1: \documentclass[10pt]{amsart}

2: \usepackage{amsthm,amssymb}

3: \usepackage{graphicx}

4: %\usepackage{fullpage}

5: 

6: 

7: \textwidth 150mm

8: \oddsidemargin 5mm

9: \evensidemargin 5mm

10: \textheight 230mm

11: \topmargin -10mm

12: 

13: \newtheorem{lemma}{Lemma}

14: \newtheorem{theorem}{Theorem}

15: \newtheorem{definition}{\rm D e f i n i t i o n}

16: \newtheorem{corollary}{corollary}

17: 

18: \theoremstyle{definition}

19: 

20: 

21: \newtheorem{example}{\rm E x a m p l e}

22: \newtheorem{property}{Property}

23: \newtheorem{remark}{\rm R e m a r k}

24: 

25: \newcommand{\supp}{\mbox{supp}}

26: \newcommand{\ordin}{\mbox{ord}}

27: \newcommand{\const}{\mbox{const}}

28: \newcommand{\sign}{\mbox{sign}}

29: \newcommand{\diag}{\mbox{diag}}

30: \newcommand{\cl}{\mbox{cl}}

31: \newcommand{\dm}{\mbox{dm}}

32: \newcommand{\loc}{\mbox{loc}}

33: \newcommand{\var}{\mbox{var}}

34: \newcommand{\ang}{\mbox{ang}}

35: \newcommand{\dyn}{D}

36: \newcommand{\us}{U}

37: 

38: 

39: 

40: \begin{document}

41: 

42: 

43: 

44: \title[Distributions with dynamic test functions]{Distributions with dynamic test functions and 

45: multiplication by discontinuous functions}

46: 

47: 

48: \author[V.Derr] {V.~Derr}

49: 

50: \address{Faculty of Mathematics, Udmurtia State University \\ Universitetskya St., 1 (building 4), Izhevsk, 426034, Russia}

51: 

52: \email{derr@uni.udm.ru}

53: 

54: \author[D.Kinzebulatov] {D.~Kinzebulatov}

55: 

56: \address{Department of Mathematics and Statistics, University of Calgary \\ 2500 University Drive N.W., Calgary, Alberta, Canada T2N 1N4}

57: 

58: \email{knzbltv@udm.net}

59: 

60: \subjclass[2000]{46F10, 34A36}

61: 

62: 

63: \keywords{Distribution theory, product of distributions, Colombeau generalized functions algebra, distributions with discontinuous test functions, ordinary differential equations with distributions}

64: 

65: \begin{abstract}

66: As follows from the Schwartz Impossibility Theorem, multiplication of two distributions is in general impossible.

67: Nevertheless, often one needs to multiply a distribution by a discontinuous function, not by an arbitrary distribution. In the present paper we construct a space of distributions where the general operation of multiplication by a discontinuous function is defined, continuous, commutative, associative and for which the Leibniz product rule holds. In the new space of distributions, the classical delta-function $\delta_\tau$ extends to a family of delta-functions $\delta_\tau^\alpha$, dependent on the \textit{shape} $\alpha$.

68: We show that the various known definitions of the product of the Heaviside function and the  delta-function in the classical space of distributions $\mathcal D'$ become particular cases of the multiplication in the new space of distributions, and provide the applications of the new space of distributions to the ordinary differential equations which arise in optimal control theory. 

69: Also, we

70: compare our approach of the Schwartz distribution theory with the approach of the

71: Colombeau generalized functions algebra, where the general operation of multiplication of two distributions is defined.

72: %It is shown that the multiplication in the new space of distributions has certain advantages in comparison with the multiplication in the Colombeau generalized functions algebra, which are important for the

73: %applications to ordinary differential equations with distributions.

74: 

75: 

76: 

77: 

78: \end{abstract}

79: 

80: 

81: \maketitle

82: 

83: 

84: \section{Introduction}

85: 

86: %\subsection{Schwartz Impossibility Theorem and multiplication of distributions}

87: The theory of distributions created by L.~Schwartz \cite{Scw} in the 1950s is an important part of modern mathematics.

88: However, in 1962 R.~Courant pointed out certain insufficiencies of the Schwartz distributions as generalizations of functions, in particular, the absence of a general multiplication of distributions \cite{Cour}. This multiplication is important for the definition of a solution for a wide class of ordinary and partial differential equations \cite{Col1}. It was mentioned  by J.F.Colombeau in \cite{Col1} that there is a strong belief by many mathematicians that a general multiplication of distributions is impossible in the Schwartz distribution theory, unless we drop essential properties (such as continuity), which is not acceptable for applications to differential equations.

89: %

90: %This defect of the Schwartz distribution theory leads to the construction of the Colombeau generalized functions algebra \cite{Col1}, containing the space of distributions as a subspace, where such general multiplication exists, though the result of this multiplication is not a distribution, in the general case (for example, the product of the Heaviside function $\theta_\tau$, which is discontinuous at $\tau$, and the delta-function $\delta_\tau$, is no longer a distribution). The algebra of Colombeau generalized functions was applied successfully to many problems related to ordinary and partial differential equations \cite{Col1,Col2,Col3,Col4}. However, the Colombeau generalized functions algebra approach is insufficient for some important classes of ordinary differential equations containing products of distributions and discontinuous functions, which arise, for example, in optimal control theory (e.g. see \cite{Silv,Ramp,Ses}). 

91: 

92: Along with that, in many cases in the theory of differential equations one needs to be able to multiply a distribution not by an arbitrary distribution, but by a discontinuous function only (which is also a distribution). For instance, the product of a distribution and a discontinuous function arises in ordinary differential equations and partial differential equations of optimal control theory, e.g. see \cite{Ramp,Silv,Mil,Der3}, where a distribution represents an impulsive control. 

93: In our paper we construct a new space of distributions which are extensions of the classical distributions from the space of continuous test functions to the space of \textit{dynamic test functions}. In this space the

94: operation of multiplication of a distribution by a discontinuous function is defined, commutative, associative and continuous. 

95: %Our study is motivated by the problems related to the nonlinear ordinary differential equations with distributions, which arise, in particular, in optimal control theory \cite{Ses,Ramp,Silv,Mil}. 

96: 

97: Further we provide some applications of a new space of distributions to the theory of ordinary differential equations. 

98: Also, let us mention that since the operation of multiplication of a distribution and a discontinuous function can be induced by the operation of multiplication of two distributions, its properties allow us to compare, in a certain sense, various definitions of the product of two distributions in the classical space of distributions with continuous test functions $\mathcal D'$, which were proposed, in particular, in \cite{Bag1,Bag2,Sar,Sar2,Sar3}, in attempt to obtain a better definition.

99: 

100: 

101: 

102: The operation of multiplication of a distribution by a discontinuous function (as a particular case of a general operation of multiplication of two distributions) is defined in the Colombeau generalized functions algebra, which was applied successfully to many problems related to ordinary and partial differential equations \cite{Col1,ColM1,ColM2,ColM3, Col3, Col2,Col4}.

103: In the present paper we show that the commutative, associative and continuous operation of multiplication of a distribution by a discontinuous function can be also defined in the classical setting of the Schwartz distribution theory. Below we

104: compare our approach based on the Schwartz distribution theory with the approach of the

105: Colombeau generalized functions algebra (Section 4).

106: 

107: %The distributions from the constructed space of distributions are applied to problems related to ordinary differential equations with distributions.

108: 

109: %Further we show some advantages of our space of distributions in comparison with the Colombeau generalized functions algebra, where multiplication by a discontinuous function is also defined. Namely,  solutions of some differential equations in the Colombeau generalized functions algebra may be generalized functions and as a result often do not have a natural physical interpretation \cite{Col1}, whereas in our space of distributions solutions of differential equations are the ordinary functions (see Appendix).

110: 

111: %In the subsequent paper we present some applications of the new space of distributions to problems related to the theory of ordinary differential equations with distributions.

112: 

113: 

114: \subsection{Continuity properties of the operation of multiplication}

115: %In the present paper we construct a space of distributions where the operation of multiplication of distributions and discontinuous functions is defined. 

116: 

117: Let us describe some natural requirements imposed on continuity properties of the operation of multiplication.

118: % on discontinuous function imposed by ordinary differential equations arising in control theory. 

119: Below the term ``regular distribution'' refers to a distribution which corresponds to a locally-summable function.

120: 

121: The operation of multiplication of a distribution by a discontinuous function has to have the following property: \\ [3mm]

122: % motivation for construction of this space of distributions were ordianry differential equations with distributions

123: %containing product of distribution and discontinuous function \cite{Ses}. 

124: %Hence, the 

125: %necessity of certain properties of operation of multiplication is motivated by 

126: %consideration of differential equations with distributions of a certain type, namely which are affine by distribution. \\ [3mm]

127: %In the definitions of properties below it is assumed that operation of multiplication

128: %has the property if it has it in certain topologies. Since the space of distributions is not constructed yet, term \textit{regular distribution} refers 

129: %to a distribution corresponding to a usual function in this space of distributions. \\ [3mm]

130: \textbf{Property $\mathcal A$}.                                             

131: \textit{If $g$ is a discontinuous function, $v$ is a distribution, $\{v_n\}_{n=1}^\infty$ is a sequence of regular distributions, $v_n \to v$,

132: then $gv_n \to gv$. If $v$ is a regular distribution, then $gv$ 

133: is regular and corresponds to the usual product of $g$ and $v$}. \\ [-2mm]

134: 

135: A space of distributions where the operation of multiplication by a discontinuous function has property $\mathcal A$ was constructed in \cite{DerKin4}. 

136: This operation of multiplication arises from the following ordinary differential equation \cite{Mil}, \\ [-4mm]

137: \begin{equation}

138: \label{intro_eq1}

139: \dot{x}=f(t,x)+g(t,u(t))v

140: \end{equation}

141: where $v$ is a distribution (impulsive control), $f$, $g$ are continuously differentiable, $u$ is a control, which is in general discontinuous function \cite{Mil}. 

142: %Since the product of distribution and discontinuous function does not exists in the classic space of distributions, solution of Cauchy problem for (\ref{intro_eq1}) is defined as a limit in the weak topology of space of functions of bounded variation of the sequence

143: %of solutions of approximating Cauchy problems with continuous functions.

144: 

145: However, property $\mathcal A$ is too weak for consideration in the following ordinary differential equation with distributions which arises in many problems of optimal control theory \cite{Ses,Silv,Ramp}, \\ [-4mm]

146: \begin{equation}

147: \label{intro_eq2}

148: \dot{x}=f(t,x)+g(t,x)v, 

149: \end{equation}

150: where $v$ is a distribution, $f$, $g$ are continuously differentiable. %solution of a Cauchy problem for (\ref{intro_eq2})

151: %is defined similarly as a limit in the weak topology of the space of %functions of bounded variation of the sequence

152: %of solutions of approximating Cauchy problems with usual functions \cite{Dyh,Ses,Vin}, 

153: In the general case, if the distribution $v$ is not regular, it is natural to expect that a solution $x$ is discontinuous \cite{Ses,Silv,Ramp,Mil}. Thus, 

154: the product of a distribution and a discontinuous function in (\ref{intro_eq2}) must have the following property.  \\ [1mm]

155: \textbf{Property $\mathcal B$}.                                                                                                                                                                                   

156: \textit{If $g$ is a discontinuous function, $v$ a distribution, $\{v_n\}_{n=1}^\infty$ a sequence of regular distributions, $v_n \to v$, and $\{g_n\}_{n=1}^\infty$ 

157: is a sequence of continuous functions, $g_n \to g$,

158: then $g_nv_n \to gv$. If $v$ is a regular distribution, then 

159: $gv$ is regular and corresponds to the usual product of $g$ and $v$}. \\ [-2mm]

160: %The approach to consideration of (\ref{intro_eq1}) and (\ref{intro_eq2}) used in \cite{Ses} is to consider (\ref{intro_eq1}) and (\ref{intro_eq2}) in 

161: %$\mathcal D'$ and to define solution as above results in incorrectness of notation since the product of dsitribution and discontinuous functions is not defined in $\mathcal D'$;

162: %consequently, in \cite{Ses} delta-function in $v$ in (\ref{intro_eq1}) and (\ref{intro_eq2}) is a symbol.

163: %Notice, that if $\mathcal D'$ is a Schwartz space of distributions with continuous test functions, then $v=\dot{u}$,

164: %$u$ has locally bounded variation, and problem of existence of product of distribution and discontinuous function is in certain sense equavalent to existence of 

165: %measure with non-zero singular component having discontinuous density.

166: 

167: 

168: Thus, in order to define the continuous operation of multiplication of distributions by discontinuous functions possessing property $\mathcal B$, which is important for the applications to ordinary differential equations, one needs to distinguish different approximations $\{v_n\}_{n=1}^\infty$ of the distribution $v$ in $\mathcal D'$, and different approximations $\{g_n\}_{n=1}^\infty$ of the discontinuous function $g$. 

169: 

170: This idea was formalized in \cite{ColM1,ColM2,ColM3} (see the references therein)

171: in the framework of the Colombeau generalized functions theory

172: %%% ubr

173: (see more detailed discussion in Section 4) and, in less generality, in \cite{Mil2,Mil,Bre,Ses} (see further references therein),

174: where the consideration of the sequence $\{v_n\}_{n=1}^\infty$ as a part of the system (\ref{intro_eq2}) in order to provide the uniqueness of solution was proposed. 

175: 

176: %Also, we should mention paper \cite{Ole}, where the idea of the choice of approximation of delta-function is formalized in the framework of the nonstandard analysis.

177: 

178: In the present paper we formalize this idea in the classical setting of the Schwartz distribution theory, i.e., in the \textit{space of distributions with dynamic test functions}, where

179: the operation of multiplication by discontinuous function has property $\mathcal B$. %and show certain advantages of our approach in comparison with some known approaches.

180: % dob

181: 

182: 

183: 

184: %Notice that if $v$ can be represented as a series of delta-functions, then the set of points of concentration of delta-functions in $v$ and even its cardinality, as well as the set of points of discontinuity of $x$, can not be fixed a priori, since in many optimal control problems it is a subject of optimization \cite{Ses,Dyh}. This leads to a necessity of consideration of the space of functions of bounded variation, and, as a result, in certain complication in contrast to the case when the set of points of discontinuity of a function, which is multiplied on distribution, is finite and fixed a priori.

185: 

186: 

187: %In subsequent sections we construct the space of distributions where the operation of multiplication of its 

188: %element on discontinuous function having general property B is defined. 

189: %Also, we compare the operation of multiplication in this space with definitions of such multiplication based on Schwartz space of distributions $\mathcal D'$, and with multiplication in

190: %the space of Colombeau new generalized functions. 

191: %The major approach to consideration of these differential equations in the space of Schwartz distributions $\mathcal D'$ results in incorrectness of notation since

192: %there is no operation of multiplication of distribution on discontinuous function defined in $\mathcal D'$.                            

