establishing the optimal discrimi- the total energy scales as nN. We may write the following general bound for the error probability pclass achievable nation of bosonic loss is still an open problem. What we 35 know currently is that block-assisted strategies based on by any classical transmitter entangled states may greatly outperform block unassisted 1 p1 F (N)2n strategies, especially when we employ a small number of pclass (n,N) := − − , (13) signal photons in the presence of thermal noise. This is ≥C 2 the observation which is at the basis of the applications where F (N) is Bures fidelity between the two possible of quantum reading and quantum illumination. outputs 0( √N √N ) and 1( √N √N ) generated by Quantum reading of classical data a single-modeE | coherentih | stateE with| Nihmean| photons. This In 2011, ref. 35 showed how the readout of classical data leads to the upper bound Jclass(n,N) 1 H2[ (n,N)]. from an optical digital memory can be modelled as a For the EPR transmitter, we may exploit≤ the− QCBC which 5

Quantum Reading Quantum Illumination classical transmitters, up to extremely large bandwidths. Because a few entangled photons can retrieve more infor- ≃ 0 = 0 OR mation than any classical source of light, we may indeed Target Bath work at very low energies. This regime may be mapped into much faster optical readers and denser memories35. Quantum reading has been extensively studied39–49. Bath Ref. 39 extended the model to multi-cell error correc- Cells tion coding and defined the notion of quantum reading capacity, a quantity that was later shown to be super- additive40. Guha and Shapiro41 also defined a similar notion of capacity42 and studied the error exponent for Nn, idlers Nn, idlers quantum reading. For single-cell reading of an ideal Transmitter Receiver Transmitter Receiver memory in noiseless conditions, Nair43 showed that Fock states are optimal. More generally, entangled states with FIG. 2: Quantum reading (left) and Gaussian quantum illu- the signal beam in a number-diagonal reduced state may minantion (right). In the basic formulation, these are both 43 based on an EPR transmitter, so that n two-mode squeezed also provide a positive quantum advantage . This class vacuum (TMSV) states irradiate N mean photons per mode of states is optimal for non-adaptive discrimination with 44 over the cell/target (where N is typically low). The reflected single-mode and multi-mode pure-loss channels . signals are combined with the retained idler (reference) modes Hirota45 proposed an alternative model based on a in a joint detection, whose output u discriminates between binary phase encoding and showed how entangled co- two hypotheses. In quantum reading, u is the information bit herent states may achieve error-free quantum reading. encoded into a cell with reflectivity ηu and subject to ther- Non-Gaussian entangled states were also considered by mal noisen ¯. In quantum illumination, u is related with the Prabhu Tej et al.46. Then, Bisio et al.47 studied a noise- absence (η0 = 0) or the presence (η1 ≃ 0) of a low-reflectivity target. The reflection is mixed with an environment with free unitary model of quantum reading where both the in- bright thermal noisen ¯ ≫ 1. These schemes are examples puts of the unknown beamsplitter are accessible for prob- of block-assisted protocols for quantum channel discrimina- ing and both its outputs for detection. Assuming a single tion. In the regimes considered, they largely outperform clas- probe (n = 1), they found that the optimal two-mode in- sical strategies, i.e., corresponding schemes based on classical put is the superposition of a N00N state and the vacuum transmitters that are not entangled but composed of mixtures 00 . This approach was later extended by Dall’Arno et of coherent states. al.| 48i to unambiguous quantum reading, where the probability of error is replaced by an inconclusive result. Similar to Hirota45, Dall’Arno et al.48 also considered µ µ n gives the lower bound JEPR 1 H2[Q(ρE0 ,ρE1 ) /2], a version of perfect quantum reading with zero discrim- µ ≥ − where ρEu is the quasi-Choi matrix of the unknown chan- ination error. This is possible by designing an ideal cell µ nel u probed by a TMSV state Φ with µ = 2N + 1. which is either a beamsplitter with perfect reflectivity E sr Using these bounds we construct a sufficient condition to (η1 = 1) or a beamsplitter with lower reflectivity η0 < 1 prove the quantum advantage ∆ > 0. and suitable π/2 phase shifters at the input and output Numerical investigations show that positive values of ports. This± scheme was experimentally implemented48. ∆ are typical for low signal photons and high reflectiv- As shown in Fig. 3(a), the setup consisted of a Mach- ities in wide ranges for the thermal noise. A positive Zehnder interferometer with the variable beamsplitter quantum advantage can already be achieved by a single situated in one arm. A heralded single photon source probe per cell (n = 1). When the land-reflectivity is very based on spontaneous parametric down-conversion in a high η1 1 (ideal memory), one may derive analytical 2mm long beta barium borate (BBO) crystal pumped expressions→ 37 and find regimes of parameters for which with a continuous-wave laser diode at 405nm served as ∆ 1 bit per cell. This extremal value means that the the quantum state source. The photon pairs generated EPR→ transmitter is able to fully