Asian Journal of Control, Vol. 21, No. 1, pp.

323–338, January 2019

Published online 9 October 2018 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/asjc.1924

CONTINUOUS FIXED-TIME CONVERGENT SUPER-TWISTING

ALGORITHM IN CASE OF UNKNOWN STATE AND
DISTURBANCE INITIAL CONDITIONS
Michael Basin , Pablo Rodriguez-Ramirez and Alison Garza-Alonso

ABSTRACT
This paper presents a fixed-time convergent super-twisting-like algorithm designed to provide a direct exten-
sion, without any additional terms, of the conventional super-twisting control system, whose state initial condition is
unknown and the disturbance initial condition is bounded by a known constant or even completely unknown. The
fixed-time convergent super-twisting-like algorithm is first designed for a scalar system and then generalized to a multi-
variable one. An upper estimate of its convergence (settling) time is calculated in each case. Several examples are provided
to illustrate the obtained theoretical results.
Key Words: Super-twisting algorithm, fixed-time convergence, settling time estimation.

I. INTRODUCTION rithm, called super-twisting, drives a system state and its

first derivative at zero in finite time. The first estimates of
The detailed study of finite-time convergence and convergence times for twisting and super-twisting algo-
stability for equilibrium points of continuous but rithms, based on the Lyapunov function approach, were
non-Lipschitzian autonomous systems, as well as prop- proposed in [17] and [18]. A detailed study of conver-
erties of the finite-time convergence time functions and gence properties of the super-twisting system, as well as
their robustness to perturbations was initiated by [1] estimation of its convergence rate, was provided in [19].
and [2]. The first result regarding finite-time regulators The multivariable super-twisting control was developed
was obtained in [3], which consists in establishing exis- and the corresponding convergence time was estimated
tence of a continuous regulator driving all states of a in [20].
chain of integrators at the origin in finite time. The In most practical cases, it is also desirable to
authors demonstrated that all states of a controllable lin- ensure that the convergence time function is uniformly
ear system can be driven at the origin in finite time by a bounded regardless of the initial condition. The prob-
continuous state feedback. Finite-time convergent algo- lem of designing continuous control laws convergent for
rithms have recently been applied to various classes of a fixed time was considered in [21] and [22]. The paper
dynamic systems, such as finite-time state-bounded sys- [23] presents comprehensive studies of fixed-time stability
tems [4], automotive suspension systems [5], rigid space- and control design for non-autonomous systems, includ-
crafts [6], power converters [7], Takagi–Sugeno [8] and ing discontinuous ones. Various approaches to estimating
Markovian jump systems [9–11], event-triggered con- finite-convergence times for different control algorithms
trol systems [12], underactuated [13] and polynomial [14] can be found in [24,25]. A fixed-time convergent mul-
systems, disturbance rejection control systems [15], and tivariable super-twisting-like control law was proposed
many others. [26], where an upper estimate of its convergence time
A continuous finite-time convergent algorithm for was calculated. However, the last result exposes a sig-
a second-order system was designed in [16]. This algo- nificant drawback: it is required that the disturbance
initial value should be equal to zero, which is restric-
tive and impractical. The research presented in this paper
Manuscript received February 28, 2018; revised June 10, 2018; accepted June
26, 2018. aims at overcoming this disadvantage and admits a dis-
The authors are with Department of Physical and Mathematical Sciences, turbance, whose initial value is unknown but bounded
Autonomous University of Nuevo Leon, San Nicolas de los Garza, Nuevo Leon,
Mexico.
by a known constant. The last assumption is conven-
Michael Basin (corresponding author, e-mail: [email protected]) is also tionally acceptable for real engineering systems, given
with ITMO University, St. Petersburg, Russia. that a superior bound of disturbances, which the sys-
The authors thank the Mexican National Science and Technology Council
(CONACyT) for financial support under Grant 250611. tem may hold without being broken, is usually known

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
324 Asian Journal of Control, Vol. 21, No. 1, pp. 323–338, January 2019

from technical specifications or common sense. Further II. PROBLEM STATEMENT

on, the research is focused on the case where a distur-
bance initial value is completely unknown, that is, no Consider a vector control system
information on its upper bound is available. The designed
algorithm is based on using a fixed-time convergent ̇
x(t) = u(t) + 𝜁(t), x(t0 ) = x0 , (1)
smooth second-order observer. Some recent results con-
where x(t) ∈ Rn is a system state, u(t) ∈ Rn is a control
cerning the observer and observer-based design problems
input, 𝜁(t) ∈ Rn is a disturbance satisfying the Lipschitz
for systems with unknown disturbances are presented
condition
in [27–32].
The contributions of this paper are as follows: ‖𝜁(t1 ) − 𝜁(t2 )‖ ≤ L|t1 − t2 |

• A fixed-time convergent algorithm is designed to for any t1 , t2 ≤ t0 , with a certain constant L. Two cases
provide a direct extension, without any additional of disturbance initial conditions are considered: a. The
terms, of the conventional super-twisting control initial value of the disturbance, 𝜁(t0 ), is bounded by a cer-
system, whose state initial condition is unknown and tain known constant K, |𝜁(t0 )| ≤ K. b. No known bound
the disturbance initial condition is bounded by a for 𝜁(t0 ) is available.
known constant. The control problem is to design a continuous con-
• A fixed-time convergent algorithm is also designed trol law driving the states of the resulting closed-loop
for a super-twisting-like control system, whose both system at the origin for a fixed time in the sense of the
state and disturbance initial conditions are com- following definition and estimate the corresponding con-
pletely unknown, that is, no information on their vergence (settling) time. This control law would general-
upper bounds is available. ize the super-twisting-like control proposed in [26] to the
• The fixed-time convergent algorithms are first cases when the disturbance initial condition is unknown
designed for scalar super-twisting-like systems and but bounded by a known constant or even completely
then generalized to multivariable ones. unknown. In [26], the disturbance initial value is assumed
• An upper estimate of the convergence (settling) time to be zero, 𝜁(t0 ) = 0.
of the obtained super-twisting-like algorithm is cal-
culated in each case. Definition 1 [26]. The vector control system (1) is called
fixed-time convergent to the origin, if there exists a time
The paper is organized as follows. The problem moment T such that the system state x(t) ∈ Rn is equal
statement is given in Section II. Section III presents to zero, x(t) = 0, for all t ≥ T, starting from any initial
a fixed-time convergent super-twisting-like algorithm in condition x0 ∈ Rn .
case of a disturbance initial condition is bounded by
a known constant and calculates the uniform upper III. CONTROL DESIGN: KNOWN BOUND
estimate for the settling time. The corresponding exam- FOR INITIAL VALUE OF DISTURBANCE
ples are given in Section IV. Section V provides a
generalization of the designed fixed-time convergent Consider first a scalar system (1), n = 1. Introduce
super-twisting-like algorithm to a multivariable case. a continuous control law
The corresponding multivariable examples are given
in Section VI. Section VII presents fixed-time con- u(t) = −𝜆1 ∣ x(t) ∣1∕2 sign(x(t))
vergent super-twisting-like algorithms for completely t

unknown disturbance initial conditions and calculates − 𝜆2 ∣ x(t) ∣p sign(x(t)) − 𝛼 sign(x(s))ds,

