Picard iterations for diffusions on symmetric matrices (Q501822): Difference between revisions

Let \(\mathcal{S}_{d\times d}\) be the set of symmetric \(d\times d\) matrices, and let \(b, g, f:\mathbb{R}\to\mathbb{R}\) be bounded functions that are Lipschitz in matrix sense, i.e., there exists a constant \(c>0\) such that for any pair \(A_1,A_2\in \mathcal{S}_{d\times d}\) and any unit vector \(x\in{\mathbb R}^d\), we have \[ \begin{aligned} x^\top(b(A_1) - b(A_2))^2 x \leq c x^\top (A_1-A_2)^2 x,\\ x^\top(g(A_1) - g(A_2))^2 x \leq c x^\top (A_1-A_2)^2 x,\\ x^\top(f(A_1) - f(A_2))^2 x \leq c x^\top (A_1-A_2)^2 x. \end{aligned} \] Here, if \(A\in \mathcal{S}_{d\times d}\), by \(b(A)\) one means \(Hb(\Lambda)H^\top\), where \(H\Lambda H^\top\) is the spectral decomposition of \(A\) and \(b(\Lambda)\) is the diagonal matrix with diagonal entries \(b(\lambda_1),\ldots,b(\lambda_d)\) and \(\lambda_1\leq \lambda_2\leq \cdots\leq \lambda_d\) are the eigenvalues of \(A\) increasingly ordered. The author proves that the stochastic differential equation \[ X_t = X_0 + \int_0^t b(X_s)\,\mathrm{d} s + \int_0^t g(X_s) \,\mathrm{d} B_s \, f(X_s) + \int_0^t f(X_s)\, \mathrm{d} B_s^\top \, g(X_s), \quad t\geq 0, \] has a pathwise unique strong solution in \(\mathcal{S}_{d\times d}\), where the initial value \(X_0\in \mathcal{S}_{d\times d}\) is positive semidefinite, and \((B_t)_{t\geq 0}\) is a \(d\times d\) Brownian motion, i.e., a \(d\times d\)-valued stochastic process of which the entry processes are independent Brownian motions. In the appendix of the paper, one can find the definition and some properties of the matrix stochastic integral \(\int_0^t A_s\,\mathrm{d} B_s \, C_s\), \(t\geq 0\), for \(d\times d\) matrix valued stochastic processes \((A_t)_{t\geq 0}\) and \((C_t)_{t\geq 0}\). The author applies the so-called Picard iteration method to establish the existence of a solution of the stochastic differential equation in question.