Nonlinear Dyn (2020) 102:713–732

https://doi.org/10.1007/s11071-020-05856-4

ORIGINAL PAPER

The Lorenz system: hidden boundary of practical stability

and the Lyapunov dimension
N. V. Kuznetsov · T. N. Mokaev ·
O. A. Kuznetsova · E. V. Kudryashova

Received: 16 March 2020 / Accepted: 29 July 2020 / Published online: 11 August 2020
© The Author(s) 2020

Abstract On the example of the famous Lorenz Keywords Global stability · Chaos · Hidden attractor ·
system, the difficulties and opportunities of reliable Transient set · Lyapunov exponents · Lyapunov
numerical analysis of chaotic dynamical systems are dimension · Unstable periodic orbit · Time-delayed
discussed in this article. For the Lorenz system, the feedback control
boundaries of global stability are estimated and the
difficulties of numerically studying the birth of self-
excited and hidden attractors, caused by the loss of
1 Introduction
global stability, are discussed. The problem of reliable
numerical computation of the finite-time Lyapunov
In 1963, meteorologist Edward Lorenz suggested an
dimension along the trajectories over large time inter-
approximate mathematical model (the Lorenz system)
vals is discussed. Estimating the Lyapunov dimension
for the Rayleigh–Bénard convection and discovered
of attractors via the Pyragas time-delayed feedback
numerically a chaotic attractor in this model [76]. This
control technique and the Leonov method is demon-
discovery stimulated rapid development of the chaos
strated. Taking into account the problems of reliable
theory, numerical methods for attractor investigation,
numerical experiments in the context of the shadowing
and still attracts much attention of scientists from dif-
and hyperbolicity theories, experiments are carried out
ferent fields1 (see, e.g., [26,89,96,102,104]). The aim
on small time intervals and for trajectories on a grid of
of this work is to discuss the estimates of the global
initial points in the attractor’s basin of attraction.
stability and the difficulties of numerically studying
the birth of self-excited and hidden attractors, caused
Dedicated to the memory of Gennady A. Leonov. by the loss of global stability.
Consider the classical Lorenz system
N. V. Kuznetsov (B)· ⎧
T. N. Mokaev · O.A. Kuznetsova · E. V. Kudryashova ⎪
⎨ẋ = −σ (x − y),
Mathematics and Mechanics Faculty, Saint Petersburg
ẏ = r x − y − x z, (1)
State University, Saint Petersburg, Russia ⎪
⎩
e-mail: [email protected] ż = −bz + x y,
N. V. Kuznetsov
Faculty of Information Technology, University of with physically sound parameters σ, r > 0, b ∈ (0, 4]
Jyväskylä, Jyväskylä, Finland [76].
N. V. Kuznetsov
Institute for Problems in Mechanical Engineering, RAS, Saint- 1 The original celebrated work by Lorenz [76] has gained more
Petersburg, Russia than 21,500 citations according to Google Scholar.

123
714 N. V. Kuznetsov et al.

An important task in both qualitative and quanti- the system is called globally stable.2 In this definition,
tative investigation of dynamical systems is the study the stationary set can contain both stable (trivial oscil-
of limiting behavior of a system after a transient pro- lations) and unstable equilibrium states, i.e., the local
cess, i.e., the problem of revealing and analysis of all stability of all equilibria is not required3 . Various ana-
possible limiting oscillations. In a theoretical perspec- lytical methods, based on the Lyapunov ideas, have
tive, this problem relates to a generalization [44,63] been developed (see survey, e.g., [47]) for the study of
of the second part of the celebrated Hilbert’s 16th global stability and estimates of its boundaries in the
problem [34] on the number and mutual disposition space of parameters.
of attractors and repellers in the chaotic multidimen- Farther, the birth of nontrivial oscillations in a sys-
sional dynamical systems, and, in particular, their tem can be observed when crossing the global stability
dependence on the degree of polynomials in the model boundary in the space of parameters. In the simplest
(see also corresponding discussion, e.g., in [99,113]). case, this is due to the loss of stability of the station-
Remark that 16th Hilbert problem is far from com- ary set. Within this framework, it is naturally to clas-
plete solution even for two-dimensional polynomial sify oscillations in systems as self-excited or hidden
quadratic systems admitting only periodic attractors [39,62,67,68]. Basin of attraction of a hidden oscil-
(see, e.g., [42,62]). lation in the phase space does not intersect with small
A limiting oscillation (or a set of oscillations) is neighborhoods of any equilibria, whereas a self-excited
called an attractor, if the initial data from its open oscillation is excited from an unstable equilibrium. A
neighborhood in the phase space (with the exception self-excited oscillation is a nontrivial one if it does not
of a minor set of points) lead to a long-term behavior approach the stationary states (i.e., ω-limit set of cor-
that approaches it, and its attracting set is called the responding trajectory does not consist of equilibria).
basin of attraction (i.e., a set of initial data for which In the general case, when the boundaries of global
the trajectories tend to the attractor). stability are violated, the birth of oscillations can occur
The examination of all limiting oscillations could be due to local bifurcations in the vicinity of the stationary
facilitated if for a system it is possible to prove the dis- set (trivial boundary of the global stability)4 or nonlocal
sipativeness in the sense of Levinson (or D-property) bifurcations (hidden boundary of the global stability).
([74], see also [67])—the existence of a bounded con- If an attractor is born via such a nonlocal bifurcation of
vex globally absorbing set containing all such oscilla- the loss of global stability, then the attractor is a hidden
tions. The Lorenz system (1) is dissipative in the sense one since the basin of its attraction is separated from
of Levinson, and one can consider, e.g., the following the locally attractive stationary set.
absorbing set [60,61]: In practice, only problem of the birth of new non-
  trivial attractors is of interest because there are round-
B = (x, y, z) ∈ R3  (x 2 + y 2 + (z − r − σ )2 ) off and truncation errors in numerical experiments and
+r )2  there is noise in physical experiments. That is, if the
≤ b(σ2c , c = min(σ, 1, b2 ). (2)
2 We use the term “global stability” for simplicity of further
This implies that all solutions of (1) exist for t ∈ discussion, while in the literature there are used different terms
[0, +∞) and, thus, system (1) generates a dynamical like “globally asymptotically stable” [108, p. 137], [31, p. 144],
system. For considered assumptions on parameters, if “gradient-like” [72, p. 2], [111, p. 56], “quasi-gradient-like” [72,
r < 1, then system (1) has a unique equilibrium p. 2], [111, p. 56] and others, reflecting the features of the sta-
tionary set and the convergence of trajectories to it.
S0 = (0, 0, 0), and if r > 1, then system (1) has three 3 This can be demonstrated on the example suggested by Vino-
equilibria: S0 = (0, 0, 0) and two symmetric equilib-
grad [109] (see also [32, p. 191]): ẋ = (x 2 +yx 2(y−x)+y
2 5
)(1+(x 2 +y 2 )2 )
,
ria: 2 (y−2x)
ẏ = (x 2 +y 2y)(1+(x 2 +y 2 )2 ) . Here, a planar system with a unique

S± = (± (r − 1)b, ± (r − 1)b, r − 1). globally attractive equilibrium of the saddle type has a family of
homoclinic orbits. Also, global stability in the presence of locally
unstable equilibria is a typical case for systems describing pen-
For a system that is dissipative in the sense of Levin- dulums, PLLs [11,21,46,69], and electric machines [64].
son, further study of existence of oscillations is usually 4 See, e.g., Andronov–Hopf bifurcation [2,5] and Bautin’s
connected with the stability of a stationary set. If any “safe” and “dangerous” boundaries of stability [9], and corre-
trajectory of a system tends to the stationary set, then sponding birth of hidden Chua attractors [43,101].

123
The Lorenz system: hidden boundary of practical stability and the Lyapunov dimension 715

model has a global bounded convex absorbing set, then the Barbashin–Krasovskii theorem is not applica-
over time, the state of the system, observed experimen- ble. In this case, system (1) may possess nontriv-
tally, will be attracted to the local attractor contained in ial self-excited oscillations with respect to unstable
the absorbing set. This problem can be considered as a equilibria and also hidden oscillations.
practical interpretation of the problem of determining (2) For r > 1 and 2σ < b, the absence of nontrivial
the boundary of global stability. oscillations (and, thus, the global stability of the
stationary set {S0 , S± }) can be demonstrated (see,
e.g., [59]) by the LaSalle principal [54]. For that by
2 Inner estimation: the global stability and trivial the time and coordinate transformations
attractors t → τ,√ ψ : (x, y, z) → (χ , ϑ, υ),
τ = t σ (r − 1), χ = √b(rx −1) ,
For the Lorenz system (1), using several approaches √
σ (y−x) bz−x 2
based on the construction of Lyapunov functions it is ϑ= √ , υ= b(r −1)
b(r −1)
possible to obtain the following result. we transform system (1) to the system
⎧ dχ
Theorem 1 (criterion for the absence of self-excited ⎨χ̇ = dτ = ϑ,
⎪
and hidden oscillations) If for parameters of system (1) ϑ̇ = dϑ
= −λϑ − χ υ + χ − χ 3 , (4)
one of the following cases holds: ⎪
⎩
dτ
υ̇ = dυ
dτ = −αυ − βχ ϑ,
2σ > b, with λ = √(σ +1)
, α = √σ (rb −1) > 0, β = 2σσ−b <
2σ ≤ b, or (σ +b)(b+1) (3) σ (r −1)
r < rgs = σ , 0, and consider the following Lyapunov function:
then there are no nontrivial self-excited and hidden υ2 χ4
V (χ , ϑ, υ) = ϑ 2 − − χ2 + . (5)
oscillations in the phase space of system (1), and any β 2
of its solution (x(t), y(t), z(t)) tends to the stationary
set as t → +∞.
Note that the LaSalle principle requires the com-
Proof Let us present the sketch of the proof (see [58, pactness of the set, where the Lyapunov function V is
63,65,73]) and consider the following three cases. defined to show its boundedness from below. In our
(1) For r ≤ 1, system (1) has the unique equilibrium case, one can show that the inequality V (χ , ϑ, υ) >
S0 . Thus, the absence of self-excited oscillations − 21 is valid for any (χ , ϑ, υ) ∈ R3 . Since β < 0, func-
follows from the Routh–Hurwitz criterion on local tion (5) satisfies the following relation on the derivative
stability for the equilibrium S0 . The absence of with respect to system (4):

hidden oscillations can be obtained by Barbashin– α
V̇ (χ , ϑ, υ) = 2 −λϑ 2 + υ 2 ≤ 0,
Krasovskii theorem (see, e.g., [7,47]) and the Lya- β
2
punov function V (x, y, z) = 21 xσ + y 2 + z 2 . For ∀χ , ϑ, υ ∈ R3 .
r ≤ 1, this function satisfies the following relation
From the relation V̇ (χ , ϑ, υ) = 0, it follows
 that the
on the derivative with respect to system (1):
largest invariant set M ⊂ {(χ , ϑ, υ) ∈ R3  V̇ (χ , ϑ, υ)
r +1 2
V̇ (x(t), y(t), z(t)) = − x(t) − 2 y(t) = 0} consists of the equilibrium points of system (4).
Then, according to LaSalle principle any solution of
− 1− (r +1)
2
4 y(t)2 −bz(t)2 <0, ∀x, y, z = 0.
system (4) (and thus system (1)) tends to an equilibrium
Also, V (x, y, z) ≥ 0, V (0, 0, 0) = 0 and state as τ → +∞. The level lines of Lyapunov func-
V (x, y, z) → ∞ as |(x, y, z)| → ∞; thus, all con- tion (5) surrounding the stationary set are presented in
ditions of Barbashin–Krasovskii theorem are satis- Fig. 1.
fied implying global stability of the unique equilib- For 2σ = b and β = 0, the third equation in (4)
rium S0 . yields υ(t) = υ0 exp(−αt) and thus any trajectory
For r > 1, system (1) has three equilibria: the sad- approaches an equilibrium as τ → +∞.
dle equilibrium S0 and the equilibria S± , whose sta- (3) For r > 1, 2σ > b, the previous approach with
bility depends on the parameters r , σ , b. Thus, here Lyapunov function (5) does not work, and we use a

123
716 N. V. Kuznetsov et al.

