Canonical: https://knowlpedia.com/differential-geometry/elliptic-differential-operator/

# Elliptic differential operator

A differential operator whose principal symbol is invertible at every nonzero cotangent vector.

Let \(D:\Gamma^\infty(E)\to\Gamma^\infty(F)\) be a differential operator of order \(m\geq0\) between smooth [vector bundles](https://knowlpedia.com/fiber-bundles/vector-bundle/) over \(M\). The operator \(D\) is **elliptic** if, for every \(x\in M\) and every nonzero \(\xi\in T_x^*M\), its [principal symbol](https://knowlpedia.com/differential-geometry/principal-symbol/)
\[
\sigma_m(D)(x,\xi):E_x\longrightarrow F_x
\]
is a [linear isomorphism](https://knowlpedia.com/convex-analysis/image-and-kernel-linear-isomorphism/). Ellipticity is therefore a condition on the highest-order part alone. It implies that \(E\) and \(F\) have equal rank on each [connected component](https://knowlpedia.com/topology/connected-component/), but it does not by itself impose self-adjointness or compactness of the base manifold.

## Analytic consequences

Elliptic regularity says that distributional solutions gain smoothness wherever the right-hand side is smooth. On a [closed manifold](https://knowlpedia.com/topology/closed-manifold/), the Sobolev extension
\[
D:H^{s+m}(E)\longrightarrow H^s(F)
\]
is [Fredholm](https://knowlpedia.com/functional-analysis/fredholm-operator/) for every real \(s\); hence its kernel and cokernel are finite-dimensional.

## Examples and non-examples

The scalar Laplace–Beltrami operator has symbol \(|\xi|^2\) and is elliptic. [Dirac-type operators](https://knowlpedia.com/noncommutative-geometry/dirac-type-operator/) have Clifford-linear symbol and are first-order elliptic operators. By contrast, \(\partial/\partial x_1\) on \(\mathbb R^n\) with \(n>1\) has symbol \(\xi_1\), which vanishes for many nonzero covectors, so it is not elliptic.

## Conventions and scope

**Warning.** Boundary-value problems require an additional elliptic boundary condition; ellipticity of the interior operator alone does not make such a problem Fredholm. On a noncompact manifold, ellipticity alone likewise need not give a Fredholm operator or discrete spectrum.

## References

1. R. O. Wells Jr., *Differential Analysis on Complex Manifolds*, 3rd ed., Springer, 2008. [Publisher record](https://doi.org/10.1007/978-0-387-73892-5). Relevant: chapter IV, elliptic operator theory.
2. H. B. Lawson Jr. and M.-L. Michelsohn, *Spin Geometry*, Princeton University Press, 1989. [Publisher record](https://doi.org/10.1515/9781400883912). Relevant: chapter III, ellipticity and analytic properties.