∫t0
the uniform upper estimates for the settling times in
(2)
both, scalar and multivariable, cases. The correspond-
ing examples are given in Section VIII. Section IX which provides fixed-time convergence to the origin for
concludes this study. Proofs of all theorems are given the scalar system (1). Here, 𝜆1 , 𝜆2 , 𝛼 > 0 and p > 1. The
in Appendix. resulting closed-loop super-twisting-like control system
The standard notation is used: |x(t)| denotes the is represented in the form
absolute value of a scalar x(t), ‖x(t)‖ denotes the
Euclidean norm of a vector x(t), ap corresponds to a ̇
x(t) = −𝜆1 ∣ x(t) ∣1∕2 sign(x(t))
value a raised to a power p. The function sign(x) is − 𝜆2 ∣ x(t) ∣p sign(x(t)) + y(t), x(t0 ) = x0 , (3)
introduced according to Filippov’s definition [33]. ̇ = −𝛼sign(x(t)) + 𝜉(t), y(t0 ) = 𝜁(t0 ).
y(t)

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
 M. Basin et al.: Continuous Fixed-Time Convergent Super-Twisting Algorithm 325

where 𝜉(t) = d𝜁(t)∕dt is bounded by the constant L. If The obtained convergence finite time estimate depends
𝜆2 = 0, the conventional super-twisting algorithm [16] on the initial conditions x0 and y0 , which are assumed
is recovered. The main result for fixed-time convergent known in [16].
control law (2) is given as follows.

Theorem 1. Consider a dynamic system (3) in the pres- IV. EXAMPLES: SCALAR CASE
ence of a disturbance 𝜁(t) satisfying Lipschitz condition
with constant L and an initial condition y(t0 ) = 𝜁(t0 ) Consider a super-twisting-like system governed by
bounded by a constant K. Then, both states x(t) and y(t) equations (3), which is similar to that in [26]. Formula (4)
converge to the origin uniformly in fixed time is applied to calculate the convergence fixed time upper
( ) bound estimate using the following parameter values:
1 2𝜖 1∕2 K
Tf ≤ + + p = 3∕2, 𝜆1 = 20, 𝜆2 = 1, L = 2.5, 𝛼 = 7; therefore,
𝜆2 (p − 1)𝜖 p−1 𝜆1 M
M = 9.5 and m = 4.5. The control gains are selected to
⎛ ⎞ (4) satisfy the conditions of Theorem 1. The disturbance is
⎜ 1 ⎟ K
× ⎜1 + ( )⎟ + , assigned as 𝜉(t) = −sin(t) − cos(10t). The bound K is set
⎜ m M1 −
h(𝜆1 )
⎟ m to K = 100. The treated initial conditions and the corre-
⎝ 𝜆1 ⎠ sponding settling times obtained by simulation are given
where 𝜖 > 0, M = 𝛼 + L, m = 𝛼 − L, h(𝜆1 ) = in Table I.
1∕𝜆1 + (2e∕m𝜆1 )1∕3 , and e is the base of natural loga- The convergence fixed time estimate given by (4) is
rithms, provided that the following conditions hold for Tf = 65.13, whereas the real convergence times for the
control gains: 𝛼 > L, 𝜆1 h−1 (𝜆1 ) > M. The minimum system (3) vary from T = 0.88 for small initial condi-
1 tions to T = 15.4 for larger ones and always approach
value of Tf (𝜖) is reached for 𝜖 = (𝜆1 ∕𝜆2 ) p+1∕2 . a certain limit as x0 tends to infinity. Therefore, the
Proof. Proofs of all theorems are given in the Appendix. estimate given by (4) well approximates the real conver-
gence times and can be useful for practical settling time
Note that the value of K affects the theoretical evaluation.
upper bound for the convergence time of the algorithm The following figures show phase portraits of the
given by (4): the less K, the less the convergence time is. trajectories of the super-twisting-like system (3) in coor-
dinates x, y. Figs 1–3 show the phase portraits for ini-
Remark 1. The right-hand side of (4) presents a con- tial conditions x0 = 100, y0 = −50 and Figs 4–6
vergence fixed time upper bound estimate for the correspond to initial conditions x0 = 100, y0 =
super-twisting-like system (3), which is independent of an 50. The given plots visually confirm finite-time (not
unknown initial conditions x0 and y0 . This means that for only asymptotic) convergence of both state variables to
any initial conditions x0 and y0 both states, x(t) and y(t), the origin.
of the super-twisting-like system (3) converge to the ori- Finally, note that the behavior of the state trajecto-
gin for a finite time Tf no greater than the upper estimate. ries corresponding to initial conditions of the opposite
The result of Theorem 1 generalizes the result obtained in and same signs is well consistent with the cases A and C
[26], where the initial condition y0 is assumed to be zero, in the proof of Theorem 1 (please see Appendix).
y0 = 0.
( )
( 1∕2 )
2𝜖 K
Remark 2. The term 𝜆 + M 1 + ( 1 h(𝜆1 ) ) + Km
1 Table I. Settling times vs. initial conditions.
1 m M
− 𝜆1

in (4) can be used to estimate the convergence time in the y0 ∖x0 10 100 1000 105 106 107
conventional super-twisting algorithm [16], setting 𝜖 =∣ −100 14.8 15.3 15.4 15.4 15.4 15.4
x0 ∣ and K =∣ y0 ∣. The resulting formula takes the form −50 8.5 8.5 8.5 8.5 8.5 8.5
−10 2.2 2.71 2.98 2.98 2.98 2.98
( )⎛ ⎞ −5 2.17 2.17 2.17 2.17 2.17 2.17
2 ∣ x0 ∣1∕2 ∣ y0 ∣ ⎜ 1 ⎟ −1 0.88 2.15 2.15 2.15 2.15 2.15
Tnf (x0 ) ≤ + ⎜1 + ( 1 )⎟
𝜆1 M ⎜ m M−
h(𝜆1 )
⎟ 1 0.4 0.95 2.15 2.15 2.15 2.15
⎝ 𝜆1 ⎠ 5 0.88 0.88 0.88 0.88 0.88 0.88
∣ y0 ∣ 10 2.15 2.15 2.15 2.15 2.15 2.15
+ . 50 7.2 7.2 7.2 7.2 7.2 7.2
m
100 14.8 14.8 14.8 14.8 14.8 14.8
(5)

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
326 Asian Journal of Control, Vol. 21, No. 1, pp. 323–338, January 2019

Fig. 4. Phase portrait for initial conditions x0 = 100, y0 = 50.