2 2 2

2
1.5 1.5 2
1.5
101

2
101 1
101 1 1 101
1 1

1
1

10
10

100.1 100.1
99.9 99.9
0.5 99.9 0.5 0
0 0.5 .1 99.9

100
.1 100

100
100

.1
.1

99.6
99.6

99.6
-0.4
ϑ
99.6

-0.4
ϑ 0

0
0 0

-0.3
-0.3

101
101

-0.3

-0.3

1
0

.9
.9

10
10

.9

99
.9 -0.5 99 100.1

99
-0.5 99.9 99 0 .1 -0.5 99.9 99.9

0.1
99.9
0.1

10

101
1
101

0 100.
100. 1
1

-1 -1 -1
101 1 101
101 1 101

2
2
-1.5 -1.5 2 -1.5

-2 -2 -2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2
χ χ χ

(a) υ = −1 (b) υ = 0 (c) υ = 1

Fig. 1 Level lines of Lyapunov function (5) surrounding the stationary set (red, green) of system (4) with parameters r = 24, σ = 10,
b = 2σ + 0.1 in three different projections: υ = −1, υ = 0, υ = 1. (Color figure online)

 
 −σ + μ   
special analytical method based on [40,41,56,95]. The  3  3
  
main idea behind this method is choosing a specific 3 =
−1 + μ 0  > 0, (8)
 3 
nonsingular (3 × 3) matrix S and differentiable scalar  0 −b + μ 
3
function V : U ⊆ R3 → R1 , to satisfy the following  
 √ 
condition: where 3 = r σ − 2z σr , 3 = y σ
2 r . Condition (8)
λ1 (u 0 , S) + λ2 (u 0 , S) + V̇ (u 0 ) < 0, u 0 ∈ B, (6) can be rewritten as follows:

where λ1 (u 0 , S) > λ2 (u 0 , S) > λ3 (u 0 , S) are the r z2

((μ − σ ) (μ − 1) − r σ ) −
eigenvalues of the following symmetrized matrix: σ 4
y 2 (μ − 1)
1 
− + r z > 0. (9)
Q= S J (u 0 )S −1 + (S J (u 0 )S −1 )∗ 4 (μ − b)
2
and For the solutions of system (1), the following relation
⎛ ⎞ is known [60,61]:
−σ σ 0
J (u 0 ) = J (x0 , y0 , z 0 ) = ⎝ r − z 0 −1 −x0 ⎠ lim sup[y 2 (t) + (z(t) − r )2 ] ≤ l 2 r 2 ,
y0 x0 −b t→+∞

is the 3 × 3 Jacobian matrix of system (1). 1, if b < 2,

l= (10)
Condition (6) yields the inequality 1 < r ≤ √b , if b ≥ 2.
2 b−1
(σ +b)(b+1)
σ in (3).
Using relations (9) and (10), we conclude that inequal-
(3.1) For b ∈ [0, 2] or b ∈ [2, 4]⎛and
 1 < r ⎞
<
r
ity (6) is true, when
0 0
⎜ σ ⎟ 2(μ − σ )(μ − 1)
r0 = 2(σσ+b)(b+1)
(1+l )
2 , we consider S = ⎝ 0 1 0⎠ σ + 2 − b ≥ 0, r< ,
σ (1 + l 2 )
0 0 1
and V ≡ 0. Then, condition (6) is equivalent to the which matches the initial conditions on parameter b
following relation: and r .
(3.2) For b ∈ [2, 4] and r0 ≤ r ≤ (σ +b)(b+1) σ ,
Q + μI > 0, (7) ⎛ −1 ⎞
−A 00
where μ = − Tr J (u 0 ) = σ + b + 1 > 0. Condition we consider S = ⎝ − (b−1) σ 1 0 ⎠ and V (x, y, z) =
(7) means that all leading principal minors 1,2,3 of 0 01
the corresponding matrix are positive. For the chosen V1 (x,y,z)
[(σ −1)2 +4σ r ]1/2
, where A = σ r +(σσ−b)(b−1) and
√
matrix S, we have 1 = −σ + μ > 0, 2 = (−b+μ) 3
+
 2
and relation (7) can be expressed in the following V1 (x, y, z) = γ1 (y 2 + z 2 ) + γ2 x 2
3
− y 2 − 2x y) − σb z.
4 x2z
way: +γ3 ( 4σ
x
2 − σ

123
The Lorenz system: hidden boundary of practical stability and the Lyapunov dimension 717

If 2σ ≥ b, then it is known [60,61] the following (0, 0, z0 exp(−bt))

relation for the solutions of system (1): S+

x 2 (t)
 S−
lim inf z(t) − ≥ 0.
8
(11)
t→+∞ 2σ 7

6
Using this relation, one can obtain the following esti-
x2
5
mation for all z ≥ 2σ :
z 4

3
V̇1 ≤ [2r x y − 2dy 2 − 2bz 2 ]γ1 + [2σ x y − 2σ x 2 ]γ2 +
2 5
+[(4σ + 2b)z 2 + 2(σ − r + d)x y − 2(σ − d)y 2 − 2r x 2 ]γ 3
S0
1

+σ z − σb x y. (12) 0
-8
0
y
-6 -5

x
-4 -2 0 2 4
The matrix Q has the following eigenvalues: 6 8

λ2 (u, S) = −b, Fig. 2 The absence of self-excited and hidden attractors and

1 the global stability of the stationary set {S0 , S± } in the Lorenz
λ1,3 (u, S) = − (σ +1)
2 2
± (σ +1−2b)2 + Az− 2σ +A2 y+ b−1
σ x ,
2 2 A
system (1) with parameters σ = 16, b = 4, and r = 6.24 <
(13) (σ +b)(b+1)
σ . Trajectories (blue, purple) in small neighborhoods
of the unstable equilibrium S0 attractors to stable equilibria S±
for which relation λ1 (u, S) > λ2 (u, S) > λ3 (u, S) (trivial attractors). (Color figure online)
holds.
Then, condition (6) takes the following form:
then conditions (14) are equivalent to the following
2(λ1 (u, S) + λ2 (u, S) + V̇ ) conditions:
≤ −(σ + 1 + 2b) + [(σ − 1)2 + 4σ r ]1/2 + b ≥ 1, r σ + (σ − b)(b − 1) > 0,
 2 2 (2r −1)σ (8σ +bσ +b2 −b)
+ [(σ −1)2 2+4σ r ]1/2 −σ z + A 4z b(b+1) > −3σ 2 + 6σ (b − 1) + (b − 1)2 ,

2
+ A4 y + b−1
2
+ V̇1 . which are satisfied for b ∈ [2, 4] and r ≥ r0 .
σ x
Thus, for these values condition (6) takes the fol-
2
If γ1 ≥ 8b + γ3 ( 2σ lowing form:
b + 1), then using (12), one gets the
A

following estimation: 2(λ1 (u, S) + λ2 (u, S) + V̇ )

A2 z 2 A2 ≤ −(σ + 1 + 2b)
−σ z + 4 + 4 (y + σ x )
b−1 2

+V̇1 ≤ B1 x 2 + B2 x y + B3 y 2 , ∀x, y, z ≥ x2 +[(σ − 1)2 + 4σ r ]1/2 < 0,

2σ ,
where which yields estimation r < (σ +b)(b+1)
σ in (3).
A2 (b−1)2
This completes the proof.
B1 = −2γ2 σ + 4σ 2
− 2γ3r, The case corresponding to conditions (3) of The-
A2 orem 1, when all trajectories tend to the stationary
B3 = −2γ1 + 4 − 2(σ − 1)γ3 ,
σ set, however, not all equilibria of the stationary set are
A2 (b−1)
B2 = 2 r γ1 + 4σ − + σ γ2 + (σ + 1 − r )γ3 . locally stable, is illustrated in Fig. 2.
2b
Beyond the estimate (3) in Theorem 1, the analysis
The later inequality B1 x 2 + B2 x y + B3 y 2 ≤ 0 holds of global stability and the birth of nontrivial attractors
for any x, y, if can be performed numerically. It is further known that
B1 ≤ 0, B3 ≤ 0, 4B1 B3 − B22 ≥ 0. (14) the separatrix of saddle S0 can form a homoclinic loop
from which unstable cycles can arise and violate the
If one considers the following coefficients (see, e.g.,
global stability (however, a set of measure zero does
[58]):
not affect the global attraction on a stationary set from
+b−1
γ1 = A2 3σ
8σ (b+2) ,
a practical point of view). Using the Fishing principle
[57,67,71] for the Lorenz system (1), it is possible to
γ3 = A2 8σ +2 ,
b−1
prove the following:
γ2 = r −2
σ (γ3 − γ1 ) − A2 (b+σ −1)
4σ 2
+ 2b ,
1

123
718 N. V. Kuznetsov et al.

equilibria S0 , S± (Fig. 4). This gives an outer estimation

of the practical global stability.
25
S− S+ In this manner, Lorenz had discovered numerically
in the phase space of system (1) the existence of the
20 celebrated chaotic Lorenz attractor from a vicinity of
unstable zero equilibrium.
z
15

10
20
5 4 The boundary of practical stability and absence
0
−10
S0 0
y of nontrivial attractors

x
−5 0 5 −20
10 The presence of an absorbing set (Fig. 5) implies the
existence of a globally attractor Aglob , which contains
Fig. 3 Two symmetric homoclinic orbits (homoclinic butterfly) all local self-excited and hidden attractors, and a sta-
in the Lorenz system (1) with parameters σ = 10, b = 8/3, and tionary set.
r = rh ≈ 13.926
Thus, inside the set B it is possible to study numeri-
cally the presence of nontrivial self-excited and hidden
Theorem 2 For σ and b fixed, there exists r = rh ∈ attractors for parameters r , σ , b not satisfying con-
(1, +∞) corresponding to the homoclinic orbit of the ditions (3) of global stability, i.e., by fixing σ and b
saddle equilibrium S0 if and only if 3σ > 2b + 1. and by decreasing r from rcr . For σ = 10, b = 8/3,
this gives us the following region r ∈ (rgs , rcr ), where
For instance, for the classical values of parameters rgs ≈ 4.64, rcr ≈ 24.74.
σ = 10, b = 8/3 of system (1), it is possible to A nontrivial self-exited attractor can be observed
find numerically the approximate value of such homo- numerically for 24.06  r < rcr ≈ 24.74 (see, e.g.,
clinic bifurcation rh ≈ 13.926, when two symmetric [96]). In this case of nontrivial multistability, system (1)
homoclinic orbits appear forming a homoclinic butter- possesses a local chaotic attractor A which is self-
fly (Fig. 3). A further increase in the parameter r leads excited with respect to equilibrium S0 and coexists with
to the birth of two saddle periodic orbits from each the trivial attractors S± (Fig. 6).
homoclinic orbit [93].

