Driving quantum systems with repeated conditional measurements

Quancheng Liu,1, ∗ Klaus Ziegler,2, † David A. Kessler,3, ‡ and Eli Barkai1, § 1Department of Physics, Institute of Nanotechnology and Advanced Materials, Bar-Ilan University, Ramat-Gan 52900, Israel 2Institut f¨urPhysik, Universit¨atAugsburg, D − 86135 Augsburg, Germany 3Department of Physics, Bar-Ilan University, Ramat-Gan 52900, Israel (Dated: April 14, 2021) We investigate the effect of conditional null measurements on a quantum system and find a rich variety of behaviors. Specifically, with a time independent H in a finite dimensional are considered with repeated strong null measurements of a specified state. We discuss four generic behaviors that emerge in these monitored systems. The first arises in systems without symmetry, along with their associated degeneracies in the energy spectrum, and hence in the absence of dark states as well. In this case, a unique final state can be found which is determined by the largest eigenvalue of the survival , the non-unitary operator encoding both the unitary evolution between measurements and the measurement itself. For a three-level system, this is similar to the well known shelving effect. Secondly, for systems with built- in symmetry and correspondingly a degenerate energy spectrum, the null measurements dynamically select the degenerate energy levels, while the non-degenerate levels are effectively wiped out. Thirdly, in the absence of dark states, and for specific choices of parameters, two or more eigenvalues of the survival operator match in magnitude, and this leads to an oscillatory behavior controlled by the measurement rate and not solely by the energy levels. Finally, when the control parameters are tuned, such that the eigenvalues of the survival operator all coalesce to zero, one has exceptional points that corresponds to situations that violate the null measurement condition, making the conditional measurement process impossible.

I. INTRODUCTION ing on null emission, the system is shelved in an coupled to the (see Fig.2, below). In The concept of quantum steering [1], namely harness- the experiment, a feedback mechanism was used to pre- ing action at a distance to manipulate quantum states vent the excursions into the non-emissive state. Thus the via measurements was introduced by Schr¨odinger[2–4]. conditional measurements, i.e. non-detection, in this case Here we investigate theoretically the effect of repeated serve as a warning that the system is moving away from measurements on a quantum system. A finite dimen- the bright state |Bi defined in Fig.2. Similar experi- sional Hilbert space is considered, and the Hamiltonian ments can be found in [11–14]. These form remarkable of the system is H. This can represent a tight-binding demonstration of quantum trajectory concepts [15–25]. quantum walk on a graph, a two- or three- level atom, More recently the interplay between unitary dynamics two entangled particles in the EPR setting [1], or a many and measurements have been emphasized [26–30], in the body system. The system is prepared at time zero in context of . It was shown how the measurement rate can modify the emerging phases of the a pure state |ψini, and then every τ units of time we perform a conditional local measurement (see Fig.1). system [31–34]. These conditional measurements drive the system, and the basic goal is to find the behavior of the system in the For a quantum system with a time independent Hamil- long-time limit. It is shown below that one may find a tonian H the energy spectrum determines the time evo- rich panoply of dynamics that depend essentially on the lution of the system, further energy is conserved. What sampling rate, the symmetry of H and the presence or will be the effect of a large number conditional null mea- absence of in the system. The possibility of surements? The latter are measurements where we do arXiv:2104.06232v1 [quant-ph] 13 Apr 2021 implementing such ideas in the laboratory is made pos- not detect the probed state. Not surprising in this pro- sible with the observation of single quantum trajectories cess of measurement the energy is not conserved, and [5–10], for example for superconducting qubits [11]. hence it is natural to wonder what it is in the long-time An example for a related experiment is the quantum limit, similarly, in principle, for other . We trajectory in the process of shelving which can be used to will show that when H has some symmetry built into reverse and reduce intermittency. Minev et al. studied it, the energy levels of the system do indeed control the a superconducting V -shaped artificial atom. Condition- long-time dynamics. However, not all the energy levels are contributing, instead only those energy levels that are degenerate. In this sense the mentioned three-level atom, which is non-degenerate can be considered as belonging ∗ [email protected] to the next class of systems. In the absence of degener- † [email protected] acy, we find other physical scenarios. For example the ‡ [email protected] system under conditions given below, can be driven to a § [email protected] unique state, like shelving in the mentioned experiment. 2

However, we identify this state in generality not as an energy state of the system, but rather it is an eigenstate No No No No of the survival operator, more specifically the state with the largest eigenvalue (see below). In other cases to be classified below, we find that the system exhibits peri- M M M M odicities. However, unlike unitary , U(τ) U(τ) U(τ) these oscillations are not determined by the energy levels t = 0 t = τ t = 2τ t = 3τ t = nτ alone. Instead the natural frequencies are controlled by ⋯ ⋯ the measurement rate as we shall show. Finally, there is a FIG. 1. Schematic depiction of the quantum driving via null fourth class of systems, where we find exceptional points detections. The quantum system is initially prepared at some [35–44]. The latter are well investigated, for example pure state. Local measurements of a target state |ψdi are in the context of photonics [45–49] and they describe in performed at discrete times t = τ, 2τ, ··· , nτ, ··· , whose out- general degeneracies of non-Hermitian system [50, 51]. comes are pinned down to null. We investigate the state of Here we show how at these points, corresponding to a the quantum system under such repeated steering. special choice of sampling time τ, eigenvalues of the sur- vival operator coalesce. Interestingly exceptional points with a summary in Sec.X. are found for systems where the very basic condition of null measurements is not a possibility. This means that when we have such exceptional points we cannot impose II. THE SURVIVAL OPERATOR the condition in the first place. Instead we may work only close to this limit. As mentioned, we are interested in the evolution of At the heart of the analysis is the investigation of the the of a system conditioned on null mea- left and right eigenvectors and eigenvalues of the survival surements. The model describes unitary dynamics inter- operator. This operator describes the unitary evolution leaved with measurements. The system is discrete and followed by the local measurement. This operator has finite, and H is the Hamiltonian, which governs the dy- been previously investigated in the context of the quan- namics between measurement events. At the times τ, tum first detection problem [52–60]. Roughly speaking, 2τ, 3τ, ··· we perform measurements in an attempt to in the latter case one is interested in the statistics of detect the system in the target state |ψdi. For example, the first detection event, namely when do we detect the for a tight-binding quantum walk on a tree (see Fig.7), system for the first time? Here we are interested in the discussed below, |ψdi describes a localized node on the state of the system itself after many measurements condi- graph, or |ψdi can be an energy level (see Fig.2), etc. τ is tioned that we never detect. Similar to the original idea a free parameter. The outcome of a given measurement, of Schr¨odinger,we consider two sets of measurements, the at least in principle, is either the system is detected at first is a set of many local measurements, this then drives the target state |ψdi or not. We condition the measure- the system to some state. The second measurement is of ments and consider only the realization of the process another operator, say H itself, and it is neither condi- described by the string: no, no, ··· , no, which repeats tioned nor local. Now if we had n null measurements, n times, namely the particle is never found in the target prior to the final one, then obviously this influences the state, e.g. the quantum walker is not detected on the statistics of the final unconditional measurement. The specified node of the graph. focus on the final state of the system is motivated in Therefore, the wave function after the first measure- part from our interest in statistical physics, where the ment is |ψ1i = N1(1 − D)Uˆ(τ)|ψini. Here, D = |ψdihψd| long-time limit of a driven system is of special interest, is the projection operator onto the measured state, while and energy is the natural choice of . Uˆ(τ) = exp(−iHτ) is the unitary evolution operator, and The manuscript is organized as follows. Secs.II and we have set ~ = 1. The operator III give the model, general discussion of the mathemati- ˆ ˆ cal properties of the survival operator, a mapping of the S = (1 − D)U(τ) (1) quantum problem to classical charge theory, and a simple incorporates the unitary evolution in the time interval discussion on the exceptional points. In Secs.IV andV τ followed by the projection described by 1 − D. The we present the main results, namely the state function latter, when acting on the state function, wipes out the under conditional measurements, for finite and infinite amplitude on the detected state, as the result of the mea- measurement times respectively. In Secs.VI andVII, surement is null. The operator Sˆ is called the survival we present three examples, a three-level system, an arti- operator as it describes the wave function that survives ficial atom that can shelve, and a glued binary tree with the detection (if the particle had been detected, the wave degeneracy. In Sec.VIII, we present general insights function would have collapsed on |ψdi). Clearly, it is a on the final state of the system with a perturbation ap- non-unitary operator. Immediately after the nth mea- proach. Then in Sec.IX, we discuss exceptional points in surement, the wave function |ψni reads: a more general setting from the point of view of the sym- ˆ n metry of charge configuration. We close the manuscript |ψni = NnS |ψini. (2) 3

These right eigenstates satisfy

ˆ R S|ξ i = exp(iθ) |ξRi. (5) | {z } ξ

We express these states in terms of linear combinations of stationary states of the time independent Hamiltonian H. For that denote H|Ek,li = Ek|Ek,li, where k is the index of the distinct energy levels, while l is a quantum FIG. 2. Three-level V -shaped system. We initially prepare number denoting the degenerate sub-levels, so that l = the system in the ground state |Gi. The matrix element of 1, ··· , gk, where gk is the degeneracy of the energy level H describing jumping between |Gi and |Di (|Bi) is γ1 (γ2), E . Degeneracy and hence symmetry play a crucial role where γ  γ . The detection state is the state |Bi. k 2 1 here. ˆ An obvious eigenstate of S is an energy state |Ek,li which is orthogonal with respect to the detected state, Below we suppress the index n in the normalization Nn. namely if we have in our system a state hψd|Ek,li = 0 we We are interested in the large-n limit. In this limit, get we encounter a variety of dynamical features explored below. In one case, the dynamics are determined by a (1 − D)Uˆ(τ)|Ek,li = exp(−iEkτ)|Ek,li. (6) sub-set of energy levels, in another by a fixed point where the system has no dynamics at all. Finally, we observe Hence all energy states which are orthogonal to the de- oscillatory behavior; however, unlike the first case sce- tected state are right eigenvectors of the survival operator R nario, the eigenvalues determining this behavior are not |ξ i = |Ek,li with an eigenvalue ξ = exp(−iEkτ). Phys- standard energy levels of H. They are controlled by H, ically this state corresponds to a dark state [8, 60, 62]. D, and τ. Our goal is to classify these behaviors and Namely if such a state exists, and if it is chosen as the see when they emerge. What is clear is that the large-n initial condition, it will never be detected. So, in this limit is described by the set of eigenvalues of Sˆ of largest case, the condition of null measurements is guaranteed magnitude; hence this is what we study next. from the start. Similarly, it is easy to see that under the As usual, since the operator Sˆ is non-Hermitian it has same condition, i.e., hψd|Ek,li = 0, the corresponding left L left and right eigenvalues and eigenvectors and these are eigenvector is hξ | = hEk,l|. We see in this special exam- denoted ple a property which is unique to states with eigenvalues on the unit circle, namely |ξLi = |ξRi. This feature is of Sˆ |ξRi = ξ|ξRi, hξL|Sˆ = hξL|ξ. (3) some importance later. Our goal now is to consider other dark states, that are Here we used the fact, well known from linear algebra, eigenstates of the survival operator, with eigenvalues on that the left and right eigenvalues are equal; hence they the unit circle. Recall that such states obviously have are denoted with ξ. These are complex numbers, and for the largest possible eigenvalue and hence dominate the the problem at hand, it was shown that these eigenvalues large-n limit, if they exist. We use the energy represen- are all on the unit disk |ξ| ≤ 1 [60, 61]. We will use tation and consider a degenerate subspace in the Hilbert R R L L space namely {|E i, ··· , |E i} and by definition all normalized states hξ |ξ i = hξ |ξ i = 1. k,1 k,gk these states have the same energy E . If some of them We start the analysis by considering three types of k satisfy hψ |E i = 0 they are clearly dark as we have just eigenvalues. The case ξ = 0, the case 0<|ξ|<1, and then d k,l explained. Consider now a of two en- |ξ| = 1. When the eigenvalue is zero, we have ergy states and we index these with l = 1 and l = 2. We ˆ R −1 L then find the eigenvector of S |ξ i = Uˆ |ψdi, hξ | = hψd|, (4) R |ξ i = N(hψd|Ek,2i|Ek,1i − hψd|Ek,1i|Ek,2i). (7) −1 which is obvious due to (1 − D)UU |ψdi = 0 and R ˆ R R R hψd|(1 − D) = 0. The physical meaning of the |ξ i state Here clearly S|ξ i = exp(−iEkτ)|ξ i since U|ξ i = R R is clear, as it is a state that is detected with probabil- exp(−iEkτ)|ξ i and D|ξ i = 0, and so the eigenvalue ity one in the first measurement attempt. In fact, if we of the survival operator is ξ = exp(−iEkτ), namely it −1 start in this state, namely |ψini = U |ψdi we cannot is just a phase determined by the energy Ek. Similarly, achieve the desired conditional measurement, since the it is easy to see that hξL| = hξR| is a left eigenvector of first measurement is always a successful. This implies the survival operator, and hence hξL|ξRi = 1. So these that not all initial conditions can yield a string of n null vectors are parallel, unlike the left and right vectors we measurements, an issue we will return to below. found above with zero eigenvalue ξ = 0 (and unlike the More profound are the states with eigenvalues |ξ| = 1 eigenvectors we discuss below with |ξ|<1). The state in on the unit circle. This set can be empty as we show Eq. (7) is clearly dark in the sense that if we start in this below, but for now we assume that such states exist. state, the amplitude on the detected state is always zero. 4

