Quantum backflow in a ring

Arseni Goussev School of Mathematics and Physics, University of Portsmouth, Portsmouth PO1 3HF, United Kingdom (Dated: February 25, 2021) Free motion of a quantum particle with the entirely comprised of plane waves with non-negative momenta may be accompanied by negative probability current, an effect called quantum backflow. The effect is weak and fragile, and has not yet been observed experimentally. Here we show that quantum backflow becomes significantly more pronounced and more amenable to experimental observation if, instead of letting the particle move along a straight line, one forces it to move in a circular ring.

I. INTRODUCTION is the so-called Bracken-Melloy bound. Finding the exact value of cline remains an open challenge. It is interesting The probability density of a quantum particle may flow to note that cline is independent of the time window T , in the direction opposite to that of the particle’s momen- the particle’s mass µ, or Planck’s constant ~. tum [1, 2], an effect called quantum backflow (QB) [3] [4]. Many questions related to QB have been addressed The effect is inconceivable from the viewpoint of classi- in the literature. These include QB against a constant cal physics, and in this respect can be paralleled with force [7], the pertinence of QB to the arrival-time prob- other genuinely quantum phenomena such as tunneling lem [8–11], position dependence of the backflow cur- or Schr¨odinger’scat states. Unlike the latter, however, rent [5, 12, 13], probability backflow in relativistic quan- QB is relatively unexplored and yet to be observed ex- tum systems [14–16], QB in escape problems [17, 18], and perimentally. QB in many-particle systems [19]. Recently, the prob- The QB effect can be formulated as follows. Consider lem of QB has been generalized to states with position- a nonrelativistic particle moving freely along a straight momentum correlations [20]. line, the x-axis. Let the particle’s wave function be com- As of today, QB has not been experimentally observed prised only of plane waves √1 eikx with nonnegative mo- 2π in any true quantum system [21]. One of the difficul- menta, ~k 0, so that at time t the wave function reads ties hindering experimental observation of QB is a rela- ≥ ∞ ikx tively small value of cline [22]. As pointed out in Ref. [3], Z 2 e ψ(x, t) = dk φ(k)e−i~k t/2µ , (1) one natural strategy for detecting QB would be to work 0 √2π with an electrically charge particle, for which a mea- where µ is the particle’s mass, and φ(k) is a complex- surement of the probability current is equivalent to that R ∞ 2 of the electric current. If the charge of the particle is valued function normalized according to 0 dk φ(k) = +∞ | | q (for concreteness taken to be positive) and the cur- 1 = R dx ψ(x, t) 2. The associated probability cur- −∞ | | rent measurement time is T , then the magnitude of the rent jψ(x, t) is given by detected backflow electric current approximately equals   q R T/2 ~ ∗ ∂ψ T −T/2 dt jψ, which cannot exceed clineq/T . The fact jψ = Im ψ . (2) − µ ∂x that cline is less than 4% hampers the direct detection of the backflow current. The QB effect consists in the fact that jψ can be negative, for some x and t, in spite of the particle’s momentum Another obstacle to observing QB experimentally is being nonnegative with certainty. In other words, even the difficulty of preparing a state with an appreciable though the momentum of a particle is pointing “to the backflow current. Theoretical considerations show [23] right”, the probability density can (locally in space and that states whose integrated backflow current is close to the Bracken-Melloy bound, c , are characterized by in- arXiv:2008.08022v2 [quant-ph] 24 Feb 2021 time) flow “to the left”; this is clearly impossible in the line classical world. finite position width and infinite mean , and there- One of the most surprising features of QB is that the fore are not realizable in a laboratory setting. The value effect has a nontrivial dimensionless scale associated with of the integrated backflow current seems to become signif- it: The probability current through a given point, say x = icantly smaller than cline if one restricts their attention to 0, integrated over an arbitrary time window, T/2 < t < the