193: %(the \textit{space of distributions with discontinuous test functions} with 

194: %operation of multiplication having property A is constructed in \cite{DerKin}, see also \cite{Kur}.

195: 

196: 

197: 

198: \subsection{Distributions with dynamic test functions}

199: Let us briefly describe our approach. In distribution theory (as a theory of space $\mathcal D'$) the product of a distribution and a continuous function is defined by

200: \begin{equation}

201: \label{proddef}

202: (gf,\varphi)=(f,g\varphi)

203: \end{equation}

204: where $f$ is a distribution, $g$ is a continuous function and $\varphi$ is a continuous test function. If $g$ is discontinuous, then $g\varphi$ is no longer a test function, thus the value of the right-hand side of (\ref{proddef}) is undefined. In order to employ the definition (\ref{proddef}), we need to extend the space of test functions so that it contains discontinuous functions. In this case the value of the right-hand side of (\ref{proddef}) would be defined for discontinuous $g$ also. 

205: 

206: In the present paper we construct the space of distributions with the test functions possibly discontinuous.

207: %Thus, we need to consider the space of distributions with discontinuous test functions, which is constructed in this paper. 

208: We show that any classical distribution can be extended from the space of continuous test functions to our space of test functions. 

209: %In this sense our space of distributions does not contain any new elements in comparison with the classical space of distributions \cite{Scw}.

210: In particular, the classical delta-function $\delta_\tau$ extends into a family of delta-functions $\{\delta_\tau^\alpha\}$, where the parameter 

211: $\alpha$

212: such that 

213: $\int_{-1/2}^{1/2} \alpha(t)dt=1$, 

214: is called the \textit{shape} of a delta-function $\delta_\tau^\alpha$.

215: In the new space of distributions the product of the Heaviside function $\theta_\tau$ and the delta-function $\delta_\tau^\alpha$ is given by \\ [-2mm]

216: \begin{equation}

217: \label{proddef2}

218: \theta_\tau\delta_\tau^\alpha=\biggl(\int_0^{1/2}\alpha(t)dt \biggr)\delta_\tau^\beta, \\ [-1mm]

219: \end{equation}

220: where $\int_0^{1/2}\alpha(t)dt \in \bf R$ is a constant, and $\beta$ is another shape. Similarly, the family of delta-sequences extends into subfamilies of delta-sequences having different shapes. Then a result similar to (\ref{proddef2}) can be obtained if delta-functions in (\ref{proddef2}) are replaced by terms of delta-sequences having the same shapes.

221: 

222: We should mention that (\ref{proddef2}) generalizes known definitions of the product of the Heaviside function and the classical delta-function in $\mathcal D'$ given by

223: \begin{equation}

224: \label{stupidmult}

225: \theta_\tau\delta_\tau=c\delta_\tau,

226: \end{equation}

227: where $c \in \bf R$, which were proposed (in particular, through the multiplication of two distributions) in \cite{Kur,Fil,Tvr1,Ses,Sar,Sar2,Sar3}, in order to obtain the optimal (from a certain point of view) value of $c \in \bf R$, e.g. definition (\ref{stupidmult}) for $c=1/2$ satisfies the formal Leibniz rule for the differentiation of the product. 

228: 

229: In contrast to definitions of form (\ref{stupidmult}), the operation of multiplication in the new space of distributions is continuous, commutative and associative, the operation of multiplication in the new space of distributions satisfies the Leibniz product rule.

230: 

231: %We pay special attention to the fact that in the new space of distributions the product of the Heaviside function and the delta-function (\ref{proddef2}) is again a delta-function (with a certain coefficient), which is not the case for the product $\theta_\tau\delta_\tau$ in the Colombeau generalized functions algebra. This means that in contrast to the Colombeau generalized functions theory, in our space of distributions the product $\theta_\tau\delta_\tau^\alpha$ can be represented as a limit of a delta-sequence (with certain coefficient), which is important for the theory of the ordinary differential equations with distributions (see Appendix). 

232: 

233: 

234: 

235: 

236: Apart of the problem of multiplication of distributions, our study is motivated by the problem of the correct definition of the solution of an ordinary differential equation with distributions. Further we provide an application of our space of distributions to nonlinear ordinary differential equations containing one delta-function $\delta_\tau^\alpha$, which arise in optimal control theory \cite{Ses,Ramp,Silv,Mil}. 

237: %Consideration of general differential equation with distributions from the constructed space (which allows us to solve a number of problems related to differential equations with distributions (see the Appendix)) will be the subject of a future study.

238: 

239: Note that the set of points of concentration of delta-functions in $v$ and even its cardinality, as well as the set of points of discontinuity of $x$, can not be fixed a priori, since in many optimal control problems it is a subject of optimization \cite{Silv,Ramp,Ses}. This leads to the necessity of considering the space of functions of bounded variation, and, as a result, to certain complications in contrast to the case when the set of points of discontinuity of a function, which is multiplied by a distribution, is finite and fixed. 

240: 

241: In order to construct this space of distributions

242: %the space of distributions where the operation of multiplication has property $\mathcal B$, which is important for the differential equations, 

243: we introduce the supplementary notion of the \textit{dynamic function}. Most of the definitions for the ordinary functions (such as limit at point, support, boundedness, continuity, bounded variation) are transferred to dynamic functions without significant changes.

244:  

245: \section{Dynamic test functions \\ [3mm]} 

246: 

247: \subsection{Notations}

248: Let $I=(a,b) \subset \bf R$ be a fixed open interval, in general unbounded.

249: %, including the case $a=-\infty$,

250: %$b=+\infty$. 

251: By $\textbf{R}^n$ and $\textbf{M}_n$ we denote the space of vectors

252: and the space of square matrices of order $n$ with real elements, respectively. We denote by $T(g) \subset I$

253: the set of points of discontinuity of a function $g:I \to \bf R$.                      

254: 

255: 

256:  

257: 

258: Consider the algebra of functions $g:I \to \bf R$ possessing one-sided limits 

259: \begin{equation*}

260: g(a+)=\lim_{t \to a+}g(t), \quad g(b-)=\lim_{t \to b-}g(t), \quad g(\tau+)=\lim_{t \to \tau+}g(t), \quad g(\tau-)=\lim_{t \to \tau-}g(t)

261: \end{equation*}

262: %$g(a+)=\lim_{t \to a+}g(t)$, $g(b-)=\lim_{t \to b-}g(t)$, $g(\tau+)=\lim_{t \to \tau+}g(t)$, $g(\tau-)=\lim_{t \to \tau-}g(t)$ 

263: for any $\tau \in I$.

264: We identify functions having the same one-sided limits, and possibly having different values in their points of discontinuity.

265: We denote this algebra of functions (i.e., equivalence classes) by $\mathbb G=\mathbb G(I)$ and endow it with the norm

266: \begin{equation*}

267: \|g\|_{\mathbb G}=\sup_{t \in I}\max\{|g(t+)|,|g(t-)|\}.

268: \end{equation*}

269: With this norm $\mathbb G$ is a Banach algebra \cite{Der2}.

270: The algebra $\mathbb G$ is called the \textit{algebra of regulated functions} \cite{Dieu}.

271: %\begin{lemma}[\cite{Der2}]

272: %\label{lem0}

273: The set of points of discontinuity $T(g)=\{t \in I: g(t+) \ne g(t-)\}$ of a regulated function $g \in \mathbb G$ is at most countable \cite{Dieu}.

274: % \cite{Der2}.

275: %\end{lemma}                                                        

276: 

277:                   

278: 

279: For a function $g \in \mathbb{G}$ let us consider a partition $d=\{t_i\}_{i=1}^n \subset I$, 

280: $t_1<\dots<t_n$ of the interval $I$.

281: % such that $d \cap T(g)=\varnothing$.

282: Denote 

283: \begin{equation*}

284: \var_{I}(g) \doteq \sup_d \sum_{i=1}^n|g(t_i-)-g(t_{i-1}+)| \geqslant 0

285: \end{equation*}

286: (\textit{total variation}). If $\var_{I}(g)<\infty$, we say that $g$ is a \textit{function of bounded variation} on $I$.

287: We denote this algebra by

288: $\mathbb{BV}=\mathbb{BV}(I)$, and define the norm

289: \begin{equation*}

290: \|g\|_{\mathbb{BV}}=|g(a+)|+\var_{I}(g) \\ [3mm].

291: \end{equation*}

292: It follows that $\mathbb{BV}$ is a Banach algebra. 

293: 

294: 

295: We denote by $\mathbb C=\mathbb C(I)$ the subalgebra of continuous elements of $\mathbb G$. Denote by $\mathbb{CBV}$ the subalgebra of continuous elements of $\mathbb{BV}$.

296: Denote by $\mathbb L=\mathbb L(I)$ the Banach algebra of functions Lebesgue summable on $I$ with the integral norm.

297: Further $\mathbb{AC}=\mathbb{AC}(I)$ stands for the Banach algebra of absolutely continuous functions \cite{Dun}.

298: Let $\mathbb L(\loc)$ and $\mathbb{AC}(\loc)$ be the algebras of locally summable and locally absolutely continuous functions, respectively.   

299: 

300: We denote $\sigma_t(g) \doteq g(t+)-g(t-)$.

301: 

302: Let us denote by $\mathbb F=\mathbb F(I)$ the algebra of functions $I \to \mathbf R$ with the operations of point-wise addition, multiplication by an element of $\bf R$ and multiplication. The elements of $\mathbb F$ are called the \textit{ordinary} functions. 

303: 

304: The definitions given above are transferred without significant changes to the case of a finite closed interval.

305: 

306: %The values of functions from $\mathbb G$ in their points of discontinuity are ignored since factor-space $\mathbb G$ admits relatively simple represenation of the elements of adjoint space, which is important for the construction of space of distributions. 

307: 

308: Notice, that ignoring  values of discontinuous functions at their points of discontinuity leads to a further simplification of the space of distributions.

309: 

310: 

311: 

312: 

313: \subsection{Dynamic functions} In this subsection we introduce the supplementary notion of a \textit{dynamic function} which is needed for the construction of the space of test function.

314: It generalizes the notion of an ordinary function. Let $J=[-\frac{1}{2},\frac{1}{2}]$. 

315: 

316: \begin{definition} 

317: A map

318: %\begin{equation}

319: %\label{dynfunc}

320: $f:I \to \mathbb F(J)$

321: %\end{equation}

322: is called the dynamic function.

323: \end{definition}

324: 

325: We denote the value of a dynamic 

326: function $f$ at a point $t$ by $f(t)(\cdot)$. 

327: 

328: Denote the set of dynamic functions by $d\mathbb F=d\mathbb F(I)$.                     

329: We say that dynamic functions $f_1$, $f_2$ are equal, if $f_1(t)(\cdot)=f_2(t)(\cdot)$ in $\mathbb F(J)$ for all

330: $t \in I$. 

331: Also, the operations of addition, multiplication by an element of $\bf R$ and multiplication are defined in $d\mathbb F$ as follows:

332: for dynamic functions $f_1,f_2 \in d\mathbb F$ we define their sum $f_1+f_2 \in d\mathbb F$ by the formula

333: \begin{equation*}

334: (f_1+f_2)(t)(\cdot)=f_1(t)(\cdot)+f_2(t)(\cdot)

335: \end{equation*}

336: for all $t \in I$;

337: similarly, we define the product

338: \begin{equation*}

339: (f_1f_2)(t)(\cdot)=f_1(t)(\cdot)f_2(t)(\cdot);

340: \end{equation*}

341: for a given dynamic function $f \in d\mathbb F$ and $\lambda \in \bf R$ we define 

342: \begin{equation*}

343: (\lambda f)(t)(\cdot)=\lambda f(t)(\cdot) 

344: \end{equation*}

345: for all $t \in I$. Thus, $d\mathbb F$ form an algebra. 

346: 

347: %\subsection{Ordinary functions as dynamic functions}

348: We define the inclusion of the algebra of ordinary functions $\mathbb F$ into the algebra of dynamic functions $d\mathbb F$ by the map:

349: with an ordinary function $\hat{f} \in \mathbb F$ we associate a dynamic function $f \in d\mathbb F$ defined by

350: \begin{equation*}

351: f(t)(\cdot) \equiv \hat{f}(t)

352: \end{equation*}

353: for all $t \in I$.                             

354: %The elements of $\mathbb F \subset d\mathbb F$ are also called  functions.

355: 

356: %\begin{definition}                                         

357: For a given $t$, we call $f(t)(\cdot)$ a \textit{dynamic value} of $f$ at point $t$.

358: A dynamic value of $f$ which is identically equal to a constant function is called an \textit{ordinary value}. If $f \in d\mathbb F$ has an ordinary value at a point $t$,

359: we denote it by $f(t) \in \bf R$. 

360: %\end{definition}

361: Then a dynamic function is ordinary if and only if it has ordinary values at all points. 

362: We denote the set of points, where $f$ has ordinary values, by $\us(f) \subset I$. Let $\dyn(f) \doteq I \setminus \us(f)$. \\

363: 

364: Let $f \in d\mathbb F$, $f(t)=0$ ($t \ne \tau$), with dynamic value $f(\tau)(\cdot)=\gamma(\cdot)$. Construct a sequence of ordinary functions $\{f_n\}_{n=1}^\infty$ defined by

365: \begin{equation}

366: f_n(t)=\left\{

367: \begin{array}{l}

368: \gamma\bigl(n(t-\tau)\bigr), \text{ if } t \in \bigl(\tau-\frac{1}{2n},\tau+\frac{1}{2n}\bigr), \\ [2mm]

369: 0, \quad \text{ otherwise} \\ [2mm]

370: \end{array}

371: \right.

372: \end{equation}

373: %(i.e. dynamic value $\gamma(\cdot)=f(\tau)(\cdot)$ in point $\tau$ become shrinked in point $\tau$ when $n \to \infty$). 

374: The sequence $\{f_n\}_{n=1}^\infty \subset \mathbb F$ is called the \textit{sequential representation} of $f \in d\mathbb{F}$.

375: 

376: %Such a representation can be defined, in particular, for an arbitrary dynamic function which is ordinary except some finite set of points.

377: %All operations applied to such dynamic functions are coherent with the operations applied to terms of their sequential representations.

378: %Thus, any dynamic function with a finite number of points where it has non-ordinary values can be identified with its sequential representation.

379: %This explains the use of the term \textit{dynamic} above.

380: 

381: 

382: 

383: 

384: 

385: %\subsection{Supremum and infinum, composition, support and limit at point}

386: We define the composition $g \circ f \in d\mathbb F$ of a dynamic function $f \in d\mathbb F$ and an ordinary function $g \in \mathbb F$ by

387: \begin{equation*}

388: (g \circ f)(t)(\cdot)=g \circ (f(t)(\cdot))

389: \end{equation*}

390: for all $t \in I$. 

