Recovery of damaged information and the out-of-time-ordered correlators

Bin Yan
Center for Nonlinear Studies, Los Alamos National Laboratory, Los Alamos, New Mexico 87545 and
Theoretical Division, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, USA

Nikolai A. Sinitsyn
Theoretical Division, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, USA
(Dated: March 17, 2020)
A time-reversed dynamics unwinds information scrambling, which is induced during the time-
forward evolution with a complex Hamiltonian. We show that if the scrambled information is, in
addition, partially damaged by a local measurement, then such a damage can still be treated by
application of the time-reversed protocol. This information recovery is described by the long-time
arXiv:2003.07267v1 [quant-ph] 16 Mar 2020

saturation value of a certain out-of-time-ordered correlator of local variables. We also propose a

simple test that distinguishes between quantum and reversible classical chaotic information scram-
bling.

In complex strongly correlated systems, local informa-

tion spreads quickly over the whole system, hindering the
scrambled information from local measurements. This
process is characterized by exponentially fast changes of
the out-of-time ordered correlators (OTOCs) [1–4], such
as the following correlator of local operators W and V
with specific time-ordering:
FIG. 1. The protocol: Alice prepares the central qubit with
F (t) = hW † (t)V † (0)W (t)V (0)i. (1) the gate P , and applies the scrambling unitary Û . Bob mea-
sures the central qubit in any basis defined by the gates R and
Recovering the scrambled information is a problem of S. Alice is still able to reconstruct the encoded information
considerable interest [5–7]. Scrambling can encode infor- via a single decoding unitary Û † .
mation into global entanglement, and protect it against
local perturbations. For instance, a single qubit thrown
into a black hole is quickly dispersed and lost behind the correlations between this qubit and the rest of the sys-
horizon. With the resource of early Hawking radiation, tem. According to the no-hiding theorem [12], the useful
only a few qubits of information emitted from the black information is still in the bath, but Alice does not have
hole are needed to reconstruct the lost qubit [8]. How- knowledge of the bath state at any time. How can she
ever, there is usually no simple recipe for how to recover recover the useful information in this case?
the desired information. In this Letter, we show that even after Bob’s measure-
Here, we consider a more accessible scenario for infor- ment, Alice can still recover her information by applica-
mation scrambling and unscrambling. Let us describe tion of the time-reversed protocol. This effect is enabled
it as a hypothetical application of a quantum processor, by the no-hiding theorem and hence distinguishes quan-
such as the one in the quantum supremacy test [9], for tum information scrambling from the scrambling that
hiding quantum information. Our processor can be sim- could be achieved using classical chaotic evolution. In ad-
pler than that in [9] because we will assume that only dition, we reveal the importance of the long-time OTOC
the state of one of the qubits can be prepared and mea- saturation values for quantifying such effects, suggesting
sured, which is suitable for experiments with liquid-NMR a new domain for the OTOC applications.
quantum computers [10, 11]. To explore the game played by Alice and Bob, we con-
Let Alice have such a processor that implements a re- sider the control sequence in Fig. 1 applied to a system
versible but pseudochaotic evolution of many interacting of qubits (spins-1/20 s). The system starts from an initial
qubits. She uses it to hide an original state of one of her product state, ρ0 = |iihi| ⊗ ρB . Here, the central qubit
qubits, which we will call the central qubit. In order to state |ii encodes the information to be scrambled and re-
recover the initial state of this qubit, Alice can apply a covered, whereas the state ρB of the bath qubits can be
time-reversed protocol. arbitrary.
Let Bob be an intruder who can measure the state After a unitary evolution during time t1 , a projective
of the central qubit in any basis unknown to Alice, as measurement along a random axis is applied to the cen-
shown in Fig. 1. If her processor has already scrambled tral spin, without collecting any data from the measure-
the information, Alice is sure that Bob cannot get any- ment outcome. Then, the system evolves backward in
thing useful. However, Bob’s measurement changes the time during t2 , followed by a state tomography for the
state of the central qubit and also destroys all quantum central spin. We claim that when t2 = t1 , the final mea-
 2

