Generalized variational inequalities

1 2 3 4
Pavel Krejčı́ and Philippe Laurençot

Abstract

We consider a rate independent evolution variational inequality with an arbitrary

convex closed constraint Z in a Hilbert space X . The main results consist in proving
that it is well-posed in the Young integral setting in the space of functions of essentially
bounded variation for every Z and in the space of regulated functions provided 0 lies in
the interior of Z .

MSC 2000 : 34C55, 26A45, 49J40

Keywords: hysteresis, evolution variational inequality, Young integral, play operator

Introduction
We consider a real Hilbert space X endowed with a scalar product h·, ·i and norm |x| :=
hx, xi1/2 for x ∈ X . Throughout the paper we assume that

Z is a convex closed subset of X such that 0 ∈ Z . (0.1)

We mainly work with the so-called regulated functions (cf. [1]), that is, functions of real
variable which at each point of their domain of definition admit both finite one-sided limits,
see Definition 1.1 below. The space of regulated functions [0, T ] → X will be denoted by
G(0, T ; X) according to [21].
We assume that an initial condition x0 ∈ Z and an input u ∈ G(0, T ; X) are given, and we
look for a function ξ ∈ G(0, T ; X) such that

( P ) (i) u(t) − ξ(t) ∈ Z ∀t ∈ [0, T ] ,

(ii) u(0) − ξ(0) = x0 ,
(iii) for every y ∈ G(0, T ; Z) , the function τ 7→ u(τ +)−ξ(τ +)−y(τ ) is Young integrable
on [0, T ] with respect to ξ according to Definition 3.1, and
Z t
hu(τ +) − ξ(τ +) − y(τ ), dξ(τ )i ≥ 0 ∀t ∈ [0, T ] .
0
1
Mathematical Institute, Academy of Sciences of the Czech Republic, Žitná 25, CZ-11567 Praha 1, Czech
Republic, E-Mail: krejci@math.cas.cz.
2
This work has been done during the first author’s stay at the Weierstrass Institute for Applied Analysis
and Stochastics (WIAS) in Berlin under the support of the Deutsche Forschungsgemeinschaft (DFG), and at
the Institut Elie Cartan in Nancy.
3
Mathématiques pour l’Industrie et la Physique, UMR CNRS 5640, Université Paul Sabatier – Toulouse 3,
118 route de Narbonne, F-31062 Toulouse Cédex 4, France, E-Mail: laurencot@mip.ups-tlse.fr.
4
Partially supported by Procope, Project No. 98158.

1
If u is continuous, then Problem ( P ) can be stated either as a limit of classical variational
inequalities with smooth inputs, see [10, 22], or in the context of the Riemann-Stieltjes integral,
see [12, 5]. The solution operator (x0 , u) 7→ ξ called the play is one of the main building blocks
of the theory of hysteresis operators and its properties have been extensively studied in various
settings. If u is of bounded variation, Problem ( P ) can also be interpreted as a special case
of a sweeping process, see [16], defined as a limit of time-discrete approximations. Theorem
2.3 and Proposition 4.3 below not only illustrate this property, but also show that the time-
discrete approximations in the sense of [16] coincide with the exact solutions of Problem ( P )
for piecewise constant inputs.
The aim of this paper is to propose an extension of the play onto the space of regulated functions
via the Young integral in the form given in [8]. An alternative, which we however do not pursue
here, would be to use Kurzweil’s integral introduced in [14]. Although the Kurzweil integral
calculus is in general simpler, its main drawback in connection with Problem ( P ) consists in
the fact that one of the key lemmas (Lemma 3.3 below) which is nearly trivial for the Young
integral, does not hold for the Kurzweil integral, and the analysis would have to be restricted
to, say, left-continuous inputs.
Another approach can be found in [4] in the scalar case X = R : the rate independence
makes it possible to use directly the ‘continuous’ methods by ‘filling in’ the discontinuities with
segments traversed with an infinite speed. In the case dim X > 1 , this procedure turns out to
be trajectory-dependent which makes the analysis difficult even if we restrict ourselves to some
canonical (the shortest, say) trajectories filling in the jumps.
As the main results of this paper (Theorems 2.3, 2.4), we show that Problem ( P ) always
defines an input-output mapping pZ : Z × BV (0, T ; X) → BV (0, T ; X) : (x0 , u) 7→ ξ which
is continuous with respect to the uniform convergence. Here, BV (0, T ; X) denotes the space
of functions with essentially bounded variation, see (1.7) below. Moreover, if 0 ∈ Int Z , then
the output ξ is well-defined in BV (0, T ; X) for every u ∈ G(0, T ; X) , and the operator
pZ : Z × G(0, T ; X) → BV (0, T ; X) is continuous with respect to the uniform convergence. It
is interesting to note that inputs u1 , u2 which are equivalent in the sense that u1 (t−) = u2 (t−)
for every t ∈ [0, T ] , generate equivalent outputs ξ1 , ξ2 .
The paper is organized as follows. In order to fix the notation and to keep the presentation
consistent, we list in Section 1 the main concepts from convex analysis and vector-valued
functions that are used throughout the text. In Section 2 we state the main results. Section
3 is devoted to a self-contained extension of the Young integration theory to functions with
values in a Hilbert space. Detailed proofs of statements from Section 2 are given in Section 4.
In Section 5 we illustrate the connection between Problem ( P ) and the concept of ε -variation
introduced by Fraňková in [7].

Acknowledgement. The authors wish to thank J. Kurzweil, Š. Schwabik and M. Tvrdý for
stimulating suggestions and comments.

1 Preliminaries
The aim of this section is to recall some basic facts about the convex analysis in Hilbert spaces
and vector-valued functions of a real variable. Most of the results are well-known and we refer

2
the reader e. g. to the monographs [2, 17] for more information.
For a given convex closed set Z ⊂ X such that 0 ∈ Z we fix the number

ρ := dist (0, ∂Z) := inf {|z| ; z ∈ ∂Z} ≥ 0. (1.1)

It is clear that ρ > 0 if and only if 0 ∈ Int Z . In this case we have Bρ (0) ⊂ Z , where

Br (x0 ) := {x ∈ X ; |x − x0 | ≤ r} (1.2)

denotes the ball centered at x0 with radius r .

We introduce in the usual way the projection QZ : X → Z onto Z and its complement
P Z := I − QZ ( I is the identity) by the formula

QZ x ∈ Z , |P Z x| = dist (x, Z) for x ∈ X . (1.3)

In the sequel, we call (P Z , QZ ) the projection pair associated with Z . The projection can be
characterized by the variational inequality

q = QZ x ⇔ hx − q, q − zi ≥ 0 ∀z ∈ Z . (1.4)

Let now [a, b] ⊂ R be a nondegenerate closed interval. We denote by Da,b the set of all
partitions of the form

d = {t0 , . . . , tm } , a = t0 < t1 < . . . < tm = b .

For a given function g : [a, b] → X and a given partition d ∈ Da,b we define the variation
Vd (g) of g on d by the formula
m
X
Vd (g) := |g(tj ) − g(tj−1 )|
j=1

and the total variation Var [a,b] g of g by

Var g := sup{Vd (g) ; d ∈ Da,b } .

[a,b]

In a standard way (cf. [2]) we denote the set of functions of bounded variation by

BV (a, b ; X) := {g : [a, b] → X ; Var g < ∞} . (1.5)

[a,b]

Let us further introduce the set S(a, b ; X) of all step functions of the form
m
X m
X
w(t) := ĉk χ {tk } (t) + ck χ ]tk−1 ,tk [ (t) , t ∈ [a, b] , (1.6)
k=0 k=1

where d = {t0 , . . . , tm } ∈ Da,b is a given partition, χ A for A ⊂ [a, b] is the characteristic

function of the set A and ĉ0 , . . . , ĉm , c1 , . . . , cm are given elements from X .
It is well-known (see e. g. the Appendix of [2]) that every function of bounded variation with
values in a Banach space admits one-sided limits at each point of its domain of definition.
Following [1], we separate this property from the notion of total variation and introduce the
following definition.

3
Definition 1.1 We say that a function f : [a, b] → X is regulated if for every t ∈ [a, b] there
exist both one-sided limits f (t+), f (t−) ∈ X with the convention f (a−) = f (a) , f (b+) = f (b) .

According to [21], we denote by G(a, b ; X) the set of all regulated functions f : [a, b] → X .
For a given function g ∈ G(a, b ; X) and a given partition d ∈ Da,b we define the essential
variation Vd (g) of g on d by the formula
m
X m
X
Vd (g) := |g(tj −) − g(tj−1 +)| + |g(tj +) − g(tj −)|
j=1 j=0

and the total essential variation Var [a,b] g of g by

Var g := sup{Vd (g) ; d ∈ Da,b } .

[a,b]

We denote the space of functions of essentially bounded variation by

BV (a, b ; X) := {g : [a, b] → X ; Var g < ∞} . (1.7)

[a,b]

The terminology has been taken from [6], although we restrict ourselves a priori to regulated
functions which makes the analysis easier. This however means here in particular that Vd (g)
is defined for every function g : [a, b] → X , but Vd (g) only for a regulated function g .
We summarize some easy basic properties of the above spaces in Lemma 1.2 below the proof
of which is left to the reader.

Lemma 1.2

(i) Every regulated function is bounded.

(ii) For every g ∈ G(a, b ; X) we have Var [a,b] g ≤ Var [a,b] g .

(iii) The sets S(a, b ; X) , BV (a, b ; X) , BV (a, b ; X) , G(a, b ; X) are vector spaces satisfying
the inclusion

S(a, b ; X) ⊂ BV (a, b ; X) ⊂ BV (a, b ; X) ⊂ G(a, b ; X) .

