You are on page 1of 6

Automatica 157 (2023) 111249

Contents lists available at ScienceDirect

Automatica
journal homepage: www.elsevier.com/locate/automatica

Technical communique

Generalized Filippov solutions for systems with prescribed-time


convergence✩
Richard Seeber
Christian Doppler Laboratory for Model Based Control of Complex Test Bed Systems, Institute of Automation and Control,
Graz University of Technology, Graz, Austria

article info a b s t r a c t

Article history: Dynamical systems with prescribed-time convergence sometimes feature a right-hand side exhibiting
Received 20 December 2022 a singularity at the prescribed convergence time instant. In an open neighborhood of this singularity,
Received in revised form 27 May 2023 classical absolutely continuous Filippov solutions may fail to exist, preventing indefinite continuation
Accepted 22 June 2023
of such solutions. This note introduces a generalized Filippov solution definition based on the notion
Available online xxxx
of generalized absolute continuity in the restricted sense. Conditions for the continuability of such
Keywords: generalized solutions are presented and it is shown, in particular, that generalized Filippov solutions
Filippov solution of systems with an equilibrium that is attractive, in prescribed time or otherwise, can always be
Prescribed-time system continued indefinitely. The results are demonstrated by applying them to a prescribed-time control
Unbounded right-hand side design example for a perturbed second-order integrator chain
Time-varying systems
© 2023 Elsevier Ltd. All rights reserved.

1. Introduction example by switching to a sliding mode approach after the con-


vergence time instant. From a theoretical point of view, this raises
Regulation of a dynamical system’s state to the origin is an the issue of continuability of solutions as defined by Filippov
important topic in control theory. In that regard, the type of (1988).
convergence of the state is of particular relevance. While lin- The present communique shows that Filippov solutions of
ear control achieves exponential convergence, more advanced systems featuring prescribed-time convergence may indeed fail
approaches, such as homogeneous or sliding-mode controllers to exist beyond the convergence time instant. A generalized Fil-
or observers, can achieve convergence in finite time (Levant, ippov solution definition is then suggested that eliminates this
2005; Roxin, 1966) or fixed time (Polyakov, 2012). More recently, issue. In particular, it is shown that, with the generalized solution
Holloway and Krstic (2019) and Song et al. (2017) proposed ap- definition, trajectories of prescribed-time systems may always be
proaches achieving so-called prescribed-time convergence using continued indefinitely, i.e., to the time interval [t0 , ∞), paving
time-varying gains that tend to infinity at the prescribed conver- the way for theoretically sound combinations of prescribed-time
gence time instant. While not all prescribed-time approaches use systems with other control approaches.
such unbounded gains, see e.g. Zhou et al. (2022), several other After introducing Filippov solutions in Section 2, Section 3 mo-
prescribed-time approaches based on possibly unbounded time- tivates the discussion with two illustrative examples.
varying gains have since been proposed, e.g., by Aldana-López Section 4 then defines generalized Filippov solutions and dis-
et al. (2022), Espitia and Perruquetti (2022) and Zhou and Shi cusses their properties. Main Theorems 5.2 and 5.7 in Section 5
(2021). show formal results on the continuability of such solutions, and
In Song et al. (2017) and many other works, trajectories are Section 6 applies them in a control design example. Section 7
only considered until the convergence time instant T , i.e., on a draws conclusions.
time interval of the form [t0 , T ). In some scenarios, continuing
trajectories beyond the convergence time instant is desirable, for Notation. R, R≥0 , R>0 , and N denote reals, nonnegative reals,
positive reals, and positive integers, respectively. Convex closure
✩ The financial support by the Christian Doppler Research Association, the and Lebesgue measure of a set S ⊆ Rn are denoted by co S and
Austrian Federal Ministry of Labour and Economy and the National Foundation µ(S ), respectively. An interval is any connected set I ⊆ R with
for Research, Technology and Development is gratefully acknowledged. The µ(I ) ̸= 0. For a function f with domain I and J ⊆ I , f |J is
material in this paper was not presented at any conference. This paper was
recommended for publication in revised form by Associate Editor Maria Letizia
the restriction of f to the domain J . For y ∈ R, ⌊y⌋ and ⌈y⌉ are
Corradini under the direction of Editor André Tits. the largest and smallest integer, respectively, not greater or not
E-mail address: richard.seeber@tugraz.at. smaller than y. The set of all subsets of a set S is denoted by 2S .