This, in turn, is because this is a stationary state of H, gk − 1 such eigenvectors that are given in Eq. (8) and hence between measurements, the dynamics do not al- they all share the same eigenvalue exp(−iEkτ). low the leakage of to the detected For example, consider a system with three distinct en- state, so we never detect the system. Once again, by a ergy levels, the first with degeneracy 10, the second with dark state, we mean that even if we do not condition the degeneracy 5, and another with no degeneracy. Further measurements to be null, the sequence of measurements assume for simplicity that all the energy states have fi- will never detect the particle. nite amplitude on the detected state. Eq. (8) specifies 13 We can easily construct other normalized dark states orthonormal eigenstates of the survival operator, all with using similar methods. We denote these states, i.e. states eigenvalues on the unit circle. These are dark states as that cannot be detected as |δk,mi (δ is for dark). Fol- mentioned, namely, states that cannot be detected even lowing [60], we find that in an energy subspace with if we do not condition the measurements. However, if we gk states, there are gk − 1 dark states {|δk,mi} with consider the case where we choose the detected state as m = 1, ··· , gk − 1. These states are given by a Gram the energy eigenstate of the non-degenerate level, it then Schmidt procedure [60, 62–65] immediately follows from standard quantum mechanics that the remaining 15 states are all dark as they are or- m " # X thogonal to the detected state. Hence, these 15 states |δ i = N |α |2|E i − α∗ α |E i , k,m k,j k,m+1 k,m+1 k,j k,j are dark, and they are also eigenvectors of the survival j=1 operator Sˆ . (8) If the system has no degeneracy, for example, a typ- where α = hE |ψ i and m goes from 1 to g − 1. k,m k,m d k ical random system with no symmetry in H, and if all Here |E − E |τ 6= 2πj, where j is an integer. All these k i the energy states have finite overlaps with the detected states are orthonormal with respect to each other, and one (even small), we reach the conclusion that we have again they are stationary states of the system, namely no dark states, and hence we do not have eigenstates of H|δ i = E |δ i. It is easy to see that these states k,m k k,m the survival operator with eigenvalues |ξ| = 1. This in- are eigenstates of the survival operator with eigenvalue dicates that the dynamics of systems with disorder and ξ = exp(−iEkτ), namely these eigenvalues all fall on the ˆ those without, under the condition of repeated null mea- unit circle, since S|δk,mi = exp(−iEkτ)|δk,mi. It is also L R surements, can have very different properties. easy to show that the left eigenvectors satisfy hξ | = hξ |. There exists, of course, a third family of states, which We have seen how dark states can be presented as are those with 0<|ξ|<1, namely the eigenvalues are all in energy eigenstates of H but also as eigenstates of the the unit disk. There is no simple way to determine the survival operator Sˆ . The moduli of the corresponding ˆ values of ξ and find the corresponding states. However, eigenvalues of S are unity. As mentioned, with a gk-fold there is an elegant method to find these eigenvalues in degenerate energy subspace {|E i, ··· , |E i}, we can k,1 k,gk principle based on a classical charge theory [55], which generate gk −1 dark states. In this subspace, there exists is discussed below. With this classical theory, we will be one additional state, which we call the bright state, and able to gain physical insight on the largest eigenvalues it is given by [60] of systems with no dark states, which in turn will give the long-time dynamics of the conditioned measurement |β i = NPˆ |ψ i, (9) k k d process. ˆ Pgk where Pk := m=1 |Ek,mihEk,m| is the eigenspace pro- jector. This state is not an eigenvector of the survival operator. Its mathematical property is that it is orthog- III. CLASSICAL CHARGE PICTURE MAPPING AND BIORTHOGONAL EIGENSTATES onal to all the gk − 1 dark states in its energy sector. This state, under repeated measurements, without con- ditioning the outcome, will be eventually detected with As previously mentioned, the dark states are eigen- probability one; hence it is called a bright state. In the functions of survival operator with a special property; next section, we will show how to use these bright states their eigenvalues are on the unit circle. However, the to construct the eigenstates of the survival operator Sˆ dark states are certainly not the only part of the eigen- with eigenvalues in the unit disk, i.e., 0<|ξ|<1. states of the survival operator, namely, the dark space To conclude this part, we see that the operator Sˆ has HD := Span({|δk,mi}) is a subspace of the full Hilbert an eigenvalue zero and we have formally found the nor- space H, so we need to search for other eigenstates. The malized left and right eigenvectors which are not parallel eigenvalues of the remaining states, which are now in- hξL|ξRi= 6 1. Using the energy presentation, we see that vestigated, all fall in the unit disk. Using the matrix dark states of the system are eigenvectors of the sur- determinant lemma [66], the eigenvalues of the survival ˆ vival operator, with eigenvalue ξ = exp(−iEkτ). Clearly operator S are given by (see AppendixA) |ξ| = 1, so these have the maximum possible eigenvalue ˆ −1 magnitude, since in general |ξ| ≤ 1. In this case the det[ξ1 − S] = det[ξ1 − Uˆ(τ)]hψd|[ξ1 − Uˆ(τ)] |ψdiξ = 0. left and right eigenvalues are parallel hξL|ξRi = 1. In a (10) degenerate sector {|Ek,li} of the Hilbert space, we have We observe the following: 5

FIG. 3. Schematic depiction of the classical charge mapping [55](a) and the eigenvalues of survival operator Sˆ (b). We plot a system with energy levels E0, ··· ,E4, where all the energy levels are two-fold degenerate, except for E0 which has not degeneracy. According the the charge picture, we have five charges on the unit circle, as shown on the left, the charges are p0, ··· , p4. These charges are located on exp(−iEkτ) from Eq. (16). All these charges are positive, and hence this gives 4 stationary points in the unit disk denoted ξ1, ··· , ξ4 in (a). These, as explained in the text, give the non-zero eigenvalues of the survival operator in the unit disk 0<|ξi|<1. In (b), we plot all the eigenvalues of Sˆ . There are three types of eigenvalues, i.e., |ξ| = 1, 0<|ξ|<1, and ξ0 = 0. The degeneracy of the energy levels leads to the dark states, which are eigenstates of Sˆ with eigenvalues exp(−iEkτ) on the unit circle (b). The eigenvalues in the unit circle are given by the charge picture, and finally there is one eigenvalue that is at the center of the unit disk. Note that the non-degenerate energy level E0 does not contribute a dark state.

(i) There is a stationary point, denoted ξ0, located at Using Eqs. (10, 11, 12, 13) , the equation for ξs reads: the origin, namely ξ0 = 0 (see the last term in Eq. (10)). We have found its left and right eigenvectors in the former w−1 X 2 −iEkτ gk−1 −iEiτ gi section. |hβk|ψdi| (ξ − e ) Πi6=k(ξ − e ) = 0. (ii) For simplicity, let us assume all elements of the k=0 (14) energy basis {|Ek,li} have finite overlaps with |ψdi. Using this energy basis, we find: From here we see that if an energy level is degenerate ˆ gk>1, then we find the survival operator S has eigenval- w−1 −iEkτ gk det[ξ1 − U(τ)] = Πk=0 (ξ − e ) . (11) ues ξ = exp(−iEkτ), which is (gk − 1)-fold degenerate. We have here a multiplication of $ terms (ξ − e−iEkτ ), In contrast, if gk = 1, ξ = exp(−iEkτ) is clearly not ˆ where $ is the size of the full Hilbert space and w is the an eigenvalue of S. This coincides with what we have Pw−1 found in the previous section—any degenerate subspace number of distinct energy levels, namely gk = $. k=0 with energy E has g − 1 dark states and one bright Only in the absence of degeneracy, gk = 1, we have $ = k k w. Here following Gr¨unbaum et al. [55, 56], we use the state. The (gk − 1)-fold eigenvalues of the survival op- symbol w, as this is actually describing a certain winding erator ξ = exp(−iEkτ) correspond to gk − 1 dark states number of the problem. Naively, it seems from Eq. (11) found in Eq. (8). The total number of dark states is Pw−1 that we already have all $ ξs that fit Eq. (10). However, k=0 (gk − 1) = $ − w. These states, have the largest −1 ˆ this intuition is not true, because hψd|[ξ1 − Uˆ(τ)] |ψdi possible eigenvalue magnitude |ξ| = 1 of S, hence they cancels part of them, as we now show. will control the long-time dynamics of the conditional w−1 (iii) The bright and dark space, i.e., {|βiii=0 } and measurements process, see below. For disordered, in- {|δi}, form a complete basis. Importantly |ψdi is orthog- teracting, or chaotic systems, where repulsion of energy onal to the dark states by definition. It follows that we levels arises, namely gk = 1 for all the energy levels, we can expand the detected state in terms of bright subspace cannot find eigenvalues on the unit circle in general. Now Pw−1 only $ = w and k=0 (gk − 1) = 0. w−1 (iv) The other eigenvalues 0<|ξι|<1, are given by X |ψ i = hβ |ψ i|β i. (12) d k d k ˆ −1 k=0 hψd|[ξ1 − U(τ)] |ψdi = 0. (15) The bright states {|β iw−1} are also eigenstates of H and k l=0 It turns out that there are w−1 ξs, that satisfy 0<|ξι|<1. hence of Uˆ(τ). It follows that So we see, we have $ − w eigenvalues on the unit circle |ξ| = 1, w − 1 in the unit disk 0<|ξ|<1, and one with w−1 2 −1 X |hβk|ψdi| ξ = 0. ($ − w) + (w − 1) + 1 = $, all the eigenvalues of hψd|[ξ1 − Uˆ(τ)] |ψdi = . (13) ξ − e−iEkτ ˆ k=0 survival operator S are found. 6

The question remains, how to find the eigenvalues sat- Uˆ(τ). Note that the right eigenvectors of Sˆ do not have R R isfying 0<|ξι|<1? Here we exploit a beautiful mapping to be orthogonal with each other, i.e., hξι |ξι0 i= 6 δι,ι0 . R of the problem to a classical charge theory, following In Eq. (17), the {|ξι i} are presented in an energy ba- the work of Gr¨unbaum et al. [55, 67, 68]. By defin- sis, in which the evolution operator Uˆ(τ) is diagonal. Us- ing p = hψ |Pˆ |ψ i = Pgk |hE |ψ i|, together with Pw−1 ˆ ˆ −1 k d k d i=1 k,i d ing the identity j=0 Pj = 1, we have [ξι1 − U(τ)] = w−1 w−1 Eqs. (9), (13) and (15), we get P ˆ ˆ −1 ˆ P ˆ −iEj τ j,j0=0 Pj[ξι − U(τ)] Pj0 = j=0 Pj/(ξι − e ) = w−1 † w−1 P Rˆ . So X pk j=0 ι,j F(ξ) = = 0. (16) ξ − e−iEkτ k=0 R ˆ −1 |ξι i = N[ξι1 − U(τ)] |ψdi, (18)

F(ξ) can be considered as a two-dimensional electrostatic independent of any representation. The geometry of the −iEkτ field created by point charges pk at positions e (see right eigenstates is that they are all orthogonal with re- Fig.3). These charges produce a log potential, namely spect to the detection state. This is easy to see using Eq. they can be viewed as long wires piercing the unit disk. (1), and hψd|(1 − D) = 0. Eq. (16) defines the stationary points of this classical Here we present a direct proof to see that Eq. (18) field. Namely, when we put a test charge into this field, is indeed the right eigenvector of Sˆ . We start from the the stationary points {ξ } are the locations where the net ι definition of the right eigenvectors. Using Eq. (3), we force on the test charge is zero. All the {ξ } are inside ι have: the unit disk (|ξι| < 1), which is rather obvious since all the charges p are positive. k (Uˆ − ξ 1)|ξRi = |ψ ihψ |Uˆ|ξRi. (19) To summarize to find the eigenvalues according to the ι ι d d ι classical charge picture, we consider: i) w charges that Inserting Eq. (18) into Eq. (19), we get are placed on the unit circle, where w is the number of distinct energy levels of H; ii) these charges are located Uˆ at the phases exp(−iEkτ), where Ek are the energy levels − |ψdi = |ψdihψd| |ψdi, (20) ξ 1 − Uˆ of H; (iv) the charges have magnitude pk; (v) with these ι positive charges, we have w −1 zeros of the force field, all ˆ ˆ inside the unit disk; (vii) once we calculate these zeros, which leads to −1 = hψd|U/(ξι1 − U)|ψdi. Using ˆ ˆ ˆ these are the eigenvalues 0<|ξ|<1 we are looking for. The U/(ξι1 − U) + 1 = ξι1/(ξι1 − U), we get for ξι 6= 0 charge theory is useful as it allows us to easily identify −1 the largest eigenvalue of the survival operator, in certain hψd|[ξι1 − Uˆ] |ψdi = 0. (21) limits of the problem discussed below. From Eq. (15), this is the equation for the eigenvalues of the survival operator Sˆ that are in the unit disk, i.e., A. Eigenvectors of the survival operator 0<|ξ|<1. Hence, Eq. (18) is correct. Following the same procedure, the left eigenvectors of Now we formally find the eigenvectors of Sˆ that corre- survival operator read: spond to the eigenvalues that are in the unit disk. Here w−1 ˆ −iEj τ we take advantage that the dark states are already eigen- L X Pje ˆ hξι | = N hψd|Lι,j,Lι,j = −iE τ . (22) states of S that correspond to eigenvalues on the unit ξι − e j circle. Furthermore, as mentioned, the dark and bright j=0 subspace form a complete basis. With these two pre- Similar to Eq. (18), we have conditions, we expand |ξRi in terms of the bright states, under the condition that the eigenvalue of |ξRi is in the hξL| = Nhψ |Uˆ(τ)[ξ 1 − Uˆ(τ)]−1. (23) unit disk. As presented in AppendixB, we find: ι d ι

w−1 ˆ The right/left eigenvectors obey the biorthogonal rela- X † † Pj L R R ˆ ˆ tion hξ |ξ 0 i = Nιδι,ι0 . Here Nι is a constant which de- |ξι i = N Rι,j|ψdi, Rι,j = . (17) ι ι ξ − e−iEj τ j=0 ι pends on how we normalize the two right/left vectors. Furthermore, the right/left eigenvectors are orthogonal R to the dark states hδ |ξRi = 0 and hξL|δ i = 0, since Using Eq. (9), we see that the states {|ξι i} indeed are k,j ι ι k,j linear combinations of the bright states {|βki} as stated. the dark subspace is orthogonal to the bright subspace. So |ξRi is a bright state as well, namely in the absences of conditioning, starting in this state, the detection (a click yes) is eventually guaranteed. Though both {|βki} and B. Phase gained under repeated measurements R {|ξι i} are bright states, the latter are eigenstates of the ˆ survival operator S with corresponding eigenvalues {ξι} We now explain the physical meaning of ξ. Using Eqs. in the unit disk, while the former are eigenstates of H and (18) and (23), if the system starts with the normalized 7