states with a finite position width and/or finite mean T/2, has a (finite) greatest lower bound. More− precisely energy (see Refs. [3, 24] for some examples), although no [3, 5, 6], systematic study of this question has yet been carried out. Z T/2 In this paper we show that the QB effect becomes much inf dt jψ(0, t) = cline , (3) ψ −T/2 − more pronounced and more amenable to experimental in- vestigation when considered for a quantum particle mov- where ing in a circular ring. In particular, we show that, for the cline 0.0384517 (4) particle-in-a-ring system, the integrated backflow current ' 2 can be over three times larger than the Bracken-Melloy The Schr¨odingerequation, probability density, and prob- bound, and that the corresponding backflow maximizing ability current are invariant under the gauge transforma- state has finite energy (and, by construction, finite spa- tion β β + ∂χ and Ψ eiχΨ, where χ(θ) is an ar- → ∂θ → tial extent). Some space-related aspects of QB in systems bitrary real function. Eigenstates ψm and eigenenergies with rotational motion, such as an in a constant Em of the Hamiltonian satisfy Hψm = Emψm and are magnetic field, have been addressed in Ref. [25] and very given by recently in Ref. [26]. Here however we are interested in imθ 2 the time-dependence of QB, and more specifically look e ~ 2 ψm(θ) = ,Em = (m β) (m Z) . for states maximizing the integrated backflow current. √2π 2µR2 − ∈ (10) The set of eigenstates is orthonormal and complete. II. PARTICLE IN A RING All Hamiltonian eigenstates ψm with m β , where is the ceiling function, have nonnegative≥ d e (gauge- We consider a nonrelativistic particle of mass µ con- invariant)d·e kinetic and probability strained to move in a circular ring of radius R. The ring current. Indeed, ψm is an eigenstate of the kinetic an- lies in the xy-plane of a Cartesian coordinate frame and gular momentum operator ~(`z β) with the eigenvalue − is centered around the origin. The unit vectors along ~(m β) 0, and the probability current corresponding − ≥ ~ the x-, y- and z-axes are denoted by e , e , and e , re- to ψm is Jψ = 2 (m β) 0. Now, in the spirit of x y z m 2πµR − ≥ spectively. The triplet (ex, ey, ez) is right-handed. We the original QB problem, we consider states Ψ(θ, t) com- further assume that the particle has an electric charge prised of the Hamiltonian eigenstates with nonnegative q, for concreteness taken to be positive, and that there kinetic angular momentum: is a constant spatially-uniform magnetic field B pointing ∞ along the z-axis, i.e., B = Bez. X −iEmt/ The particle is described by a time-dependent wave Ψ(θ, t) = cmψm(θ)e ~ , (11) function Ψ(θ, t), where θ is the polar angle between ex m=dβe and the position radius vector of the particle. The wave function is periodic, Ψ(θ + 2π, t) = Ψ(θ, t), and satisfies where, in view of Eq. (7), complex amplitudes cm satisfy ∂Ψ the normalization condition the Schr¨odingerequation i~ ∂t = HΨ with the Hamilto- nian [27, 28] ∞ X 2 2 cm = 1 . (12) ~ 2 | | H = (`z β) , (5) m=dβe 2µR2 − The corresponding probability current is obtained by ∂ where `z = i , so that ~`z is the projection of the substituting Eq. (11) into Eq. (9), and reads − ∂θ canonical angular momentum on ez. Here, ∞ 2 ~ X qR B JΨ(θ, t) = (m + n 2β) β = (6) 2µR2 − 2~c m,n=dβe ∗ ∗ i(Em−En)t/~ is the dimensionless magnetic flux through the ring, with cmcnψm(θ)ψn(θ)e . (13) 2 × c denoting the speed of light. (A constant term ~ has 8µR2 We now focus on the probability current through a to be added to the Hamiltonian, Eq. (5), if the latter is fixed point on the ring, say θ = 0, integrated over a time derived using the Dirac method [27]. This term however window T/2 < t < T/2, and define the dimensionless plays no role in the context of the present work.) The quantity− wave function is normalized to unity, T/2 Z 2π Z dθ Ψ(θ, t) 2 = 1 . (7) PΨ = dt JΨ(0, t) . (14) −T/2 0 | | The probability density, Ψ(θ, t) 2, satisfies the continuity Substituting Eqs. (13) and (10) into Eq. (14), and eval- equation: | | uating the time integral, we find

2 ∞ ∂ Ψ ∂JΨ X ∗ | | + = 0 , (8) PΨ = c Kmncn (15) ∂t ∂θ m m,n=dβe where JΨ(θ, t) is the probability current, defined as with  ∂Ψ β J = ~ Im Ψ∗ ~ Ψ 2 . (9) α   Ψ 2 2 Kmn = (m+n 2β) sinc α(m+n 2β)(m n) . (16) µR ∂θ − µR | | π − − − 3