[Color figure can be viewed at wileyonlinelibrary.com]
Fig. 1. Phase portrait for initial conditions x0 = 100,
y0 = −50. [Color figure can be viewed at
wileyonlinelibrary.com]

Fig. 5. Phase portrait for initial conditions x0 = 100, y0 = 50

Fig. 2. Phase portrait for initial conditions x0 = 100, (zoomed). [Color figure can be viewed at
y0 = −50 (zoomed). [Color figure can be viewed at wileyonlinelibrary.com]
wileyonlinelibrary.com]

Fig. 3. Phase portrait for initial conditions x0 = 100, Fig. 6. Phase portrait for initial conditions x0 = 100, y0 = 50
y0 = −50 (detailed). [Color figure can be viewed at (detailed). [Color figure can be viewed at
wileyonlinelibrary.com] wileyonlinelibrary.com]

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
 M. Basin et al.: Continuous Fixed-Time Convergent Super-Twisting Algorithm 327

V. CONTROL DESIGN: KNOWN BOUND The resulting formula takes the form
FOR INITIAL VALUE OF DISTURBANCE ( √ )
MULTIVARIABLE CASE 2( n‖x0 ‖)1∕2 ‖y0 ‖
Tmnf (x0 ) ≤ +
𝜆1 M
The proposed scalar continuous fixed-time conver- ( ) √
gent control law (2) is generalized to a vector case as M n‖y0 ‖
× 1+ √ + ,
m(1 − 2𝛼∕𝜆1 ) m
t
x(t) x(s) (9)
u(t) = −𝜆1 −𝜆 x(t)‖x(t)‖p−1
− 𝛼 ds,
‖x(t)‖1∕2 2
∫t0 ‖x(s)‖
The formulas (8) and (9) are further illustrated in the
(6)
examples.
where p > 1. Then, the resulting closed-loop system takes
the form VI. EXAMPLES: MULTIVARIABLE CASE

x(t) Consider a 3D (n = 3) super-twisting-like system

̇
x(t) = −𝜆1 − 𝜆2 x(t)‖x(t)‖p−1 + y(t), x(t0 ) = x0 ,
‖x(t)‖1∕2 (7), which is also similar to that in [26]. Formula (8)
x(t) is applied to calculate the convergence fixed time upper
̇ = −𝛼
y(t) + 𝜉(t), y(t0 ) = 𝜁(t0 ). bound estimate using the following parameter values:
‖x(t)‖
(7) p = 3∕2, 𝜆1 = 5, 𝜆2 = 5, L = 1, 𝛼 = 4.5; therefore,
M = 5.5 and m = 3.5. The control gains are selected to
The fixed-time convergence problem is considered for the satisfy the conditions of Theorem 2. The 3D disturbance
vector system (7) and consists in determining conditions is assigned as 𝜉(t) = [0.1cos(0.5t), −0.3sin(t), 0.1cos(2t)].
for control gains 𝜆1 , 𝜆2 , 𝛼 > 0 such that both states of The bound K is set to K = 100. The treated initial condi-
the system (7) converge to the origin for a finite time tions for each component of x(t) = [x1 (t), x2 (t), x3 (t)] and
smaller than a certain fixed value Tmf > 0, which is y(t) = [y1 (t), y2 (t), y3 (t)] and the corresponding settling
pre-established independently of initial conditions. The times obtained by simulation are given in Table II. The
main result for the multivariable fixed-time convergent initial values for all components of the same vector are
control law (6) and the multivariable system (7) is given considered equal: x10 = x20 = x30 and y10 = y20 = y30 .
as follows. The convergence fixed time estimate given by (7) is
Tf = 143.72, whereas the real convergence times for the
Theorem 2. Consider a dynamic system (7) in the pres- system (7) vary from T = 0.74 for small initial conditions
ence of a disturbance 𝜉(t), whose norm is bounded by a to T = 39 for larger ones and always approach a cer-
constant L, and an initial condition y(t0 ) = 𝜁(t0 ), whose tain limit as x0 tends to infinity. Therefore, similarly to
norm is bounded by a constant K. Then, both vector the scalar case of Section IV, the estimate given by (7) is
states x(t) and y(t) converge to the origin uniformly in non-conservative and can be useful for practical settling
fixed time time evaluation.
( √ ) The following figures show phase portraits of
1 2( n𝜖)1∕2 K the trajectories of the components x1 (t), y1 (t) of the
Tmf ≤ + + super-twisting-like system (7). Figs 7–9 show the phase
𝜆2 (p − 1)𝜖 p−1 𝜆1 M
( ) √ (8) Table II. Settling times vs. initial conditions.
M nK
× 1+ √ + , y0 ∖ x0 10 100 1000 105 106 107
m(1 − 2𝛼∕𝜆 ) m
1
−100 38.62 38.735 38.776 38.82 38.82 38.82
where 𝜖 > 0, M = 𝛼 + L, and m = 𝛼 − L, provided that −50 19.45 19.58 19.62 19.64 19.64 19.64
the following conditions hold for control gains: 𝛼 > 4L, −10 4.28 4.4 4.44 4.46 4.46 4.46
√ −5 2.48 2.6 2.64 2.65 2.65 3.65
𝜆1 > 2𝛼. The minimum value of Tf (𝜖) is reached for −1 1.08 1.2 1.23 1.22 1.22 1.22
1
𝜖 = (n1∕4 𝜆1 ∕𝜆2 ) p+1∕2 . 1 0.62 0.7 0.725 0.74 0.74 0.74
5 1.91 1.91 1.91 1.91 1.91 1.91
Remark 3. The last terms in (8) can be used to estimate 10 3.82 3.82 3.82 3.82 3.82 3.82
50 19.25 19.25 19.25 19.25 19.25 19.25
the convergence time for the multivariable super-twisting
100 38.51 38.51 38.51 38.51 38.51 38.51
algorithm (6),(7), setting 𝜖 = ‖x0 ‖ and K = ‖y0 ‖.

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
328 Asian Journal of Control, Vol. 21, No. 1, pp. 323–338, January 2019

Fig. 7. Phase portrait of x1 (t), y1 (t) for initial conditions Fig. 10. Phase portrait of x1 (t), y1 (t) for initial conditions
x10 = 100, y10 = −50. [Color figure can be viewed at x10 = 100, y10 = 50. [Color figure can be viewed at
wileyonlinelibrary.com] wileyonlinelibrary.com]

Fig. 8. Phase portrait of x1 (t), y1 (t) for initial conditions Fig. 11. Phase portrait of x1 (t), y1 (t) for initial conditions
x10 = 100, y10 = −50 (zoomed). [Color figure can be x10 = 100, y10 = 50 (zoomed). [Color figure can be
viewed at wileyonlinelibrary.com] viewed at wileyonlinelibrary.com]

Fig. 9. Phase portrait of x1 (t), y1 (t) for initial conditions Fig. 12. Phase portrait of x1 (t), y1 (t) for initial conditions
x10 = 100, y10 = −50 (detailed). [Color figure can be x10 = 100, y10 = 50 (detailed). [Color figure can be
viewed at wileyonlinelibrary.com] viewed at wileyonlinelibrary.com]

