---
title: Generalized Sectorial Decomposition
url: https://www.emergentmind.com/topics/generalized-sectorial-decomposition
type: topic
---

# Generalized Sectorial Decomposition

Generalized sectorial decomposition denotes a family of constructions in which an operator, matrix, flow, symplectic manifold, or integral is partitioned by rays, sectors, or sectorial hypersurfaces so that spectral, asymptotic, or microlocal structure becomes explicit. In operator theory this language is tied to sectorial projections and logarithmic functional calculus; in matrix analysis it is tied to congruence decompositions and phase data; in harmonic analysis it is tied to Paley–Littlewood and Triebel–Lizorkin decompositions attached to sectorial operators; and in geometric and dynamical settings it appears in Liouville sectorial coverings and in higher-dimensional analogues of Poincaré’s planar sectorial decomposition [1102.4051][2205.07607][1407.0821][2511.16584][2306.15822].

## 1. Elliptic operators, logarithms, and sectorial projections

For a classical elliptic pseudodifferential operator \(P\) of order \(m>0\) on a closed manifold, sectorial decomposition begins with two rays \(e^{i\theta}\mathbb R_+\) and \(e^{i\varphi}\mathbb R_+\) that separate the eigenvalues of the principal symbol \(p_m(x,\xi)\). Under the hypotheses that these rays are rays of minimal growth and contain no eigenvalues of \(P\), the sectorial projection is defined by the contour integral
\[
\Pi_{\theta,\varphi}(P)
= \frac{i}{2\pi}\int_{\Gamma_{\theta,\varphi}} \lambda^{-1}P(P-\lambda)^{-1}\,d\lambda.
\]
The decisive structural identity is
\[
\Pi_{\theta,\varphi}(P)=\frac{i}{2\pi}\bigl(\log_\theta P-\log_\varphi P\bigr),
\]
which converts the projection problem into a comparison of two logarithms with different branch cuts [1102.4051].

This formula resolves a flaw in earlier direct proofs. Booss–Bavnbek, Chen, Lesch, and Zhu had observed that one cannot in general modify the principal symbol \(p_m(x,\xi)\) so that \((p_m(x,\xi)-\lambda)^{-1}\) is uniformly well defined along both rays for all \(\xi\). The logarithmic approach avoids that obstruction because in the construction of \(\log_\theta P\) one only modifies \(p_m(x,\xi)\) near a single ray, as in Seeley’s theory of complex powers. The resulting symbol has the form
\[
\sigma(\log_\theta P)(x,\xi)=m\log[\xi]+l_\theta(x,\xi),
\]
so the logarithmic terms cancel in the difference, and \(\Pi_{\theta,\varphi}(P)\) becomes a classical pseudodifferential operator of order \(0\) [1102.4051].

At symbol level, the principal symbol of the projection is the matrix Riesz projection of \(p_m(x,\xi)\) onto the eigenvalues in the open sector \(\Lambda_{\theta,\varphi}\):
\[
\pi_{\theta,\varphi,0}(x,\xi)
= \frac{i}{2\pi}\int_{\mathcal C_{\theta,\varphi}(x,\xi)}(p_m(x,\xi)-\lambda)^{-1}\,d\lambda.
\]
This gives the decomposition a microlocal meaning: the operator splits phase space according to sectorially separated branches of the principal symbol. The same framework formally extends to several rays \(\theta_0<\theta_1<\cdots<\theta_k<\theta_0+2\pi\), yielding projections associated with adjacent sectors and a decomposition of the identity into sectorial pieces. This suggests a generalized sectorial decomposition in the non-selfadjoint elliptic calculus, with logarithms replacing direct resolvent integration as the robust construction.

## 2. Matrix decompositions, phases, and semi-sectorial normal forms

In matrix analysis, sectorial decomposition is first a congruence normal form. If \(C\) is sectorial, there exist a nonsingular matrix \(T\) and a diagonal unitary matrix \(D\) such that
\[
C=T^*DT.
\]
The diagonal entries of \(D\) determine the phases \(\phi_1(C)\ge\cdots\ge\phi_n(C)\), which coincide with the canonical angles of \(C\). The unitary factor is unique up to permutation, and two canonical refinements of this decomposition are the symmetric polar decomposition
\[
C=PUP
\]
with \(P>0\) and \(U\) unitary, and the generalized Cholesky factorization
\[
C=R^*WR
\]
with \(R\) upper triangular with positive diagonal and \(W\) unitary; both are unique [1904.07211].

The semi-sectorial theory extends this picture to singular matrices and to numerical ranges that may touch the origin. For a quasi-sectorial matrix \(C\) of rank \(r\),
\[
C=U\begin{bmatrix}0&0\\0&C_s\end{bmatrix}U^*,
\]
where \(C_s\) is sectorial. In the generic semi-sectorial case with nonempty interior, one has the generalized sectorial decomposition
\[
C=T^*
\begin{bmatrix}
0_{n-r}&0&0\\
0&D&0\\
0&0&E
\end{bmatrix}
T,
\]
where \(D=\operatorname{diag}(e^{j\theta_1},\dots,e^{j\theta_m})\) and \(E\) is a block diagonal matrix built from rotated Jordan blocks
\[
e^{j\theta_0}\begin{bmatrix}1&2\\0&1\end{bmatrix}.
\]
The phases of \(C\) are the angles from the \(D\)-block together with copies of the extremal boundary angles \(\theta_0\pm\pi/2\). This is the canonical extension of sectorial phase data to semi-sectorial matrices [2205.07607].

This decomposition is stable under several fundamental operations. For a semi-sectorial matrix \(C\) of rank \(r\), the phases of the Moore–Penrose inverse satisfy
\[
\phi_i(C^\dagger)=-\phi_{r-i+1}(C).
\]
Compressions and generalized Schur complements satisfy phase interlacing inequalities, and for quasi-sectorial \(A\) and semi-sectorial \(B\) there exists an isometry \(U\) such that
\[
\angle\lambda_{\neq 0}(AB)\prec \phi(U^*AU)+\phi(U^*BU),
\]
which yields the generalized matrix small phase theorem:
\[
\det(I+AB)\neq 0 \quad \text{for all } B\in\mathcal C[\alpha,\beta]
\]
if and only if
\[
[\alpha,\beta]\subset (-\pi-\underline{\phi}(A),\,\pi-\overline{\phi}(A)) \quad (\mathrm{mod}\ 2\pi).
\]
The theory also introduces essential phases via positive diagonal similarity and proves that for a strongly connected directed graph, the Laplacian becomes quasi-sectorial after the canonical scaling \(D_0=V^{-1/2}\), where \(v>0\) is the positive left eigenvector satisfying \(v^TL=0\) [2205.07607].

A related inequality-oriented viewpoint treats a sectorial matrix as a positive real part plus a bounded-angle phase. Using the decomposition \(A=XZX^*\), one obtains sharp control of singular values, determinants, inverses, and operator means in terms of \(R(A)=(A+A^*)/2\) and the sector angle \(\alpha\) [2001.00687]. Numerical-radius refinements use the same geometry: if \(W(A)\subseteq\{re^{\pm i\theta}:\theta_1\le\theta\le\theta_2\}\) and \(\gamma=\max\{\theta_2,\pi/2-\theta_1\}\), then
\[
w(A)\ge \frac{\csc\gamma}{2}\|A\|+\frac{\csc\gamma}{2}\bigl|\|\Im(A)\|-\|\Re(A)\|\bigr|,
\]
and if \(A,B\) are double commuting sectorial matrices with sectorial index \(\gamma\in[0,\pi/2)\), then
\[
w(AB)\le (1+\sin^2\gamma)w(A)w(B)
\]
[2208.09816]. These estimates quantify how decomposition by sector or phase controls non-normality.

## 3. Functional-calculus decompositions and abstract smoothness scales

For Banach-function-space operators, generalized sectorial decomposition takes the form of scale decompositions generated by holomorphic functional calculus. If \(A\) is \(\mathcal R_s\)-sectorial and \(\varphi\in\Phi_{\sigma,\theta}\), the generalized Triebel–Lizorkin norm is
\[
\|x\|_{\theta,s,A,\varphi}
=
\Big\|
\Big(
\int_0^\infty |t^{-\theta}\varphi(tA)x|^s\,\frac{dt}{t}
\Big)^{1/s}
\Big\|_X,
\]
with discrete variants
\[
\|x\|_{\theta,s,A,\varphi}^{\Sigma}
=
\Big\|
\Big(
\sum_{j\in\mathbb Z}|2^{-j\theta}\varphi(2^jA)x|^s
\Big)^{1/s}
\Big\|_X.
\]
The resulting homogeneous and inhomogeneous spaces \(X^\theta_{s,A}\) are independent of the auxiliary cutoff \(\varphi\), interpolate well, and the part of \(A\) in these spaces has a bounded \(H^\infty\)-functional calculus [1207.4217].

When \(A=(-\Delta)^m\) on \(L^p(\mathbb R^d)\), this abstract construction reproduces the classical spaces:
\[
X^\theta_{s,A}=\dot F^{2m\theta}_{p,s}(\mathbb R^d),
\qquad
X^\theta_{s,A}=F^{2m\theta}_{p,s}(\mathbb R^d)
\ \text{for } \theta>0.
\]
In this sense, sectorial calculus replaces Fourier decomposition by a holomorphic spectral decomposition adapted to \(A\) [1207.4217].

A complementary dyadic version is the Paley–Littlewood decomposition for \(0\)-sectorial operators with bounded \(M^a\)-calculus. For a homogeneous dyadic partition \((\varphi_n)\), the series
\[
x=\sum_{n\in\mathbb Z}\varphi_n(A)x
\]
converges unconditionally, and the fractional-domain norms satisfy
\[
\|x\|_{X^\circ_\theta}\approx
\Big(\sum_{n\in\mathbb Z}2^{2n\theta}\|\varphi_n(A)x\|^2\Big)^{1/2},
\qquad
\|x\|_{X_\theta}\approx
\Big(\sum_{n\ge 0}2^{2n\theta}\|\phi_n(A)x\|^2\Big)^{1/2}.
\]
Real interpolation of these fractional-domain scales yields abstract Besov spaces \(B_q^\theta(A)\) and \(\widetilde B_q^\theta(A)\), while variants for bisectorial operators and strip-type generators extend the same principle beyond single sectors [1407.0821].

The unifying point is that the dyadic pieces \(\varphi_n(A)x\) or the continuous pieces \(t^\theta v(tA)x\) function as sectorially localized frequency bands. This is a genuine decomposition of regularity by the spectrum of a sectorial operator rather than by Euclidean frequency.

## 4. L-systems, non-selfadjoint spectral expansions, and fractional evolution

In the theory of L-systems, sectorial decomposition is encoded by the relation between a dissipative or accumulative main operator \(T\), a state-space \((*)\)-extension \(\mathbb A\), and the impedance function
\[
V_\Theta(z)=K^*(\mathbb A-zI)^{-1}K.
\]
For minimal L-systems with nonnegative symmetric part, the classes \(S^\alpha\), \(S^{\alpha_1,\alpha_2}\), \(S^{-1,\alpha}\), and \(S^{-1,\alpha_1,\alpha_2}\) of sectorial Stieltjes and inverse Stieltjes functions govern the exact angle of sectoriality of \(T\) and of \(\mathbb A\) or the associated operator \(\widetilde{\mathbb A}=2\mathbb A^*-\mathbb A\). In particular, if \(V_\Theta\in S^{0,\alpha}\) or \(V_\Theta\in S^{-1,0,\alpha}\), then the main and state-space operators have the same exact angle \(\alpha\); in the general two-parameter case,
\[
\tan\alpha
=
\tan\alpha_2
+
2\sqrt{\tan\alpha_1(\tan\alpha_2-\tan\alpha_1)}.
\]
For one-dimensional Schrödinger operators on the half-line, the theory makes these relations explicit in terms of the Weyl–Titchmarsh function \(m_\infty\) and the boundary parameter \(h\) [1709.07370].

A different spectral use of sectoriality appears in the monograph on fractional calculus and Abel–Lidskii theory. There the central object is a sectorial operator \(W\) with compact resolvent and compact inverse \(B=W^{-1}\). The monograph formulates sufficient conditions for the Abel–Lidskii basis property, strengthens the conditions regarding the semi-angle of the sector, weakens the conditions regarding the involved parameters, introduces the Schatten-von Neumann class of the convergence exponent, and develops a method of contour integration based on a sequence of contours of the power type [2402.03331].

The outcome is a generalized root-vector decomposition for non-selfadjoint sectorial operators. The Hilbert space is expanded in Jordan chains of root vectors and biorthogonal vectors, and contour integrals defining \(\varphi(W)\) produce Abel–Lidskii expansions of the form
\[
u(t)=\sum_{v=0}^\infty P_v(\varphi^\alpha,t)f.
\]
These expansions are then used for evolution equations in abstract Hilbert space and for operators arising from fractional derivatives, including Kipriyanov-type constructions [2402.03331]. In this setting, generalized sectorial decomposition is not a projection calculus but a regularized decomposition of the space along root subspaces of a non-normal sectorial operator.

## 5. Liouville sectors and symmetric products of surfaces

In symplectic topology, sectorial decomposition refers to covers by Liouville sectors whose boundaries form a sectorial collection of hypersurfaces. For a Riemann surface \(\Sigma\) with a quadratic Stein structure, cutting \(\Sigma\) along stable arcs of saddle points produces a decomposition that lifts to the second symmetric product. The central result is that there exists a collection of smooth hypersurfaces \(H_{s_i}\subset \operatorname{Sym}^2(\Sigma)\) and a decomposition
\[
\operatorname{Sym}^2(\Sigma)=\bigcup_{i\le j} U_{m_i,m_j},
\]
where each \(U_{m_i,m_j}\) is a Liouville sector with corners, obtained as the closure of the stable manifold of the critical point \((m_i,m_j)\) of a Stein potential \(\widetilde\Phi\) on \(\operatorname{Sym}^2(\Sigma)\) [2511.16584].

Locally, the model is built on \(\operatorname{Sym}^2(\mathbb C)\), where stable manifolds of asymptotic regions define hypersurfaces \(H_{0,-}\) and \(H_{0,+}\), and one gets a sectorial decomposition
\[
\operatorname{Sym}^2(\mathbb C)
=
U_{-,-}\cup_{H_{0,-}} U_{-,+}\cup_{H_{0,+}} U_{+,+}.
\]
Globally, the functions \(I_i\) required in the definition of sectorial coverings are produced by transporting \(\Im(z_1)\) along the Liouville flow; the resulting hypersurfaces satisfy the orthogonality and Poisson-commutation conditions that define a sectorial collection [2511.16584].

This construction has categorical consequences. Sectorial descent reconstructs the wrapped Fukaya category of \(\operatorname{Sym}^2(\Sigma)\) from the categories of the pieces \(U_{m_i,m_j}\) and their overlaps. Applied to \(\Sigma=\mathbb P^1-\{\text{4 points}\}\), the decomposition of \(\operatorname{Sym}^2(\Sigma)\) into \(U_{-,-}\), \(U_{-,+}\), and \(U_{+,+}\) yields a new geometric proof of Homological Mirror Symmetry for the complex two-dimensional pair of pants [2511.16584]. Here generalized sectorial decomposition is a geometric gluing formalism rather than a spectral splitting.

## 6. Dynamical stratifications and adjacent algorithmic usages

For analytic vector fields in \(\mathbb R^3\) with an isolated singularity at the origin, a reduction of singularities \(\pi:M\to\mathbb R^3\) produces a three-dimensional analogue of Poincaré’s planar sectorial decomposition. Under the hypotheses that the induced foliated manifold \((M,D,L)\) is spherical, non-dicritical, hyperbolic, acyclic, non s-resonant, and of Morse–Smale type, there exists a fundamental system of neighborhoods \(U_n\) of the divisor \(D\cong S^2\) such that each \(U_n\) admits an analytic \(L\)-stratification into elementary dynamical pieces. These pieces are classified as elliptic, hyperbolic, parabolic attracting, or parabolic repelling according to their \(\alpha\)- and \(\omega\)-limit sets, and they occur in dimensions \(0,1,2,\) and \(3\) [2306.15822].

The construction uses a graph \(\Omega\) on the divisor, fitting domains with carefully controlled frontier types, and two classes of invariant two-dimensional surfaces: fixed separating surfaces and mobile separating surfaces. Their mutual disjointness and their traces on \(D\) provide the walls of the stratification, while the three-dimensional complementary components play the role of higher-dimensional sectors. The result is explicitly presented as a generalization to dimension three of Poincaré’s decomposition of planar flows into parabolic, elliptic, and hyperbolic invariant sectors [2306.15822].

A distinct but terminologically adjacent usage occurs in perturbative quantum field theory. There, sector decomposition means subdivision of the Feynman-parameter domain so that infrared singularities factor into monomials. A non-iterative version recasts the factorization step as a problem in computational geometry: exponent vectors of the relevant polynomials define convex polyhedral cones
\[
\Delta_b^F
=
\{y\in\mathbb R_{\ge 0}^{N-1} : (c-b,y)>0,\ \forall c\in Z_l^F\},
\]
which are then triangulated to produce a finite set of sectors with monomial changes of variables. This method is deterministic, avoids infinite loops, and in the examples reported produces fewer sectors than several iterative strategies [1004.5490]. Although this is sector decomposition rather than sectoriality in the operator-theoretic sense, it belongs to the broader history of decompositions organized by asymptotic cones and sectors.

Across these settings, generalized sectorial decomposition has no single invariant definition. Its stable content is structural: a complicated object is cut into sectorially organized pieces—by logarithmic branch cuts, congruence phases, dyadic functional-calculus bands, Liouville hypersurfaces, invariant separating surfaces, or asymptotic cones—so that non-selfadjoint, non-normal, or singular behavior becomes tractable.

Source: https://www.emergentmind.com/topics/generalized-sectorial-decomposition