Here, 0.4 0.1 T = − α = ~ (17) β 4µR2 0.3 0.075 β = − is a (positive) dimensionless parameter, and the sinc 0.2

sin z 1) 0.05 , β = function is defined as sinc z = z if z = 0 and sinc 0 = 1. − Our aim is to investigate the integrated6 probability (0 P 0.1 current, Eq. (15), in view of the normalization condition, β = 0.025 − Eq. (12). Since PΨ is invariant with respect to the trans- formation β β + 1 and c c , m , it is 0.0 m m−1 Z β = 0 sufficient to only→ consider the parametric→ interval∈ -0.1 1 < β 0 . (18) 0.0 0.5 1.0 1.5 2.0 2.5 − ≤ α/π On this interval β = 0, and so Eqs. (15) and (12) take the form d e FIG. 1. Minimum of the time-integrated probability current, ∞ X ∗ Eq. (25), for m1 = 0 and m2 = 1, as a function of α for five PΨ = cmKmncn (19) different values of β. m,n=0 and Figure 1 shows the dependence of (0,1) (correspond- ∞ ing to m = 0 and m = 1) on α for fiveP different values X 2 1 2 cm = 1 , (20) of β. There are two main messages conveyed by this | | m=0 figure. First, it confirms that the integrated probability respectively. Hereafter, we rely on Eqs. (18)–(20). current can indeed be negative. Second, it shows that al- ready for some very simple states (such as a superposition It is worth nothing that PΨ is unbounded from above. This is readily established by taking c = δ , with of ψ0 and ψ1) the magnitude of the integrated negative m mm1 probability current can significantly exceed the Bracken- m1 0, for which PΨ = 2α(m1 β)/π, and observing ≥ − Melloy bound, cline. In fact, numerical evaluation shows that PΨ as m1 . However, the nontrivial (0,1) → ∞ → ∞ that minα,β 0.101727 2.6 cline. questions are whether PΨ can be negative, and whether P ' − ' − × Consideration of cases other than (m1, m2) = (0, 1) inf PΨ is finite. We begin our study by considering an example scenario does not reveal a more pronounced backflow. It is easy to verify that for any (m1, m2), such that 0 m1 < m2, in which Ψ is comprised of only two eigenstates, ψm1 and ≤ ψ with 0 m < m . Thus, we take m2 1 2 (m1,m2)(α, β) ≤ P  ϕ   cos if m = m 0 1 β m1  2 1 = (0,1) α(m m )2, . (26)  ϕ ≥ 2 1 − c = eiγ sin if m = m > m (21) m2 m1 P − m2 m1 m 2 2 1 − −   0 otherwise This scaling relation, in conjunction with the bound on (0,1) established above, implies that with 0 ϕ π and 0 γ < 2π. This parametrization P (m1,m2) ≤ ≤ ≤ minα,β,m1,m2 (α, β) 0.101727. ensures Eq. (20) is fulfilled. Substituting Eq. (21) into We now turnP to the general' − case and minimize the Eq. (19), we obtain integrated probability current PΨ, Eq. (19), subject to α  the normalization constraint on cm, Eq. (20). This PΨ = A B cos ϕ + A sinc(αAB) cos γ sin ϕ , (22) π − problem is equivalent to unconstrained minimization of P∞ ∗ a real-valued functional I[cm] = m,n=0 cnKnmcm where P∞ ∗ − λ n=0 cncn, where λ is a Lagrange multiplier. The cor- responding Euler-Lagrange equation reads A = m1 + m2 2β , B = m2 m1 . (23) − − ∞ We now look for the minimum of P with respect to ϕ X Ψ Kmncn = λcm . (27) and γ (for fixed values of α, β, m1, and m2), i.e. n=0

(m1,m2) (α, β) = min PΨ . (24) Note that both the matrix Knm, defined by Eq. (16), and P ϕ,γ its eigenvalue spectrum λ depend parametrically on α { } A straightforward calculation yields and β. Then, the infimum of PΨ is given by that of the eigenvalue spectrum, i.e. α  q  (m1,m2) 2 2 2 = A B + A sinc (αAB) . (25) (α, β) inf PΨ = inf λ . (28) P π − P ≡ Ψ { } 4

