The Jacobian problem for one class of nonpolynomial mappings (Q2123089)

Much work has been devoted to the Jacobian conjecture, which asks if every polynomial map \(f: \mathbb C^n \to \mathbb C^n\) with non-vanishing Jacobian determinant \(J_f\) is bijective with polynomial inverse. More generally one can ask under what conditions a local diffeomorphism \(F: \mathbb R^n \to \mathbb R^n\) is globally injective (for example [\textit{S. Nollet} and \textit{F. Xavier}, Discrete Contin. Dyn. Syst. 8, No. 1, 17--28 (2002; Zbl 1008.58009); Geom. Funct. Anal. 14, No. 6, 1339--1351 (2004; Zbl 1073.32005); Arch. Math. 89, No. 5, 385--389 (2007; Zbl 1135.14049)] or [\textit{S. Nollet} et al., J. Reine Angew. Math. 627, 83--95 (2009; Zbl 1161.14042)]). After discussing the Jacobian conjecture, the author gives examples of injective smooth functions \(f: D \to \mathbb R^n\), where \(D \subset \mathbb R^n\) is a convex domain. The most general is the following. Let \(0 \neq B \in \mathbb R^n\) be vector and \(\Lambda \subset \mathbb R\) the interval \(\{X \cdot B: X \in D\}\). Let \(g: \Lambda \to \mathbb R^n\) be a smooth function for which \(\displaystyle \frac{g(u)-g(v)}{u-v} \cdot B \neq -1\) for all \(u,v \in \Lambda\). Then the function \(\Phi (X) = X + g(X \cdot B)\) is injective on \(D\) and \(J_{\Phi} \neq 0\) on \(D\). Using the Lagrange formula, the hypothesis \(\displaystyle \frac{g(u)-g(v)}{u-v} \cdot B \neq -1\) for all \(u,v \in \Lambda\) may be replaced with \(g^\prime (w) \cdot B \neq -1\) for \(w \in \Lambda\). The proof reduces the argument to the special case when \(B = (1, \dots, 1)^T\), which is proven using a result of \textit{S. Ponnusamy} and \textit{V. V. Starkov} [Bull. Malays. Math. Sci. Soc. (2) 41, No. 4, 2099--2115 (2018; Zbl 1406.14044)]. Namely, let \(D \subset \mathbb R^n\) be a convex domain and \(F = (f_1, \dots, f_n): D \to \mathbb R^n\) be continuously differentiable. For \(P,Q \in D\), let \(\gamma (t) = P + t (Q-P)\) be the line segment from \(P\) to \(Q\) with \(0 \leq t \leq 1\). If the matrix \(A = (a_{i,j})\) with entries \(a_{i,j} = \int_0^1 \frac{\partial f_i}{\partial x_j} (\gamma (t)) \; dt\) is invertible for all \(P \neq Q \in D\), then \(F\) is injective. While these results are proved for convex domains, the author notes that they are valid for any open domain. There are variations for functions \(F: D \to \mathbb C^n\), where \(D \subset \mathbb C^n\) is a domain.