391: Accordingly, define the absolute value of $f \in d\mathbb F$ by

392: \begin{equation*}

393: |f|(t)(\cdot) \doteq |f(t)(\cdot)|

394: \end{equation*}

395: for all $t \in I$. Also, we define the support \\

396: \begin{equation*}

397: \supp f=\cl\{t \in I: f(t)(\cdot) \text{ is not ordinary or } f(t) \ne 0\} \\ [2mm]

398: \end{equation*}

399: where $\cl$ stands for closure in $I$. For a given set $M \subset I$ let us define

400: \begin{equation*}

401: \sup_M f \doteq \sup_{t \in M} \sup_{s \in J} f(t)(s), \quad

402: \inf_M f \doteq \inf_{t \in I} \inf_{s \in J} f(t)(s). 

403: \end{equation*}

404: We say that $f$ is bounded on $M$, if $\sup_M|f|<\infty$.

405: 

406: A dynamic function $f$ is called nonnegative (positive) (denote $f \geqslant 0$ ($f>0$)) if $f(t)(s) \geqslant 0$ ($f(t)(s)>0$) for all $t \in I$.

407: A dynamic function $f$ is called nonpositive (negative) if $-f$ is nonnegative (positive).

408: 

409: %\subsection{Limit at point} 

410: %\begin{definition}

411: Let $f \in d\mathbb F$ be bounded in a right neighborhood of a point $\tau \in I$.

412: We say that $f$ tends to $c \in \bf R$ from the right as $t \to \tau+$ if for any $\varepsilon>0$ there exists an $\eta>0$ such that

413: \begin{equation}

414: \sup_{(\tau,\tau+\eta)} |f-c|<\varepsilon.

415: \end{equation}

416: %\end{definition}

417: We denote the right-sided limit $c$ by $f(\tau+)$. The left-sided limit and the limit of a dynamic functions at a point are defined similarly. Analogously, define the one-sided limits at $a$, $b$.

418: 

419: %\begin{definition}

420: We say that $f \in d\mathbb F$ is continuous at $\tau \in I$, if

421: \begin{equation}

422: f(\tau)=f(\tau+)=f(\tau-). 

423: \end{equation}

424: 

425: %\end{definition}

426: Notice that according to the notational convention above the notations used in this definition imply that $f$ has an ordinary value $f(\tau) \in \bf R$.

427: If $f$ is not continuous at  $\tau$, then we call $f$ discontinuous at this point.

428: 

429: Clearly, if in the definitions given above a dynamic function $f$ is an ordinary function, then these definitions coincide with the usual ones. 

430: 

431: %\subsection{Algebras of dynamic functions posessing onesided limits}

432: \subsection{Special algebras of dynamic functions}

433: %Our main interest related to a space of distributions and differential equations is in the following subalgebras of the algebra of dynamic functions.

434: Motivated by our interest in ordinary differential equations, we consider the following subalgebras of the algebra of dynamic functions.

435: 

436: 

437: Denote by $d\mathbb G$ an algebra of dynamic functions $f$ having dynamic values $f(t)(\cdot) \in \mathbb G(J)$ and such that the

438: one-sided limits 

439: \begin{equation*}

440: f(a+), \quad f(b-), \quad f(\tau+), \quad f(\tau-)

441: \end{equation*}

442: exist for all $\tau \in I$. We call the elements of $d\mathbb G$ the \textit{dynamic regulated functions} (since $\mathbb G$ is a factor-space, $d\mathbb G$ is also a factor-space; definitions of the supremum and infimum are changed respectively; if there is no need, we will not make special remarks, and call the elements of $d\mathbb G$ the dynamic functions).

443: Let us define the norm in the algebra of dynamic regulated functions $d\mathbb G$ by

444: \begin{equation*}

445: \|f\|_{d\mathbb G}=\sup_{I}|f|.

446: \end{equation*}

447: %It can be shown that $d\mathbb G$ is a Banach algebra. 

448: %The proof of the following lemma has technical character and thus can be omitted.

449: 

450: We have the following properties of the elements of $d\mathbb G$.

451: 

452: \begin{lemma}

453: \label{lem1}

454: Any dynamic regulated function $f \in d\mathbb G$ has ordinary values on $I$ except for at most a countable set.

455: \end{lemma}

456: 

457: The proof of Lemma \ref{lem1} and other statements of Sections 2 and 3 are provided in Section 5.

458: 

459: Given $f \in d\mathbb G$, define a function $\hat{f}$ by $\hat{f}(t)=f(t)$ for any $t \in \us(f)$.

460: We call $\hat{f}$ the \textit{ordinary part} of $f$.

461: As follows from Lemma \ref{lem1}, the ordinary part $\hat{f}$ is defined everywhere on $I$ 

462: except for at most a countable set. Denote $\ordin(f) \doteq \hat{f}$.

463: %The following lemma states that the usual part of dynamic regulated function from $d\mathbb G$ is a usual regulated function from $\mathbb G$.

464: 

465: \begin{lemma}

466: \label{lem2}

467: The ordinary part $\hat{f}$ of a dynamic regulated function $f \in d\mathbb G$ 

468: belongs to $\mathbb G$, and the equalities $f(t+)=\hat{f}(t+)$, $f(t-)=\hat{f}(t-)$ hold for all $t \in I$.

469: \end{lemma}

470: 

471: %As it follows from lemma \ref{lem2}, we can define linear and continuous operator $\ordin:d\mathbb G \to \mathbb G$ by $\ordin(f)=\hat{f}$ -- usual part. 

472: 

473: \begin{lemma}

474: \label{lem3}

475: The set of points of discontinuity $T(f)$ of any dynamic regulated function $f \in d\mathbb G$ is at most countable.

476: \end{lemma}

477: 

478: Let us define an inclusion of $\mathbb G$ into $d\mathbb G$. To do this we associate with any regulated function $\hat{f} \in \mathbb G$

479: a dynamic regulated function $f \in d\mathbb G$ having dynamic values

480: \begin{equation}

481: f(t)(s)=\hat{f}(\tau-) \text{ for } s \in [-1/2,0), \quad 

482: f(t)(s)=\hat{f}(\tau+) \text{ for } s \in (0,1/2] 

483: \end{equation}

484: for all $t \in I$. %Then, as can be shown, $f$ belongs to $d\mathbb G$ (notice that since $\mathbb G$ is a factor-algebra, images of its elements in $d\mathbb G$ are not the ordinary functions).

485: 

486: Let $f \in d\mathbb G$ be such that

487: \begin{equation*}

488: f(t)(\cdot) \in \mathbb{AC}(J), \quad f(t)(-1/2)=f(t-), \quad f(t)(1/2)=f(t+)

489: \end{equation*}

490: for all $t \in I$.

491: The set of such dynamic regulated functions with operations induced from $d\mathbb G$ form a

492: subalgebra $s\mathbb G \subset d\mathbb G$. Notice that $T(g)=\dyn(g)$ ($g \in s\mathbb G$).

493: 

494: 

495: 

496: We say that $f \in s\mathbb G$ is a \textit{dynamic function of bounded variation}, if $\ordin(f) \in \mathbb{BV}$ and

497: \begin{equation*}

498: \sum_{t \in \dyn(f)}\var_{s \in J}f(t)(s)<\infty, 

499: \end{equation*}

500: where $\dyn(f) \subset I$ is at most countable according to Lemma \ref{lem2}.

501: The set of dynamic functions of bounded variation with operations induced from $s\mathbb G$ forms a subalgebra.

502: The subalgebra of such dynamic functions is denoted by $s\mathbb{BV}$ and endowed with the norm

503: \begin{equation*}                          

504: \|f\|_{s\mathbb{BV}}=|f(a+)|+\|\hat{f}_c\|_{\mathbb{BV}}+\sum_{t \in \dyn(f)}\var_J (f(t)(\cdot)) \\ [1mm]

505: \end{equation*}

506: where $\hat{f}_c \in \mathbb{CBV}$ is a continuous part of  $\hat{f}=\ordin(f) \in \mathbb{BV}$ (the \textit{Jordan decomposition of functions of bounded variation}). 

507: 

508: \begin{example} The dynamic Heaviside function $\theta_\tau^\beta \in s\mathbb{BV}$ is defined by

509: \begin{equation*}

510: \theta_\tau^\beta(t)=\left\{

511: \begin{array}{l}

512: 1, \quad t>\tau, \\

513: 0, \quad t<\tau,

514: \end{array}

515: \right. \qquad

516: \theta_\tau^\beta(\tau)(\cdot)=\beta(\cdot), \\

517: \end{equation*}

518: where $\beta \in \mathbb{AC}(J)$ is such that $\beta(-1/2)=0$, $\beta(1/2)=1$.

519: %$\theta_\tau^\beta(t)=0$ for $t \in (a,\tau)$, $\theta_\tau^\beta(t)=1$ for $t \in (\tau,b)$,

520: %dynamic value of $\theta_\tau^\beta$ at a point $\tau \in I$ is defined by $\theta_\tau^\beta(\tau)(\cdot)=\beta(\cdot) \in \mathbb{AC}(I)$,

521: \end{example}

522: 

523: \begin{example}

524: Since $\mathbb G \subset d\mathbb G$, the usual Heaviside function $\theta_\tau \in \mathbb G$ is in $d\mathbb G$. According to the definition of the inclusion of

525: $\mathbb G$ to $d\mathbb G$, we have that 

526: \begin{equation*}

527: \theta_\tau(t)=\left\{

528: \begin{array}{l}

529: 0, \quad t<\tau, \\

530: 1, \quad t>\tau,

531: \end{array}

532: \right. \qquad

533: \theta_\tau(\tau)(s)=\left\{

534: \begin{array}{l}

535: 0, \quad s \in [-\frac{1}{2},0), \\

536: 1, \quad s \in (0,\frac{1}{2}].

537: \end{array}

538: \right.

539: \end{equation*}

540: \end{example}

541: 

542: %\subsection{Representation of a dynamic function by a sequence of ordinary functions}

543: 

544: \subsection{Test functions}        

545: We denote by $\mathcal D$ the space of classical test functions, i.e., the space of the real-valued continuous functions having compact support in $I \subset \bf R$, which is endowed with the standard topology \cite{Shi}. 

546: 

547: We denote by $\mathcal T$ the space of elements $\varphi \in d\mathbb G$ having compact support $\supp\varphi \subset I$.

548: We say that $\{\varphi_n\}_{n=1}^\infty \subset \mathcal T$ converges to $\varphi \in \mathcal T$ if $\varphi_n \to \varphi$ in $d\mathbb G$ and 

549: there exists a closed interval $[c,d] \subset I$ such that $\supp\varphi_n \subset [c,d]$ for all $n=1,2,\dots$ Thus,

550: the classical space of continuous test functions $\mathcal D$ is contained in $\mathcal T$ as a subspace.

551: 

552: Notice that since for $\varphi \in \mathcal T$, $g \in d\mathbb G$ their product $g\varphi$ belongs to $\mathcal T$,

553: algebra $\mathcal T$ is an ideal in $d\mathbb G$. Further, if $\{\varphi_n\}_{n=1}^\infty \subset \mathcal T$ and

554: $\varphi_n \to \varphi$, then $g\varphi_n \to g\varphi$ in $\mathcal T$. 

555: 

556: 

557: 

558: 

559: 

560: We call $\mathcal T$  the space of \textit{dynamic test functions}.

561: %\end{definition}

562: The proof of the following theorem is similar to the proof of an analogous statement for

563: $\mathcal D$ \cite{Ios}.  

564: 

565: \begin{theorem}

566: \label{topteo}

567: $\mathcal T$ is a locally convex topological vector space.

568: \end{theorem}

569: \section{Distributions with dynamic test functions \\ [3mm]}

570: 

571: Following standard notation \cite{Shi}, we denote by $\mathcal D'$ the space of distributions with continuous test functions, i.e., the space of continuous linear functionals

572: $\mathcal D \to \bf R$ \cite{Shi}.

573: 

574: We denote by $\mathcal T'$ the space of \textit{distributions with dynamic test functions}, i.e., the space of continuous linear functionals $\mathcal T \to \bf R$. 

575: The value of a distribution $f \in \mathcal T'$ on a test function $\varphi \in \mathcal T$ is denoted by $(f,\varphi) \in \bf R$.

576: 

577: \begin{example}

578: %\subsection{Regular distributions}

579: Let $f \in d\mathbb G$ (in particular, $f \in \mathbb G$). 

580: We define a distribution, which is also denoted by $f$, by the formula

581: \begin{equation}

582: \label{reg}

583: %(f,\varphi)=\int_If(t)\hat{\varphi}(t)dt \quad \text{  or  } \quad

584: (f,\varphi)=\int_I\hat{f}(t)\hat{\varphi}(t)dt

585: \end{equation}        

586: where $\varphi \in \mathcal T$, $\hat{\varphi}=\ordin(\varphi)$,

587: $\hat{f}=\ordin(f)$.

588: The linearity and continuity of $f$ follows from the properties of the integral and the definition of 

589: the ordinary part of a dynamic regulated function; a distribution $f \in \mathcal T'$ is called a \textit{regular distribution}.

590: % corresponding to locally 

591: %summable function $f \in \mathbb L(\loc)$.

592: % or dynamic regulated function $f \in d\mathbb G$ respectively.

593: %Since $\mathcal D \subset \mathcal T$, then according to known statement for $\mathcal D'$ \cite{Shi},

594: %linear space $\mathbb L(\loc)$ is isomorphic to linear manifold of regular distributions

595: %in $\mathcal T'$. 

596: \end{example}

597: 

598: %\subsection{Delta-function} 

599: \begin{example}

600: Define a distribution $\delta_\tau^\alpha$ by the formula

601: \begin{equation}

602: \label{delta}

603: (\delta_\tau^\alpha,\varphi)=\int_J\varphi(\tau)(s)\alpha(s)ds

604: \end{equation}

605: where $\varphi \in \mathcal T$ and $\alpha\in \mathbb L(J)$ is such that

606: \begin{equation}

607: \label{normalization}

608: \int_J \alpha(s)ds=1.

609: \end{equation}                                 

610: The linearity and continuity of $\delta_\tau^\alpha$ follows from the properties of the integral and the definition of the convergence in $\mathcal T$, so

611: $\delta_\tau^\alpha \in \mathcal T'$. 

612: The distribution $\delta_\tau^\alpha \in \mathcal T'$ is called the \textit{delta-function} concentrated at the point $\tau \in I$ and having \textit{shape} $\alpha \in \mathbb L(J)$.

613: %\end{definition}

614: Now, for continuous test functions $\varphi \in \mathcal D$ 

615: \begin{equation*} (\delta_\tau^\alpha,\varphi)=\int_J\varphi(\tau)\alpha(s)ds=\varphi(\tau)\int_J\alpha(s)ds=\varphi(\tau).

616: \end{equation*}

617: %Thus for $\varphi \in \mathcal D \subset \mathcal T$ the value of $\delta_\tau^\alpha \in \mathcal T'$ coincides with value of delta-function $\delta_\tau \in \mathcal D'$.