surements contain information that can fully reconstruct The fact that it is the projection rather than the Pauli
the initial state of the central qubit. operator that describes Bob’s interference in (5) is impor-
Two remarks are in order. First, we assume that the tant. However, it is useful now to express this OTOC in
unitary evolution is complex enough to scramble the in- terms of Pauli operators, using the identity σ̂φ ≡ 2P̂φ − Î.
formation. In other words, at long times, the reduced We are interested in long scrambling times. The second
density matrix of the central qubit becomes maximally order spin correlators, such as hσ̂f (t1 )σ̂i i, decay quickly
mixed. For instance, the unitary can be drawn from an with time. Hence, we can safely neglect all such correla-
ensemble of random few-qubit unitaries. Without a large tors except the ones that depend on t1 − t2  t1 . The
spin bath the described effect is suppressed. Second, we joint probability is then
recover the initial information at the time t2 = t1 , i.e.,
right after the backward unitary becomes conjugated to 1 1
P rob(P̂f , P̂r ) = + hσ̂f (t1 − t2 )σ̂i i+
the forward unitary but it will be useful to explore what 4 16 (6)
happens for t2 6= t1 . 1
+ hσ̂r (t1 )σ̂f (t1 − t2 )σ̂r (t1 )σ̂i i.
Let, for the central qubit, P̂r be the projection operator 16
for Bob’s measurement. There are two complementary
For t1 = t2 , the second term on the right hand side in (6)
histories in which, after Bob’s measurement, the central
is independent of the evolution unitary. All such details
spin state is projected to the subspace described by ei-
are hidden in the third term. At t1 = t2 , this four-point
ther P̂r or Î − P̂r . The probability of a nonzero result
correlator becomes a standard spin OTOC, i.e,
for Alice’s projective measurement P̂f at the final time
moment is F (t) = hσ̂r (t)σ̂i σ̂r (t)σ̂f i. (7)
Z
P rob(P̂f ) = [dr] P rob(P̂f , P̂r ) + P rob(P̂f , Î − P̂r ), (2)
For finite t and for a small bath this correlator has a
nontrivial system-specific behavior that obscures the con-
where P rob(P̂f , P̂r ) is the joint probability that both Al- tribution of hσ̂f (t1 − t2 )σ̂i i in (6). However, we are in-
ice and Bob find unit measurement outcomes for their terested in the typical complex unitary evolution that
measurement operators. Note that, since we do not scrambles information. Hence, we claim that this corre-
collect results of the Bob’s intermediate measurements, lator saturates to a universal value that is described by
his complementary measurement outcome contributes to its average over an ensemble of random unitaries. This
P rob(P̂f ) as well. The integral over r in (2) accounts average can be evaluated as an integral over all unitaries
for averaging over an arbitrary distribution of possible with respect to the Haar measure [13, 14], i.e.,
directions for Bob’s measurement axes.
Denote Û (t1 ) the evolution operator for the time-
Z
dU tr U † σr U σi U † σr U σf ρB ,
 
forward protocol during time t1 and Û † (t2 ) the evolu- F̄ = (8)
Haar
tion operator for the time-reversed protocol during time
t2 . The probability of the nonzero outcome for the in- where ρB is the initial state of bath qubits. The inte-
termediate measurement P̂r and the corresponding post gral can be further calculated using the identity for Haar
measurement state ρr are then given by unitaries [6]:
P rob(P̂r ) = tr P̂r ρ(t1 )P̂r , ∗
Um1 n1 Um ∗
0 n0 Um2 n2 Um0 n0
(3) 1 1 1 1
ρr = P̂r ρ(t1 )P̂r /P rob(P̂r ), δm1 m01 δm2 m02 δn1 n01 δn2 n02 + δm1 m02 δm2 m01 δn1 n02 δn2 n01
=
where ρ(t1 ) = Û (t1 )ρ0 Û † (t1 ) is the system state at time N2 − 1
t1 . The probability for the final measurement P̂f , condi- δm1 m01 δm2 m02 δn1 n02 δn2 n01 + δm1 m02 δm2 m01 δn1 n01 δn2 n02
+ ,
tioning on the system being projected to the post mea- N (N 2 − 1)
surement state ρr , is (9)
  where N is the dimension of the Hilbert space. Af-
P rob(P̂f |P̂r ) = tr P̂f Û (t2 )ρr Û † (t2 )P̂f . (4) ter summing over all indices and using a trivial identity
tr(σ̂f,i,r ) = 0, the average reduces to
This gives the desired joint probability P rob(P̂f , P̂r ) =
P rob(P̂f |P̂r )P rob(P̂r ). Since the system is initially in F̄ = hσi σf i/(N 2 − 1) ≡ tr(σi σf ⊗ ρB )/(N 2 − 1). (10)
the state ρ0 = |iihi| ⊗ ρB , the joint probability can be
expressed in a compact form: We need this formula only for N  1 because then a
P rob(P̂f , P̂r ) = hP̂r (t1 )P̂f (t1 − t2 )P̂r (t1 )P̂i i. (5) single typical unitary produces the effect that coincides
with the average of the OTOC. The large denominator,
Here, the ensemble average is defined as h•i ≡ tr(• Î ⊗ for N → ∞, makes this OTOC decay to zero. Since
ρB ). Equation (5) shows that the effect of Bob’s interfer- this happens for random unitary evolution, the same is
ence on the information that Alice obtains after apply- true for sufficiently long scrambling times and sufficiently
ing the time-reversed protocol is described by a two-time large baths, so the fourth order correlator in Eq. (6) is
OTOC of projection operators. also vanishing.
 3