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
 M. Basin et al.: Continuous Fixed-Time Convergent Super-Twisting Algorithm 329

portraits for initial conditions x10 = 100, y10 = −50, and where 𝜖 > 0, M = 𝛼 + L, m = 𝛼 − L, h(𝜆1 ) =
Figs 10–12 correspond to initial conditions x10 = 100, 1∕𝜆1 + (2e∕m𝜆1 )1∕3 , and e is the base of natural loga-
y10 = 50. The given plots visually confirm finite-time (not rithms, provided that the following conditions hold for
only asymptotic) convergence of all state variables to the control gains: 𝛼 > L, 𝜆1 h−1 (𝜆1 ) > M. The minimum
1
origin. value of Tf (𝜖) is reached for 𝜖 = (𝜆1 ∕𝜆2 ) p+1∕2 . The time
parameter TBBF is defined by the observer (10) as
VII. CONTROL DESIGN: UNKNOWN BOUND
FOR INITIAL VALUE OF DISTURBANCE 𝜆𝜌max (P) 1
TBBF = + , (12)
r𝜌 r1 𝜎Υ𝜎
In this section, it is assumed that no known bound 𝜆 (Q) 𝜆 (Q )
K is available for the initial value of disturbance. In this where 𝜌 = 1 − 𝛼, 𝜎 = 𝛽 − 1, r = 𝜆 min (P) , r1 = 𝜆 min (P1 ) ,
max max 1
case, the fixed-time convergence can still be reached using Υ ≤ 𝜆min (P1 ) is a positive number, the symmetric positive
a fixed-time observer for the variable y(t). Consider the definite matrix P satisfies a Lyapunov equation
control law (2) combined with the fixed-time observer
PA + AT P = −Q,
ŷ (t) = z2 (t) + 𝜆1 ∣ z1 (t) ∣1∕2 sign(z1 (t))
+ 𝜆2 ∣ z1 (t) ∣p sign(z1 (t)), 𝜆min (Q) > 0 is the minimum eigenvalue of the matrix Q,
and 𝜆max (P) > 0 is maximum eigenvalue of the matrix
ż 1 (t) = z2 (t) − k1 ∣ z1 (t) − x(t) ∣𝛼1 sgn(z1 (t) − x(t))
P, the symmetric positive definite matrix P1 satisfies a
− 𝜅1 ∣ z1 (t) − x(t) ∣𝛽1 sgn(z1 (t) − x(t)), Lyapunov equation
ż 2 (t) = −k2 ∣ z1 (t) − x(t) ∣𝛼2 sgn(z1 (t) − x(t)) P1 A1 + AT1 P1 = −Q1 ,
𝛽2
− 𝜅2 ∣ z1 (t) − x(t) ∣ sgn(z1 (t) − x(t)),
and Q, Q1 ∈ Rn×n are symmetric positive definite matri-
k1 , k2 , 𝜅1 , 𝜅2 > 0.
ces.
(10) The obtained result can be generalized to a mul-
tivariable case as follows. Consider the control law (6)
Here, the exponents 𝛼i , i = 1, 2, are selected as 𝛼i ∈ (0, 1), combined with the fixed-time observer
i = 1, 2 satisfying the recurrent relations 𝛼i = i𝛼 − (i − 1),
z1 (t)
i = 2, … , n, where 𝛼1 belongs to an interval (1 − 𝜖, 1) for ŷ (t) = z2 (t) + 𝜆1 + 𝜆2 z1 (t)‖z1 (t)‖p−1 ,
a sufficiently small 𝜖 > 0. The exponents 𝛽i , i = 1, … , 2, ‖z1 (t)‖1∕2
are selected as 𝛽i > 1, i = 1, … , 2 satisfying the recurrent z1 (t) − x(t)
ż 1 (t) = z2 (t) − k1
relations 𝛽i = i𝛽 − (i − 1), i = 2, … , 2, where 𝛽1 belongs ‖z1 (t) − x(t)‖1−𝛼1
to an interval (1, 1 + 𝜖1 ) for a sufficiently small 𝜖 > 0. − 𝜅1 (z1 (t) − x(t))‖z1 (t) − x(t)‖𝛽1 −1 ,
Observer gains ki > 0, 𝜅i > 0, i = 1, … , 2, are positive
z1 (t) − x(t)
numbers. ż 2 (t) = −k2
The main result for fixed-time convergent control ‖z1 (t) − x(t)‖1−𝛼2
law (2) combined with the fixed-time observer (10) is − 𝜅2 (z1 (t) − x(t))‖z1 (t) − x(t)‖𝛽2 −1 ,
given as follows. k1 , k2 , 𝜅1 , 𝜅2 > 0.
Theorem 3. Consider a dynamic system (3) combined (13)
with a fixed-time observer (10) in the presence of a distur-
Then, the following theorem holds.
bance 𝜁(t) satisfying Lipschitz condition with constant L
and an unknown initial condition y(t0 ) = 𝜁(t0 ). Then, Theorem 4. Consider a dynamic system (7) combined
both states x(t) and y(t) converge to the origin uniformly with a fixed-time observer (13) in the presence of a dis-
in fixed time turbance 𝜉(t) bounded by a constant L. Then, both states
( ) x(t) and y(t) converge to the origin uniformly in fixed time
1 2𝜖 1∕2 ∣ ŷ (TBBF ) ∣ ( √ )
Tf ≤ + + 2( n𝜖)1∕2 ‖̂y(TBBF )‖
𝜆2 (p − 1)𝜖 p−1 𝜆1 M 1
Tmf ≤ + +
⎛ ⎞ 𝜆2 (p − 1)𝜖 p−1 𝜆1 M
⎜ 1 ⎟ ∣ ŷ (TBBF ) ∣ ( ) √
× ⎜1 + ( )⎟ + + TBBF , M n‖̂y(TBBF )‖ (14)
⎜ 1
m M−
h(𝜆1 )
⎟ m × 1+ √ +
⎝ 𝜆1 ⎠ m(1 − 2𝛼∕𝜆 ) m
1
(11) + TBBF ,

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
330 Asian Journal of Control, Vol. 21, No. 1, pp. 323–338, January 2019

where 𝜖 > 0, M = 𝛼 + L, and m = 𝛼 − L, provided that moment t0 : they are based on the value of the esti-
the following
√ conditions hold for control gains: 𝛼 > 4L, mate ŷ (TBBF ) at time TBBF , which cannot be calculated
𝜆1 > 2𝛼. The minimum value of Tf (𝜖) is reached for a priori. This presents a certain disadvantage of the pro-
1 posed method, which is however consistent with the fact
𝜖 = (n1∕4 𝜆1 ∕𝜆2 ) p+1∕2 . The time parameter TBBF is defined that the value of the disturbance 𝜁(t) is unknown at
by (12). the initial time moment and cannot be estimated until
Remark 4. Note that the convergence time upper bounds time TBBF , when the estimate ŷ (t) certainly converges
(11) and (14) cannot be determined at an initial time to y(t).

Table III. Settling times vs. initial conditions.

y0 ∖ x0 10 100 1000 105 107

−100 (Sim) 49.6 50 50 50 50
−100 (Tf) 489.91 496.38 501.53 487.65 492.28
−50 (Sim) 28 28 28 28 28
−50 (Tf) 28.54 28.54 28.56 28.56 28.56
−10 (Sim) 7.8 7.8 10.5 15.4 17.8
−10 (Tf) 28.56 28.56 28.56 28.56 28.56
−5 (Sim) 4.8 7.2 10.3 15.5 17.75
−5 (Tf) 28.67 28.67 28.67 28.68 28.68
−1 (Sim) 6.5 7.2 10.4 16.5 18.2
−1 (Tf) 28.57 28.57 28.57 28.57 28.57
1 (Sim) 5 6.3 10.3 15.6 18
1 (Tf) 28.67 27.67 27.67 27.67 27.67
5 (Sim) 5.5 9.2 10.3 15.35 18.6
5 (Tf) 28.65 28.65 28.65 28.65 28.65
10 (Sim) 5.6 6.8 10.3 16.2 18.55
10 (Tf) 28.54 28.54 28.66 28.66 28.66
50 (Sim) 25.5 25.8 24.5 26 27.5
50 (Tf) 28.7 28.7 28.7 28.7 28.9
100 (Sim) 49 49 49 49 49
100 (Tf) 459.36 459.36 459.36 477.875 482.67