read the cell, while classi- in orthogonal polarizations in the type-II phase match- cal transmitters do not retrieve any information. This ad- ing process were separated by a polarizing beamsplitter. vantage may also be used to design cryptographic mem- While one photon was used to herald the process, the ories whose data can only be read by entanglement38. other was fed into the Mach-Zehnder interferometer. Another type of energy constraint is “global”, meaning The setup discriminated between two beamsplitter that we fix the total mean number of photons NT over configurations, the one with reflectivity η1 = 1 and the all the n uses, so that we employ an average of NT/n other with reflectivity η0 < 1 and an additional phase photons per probing. Let us call n the bandwidth of the shift in the Mach-Zehnder interferometer arm. The mea- transmitter, due to the fact that this can be physically re- surement consisted of three photon counters, one at the lated with the number of effective frequencies used in the output of the variable beamsplitter under test and the readout. One can then show that, at sufficiently low en- other two at each output of the interferometer. Coin- ergies NT . 10 photons, a narrowband EPR transmitter cidence counts between each of the three detectors and (even monochromatic nEPR = 1) is able to beat arbitrary the trigger detector were measured using a 3ns coinci- 6 dence window. With the perfect beamsplitter under test, quantum illumination advantage in terms of detection only one of the detectors at the output of the Mach- probability versus false-alarm probability54. Zehnder interferometer would detect the photon (Hong- In 2015, Gaussian quantum illumination was ex- Ou-Mandel effect), while for the beamsplitter with η0 < 1 tended to the microwave regime, thus providing a then any of the two other detectors would detect the pho- prototype of quantum radar55. In this scheme, an ton (due to the additional phase shift). electro-optomechanical converter97–99 transforms an op- Quantum illumination of targets tical mode into microwave. If this transducer has high Quantum sensing can be used not only to enhance the quantum efficiency, then optical-optical entanglement is readout of information from classical systems, but also translated into microwave-optical entanglement. The mi- to boost the standoff detection of remote objects. This crowave signal is sent to probe the target region, while idea was first pushed forward by the efforts of Lloyd the optical idler is retained. The microwave radiation and Shapiro at MIT50–52. In 2008, Lloyd50 designed a collected from the target region is then phase conjugated qubit-based protocol of quantum illumination, showing and upconverted into an optical field by a second use of how the detection of a low-reflectivity target object can the transducer. The optical output is finally combined be enhanced by using quantum entanglement. The ad- with the retained idler in a joint detection, following the 56 vantage of the entangled transmitter over non-entangled ideas behind the receiver design by Guha and Erkmen . ones is achieved even if the entanglement itself is com- In this way, ref. 55 found that the error probability of mi- pletely lost after reflection from the target. In fact, the crowave quantum illumination is superior to that of any initial signal-idler entanglement is mapped into residual classical radar of equal transmitted energy. A followup 57 but yet quantum correlations between the reflected sig- analysis has been carried out by Xiong et al. . nal and the kept idler that a suitably-designed quantum More recently, Sanz et al.58 studied the protocol of detector may “amplify” with respect to the uncorrelated quantum illumination using the tools of quantum metrol- thermal background. ogy so as to measure the reflectivity of the target. They In the same year, Shapiro’s team51 proposed a prac- employed the QFI to bound the error probability show- tical formulation of quantum illumination based on ing a 3dB-enhancement of the signal-to-noise ratio with continuous-variable systems5. Ref. 51 designed a Gaus- respect to the use of local measurements. They also sian version where bosonic modes are prepared in Gaus- considered non-Gaussian Schrödinger’s cat states. Other sian states and sent to detect an object with low reflec- studies have quantified the quantum illumination advan- tivity η 0 in a region with bright thermal noise, i.e., tage in terms of “consumption” of discord96 associated withn ¯ ≃ 1 mean thermal photons. The detection pro- with the target59, and in terms of mutual information60. cess can≫ be modelled as the discrimination between a Finally note that quantum illumination has been also zero-reflectivity thermal-loss channel η=0,n¯ (target ab- studied as an asymmetric Gaussian discrimination prob- E 25,54,61,62 sent) and a low-reflectivity thermal-loss channel η,n¯′ lem by various authors . In this setting, TMSV with η 0 andn ¯′ =n/ ¯ (1 η) (target present). HereE states have been identified as optimal probes62. Finding the factor≃ (1 η)−1 excludes− the possibility of a “passive the ultimate performance achievable by an adaptive ver- signature” which− means the possibility of detecting the sion of quantum illumination remains an open question. target without transmitting any radiation by just mea- Several experimental implementations of quantum il- suring a lower