4.1 Hidden attractor or hidden transient chaotic sets?

3 Outer estimation: the absence of trivial
attractors For the Lorenz system (1), it is still an open question
[39, p. 14], whether for some parameters there exists a
For systems with a global absorbing set and an unstable hidden chaotic attractor, i.e., whether it is possible by
stationary set, the existence of self-excited attractors changing parameters to disconnect the basin of attrac-
is obvious. From a computational point of view, this tion from equilibria S0 , S± (e.g., for the parameters
allows one to use a standard computational procedure, σ = 10, b = 83 : if r = 28, then attractor is connected
in which after a transient process, a trajectory, starting with S0 , S± ; if r = 24.5, then attractor is connected
from a point of unstable manifold in a neighborhood with only S0 ). The current results on the existence of the
of equilibrium, reaches a state of oscillation; therefore, hidden attractors in the Lorenz system are the follow-
one can easily identify it. ing. Recently reported hidden attractors in the Lorenz
System (1) possesses the absorbing set B (defined system with r < rcr and locally stable equilibria S±
by Eq. (2)) and for σ > b + 1, r > rcr = σ ( σσ +b+3
−b−1 ) all turn out to be a transient chaotic set (a set in the phase
equilibria are unstable. Thus, in this case, system (1) space, which can persist for a long time, but after all
has a nontrivial self-excited attractor. If we consider collapses), but not a sustained hidden chaotic attractor
classical values of parameters σ = 10, b = 8/3, then [79,112].
for r > rcr , e.g., for r = 28, it is possible to observe Since in a numerical computation of a trajectory
the self-excited chaotic attractor with respect to all three over a finite-time interval it is difficult to distinguish

123
The Lorenz system: hidden boundary of practical stability and the Lyapunov dimension 719

S− S+ S− S+ S− S+
50 50 50

40 40 40

z z z
30 30 30

20 20 20
50 50 50
10 10 10

0 S0 0
y 0 S0 0
y 0 S0 0
y
x x x
−20 −10 −20 −20
0 −50 −10 0 −10 0
10 20 10 −50 10 −50
20 20

(a) Initial data near the equilib- (b) Initial data near the equilib- (c) Initial data near the equilib-
rium S0 rium S+ rium S−

Fig. 4 Numerical visualization of the classical self-excited local chaotic attractor in the Lorenz system (1) with r = 28, σ = 10, b = 8
3
by integrating the trajectories with initial data from small neighborhoods of the unstable equilibria S0 , S±

For the Lorenz system (1), suppose that σ = 10,

b = 83 are fixed and r varies. Near the point r ≈ 24.06,
it is possible to observe a long living transient chaotic
100 B
Ω set, which is hidden since its basin of attraction does
80
not intersect with the small vicinities of equilibrium S0 .
z
60

40
For example, for r = 24 a hidden transient chaotic set
20
can be visualized [112] from the initial point (2, 2, 2)
0
(Fig. 7). In [79], hidden transient chaotic set was
-20
-50 obtained in system (1) with r = 29, σ = 4, b = 2.
50
In our experiments, consider system (1) with r = 24.
y
0
0 For a trajectory with a certain initial point, which is
x -50 50 computed by a certain solver with specific parameters,
we estimate the moment of the end of transient behavior
 absorbing setB (see Eq. (2)) and positive invariant
Fig. 5 Global  as the moment of time when the trajectory falls into a
set  = (x, y, z) ∈ R3  [y 2 (t) + (z(t) − r )2 ] ≤ r 2 (see
Eq. (10)) containing self-excited chaotic attractor in the Lorenz small vicinity of one of the stable equilibria S± . Using
system (1) with r = 28, σ = 10, b = 83 , when the stationary set MATLAB’s standard procedure ode45 with default
{S0 , S± } is unstable parameters (relative tolerance 10−3 , absolute tolerance
10−6 ) for system (1) with parameters r = 24, σ = 10,
b = 8/3 and for initial point u 0 = (20, 20, 20), a tran-
a sustained chaos from a transient chaos [30,52], it is sient chaotic behavior is observed on the time interval
reasonable to give a similar classification for transient [0, 1.8 × 104 ], for initial point u 0 = (−7, 8, 22)—
chaotic sets [14,18]. A transient chaotic set is called a on the time interval [0, 7.2 × 104 ], for initial point
hidden transient chaotic set if it does not involve and u 0 = (2, 2, 2)—on the time interval [0, 2.26 × 105 ],
attract trajectories from a small neighborhood of equi- and for initial point u 0 = (0, −0.5, 0.5) a transient
libria; otherwise, it is called self-excited. chaotic behavior continues over a time interval of more
For an arbitrary system possessing a transient than [0, 107 ]. Remark that, if we consider the same ini-
chaotic set, the time of transient process depends tial points, but use MATLAB’s procedure ode45 with
strongly on the choice of initial data in the phase space relative tolerance 10−6 , for all these initial points the
and also on the parameters of numerical solvers to inte- chaotic transient behavior will last over a time inter-
grate a trajectory (e.g., order of the method, step of inte- val of more than [0, 106 ], and corresponding transient
gration, relative and absolute tolerances). This compli- chaotic sets will not collapse. As a conclusion from
cates the task of distinguishing a transient chaotic set this numerical study of transient chaotic behavior, we
from a sustained chaotic set (attractor) in numerical may suggest to specify precisely numerical solver, its
experiments.

123
720 N. V. Kuznetsov et al.

S− S+
50 50 50

40 40
S− S+ 40
S− S+
z z z
30 30 30

20 20 20
50 50 50

S0
10

S0
10 10
0
0 S0 0
y 0
0
y 0
0
y
x x x
−20 −10 -20 -10
−20 −10
0 10 −50 Y 0 10 -50 0 10 −50 Y
20 20 Y 20

(a) Initial data near the equilib- (b) Initial data near the equilib- (c) Initial data near the equilib-
rium S0 rium S+ rium S−

Fig. 6 Numerical visualization of the self-excited local chaotic attractor in the Lorenz system (1) r = 24.5, σ = 10, b = 83 by a
trajectory that starts in the vicinity of the unstable equilibrium S0 . This attractor coexists with stable equilibria S± (trivial attractors)

45

40

35

30

z z z 25

20

15

S− S+ S− S+ 10 S− S+
5

S0 S0 -20
-10 S0 20
30
10
0 0

x y x y x 10
20 -30
-20
-10
y

5
(a) u0 = (2, 2, 2), t ∈ [0, 1000] (b) u0 = (2, 2, 2), t ∈ [0, 2.3 · 10 ] (c) u0 in vicinity of S0

Fig. 7 Visualization of the hidden transient chaotic set in the Lorenz system (1) with r = 24, σ = 10, b = 8
3

parameters, initial data and time interval, along which with parameters σ, r, b > 0 and parameter a = 0.
the transient behavior continues. Consider
b = 1, a > 0, σ > ar. (16)

4.2 The study of existence of the hidden transient For the following linear transformation (see, e.g., [60]):

chaotic sets and hidden attractors via the ζ ζ
(x, y, z) → x, ,r − y , (17)
continuation method σ − ar σ − ar
system (1) is transformed to the Glukhovsky–Dolzhansky
In [79,112], the study of existence of the hidden tran- system [27]:
sient chaotic sets and hidden attractors in the Lorenz ⎧
system was done by the numerical study of basins of ⎨ẋ = −σ x + ζ z + αyz,
⎪
attraction on some 2d cross sections. Next, we demon- ẏ = ρ − y − x z, (18)
⎪
⎩
strate an approach based on the extension of the param- ż = −z + x y,
eter space and the numerical continuation method (see where
“Appendix A”).
r (σ − ar ) ζ 2a
Consider the following generalization of the classi- ζ > 0, ρ= >0, α = > 0.
cal Lorenz system [76]: ζ (σ − ar )2
⎧ (19)
⎨ẋ = −σ (x − y) − ayz,
⎪ The Glukhovsky–Dolzhansky system describes the
ẏ = r x − y − x z, (15) convective fluid motion inside a rotating ellipsoidal
⎪
⎩
ż = −bz + x y, cavity.

123
The Lorenz system: hidden boundary of practical stability and the Lyapunov dimension 721

If we set Table 1 Initial point (x0 , y0 , z 0 ) and time interval [0, T ] of

numerical integration for each part of the path
a < 0, σ = −ar, (20)
Path (x0 , y0 , z 0 ) T
then after the linear transformation (see, e.g., [60]):
 GD → L (10.64, 60.78, 390) 104
(x, y, z) → ν1−1 y, ν1−1 ν2−1 h x, ν1−1 ν2−1 h z) , t → ν1 t L→R (0.2, 0.2, 0.35) 1.1 × 104

