Abstract
Using classical techniques related to the so-called Hardy–Vitali variation, we present the class of X-valued functions of bounded \(\Phi \)-variation in several variables, where \((X,d,+ )\) is a metric semigroup. We exhibit some of the main properties of this class; among them, we show that this class can be made into a normed space and present a counterpart of the renowned Riesz’s Lemma for the case in which \(X=\mathbb {R}\) with its usual metric.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
BV functions of a single variable were first introduced by Camille Jordan, in a paper [21] that deals with the convergence of Fourier series. Soon after Jordan’s work, many mathematicians began to study notions of bounded variation for functions of several variables. There are various approaches to the notion of variation for functions of several variables. We can mention those belonging to Vitali, Hardy, Krause, Arzela, Frechét, Tonelli, Hahn, Kronrod-Vitushkin, Minlos, and others. Functions of bounded variation in n variables (\(n>1\)) belonging to each of these classes have more or less the same properties as the functions of bounded variation of one variable. However, there are some properties in the one-dimensional case that cannot be transferred automatically to the multidimensional one (see [22]). On the other hand, functions of bounded variation in \(\mathbb {R}^n\) can be identified with n-dimensional normal currents in \(\mathbb {R}^n\). This point of view is due to Federer [18].
In the literature, the many notions of bounded variation are mainly studied for functions defined on a rectangle \(J \subset \mathbb {R}^n\). A definition of the variation in the sense of Hardy and Krause is given in [22]. Let \(f: [0,1]^n \rightarrow \mathbb {R}\). Let \(\mathbf {a}=(a_1,\ldots ,a_n)\) and \(\mathbf {b}=(b_1,\ldots ,b_n)\) be elements of \([0,1]^n\) such that \(\mathbf {a} < \mathbf {b}\) (see Sect. 2). The n-dimensional difference operator \(\Delta ^n\), which assigns to the axis-parallel rectangle \([\mathbf {a}, \mathbf {b}]\) a n-dimensional quasi-volume
For \(s= 1,\ldots , n\) let \(0 = x_0^{(s)}< x_1^{(s)} < \ldots x_{m_s}^{(s)} = 1 \) be a partition of [0, 1], and \(\mathcal {P}\) be the partition of \([0,1]^n\) which is given by
Then the variation of f on \([0,1]^n\) in the sense of Vitali is given by
where the supremum is extended over all partitions of \([0,1]^n\) into axis-parallel boxes generated by d one-dimensional partitions of [0, 1], as in (1.1).
If the same functions restricted to the various faces of \([0,1]^s\) with \(s=1,\ldots ,n\) are of bounded variation in the sense of Vitali over each of them, then f is said to be of bounded variation on \([0,1]^s\) in the sense of Hardy and Krause, that is, for \(1 \le s \le n\) and \(1 \le i_1< \cdots < i_s \le n,\) let \(V^{(s)}(f; i_1, \ldots , i_s; [0,1]^n)\) denote the s-dimensional variation in the sense of Vitali of the restriction of f to the face
of \([0,1]^n\). Then the variation of f on \([0, 1]^n\) in the sense of Hardy and Krause anchored at \(\mathbf {1}\), abbreviated by HK-variation, is given by
Note that for the definition of the HK-variation in (1.2), we add the n-dimensional variation in the sense of Vitali plus the variation in the sense of Vitali on all lower dimensional faces of \([0, 1]^n\) which are adjacent to \(\mathbf {1}\).
On the other hand, a function \(f: \mathbb {R}^n \rightarrow \mathbb {R}\) is said to be of bounded Tonelli-variation if a.e. in \((x_1, \ldots , x_{j-1}, x_{j+1},\ldots ,x_n)\) it is of bounded variation in each variable \(x_j\) for all \(1 \le j \le n\) and if these variations \(\mathrm{BV}_j(f(x)):= \mathrm{BV}_{x_j \in \mathbb {R}(f(x))}\) are Lebesgue integrable as functions of the other \(n-1\) variables \(x_1, x_2,\ldots , x_{j-1}, x_{j+1},\ldots , x_n\):
which for a smooth enough function f, it is equal to
Among the sources dealing with the Tonelli variation, let us mention [2, 7, 9, 16, 27].
Until now it seems ([20, 22]) that only the approach due to Vitali–Hardy–Krause gives a notion of variation for real-valued functions of several variables such that a complete analogue of the Helly theorem holds with respect to the pointwise convergence of extracted subsequences. However, the point of view which is nowadays accepted in the literature as most efficient generalization of the 1-dimensional theory is due to De Giorgi and Fichera (see [17, 19]).
Thus, the unvarying interest generated by the classical notion of function of bounded variation has led to some generalizations of the concept, mainly, intended to the search of bigger classes of functions whose elements have pointwise convergent Fourier series or to find applications in geometric measure theory, calculus of variations, and mathematical physics. As in the classical case, these generalizations have found many applications in the study of certain differential and integral equations (see e.g., [8]).
In this paper, we present a detailed study of the space of functions of bounded Riesz-\(\Phi \)-variation, which was introduced previously in [5, 6], for real-valued functions of several variables.
This extends the work done in [3] (resp. in [4]), in which the authors present the notion of real-valued function (resp. vector valued) of bounded Riesz-\(\Phi \)-variation, that are defined on a rectangle of \(\mathbb {R}^2\). In particular, we extend some results due to Chistyakov ([15]) and give a version of Riesz’s Lemma for the case of functions of several real variables which take values in a reflexive Banach space.
2 Notation and preliminary
To start, we give some notations and definitions that will be used throughout the rest of this paper (see [5, 12, 13]).