received background level. As also de- lumination have been reported63,65,66. As depicted in picted in Fig. 2, one assumes that the detector’s noise Fig. 3(b), Lopaeva et al.63 exploited a parametric down- does not depend on the presence of the target. conversion source using a BBO crystal to generate two In this setup, we assume a local energy constraint, so intensity-correlated light beams in orthogonal polariza- that N mean photons are irradiated by each of the n tions at 710nm. Both beams were detected by a photon- bosonic modes sent over the target object. Under this counting high-quantum efficiency CCD camera. The tar- assumption, we compute the error probability associated get object, a 50:50 beamsplitter in the experiment, was with the various transmitters. In particular, we exploit placed in one of the two entangled beams before detec- the bounds in Eqs. (10) and (11) to compare the perfor- tion. The beamsplitter object was illuminated by pho- mance of the EPR transmitter (based on TMSV states) tons scattered on an Arecchi’s rotating ground glass to with that of the classical transmitter (based on coherent simulate a thermal environment. A single captured im- states). In the regime of low-energy signals (N 1) age was used to measure the second-order correlations and many modes (n 1), the EPR transmitter has≪ the between the two beams. The implementation shows ro- 51 err ≫ scaling pEPR exp( nηN/n¯)/2, which clearly beats bustness against noise and losses, and demonstrates a ≃ − err the classical transmitter pclass exp( nηN/2¯n)/4. In quantum enhancement in target detection in thermal en- err ≥ − particular, pEPR realizes a 6dB advantage in the error- vironments even when nonclassicality is lost. However, probability exponent over the coherent-state transmitter coincidence detection of spontaneous parametric down- err 53 pCS exp( nηN/4¯n)/2. Zhuang et al. proved that conversion is not the optimal detection method to ex- ≃ − err the theoretical limit pEPR can be achieved by an explicit tract the most information from the signal-idler entan- quantum receiver based on feed-forward sum-frequency gled modes, and the implemented classical scheme using generation. This receiver has been also used to show the weakly thermal states is also non-optimal. 7

a) b) CCD

Object

BBO

Mirror Arecchi disk

FIG. 3: Experimental demonstrations of quantum reading and quantum illumination. a) Experimental setup of perfect quantum 48 reading . A photon pair source is used to generate a heralded single photon using a trigger detector (DT ). The heralded single photon is fed into an interferometer with variable ratio couplers (VRCs) and phase modulators (PMs) to add additional phase shifts. Coincidence detection of the outputs are used to discriminate between two possible splitting ratios of VRC- mid. b) Quantum illumination experiment of Lopaeva et al.63. See also Ref. 64. Both beams of a photon pair source are detected by a photon counting CCD camera. In the experiment the target is a 50:50 beamsplitter placed in one of the beams. The beamsplitter is simulated to be in a thermal environment by illuminating it with scattered light from an Arrechi disk. c) Quantum illumination experiment of Zhang et al.66. Photon pairs generated by spontaneous parametric down-conversion (SPDC) at two different wavelengths are split using a dichroic mirror (DM). One of the photons is stored in a delay line using a dispersion-shifted LEAF fiber (DSF). The other photon is phase modulated (PM). A lossy and noisy environment is simulated by a beamsplitter (BS) and amplified spontaneous emission (ASE) from a erbium doped fiber amplifier (EDFA). The joint detection is implemented using an optical parameteric amplifier (OPA) whose output is detected by a PIN photo detector (D). DCF: dispersion-compensating fiber, POL: polarizer, CWDM: coarse wavelength-division multiplexer. Thin lines are optical fiber, thick lines are unguided propagation. Figures adapted with permission from: a), ref. 48, c 2012 APS; b), ref. 63, c 2013 APS; c) ref. 66, c 2015 APS.

Adopting a different approach, in 2013 Zhang et al.65 the Rayleigh length λ/a, where λ is the wavelength of reported a secure communication experiment based on the emitted light and≃a is the numerical aperture of the quantum illumination. More recently, Zhang et al.66 observing lens. For this reason, if we use a converging op- demonstrated the advantage of quantum illumination tical system to focus light on a screen and an array of de- over coherent states by using broadband entangled Gaus- tectors to measure the intensity, the Rayleigh’s criterion sian states, as produced by continuous-wave spontaneous together with the presence of photon shot noise, can lead parametric down-conversion. In the experiment shown to severe limitations in resolving point-like sources. The in Fig. 3(c), the signal modes were phase modulated be- situation changes completely if we consider a fully quan- fore probing the weakly-reflecting target, while the idler tum description of the light and the measurement appa- modes were stored in a delay line. The joint measurement ratus. In this way, Tsang et al.69 showed the existence of was performed by combining the reflected signal modes a quantum detection scheme able to measure the distance and the idler modes with a pump in another optical para- between two point-like sources with a constant accuracy, metric