PHYSICAL REVIEW LETTERS 123, 254101 (2019)

Universal Signature from Integrability to Chaos in Dissipative Open Quantum Systems

*
Gernot Akemann
Faculty of Physics, Bielefeld University, Postfach 100131, 33501 Bielefeld, Germany
and Department of Mathematics, Royal Institute of Technology (KTH),
Brinellvägen 8, 114 28 Stockholm, Sweden
†
Mario Kieburg
School of Mathematics and Statistics, University of Melbourne,
813 Swanston Street, Parkville, Melbourne, Victoria 3010, Australia

Adam Mielke‡
Faculty of Physics, Bielefeld University, Postfach 100131, 33501 Bielefeld, Germany

Tomaž Prosen§
Physics Department, Faculty of Mathematics and Physics, University of Ljubljana, Ljubljana 1000, Slovenia
(Received 15 October 2019; published 18 December 2019)

We study the transition between integrable and chaotic behavior in dissipative open quantum systems,
exemplified by a boundary driven quantum spin chain. The repulsion between the complex eigenvalues of the
corresponding Liouville operator in radial distance s is used as a universal measure. The corresponding level
spacing distribution is well fitted by that of a static two-dimensional Coulomb gas with harmonic potential at
inverse temperature β ∈ ½0; 2. Here, β ¼ 0 yields the two-dimensional Poisson distribution, matching the
integrable limit of the system, and β ¼ 2 equals the distribution obtained from the complex Ginibre
ensemble, describing the fully chaotic limit. Our findings generalize the results of Grobe, Haake, and
Sommers, who derived a universal cubic level repulsion for small spacings s. We collect mathematical
evidence for the universality of the full level spacing distribution in the fully chaotic limit at β ¼ 2. It holds for
all three Ginibre ensembles of random matrices with independent real, complex, or quaternion matrix
elements.

DOI: 10.1103/PhysRevLett.123.254101

Introduction.—It has been a long discussed question Non-Hermitian operators play an equally important role in
how classically integrable and chaotic behavior carries physics, e.g., in disordered systems [12] or quantum chromo-
over to the quantized world. A simple spectral measure dynamics (QCD) with chemical potential [13]. Shortly after
was found in the spacing between neighboring eigenvalues of BGS, the above spectral distinction between integrable and
the corresponding Hamiltonian H. For closed systems it is chaotic was extended by Grobe, Haake, and Sommers (GHS)
Hermitian, H ¼ H† , with real eigenvalues. Berry and Tabor [14] to Markovian dissipative open quantum systems. These
conjectured [1] for quantum integrable systems to generi- follow a Lindblad master equation,
cally follow the one-dimensional (1D) Poisson distribution
ð1DÞ
pP ðsÞ ¼ e−s . In contrast, Bohigas, Giannoni, and Schmit dρ
ðtÞ ¼ LρðtÞ; ð1Þ
(BGS) conjectured [2] (see Ref. [3]) chaotic systems [4] to dt
follow random matrix theory (RMT) statistics in the corre-
sponding symmetry class. Initially Dyson [5] had offered a with L being the Liouville and ρ the density operator; see
first classification within RMT, distinguishing systems with- Ref. [15]. Postponing a detailed discussion for our example of
out or with time reversal at (half-)integer spin which is the a quantum XXZ spin chain—see Refs. [16,17]—the eigen-
celebrated “threefold way.” Much evidence has been given to values of L are real or come in complex conjugate pairs and
support this spectral classification in quantum systems, can be used to characterize integrable or chaotic behavior; see
including neutron scattering, quantum billiards [2], and below. Indeed this has been observed in many examples: for
the hydrogen atom in a magnetic field [6], to name a few; dissipative chaotic systems [18], the QCD Dirac operator
see Refs. [7,8] for standard references. Starting from Berry’s [19], the adjacency matrix of directed graphs [20], and hard-
diagonal approximation [9] the BGS conjecture is now well core bosons with asymmetric hopping on a one-dimensional
understood from a semiclassical expansion [10,11]. lattice at weak disorder [21]. In Ref. [14] GHS studied