As usual, \(\mathbb {N}, \mathbb {N}_0\) and \(\mathbb {R}\) denotes the set of all positive integers, non-negative integers and real numbers, respectively. A typical point of \(\mathbb {X}^{n}\) (\(\mathbb {N}, \mathbb {N}_0\) or \(\mathbb {R}\)) is denoted as \( \mathbf {x} = (x_1, x_2, \ldots , x_n) := (x_i)_{i=1}^{n}\), but the canonical unit vectors of \(\mathbb {R}^{n}\) are denoted by \(\mathbf {e_j}\) (\(j = 1, 2, \dots , n\)); that is, \(\mathbf {e_j} := (e_1^{j}, e_2^{j},\ldots , e_n^{j})\) where, \( e_r^{j} := \left\{ \begin{array}{ll} 0 &{} \text { if } \, r\ne j \\ 1 &{} \text { if } \, r= j. \end{array} \right. \). The zero n-tuple \((0,0, \dots , 0)\) will be denoted by \(\mathbf {0}\), and by \(\mathbf {1}\) we will denote the n-tuple \(\mathbf {1}= (1,1, \dots , 1)\).
If \( {{\varvec{\alpha }}} = (\alpha _1, \alpha _2, \ldots , \alpha _n)\) is a n-tuple of non-negative integers then we call \( {\varvec{\alpha }}\) a multi-index ([1]).
If \( \mathbf {a}=(a_1, a_2, \dots , a_n) ,\mathbf {b}= (b_1, b_2, \dots , b_n) \in \mathbb {R}^{n}\) we use the notation \( \mathbf {a} < \mathbf {b}\) to mean that \(a_i < b_i\) for each \(i=1,\ldots , n\) and similarly are defined \( \mathbf {a} = \mathbf {b}\), \( \mathbf {a} \le \mathbf {b}\), \( \mathbf {a} \ge \mathbf {b}\) and \( \mathbf {a} > \mathbf {b}\). If \( \mathbf {a} < \mathbf {b}\), the set \(\mathbf {J}:= [ \mathbf {a}, \mathbf {b}] = \prod \nolimits _{i=1}^{n} [a_i, b_i]\) will be called a n-dimensional closed interval. The Euclidean volume of an n-dimensional closed interval will be denoted by \({{\mathrm{Vol}}}[ \mathbf {a}, \mathbf {b}] \); that is, \({{\mathrm{Vol}}}[ \mathbf {a}, \mathbf {b}] = \prod \nolimits _{i=1}^{n} (b_i - a_i). \)
In addition, for \( {\varvec{\alpha }}= (\alpha _1, \alpha _2, \ldots , \alpha _n) \in \mathbb {N}_0^{n}\) and \( \mathbf {x} = (x_1, x_2, \ldots , x_n) \in \mathbb {R}^{n}\) we will use the notations \(| {\varvec{\alpha }}| := \alpha _1 + \alpha _2 + \cdots + \alpha _n \) and \(\alpha \mathbf {x} := (\alpha _1 x_1, \alpha _2 x_2, \ldots , \alpha _n x_n).\)
In this work, we will consider functions whose domain is a n-dimensional closed interval \([ \mathbf {a}, \mathbf {b}] \) and whose range is an invariant metric semigroup \((X,d,+,\cdot )\); i.e., (X, d) is a complete metric space, d is a translation invariant metric on X, \((X,+)\) is an commutative semigroup. In particular, the triangle inequality implies that, for all \(u,v, p, q \in X\),
Clearly any normed space is a metric semigroup.
The following standard notation (see [14]) will be used: \(\mathcal {N}\) will denote the set of all continuous convex functions \(\Phi :[0,+\infty ) \rightarrow [0,+\infty )\) such that \(\Phi (t)= 0\) if and only if \(t=0\), and \(\mathcal {N}_{\infty }\) the set of all functions \(\Phi \in \mathcal {N}\), for which the Orlicz condition (also called \(\infty _1 \) condition) holds: \( \lim \nolimits _{t \rightarrow \infty } \dfrac{\Phi (t)}{t} = + \infty \). Following [23], functions from \(\mathcal {N}\) are called \(\varphi \)-functions. Any function \(\Phi \in \mathcal {N}\) is strictly increasing, and so, its inverse \(\Phi ^{-1}\) is continuous and concave; besides, the functions \(t \longmapsto \dfrac{\Phi (t)}{t}\) and \(t \longmapsto t \Phi ^{-1}\left( \dfrac{1}{t} \right) \) are nondecreasing for \(t > 0\).
One says that a function \(\Phi \in \mathcal {N}\) satisfies a \(\Delta _2\) condition, and writes \(\Phi \in \Delta _2\), if there are constants \(\, K > 2\, \) and \(\, t_0 >0 \in \mathbb {R} \, \) such that
For instance, if \( \Phi (x) := t^p, \, p > 1, \) one may choose the optimal constant \(K= 2^p.\)
Now we define two sets that will play an important role in this work:
Notice that these sets are related in one-to-one correspondence; indeed, if \(\theta = (\theta _1, \ldots , \theta _n) \in \mathcal {E}(n)\) then we can define \(\widetilde{\mathbf {\theta }} := (\theta _1, \ldots ,\theta _{i-1}, 1 - \theta _i,\theta _{i+1},\ldots \theta _n) \in \mathcal {O}(n)\), and this operation is clearly invertible.