amplifier. The output in the order of nW was then even when the sources have sub-wavelength separation. detected by a PIN photo detector with high gain and low This ground-breaking result was achieved by relating the noise. They showed a 20% improvement of the signal-to- problem of resolving two incoherent point-like sources to noise ratio in comparison to the optimal classical scheme quantum estimation theory and using the QCRB. in an environment exhibiting 14dB loss and a thermal The theory beyond these results were extended from background 75dB above the returned probe light. incoherent sources emitting faint pulses to thermal Optical resolution beyond the Rayleigh limit sources of arbitrary brightness70,71. In general, ref. 70 The Rayleigh criterion is a well-known result in clas- established a connection between optical resolution and sical imaging. Two point-like sources cannot be opti- bosonic channel estimation, so that measuring the sep- cally resolved (in the far field) if they are closer than aration between two point-like sources is equivalent to 8 estimating the loss parameters of two lossy channels. In Paúr et al.80 simulated Gaussian and slit apertures by a this way, ref. 70 developed a theory of super-resolution for digital micromirror chip illuminated by a laser. Projec- point-like sources emitting light in a generic state, i.e., at- tion onto different modes was performed by a spatial light tenuated or bright, classical, coherent, incoherent, as well modulator which creates a digital hologram measured by as entangled (e.g., in a microscope setup). The ultimate an electron-multiplying CCD. See Fig. 4(d). Let us con- resolution was found as a function of the optical proper- clude that super-resolving quantum imaging is a hot topic ties of the two sources and their separation. In particular, and many other experiments could be mentioned81–83. super-resolution can be enhanced when the sources emit Discussion and outlook entangled or quantum-correlated (discordant) light. Quantum sensing is a rapidly evolving field with many 72 More recently, Kerviche et al. extended Tsang et potential implications and technological applications. al.’s analysis from a Gaussian point spread function to a Despite the great advances that have been achieved in re- hard-aperture pupil, proving the information optimality cent years, a number of problems and experimental chal- of image-plane sinc-Bessel modes. They also generalized lenges remain open. From the point of view of the basic the result to an arbitrary point spread function. Rehacek theoretical models of quantum metrology and quantum 73 et al. carried out further work on the optimal measure- hypothesis testing, we may often compute the ultimate 74 ments for beating the Rayleigh limit. Yang et al. ex- performances allowed by quantum mechanics. However, plored the use of homodyne or heterodyne detection. Fi- we do not know in general how to implement the opti- 75 nally, Lu et al. studied the quantum-optimal detection mal measurements achieving these performances and/or of one-versus-two incoherent optical sources with arbi- what optimal states we need to prepare at the input of trary separation. the unknown quantum channel. Then, do we need to Shortly after the idea of Tsang et al.69 was pre- consider feedback and perform adaptive protocols? For sented, it was experimentally verified in several proof- instance, this is an open question for both estimation and of-principle experiments. The first experiment by Tang discrimination of bosonic loss26, which is at the basis of et al.76 was based on super-localization by image inver- quantum reading, quantum illumination and quantum- sion interferometry77. As shown in Fig. 4(a), they used enhanced optical super-resolution. an image inversion interferometer to determine the sepa- From a more practical and experimental point of view, ration of two incoherent point sources, generated by two there are non-trivial challenges as well. Despite a first laser beams in orthogonal polarizations stemming from proof-of-principle demonstration48 based on the unitary the same HeNe laser. Using the light from the simulated discrimination of beamsplitters, we do not have yet a sources as the input, the interferometer was implemented truly quantum reading experiment where a single out- as a Mach-Zehnder interferometer with image inversion put of the cells is effectively accessed for the readout. A generated by a lens system in one arm. The other arm full demonstration would involve an actual (one- or two- was delayed appropriately so that the detector at the dimensional) array of cells, where information is stored output of the interferometer ideally showed no response with classical codes and the quantum readout is perfor zero separation due to destructive interference. With formed by simultaneously probing blocks of cells. This growing separation of the two sources the destructive in- idea may be further developed into a full experiment of terference becomes more and more imperfect, yielding an bosonic quantum pattern recognition where the use of optical resolution beyond the Abbe-Rayleigh limit. entanglement across an array may boost the resolution Yang et al.78 used heterodyne detection with a local os- of unsupervised problems of data clustering. cillator in TEM01 mode to detect the separation of the Quantum illumination has had various experimental