R R right eigenstate |ξι i, i.e, |ψini = |ξι i, we have (Ap- and hence for the second eigenvalue ξ = cos(γτ) and pendixC) vector |ξRi = |ri, we have Uˆ(τ)|ξRi = i sin(γτ)|li + L ∗ ˆ cos(γτ)|ri = |ξ i . Then the conditional measurement R U(τ) L ∗ (1−D) R R Sˆ R |ξι i −→ |ξι i −→ |ξι i, |ξι i −→ |ξι i, (24) (1 − D) wipes the amplitude on |li, which leads to (1 − D)|ξLi∗ → |ri = |ξRi. This is an illustration of here ∗ is the complex conjugate. And we use the fact Eq. (24). that |ξRi is an eigenstate of Sˆ , so Sˆ |ξRi → |ξRi. Using ι ι ι According to our formalism, in the large-n limit the Eq. (2), the wave function of the system after n mea- state of the system is determined by the eigenvector cor- surements reads: responding to the eigenvalue closest to the unit circle, n R |ψni = Nnξι |ξι i, (25) i.e. max(0, | cos(γτ)|) which, unless | cos(γτ)| = 0, means limn→∞ |ψni = |ri. This is easy to understand, as af- where Nn is normalization. In this process, the opera- ter the null measurements, the system cannot be on |li. ˆ tions U(τ) and (1 − D) send the wave function to the Since there are only two states, we find the system after L ∗ R |ξι i and |ξι i n times. Importantly, we gain an unusual the n-th measurement on |ri. This holds true for any n. phase, namely The exceptional point is found when cos(γτ) = 0, and for this case Sˆ cannot be diagonalized at all. Three |ψ i = exp(inφ)|ξRi = exp(inφ)|ψ i , (26) n ι in features are found: a) in this limit, Eq. (28) gives hξL|ξRi = 0, b) the two eigenvalues coalesce on zero, where φ = −i ln(ξι/|ξι|). This phase is not determined by the energy times evolution time, but is given by the phase i.e. ξ = 0, finally c) the two pairs of vectors, both L and ˆ R, in Eq. (28) become parallel (in our representation of the complex eigenvalue ξι of the survival operator S. As we repeat the measurements, the system is driven besides an i they are identical). Thus at the exceptional periodically, and the phase accumulated is nφ. sampling time τ = (π/2 + kπ)/γ, where k is an integer, we have only one eigenvector instead of the familiar two. This means that both the eigenvalues and the eigenvec- C. Exceptional points: example of a two-level tors coalesce, which implies an exceptional point. Math- system ematically, when cos(γτ) = 0, the survival operator now reads In the above discussion we assumed that the system 0 0 Sˆ = (1 − D)Uˆ(τ) = (30) has no exceptional points. Exceptional points are cases, i 0 where for specific τ, or for certain control parameters of H, we have two (or more) eigenvalues ξ (|ξ|<1) of the and then the eigenvalues are clearly zero, and we have survival operator Sˆ merging, a kind of degeneracy found only one eigenvector solution instead of two. for non-Hermitian physical systems [69]. In this subsec- How can we search for these exceptional points without tion, we illustrate part of the effects with the simplest solving the problem exactly? We claim that we can use example—the two-level system, also presenting the ex- the classical charge picture, which is very useful in deter- ceptional point and its meaning. More general discussion mining the exceptional points in general. Let us present about the exceptional points is presented in Sec.IX. it for the two-level system so that the reader can get For a two-level system with state |li (left) and |ri familiarized with the basic ideas. We allocate the classi- (right), the Hamiltonian reads H = −γ(|lihr| + |rihl|) cal charges on the unit circle. The location of the pair where γ is the hopping amplitude between these two of charges, are on the phases exp(−iE1τ) = exp(iγτ) states. The energies in this system are E1 = −γ and and exp(−iE2τ) = exp(−iγτ). The magnitudes of the E2 = γ. We assume that we detect the particle on the charges in our example are identical and positive. Then left node, so D = |lihl|. In this basis the survival operator we search for the stable point, namely the point in space is where the classical force field of the charges vanishes. This is clearly on the midpoint connecting two charges.  0 0  Sˆ = (1 − D)Uˆ(τ) = . (27) From basic geometry, we realize that this classical sta- i sin(γτ) cos(γτ) ble point is found in this example on ξ = cos(γτ) in We easily find the following eigenvalues and vectors of Sˆ the complex plane. Especially, if we choose τ, such that these charges are on the north and south poles, the stable ξ = 0 : hξL| = hl|, |ξRi = cos(γτ)|li − i sin(γτ)|ri; point is on the origin, in agreement with Eq. (28). This ξ = cos(γτ): hξL| = i sin(γτ)hl| + cos(γτ)hr|, |ξRi = |ri. describes the coalescence at the stationary point. Dy- (28) namically we imagine the eigenvalue cos(γτ), moving to We see from here the features discussed above, for ex- the origin where it fuses with the ever-present eigenvalue ample for the eigenvalue ξ = 0 the left eigenstate is ξ = 0 there. By moving, we mean that we think of the hξL| = hψ |, which in this model is hξL| = hl|. Further process as we tune γτ. We may control this fusing process d with the rearrangement of the charges on the unit circle,  cos(γτ) i sin(γτ) i.e., moving them to the south and north pole. The map- Uˆ(τ) = , (29) i sin(γτ) cos(γτ) ping of the problem to a classical charge theory allows 8 for an intuitive understanding when exceptional points tioned the measurements to give a no when we detect on emerge, in generality, and shows how this is related to |li). However, when cos(γτ) = 0, we will find the state the symmetry of the problem. of the system just before the second measurement in the state |li. We then perform the second measurement and For the two-level system, what is the physical meaning find the system in the detected state |li with probability of the exceptional point? When the system is tuned to one. It follows that we tried to impose the sequence of this point, and we attempt to follow the null measure- null measurements. However, at the exceptional points, ments, we encounter a problem. After the first measure- this is impossible since we always get a yes in the second ment, we get a null result, as this is the condition im- measurement. This feature is generic, and found below posed by the rules of the model. But now, after the first in systems whose complexity exceeds that of a two-level measurement, the system is in state |ri (since we condi- example.

IV. STATE FUNCTION UNDER CONDITIONED MEASUREMENTS

A useful identity, which holds for non-Hermitian operators, and not limited to our case reads [70–72]

X |ξRihξL| = 1. (31) hξL|ξRi ξ This formula is valid provided that we do not have exceptional points in the system. The summation is over all the states in the system, denoted by the sum over the index ξ. It follows from Eq. (31) that a generic initial condition can be expanded like

L R X hξ |ψini|ξ i |ψ i = . (32) in hξL|ξRi ξ For the case under study, we split the sum into three parts, those with eigenvalues on the unit circle |ξ| = 1, ξ = 0, and all the rest. L hψd|ψini −1 X hξ |ψini R X |ψini = −1 U |ψdi + L R |ξ i + hδ|ψini|δi. (33) hψd|U |ψdi hξ |ξ i | {z } 0<|ξ|<1 δ,|ξ|=1 ξ=0

−1 The first term, proportional to U(τ) |ψdi, has a clear physical meaning, since after the first time interval τ, the −1 unitary evolution yields UU |ψdi = |ψdi. The first detection, is modeled with the projector (1 − D) and hence this −1 first term is wiped out by the first measurement. As mentioned, if |ψini = Uˆ |ψdi, we cannot get the conditional measurements, namely for this initial condition, we record a yes at the first measurement attempt, so it is important −1 for our discussion, that the second and third terms are non-zero. In Eq. (33), we assume hψd|U |ψdi= 6 0 and hξL|ξRi= 6 0, otherwise we get an exceptional point. Using Eqs. (2), (3) and (33), the wave function after n ≥ 1 measurements reads   L  X hξ |ψini X  |ψ i = N ξn |ξRi + exp(−iE nτ)hδ|ψ i|δi , (34) n hξL|ξRi δ in 0<|ξ|<1 δ,|ξ|=1  and N is the normalization. In the next section we use this equation to explore the large-n limit of the system. Here we see that in general the state of the wave function is composed of dark states |δi with eigenvalues exp(−iEδτ), and bright states |ξRi with eigenvalues 0 < |ξ| < 1. The effect of the state with eigenvalue ξ = 0 is washed out after the first measurement. We now consider the mean energy of the system. Even though the particle is not detected, there must be an exchange of energy between the system and the detector, hence the energy of the particle is not constant. Initially, the expectation of the energy of the system is E0 = hψin|H|ψini. With the measurements going on, the averaged energy of the system after the n-th measurement is hEni = hψn|H|ψni. Using Eq. (34), we have:

P ∗ n 0∗ n 0R R P 2 0<|ξ,ξ0|<1 aξaξ0 ξ (ξ ) hξ |H|ξ i + δ,|ξ|=1 |hδ|ψini| Eδ hEni = P ∗ n 0∗ n 0R R P 2 , (35) 0<|ξ,ξ0|<1 aξaξ0 ξ (ξ ) hξ |ξ i + δ,|ξ|=1 |hδ|ψini| 9

L L R 0R R Pw−1 0 iEkτ −iEkτ where aξ = hξ |ψini/hξ |ξ i, and hξe |H|ξ i = k=0 pkEk/[(ξe − e )(ξ − e )]. If initially the system is prepared in a linear combination of dark states, then clearly the energy is a constant of motion, but otherwise it is not. In particular if we have no dark states in our system, namely no symmetry and hence no degeneracy, the measurements never conserve energy. Note that in this case the mean energy can saturate for large n, or exhibit dynamical oscillations, which are explored below.