0031-9007=19=123(25)=254101(6) 254101-1 © 2019 American Physical Society

 PHYSICAL REVIEW LETTERS 123, 254101 (2019)

periodically kicked tops with damping and the corresponding critical points (here, the real line) of the spectrum. This
discrete quantum map. In the integrable limit they found universality holds for the complex, real [31], and quatern-
agreement between the nearest neighbor spacing in radial ion Ginibre ensemble—to be presented here—and for non-
distance s of its complex bulk eigenvalues and the two- Gaussian extensions [32] of the two former ones. This is in
dimensional (2D) Poisson distribution contrast to random matrices with real spectra, where
quantum chaotic behavior is distinct for the three Dyson
ð2DÞ π 2
pP ðsÞ ¼ se−πs =4 ; ð2Þ classes, corresponding to a 1D log gas at different values
2 β ¼ 1, 2, 4. For complex bulk eigenvalues of chaotic
systems, the possibility of distinguishing their global
which are local quantities. In the fully chaotic limit the spacing
symmetry is thus lost. To prepare our 2D data from the
distribution agrees with the corresponding distribution of the
Liouville operator L for a comparison, we need to unfold
Ginibre ensemble [22] of complex Gaussian non-Hermitian
the complex spectrum. While this is straightforward for real
random matrices (GinUE), given by Ref. [14],
spectra [8], we discuss the literature [19] and present our

Y  ∞ method below.
∞
Γð1 þ k; s2 Þ X
2
2s2jþ1 e−s Integrable and nonintegrable quantum spin chains.—
pGinUE ðsÞ ¼ ; ð3Þ
k¼1
k! j¼1
Γð1 þ j; s2 Þ The system we consider is a Heisenberg XXZ Hamiltonian
H of N spins 1=2, comprising nearest and next-to-nearest
R
with Γð1 þ k; s2 Þ ¼ s∞2 tk e−t dt being the incomplete Γ neighbor interactions,
function. GHS conjectured that the local spectrum of a
generic chaotic dissipative open quantum system in the bulk X
N −1
H¼J ðσ xl σ xlþ1 þ σ yl σ ylþ1 þ Δσ zl σ zlþ1 Þ
should follow the same statistics. This was somewhat l¼1
surprising, as they showed that the complex Ginibre ensemble
leading to Eq. (3) does not satisfy the global symmetries of X
N −2
þ J0 ðσ xl σ xlþ2 þ σ yl σ ylþ2 þ Δ0 σ zl σ zlþ2 Þ; ð5Þ
dissipative open quantum systems [14], unlike its real l¼1
counterpart. Grobe and Haake showed in Ref. [23] that,
based on these symmetries, using perturbative arguments for with J, J0 , Δ, Δ0 ∈ R. We denote the three Pauli matrices by
small distance s, the repulsion is universally cubic. This σ αl , α ¼ x, y, z, for each single spin l ¼ 1; …; N. With each
repulsion is shared by the complex Ginibre ensemble (3), as spin a dephasing operator
well as by a larger class of complex normal random matrices
[24]. The global statistics of Lindblad operators has also been pffiffiffi z
Ll ¼ γσl ; l ¼ 1; …; N and γ > 0 ð6Þ
compared to random matrices; see Refs. [25–28].
Our goals are, first, to provide a further example for the
GHS conjecture for complex spectra of integrable or is associated. Additionally, we introduce dissipation of
quantum chaotic systems to be true, given by boundary polarization at the two ends of the spin chain via the
driven quantum spin chains. These are many-body systems Lindblad operators
with no meaningful semiclassical limit, so the term quan- qffiffiffiffiffi
pffiffiffiffiffi
tum chaos is understood as the absence of integrability or L−1 ¼ γþ þ
L σ1 ; L0 ¼ γ −L σ −1 ;
weak coupling thereof, while its rigorous definition is qffiffiffiffiffi
pffiffiffiffiffi
still lacking. Second, we will show that in the intermediate LNþ1 ¼ γþ þ
R σN ; LNþ2 ¼ γ −R σ −N ; ð7Þ
regime the full spacing distribution is very well described
by a static 2D Coulomb gas at inverse temperature where γ    y
L , γ R > 0 and σ l ¼ σ l  iσ l . The Liouville
x
β ∈ ð0; 2 in a harmonic potential. Its joint distribution operator L acting on a density operator ρ in the master
of theffi set z of N point charges at rescaled positions
pffiffiffiffiffiffiffi equation (1) is given by [16,17]
2=βzi¼1;…;N ∈ C [29] reads [30]
X
N þ2