(iv) Let GL (a, b ; X) be the subset of left-continuous functions in G(a, b ; X) . Then we have
GL (a, b ; X) ∩ BV (a, b ; X) ⊂ BV (a, b ; X) .

We introduce in G(a, b ; X) a system of seminorms

kf k[s,t] := sup{|f (τ )| ; τ ∈ [s, t]} (1.8)

for any subinterval [s, t] ⊂ [a, b] . Indeed, k·k[a,b] is a norm.

Let us note that the space C(a, b ; X) of continuous functions f : [a, b] → X is a closed
subspace of G(a, b ; X) with respect to the norm k·k[a,b] .
We now list some characteristic properties of regulated functions which are needed in the sequel.

4
Proposition 1.3

(i) The space G(a, b ; X) is complete with respect to the norm k·k[a,b] .

(ii) Given C > 0 , the set VC := {g ∈ BV (a, b ; X) ; Var [a,b] g ≤ C} is closed in G(a, b ; X) .
(iii) For f ∈ G(a, b ; X) and t ∈ [a, b] put

osc f (t) := max{|f (t) − f (t−)| , |f (t+) − f (t)| , |f (t+) − f (t−)|} .

Then for every ε > 0 the set

Ufε := {t ∈ [a, b] ; osc f (t) ≥ ε} (1.9)

is finite, and f is continuous except in a countable number of points.

(iv) Let f ∈ G(a, b ; X) and ε > 0 be given and let Ufε be the set defined by (1.9). Then
there exists h > 0 such that for every [s, t] ⊂ [a, b] , [s, t] ∩ Ufε = ∅ , 0 < t − s < h we
have |f (t) − f (s)| < ε .
(v) For every f ∈ G(a, b ; X) and ε > 0 there exists w ∈ S(a, b ; X) such that kf − wk[a,b] ≤
ε , w(t) ∈ ∪τ ∈[a,b] {f (τ )} for every t ∈ [a, b] , Var [a,b] w ≤ Var [a,b] f and Var [a,b] w ≤
Var [a,b] f .
(vi) Let f ∈ G(a, b ; X) be such that
1
 
∃C > 0 ∀h ∈ 0, (b − a) : Var f ≤ C.
2 [a+h, b−h]

Then f ∈ BV (a, b ; X) and Var [a,b] f ≤ C + |f (a+) − f (a)| + |f (b) − f (b−)| .

Proof.
(i) Let {fn } be a Cauchy sequence in G(a, b ; X) . For t ∈ [a, b] put f (t) := limn→∞ fn (t) .
For every t ∈ ]a, b] and every sequence tk % t we have

|f (tk ) − f (t` )| ≤ |f (tk ) − fn (tk )| + |fn (tk ) − fn (t` )| + |f (t` ) − fn (t` )| ,

hence {f (tk ) ; k ∈ N} is a Cauchy sequence whose limit is independent of the choice of the
sequence tk , and we conclude that f (t−) exists. In the same way we check that f (t+) exists
for t ∈ [a, b[ .
(ii) Let {gn } be a sequence in VC which converges uniformly to g in [a, b] . Then, for d ∈ Da,b ,
Vd (gn ) converge to Vd (g) , hence (ii).
(iii) For every t ∈ [a, b] there exists δ(t) > 0 such that |f (τ ) − f (t+)| < ε/4 for τ ∈
]t, t + δ(t)[ ∩[a, b] , |f (τ ) − f (t−)| < ε/4 for τ ∈ ]t − δ(t), t[ ∩[a, b] . In particular,
 
]t − δ(t), t + δ(t)[ \ {t} ∩ Ufε = ∅ ∀ t ∈ [a, b] .

As [a, b] is compact, we can select from the covering

[
[a, b] ⊂ ]t − δ(t), t + δ(t)[
t∈[a,b]

5
a finite covering
n
[
[a, b] ⊂ ]tj − δ(tj ), tj + δ(tj )[ ,
j=1

hence Ufε ⊂ {t1 , . . . , tn } .

(iv) Let Ufε = {t1 , . . . , tn } , tj+1 > tj for j = 1, . . . , n − 1 , and put I0 := [a, t1 [ , In := ]tn , b]
(which might possibly be empty), Ij := ]tj , tj+1 [ for j = 1, . . . , n − 1 . Let fj , j = 0, . . . , n be
the functions f |Ij continuously extended to Ij . The assertion holds provided we put
n o
h := inf |t − s| ; [s, t] ⊂ Ij , |fj (t) − fj (s)| ≥ ε , j = 0, . . . , n ,
and we easily check that h > 0 .
(v) Let ε > 0 be given and let Ufε , h > 0 and Ij , j = 0, . . . , n be as in the proof of (iv). In
`
each interval Ij we find a partition tj = s0j < s1j < . . . < sjj = tj+1 (with the convention t0 = a ,
` −1
tn+1 = b ) such that s1j , . . . , sjj are continuity points of f , skj − sk−1
j < h for k = 1, . . . , `j .
For j = 0, . . . , n we now put
w(tj ) := f (tj )

 f (sk )
j for t ∈ ]sk−1
j , skj ] , k = 1, . . . , `j − 2 ,
w(t) :=  ` −1 ` −2
f (sjj ) for t ∈ ]sjj , tj+1 [ .
Then w ∈ S(a, b ; X) and from (iii) we immediately obtain that |f (t) − w(t)| ≤ ε for every
t ∈ [a, b] . Moreover, putting
 
n n
` −1 
o
s1j , . . . , sjj
[
d := {a} ∪ {b} ∪ {t1 , . . . , tn } ∪  ∈ Da,b
j=0

we have Var [a,b] w = Vd (f ) , Var [a,b] w ≤ Vd (f ) , and the assertion follows.

(vi) Let d = {t0 , . . . , tm } ∈ Da,b be an arbitrary partition. For 0 < h < min{t1 − a, b − tm−1 }
we have
Vd (f ) ≤ Var f + |f (a+)−f (a)| + |f (a+h)−f (a+)| + |f (b)−f (b−)| + |f (b−)−f (b−h)| ,
[a+h, b−h]

and letting h tend to 0+ we obtain the assertion. The proof of Proposition 1.3 is complete.


2 Main results
We will see in Section 3 that the integral in Problem ( P ) is meaningful if ξ ∈ BV (0, T ; X) .
Let us denote by Dom (P) ⊂ Z × G(0, T ; X) the set of all (x0 , u) ∈ Z × G(0, T ; X) such that
there exists a solution ξ ∈ BV (0, T ; X) to Problem ( P ) . We first show that the solution ξ
is unique for every (x0 , u) ∈ Dom (P) . In fact, we prove more, namely

Lemma 2.1 Let (x0 , u), (y0 , v) ∈ Dom (P) be given and let ξ, η ∈ BV (0, T ; X) be respective
solutions to Problem ( P ) . Then for every t ∈ [a, b] we have
|ξ(t) − η(t)|2 ≤ |ξ(0) − η(0)|2 + 2 ku − vk[0,t] Var (ξ − η) . (2.1)
[0,t]

6
Proof. Putting y(τ ) := (1/2) (u(τ +) + v(τ +) − ξ(τ +) − η(τ +)) in the inequalities ( P ) (iii)
for ξ and for η and summing them up we obtain
Z t
hu(τ +) − v(τ +) − ξ(τ +) + η(τ +), d(ξ − η)(τ )i ≥ 0 .
0

Corollary 3.15 then yields that

Z t
1 
|ξ(t) − η(t)|2 − |ξ(0) − η(0)|2 ≤ hu(τ +) − v(τ +), d(ξ − η)(τ )i
2 0

and (2.1) follows from (3.16). 

Lemma 2.1 enables us to define the operator

pZ : Dom (P) → BV (0, T ; X) : (x0 , u) 7→ ξ , (2.2)
where ξ =: pZ [x0 , u] is the unique solution to Problem ( P ) .
We summarize the main results of this paper as Proposition 2.2 and Theorems 2.3 – 2.4 below.
The proofs will be given in Section 4.

Proposition 2.2 Let x0 ∈ Z be given, and for u ∈ G(0, T ; X) , t ∈ [0, T ] put u− (t) := u(t−) .
Then (x0 , u) ∈ Dom (P) if and only if (x0 , u− ) ∈ Dom (P) , and in this case we have
pZ [x0 , u](t−) = pZ [x0 , u− ](t) , pZ [x0 , u](t) − pZ [x0 , u](t−) = P Z (u(t) − pZ [x0 , u](t−)) (2.3)
for every t ∈ [0, T ] , with P Z defined by (1.3).

Theorem 2.3 For each set Z satisfying (0.1) we have Z × BV (0, T ; X) ⊂ Dom (P) and
Var [0,T ] pZ [x0 , u] ≤ Var [0,T ] u for every (x0 , u) ∈ Z × BV (0, T ; X) . Moreover, for every
(x0 , u), (y0 , v) ∈ Z × BV (0, T ; X) , ξ = pZ [x0 , u] , η = pZ [y0 , v] and every t ∈ [0, T ] we
have !
|ξ(t) − η(t)|2 ≤ |ξ(0) − η(0)|2 + 2 ku − vk[0,t] Var u + Var v . (2.4)
[0,t] [0,t]

Theorem 2.4 If 0 ∈ Int Z , then Dom (P) = Z ×G(0, T ; X) . Moreover, if {(xn0 , un ) ; n ∈ N}

is a sequence in Z × G(0, T ; X) such that |xn0 − x0 | → 0 , kun − uk[0,T ] → 0 as n → ∞ ,
then there exists a constant C > 0 independent of n such that Var [0,T ] pZ [xn0 , un ] ≤ C and
kpZ [xn0 , un ] − pZ [x0 , u]k[0,T ] → 0 .