V. THE FIXED STATE AND QUANTUM above case (B1), but also exhibits interference due to the DYNAMICS relative phases of these eigenvalues. As an example, consider two eigenvalues |ξ1| = From Eq. (34), in the large-n limit, we reach several |ξ2|>|ξi| (i 6= 1, 2) and |ξ1| = |ξ2| < 1, we denote iφ1 iφ2 general conclusions: ξ1 = re and ξ2 = re , where r = |ξ1| = |ξ2|, (A) If the dark subspace is not empty, namely if we φ1 = −i ln(ξ1/|ξ1|), and φ2 = −i ln(ξ2/|ξ2|). Using Eq. have some degenerate energy levels (implying some sym- (34), the terms with eigenvalues ξi (i 6= 1, 2) decay away, n inφ1 R metry in the system) or if one of the stationary states of and we are left with |ψf i = |ψni ∼ N(a1r e |ξ1 i + n inφ2 R n inφ1 R inφ2 R H is orthogonal to |ψdi, the second term in Eq. (34) dom- a2r e |ξ2 i) = Nr (a1e |ξ1 i + a2e |ξ2 i) (as a n inates the large-n limit provided that the initial condition reminder, ai is the overlap with the initial state). r can |ψini has some overlap with the dark states. This means be absorbed by the normalization. In the end, we have that the long-time dynamics is determined by the phases φ1+φ2 φ1−φ2 φ2−φ1 exp(−iEδnτ). Compared to unitary dynamics, where all in 2 in 2 R in 2 R |ψf i ∼ Ne (a1e |ξ i + a2e |ξ i) . the energy levels contribute, the effect of measurement 1 2 is therefore to remove the non-degenerate energy levels (38) from the long-time dynamics. The wave function then Comparing with Eq. (37), when there are two ξs that reads: dominate the large n evolution, the system not only gains a global phase (φ1 +φ2)n/2 in the measurements process,

X −iEδ nτ but also exhibits an oscillation that is controlled by the |ψf i = lim |ψni ∼ N e hδ|ψini|δi . (36) n→∞ relative phase (φ1 −φ2)n/2 of ξ1 and ξ2. We can tune this δ,|ξ|=1 relative phase by changing the sampling time interval τ (see Sec.VIII). Mean observables, like the energy, may (B) In the case when the dark subspace is empty, as thus exhibit periodic controllable oscillations, see our ex- typically found for systems without special symmetries, amples below. In this process, the system are steered we find two types of behaviors. We first need to deter- periodically by the repeated measurements, and we call mine the maximum of the set {|ξ |} and recall that these ι this interesting phenomena quantum dynamics induced absolute values are all less than unity. Then: by the measurements. (B1) If the maximum is unique, the system in the long- time limit goes to a specific state. There is no dynamics in this limit as only one term dominates. This is similar VI. THREE-LEVEL SYSTEM, V -SHAPED to a fixed point. We denote this unique largest eigenvalue ˆ SYSTEM of survival operator S as ξf . Using Eq. (34), we have The two-level system example, treated in Sec.IIIC, is |ψ i ∼ Nξn|ξRi = einφf |ξRi , (37) f f f f obviously a very special case, and misses a lot of the dif- ferent dynamical regimes. We now consider a three-level where φf = −i ln(ξf /|ξf |). The system ends up with R system, with no dark states. According to the theory de- the state |ξ i and attains a global phase nφf . In Eq. f veloped, we will have two eigenvalues ξ1 and ξ2, besides (26), the phase φi of the system is determined by the ξ0 = 0, in the unit disk and the long-time limit of the sys- initial state, while here the phase φf is determined by tem will be determined by the eigenvalue which is larger the maximum of the set {|ξι|}. Furthermore, for a generic in magnitude in absolute value sense. When we vary, say initial state, at the first stages of the processes (n is not τ, we can find different cases. When |ξ1| = |ξ2|= 6 0, large) many phases φi = −i ln(ξi/|ξi|) contribute, see the relative phase of these eigenvalues will play a special R Eq. (34). As the component |ξf i gradually wins the role, namely we then expect oscillatory behavior to per- game, the corresponding phase becomes the only term sist forever. This is investigated here with an example ˆ that dominates (n is large). Each application of S rotates using the charge picture. the phase of the system by φf . We consider a three-level system with the Hamiltonian (B2) If the maximum of the set {|ξι|} is shared by sev- eral eigenvalues of the survival operator Sˆ , all with the H = −γ(|0ih1| + |1ih0| + |1ih2| + |2ih1| + |0ih0|). (39) same absolute value but with different phases, then these eigenvalues will control the large-n limit of Eq. (34). In Note that we have added an onsite energy −γ on the this case, the wave function, for large n, will exhibit dy- node labeled |0i. The spectrum of H is given by 3 2 namic. The system not only gains a global phase like the Ei − Ei − 2Ei + 1 = 0, where we have set γ = 1. We 10

FIG. 4. The absolute values of the eigenvalues ξ1 (red) and ξ2 (blue) versus τ are presented in sub-plots (e) and (f). Plots (a)-(d) are charge configurations for τ = 0.1 (a), τ = 2 (b), τ = 4.31697 (c), and τ = 4 (d). The black points are charges located on the unit disk. The red and blue points are stationary points (eigenvalues of Sˆ ) ξ1 and ξ2. In this simple model, we have two eigenvalues ξ1 and ξ2 that are in the unit disk. The larger one, in absolute value sense, determines the final state of the system in the large-n limit. When the curves in (e) cross, namely |ξ1| = |ξ2|, we find ever lasting quantum oscillations, for example for the energy. In the Zeno limit (a), τ → 0, both eigenvalues approach unity. A comparison with the lower bound Eq. (40) (green line) is presented in (f). Note the usefulness of the charge picture. For example in (a), since τ is small, all the charges, i.e. black points on the unit disk, are closely situated, and it follows from basic electrostatics that also the eigenvalues (red and blue points) are both close to unity. This is shown clearly in subplot (e) in the small τ limit.

0.4 0.8 1 0.2 a b c d 0 0.2 0.6 0.5 -0.2 0

-0.4 -0.2 0.4 0

-0.6 -0.4 0.2 -0.5 -0.8 -0.6

-1 -0.8 0 -1 100 102 104 106 108 0 20 40 60 80 100 0 200 400 600 800 1000 0 200 400 600 800 1000

FIG. 5. The expected energy of the three-level system versus the measurements step n for τ = 0.1(a), τ = 2(b), τ = 4.31697(c), R and τ = 4(d). In subfigure (a), we have ξ1 ∼ ξ2 ∼ 1 and |ξ1| > |ξ2|, the number of steps it takes to reach the state |ξ1 i is very R large. As a comparison, in subfigure (b), |ξ1| ∼ 1 and |ξ2| ∼ 0.4, and the transition to state |ξ1 i is very fast. In (c), we have |ξ1| = |ξ2|. Hence the energy oscillates due to the competition between the two states. (d) is close to the oscillation point but not exactly on it, so the amplitude of oscillation decays to zero.

then find E0 ' −1.25, E1 ' 0.445, and E2 ' 1.80. We eventually switches to a state that is stable in time. The 5 prepare the system initially in state |ψini = |2i and we turn over is seen roughly at n = 10 (see Fig.5(a)). In perform null measurements, every τ units of time, on the contrast, in Fig.5(b), when τ = 2, we find a steady state detection state |ψdi = |0i. Using Eq. (16), there are after roughly five measurements. This corresponds to a three charges: p0 ' 0.108, p1 ' 0.349, and p2 ' 0.543, case where the eigenvalues ξ1 and ξ2 are separated, while −iE0τ −iE1τ −iE2τ located at e , e , and e (see Fig.4 left pan- in the Zeno case |ξ1| ' |ξ2| ' 1. We also see oscillatory els). This forms our charge picture with two stationary behavior for the special choice of τ = 4.31697 in Fig.5(c) points ξ1 and ξ2 inside the unit disk. Since the energy or when working close to this value, where the oscillations spectrum is non-degenerate, there are no dark states in decays away eventually, Fig.5(d). To gain insights on the system. As demonstrated in Fig.4, varying τ we these behaviors we go back to the charge picture in Fig. get different charge configurations, that yield different 4. behaviors for the measurement process. In Fig.4, we plot the absolute values of eigenvalues In Fig.5, we show the energy of the system versus ξ1 and ξ2 and the corresponding charge configurations n for four choices of τ. Notice the different scales of n for τ = 0.1, 2, 4.31697, 4. In the Zeno regime, i.e., in these plots. For small τ, we find Zeno dynamics, the τ → 0, all the three charges merge (see Fig.4(a)). Since system is lingering in one state for very long time, but the charges are closely situated, it is obvious from elec- 11 trostatics that the stationary points are all coalescing in the vicinity of the charges but in the unit disk. This is 3 1 because all the charges are positive. The eigenvalues of ˆ 2.5 S are |ξ1| ' |ξ2| ' 1. Hence the system reaches the fi- 0.8 n n nal state slowly, since clearly both |ξ1| and |ξ2| decay slowly. This behavior is generic to the Zeno limit and it 2 will be investigated in generality in Sec.VIII. A general 0.6 feature of the eigenvalues is found to be: 1.5 0.4 |ξi| ≥ cos(∆Eτ/2), (40) 1 for small τ, where ∆E is the difference between the max- 0.5 0.2 imum of the energy Emax and the minimum Emin [55]. In our case ∆E ' 3.05. We see, with this lower bound, 0 that τ → 0 all the eigenvalues of the survival operator 0 1 2 3 approach unity (and hence one cannot simply neglect one 10 10 10 10 compared to the other, unless n is really large). In Fig. 4(f) we plot this bound for demonstration. In Fig.4(b), i.e., τ = 2, we have two charges that FIG. 6. With simulations, we plot the mean energy (red tri- are merging [55]. This means that the phases satisfy angles) of the V -shaped system and the probability to be in state |Di (green crosses) versus n. Here we choose γ1 = 0.01, exp(−iE0τ) ' exp(−iE2τ). From basic electrostatics, if we have two nearby charges we expect a stationary point γ2 = 1, and τ = 0.5. The red line is the theoretical prediction of the mean energy for the large-n limit from Eq. (42). The in their vicinity, where the forces are balanced. Thus null measurements shelves the system in state |Di. once again we see that the charge picture can be used to rationalise our finding, and more importantly in Sec. VIII, we will present a more general theory based on it. This leads to |ξ1| ' 1, while |ξ2| ' 0.4. The system we set EG = 0,ED = 3,EB = 5, γ2 = 1. We keep γ1 as reaches the final state very fast, since |ξ1|  |ξ2|. In a free parameter, and we will consider the limit when it Fig.4(c), τ = 4.31697, |ξ1| = |ξ2|, which leads to the approaches zero. Then the eigenvalues Ei of H are given 3 2 2 2 oscillation in Fig.5(c) predicated in Eq. (38). by Ei − 8Ei + (14 − γ1 )Ei + 3 + 5γ1 = 0. When γ1 → 0, 2 we have (−3 + Ei)(−1 − 5Ei + Ei ) ' 0, which leads to E0 ' −0.2, E1 ' 3, and E2 ' 5.2. The eigenstates of H 2 T A. Artificial driven atom with V -shaped energy are |Eii = N{(−1 − 5Ei + Ei )/γ1,Ei − 5, 1} , where N structure is the normalization. As γ1 → 0, |E1i ' |Di, a result we will use later.

We now consider an artificial atom system with the We then choose γ1 = 0.01. By the definition of charges, 2 2 famous V -shaped energy level structure that allows for we have p0 = |hB|E0i| ' 0.03576, p1 = |hB|E1i| ' −6 2 shelving [8], see Fig.2. The system’s states are |Gi 2.040×10 , p2 = |hB|E2i| ' 0.9642. So p2  p0  p1, (ground state) and |Di and |Bi. Inspired by the experi- the charges p0 and p1 are much weaker than p2. In ment in Ref. [9], we will investigate the case where tran- the electrostatic language, we have two weak charges, sition amplitudes from |Gi to |Di and |Gi to |Bi vary and as a result, the stationary points ξ1 and ξ2 will considerably. The system starts in the ground state, and approach these weak charges separately. With simple then conditional null measurements are made in state numerical calculations, we find the eigenvalues of Sˆ |Bi. The question is where will we find the particle, are ξ1 ' 0.0707 + 0.9975i, ξ2 = 0.9292 + 0.0742i, so when n is large? It turns out that the particle is shelved |ξ1| ' |ξ2| ' 1, and |ξ1|>|ξ2|. in state |Di and the amplitude of finding the particle in We see that we have a very weak charge p1 in the sys- the ground state diminishes with n. In some sense the tem. In this case we expect and indeed find an eigenvalue fact that we gain knowledge from measurements that the of the survival operator that is very close the vicinity system is not in |Bi implies that it cannot be found in of this charge, but in the unit disk, and it is ξ1. This |Gi also, since |Gi is loosely speaking the doorway to |Bi. can be understood using basic electrostatics, namely for Thus the energy of the system is going to increase from three positive charges on the unit disk, we expect to find the energy of the ground state to the energy of state |Bi. a stationary point close to the weak charge, the remain- We now investigate this scenario in more detail. The ing charges balance the force (Similarly, the stationary Hamiltonian H reads: point between Earth and Moon is closer to the Moon).

H = ED|DihD| + EB|BihB| + EG|GihG| We will show this in more generality in Sec.VIII, us- + γ (|GihD| + |DihG|) + γ (|GihB| + |BihG|). (41) ing , but for now we point out that 1 2 to leading order we expect that the corresponding eigen- We first find the eigen energies and vectors of H using state, namely |E1i, is selected as the stationary state of the basis {|Di, |Gi, |Bi}. To simplify the calculations, the system. So for a quantum particle initially in the 12 ground state |Gi, using Eq. (47), we have for large n 4, 8 inφ −inE1τ −inE1τ 3, 4 5, 4 |ψf i ' e |ξ1i ' e |E1i ' e |Di. (42) 4, 7 2, 2 6, 2 Secondly, because we have two weak charges, i.e. both 4, 6 p0 and p1 are weak compared to p2, that leads to |ξ1| ' 3, 3 5, 3 4, 5 |ξ2| as mentioned. Therefore, for not too large n, and 1, 1 7, 1 from Eq. (34), both the eigenvalues contribute, and the 4, 4 transition to the final state Eq. (42) is slow. Only at 3, 2 5, 2 4, 3 a certain critical large number of measurements do we 2, 1 6, 1 observe a transition. 4, 2 As shown in Fig.6, the expected energy of the mon- 3, 1 5, 1 4, 1 itored system exhibits a transition from the zero energy of ground state |Gi to the energy of state |Di, which is 3. This transition is found at n ' 50, which is related FIG. 7. Glued binary tree graph G3. The detector is set on the state |1, 1i. We consider different initial states. to the relative magnitude of |ξ1| and |ξ2|, since in this example both are close to unity as mentioned. We also plot the probability of being in state |Di which exhibits a similar transition in the expected energy from zero to of parameters like γ1 we and others use in these problems. unity. The meaning of a nearly dark state is explored below in As already mentioned in the introduction, we have perturbation theory in full generality using the so called been inspired by the experiment of Minev et al. [9]. weak charge theory (see Sec.VIIIA). There the steering by conditional measurements was used to control and reverse the quantum jump in mid-flight. This, in turn, is based on concepts of quantum trajec- tories, well investigated in . There the VII. GLUED BINARY TREE emission events, or more generally quantum jumps, are an inherit to the dynamics of the system, for example, a fluorescence process. In our approach, which is dif- As another application of our general theory, we ferent from the experiment, we impose the jumps and consider the glued binary tree [73, 74]. Glued trees the conditional steering by repeated projective measure- were investigated previously as they provide exponential ments, where the condition is a null measurement (i.e., speedup for quantum search algorithms [75, 76], and this non-detection). We believe that due to the advances in was observed in a recent experiment [77]. In contrast we single-particle manipulation, the proposed method will consider the effect of null measurements on this popular be useful both in the steering of single-particle systems, model. as shown here, but also for their control. For example, in the quantum search problem, the goal of the field is to speed up the search, which in turn is presented as the time it takes for a quantum walker to reach the tar- A. Eigenstates and eigenenergies get state. Then if we detect many events of failed mea- surements (like those analyzed here), we may wish to re- store to control. By that, we mean that after observing First, let us define a sequence of graphs Gd. Gd con- many failed detections, the system in the search process sists of two balanced binary trees [73]. The total number d+1 d is pushed toward a dark state. These dark states, in sys- of vertices in Gd is 2 +2 −2. In Fig.7 we present the tems with built in symmetry, are stable. This means that G3 tree, which will serve as our example. This graph de- after the target is not detected for a while, it will be of scribes the Hamiltonian of the system, namely the nodes benefit to add control, for instance, to restart the search are the states and the links represent the hoping ampli- process or add an external perturbation. This approach tudes between states. All transitions are identical, so and its effect on the quantum search process must be that the Hamiltonian is given by the adjacency matrix of investigated in more detail in the future. the graph. Remark: The state |Di in this section is traditionally The eigenenergies and eigenstates of this model were called dark, indeed this term is obviously very physical. obtained in Refs [73, 74]. Here we briefly recap the solu- In the context of this work, and based on our definition of tion of the stationary Schr¨odingerequation. We consider the dark state |δi, note that state |Di is only nearly dark. a labeling along the “columns” and “rows” of the form We defined dark states, as such that starting in those (j, s) (see Fig.7), where j goes from 0 to 2d indicates the states the system is never detected, while here a system distance from the left root to right root along the graph, initially starting in the state |Di can be detected, at least and s goes from 0 to Nd,j −1 is the location within a given in principle (without conditioning the measurements, of column. Nd,j is the number of sites in a given column j, j 2d−j course). The meaning of nearly dark is related to the set and it is given by Nd,j = 2 for j ≤ d, and Nd,j = 2 13

FIG. 8. The classical charge pictures for the eigenvalues of Sˆ of G3 tree model when τ = 1.2 (left) and τ = 1.25 (right). As shown in figure, when τ = 1.2, there is only one eigenvalue ξf that is closest to the unit circle. While at τ = 1.25, two eigenvlaues ξα and ξβ that are equally close the the unit circle are found. In the latter case we anticipate quantum dynamics in the long-time limit, while in the former a unique steady state emerges. In this system we have 11 distinct energy levels, see TableI. One would naively expect to find 11 charges, however some charges vanish, namely the detected state |1, 1i is orthogonal to some of the energy levels, due to symmetry. As a result we have 7 charges.

√ E0 E√1 E2 E√3 √ E4 E5 E6 E7 E8 E9 E10 Energy −2 2 cos(π/8) − 6 −2 − 2 −2 2 sin(π/8) 0 −E4 −E3 −E2 −E1 −E0 Degeneracy 1 131 1 811311 hψd|Ek,ii = 0 0 101 0 501010 hψd|Ek,ii 6= 0 1 030 1 310301

TABLE I. The energy spectrum of G3 tree. The detection state |ψdi = |1, 1i. The number of states with zero and non-zero overlap with the detection state |ψdi is listed in the last two lines. for j>d. Then the eigenstates of H are [74]: particle has some overlap with the dark subspace, it can be trapped in the dark subspace. In this case the dynam- 2(d−v) 1 X k(j + 1)π ics will be determined by the degenerate eigenenergies as |E i = √ sin[ ]|scol j; α, vi k,d−v,α 2(d − v + 1) mentioned. d − v + 1 j=0 (43) with

(2α+1)Nj,d−v −1 X |j + v, si − |j + v, s + Nj,d−vi |scol j; α, vi = p To present the quantum dynamics via measurements, 2Nj,d−v s=2αNj,d−v we treat a specific model. We choose the G3 tree as in which |j, si denote the Hilbert space vector associated shown in Fig.7, which consists of 22 nodes. So d = 3. with vertex (j, s); and v = 0, ··· , d; and k = 1, ··· , 2(d− We use Eq. (44) and for simplicity we label the energy v) + 1. In |scol j; α, vi, j = 0, ··· , 2(d − v) and α = levels according to E0

FIG. 9. The expected energy of the G tree system versus 3 FIG. 10. The expected energy of the G3 tree system ver- the measurement steps n conditioned the detections are null. sus the measurement steps n. The blue crosses are numerical√ We prepare the system in its ground state (hEi = −2.61313), simulations for the initial state |ψini = (|E2,1i + |E10i)/ 2. and then perform repeated stroboscopic measurements at the The selection rule for the final energy of the system indi- state |ψdi = |1, 1i (see Fig.7). As demonstrated in Fig.8 cates, that this final energy is the same as that of state ˆ when τ = 1.2, there is only a unique eigenvalue ξf of S that |E2,1i since that state belongs to a degenerate subspace, while R is closest to the unit circle, corresponding to the state |ξf i, |E10i does not. And the theoretical value of E in the large-n with expected energy hEf i = 0 (red circles). As shown here, limit is the blue line. The yellow triangles√ are for the initial the energy of the system (red crosses, numerical simulations) state |ψini = (|E2,1i + |E10i + |E5,1i)/ 3 (here both |E2,1i reaches the expected energy when n ' 20. The detections and |E5,1i are degenerate, hence the final state is different pump the energy of the system and send it to the new steady from the previous case). We also show the mean energy for state. When τ = 1.25, there are two eigenvalues ξα and ξβ |ψini = (|E2,1i + |E10i + |E5,1i + |E6i)/2 in red pentagrams. that are equal in magnitude |ξb| = |ξa| (see Fig.8), while The theoretical predictions of the mean energy E is the green they are also the largest eigenvalues (besides the darks states line. Note that red and yellow curves merge in the large-n which are not relevant due to the initial condition). Now the limit. This is because the difference between these two initial total energy of the system oscillates. Our theory (cyan circles) conditions, is an energy eigenstate which is non degenerate, predicts this oscillation and perfectly matches the numerical which is unimportant in the long-time limit. simulations (black crosses) when n is large.

there are two eigenvalues ξα ' −0.894962 + 0.108282i 1. Initial condition orthogonal to the dark states and ξβ ' −0.894962 − 0.108282i that are equally close to the unit circle (see Fig.8). The expected energy of the corresponding eigenstates are Eα ' 1.46103 and Now, based on the charge picture approach, we show Eβ ' −1.46103. Since there are two eigenvalues equally different quantum dynamics driven by repeated measure- close to unit circle, the system will oscillate in time. Due ments. We start from the ground state of the system, R R to the competition between the states |ξα i and |ξβ i dur- namely, |ψini = |E0i. The ground state is an eigen- ing the measurements process the energy oscillates in the state of Hamiltonian H and non-degenerate, hence the detection process (see Fig.9). The measurement trans- initial state has no overlap with all the dark states, i.e., fers energy to the quantum system and back in an os- hψin|δk,ji = hE0|δk,ji = 0. This is obvious since the en- cillatory way, and the system never reaches a stationary ergy eigenstates are orthogonal with respect one to an- state. These interesting phenomena reveal the effect of other. For the first case, we consider τ = 1.2. The eigen- the measurements and are also instructive for the quan- ˆ values ξi of survival operator S are stationary points of tum dynamics out of equilibrium. the charge field in Fig.8. There is a unique maximum ξf ' −0.999767 of the set {|ξi|} that is closest to the unit circle (see Fig.8). So the system will approach this 2. Initial condition overlapping with the dark states fixed point when the detection number is large. From Eq. inφf R (37), the final state is |ψf i = e |ξf i, whose expected Now we consider the cases that the initial state has R R energy is E = hξf |H|ξf i = 0, while initially the energy is some overlap with the dark states. According to Eq. E0 ' −2.61313. The measurements transfer energy to the (36), the final state of the system is determined by the quantum system continuously until the system reaches its dark states, which are constructed from the degener- equilibrium state. We present this in Fig.9. ate energy levels. In this case we expect that the non- For the second scenario, we choose τ = 1.25, where degenerate eigenenergies will be irrelevant in the long- 15 time limit. In this part, we exploit the degeneracy of H We find a stationary point close to this charge, denoted and explore the selection rule for the degenerate states. ξf ' exp(−iE0τ). At ξf , the electrostatic force vanishes, We first consider√ the case that the initial state |ψini = because the force of the weak charge balances all other (|E2,1i + |E10i)/ 2, which is a linear combination of the forces. By analogy, the stationary point of the Moon- degenerate energy level E2 and non-degenerate energy Earth system is much closer to the Moon than to the level E10. The mean energy of the G3 tree system at the Earth. Using Eq. (16) together with perturbation the- beginning is E0 = E2/2 + E10/2 ' 0.31. Conditioning ory, we get [67] the measurements to be null, the system converges to −iE0τ the degenerate energy level only and the component of ξf ∼ e − , (45) the non-degenerate energy level is eventually wiped out. So, the expected energy of system in the final state is where E = E = −2. In Fig. 10, our numerical simulations p 2  ∼ 0 . (46) show that the energy of the system is indeed E when n P −iτE0 −iτEk 2 k6=0 pk/(e − e ) is large.