618: Thus, the delta-function $\delta_\tau \in \mathcal D'$ extends to the family of delta-functions $\delta_\tau^\alpha \in \mathcal T'$. 

619: For a given delta-function $\delta_\tau^\alpha$ we construct a sequence $\{\omega_n^\alpha\}_{n=1}^\infty$,

620: \begin{equation*}

621: \omega^\alpha_n(t)=\left\{

622: \begin{array}{l}

623: n\alpha(n(t-\tau)), \text{ if } t \in (\tau-\frac{1}{2n},\tau+\frac{1}{2n}), \\ [3mm]

624: 0, \text{ otherwise }

625: \end{array}

626: \right.

627: \end{equation*}

628: We call $\{\omega^\alpha_n\}_{n=1}^\infty$ the \textit{delta-sequence having shape} $\alpha \in \mathbb L(J)$.

629: \end{example}

630: 

631: \begin{example}

632: \label{ex_newdelta}

633: Let us define a distribution $\delta_\tau^\lambda \in \mathcal T'$, which is also called \textit{delta-function}, by 

634: \begin{equation}

635: \label{newdelta}

636: (\delta_\tau^\lambda,\varphi)=\lambda\varphi(\tau+)+(1-\lambda)\varphi(\tau-),

637: \end{equation}

638: where $\lambda \in \bf R$ and the restriction $\delta_\tau^\lambda|_{\mathcal D}=\delta_\tau \in \mathcal D'$.

639: Clearly, (\ref{newdelta}) can not be obtained from (\ref{delta}). The notion of a delta-sequence is not defined for the delta-function (\ref{newdelta}). 

640: \end{example}

641:             

642: \begin{remark}

643: \label{rm_newdelta}

644: In the general case, the delta-function in $\mathcal T'$ is defined as an affine combination of delta-functions (\ref{delta}) and (\ref{newdelta}). Below we consider mainly the delta-functions of the form (\ref{delta}).

645: \end{remark}              

646:                                                                                                        

647: %\subsection{Operations on distributions} 

648: In the space of distributions $\mathcal T'$ the linear operations of addition and multiplication by elements of $\bf R$ are

649: introduced in a standard way, so $\mathcal T'$ is a linear space. Let

650: $f_n \to f$ ($f_n,f \in \mathcal T'$) in $\mathcal T'$, if

651: $(f_n,\varphi) \to (f,\varphi)$ as $n \to \infty$ for any test function $\varphi \in \mathcal T$. The proof of the following lemma is similar to the proof of analogous theorem for $\mathcal D'$ \cite[p.65]{Shi}.

652: 

653: \begin{lemma} 

654: \label{lem4}                                         

655: If $\{f_n\}_{n=1}^\infty$ converges in $\mathcal T'$, and $\varphi_n \to 0$ in $\mathcal T$,

656: then $(f_n,\varphi) \to 0$.

657: \end{lemma}

658: 

659: \begin{theorem}

660: \label{limitteo}

661: Let the sequence $\{f_n\}_{n=1}^\infty \subset \mathcal T'$ be such that for any $\varphi \in \mathcal T$ the

662: sequence $\{(f_n,\varphi_n)\}_{n=1}^\infty$ converges as $n \to \infty$. The functional $f$ defined on $\mathcal T$ by 

663: \begin{equation}

664: \label{limit}

665: (f,\varphi)=\lim\limits_{n \to \infty} (f_n,\varphi)

666: \end{equation}

667: is linear and continuous and thus belongs to $\mathcal T'$.

668: \end{theorem}

669: 

670: \begin{theorem}

671: \label{ext_teo}

672: Any distribution in $\mathcal D'$ can be extended from $\mathcal D$ to $\mathcal T$.

673: \end{theorem}

674: 

675: Thus, the space $\mathcal T'$ is a space of extensions of distributions in $\mathcal D'$ from $\mathcal D$ to $\mathcal T$.

676: 

677: Notice, that the restriction of the distribution $\delta_\tau^\alpha-\delta_\tau^\gamma \in \mathcal T'$, where $\alpha$, $\gamma$ are the shapes of delta-functions, from $\mathcal T$ to $\mathcal D$, is a zero distribution in $\mathcal D'$. Hence, since the shapes $\alpha$, $\gamma$ can be chosen arbitrarily, 

678: addition of $\delta_\tau^\alpha-\delta_\tau^\gamma$ to an extension from $\mathcal D$ to $\mathcal T$ of any distribution in $\mathcal D'$, gives another extension.

679: Consequently, any distribution in $\mathcal D'$ has infinitely many extensions from $\mathcal D$ to $\mathcal T$.

680: 

681: A distribution $f \in \mathcal T'$ is called nonnegative (nonpositive) if for any $\varphi \in \mathcal T$ such that $\varphi \geq 0$ we have

682: $(f,\varphi) \geq 0$ (or $(f,\varphi) \leq 0$, respectively). 

683: 

684: %\subsection{Multiplication by discontinuous functions} 

685: 

686: Let us define in $\mathcal T'$ the operation of multiplication of distributions by the elements of $d\mathbb G$.

687: Since for any $g \in d\mathbb G$, $\varphi \in \mathcal T$ their product $g\varphi$ belongs to $\mathcal T$, we may define

688: \begin{equation}

689: \label{multdef}

690: (gf,\varphi) \doteq (f,g\varphi). \\ [2mm]

691: \end{equation}

692: where $f \in \mathcal T'$.

693: In particular, since $\mathbb G \subset d\mathbb G$,

694: any distribution in $\mathcal T'$ can be multiplied by any piece-wise continuous function.

695: 

696: 

697: 

698: \begin{theorem}

699: \label{cont_teo}

700: Let $f_n \to f$ in $\mathcal T'$, $g_n \to g$ in $d\mathbb G$. Then $g_nf_n \to gf$ in $\mathcal T'$.

701: \end{theorem}

702: 

703: The operation of multiplication in $\mathcal T'$ defined by (\ref{multdef}) is continuous, commutative and associative in the sense that

704: $(gh)f=g(hf)$ in $\mathcal T'$ for any $g,h \in d\mathbb G$, $f \in \mathcal T'$. 

705: 

706: \begin{example}

707: \label{ex21}

708: The product of the delta-function $\delta_\tau^\alpha$ and the Heaviside function $\theta_\tau^\beta \in s\mathbb G$ is given by the formula

709: \begin{equation}

710: \label{prod_w1}

711: (\theta_\tau^\beta\delta_\tau^\alpha,\varphi)=(\delta_\tau^\alpha,\theta_\tau^\beta\varphi)=\int_J\varphi(\tau)(s)\beta(s)\alpha(s)ds.

712: \end{equation}

713: If $\int_J\beta(s)\alpha(s)ds \ne 0$, then (\ref{prod_w1}) can be rewritten as

714: \begin{equation}

715: \label{prod33}

716: \theta_\tau^\beta\delta_\tau^\alpha=\biggl(\int_J\beta(s)\alpha(s)ds \biggr)\delta_\tau^\gamma,

717: \end{equation}

718: where 

719: $\gamma(\cdot)=\beta(\cdot)\alpha(\cdot)/\int_J\beta(s)\alpha(s)ds$ satisfies (\ref{normalization}). 

720: \end{example}

721: 

722: \begin{example}

723: \label{ex2}

724: The product of the delta-function $\delta_\tau^\alpha$ and the Heaviside function $\theta_\tau \in \mathbb G$ is given by the formula \\ [-5mm]

725: \begin{equation}

726: \label{prod_w2}

727: (\theta_\tau\delta_\tau^\alpha,\varphi)=(\delta_\tau^\alpha,\theta_\tau\varphi)=\int_{0}^{1/2}\varphi(\tau)(s)\alpha(s)ds.

728: \end{equation}

729: If $\int_{0}^{1/2}\alpha(r)dr \ne 0$, then (\ref{prod_w2}) can be rewritten as 

730: \begin{equation*}

731: \theta_\tau\delta_\tau^\alpha=\biggl(\int_0^{1/2}\alpha(s)ds\biggr)\delta_\tau^\gamma,

732: \end{equation*}

733: where the shape of the delta-function $\delta_\tau^\gamma$ is defined by $\gamma(s)=0$ for $s \in [-1/2,0)$, $\gamma(s)=\alpha(s)/\int_{0}^{1/2}\alpha(r)dr$ for $s \in (0,1/2]$.

734: \end{example}

735: 

736: \begin{remark}

737: \label{rem1}

738: %The same results can be obtained is a delta-function is replaced by the terms of the delta-sequence having the same shape, and a Heaviside function is replaced by the terms of its sequential representation. In this sense the operation of multiplication in $\mathcal T'$ has Property $\mathcal B$ (see the Introduction). 

739: Let us show that the equality (\ref{prod33}) can be obtained if $\delta_\tau^\alpha$ and $\theta_\tau^\beta$ are replaced by the terms of the corresponding delta-sequence $\{\omega^\alpha_n\}_{n=1}^\infty$ and sequential representation $\{f_n\}_{n=1}^\infty$, respectively. 

740: Let $\int_J\beta(s)\alpha(s)ds \ne 0$.

741: We have that

742: \begin{equation*}

743: f_n(t)\omega_n^\alpha(t)=\left\{

744: \begin{array}{l}

745: \beta(n(t-\tau))n\alpha(n(t-\tau)), \text{ if } t \in (\tau-\frac{1}{2n},\tau+\frac{1}{2n}), \\

746: 0, \text{ otherwise},

747: \end{array}

748: \right.

749: \end{equation*}

750: i.e., \\ [-5mm]

751: \begin{equation*}

752: f_n(t)\omega_n^\alpha(t)=\int_J \beta(s)\alpha(s)ds\left\{

753: \begin{array}{l}

754: n\frac{\beta(n(t-\tau))\alpha(n(t-\tau))}{\int_J \beta(s)\alpha(s)ds}, \text{ if } t \in (\tau-\frac{1}{2n},\tau+\frac{1}{2n}), \\

755: 0, \text{ otherwise},

756: \end{array}

757: \right.

758: \end{equation*}

759: where $\gamma(s)=\beta(s)\alpha(s)/\int_J\alpha(s)\beta(s)ds$ satisfies (\ref{normalization}),

760: \begin{equation*}

761: f_n\omega_n^\alpha=\biggl(\int_J \beta(s)\alpha(s)ds\biggr)\omega_n^\gamma

762: \end{equation*}

763: for any $n \in \bf N$. In this sense the operation of multiplication in $\mathcal T'$ has Property $\mathcal B$ (see the Introduction).

764: \end{remark}

765: 

766: \begin{example}

767: \label{ex3}

768: In the general case, let us consider the product of the delta-function $\delta_\tau^\alpha$ and a dynamic regulated function $f \in d\mathbb G$: \\ [-5mm]

769: \begin{equation*}

770: (f\delta_\tau^\alpha,\varphi)=(\delta_\tau^\alpha,f\varphi)=\int_J\varphi(\tau)(s)f(\tau)(s)\alpha(s)ds.

771: \end{equation*}

772: If $f$ has an ordinary value at $\tau$, then

773: the result above can be rewritten as

774: \begin{equation*}

775: f\delta_\tau^\alpha=f(\tau)\delta_\tau^\alpha.

776: \end{equation*}

777: \end{example}

778: 

779: \begin{remark}

780: The intersection $\mathcal T \cap \mathbb G$ with the topology induced by $\mathcal T$ determines a space of \textit{discontinuous test functions}. Since this is a subspace of $\mathcal T$, we obtain a factor-space of $\mathcal T'$, which is called the space of \textit{distributions with discontinuous test functions} \cite{DerKin4} (see the Introduction; also, see \cite{Kur,Kur2}).

781: \end{remark}

782: 

783: 

784: 

785: %\subsection{Differentiation} 

786: Let

787: $g \in s\mathbb{BV}$. Define the derivative $\dot{g} \in \mathcal T'$ by the formula

788: \begin{equation}

789: \label{dyn_deriv}

790: (\dot{g},\varphi)\doteq \int_I \hat{\varphi}(t)dg_c(t)+\sum_{\tau \in T(g)}\int_J\varphi(\tau)(s)(g(\tau)(s))^{\cdot}_sds.

791: \end{equation}

792: where $\varphi \in \mathcal T$, $g_c \in \mathbb{CBV}$ is a continuous part of $\ordin(g) \in \mathbb{BV}$. 

793: If for any $\tau \in T(g)$ we have $\sigma_\tau(g)=g(\tau+)-g(\tau-) \ne 0$, then \\ [-5mm]

794: \begin{equation}

795: \label{deriv2}

796: \dot{g}=\dot{g}_c+\sum_{\tau_k \in T(g)}\sigma_{\tau_k}(g)\delta_{\tau_k}^{\alpha_k}, \\ [-2mm] 

797: \end{equation}

798: where \\ [-5mm]

799: \begin{equation}

800: \label{cderiv}

801: (\dot{g}_c,\varphi)=\int_I \hat{\varphi}(t)dg_c(t),

802: \end{equation}

803: the shapes $\alpha_k$ are defined by

804: $\alpha_k(s)=(g(\tau_k)(s))^{\cdot}_s/\sigma_{\tau_k}(g)$ 

805: ($s \in J$, $k \in \bf N$). 

806: 

807: %For test functions from $\mathcal D$ the value of the derivative $\dot{g} \in \mathcal T'$

808: %coincides with the value of the derivative $\dot{\hat{g}} \in \mathcal D'$.

809: 

810: \begin{theorem}

811: \label{dercorteo}

812: The functional defined by (\ref{dyn_deriv}) is linear, continuous and thus determines a distribution in $\mathcal T'$.

813: \end{theorem}

814: 

815: %

816: %\begin{theorem}[Linearity]

817: %\label{linteo}

818: %For any $f,g \in s\mathbb{BV}$, $\lambda, \mu \in \bf R$ we have

819: %\begin{equation*}

820: %(\lambda f+\mu g)^{\cdot}=\lambda \dot{f}+\mu \dot{g} \in \mathcal T'

821: %\end{equation*}

822: %\end{theorem}

823: 

824: \begin{theorem}[Leibniz product rule]

825: \label{leibnitzteo}

826: For any $f,g \in s\mathbb{BV}$ we have

827: \begin{equation}

828: \label{dyn_leibnitz}

829: (fg)^{\cdot}=\dot{f}g+f\dot{g} \in \mathcal T'.

830: \end{equation}

831: \end{theorem}

832: 

833: \begin{example}

834: \label{dist_exdiff}

835: The derivative of the Heaviside function $\theta_\tau^\beta \in s\mathbb {BV}$ is given by

836: \begin{equation*}

837: \dot{\theta}_\tau^\beta=\delta_\tau^\alpha

838: \end{equation*}

839: where $\alpha=\dot{\beta}$ is a shape of the delta-function.

840: \end{example}

841: 

842: %\begin{remark}

843: %The higher order derivatives are not defined in $\mathcal T'$.