Definition 2.1
[10, 11, 26] Given \(f : [ \mathbf {a}, \mathbf {b}] \rightarrow X\), we define the n-dimensional Vitali difference of f over an n-dimensional interval \([ \mathbf {x}, \mathbf {y}] \subseteq [ \mathbf {a}, \mathbf {b}] \), by
Note that in the case \(n=2\), we have \(\mathcal {E}(2):= \{(0,0),(1,1)\}\) and \(\mathcal {O}(2)=\{(1,0),(0,1)\},\) thus \( \Delta _2(f,[ \mathbf {x}, \mathbf {y }] ) = d(f(x_1,x_2) + f(y_1,y_2), f(y_1, x_2) + f(x_1,y_2)).\)
Even when \(\Delta _n (f, [ \mathbf {x}, \mathbf {y}] ),\) in (2.3), is defined for \( \mathbf {x} < \mathbf {y}\), note that if \(x_i = y_i\) for some i, then the right-hand side of (2.3) is equal to zero for all maps \(f : [ \mathbf {a}, \mathbf {b}] \rightarrow X\). This difference is also called mixed difference and it is usually associated to the names of Vitali, Lebesgue, Hardy, Krause, Fréchet and De la Vallée Poussin ([10, 11, 20]).
Now, we are going to define the \(\Phi \)-variation of a function \(f:[ \mathbf {a}, \mathbf {b}] \rightarrow X\) (see [5, 6]). To that end, we consider net partitions of \([ \mathbf {a}, \mathbf {b}] = \prod \nolimits _{i=1}^{n} [a_i, b_i]\); that is, partitions of the kind
where \(\{k_i\}_{i=1}^n \subset \mathbb {N}\) and for each i, \(\mathbf {\xi } _i \) is a partition of \([a_i, b_i]\). The set of all net partitions of an interval \([ \mathbf {a}, \mathbf {b}]\) will be denoted by \( \Lambda ([ \mathbf {a}, \mathbf {b}])\).
A point in a net partition \(\mathbf {\xi } \) is called a node ([25]) and it is of the form
where \(\mathbf 0 \le {\varvec{\alpha }} \, = \, (\, \alpha _i \,)_{i=1}^{n} \, \le \, \mathbf {\kappa } \), with \( \mathbf {\kappa } := (k_i)_{i=1}^{n},\) as a result, \(t_{\alpha _i}^{(j)} \in [a_j, b_j]\).
For the sake of simplicity in notation, we will simply write \(\mathbf {\xi } =\{\mathbf {t}_{{ {\varvec{\alpha }}} }\}, \) to refer to all nodes determined by a given partition \(\mathbf {\xi } .\)
A cell of an n-dimensional interval \([ \mathbf {a}, \mathbf {b}]\) is an n-dimensional subinterval of the form \([ \mathbf {t}_{ {\varvec{\alpha }}- \mathbf {1} }, \mathbf {t}_{ {\varvec{\alpha }} }], \) for \(\mathbf 0 < {\varvec{\alpha }}\le \mathbf {\kappa }\).
Note that
3 \( \mathrm{RV}^{n}_{\Phi } ([ \mathbf {a}, \mathbf {b}] ; X)\)
Now we introduce the Riesz-\(\Phi \)-variation of a function f.
Definition 3.1
Let \(f : [ \mathbf {a}, \mathbf {b}] \rightarrow X\) and \(\Phi \in \mathcal {N} \). The \(\Phi \)-variation, in the sense of Vitali–Riesz of f is defined as
where \(\mathbf {\xi } =\{\mathbf {t}_{{ {\varvec{\alpha }}} }\}, \) and
The main objective of this section is to define the Riesz-\(\Phi \)-variation of a function f. To do that it will be necessary to define the variation of a function when we consider that certain variables are fixed, thus as it was done in [11], we now define the truncation of a point, of an interval and of a function, by a given multi-index \(\mathbf {0}\le \eta \le \mathbf {1}\), with \(\mathbf {0} \ne \eta \). Notice that in this case, the entries of a such \(\eta \) are either 0 or 1.
-
The truncation of a point \( \mathbf {x} \in \mathbb {R}^{n}\) by a multi-index \(\mathbf {0}\le \eta \le \mathbf {1}\) with \(\mathbf {0} \ne \eta \), which is denoted by \( \mathbf {x} \lfloor \mathbf {\eta }\), is defined as the \(| \mathbf {\eta }|\)-tuple that is obtained if we suppress from \(\mathbf {x}\) the entries for which the corresponding entries of \(\eta \) are equal to 0. That is, \( \mathbf {x}\lfloor \mathbf {\eta } = (x_i : i \in \{1,2,\ldots ,n\}, \eta _i = 1) \). For instance, if \( \mathbf {x} = (x_1, x_2, x_3, x_4, x_5)\) and \( \mathbf {\eta } = ( 0, 1, 1, 0, 1)\) then \( \mathbf {x}\lfloor \mathbf {\eta } = ( x_2, x_3, x_5).\)
-
The truncation of an n-dimensional interval \([ \mathbf {a}, \mathbf {b}]\) by a multi-index \(\mathbf {0}\le \eta \le \mathbf {1}\) with \(\mathbf {0} \ne \eta \), is defined as \([ \mathbf {a}, \mathbf {b}] \lfloor \mathbf {\eta } := [{ \mathbf {a}\lfloor \mathbf {\eta }},{{ \mathbf {b}\lfloor \mathbf {\eta } }}] \).
-
Given a function \(f: [ \mathbf {a}, \mathbf {b}] \rightarrow X\), a multi-index \(\mathbf {0}\le \eta \le \mathbf {1}\) with \(\mathbf {0} \ne \eta \) and a point \( \mathbf {z} \in [ \mathbf {a}, \mathbf {b}] \), we define \(f_{ \mathbf {\eta }}^{ \mathbf {z}} : [ \mathbf {a}, \mathbf {b}] \lfloor \mathbf {\eta } \rightarrow X \), the truncation of f by \(\eta \), by the formula
$$\begin{aligned} f_{ \mathbf {\eta }}^{ \mathbf {z}}( \mathbf {x}\lfloor \mathbf {\eta }) : = f( \mathbf {\eta } \mathbf {x} + (\mathbf 1 - \mathbf {\eta } ) \mathbf {z}), \, \, \, x \in [ \mathbf {a}, \mathbf {b}]. \end{aligned}$$
Note that the function \(f_{ \mathbf {\eta }}^{ \mathbf {z}}\) depends only on the \(| \mathbf {\eta }|\) variables \(x_i\) for which \(\eta _i = 1\).
Remark 3.2
Given a function \(f:[\mathbf {a},\mathbf {b}] \rightarrow X \) and a multi-index \(\eta \ne \mathbf {0}\), then the \(|\eta |\)-dimensional Vitali difference for \(f_{\eta }^{\mathbf {a}}\) (cf. (2.3)), is given by
The Vitali-type nth variation of \(f: [\mathbf {a},\mathbf {b} ] \rightarrow X\) is defined by
the supremum being taken over all multiindices \(\kappa \) and all net partitions of \([\mathbf {a},\mathbf {b} ]\).
The total variation of \(f: [\mathbf {a}, \mathbf {b}]\rightarrow X\) in the sense of Hildebrandt and Leonov (see [20, 22]) is defined by
the summations here and throughout the paper being taken over n-dimensional multiindices in the ranges specified under the summation sign.
Definition 3.3
Let \(\Phi \in \mathcal {N}\) and let \(f: [ \mathbf {a}, \mathbf {b}] \rightarrow X\) be a function. We call
the \(\Phi \)-variation of f in the sense of Vitali–Hardy–Riesz, briefly: Riesz-\(\Phi \)-variation of f, in \([ \mathbf {a}, \mathbf {b}]\). The set of all functions f satisfying the condition \( \mathrm{TRV}_{\Phi } (f,[ \mathbf {a}, \mathbf {b}] ) < \infty \) will be denoted by \( \mathrm{RV}^{n}_{\Phi } ([ \mathbf {a}, \mathbf {b}] ; X)\).
It is easy to check that if f is a constant function then \( \Delta _n(f,[ \mathbf {x}, \mathbf {y}] )= 0\) and consequently \(\mathrm{TRV}_{\Phi } (f,[ \mathbf {a}, \mathbf {b}] )=0\). In fact, \(\mathrm{TRV}_{\Phi } (f,[ \mathbf {a}, \mathbf {b}] )=0\) if and only if f is constant, as we show next.
Theorem 3.4
\(\mathrm{TRV}_{\Phi } (f,[ \mathbf {a}, \mathbf {b}] )=0\) if and only if f is a constant function.
Proof
We just prove the necessity of the condition. Suppose \(\mathrm{TRV}_{\Phi } (f,[ \mathbf {a}, \mathbf {b}] )=0\) and let \( \mathbf {x}= (x_1,\ldots , x_n)\) be a point in \([ \mathbf {a}, \mathbf {b}]\). Then, \(\mathbf {x}\) determines, for each \(1 \le i \le n\), the partitions \(\mathbf {\xi } _i := \{a_i, x_i, b_i\}:= \{t_0^{(i)},t_1^{(i)}, t_2^{(i)}\} \).
Since \(\mathrm{TRV}_{\Phi } (f,[ \mathbf {a}, \mathbf {b}] ) =0\), we must have \( \Delta _{|\eta |}(f_{\eta }^{\mathbf {a}},[ \mathbf {t}_{\alpha -\mathbf {1}}, \mathbf {t}_{\alpha }] ) = 0 \) for every \(\mathbf {1} \le {\varvec{\alpha }}\le \mathbf {2} \) and every \(\mathbf {0} \ne \eta \le \mathbf {1} \). Consequently, if \(\eta = \mathbf {e}_i\) and \(\alpha = \mathbf {1}\) we obtain
Hence,
On the other hand, if \(\eta = \mathbf {e}_i + \mathbf {e}_j\) with \(i < j\) then
Thus, using (3.3) and (3.2) we have
Equivalently
Now, suppose that (3.4) holds for any multi-index \(\eta \), \(\mathbf {0} \le \eta \le \mathbf {1}\), \(\mathbf {0} \ne \eta \), with k non-zero entries.
Then, if \( \lambda \) is a multi-index such that \(\mathbf {0} \le \lambda \le \mathbf {1}\), \(\mathbf {0} \ne \lambda \), with \(k+1\) non-zero entries and \(\Delta _{|\lambda |}(f_{\lambda }^{\mathbf {a}},[ \mathbf {t}_{\alpha -\mathbf {1}}, \mathbf {t}_{\alpha }] ) = 0 \) for all \(\mathbf {1} \le {\varvec{\alpha }}\le \mathbf {2} \), then
Notice that if \(\theta \ne \mathbf {0}\) then \( \lambda ( \theta \mathbf {t}_\mathbf {0} + (\mathbf {1} - \theta ) \mathbf {t}_{\mathbf {1}}) + (\mathbf {1}-\lambda ) \mathbf {a} \) has at most k entries equal to the corresponding entries of \( \mathbf {x}\) and the remaining entries equal to the corresponding entries of \(\mathbf {a}.\) In this case, (3.4) implies that \(f( \lambda ( \theta \mathbf {t}_\mathbf {0} + (\mathbf {1} - \theta ) \mathbf {t}_{\mathbf {1}}) + (\mathbf {1}-\lambda ) \mathbf {a}) = f(\mathbf {a})\). Hence, since \( \mathcal {E}(n) \) has the same number of elements as \(\mathcal {O}(n)\), it follows from identity (3.4) that \(f\left( \lambda \mathbf {t}_{\mathbf {1}} + (\mathbf {1}-\lambda ) \mathbf {a}\right) = f(\mathbf {a})\). We conclude that f is a constant function. \(\square \)
Remark 3.5
Note that if X is a normed space, then (2.3) can be replaced by
Example 3.6
Let \(c_0\) be the space of all null sequences with the \({\infty }\)-norm, and let \(\, f: [0, 1] \times [0, 1] \times [0, 1] \rightarrow c_0\) be defined by
If \(\mathbf {\xi } = \mathbf {\xi } _1 \times \mathbf {\xi } _2 \times \mathbf {\xi } _3 \), where \(\mathbf {\xi } _i := \{t_1^{(i)},t_2^{(i)},\ldots , t_{k_i}^{(i)} \}\), \(i=1,2,3\). Then,
Hence,
Consequently, for each \( i=1,2,3 \) we must have \( \mathrm{RV}_{\Phi }^{1}(f_{\mathbf {e}_i}^{0},[ \mathbf {a}, \mathbf {b}] ) := \Phi (1).\) In addition,
and therefore, \( \mathrm{RV}_{\Phi }^{2}(f_{(1,1,0)}^{0},[ \mathbf {a}, \mathbf {b}] )= 0.\)
It can be verified similarly that \(\mathrm{RV}_{\Phi }^{2}(f_{\mathbf {e}_1+ \mathbf {e}_2+\mathbf {e}_3}^{\mathbf {0}},[ \mathbf {a}, \mathbf {b}] )=\mathrm{RV}_{\Phi }^{2}(f_{\mathbf {e}_i+ \mathbf {e}_j}^{\mathbf {0}},[ \mathbf {a}, \mathbf {b}] )= 0\), where \(i, j =1,2,3\) with \(i\ne j\).
From (3.2) we conclude that \(\mathrm{TRV}_{\Phi } (f,[ \mathbf {a}, \mathbf {b}] ) = 3 \Phi (1).\)
A proof of the following lemma can be found in [5] or in [6].
Lemma 3.7
[5, 6] If X is a normed space, then the functional \( \mathrm{TRV}_{\Phi } (\cdot ,[ \mathbf {a}, \mathbf {b}] )\) is convex.
4 Further properties
Theorem 4.1
Let X be a normed space. If \(f \in \mathrm{RV}_{\Phi } ([ \mathbf {a}, \mathbf {b}] ; X)\), then f is bounded.
Proof
It is well known that if \(f: [a,b] \rightarrow X, \, [a,b] \subset \mathbb {R}, \) is a function of bounded Riesz-\(\Phi \)-variation, then f is bounded (see [15]). Hence, if \(f: [ \mathbf {a}, \mathbf {b}] \rightarrow X\) is a function of bounded \(\Phi \)-variation on \([ \mathbf {a}, \mathbf {b}]:= [a_1, b_1] \times [a_2,b_2] \) then there are constants \(C_1\) and \(C_2\) such that
Suppose that f is not bounded. Then, for each \(m \in \mathbb {N}\) that satisfies \(m > C_1 + C_2\), there exists \( \mathbf {x}_m = (x_1^{m}, x_2^m) \in [ \mathbf {a}, \mathbf {b}]\) such that \(\Vert f( \mathbf {x}_m) + f( \mathbf {a})\Vert \ge m\). Hence, for all \(m > C_1 + C_2\),
If \(\mathrm{Vol} \, [ \mathbf {a}, \mathbf {x}] \le 1\), then (4.1) implies
a contradiction (since \( \lim \nolimits _{m\rightarrow \infty }\Phi (m - C_1 - C_2) =\infty \)).
On the other hand, if \(\mathrm{Vol} \, [ \mathbf {a}, \mathbf {x}]> 1\) then from (4.1) we obtain
which again leads to a contradiction; thus, f is a bounded function.
Suppose that the result holds for all the cases in which \([ \mathbf {a}, \mathbf {b}]\) is a k-dimensional interval with \(k < n\).
Consider now the case in which \(f: [ \mathbf {a}, \mathbf {b}] \rightarrow X\) where \([ \mathbf {a}, \mathbf {b}]\) is an n-dimensional interval. Then, for all multi-index \(\mathbf {0}< \mathbf {\eta }\le \mathbf {1} (\mathbf {0}< \mathbf {\eta })\) that satisfies \(| \mathbf {\eta }| < n\) there exists a constant \(M_{\eta }\) such that
In this case, if we suppose that f is not bounded, then for all \(m > \sum \nolimits _{\mathbf {0}< \mathbf {\eta }< \mathbf {1}} M_{\eta }\) there is a point \( \mathbf {x}_m = (x_1^{m}, x_2^{m}, \ldots , x_n^{m})\) such that
and hence
The result now follows, as in the \(n=2\) case, from the fact that
again, by considering the two cases \(\mathrm{Vol}([a, x_m]) \le 1\) and \(\mathrm{Vol}([a, x_m]) > 1\). \(\square \)
Theorem 4.2
Let X be a normed space. Let \(\Phi _1, \Phi _2 \in \mathcal {N}\) such that \(\Phi _1(x) \le K \Phi _2(x) \) for all x and some constant K, then \(\mathrm{RV}_{\Phi _2} ([ \mathbf {a}, \mathbf {b}] ; X) \subseteq \mathrm{RV}_{\Phi _1} ([ \mathbf {a}, \mathbf {b}] ; X).\)
Proof
Let \(f \in \mathrm{RV}_{\Phi _2} ([ \mathbf {a}, \mathbf {b}] ; X)\), then for all net partition \(\mathbf {\xi } =\{\mathbf {t}_{{ {\varvec{\alpha }}} }\} \in \Lambda ([ \mathbf {a}, \mathbf {b}])\) we have
from which the proposition follows. \(\square \)
Theorem 4.3
Let \(\Phi \in \mathcal {N}\) satisfying condition \(\Delta _2\) and let X be a normed space. Then \(\mathrm{RV}_{\Phi } ([ \mathbf {a}, \mathbf {b}] ; X)\) is a linear space.
