Uncertain differential equations (Q318838)

The theory of uncertain differential equations is an attempt of describing indeterministic dynamic systems, different from the Itô stochastic calculus. The book under review is the first textbook on this topic. The author, Kai Yao, is an associate professor in the School of Economics and Management, University of Chinese Academy of Sciences in Beijing. He has numerous papers on uncertain differential equations and their applications. The uncertainty theory was invented and developed by \textit{B. Liu} [Uncertainty theory. 4th ed. Berlin: Springer (2015; Zbl 1309.28001)]. The author starts with the presentation of basic notions of uncertainty theory. An uncertainty space is the triple \((\Gamma ,\mathscr{L}, \mathscr{M})\), where \(\Gamma \) is a set, \(\mathscr{L}\) is a \(\sigma \)-algebra, and \(\mathscr{M}\) is a set function on \(\mathscr{L}\) which is non-negative, subadditive and satisfies \(\mathscr{M}(\Gamma)=1\), and \(\mathscr{M}(\Lambda )+\mathscr(\Lambda ^c)=1\) for \(\Lambda \in \mathscr{L}\). The number \(\mathscr{M}(\Lambda )\) is interpreted as the belief degree that the event \(\Lambda \) will occur, and \(\mathscr{M}\) is called an uncertain measure. The definitions of uncertain variable, its distribution function and uncertain process are analogous as in the probability. The main difference between probability theory and uncertainty theory lies in definitions of product measure and independence. For uncertain measures \(\mathscr{M}_1\) and \(\mathscr{M}_2\) the product measure of ``a rectangle'' \(\Lambda _1\times \Lambda _2\) is the minimum of \(\mathscr{M}_1(\Lambda _1)\) and \(\mathscr{M}_2(\Lambda _2)\). The independence of uncertain variables is also defined by the minimum operation. The author introduces an interesting notion of a contour process and its \(\alpha \)-paths. Inverse distributions, expected values, extreme values and time-integrals of a contour process can be obtained via \(\alpha \)-paths. An uncertain process \(C_t\), \(t\geq 0\), is called a canonical Liu process if it has stationary independent increments being normal uncertain variables, and its almost all sample paths are Lipschitz continuous. Note, that here ``normal'' has a different meaning than in the probability theory. Liu [loc. cit.] proves that such a process there exists. This an analogue of the Wiener process in the uncertainty theory. Uncertain processes can be integrated with respect to \(C_t\), and the result is called the Liu integral. Sample paths of \(C_t\) have bounded variation on bounded intervals. Thus the Liu integral can be defined path-by-path as the Riemann-Stieltjes integral. It is not required that an integrated process is adapted to the filtration defined by \(C_t\). By a general Liu process, the author means an uncertain process \(Z_t\) which can be represented as \(Z_t=Z_0+\int_0^t \mu _sds+\int_0^t \sigma _sdC_s\), where \(\mu _t\) is a time integrable process and \(\sigma _t\) is a Liu integrable process. The last formula is shortly written in the differential form \(dZ_t=\mu _tdt+\sigma _tdC_t\). For such processes, the author gives basic rules of calculus. The central part of the book is Chapter 6 devoted to uncertain differential equations (UDEs in short). Such an equation, in the integral notation, has the form \(X_s=X_0+\int_0^s f(t,X_t)dt+\int_0^s g(t,X_t)dC_t\), where \(C_t\) is a canonical Liu process, and \(f\), \(g\) are measurable functions. An uncertain process \(X_t\) that identically satisfies this equation is called a solution. The solution of an UDE is a Liu process. The author presents analytic methods of solving some types of UDEs. Then, he gives the Yao-Chen formula, which relates UDEs and ordinary differential equations. It appears that a solution of an UDE is a contour process and its \(\alpha \)-paths solve a family of ordinary differential equations. This provides a numerical method of solving of UDS's. Next, he discusses existence, uniqueness and different types of stability of solutions of UDEs. The chapter is concluded with applications to stock markets. Subsequently, the author builds uncertain calculus with renewal processes. He defines the integral with respect to an uncertain renewal process (path-by-path), and calls it the Yao integral. Then, he considers UDEs with jumps \(X_t=X_0+\int_0^t f(s,X_s)ds+\int _0^t g(s,X_s)dC_s+\int_0^t h(s,X_s)dN_s\), where the last term is the Yao integral with respect to a renewal process \(N_t\). He presents existence, uniqueness and stability theorems for such equations, and their applications to stock markets. This is an extension of the theory from Chapter 6. The final part of the book is devoted to multi-dimensional and high-order UDEs. Again, the author discusses existence, uniqueness and stability of solutions. The book is well organized and endowed with many illustrative examples. It is suitable for mathematicians interested in the finance theory. Baoding Liu [loc. cit.] argues that UDEs better model stock price than Itô's stochastic differential equations. Reviewer's remarks: 1. The author neglects some technical details, e.g., the measurabilility of extreme values of an uncertain process, and intersections in the definition of a contour process. There is no problem if we deal with processes with continuous sample paths. 2. The proof of the Yao-Chen formula uses the comparison theorem for ordinary differential equations, but the author does not refer to any particular result.