with positive ν1 , ν2 , h, we obtain the Rabinovich sys-

tem [83,87], describing the interaction of three reso-
Consider two line segments, PGD → PL and PL →
nantly coupled waves, two of which being parametri-
PR , defining two parts of the path in the continuation
cally excited:
⎧ procedure. Choose the partition of the line segments
⎨ẋ = hy − ν1 x − yz,
⎪ into Nst = 10 parts and define intermediate values of
ẏ = hx − ν2 y + x z, (21)
i
parameters as follows: PGD→L = PGD + Nist (PL −PGD )
⎪
⎩
ż = −z + x y,
i
and PL→R = PL + Nist (PR −PL ), where i = 1, . . . , Nst .
Initial points for trajectories of system (15) that define
where
hidden chaotic sets are presented in Table 1. At each
σ = ν1−1 ν2 , b = ν1−1 , a = −ν22 h −2 , r = ν1−1 ν2−1 h 2 . iteration of the procedure, a chaotic set defined by the
(22) trajectory in the phase space of system (15) is computed
using MATLAB’s procedure ode45 with relative and
Following [14], let us briefly describe the exper- absolute tolerances equal to 10−8 . The last computed
iment, which allows to organize the transition from point of the trajectory at the previous step is used as the
the hidden attractors in Glukhovsky–Dolzhansky and initial point for computation at the next step.
Rabinovich systems to the hidden transient chaotic set Thus, using numerical continuation procedure and
in the Lorenz system. starting from the hidden attractor in the Glukhovsky–
In this experiment for system (15), we consider three Dolzhansky system (18), it is possible to organize a
sets of parameters: PGD r = 346, a = 0.01, σ = transition and localize numerically the hidden transient
4, b = 1 (for the Glukhovsky–Dolzhansky system chaotic sets in both Lorenz (1) and Rabinovich (21) sys-
— GD), PL r = 24, a = 0, σ = 10, b = 8/3 tems. One can ensure that in all three cases the trajec-
(for the Lorenz system — L), and PR r = 24, a = tories (including separatrices), starting in small neigh-
−1/r − 0.01, σ = −ar, b = bcr + 0.14 (for the borhoods of the unstable equilibrium S0 = (0, 0, 0),
Rabinovich system — R). Here, in contrast to the case are not attracted by the computed chaotic set, but tend
of classical Lorenz system considered in previous sec- to the symmetric stable equilibria S± .
tion, we change the parameters in such a way that hid-
den Glukhovsky–Dolzhansky and Rabinovich attrac-
tors are located not too close to the unstable zero equi- 5 Exact and finite-time Lyapunov dimension &
librium so as to avoid a situation that numerically inte- Lyapunov exponents: global attractor, local
grated trajectory persists for a long time and then falls attractors, and transient chaotic sets
on an unstable manifold of the unstable zero equilib-
rium, then leaves the transient chaotic set, and finally In this section, we will demonstrate the difficulties in
tends to one of the stable equilibria. the reliable numerical computation of the finite-time
Hidden chaotic attractors in the Glukhovsky– Lyapunov exponents (FTLE) and finite-time Lyapunov
Dolzhansky system with PGD r = 346, a = 0.01, σ = dimension (FTLD)5 , as well as distinguishing with
4, b = 1 [66,67] and in the Rabinovich system with their help an attractor from a transition set.
PR r = 24, a = −1/r − 0.01, σ = −ar, b = Before proceeding to the description and discus-
bcr +0.14 [44] could be obtained from the correspond- sion of these numerical experiments, let us remark that
ing self-excited attractors by the numerical continua- investigations of the numerical reliability of computer-
tion method (see “Appendix A”). generated chaotic trajectories have been extensively
These sets of parameters define three points, PGD ,
PL , and PR , in the 4D parameter space (r, a, σ, b). 5 See corresponding definitions in “Appendix B.”

123
722 N. V. Kuznetsov et al.

45
S+ S−
700
40

S+ S−
S+ S− 35
45

30
600 40
25
35

z
500 20

15 30
400 10
25
z

z
5

300
0
20 15 10
S0 0
-20
-40 20

5 0 20
-5 -10 15
-15 -20 40
200
x y
10
100

0 S0 -200
-400 x20 5

0 S0 -40
-20
40 30 0 15 0
20 10
10 5
0 -10 200 0 20
-20 -5 -10
-30 -40 400 -15 40

x y x y

AGD x2T S+ S− ALhid S+ S−

S+ hid
S−
S0

S+ S−
AR
x1T hid
x30
S0 A Lhid S0
x 2T
S0

Step 1 : r = 346, a = 0.01, σ = 4, b = 1, Step 2 : r = 24, a = 0, σ = 10, b = 8/3, Step 3 : r = 24, a = 1/r − 0.01, σ = −ar,
Glukhovsky-Dolzhansky system, Lorenz system,separatrices of S0 → S± ; b = bcr + 0.14, Rabinovich system,
separatrices of S0 → S± ; separatrices of S0 → S± ;

Fig. 8 Localization of hidden chaotic sets in Glukhovsky– point on the (i + 1)th iteration (yellow) is defined as x0i+1 := xiT
Dolzhansky, Lorenz and Rabinovich systems defined by equa- (light green arrows), where xiT = xi (T ) is the final point (yel-
tion (1) using numerical continuation method. Here, trajectories low). Outgoing separatrices of unstable zero equilibrium tend to
xi (t) = (x i (t), y i (t), z i (t) (blue) are defined on the time interval two symmetric stable equilibria. (Color figure online)
[0, T ], (GD → L : T = 104 ; L → R : T = 1.1 × 104 ) and initial

performed in the context of the shadowing theory and existence of fluctuations of the FTLE is a numerical
hyperbolicity breakdown (see, e.g., [29,33,81,84,92]). fingerprint of shadowability breakdown via unstable
A computer-generated chaotic trajectory shadows a dimension variability (see, e.g., [90,105–107]).
“true” chaotic trajectory if the former stays uniformly Since the classical Lorenz system (1) has a property
close to the latter and vice versa. Under this assump- of singular hyperbolicity (or quasi-hyperbolicity), only
tion, the shadowing of numerical trajectories for a rea- finite shadowability is possible for it (see, e.g., [4,16]).
sonable time span would be a minimum requirement Taking into account these problems, the experiments
for a meaningful computer simulation of a dynamical are carried out on small time intervals and for trajecto-
system. The obstructions to shadowability vary from ries on a grid of initial points in the attractor’s basin of
mild to severe. For example, hyperbolic chaotic sys- attraction.
tems have been proved by Anosov to be shadowable
for an infinite period of time [3]. However, if non-
hyperbolic behavior is caused by manifold tangencies, 5.1 The problem of choosing a time interval: attractor
it has been proved that computer-generated chaotic tra- vs transient set
jectories shadow “true” trajectories for a finite period
of time [91]. Finally, strongly non-hyperbolic systems, Consider system (1) with parameters r = 24, σ =
like those with unstable dimension variability [38], can 10, b = 8/3 and integrate numerically the trajec-
be shown to be practically nonshadowable since the tory with initial data u 0 = (20, 20, 20) using MAT-
shadowing time may be very small. In particular, the LAB’s standard procedure ode45 with default param-

123
The Lorenz system: hidden boundary of practical stability and the Lyapunov dimension 723

S− S+ S− S+

z z

u0 u0
S0 S0
x y x y

(a) Trajectory u(t, uinit ) for t ∈ [0, T1 ], (b) Trajectory u(t, uinit ) for t ∈ [0, 106 ].
T1 ≈ 18082.

Fig. 9 The trajectory forms a chaotic set, which looks like an “attractor” (navy blue) and then tends to S+ (cyan). (Color figure online)

eters (relative tolerance 10−3 , absolute tolerance 10−6 ). a time interval of more than [0, 106 ], and the transient
We numerically approximate the finite-time Lyapunov chaotic set will not collapse. If now we choose a new
exponents and finite-time Lyapunov dimension, using initial point u 0 = (10, 10, 10), it would be possible to
the algorithm from [44]. Integration with t > T1 ≈ observe a chaotic transient process over the time inter-
18082 leads to the collapse of the “attractor,” i.e., the val [0, T1 ≈ 6.85 × 105 ], while LE1 (t, u 0 ) > 0 and
“attractor” turns out to be a transient chaotic set. How- dimL (t, u 0 ) > 2 would be satisfied on the time inter-
ever, on the time interval t ∈ [0, T3 ≈ 6.9 × 105 ] vals t ∈ [0, T2,3 ], where T2,3  106 .
we have LE1 (t, u 0 ) > 0 and, thus, may conclude
that the behavior is chaotic, and for the time interval
t ∈ [0, T2 ≈ 3.443 × 105 ] we have dimL (t, u 0 ) > 2. 5.2 The problem of choosing the time interval and
Thus, the estimation of duration of the transient behav- initial data: attractor and embedded unstable
ior by analyzing the sign of the largest Lyapunov expo- periodic orbits
nent could lead to the wrong conclusion.
This numerical phenomenon can be explained due Consider the challenges of the finite-time Lyapunov
to the fact that the finite-time Lyapunov exponents and dimension computation along the trajectories over large
Lyapunov dimension are averaged during computation time intervals (see, e.g., [48–51]), which is connected
over the considered time interval. Since the “lifetime” with the existence of unstable periodic orbits (UPOs)
of a transient chaotic process can be extremely long embedded in chaotic attractor. The “skeleton” of a
and in view of the limitations of reliable integration chaotic attractor comprises embedded unstable peri-
of chaotic ODEs (which we will discuss in the next odic orbits (UPOs) (see, e.g., [1,6,17]), and one of
subsection), even longtime computation of the finite- the effective methods among others for the computa-
time Lyapunov exponents and the finite-time Lyapunov tion of UPOs is the delayed feedback control (DFC)
dimension does not guarantee a relevant approximation approach, suggested by K. Pyragas [85] (see also dis-
of the Lyapunov exponents and the Lyapunov dimen- cussions in [15,45,55]). This approach allows Pyragas
sion. and his progeny to stabilize and study UPOs in various
Remark that, if we choose the same initial point chaotic dynamical systems. Nevertheless, some gen-
u 0 = (20, 20, 20), as in the previous experiment, but eral analytical results have been obtained [35], showing
use MATLAB’s procedure ode45 with relative toler- that DFC has a certain limitation, called the odd num-
ance 10−6 , the chaotic transient process will last over ber limitation (ONL), which is connected with an odd
number of real Floquet multipliers larger than unity.

123
724 N. V. Kuznetsov et al.

1 2.5

0.9

0.8 2

0.7

dimL (t, u0 )
LE1 (t, u0 )

2.055
0.6 1.5
0.82 2.054
0.5
0.81
2.053
0.4 0.8 1
0.79 2.052

0.3 0.78
2.051

0.2 0.77 0.5

0.76 2.05
0 0.5 1 1.5 2 0 0.5 1 1.5 2
10 4 104
0.1

0 t 0
0 0.5 1 1.5 2
t
0 0.5 1 1.5 2
10 4 10 4

(a) LE 1 (t, u0 ), t ∈ [0, T1 ], T1 ≈ 18082. (b) dim L (t, u0 ), t ∈ [0, T1 ], T1 ≈ 18082.

Fig. 10 Numerical computation of the largest FTLE LE1 (t, u 0 ) and FTLD dimL (t, u 0 ) for the time interval [0, T1 ≈ 18082]

2.5
1

0.9
2
0.8 0.01
dimL (t, u0 )

0.7
LE1 (t, u0 )

1.5
0.6
0
6 6.5 7 7.5 8
0.5 10 5

0.4 1

0.3 -0.01

0.2 0.5

0.1

0 t 0
0 2 4 6 8 10
t
0 2 4 6 8 10
10 5 10 5
LE 1 > 0, LD > 2 LE 1 > 0, LD < 2 LE 1 > 0, LD > 2 LE 1 > 0, LD < 2
T1 T2 T3 T1 T2 T3
(a) LE 1 (t, u0 ), t ∈ [0, T = 106 ]. (b) dim L (t, u0 ), t ∈ [0, T = 106 ].

Fig. 11 Numerical computation of the largest FTLE LE1 (t, u 0 ) and FTLD dimL (t, u 0 ) for the time interval [0, 106 ]

 
In order to overcome ONL, later Pyragas suggested a u̇(t) = f (u(t)) + K B FN (t) + w(t) ,
modification of the classical DFC technique, which was ẇ(t) = λ0c w(t) + (λ0c − λ∞
c )FN (t),
called the unstable delayed feedback control (UDFC) N
FN (t) = C ∗ u(t) − (1− R) k=1 R k−1 C ∗ u(t − kT ), (24)
[86].
Rewrite system (1) in a general form where 0 ≤ R < 1 is an extended DFC parameter,
N = 1, 2, . . . , ∞ defines the number of previous states
u̇ = f (u). (23)
involved in delayed feedback function FN (t), λ0c > 0,
upo
Let u upo (t, u 0 1 ) be its UPO with period τ > 0, and λ∞c < 0 are additional unstable degree of freedom
upo upo
u upo (t − τ, u 0 1 ) = u upo (t, u 0 1 ), and initial condi- parameters, B, C are vectors and K > 0 is a feedback
upo
upo1 upo1
tion u 0 = u (0, u 0 ). To compute the UPO and
upo gain. For initial condition u 0 1 and T = τ , we have
overcome ONL, we add the UDFC in the following
form: FN (t) ≡ 0, w(t) ≡ 0,

123
The Lorenz system: hidden boundary of practical stability and the Lyapunov dimension 725

50 S− u(t, uupo1
) S+
u(t, uupo
0
1
) 0

40

z
30
z
S− S+
20

10

0 uupo
0
1
uupo
0
1
ũ(t, uupo
0
1
)
-20
-10 S0 20 S0
0 10
x 10 -10
0
y
20 -20
x y
(a) (b)
upo1
Fig. 12 Period-1 UPO u upo1 (t) (red, period τ1 = 1.5586) stabilized using UDFC method, and pseudo-trajectory ũ(t, u 0 ) (blue,
t ∈ [0, 100]) in system (1) with parameters r = 28, σ = 10, b = 8/3. (Color figure online)

and, thus, the solution of system (24) coincides with We use the adaptive algorithm for the computation
the periodic solution of initial system (23). of the finite-time Lyapunov dimension and exponents
For the Lorenz system (1) with parameters r = 28, for trajectories on the local attractor A [44]. In order
σ = 10, b = 8/3 using (24) with B ∗ = (0, 1, 0), to distinguish the corresponding values for the stabi-
C ∗ = (0, 1, 0), R = 0.7, N = 100, K = 3.5,
upo
lized UPO u(t, u 0 1 ) and for the pseudo-trajectory
λ0c = 0.1, λ∞
upo1
c = −2, one can stabilize a period-1 UPO ũ(t, u 0 ) computed without Pyragas stabilization in
u upo1 (t, u 0 ) with period τ1 = 1.5586... from the ini- our experiment, we use the following notations for
upo
tial point u 0 = (1, 1, 1), w0 = 0 (Fig. 12). Results of finite-time Lyapunov dimensions: dimL (u(t, ·), u 0 1 )
upo1
this experiment could be repeated using various other and dimL (ũ(t, ·), u 0 ), respectively.
numerical approaches (see, e.g., [13,82,110]) and are The comparison of the obtained values of finite-
in agreement with similar results on the existence of time Lyapunov dimensions computed along the sta-
UPOs embedded in the Lorenz attractor [8,25]. How- bilized UPO and the trajectory without stabilization
ever, the Pyragas procedure, in general, is more con- gives us the following results. On the initial small part
venient for UPOs numerical visualization and finds of the time interval, one can observe the coincidence
widespread application in a rich variety of chaos con- of these values with a sufficiently high accuracy. For
trol problems (see, e.g., [12,23,94]). the UPO and for the unstabilized trajectory, the finite-
upo upo
For the initial point u 0 1 ≈ (−6.2262, −11.0027, time local Lyapunov dimensions dimL (u(t, ·), u 0 1 )
upo upo1
13.0515) on the UPO u upo1 (t) = u(t, u 0 1 ), we and dimL (ũ(t, ·), u 0 ) coincide up to the 4th deci-
numerically compute the trajectory of system (24) mal place inclusive on the interval [0, tm1 ≈ 7τ1 ]. After
without the stabilization (i.e., with K = 0) on the t > tm1 , the difference in values becomes significant
time interval [0, T = 100] (Fig. 12b). We denote it by and the corresponding graphics diverge in such a way
upo
ũ(t, u 0 1 ) to distinguish this pseudo-trajectory from that the part of the graph corresponding to the unsta-
upo
the periodic orbit u(t, u 0 1 ). One can see that on the bilized trajectory is lower than the part of the graph
initial small time interval [0, T1 ≈ 11], even without corresponding to the UPO (see Figs. 13b, 14).
upo
the control, the obtained trajectory ũ(t, u 0 1 ) traces The Jacobi matrix at the saddle-foci equilibria S±
upo
approximately the “true” periodic orbit u(t, u 0 1 ). But has simple eigenvalues, which give the following:
upo
for t > T1 , without a control, the trajectory ũ(t, u 0 1 ) dimL S± = 2.0136. The UPO u upo1 with period
upo
diverges from u upo1 (t, u 0 1 ) and visualizes a local τ1 = 1.5586 has the following Floquet multipliers:
chaotic attractor A. ρ1 = 4.7127, ρ2 = 1, ρ3 = −1.19 × 10−10 and
corresponding Lyapunov exponents: { τ11 log ρi }i=1 3 .

123
726 N. V. Kuznetsov et al.

S− u(t, uupo1 analytical value dimL Aglob = dimL S0

0 ) S+ 2.4

2.35 2.075

2.3 2.07

z 2.25 2.065

2.2
2.06

2.15
50 60 70 80 90 100

uupo
2.1
0
1
ũ(t, uupo
0
1
)
S0 2.05

y dimL S±
x 2
0 20 40 60 80 100

(a) (b)
upo upo
Fig. 13 Evolution of FTLDs dimL (u(t, ·), u 0 1 ) (red) and trajectory ũ(t, u 0 1 ) (blue) integrated without stabilization,
upo upo
dimL (ũ(t, ·), u 0 1 ) (blue) computed on the time interval t ∈ respectively. Both trajectories start from the point u 0 1 =
upo
[0, 100] along the UPO u upo1 (t) = u(t, u 0 1 ) (red) and the (−6.2262, −11.0027, 13.0515). (Color figure online)

Fig. 14 Evolution of 2.095 Lorenz system

upo
FTLDs dimL (u(t, ·), u 0 1 ) Finite-time Lyapunov dimension
2.09 along UPO (with Pyragas stabilization, dde23)
(red) and
upo along pseudo-trajectory (no stabilization, ode45)
dimL (ũ(t, ·), u 0 1 ) (blue) 2.085 anlytical value via Floquet multipliers
computed on the long time
2.08
interval t ∈ [0, 10000]
along the UPO 2.075
upo
u upo1 (t) = u(t, u 0 1 ) (red)
and the trajectory
2.07
...
upo
ũ(t, u 0 1 ) (blue) integrated 2.065
without stabilization, ...
2.06
respectively. Both
trajectories start from the 2.055
upo
point u 0 1 =
2.05
(−6.2262, −11.0027, 13.0515).
(Color figure online) 2.045
0 tm
1
20 ts 40 60 80 100 ... 9960 9980 10000

Thus, for the local Lyapunov dimension of this UPO, aging of these values across the attractor, but to tending
we obtain: dimL u upo1 = 2.0678  2.0679 = of these values to the finite-time local Lyapunov expo-
upo
dimL (u(100, ·), u 0 1 ). nents of S0 .
Also remark that system (1) has the analytical solu-
tion u(t) = (0, 0, z 0 exp(−bt)) which tends to the
equilibrium S0 = (0, 0, 0) from any initial point 5.3 Rigorous analytical computation for the global
(0, 0, z 0 ) ∈ R3 . In general, the existence of such solu- attractor
tions in the phase space complicates the procedure of
visualization of a chaotic attractor (pseudo-attractor) by Using an effective analytical technique, proposed by
one pseudo-trajectory with arbitrary initial data com- Leonov [40,56], it is possible to obtain [65,70] the
puted for a sufficiently large time interval (see, e.g., exact formula of the Lyapunov dimension for the global
[79,97]). In particular, the numerical computation of attractor Aglob of the Lorenz system (1):
finite-time local Lyapunov exponents along this trajec-
tory during for any time interval does not lead to aver- dimL Aglob = 3 − √ +b+1)
2(σ
(25)
σ +1+ (σ −1)2 +4σ r