Now we add the component of another degenerate −iE0τ Since p0  1,  ∼ 0. The leading term of ξf is e , energy level E to the initial state. The initial state 5 √ hence |ξf | ∼ 1 and |ξf | < 1. From Eq. (37), for then reads |ψini = (|E2,1i + |E10i + |E5,1i)/ 3. This a system with such a weak charge, the final state is is a linear combination of two degenerate energy lev- inφf R |ψf i ∼ e |ξf i. Substituting ξf into Eq. (17), the els and one non-degenerate energy level. Following the R expression for the right eigenstate |ξf i can be highly sim- selection rule, the two degenerate energy levels deter- plified, which leads to mine the final state of the system, and the mean en- ergy of the system will reach the value E = −1 eventu- inφf |ψf i ∼ e |E0i, (47) ally. If we add another non-degenerate energy level E6, i.e., |ψini = (|E2,1i + |E10i + |E5,1i + |E6i)/2, this does where φf = −i ln(ξf /|ξf |) ∼ −E0τ. Eq. (47) indi- not change the final state, as the selection rule of the cates that the final state is the energy eigenstate |E0i. measurements wipes out both the |E10i and |E6i compo- And the global phase accumulated is approximately the nents. We present the numerical simulations in Fig. 10, energy E0 multiplied by the evolution time. The re- and the results are consistent with our predictions. peated measurements drive the system to this specific state. To get a deeper understanding of this result, we go back to the definition of the weak charge p0, where VIII. NULL MEASUREMENTS, INSIGHTS 2 p0 = |hψd|E0i|  1. Actually, the “weakness” of the FROM THE CHARGE THEORY charge means the overlap of the energy state |E0i and the detection state |ψdi is nearly zero, i.e., hψd|E0i ∼ 0. We now provide general insights from the charge the- Hence the bright state |E0i is acting like a nearly dark ory, which are used to find the largest eigenvalue of Sˆ in sate, due to the small overlap. The expected energy of generic situations. The technique here presented follows, the final state is hEi = hψf |H|ψf i ∼ E0. and in some cases extends the ideas in [55, 67, 68], that This picture, was demonstrated already in specific V - were developed in the context of the first detection prob- shaped system in Sec.VIA. More specifically, in this −6 lem. We consider cases where the dark space is empty, example, the weak charge p0 ' 2.040×10 (note in Sec. simply because a system with a non-empty dark space VI A, we denoted it p1). Using Eq. (45), we have  ' −6 −6 can be treated exactly with the tools given in Eq. (8). 1.945 × 10 − 1.077 × 10 i, from here ξ1 ' 0.0707353 − So, in this section |ξi|<1. Furthermore, we consider cases 0.997496i in excellent agreement with the exact value, where one or several eigenvalues of the survival opera- ξ1 = 0.0707353 − 0.997494i. Moreover, the general Eq. tor are close to the unit circle, and hence these are the (47) is directly demonstrated with Eq. (42). largest. We develop an approximate expression for the eigenvalue(s), and also give insight on the eigenvectors. We consider four cases: i) a weak charge scenario, ii) two B. Two merging charges charges merging picture, iii) triple-charge theory, and fi- nally iv) quantum Zeno regime, where all the ξs approach Another mechanism leading to the eigenvalue of Sˆ be- the unit circle. ing close to the unit circle is the case when two energy levels, denoted Ea and Eb, satisfy the resonance condition exp (−iEaτ) ' exp (−iEbτ)[55]. Note that this can be A. Weak charge theory achieved by modifying τ or some other parameter of H. We then have two charges pa and pb close to each other, Assume that one of the overlaps denoted p0, associ- both located on the unit circle. So we expect to find a ated with energy level E0, is small, p0  1, and in par- stationary point, denoted ξf in their neighborhood. This ticular much smaller than all the others. In the electro- is because the point of zero force is largely determined static language, we have a weak charge at exp(−iE0τ). by this pair. In analogy, the equilibrium point between 16 two neighboring stars is determined to leading order by C. Triple-charge theory these and not by other distant stars. An example was shown in Fig.4(b), however now we treat the problem in An interesting case is found when three charges merge generality. We need to find an approximation for ξf as on the unit circle. In this case we will have two eigen- δ → 0, where δ = (Ebτ −Eaτ)/2 mod 2π, which measures values of the survival operator Sˆ in the vicinity of these the angular distance between the two phases. Using Eq. three charges, and in that sense this example is different (16), we find [68]: from those considered in the previous two subsections. Once we have two eigenvalues which have the same mag- −iEbτ −iEaτ pae + pbe 2 nitude in absolute value sense, we expect to find oscilla- ξf ∼ + O(δ ). (48) pa + pb tory behavior, induced by the phase differences, which in turn are controlled in principle by the measurements time As shown in Eq. (48), the charges pa and pb determine τ. For simplicity we consider systems with commensurate the location of the stationary point ξp, and the other energy levels, the three merging energy levels are E0 = 0 0 ±iEτ i2πk±iδ charges give only a second order perturbation. Since and E± = ±E with phases e and e = e , exp (−iEaτ) ' exp (−iEbτ), |ξf | ∼ 1. As a demonstra- where k is an integer and δ is our small parameter. This tion of Eq. (48), we consider the charge configuration configuration of charges or phases yields two complex iθ± in Fig.4(b). Using Eq. (48), we have ξf ' −0.8170 + eigenvalues denoted ξ± = r±e . They are located in the 0.5726i. While the exact value is −0.8158+0.5722i. Here vicinity of the unit circle, as expected from basic electro- 2 2 δ = 0.09, and the error comes from terms which are sec- statics. We denote p0 = |hE0|ψdi| and p = |hE±|ψdi| . ond order in δ. As shown in AppendixD, a third-order expansion of Eq. From Eq. (37), the final state of the system then is (16) in δ yields the ξ+ and ξ− [67] (see Eqs. (D1) and inφf R ∗ 2 |ψf i ∼ e |ξf i. Substituting ξf into Eq. (17), we have (D2)). We find ξ+ = ξ− up to the order O(δ ). As mentioned, such a system, will yield quantum dynam- R hEa|ψdi hEb|ψdi ics even in the long-time limit, since other eigenvalues |ξf i ∼ N( |Eai − |Ebi), (49) pa pb of the survival operator are decaying faster, as they are smaller, which is due to the fact that the three merging 1 R where N is for normalization . The eigenstate |ξf i is charges are situated one next to each other. In principle a linear combination of the energy eigenstates |Eai and the life time of these quantum oscillations depends on |Ebi that are merging. As a reminder, when the energy the other background charges, that can break symmetry level E0 is 2-fold degenerate, the dark state is given by (to higher order in delta) creating a situation where one |δ0,1i = N(hψd|E0,2i|E0,1i − hψd|E0,1i|E0,2i). Since the of the eigenvalues ξ+ or ξ− is actually closer to the unit energy levels Ea and Eb satisfy the resonance condition circle. exp (iEaτ) ' exp (iEbτ), they now act like one degener- Symmetric background: For symmetric back- ate energy level. However, now they are weighted by the ground charges, we have r+ = r− and θ+ = −θ−, and magnitude of the corresponding charges for each energy hence, according to our theory, the measurements induce level. The expected energy of the final state is dynamics forever. From Eq. (38), the final state is de- R R termined by the states |ξ+i and |ξ−i. Using Eq. (17), p E + p E we have: hEi ∼ b a a b . (50) pa + pb   R hE−|ψdi hE+|ψdi hE0|ψdi |ξ+i ∼ N |E−i+ |E+i+ |E0i , Following the example in Fig.4(b), we have τ = 2 and C − 1 C + 1 C (52) the merging charges are p0 ' 0.108 and p2 ' 0.543 corre- p R sponding to the energy levels E0 ' −1.25 and E2 ' 1.80. where C = p0/(p0 + 2p). Similarly, |ξ−i reads: Using Eq. (50), the mean energy of the system in the hE |ψ i hE |ψ i hE |ψ i  large-n limit is E ' −0.744. The exact mean energy is |ξRi ∼ N − d |E i+ + d |E i+ 0 d |E i . −0.7511, which is presented in Fig.5(b). − C + 1 − C − 1 + C 0 The global phase accumulated in the measurements is: (53) Using Eq. (38), since θ+ = −θ− = θ, the final state of inθ R −inθ R Ea + Eb pb − pa the system is |ψf i = N(a1e |ξ+i + a2e |ξ−i), where φf ∼ − τ + δ. (51) 2 pa + pb a1 and a2 are overlaps with the initial state. Specially, we choose a1 = a2, which leads to The first term on the lhs is the leading part of φf . Like ∗ the final state, the phase is also determined by the two |ψf i ∼ N{D(n)hE−|ψdi|E−i + D (n)hE+|ψdi|E+i merging energy levels. − 2 cos(nθ)hE0|ψdi|E0i}, (54) p where D(n) = [p0 cos(nθ)+i p0(p0 + 2p) sin(nθ)]/p and iθ+ iθ √ θ is defined by ξ+ = r+e = re . Since the first order 1 When hE |ψ i is real we have of course hE |ψ i/p = 1/ p . 2 a d a d a a approximation of ξ± is totally imaginary, when n  1/δ , 17

FIG. 11. The classical charge pictures for the eigenvalues of Sˆ of G3 tree model when τ = 2.3 (left) and τ = 2.35 (right). Here |ψdi = |1, 1i. As shown in figure, on the east part, on and just above and below the equator, we see three charges that are close one to another (charges are in gray, on the unit circle). As a result, there are two eigenvalues ξ+ and ξ− that are approaching ∗ ˆ the unit circle and ξ+ = ξ− (orange for τ = 2.3 and red for τ = 2.35). Other eigenvalues of S, in the unit disk, are colored blue, they will not contribute to the long-time limit as they are further away from the unit circle. Similar to Fig.8 some of the charges in this system are zero, hence while the number of distinct energy levels is 11 (TableI) the number of non zero charges, is 7.

p we can express θ ∼ Aδ,(A = p0/(p0 + 2p), see Eq. system is driven periodically due to the measurements. (D3)). Here, we choose the initial state to be the ground state, Random background: When the background where the effects of the dark states can be neglected. charges are not strictly symmetric, |ξ+| 6= |ξ−|. Then, As shown in Fig. 12, the mean energy of the G3 sys- the quantum dynamics will decay away for very large- tem oscillates periodically with the number of measure- n and the system will go to a fixed state determined ment steps, n, and our theory (lines) fits the numerical by the larger of |ξ+| and |ξ−|. Nevertheless, the sys- simulations (crosses). More importantly, the oscillation tem will exhibit nontrivial dynamics in a certain time frequency of the mean energy changes when we change regime because of the charge configuration we set (the τ. As shown in the figure, the oscillation frequency at three charges we consider are symmetric). So the theo- τ = 2.3 is faster than the frequency at τ = 2.35. This retical question is, how long will the quantum dynamics follows from the charge picture, as the relative phase, last? From the expressions of ξ+ and ξ−, we see that i.e., Arg(ξ+) − Arg(ξ−), at τ = 2.3 is larger than that at 2 |ξ+| = |ξ−| up to order O(δ ). The background effect τ = 2.35 (see Fig. 11). The oscillation frequency is con- comes in order O(δ3). So, at least, the quantum dynam- trolled by this relative phase, and hence the frequency at ics will last until n ∼ 1/δ2. When the number of mea- τ = 2.3 is faster. surements becomes large than 1/δ2, the system gradually goes to a fixed state due to the symmetry breaking. D. Quantum Zeno regime

1. Example As we increase the number of charges merging to the vicinity of a point on the unit circle, more eigenvalues of To demonstrate the triple-charge theory, we use the the survival operator approach the unit circle, and they glued binary G3 tree example and tune the measurement are also all close to each other. An example is the quan- time interval τ. As shown in Fig. 11, when τ = 2.3 and tum Zeno regime [58, 78–84], where τ ∼ 0 and all phases τ = 2.35, we have three charges that are close to each exp(−iEkτ) coalesce (as an example, see Fig.4(a)). In other and far away from other charges (the background the quantum Zeno regime, due to the fast measurements, charges). As a result, there are two eigenvalues (ξ+ and the dynamics of the monitored system is slowed down [84] ξ−) of the survival operator that are near the unit circle and our goal is to characterize this behavior. In general, (orange for τ = 2.3 and red for τ = 2.35 in Fig. 11). we have a set of eigenvalues of the survival operator (ex- Because of the symmetry of the system, they also have cluding the trivial one on zero) and we consider their the same absolute value, so |ξ+| = |ξ−| ' 1. The exact absolute value |ξ1|, |ξ2|, ··· , |ξw|, which are approaching numerical values, when τ = 2.3, are |ξ+| = |ξ−| = 0.9873, unity as τ ∼ 0. The fastest decay mode is given by the while using Eqs. (D1) and (D2), we have |ξ+| ' |ξ−| ' minimum of the set |ξmin|. Our goal is to find a rough 0.9869. estimate for this eigenvalue, controlling the rate of the n The triple-charge configuration leads to persistent decay of this component |ξmin| , see Eq. (34). quantum dynamics. For instance, the mean energy of the Basic electrostatics tells us that all the stationary 18

of the energy band also affects this lower bound. The bound of the number of measurements then is nb = tb/τ, which is proportional to τ −2. So when τ ∼ 0, the re- laxation is therefore extremely slow, see example in Fig. 5(a).

IX. EXCEPTIONAL POINTS

Previously in Sec.IIIC we presented a simple example of an exceptional point of a two-level system. We now briefly explain how similar effects can be found in larger systems. In particular, we are searching for cases where all the eigenvalues of survival operator ξ = 0. Then, clearly, the right hand side of our main Eq. (34) is equal to zero and the whole approach is invalid. As mentioned, this corresponds to a situation where we cannot satisfy the condition of null measurements to begin with. The basic question is how to construct such systems. Here the charge picture is very useful. Consider a three- level system. Then if we have three charges of the same magnitude on −120°, 0°, and 120° on the unit circle, clearly from symmetry all the eigenvalues of Sˆ are zero. This is because the stationary point of this charge config- uration is in the centre of the unit disk. Assume this sys- tem has energy levels −E, 0,E, further assume that cor- responding wave functions have the same overlaps with the detected state (so the charges are the same). In this case, if we choose Eτ = 2π/3+2πk with k an integer, we get a charge configuration that will exhibit the desired FIG. 12. The mean energy of G3 tree system versus the mea- surement steps n. Initially, the system is at its ground state. result. It is now rather easy to construct a Hamiltonian The measurements drive the mean energy of the system pe- that meets this demand, and we present an example be- riodically. The upper plot is for τ = 2.3, where the green low. In fact, this method can be extended to systems crosses are numerical simulations and the orange line is the beyond three or two-level systems rather easily. theory. The lower plot is for τ = 2.35, and the theory (red We also note that one may have exceptional points that line) perfectly matches the numerical simulations (blue pen- are not zero, i.e. 0<|ξi| = |ξj|<1. We do not treat that tagrams). The oscillation frequency at τ = 2.3 is faster than case here, but leave it for future work. the frequency at τ = 2.35, because the relative phase of ξ+ Remark: In the dark subspace, the eigenvalues and ξ− at τ = 2.3 is larger than the phase at τ = 2.35 (see ˆ exp(−iEkτ) of the survival operator S is (gk − 1)-fold Fig. 11). degenerate. When gk ≥ 3, using Eq. (8), there are two ˆ or more eigenstates |δk,ii of the survival operator S. As points are located in the convex hull of the charges [55], they are constructed by a Gram Schmidt procedure, these the area of which vanishes as τ ∼ 0. We use this to our dark states are orthogonal. Using Eqs. (17) and (22), for ˆ advantage and obtain a lower bound for |ξ|: the eigenstates of S that correspond to the eigenvalues that lie in the unit disk, i.e., 0<|ξ|<1, when two ξs co- |ξmin| ≥ cos(∆Eτ/2), (55) alesce, both left and right eigenvectors become parallel. Thus we only have one eigenvector instead of two. where ∆E is the width of the energy spectrum, namely ∆E = Emax − Emin. An example of Eq. (55) is Eq. (40) for the three-level system, which is demonstrated A. Example n −1 in Fig.4(f). Then we let |ξmin| b = e , which leads to nb = −1/ ln[cos(∆Eτ/2)]. The lower bound of the Here we find the Hamiltonian of the three-level system evolution time tb = nbτ is: with the approach mentioned above. As mentioned, we −τ 8 want the three charges to be located on angles −120°, 0°, t ∼ ∼ . (56) b ln[cos(∆Eτ/2)] (∆E)2τ and 120°. This can be easily achieved by choosing equally spaced energy levels, for instance, E0 = −γ, E1 = 0, and From Eq. (56), the evolution of the system is repressed, E2 = γ, where γ>0. When τγ = 2π/3 + 2πk, we get the −1 and the lower bound tb is proportional to τ . The width desired charge configuration. 19