X  Lρ ¼ −i½H; ρ þ ð2Ll ρL†l − fL†l Ll ; ρgÞ: ð8Þ
N
2 βX N
P β ðzÞ ∝ exp − jzi j þ ln jzi − zj j : ð4Þ l¼−1
i¼1
2 i≠j
The commutator and anticommutator are denoted by ½·; ·
For β ¼ 0 this leads to the Poisson distribution (2) [18], and f·; ·g, respectively; see Ref. [15].
whereas β ¼ 2 corresponds to the level spacing distribution What we are interested in is the spectral statistics of the
(3) [14]. Third, we collect mathematical evidence for the Liouville operator L considered as a ð4N − 1Þ × ð4N − 1Þ
fully chaotic case (3) at β ¼ 2 to be universal in the bulk of real matrix, acting on the vector space of density operators.
the spectrum, regardless of the constraints [14]. With bulk The reduction in dimension by 1 results from the fixed trace
we mean to stay macroscopically away from any edge or condition on ρ and is represented by the identity matrix.

254101-2
 PHYSICAL REVIEW LETTERS 123, 254101 (2019)

The operator L is real because ρ → Lρ preserves the

Hermiticity. The statistics of L should indicate whether
the Lindblad master equation (1) behaves in an integrable
or chaotic way. For this purpose we recall some properties
of the operator L in our example.
Switching off all incoherent processes γ ¼ γ  
L ¼ γ R ¼ 0,
the operator L becomes a real antisymmetric (because of
Trρ1 ½H; ρ2  ¼ −Tr½H; ρ1 ρ2 ) and chiral (due to ½H; ρT ¼ (a) (b)
−½H; ρT ) matrix so that the spectrum becomes 1D and is
purely imaginary and symmetric about the origin. When
also suppressing the next-to-nearest neighbor interactions
(J 0 ¼ 0), the spectrum is completely integrable. With
increasing J0 ≠ 0 chaotic behavior will take over, and
Wigner’s β ¼ 1 statistics in the bulk of the spectrum
applies; see Ref. [33] for a review of the standard 1D
RMT analysis of this setup.
The situation changes drastically when the dissipative (c) (d)
processes are switched on (γ, γ  
L , γ R ≠ 0). Then, the
Liouville operator L becomes a real nonsymmetric FIG. 1. Comparison of the level spacing distributions for
matrix, and its eigenvalues spread into the complex plane. various Liouville operators (8), the analytical spacing distribu-
Nonetheless, there is still a good quantum number tions (2) (Poisson β ¼ 0, dotted line) and (3) (Ginibre β ¼ 2,
which has to be takenP into account—namely, the total dashed line) as well as fits to general Coulomb gas (4)
simulations [(b) and (d), solid lines]. Unfolding (13) is used
spin polarization S ¼ Nl¼1 σ zl . It keeps the coherent
with the smearing parameter σ ¼ 4.5s̄ [see Eq. (13)], where the
processes invariant due to ½H; S ¼ 0, while all additional mean spacing varies from s̄ ¼ 0.0036 to 0.0045 for the datasets
incoherent dissipative processes result in the following (a)–(d). The first moment of all spacings is normalized to unity.
weak symmetry of the Liouvillian [17],