After averaging over random unitaries, the joint proba-

bility P rob(P̂f , P̂r ) is the same as P rob(P̂f , Î− P̂r ), so the
final probability is twice of P rob(P̂f , P̂r ) for any distri-
bution of Bob’s measurement axes. Equation (6) reduces
then to
1 1
P rob(P̂f ) = + hσ̂f (t1 − t2 )σ̂i i. (11)
2 8
This is the main result of our work. It shows that Alice’s
measurement probability at t2 = t1 depends only on the
initial state of the central qubit. In addition, Eq. (11)
shows that this information returns to the central qubit
during a short central spin lifetime, as it is described by
a second order spin correlator.
Thus an echo signal shows up in Alice’s measurement
probability near t2 = t1 . The full information about the FIG. 2. Top: The left and right panels correspond to the joint
initial state, i.e., σ̂i , can be reconstructed by detecting measurement probabilities (5) and (13), respectively, in the
this echo. model of a quantum spin bath (12). Bottom: Dynamics of
Finally, we note that the central qubit has not been re- Z-component of the central spin vector in a model (12) with
all classical spins. The green bar marks time t1 with invasive
versed to a completely separable state – this qubit ends
measurement of the central spin. Black dashed curve and
up in a partially mixed state entangled with the bath. blue solid curve correspond to the cases with and without in-
What has been unscrambled is the initial state informa- termediate invasive measurement, respectively. Ns = 10 and
tion, which can be fully extracted either by repeating the Ns = 30 for quantum and classical simulations, respectively.
experiment or by working with a relatively small ensem-
ble of our system with the same initial state of the central
qubit but not necessarily of the bath and scrambling uni- It is instructive to compare the echo in OTOC with
taries. another type of spin echo, which is induced by a similar
The model of a nuclear spin bath. In order to verify protocol, in which the backward evolution, Û † (t2 ), is re-
that the Haar random unitary provides correct descrip- placed by the forward one, Û (t2 ). The joint probability
tion of the universal behavior of OTOCs in realistic sys- is then described by the correlator (5) with t2 replaced
tems, we studied a spin-bath model numerically. This by −t2 , i.e.,
model has been frequently used to describe interactions
of a solid state qubit with nuclear spins [15, 16]. Its P rob2 (P̂f , P̂r ) = hP̂r (t1 )P̂f (t1 + t2 )P̂r (t1 )P̂i i. (13)
Hamiltonian is This correlator can be measured in solid state systems
Ns X by standard means, for example, it was studied exper-
imentally in semiconductor quantum dots [17] (see also
X
H= Jiα S α sα
i , α = x, y, z, (12)
i=1 α
Ref. [18]). As we show in Figure 2 (top right), the final
probability (13) also shows an echo effect near t2 = t1 .
where the couplings Jiα are independent Gaussian dis- However, this echo originates from a finite scrambling
tributed random numbers with zero mean and variance rate and therefore decays at large times, in sharp con-
J. Here, the central spin-1/2, S, interacts with the bath trast to the echo in the OTOC.
of Ns spins-1/2’s, si . We simulated the Schrödinger equa- Quantum vs classical scrambling. Our observation can
tion with this Hamiltonian numerically to obtain the ef- be used to additionally validate quantum supremacy tests
fect of the unitary evolution. such as in [9]. There is only a finite depth for precise sim-
In our simulations, the central spin starts from the | ↑i ulations of classical chaotic motion with classical digital
state, i.e., σi = σz in Eq. (11). The bath spins are pre- computers because of the exponential increase of round-
pared in a maximally mixed state. This corresponds to a off errors [19]. Therefore, a classical analog computer
common situation of practically infinite temperature for with nonlinear interactions between its components can
nuclear spins. Figure 2 (top left) shows the obtained final evolve reversibly to a state that cannot be predicted with
Alice’s probability of a nonzero measurement result for classical digital computers. Hence, to validate quantum
σf = σz and a fixed randomly chosen σr . It does show supremacy, an additional test may be needed to prove
the echo effect along t1 = t2 line, in perfect quantitative that we deal with a quantum scrambled state at the end
agreement with Eq. (11). No additional averaging over rather than with a state generated by classical chaos [20].
Bob’s measurement axes and parameters Jiα is needed for In the classical case, a small change of the state vec-
quantitative agreement with Eq. (11). We also verified tor at the end of the forward protocol would be quickly
numerically for Ns = 10 (not shown) that the same uni- magnified during the backward time evolution. Thus, the
versal result is obtained for several other choices of the state at the end of our control protocol would be strongly
initial density matrix of the spin bath. different from the initial one.
 4