https://doi.org/10.1016/j.automatica.2023.111249
0005-1098/© 2023 Elsevier Ltd. All rights reserved.
R. Seeber Automatica 157 (2023) 111249

2. Filippov solutions

This communique studies definitions and existence of solu-


tions of differential equations of the form
ẋ = f(x, t) (1)
n n
where f : R × R≥0 → R may be discontinuous in x and
may, additionally, exhibit singularities in t. Systems of this form Fig. 1. Classical Filippov solution from Example 3.1 with T = 1 and initial
commonly occur in the context of prescribed-time systems, cf. conditions t0 = 0, x(t0 ) = x0 = [1 0]T .
e.g. Holloway and Krstic (2019), Song et al. (2017) and Zhou and
Shi (2021), specifically in the form of time-varying gains that
exhibit a singularity at a prescribed time instant T ∈ R>0 . e.g., by Holloway and Krstic (2019), which at t = T is switched
Solutions of (1) may be studied in terms of the associated to the robust exact super-twisting differentiator due to Levant
differential inclusion
(1998). The next example shows that, beyond the formal simi-
ẋ ∈ F (x, t) (2) larity to the systems in Holloway and Krstic (2019), this system
n Rn
indeed features prescribed-time convergence, but classical Filip-
with F : R × R≥0 → 2 obtained by applying the Filip- pov solutions fail to exist in any (two-sided) neighborhood of the
pov procedure, cf. Filippov (1988), to the right-hand side of (1);
prescribed convergence time instant T , where f (and thus also F )
specifically,
⋂ ⋂ exhibits a singularity.
F (x, t) = co {f(x̃, t) : x̃ ∈ Bδ (x) \ S } (3)
δ∈R>0 S ⊂Rn
Example 3.1. Consider system (1) with previously defined right-
µ(S )=0 hand side (4)–(5). Note that f(·, t) is globally Lipschitz continuous
with Bδ (x) = {x̃ ∈ Rn : ∥x − x̃∥ < δ}. The definition of on every compact subset of [0, T ). Hence, for every x0 ∈ R2
Filippov solutions of (1) is based on the notion of (local) absolute and every t0 ∈ [0, T ), the function x : [t0 , T ) → R2 defined as
continuity. x(t) = M(T − t)M(T − t0 )−1 x0 with
[ ]
τ 3 sin τ1 τ 3 cos τ1
Definition 2.1. Let n ∈ N. A function g : R → Rn is called M(τ ) = (6)
τ 2 sin τ1 + τ cos τ1 τ 2 cos τ1 − τ sin τ1
(a) absolutely continuous (AC) on a closed ∑ interval I , if for each
ε > 0 there exists δ > 0 such that m i=1 ∥g(bi ) − g(ai )∥ < ε is the unique classical solution with x(t0 ) = x0 ; for T = 1, it
holds for each finite collection of non-overlapping intervals is depicted in Fig. 1. However, its only continuous extension to
[ai , bi ] ⊆ I , i = 1, . . . , m, that satisfies m i=1 |bi − ai | < δ ; [t0 , T ] with x(T ) = 0 is not absolutely continuous, because it

(b) locally absolutely continuous (LAC) on an interval I , if it is does not have bounded variation. Indeed, the total variation of
absolutely continuous on each compact interval J ⊆ I . the term g(τ ) = τ cos τ1 with τ = T − t ∈ (0, T − t0 ], for example,
is bounded from below by
It is also worth to recall that a function is absolutely contin-
∞ ⏐ ( ∞ ⏐
⏐ (−1)i (−1)i+1 ⏐

uous if and only if it is continuous, has bounded variation, and
) ( )⏐ ∑
⏐g 1 − g 1
∑ ⏐ ⏐
⏐= −
satisfies Lusin’s N-property (Gordon, 1994). According to Filip-
⏐ ⏐
⏐ iπ (i + 1)π ⏐ ⏐ iπ (i + 1)π ⏐
pov (1988), solutions of (1)—which are called classical Filippov i=i0 i=i0