two slits in a double slit configuration beyond the classi- demonstrations63,65,66 but still remain the issue of design- cal resolution limit. As also depicted in Fig. 4(b), they ing a practical receiver that would allow one to closely use a piece of paper to achieve incoherence and diffuse approximate the theoretical limit forseen by the Hel- transmission. Measuring at a frequency of some MHz to strom bound. Challenges become more serious when avoid noise at lower frequencies, the beat between the we consider the implementation of quantum illumina- local oscillator and the beam illuminating the slits be- tion in the microwave regime55. Here the development comes zero if the separation is 0 as both spatial parts of highly-efficient microwave-optical quantum converters of the TEM01 mode have a phase shift of π. Separat- could mitigate experimental issues related with the gen- ing the two slits yields a measurement beyond the Abbe eration of microwave entanglement and the detection of limit. While the scheme requires the two sources to be microwave fields at the single-photon level. Furthermore exactly aligned to the center of the TEM01 mode, us- these converters are highly desired for other applications, ing higher-order TEM modes will provide general sub- in particular as interfaces for connecting superconduct- Rayleigh imaging. ing quantum chips and optical fibers in a potential hybrid In another experiment, Tham et al.79 inserted a phase design of a future quantum Internet89. shift of π into the beam resembling the two incoherent It is clear that other designs of a quantum radar are sources (generated by partially overlapping two beams in possible. For instance, a fully microwave implementation orthogonal polarizations) and projecting it onto the guid- of quantum illumination (without the use of converters) ing mode of a single mode fiber, as in Fig. 4(c). Finally, may be achieved by using a superconducting Josephson 9

a) c)

b) d)

FIG. 4: Proof-of-principle experiments demonstrating a quantum detection scheme able to measure a distance of two incoherent point sources better than the Rayleigh limit. a) Experiment of Tang et al.76. A HeNe laser with fiber coupled output is split at a polarizing beamsplitter (PBS) into two beams of orthogonal polarization. They are recombined at a beamsplitter (BS) with a slight lateral displacement to simulate two incoherent light sources. The light sources are imaged using an image inversion interferometer, a Mach-Zehnder interferometer with an image inverter consisting of two lenses in one arm. One output of the interferometer is detected by a photo detector (D2). b) Experiment of Yang et al.78. The signal beam is frequency shifted by an acousto-optical modulator (AOM) and illuminates slits. A paper card is placed in front of the slits to make the illumination incoherent. The signal beam is measured by heterodyne detection using a local oscillator prepared in TEM01 mode by means of an optical cavity. c) Experiment of Tham et al.79. Two partially overlapping beams as shown in the upper inset are generated by coupling laser light out of a fiber and combining them on a beamsplitter. The distance between the beams can be controlled by the position of the upper mirror. The separation of the two beams is detected by projecting the beams onto a mode orthogonal to TEM00, in their case a spatially antisymmetric field mode. This is performed by passing the two beams through a phase plate which is built in such a way that it introduces different phase shifts between opposite halves of the beam and aligned such that coupling into a well-aligned fiber coupler is minimized. The coupling into a single mode fiber corresponds to a projection onto the TEM00 mode, thus together with the phase plate the beams are projected onto a mode orthogonal to TEM00. d) Experiment of Paúr et al.80. Using a high frequency switched digital micro mirror chip (DMD) illuminated by a HeNe laser two closely spaced incoherent beams are generated. The beam is projected onto different modes by an amplitude spatial light modulator (SLM) generating a digital hologram. The first-order diffraction spectrum is detected by an electron multiplying CCD (EMCCD) camera. Figures adapted with permission from: a), ref. 76, c 2016 OSA; b), ref. 78, c 2016 OSA; c) ref. 79, c 2017 APS; d) ref. 80, c 2016 OSA. parametric amplifier to generate signal-idler microwave ence times, until the final joint detection. Then, unlike entanglement. Reflected signals could then be phase- classical radars, whose performance improves as the sig- conjugated via another parametric amplifier, recombined nal power is increased at constant bandwidth, the quan- with the idlers, and finally measured, e.g., by using a tum counterpart needs to increase bandwidth at constant transmon qubit as single-photon detector. The idea of signal brightness. The challenge is therefore to generate using Josephson mixers and photocounters has been also microwave pulses with a time-bandwidth product of 106 studied by Las Heras et al.67 in the context of using mi- modes or more. Furthemore, classical radars can interro- crowave quantum illumination to reveal phase-shift in- gate many potential target bins with a single pulse, while ducing cloaking, also known as “invisible cloak”. present models of quantum radar may only query a single Other experimental challenges need to be addressed in