---
title: Finite-Dimensional Cokernel (FC) in PDEs
url: https://www.emergentmind.com/topics/finite-dimensional-cokernel-condition-fc
type: topic
---

# Finite-Dimensional Cokernel (FC) in PDEs

The finite-dimensional cokernel condition, abbreviated FC, is an algebraic condition on the principal symbol of a linear differential operator that was introduced to control the formal cokernel and to enable the construction of integral solution operators with prescribed kernel support properties. In the framework of underdetermined and overdetermined PDEs, FC requires that the principal symbol have full rank for every non-zero complex covector, not merely for real covectors as in classical ellipticity. For operators with constant-coefficient principal part, FC is equivalent to recovery on curves (RC) and to finite-dimensionality of the formal cokernel without boundary conditions; this equivalence underlies a unified treatment of right-inverses and left-inverses up to finite rank, curve-supported Green kernels, and Poincaré-, Friedrich-, and Korn-type inequalities [2509.04617].

## 1. Definition and formal setting

Let $P$ be a linear differential operator of order $m$ acting between sections of vector bundles over an open set $U \subset \mathbb{R}^n$, or in coordinates between spaces $E \to F$ of fixed finite dimensions $r_0, s_0$. Writing $u=(u^K)\in C^\infty(U;\mathbb{C}^{s_0})$, the operator has the form
$$
(Pu)^J(x)=\sum_{K=1}^{s_0}\sum_{|\alpha|\le m_K} c[P]^{(\alpha,J)}_K(x)\,\partial_x^\alpha u^K,
$$
where the component orders $m_K$ allow matrix operators with differing orders. Its principal symbol is
$$
p^J_K(x,\xi)=\sum_{|\alpha|=m_K} c[P]^{(\alpha,J)}_K(x)\,(i\xi)^\alpha,
$$
for $x\in U$ and $\xi\in \mathbb{C}^n$ [2509.04617].

The adjoint $P^*$ is taken with respect to the $L^2$ inner product on $U$, and its principal symbol satisfies $p^*(x,\xi)=p(x,\xi)^\dagger$ at the principal level. In this terminology, $P$ is underdetermined if $r_0\le s_0$ and $p(x,\xi)$ is surjective of full rank $r_0$, while $P^*$ is overdetermined if $r_0\ge s_0$ and $p^*(x,\xi)$ is injective of full rank $s_0$ [2509.04617].

The formal cokernel of $P$ over $U$, without boundary conditions, is
$$
\ker P^*:=\{\phi\in C^\infty(U;\mathbb{C}^{r_0}) : P^*\phi=0 \text{ in }\mathcal{D}'(U)\}.
$$
The phrase “finite-dimensional” means $\dim \ker P^*<\infty$. FC is then defined symbolically: for all $x\in U$ and all non-zero complex covectors $\xi\in \mathbb{C}^n\setminus\{0\}$, the principal symbol has full rank,
$$
\operatorname{rank} p(x,\xi)=\min\{\dim E_x,\dim F_x\},
$$
or equivalently
$$
\operatorname{rank} p^*(x,\xi)=\min\{\dim F_x,\dim E_x\}.
$$
For a matrix-valued operator this means $\operatorname{rank} p(x,\xi)=r_0$ for all $\xi\ne 0$ when $r_0\le s_0$, or $\operatorname{rank} p^*(x,\xi)=s_0$ for all $\xi\ne 0$ when $r_0\ge s_0$ [2509.04617].

A central point is that FC strengthens standard ellipticity. Classical ellipticity typically asks for full rank or invertibility only for real covectors $\xi\in \mathbb{R}^n\setminus\{0\}$, whereas FC imposes the condition for complex covectors. In the terminology of the paper, this stronger requirement eliminates complex characteristic directions and suppresses oscillatory obstructions encoded by complex zeros of the principal symbol; for constant-coefficient principal symbols, that is precisely what ensures that the formal cokernel is finite-dimensional [2509.04617].

## 2. Equivalence with recovery on curves and finite-dimensional formal cokernel

The paper’s main structural result identifies the exact scope of FC when the principal part has constant coefficients. If $P_{\mathrm{prin}}$ denotes the principal part with constant coefficients, then the following are equivalent: FC, recovery on curves (RC) for any admissible family of curves, and finite-dimensionality of the formal cokernel of $P_{\mathrm{prin}}^*$ on $U$ [2509.04617].

More precisely, Theorem 1.7 states that if $U$ is connected, $P$ has smooth coefficients, and $p^*(x,\xi)$ satisfies FC, then there exists a maximal graded augmented system of variables $\Phi$, indexed by all partial derivatives of the data up to a fixed order $N_0-1$, together with a first-order PDE system whose ODE reduction along curves yields RC. When $p^*$ is independent of $x$, the equivalence
$$
\mathrm{FC}\iff \mathrm{RC}\iff \dim \ker P_{\mathrm{prin}}^*(U)<\infty
$$
holds [2509.04617].

The proof mechanism is algebraic as well as analytic. For constant coefficients, Hilbert’s Nullstellensatz is used to produce polynomial identities of the form
$$
(i\xi)^\alpha I = G_\alpha(\xi)\,p^*(\xi),
$$
which generate an augmented system with variables $\Phi_{(\alpha,J)}=\partial^\alpha \phi_J$. Along any admissible curve $\gamma$, this augmented system becomes an ODE whose fundamental matrix is explicitly controlled. Duhamel’s principle then recovers $\phi(y)$ from the jet of $P^*\phi$ along the curve and endpoint values. Conversely, if $p^*$ fails to have full rank at some complex $\xi\ne 0$, one can construct infinite families of solutions to $P^*\phi=0$, for instance plane waves, and the formal cokernel is then infinite-dimensional [2509.04617].

This equivalence is significant because it translates a symbolic rank condition into both a geometric recovery principle and a functional-analytic statement about cokernel size. A plausible implication is that FC serves as the decisive bridge between local symbol algebra and global integral representation formulas in the class of operators considered in the paper.

## 3. Recovery on curves, curve-supported kernels, and inverse operators

Recovery on curves is the starting point for the integral constructions. For any admissible curve $\gamma(s)=(y,y_1,s)$, $s\in[0,1]$, RC asserts the existence of linear functionals such that for all $\phi\in C_c^\infty(U)$,
$$
\phi(y)=\int_0^1 \sum_{|\alpha|\le m_K'} S_J^{(\alpha,K)}(y,y_1,s)\,\partial_x^\alpha[(P^*\phi)_K](\gamma(s))\,ds+\langle b_{y_1}(\cdot,y),\phi\rangle,
$$
where $b_{y_1}$ is a point-supported distribution at $y_1$ and the coefficients $S_J^{(\alpha,K)}$ are smooth with appropriate graded bounds. Thus $\phi(y)$ is recovered from the jet of $P^*\phi$ along the curve together with an endpoint contribution [2509.04617].

This recovery formula yields Green kernels supported on prescribed curves. For each $y_1$, one defines a distributional kernel $K_{y_1}(x,y)$ supported on the image of the curve and satisfying
$$
P_x K_{y_1}(x,y)=\delta_y(x)-b_{y_1}(x,y)
$$
in $U$, with
$$
\operatorname{supp} K_{y_1}(\cdot,y)\subset \gamma(y,y_1,[0,1]).
$$
Averaging over $y_1$ against a smooth weight $\eta(y,y_1)$ of unit mass produces a smoother kernel
$$
K_\eta(x,y)=\int K_{y_1}(x,y)\,\eta(y,y_1)\,dy_1,
$$
whose support is constrained by the support of $\eta$:
$$
\operatorname{supp} K_\eta(\cdot,y)\subset \bigcup_{y_1\in \operatorname{supp}\eta(y,\cdot)} \gamma(y,y_1,[0,1]).
$$
This is the support-prescription mechanism [2509.04617].

The resulting integral operator
$$
Sf(x)=\int_U K_\eta(x,y)\,f(y)\,dy
$$
satisfies
$$
PSf=f-\int_U b_\eta(x,y)f(y)\,dy.
$$
Under nontrapping and admissibility hypotheses, it has optimal regularization:
$$
S:H^s(U)\to H^{s+m}(U), \qquad S:W^{s,p}(U)\to W^{s+m,p}(U),
$$
and in bounded star-shaped domains one has
$$
PS=\operatorname{Id}-\sum_{j=1}^N \langle f,\psi_j\rangle \phi_j.
$$
The paper also states that the resulting $S$ is a pseudodifferential operator of order $-m$ [2509.04617].

By duality, the same construction gives left-inverses up to finite rank for overdetermined operators:
$$
S^*P^*\varphi=\varphi-\sum_{j=1}^N \langle \varphi,\varphi_j\rangle \eta_j.
$$
This yields inequalities of Friedrich, Poincaré, and Korn type. A typical estimate is
$$
\|\varphi\|_{W^{-s,p'}(U)}
\le C\sum_K \|(P^*\varphi)_K\|_{W^{-s-m_K,p'}(U)}+\text{lower-order terms},
$$
and in bounded star-shaped domains, after orthogonality to the cokernel,
$$
\|\varphi\|_{W^{-s,p'}(U)}
\le C\sum_K \|(P^*\varphi)_K\|_{W^{-s-m_K,p'}(U)}.
$$
For bounded Lipschitz star-shaped domains, the paper also records optimal Sobolev-scale estimates such as
$$
\|u\|_{H^s(U)} \le C\Bigl(\|Pu\|_{H^{s-m}(U)}+\sum_{j=1}^N |\langle u,\varphi_j\rangle|\Bigr)
$$
and
$$
\Bigl\|v-\sum_{j=1}^N \langle v,w_j\rangle \psi_j\Bigr\|_{H^{-s}(U)}
\le C\|P^*v\|_{H^{-s-m}(U)}.
$$
These statements are the principal analytic consequences of FC via RC [2509.04617].

