On Discontinuous Differential Equations

On Discontinuous Differential Equations
Iguer Luis Domini dos Santos
Universidade Estadual Paulista

[email protected]
February 2019
Iguer Luis Domini dos Santos (UNESP) Discontinuous Differential Equations February 2019 1 / 33
Summary
1 Introduction
2 Preliminaries
Lebesgue integral
Absolutely continuous functions
First fundamental theorem of calculus
3 Generalized solutions
Euler and Hermes solutions
Filippov and Krasovskii solutions
Sentis solutions
Introduction
Discontinuous differential equations: ordinary differential equations

with a discontinuous right hand side
Generalized solutions to discontinuous differential equations
Discontinuous differential equations: Bacciotti (2005), Bacciotti
(2004), da Silveira (2009), Hajek (1979), Sentis (1978), dos Santos
(2015), Clarke et al. (1998)
Euler solutions: Bacciotti (2004), da Silveira (2009), dos Santos
(2015), Clarke et al. (1998)
Hermes solutions: da Silveira (2009), Hajek (1979), dos Santos
(2015)
Filippov solutions: Bacciotti (2005), da Silveira (2009), Hajek (1979),
Sentis (1978), dos Santos (2015)
Krasovskii solutions: da Silveira (2009), Hajek (1979), dos Santos
(2015)
Sentis solutions: Sentis (1978), Bacciotti (2005), Bacciotti (2004)
Consider the initial value problems
ẋ(t) = f (t, x(t)), x(a) = x0 (1)
and
ẋ(t) = g (x(t)), x(a) = x0 (2)
where f : [a, b] × Rn → Rn and g : Rn → Rn .
Properties of the Euler and Hermes solutions to Eq. (1)
The study of Filippov, Krasovskii and Sentis solutions are related to
Eq. (2)
Here we obtain an analogue of Theorem 1.7. (Clarke et al. (1998))
for Hermes solutions
Motivated by Bacciotti (2004) (by the study on relationships between
generalized solutions to Eq. (2)), we study relationships between the
Euler, Filippov, Hermes, Krasovskii, and Sentis generalized solutions
to Eq. (2)
Lebesgue integral
The Banach space of Lebesgue integrable functions x : [a, b] → R with the

norm Z b
kx(·)kL1 = |x(t)|dt
a
will be denoted by L1 ([a, b]).
Absolutely continuous functions
Definition
A function x : [a, b] → Rn is called absolutely continuous if ∀ε > 0, there
exists δ > 0 such that, for any countable collection of disjoint subintervals
[ak , bk ] of [a, b] such that
X
(bk − ak ) < δ,
we have X
|x(bk ) − x(ak )| < ε.
An absolutely continuous function x : [a, b] → Rn is differentiable
almost everywhere, and its derivative ẋ(·) is a Lebesgue integrable
function. Moreover, the Newton-Leibniz formula is true; that is,
Z t2
x(t2 ) − x(t1 ) = ẋ(t)dt
t1
for all t1 , t2 ∈ [a, b], t1 < t2 .

Any absolutely continuous function x : [a, b] → Rn can be represented
in the form Z t
x(t) = x(a) + ẋ(s)ds
a
We will refer to any absolutely continuous function x : [a, b] → Rn as
an arc on [a,b]
Gronwall’s Lemma
Proposition
Let x be an arc on [a, b] which satisfies
kẋ(t)k ≤ γkx(t)k + c(t) a.e., t ∈ [a, b],
where γ is a nonnegative constant and where c(·) ∈ L1 ([a, b]) is a