The second step is to have all three charges with equal which are controlled by the measurements, and not magnitude, i.e. p0 = p1 = p2 = 1/3. We choose the de- by the energy levels alone (unlike standard quan- T tection state |ψdi = (1, 0, 0) . We then construct a group tum mechanics). 2 of orthogonal eigenstates of H that |hψd|Eii| = 1/3. A possible√ √ choice√ of such eigenstates√ |Eii√is |E√0i = ˆ T T 4. In the Zeno limit, all the eigenvalues of S approach (1/ 3, −1/ 6, 1√/ 2) , |E1i = (−1/ 3, 1/ 6, 1/ 2) , p T the unit circle. This means that oscillations are ef- and |E2i = (1/ 3, 2/3, 0) . Such eigenstates of H fectively undetectable, and the relaxations are slow have identical overlaps with the detection state, and (as well known). We have presented a lower bound hence all the charges have the same magnitude. for the eigenvalues, showing their vicinity to the The last step is to find the concrete form of H. Since unit circle. |Eii is an eigenstate of H with eigenvalues Ei, we have H|Eii = Ei|Eii, which leads to 5. An interesting case is a situation where all the  √ √  ˆ 0√ −1/ 2 1/ √6 eigenvalues of S are equal to zero. This corre- sponds to widely investigated exceptional points of H = −γ −1/√ 2 −1/√2 −1/(2 3) . (57) 1/ 6 −1/(2 3) 1/2 non-Hermitian systems. The physics in this case implies that the condition of null measurements is For a quantum system with such a Hamiltonian, when not realizable. Rather the system is detected with τγ = 2π/3 + 2πk, we cannot satisfy the condition of null probability one, and hence the condition we impose measurements. In other words, the system is detected is violated. The classical charge picture was proven with probability one by the local measurements. ξ = 0 to be very useful here, in the sense that we can ex- is an exceptional point of the survival operator of order ploit the symmetry of the charge picture to direct three. With the same procedure, this can be generalized the eigenvalues of Sˆ to a unique stable point in for larger systems. the centre of the unit disk, and hence exhibit the exceptional physics.

X. SUMMARY We have considered the examples of two, and three- level systems, as well as a binary glued tree. We use the We have investigated the properties of the quantum mean energy of the system to characterize the state of the survival operator Sˆ , which in turn gives the state of the systems, while other observables are also possible. The system after a large number of conditional measurements. two-level system was used to exhibit an exceptional point We have classified five types of generic behaviors: (effect 5). However, with a two-level system one cannot detect the measurement induced dynamics (effect 3) sim- 1. In the presence of symmetry and hence a spectrum ply because in a two-level system one has only one non- which is degenerate, the system will exhibit dynam- trivial eigenvalue of the survival operator. In a three-level ics determined by the energy levels of the systems. V -shaped artificial atom (see Fig.2), we detected shelv- However, only the degenerate energy levels partici- ing and effect 2 (see Fig.6), which is actually well known pate. Of course, this is the case under the condition [8]. The theory developed here, in particular the charge that initially |ψini has some overlap with the dark theory, makes it possible to realize the long-time behav- part of the Hilbert space. A non-degenerate energy ior using an intuitive electrostatic analogy. For example, level may contribute only if it is orthogonal to the a single weak charge, corresponding to a weak overlap of detected state. a stationary energy with the detected state, or merging 2. In the absence of a dark subspace, for example, for charges (phases) on the unit circle, imply an eigenvalue ˆ systems with no-degeneracy, e.g. disordered, inter- of S which is close to unity in magnitude, and its precise acting or chaotic systems, typically the final state of value can be estimated in generality. Maybe more inter- the system is unique. This means that in the long- esting is the merging of three charges on the unit circle, ˆ time limit, we have no dynamics at all, and we can since here we get two eigenvalues of S approaching the say that the system reaches a kind of equilibrium, unit charges, and then we get the dynamical effect due induced by the repeated measurement. This state to the phase difference (effect 3). These effects was ex- corresponds the eigenvalue of the survival operator plored with the glued tree example (see Fig. 12). Clearly closest to the unit circle. In the second part of the the high degree of symmetry in this case, and the energy paper we developed tools to identify this state. degeneracy of the system, imply a large dark subspace (unlike the three-level systems under study) and hence 3. Under certain conditions, we find a pair of eigen- this type of example exhibits physical behaviors drasti- values of Sˆ which together are also the largest in cally different from the simple three-level systems. As magnitude. We find that these eigenvalues then mentioned, the charge picture was used also to find a have a relative phase that is controlled by the rate lower bound for the Zeno limit, and helps tremendously 1/τ. In this case, the system exhibits oscillation in the identification of exceptional points. 20

XI. ACKNOWLEDGEMENTS Substituting Eq. (B6) into Eq. (B5), we have

√ [e−iτEj (1 − p ) − ξ ] X −iτEi j l The support of Israel Science Foundation’s grant aie pi = aj √ . (B7) pj 1898/17 as well as the support by the Julian Schwinger {B},i6=j Foundation (KZ) are acknowledged. √ √ −iτEi −iτEi Now we define bi = aie pi, so ai = bi/(e pi). iτEj We also define ζj = 1 − (1 − e ξ)/pj. Eq. (B7) then Appendix A: Matrix determinant lemma can be simplified as: X In this section, we present the details for the calcula- bi = bjζj. (B8) tion using the matrix determinant lemma. The formula {B},i6=j of the matrix determinant lemma is: For j = 1, 2, ··· , w, Eq. (B8) is an equation set, which det(A + uvT ) = (1 + vT A−1u)det(A), (A1) contains w terms. We want to rewrite this equation set where A is an invertible square matrix and u, v are col- into the matrix form. We define the vector B as: umn vectors. For the eigenvalues of the survival operator † ∗ ∗ ∗ Sˆ , we have: B = (b1, b2, ··· , bw). (B9) ˆ det(ξ − S) = det[ξ − Uˆ(τ) + |ψdihψd|Uˆ(τ)]. (A2) Then the matrix is: ˆ T ˆ ζ 1 1 ··· 1  We let A = ξ − U(τ), u = |ψdi, and v = hψd|U(τ),  1   1 ζ2 1 ··· 1  which leads to:   1 1 ζ3 ··· 1 ˆ ˆ ˆ −1 M = . (B10) det(ξ − S) = det[ξ − U(τ)]hψd|[ξ − U(τ)] |ψdiξ. (A3)  . . . . .   . . . . .    This is the formula we used in the main text. 1 1 1 ··· ζw.

Eq. (B8) then can be rewritten as: Appendix B: Calculation for the right eigenvectors of survival operator with eigenvalues in the unit disk MB = 0. (B11)

In this section, we present the details for Eq. (17) in Using Eq. (B11) or Eq. (B8), we have: the main text. As discussed in the main text, we expand ζ1 − 1 the right eigenstates in the bright subspace. bi = b1 . (B12) ζi − 1 R X |ξ i = ai|βii, (B1) Substituting Eq. (B12) into the expressions for bi and ζi, {B} we get the expression for ai: where ai is the index we are looking for and {B} repre- √ r −iτE1 −iτE1 sents the summation in the bright subspace. Substituting pi e − ξ a1(e − ξ) pi ai = a1 −iτE = √ −iτE . Eq. (B1) into Eq. (3), we have: p1 e i − ξ p1 e i − ξ (B13) ˆ R X ˆ S|ξ i = ai(1 − D)U(τ)|βii, (B2) From Eq. (B13), we see that a2, a3, ··· , aw can be ex- {B} pressed by the a1. We can choose a1 freely, and in the

X −iτEi end it is a global constant that can be neglected by the = ai(1 − D)e |βii, (B3) normalization. Here we choose {B} √ √ X p1 pi = ξ ai|βii. (B4) a1 = , then ai = . (B14) e−iτE1 − ξ e−iτEi − ξ {B} Using Eq. (B1), we have: Multiplying Eqs. (B3, B4) hβj|, we have: √ X −iτEi R X pi ξaj = aie hβj|(1 − D)|βii. (B5) |ξ i = |βii. (B15) e−iτEi − ξ {B} {B}

Following the definition of the charges, the matrix ele- ˆ √ ments on the rhs of Eq. (B5) can be simplified. Using Eq. (9), the bright states |βii = Pi|ψdi/ pi. So together with Eq. (B15) we have:  1 − hβj|D|βji = 1 − pj i = j w hβj|(1 − D)|βii = √ . ˆ −hβ |D|β i = − p p i 6= j R X Pi j i j i |ξ i = |ψdi. (B16) e−iτEi − ξ (B6) i=1 21

Appendix C: Relation between right and left tion approach applied here is that we perform a third- eigenstates order expansion of Eq. (16) in small parameter δ. The eigenvalues ξ+ and ξ− then reads: In this section we derive a relation that connects the iθ+ 2 3 right eigenvectors to the left eigenvectors. Using Eqs. ξ+ = r+e ∼ 1 + iAδ − Bδ + O(δ ); (D1) (18) and (23), we find the right and left eigenstates can be related.

ˆ R L ∗ −iθ− 2 3 U(τ)|ξι i = |ξι i , (C1) ξ− = r−e ∼ 1 − iAδ − Bδ + O(δ ), (D2) here ∗ is the complex conjugate. For a two-level system, L T where if |ξι i = {a + ib, c + id} (a, b, c, d are real numbers), then |ξLi∗ = {a − ib, c − id}T . Eq. (C1) shows that ι r p the evolution operator Uˆ(τ) maps the right eigenstate of A = 0 ; (D3) p + 2p survival operator Sˆ to the corresponding left one. 0

Appendix D: Triple-charge theory p0 + p p X pj B = − 2 −iE τ . (D4) 2p0 + 4p (p0 + 2p) 1 − e j j6=0,± In this section we present the approximate expressions of ξ+ and ξ− for the triple-charge theory. The perturba-

[1] R. Uola, A. C.S. Costa, H. C. Nguyen, and O. G¨uhne, 573 (2014). “Quantum steering,” Rev. Mod. Phys. 92, 15001 (2020). [13] D. Tan, S. J. Weber, I. Siddiqi, K. Mølmer, and K. W. [2] E. Schr¨odinger,“Die gegenw¨artige situation in der quan- Murch, “Prediction and Retrodiction for a Continuously tenmechanik,” Naturwissenschaften 23, 844–849 (1935). Monitored Superconducting Qubit,” Phys. Rev. Lett. [3] E. Schr¨odinger, “Discussion of probability relations be- 114, 090403 (2015). tween separated systems,” Math. Proc. Cambridge Phi- [14] N. Foroozani, M. Naghiloo, D. Tan, K. Mølmer, and los. Soc. 31, 555–563 (1935). K. W. Murch, “Correlations of the Time Dependent Sig- [4] E. Schr¨odinger, “Probability relations between sepa- nal and the State of a Continuously Monitored Quantum rated systems,” Math. Proc. Cambridge Philos. Soc. 32, System,” Phys. Rev. Lett. 116, 110401 (2016). 446–452 (1936). [15] V. P. Belavkin, “A stochastic posterior schr¨odingerequa- [5] W. Nagourney, J. Sandberg, and H. Dehmelt, “Shelved tion for counting nondemolition measurement,” Lett. optical electron amplifier: Observation of quantum Math. Phys. 20, 85–89 (1990). jumps,” Phys. Rev. Lett. 56, 2797–2799 (1986). [16] J. Dalibard, Y. Castin, and K. Mølmer, “Wave-function [6] T. Sauter, W. Neuhauser, R. Blatt, and P. E. Toschek, approach to dissipative processes in quantum optics,” “Observation of quantum jumps,” Phys. Rev. Lett. 57, Phys. Rev. Lett. 68, 580–583 (1992). 1696–1698 (1986). [17] C. W. Gardiner, A. S. Parkins, and P. Zoller, “Wave- [7] J. C. Bergquist, R. G. Hulet, W. M. Itano, and D. J. function quantum stochastic differential equations and Wineland, “Observation of quantum jumps in a single quantum-jump simulation methods,” Phys. Rev. A 46, atom,” Phys. Rev. Lett. 57, 1699–1702 (1986). 4363–4381 (1992). [8] M. B. Plenio and P. L. Knight, “The quantum-jump ap- [18] H. Carmichael, An Open Systems Approach to Quantum proach to dissipative dynamics in quantum optics,” Rev. Optics, Lecture Notes in Physics Monographs, Vol. 18 Mod. Phys. 70, 101–144 (1998). (Springer Berlin Heidelberg, Berlin, Heidelberg, 1993). [9] Z. K. Minev, S. O. Mundhada, S. Shankar, P. Reinhold, [19] H. Mabuchi and P. Zoller, “Inversion of Quantum Jumps R. Guti´errez-J´auregui,R. J. Schoelkopf, M. Mirrahimi, in Quantum Optical Systems under Continuous Obser- H. J. Carmichael, and M. H. Devoret, “To catch and vation,” Phys. Rev. Lett. 76, 3108–3111 (1996). reverse a quantum jump mid-flight,” Nature 570, 200– [20] H. M. Wiseman, “Weak values, quantum trajectories, 204 (2019). and the cavity-QED experiment on wave-particle corre- [10] F. Pokorny, C. Zhang, G. Higgins, A. Cabello, M. Klein- lation,” Phys. Rev. A 65, 032111 (2002). mann, and M. Hennrich, “Tracking the Dynamics of [21] J. Gambetta, A. Blais, M. Boissonneault, A. A. Houck, an Ideal Quantum Measurement,” Phys. Rev. Lett. 124, D. I. Schuster, and S. M. Girvin, “Quantum trajectory 080401 (2020). approach to circuit QED: Quantum jumps and the Zeno [11] K. W. Murch, S. J. Weber, C. Macklin, and I. Siddiqi, effect,” Phys. Rev. A 77, 012112 (2008). “Observing single quantum trajectories of a supercon- [22] J. P. Garrahan and M. Gut¸˘a,“Catching and reversing ducting quantum bit,” Nature 502, 211–214 (2013). quantum jumps and thermodynamics of quantum trajec- [12] S. J. Weber, A. Chantasri, J. Dressel, A. N. Jordan, tories,” Phys. Rev. A 98, 052137 (2018). K. W. Murch, and I. Siddiqi, “Mapping the optimal [23] H. M. Wiseman and G. J. Milburn, Quantum Measure- route between two quantum states,” Nature 511, 570– ment and Control (Cambridge University Press, Cam- 22