solutions hereafter—are typically defined as follows. ∞


∑ 2
≥ = ∞, (7)
n (i + 1)π
Definition 2.2. Let I ⊆ R≥0 be an interval. A function x : I → R i=i0
is said to be a classical solution of (2) or a classical Filippov solution
of (1), if it is LAC on I and satisfies (2) almost everywhere on I . wherein i0 = ⌈ (T −1t )π ⌉. Hence, every (nontrivial) continuous
0
function defined on (T − ε, T + ε ) with ε > 0 that satisfies the
3. Motivating examples corresponding inclusion (2) almost everywhere is not locally ab-
solutely continuous. Consequently, no nontrivial classical Filippov
To motivate the discussion, consider the error dynamics of a solutions exist on intervals of this form, i.e., classical solutions
prescribed-time differentiator given by (1) with x = [x1 x2 ]T , cannot be continued beyond the singularity occurring at T . △
−h1 (x1 , t) + x2
[ ]
The next example shows that, for systems that do not feature
f(x, t) = , (4)
−h2 (x1 , t) prescribed-time convergence, continuability of bounded solutions
can fail due to yet another problem.
correction functions h1 , h2 : R × R≥0 → R defined as
{ 4
x t ∈ [0, T ) Example 3.2. Consider system (1) with
h1 (x1 , t) = T −t 1
√ (5a)
|x1 | sign(x1 ) t ∈ [T , ∞)
{ [ ]
k1 A(T − t)x t ∈ [0, T ) − τ1 1
{ f(x, t) = A(τ ) = , (8)
2
x + 1
x t ∈ [0, T ) A(T )x t ∈ [T , ∞), − τ12 0
h2 (x1 , t) = (T −t)2 1 (T −t)4 1 (5b)
k2 sign(x1 ) t ∈ [T , ∞), with T ∈ R>0 . For x0 ∈ R2 , t0 ∈ [0, T ), the unique (and bounded)
and positive constants k1 , k2 , T ∈ R>0 . On the time interval classical solution is given by the function x : [t0 , T ) → R2 defined
[0, T ), these error dynamics correspond to a linear time-varying as x(t) = M(T − t)M(T − t0 )−1 x0 with
differentiator of similar structure1 as the observers proposed, [
τ sin ln τ τ cos ln τ
]
M(τ ) = . (9)
− cos ln τ sin ln τ
1 Note that the example is not a special case of the observers in Holloway and
Krstic (2019), but nonetheless corresponds to an observer with prescribed-time Fig. 2 exemplarily depicts it for T = 1. Since limt →T x(t) does not
convergence. exist for x0 ̸ = 0, every nontrivial function that satisfies (1) almost
2
R. Seeber Automatica 157 (2023) 111249

and Fridman (2014) for an overview—by analogously relaxing the


absolute continuity requirement in those definitions to general-
ized absolute continuity in the restricted sense.