polarization, azimuth, elevation, range, Doppler bin at a order to build an actual quantum radar68. An important time. This is an area of development with very promising 100 aspect is the preservation of the idler modes while the sig- steps forward . Finally, more advanced technical issues nals are being propagated forward and back from the tar- related to random-amplitude targets and radar clutters get. The idlers should be kept in a low-loss delay line or should also be addressed. stored in quantum registers with sufficiently-long coher- Regarding the experimental challenges for super- 10 resolution76, most of the current schemes, from spatial- Lvovsky, M. Miková, M. Tsang, Z. Zhang. S.P. would mode demultiplexing to super-localization by image in- like to specifically thank J. H. Shapiro for discussions version and heterodyne, rely on the assumption that we on the experimental challenges related with a quantum need to know the location of the centroid of the sources in radar, and R. Nair for the feedback on the experimen- order to get full quantum-optimal resolution. In general, tal challenges in optical super-resolution. S. P. thanks this location is not exactly known76,78–80, so that achiev- support from the EPSRC via the ‘UK Quantum Com- ing maximum alignment before estimating the separation munications Hub’ (EP/M013472/1). T. G. would like to becomes an important step in order to optimize the per- acknowledge support from the Danish Research Coun- formance in a realistic implementation. cil for Independent Research (Sapere Aude 4184-00338B) Acknowledgments as well as the Innovation Fund Denmark (Qubiz) and the The authors would like to thank feedback from U. L. Danish National Research Foundation (Center for Macro- Andersen, L. Banchi, Sh. Barzanjeh, J. Borregaard, S. scopic Quantum States). L. Braunstein, V. Giovannetti, S. Guha, C. Lupo, A.

1 Nielsen, M. A. & Chuang, I. L. Quantum Computation 19 Bergou, J. A. , Herzog, U. & Hillery, M. Discrimination and Quantum Information (Cambridge University Press, of Quantum States, in Quantum state estimation, Lecture Cambridge, 2000). Notes in Physics, 649, 417-465 (Springer, Berlin, Heidel- 2 Hayashi, M. Quantum Information Theory: Mathematical berg, 2004). Foundation (Springer-Verlag, Berlin, 2017). 20 Audenaert, K. M. R. et al. Discriminating States: The 3 U. L. Andersen, U. L., Neergaard-Nielsen, J. S., Quantum Chernoff Bound. Phys. Rev. Lett. 98, 160501 van Loock, P. & Furusawa, A. Hybrid discrete- and (2007). continuous-variable quantum information. Nature Phys. 21 Calsamiglia, J., Munoz-Tapia, R., Masanes, L., Acin, A. 11, 713-719 (2015). & Bagan, E., Quantum Chernoff bound as a measure of 4 Serafini, A. Quantum Continuous Variables: A Primer of distinguishability between density matrices: Application Theoretical Methods (Taylor & Francis, Oxford, 2017). to qubit and Gaussian states. Phys. Rev. A 77, 032311 5 Weedbrook, C. et al. Gaussian quantum information. Rev. (2008). Mod. Phys. 84, 621 (2012). 22 Pirandola, S. & Lloyd, S. Computable bounds for the dis- 6 Braunstein, S. L. & Van Loock, P. Quantum informa- crimination of Gaussian states. Phys. Rev. A 78, 012331 tion with continuous variables. Rev. Mod. Phys. 77, 513 (2008). (2005). 23 K. M. R. Audenaert, M. Nussbaum, A. Szkola, and F. 7 Adesso, G., Ragy, S. & Lee, A. R. Continuous variable Verstraete, Commun. Math. Phys. 279, 251 (2008). quantum information: Gaussian states and beyond. Open 24 Invernizzi, C., Paris, M. G. A. & Pirandola, S. Optimal Systems and Information Dynamics 21, 1440001 (2014). detection of losses by thermal probes. Phys. Rev. A 84, 8 Degen, C. L., Reinhard, F. & Cappellaro, P. Quantum 022334 (2011). sensing. Rev. Mod. Phys. 89, 035002 (2017). 25 Spedalieri, G. & Braunstein, S. L. Asymmetric quantum 9 Braunstein, S. L. & Caves, C. M. Statistical distance and hypothesis testing with Gaussian states. Phys. Rev. A 90, the geometry of quantum states. Phys. Rev. Lett. 72, 3439 052307 (2014). (1994). 26 Laurenza, R., Lupo, C., Spedalieri, G., Braunstein, S. L. 10 Braunstein, S. L., Caves, C. M. & Milburn, G. J. Gen- & Pirandola, S. Channel Simulation in Quantum Metrol- eralized Uncertainty Relations: Theory, Examples, and ogy. Preprint at https://arxiv.org/abs/1712.06603 (2017). Lorentz Invariance. Ann. Phys. 247, 135-173 (1996). 27 Pirandola, S. & Lupo, C. Ultimate precision of adaptive 11 Giovannetti, V., Lloyd, S. & Maccone, L. Quantum- noise estimation. Phys. Rev. Lett. 118, 100502 (2017). enhanced measurements: beating the standard quantum 28 Nielsen, M. A. & Chuang, I. L. Programmable Quantum limit. Science 306, 1330-1336 (2004). Gate Arrays. Phys. Rev. Lett. 79, 321 (1997). 12 Giovannetti, V., Lloyd, S. & Maccone, L. Quantum 29 Ji, Z., Wang, G., Duan, R., Feng, Y. & Ying, M. Pa- metrology. Phys. Rev. Lett. 96, 010401 (2006). rameter Estimation of Quantum Channels. IEEE Trans. 13 Paris, M. G. A. Quantum estimation for quantum tech- Inform. Theory 54, 5172–5185 (2008). nology. Int. J. Quant. Inf. 7, 125-137 (2009). 30 Kolodynski, J. & Demkowicz-Dobrzański, R. Efficient 14 Giovannetti, V., Lloyd S. & Maccone, L. Advances in tools for quantum metrology with uncorrelated noise. New quantum metrology. Nature Photon. 5, 222 (2011). J. Phys. 15, 073043 (2013). 15 Braun, D. et al. Quantum enhanced mea- 31 Demkowicz-Dobrzański, R. & Maccone, L. Using Entan- surements without entanglement. Preprint at glement Against Noise in Quantum Metrology. Phys. Rev. https://arxiv.org/abs/1701.05152 (2018). Lett. 113, 250801 (2014). 16 Helstrom, C. W. Quantum Detection and Estimation The- 32 Harrow, A. W., Hassidim, A., Leung, D. W. & Watrous, ory (Academic, New York, 1976). J. Adaptive versus nonadaptive strategies for quantum 17 Chefles, A. Quantum state discrimination. Contemp. channel discrimination. Phys. Rev. A 81, 032339 (2010). Phys. 41, 401 (2000). 33 Pirandola, S., Laurenza, R., Ottaviani, C. & Banchi, L. 18 Barnett, S. M. & Croke, S. Quantum state discrimination. Fundamental limits of repeaterless quantum communica- Adv. Opt. Photonics 1, 238-278 (2009). tions. Nat. Commun. 8, 15043 (2017). See also Preprint 11