nonnegative function. Then, for all t ∈ [a, b], we have
Z t
γ(t−a)
kx(t) − x(a)k ≤ (e − 1)kx(a)k + e γ(t−s) c(s)ds.
a
If the function c is constant and γ > 0, this becomes
kx(t) − x(a)k ≤ (e γ(t−a) − 1)(kx(a)k + c/γ).
First fundamental theorem of calculus
Theorem
Let f be a function that is integrable on [a, x] for each x in [a, b]. Let c be
such that a ≤ c ≤ b and define a new function A as follows:
Z x
A(x) = f (t)dt if a ≤ x ≤ b.
c
Then the derivative A0 (x) exists at each point x in the open interval (a, b)
where f is continuous, and for such x we have
A0 (x) = f (x).
Euler solutions to Eq. (1)
Let
π = {t0 , t1 , ..., tN−1 , tN }
be a partition of [a, b], where t0 = a and tN = b. The diameter of the
partition π is given by
µπ := max{ti − ti−1 : 1 ≤ i ≤ N}.
Definition
We define the Euler polygonal arc for Eq. (1) and corresponding to the
partition π, by the arc xπ : [a, b] → Rn given by:
xπ (t0 ) = x0 , xπ (t) = x0 + (t − t0 )f (t0 , x0 ), t ∈ [t0 , t1 ]

xπ (t1 ) = x1 , xπ (t) = x1 + (t − t1 )f (t1 , x1 ), t ∈ [t1 , t2 ]
and by induction
xπ (ti ) = xi , xπ (t) = xi + (t − ti )f (ti , xi ), t ∈ [ti , ti+1 ]
when i ∈ {0, 1, . . . , N − 1}.
Definition
We say that the arc x : [a, b] → Rn is an Euler solution for (1), if x is the
uniform limit of Euler polygonal arcs xπj , corresponding to some sequence
πj such that µπj → 0.
Below we state the Theorem 1.7. (Clarke et al. (1998)).
Theorem
Suppose that for positive constants γ and c and for all (t, x) ∈ [a, b] × Rn ,
we have the linear growth condition
kf (t, x)k ≤ γkxk + c,
where f is otherwise arbitrary. Then:

(a) At least one Euler solution x to the initial-value problem (1) exists on
[a, b], and any Euler solution is Lipschitz.
(b) Any Euler arc x for f on [a, b] satisfies
kx(t) − x(a)k ≤ (t − a)e γ(t−a) (c + γkx(a)k), a ≤ t ≤ b.
(c) If f is continuous, then any Euler arc x for f on [a, b] is continuously

differentiable on (a, b) and satisfies ẋ(t) = f (t, x(t)) ∀t ∈ (a, b).
Example
Let f : [0, 1] × R → R be the function given by

1, x = 0
f (t, x) =
0, x 6= 0
and let x0 = 0. Then x(t) = 0 is the only Euler solution for (1).
Example
Consider the function f : [0, 1] × R → R given by

0, x = 0
f (t, x) =
1, x 6= 0
and let x0 = 0. Then x(t) = 0 is the only Euler solution for (1).
Example
Consider the function f : [0, 1] × R → R given by

0, x = 0
f (t, x) =
1, x 6= 0
and let x0 = 1. Then x(t) = 1 + t is the only Euler solution for (1).
Hermes solutions to Eq. (1)
Definition
An arc x : [a, b] → Rn is a Carathodory solution to (1) if, and only if, x
satisfies the differential equation given in Eq. (1) for a.e. t ∈ [a, b] and
x(a) = x0 .
Below we define Hermes solutions as dos Santos (2015).
Definition
Let x : [a, b] → Rn be an arc. We say that x is a Hermes solution to Eq.
(1) if, and only if, there exist functions Lebesgue measurable
pj : [a, b] → Rn and Carathodory solutions xj to the initial value problem
ẏ (t) = f (t, y (t) + pj (t)), y (a) = x0 (3)
such that pj ⇒ 0 and xj ⇒ x.
Proposition (da Silveira (2009))

Suppose that there are positive constants γ and c such that
kf (t, x)k ≤ γkxk + c (4)
for every (t, x) ∈ [a, b] × Rn . Then every Euler solution of Eq. (1) is a
Hermes solution.
Example
Consider again the function f : [0, 1] × R → R given by

0, x = 0
f (t, x) =
1, x 6= 0
and let x0 = 0. Then x(t) = t is a Hermes solution to Eq. (1). Since in

this case x(t) = 0 is the only Euler solution to Eq. (1), we conclude that
in general a Hermes solution is not an Euler solution.
Below we get properties for Hermes solutions that make an analogy with
the Theorem 1.7. (Clarke et al. (1998)).
Theorem
Suppose f is Lebesgue measurable in t for each x fixed. Assume also that
f satisfies the linear growth condition given in Eq. (4). Then:
(a) There exists at least one Hermes solution y to Eq. (1). In addition,
any Hermes solution is Lipschitz continuous.
(b) There exists a positive constant K such that every Hermes solution
satisfies
ky (t) − y (a)k ≤ (e γ(t−a) − 1)(ky (a)k + K /γ)
for all t ∈ [a, b].

(c) If f is a continuous function, then any Hermes solution y to Eq. (1)
is continuously differentiable on (a, b) and satisfies ẏ (t) = f (t, y (t))
∀t ∈ (a, b).
Filippov and Krasovskii solutions to Eq. (2)
It is assumed that the vector field g : Rn → Rn is Lebesgue

measurable and locally bounded
We replace the differential equation given in Eq. (2) by a suitable
differential inclusion:
ẋ(t) ∈ G (x(t)) (5)
By ordinary solution of Eq. (5) we mean any arc x : [a, b] → Rn
satisfiyng ẋ(t) ∈ G (x(t)) a.e. on the interval [a, b]
Filippov solutions to Eq. (2) are the ordinary solutions of Eq. (5)
obeying x(a) = x0 , where
\ \
G (x) = GF (x) = co{g (B(x, δ) \ N)}
δ>0 µ(N)=0
Krasovskii solutions of Eq. (2) are the ordinary solutions of Eq. (5)
obeying x(a) = x0 , where
\
G (x) = GK (x) = co{g (B(x, δ))}
δ>0
µ is the Lebesgue measure of Rn , co denotes the closure of the

convex hull, and B(x, r ) is the open ball of radius r centered at x
Since GF (x) ⊂ GK (x), every Filippov solution is a Krasovskii solution
The set of Filippov (respectively, Krasovskii) solutions of Eq. (2) will
be denoted by F (respectively, K)
We denote the sets of Euler, Hermes, and Carathodory solutions to
Eq. (2) respectively by E, H, and C.
Since g is measurable and locally bounded, then the set valued map
GF (x) is upper semicontinuous, locally bounded, compact and convex
valued. The same holds true for GK (x)
The Eq. (2) has a Filippov solution (and hence a Krasowskii solution)
on some interval [a, c] (Filippov (1988))
GK (x) = GF (x) = g (x) whenever the function g is continuous at x
F = K = C when the function g is continuous
K = H (Hajek (1979))
Sentis solutions to Eq. (2)
Sentis solutions to Eq. (2) will be defined as g -solutions of Eq. (5)

g -solutions are defined below
It is assumed that the vector field g : Rn → Rn is Lebesgue
measurable and locally bounded
We replace the Eq. (2) by a differential inclusion given in Eq. (5)
where \ \
G (x) = GS (x) = {g (B(x, δ) \ N)} (6)
δ>0 µ(N)=0
The set valued map GS (x) is upper semicontinuous, locally bounded

and compact (but in general not convex) valued
Let x0 be a fixed point of Rn . Given m ∈ N, consider a partition of [a, b]
a = tm,0 < tm,1 < · · · < tm,km −1 < tm,km = b
where km is some positive integer, and let
lm = max{tm,i+1 − tm,i , i = 0, . . . , km − 1}.
Then, for each i = 0, . . . , km − 1 take εm,i ∈ Rn and construct a piecewise

linear function ψm (t) on the interval [a, b] satisfying ψm (tm,0 ) = x0 , and
ψm (tm,i+1 ) = ψm (tm,i ) + vm,i (tm,i+1 − tm,i ) + εm,i
where vm,i ∈ G (ψm (tm,i )). The function ψm (t) will be called a polygonal
approximation.
Definition
A function ϕ : [a, b] → Rn is a g -solution of Eq. (5) if for each σ > 0
there exists an integer m and a polygonal approximation obeying
kϕ(t) − ψm (t)k < σ ∀t ∈ [a, b]

P m −1
0 < lm < σ and 0 ≤ ki=0 kεm,i k < σ.
By construction, ϕ(a) = x0
If G (x) is upper semicontinuous, compact valued and locally
bounded, then for each x0 there exist b > a and a g -solution
ϕ : [a, b] → Rn with ϕ(a) = x0 (Sentis (1978))
Ordinary solutions of Eq. (5), when they exists, are g -solutions

(Sentis (1978))
We say that a function ϕ(t) is a Sentis solution of Eq. (2) if it is a
g -solution of Eq. (5) with G (x) = GS (x)
The set of Sentis solutions of Eq. (2) will be denoted by S
Proposition (Sentis (1978))

Suppose that the function g : Rn → Rn is Lebesgue measurable and locally
bounded. Then every Sentis solution of Eq. (2) is a Filippov solution.
As a consequence from the relationships K = H and F ⊂ K, we have

the following lemma
Lemma
Suppose that the function g : Rn → Rn is Lebesgue measurable and
locally bounded. Then S ⊂ H.
Example
Consider the function f : R → R given by f (x) = 1 if x > 0, f (x) = −1 if
x < 0 and f (x) = 0 if x = 0. Suppose that a = 0, b = 1 and x0 = 0.
Then S = {t, −t}. On the other hand, x(t) = 0 is a Hermes solution.
Now consider the differential equation

3
ẋ(t) = x 1/3 , x(0) = 0 (7)
2
on the interval [0, 1]
The Eq. (7) has three distinct Carathodory solutions: x(t) = t 3/2 ,
x(t) = −t 3/2 and x(t) = 0
However the only Euler solution to Eq. (7) is x(t) = 0
Since g (x) = 32 x 1/3 is a continuous function, F = K = H = C
Therefore, even in the case where g is a continuous function, in
general a Hermes solution is not an Euler solution
Thus, we get the following proposition
Proposition
Suppose that the function g : Rn → Rn is continuous. Then
E ⊂ H = K = F = C.
References
A. Bacciotti. Generalized solutions of differential inclusions and stability, Ital. J.

Pure Appl. Math., 183–192, 2005.
A. Bacciotti. Some remarks on generalized solutions of discontinuous differential

equations, Int. J. Pure Appl. Math., 10:257–266, 2004.
D. A. R. da Silveira, Equações diferenciais descontı́nuas e inclusões diferenciais,

Master’s thesis, Universidade de Aveiro, 2009.
O. Hájek. Discontinuous differential equations, I, J. Differential Equations,

32:149–170, 1979.
R. Sentis. Équations différentielles à second membre mesurable, Boll. Un. Mat. Ital.
B (5), 15:724–742, 1978.
I. L. D. dos Santos. Discontinuous dynamic equations on time scales, Rend. Circ.

Mat. Palermo (2), 64:383–402, 2015.
References
F. H. Clarke, Yu. S. Ledyaev, R. J. Stern, and P. R. Wolenski. Nonsmooth analysis

and control theory. In Graduate Texts in Mathematics. Springer-Verlag, New York,
1998. ISSN: 0072-5285.
H. L. Royden. Real Analysis. The Macmillan Co., New York; Collier-Macmillan
Ltd., London, 1963.
J.-P. Aubin and A. Cellina. Differential Inclusions. Springer-Verlag, Berlin, 1984.

Set-valued maps and viability theory.
G. V. Smirnov. Introduction to the theory of differential inclusions. In Graduate

Studies in Mathematics. American Mathematical Society, Providence, RI, 2002.
ISSN: 1065-7339.
T. M. Apostol. Calculus. Vol. I: One-variable calculus, with an introduction to
linear algebra. Blaisdell Publishing Co. Ginn and Co., Waltham, Mass.-Toronto,
Ont.-London, 1967.
References
A. F. Filippov. Differential equations with discontinuous righthand sides, vol. 18 of

Mathematics and its Applications (Soviet Series). Kluwer Academic Publishers
Group, Dordrecht, 1988. Translated from the Russian.

On Discontinuous Differential Equations

Uploaded by

Copyright:

Available Formats

On Discontinuous Differential Equations

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

On Discontinuous Differential Equations

Uploaded by

Copyright:

Available Formats

On Discontinuous Differential Equations

Iguer Luis Domini dos Santos

Universidade Estadual Paulista

Discontinuous differential equations: ordinary differential equations

Consider the initial value problems

ẋ(t) = f (t, x(t)), x(a) = x0 (1)

The Banach space of Lebesgue integrable functions x : [a, b] → R with the

for all t1 , t2 ∈ [a, b], t1 < t2 .

kẋ(t)k ≤ γkx(t)k + c(t) a.e., t ∈ [a, b],

where γ is a nonnegative constant and where c(·) ∈ L1 ([a, b]) is a

If the function c is constant and γ > 0, this becomes

kx(t) − x(a)k ≤ (e γ(t−a) − 1)(kx(a)k + c/γ).

µπ := max{ti − ti−1 : 1 ≤ i ≤ N}.

xπ (t0 ) = x0 , xπ (t) = x0 + (t − t0 )f (t0 , x0 ), t ∈ [t0 , t1 ]

xπ (ti ) = xi , xπ (t) = xi + (t − ti )f (ti , xi ), t ∈ [ti , ti+1 ]

when i ∈ {0, 1, . . . , N − 1}.

Below we state the Theorem 1.7. (Clarke et al. (1998)).

kf (t, x)k ≤ γkxk + c,

where f is otherwise arbitrary. Then:

kx(t) − x(a)k ≤ (t − a)e γ(t−a) (c + γkx(a)k), a ≤ t ≤ b.

(c) If f is continuous, then any Euler arc x for f on [a, b] is continuously

Below we define Hermes solutions as dos Santos (2015).

ẏ (t) = f (t, y (t) + pj (t)), y (a) = x0 (3)

such that pj ⇒ 0 and xj ⇒ x.

Proposition (da Silveira (2009))

kf (t, x)k ≤ γkxk + c (4)

and let x0 = 0. Then x(t) = t is a Hermes solution to Eq. (1). Since in

ky (t) − y (a)k ≤ (e γ(t−a) − 1)(ky (a)k + K /γ)

for all t ∈ [a, b].

It is assumed that the vector field g : Rn → Rn is Lebesgue

µ is the Lebesgue measure of Rn , co denotes the closure of the

Sentis solutions to Eq. (2) will be defined as g -solutions of Eq. (5)

The set valued map GS (x) is upper semicontinuous, locally bounded

Let x0 be a fixed point of Rn . Given m ∈ N, consider a partition of [a, b]

a = tm,0 < tm,1 < · · · < tm,km −1 < tm,km = b

where km is some positive integer, and let

lm = max{tm,i+1 − tm,i , i = 0, . . . , km − 1}.

Then, for each i = 0, . . . , km − 1 take εm,i ∈ Rn and construct a piecewise

ψm (tm,i+1 ) = ψm (tm,i ) + vm,i (tm,i+1 − tm,i ) + εm,i

kϕ(t) − ψm (t)k < σ ∀t ∈ [a, b]

Ordinary solutions of Eq. (5), when they exists, are g -solutions

Proposition (Sentis (1978))

As a consequence from the relationships K = H and F ⊂ K, we have

Now consider the differential equation

Thus, we get the following proposition

A. Bacciotti. Generalized solutions of differential inclusions and stability, Ital. J.

A. Bacciotti. Some remarks on generalized solutions of discontinuous differential

D. A. R. da Silveira, Equações diferenciais descontı́nuas e inclusões diferenciais,

O. Hájek. Discontinuous differential equations, I, J. Differential Equations,

I. L. D. dos Santos. Discontinuous dynamic equations on time scales, Rend. Circ.

F. H. Clarke, Yu. S. Ledyaev, R. J. Stern, and P. R. Wolenski. Nonsmooth analysis

J.-P. Aubin and A. Cellina. Differential Inclusions. Springer-Verlag, Berlin, 1984.

G. V. Smirnov. Introduction to the theory of differential inclusions. In Graduate

A. F. Filippov. Differential equations with discontinuous righthand sides, vol. 18 of

You might also like