844: %Suppose that the derivative of the delta-function $\dot{\delta}_\tau \in \mathcal T'$ is defined, and the Leibniz product rule holds. Then

845: %$(\theta_\tau^\beta\delta_\tau^\alpha)^{\cdot}=\delta_\tau^{\dot{\beta}}\delta_\tau^\alpha+\theta_\tau^\beta\dot{\delta}_\tau^\alpha$,

846: %i.e.,

847: %$\delta_\tau^{\dot{\beta}}\delta_\tau^\alpha=(\theta_\tau^\beta\delta_\tau^\alpha)^{\cdot}-\theta_\tau^\beta\dot{\delta}_\tau^\alpha$.

848: %Consequently, the existence of the higher-order derivatives is equivalent to the existence of the product of two delta-functions. 

849: %However, as it is mentioned in \cite{Col1} (see the Introduction), there is a strong belief that the definition of continuous operation of multiplication of two delta-functions is impossible in the Schwartz distribution theory approach.

850: %\end{remark}

851: 

852: 

853: \subsection{Ordinary differential equations with distributions}

854: 

855: %In what follows, notations $\mathcal T'_n$ and $\mathcal D_n'$

856: %stand for the spaces of $n$-valued distributions with components in $\mathcal T'$ and $\mathcal D'$, respectively, where convergence, linear operations, operations of multiplication and differentiation are defined component-wise. We introduce analogous notations for the spaces of ordinary functions and dynamic functions $\mathbb{BV}_n$, $\mathbb{AC}_n$ and $s\mathbb{BV}_n$, respectively.

857: %

858: %In a subsequent paper we consider the initial value problem

859: %\begin{equation}

860: %\label{eq10}

861: %\dot{x}=f(t,x)+g(t,x)v, \quad x(t_0-)=x_0,

862: %\end{equation}

863: %where $t_0 \in I$, $x_0 \in \mathbf R^n$, and the distribution $v \in \mathcal T_n'$ is given by

864: %\begin{equation*}

865: %v=\dot{u},

866: %\end{equation*}

867: %where $u \in s\mathbb{BV}_n$ is such that $\sigma_\tau(u) \ne 0$ ($\tau \in T(u)$).

868: %function $f:I \times \mathbf{R}^n \mapsto \textbf{R}^n$ is continuous in $t$ and locally Lipschitz in $x$ with a constant $K_f>0$ for all $t \in I$, and function $g:I \times \mathbf{R}^n \mapsto \textbf{R}^{n \times n}$ is

869: %continuous in $t$ and locally Lipschitz in $x$ with a constant $K_g>0$ for all $t \in I$.

870: %Here according to (\ref{deriv2})

871: %\begin{equation*}

872: %v=\dot{u}_c+\sum_{\tau \in T(u)}\langle \sigma_\tau(u),\delta_\tau^{\alpha_\tau}\rangle,

873: %\end{equation*}

874: %where $\langle,\rangle$ is a component-wise product in $\mathbf R^n$, $\delta_\tau^{\alpha_\tau} \in \mathcal T_n'$ is a vector-valued delta-function

875: %defined by

876: %\begin{equation}

877: %(\delta_\tau^{\alpha_\tau},\varphi)\doteq\biggl((\delta_\tau^{\alpha_\tau^1},\varphi),\dots,(\delta_\tau^{\alpha_\tau^n},\varphi)\biggr)^{\top}

878: %\end{equation}

879: %where $\varphi \in \mathcal T$, $\alpha_\tau=(\alpha_\tau^1,\dots,\alpha_\tau^n) \in \mathbb L_n(J)$ is a vector of shapes.

880: %

881: %A \textit{solution} of the initial value problem (\ref{eq10}) is a dynamic function $x \in s\mathbb{BV}_n$ which satisfies (\ref{eq10}) in $\mathcal T_n'$. 

882: %An ordinary part $\hat{x} \in \mathbb{BV}_n$ of a solution $x$ is called an \textit{ordinary solution} of (\ref{eq10}).

883: %

884: %\begin{theorem}

885: %\label{existsteo2}

886: %There are $h>0$ and a solution $x \in s\mathbb{BV}_n(t_0-h,t_0+h)$ of \rm{(}\ref{eq10}\rm{)}, \it which is unique in the sense that it coincides with any other solution of \rm{(}\ref{eq10}\rm{)} \it on the common interval of definition.

887: %\end{theorem}

888: %

889: %\begin{theorem}

890: %\label{teo1}

891: %Let $x \in s\mathbb{BV}_n$ be a solution of \rm{(}\ref{eq10}\rm{)}\it, $\hat{x} \in \mathbb{BV}_n$ be an ordinary solution of \rm{(}\ref{eq10}\rm{)}\it. Then

892: %\begin{multline}

893: %\label{inteq}

894: %\hat{x}(t)=x_0+\int_{t_0}^t f\bigl(r,\hat{x}(r)\bigr)dr+\int_{t_0}^t g\bigl(r,\hat{x}(r)\bigr)du_c(r)+

895: %\sum\limits_{\tau<t} \bigl(\gamma_\tau(1/2)-\hat{x}(\tau-)\bigr)-\sum_{\tau<t_0}(\gamma_\tau(1/2)-\hat{x}(\tau-)),

896: %\end{multline}

897: %and dynamic value $\gamma_\tau(\cdot) \doteq x(\tau)(\cdot)$ satisfies \\ [-1mm]

898: %\begin{equation}

899: %\label{limitsystem}

900: %\dot{\gamma}_\tau(s)=g\bigl(\tau,\gamma_\tau(s)\bigr)\bigl\langle \sigma_\tau(u),\alpha_\tau(s)\bigr\rangle, \quad \gamma_\tau(-1/2)=x(\tau-) \\ [2mm]

901: %\end{equation}

902: %where $T(u)=\{\tau\}$ is at most countable by Lemma \ref{lem3}, $x(\tau+)=\hat{x}(\tau+)=\gamma_\tau(1/2)$.

903: %Conversely, any $x \in s\mathbb{BV}_n$ satisfying \rm{(}\ref{inteq}\rm{)}\rm{(}\ref{limitsystem}\rm{)} \it is a solution of \rm{(}\ref{eq10}\rm{)}.

904: %\end{theorem}

905: %

906: %Let us note that if $v=\delta_\tau^\alpha$, then

907: %\begin{equation}

908: %\dot{x}(t)=f\bigl(t,x(t)\bigr), \quad (t \ne \tau),

909: %\end{equation}

910: %\begin{equation}

911: %\label{2}

912: %\dot{\gamma}(s)=g\bigl(\tau,\gamma(s)\bigr)\alpha(s), \quad \gamma(-1/2)=x(\tau-),

913: %\end{equation}

914: %where $x(\tau+)=\gamma(1/2)$.

915: %

916: %Notice that the notations in (\ref{eq10}) are correct from the point of view of distribution theory, i.e., we can substitute $x \in s\mathbb{BV}_n$ into (\ref{eq10}), because the operations of differentiation, composition and multiplication (Example \ref{ex3}), which arise in (\ref{eq10}), are

917: %correctly defined in $\mathcal T'_n$, in contrast to the classical space $\mathcal D_n'$.

918: %

919: %\begin{remark}

920: %Consideration of ordinary differential equations with distributions in $\mathcal D_n'$ is presented in \cite{Silv,Ramp,Ses}, where, in particular, the following Cauchy problem is considered: 

921: %\begin{equation}

922: %\label{bad}

923: %\dot{x}=f(t,x)+\bigl(g(t,x)\iota\bigr)\delta_\tau, \quad x(t_0)=x_0,

924: %\end{equation}

925: %where $f$, $g$ are the same as above, $\delta_\tau \in \mathcal D'$ is a scalar delta-function, $\iota=(1,\dots,1)^{\top}$.

926: %

927: %As is mentioned in \cite{Ses}, it is reasonable to expect that in the general case a solution $x$ of (\ref{bad}) is discontinuous at the point $\tau$. Thus, in the general case, the notation in (\ref{bad}) is incorrect from the point of view of the distribution theory, since (\ref{bad}) contains the product of a discontinuous function $g(\tau,x(\cdot))$ and a distribution $\delta_\tau \in \mathcal D'$, which is undefined in  $\mathcal D_n'$.

928: %In \cite{Silv,Ramp,Ses} the following definition of a solution of (\ref{bad}) is proposed. The left-continuous function $x \in \mathbb{BV}_n$ is said to be a solution of (\ref{bad}), if there exists a delta-sequence $\{\omega_m\}_{m=1}^\infty$, $\omega_m \to \delta_\tau$ in $\mathcal D'$, such that 

929: %\begin{equation}

930: %\label{xconv}

931: %x_m \to x

932: %\end{equation}

933: %($m \to \infty$) in the weak topology of $\mathbb{BV}_n$, where $x_m \in \mathbb{AC}_n$ is a solution of the approximation problem  

934: %\begin{equation}

935: %\dot{x}=f(t,x)+\bigl(g(t,x)\iota\bigr)\omega_m(t), \quad x(t_0)=x_0.

936: %\end{equation}

937: %

938: %

939: %The necessary and sufficient condition for the uniqueness of the solution of (\ref{bad}), i.e., its independence of the choice of $\{\omega_m\}_{m=1}^\infty$, is the Frobenius condition

940: %\begin{equation}

941: %\label{frob}

942: %%[g^k,g^l]_x \equiv 0 

943: %\sum_{i=1}^n \frac{\partial g_{ij}}{\partial x_k}g_{km}=

944: %\sum_{i=1}^n \frac{\partial g_{im}}{\partial x_k}g_{kj}

945: %\end{equation}

946: %for all $t \in I$, $x \in \bf R^n$, $1 \leqslant k,j,m \leqslant n$, $g=(g_{ij})_{ij=1}^n$, see \cite{Ses,Mil}. If (\ref{frob}) is satisfied,

947: %then there exists the unique solution $x \in \mathbb{BV}_n$ of (\ref{bad}), such that 

948: %\begin{equation}

949: %\dot{x}=f(t,x(t)) \quad (t \ne \tau),

950: %\end{equation}

951: %\begin{equation}

952: %\label{bad2}

953: %\dot{\eta}(s)=g(\tau,\eta(s))\iota, \quad \eta(-1/2)=x(\tau-),

954: %\end{equation}

955: %where $x(\tau+)=\eta(1/2)$, see \cite{Ses}.

956: %Notice that (\ref{bad2}) coincides with (\ref{2}) for $\alpha \equiv \iota$. Indeed, for (\ref{2}) Frobenius condition (\ref{frob}) is the necessary and sufficient condition for the independence of the value $\gamma(1/2)$ of the choice of $\alpha$ satisfying (\ref{normalization}) \cite{Guyshun}.

957: %Also, if (\ref{frob}) is satisfied, then any solution of (\ref{bad}) coincides with the ordinary solution of (\ref{eq10}).

958: %Comparison of the results for equation (\ref{eq10}) considered in the space of distributions $\mathcal T_n'$, and the results for equation (\ref{bad}) considered in the classical space of distributions $\mathcal D_n'$, shows that in contrast to the approach based on the space $\mathcal D_n'$ \cite{Silv,Ramp,Ses}, the use of the space of distributions $\mathcal T_n'$ allows us to make definitions correct from the point of view of the distribution theory, and to eliminate the restrictive Frobenius condition (\ref{frob}), which, as it is mentioned in \cite{Mil}, can not be satisfied in many problems of the optimal control theory.

959: %\end{remark}

960: %

961: %In a subsequent paper we also consider the property of viability of solutions of ordinary differential equations with distributions.

962: %Consider in $\textbf{R}^n$ an ordinary differential equation

963: %\begin{equation}

964: %\label{ieqviab}

965: %\dot{x}=f(t,x),

966: %\end{equation}

967: %where $f$ is continuous in $t$ and Lipschitz in $x$.

968: %Let $M \subset \textbf{R}^n$ be closed, $t_0 \in I$, $\Omega=(t_0,T) \subset I$ is an open interval, in the general case unbounded. Following \cite{Aub}, we give the next definition.

969: %

970: %\begin{definition}

971: %A solution of \rm (\ref{ieqviab}) \it such that 

972: %$x(t_0) \in M$ and

973: %\begin{equation}

974: %x(t) \in M 

975: %\end{equation}

976: %for all $t \in \Omega$ is said to be \textit{viable} in $M$ \rm (on $\Omega$). \it The set $M$ is said to have the \textit{property of viability} \rm(\it on $\Omega$\rm) \it for \rm (\ref{ieqviab}), \it if any solution such that $x(t_0) \in M$ is viable in $M$ on $\Omega$. 

977: %\end{definition}

978: %

979: %The property of viability is closely related to the problems of existence of equilibria, construction of nonsmooth Liapunov functionals, and the problems of optimal control theory \cite{Aub5}, in particular, 

980: %the problem of construction of a map

981: %$M \ni x_0 \mapsto v_{x_0} \in \mathcal V$,

982: %where $v_{x_0} \in \mathcal V$ is an admissible control in a controlled system

983: %\begin{equation}

984: %\label{intro_contr}

985: %\dot{x}=f(t,x,v), \quad v \in \mathcal V,

986: %\end{equation}

987: %such that $M$ has the property of viability for (\ref{intro_contr}) if $v=v_{x_0}$, where $x_0 \in M$ (the \textit{viability problem} \cite{Aub}), and the problem of construction of an admissible control $v_* \in \mathcal V$ for a controlled system

988: %\begin{equation}

989: %\label{intro_contr2}

990: %\dot{x}=f(t,x,v), \quad x(t_0)=x_0, \quad v \in \mathcal V,

991: %\end{equation}

992: %such that solution of (\ref{intro_contr2}) for $v=v_*$ is viable in $M$ on $\Omega=(t_0,T_*)$, where $T_*=\max_{v \in \mathcal V} T$ (the \textit{problem of avoiding of encounters} with the set $\mathbf{R}^n \setminus M$ \cite{Faz}). 

993: %For the ordinary differential equations the condition for viability was given by M.~Nagumo \cite{Nag} in terms of the Bouligand's contingent cones \cite{Boul}. 

994: %\begin{definition}

995: %Let $\rho$ be the Euclidean distance between the sets in $\mathbf{R}^n$.

996: %The set 

997: %\begin{equation}

998: %T_M(x)=\{y \in {\bf R}^n:~ \liminf_{\varepsilon \to 0+} \bigl(\rho \bigl(\{x+\varepsilon y\},M\bigr)/\varepsilon\bigr)=0\}

999: %\end{equation}

1000: %is called the contingent cone to $M$ at $x \in \mathbf{R}^n$.