at https://arxiv.org/abs/1510.08863 (2015). 55 Barzanjeh, Sh. et al. Microwave quantum illumination. 34 Pirandola, S., Braunstein, S. L., Laurenza, R., Ottaviani, Phys. Rev. Lett. 114, 080503 (2015). C., Cope, T. P. W., Spedalieri, G. & Banchi, L. The- 56 Guha, S. & Erkmen, B. I. Gaussian-state quantum- ory of channel simulation and bounds for private commu- illumination receivers for target detection. Phys. Rev. A nication. Preprint at https://arxiv.org/abs/1711.09909 80, 052310 (2009). (2017). 57 Xiong, B., Li, X., Wang, X-.Y. & Zhou, L. Improve mi- 35 Pirandola, S., Quantum Reading of a Classical Digital crowave quantum illumination via optical parametric am- Memory. Phys. Rev. Lett. 106, 090504 (2011). plifier. Ann. Phys. 385, 757-768 (2017). 36 Lupo, C., Pirandola, S., Giovannetti, V. & Mancini, S. 58 Sanz, M., Las Heras, U., Garc´ıa-Ripoll, J. J., Solano, E. & Quantum reading capacity under thermal and correlated Di Candia, R. Quantum estimation methods for quantum noise. Phys. Rev. A 87, 062310 (2013). illumination. Phys. Rev. Lett. 118, 070803 (2017). 37 Spedalieri, G., Lupo, C., Mancini, S., Braunstein, S. L. 59 Weedbrook, C., Pirandola, S., Thompson, J., Vedral, V. & Pirandola, S. Quantum reading under a local energy & Gu, M. How discord underlies the noise resilience of constraint. Phys. Rev. A 86, 012315 (2012). quantum illumination. New J. Phys. 18, 043027 (2016). 38 Spedalieri, G. Cryptographic aspects of quantum reading. 60 Ragy, S. et al., Quantifying the source of enhancement in Entropy 17, 2218-2227 (2015). experimental continuous variable quantum illumination, 39 Pirandola, S., Lupo, C., Giovannetti, V., Mancini, S. & J. Opt. Soc. Am. B 31, 2045–2050 (2014). Braunstein, S. L. Quantum Reading Capacity. New J. 61 Wilde, M. M., Tomamichel, M., Lloyd, S., & Berta, M. Phys. 13, 113012 (2011). Gaussian Hypothesis Testing and Quantum Illumination. 40 Lupo, C. & Pirandola, S. Super-additivity and entangle- Phys. Rev. Lett. 119, 120501 (2017). ment assistance in quantum reading. Quantum Inf. Com- 62 De Palma, G. & Borregaard, J. The ultimate precision of put. 17, 0611 (2017). quantum illumination. arXiv:1802.02158 (2018). 41 Guha, S. & Shapiro, J. H. Reading boundless error-free 63 Lopaeva et al. Experimental realization of quantum illu- bits using a single photon. Phys. Rev. A 87, 062306 mination. Phys. Rev. Lett. 110, 153603 (2013). (2013). 64 Meda, A., Losero, E., Samantaray, N., Scafirimuto, F., 42 Guha, S., Dutton, Z., Nair, R., Shapiro, J. H. & Yen, B. Pradyumna, S., Avella, A., Ruo-Berchera, I. & Gen- Information Capacity of Quantum Reading in Laser Sci- ovese, M. Photon-number correlation for quantum en- ence, OSA Technical Digest,Optical Society of America) hanced imaging and sensing. J. Opt. 19, 094002 (2017). paper LTuF2 (2011). 65 Zhang, Z., Tengner, M., Zhong, T., Wong, F. N. C. 43 Nair, R. Discriminating quantum-optical beam-splitter & Shapiro, J. H. Entanglement’s benefit survives an channels with number-diagonal signal states: Applica- entanglement-breaking channel. Phys. Rev. Lett. 111, tions to quantum reading and target detection. Phys. Rev. 010501 (2013). A 84, 032312 (2011). 66 Zhang, Z., Mouradian, S., Wong, F. N. C. & Shapiro, J. 44 Nair, R. & Yen, B. J. Optimal Quantum States for Image H. Entanglement-enhanced sensing in a lossy and noisy Sensing in Loss. Phis. Rev. Lett. 107, 193602 (2011) environment. Phys. Rev. Lett. 114, 110506 (2015). 45 Hirota, O., Error Free Quantum Reading by Quasi Bell 67 Las Heras, U., Di Candia, R., Fedorov, K. G., Deppe, State of Entangled Coherent States. Quantum Measure- F., Sanz, M. & Solano, E. Quantum illumination reveals ments and Quantum Metrology 4, 70-73 (2017). phase-shift inducing cloaking. Sci. Rep. 7, 9333 (2017). 46 Prabhu Tej, J., Usha Devi, A. R. & Rajagopal A. K. 68 M. Lanzagorta, Quantum Radar (Synthesis Lectures on Quantum reading of digital memory with non-Gaussian Quantum Computing), Vol. 3, No. 1, pp. 1-139 (Morgan entangled light. Phys. Rev. A 87, 052308 (2013). & Claypool, 2011). 47 Bisio, A., Dall’Arno, M. & D’Ariano, G. M. Tradeoff be- 69 Tsang, M., Nair, R. & Lu, X.-M. Quantum theory of tween energy and error in the discrimination of quantum- superresolution for two incoherent optical point sources. optical devices. Phys. Rev. A 84, 012310 (2011). Phys. Rev. X 6, 031033 (2016). 48 Dall’Arno, M., Bisio, A., D’Ariano, G. M., Miková, M., 70 Lupo, C. & Pirandola, S. Ultimate precision bound of Jeˇzek, M. & Duˇsek, M. Experimental implementation of quantum and subwavelength imaging. Phys. Rev. Lett. unambiguous quantum reading. Phys. Rev. A 85, 012308 117, 190802 (2016). (2012). 