It is worth noting that, while LAC functions are associated


to the Lebesgue integral, ACG∗ functions are associated to the
Denjoy integral.2 Specifically, a function h : I → Rn is Denjoy
Fig. 2. Classical Filippov solution from Example 3.2 with T = 1 and initial
integrable if and only if there exists an ACG∗ function g : I → Rn
conditions t0 = 0, x(t0 ) = x0 = [1 0]T . such that ġ(t) = h(t) holds almost everywhere on I , cf. for
example Gordon (1994, Definition 7.1).
A similar solution definition for ordinary differential equations,
everywhere on (T − ε, T + ε ) with ε > 0 is discontinuous at t = T . using the Perron integral, is considered by Kurzweil and Schwabik
As a consequence, no nontrivial classical Filippov solutions exist (1990), who also study some general criteria for the existence
on intervals of this form. △ of such solutions, cf. Schwabik (1990) and references therein.
Here, the considered generalized solution definition for Filippov
Recall again that absolute continuity is equivalent to the systems is studied in the context of prescribed-time systems,
three properties continuity, boundedness of variation, and Lusin’s the main result being conditions for continuability of generalized
N-property. While Lusin’s property cannot conceivably be lost, solutions for all systems of this type.
solutions near t = T fail to exist due to a lack of bounded
variation in Example 3.1 and due to a lack of continuity in
Example 3.2. It is clear that, in the latter example, any attempt to 4.2. Properties of generalized solutions
continuously continue solutions must fail. However, intuitively,
solutions in the former example (as shown in Fig. 1) should be The following lemma, proven in the appendix, shows some
continuable by the zero solution. connections between the different classes of functions (cf. also
The present communique addresses this issue by proposing a Gordon, 1994).
generalized Filippov solution definition, which guarantees exis-
tence and continuability of solutions under very mild conditions. Lemma 4.4. Consider an interval I ⊆ R, a continuous function
In particular, it will be shown that the proposed generalized g : I → Rn , and a countable collection of intervals (Ij ), j ∈ N, with
Filippov solutions are always continuable to the interval [t0 , ∞) I = ∪∞ 3
j=1 Ij . Then, the following statements are true :
for systems whose origin is attractive, in prescribed time or
otherwise. (a) if I is closed, then g is AC on I if and only if it is AC∗ on I ;
(b) if g is LAC on I , then it is ACG∗ on I ;
4. Generalized Filippov solutions (c) if g is AC∗ on I , then it is LAC on I ;
(d) if g is ACG∗ on each Ij , then it is ACG∗ on I .
In the following, a generalized solution definition for Filippov
From item (b), the following proposition is obvious.
type systems is introduced based on the class of functions that are
generalized absolutely continuous in the restricted sense (ACG∗ )
Proposition 4.5. Every classical solution of (2) is also a generalized
as originally introduced by Lusin (1912), cf. Saks (1937).
solution.
Definition 4.1 (Gordon, 1994, Definition 6.1). Let n ∈ N. A function The next proposition shows how classical solutions can be
g : R → Rn is said to be recovered from generalized solutions using item (c).
(a) absolutely continuous in the restricted sense (AC∗ ) on a set
S ⊆ R, if for each ε > 0 there exists δ > 0 such that Proposition 4.6. Let x : I → Rn be a generalized solution of (2).
m
i=1 supa,b∈[ai ,bi ] ∥g(b) − g(a)∥ < ε for each finite collection
Then, for every interval J ⊆ I , an open interval H ⊆ J exists such

of non-overlapping intervals ∑m [ai , bi ], i = 1, . . . , m, that that x|H is a classical solution.
satisfies ai , bi ∈ S and i=1 |bi − ai | < δ ;
(b) generalized absolutely continuous in the restricted sense Proof. Let J˜ be a closed interval contained in J . Since J˜ is
(ACG∗ ) on a set S ⊆ R, if g|S is continuous on S and S can a perfect set, it contains an open subinterval H on which x is
be written as the union of countably many sets on each of AC∗ according to Gordon (1994, Theorem 6.1). Hence, x|H is
which g is AC∗ . LAC according to Lemma 4.4, item (c), and is thus a classical
solution. □
4.1. Definition of generalized solutions
Item (d) shows that concatenating countably many general-
ized solutions again yields a generalized solution.
Like LAC functions, also ACG∗ functions are differentiable al-
most everywhere, cf. Gordon (1994, Corollary 6.19). This leads to
Proposition 4.7. Let (Ij ), (xj ), j ∈ N, be a countable collection
the following definition.
of intervals and corresponding classical or generalized solutions
xj : Ij → Rn of (2) with the property that I = ∪∞ j=1 Ij is an interval
Definition 4.2. Let I ⊆ R≥0 be an interval. A function x : I → Rn
is said to be a generalized solution of (2) or a generalized Filip- and that, for every pair of integers i, j ∈ N, limt →τ [xj (t) − xi (t)] = 0
pov solution of (1), if it is ACG∗ on I and satisfies (2) almost for all τ ∈ co Ii ∩ co Ij . Then, x : I → Rn defined as x(τ ) = xj (τ )
everywhere on I . for τ ∈ Ij is a generalized solution of (2).

Remark 4.3. The proposed definition may also be extended to 2 Or to the equivalent Perron or Henstock–Kurzweil integral.
different solution definitions of discontinuous systems, such as 3 Note that item (d) of Lemma 4.4, and consequently also Proposition 4.7,
Utkin solutions or Aizerman–Pyatnitskii solutions—cf. Polyakov depend on the Axiom of Choice unless the collection (Ij ) is finite.

3
R. Seeber Automatica 157 (2023) 111249

Proof. Clearly, x is well-defined and continuous by construction. Proof. Since the zero solution x̄(t) = 0 is a solution, the inclusion
Moreover, from Proposition 4.5, each xj is a generalized solution 0 ∈ F (0, t) holds for almost all t ∈ R≥0 . Consider now any
and hence ACG∗ on Ij . Thus, x is ACG∗ on I due to Lemma 4.4, maximal solution x : I → Rn satisfying limt →T x(t) = 0
item (d). Since it also satisfies the inclusion almost everywhere by assumption. If T = sup I , then there is nothing to prove.
on each Ij , it is a generalized solution of (2). □ Otherwise, T ∈ I is finite and x(T ) = 0 holds by continuity of
x. Since the zero solution is unique, then x(t) = 0 holds for all
5. Continuability of generalized solutions t ∈ [T , ∞) ∩ I , implying limt →sup I x(t) = 0. □

Formal conditions for the continuability of generalized Fil- Denote by D the set of time instants where (2) exhibits a
ippov solutions are now shown. First, general systems with a singularity, i.e., where F is not locally bounded.4 It will be shown
singularity at some time instant T ∈ R>0 are considered. Then, that if D does not have cluster points, then generalized solutions
prescribed-time systems are addressed, by showing indefinite of systems with attractive equilibria in the classical sense are
continuability of generalized solutions for systems with an attrac- always continuable indefinitely. If D is moreover finite, attractive
tive equilibrium. equilibria in the classical and generalized sense will be shown to
be equivalent. To that end, the following auxiliary lemma is used,
5.1. General systems with singularities which is proven in the appendix.

Definition 5.1. A solution x : I → Rn is said to be maximal, Lemma 5.6. Let D ⊂ R≥0 and consider the inclusion (2). Suppose
if there exists no solution x̄ : Ī → Rn with Ī ∩ I ̸ = {} and that the set D has no cluster points and that F is locally bounded
sup Ī > sup I satisfying x(t) = x̄(t) for all t ∈ I ∩ Ī . on G = Rn × (R≥0 \ D). Let x : I → Rn be a maximal generalized
solution. Then, if either sup I or sup D are finite, an interval J ⊆ I
The next theorem shows that generalized solutions that are
with sup J = sup I exists such that x|J is a maximal classical
continuous at a singularity at some time instant T can be contin-
solution.
ued beyond T subject to the usual Filippov conditions, cf. Filippov
(1988).
Theorem 5.7. Let D ⊂ R≥0 and the inclusion (2) satisfy the
conditions of Lemma 5.6, and suppose that the origin is an attractive
Theorem 5.2. Let T , ε ∈ R>0 and consider the inclusion (2).
n equilibrium in the classical sense. Then, every maximal generalized
Suppose that F : Rn × R≥0 → 2R is upper semicontinuous on every
solution x : I → Rn satisfies sup I = ∞. If, moreover, sup D is
compact subset of G = Rn × [T , T + ε ) and that the set F (x, t) is
finite, then the origin is an attractive equilibrium in the generalized
nonempty, compact, and convex for all (x, t) ∈ G . Let x : I → Rn be
sense.
a maximal generalized solution of (2). If sup I ≥ T and limt →T x(t)
exists, then there exists δ ∈ R>0 such that sup I ≥ T + δ .
Proof. Let x : I → Rn be a maximal generalized solution and
suppose to the contrary that sup I = T is finite. From Lemma 5.6
Remark 5.3. Note that the theorem imposes conditions on F only
after the singularity, and requires existence of the generalized there exists an interval J with sup J = T such that x|J is a
solution only before and its continuity at the singularity (cf., maximal classical solution; hence, limt →T x(t) = 0. But then the
however, Example 3.2). generalized solution x can be continued with the zero solution
using Proposition 4.7, a contradiction. Hence, sup I = ∞. If,
Proof. Suppose to the contrary that sup I = T , and define moreover, sup D is finite, then from Lemma 5.6 there exists an
x̄0 = limt →T x(t). According to Filippov (1988, Theorem 7.1), there interval J with sup J = ∞ such that x|J is a maximal classical
exists a classical solution x̄ : [T , T + δ] → Rn with x̄(T ) = x̄0 solution. Hence, limt →∞ x(t) = limt →∞ x|J (t) = 0. □
and some δ ∈ R>0 . Due to Proposition 4.7, there then exists a
generalized solution defined on I ∪ [T , T + δ] which coincides 6. Design example: Prescribed-time control
with x on I . This contradicts the fact that x is maximal. □
In the following, a class of prescribed-time controllers for a
Theorem 5.2 already shows local continuability of solutions
second-order system is designed, utilizing the proposed solu-
in Example 3.1. For systems with attractive equilibria, a more
tion definition. Consider the perturbed double integrator ẋ1 = x2 ,
powerful result, allowing to continue solutions indefinitely, is
shown in the next section. ẋ2 = u + w with a time-varying perturbation w bounded accord-
ing to |w (t)| ≤ W ∈ R≥0 for all t ∈ R≥0 subject to the control law
5.2. Systems with attractive equilibria
⎧ [ ]
⎨− γ 2 + 6 4
x1 − T − x t<T
4 2 t 2
u(x, t) = (T −t) (T −t)
(10)
Definition 5.4. The origin of the inclusion (2) is said to be ⎩−k sign(x ) − k sign(x ) t≥T
1 1 2 2
an attractive equilibrium in the classical (generalized) sense, if
0 ∈ F (0, t) holds for almost all t ∈ R≥0 , and all maximal classical with prescribed time T ∈ R>0 and parameters γ , k1 , k2 ∈ R>0 .
(generalized) solutions x : I → Rn satisfy limt →sup I x(t) = 0. Without perturbation, i.e., for w (t) ≡ 0, one non-trivial Filip-
pov solution of the closed loop on the time interval [0, T ) is, for
To establish the connection to prescribed-time systems, it is
example, given by
first shown that such systems always have an attractive equilib-
γ
rium if the zero solution is unique. x1 (t) = (T − t)3 sin , (11)
T −t
γ γ
Proposition 5.5. Consider the inclusion (2). Suppose that for all x2 (t) = (T − t)γ cos − 3(T − t)2 sin .
t0 ∈ R≥0 the unique classical (generalized) solution x̄ with x̄(t0 ) = 0 T −t T −t
satisfies x̄(t) = 0 for all t ≥ t0 , and that every maximal classical
(generalized) solution x : I → Rn fulfills limt →T x(t) = 0 for some 4 Note that a set-valued function F is called locally bounded on a set G , if
T ∈ R≥0 ∪ {∞}. Then, the origin is an attractive equilibrium in the for all y ∈ G there exist constants ε, M ∈ R>0 such that supz∈F (ỹ) ∥z∥ ≤ M holds
classical (generalized) sense. for all ỹ ∈ G with ∥y − ỹ∥ < ε .