Fig. 13. Time histories of x(t) and its estimate z1 (t) for initial conditions x0 = 100, y0 = −50. [Color figure can be viewed at
wileyonlinelibrary.com]

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
 M. Basin et al.: Continuous Fixed-Time Convergent Super-Twisting Algorithm 331

VIII. EXAMPLES: UNKNOWN BOUND FOR 𝛽1 = 10∕9, k1 = 𝜅1 = 100, k2 = 𝜅2 = 80, L = 2,

INITIAL VALUE OF DISTURBANCE 𝛼 = 2.1; therefore, M = 4.1 and m = 0.1. The control
gains are selected to satisfy the conditions of Theorem 3.
Consider a super-twisting-like system governed by The disturbance is assigned as 𝜉(t) = −sin(t) − cos(10t).
equations (3) with an unknown initial condition y(t0 ) = The treated initial conditions for the state variables
𝜁(t0 ). Formula (11) is applied to calculate the convergence and the corresponding settling times are given in Table
fixed time upper bound estimate using the following III. The zero initial conditions are assumed for the
parameter values: p = 3∕2, 𝜆1 = 20, 𝜆2 = 1, 𝛼1 = 9∕10, observer (10).

Fig. 14. Time histories of y(t) and its estimate ŷ (t) for initial conditions x0 = 100, y0 = −50. [Color figure can be viewed at
wileyonlinelibrary.com]

Fig. 15. Time histories of x(t) and its estimate z1 (t) for initial conditions x0 = 100, y0 = 50. [Color figure can be viewed at
wileyonlinelibrary.com]

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
332 Asian Journal of Control, Vol. 21, No. 1, pp. 323–338, January 2019

It can be observed that the ratios between the con- their estimates produced by the fixed-time observer (10).
vergence time estimates given by (11) and the real con- Figs 13–14 show the time histories for initial conditions
vergence times for the system (3) vary from 1.5 for small x0 = 100, y0 = −50 and Figs 15–16 correspond to
initial conditions to 10 for larger ones. Therefore, the esti- initial conditions x0 = 100, y0 = 50. The given plots
mates given by (11) approximates the real convergence visually confirm finite-time (not only asymptotic) con-
times even better that the estimate given by (4) and can vergence of the state variables and their estimates to
as well be useful for practical settling time evaluation. the origin.
The following figures show time histories of the Finally, note again that the behavior of the state
trajectories of the super-twisting-like system (3) and trajectories corresponding to initial conditions of the

Fig. 16. Time histories of y(t) and its estimate ŷ (t) for initial conditions x0 = 100, y0 = 50. [Color figure can be viewed at
wileyonlinelibrary.com]

Table IV. Settling times vs. initial conditions.

y0 ∖x0 10 100 1000 105 107

−100 (Sim) 68 68 68 68 68
−100 (Tf) 370.09 370.57 369.21 370.22 370.10
−50 (Sim) 38 38 38 38 38
−50 (Tf) 79.56 80.97 81 77.86 77.9
−10 (Sim) 14 14 14 17 20
−10 (Tf) 40.99 40.99 41 41 41
−5 (Sim) 14 11.5 12.5 17 23
−5 (Tf) 41 41 41 41 41
−1(Sim) 5.5 8.5 14.6 17.5 20.5
−1 (Tf) 41 41 41 41 41
1 (Sim) 7.4 17 13.4 16 20
1 (Tf) 41 41 41 41 41
5 (Sim) 14 8.4 14 16 21.5
5 (Tf) 41 41 41 41 41
10 (Sim) 11.8 11.8 14.8 17 19
10 (Tf) 41 41 41 41 41
50 (Sim) 37 37 37 37 40.5
50 (Tf) 76.35 76.35 76.35 81.12 84.89
100 (Sim) 68 68 68 68 71.5
100 (Tf) 367.72 368.36 368.33 371.25 395.15

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
 M. Basin et al.: Continuous Fixed-Time Convergent Super-Twisting Algorithm 333

Fig. 17. Time histories of x1 (t) and and its estimate z11 (t) for initial conditions x10 = 100, y10 = 50. [Color figure can be viewed at
wileyonlinelibrary.com]

Fig. 18. Time histories of y1 (t) and its estimate ŷ 1 (t) for initial conditions x10 = 100, y10 = 50. [Color figure can be viewed at
wileyonlinelibrary.com]

opposite and same signs is well consistent with the and m = 2.11. The control gains are selected to sat-
cases A and C in the proof of Theorem 1 (please see isfy the conditions of Theorem 4. The 3D disturbance
Appendix). is assigned as 𝜉(t) = [0.1cos(0.5t), −0.3sin(t), 0.1cos(2t)].
In the multivariable case, consider a super-twisting- The treated initial conditions for each component of
like system governed by equations (7) with an unknown x(t) = [x1 (t), x2 (t), x3 (t)] and y(t) = [y1 (t), y2 (t), y3 (t)]
initial condition y(t0 ) = 𝜁(t0 ). Formula (14) is applied and the corresponding settling times obtained by sim-
to calculate the convergence fixed time upper bound esti- ulation are given in Table IV. The initial values for all
mate using the following parameter values: p = 3∕2, components of the same vector are considered equal:
𝜆1 = 3, 𝜆2 = 3, 𝛼1 = 9∕10, 𝛽1 = 10∕9, k1 = 𝜅1 = 100, x10 = x20 = x30 and y10 = y20 = y30 . The zero initial
k2 = 𝜅2 = 80, L = 0.7, 𝛼 = 2.81; therefore, M = 3.51 conditions are assumed for the observer (14).

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
334 Asian Journal of Control, Vol. 21, No. 1, pp. 323–338, January 2019