Remark 2.5 We cannot replace the variational inequality (iii) in Problem ( P ) by

Z t
(iii)’ hu(τ ) − ξ(τ ) − y(τ ), dξ(τ )i ≥ 0 ∀y ∈ G(0, T ; Z) ∀t ∈ [0, T ]
0

which might seem to be a natural extension of the continuous case in [12]. It suffices to consider
the scalar case X = R , Z = [−r, r] for some r > 0 , x0 = 0 , u(τ ) = ū χ ]0,T ] (τ ) with some
ū > r . Assume that there exists ξ ∈ BV (0, T ; X) satisfying (iii)’, ku − ξk[0,T ] ≤ r , ξ(0) = 0 .
Putting y(τ ) := r χ {0} (τ ) + (u(τ ) − ξ(τ )) χ ]0,T ] (τ ) we obtain from (iii)’ and Proposition 3.7
that Z t
0 ≤ (u(τ ) − ξ(τ ) − r) χ {0} (τ ) dξ(τ ) = −r ξ(0+) ,
0
hence ξ(0+) ≤ 0 and u(0+) − ξ(0+) ≥ ū > r , which is a contradiction.

7
3 The Young integral
We give here a survey of those elements of the Young integral calculus that are related to
Problem ( P ) using the ideas of [8, 19, 20, 21].
We fix a compact interval [a, b] ⊂ R and as in Section 1, we denote by Da,b the set of all
partitions d = {t0 , . . . , tm } , a = t0 < t1 < . . . < tm = b of [a, b] .
We say that a partition dˆ is a refinement of d ∈ Da,b and write dˆ  d if dˆ ∈ Da,b and d ⊂ dˆ.
Let d = {t0 , . . . , tm } ∈ Da,b be a partition. We denote by B(d) the set of special refinements
D of d (the so-called P -partitions, see [19]), of the form
D = {t0 , %1 , t1 , . . . , tm−1 , %m , tm } , a = t0 < %1 < t1 < %2 < t2 < . . . tm−1 < %m < tm = b . (3.1)

For given functions f : [a, b] → X , g ∈ G(a, b ; X) and partitions d ∈ Da,b , D ∈ B(d) of the
form (3.1) we define the integral sum SD (f ∆g) by the formula
m
X m
X
SD (f ∆g) = hf (%j ), g(tj −) − g(tj−1 +)i + hf (tj ), g(tj +) − g(tj −)i , (3.2)
j=1 j=0

again with the convention g(a−) = g(a) , g(b+) = g(b) .

Definition 3.1 We say that J ∈ R is the Young integral over [a, b] of f with respect to g
and denote Z b
J = hf (t), dg(t)i , (3.3)
a
if for every ε > 0 there exists dε ∈ Da,b such that for every d  dε and D ∈ B(d) we have
|J − SD (f ∆g)| ≤ ε . (3.4)

It is an easy exercise to check that if the value JR in Definition 3.1 exists, then it is uniquely
determined. In what follows, whenever we write ab hf (t), dg(t)i = J , we interpret it as ‘the
function f is Young integrable with respect to g in [a, b] and the integral equals to J .’
Similarly as for other integration theories (cf. e. g. [19] for the Perron-Stieltjes or Kurzweil
integral), the Young integral admits the following ‘Bolzano-Cauchy-type’ characterization.

Lemma 3.2 Consider f : [a, b] → X and g ∈ G(a, b ; X) . Then f is Young integrable with
respect to g in [a, b] if and only if for every ε > 0 there exists dε ∈ Da,b such that for every
di  dε and Di ∈ B(di ) , i = 1, 2 we have
|SD1 (f ∆g) − SD2 (f ∆g)| ≤ ε . (3.5)

Proof. If ab hf (t), dg(t)i exists, then (3.5) obviously holds. Conversely, let (3.5) hold. For
R

every n ∈ N we find a partition d1/n ∈ Da,b such that (3.5) holds with ε = 1/n , and put
d∗1 := d1 , d∗n := d∗n−1 ∪ d1/n for n = 2, 3, . . . . For each n ∈ N we fix some Dn ∈ B(d∗n ) and put
Jn := SDn (f ∆g) . By (3.5), {Jn } is a Cauchy sequence in R , and putting J := limn→∞ Jn we
easily check that J = ab hf (t), dg(t)i by Definition 3.1.
R


The reason we decided for the Young integral is its following property which is an immediate
consequence of the definition and which plays an important role in our arguments. Surprisingly
enough, identity (3.6) does not hold for the Kurzweil integral in general [13].

8
Lemma 3.3 Consider f : [a, b] → X and g ∈ G(a, b ; X) , and assume that there exists a
countable set A ⊂ [a, b] such that g(t) = 0 for every t ∈ [a, b] \ A . Then we have
Z b
hf (t), dg(t)i = hf (b), g(b)i − hf (a), g(a)i . (3.6)
a

The Young integral is linear with respect to both functions f and g . For the sake of complete-
ness, we state this result explicitly.

Proposition 3.4
Rb Rb
(i) Let a hf1 (t), dg(t)i , a hf2 (t), dg(t)i exist. Then we have
Z b Z b Z b
h(f1 + f2 )(t), dg(t)i = hf1 (t), dg(t)i + hf2 (t), dg(t)i . (3.7)
a a a

Rb Rb
(ii) Let a hf (t), dg1 (t)i , a hf (t), dg2 (t)i exist. Then we have
Z b Z b Z b
hf (t), d(g1 + g2 )(t)i = hf (t), dg1 (t)i + hf (t), dg2 (t)i . (3.8)
a a a

Rb
(iii) Let a hf (t), dg(t)i exist. Then for every constant λ ∈ R we have
Z b Z b Z b
hλf (t), dg(t)i = hf (t), d(λg)(t)i = λ hf (t), dg(t)i . (3.9)
a a a

Proof. (i) Let ε > 0 be given. We find d1ε/2 , d2ε/2 ∈ Da,b such that for all di  diε/2 ,
Di ∈ B(di ) , i = 1, 2 we have
Z b ε
hfi (t), dg(t)i − SDi (fi ∆g) < . (3.10)
a 2

Putting dε := d1ε/2 ∪ d2ε/2 we obtain (3.7) immediately from (3.10). The same argument applies
to the case (ii), while (iii) is obvious. 

The Young integral behaves in the following way with respect to the variation of the integration
domain.

Proposition 3.5 Let f : [a, b] → X , g ∈ G(a, b ; X) be given functions and let [r, s] ⊂ [a, b]
be a nondegenerate interval.
Rb Rs
(i) Assume that a hf (t), dg(t)i exists. Then r hf (t), dg(t)i exists.
Rs
(ii) Assume that r hf (t), dg(t)i exists. Then we have
Z b D  E Z s
f χ ]r,s[ (t), dg(t) = hf (t), dg(t)i − hf (r), g(r+) − g(r)i − hf (s), g(s) − g(s−)i .
a r
(3.11)

9
Remark 3.6 Proposition 3.5 needs some comment. Here and in the sequel, whenever we
integrate functions f, g defined in [a, b] over an interval [r, s] ⊂ [a, b] , we implicitly consider
their restrictions f |[r,s] , g|[r,s] . In particular, we have e. g. f |[r,s] (s+) = f (s) , f |[r,s] (r−) = f (r) ,
similarly as in Definition 1.1.

Proof of Proposition 3.5.

(i) Let ε > 0 be given. By Lemma 3.2 we find dε ∈ Da,b such that for every di  dε and
Di ∈ B(di ) , i = 1, 2 we have

|SD1 (f ∆g) − SD2 (f ∆g)| < ε ,

and put d∗ε := (dε ∩ ]r, s[ ) ∪ {r} ∪ {s} . Then d∗ε ∈ Dr,s , and we arbitrarily fix di  d∗ε and
Di ∈ B(di ) , i = 1, 2 . Put dˆi := di ∪ dε . Then dˆ1 , dˆ2 may be written in the form

dˆ1 = {a = t0 < t1 < . . . < tk = r < t1k+1 < . . . < s = t1m1 −` < . . . < t1m1 = b} ,
dˆ2 = {a = t0 < t1 < . . . < tk = r < t2k+1 < . . . < s = t2m2 −` < . . . < t2m2 = b} ,

where t1m1 −j = t2m2 −j for j = 0, . . . , ` . We now fix arbitrary %i ∈ ]ti−1 , ti [ for i = 1, . . . , k and
%̂j ∈ ]t1m1 −j , t1m1 −j+1 [ for j = 1, . . . , ` , and put
n o
D̂1 = t0 , %1 , t1 , . . . , tk−1 , %k , D1 , %̂` , t1m1 −`+1 , . . . , t1m1 −1 , %̂1 , t1m1 ,
n o
D̂2 = t0 , %1 , t1 , . . . , tk−1 , %k , D2 , %̂` , t2m2 −`+1 , . . . , t2m2 −1 , %̂1 , t2m2 .

Then D̂i ∈ B(dˆi ) , i = 1, 2 , and we have

SD1 (f ∆g) − SD2 (f ∆g) = SD̂1 (f ∆g) − SD̂2 (f ∆g) ,

hence the assertion follows from Lemma 3.2.

(ii) Let ε > 0 be given and let dε = {t0 , . . . , tm } ∈ Dr,s be such that for every d  dε and
D ∈ B(d) we have Z s
hf (t), dg(t)i − SD (f ∆g) < ε . (3.12)
r

Put d∗ε := dε ∪ {a, b} . Then d∗ε ∈ Da,b , and every D∗ ∈ B(d∗ ) with d∗  d∗ε is of the form

D∗ = {t∗0 , %∗1 , t∗1 , . . . , t∗m∗ −1 , %∗m∗ , t∗m∗ } ,

where r = t∗i , s = t∗k for some 0 ≤ i < k ≤ m∗ . By construction, d := {t∗i , t∗i+1 , . . . , t∗k } is a
refinement of dε and D := {t∗i , %∗i+1 , t∗i+1 , . . . , %∗k , t∗k } belongs to B(d) . On the other hand, for
j ≤ i and j > k we have f χ ]r,s[ (%∗j ) = 0 , hence
  
SD ∗ f χ ]r,s[ ∆g = SD (f ∆g) − hf (r), g(r+) − g(r)i − hf (s), g(s) − g(s−)i ,

and the assertion follows. 

We next investigate some typical cases.

Proposition 3.7 For every f : [a, b] → X , g ∈ G(a, b ; X) , a ≤ r ≤ b and v ∈ X we have

10
 Z b D E
(i) v χ {r} (t), dg(t) = hv, g(r+) − g(r−)i ,
a

 0 if r ∈ ]a, b[ ,
b
Z D   E 

(ii) f (t), d v χ {r} (t) = −hf (a), vi if r = a ,
a 
hf (b), vi


if r = b ,
Z b D E
(iii) v χ ]r,s[ (t), dg(t) = hv, g(s−) − g(r+)i ∀ s ∈ ]r, b] ,
a
Z b D   E
(iv) f (t), d v χ ]r,s[ (t) = hf (r) − f (s), vi ∀ s ∈ ]r, b] .
a

Proof. Identity (iii) follows immediately

Rs
from Proposition 3.5 for f (t) ≡ v and from the
obvious fact (cf. Remark 3.6) that r hv, dg(t)i = hv, g(s) − g(r)i . For a < r < b we obtain
(i) from (iii) and from Proposition 3.4 using the formula χ {r} = χ ]a,b[ − χ ]a,r[ − χ ]r,b[ . To
treat the case r = a , we just note that for each d ∈ Da,b and D ∈ B(d) , the integral sum
SD (v χ {a} ∆g) contains only one nonzero term, namely hv, g(a+) − g(a)i . For r = b we argue
analogously. Statement (ii) is a trivial consequence of Lemma 3.3.
To prove (iv), we set d0 := {a, r, s, b} . Then for every d  d0 and D ∈ B(d) of the form
(3.1) we have r = ti , s = tk for some 0 ≤ i < k ≤ m , where χ ]r,s[ (tj −) = χ ]r,s[ (tj−1 +) for
every j = 1, . . . , m , χ ]r,s[ (tj −) = χ ]r,s[ (tj +) for every j 6= i, k , χ ]r,s[ (ti +) − χ ]r,s[ (ti −) = 1 ,
  
χ ]r,s[ (tk +) − χ ]r,s[ (tk −) = −1 , hence SD f ∆ v χ ]r,s[ = hf (r) − f (s), vi , and Proposition
3.7 is proved. 

Rs
Corollary
Rb
3.8 Let f : [a, b] → X , g ∈ G(a, b ; X) and s ∈ ]a, b[ be such that a hf (t), dg(t)i ,
s hf (t), dg(t)i exist. Then we have
Z b Z s Z b
hf (t), dg(t)i = hf (t), dg(t)i + hf (t), dg(t)i . (3.13)
a a s

Proof. We clearly have f = f χ ]a,s[ + f χ ]s,b[ + f χ {a} + f χ {s} + f χ {b} . By Propositions 3.5
and 3.7 (i), each of these five functions is Young integrable with respect to g in [a, b] . Owing
to Proposition 3.4, f is Young integrable with respect to g in [a, b] , and (3.13) readily follows
from (3.11), Proposition 3.7 (i), and the identity
Z s Z b Z b D   E
hf (t), dg(t)i + hf (t), dg(t)i = f χ]a,s[ + χ]s,b[ (t), dg(t)
a s a
+ hf (a), g(a+) − g(a)i + hf (s), g(s+) − g(s−)i + hf (b), g(b) − g(b−)i
Z b D   E Z b
= f χ]a,s[ + χ]s,b[ + χ{a} + χ{s} + χ{b} (t), dg(t) = hf (t), dg(t)i .
a a

In order to preserve the consistency of (3.13) also in the limit cases s = a and s = b , we set
Z s
hf (t), dg(t)i = 0 ∀s ∈ [a, b] , ∀ f, g : [a, b] → X . (3.14)
s

11
Propositions 3.4 and 3.7 enable us to evaluate the integral ab hf (t), dg(t)i provided one of the
R

functions f , g belongs to S(a, b ; X) . The next strategy consists in exploiting the density of
S(a, b ; X) in G(a, b ; X) stated in Proposition 1.3 (iv). We first notice that for all functions
f, g : [a, b] → X and every P -partition D of the form (3.1) we have
m
X m
X
SD (f ∆g) = hf (%j ), g(tj −) − g(tj−1 +)i + hf (tj ), g(tj +) − g(tj −)i
j=1 j=0

= hf (b), g(b)i − hf (a), g(a)i

m
X m−1
X
+ hf (%j ) − f (tj ), g(tj −)i − hf (%j+1 ) − f (tj ), g(tj +)i ,
j=1 j=0

hence n o
|SD (f ∆g)| ≤ min kf k[a,b] Vd (g) , (|f (a)| + |f (b)| + Vd (f )) kgk[a,b] . (3.15)

The extension of the Young integral to G(a, b ; X) is based on Theorem 3.9 below.

Theorem 3.9 Consider f, fn : [a, b] → X , n ∈ N such that limn→∞ kf − fn k[a,b] = 0 . Then

the following implications hold.
Z b Z b
(i) If g ∈ BV (a, b ; X) and hfn (t), dg(t)i exists for each n ∈ N , then hf (t), dg(t)i
a a
exists and Z b Z b
hf (t), dg(t)i = lim hfn (t), dg(t)i .
a n→∞ a

Z b Z b
(ii) If g ∈ BV (a, b ; X) and hg(t), dfn (t)i exists for each n ∈ N , then hg(t), df (t)i
a a
exists and Z b Z b
hg(t), df (t)i = lim hg(t), dfn (t)i .
a n→∞ a

Proof.
(i) For n ∈ N put Jn := ab hfn (t), dg(t)i . For each n we find dn ∈ Da,b such that for every
R

d  dn and D ∈ B(d) we have

1
|SD (fn ∆g) − Jn | < .
n
For m, n ∈ N put dmn := dn ∪ dm . For every d  dmn and D ∈ B(d) we then have
1 1
|SD (fn ∆g) − Jn | < , |SD (fm ∆g) − Jm | < ,
n m
and (3.15) with f := fn − fm , g := g implies that

|Jn − Jm | ≤ |SD (fn ∆g) − Jn − SD (fm ∆g) + Jm | + |SD ((fn − fm ) ∆g)|

1 1
≤ + + kfn − fm k[a,b] Var g ,
m n [a,b]

12
hence {Jn } is a Cauchy sequence and we may put J := limn→∞ Jn . For each d  dn and
D ∈ B(d) we obtain that

|SD (f ∆g) − J| ≤ |SD ((f − fn ) ∆g)| + |SD (fn ∆g) − Jn | + |Jn − J|

≤ kf − fn k[a,b] Var g + 1/n + |Jn − J| ,
[a,b]

Rb
hence a hf (t), dg(t)i = J and (i) is proved.
The same argument based on (3.15) with f := g , g := fn − fm yields (ii). 

From Propositions 3.4 and 3.7 it follows that the integral ab hf (t), dg(t)i exists whenever one
R

of the functions f , g belongs to S(a, b ; X) . Using the fact that every regulated function can
be uniformly approximated by step functions (cf. Proposition 1.3 (v)), we obtain the following
result as an immediate consequence of Theorem 3.9 and inequality (3.15).

Corollary 3.10 If either

Rb
f ∈ G(a, b ; X) and g ∈ BV (a, b ; X) , or f ∈ BV (a, b ; X) and
g ∈ G(a, b ; X) , then a hf (t), dg(t)i exists and satisfies the estimate
( ! )
Z b
hf (t), dg(t)i ≤ min kf k[a,b] Var g , |f (a)| + |f (b)| + Var f kgk[a,b] . (3.16)
a [a,b] [a,b]

The estimate (3.16) is optimal in the following sense.

Theorem 3.11 For every g ∈ BV (a, b ; X) we have

(Z )
b
|g(a)| + |g(b)| + Var g = sup hg(t), df (t)i ; f ∈ S(a, b ; B1 (0)) . (3.17)
[a,b] a

For every g ∈ BV (a, b ; X) we have

(Z )
b
Var g = sup hf (t), dg(t)i ; f ∈ S(a, b ; B1 (0)) . (3.18)
[a,b] a

Proof. To prove (3.17), we fix ε > 0 and find a partition d = {t0 , . . . , tm } ∈ Da,b such that
m
X
|g(tj ) − g(tj−1 )| ≥ Var g − ε . (3.19)
[a,b]
j=1

Let σ : X → X be the function

σ(x) = x/|x| for x 6= 0 , σ(0) = 0 ,

and for t ∈ [a, b] put

m
X
f1 (t) := σ(g(b)) χ {b} (t) − σ(g(a)) χ {a} (t) − σ(g(tj ) − g(tj−1 )) χ ]tj−1 ,tj [ (t) . (3.20)
j=1

13
We then infer from (3.19), Proposition 3.7 (ii), (iv) and Proposition 3.4 (ii) that
Z b m
X
hg(t), df1 (t)i = |g(a)| + |g(b)| + |g(tj ) − g(tj−1 )| ≥ |g(a)| + |g(b)| + Var g − ε , (3.21)
a [a,b]
j=1

which, together with Corollary 3.10, yields (3.17).

We now turn to the proof of (3.18) and consider g ∈ BV (a, b ; X) and ε > 0 . We find again a
partition d = {t0 , . . . , tm } ∈ Da,b such that
m
X m
X
|g(tj −) − g(tj−1 +)| + |g(tj +) − g(tj −)| ≥ Var g − ε , (3.22)
[a,b]
j=1 j=0

and put
m
X m
X
f2 (t) := σ(g(tj −) − g(tj−1 +)) χ ]tj−1 ,tj [ (t) + σ(g(tj +) − g(tj −)) χ {tj } (t) . (3.23)
j=1 j=0

Then f2 ∈ S(a, b ; B1 (0)) and it follows from Propositions 3.4 and 3.7 (i), (iii) that
Z b m
X m
X
hf2 (t), dg(t)i = |g(tj −) − g(tj−1 +)| + |g(tj +) − g(tj −)| . (3.24)
a j=1 j=0

The assertion (3.18) is then a consequence of (3.22), (3.24), and Corollary 3.10. 

As an easy extension of Theorem 3.9 we have the following convergence result.

Proposition 3.12 Consider f, fn ∈ G(a, b ; X) , g, gn ∈ BV (a, b ; X) , n ∈ N such that

lim kf − fn k[a,b] = 0 ,
n→∞
lim kg − gn k[a,b] = 0 ,
n→∞
sup Var gn = C < ∞ .
n∈N [a,b]

Then Z b Z b
hf (t), dg(t)i = lim hfn (t), dgn (t)i . (3.25)
a n→∞ a

Proof. For any w ∈ S(a, b ; X) we have by Corollary 3.10 that

Z b Z b Z b
hf (t), dg(t)i − hfn (t), dgn (t)i ≤ h(f − fn )(t), dgn (t)i
a a a
Z b Z b
+ h(f − w)(t), d(g − gn )(t)i + hw(t), d(g − gn )(t)i
a a
!
≤ C kf − fn k[a,b] + 2 C kf − wk[a,b] + 2 kwk[a,b] + Var w kg − gn k[a,b]
[a,b]

and the assertion follows from Proposition 1.3 (v). 

Example 3.13

14
 (i) Notice that the pointwise convergence gn (t) → g(t) for every t ∈ [a, b] is not sufficient in
Proposition 3.12 as in the case of the Riemann-Stieltjes integral. In the example X = R ,

fn (t) = f (t) = χ {0} (t) , g(t) ≡ 0 , gn (t) = χ ]0,1/n[ (t) for t ∈ [0, 1] (3.26)

we have 01 fn (t) dgn (t) = 1 for every n ∈ N ,

R R1
0 f (t) dg(t) = 0 , hence the assertion of
Proposition 3.12 does not hold.

(ii) Similarly, the pointwise convergence of {fn } is not sufficient for Proposition 3.12 to hold.
Indeed, putting
n
(−1)k−1 χ{k/n2 } (t)
X
fn (t) := for t ∈ [0, 1] , (3.27)
k=1
   h h
1 k−1 k


2n
(−1)k + 1 for t ∈ ,
n2 n2
, k = 1, . . . , n ,
gn (t) := h i (3.28)
1
 0

for t ∈ n
,1 ,

for n ∈ N , we see that fn , gn ∈ S(a, b ; X) , kfn k[a,b] ≤ 1 , Var [a,b] gn ≤ 1 , kgn k[a,b] → 0
and fn (t) → 0 for every t ∈ [a, b] as n → ∞ , while ab fn (t) dgn (t) → 1 .
R

The uniform convergence of fn towards f in Proposition 3.12 can however be relaxed, as we will
see in Section 5, Proposition 5.4. To conclude this section, we derive two integration-by-parts
formulas.

Theorem 3.14 For every f ∈ G(a, b ; X) , g ∈ BV (a, b ; X) we have

Z b Z b
hf (t), dg(t)i + hg(t), df (t)i = hf (b), g(b)i − hf (a), g(a)i (3.29)
a a
X  
+ hf (t) − f (t−) , g(t) − g(t−)i − hf (t+) − f (t) , g(t+) − g(t)i .
t∈[a,b]

Proof. From Proposition 1.3 (ii) it follows that the sum on the right-hand side of (3.29)
is at most countable, hence the formula is meaningful. Using Proposition 3.7 we check in a
straightforward way that (3.29) holds for every g ∈ BV (a, b ; X) whenever f is of the form
v χ {r} or v χ ]r,s[ , hence also for every f ∈ S(a, b ; X) by Proposition 3.4. For f ∈ G(a, b ; X)
and n ∈ N we find fn ∈ S(a, b ; X) such that kf − fn k[a,b] → 0 as n → ∞ and pass to the
limit using Theorem 3.9 and the obvious inequality
X  
|g(t) − g(t−)| + |g(t+) − g(t)| ≤ Var g .
[a,b]
t∈[a,b]

Corollary 3.15 For every g ∈ BV (a, b ; X) we have

Z b 1   1 X
hg(t+), dg(t)i = |g(b)|2 − |g(a)|2 + |g(t+) − g(t−)|2 . (3.30)
a 2 2 t∈[a,b]

15
Proof. The function g+ (t) := g(t+) satisfies g+ (t+) = g(t+) = g+ (t) for every t ∈ [a, b] ,
g+ (t−) = g(t−) for every t ∈ ]a, b] , and belongs to BV (a, b ; X) . By Theorem 3.14 we have
Z b 1   1 X
hg+ (t), dg+ (t)i = |g(b)|2 − |g(a+)|2 + |g(t+) − g(t−)|2
a 2 2 t∈ ]a,b]

(note that the sum is taken over the semi-open interval ]a, b] ), while (3.6) yields that
Z b
hg+ (t), d(g − g+ )(t)i = hg(a+), g(a+) − g(a)i .
a

Combining the above identities we obtain the assertion. 

4 Proofs of main results

We first investigate local properties of the mapping pZ introduced in (2.2) that will enable us
to treat the general case in Theorems 2.3 – 2.4. With the notation from Section 2, we establish
the following lemma.

Lemma 4.1 Consider (x0 , u) ∈ Dom (P) and put ξ := pZ [x0 , u] . Then for every t ∈ [0, T ] we
have
ξ(t) − ξ(t−) = P Z (u(t) − ξ(t−)) , ξ(t+) − ξ(t−) = P Z (u(t+) − ξ(t−)) , (4.1)
where (P Z , QZ ) is the projection pair introduced in (1.3). In particular, the inequalities

|ξ(t) − ξ(t−)| ≤ |u(t) − u(t−)| , |ξ(t+) − ξ(t)| ≤ |u(t+) − u(t)| (4.2)

hold for every t ∈ [0, T ] .

Proof. For a given t ∈ [0, T ] and z ∈ Z put

  
y(τ ) := z χ {t} (τ ) + u(τ +) − ξ(τ +) χ [0,t[ (τ ) + χ ]t,T ] (τ )

for τ ∈ [0, T ] . Then Proposition 3.7 (i) yields (cf. Remark 3.6) that
Z t Z t D E
0 ≤ hu(τ +) − ξ(τ +) − y(τ ), dξ(τ )i = (u(t) − ξ(t) − z)χ{t} (τ ), dξ(τ ) (4.3)
0 0
= hu(t) − ξ(t) − z, ξ(t) − ξ(t−)i .

If moreover T > t , then

Z T Z T D E
0 ≤ hu(τ +) − ξ(τ +) − y(τ ), dξ(τ )i = (u(t+) − ξ(t+) − z)χ{t} (τ ), dξ(τ ) (4.4)
0 0
= hu(t+) − ξ(t+) − z, ξ(t+) − ξ(t−)i .

Since z ∈ Z is arbitrary, we obtain from (1.4), (4.3), (4.4) that u(t)−ξ(t) = QZ (u(t)− ξ(t−)) ,
u(t+) − ξ(t+) = QZ (u(t+) − ξ(t−)) , and the assertion follows. The inequalities (4.2) are
obvious: the first one follows from (4.3) by putting z := u(t−) − ξ(t−) , to prove the second

16
one we put z := u(t+) − ξ(t+) in (4.3), z := u(t) − ξ(t) in (4.4) and sum up both inequalities.


As a consequence of Lemma 4.1, we see that ξ := pZ [x0 , u] is left- (right-) continuous if u is

left- (right-) continuous, respectively. We now pass to the proof of Proposition 2.2 which shows
that we can reduce the analysis to the left-continuous case.
Proof of Proposition 2.2. Let u ∈ G(0, T ; X) , y ∈ G(0, T ; Z) , x0 ∈ Z and ξ ∈ BV (0, T ; X)
be given. For t ∈ [0, T ] put u− (t) := u(t−) , ξ− (t) := ξ(t−) , ξ ∗ (t) := ξ(t) − ξ− (t) . For all
t ∈ [0, T ] and τ ∈ [0, t[ we have u− (τ +) − ξ− (τ +) = u(τ +) − ξ(τ +) , and from Proposition
3.7 (i), Lemma 3.3 and property (1.4) it follows that
Z t Z t D E
hu(τ +) − ξ(τ +) − y(τ ), dξ(τ )i = (u(t) − ξ(t) − u− (t) + ξ− (t))χ{t} (τ ), dξ− (τ ) (4.5)
0 0
Z t Z t
+ hu− (τ +) − ξ− (τ +) − y(τ ), dξ− (τ )i + hu(τ +) − ξ(τ +) − y(τ ), dξ ∗ (τ )i
0 0
Z t
= hu− (τ +) − ξ− (τ +) − y(τ ), dξ− (τ )i + hu(t) − ξ(t) − y(t), ξ(t) − ξ(t−)i .
0

