A power matrix approach to the Witt algebra and Loewner equations (Q977105)

For an integer \(p\), consider the sets of formal complex power series \[ \mathbb{C}_p[[z]]:=\bigg\{ \sum_{n=p}^\infty a_nz^n \;\Big|\; a_p\neq 0 \bigg\} \] and \[ \mathbb{C}_p[[1/z]]:=\bigg\{ \sum_{n=-\infty}^p a_nz^n \;\Big|\; a_p\neq 0\bigg\}. \] By definition, the power matrix \([f]\), for \(f\in \mathbb{C}_p[[z]]\), is the infinite triangular matrix with entries \([f]_k^n\) being the coefficients of the powers of \(f\): \((f(z))^n =\sum_{k=pn}^\infty[f]_k^nz^n\). Let \(\mathcal{M}_p(0)\) be the set of all such \([f]\), \(\mathcal{M}(0):=\mathcal{M}_1(0)\). In the same way, we can define the power matrix \([f]\) for \(f\in \mathbb{C}_p[[1/z]]\) and the sets \(\mathcal{M}_p(\infty)\) and \(\mathcal{M}(0)\). Let \(h(z)=\sum_{n=1}^\infty a_nz^n\in \mathbb{C}_{\geq 1}[[z]]:=\sum_{p\geq 1} \mathbb{C}_{p}[[z]]\). The infinitesimal power matrix \(\langle h\rangle\) of \(h\) is the infinite triangular matrix with entries \(\langle h\rangle_m^n=mh_{n-m+1}\). Let \(\mathfrak{m}(0):=\big\{\langle h\rangle \mid h\in \mathbb{C}_{\geq 1}[[z]]\big\}\). Similarly, for the power series in \(1/z\), we can define the infinitesimal power matrix and \(\mathfrak{m}(\infty)\). The author proves that differentiable one-parametric families of power series satisfy some differential equation, which he calls formal Loewner equations. More precisely, he proves Proposition~6. Let \(f_t\in \mathbb{C}_1[[z]]\) for \(t\in (a,b)\) have coefficients which are differentiable in \(t\). \(\quad\)(i) There exists an \(h_t\in \mathbb{C}_{\geq1}[[z]]\) such that \(\partial f_t/\partial t(z)=h_t(z)f'_t(z)\). The \(n\)-th coefficient of \(h_t\) depends only on the first \(n\) coefficients of \(f_t\) and \(\partial f_t/\partial t\). \(\quad\)(ii) There exists an \(\tilde{h}_t\in \mathbb{C}_{\geq1}[[z]]\) such that \(\partial f_t/\partial t(z)= \tilde{h}_t\circ f_t(z)\). The \(n\)-th coefficient of \(\tilde{h}_t\) depends only on the first \(n\) coefficients of \(f_t\) and \(\partial f_t/\partial t\). The same equation holds for series at \(\infty\). The equation in (i) is called the formal Loewner PDE, and the equation in (ii) is called the formal Loewner ODE. Proposition~6. (i) \(f_t\in \mathbb{C}_1[[z]]\) and \(h_t\in \mathbb{C}_{\geq 1}[[z]]\) satisfy the formal Loewner PDE on an interval \((a,b)\) if and only if their power matrices satisfy \(d[f_t]/dt=[f_t]\langle h_t \rangle\) on \((a,b)\). \(\quad\)(ii) \(f_t\in \mathbb{C}_1[[z]]\) and \(h_t\in \mathbb{C}_{\geq 1}[[z]]\) satisfy the formal Loewner ODE on an interval \((a,b)\) if and only if their power matrices satisfy \(d[f_t]/dt=\langle h_t \rangle [f_t]\) on \((a,b)\). The same statement holds for the formal Loewner DE at \(\infty\). There is a connection between \(\mathcal{M}(0)\) and \(\mathfrak{m}(0)\) as between Lie groups and their Lie algebras. It can be described via an exponential mapping. Further, the author investigates the exponential map for \(\langle h\rangle\in \mathbb{C}_{\geq 1}[[z]]\) and \(\langle h\rangle\in \mathbb{C}_{\geq 1}[[1/z]]\). We describe some basic results. He proves that each coefficient of the matrix exponential \(\exp \langle h\rangle\) converges (Corollary~9). Theorem~12 shows that \(\exp \mathfrak{m}(0)\subset \mathcal{M}(0)\). Similarly, \(\exp \mathfrak{m}(\infty)\subset \mathcal{M}(\infty)\). Theorem~13 states that, for \(h\in \mathbb{C}_{\geq1}[[z]]\) and \(g\in \mathbb{C}_p[[z]]\), we have \([\exp(h\partial/\partial z)g]=[g] \langle h \rangle\). Corollary~19 shows that, for \(h\in \mathbb{C}_{\geq 1} [[z]]\) and \(m\geq 1\), the entry \([\exp \langle h\rangle]_n^m\) is an entire function of \(h_1,\ldots,h_{n-m+1}\). Theorem~20 says that the \(\exp\)-mapping is surjective, and an 'inverse' functions \(\log\) can be defined. The final part of the paper is devoted to applications. It is proved that the coefficients of solutions of the non-formal Loewner PDE and ODE, with constant infinitesimal generator \(p\) and initial condition \(f_0\), are given by \([f_0]\exp t\langle zp\rangle\) and \(\exp t\langle zp\rangle [f_0]\), respectively. This makes it possible to prove that, for the Loewner PDE and ODE, the power series of their solutions can be approximated by a matrix polynomial (Theorem~23). At last, a theorem on local solvability of the Loewner ODE is proved (Theorem~25).