Perturbation of differential operators admitting a continuous linear right inverse on ultradistributions. (Q1880093)

Let \(P(D)\) be a linear partial differential operator with constant coefficients with \(n\) variables, \(n>2\), and let \(P_m(D)\) be its principal part. \textit{R. Meise, B. A. Taylor} and \textit{D. Vogt} [Ann. Inst. Fourier 40, 619--655 (1990; Zbl 0703.46025)], solved a problem posed by L. Schwartz and determined when \(P(D)\) admits a continuous linear right inverse on the space \({\mathcal E}(G)\), resp. \({\mathcal D}'(G)\), of all \(C^{\infty}\)-functions, resp. distributions, on an open subset \(G\) of \(\mathbb{R}^n\). The case of \(\omega\)-ultradifferentiable functions \({\mathcal E}_{\omega}(G)\) and \(\omega\)-ultradistributions \({\mathcal D}'_{\omega}(G)\) of Beurling type was treated by the same authors in [Math. Nachr. 180, 213--242 (1996; Zbl 0858.46030)]. The characterizations are rather involved. In the case \(G=\mathbb{R}^n\), the existence of a continuous linear right inverse for \(P(D)\) is equivalent to a Phragmén-Lindelöf condition \(PL(\mathbb{R}^n,\omega)\) on the zero variety \(V(P)\) of the polynomial \(P(-z)\). \textit{R. Meise, B. A. Taylor} and \textit{D. Vogt} [J. Am. Math. Soc. 11, 1--39 (1998; Zbl 0896.32008)] showed that if \(P(D)\) admits a continuous linear right inverse on \({\mathcal E}_{\omega}(\mathbb{R}^n)\), so does \(P_m(D)\). Continuing work of the present authors in [Pac. J. Math. 192, 201--218 (2000; Zbl 1071.47512)] about perturbation results, the authors explain in terms of \(P_m\) and the degree \(l<m\) of a polynomial \(Q\) the optimal choice of \(\beta:=\beta(l,P_m)\) so that \((P_m+Q)(D)\) has a continuous linear right inverse on the space \({\mathcal E}_{\omega_{\beta}}(\mathbb{R}^n)\), \(\omega_{\beta}(t):= (1+t)^{\beta}\), \(0<\beta <1\), of Gevrey functions, for all polynomials \(Q\) of degree \(l\). A very neat theorem is obtained for \(n=3\). The proof requires new results about the Phragmén-Lindelöf condition \(PL(\mathbb{R}^n,\omega)\) presented in the paper. Several concrete examples are checked. In particular, they show that, for a fixed polynomial \(Q\), the operator \(P(D)=(P_m+Q)(D)\) may behave better than the behaviour predicted by the main result in which the perturbation is required for all polynomials \(Q\) of a certain degree.