123
The Lorenz system: hidden boundary of practical stability and the Lyapunov dimension 727

for the case, when conditions (3) of the Theorem 1 are good agreement with the rigorous analysis of the time
not satisfied. interval choices for reliable numerical computation of
For the Lorenz system (1) with classical values of trajectories for the Lorenz system. The time interval
parameters r = 28, σ = 10, b = 8/3, we have the for reliable computation with 16 significant digits and
following relations: error 10−4 is estimated as [0, 36], with error 10−8 is
estimated as [0, 26] (see [36,37]), and reliable compu-
dimL Aglob = dimL S0 = 3 − √ +b+1)
2(σ
= 2.4013 >
σ +1+ (σ −1)2 +4σ r tation for a longer time interval, e.g., [0, 10000] in [75],
upo upo is a challenging task that requires a significant increase
> dimL A ≥ dimL u 1 = 2.0678 > dimL (ũ(100, ·), u 0 1 )
= 2.0621 > dimL S± = 2.0136. in the precision of the floating-point representation and
the use of supercomputers.
Here, since the global Lorenz attractor contains a
period-1 UPO: Aglob ⊃ u upo1 , we have the follow-
ing lower-bound estimate for the Lyapunov dimension: 6 Conclusion
dimL Aglob ≥ 2.0678 = dimL u upo1 . Similar experi-
ment and results for the Rössler system [88] are pre- In this work, for the classical Lorenz system, the bound-
sented in [48,50]. aries of practical global stability are estimated and the
Concerning the time of integration, remark that difficulties of numerically studying the birth of self-
while the time series obtained from a physical exper- excited and hidden attractors, caused by the loss of
iment are assumed to be reliable on the whole con- global stability, are discussed. Difficulties in localiza-
sidered time interval, the time series produced by the tion and numerical analysis of hidden transient chaotic
integration of mathematical dynamical model can be sets in the Lorenz system are emphasized in the exper-
reliable on a limited time interval only due to computa- iments on computation of the largest finite-time Lya-
tional errors (caused by finite precision arithmetic and punov exponent and finite-time Lyapunov dimension
numerical integration of ODE). Thus, in general, the along trajectories with different initial data and large
closeness of the real trajectory u(t, u 0 ) and the corre- time intervals. The classical self-excited Lorenz attrac-
sponding pseudo-trajectory ũ(t, u 0 ) calculated numer- tor is considered, and the applications of the Pyragas
ically can be guaranteed on a limited short time interval time-delayed feedback control technique and Leonov
only. analytical method are demonstrated for the Lyapunov
In our experiment, if we continue computation over dimension estimation, as well as for the verification of
a long time interval [0, 10000] of FTLD along the sta- the famous Eden’s conjecture on the maximum value of
bilized UPO and the pseudo-trajectory obtained with- this characteristic that could be achieved on the attrac-
out Pyragas stabilization, as a result, completely dif- tor. The problem of reliable numerical computation of
ferent values will be obtained (Fig. 14). Evolution the finite-time Lyapunov dimension along the trajecto-
upo
of dimL (u(t, ·), u 0 1 ) along the stabilized UPO will ries over large time intervals is discussed.
tend to the analytical value dimL u upo1 = 2.0678,
computed via Floquet multipliers, while evolution of Acknowledgements Open access funding provided by Uni-
upo versity of Jyväskylä (JYU). The authors declare that they have
dimL (ũ(t, ·), u 0 1 ) along the pseudo-trajectory will no conflict of interest. We dedicate this paper to the memory
converge to the value 2.06226 . These results are in of Gennady A. Leonov (1947–2018), with whom we began this
work in 2018. This study was partially funded by the Russian
6 The following results on the dimension of the Lorenz attrac- Science Foundation (Project 19-41-02002).
tor with parameters r = 28, σ = 10, b = 8/3 can be found
in the literature. In [28, p. 193] and [10, p. 3529], the fractal Open Access This article is licensed under a Creative Com-
(box counting, capacity) dimension is estimated as 2.06 ± 0.01. mons Attribution 4.0 International License, which permits use,
For the correlation dimension, the following results are known: sharing, adaptation, distribution and reproduction in any medium
2.05 ± 0.01 in [28, p. 193] and [103, p. 456]; 2.06 ± 0.03 in [78, or format, as long as you give appropriate credit to the original
p. 47]; 2.049 ± 0.096 in [100, p. 1874]; 2.05 in [24, p. 80]. For author(s) and the source, provide a link to the Creative Com-
the Lyapunov dimension, the following values have been com- mons licence, and indicate if changes were made. The images or
puted: 2.063 in [77, p. 92] and [80, p. 1957]; 2.05 in [19, p. 267]; other third party material in this article are included in the article’s
2.062 in [100, p. 1874], [97, p. 115] and [53, p. 53]; 2.06215 [98, Creative Commons licence, unless indicated otherwise in a credit
p. 033124-3] and [24, p. 83]. Also, let us mention estimates for line to the material. If material is not included in the article’s Cre-
the global attractor: 2.401 ≤ dimL Aglob ≤ 2.409 [22, p. 170] ative Commons licence and your intended use is not permitted by
and dimL Aglob ≈ 2.401... in [19, p. 267]. statutory regulation or exceeds the permitted use, you will need

123
728 N. V. Kuznetsov et al.

to obtain permission directly from the copyright holder. To view to system (23) with λ = λ j+1 , a loss of stability bifur-
a copy of this licence, visit http://creativecommons.org/licenses/ cation is observed and attractor A j vanishes. If, while
by/4.0/.
changing λ from λbegin to λend , there is no loss of sta-
bility bifurcation of the considered attractors, then a
hidden attractor for λk = λend (at the end of the proce-
Appendix A: Hidden attractors localization via dure) is localized.
numerical continuation method

One of the effective methods for numerical localiza- Appendix B: Finite-time and limit values of Lya-
tion of hidden attractors in multidimensional dynam- punov exponents and Lyapunov dimension
ical systems is based on the homotopy and numerical
continuation method (NCM). The idea is to construct a Following [40,44], let us outline the concept of the
sequence of similar systems such that for the first (start- finite-time Lyapunov dimension, which is convenient
ing) system the initial point for numerical computa- for carrying out numerical experiments with finite time.
tion of oscillating solution (starting oscillation) can be For a fixed t ≥ 0, let us consider the map u(t, ·) :
obtained analytically. Thus, it is often possible to con- R → R3 defined by the shift operator along the solu-
3

sider the starting system with self-excited starting oscil- tions of system (1): u(t, u 0 ), u 0 ∈ R3 . Since system (1)
lation; then, the transformation of this starting oscilla- possesses an absorbing set, the existence and unique-
tion in the phase space is tracked numerically while ness of solutions of system (1) for t ∈ [0, +∞) take
passing from one system to another; the last system place and, therefore, the system generates a dynamical
corresponds to the system in which a hidden attractor system {u(t, ·)}t≥0 , (R3 , | · |) .
is searched. Consider linearization of system (1) along the solu-
For studying the scenario of transition to chaos, we tion u(t, u 0 ) and its 3 × 3 fundamental matrix of
consider system (23) with f (u) = f (u, λ), where solutions (t, u 0 ): (t,˙ u 0 ) = D f (u(t, u 0 ))(t, u 0 ),
λ ∈  ⊂ Rd is a vector of parameters, whose variation where (0, u 0 ) = I is a unit 3 × 3 matrix. Denote
in the parameter space  determines the scenario. Let by σi (t, u 0 ) = σi ((t, u 0 )), i = 1, 2, 3, the singular
λend ∈  define a point corresponding to the system, values of (t, u 0 ) (i.e., the square roots of the eigen-
where a hidden attractor is searched. Choose a point values of the symmetric matrix (t, u 0 )∗ (t, u 0 ) with
λbegin ∈  such that we can analytically or numeri- respect to their algebraic multiplicity)8 , ordered so that
cally localize a certain nontrivial (oscillating) attractor σ1 (t, u 0 ) ≥ σ2 (t, u 0 ) ≥ σ3 (t, u 0 ) > 0 for any u 0 ∈ R3
A1 in system (23) with λ = λbegin (e.g., one can con- and t > 0.
sider an initial self-excited attractor defined by a tra- Consider a set of finite-time Lyapunov exponents at
jectory u 1 (t) numerically integrated on a sufficiently the point u 0 :
large time interval t ∈ [0, T ] with initial point u 1 (0) 1
in the vicinity of an unstable equilibrium). Consider LEi (t, u 0 ) = ln σi (t, u 0 ), t > 0, i = 1, 2, 3. (26)
t
a path7 in the parameter space  , i.e., a continuous
function γ : [0, 1] → , for which γ (0) = λbegin Here, the set {LEi (t, u 0 )}i=1
3 is ordered by decreasing
and γ (1) = λend , and a sequence of points {λ j }kj=1 (i.e., LE1 (t, u 0 ) ≥ LE2 (t, u 0 ) ≥ LE3 (t, u 0 ) for all
t > 0). The finite-time local Lyapunov dimension [40,
on the path, where λ1 = λbegin , λk = λend , such
44] can be defined via an analog of the Kaplan–Yorke
that the distance between λ j and λ j+1 is sufficiently
formula with respect to the set of ordered finite-time
small. On each next step of the procedure, the initial
Lyapunov exponents {LEi (t, u 0 )}i=1 3 :
point for a trajectory to be integrated is chosen as the
last point of the trajectory integrated on the previous LE1(t,u 0 )+···+LE j (,u 0 )(t,u 0 )
dimL (t, u 0 ) = j (t, u 0 ) + | LE j (t,u 0 )+1 (t,u 0 )| ,
step: u j+1 (0) = u j (T ). Following this procedure and
sequentially increasing j, two alternatives are possible: (27)
the point of A j are in the basin of attraction of attractor m
where j (t, u 0 ) = max{m : i=1 LEi (t, u 0 ) ≥ 0}.
A j+1 , or while passing from system (23) with λ = λ j Then, the finite-time Lyapunov dimension of dynamical
7 In the simplest case, when d = 1, the path is a line segment. 8 Symbol ∗ denotes the transposition of matrix.

