A multivariate Lagrange inversion formula for asymptotic calculations (Q1389777)

Bold face letters denote vectors and a vector to a vector power is the product of component-wise exponentiation: if \({\mathbf x}= (x_1,\dots, x_d)\) and \({\mathbf n}= (n_1,\dots, n_d)\), then \({\mathbf x}^{\mathbf n}= x^{n_1}_1 x^{n_2}_2\cdots x^{n_d}_d\). Finally, if \(h({\mathbf x})\) is a formal power series in the vector variable \({\mathbf x}\), \([{\mathbf x}^{\mathbf n}]h({\mathbf x})\) denotes the coefficient of \({\mathbf x}^{\mathbf n}\) in \(h({\mathbf x})\). This paper is devoted to a proof of the following theorem. Theorem 2. Suppose that \(g({\mathbf x}),f_1({\mathbf x}),\dots, f_d({\mathbf x})\) are formal power series in \({\mathbf x}\) such that \(f_i(\text{\textbf{0}})\neq 0\) for \(1\leq i\leq d\). Then the set of equations \(w_i= t_if_i({\mathbf w})\) for \(1\leq i\leq d\) uniquely determine the \(w_i\) as formal power series in \({\mathbf t}\) and \[ [{\mathbf t}^{\mathbf n}]g({\mathbf w}({\mathbf t}))= {1\over \prod n_i} [{\mathbf x}^{{\mathbf n}-\mathbf{1}}] \sum_\tau {\partial(g, f^{n_1}_1,\dots, f^{n_d}_d)\over \partial\tau}, \] where \(\text{\textbf{1}}= (1,\dots, 1)\), the sum is over all trees \(\tau\) with \(V= \{0,1,\dots, d\}\) and edges directed toward \(0\), and the vector in \(\partial/\partial\tau\) is indexed from \(0\) to \(d\).