It can be observed that the ratios between the 5. Rath, J. J., M. Defoort, H. R. Karimi, and K. C.
convergence time estimates given by (14) and the real Veluvolu, “Output feedback active suspension con-
convergence times for the system (7) vary from 1 to trol with higher order terminal sliding mode,” IEEE
1.8. Therefore, the estimates given by (14) provide even Trans. Ind. Electron., Vol. 64, No. 2, pp. 1392–1403
better approximation of the real convergence times than (2017).
it was registered in the previous simulations by means of 6. Lu, K. and Y. Xia, “Finite-time attitude stabiliza-
assigning appropriate observer and controller gains. tion for rigid spacecraft,” Int. J. Robust Nonlinear
The following figures show time histories of the Control, Vol. 25, No. 1, pp. 32–51 (2015).
trajectories of the of the components x1 (t), y1 (t) of the 7. Liu, J., S. Vasquez, L. Wu, A. Marquez, H. Gao,
super-twisting-like system (7) and their estimates pro- and L. Franquelo, “Extended state observer-based
duced by the fixed-time observer (14). Figs 17 and 18 sliding-mode control for three-phase power convert-
show the time histories for initial conditions x0 = 100, ers,” IEEE Trans. Ind. Electron., Vol. 64, No. 1,
y0 = 50. The given plots visually confirm finite-time (not pp. 22–31 (2017).
only asymptotic) convergence of the state variables and 8. Li, H., J. Wang, H. Du, and H. R. Karimi, “Adaptive
their estimates to the origin. sliding mode control for Takagi-Sugeno fuzzy sys-
tems and its applications,” IEEE Trans. Fuzzy Syst.,
Vol. 26, No. 2, pp. 531–543 (2018).
IX. CONCLUSIONS 9. Wu, L., X. Su, and P. Shi, “Sliding mode control with
bounded L2 gain performance of Markovian jump
This paper has presented a fixed-time convergent singular time-delay systems,” Automatica, Vol. 48,
control algorithm for a multivariable super-twisting sys- No. 8, pp. 1929–1933 (2012).
tem, whose state initial condition is unknown and distur- 10. Li, H., H. Gao, P. Shi, and X. Zhao, “Fault-tolerant
bance initial condition is bounded by a known constant control of Markovian jump stochastic systems via the
or even unknown. An explicit formula has been derived augmented sliding mode observer approach,” Auto-
for calculating an upper estimate for the fixed conver- matica, Vol. 50, No. 7, pp. 1825–1834 (2014).
gence time to the origin of the closed-loop system states 11. Li, H., P. Shi, D. Yao, and L. Wu, “Observer-based
in both situations. The obtained simulation results show adaptive sliding mode control for nonlinear Marko-
close approximation of the calculated upper estimates vian jump systems,” Automatica, Vol. 64, No. C,
to real convergence times and their non-conservativeness pp. 133–142 (2016).
in both, scalar and multivariable, cases. The displayed 12. Wu, L., Y. Gao, J. Liu, and H. Li, “Event-triggered
time histories also confirm the behavior of the state vari- sliding mode control of stochastic systems via output
ables of the super-twisting system, which is theoretically feedback,” Automatica, Vol. 82, pp. 79–92 (2017).
predicted in the theorem proofs. The ongoing research 13. Wang, H., Z. Zhou, C. Hao, Z. Hu, and W. Zheng,
focuses on designing adaptive control algorithms for “FTESO-based finite time control for underactu-
super-twisting systems with unknown state and distur- ated system within a bounded input,” Asian J. Con-
bance initial conditions. trol, Vol. 20, pp. 1427–1439 (2018). https://doi.org/10.
1002/asjc.1624.
REFERENCES 14. Mardani, M. M., N. Vafamand, M.
Shokrian Zeini, M. Shasadeghi, and A. Khayatian,
1. Bhat, S. P. and D. S. Bernstein, “Finite-time stabil- “Sum-of-squares-based finite-time adaptive slid-
ity of continuous autonomous systems,” SIAM J. ing mode control of uncertain polynomial systems
Control Optim., Vol. 38, No. 3, pp. 751–766 (2000). with input nonlinearities,” Asian J. Control, Vol. 20,
2. Bhat, S. P. and D. S. Bernstein, “Finite-time stability pp. 1658–1662 (2018). https://doi.org/10.1002/asjc.
of homogeneous systems,” Proc. 1997 Amer. Con- 1625.
trol Conf., Albuquerque, NM, USA, pp. 2513–2514 15. Wang, Q., M. Ran, and C. Dong, “On finite-time
(1997). stabilization of active disturbance rejection control
3. Bhat, S. P. and D. S. Bernstein, “Geometric homo- for uncertain nonlinear systems,” Asian J. Control,
geneity with applications to finite-time stability,” Vol. 20, No. 1, pp. 415–424 (2018).
Math. Control Signals Syst., Vol. 17, No. 2, 16. Levant, A., “Robust exact differentiation via slid-
pp. 101–127 (2005). ing mode technique,” Automatica, Vol. 34, No. 3,
4. Song, J., Y. Niu, and Y. Zou, “Finite-time stabiliza- pp. 379–384 (1998).
tion via sliding mode control,” IEEE Trans. Autom. 17. Polyakov, A. and A. Poznyak, “Lyapunov func-
Control, Vol. 62, No. 3, pp. 1478–1483 (2017). tion design for finite-time convergence analysis:

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
 M. Basin et al.: Continuous Fixed-Time Convergent Super-Twisting Algorithm 335

Twisting controller for second-order sliding mode with sensor networks in presence of packet dropouts
realization,” Automatica, Vol. 45, No. 2, pp. 444–448 and quantization,” IEEE Trans. Circuits Syst. I:
(2009). Regul. Pap., Vol. 64, No. 10, pp. 2783–2796 (2017).
18. Polyakov, A. and A. Poznyak, “Reaching time esti- 31. Zhang, D., Z. Xu, H. R. Karimi, Q. G. Wang,
mation for super-twisting second order sliding mode and L. Yu, “Distributed H∞ output-feedback con-
controller via Lyapunov function designing,” IEEE trol for consensus of heterogeneous linear multiagent
Trans. Autom. Control, Vol. 54, No. 9, pp. 1951–1955 systems with aperiodic sampled-data communica-
(2009). tions,” IEEE Trans. Ind. Electron., Vol. 65, No. 5,
19. Utkin, V., “On convergence time and disturbance pp. 4145–4155 (2018).
rejection of super-twisting control,” IEEE Trans. 32. Lau, J. Y., W. Liang, H. C. Liaw, and K. K. Tan,
Autom. Control, Vol. 58, No. 8, pp. 2013–2017 (2013). “Sliding mode disturbance observer-based motion
20. Nagesh, I. and C. Edwards, “A multivariable control for a piezoelectric actuator-based surgi-
super-twisting sliding mode approach,” Automatica, cal device,” Asian J. Control, Vol. 20, No. 3,
Vol. 50, No. 3, pp. 984–988 (2014). pp. 1194–1203 (2018).
21. Andrieu, V., L. Praly, and A. Astolfi, “Homogeneous 33. Filippov, A. F., Differential Equations with Discontin-
approximation, recursive observer design, and out- uous Righthand Sides, Kluwer, Dordrecht (1988).
put feedback,” SIAM J. Control Optim., Vol. 47, 34. Basin, M., P. Yu, and Y. Shtessel, “Finite- and
No. 4, pp. 1814–1850 (2008). fixed-time differentiators utilising HOSM tech-
22. Polyakov, A., “Nonlinear feedback design for niques,” IET Contr. Theory Appl., Vol. 11, No. 8,
fixed-time stabilization of linear control systems,” pp. 1144–1152 (2017).
IEEE Trans. Autom. Control, Vol. 57, No. 8,
pp. 2106–2110 (2012).
23. Polyakov, A. and L. Fridman, “Stability notions and X. APPENDIX
Lyapunov functions for sliding mode control sys-
tems,” J. Frankl. Inst., Vol. 351, No. 4, pp. 1831–1865 10.1 Proof of Theorem 1
(2014). A. Consider first that sign(y0 ) is opposite to sign(x0 )
24. Levant, A., “Sliding order and sliding accuracy in or y0 = 0. Let ∣ x0 ∣> 𝜖, where 𝜖 > 0 is a given
sliding mode control,” Int. J. Control, Vol. 58, No. 6, constant. The first equation in (3) yields
pp. 1247–1263 (1993).
25. Shtessel, Y., C. Edwards, L. Fridman, and A. Levant, d ∣ x(t) ∣
Sliding Mode Control and Observation, Birkhäuser, ≤ −𝜆2 ∣ x(t) ∣p , (15)
dt
Basel (2014).
26. Basin, M., C. B. Panathula, and Y. Shtessel, “Mul- taking into account that sign(y(t)) remains opposite
tivariable continuous fixed-time second-order sliding to sign(x(t)) for t > t0 , while x(t) does not cross the
mode control: Design and convergence time esti- axis x = 0, and 𝛼 > L. Rewriting (15) as
mation,” IET Contr. Theory Appl., Vol. 11, No. 8,
pp. 1104–1111 (2017). d ∣ x(t) ∣
=∣ x(t) ∣−p d ∣ x(t) ∣≤ −𝜆2 dt
27. Zhang, D., Q. G. Wang, D. Srinivasan, H. Li, and L. ∣ x(t) ∣p
Yu, “Asynchronous state estimation for discrete-time
and integrating this expression with the initial con-
switched complex networks with communication
dition x(t0 ) = x0 yields
constraints,” IEEE Trans. Neural Netw. Learn. Syst.,
Vol. 29, No. 5, pp. 1732–1746 (2018).
∣ x(t) ∣1−p ∣ x0 ∣1−p
28. Wei, Y., J. Qiu, and H. K. Lam, “A novel approach − ≤ −𝜆2 (t − t0 ),
to reliable output feedback control of fuzzy-affine 1−p 1−p
systems with time delays and sensor faults,” IEEE
which leads to
Trans. Fuzzy Syst., Vol. 25, No. 6, pp. 1808–1823
(2017). ∣ x(t) ∣1−p ∣ x ∣1−p
29. Wei, Y., J. Qiu, and H. R. Karimi, “Reliable output ≤ −𝜆2 (t−t0 )+ 0 ≤ −𝜆2 (t−t0 ),
1−p 1−p
feedback control of discrete-time fuzzy affine systems
with actuator faults,” IEEE Trans. Circuits Syst. I: taking into account 1 − p < 0. Multiplying both
Regul. Pap., Vol. 64, No. 1, pp. 170–181 (2017). parts by p − 1 > 0 results in
30. Zhang, D., Z. Xu, H. R. Karimi, and Q. G. Wang,
“Distributed filtering for switched linear systems − ∣ x(t) ∣1−p ≤ −𝜆2 (t − t0 )(p − 1).