1001: %\end{definition}

1002: %

1003: %\begin{theorem}[Nagumo Theorem]

1004: %\label{nagteo2}

1005: %Let $M \subset \mathbf{R}^n$ be closed. The sufficient condition for $M$ to have the property of viability for \rm{(}\ref{ieqviab}\rm{)} \it \rm(\it on $\Omega$\rm)\it~is the tangential condition

1006: %\begin{equation}

1007: %\label{itang}

1008: %f(t,x) \in T_M(x)

1009: %\end{equation}

1010: %for all $t \in \Omega$, $x \in \partial M$, where $\partial M$ is a boundary of $M$. If \rm (\ref{ieqviab}) \it is autonomous, then \rm (\ref{itang}) \it is a necessary and sufficient condition for viability.

1011: %\end{theorem}

1012: %

1013: %After ordinary differential equations, conditions for viability were obtained for differential inclusions \cite{Aub2,Aub3}, stochastic differential equations \cite{Aub4}, differential equations and differential inclusions with aftereffect \cite{Had,Bar}.

1014: %

1015: %

1016: %In a subsequent paper we consider viable solutions of the systems with distributions

1017: %\begin{equation}

1018: %\label{viab_eq5}

1019: %\dot{x}=f(t,x)+g(t,x)v, \quad v \in \mathcal T_n',

1020: %\end{equation}

1021: %and provide sufficient conditions for a closed set $M \subset \mathbf R^n$ given by analytical constraints

1022: %\begin{equation}

1023: %M=\{x \in \mathbf{R}^n: \eta_i(x) \leqslant 0,~ 1 \leqslant i \leqslant m\},

1024: %\end{equation}

1025: %where $\eta_i:\mathbf R^n \mapsto \mathbf R$ are continuously differentiable, to have the property of viability for system (\ref{viab_eq5}) (a generalization of Nagumo Theorem). Consideration of the property of viability for system (\ref{viab_eq5}) is necessary for the problem of avoiding of encounters with the set $\mathbf{R}^n \setminus M$. Namely, the problem of avoiding of encounters may have no solution for system with ordinary right-hand side

1026: %\begin{equation}

1027: %\label{viab_eq6}

1028: %\dot{x}=f(t,x)+g(t,x)v, \quad v \in \mathbb L_n,

1029: %\end{equation}

1030: %which might be viewed as a restriction of system (\ref{viab_eq5}) to the set of regular controls $v$, and have solution for extended system (\ref{viab_eq5}), i.e., when control is allowed to be distributional. Thus, consideration of the problem of avoiding of encounters for system with distributions (\ref{viab_eq5}) allows us to provide existence of optimal solution, which has the natural interpretation for system (\ref{viab_eq6}) as the limit of its solutions. Also, consideration of the property of viability for (\ref{viab_eq5}) in the space $\mathcal T_n'$ (in contrast to approach based on the Schwartz space $\mathcal D_n'$ \cite{Ses}) allows us to define the trajectory at the moment of discontinuity (i.e., the moment of concentration of delta-function at $v$), which is important for the property of viability. 

1031: 

1032: 

1033: In what follows, notations $\mathcal T'_n$ and $\mathcal D_n'$

1034: stand for the spaces of $n$-valued distributions with components in $\mathcal T'$ and $\mathcal D'$ respectively, where convergence, linear operations, operations of multiplication and differentiation are defined componentwise. We introduce analogous notations for the spaces of ordinary functions and dynamic functions $\mathbb{BV}_n$, $\mathbb{AC}_n$ and $s\mathbb{BV}_n$, respectively.

1035: 

1036: Consider in $\mathcal T_n'$

1037: a Cauchy problem

1038: \begin{equation}

1039: \label{eq10}

1040: \dot{x}=f(t,x)+g(t,x)\delta_\tau^\alpha, \quad x(t_0-)=x_0,

1041: \end{equation}

1042: where $t_0 \in I$, $x_0 \in \textbf{R}^n$, and the functions $f$ and $g$ with values in $\bf R^n$ and $\bf M_n$, respectively, are continuously differentiable in both variables, $\delta_\tau^\alpha \in \mathcal T_n'$ is a vector-valued delta-function,

1043: \begin{equation*}

1044: \alpha=(\alpha_1,\dots,\alpha_n)^{\top}, 

1045: \end{equation*}

1046: the value of $\delta_\tau^\alpha$ on $\varphi \in \mathcal T$ is defined by

1047: \begin{equation}

1048: (\delta_\tau^\alpha,\varphi)\doteq\biggl((\delta_\tau^{\alpha_1},\varphi),\dots,(\delta_\tau^{\alpha_n},\varphi)\biggr)^{\top}.

1049: \end{equation}

1050: %In this example we restrict ourselves to the case of a differential equation with one delta-function. The results below can be transferred to the case of linear combination of delta-functions in (\ref{eq10}). However, the study of the equation (\ref{eq10}) with an arbitrary distribution from $\mathcal T'$ in the right-hand side is necessary, since in many optimal control problems the set of the points of concentration of delta-functions and its cardinality are the subject of optimization (see the Introduction).

1051: 

1052: 

1053: 

1054: %\begin{theorem}[existence theorem]

1055: We define a solution of the Cauchy problem (\ref{eq10}) to be a dynamic function $x \in s\mathbb{BV}_n$ which satisfies (\ref{eq10}) in $\mathcal T_n'$. Define an \textit{ordinary solution} $\hat{x} \doteq \ordin(x) \in \mathbb{BV}_n$.

1056: 

1057: 

1058: Then there exists a solution of the Cauchy problem (\ref{eq10}) $x \in s\mathbb{BV}_n$ having ordinary values $x(t)$ for all $t \ne \tau$, and having dynamic value $x(\tau)(\cdot)$ (denote $\gamma(\cdot) \doteq x(\tau)(\cdot)$), such that

1059: \begin{equation}

1060: \dot{x}(t)=f\bigl(t,x(t)\bigr), 

1061: \end{equation}

1062: for all $t \ne \tau$, and

1063: \begin{equation}

1064: \label{2}

1065: \dot{\gamma}(s)=g\bigl(\tau,\gamma(s)\bigr)\alpha(s), \quad \gamma(-1/2)=x(\tau-).

1066: \end{equation}

1067: %\end{theorem}

1068: (see Example \ref{ex3}), where $x(\tau+)=\gamma(1/2)$. 

1069: %\begin{proof}

1070: 

1071: %\end{proof}

1072: 

1073: Notice that in the result of the substitution of $x \in s\mathbb{BV}_n$ into (\ref{eq10}), the operations of differentiation, composition and multiplication (see example \ref{ex3}), which arise in (\ref{eq10}), are

1074: correctly defined in $\mathcal T'_n$, in contrast to the space $\mathcal D_n'$ (see the Introduction).

1075: 

1076: \begin{remark}

1077: Consideration of ordinary differential equations with distributions in $\mathcal D_n'$ is presented in \cite{Silv,Ramp,Ses}, where, in particular, the following Cauchy problem is considered: 

1078: \begin{equation}

1079: \label{bad}

1080: \dot{x}=f(t,x)+\bigl(g(t,x)\iota\bigr)\delta_\tau, \quad x(t_0)=x_0,

1081: \end{equation}

1082: where $f$, $g$ are the same as above, $\delta_\tau \in \mathcal D'$ is a scalar delta-function, $\iota=(1,\dots,1)^{\top}$.

1083: 

1084: As is mentioned in \cite{Ses}, it is reasonable to expect that in the general case a solution $x$ of (\ref{bad}) is discontinuous at point $\tau$. Thus, in the general case, the notation in (\ref{bad}) is incorrect from the point of view of distribution theory, since (\ref{bad}) contains the product of a discontinuous function $g(\tau,x(\cdot))$ and a distribution $\delta_\tau \in \mathcal D'$, which is undefined in  $\mathcal D_n'$.

1085: In \cite{Silv,Ramp,Ses} the following definition of a solution of (\ref{bad}) is proposed. The left-continuous function $x \in \mathbb{BV}_n$ is said to be a solution of (\ref{bad}), if there exists a delta-sequence $\{\omega_m\}_{m=1}^\infty$, $\omega_m \to \delta_\tau$ in $\mathcal D'$, such that 

1086: \begin{equation}

1087: \label{xconv}

1088: x_m \to x

1089: \end{equation}

1090: ($m \to \infty$) in a weak topology of $\mathbb{BV}_n$, where $x_m \in \mathbb{AC}_n$ is a solution of the approximating problem  

1091: \begin{equation}

1092: \dot{x}=f(t,x)+\bigl(g(t,x)\iota\bigr)\omega_m(t), \quad x(t_0)=x_0.

1093: \end{equation}

1094: 

1095: The necessary and sufficient condition for the uniqueness of solution of (\ref{bad}), i.e., its independence on the choice of $\{\omega_m\}_{m=1}^\infty$, is a Frobenius condition

1096: \begin{equation}

1097: \label{frob}

1098: %[g^k,g^l]_x \equiv 0 

1099: \sum_{i=1}^n \frac{\partial g_{ij}}{\partial x_k}g_{km}=

1100: \sum_{i=1}^n \frac{\partial g_{im}}{\partial x_k}g_{kj}

1101: \end{equation}

1102: for all $t \in I$, $x \in \bf R^n$, $k,j,m=1,\dots,n$, $g=(g_{ij})_{ij=1}^n$ \cite{Ses,Mil}. If (\ref{frob}) is satisfied,

1103: then there exists a unique solution $x \in \mathbb{BV}_n$ of (\ref{bad}), such that 

1104: \begin{equation}

1105: \dot{x}=f(t,x(t)) 

1106: \end{equation}

1107: for all $t \ne \tau$, 

1108: and

1109: \begin{equation}

1110: \label{bad2}

1111: \dot{\eta}(s)=g(\tau,\eta(s))\iota, \quad \eta(-1/2)=x(\tau-),

1112: \end{equation}

1113: where $x(\tau+)=\eta(1/2)$ \cite{Ses}.

1114: 

1115: Notice that (\ref{bad2}) coincides with (\ref{2}) for $\alpha \equiv \iota$. Indeed, for (\ref{2}) the Frobenius condition (\ref{frob}) is a necessary and sufficient condition for the independence of the value $\gamma(1/2)$ on the choice of $\alpha$ satisfying (\ref{normalization}) \cite{Guyshun}.

1116: Also, if (\ref{frob}) is satisfied, then a solution of (\ref{bad}) coincides with the ordinary solution of (\ref{eq10}).

1117: 

1118: The comparison of the results for equation (\ref{eq10}), which is considered in the space of distributions $\mathcal T_n'$, and the results for equation (\ref{bad}), which is considered in the classical space of distributions $\mathcal D_n'$, shows, that in contrast to the approach based on the space $\mathcal D_n'$ \cite{Silv,Ramp,Ses}, the use of the space of distributions $\mathcal T_n'$ allows us to make notations correct from the point of view of distribution theory, and to eliminate the restrictive Frobenius condition (\ref{frob}), which can not be satisfied in many problems of optimal control theory \cite{Mil}.

1119: \end{remark}

1120: 

1121: %Let us note that the dynamic value $x(\tau)(\cdot)$ describes the

1122: %approximating sequence for the solution.

1123: 

1124: 

1125: 

1126: 

1127: 

1128: 

1129: 

1130: \section{Multiplication of distributions by discontinuous functions in other approaches \\ [3mm]}

1131: 

1132: 

1133: 

1134: 

1135: \subsection{Multiplication by discontinuous functions in the Schwartz space of distributions}

1136: 

1137: %\subsection{Distributions with continuous test functions}

1138: Various definitions of the product of a distribution and a discontinuous function in $\mathcal D'$ were proposed in \cite{Ses,Fil,Kur,Kur2,Tvr1}, and also in \cite{Sar,Sar2,Sar3}, where the operation of multiplication of a distribution by a discontinuous function is induced by the operation of multiplication of two distributions.

1139: 

1140: 

1141: Furthermore, we concentrate on the multiplication of the Heaviside function $\theta_\tau$ and the delta-function. The product $\theta_\tau\delta_\tau \in \mathcal D'$ is defined in \cite{Ses,Fil,Kur,Kur2,Tvr1,Sar,Sar2,Sar3} by the formula

1142: \begin{equation}

1143: \label{prod100}

1144: \theta_\tau\delta_\tau=c\delta_\tau,

1145: \end{equation}

1146: where $c \in \bf R$ is different in different definitions. In particular, 

1147: in \cite{Sar,Sar2,Sar3}, where the family of the products of distributions $\{\cdot_\sigma\}_{\sigma \in \mathcal D}$ is defined (here $\mathcal D$ is a space of infinitely differentiable test functions),

1148: %in the context of studies of the ordinary and partial differential equations, 

1149: for any $\sigma \in \mathcal D$ we have \\ [-6mm]

1150: \begin{equation}

1151: \theta_\tau \cdot_{\sigma}\delta_\tau=\sigma(0)\delta_\tau.

1152: \end{equation}

1153: 

1154: The value $c=1/2$ is considered in \cite{Fil,Kur,Kur2}). If $c=1/2$, then we have that $(\theta_\tau \theta_\tau)^{\cdot}=\theta_\tau\delta_\tau+\delta_\tau\theta_\tau$, i.e., Leibniz product rule holds.

1155: Also, the choice of the value $c=1/2$ is motivated by the symmetry of the delta-function $\delta_\tau \in \mathcal D'$.

1156: Another definition of the form (\ref{prod100}) was proposed in \cite{Ses}, so the solution of the Cauchy problem (\ref{bad}) in $\mathcal D_n'$ coincides with the solution (\ref{xconv}). In \cite{Ses} the value of $c \in \bf R$ is determined by the right-hand side of the equation (\ref{bad}).

1157: 

1158: Nevertheless, any definition of the product of the form (\ref{prod100}) leads to contradictions with distribution theory, which are unavoidable in $\mathcal D'$.

1159: Namely, let $c \in \bf R$ be given. Then we have to be able to replace delta-function in (\ref{prod100}) by the terms of arbitrary delta-sequences $\{\omega_n\}_{n=1}^\infty$, $\{\lambda_n\}_{n=1}^\infty$ in $\mathcal D'$ so that

1160: \begin{equation}

1161: \label{prod101}

1162: \lim_{n \to \infty} \theta_\tau \omega_n=c\lim_{n \to \infty} \lambda_n

1163: \end{equation}

1164: in $\mathcal D'$. Nevertheless, for a given $c \in \bf R$ one can easily find delta-sequences $\{\omega_n\}_{n=1}^\infty$, $\{\lambda_n\}_{n=1}^\infty$ such that (\ref{prod101}) is not true. This implies that the operation of multiplication (\ref{prod100}) is not continuous in $\mathcal D'$.

1165: Also, if $c \ne 0$, $c \ne 1$, then the operation of multiplication (\ref{prod100}) is not associative. 