123
The Lorenz system: hidden boundary of practical stability and the Lyapunov dimension 729

system with respect to a set A is defined as: 14. Chen, G., Kuznetsov, N., Leonov, G., Mokaev, T.: Hidden
attractors on one path: Glukhovsky–Dolzhansky, Lorenz,
dimL (t, A) = sup dimL (t, u 0 ). (28) and Rabinovich systems. Int. J. Bifurc. Chaos Appl. Sci.
u 0 ∈A Eng. 27(8), 1750115 (2017)
15. Chen, G., Yu, X.: On time-delayed feedback control of
The Douady–Oesterlé theorem [20] implies that for
chaotic systems. IEEE Trans. Circuits Syst. I Fundam. The-
any fixed t > 0 the finite-time Lyapunov dimension ory Appl. 46(6), 767–772 (1999)
on a compact invariant set A, defined by (28), is an 16. Coomes, B., Kocak, H., Palmer, K.: Rigorous computa-
upper estimate of the Hausdorff dimension: dimH A ≤ tional shadowing of orbits of ordinary differential equa-
tions. Numer. Math. 69(4), 401–421 (1995)
dimL (t, A). The best estimation is called the Lyapunov
17. Cvitanović, P.: Periodic orbits as the skeleton of classical
dimension [40] and quantum chaos. Phys. D Nonlinear Phenom. 51(1–3),
138–151 (1991)
dimL A= inf sup dimL (t, u 0 )= lim inf sup dimL (t, u 0 ). 18. Danca, M.F., Kuznetsov, N.: Hidden chaotic sets in a Hop-
t>0 u ∈A t→+∞ u 0 ∈A
0 field neural system. Chaos Solitons Fractals 103, 144–150
(2017). https://doi.org/10.1016/j.chaos.2017.06.002
19. Doering, C.R., Gibbon, J.: On the shape and dimension
of the Lorenz attractor. Dyn. Stab. Syst. 10(3), 255–268
(1995)
References 20. Douady, A., Oesterle, J.: Dimension de Hausdorff des
attracteurs. C.R. Acad. Sci. Paris, Ser. A. (in French)
1. Afraımovic, V., Bykov, V., Silnikov, L.: On the origin and 290(24), 1135–1138 (1980)
structure of the Lorenz attractor. Dokl. Akad. Nauk SSSR 21. Dudkowski, D., Jafari, S., Kapitaniak, T., Kuznetsov, N.,
234(2), 336–339 (1977) Leonov, G., Prasad, A.: Hidden attractors in dynamical
2. Andronov, A., Vitt, E., Khaikin, S.: Theory of Oscillators. systems. Phys. Rep. 637, 1–50 (2016). https://doi.org/10.
Pergamon Press, Oxford (1966) 1016/j.physrep.2016.05.002
3. Anosov, D.: Geodesic flows on closed Riemannian mani- 22. Eden, A., Foias, C., Temam, R.: Local and global Lyapunov
folds of negative curvature. Proc. Steklov Inst. Math. 90(1) exponents. J. Dyn. Differ. Equ. 3(1), 133–177 (1991).
(1967) https://doi.org/10.1007/BF01049491. [Preprint No. 8804,
4. Araujo, V., Pacifico, M., Pujals, E., Viana, M.: Singular- The Institute for Applied Mathematics and Scientific Com-
hyperbolic attractors are chaotic. Trans. Am. Math. Soc. puting, Indiana University, 1988]
361(5), 2431–2485 (2009) 23. Fradkov, A., Evans, R.: Control of chaos: methods and
5. Arnold, V., Afraimovich, V., Ilyashenko, Y., Shilnikov, L.: applications in engineering. Annu. Rev. Control 29(1), 33–
Dynamical Systems V: Bifurcation Theory and Catastrophe 56 (2005)
Theory. Encyclopaedia of Mathematical Sciences, Vol. 5, 24. Fuchs, A.: Nonlinear Dynamics in Complex Systems.
p. 274, Springer, Berlin (1994). https://doi.org/10.1007/ Springer, Berlin (2013)
978-3-642-57884-7 25. Galias, Z., Tucker, W.: Short periodic orbits for the Lorenz
6. Auerbach, D., Cvitanović, P., Eckmann, J.P., Gunaratne, G., system. In: 2008 International Conference on Signals and
Procaccia, I.: Exploring chaotic motion through periodic Electronic Systems, pp. 285–288. IEEE (2008)
orbits. Phys. Rev. Lett. 58(23), 2387 (1987) 26. Galias, Z., Zgliczyński, P.: Computer assisted proof of
7. Barbashin, E., Krasovsky, N.: On the stability of a motion in chaos in the Lorenz equations. Phys. D 115, 165–188
the large. Dokl. Akad. Nauk SSSR 86(3), 453–456 (1952). (1998)
(In Russian) 27. Glukhovsky, A.B., Dolzhansky, F.V.: Three component
8. Barrio, R., Dena, A., Tucker, W.: A database of rigorous models of convection in a rotating fluid. Izv. Acad. Sci.
and high-precision periodic orbits of the Lorenz model. USSR, Atmos. Oceanic Phys. 16, 311–318 (1980)
Comput. Phys. Commun. 194, 76–83 (2015) 28. Grassberger, P., Procaccia, I.: Measuring the strangeness
9. Bautin, N.: The Behaviour of Dynamical Systems close of strange attractors. Phys. D Nonlinear Phenom. 9(1–2),
to the Boundaries of a Stability Domain. Gostekhizdat (in 189–208 (1983)
Russian), Leningrad (1949) 29. Grebogi, C., Hammel, S., Yorke, J., Sauer, T.: Shadowing
10. Benzi, R., Paladin, G., Parisi, G., Vulpiani, A.: On the mul- of physical trajectories in chaotic dynamics: containment
tifractal nature of fully developed turbulence and chaotic and refinement. Phys. Rev. Lett. 65(13), 1527 (1990)
systems. J. Phys. A Math. Gen. 17(18), 3521 (1984) 30. Grebogi, C., Ott, E., Yorke, J.: Fractal basin boundaries,
11. Best, R., Kuznetsov, N., Leonov, G., Yuldashev, M., Yulda- long-lived chaotic transients, and unstable-unstable pair
shev, R.: Tutorial on dynamic analysis of the Costas loop. bifurcation. Phys. Rev. Lett. 50(13), 935–938 (1983)
IFAC Annu. Rev. Control 42, 27–49 (2016). https://doi. 31. Haddad, W., Chellaboina, V.: Nonlinear Dynamical Sys-
org/10.1016/j.arcontrol.2016.08.003 tems and Control: A Lyapunov-Based Approach. Princeton
12. Boccaletti, S., Grebogi, C., Lai, Y.C., Mancini, H., Maza, University Press, Princeton (2011)
D.: The control of chaos: theory and applications. Phys. 32. Hahn, W.: Stability of Motion. Springer, Berlin (1967)
Rep. 329(3), 103–197 (2000) 33. Hammel, S., Yorke, J., Grebogi, C.: Numerical orbits of
13. Budanov, V.: Undefined frequencies method. Fundam. chaotic processes represent true orbits. Bull. Am. Math.
Prikl. Mat. 22, 59–71 (2018). (in Russian) Soc. 19(2), 465–469 (1988)

123
730 N. V. Kuznetsov et al.

34. Hilbert, D.: Mathematical problems. Bull. Am. Math. Soc. 49. Kuznetsov, N., Mokaev, T., Alexeeva, T.: On lower-bound
8, 437–479 (1902) estimates of the Lyapunov dimension and topological
35. Hooton, E., Amann, A.: Analytical limitation for time- entropy via Pyragas control. In: International Conference
delayed feedback control in autonomous systems. Phys. “Stability, Control, Differential Games” devoted to the 95th
Rev. Lett. 109, 101–154 (2012) anniversary of Academician N.N. Krasovskii (Yekaterin-
36. Kehlet, B., Logg, A.: Quantifying the computability of the burg), pp. 425–428 (2019)
Lorenz system using a posteriori analysis. In: Proceedings 50. Kuznetsov, N., Mokaev, T., Kudryashova, E., Kuznetsova,
of the VI International Conference on Adaptive Modeling O., Danca, M.F.: On lower-bound estimates of the Lya-
and Simulation (ADMOS 2013) (2013) punov dimension and topological entropy for the Rossler
37. Kehlet, B., Logg, A.: A posteriori error analysis of round- systems. IFAC-PapersOnLine pp. 97–102 (2019). (15th
off errors in the numerical solution of ordinary differential IFAC Workshop on Time Delay Systems)
equations. Numer. Algorithms 76(1), 191–210 (2017) 51. Kuznetsov, N., Mokaev, T., Mokaev, R., Kuznetsova, O.,
38. Kostelich, E., Kan, I., Grebogi, C., Ott, E., Yorke, J.: Unsta- Kudryashova, E.: A lower-bound estimate of the Lyapunov
ble dimension variability: a source of nonhyperbolicity in dimension for the global attractor of the Lorenz system.
chaotic systems. Phys. D Nonlinear Phenom. 109(1–2), 81– arXiv preprint arXiv:1910.08740 (2019)
90 (1997) 52. Lai, Y., Tel, T.: Transient Chaos: Complex Dynamics on
39. Kuznetsov, N.: Hidden attractors in fundamental problems Finite Time Scales. Springer, New York (2011)
and engineering models. A short survey. Lecture Notes in 53. Lappa, M.: Thermal Convection: Patterns, Evolution and
Electrical Engineering 371, 13–25 (2016). (Plenary lec- Stability. Wiley, Hoboken (2009)
ture at International Conference on Advanced Engineering 54. LaSalle, J.: Some extensions of Liapunov’s second method.
Theory and Applications 2015). https://doi.org/10.1007/ IRE Trans. Circuit Theory 7(4), 520–527 (1960)
978-3-319-27247-4_2 55. Lehnert, J., Hövel, P., Flunkert, V., Guzenko, P., Fradkov,
40. Kuznetsov, N.: The Lyapunov dimension and its estimation A., Schöll, E.: Adaptive tuning of feedback gain in time-
via the Leonov method. Phys. Lett. A 380, 25–26 (2016). delayed feedback control. Chaos Interdiscip. J. Nonlinear
https://doi.org/10.1016/j.physleta.2016.04.036 Sci. 21(4), 43–111 (2011)
41. Kuznetsov, N., Alexeeva, T., Leonov, G.: Invariance of 56. Leonov, G.: On estimations of Hausdorff dimension of
Lyapunov exponents and Lyapunov dimension for regular attractors. Vestnik St. Petersburg University: Mathemat-
and irregular linearizations. Nonlinear Dyn. 85(1), 195– ics 24(3), 38–41 (1991). [Transl. from Russian: Vestnik
201 (2016). https://doi.org/10.1007/s11071-016-2678-4 Leningradskogo Universiteta. Mathematika, 24(3), 1991,
42. Kuznetsov, N., Kuznetsova, O., Leonov, G.: Visualization pp. 41-44]
of four normal size limit cycles in two-dimensional polyno- 57. Leonov, G.: Shilnikov chaos in Lorenz-like systems.
mial quadratic system. Differ. Equ. Dyn. Syst. 21(1–2), 29– Int. J. Bifurc. Chaos (2013). https://doi.org/10.1142/
34 (2013). https://doi.org/10.1007/s12591-012-0118-6 S0218127413500582
43. Kuznetsov, N., Kuznetsova, O., Leonov, G., Mokaev, 58. Leonov, G.: Lyapunov functions in the global analysis of
T., Stankevich, N.: Hidden attractors localization in chaotic systems. Ukr. Mat. J. 70(1), 40–62 (2018). (in Rus-
Chua circuit via the describing function method. IFAC- sian)
PapersOnLine 50(1), 2651–2656 (2017) 59. Leonov, G., Andrievskiy, B., Mokaev, R.: Asymptotic
44. Kuznetsov, N., Leonov, G., Mokaev, T., Prasad, A., behavior of solutions of Lorenz-like systems: Analytical
Shrimali, M.: Finite-time Lyapunov dimension and hid- results and computer error structures. Vestnik St. Peters-
den attractor of the Rabinovich system. Nonlinear burg University. Mathematics 50(1), 15–23 (2017)
Dyn. 92(2), 267–285 (2018). https://doi.org/10.1007/ 60. Leonov, G., Boichenko, V.: Lyapunov’s direct method in
s11071-018-4054-z the estimation of the Hausdorff dimension of attractors.
45. Kuznetsov, N., Leonov, G., Shumafov, M.: A short sur- Acta Applicandae Math. 26(1), 1–60 (1992)
vey on Pyragas time-delay feedback stabilization and odd 61. Leonov, G., Bunin, A., Koksch, N.: Attraktorlokalisierung
number limitation. IFAC-PapersOnLine 48(11), 706–709 des Lorenz-systems. ZAMM J. Appl. Math. Mech.
(2015). https://doi.org/10.1016/j.ifacol.2015.09.271 Zeitschrift fur Angewandte Mathematik und Mechanik
46. Kuznetsov, N., Leonov, G., Yuldashev, M., Yuldashev, R.: 67(12), 649–656 (1987)
Hidden attractors in dynamical models of phase-locked 62. Leonov, G., Kuznetsov, N.: Hidden attractors in dynamical
loop circuits: limitations of simulation in MATLAB and systems from hidden oscillations in Hilbert-Kolmogorov,
SPICE. Commun. Nonlinear Sci. Numer. Simul. 51, 39– Aizerman, and Kalman problems to hidden chaotic
49 (2017). https://doi.org/10.1016/j.cnsns.2017.03.010 attractors in Chua circuits. Int. J. Bifurc. Chaos
47. Kuznetsov, N., Lobachev, M., Yuldashev, M., Yuldashev, Appl. Sci. Eng.23(1) (2013). https://doi.org/10.1142/
R., Kudryashova, E., Kuznetsova, O., Rosenwasser, E., S0218127413300024. art. no. 1330002
Abramovich, S.: The birth of the global stability theory 63. Leonov, G., Kuznetsov, N.: On differences and similarities
and the theory of hidden oscillations. In: 2020 European in the analysis of Lorenz, Chen, and Lu systems. Appl.
Control Conference Proceedings, pp. 769–774 (2020) Math. Comput. 256, 334–343 (2015). https://doi.org/10.
48. Kuznetsov, N., Mokaev, T.: Numerical analysis of dynam- 1016/j.amc.2014.12.132
ical systems: unstable periodic orbits, hidden transient 64. Leonov, G., Kuznetsov, N., Kiseleva, M., Solovyeva, E.,
chaotic sets, hidden attractors, and finite-time Lyapunov Zaretskiy, A.: Hidden oscillations in mathematical model
dimension. J. Phys. Conf. Ser. (2019). https://doi.org/10. of drilling system actuated by induction motor with a
1088/1742-6596/1205/1/012034