4
R. Seeber Automatica 157 (2023) 111249

Similar to Example 3.1, boundedness of variation is lost near T , 7. Conclusion


and hence this solution cannot cross that singularity as a clas-
sical Filippov solution. Nevertheless, the following proposition It was shown that Filippov solutions of prescribed-time sys-
and its proof show that the considered control law achieves tems may in some cases fail to be continuable beyond the pre-
prescribed-time convergence on [0, T ) and finite-time conver- scribed convergence time instant. To overcome this restriction,
gence on [T , ∞), and thus renders the origin attractive in the a definition of generalized Filippov solutions was introduced.
classical sense, permitting the application of Theorem 5.7. Indefinite continuability of such generalized solutions was shown
for systems with an attractive equilibrium and thus, in particu-
Proposition 6.1. Let γ , T ∈ R>0 , W ∈ R≥0 and suppose that lar, for all systems featuring prescribed-time convergence. These
k1 > k2 + W > 2W . Consider the closed loop ẋ = f(x, t) defined via results pave the way for theoretically sound combinations of
ẋ1 = x2 , ẋ2 = u(x, t) + w (t) with u defined in (10) and arbitrary new or existing prescribed-time approaches with other control
Lebesgue measurable disturbance w : R≥0 → [−W , W ]. Then, the techniques, such as sliding mode control. This was demonstrated
closed loop’s origin is an attractive equilibrium in the classical sense. by designing a combined prescribed-time/sliding-mode controller
for a perturbed second-order integrator chain.
Remark 6.2. Using this proposition and Theorem 5.7 with the
set D = {T }, indefinite continuability of generalized solutions Appendix. Proofs
and attractivity of the origin in the generalized sense may be
concluded for the perturbed closed loop.
Proof of Lemma 4.4. For item (a), note that if S is a closed
Proof. Let t0 ∈ R≥0 and consider maximal classical Filippov interval in Definition 4.1, then [ai , bi ] ⊆ S if and only if ai , bi ∈ S .
solutions x : I → R of the closed loop with t0 ∈ I . Distinguish Thus, ∥g(bi ) − g(ai )∥ ≤ supa,b∈[ai ,bi ] ∥g(b) − g(a)∥ shows that AC∗
the cases t0 ≥ T and t0 ∈ [0, T ). In the first case, sup I = ∞ and implies AC. Conversely, suppose that f is AC, let ε > 0 and
limt →∞ x(t) = 0 follows from finite-time stability of the twisting let δ > 0 be as in Definition 2.1. Consider any finite collection
algorithm, cf. Levant (2007). In the second case, convergence in Ii = [ai , bi ] of non-overlapping intervals as in Definition 4.1.
prescribed time T , i.e., [t0 , T ) ⊆ I and limt →T x(t) = 0, will Since f is continuous and Ii is compact, there exist ci , di ∈ Ii ⊆ S
be shown. Solution existence on each compact subinterval of with ci < di such that supa,b∈Ii ∥g(b) − g(a)∥ = ∥g(di ) − g(ci )∥.
[t0 , T ) follows from the fact that the closed loop is an affine time-
∑m ∑m
Noting
∑m that i=1 |di − ci | ≤ i=1 |bi − ai | < δ then yields
i=1 supa,b∈Ii ∥g(b) − g(a)∥ < ε from Definition 2.1. For item (b),
varying system satisfying a global Lipschitz condition on such
a subinterval. To show prescribed-time convergence, define the since I is an interval, it can be written as the countable union of
abbreviation q(x, t) = (T −1t)2 + T −2 t and consider the function
3x x
compact intervals Jj , on each of which g is AC. Since each interval
Q : R2 × [0, T ) → R≥0 defined as Jj is closed, g being AC implies g being AC∗ on Jj due to item (a).
Hence, g is ACG∗ on I by definition. Item (c) follows from the fact
γ 2 x21
Q (x, t) = + q(x, t)2 , (12) that g is AC∗ on every subset of I , and hence AC on all compact
(T − t)6 subintervals of I due to item (a). For item (d), for each j, define Hj
whose time derivative Q̇ = ∂∂Qx f + ∂∂Qt along closed-loop trajec- as the countable set of sets according to Definition 4.1, with union
2w (t)q(x,t) Ij = ∪S ∈Hj S , on which g is AC∗ . Then, the set of sets H = ∪∞ j=1 Hj ,
tories on the time interval [0, T ) is given by Q̇ (x, t) = .
T −t being the countable union of countable sets, is countable, and g
Define V : R2 × [0, T ) → R≥0 as V (x, t) = (T − t)Q (x, t), whose is AC∗ on each S ∈ H. Since ∪S ∈H S = ∪∞ j=1 Ij = I , g is ACG∗ on
time-derivative satisfies I by definition. □
V̇ (x, t) = −Q (x, t) + 2w (t)q(x, t)
]2 Proof of Lemma 5.6. Let T = sup I ∈ R≥0 ∪ {∞} and choose
q(x, t)
[
≤ −Q (x, t) + 2w(t)q(x, t) + 2w(t) − θ ∈ I such that D ∩ (θ, T ) = ∅, which is possible due to the
2 conditions imposed on D. From Proposition 4.6, there exists an
3 3 V (x, t) open interval H ⊆ (θ, T ), such that x|H is a classical solution.
≤ − Q (x, t) + 4w (t)2 ≤ − + 4W 2 Choose any τ1 ∈ H and the largest τ2 ∈ (τ1 , T ] such that x is
4 4 T −t
3 V (x, t) LAC on J = [τ1 , τ2 ). To show sup J = sup I , which also implies
≤− + 4W 2 . (13) that x|J is maximal because otherwise x could be extended using
4 T
Proposition 4.7, suppose to the contrary that τ2 < T . Consider
As a consequence, V is uniformly bounded on [t0 , T ); specifically, the compact interval J¯ = [τ1 , τ2 ], on which x is not AC by
V (x(t), t) ≤ max{V (x(t0 ), t0 ), √4W } holds for all t ∈ [t0 , T ). construction. Then, x|J¯ is continuous and has Lusin’s N-property
3T
The claim limt →T x(t) = 0 is now obtained by contradiction as according to Gordon (1994, Theorem 6.12), because x is ACG∗ on
follows. Suppose that a sequence (tk ), tk ∈ [t0 , T ) and a nonzero J¯ ⊂ I . Let M = supt ∈J¯ ∥x(t)∥ be the uniform bound of x on
vector c ∈ R2 exist with limk→∞ tk = T and limk→∞ x(tk ) = c. J¯ , and let Q be the uniform bound of F on the compact subset
Since lim supt →T V (c, t) = ∞ for each fixed, nonzero vector c, {z ∈ Rn : ∥z∥ ≤ M } × J¯ of G . Then, the total variation of x on J¯
lim sup V (x(t), t) ≥ lim sup V (x(tk ), tk ) is bounded from above by (τ2 − τ1 )Q . But this implies that x is
t →T k→∞ AC on J¯ , a contradiction. □
≥ lim sup V (c, tk ) = ∞ (14)
k→∞ References
then follows, which contradicts the fact that V (x(t), t) is uni-
Aldana-López, R., Seeber, R., Gómez-Gutiérrez, D., Angulo, M. T., & Defoort, M.
formly bounded on [t0 , T ). □ (2022). A redesign methodology generating predefined-time differentia-
tors with bounded time-varying gains. International Journal of Robust and
Remark 6.3. It is worth noting that the arguments of the previous Nonlinear Control, http://dx.doi.org/10.1002/rnc.6315.
proof may also be used to show (non-uniform) Lyapunov stabil- Espitia, N., & Perruquetti, W. (2022). Predictor-feedback prescribed-time sta-
bilization of LTI systems with input delay. IEEE Transactions on Automatic
ity of the closed-loop system’s origin, provided that generalized Control, 67(6), 2784–2799.
Filippov solutions are considered. Arguing this in formal detail is Filippov, A. F. (1988). Differential equations with discontinuous right-hand side.
beyond the scope of the present paper, however. Dordrecht, The Netherlands: Kluwer Academic Publishing.