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
336 Asian Journal of Control, Vol. 21, No. 1, pp. 323–338, January 2019

∑
Noting that −x ≤ −y implies x1 ≤ 1y for positive ̇ i )∕m, where x(t
i x(t ̇ i ), i = 1, 2, ..., are derivatives
x > 0 and y > 0 and applying this deduction to the of x(t) at subsequent time moments ti , such that
last expression yields x(ti ) = 0 and x(t ̇ i ) = y(ti ), if qc =∣ x(t ̇ 2) ∣
∕ ∣ x(ṫ 1 ) ∣< 1. Note that Step D begins at one
1 of such time moments, T2 + T3 . Therefore TSTW
∣ x(t) ∣p−1 ≤ .
𝜆2 (p − 1)(t − t0 ) is calculated as TSTW = y2 ∕(1 − qc )m, where the
value of qc can be estimated using the formula
Therefore, ∣ x(t) ∣ decreases and reaches the value qc ≤ Mh(𝜆1 )∕𝜆1 derived in [19]. The necessary
1
∣ x(t) ∣= 𝜖 for a time T1 ≤ 𝜆 (p−1)𝜖 p−1
, which corre- condition qc = Mh(𝜆1 )∕𝜆1 < 1 corresponds
2
sponds to the first term in (4). Note that this term to the second condition for control gains in this
expression is independent of an unknown initial theorem. As noted in [19], the first condition
condition x0 . 𝛼 > L is a mandatory condition for conver-
Step A ends with ∣ x(T1 ) ∣= 𝜖 > 0. If ∣ x0 ∣≤ 𝜖, gence of the super-twisting algorithm. Substituting
1 the obtained estimates
Step A is not executed; therefore, the term 𝜆 (p−1)𝜖
2
p−1 ( for y2 =∣ y(T2 ) ∣ )and qc
1∕2
1
would be absent in (4). yields TSTW < M 𝜆 (p−1)𝜖 p−1
+ 2𝜖𝜆 + MK
∕(1 −
2 1
B. For t > T1 , ∣ x(t) ∣ continues decreasing to zero, Mh(𝜆1 )∕𝜆1 )m, which, after dividing both parts of
reaching it at a certain time T2 . The first equation the fraction by M, corresponds to the former
in (3) yields terms in (4). The optimal value of 𝜖 is determined
[26] minimizing the first two terms in (4) with
d ∣ x(t) ∣
≤ −𝜆1 ∣ x(t) ∣1∕2 , (16) respect to 𝜖.
dt
taking into account that sign(y(t)) remains oppo- 10.2 Proof of Theorem 2
site to sign(x(t)) for t ∈ [T1 , T2 ], until x(t) becomes
equal to zero. Solving (16) implies that A. Consider first that sign(yi (0)) is opposite to
sign(xi (0)) or yi (0) = 0 for any i = 1, … , n. Let
2 ∣ x(t) ∣1∕2 ≤ −𝜆1 (t − T1 ) + 2 ∣ x(T1 ) ∣1∕2 ‖x0 ‖ > 𝜖, where 𝜖 > 0 is a given constant. The first
= −𝜆1 (t − T1 ) + 2𝜖 1∕2 . equation in (7) yields
d‖x(t)‖
Therefore, ∣ x(t) ∣ decreases and reaches zero for ≤ −𝜆2 ‖x(t)‖p , (17)
dt
a time T2 ≤ 2𝜖 1∕2 ∕𝜆1 , which corresponds to the
second term in (4). taking into account that sign(yi (t)) remains oppo-
Step B ends with ∣ x(T2 ) ∣= 0. Given that site to sign(xi (t)) for t > t0 , while xi (t) > 0, i =
sign(y(t)) is opposite to sign(x(t)) for t ∈ [t0 , T2 ] and 1, … , n, and 𝛼 > L. Applying the transformations
𝛼 > L, ∣ y(t) ∣ increases for t ∈ [t0 , T2 ]. The value following the equation (15) with ‖x(t)‖ in place of
∣ y(T2 ) ∣ (
is bounded by ∣ y(T
) 2 ) ∣< K + M(T2 − t0 ) = ∣ x(t) ∣ and the initial condition ‖x(t0 )‖ = ‖x0 ‖, one
K + M 𝜆 (p−1)𝜖1
+ 2𝜖𝜆
1∕2
in view of the second obtains that
p−1
2 1
equation in (3). ‖x(t)‖1−p ‖x ‖1−p
≤ −𝜆2 (t−t0 )+ 0 ≤ −𝜆2 (t−t0 ),
C. Consider now that sign(y0 ) is equal to sign(x0 ). 1−p 1−p
Then, x(t) cannot reach zero until sign(y(t))
becomes opposite to sign(x(t)), that is, y(t) has to which yields
reach zero first. In view of the second equation in 1
(3), the convergence time of y(t) to zero can be esti- ‖x(t)‖p−1 ≤ .
𝜆2 (p − 1)(t − t0 )
mated from above as T3 = K∕m, which should be
added to time T2 calculated at step B. Therefore, each ‖x(t)‖ decreases and reaches the
1
D. It is evident that the trajectory of the system (3) value ‖xi (t)‖ = 𝜖 for a time T1 ≤ 𝜆 (p−1)𝜖 q−1
,
2
starting at (0, y(T2 )) is dominated by the trajec- which corresponds to the first term in (8). Note that
tory
( of the system )(3) starting at (0, y2 = K + this term expression is independent of an unknown
1 1∕2
M 𝜆 (p−1)𝜖 p−1
+ 2𝜖𝜆 and converges to the origin initial condition x0 .
2 1
faster. Following the proof of Theorem 4.5 in [25], Step A ends with ‖x(T1 )‖ = 𝜖 > 0. If ‖x0 ‖ ≤ 𝜖,
1
the finite convergence time for the latter trajec- Step A is not executed; therefore, the term 𝜆 (p−1)𝜖 q−1
2
tory can be estimated by the formula TSTW ≤ would be absent in (8).