## 4. Operators known to satisfy FC

The paper verifies FC by direct symbol-rank arguments for a broad class of operators. The common pattern is injectivity of $p^*(x,\xi)$ over every $\xi\in \mathbb{C}^n\setminus\{0\}$, which is the overdetermined formulation of FC [2509.04617].

For divergence and gradient, with $P=\operatorname{div}:\mathbb{C}^n\to \mathbb{C}$ and $P^*$ the scalar gradient, the principal symbol is
$$
p_j^*(\xi)\phi=-i\xi_j\phi.
$$
If $p^*(\xi)\phi=0$ for all $j$, then $\xi_j\phi=0$ for all $j$; since some $\xi_j\ne 0$, one gets $\phi=0$. FC therefore holds [2509.04617].

For the Hessian and the double divergence, which is the principal part of the linearized scalar curvature operator, one has
$$
(P^*\phi)_{jk}=\partial_j\partial_k\phi,\qquad
p_{jk}^*(\xi)\phi=-\xi_j\xi_k\phi.
$$
If all components vanish, choosing an index with $\xi_j\ne 0$ gives $\xi_j^2\phi=0$, hence $\phi=0$. FC follows [2509.04617].

For the trace-free Hessian,
$$
(P^*\phi)_{jk}=\partial_j\partial_k\phi-\frac{1}{d}g_{jk}\Delta \phi,
$$
with principal symbol
$$
p_{jk}^*(x,\xi)\phi=-\xi_j\xi_k\phi+\frac{1}{d}g_{jk}(x)g^{\ell m}(x)\xi_\ell\xi_m\phi.
$$
The paper’s argument contracts with $\xi_j$ and shows that for $d>1$ one gets $\sum_\ell \xi_\ell^2\phi=0$, then $\xi_j^2\phi=0$, so $\phi=0$. FC holds for $d\ge 2$ [2509.04617].

For the Killing operator, the adjoint of symmetric divergence,
$$
(P^*\omega)_{jk}=-\frac12(\partial_j\omega_k+\partial_k\omega_j),
$$
and
$$
p_{jk}^{*\ell}(\xi)\omega_\ell
=-\frac{i}{2}(\xi_j\delta_k^\ell+\xi_k\delta_j^\ell)\omega_\ell.
$$
From $p^*(\xi)\omega=0$ one gets $\xi_j\omega_k=-\xi_k\omega_j$ for all $j,k$, hence $\xi_j\omega_j=0$ and then $\xi_j^2\omega_k=0$; choosing $\xi_j\ne 0$ forces $\omega=0$. FC holds for all $d\ge 1$ [2509.04617].

For the conformal Killing operator,
$$
(P^*\omega)_{jk}
=-\frac12(\partial_j\omega_k+\partial_k\omega_j)
+\frac{1}{d}g_{jk}g^{\ell m}\partial_\ell\omega_m,
$$
with
$$
p_{jk}^{*\ell}(x,\xi)\omega_\ell
=-\frac{i}{2}(\xi_j\delta_k^\ell+\xi_k\delta_j^\ell)\omega_\ell
+\frac{i}{d}g_{jk}(x)g^{\ell m}(x)\xi_m\omega_\ell.
$$
Setting $w:=(1/d)\sum_\ell \xi_\ell\omega_\ell$, the paper deduces first a relation $\xi_j\omega_k=-\xi_k\omega_j+2wg_{jk}$, then $\sum_\ell \xi_\ell^2w=0$, then $w=0$ for $d\ge 3$, and finally $\omega=0$. FC therefore holds for $d\ge 3$ [2509.04617].

The linearized Einstein vacuum constraint operator is handled by a block decomposition. Its principal symbol is block diagonal, with a Hessian block for the scalar constraint and a Killing block for the momentum constraint; since each block satisfies FC, the full operator does as well. In constant mean curvature gauge, the momentum block is conformal Killing, which requires $d\ge 3$ [2509.04617].

These verifications establish FC for divergence, gradient, Hessian, trace-free Hessian, Killing, conformal Killing, and linearized Einstein constraint operators. This list is central because it shows that FC is not confined to a single model problem but applies to several operators already associated with Bogovskii-, Reshetnyak-, and geometric-constraint constructions [2509.04617].

## 5. Analytic and geometric consequences

FC has two immediate consequences in the paper’s framework: it implies RC, and RC yields curve-supported Green kernels. Through smooth averaging, one then obtains integral operators with prescribed support and optimal regularization. This support control is not incidental; it is one of the main motivations for imposing FC rather than a weaker real-symbol condition [2509.04617].

The support prescription is flexible. By choosing the averaging weight $\eta$ to be supported outside $U$, or within cones or star-shaped sets, one prescribes the support of $K_\eta(\cdot,y)$ and hence of $Sf$. The paper states explicitly that this generalizes Bogovskii’s construction based on line segments to a star center and the conic operators built from rays over directions on the sphere [2509.04617].

The method is also presented as a unification of several earlier constructions. For divergence, straight-segment curve families recover the classical Bogovskii operator. For divergence with rays in prescribed directions, conic-type averaging reproduces the Oh–Tataru construction. For overdetermined operators such as Killing and conformal Killing, the framework recovers and extends Reshetnyak’s integral representation formulas with kernels supported on line segments. The paper further states that the new augmented-system and RC formalism extends these constructions to variable-coefficient backgrounds and derives Poincaré- and Korn-type inequalities in a unified way [2509.04617].

The geometric scope includes Riemannian manifolds, where the adjoint is taken with respect to the Riemannian volume form $dV$. The examples include linearized scalar curvature and Einstein constraints on constant-curvature backgrounds, for which the paper states that the augmented systems are completely integrable and that explicit kernels can be computed [2509.04617].

At the level of functional analysis, the method is notable for avoiding Fredholm theory while treating non-Fredholm operators. In the underdetermined case, it yields right-inverses up to finite rank; in the overdetermined case, left-inverses up to finite rank and coercive inequalities arise dually. Boundary conditions enter through the function-space choice and through geometric hypotheses such as star-shapedness or nontrapping, rather than through an abstract Fredholm boundary-value framework [2509.04617].

## 6. Scope, limitations, and terminological overlap

The paper distinguishes sharply between constant- and variable-coefficient settings. For constant coefficients, one has the three-way equivalence
$$
\mathrm{FC}\iff \mathrm{RC}\iff \dim \ker P_{\mathrm{prin}}^*(U)<\infty.
$$
For variable coefficients, FC still implies RC, and it implies finite-dimensional cokernel in Sobolev dual scales $H^{-s}(U)$ together with invariance of dimension under restriction and orthogonality conditions, but the full equivalence is only asserted for constant-coefficient principal symbols [2509.04617].

The construction also requires admissible curve families and coefficient bounds sufficient to control the ODE fundamental matrix in the graded augmented system. The paper lists star-shaped geometry for Bogovskii-type operators, nontrapping assumptions for conic-type operators, and boundedness of coefficients as part of the analytic framework. These are limitations of the integral construction, not of the symbolic definition of FC itself [2509.04617].

A separate source of ambiguity is terminological. The acronym “FC” is not uniform across the arXiv literature. In commutative algebra, the paper "“(FC)-Sequences, Mixed Multiplicities and Reductions of Modules”" uses “(FC)-element” and “(FC)-sequence” as labels for elements satisfying the formal conditions (FC1), (FC2), and (FC3); the authors explicitly do not expand the acronym in the text, and it is unrelated to finite-dimensional formal cokernels of differential operators [1109.5058]. In operator algebras, the paper "“C*-algebras of endomorphisms of groups with finite cokernel and partial actions”" uses the “finite cokernel” condition for an injective group endomorphism $p:G\to G$, namely $[G:p(G)]<\infty$ [1707.03056]. In random-matrix theory over $\mathbb{Z}_p$, the paper "“Generalizations of results of Friedman and Washington on cokernels of random $p$-adic matrices”" uses FC to denote finiteness of $\operatorname{cok}(P(X))$, equivalent there to $\det(P(X))\ne 0$ and hence holding almost surely for Haar-random $X$ under the paper’s assumptions [2201.08777].

Within PDE theory, however, the finite-dimensional cokernel condition has a specific meaning: full-rank principal symbol on all non-zero complex covectors, with constant-coefficient equivalence to recovery on curves and to finite-dimensionality of the formal cokernel. In that sense, FC is a symbolic criterion with geometric, analytic, and integral-representation consequences rather than a generic label for cokernel finiteness in unrelated settings [2509.04617].

Source: https://www.emergentmind.com/topics/finite-dimensional-cokernel-condition-fc