Assume first that (x0 , u− ) ∈ Dom (P) , and put η := pZ [x0 , u− ] . By Lemma 4.1, η is left-
continuous, and putting
ξ(t) := η(t) + P Z (u(t) − η(t)) (4.6)
we obtain u(t) − ξ(t) ∈ Z for every t ∈ [0, T ] , ξ(t−) = η(t) + P Z (u− (t) − η(t)) = η(t) , hence
u(0) − ξ(0) = u− (0) − η(0) = x0 . Moreover, the first term on the rightmost side of (4.5) is
non-negative by hypothesis, the second term is non-negative by (4.6), hence ξ = pZ [x0 , u] .
Conversely, assume that (x0 , u) ∈ Dom (P) , and for a fixed y ∈ G(0, T ; Z) and t ∈ [0, T ] put
y ∗ (τ ) := y(τ ) + χ {t} (τ ) (u(t) − ξ(t) − y(t)) for τ ∈ [0, T ] . Then y ∗ ∈ G(0, T ; Z) and identity
(4.5) with y replaced by y ∗ yields
Z t
0 ≤ hu(τ +) − ξ(τ +) − y ∗ (τ ), dξ(τ )i
0
Z t D E
= u− (τ +) − ξ− (τ +) − y(τ ) + χ{t} (τ ) (y(t) − y ∗ (t)), dξ− (τ )
0
Z t
= hu− (τ +) − ξ− (τ +) − y(τ ), dξ− (τ )i ,
0

where we used Proposition 3.7 (i) and the left-continuity of ξ− . We have indeed u− (t)−ξ− (t) ∈
Z for every t , u− (0) − ξ− (0) = u(0) − ξ(0) , and Proposition 2.2 is proved. 

In the sequel, we denote by BVL (0, T ; X) the space of functions from BV (0, T ; X) which are
left-continuous. Besides the fact that Var [a,b] f = Var [a,b] f for every f ∈ GL (0, T ; X) and
every [a, b] ⊂ [0, T ] , the restriction to left-continuous functions has the following advantage.

Lemma 4.2 Let u ∈ G(0, T ; X) and ξ ∈ BVL (0, T ; X) be such that u(t) − ξ(t) ∈ Z for
every t ∈ [0, T ] , u(0) − ξ(0) = x0 . Assume that
Z T
hu(τ +) − ξ(τ +) − y(τ ), dξ(τ )i ≥ 0 ∀y ∈ G(0, T ; Z) . (4.7)
0

17
Then for every 0 ≤ s < t ≤ T we have
Z t
hu(τ +) − ξ(τ +) − y(τ ), dξ(τ )i ≥ 0 ∀y ∈ G(0, T ; Z) , (4.8)
s

in particular ξ = pZ [x0 , u] .

Proof. For y ∈ G(0, T ; Z) and τ ∈ [0, T ] put

 
ŷ(τ ) := (u(τ +) − ξ(τ +)) χ [0,s[ (τ ) + χ [t,T ] (τ ) + y(τ ) χ [s,t[ (τ ) .

Then ŷ ∈ G(0, T ; Z) and combining (4.7) with Propositions 3.5, 3.7 (i) we obtain using the
left-continuity of ξ that
Z T Z T D E
0 ≤ hu(τ +) − ξ(τ +) − ŷ(τ ), dξ(τ )i = (u(τ +) − ξ(τ +) − y(τ ))χ[s,t[ (τ ), dξ(τ )
0 0
Z t
= hu(τ +) − ξ(τ +) − y(τ ), dξ(τ )i
s

and Lemma 4.2 is proved. 

Lemma 4.2 and Proposition 2.2 enable us to construct easily ξ = pZ [x0 , u] ∈ S(0, T ; X)
whenever u ∈ S(0, T ; X) . As it has already been mentioned, the explicit formula coincides
with the time-discrete scheme of [16, 12].

Proposition 4.3 Let x0 ∈ Z be given and let u ∈ S(0, T ; X) be of the form

m
X
u(t) = u0 χ {0} (t) + uk χ ]tk−1 ,tk ] (t) for t ∈ [0, T ] , (4.9)
k=1

with {t0 , . . . , tm } ∈ Da,b . For k = 1, . . . , m put ξ0 := u0 − x0 , ξk := ξk−1 + P Z (uk − ξk−1 ) .

Then ξ = pZ [x0 , u] has the form
m
X
ξ(t) = ξ0 χ {0} (t) + ξk χ ]tk−1 ,tk ] (t) for t ∈ [0, T ] (4.10)
k=1

and we have Var [0,T ] ξ = Var [0,T ] ξ ≤ Var [0,T ] u = Var [0,T ] u .

Proof. For every k = 1, . . . , m we have uk − ξk = QZ (uk − ξk−1 ) ∈ Z , hence u(t) − ξ(t) ∈ Z

for every t ∈ [0, T ] . Moreover, by (1.4) and Proposition 3.7, the function ξ given by (4.10)
satisfies the inequality
Z T m
X
hu(τ +) − ξ(τ +) − y(τ ), dξ(τ )i = hξk − ξk−1 , uk − ξk − y(tk−1 )i ≥ 0
0 k=1

for each y ∈ G(0, T ; Z) , hence ξ = pZ [x0 , u] by Lemma 4.2. To complete the proof, we
use again (1.4) which entails that hξk − ξk−1 , uk − ξk − uk−1 + ξk−1 i ≥ 0 , hence |ξk − ξk−1 | ≤
|uk − uk−1 | for every k = 1, . . . , m and Var [0,T ] ξ ≤ Var [0,T ] u . 

18
We are now ready to prove Theorem 2.3.
Proof of Theorem 2.3. From Propositions 4.3 and 2.2 it follows that for each function
m
X m
X
u(t) = ûk χ {tk } (t) + uk χ ]tk−1 ,tk [ (t)
k=0 k=1

we have ξ = pZ [x0 , u] if and only if

m m
ξˆk χ {tk } (t) +
X X
ξ(t) = ξk χ ]tk−1 ,tk [ (t)
k=0 k=1

for t ∈ [a, b] with ξˆ0 = û0 − x0 , ξk − ξk−1 = P Z (uk − ξk−1 ) , ξˆk − ξk = P Z (ûk − ξk ) for
k = 1, . . . , m . By Lemma 4.1 we have |ξk − ξk−1 | ≤ |uk − uk−1 | , hence Var [0,T ] ξ ≤ Var [0,T ] u .
Consider now u ∈ BV (0, T ; X) and x0 ∈ Z . By Proposition 1.3 (iv) we find un ∈ S(0, T ; X)
such that ku − un k[0,T ] → 0 as n → ∞ , Var [0,T ] un ≤ Var [0,T ] u . Put ξn := pZ [x0 , un ] . By
Lemma 2.1 we have for n, m ∈ N that
!
2 2
|ξn (t) − ξm (t)| ≤ |ξn (0) − ξm (0)| + 2 kun − um k[0,t] Var (ξn ) + Var (ξm ) (4.11)
[0,t] [0,t]

≤ |un (0) − um (0)|2 + 4 kun − um k[0,t] Var u ,

[0,t]

hence {ξn } is a Cauchy sequence in G(0, T ; X) , Var [0,T ] ξn ≤ Var [0,T ] u for every n ∈ N . Let
ξ ∈ BV (0, T ; X) be its limit, Var [0,T ] ξ ≤ Var [0,T ] u . From Proposition 3.12 we infer that
ξ = pZ [x0 , u] , and the estimate (2.4) follows immediately from Lemma 2.1. Theorem 2.3 is
proved. 

For the proof of Theorem 2.4 we need the following crucial Lemma the idea of which (for
bounded domains Z ) goes back to A. Vladimirov, see Sect. 19 of [10], cf. also Chapter 2 of
[15].

Lemma 4.4 Let 0 ∈ Int Z and {(xn0 , un ) ; n ∈ N} be an arbitrary sequence in Dom (P) ∩
(Z × GL (0, T ; X)) such that |xn0 − x0 | → 0 , kun − uk[0,T ] → 0 as n → ∞ . Then there exists
a constant C > 0 independent of n such that Var [0,T ] pZ [xn0 , un ] ≤ C .

Proof. Notice first that the uniform convergence of {un } and Proposition 1.3 (i) guarantee that
u ∈ GL (0, T ; X) . Next, let ρ > 0 be as in (1.1) and let us denote Uρ := {t ∈ [0, T ] ; |u(t+) −
u(t)| ≥ ρ/6} . By Proposition 1.3 (iii), (iv), the set Uρ is finite and there exists h > 0 such
that for every [s, t] ⊂ [0, T ] with |t − s| < h and [s, t] ∩ Uρ = ∅ , we have |u(t) − u(s)| ≤ ρ/6 .
We fix a partition d = {t0 , . . . , tm } such that Uρ ⊂ d , tk − tk−1 < h for k = 1, . . . , m , and a
number n0 ∈ N such that kun − uk[0,T ] ≤ ρ/6 for n ≥ n0 .
For ξn = pZ [xn0 , un ] and k ∈ {0, . . . , m} put
xkn := un (tk +) − ξn (tk +) ∈ Z . (4.12)
Consider now a fixed k ∈ {1, . . . , m} . By Lemma 4.2 we have for every y ∈ G(0, T ; Z) , n ∈ N
and ε ∈ ]0, (tk − tk−1 )/2[ that
Z tk −ε
hun (τ +) − ξn (τ +) − y(τ ), dξn (τ )i ≥ 0 . (4.13)
tk−1 +ε

19
In particular, for n ≥ n0 we may put in (4.13)
ρ
 
y(τ ) = un (τ +) − un (tk−1 +) + w(τ ) χ ]tk−1 ,tk [ (τ )
2
for an arbitrary w ∈ S(0, T ; B1 (0)) , since for τ ∈ ]tk−1 , tk [ we have |un (τ +) − un (tk−1 +)| ≤
ρ/2 , hence y(τ ) ∈ Bρ (0) ⊂ Z . Then (4.13) yields
ρ Z tk −ε Z tk −ε
hw(τ ), dξn (τ )i ≤ hun (tk−1 +) − ξn (τ +), dξn (τ )i
2 tk−1 +ε tk−1 +ε

and from Theorem 3.11 and Corollary 3.15 it follows that

ρ 1 
Var ξn ≤ |un (tk−1 +) − ξn (tk−1 + ε)|2 − |un (tk−1 +) − ξn (tk − ε)|2 .
2 [tk−1 +ε,tk −ε] 2
Letting ε → 0 (note that ξn is left-continuous), we obtain on the one hand that

|un (tk−1 +) − ξn (tk−1 +)| ≥ |un (tk−1 +) − ξn (tk )| , (4.14)

and, on the other hand, Proposition 1.3 (vi) yields that

ρ Var ξn = ρ Var ξn ≤ ρ |ξn (tk−1 +) − ξn (tk−1 )| (4.15)

[tk−1 ,tk ] [tk−1 ,tk ]

+ |un (tk−1 +) − ξn (tk−1 +)|2 − |un (tk−1 +) − ξn (tk )|2 .

Lemma 4.1, inequality (4.14) and the triangle inequality imply that

|xkn | ≤ |un (tk−1 +) − ξn (tk )| + |un (tk +) − un (tk−1 +)| + |ξn (tk +) − ξn (tk )|
≤ |xk−1
n | + 2 |un (tk +) − un (tk )| + ρ/2

≤ |xk−1
n | + 2 |u(tk +) − u(tk )| + 3ρ/2 .

Now consider ` ∈ {1, . . . , m} . Summing up the above inequalities from k = 1 to k = ` , we

obtain that m
|x`n | ≤ |x0n | + 2
X
|u(tk +) − u(tk )| + 3m ρ/2 .
k=1

Observe further that Lemma 4.1 and the triangle inequality ensure that

|x0n | ≤ |xn0 | + 2 |un (0+) − un (0)| ≤ sup {|xn0 |} + 2 |u(0+) − u(0)| + ρ/2 .
n

Consequently, for ` ∈ {0, . . . , m} ,

m
|x`n | ≤ sup {|xn0 |} + 2
X
|u(tk +) − u(tk )| + (3m + 1) ρ/2 =: C1 . (4.16)
n k=0

From (4.15), (4.16) and Lemma 4.1, we thus obtain that

ρ Var ξn ≤ ρ |un (tk−1 +) − un (tk−1 )| + |xk−1 2 2 2

n | ≤ ρ /2 + C1 + ρ |u(tk−1 +) − u(tk−1 )| ,
[tk−1 ,tk ]

hence Var [0,T ] ξn ≤ mρ/2 + mC12 /ρ + C1 /2 , and Lemma 4.4 is proved. 

20
Proof of Theorem 2.4. For an arbitrary (x0 , u) ∈ Z ×G(0, T ; X) we find a sequence {un }∞ n=1 of
n
step functions such that kun − uk[0,T ] → 0 as n → ∞ . For t ∈ [0, T ] and n ∈ N put u− (t) :=
un (t−) , u− (t) := u(t−) , ξn (t) := pZ [xn0 , un ](t) , ξ−
n
(t) := pZ [xn0 , un− ](t) . From Proposition 2.2
n
it follows that ξn (t−) = ξ− (t) for every t ∈ [0, T ] and n ∈ N . Using Lemma 4.4 we find a
n
constant C > 0 such that Var [0,T ] ξ− ≤ C , and Lemma 4.1 yields
Var ξn ≤ C + |ξn (T ) − ξn (T −)| ≤ C + |un (T ) − un (T −)| ≤ C̄
[0,T ]

for some constant C̄ . The same argument as in (4.11) implies that {ξn } is a Cauchy sequence
in G(0, T ; X) . Denoting its limit by ξ , we obtain from Proposition 3.12 that ξ := pZ [x0 , u] .
Repeating the same argument for an arbitrary sequence kun − uk[0,T ] → 0 , un ∈ G(0, T ; X) ,
we complete the proof. 

5 Functions of bounded ε-variation

In this section, we give an extension of Proposition 3.12 to the case where the sequence {fn }
does not converge uniformly. The idea is based on the following concept introduced in [7], Def.
3.3.

Definition 5.1 We say that a set A ⊂ G(a, b ; X) has uniformly bounded ε -variation, if
( )
∀ ε > 0 ∃ Lε > 0 ∀ f ∈ A : inf Var ψ ; ψ ∈ BV (a, b ; X) , kf − ψk[a,b] < ε ≤ Lε .
[a,b]

We will see in Proposition 5.6 below that every uniformly convergent sequence in G(a, b ; X) has
uniformly bounded ε -variation. The converse is obviously false, as we can see from the example
fn (t) = χ [0,1/n] (t) for t ∈ [0, 1] . On the other hand, we prove the following generalization of
Helly’s Selection Principle as an extension of Theorem 3.8 of [7] to the infinite dimensional
case.

Theorem 5.2 Let X be a real separable Hilbert space and let {fn ; n ∈ N} be a bounded
sequence of functions from G(a, b ; X) which has uniformly bounded ε -variation. Then there
exist f ∈ G(a, b ; X) and a subsequence {fnk } of {fn } such that fnk (t) converges weakly to
f (t) as k → ∞ for every t ∈ [a, b] .

The proof of Theorem 5.2 consists in a gradual selection of subsequences similar to the proof
of the classical Helly Selection Principle (see e. g. [9], pp. 372 – 374). In order to make the
diagonalization argument more transparent, we introduce the following notation.
By G(N) we denote the set of all infinite subsets M ⊂ N . We say that a sequence {xn ; n ∈ N}
of elements of a topological space M -converges to x if for every neighborhood U(x) of x there
exists n0 such that xn ∈ U(x) for every n ∈ M , n ≥ n0 .
We start with the following Lemma as the Hilbert-space version of [3], Theorem I.3.5.

Lemma 5.3 Let {ψn ; n ∈ N} be a bounded sequence in BV (a, b ; X) such that Var [a,b] ψn ≤
C for every n ∈ N . Then there exist ψ ∈ BV (a, b ; X) and a set M ∈ G(N) such that
Var [a,b] ψ ≤ C and the sequence ψn (t) weakly M -converges in X to ψ(t) for every t ∈ [a, b] .

21
Proof. Let {wj ; j ∈ N} be a countable dense subset of X . The functions t 7→ hψn (t), w1 i have
uniformly bounded variation, and according to the one-dimensional Helly Selection Principle
we find N1 ∈ G(N) such that the sequence {hψn (t), w1 i} N1 -converges to a limit v1 (t) for every
t ∈ [a, b] . By induction we construct a sequence {Nk ; k ∈ N} of sets in G(N) , N1 ⊃ N2 ⊃ . . . ,
such that the sequence {hψn (t), wj i} Nj -converges to a limit vj (t) for every t ∈ [a, b] . We
now put n1 := min N1 , nk := min{n ∈ Nk ; n > nk−1 } for k = 2, 3, . . . , and define the set
M := {nk ; k ∈ N} ∈ G(N) . By construction, every Nj -convergent sequence is M -convergent,
hence {hψn (t), wj i} M -converges to vj (t) for every t ∈ [a, b] and j ∈ N .
For a fixed t ∈ [a, b] , the mapping wj 7→ vj (t) can be extended in a unique way to a bounded
linear functional on X . By the Riesz Representation Theorem, there exists an element ψ(t) ∈
X such that vj (t) = hψ(t), wj i for every j ∈ N . Since the system {wj } is dense in X , we
obtain that
lim hψnk (t), wi = hψ(t), wi
k→∞
for every w ∈ X and t ∈ [a, b] . Moreover, for a fixed partition a = t0 < t1 < . . . < tm = b we
have m m
X X
|ψ(ti ) − ψ(ti−1 )| ≤ lim inf |ψnk (ti ) − ψnk (ti−1 )| ≤ C ,
k→∞
i=1 i=1
and the assertion follows. 

We now use Lemma 5.3 to prove Theorem 5.2 by an argument similar to the one used in [7] in
the case dim X < ∞ .
Proof of Theorem 5.2. We fix a sequence εi → 0 and for every n, i ∈ N we find ψni ∈
BV (a, b ; X) such that kψni − fn k[a,b] < εi , Var [a,b] ψni ≤ Lεi + 1 . We now apply Lemma 5.3
to find M1 ∈ G(N) and ψ 1 ∈ BV (a, b ; X) such that Var [a,b] ψ 1 ≤ Lε1 + 1 and ψn1 (t) weakly
M1 -converges to ψ 1 (t) for every t ∈ [a, b] . We continue by induction and construct a sequence
{Mi } of sets in G(N) , M1 ⊃ M2 ⊃ . . . , such that the sequence {ψni (t)} weakly Mi -converges
to ψ i (t) for every t ∈ [a, b] and i ∈ N , ψ i ∈ BV (a, b ; X) , Var [a,b] ψ i ≤ Lεi + 1 . Putting
n1 := min M1 , nk := min{n ∈ Mk ; n > nk−1 } for k = 2, 3, . . . , M ∗ := {nk ; k ∈ N} we argue
as in the proof of Lemma 5.3 to obtain that ψni (t) weakly M ∗ -converges to ψ i (t) for every
t ∈ [a, b] and i ∈ N .
We now check that {ψ i } is a Cauchy sequence in G(a, b ; X) . For i, j, n ∈ N we have

ψni − ψnj ≤ ψni − fn + fn − ψnj ≤ εi + εj .