Proof
Let K and \(x_0 \in \mathbb {R}\) be as in (2.2) and suppose \(f,g \in \mathrm{RV}_{\Phi } ([ \mathbf {a}, \mathbf {b}] ; X) \). Then, for any net partition \(\mathbf {\xi } =\{\mathbf {t}_{{ {\varvec{\alpha }}} }\} \in \Lambda ([ \mathbf {a}, \mathbf {b}])\)
Put \( A_n = \dfrac{\Delta _n \left( f, [ \mathbf {t}_{ {\varvec{\alpha }} - \mathbf {1}}, \mathbf {t}_{ {\varvec{\alpha }}}] \right) + \Delta _n \left( g, [ \mathbf {t}_{ {\varvec{\alpha }} - \mathbf {1}}, \mathbf {t}_{ {\varvec{\alpha }}}] \right) }{\mathrm{Vol} \, [ \mathbf {t}_{ {\varvec{\alpha }} - \mathbf {1}}, \mathbf {t}_{ {\varvec{\alpha }}}]}\), then
Since this holds for all n, it follows that
from which we conclude that \(f+g \in \mathrm{RV}_{\Phi } ([ \mathbf {a}, \mathbf {b}] ; X)\). On the other hand, if \(\gamma \) is any scalar, then
Thus,
If \(|\gamma | \le 1\), then
On the other hand, if \(|\gamma | > 1\) then, again by (2.2), there is a constant \(K'\) and a point \(x_0\) such that
It follows that \(\gamma f \in \mathrm{RV}_{\Phi } ([ \mathbf {a}, \mathbf {b}] ; X)\). We conclude that \(\mathrm{RV}_{\Phi } ([ \mathbf {a}, \mathbf {b}] ; X)\) is a linear space. \(\square \)
Lemma 4.4
Let X be a metric semigroup and suppose \(f \in \mathrm{RV}_{\Phi } ([ \mathbf {a}, \mathbf {b}] ; X)\). If \( \mathbf {x}\),\( \mathbf {y} \in [ \mathbf {a}, \mathbf {b}]\) are such that \(x_k = y_k\) for some \(0 \le k \le n\), then
Proof
Let f, \( \mathbf {x}\) and \( \mathbf {y}\) be as in the statement. If \(\theta = (\theta _1, \theta _2, \ldots , \theta _n) \in \mathcal {E}(n)\) then \(\widetilde{\theta } := (\theta _1, \theta _2, \ldots , 1-\theta _k, \ldots , \theta _n) \in \mathcal {O}(n)\), thus, \(( \theta \, \mathbf {x} + (\mathbf {1} - \theta ) \mathbf {y}) \) has the same entries as \(( \widetilde{\theta } \, \mathbf {x} + (\mathbf {1} - \widetilde{\theta }) \mathbf {y})\) since the kth entry \(\widetilde{\theta }_k \, x_k + (1 - \widetilde{\theta }_k) y_k = (1 - \theta _k) \, x_k + \theta _k y_k = (1 - \theta _k) \, y_k + \theta _k x_k.\) Hence,
\(\square \)
Lemma 4.5
Let X be a metric semigroup and let \([ \mathbf {x}, \mathbf {y}]\) be an n-dimensional interval. Suppose that \(t^{(j)} \in [x_j, y_j]\) for some \(1 \le j \le n\). Then
where \(\mathbf {\widetilde{x}}:=(\widetilde{x}_1, \ldots , \widetilde{x}_n)\) with \(\widetilde{x}_i = x_i\) if \(i \ne j\) and \(\widetilde{x}_j = t^{(j)}\), and \(\mathbf {\widetilde{y}}:=(\widetilde{y}_1, \ldots , \widetilde{y}_n)\) with \(\widetilde{y}_i = y_i\) if \(i \ne j\) and \(\widetilde{y}_j = t^{(j)}\).
Proof
Note that the interval \([ \mathbf {x}, \mathbf {y}]\) can be divided into two intervals, namely \([ \mathbf {x}, \widetilde{\mathbf {y}}]\) and \([\widetilde{\mathbf {x}}, \mathbf {y}]\), thus, by virtue of property (2.1) and lemma 4.4 we have
\(\square \)
Theorem 4.6
Let X be a metric semigroup. If \(f:[ \mathbf {a}, \mathbf {b}] \rightarrow X\) is a function, then
for any net partition \(\mathbf {\xi } =\prod \nolimits _{i=1}^{n} \mathbf {\xi } _i\) of \([ \mathbf {a}, \mathbf {b}]\).