1166: 

1167: In $\mathcal T'$, the product of the Heaviside function $\theta_\tau$ and the delta-function $\delta_\tau^\alpha \in \mathcal T'$ is given by the formula

1168: \begin{equation}

1169: \label{prod102}

1170: \theta_\tau\delta_\tau^\alpha=\biggl(\int_0^{1/2}\alpha(s)ds\biggr)\delta_\tau^\gamma,

1171: \end{equation}

1172: (Example \ref{ex2}), where $\delta_\tau^\gamma \in \mathcal T'$. Let us notice the following:

1173: 

1174: 1) Any definition of the form (\ref{prod100}) can be obtained from (\ref{prod102}) for the delta-functions $\delta_\tau^\alpha \in \mathcal T'$ having the shape $\alpha$ such that \\ [-5mm]

1175: \begin{equation*}

1176: \int_0^{1/2}\alpha(s)ds=c.

1177: \end{equation*}

1178: 

1179: 2) In contrast to (\ref{prod100}), the operation of multiplication (\ref{prod102}) is associative, satisfies the Leibniz product rule (Theorem \ref{leibnitzteo}), continuous and, furthermore, has Property $\mathcal{B}$ (see the Introduction and Remark \ref{rem1}). 

1180: 

1181: Thus, consideration of the product of the Heaviside function and the delta-function in $\mathcal T'$ allows us to avoid the aforementioned contradictions with distribution theory. 

1182: 

1183: %In \cite{Col1} the differential algebra of 

1184: %generalized functions $\mathcal G(\mathbf R^n)$ was introduced, which contains the space of distributions $\mathcal D'(\mathbf R^n)$ as a subspace (further $\mathcal D'(\mathbf R^n)$ is a space of distributions with infinitely differentiable test functions defined on $\mathbf R^n$).

1185: %Any generalized function $f \in \mathcal G(\mathbf R^n)$ has value at any point of $\mathbf R^n$, which is the element of the generalized real numbers algebra $\hat{\mathbf R} \supset \mathbf R$.

1186: %In $\mathcal G(\mathbf R^n)$ the product of any two generalized functions exists, so, in particular, in $\mathcal G(\mathbf R^n)$ exists the product of a distribution and a discontinuous function.

1187: %In the general case, the product of two distributions is a generalized function,

1188: %but not a distribution, e.g. the product $\theta_\tau\delta_\tau \in \mathcal G(\mathbf R^1)$ but $\theta_\tau\delta_\tau \not\in \mathcal D'(\mathbf R^1) \subset \mathcal G(\mathbf R^1)$. As a result, in contrast to multiplication in $\mathcal T'$ (Examples \ref{ex21} and \ref{ex2}), the product of the Heaviside function and the delta-function in $\mathcal G(\mathbf R^n)$ can not be represented as a limit of an arbitrary delta-sequence (multiplied by-term by certain coefficient). 

1189: %

1190: %

1191: %

1192: %Also, in \cite{Col1} the algebra of tempered generalized functions $\mathcal G^{T}_n(\mathbf R^n)$ was introduced, which contains the space of tempered distributions $\mathcal S'(\mathbf R^n)$ \cite{Shi} as a subspace. In what follows, we denote by $\mathcal G_n(\mathbf R^n)$, $\mathcal G_{n \times n}(\mathbf R^n)$ ($\mathcal G^T_n(\mathbf R^n)$, $\mathcal G^T_{n \times n}(\mathbf R^n)$) the spaces of generalized functions with values in $\mathbf R^n$, $\mathbf R^{n \times n}$, respectively. 

1193: %

1194: %As a major difference between the space of distributions $\mathcal T'(\mathbf R)$ and the Colombeau generalized functions algebra $\mathcal G(\mathbf R)$ we may point out the existence in $\mathcal T'(\mathbf R)$ of infinitely many delta-functions concentrated at the same point (having different \textit{shapes}).

1195: %

1196: %In \cite{Obe} the following initial value problem 

1197: %\begin{equation}

1198: %\label{obeeq}

1199: %\dot{x}=f(t,x), \quad x(t_0)=x_0,

1200: %\end{equation}

1201: %is considered, where $t_0 \in \mathbf R$, $x_0 \in \hat{\mathbf R}^n$, generalized function $f \in \mathcal G^T_n(\mathbf R \times \mathbf R^n)$ is of $L^\infty$-log type (see \cite{Obe} for details). Then solution of the problem (\ref{obeeq}) exists and unique in $\mathcal G_n(\mathbf R)$. In particular, consideration of the nonlinear ordinary differential equation (\ref{obeeq}) in $\mathcal G_n(\mathbf R^n)$ allows to provide existence of solution to the initial value problem

1202: %\begin{equation}

1203: %\ddot{x}(t)+\dot{\delta}(x(t))=0,

1204: %\end{equation}

1205: %\begin{equation}

1206: %x(0)=x_0, \quad \dot{x}(0)=x_1,

1207: %\end{equation}

1208: %where $x_0$, $x_1 \in \mathbf R$.

1209: %

1210: %As is mentioned in \cite[p.71]{Col1}, generalized solution $x \in \mathcal G_n(\mathbf R^n)$ may have no obvious physical interpretation. In our approach, which is less general that the approach of \cite{Col1,Obe}, the ordinary solution of initial value problem (\ref{eq10}), which is a particular case of (\ref{obeeq}), is always a function of bounded variation, and thus have a physical interpretation \cite{Ses,Silv,Ramp,Mil}. \\

1211: 

1212: 

1213: \subsection{Multiplication by discontinuous functions in the Colombeau generalized functions algebra} 

1214: 

1215: As is mentioned in the Introduction, the general multiplication of distributions is impossible in the classical space of Schwartz distributions with continuous test functions.

1216: This defect of the Schwartz distribution theory lead to the construction of the Colombeau generalized functions algebra $\mathcal G$, containing the space of distributions $\mathcal D'$ as a subspace (here $\mathcal D'$ is a space of distributions with infinitely differentiable test functions), where such general multiplication exists \cite{Col1} (see description of the Colombeau algebra in \cite{ColM3,Obe}). The algebra of Colombeau generalized functions was applied successfully to many problems related to ordinary and partial differential equations, e.g., see \cite{Obe,ColM1,ColM2,ColM3}. 

1217: In the classical approach the Heaviside function $\theta_\tau \in \mathcal G$ and the Dirac delta-function $\delta_\tau \in \mathcal G$ are defined as the derivatives of continuous functions in $\mathcal G$, so

1218: $\theta_\tau$, $\delta_\tau \in \mathcal D' \subset \mathcal G$ and the product

1219: \begin{equation*}

1220: \theta_\tau\delta_\tau \in \mathcal G \setminus \mathcal D'

1221: \end{equation*}

1222: is not a distribution \cite{Col1}. However, as was observed in \cite{ColM1},\footnote{We would like to thank Professor J.F.Colombeau who kindly mentioned these results to us.} instead of consideration of the elements of the canonical embedding $\mathcal D' \subset \mathcal G$, often it is natural to work with the algebra $\mathcal G$ itself. Namely, a generalized function $f \in \mathcal G$ is said to have $\theta_\tau$ or $\delta_\tau$ as its \textit{macroscopic profile}, if 

1223: \begin{equation*}

1224: f \approx \theta_\tau \text{ or } f \approx \delta_\tau, 

1225: \end{equation*}

1226: respectively, where $\approx$ stands for the \textit{association} in $\mathcal G$. If $f$ has $\theta_\tau$ or $\delta_\tau$ as a macroscopic profile, then $f$ is called the \textit{Heaviside generalized function} or the \textit{Dirac generalized function}, respectively \cite{ColM1}. According to \cite{ColM1}, there exist different Heaviside generalized functions and Dirac generalized functions, which differ by their \textit{microscopic profiles}. The equalities of the form 

1227: \begin{equation}

1228: \label{colomeq}

1229: \theta_{1,\tau}\delta_{2,\tau}=c\delta_{3,\tau}, \\ [2mm]

1230: \end{equation}

1231: where $\theta_{1,\tau} \in \mathcal G$ is a Heaviside generalized function,

1232: $\delta_{2,\tau}$, $\delta_{3,\tau} \in \mathcal G$ are two Dirac generalized functions, were studied in \cite{ColM3}, where the value of constant $c \in \mathbf R$ was derived from the supplementary partial differential equation, arising from the statement of the physical problem. 

1233: As follows from the definition of the algebra $\mathcal G$ \cite{Obe}, equality (\ref{colomeq}) can be also obtained if the Heaviside generalized function and the Dirac generalized functions in (\ref{colomeq}) are approximated by continuous functions.

1234: 

1235: The equality (\ref{colomeq}) is similar to the equality (\ref{prod_w1}) in the space of distributions $\mathcal T'$, where the constant $c \in \mathbf R$ is given explicitly (see also Remark \ref{rem1}). 

1236: However, despite this similarity, space $\mathcal T'$ also possesses other properties. In particular, $\mathcal T'$ contains delta-functions (\ref{newdelta}) (see Remark \ref{rm_newdelta}), which allow us to unify in the subsequent paper different definitions of solution of the ordinary differential equations with distributions (see \cite{Fil} on ambiguities arising in the definition of the notion of solution of the differential equations with distributions). \\ [1mm]

1237: 

1238: 

1239: 

1240: 

1241: \section{Proofs of statements of Sections 2 and 3 \\ [3mm]}

1242: 

1243: \begin{proof}[Proof of Lemma \ref{lem1}]

1244: Suppose that there exists an uncountable set $A \subset I$ such that

1245: $f(t)(\cdot)$ is not ordinary for all $t \in A$. Let us define a function $\hat{g}:I \to \bf R$ by

1246: \begin{equation}

1247: \label{dyn_g}

1248: \hat{g}(t)=\sup_{s \in J} f(t)(s)-\inf_{s \in J} f(t)(s)

1249: \end{equation}

1250: Since $f \in d\mathbb G$ is bounded, $0 \leqslant \hat{g}(t)< \infty$ for all $t \in I$. Obviously, 

1251: $f$ has an ordinary value at $t$ if and only if $\hat{g}(t)=0$, so $\hat{g}(t)>0$ for all $t \in A$. We define

1252: \begin{equation*}

1253: A_n=\biggl\{t \in A: \frac{1}{n+1}<\hat{g}(t)\leqslant \frac{1}{n}\biggr\} \quad (n=1,2,\dots),

1254: A_0=\{t \in A: \hat{g}(t)>1\}.

1255: \end{equation*}

1256: Then $A=\cup_{n=0}^\infty A_n$. Since $A$ is uncountable, there exists $n_0 \geqslant 0$ such that $A_{n_0}$ is uncountable. Without loss of generality we may assume that $n_0 \ne 0$. Since $A_{n_0}$ is uncountable, there exists a closed interval $[c,d] \subset I$ such that $A_{n_0}' \doteq A_{n_0} \cap [c,d]$ is uncountable. Consequently, due to compactness of $[c,d]$, the set $A_{n_0}'$ has a limit point $\xi \in [c,d]$. Without loss of generality we may assume that there exists a right-sided neighborhood of $\xi$ which contains infinitely many points of $A_{n_0}'$. Let us show that this implies that $f(\xi+)$ does not exist.

1257: Suppose that $p=f(\xi+) \in \bf R$ exists. Then by definition for any $\varepsilon>0$ there exists $\eta>0$ such that $\|f-p\|_{(\xi,\xi+\eta)}<\varepsilon$, i.e.,

1258: $\sup_{t \in (a,b}\sup_{s \in J}|f(t)(s)-p|<\varepsilon$.

1259: Now consider the value of 

1260: $\sup_{s \in J}|f(t)(s)-p|$, 

1261: for $t \in (\xi,\xi+\eta) \cap A_{n_0}'$,

1262: where $(\xi,\xi+\eta) \cap A_{n_0}' \ne \varnothing$. Then the inequality 

1263: %\begin{multline}

1264: %\label{dyn_ineq1}

1265: %\sup\limits_{s \in J}|f(t)(s)-p| \geqslant \frac{1}{2} \biggl(\sup_{s \in J}(f(t)(s)-p)-\inf_{s \in J}(f(t)(s)-p) \biggr)=\\

1266: %=\frac{1}{2} \biggl(\sup_{s \in J]}(f(t)(s))-\inf_{s \in J}(f(t)(s)) \bigr)

1267: %\end{multline}

1268: \begin{equation}

1269: \label{dyn_ineq1}

1270: \sup\limits_{s \in J}|f(t)(s)-p| \geqslant 

1271: \frac{1}{2} \biggl(\sup_{s \in J}(f(t)(s))-\inf_{s \in J}(f(t)(s)) \bigr)

1272: \end{equation}

1273: holds (see proof below). As follows from the definitions of $A_{n_0}'$ and $\hat{g}$, 

1274: \begin{equation*}

1275: \frac{1}{2}(\sup_{s \in J}(f(t)(s))-\inf_{s \in J}(f(t)(s))) \geqslant 1/n_0 

1276: \end{equation*}

1277: for all $t \in A_{n_0}' \cap (\xi,\xi+\eta)$. Consequently, 

1278: $\sup_{t \in (\xi,\xi+\eta)}\sup_{s \in J}|f(t)(s)-p| \geqslant \frac{1}{2n_0}$.

1279: Since for any $\eta>0$ $A_{n_0}' \cap (\xi,\xi+\eta) \ne \varnothing$, the last inequality holds for any $\eta>0$. This contradicts to the definition of the right-sided limit $p=f(\xi+)$.

1280: 

1281: Let us show that the inequality (\ref{dyn_ineq1}) holds. If $f(t)(s)-p \geqslant 0$ for all $s \in J$, then $\sup_{s \in J}|f(t)(s)-p|=\sup_{s \in J}(f(t)(s)-p)$ and (\ref{dyn_ineq1}) holds, where  $\int_{s \in J}(f(t)(s)-p) \geqslant 0$. In the general case we can represent $f(t)(s)=q^+(s)-q^-(s)$, where $q^+(s)$, $q^-(s) \geqslant 0$. Then

1282: \begin{equation*}

1283: \sup_{s \in J}|f(t)(s)-p|=\max\{\sup_{s \in J}q^+(s),\sup_{s \in J}q^-(s)\}

1284: \geqslant 1/2(\sup_{s \in J}q^+(s)+\sup_{s \in J}q^-(s)).

1285: \end{equation*}

1286: Since $\sup_{s \in J}(f(t)(s)-p)=\sup_{s \in J} q^+(s)$, $-\inf_{s \in J}(f(t)(s)-p)=\sup_{s \in J} q^-(s)$, we obtain that (\ref{dyn_ineq1}) is true.