two bath qubits. We also added an ancillary qubit to

effectively realize measurement by entanglement of this
and the central qubits.
The quantum circuit is the same as in Fig. 1. Here,
the three qubit unitary is given by

Û = I ⊗ |01ih00| + σx ⊗ |00ih01|
(14)
− iσy ⊗ |11ih10| − σz ⊗ |10ih11|,

with the corresponding quantum circuit realization

shown in Fig. 3 (top) [22]. It maps any initial state
|ii of the target qubit to the maximally mixed state,
provided that the√ input states of the bath qubits are
|+i = (|0i + |1i)/ 2. This unitary is not a typical Haar
FIG. 3. Top: The realization of the unitary (14). Bottom: random unitary, since its scrambling effect depends on
Statistics of the measurement probabilities obtained with the the bath initial state. Consequently, the four point cor-
IBM-Q processor. Colors label orthogonal final measure- relator at t2 = t1 in Eq. (6), i.e., hÛ σ̂r Û † σ̂f Û σ̂r Û † σ̂i i,
ments of P̂f . Three groups of chart correspond to orthog- is not zero, even after averaging σ̂r over random Pauli
onal intermediate measurements
√ |Zi = |0i,
P̂r . The basis is: √ matrices (corresponding to random intermediate projec-
|Xi = (|0i + |1i)/ 2, and |Y i = (|0i + i|1i)/ 2. The tive measurements). Instead, it makes the above four
final measurement probabilities, after averaging over inter- point correlator vanish when σ̂r is averaged over the Pauli
mediate outcomes, read as 0.5426, 0.5522, and 0.7529, for
group. Thus, if we include the identity operator as part
P̂f = |XihX|, |Y ihY |, and |ZihZ|, respectively.
of the intermediate measurement, the rest of the protocol
and the final result shall not change.
The central qubit starts at |0i. We perform the final
To illustrate this, we simulated the evolution of in-
measurements along three √ orthogonal directions, √ corre-
teracting classical spins with the same Hamiltonian (12)
sponding to |0i, (|0i + |1i)/ 2, and (|0i + i|1i)/ 2. The
subject to the classical Landau-Lifshitz equations [21].
theoretical probabilities for these final measurements are
The state of a classical spin is specified by a three di-
0.75, 0.5, and 0.5, respectively. Figure (3) shows the
mensional unit vector. Our initial state for the central
statistics of the final measurements, obtained with the
spin is a unit vector pointing along the z-axis. Bath
IBM 5-qubit processor, from which we inferred the ini-
spins start from random directions. Instead of quantum
tial state 0.992|0i − (0.082 − 0.101i)|1i. This corresponds
projective measurements, we assumed that classical mea-
to 0.983 fidelity, which means that natural decoherence
surements were invasive. Namely, a measurement resets
was not detrimental.
the classical spin to be directed either along or opposite
In conclusion, OTOCs are unusual correlators that
to the measurement axis with probabilities cos2 (θ/2) or
have no counterpart in classical physics. We showed that
1 − cos2 (θ/2), respectively, where θ ∈ [0, π] is the angle
such correlators quantify not only the information scram-
between the central spin vector and the direction of the
bling rate but also the purely quantum effect of damaged
measurement axis. The time-reversed dynamics was in-
information recovery. This effect must be testable using
duced by changing the sign of all spin coupling constants.
experimental capabilities for the generation of random
Figure 2 (bottom) shows the evolution of the z- unitary matrices.
component of the central classical spin during the time
of the protocol. As expected, an intermediate invasive
measurement of only the central spin has lead then to ACKNOWLEDGMENTS
a state with unrecoverable initial information, in sharp
contrast to the quantum case.
We would like to thank Avadh B. Saxena and Wojciech
IBM-Q experiment. To verify that our predictions are H. Zurek for comments and suggestions. This work was
robust against weak natural decoherence that is always supported by the U.S. Department of Energy, Office of
present in modern quantum computers, we performed an Science, Basic Energy Sciences, Materials Sciences and
experiment with the IBM-Q five-qubit processor. The Engineering Division, Condensed Matter Theory Pro-
main programmed system consisted of one central and gram.

[1] A. Larkin and Y. N. Ovchinnikov, Sov. Phys. JETP 28, Proceedings of the KITP Program: Entanglement in
1200 (1969). Strongly-Correlated Quantum Matter, 2015 (Kavli Insti-
[2] A. Kitaev, A simple model of quantum holography, in tute for Theoretical Physics, Santa Barbara, 2015), Vol.
 5

7. [22] This unitary was used before to demonstrate the no-

[3] Juan Maldacena, Stephen H Shenker, and Douglas hiding theorem on the quantum computers [23].
Stanford, “A bound on chaos,” Journal of High Energy [23] Amolak Ratan Kalra, Navya Gupta, Bikash K Behera,
Physics 2016, 106 (2016). Shiroman Prakash, and Prasanta K Panigrahi, “Demon-
[4] Brian Swingle, “Unscrambling the physics of out-of-time- stration of the no-hiding theorem on the 5-qubit IBM
order correlators,” Nature physics 14, 988–990 (2018). quantum computer in a category-theoretic framework,”
[5] K A Landsman, C Figgatt, T Schuster, N M Linke, Quantum Information Processing 18, 170 (2019).
B Yoshida, N Y Yao, and C Monroe, “Verified quan-
tum information scrambling,” Nature 567, 61–65 (2019).
[6] B Vermersch, A Elben, L M Sieberer, N Y Yao, and
P Zoller, “Probing scrambling using statistical correla-
tions between randomized measurements,” Physical Re-
view X 9, 021061 (2019).
[7] Beni Yoshida and Norman Y Yao, “Disentangling scram-
bling and decoherence via quantum teleportation,” Phys-
ical Review X 9, 011006 (2019).
[8] Patrick Hayden and John Preskill, “Black holes as mir-
rors: quantum information in random subsystems,” Jour-
nal of High Energy Physics 2007, 120 (2007).
[9] Frank Arute et al., “Quantum supremacy using a pro-
grammable superconducting processor,” Nature 574,
505–510 (2019).
[10] Jun Li, Ruihua Fan, Hengyan Wang, Bingtian Ye, Bei
Zeng, Hui Zhai, Xinhua Peng, and Jiangfeng Du, “Mea-
suring Out-of-Time-Order correlators on a nuclear mag-
netic resonance quantum simulator,” Physical Review X
7, 031011 (2017).
[11] Jharana Rani Samal, Arun K Pati, and Anil Kumar,
“Experimental test of the quantum no-hiding theorem,”
Physical Review Letters 106, 080401 (2011).
[12] Samuel L Braunstein and Arun K Pati, “Quantum infor-
mation cannot be completely hidden in correlations: im-
plications for the black-hole information paradox,” Phys-
ical Review Letters 98, 080502 (2007).
[13] J. Cotler, N. Hunter-Jones, J. Liu, and B. Yoshida,
“Chaos, complexity, and random matrices,” J. High En-
ergy Phys. 2017, 48 (2017).
[14] Bin Yan, Lukasz Cincio, and Wojciech H Zurek, “In-
formation scrambling and loschmidt echo,” (2019),
arXiv:1903.02651 [quant-ph].
[15] Nina Fröhling and Frithjof B Anders, “Long-time coher-
ence in fourth-order spin correlation functions,” Physical
Review B 96, 045441 (2017).
[16] Nina Fröhling, Frithjof B Anders, and Mikhail Glazov,
“Nuclear spin noise in the central spin model,” Physical
Review B 97, 195311 (2018).
[17] A Bechtold, F Li, K Müller, T Simmet, P-L Ardelt, J J
Finley, and N A Sinitsyn, “Quantum effects in Higher-
Order correlators of a Quantum-Dot spin qubit,” Physi-
cal Review Letters 117, 027402 (2016).
[18] R-B Liu, S-H Fung, H-K Fung, A N Korotkov, and
L J Sham, “Dynamics revealed by correlations of time-
distributed weak measurements of a single spin,” New
Journal of Physics 12, 013018 (2010).
[19] Pengliang Shi, “A relation on round-off error, attractor
size and its dynamics in driven or coupled logistic map
system,” Chaos 18, 013122 (2008).
[20] Robert Alicki, “On the meaning of “quantum
supremacy” experiments,” (2020), arXiv:2001.00791
[quant-ph].
[21] N A Sinitsyn, Yan Li, S A Crooker, A Saxena, and D L
Smith, “Role of nuclear quadrupole coupling on decoher-
ence and relaxation of central spins in quantum dots,”
Physical review letters 109, 166605 (2012).