Proof
Suppose \(\mathbf {\xi } = \prod \nolimits _{i=1}^{n} \mathbf {\xi } _i\) where \(\mathbf {\xi } _i = \{a_i= t_0^{(i)}, t_1^{(i)}, \ldots , t_{k_i}^{(i)} = b_i\}\). Assume \(z \in \{1, \ldots , n\}\) and let \( t^{(z)} \) be such that \(t_0^{(z)}< t_1^{(z)}<\dots< t_{r-1}^{(z)}< t^{(z)}< t_r^{(z)}< \dots < t_{k_j}^{(z)}.\) Put \(\varrho := \prod \nolimits _{i=1}^{n} \varrho _i\) where \(\varrho _i = \{a_i= s_0^{(i)}, s_2^{(i)}, \ldots , s_{\widehat{k}_i}^{(i)} = b_i\}\), with
Then,
Hence,
Now if \(\widetilde{t}_{\alpha _i}^{(i)}=t_{\alpha _i}^{(i)}\) for \(i\ne z\) and \(\widetilde{t}_{\alpha _z}^{(z)}=t^{(z)}\), by Lemma 4.5, we obtain
and, by the convexity of \(\Phi \) we get:
It follows that
and therefore, \( \mathrm{RV}^{n}_{\Phi }\left( f,[ \mathbf {a}, \mathbf {b}], \mathbf {\xi }_1 \times \cdots \times \mathbf {\xi } _{z}\cup \{t^{(z)}\} \times \cdots \times \mathbf {\xi } _n \right) \ge \mathrm{RV}^{n}_{\Phi } (f,[ \mathbf {a}, \mathbf {b}], \prod \nolimits _{i=1}^{n} \mathbf {\xi } _i ). \) \(\square \)
Corollary 4.7
Let X be a metric semigroup. If \(f:[ \mathbf {a}, \mathbf {b}] \rightarrow X\), \(\mathbf {\xi } \) and \(\delta \) are any net partitions of \([ \mathbf {a}, \mathbf {b}]\), such that \(\mathbf {\xi } \subseteq \delta , \) then \( V^{n}(f,[ \mathbf {a}, \mathbf {b}], \mathbf {\xi } ) \le V^{n}\left( f,[ \mathbf {a}, \mathbf {b}], \delta \right) . \)
Proof
It suffices to apply Theorem 4.6 finitely many times. \(\square \)
5 A representation theorem
In our next result, we present a counterpart of the classical Riesz’s lemma (cf., [3, 24]) in the n-dimensional case. It will be assumed that \(X=\mathbb {R}\) and the following well-known notation will be used: given any multi-index \(\beta = (\beta _1, \ldots , \beta _k)\) we define
Theorem 5.1
Let \(\, \Phi \in \mathcal {N}.\, \) If \( \, f \in \mathrm{RV}_{\Phi } ([ \mathbf {a}, \mathbf {b}] ; \mathbb {R})\) and is of class \(C^{n}\), then
Proof
Let \(\mathbf {\xi } _i := \{t_1^{(i)},t_2^{(i)}, \ldots , t_{k_i}^{(i)}\}\), \(i=1,\ldots ,n\) and let \(\mathbf {\xi } = \prod \nolimits _{i=1}^{n} \mathbf {\xi } _i= \{\mathbf {t}_{\alpha }\}\) be a net partition of \([ \mathbf {a}, \mathbf {b}]\) with \( \mathbf {\kappa } = (k_1, k_2, \ldots , k_n)\). Then, for all \( {\varvec{\alpha }} \le \mathbf {\kappa }\) we have
where \( z \le k_z\).
Let \( \mathbf {x} = (x_1, x_2, \ldots , x_n)\) , \( \mathbf {v}= \theta \mathbf {t}_{( {\varvec{\alpha }}-1)} + (\mathbf {1} - \theta ) \mathbf {t}_{ {\varvec{\alpha }}}\) and define the function
(\(\mathbf {e}_i\) denotes the canonical unit vectors of \(\mathbb {R}^{n}\)). Then, since f is differentiable, \(g_1: [t_{\alpha _1-1}^{1}, t_{\alpha _1}^{1}] \rightarrow \mathbb {R}\), satisfies the conditions of the ordinary mean value theorem and thus, there is an \(x_{\alpha _1}^{1} \in (t_{\alpha _1-1}^{1}, t_{\alpha _1}^{1}) \) such that,
That is,
Now define \(g_2: [t_{\alpha _2-1}^{2}, t_{\alpha _2}^{2}] \rightarrow \mathbb {R}\), by
Then, as before, since \(g_2\) depends only in the second variable, \(x_2\), an application of the mean value theorem implies that there is \(x_{\alpha _2}^{2} \in (t_{\alpha _2-1}^{2}, t_{\alpha _2}^{2})\) such that
Thus,
By repeating this procedure n times, we obtain that there is \( \mathbf {x}_{\alpha } \in [ \mathbf {t}_{\alpha -1}, \mathbf {t}_{\alpha }]\) such that
and hence
Since this holds for each \(\mathbf {t}_{\alpha }\), \( {\varvec{\alpha }}\le \mathbf {\kappa }\), of any net partition \(\mathbf {\xi } \) of \([ \mathbf {a}, \mathbf {b}]\), we must have
Now define,
then
Notice that the lower sum, \(\underline{S}\), and \(\mathrm{RV}_{\Phi }^{n}(f,[ \mathbf {a}, \mathbf {b}],\mathbf {\xi } )\) are increasing with respect to refinements of the partition \(\mathbf {\xi } \) while the upper sums, \( \overline{S}\), are decreasing. This means that if \(k_i \rightarrow \infty \) then the upper sums decrease to the limit
and the lower sums increase to the limit
whereas \(\mathrm{RV}_{\Phi }^{n}(f,[ \mathbf {a}, \mathbf {b}],\mathbf {\xi } )\) increases to the limit \(\mathrm{RV}_{\Phi }^{n}(f,[ \mathbf {a}, \mathbf {b}])\), and consequently
Now, since this holds for any function of n variables, with \(n\ge 1\), in particular it holds for any truncated function \(f_{\eta }^{ \mathbf {a}}\), where \(\eta \le \mathbf {1}\), which yields (5.1), since there are \((n-|\eta |)!\) truncations. \(\square \)
References
Adams, C.R., Clarkson, J.A.: Properties of functions \(f(x, y)\) of bounded variation. Trans. Am. Math. Soc. 36, 711 (1934)
Ambrosio, L., Fusco, N., Pallara, D.: Functions of Bounded Variation and Free Discontinuity Problems, Oxford Mathematical Monographs. The Clarendon Press, Oxford University Press, New York (2000)