In the limit α 0, which corresponds to R value. A careful numerical investigation yields the follow- and/or T 0, we→ recover the Bracken-Melloy bound:→ ∞ ing estimate for the infimum of the integrated probability → current: lim (α, β) cline . (29) α→0 P → − inf = cring , cring 0.116816 . (31) α,β P − ' This can be readily seen by defining u = m√α and 1/4 It is interesting to note that c is more than three times f(u) = cm/α , and, for α 0 and β fixed, rewrit- ring ing Eq. (27) as → larger than the Bracken-Melloy constant, cline. Figure 2(c) is another blow-up of the curve (α, 0). Z ∞ P 1 2 2 It illustrates the fact, also evident in Fig. 2(b), that the dv (u + v) sinc u v f(v) = λf(u) . (30) dependence of on α has an intricate structure on very π 0 − small scale, asP well as some degree of self-similarity. In The last equation is the integral eigenvalue problem orig- fact, it might be the case that this dependence has a inally formulated by Bracken and Melloy [3], and the in- fractal nature. fimum of its eigenvalue spectrum is cline. We now return to the eigenproblem defined by Eq. (27) In general, for arbitrary α and β,− we compute (α, β) and use it to find a numerical approximation to the numerically. To this end, we truncate the sum in Eq.P (27) backflow-maximizing state. More concretely, we set at a large value n = N (of the order of 1000–10000), com- α/π = 0.3703965 and β = 0 (corresponding to  (N) P' pute the spectrum λ of the corresponding finite- cring), truncate the sum in Eq. (27) at N = 2000, and PN (N) compute− the eigenvector (c , c , . . . , c ). The sought ap- dimensional problem, Kmncn = λ cm, and find 0 1 N n=0 proximation to the backflow-maximizing state, at t = 0, its minimum λ(N) = min λ(N) . We repeat this calcu- min is given by Ψ = PN c ψ [cf. Eq. (11)]. lation for a sequence of increasing N, and extrapolate m=0 m m (N) λmin to N . This procedure yields a numerical esti- 0 → ∞ (N) 10 mate for (α, β) = limN→∞ λmin. Further details on the numericalP evaluation of are provided in Appendix A.

P | 3

0 10− 2 0.4 /c m− (a) 0.1 m c 6 = | 10− β − 0.3 0.075 (a) = 9 β − 10− 0.2 100 101 102 103 = 0.05 m β −

P 0.4 0.1 β = 0.025 − 0.2 T 0.0 ) , t β = 0 0.0 (0 Ψ

-0.1 J 0.2 − 0.0 0.5 1.0 1.5 2.0 2.5 (b) α/π 0.4 − 0.75 0.50 0.25 0.00 0.25 0.50 0.75 0.1167 0.0349 − − − − − t/T P β = 0 β = 0 0.1168 (b) 0.0354 (c) − − FIG. 3. Characteristics of the backflow-maximizing state. (a) 0.364 0.369 0.374 1.47 1.50 1.53 Blue circles show the magnitude of the expansion coefficients α/π α/π cm, for m ≥ 1, in the units of |c0|. The black solid line −2 represents the curve |cm/c0| = m . (b) Probability current FIG. 2. Infimum of the time-integrated probability current, JΨ(0, t) (in units of 1/T ) as a function of time t. The interval Eq. (28), as a function of α. (a) P(α, β) for five different −T/2 < t < T/2 is identified by two red vertical lines. values of β. (b,c) Zoom-ins into P(α, 0). Figure 3(a) shows, on the log-log scale, the dependence Figure 2(a) shows the dependence of on α for the of cm on m for the backflow-maximizing state Ψ. We | | same five values of β as in Fig. 1. We seeP that (α, β) clearly see that decreases as β approaches 0. It is easy to showP that c0 (α, 0) = 0 if α is an integer multiple of π. For all other cm < | 2| m 1 . (32) valuesP of α, the value of (α, 0) appears to be negative. | | m ∀ ≥ Figure 2(b) shows theP curve (α, 0) in a small inter- This inequality ensures that Ψ has a finite mean energy P P 2 2 val around α/π 0.3703965, where attains its smallest E = cm Em with Em m , as given by Eq. (10). ' P h i m | | ∼ 5

More precisely, we find make the description concrete, we take α/π = 0.3703965 and β = 0. (These parameter values correspond to E T cring.) All numerical calculations were performed h i 0.3855 . (33) P' ~ ' in Python and utilized the numpy package. First, using the function numpy.linalg.eigh, we This result is in stark contrast to the fact that mean compute eigenvalues λ(N) of the N-dimensional system energy of the state maximizing probability backflow on a PN K c = λ(N)c , and select the smallest eigen- line is infinite. n=0 mn n m (N) We also compute the time-dependent probability cur- value λmin. We perform this computation for 15 different rent for the backflow-maximizing state Ψ. We do this values of N, ranging between 800 and 10000; the results by numerically evaluating the double sum in Eq. (13) for are as follows: α/π = 0.3703965, β = 0, and θ = 0. Figure 3(b) shows J (0, t) (in units of 1/T ) as a function of t/T . The inte- λ(800) = 0.11681560946083251 , Ψ min − gral of JΨ(0, t) over the time interval T/2 < t < T/2, (1000) − λ = 0.11681562375295221 , identified in the figure by two red vertical lines, gives a min − (1200) value close to cring. It is interesting to observe that, λ = 0.11681563170026898 , − min − unlike in the problem of QB on a line, JΨ(0, t) displays λ(1400) = 0.11681563657782222 , an erratic dependence on time and fails to be everywhere min − negative on the interval T/2 < t < T/2. λ(1600) = 0.11681563974451246 , − min − λ(1800) = 0.11681564184588990 , min − III. CONCLUSION λ(2000) = 0.11681564340085021 , min − λ(2200) = 0.11681564437173106 , In conclusion, we have shown that the backflow effect min − is more pronounced and more amenable to experimental λ(2400) = 0.11681564524093137 , investigation when considered for a particle moving in a min − (3000) circular ring rather than along a straight line. In partic- λ = 0.11681564684342790 , min − ular, the integrated backflow current in the ring scenario (4000) λmin = 0.11681564811514884 , can be as high as cring 0.116816, which is more than − three times larger than' the corresponding bound in the λ(5000) = 0.11681564868561355 , min − case of a line, cline 0.0384517. Also, in the ring case, (6000) ' λ = 0.11681564900305073 , the energy and spatial extent of the backflow-maximizing min − (8000) state are finite; this gives a significant advantage over λmin = 0.11681564932805089 , the line case in which both of these quantities diverge. − λ(10000) = 0.11681564947322964 . Moreover, in the ring case, even very simple states can min − generate substantial backflow: e.g., a superposition of the ground and first-excited states can yield backflow as (N) Then, using the function numpy.polyfit, λmin is fit- high as 87% of the overall bound, cring. These definite ted by the following quadratic polynomial in 1/N: advantages of the particle-in-a-ring system open a new possibility for the first experimental observation of the (N) a1 a2 λ a0 + + (A1) QB effect. min ' N N 2

with ACKNOWLEDGMENTS

a0 = 0.11681564972831678 , − The author would like to thank Maximilien Barbier −8 a1 = 5.3630711822449864 10 , and Remy Dubertrand for helpful discussions. − × a2 = 0.02587490326775755 .

Appendix A: Details of the numerical evaluation The fit, shown in Fig. 4, is very accurate: the corre- of P (α, β) sponding residual (i.e. the sum of the squares of the fit errors) is approximately equal to 7.3 10−20. This allows × (N) Here we give a detailed description of the computa- us to approximate the sought value of limN→∞ λ P ≡ min tional procedure used to evaluate (α, β). In order to by a0. P

[1] G. R. Allcock, “The time of arrival in 311 (1969). III. The measurement ensemble,” Ann. Phys. (N. Y). 53, 6

[14] G. F. Melloy and A. J. Bracken, “Probability Backflow for a Dirac Particle,” Found. Phys. 28, 505 (1998). [15] H. Su and J. Chen, “Quantum backflow in solutions to the Dirac equation of the spin-1/2 free particle,” Mod. Phys. Lett. A 33, 1850186 (2018). [16] J. M. Ashfaque, J. Lynch, and P. Strange, “Relativistic quantum backflow,” Phys. Scr. 94, 125107 (2019). [17] A. Goussev, “Equivalence between quantum backflow and classically forbidden probability flow in a diffraction- in-time problem,” Phys. Rev. A 99, 043626 (2019). [18] W. van Dijk and F. M. Toyama, “Decay of a quasistable quantum system and quantum backflow,” Phys. Rev. A 100, 052101 (2019). [19] M. Barbier, “Quantum backflow for many-particle sys- tems,” arXiv:2005.14685 (2020). [20] A. Goussev, “Probability backflow for correlated quan- tum states,” Phys. Rev. Research 2, 033206 (2020). (N) [21] An experimental scheme for observing QB in Bose- FIG. 4. The fit of λmin (blue circles) by the quadratic poly- nomial in 1/N defined by Eq. (A1) (red curve). Einstein condensates was proposed in Ref. [30]. Also, there has been a recent experimental realization of an optical analog of the QB effect [31]. [22] This difficulty is aggravated by the fragility of the QB [2] J. Kijowski, “On the time operator in quantum mechan- effect in the presence of thermal noise. See: F. Al- ics and the Heisenberg uncertainty relation for energy barelli, T. Guaita, and M. G. A. Paris, “Quantum back- and time,” Rep. Math. Phys. 6, 361 (1974). flow effect and nonclassicality,” Int. J. Quantum Inf. 14, [3] A. J. Bracken and G. F. Melloy, “Probability backflow 1650032 (2016). and a new dimensionless quantum number,” J. Phys. A: [23] J. M. Yearsley, J. J. Halliwell, R. Hartshorn, and Math. Gen. 27, 2197 (1994). A. Whitby, “Analytical examples, measurement models, [4] See Ref. [29] for a short introduction to the quantum and classical limit of quantum backflow,” Phys. Rev. A backflow effect. 86, 042116 (2012). [5] S. P. Eveson, C. J. Fewster, and R. Verch, “Quantum In- [24] J. J. Halliwell, E. Gillman, O. Lennon, M. Patel, and equalities in Quantum Mechanics,” Ann. Henri Poincar´e I. Ramirez, “Quantum backflow states from eigenstates 6, 1 (2005). of the regularized current operator,” J. Phys. A: Math. [6] M. Penz, G. Gr¨ubl,S. Kreidl, and P. Wagner, “A new Theor. 46, 475303 (2013). approach to quantum backflow,” J. Phys. A: Math. Gen. [25] P. Strange, “Large quantum probability backflow and the 39, 423 (2006). azimuthal angle–angular momentum uncertainty relation [7] G. F. Melloy and A. J. Bracken, “The velocity of prob- for an electron in a constant magnetic field,” Eur. J. ability transport in quantum mechanics,” Ann. Phys. Phys. 33, 1147 (2012). (Leipzig) 7, 726 (1998). [26] V. D. Paccoia, O. Panella, and P. Roy, “Angular momen- [8] J. G. Muga, J. P. Palao, and C. R. Leavens, “Arrival tum quantum backflow in the noncommutative plane,” time distributions and perfect absorption in classical and Phys. Rev. A (in press), arXiv:2011.13644 (2020). quantum mechanics,” Phys. Lett. A 253, 21 (1999). [27] A. Scardicchio, “Classical and quantum dynamics of a [9] J. G. Muga and C. R. Leavens, “Arrival time in quantum particle constrained on a circle,” Phys. Lett. A 300, 7 mechanics,” Phys. Rep. 338, 353 (2000). (2002). [10] J. A. Damborenea, I. L. Egusquiza, G. C. Hegerfeldt, and [28] G. A. Vugalter, A. K. Das, and V. A. Sorokin, “A J. G. Muga, “Measurement-based approach to quantum charged particle on a ring in a magnetic field: quantum arrival times,” Phys. Rev. A 66, 052104 (2002). revivals,” Eur. J. Phys. 25, 157 (2004). [11] J. J. Halliwell, H. Beck, B. K. B. Lee, and S. O’Brien, [29] J. M. Yearsley and J. J. Halliwell, “An introduction to “Quasiprobability for the arrival-time problem with links the quantum backflow effect,” J. Phys. Conf. Ser. 442, to backflow and the Leggett-Garg inequalities,” Phys. 012055 (2013). Rev. A 99, 012124 (2019). [30] M. Palmero, E. Torrontegui, J. G. Muga, and M. Mod- [12] M. V. Berry, “Quantum backflow, negative kinetic en- ugno, “Detecting quantum backflow by the density of ergy, and optical retro-propagation,” J. Phys. A: Math. a Bose-Einstein condensate,” Phys. Rev. A 87, 053618 Theor. 43, 415302 (2010). (2013). [13] H. Bostelmann, D. Cadamuro, and G. Lechner, “Quan- [31] Y. Eliezer, T. Zacharias, and A. Bahabad, “Observation tum backflow and scattering,” Phys. Rev. A 96, 012112 of optical backflow,” Optica 7, 72 (2020). (2017).