71 Nair, R. & Tsang, M. Far-Field Superresolution of ther- 49 Dall’Arno, M., Bisio, A., D’Ariano, G.M. Ideal quan- mal electromagnetic sources at the quantum limit. Phys. tum reading of optical memories. Int. J. Quant. Inf. 10, Rev. Lett. 117, 190801 (2016). 1241010 (2012). 72 Kerviche, R., Guha, S. & Ashok, A. Fundamen- 50 Lloyd, S. Enhanced sensitivity of photodetection via tal limit of resolving two point sources limited quantum illumination. Science 321, 1463 (2008). by an arbitrary point spread function. Preprint at 51 Tan, S.-H. et al. Quantum illumination with Gaussian https://arxiv.org/abs/1701.04913 (2017). States. Phys. Rev. Lett. 101, 253601 (2008). 73 Rehacek, J., Paúr, M., Stoklasa, B., Hradil, Z. & Sánchez- 52 Shapiro, J. H. & Lloyd, S. Quantum illumination versus Soto, L. L. Optimal measurements for resolution beyond coherent-state target detection. New J. Phys. 11, 063045 the Rayleigh limit. Opt. Letters 42, 231-234 (2017). (2009). 74 Yang, F., Nair, R., Tsang, M., Simon, C. & Lvovsky, 53 Zhuang, Q., Zhang, Z. & Shapiro, J. H. Optimum Mixed- A. I. Fisher information for far-field linear optical su- State Discrimination for Noisy Entanglement-Enhanced perresolution via homodyne or heterodyne detection in Sensing. Phys. Rev. Lett. 118, 040801 (2017). a higher-order local oscillator mode. Phys. Rev. A 96, 54 Zhuang, Z., Zhang, Z. & Shapiro, J. H. Entanglement- 063829 (2017). enhanced Neyman–Pearson target detection using quan- 75 Lu, X.-M., Krovi, H., Nair, R., Guha, S., Shapiro, tum illumination. J. Opt. Soc. Am. B 34, 1567 (2017). J. H. Quantum-optimal detection of one-versus-two 12

incoherent optical sources with arbitrary separation, (1998). arXiv:1802.02300 (2018). 89 Pirandola, S., Eisert, J., Weedbrook, C., Furusawa, A 76 Tang, Z. S., Durak, K. & Ling, A. Fault-tolerant and & Braunstein, S. L. Advances in quantum teleportation. finite-error localization for point emitters within the Nat. Photonics 9, 641-652 (2015). diffraction limit. Opt. Express 24, 22004 (2016). 90 Monras, A. & Paris, M. G. A. Optimal quantum esti- 77 Nair, R. & Tsang, M. Interferometric superlocalization mation of loss in bosonic channels. Phys. Rev. Lett. 98, of two incoherent optical point sources. Opt. Express 24, 160401 (2007). 3684-3701 (2016). 91 Banchi, L., Braunstein, S. L. & Pirandola S. Quantum fi- 78 Yang, F., Taschilina, A., Moiseev, E. S., Simon, C. & delity for arbitrary Gaussian states. Phys. Rev. Lett. 115, Lvovsky, A. I. Far-field linear optical superresolution via 260501 (2015). heterodyne detection in a higher-order local oscillator 92 Adesso, G., Dell’Anno, F., Siena, S. D., Illuminati, F. mode. Optica 3, 1148 (2016). & Souza, L. A. M. Optimal estimation of losses at the 79 Tham, W. K., Ferretti, H. & Steinberg, A. M. Beating ultimate quantum limit with non-Gaussian states. Phys. Rayleigh’s Curse by Imaging Using Phase Information. Rev. A 79, 040305(R) (2009). Phys. Rev. Lett. 118, 070801 (2017). 93 Fuchs, C. A. & van de Graaf, J. Cryptographic dis- 80 Paúr, M., Stoklasa, B., Hradil, Z., Sanchez-Soto, L. L. tinguishability measures for quantum-mechanical states. & Reháˇcek,ˇ J. Achieving the ultimate optical resolution. IEEE Trans. Inf. Theory 45, 1216 (1999). Optica 3, 1144-1147 (2016). 94 Sudarshan, E. C. G. Equivalence of Semiclassical and 81 Gatto Monticone, D. et al. Beating the Abbe Diffraction Quantum Mechanical Descriptions of Statistical Light Limit in Confocal Microscopy via Nonclassical Photon Beams. Phys. Rev. Lett. 10, 277 (1963). Statistics. Phys. Rev. Lett. 113, 143602 (2014). 95 Glauber, R. J. Coherent and Incoherent States of the Ra- 82 Treps, N., Andersen, U. L., Buchler, B., Lam, P. K., diation Field. Phys. Rev. 131, 2766 (1963). Maˆıtre, A., Bachor, H.-A. & Fabre, C. Surpassing the 96 Modi, K., Brodutch, A., Cable, H., Paterek, T. & Ve- Standard Quantum Limit for Optical Imaging Using Non- dral, V. The classical-quantum boundary for correlations: classical Multimode Light. Phys. Rev. Lett. 88, 203601 Discord and related measures. Rev. Mod. Phys. 84, 1655 (2014). (2012). 83 Classen, A. et al. Superresolving imaging of arbitrary one- 97 Barzanjeh, Sh., Vitali, D., Tombesi, P. & Milburn, G. J. dimensional arrays of thermal light sources using multi- Entangling optical and microwave cavity modes by means photon interference. Phys. Rev. Lett. 117, 253601 (2016). of a nanomechanical resonator. Phys. Rev. A 84, 042342 84 Tsang, M. Quantum imaging beyond the diffraction limit (2011). by optical centroid measurements. Phys. Rev. Lett. 102, 98 Barzanjeh, Sh., Abdi, M., Milburn, G. J., Tombesi, P. & 253601 (2009). Vitali. Reversible Optical-to-Microwave Quantum Inter- 85 Rozema, L. A. et al. Scalable spatial superresolution using face. D. Phys. Rev. Lett. 109, 130503 (2012). entangled photons. Phys. Rev. Lett. 112, 223602 (2014). 99 Andrews, R. W. et al. Bidirectional and efficient conver- 86 Chiribella, G., D’Ariano, G. M. & Perinotti, P. Quantum sion between microwave and optical light. Nat. Phys. 10, circuit architecture. Phys. Rev. Lett. 101, 060401 (2008). 321 (2014). 87 Bennett, C. H. et al. Teleporting an unknown quan- 100 Zhuang, Q., Zhang, Z. & Shapiro, J. H. Entanglement- tum state via dual classical and Einstein-Podolsky-Rosen enhanced lidars for simultaneous range and velocity mea- channels. Phys. Rev. Lett. 70, 1895 (1993). surements. Phys. Rev. A 96, 040304(R) (2017). 88 Braunstein, S. L. & Kimble, H. J. Teleportation of continuous quantum variables. Phys. Rev. Lett. 80, 869-872