Azíz, W.: Algunas extensiones a \(\mathbb{R}^{2}\) de la noción de funciones con \(\varphi \)-variación acotada en el sentido de Riesz y controlabilidad de las RNC, Tesis Doctoral para optar la título de Doctor en Ciencias, Facultad de Ciencias-UCV (2009)
Bracamonte, M., Giménez, J., Merentes, N.: Vector valued functions of bounded bidimensiona \(\Phi \)-Variation. Ann. Funct. Anal. 4(1), 89–108 (2013)
Bracamonte, M., Ereú, J., Giménez, J., Merentes, N.: Uniformly continuous superposition operators in the space of functions of bounded \(n\)-dimensional \(\Phi \)-variation. Demonstr. Math. 47(1), 56–68 (2014)
Bracamonte, M., Ereú, J., Giménez, J., Merentes, N.: Uniformly bounded superposition operators in the space of functions of bounded \(n\)-dimensional \(\Phi \)-variation. Extr. Math. 29(1–2), 19–33 (2014)
Brudnyi, Y.: Multivariate functions of bounded (k, p)-variation, Banach spaces and their applications in analysis, Walter de Gruyter, Berlin, pp. 3757 (2007)
Bugajewska, D.: On the superposition operator in the space of functions of bounded variation, revisted. Math. Comput. Model. 52(5–6), 791–796 (2010)
Calderón, A.-P., Zygmund, A.: On the differentiability of functions which are of bounded variation in Tonelli’s sense. Rev. Univ. Mat. Argent. 20, 102–121 (1962)
Clarkson, J.A., Adams, C.R.: On definitions of bounded variation for functions of two variables. Trans. Am. Math. Soc. 35, 824–854 (1933)
Chistyakov V.V., Tretyachenko Y.: Maps of several variables of finite total variation and Helly-type selection principles. Part I in: J. Math. Ann. Appl. 370(2), 672–686 (2010), and Part II: in J. Math. Ann. Appl. 369(1), 82–93 (2010)
Chistyakov V.V.: Functions of several variables of finite variation and superposition operators. In: Real Analysis Exchange 26th Summer Symposium, Lexington, VA, USA, pp. 61–66 (2002)
Chistyakov V.V.: A selection principle for mappings of bounded variation of several variables. In: Real Analysis Exchange 27th Summer Symposium, Opava, Czech Republic, pp. 217–222 (2003)
Chistyakov, V.V.: Selections of bounded variation. J. Appl. Anal. 10(1), 1–82 (2004)
Chistyakov, V.V.: Generalized variation of mappings with applications to composition operators and multifunctions. Positivity 5(4), 323–358 (2001)
Cohen, A., Dahmen, W., Daubechies, I., DeVore, R.: Harmonic analysis of the space BV. Rev. Mat. Iberoam. 19(1), 235–263 (2003)
De Giorgi, E.: Su una teoria generale della misura n-1n-1-dimensionale in uno spazio a rr dimensioni. Ann. Mat. Pura Appl. (4). 36, 191–213 (1954) (Zbl 0055.28504)
Federer, H.: Geometric measure theory. Volume 153 of Die Grundlehren der mathematischen Wissenschaften. Springer, New York Inc., New York (1969) (Zbl 0874.49001)
Fichera, G.: Lezioni sulle trasformazioni lineari. Istituto matematico dell’Universit di Trieste. vol. I (1954)
Hildebrandt, T.H.: Introduction to the Theory of Integration. Academic Press, New York (1963)
Jordan, C.: Sur la srie de Fourier, Comptes rendus hebdomadaires des sances de l’Acadmie des sciences 92: 228–230 (1881) (at Gallica) (This is, according to Boris Golubov, the first paper on functions of bounded variation)
Leonov, A.S.: On the total variation for functions of several variables and a multidimensional analog of Helly’s selection principle. Math. Notes 63(1), 61–71 (1998)
Maligranda L.: Orlicz Spaces and Interpolation, Seminars in Math. 5, University of Campinas, IMECC-UNICAMP, Brasil (1989)
Riesz, F.: Untersuchungen über Systeme integrierbarer Funktionen. Math. Ann. 69, 449–497 (1910)
Talalyan, F.A.: A multidimensional analogue of a theorem of F. Riesz, Sb. Math. 186(9), 1363–1374 (1995)
Vitali G.: Sui gruppi di punti e sulle funzioni di variabili reali. Atti Accad. Sci. Torino 43, 75–92(1908)
Ziemer, W.P.: Weakly Differentiable Functions, Graduate Texts in Mathematics, Sobolev spaces and functions of bounded variation. vol. 120, Springer, New York (1989)
Acknowledgments
The authors would like to thank the referee of the first version of this paper for his/her valuable comments and suggestions.
Author information
Authors and Affiliations
Corresponding author
Additional information
This research has been partly supported by the Central Bank of Venezuela.
Rights and permissions
About this article
Cite this article
Bracamonte, M., Ereú, J., Giménez, J. et al. On metric semigroups-valued functions of bounded Riesz-\(\Phi \)-variation in several variables. Bol. Soc. Mat. Mex. 24, 133–153 (2018). https://doi.org/10.1007/s40590-016-0138-2
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s40590-016-0138-2