[a,b] [a,b] [a,b]

Consequently we have for t ∈ [a, b] ,

ψ i (t) − ψ j (t) ≤ lim inf ψni k (t) − ψnj k (t) ≤ εi + εj ,

k→+∞

from which readily follows that {ψ i } is a Cauchy sequence in G(a, b ; X) . We denote by

f ∈ G(a, b ; X) its limit. For each t ∈ [a, b] , w ∈ X and k ∈ N we then have
D E D E D E
hf (t) − fnk (t), wi = f (t) − ψ i (t), w + ψ i (t) − ψni k (t), w + ψni k (t) − fnk (t), w

for a suitably chosen i , and we easily conclude that fnk (t) weakly converges to f (t) for every
t ∈ [a, b] . Theorem 5.2 is proved. 

As a complement to Theorem 5.2, the following extension of Proposition 3.12 holds true.

22
Proposition 5.4 Let f, fn ∈ G(a, b ; X) , g, gn ∈ BV (a, b ; X) for n ∈ N be such that the
sequence {fn } has uniformly bounded ε -variation and
fn (t) → f (t) weakly for every t ∈ [a, b] ,
lim kg − gn k[a,b] = 0 , sup Var gn = C < ∞ .
n→∞ n∈N [a,b]
Then (3.25) holds.

For the proof of Proposition 5.4 we need the following lemma.

Lemma 5.5 Consider w ∈ S(a, b ; X) and f˜n : [a, b] → X , f˜n (t) → 0 weakly for every
t ∈ [a, b] . Then
Z bD E
lim f˜n (t), dw(t) = 0 .
n→∞ a

Proof of Lemma 5.5. For a function w of the form (1.6) we have by Proposition 3.7
Z b D E m D E
f˜n (t), dw(t) f˜n (tk ), ck+1 − ck ,
X
=
a k=0

where we put c0 := ĉ0 , cm+1 := ĉm , and it suffices to pass to the limit as n → ∞ . 

Proof of Proposition 5.4. For each ε > 0 and n ∈ N we find {ψ ε } , {ψnε } in BV (a, b ; X)
such that kfn − ψnε k[a,b] ≤ ε , kf − ψ ε k[a,b] ≤ ε , Var [a,b] ψnε ≤ Lε + 1 , and put L̂ε :=
max{Var [a,b] ψ ε , Lε + 1} .
The sequence {fn } is obviously bounded in G(a, b ; X) . Indeed, as {fn (a)} is weakly conver-
gent, it is necessarily bounded and we have for every n and t that
|fn (t)| ≤ |fn (t) − ψnε (t)| + |ψnε (t) − ψnε (a)| + |fn (a) − ψnε (a)| + |fn (a)| ≤ 2 ε + L̂ε + |fn (a)|
and taking the infimum over ε we obtain an upper bound for kfn k[a,b] independent of n and
ε , say
kfn k[a,b] ≤ R .
Let now ε > 0 be fixed. By Proposition 1.3 (v), there exists a step function w ∈ S(a, b ; X)
such that kg − wk[a,b] ≤ ε/L̂ε , Var [a,b] w ≤ C . Using Lemma 5.5 and the uniform convergence
of {gn } , we find n0 such that for n ≥ n0 we have | ab h(f − fn )(t), dw(t)i | ≤ ε , kg − gn k[a,b] ≤
R

ε/L̂ε . Then Corollary 3.10 yields

Z b Z b Z b
hf (t), dg(t)i − hfn (t), dgn (t)i ≤ h(f − ψ ε − fn + ψnε )(t), d(g − w)(t)i
a a a
Z b Z b
+ h(f − fn )(t), dw(t)i + h(ψ ε − ψnε )(t), d(g − w)(t)i
a a
Z b Z b
+ h(fn − ψnε )(t), d(g − gn )(t)i + hψnε (t), d(g − gn )(t)i
a a

≤ 2 C kf − ψ ε − fn + ψnε k[a,b] + ε + (4 (R + ε) + 2 L̂ε ) kg − wk[a,b]

+ 2 C kfn − ψnε k[a,b] + (2 (R + ε) + L̂ε ) kg − gn k[a,b]
≤ Mε

23
for n ≥ n0 , where M is a constant independent of n and ε , hence (3.25) holds. 

To conclude the paper, we show how Theorem 2.4 can be used to prove directly the following
link between Propositions 3.12 and 5.4.

Proposition 5.6 Let {un } be a sequence in G(0, T ; X) , kun − uk[0,T ] → 0 as n → ∞ . Then

{un } has uniformly bounded ε -variation.

Proof. Let ε > 0 be given. For t ∈ [0, T ] and n ∈ N put u− (t) := u(t−) , un− (t) := un (t−) ,
n
ξ− (t) := pZ [0, un− ](t) with Z = Bε/2 (0) . By Theorem 2.4 there exists Cε > 0 such that
n
Var [0,T ] ξ− n
≤ Cε independently of n , un− − ξ− ≤ ε/2 .
[0,T ]

Let Uε ⊂ [0, T ] be the finite set of those t for which |u(t) − u− (t)| ≥ ε/4 . For t ∈ [0, T ] and
n ∈ N we now put

unε (t) := un− (t) + (un (t) − un− (t)) χ Uε (t) , ξεn (t) := pZ [0, unε ](t) . (5.1)

Clearly, unε (t−) = un− (t) for t ∈ [0, T ] and we infer from Proposition 2.2 that ξεn (t−) =
n
ξ− (t) for t ∈ [0, T ] . Consequently, ξεn (t) = ξεn (t−) = ξ− n
(t) for every t ∈ [0, T ] \ Uε , while
|ξε (t) − ξε (t−)| ≤ |u (t) − u (t−)| for t ∈ Uε by Lemma 4.1. Then ξεn ∈ BV (0, T ; X) with
n n n n

Var [0,T ] ξεn ≤ Cε +2 t∈Uε |un (t)−un− (t)| , where |un (t)−un− (t)| ≤ 2 kun − uk[0,T ] +|u(t)−u− (t)| ,
P

and we may put Lε := supn Var [0,T ] ξεn < +∞ . For every t ∈ [0, T ] and n ∈ N we have by
(5.1) that |unε (t) − ξεn (t)| ≤ ε/2 and, on the other hand,
   
|un (t) − unε (t)| = un (t) − un− (t) 1 − χUε (t)
 
≤ (u(t) − u− (t)) 1 − χUε (t) + 2 kun − uk[0,T ]
≤ ε/4 + 2 kun − uk[0,T ] ≤ ε/2

for n ≥ n0 with n0 sufficiently large. This yields that kun − ξεn k[0,T ] ≤ ε for n ≥ n0 . For
n < n0 we approximate the functions un for instance by step functions and taking a larger Lε
if necessary, we complete the proof. 

References
[1] G. Aumann: Reelle Funktionen (German). Springer-Verlag, Berlin – Göttingen – Heidelberg,
1954.

[2] H. Brézis: Opérateurs maximaux monotones. North-Holland Math. Studies, Amsterdam, 1973.

[3] V. Barbu, T. Precupanu: Convexity and optimization in Banach spaces. 2nd ed., D. Reidel,
Dordrecht – Boston – Lancaster, 1986.

[4] M. Brokate, J. Sprekels: Hysteresis and phase transitions. Appl. Math. Sci., 121, Springer-
Verlag, New York, 1996.

[5] P. Drábek, P. Krejčı́, P. Takáč: Nonlinear differential equations. Research Notes in Mathe-
matics, Vol. 404, Chapman & Hall/CRC, London, 1999.

24
 [6] L. C. Evans, R. F. Gariepy: Measure theory and fine properties of functions. CRC Press, Boca
Raton, 1992.

[7] D. Fraňková: Regulated functions. Math. Bohem. 119 (1991), 20–59.

[8] T. H. Hildebrandt: Introduction to the theory of integration. Academic Press, New York and
London, 1963.

[9] A. N. Kolmogorov, S. V. Fomin: Introductory real analysis. Prentice Hall, Inc., Englewood
Cliffs, 1970.

[10] M. A. Krasnosel’skii, A. V. Pokrovskii : Systems with hysteresis. Nauka, Moscow, 1983 (English
edition Springer 1989).

[11] P. Krejčı́: Vector hysteresis models. Euro. Jnl. Appl. Math. 2 (1991), 281–292.

[12] P. Krejčı́: Hysteresis, convexity and dissipation in hyperbolic equations. Gakuto Int. Ser.
Math. Sci. Appl., Vol. 8, Gakkōtosho, Tokyo, 1996.

[13] P. Krejčı́, J. Kurzweil : A nonexistence result for the Kurzweil integral. Math. Bohem., to
appear.

[14] J. Kurzweil : Generalized ordinary differential equations and continuous dependence on a

parameter. Czechoslovak Math. J. 7 (82) (1957), 418–449.

[15] M.D.P. Monteiro Marques: Differential inclusions in nonsmooth mechanical problems. Shocks
and dry friction. Birkhäuser, Basel, 1993.

[16] J.-J. Moreau: Evolution problem associated with a moving convex set in a Hilbert space. J.
Diff. Eq. 26 (1977), 347–374.

[17] R. T. Rockafellar : Convex analysis. Princeton University Press, 1970.

[18] Š. Schwabik : On the relation between Young’s and Kurzweil’s concept of Stieltjes integral.
Časopis pěst. mat. 98 (1973), 237–251.

[19] Š. Schwabik : Abstract Perron-Stieltjes integral. Math. Bohem. 121 (1996), 425–447.

[20] Š. Schwabik, M. Tvrdý, O. Vejvoda: Differential and Integral Equations : Boundary Value
Problems and Adjoints. Academia and D. Reidel, Praha, 1979.

[21] M. Tvrdý: Regulated functions and the Perron-Stieltjes integral. Časopis pěst. mat. 114
(1989), 187–209.

[22] A. Visintin: Differential models of hysteresis. Springer, Berlin - Heidelberg, 1994.

25