5
R. Seeber Automatica 157 (2023) 111249

Gordon, R. A. (1994). The integrals of Lebesgue, Denjoy, Perron, and Henstock. Polyakov, A., & Fridman, L. (2014). Stability notions and Lyapunov functions
Providence, RI, USA: American Mathematical Society. for sliding mode control systems. Journal of the Franklin Institute, 351(4),
Holloway, J., & Krstic, M. (2019). Prescribed-time observers for linear systems 1831–1865.
in observer canonical form. IEEE Transactions on Automatic Control, 64(9), Roxin, E. (1966). On finite stability in control systems. Rendiconti del Circolo
3905–3912. Matematico di Palermo, 15(3), 273–282.
Kurzweil, J., & Schwabik, Š. (1990). Ordinary differential equations the solution Saks, S. (1937). Theory of the integral. New York, NY, USA: Hafner Publishing
of which are ACG∗ -functions. Archivum Mathematicum, 26(2–3), 129–136. Company.
Levant, A. (1998). Robust exact differentiation via sliding mode technique. Schwabik, Š. (1990). Generalized ordinary differential equations – a survey. In
Automatica, 34(3), 379–384. Proc. of the 7th Czechoslovak conference on differential equations and their
Levant, A. (2005). Homogeneity approach to high-order sliding mode design. applications (pp. 59–70).
Automatica, 41(5), 823–830. Song, Y., Wang, Y., Holloway, J., & Krstic, M. (2017). Time-varying feedback
Levant, A. (2007). Principles of 2-sliding mode design. Automatica, 43(4), for regulation of normal-form nonlinear systems in prescribed finite time.
576–586. Automatica, 83, 243–251.
Lusin, N. (1912). Sur les propriétés des fonctions mesurables. Comptes Ren- Zhou, B., Michiels, W., & Chen, J. (2022). Fixed-time stabilization of linear delay
dus Hebdomadaires de Séances de l’Académie des Sciences de Paris, 154, systems by smooth periodic delayed feedback. IEEE Transactions on Automatic
1688–1690. Control, 67(2), 557–573.
Polyakov, A. (2012). Nonlinear feedback design for fixed-time stabilization Zhou, B., & Shi, Y. (2021). Prescribed-time stabilization of a class of nonlinear
of linear control systems. IEEE Transactions on Automatic Control, 57(8), systems by linear time-varying feedback. IEEE Transactions on Automatic
2106–2110. Control, 66(12), 6123–6130.

You might also like