123
The Lorenz system: hidden boundary of practical stability and the Lyapunov dimension 731

wound rotor. Nonlinear Dyn. 77(1–2), 277–288 (2014). 81. Palmer, K.: Shadowing in Dynamical Systems: Theory and
https://doi.org/10.1007/s11071-014-1292-6 Applications, vol. 501. Springer, Berlin (2013)
65. Leonov, G., Kuznetsov, N., Korzhemanova, N., Kusakin, 82. Pchelintsev, A., Polunovskiy, A., Yukhanova, I.: The
D.: Lyapunov dimension formula for the global attractor harmonic balance method for finding approximate peri-
of the Lorenz system. Commun. Nonlinear Sci. Numer. odic solutions of the Lorenz system. Tambov University
Simul. 41, 84–103 (2016). https://doi.org/10.1016/j.cnsns. Reports. Series: Natural and Technical Sciences 24, 187–
2016.04.032 203 (2019). (in Russian)
66. Leonov, G., Kuznetsov, N., Mokaev, T.: Hidden attractor 83. Pikovski, A., Rabinovich, M., Trakhtengerts, V.: Onset of
and homoclinic orbit in Lorenz-like system describing con- stochasticity in decay confinement of parametric instabil-
vective fluid motion in rotating cavity. Commun. Nonlinear ity. Sov. Phys. JETP 47, 715–719 (1978)
Sci. Numer. Simul. 28, 166–174 (2015). https://doi.org/10. 84. Pilyugin, S.: Theory of shadowing of pseudotrajectories in
1016/j.cnsns.2015.04.007 dynamical systems. Differencialnie Uravnenia i Protsesy
67. Leonov, G., Kuznetsov, N., Mokaev, T.: Homoclinic orbits, Upravlenia (Differential Equations and Control Processes)
and self-excited and hidden attractors in a Lorenz-like sys- 4, 96–112 (2011)
tem describing convective fluid motion. Eur. Phys. J. Spec. 85. Pyragas, K.: Continuous control of chaos by self-
Top. 224(8), 1421–1458 (2015). https://doi.org/10.1140/ controlling feedback. Phys. Lett. A 170(6), 421–428 (1992)
epjst/e2015-02470-3 86. Pyragas, K.: Control of chaos via an unstable delayed feed-
68. Leonov, G., Kuznetsov, N., Vagaitsev, V.: Localization of back controller. Phys. Rev. Lett. 86, 2265–2268 (2001)
hidden Chua’s attractors. Phys. Lett. A 375(23), 2230–2233 87. Rabinovich, M.I.: Stochastic self-oscillations and turbu-
(2011). https://doi.org/10.1016/j.physleta.2011.04.037 lence. Soviet Physics Uspekhi 21(5), 443–469 (1978)
69. Leonov, G., Kuznetsov, N., Yuldashev, M., Yuldashev, R.: 88. Rössler, O.: An equation for continuous chaos. Phys. Lett.
Hold-in, pull-in, and lock-in ranges of PLL circuits: rig- A 57(5), 397–398 (1976)
orous mathematical definitions and limitations of clas- 89. Ruelle, D.: The Lorenz attractor and the problem of tur-
sical theory. IEEE Trans. Circuits Syst. I Regul. Papers bulence. In: Turbulence and Navier Stokes Equations, pp.
62(10), 2454–2464 (2015). https://doi.org/10.1109/TCSI. 146–158. Springer, Berlin (1976)
2015.2476295 90. Sauer, T.: Shadowing breakdown and large errors in dynam-
70. Leonov, G., Lyashko, S.: Eden’s hypothesis for a Lorenz ical simulations of physical systems. Phys. Rev. E 65(3),
system. Vestnik St. Petersburg University: Mathematics 036220 (2002)
26(3), 15–18 (1993). [Transl. from Russian: Vestnik Sankt- 91. Sauer, T., Grebogi, C., Yorke, J.: How long do numerical
Peterburgskogo Universiteta. Ser 1. Matematika, 26(3), 14- chaotic solutions remain valid? Phys. Rev. Lett. 79(1), 59
16] (1997)
71. Leonov, G., Mokaev, R., Kuznetsov, N., Mokaev, T.: Homo- 92. Sauer, T., Yorke, J.: Rigorous verification of trajectories for
clinic bifurcations and chaos in the Fishing principle for the the computer simulation of dynamical systems. Nonlinear-
Lorenz-like systems. Int. J. Bifurc. Chaos 30(8), (2020). ity 4(3), 961 (1991)
https://doi.org/10.1142/S0218127420501242 93. Shilnikov, L.P., Shilnikov, A.L., Turaev, D.V., Chua, L.:
72. Leonov, G., Reitmann, V., Smirnova, V.: Nonlocal Methods Methods of Qualitative Theory in Nonlinear Dynamics:
for Pendulum-like Feedback Systems. Teubner, Stuttgart- Part 2. World Scientific, Singapore (2001)
Leipzig (1992) 94. Shinbrot, T., Grebogi, C., Ott, E., Yorke, J.: Using small
73. Leonov, G.A.: Lyapunov functions in the attractors dimen- perturbations to control chaos. Nature 363, 411–417 (1993)
sion theory. J. Appl. Math. Mech. 76(2), 129–141 (2012) 95. Smith, R.: Some application of Hausdorff dimension
74. Levinson, N.: Transformation theory of non-linear differ- inequalities for ordinary differential equation. Proc. R. Soc.
ential equations of the second order. Ann. Math. 723–737 Edinburg 104A, 235–259 (1986)
(1944) 96. Sparrow, C.: The Lorenz Equations: Bifurcations, Chaos,
75. Liao, S., Wang, P.: On the mathematically reliable long- and Strange Attractors. Applied Mathematical Sciences.
term simulation of chaotic solutions of Lorenz equation Springer, New York (1982)
in the interval [0,10000]. Sci. China Phys. Mech. Astron. 97. Sprott, J.: Chaos and Time-Series Analysis. Oxford Uni-
57(2), 330–335 (2014) versity Press, Oxford (2003)
76. Lorenz, E.: Deterministic nonperiodic flow. J. Atmos. Sci. 98. Sprott, J.: Maximally complex simple attractors. Chaos
20(2), 130–141 (1963) Interdiscip. J. Nonlinear Sci. 17(3), 033124 (2007)
77. Lorenz, E.: The local structure of a chaotic attractor in four 99. Sprott, J., Jafari, S., Khalaf, A., Kapitaniak, T.: Megastabil-
dimensions. Phys. D Nonlinear Phenom. 13(1–2), 90–104 ity: coexistence of a countable infinity of nested attractors
(1984) in a periodically-forced oscillator with spatially-periodic
78. Malinetskii, G., Potapov, A.: On calculating the dimension damping. Eur. Phys. J. Spec. Top. 226(9), 1979–1985
of strange attractors. USSR Comput. Math. Math. Phys. (2017)
28(4), 39–49 (1988) 100. Sprott, J., Rowlands, G.: Improved correlation dimension
79. Munmuangsaen, B., Srisuchinwong, B.: A hidden chaotic calculation. Int. J. Bifurc. Chaos 11(07), 1865–1880 (2001)
attractor in the classical Lorenz system. Chaos Solitons 101. Stankevich, N., Kuznetsov, N., Leonov, G., Chua, L.: Sce-
Fractals 107, 61–66 (2018) nario of the birth of hidden attractors in the Chua circuit. Int.
80. Nese, J., Dutton, J., Wells, R.: Calculated attractor dimen- J. Bifurc. Chaos Appl. Sci. Eng. 27(12), 1730038 (2017)
sions for low-order spectral models. J. Atmos. Sci. 44(15), 102. Stewart, I.: Mathematics: the Lorenz attractor exists.
1950–1972 (1987) Nature 406(6799), 948–949 (2000)

123
732 N. V. Kuznetsov et al.

103. Strogatz, H.: Nonlinear Dynamics and Chaos. With Appli- 110. Viswanath, D.: The Lindstedt–Poincaré technique as an
cations to Physics, Biology, Chemistry, and Engineering. algorithm for computing periodic orbits. SIAM Rev. 43(3),
Westview Press, Boulder (1994) 478–495 (2001)
104. Tucker, W.: The Lorenz attractor exists. Comptes Ren- 111. Yakubovich, V., Leonov, G., Gelig, A.: Stability of Station-
dus de l’Academie des Sciences - Series I - Mathematics ary Sets in Control Systems with Discontinuous Nonlinear-
328(12), 1197–1202 (1999) ities. World Scientific, Singapure (2004). [Transl from Rus-
105. Viana, R., Grebogi, C.: Unstable dimension variability and sian: A.Kh. Gelig and G.A. Leonov and V.A. Yakubovich,
synchronization of chaotic systems. Phys. Rev. E 62(1), Nauka, 1978]
462 (2000) 112. Yuan, Q., Yang, F.Y., Wang, L.: A note on hidden transient
106. Viana, R., Pinto, S.D.S., Barbosa, J., Grebogi, C.: Pseudo- chaos in the Lorenz system. Int. J. Nonlinear Sci. Numer.
deterministic chaotic systems. Int. J. Bifurc. Chaos 13(11), Simul. 18(5), 427–434 (2017)
3235–3253 (2003) 113. Zhang, X., Chen, G.: Constructing an autonomous system
107. Viana, R., Pinto, S.D.S., Grebogi, C.: Chaotic bursting at with infinitely many chaotic attractors. Chaos Interdiscip.
the onset of unstable dimension variability. Phys. Rev. E J. Nonlinear Sci. 27(7), 071101 (2017)
66(4), 046213 (2002)
108. Vidyasagar, M.: Nonlinear Systems Analysis. Prentice-
Hall, Englewood Cliffs (1978) Publisher’s Note Springer Nature remains neutral with regard
109. Vinograd, R.: The inadequacy of the method of charac- to jurisdictional claims in published maps and institutional affil-
teristic exponents for the study of nonlinear differential iations.
equations. Dokl. Akad. Nauk SSSR (In Russian) 114(2),
239–240 (1957)

123