1287: \end{proof}                                                                 

1288: 

1289: \begin{proof}[Proof of Lemma \ref{lem2}]

1290: For a given $\tau \in I$ let us denote $p=f(\tau+)$. Then for any $\varepsilon>0$ there exits $\eta>0$ 

1291: such that $\sup_{(\tau,\tau+\eta)}|p-f|<\varepsilon$. According to Lemma \ref{lem1}, the intersection

1292: $(\tau,\tau+\eta) \cap \us(f) \ne \varnothing$. Then

1293: \begin{equation*}

1294: \sup_{t \in (\tau,\tau+\eta) \cap us(f)}|p-\hat{f}(t)|=

1295: \sup_{t \in (\tau,\tau+\eta)\cap us(f)}\sup_{s \in J}|p-\hat{f}(t)| 

1296: \leqslant

1297: \sup_{(\tau,\tau+\eta)\cap us(f)}|p-f| 

1298: \end{equation*}

1299: where the first equality holds since $|p-\hat{f}(t)|$ does not depend on

1300: $s \in J$. Thus, for any $t \in (\tau,\tau+\eta)$, where $\hat{f}(t)$ is defined, i.e, for all

1301: $t \in (\tau,\tau+\eta) \cap \us(f)$, we have $|p-\hat{f}(t)|<\varepsilon$. Then the right-sided limit

1302: $\hat{f}(\tau+)$ exists, which is equal to $f(\tau+)$. The proof for the left-sided limit is analogous.

1303: Since $\tau \in I$ was chosen arbitrarily, the statements of the lemma holds.

1304: \end{proof}                           

1305: 

1306: \begin{proof}[Proof of Lemma \ref{lem3}]

1307: The proof follows from an analogous statement 

1308: for the space of regulated functions $\mathbb G$ \cite{Hon}, the definition of a point of discontinuity of a dynamic function, and Lemma \ref{lem2}.

1309: \end{proof}

1310: 

1311: \begin{proof}[Proof of Theorem \ref{limitteo}]                             

1312: The linearity of the limit functional is obvious. Without loss of generality it suffices to show that $f$ is continuous at $\varphi=0$ only.

1313: Let $\varphi_n \to 0$ in $\mathcal T$. Suppose that $(f,\varphi_n) \not\to 0$ ($n \to \infty$).

1314: Being considering, if necessary, a subsequence of $\{(f,\varphi_n)\}_{n=1}^\infty$, we may assume that

1315: for any $n=1,2,\dots$ the inequality $|(f,\varphi_n)|>\varepsilon_0$ holds for certain $\varepsilon_0>0$.

1316: Due to (\ref{limit}) for any $k=1,2,\dots$ there exists $n_k$ such that $|(f_{n_k},\varphi_k)|>\frac{\varepsilon_0}{2}$.

1317: Without loss of generality we may assume that $n_k=k$, i.e., $|(f_k,\varphi_k)|>\frac{\varepsilon_0}{2}$ for

1318: any $k=1,2,\dots$ The last inequality contradicts to the conditions of the lemma. Thus, $(f,\varphi_n) \to 0$ as $n \to \infty$, i.e., the limit functional $f$ is continuous.

1319: \end{proof}

1320: 

1321: \begin{proof}[Proof of Theorem \ref{ext_teo}]

1322: According to Theorem \ref{topteo} the space of dynamic test functions $\mathcal T$ is locally-convex;

1323: $\mathcal D$ is a subspace of $\mathcal T$. Then according to the Hahn-Banach theorem \cite{Kan},

1324: every linear continuous functional defined on a subspace $\mathcal D$ has an extension to the whole space $\mathcal T$.

1325: %Further, the restriction of distribution $\delta-\tau^\alpha-\delta_\tau^\gamma \in \mathcal T'$, where $\alpha$, $\gamma \in \mathbb L(J)$ satisfying normalization condition are norms of delta-functions, from $\mathcal T$ to $\mathcal D$ is a zero distribution in $\mathcal D'$. Thus, since forms $\alpha$, $\gamma$ can be chosen arbitrarily, 

1326: %addition of this distribution to extension of any other distribution from $\mathcal D$ to $\mathcal T$ gives different extensions.

1327: \end{proof} 

1328: 

1329: \begin{proof}[Proof of Theorem \ref{cont_teo}]

1330: Note that $g_n \varphi \xrightarrow {} g\varphi$ in $\mathcal T$

1331: for any $\varphi \in \mathcal T$. Consequently

1332: \begin{multline}

1333: \notag

1334: |(g_nf_n,\varphi)-(gf,\varphi)|=|(f_n,g_n\varphi)-(f,g\varphi)|

1335: \leqslant |(f_n,g_n\varphi)-(f_n,g\varphi)|+\\+|(f_n,g\varphi)-(f,g\varphi)|

1336: \leqslant |(f_n,g_n\varphi-g\varphi)|+|(f_n,g\varphi)-(f,g\varphi)| \xrightarrow {} 0

1337: \end{multline}

1338: (the first and the second summands tend to zero according to Lemma \ref{lem4} 

1339: and due to convergence $f_n \to f$ in $\mathcal T'$, respectively).

1340: \end{proof}

1341: 

1342: \begin{proof}[Proof of Theorem \ref{dercorteo}]

1343: The Stieltjes integral in the right-hand side of (\ref{dyn_deriv}) exists since

1344: $\hat{\varphi} \in \mathbb G$, $\hat{g}_c \in \mathbb{CBV}$ (see \cite{Der2}), and the

1345: convergence of the series

1346: follows from the inequality

1347: %\begin{multline}

1348: %\notag

1349: \begin{equation*}

1350: \biggl|\int_J\varphi(\tau)(s)(g(\tau)(s))^{\cdot}_sds\biggr| 

1351: \leqslant \sup_{s \in J}|\varphi(\tau)(s)|\int_J|g(\tau)(s)^{\cdot}_s|ds

1352: \leqslant

1353: \sup_{s \in J}|\varphi(\tau)(s)|\var_{s \in J}(g(\tau)(s))

1354: \end{equation*}

1355: %\end{multline}

1356: for all $\tau \in T(g)$.

1357: Then the following inequality holds

1358: \begin{equation*}

1359: \sum_{\tau \in T(g)}\biggl|\int_J\varphi(\tau)(s)(g(\tau)(s))^{\cdot}_sds\biggr| \leqslant \sup_{I}|\varphi| \\ \sum_{\tau \in T(g)}\var_{s \in J}(g(\tau)(s)),

1360: \end{equation*}

1361: where convergence of series in the right-hand side follows from the definition of the algebra $s\mathbb{BV}$. 

1362: Thus, the value of $\dot{g}$ is defined for all test functions $\varphi \in \mathcal T$.

1363: The linearity and continuity of $\dot{g}$ 

1364: follow from definition of the convergence in $\mathcal T$, 

1365: properties of the Stieltjes integral and properties of convergent series. 

1366: \end{proof}

1367: 

1368: %\begin{proof}[Proof of Theorem \ref{linteo}]

1369: %According to the definition of linear operations in algebra $s\mathbb{BV}$,

1370: %$(\lambda f+\mu g)(t)(s)=\lambda f(t)(s)+\mu g(t)(s)$. Then as follows from 

1371: %the definition of derivative, for any $\varphi \in \mathcal T$

1372: %\begin{multline}

1373: %\notag

1374: %((\lambda f+\mu g)^{\cdot},\varphi)=\int_I \hat{\varphi}(t)d\widehat{(\lambda f+\mu g)}_c(t)+

1375: %\sum_{\tau_k \in T(\lambda f+\mu g)} \int_J(\lambda f+\mu g)(\tau_k)(s)\varphi(\tau_k)(s)ds=\\

1376: %=\lambda \int_I \hat{\varphi}d\hat{f}_c(t)+\lambda \sum_{p_j \in T(f)} \int_Jf(p_j)(s)\varphi(p_j)(s)ds+\\

1377: %+\mu\int_I \hat{\varphi}(t)d\hat{g}_c(t)+\mu\sum_{r_l \in T(g)}

1378: %\int_J f(r_l)(s)\varphi(r_l)(s)ds

1379: %\end{multline}

1380: %Note that though in the general case $T(f) \ne T(\lambda f+\mu g)$, $T(g) \ne T(\lambda f+\mu g)$, the equality of sums above remains true (as follows from definition of 

1381: %linear operations in $s\mathbb{BV}$).

1382: %\end{proof}

1383: 

1384: \begin{proof}[Proof of Theorem \ref{leibnitzteo}]

1385: We have

1386: \begin{equation}

1387: \label{fg}

1388: (\dot{f}g,\varphi)=(\dot{f},g\varphi)=\int_I \hat{\varphi}(t)\hat{g}(t)df_c(t)+\sum_{\tau \in T(f)}\int_J g(\tau)(s)\varphi(\tau)(s)(f(\tau)(s))^{\cdot}_s ds,

1389: \end{equation}

1390: \begin{equation}

1391: \label{gf}

1392: (f\dot{g},\varphi)=(\dot{g},f\varphi)=\int_I \hat{\varphi}(t)\hat{f}(t)dg_c(t)+\sum_{\tau \in T(g)}\int_J f(\tau)(s)\varphi(\tau)(s)(g(\tau)(s))^{\cdot}_s ds.

1393: \end{equation}

1394: In (\ref{fg}) and (\ref{gf}) we can perform the summation by $T(f) \cup T(g)$, since we need to add zero summands only. Consequently,

1395: \begin{multline}

1396: \label{fggf}

1397: \sum_{\tau \in T(f)}\int_J g(\tau)(s)\varphi(\tau)(s)(f(\tau)(s))^{\cdot}_s ds+\sum_{\tau \in T(g)}\int_J f(\tau)(s)\varphi(\tau)(s)(g(\tau)(s))^{\cdot}_s ds=\\

1398: =\!\sum_{\tau \in T(f) \cup T(g)}\int_J \varphi(\tau)(s)(g(\tau)(s)f(\tau)(s))^{\cdot}_sds=\!\sum_{\tau \in T(fg)}\!\int_J \varphi(\tau)(s)(g(\tau)(s)f(\tau)(s))^{\cdot}_sds,

1399: \end{multline}

1400: where the last equality is due to $T(fg) \subset T(f) \cup T(g)$ and $\tau \in T(f) \cup T(g) \setminus T(fg)$ if and only if $f(\tau)=0$ or $g(\tau)=0$, so we may exclude in (\ref{fggf}) the summands corresponding to $\tau \in T(f) \cup T(g) \setminus T(fg)$. We have

1401: \begin{equation}

1402: \label{c2}

1403: \int_I \hat{\varphi}(t)\hat{g}(t)df_c(t)+\int_I\hat{\varphi}(t)\hat{f}(t)d_gc(t)=\int_I \hat{\varphi}(t)d\biggl(\int_a^t \hat{g}(\xi)df_c(\xi)+\int_a^t \hat{f}(\xi)dg_c(\xi) \biggr),

1404: \end{equation}

1405: where $I=(a,b)$.

1406: Let us show that the following equality holds

1407: \begin{equation}

1408: \label{c}

1409: (fg)_c(t)=\int_a^t \hat{g}(\xi)df_c(\xi)+\int_a^t \hat{f}(\xi)dg_c(\xi)+f(a+)g(a+)

1410: \end{equation}

1411: for all $t \in I$. Now $(fg)_c=(\hat{f}\hat{g})_c$, $f_c=\hat{f}_c$, $g_c=\hat{g}_c$, thus we may prove (\ref{c}) for the ordinary parts only. Hence

1412: \begin{equation*}

1413: \hat{f}(t)\hat{g}(t)=(\hat{f}_c(t)+\hat{f}_h(t))(\hat{g}_c(t)+\hat{g}_h(t))\!=\!

1414: \hat{f}_c(t)\hat{g}_c(t)+\hat{f}_c(t)\hat{g}_h(t)+\hat{g}_c(t)\hat{f}_h(t)+\hat{f}_h(t)\hat{g}_h(t)

1415: \end{equation*}

1416: for all $t \in I$. Thus,

1417: \begin{multline}

1418: \notag

1419: \int_a^t \hat{g}(\xi)d\hat{f}_c(\xi)+\int_a^t \hat{f}(\xi)d\hat{g}_c(\xi)+f(a+)g(a+)=\\=\int_a^t\hat{g}_c(\xi)d\hat{f}_c(\xi)+\int_a^t\hat{g}_h(\xi)d\hat{f}_c(\xi)+\int_a^t\hat{f}_c(\xi)d\hat{g}_c(\xi)+\int_a^t \hat{f}_h(\xi)d\hat{g}_c(\xi)=\\

1420: =\hat{f}_c(t)\hat{g}_c(t)-\hat{f}_c(a)\hat{g}_c(a)+f(a+)g(a+)+\hat{f}_c(t)\hat{g}_h(t)-\\-\sum_{a<\tau_i<t}\hat{f}_c(\tau_i)(g(\tau_i+)-g(\tau_i-))+\hat{g}_c(t)\hat{f}_h(t)-\sum_{a<\tau_i<t}\hat{g}_c(\tau_i)(f(\tau_i+)-f(\tau_i-)),

1421: \end{multline}

1422: for all $t \in I$,

1423: where $\hat{f}_c(a)\hat{f}_c(a)=f(a+)g(a+)$ and

1424: \begin{equation*}

1425: \hat{f}_c(t)\hat{g}_h(t)-\sum_{a<\tau_i<t}\hat{f}_c(\tau_i)(g(\tau_i+)-g(\tau_i-))

1426: \end{equation*}

1427: is a continuous part of $\hat{f}_c\hat{g}_h$,

1428: \begin{equation*}

1429: \hat{g}_c(t)\hat{f}_h(t)-\sum_{a<\tau_i<t}\hat{g}_c(\tau_i)(f(\tau_i+)-f(\tau_i-))

1430: \end{equation*}

1431: is a continuous part of  $\hat{g}_c\hat{f}_h$,

1432: so equality (\ref{c}) is true.

1433: The comparison of (\ref{fg}) -- (\ref{c}) with the definition of the derivative in $\mathcal T'$ gives us (\ref{dyn_leibnitz}).

1434: %As follows from the definitions of the derivative and the product,

1435: %\begin{equation*}

1436: %((fg)^{\cdot},\varphi)=\int_I \hat{\varphi}(t)d(\hat{f}_c(t)\hat{g}_c(t))+\sum_{\tau \in D(fg)}\int_J\varphi(\tau)(s)(f(\tau)(s)g(\tau)(s))^{\cdot}_sds 

1437: %\end{equation*}

1438: %and, so,

1439: %\begin{equation*}

1440: %(\dot{f}g,\varphi)=(\dot{f},g\varphi)=\int_I (\hat{\varphi}(t)\hat{g}_c(t))d\hat{f}_c(t)+\sum_{\tau \in D(f)}\int_J\varphi(\tau)(s)g(\tau)(s)(f(\tau)(s))^{\cdot}_sds 

1441: %\end{equation*}

1442: %for any $\varphi \in \mathcal T$. The value of $(f\dot{g},\varphi) \in \bf R$ is defined analogously. 

1443: %Comparison of the obtained equalities gives (\ref{dyn_leibnitz}).

1444: \end{proof}       

1445: 

1446: 

1447: 

1448: \textbf{Acknowledgments.} We would like to thank Professor J.F.~Colombeau for very important discussion and comments, Professor A.~Brudnyi for his valuable support,  Professor L.~Bates and Professor P.~Zvengrowski for their attention and help.

1449: 

1450:            

1451: \bibliographystyle{amsalpha}

1452: \bibliography{eng}

1453: 

1454: \end{document}

1455: