Carleman estimates and uniqueness of the Cauchy problem for differential operators with nonsingular characteristics: closing the pseudo-convexity gap between necessary and sufficient conditions

In a 1962 research announcement, Lars Hörmander proved weighted a priori estimates of Carleman type for linear differential operators with nonsingular characteristics (operators of principal type), and used them to prove uniqueness in the Cauchy problem for solutions vanishing on one side of a hypersurface, with smooth non-analytic coefficients. He gave a necessary condition on the weight — a non-strict positivity inequality for a quadratic form built from the Hessian of the weight and the derivatives of the principal symbol, evaluated at complex characteristic covectors — and strictly stronger sufficient conditions, since called strong pseudo-convexity, remarking that the two are 'nearly sufficient'. The open problem is to complete that characterization: (i) whether the necessary condition implies the Carleman estimate up to an explicitly describable sharp correction, so that the exact boundary between validity and failure of the estimate is known; (ii) whether pseudo-convexity is necessary for uniqueness at genuinely complex characteristic covectors, for the fixed operator rather than only up to lower-order perturbations; and (iii) the extension of the theory to operators with genuinely complex principal symbol, beyond the principally-normal and condition-(P) classes treated in the standard references. The real-principal-symbol case was substantially settled between 1983 and 1995 up to the necessary/sufficient gap and the necessity question at genuinely complex characteristics; the complex-symbol case remains open, and the 2019 standard reference still formulates related conjectures and open problems.

Background

Let $P(x,D)$ be a linear partial differential operator of order $m$ in an open set $\Omega\subset\mathbb{R}^n$, and let $S=\{x:\varphi(x)=\varphi(x^0)\}$ be a smooth hypersurface through $x^0$ (so $\nabla\varphi(x^0)\neq 0$). The uniqueness of the Cauchy problem (also called unique continuation) asks whether a solution of $P(x,D)u=0$ that vanishes on the side $\{\varphi<\varphi(x^0)\}$ of $S$ must vanish in a full neighborhood of $x^0$. For analytic coefficients this is Holmgren's theorem, a complex-analytic argument; for $C^\infty$ (non-analytic) coefficients it is a central problem of real analysis, tied to existence theory for $Pu=f$, since failure of uniqueness for the adjoint operator obstructs solvability. The classical approach, begun by Carleman in 1939 for first-order systems in two variables and extended by Calderón in 1958, is a weighted a priori estimate (a Carleman estimate) with exponential weight $e^{2\tau\varphi}$ and large parameter $\tau$: the weighted $L^2$ norms of $u$ and its derivatives of order $<m$ are controlled by the weighted $L^2$ norm of $Pu$, plus a weaker lower-order correction; letting the weight sweep through the domain forces $u=0$.

The highest-order part of $P$ is encoded in its principal symbol $p_m(x,\xi)=\sum_{|\alpha|=m}a_\alpha(x)\xi^\alpha$, a homogeneous polynomial in the covector $\xi$; a point $(x,\xi)$ with $\xi\neq 0$ is characteristic when $p_m(x,\xi)=0$. The operator has nonsingular characteristics (is of principal type, in the terminology of Hörmander's 1958-1960 theory of such operators) when $\nabla_\xi p_m$ does not vanish at characteristic points, so the characteristic variety is smooth. Bicharacteristics are the integral curves of the Hamilton vector field of $p_m$; high-frequency wave packets of solutions travel along them, so the geometry of a weight relative to the bicharacteristic flow governs both the estimates and the propagation of singularities.

The source is Hörmander's 1962 announcement 'Differential operators with nonsingular characteristics'. For real principal coefficients it proves: a necessary condition for the Carleman estimate (Theorem 2.1, equation (2.2)); sufficient conditions (Theorems 2.2-2.3, equations (2.5) and (2.8)); uniqueness of the Cauchy problem (Theorem 3.1, condition (3.1)); and unique continuation of singularities (Theorem 5.1). The conditions are positivity statements about one quadratic form, built from the second derivatives of the weight $\varphi$ contracted with the $\xi$-derivatives of $p_m$, together with a Poisson-bracket correction built from first-order derivatives of $p_m$ and $\varphi$ (a Poisson bracket measures the failure of two first-order differential expressions to commute; here it detects rotation of the characteristic directions along the flow). The sufficient conditions require strict positivity at two kinds of points: real characteristic covectors at which the Hamiltonian derivative of $\varphi$ vanishes (bicharacteristics tangent to the level surfaces of $\varphi$), and complex characteristic covectors $\zeta=\xi+i\sigma\nabla\varphi(x)$ with $\sigma\neq 0$ satisfying $p_m(x,\zeta)=0$ — the complex phases that would appear in a Holmgren-type complexification. This pair of requirements is what the modern literature calls strong pseudo-convexity of the weight with respect to the operator. For real principal symbols the source itself gives the geometric reading: the restriction of $\varphi$ to any bicharacteristic curve has positive second derivative wherever its first derivative vanishes — strict convexity of the weight along the bicharacteristic flow at tangency points. The necessary condition (2.2) is a weaker, non-strict inequality for the same form at complex characteristics; the announcement states that (2.2) and its corollary (2.4) are 'nearly sufficient' for the estimate. Theorem 3.1 deduces uniqueness from the estimate under the positivity (3.1) at complex characteristic covectors $\zeta=\xi+i\sigma N^0$ (with $N^0=\nabla\varphi(x^0)$) satisfying $p_m(x^0,\zeta)=0$ and $\sum_i\partial p_m/\partial\zeta_i(x^0,\zeta)N_i^0=0$. The announcement records two boundary facts of its own theory: counterexamples due to P. Cohen (unpublished 1960 Stanford report; documented in the published accounts of Alinhac 1984 and Lerner 2019 cited below) show the convexity assumption is 'vital at least as far as real solutions' of the complex characteristic equations are concerned; and the convexity hypothesis of the unique-continuation-of-singularities theorem is essential, by a wave-equation example with singularities propagating along a line. The whole treatment is limited to real principal coefficients, complex non-elliptic symbols being deferred to an additional hypothesis imported from Hörmander's 1960 paper.

The subsequent literature transformed each thread. The propagation of singularities theorem of Duistermaat-Hörmander (1972) settled the real-principal-symbol content of the singularities thread: wave-front sets of solutions of real principal-type operators propagate along bicharacteristics. On the sufficiency side, Lerner-Robbiano (1984/1985) proved uniqueness for second-order operators of principal type under non-strict convexity along bicharacteristics (the quadratic form merely nonnegative at tangency points), at the price of a compactness hypothesis on the part of the support of the solution meeting the surface; Hörmander's treatise The Analysis of Linear Partial Differential Operators IV (Ch. XXVIII) organized the uniqueness theory — general Carleman estimates, uniqueness under convexity conditions for principally normal operators (operators whose Poisson bracket $\{p_m,\bar p_m\}$ vanishes on the characteristic set, a complex-symbol condition descending from the 1960 hypothesis), and an improved second-order real theory — but its degenerate results still carry compactness-type hypotheses, and complex principal type is handled (Ch. XXVI) only under condition (P), the sign condition on the Poisson bracket of the principal symbol along its Hamilton flow that governs local solvability and propagation for complex principal type. On the necessity side, Alinhac (1983) proved that a pseudo-concave tangent bicharacteristic admits a smooth zeroth-order perturbation losing uniqueness; Bahouri (1983) constructed genuine non-uniqueness for second-order operators with real principal symbol, including configurations where bicharacteristics remain inside the boundary; Alinhac-Baouendi (1995) proved that for the wave equation with respect to a timelike hyperplane uniqueness holds for the operator itself yet is destroyed by a suitable smooth zeroth-order perturbation (absence of stable uniqueness), showing the weak-pseudo-convexity uniqueness theorems cannot be made perturbation-stable in that regime; Métivier (1993) gave counterexamples to Holmgren uniqueness for analytic nonlinear Cauchy problems. Lerner's 2019 monograph Carleman Inequalities synthesizes the field: new Carleman estimates for complex-coefficient principally-normal operators and condition-(P) classes (Ch. 5), and a chapter 'on the edge of pseudo-convexity' (Ch. 6) presenting these counterexamples, proving a 'compact uniqueness' result under a very weak hypothesis, and 'ends with the formulation of various conjectures and open problems'. No standard reference develops a Carleman-estimate or uniqueness theory for genuinely complex principal symbols outside the principally-normal and condition-(P) classes; this, together with the unresolved gap between the necessary condition and strong pseudo-convexity and the open necessity question at genuinely complex characteristics, forms the surviving core of the 1962 gap.

Problem Statement

Objective. Determine the exact conditions under which Carleman estimates hold and the Cauchy problem is unique for linear differential operators with nonsingular characteristics — the characterization problem raised by the gap left open in Hörmander's 1962 announcement 'Differential operators with nonsingular characteristics' (Bull. Amer. Math. Soc. 68 (1962), 354-359) between the necessary condition (2.2) and the sufficient conditions (2.5)/(2.8), and by its restriction to real principal coefficients. The single scientific objective is a complete characterization: for each operator class and weight, which side of the boundary the pair falls on, with theorem or counterexample.

Setting. $P(x,D)$ is a linear differential operator of order $m$ with $C^\infty$ coefficients in an open set $\Omega\subset\mathbb{R}^n$ (the announcement allows bounded measurable lower coefficients and $C^1$ principal coefficients; either convention may be fixed, consistently), with principal symbol $p_m$ of principal type ($\nabla_\xi p_m\neq 0$ at every characteristic point). Let $\varphi$ be a real $C^\infty$ weight with nonvanishing gradient. A Carleman estimate for the pair $(P,\varphi)$ is an inequality, valid for all $u\in C_0^\infty(\Omega)$ and all $\tau$ above a threshold, of the form (1.1) of the primary source: the $e^{2\tau\varphi}$-weighted $L^2$ norms of the derivatives $D^\alpha u$, $|\alpha|<m$, each carrying its prescribed power of the parameter $\tau$, are bounded by a constant times the weighted norm of $P(x,D)u$ plus a constant times a prescribed weaker weighted combination of lower-order norms. The variant without the lower-order term is the one that forces uniqueness of the Cauchy problem; the variant with an arbitrary lower-order constant feeds the existence theory. Let $Q_\varphi$ denote the quadratic form of the announcement — the Hessian of $\varphi$ contracted with the $\xi$-gradients of $p_m$, plus the Poisson-bracket correction displayed in equations (2.5), (2.8), (3.1) of the source. Strong pseudo-convexity means: $Q_\varphi>0$ at every real characteristic covector where the Hamiltonian derivative of $\varphi$ vanishes, and at every complex characteristic covector

$$p_m(x,\zeta)=0,\qquad \zeta=\xi+i\sigma\nabla\varphi(x),\quad \xi\in\mathbb{R}^n,\ \sigma\neq 0;$$

for real principal symbols this is strict convexity of $\varphi$ along bicharacteristics at tangency points, together with the analogous positivity at complex characteristics. The necessary condition (2.2) is the corresponding non-strict inequality.

Required determinations (faces of the single characterization gap left open in the 1962 announcement):

1. Close the necessary/sufficient gap for the estimate: either prove that the necessary condition (2.2) implies the Carleman estimate up to an explicitly describable correction — identifying exactly the geometric configurations where the correction is needed — or exhibit the sharp boundary by giving, for each configuration left between (2.2) and strong pseudo-convexity, either a proof that the estimate holds or an operator-and-weight pair for which it fails.

2. Decide the necessity of the complex-characteristic convexity (condition (3.1)) for uniqueness of the Cauchy problem for the fixed operator — the announcement's counterexamples settle only the real-characteristic case: either construct an operator with real principal symbol and a weight violating (3.1) at a genuinely complex characteristic covector for which uniqueness fails, or prove that uniqueness survives such violations, so that only lower-order-perturbation instability (as in the known zeroth-order counterexamples) remains possible; and delimit precisely when uniqueness holds but is unstable under smooth lower-order perturbations.

3. Extend the Carleman estimate and the resulting uniqueness theory to operators with genuinely complex principal symbol — neither principally normal nor satisfying condition (P) — or prove that no Carleman-estimate/uniqueness theory of pseudo-convexity type can exist for operators with genuinely complex principal symbol outside the principally-normal and condition-(P) classes.

The characterization is regime-dependent: it is relative to a specified operator class (coefficient regularity, presence or absence of principal normality and of condition (P)), a specified class of weights, and a specified solution class (distributions, $C^\infty$, or Gevrey classes — the scales between analytic and $C^\infty$ parametrized by an order $s\ge 1$); the known counterexamples show the answers depend on these choices.

References

1. Lars Hörmander, Differential operators with nonsingular characteristics, Bulletin of the American Mathematical Society 68 (1962), no. 4, 354-359. DOI: 10.1090/S0002-9904-1962-10807-6. URL: https://doi.org/10.1090/S0002-9904-1962-10807-6.
2. T. Carleman, Sur un problème d'unicité pour les systèmes d'équations aux dérivées partielles à deux variables indépendantes, Arkiv för Matematik, Astronomi och Fysik 26B (1939), no. 17, 1-9. zbMATH identifier: 0022.34201. URL: https://zbmath.org/3036525.
3. A. P. Calderón, Uniqueness in the Cauchy problem for partial differential equations, American Journal of Mathematics 80 (1958), 16-36. DOI: 10.2307/2372819. URL: https://doi.org/10.2307/2372819.
4. Lars Hörmander, On the uniqueness of the Cauchy problem. I, Mathematica Scandinavica 6 (1958), 213-225. DOI: 10.7146/math.scand.a-10546. URL: https://doi.org/10.7146/math.scand.a-10546.
5. Lars Hörmander, Differential operators of principal type, Mathematische Annalen 140 (1960), 124-146. DOI: 10.1007/BF01360085. URL: https://doi.org/10.1007/BF01360085.
6. Fritz John, Continuous dependence on data for solutions of partial differential equations with a prescribed bound, Communications on Pure and Applied Mathematics 13 (1960), 551-585. DOI: 10.1002/cpa.3160130402. URL: https://doi.org/10.1002/cpa.3160130402.
7. Bernard Malgrange, Existence et approximation des solutions des équations aux dérivées partielles et des équations de convolution, Annales de l'Institut Fourier 6 (1955-1956), 271-355. DOI: 10.5802/aif.65. URL: https://doi.org/10.5802/aif.65.
8. Lars Hörmander, Linear Partial Differential Operators, Grundlehren der mathematischen Wissenschaften 116, Springer, Berlin, 1963. DOI: 10.1007/978-3-642-46175-0. URL: https://doi.org/10.1007/978-3-642-46175-0.
9. J. J. Duistermaat and L. Hörmander, Fourier integral operators. II, Acta Mathematica 128 (1972), 183-269. DOI: 10.1007/BF02392165. URL: https://doi.org/10.1007/BF02392165.
10. Lars Hörmander, Non-uniqueness for the Cauchy problem, in: Fourier Integral Operators and Partial Differential Equations, Lecture Notes in Mathematics 459, Springer, Berlin, 1975, pp. 36-72. DOI: 10.1007/BFb0074192. URL: https://doi.org/10.1007/BFb0074192.
11. Lars Hörmander, The Analysis of Linear Partial Differential Operators IV, Classics in Mathematics, Springer, 2009. DOI: 10.1007/978-3-642-00136-9. URL: https://doi.org/10.1007/978-3-642-00136-9.
12. Nicolas Lerner and Luc Robbiano, Unicité de Cauchy pour des opérateurs de type principal, Journal d'Analyse Mathématique 44 (1984–1985), 32-66. DOI: 10.1007/BF02790189. URL: https://doi.org/10.1007/BF02790189.
13. S. Alinhac, Non-unicité du problème de Cauchy, Annals of Mathematics 117 (1983), 77-108. DOI: 10.2307/2006972. URL: https://doi.org/10.2307/2006972.
14. Hajer Bahouri, Non unicité du problème de Cauchy pour des opérateurs à symbole principal réel, Communications in Partial Differential Equations 8 (1983), 1521-1547. DOI: 10.1080/03605308308820314. URL: https://doi.org/10.1080/03605308308820314.
15. S. Alinhac, Uniqueness and nonuniqueness in the Cauchy problem, Contemporary Mathematics 27 (1984), 1-22. DOI: 10.1090/conm/027/741036. URL: https://doi.org/10.1090/conm/027/741036.
16. S. Alinhac and M. S. Baouendi, A non uniqueness result for operators of principal type, Mathematische Zeitschrift 220 (1995), 561-568. DOI: 10.1007/BF02572631. URL: https://doi.org/10.1007/BF02572631.
17. G. Métivier, Counterexamples to Holmgren's uniqueness for analytic non linear Cauchy problems, Inventiones Mathematicae 112 (1993), 217-222. DOI: 10.1007/BF01232431. URL: https://doi.org/10.1007/BF01232431.
18. Nicolas Lerner, Carleman Inequalities: An Introduction and More, Grundlehren der mathematischen Wissenschaften 353, Springer, 2019. DOI: 10.1007/978-3-030-15993-1. URL: https://doi.org/10.1007/978-3-030-15993-1.