½LðρÞ; S ¼ Lð½ρ; SÞ; ð9Þ γ −L ¼ 1.2, γ þ −

R ¼ γ ¼ 1, and γ R ¼ 0.8. The model is
equivalent to the Fermi-Hubbard chain with imaginary
which is equivalent to the vanishing commutator of the interaction U ¼ iγ with off-diagonal boundaries—see
matrix representations of L and ½S; ·. Ref. [34]—which is known to be Bethe ansatz integrable.
Let js; ni be an eigenstate of H with Sjs; ni ¼ sjs; ni, According to the GHS conjecture, we expect Poisson
and let s ¼ −N=2; −ðN − 2Þ=2; …; N=2. Then, the eigen- statistics for the Liouvillian spectrum; see Fig. 1(a).
value equation of the state js; nihs0 ; n0 j under the adjoint (b) The isotropic Heisenberg XXX chain (Δ ¼ 1)
action of S is with pure-source and pure-sink driving. The parameters
are J0 ¼ 0, γ þ − − þ
L ¼ 0.6, γ R ¼ 1.4, γ L ¼ γ R ¼ γ ¼ 0 in this
½S; js; nihs0 ; n0 j ¼ ðs − s0 Þjs; nihs0 ; n0 j: ð10Þ regime. The steady state (zero mode) of this problem is
known to be exactly solvable [16]; however, the full
Defining M ¼ N − s þ s0 ∈ f0; 1; …; 2Ng, the dimension Liouvillian spectrum shows nonintegrable behavior [see
κ of the eigenspace of the fixed quantum number s − s0 ¼ Fig. 1(b)].
N − M is given by κ ¼ ð2N M Þ − δNM, where the Kronecker (c) The XXX chain (Δ ¼ 1) with arbitrary polarizing
delta represents the identity matrix which obviously boundary driving. Here, we chose the parameters J0 ¼ 0,
belongs to the M ¼ N state space. Therefore, L decom- γþ − þ −
L ¼ 0.5, γ L ¼ 0.3, γ R ¼ 0.3, γ R ¼ 0.9, and γ ¼ 0. The
poses into block matrices, and one needs to study the bulk Hamiltonian of this model is well known to be
spectral statistics of each of these matrices separately. Since integrable via the Bethe ansatz, but with the boundary
we are interested in a good statistical error, it is favorable to driving not even the steady state seems to be exactly
choose M close to N, as then the number of eigenvalues solvable. The spectrum in Fig. 1(c) confirms that its
κ ∼ 22N =N grows exponentially quickly for large N. dynamics is fully chaotic, according to the GHS conjecture.
Comparing data with predictions.—We have generated (d) The XXZ chain with nearest neighbor and next-to-
four realizations of the Liouville operator (8) where, for all nearest neighbor interactions. We have chosen J0 ¼ 1,
four cases, N ¼ 10 and M ¼ 7, and we set the scale to Δ ¼ 0.5, and Δ0 ¼ 1.5 with the same dephasing parameters
J ¼ 1. Thus, we have had in total 77 520 eigenvalues per as in (c). This time, even the bulk Hamiltonian is non-
case to analyze. integrable (quantum chaotic), so we expect Ginibre sta-
(a) The boundary driven XX chain (Δ ¼ 0) with bulk tistics following the GHS conjecture, which is confirmed in
dephasing. The parameters are chosen as J0 ¼ 0, γ þ L ¼ 0.5, Fig. 1(d).

254101-3
 PHYSICAL REVIEW LETTERS 123, 254101 (2019)

All four datasets are depicted in Fig. 1, illustrating the symmetry of their data, Markum et al. [19] proposed
integrable [Fig. 1(a)], intermediate [Fig. 1(b)], and appa- unfolding in strips parallel
Rx to the x axis, in choosing
rently fully chaotic cases [Figs. 1(c) and 1(d)]. Note that the y0 ¼ y and thus x0 ¼ −∞ ρav ðt;yÞdt ¼ uðx;yÞ. Apparently
intermediate case [Fig. 1(b)] flows closer (and is expected for more general datasets this choice is not ideal, e.g.,
to converge) to fully chaotic statistics by increasing the for products of M Ginibre matrices where the density at the
dimension κ. We compare this with the 2D Poisson origin is singular [37]. Its local statistics is known to still
distribution (2), the distribution of the numerically gen- follow the complex Ginibre ensemble [38], making proper
erated Coulomb gas (4) with the best fit for β, and the level unfolding crucial.
spacing distribution (3) of the complex Ginibre ensemble. In fact we found a much simpler method following
The Kolmogorov-Smirnov distances [35] between the Eq. (11) by approximating ρav ðx; yÞ by a sum of Gaussian
empirical distributions of the spectrum of L, and each of distributions around each eigenvalue zj ,
these curves (after fitting β) are listed in Table I. The
spacings for the Coulomb gas are obtained by generating

1 X N
1 2
points with the distribution (4) by using the Metropolis ρav ðx; yÞ ≈ exp − 2 jz − zj j : ð13Þ
algorithm, following Ref. [36], and then determining the 2πσ 2 N j¼1 2σ
spacing numerically. Figure 1 confirms our expectations of
an extended GHS conjecture [14,23] for dissipative open The measured p spacing at ffi a point z0 is then simply
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
quantum systems to hold, even without classically chaotic multiplied by ρav ðx0 ; y0 Þ. Testing this on spectra of
correspondents. the products of random matrices, the choice σ ¼ 4.5s̄ in
Unfolding of complex spectra.—In order to compare the terms of the global mean spacing s̄ leads to very good
spectrum of L with the spectral statistics of the 2D results; see Ref. [39]. This method is applied to our datasets
Coulomb gas (2)–(4), we need to unfold the spectrum. in Figs. 1(a)–(d).
This means that we have to separate the fluctuations (fl), Random matrix universality.—The question raised by the
which are supposedly universal, from the global, averaged conjecture of GHS was why the fully chaotic case should
(av) spectral density, which is system specific: be compared with the predictions of the complex Ginibre
ensemble. Grobe et al. [14] showed that, due to Hermiticity
X
N
ρðx; yÞ ¼ δð2Þ ðz − zi Þ ¼ ρav ðx; yÞ þ ρfl ðx; yÞ; ð11Þ constraints, generic dissipative open quantum systems lead
i¼1 to a spectrum of real and complex conjugate eigenvalue
pairs. Thus, one would expect the real or quaternion
where z ¼ x þ iy. For real spectra unfolding is achieved by Ginibre ensemble (GinOE or GinSE) sharing this property
introducing the cumulative
Rx spectral function and fitting the to apply, and not the GinUE. However, Grobe et al. found
smooth part ηðxÞ ¼ −∞ ρav ðtÞdt [8]. For complex spectra an agreement of their data from periodically kicked tops
this is more involved. Following Ref. [19], unfolding is a with damping with the GinUE—results for the GinOE or
map GinSE were not available at the time.
While the results for the GinSE became available soon
z → z0 ¼ x0 þ iy0 ¼ uðx; yÞ þ ivðx; yÞ; ð12Þ after the publication of Ref. [42], including the spacing
distribution at the origin [which is different from the
to be found, that satisfies certain conditions. First, after GinUE (3)], the GinOE was independently solved much
unfolding the density has to be unity (or constant), later by three groups [31,43,44]. They are given by so-
ρav ðx0 ; y0 Þ ¼ 1, or in other words the Jacobian of the called Pfaffian point processes, with matrix valued kernels
transformation (12) has to cancel the density before as the main building block.
unfolding, dx0 dy0 ¼ ρav ðx; yÞdxdy. This is certainly not Once all density correlation functions are known, all
unique, and we believe that, second, local isotropy has to be spectral information is given, including the spacing. While
achieved, e.g., using conformal maps [19]. Following the close to the real line all three ensembles differ, it was shown
that at the edge of the spectrum the GinSE [45] and GinOE
TABLE I. The Kolmogorov distance between the empirical [31] agree with the GinUE [46]. It is therefore natural to ask
data shown in Fig. 1, the Poisson distribution (2), the fitted value whether or not this agreement continues to hold in the bulk.
for β (specified in the inset) of the Coulomb gas, and the Ginibre For the GinOE this was answered affirmatively in Ref. [31],
spacing distribution (3). and in the Supplemental Material [39] we show that this
also holds for the GinSE. Below we give a heuristic
System Poisson Fitted Coulomb β Ginibre
argument (see also Ref. [11]) for why all three symmetry
(a) 0.015  0.15 classes yield the same spacing distribution in the bulk, and
(b) 0.10 0.0092 (β ¼ 1Þ 0.058 it is thus universal.
(c) 0.15  0.012
The joint probability density function (jpdf) of eigen-
(d) 0.16 0.0094 (β ¼ 1.9) 0.012
values for all three Ginibre ensembles read [22,47]

254101-4
 PHYSICAL REVIEW LETTERS 123, 254101 (2019)

spacings around a bulk eigenvalue z0 , chosen far away

from the real axis and the edge of the support. Close to z0 ,
complex
pffiffiffiffi conjugate and real eigenvalues are of the order
Oð N Þ away from z0 and thus do not contribute to the
local spectral statistics. Hence all jpdf’s (14) become
locally proportional to
Y
∼ jzj − z0 j2 ð15Þ
j∶jzj −z0 j<R

for large N. Thus, all three ensembles coincide locally and

share the GinUE spacing distribution (3). In Fig. 2, we
illustrate this argument with Monte Carlo simulations of all
three Ginibre ensembles in the bulk, finding perfect agree-
ment. Very recently numerical evidence has been given for
FIG. 2. Comparison of the spacing distribution (3) with four further symmetry classes to follow the spacing (3) of
normalized first moment and those of the GinOE (blue crosses),
GinUE (red squares), and GinSE (purple triangles) in the bulk of
the GinUE [50]. While Hamazaki et al. identified two
the spectrum. For the latter we use the standard 2N-dimensional ensembles where the spacing differs, it remains to be seen
representation of an N-dimensional quaternionic matrix, making how many classes emerge in the bulk from the complete list
the complex eigenvalues unique; see Ref. [22]. An ensemble of of non-Hermitian ensembles [51–53].
1000 500 × 500 matrices has been generated in a Monte Carlo Conclusions.—We have studied universal spectral prop-
simulation. Here, the unfolding is trivial due to a uniform density erties of dissipative open quantum systems. Their corre-
of all three ensembles. sponding Liouville operator L generically exhibits complex
eigenvalue statistics. In our example we have numerically
ðkÞ
Y
M Y
k Y
M
diagonalized boundary driven quantum spin chains of the
P GinOE ðzÞ ∝ jΔM ðzÞj2 Δk ðxÞ ðzi −zj Þ jzj − xi j2 XXZ type, with nearest and next-to-nearest neighbor
i;j¼1 i¼1 j¼1
interactions with different sets of couplings. Depending
Y
k
2
Y
M pffiffiffi on these parameters, it is known that the system undergoes
× e−ð1=2Þxl sgn½Imðzj Þerfc½ 2Imðzj Þ a transition from integrable to chaotic behavior. The
l¼1 j¼1
spacing distribution in radial distance between the complex
−ð1=2Þðz2j þz2
j Þ eigenvalues of L has been shown to be an efficient measure
×e ;
to observe this transition. Generalizing the conjecture of
Y
N
P GinUE ðzÞ ∝ jΔN ðzÞj2
2
e−jzj j ; Grobe, Haake, and Sommers for the extreme cases, we have
j¼1 shown that the intermediate statistics is very well described
by a two-dimensional Coulomb gas with harmonic poten-
Y
M Y
M
2 tial by fitting to an inverse temperature β ∈ ½0; 2.
P GinSE ðzÞ ∝ jΔM ðzÞj2 jzi − zj j2 jzj − zj j2 e−jzj j : ð14Þ
i>j j¼1
Furthermore, we have generalized the universality argu-
ment of these authors from a cubic repulsion for small
Q spacing in the chaotic case β ¼ 2 to hold for the full
Here, ΔN ðaÞ ¼ Nj>k ðaj − ak Þ denotes the Vandermonde
determinant and erfc the complementary error function. distribution in all three Ginibre ensembles. Here, we have
The N ¼ k þ 2M eigenvalues in the GinOE are ordered to contributed analytically to the quaternion case and have
yield a positive density, and k counts the number of real illustrated this with numerical evidence.
eigenvalues; see, e.g., Refs. [31,43,44] for details, and for Several open questions deserve further study. While for
the GinSE N ¼ 2M. quantum systems with real eigenvalues the emergence of
pffiffiffiffi random matrix statistics in the chaotic regime is well
For large N there are only k ∝ N real eigenvalues xl on
average [48], and thus we consider M ∼ N=2. Raising the understood, using a semiclassical expansion, such an
Vandermonde to the exponent leads to the Coulomb gas approach is not developed here. Further examples for
picture (4) at β ¼ 2 for the GinUE. Notice that the other physical systems with complex eigenvalues should be
two ensembles are not proportional to jΔN ðzÞjβ for β ¼ 1, studied throughout the transition region from integrable
4. The limiting to chaotic behavior to see whether the description by a 2D
pffiffiffiffi spectral density is constant on a disk of Coulomb gas is indeed universal.
radius Oð N Þ for all three Ginibre ensembles, and also for
Coulomb gases (4) for all β > 0; see, e.g., Ref. [49] for a This work was partly funded by the Wallenberg foun-
review. The local bulk statistics is defined by zooming into dation (G. A.), the German Science Foundation DFG
the vicinity of radius R ¼ Oð1Þ of a few mean level within CRC1283, “Taming uncertainty and profiting from

254101-5
 PHYSICAL REVIEW LETTERS 123, 254101 (2019)

randomness and low regularity in analysis, stochastics, [23] R. Grobe and F. Haake, Phys. Rev. Lett. 62, 2893
and their applications” (G. A. and M. K.), and within (1989).
IRTG2235, “Searching for the regular in the irregular: [24] G. Oas, Phys. Rev. E 55, 205 (1997).
Analysis of singular and random systems” (A. M.), by the [25] S. Denisov, T. Laptyeva, W. Tarnowski, D. Chruściński, and
K. Życzkowski, Phys. Rev. Lett. 123, 140403 (2019).
European Research Council under Advanced Grant
[26] T. Can, arXiv:1902.01442.
No. 694544—OMNES (T. P.), and by the Slovenian [27] T. Can, V. Oganesyan, D. Ograd, and S. Gopalakrishnan,
Research Agency (ARRS) under Program No. P1-0402 arXiv:1902.01414.
(T. P.). Support from the Simons Center for Geometry [28] L. Sa, P. Ribeiro, and T. Prosen, arXiv:1905.02155.
and Physics, Stony Brook University, where part of this [29] The rescaling is made for the limit β → 0 to exist, leading to
work was completed (G. A. and M. K.), is gratefully noninteracting particles confined by a Gaussian potential.
acknowledged, as are the fruitful discussions with [30] P. J. Forrester, Log-Gases and Random Matrices (Princeton
Maurice Duits (G. A.). University Press, Princeton, NJ, 2010).
[31] A. Borodin and C. D. Sinclair, Commun. Math. Phys. 291,
177 (2009).
[32] T. Tao and V. Vu, Ann. Probab. 43, 782 (2015).
* [33] L. D’Alessio, Y. Kafri, A. Polkovnikov, and M. Rigol, Adv.
akemann@physik.uni-bielefeld.de Phys. 65, 239 (2016).
†
m.kieburg@unimelb.edu.au [34] M. V. Medvedyeva, F. H. L. Essler, and T. Prosen, Phys.
‡
amielke@math.uni-bielefeld.de Rev. Lett. 117, 137202 (2016).
§
tomaz.prosen@fmf.uni-lj.si [35] R. R. Wilcox, Introduction to Robust Estimation and
[1] M. V. Berry and M. Tabor, Proc. R. Soc. A 356, 375 (1977). Hypothesis Testing, 2nd ed. (Elsevier, Amsterdam, 2005).
[2] O. Bohigas, M. J. Giannoni, and C. Schmit, Phys. Rev. Lett. [36] D. Chafaï and G. Ferré, J. Stat. Phys. 174, 692 (2019).
52, 1 (1984); J. Phys. Lett. (Paris) 45, 1015 (1984). [37] Z. Burda, R. A. Janik, and B. Waclaw, Phys. Rev. E 81,
[3] G. Casati, F. Valz-Gris, and I. Guarneri, Lett. Nuovo 041132 (2010).
Cimento 28, 279 (1980). [38] G. Akemann and Z. Burda, J. Phys. A 45, 465201 (2012).
[4] P. Walters, An Introduction to Ergodic Theory (Springer, [39] See Supplemental Material at http://link.aps.org/
Heidelberg, 1982). supplemental/10.1103/PhysRevLett.123.254101, which in-
[5] F. J. Dyson, J. Math. Phys. (N.Y.) 3, 1199 (1962). cludes Refs. [40,41], for where details including checks of
[6] D. Wintgen and H. Friedrich, Phys. Rev. A 35, 1464 (1987). our unfolding method as well as a mathematical proof for
[7] H.-J. Stöckmann, Quantum Chaos: An Introduction the bulk universality of the GinSE can be found.
(Cambridge University Press, Cambridge, England, 1999). [40] NIST Handbook of Mathematical Functions, edited by F. W. L
[8] T. Guhr, A. Müller-Groeling, and H. A. Weidenmüller, Olver et al. (Cambridge University Press, Cambridge,
Phys. Rep. 299, 189 (1998). England, 2010).
[9] M. V. Berry, Proc. R. Soc. A 400, 229 (1985). [41] E. Kanzieper, J. Phys. A 35, 6631 (2002).
[10] M. Sieber and K. Richter, Phys. Scr. T90, 128 (2001). [42] M. L. Mehta, Random Matrices, 2nd ed. (Academic Press,
[11] S. Müller, S. Heusler, A. Altland, P. Braun, and F. Haake, New York, 1990).
New J. Phys. 11, 103025 (2009). [43] H.-J. Sommers, J. Phys. A 40, F671 (2007).
[12] N. Hatano and D. R. Nelson, Phys. Rev. Lett. 77, 570 [44] P. J. Forrester and T. Nagao, Phys. Rev. Lett. 99, 050603
(1996). (2007).
[13] M. A. Stephanov, Phys. Rev. Lett. 76, 4472 (1996). [45] B. Rider, J. Phys. A 36, 3401 (2003).
[14] R. Grobe, F. Haake, and H.-J. Sommers, Phys. Rev. Lett. 61, [46] P. J. Forrester and G. Honner, J. Phys. A 32, 2961 (1999).
1899 (1988). [47] N. Lehmann and H.-J. Sommers, Phys. Rev. Lett. 67, 941
[15] H.-P. Breuer and F. Petruccione, The Theory of Open (1991).
Quantum Systems (Oxford University Press, Oxford, 2007). [48] K. B. Efetov, Phys. Rev. Lett. 79, 491 (1997).
[16] T. Prosen, Phys. Rev. Lett. 107, 137201 (2011). [49] S. Serfaty, in Random Matrices, IAS/Park City Mathematics
[17] B. Buča and T. Prosen, New J. Phys. 14, 073007 (2012). Series Vol. 26, edited by A. Borodin, I. Corwin, and A.
[18] F. Haake, Quantum Signatures of Chaos, 3rd ed. (Springer, Guionnet (AMS, Providence, 2019), p. 341–387.
Heidelberg, 2010). [50] R. Hamazaki, K. Kawabata, N. Kura, and M. Ueda, arXiv:
[19] H. Markum, R. Pullirsch, and T. Wettig, Phys. Rev. Lett. 83, 1904.13082.
484 (1999). [51] D. Bernard and A. LeClair, in Statistical Field Theories
[20] B. Ye, L. Qiu, X. Wang, and T. Guhr, Commun. Nonlinear (Springer, Dordrecht, 2002), pp. 207–214.
Sci. Numer Simulat. 20, 1026 (2015). [52] U. Magnea, J. Phys. A 41, 045203 (2008).
[21] R. Hamazaki, K. Kawabata, and M. Ueda, Phys. Rev. Lett. [53] K. Kawabata, K. Shiozaki, M. Ueda, and M. Sato, Phys.
123, 090603 (2019). Rev. X 9, 041015 (2019).
[22] J. Ginibre, J. Math. Phys. (N.Y.) 6, 440 (1965).

254101-6