bridge, 2009). Hermitian physics and PT symmetry,” Nat. Phys. 14, [24] S. Gammelmark, B. Julsgaard, and K. Mølmer, “Past 11–19 (2018). Quantum States of a Monitored System,” Phys. Rev. [45] T. Goldzak, A. A. Mailybaev, and N. Moiseyev, “Light Lett. 111, 160401 (2013). Stops at Exceptional Points,” Phys. Rev. Lett. 120, [25] I. Guevara and H. Wiseman, “ Smooth- 013901 (2018). ing,” Phys. Rev. Lett. 115, 180407 (2015). [46] E. J. Bergholtz, J. C. Budich, and F. K. Kunst, “Excep- [26] A. Solfanelli, L. Buffoni, A. Cuccoli, and M. Campisi, tional topology of non-Hermitian systems,” Rev. Mod. “Maximal energy extraction via quantum measurement,” Phys. 93, 015005 (2021). J. Stat. Mech: Theory Exp. 2019, 094003 (2019). [47] S¸. K. Ozdemir,¨ S. Rotter, F. Nori, and L. Yang, “Par- [27] L. Buffoni, A. Solfanelli, P. Verrucchi, A. Cuccoli, and ity–time symmetry and exceptional points in photonics,” M. Campisi, “Quantum Measurement Cooling,” Phys. Nat. Mater. 18, 783–798 (2019). Rev. Lett. 122, 070603 (2019). [48] A. Bergman, R. Duggan, K. Sharma, M. Tur, A. Zadok, [28] M. H. Mohammady and A. Romito, “Conditional work and A. Al`u,“Observation of anti--time-symmetry, statistics of quantum measurements,” Quantum 3, 175 phase transitions and exceptional points in an optical (2019). fibre,” Nat. Commun. 12, 486 (2021). [29] A. Riera-Campeny, J. Oll´e, and A. Mas´o-Puigdellosas, [49] A. Ben-Asher, D. Simsa,ˇ T. Uhl´ıˇrov´a,M. Sindelka,ˇ and “Measurement-induced resetting in open quantum sys- N. Moiseyev, “Laser control of resonance tunneling via an tems,” (2020), arXiv:2011.04403. exceptional point,” Phys. Rev. Lett. 124, 253202 (2020). [30] S. Roy, J. T. Chalker, I. V. Gornyi, and Y. Gefen, [50] N. Moiseyev, Non-Hermitian Quantum Mechanics (Cam- “Measurement-induced steering of quantum systems,” bridge University Press, 2011). Phys. Rev. Res. 2, 033347 (2020). [51] Y. Ashida, Z. Gong, and M. Ueda, “Non-Hermitian [31] B. Skinner, J. Ruhman, and A. Nahum, “Measurement- Physics,” (2020), arXiv:2006.01837. Induced Phase Transitions in the Dynamics of Entangle- [52] H. Krovi and T. A. Brun, “Hitting time for quantum ment,” Phys. Rev. X 9, 031009 (2019). walks on the hypercube,” Phys. Rev. A 73, 032341 [32] A. Nahum, S. Roy, B. Skinner, and J. Ruhman, “Mea- (2006). surement and entanglement phase transitions in all-to- [53] H. Krovi and T. A. Brun, “Quantum walks with infinite all quantum circuits, on quantum trees, and in landau- hitting times,” Phys. Rev. A 74, 042334 (2006). ginsburg theory,” PRX Quantum 2, 010352 (2021). [54] M. Varbanov, H. Krovi, and T. A. Brun, “Hitting time [33] A. Lavasani, Y. Alavirad, and M. Barkeshli, for the continuous quantum walk,” Phys. Rev. A 78, “Measurement-induced topological entanglement transi- 022324 (2008). tions in symmetric random quantum circuits,” Nat. Phys. [55] F. A. Gr¨unbaum, L. Vel´azquez, A. H. Werner, and (2021). R. F. Werner, “Recurrence for discrete time unitary evo- [34] Y. Bao, S. Choi, and E. Altman, “Symmetry enriched lutions,” Commun. Math. Phys. 320, 543–569 (2013). phases of quantum circuits,” (2021), arXiv:2102.09164. [56] J. Bourgain, F. A. Gr¨unbaum, L. Vel´azquez, and [35] T. Kato, Perturbation Theory for Linear Operators, Clas- J. Wilkening, “Quantum recurrence of a subspace and sics in Mathematics, Vol. 132 (Springer Berlin Heidel- operator-valued schur functions,” Commun. Math. Phys. berg, Berlin, Heidelberg, 1995). 329, 1031–1067 (2014). [36] W. D. Heiss, “Phases of wave functions and level repul- [57] S. Dhar, S. Dasgupta, A. Dhar, and D. Sen, “Detection sion,” Eur. Phys. J. D 7, 1–4 (1999). of a quantum particle on a lattice under repeated projec- [37] M. V. Berry, “Physics of Nonhermitian Degeneracies,” tive measurements,” Phys. Rev. A 91, 062115 (2015). Czech. J. Phys. 54, 1039–1047 (2004). [58] H. Friedman, D. A. Kessler, and E. Barkai, “Quantum [38] E. Persson, I. Rotter, H.-J. St¨ockmann, and M. Barth, walks: The first detected passage time problem,” Phys. “Observation of Resonance Trapping in an Open Mi- Rev. E 95, 032141 (2017). crowave Cavity,” Phys. Rev. Lett. 85, 2478–2481 (2000). [59] F. Thiel, E. Barkai, and D. A. Kessler, “First detected [39] C. Dembowski, H.-D. Gr¨af, H. L. Harney, A. Heine, arrival of a quantum walker on an infinite line,” Phys. W. D. Heiss, H. Rehfeld, and A. Richter, “Experimental Rev. Lett. 120, 040502 (2018). Observation of the Topological Structure of Exceptional [60] F. Thiel, I. Mualem, D. Meidan, E. Barkai, and D. A. Points,” Phys. Rev. Lett. 86, 787–790 (2001). Kessler, “Dark states of quantum search cause imperfect [40] K. G. Makris, R. El-Ganainy, D. N. Christodoulides, and detection,” Phys. Rev. Res. 2, 043107 (2020). Z. H. Musslimani, “Beam dynamics in PT symmetric [61] P. Kuklinski, “Conditional probability distributions of optical lattices,” Phys. Rev. Lett. 100, 103904 (2008). finite absorbing quantum walks,” Phys. Rev. A 101, [41] S. Klaiman, U. G¨unther, and N. Moiseyev, “Visual- 032309 (2020). ization of branch points in PT -symmetric waveguides,” [62] F. Caruso, A. W. Chin, A. Datta, S. F. Huelga, and Phys. Rev. Lett. 101, 080402 (2008). M. B. Plenio, “Highly efficient energy excitation transfer [42] B. Zhen, C. W. Hsu, Y. Igarashi, L. Lu, I. Kaminer, in light-harvesting complexes: The fundamental role of A. Pick, S. Chua, J. D. Joannopoulos, and M. Soljaˇci´c, noise-assisted transport,” J. Chem. Phys. 131, 105106 “Spawning rings of exceptional points out of Dirac (2009). cones,” Nature 525, 354–358 (2015). [63] P. Facchi and S. Pascazio, “Quantum zeno subspaces,” [43] A. Cerjan, A. Raman, and S. Fan, “Exceptional Con- Phys. Rev. Lett. 89, 080401 (2002). tours and Band Structure Design in Parity-Time Sym- [64] P. Facchi and S. Pascazio, “Quantum zeno dynamics: metric Photonic Crystals,” Phys. Rev. Lett. 116, 203902 mathematical and physical aspects,” J. Phys. A: Math. (2016). Theor. 41, 493001 (2008). [44] R. El-Ganainy, K. G. Makris, M. Khajavikhan, Z. H. Musslimani, S. Rotter, and D. N. Christodoulides, “Non- 23

[65] M. M. M¨uller,S. Gherardini, and F. Caruso, “Quantum Annual ACM Symposium on Theory of Computing (As- zeno dynamics through stochastic protocols,” Ann. Phys. sociation for Computing Machinery, New York, NY, 529, 1600206 (2017). USA, 2003). [66] D. A. Harville, Matrix Algebra From a Statistician’s Per- [77] Z. Shi, H. Tang, Z. Feng, Y. Wang, Z. Li, J. Gao, spective (Springer New York, 2006). Y. Chang, T. Wang, J. Dou, Z. Zhang, Z. Jiao, W. Zhou, [67] R. Yin, K. Ziegler, F. Thiel, and E. Barkai, “Large and X. Jin, “Quantum fast hitting on glued trees mapped fluctuations of the first detected quantum return time,” on a photonic chip,” Optica 7, 613–618 (2020). Phys. Rev. Res. 1, 033086 (2019). [78] B. Misra and E. C. G. Sudarshan, “The Zeno’s paradox [68] Q. Liu, R. Yin, K. Ziegler, and E. Barkai, “Quantum in quantum theory,” J. Math. Phys. 18, 756–763 (1977). walks: The mean first detected transition time,” Phys. [79] M. C. Fischer, B. Guti´errez-Medina, and M. G. Raizen, Rev. Res. 2, 033113 (2020). “Observation of the quantum zeno and anti-zeno effects [69] M. Miri and A. Al`u,“Exceptional points in optics and in an unstable system,” Phys. Rev. Lett. 87, 040402 photonics,” Science 363 (2019). (2001). [70] K. B. Datta, Matrix and Linear Algebra : Aided with [80] F. Sch¨afer,I. Herrera, S. Cherukattil, C. Lovecchio, F. S. Matlab, Eastern Economy Condition (Prentice Hall India Cataliotti, F. Caruso, and A. Smerzi, “Experimental re- Pvt., Limited, 2009). alization of quantum Zeno dynamics,” Nat. Commun. 5, [71] D. C. Brody, “Biorthogonal quantum mechanics,”J. 3194 (2014). Phys. A: Math. Theor. 47, 35305 (2014). [81] S. Lahiri and A. Dhar, “Return to the origin problem for [72] A. Ghatak and T. Das, “New topological invariants in a particle on a one-dimensional lattice with quasi-Zeno non-Hermitian systems,” J. Phys.: Condens. Matter 31, dynamics,” Phys. Rev. A 99, 012101 (2019). 263001 (2019). [82] S. Belan and V. Parfenyev, “Optimality and universality [73] A. M. Childs, E. Farhi, and S. Gutmann, “An example in quantum Zeno dynamics,” New J. Phys. 22, 073065 of the difference between quantum and classical random (2020). walks,” Quantum Inf. Process. 1, 35–43 (2002). [83] V. Dubey, C. Bernardin, and A. Dhar, “Quantum Dy- [74] S. R. Jackson, T. J. Khoo, and F. W. Strauch, “Quan- namics under continuous projective measurements: non- tum walks on trees with disorder: Decay, diffusion, and Hermitian description and the continuous space limit,” localization,” Phys. Rev. A 86, 022335 (2012). (2020), arXiv:2012.01196. [75] E. Farhi and S. Gutmann, “Quantum computation and [84] F. Thiel and D. A. Kessler, “Non-Hermitian and Zeno decision trees,” Phys. Rev. A 58, 915–928 (1998). limit of quantum systems under rapid measurements,” [76] A. M. Childs, R. Cleve, E. Deotto, E. Farhi, S. Gutmann, Phys. Rev. A 102, 012218 (2020). and D. A. Spielman, “Exponential algorithmic speedup by a quantum walk,” in Proceedings of the Thirty-Fifth