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
 M. Basin et al.: Continuous Fixed-Time Convergent Super-Twisting Algorithm 337

√
B. For t > T1 , ∣ xi (t) ∣, i = 1, … , n continue decreasing 2𝛼∕𝜆1 )m, which corresponds to the former terms
to zero, reaching it at no later than a certain time T2 . in (8). The optimal value of 𝜖 is determined in the
The last fact readily follows from the component same way as in Theorem 1.
form of (7)
xi (t)
ẋ i (t) ≤ −𝜆1 + yi (t), 10.3 Proof of Theorem 3
‖x(t)‖1∕2
x (t) In accordance with Theorem 3 in [34], the variables
ẏ i (t) = −𝛼 i + 𝜉i (t), i = 1, … , n.
‖x(t)‖ of the fixed-time observer (10), z1 (t) and z2 (t), converge
̇ for a fixed time
to state variable x(t) and its derivative x(t)
taking account that sign(yi (t)) = −sign(xi (t)), i = no greater than TBBF given by (12). Therefore, z1 (TBBF ) =
1, … , n, for t ∈ [T1 , T2 ], until xi (t) becomes equal ̇ BBF ). Also, in view of the first
x(TBBF ) and z2 (TBBF ) = x(T
to zero. equations in (3) and (10), ŷ (TBBF ) = y(TBBF ). Finally,
Then, the first equation in (7) yields the theorem assertion follows from the formula (4) in
d‖x(t)‖ Theorem 1, setting K =∣ ŷ (TBBF ) ∣.
≤ −𝜆1 ‖x(t)‖1∕2 . (18)
dt
Solving (18) implies that 10.4 Proof of Theorem 4
2‖x(t)‖ 1∕2
≤ −𝜆1 (t − T1 ) + 2‖x(T1 )‖ 1∕2
The proof of Theorem 4 repeats the lines of the
= −𝜆1 (t − T1 ) + 2𝜖 1∕2
. proof of Theorem 3 in the multivariable case.

Therefore,
√ each ∣ xi (t) ∣ reaches zero for a time T2 ≤ Michael Basin (SM’07, M’95) received his
2( n𝜖)1∕2 ∕𝜆1 , which corresponds to the worst case Ph.D. degree in Physical and Mathemat-
of ∣ xi (T1 ) ∣= 𝜖 at the beginning of Step B. This ical Sciences with major in Automatic
expression is equal to the second term in (11). Control and System Analysis from the
Step B ends with all components of x(t) Moscow Aviation University (MAI) in
already crossed or equal to zero. The value ‖y(T2 )‖ 1992. He is currently Full Professor with
is bounded by ‖y(T2√ )‖ < K + M(T2 − t0 ) = the Autonomous University of Nuevo
( )
1 2( n𝜖)1∕2 Leon, Mexico, and Leading Researcher with ITMO
K + M 𝜆 (p−1)𝜖p−1 + 𝜆 in view of the second
2 1 University, St. Petersburg, Russia. Starting from 1992,
equation in (7). Dr. Basin published more than 300 research papers
C. Consider now that sign(yi (t0 )) is equal to sign(xi (t0 )) in international referred journals and conference pro-
at least for one i, i = 1, … , n. Then, xi (t) can- ceedings. He is the author of the monograph “New
not reach zero until sign(yi (t)) becomes opposite to Trends in Optimal Filtering and Control for Polyno-
sign(xi (t)), that is, yi (t) has to reach zero first. In mial and Time-Delay Systems,” published by Springer.
view of the second equation in (7), the convergence His works are cited more than 3000 times (h index =
time of yi (t) to zero can be estimated from above as 32). Dr. Basin has supervised 14 doctoral and 7 mas-
T3 = K∕m, which corresponds to the worst case of ter’s theses. He has served as the Editor-in-Chief and
∣ yi (t0 ) ∣= K. The term for T3 should be added to serves as the Co-Editor-in-Chief in Control of Jour-
time T2 calculated at step B. nal of The Franklin Institute, a Technical Editor of
D. The trajectory of the system (7) starting at (0, y(T2 )) IEEE/ASME Transactions on Mechatronics, an Asso-
is dominated by the trajectory ( of the system√
(7)
) ciate Editor of Automatica, IEEE Transactions on Sys-
1 2( n𝜖)1∕2
starting at (0, y2 = K + M 𝜆 (p−1)𝜖 p−1
+ 𝜆 tems, Man and Cybernetics: Systems, IET-Control The-
2 1
and converges to the origin faster. Following the ory and Applications, International Journal of Systems
proof of Theorem 4.5 in [25], the finite conver- Science, Neural Networks. Dr. Basin was awarded a
gence time for the latter trajectory is calculated title of Highly Cited Researcher by Thomson Reuters,
as TMSTW = ‖y2 ‖∕(1 − qc )m, where the value the publisher of Science Citation Index, in 2009; he
of is a regular member of the Mexican Academy of Sci-
√ qc can be estimated using the relation qc = ences. His research interests include optimal filtering
2𝛼∕𝜆1 obtained in Lemma 1 of [26]. Substituting
and control problems, stochastic systems, time-delay sys-
the obtained estimates for ‖y2 ‖ = ‖y(T2 )‖ and qc
( √ ) tems, identification, sliding mode control and variable
1 2( n𝜖)1∕2 K
yields TMSTW < M 𝜆 (p−1)𝜖p−1 + 𝜆 +M ∕(1− structure systems.
2 1

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd
338 Asian Journal of Control, Vol. 21, No. 1, pp. 323–338, January 2019

Pablo Rodriguez-Ramirez received his Alison Garza-Alonso received her B.S.

Ph.D. degree in Industrial Physics Engi- degree in Mathematics from the School
neering from the School of Physical of Physical and Mathematical Sciences
and Mathematical Sciences at the at the Autonomous University of Nuevo
Autonomous University of Nuevo Leon, Leon, Mexico, in 2017. She is currently
Mexico in 2013 and finished his postdoc- studying towards her M.S. degree at
toral study with the Center for Research the same university. Her research inter-
and Graduate Studies (CINVESTAV) of the National ests include nonlinear dynamical systems including
Polytechnic Institute (IPN), Campus Monterrey, in 2014. finite-time dynamics, super-twisting control and observer
He has been a professor with the Autonomous Univer- design.
sity of Nuevo Leon from 2015. His research interests
include optimal filtering and control problems, stochas-
tic systems, identification and sliding mode control,
nonlinear dynamical systems and control.

© 2018 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd