Glocal Spectral Subspaces

• Glocal spectral subspaces are frameworks that use local spectral data (e.g., eigenprojectors, local transforms) to construct global spectral decompositions.
• They integrate techniques from microlocal analysis, automorphic forms, graph theory, and Banach space theory to bridge local computations with global structures.
• These methods enable precise analytic control and recovery of global spectral properties, with applications ranging from elliptic systems to modular ultrapowers.

Glocal spectral subspaces are spectral constructions in which local spectral data determine, approximate, or obstruct a globally meaningful subspace, projector, or decomposition. In the literature, the expression does not denote a single universal definition. It appears in microlocal spectral theory for elliptic systems, in tensorially assembled automorphic projectors, in graph-spectral and harmonic-analysis factorization frameworks, in Banach-space local spectral theory for semigroups and subordinated operators, in interval-restricted frame theory, and in modular spectral theory of ultrapowers (Capoferri et al., 2021, Blomer et al., 2024, Fernández-Menduiña et al., 21 Feb 2026, Gallardo-Gutiérrez et al., 6 Aug 2025, Ando et al., 21 May 2026).

1. Terminological scope and common structural pattern

Across the surveyed works, the common feature is a local-to-global spectral mechanism: one begins with local eigenprojectors, local transforms, local fibers, local samples, or coordinatewise spectral constraints, and one obtains a global spectral object such as a decomposition, an invariant subspace, a projector, or a failure-of-commutation phenomenon. The local ingredient is not uniform across fields, and the meaning of “spectral subspace” itself varies between Hilbert spectral bands, Arveson spectral subspaces, local spectral theory, and invariant subspaces of group representations.

Setting	Local ingredient	Global spectral object
Elliptic systems	$p_j(x,\xi)$, $P_j$	$H_j=\operatorname{Ran}(P_j)$
Automorphic $PGL_2$	$h_v^\vee$, $h_v^\#$	$\prod_{v\in S} h_v^\vee$ or $\prod_{v\in S} h_v^\#$
Graphs and LCA groups	local eigenspaces, anchors, fibers	global GFT factorization or fiberwise invariant $V$
Operator and modular theory	local resolvent or coordinatewise spectrum	$X_T(F)$ or $P_j$0

In microlocal analysis, the subspaces are almost-invariant and almost-orthogonal modulo $P_j$1. In automorphic analysis, they are exact tensorial projectors built from local integral transforms. In graph settings, they can be exact factorizations of the global graph Fourier transform or local-recovery mechanisms for a global spectral support. In Banach local spectral theory, the subspace is defined by analytic solvability on $P_j$2. In modular ultrapowers, glocality names the mismatch between coordinatewise spectral constraints and the spectral subspace of the ultraproduct itself (Capoferri et al., 2021, Cabrelli et al., 2011, Jorgensen et al., 2015, Ando et al., 21 May 2026).

2. Microlocal and analytic realizations

For elliptic self-adjoint pseudodifferential systems, glocal spectral subspaces arise from microlocally defined eigenprojectors of the principal symbol. If $P_j$3 is elliptic and self-adjoint on $P_j$4-columns of half-densities, and the principal symbol $P_j$5 has simple eigenvalues $P_j$6 with eigenprojectors $P_j$7, then there exist pseudodifferential projections $P_j$8 satisfying

$P_j$9

and

$H_j=\operatorname{Ran}(P_j)$0

The associated glocal spectral subspaces are

$H_j=\operatorname{Ran}(P_j)$1

They are local because $H_j=\operatorname{Ran}(P_j)$2 is defined from $H_j=\operatorname{Ran}(P_j)$3 on $H_j=\operatorname{Ran}(P_j)$4, and global because $H_j=\operatorname{Ran}(P_j)$5 is an $H_j=\operatorname{Ran}(P_j)$6-subspace. Spectrally, the positive spectrum of $H_j=\operatorname{Ran}(P_j)$7 decomposes, up to superpolynomially small errors, into $H_j=\operatorname{Ran}(P_j)$8 series associated with the positive principal branches, with

$H_j=\operatorname{Ran}(P_j)$9

Dynamically, the propagator decomposes as

$PGL_2$0

and singularities in $PGL_2$1 propagate along the Hamiltonian flow of $PGL_2$2 (Capoferri et al., 2021).

A second analytic realization occurs for the Sturm–Liouville operator $PGL_2$3 on $PGL_2$4, where $PGL_2$5 is positive and piecewise constant. The spectral subspace is

$PGL_2$6

and its local bandwidth is determined by

$PGL_2$7

This is glocal in a different sense: the subspace is defined globally by the spectral theorem for $PGL_2$8, but within each interval where $PGL_2$9, functions behave like classical bandlimited functions with local bandwidth $h_v^\vee$0. The reproducing kernel is explicit,

$h_v^\vee$1

and the sampling density theorem is governed by the weighted measure

$h_v^\vee$2

For finite-measure $h_v^\vee$3,

$h_v^\vee$4

The local sinc-like behavior is therefore globally corrected by transmission and reflection across interfaces, through the explicit kernel and the factor $h_v^\vee$5 (Celiz et al., 2023).

3. Automorphic and arithmetic constructions

In the automorphic setting of $h_v^\vee$6 over a number field $h_v^\vee$7, glocal spectral subspaces are produced by tensoring explicit local integral transforms into global spectral projectors. The local inputs are Whittaker/Kirillov test functions $h_v^\vee$8, from which one forms local kernels

$h_v^\vee$9

The associated local transform weights are $h_v^\#$0 for shifted convolution and $h_v^\#$1 for the second-moment side. These are given by Mellin-type integrals against local characters with kernels involving local gamma-factors, and they admit explicit inversion formulae. At archimedean places the kernels are hypergeometric; at non-archimedean places the transforms are expressed uniformly in terms of local $h_v^\#$2-factors and the Kirillov model. Lemma 2.5 yields rapid decay in the conductor $h_v^\#$3.

The global projector is then defined by restricted tensor product. For a finite set $h_v^\#$4 of places,

$h_v^\#$5

for shifted convolution, or

$h_v^\#$6

for the second moment. These weights appear in exact global spectral decompositions on $h_v^\#$7. For shifted convolution sums, the paper gives

$h_v^\#$8

For the second moment of automorphic $h_v^\#$9-functions,

$\prod_{v\in S} h_v^\vee$0

The characterization stated in the paper is explicitly spectral: the glocal subspace consists of those $\prod_{v\in S} h_v^\vee$1 whose local spectral parameters lie in the support of $\prod_{v\in S} h_v^\vee$2, where $\prod_{v\in S} h_v^\vee$3 denotes $\prod_{v\in S} h_v^\vee$4 or $\prod_{v\in S} h_v^\vee$5. At $\prod_{v\in S} h_v^\vee$6, choosing $\prod_{v\in S} h_v^\vee$7 in a window of $\prod_{v\in S} h_v^\vee$8-values produces a hypergeometric kernel concentrating on $\prod_{v\in S} h_v^\vee$9. At $\prod_{v\in S} h_v^\#$0, support on $\prod_{v\in S} h_v^\#$1 or a coset controls ramification or Satake ranges, and the decay in $\prod_{v\in S} h_v^\#$2 filters $\prod_{v\in S} h_v^\#$3-adic depth. The resulting projector is therefore local at each place and global at the automorphic level (Blomer et al., 2024).

4. Graphs, local sampling, and harmonic-analysis fiberizations

In spectral graph learning, glocal spectral subspaces are used to reconstruct global graph spectral structure from local components. In L2G-Net, the graph is partitioned into connected subgraphs with local Laplacians $\prod_{v\in S} h_v^\#$4 and local eigenbases $\prod_{v\in S} h_v^\#$5. Adding a bridge edge is a rank-one update of the Laplacian, and the change of eigenvectors is encoded by orthogonal Cauchy-like matrices. The global graph Fourier transform admits the exact factorization

$\prod_{v\in S} h_v^\#$6

The “glocal spectral basis” is built by composing local bases and these Cauchy factors across the hierarchy. With all interfaces retained, the factorization is exact; with interface sparsification, the Laplacian quadratic form is preserved within $\prod_{v\in S} h_v^\#$7, and for Lipschitz spectral filters the output deviation is $\prod_{v\in S} h_v^\#$8. The construction yields global receptive fields together with local spectral inductive bias (Fernández-Menduiña et al., 21 Feb 2026).

For frequency-sparse graph signals, the same local-to-global principle appears in recovery rather than factorization. If

$\prod_{v\in S} h_v^\#$9

is $V$0-sparse in the graph Fourier domain, then local samples on $V$1 or on a union of small neighborhoods can recover the active spectral support under zero-free or rank conditions such as $V$2 or $V$3. The local operator moments

$V$4

form Hankel or stacked Hankel matrices whose nullspace yields the annihilating polynomial for the active eigenvalues. Once the support is identified, the projector

$V$5

recovers

$V$6

locally. The paper explicitly describes this as a glocal mechanism: local aggregation of powers of $V$7 reveals and isolates components of the global spectral atoms (Emmrich et al., 2023).

In harmonic analysis on LCA groups, glocal spectral subspaces arise as $V$8-shift-modulation invariant spaces. With $V$9 and $X_T(F)$0 uniform lattices, the unitary representation is

$X_T(F)$1

A Zak-type fiberization

$X_T(F)$2

reduces simultaneous shift and modulation invariance to a measurable range function $X_T(F)$3 with periodicity in $X_T(F)$4. The classification theorem states that a closed subspace $X_T(F)$5 is $X_T(F)$6-invariant if and only if

$X_T(F)$7

Here the global invariant space is a direct integral of fiber subspaces, selected by local data in $X_T(F)$8 and constrained by the representation $X_T(F)$9 (Cabrelli et al., 2011).

5. Local spectral theory, semigroups, and interval-restricted frames

In local spectral theory for subordinated operators, glocal spectral subspaces are defined by analytic solvability on the complement of a closed set. For $P_j$00 and closed $P_j$01, the paper defines

$P_j$02

This is stronger than mere local spectral inclusion when SVEP fails. For a subordinated operator

$P_j$03

built from a $P_j$04-semigroup with a dense analytic eigenvector field, the main theorem shows that $P_j$05 does not have SVEP, that $P_j$06 is dense for every nonempty relatively open $P_j$07, and that on reflexive spaces the adjoint $P_j$08 has trivial spectral subspaces for proper closed $P_j$09 and enjoys Dunford property $P_j$10. For the Cesàro operator $P_j$11 on $P_j$12, $P_j$13, the same machinery yields

$P_j$14

and for every nonzero $P_j$15,

$P_j$16

Moreover, if $P_j$17 is any nontrivial closed $P_j$18-invariant subspace, then $P_j$19. In this setting, glocality is mediated by the Hille–Phillips functional calculus, Koenigs-domain geometry, and the adjoint annihilator relation from Laursen–Neumann; the paper also situates the Cesàro results relative to Siskakis, Brown–Halmos–Shields, Miller–Miller–Smith, Betsakos, and Bracci–Gallardo–Yakubovich (Gallardo-Gutiérrez et al., 6 Aug 2025).

For polynomially bounded $P_j$20-groups, glocal spectral subspaces are characterized by resolvent boundary behavior. If $P_j$21 is the generator and

$P_j$22

then for a closed $P_j$23,

$P_j$24

If $P_j$25 is reflexive, $P_j$26, and the polynomial growth exponent lies in $P_j$27, then

$P_j$28

The same theorem gives algebraic range characterizations: $P_j$29 and

$P_j$30

Here glocality consists in converting local spectral support on $P_j$31 into global harmonic and analytic control in $P_j$32 (Borichev et al., 2010).

A third variant appears in the generalized Gramian framework of Jorgensen–Tian. Let $P_j$33 be the selfadjoint nonnegative operator attached to a countable system $P_j$34 with possibly unbounded Gramian. For every finite interval $P_j$35, the spectral subspace

$P_j$36

is a maximal closed subspace on which the original global system becomes a standard frame with bounds $P_j$37 and $P_j$38: $P_j$39 Equivalently, the projected vectors

$P_j$40

form a frame for $P_j$41. This is glocal in the sense that the analysis dictionary is global, while stability is localized to spectral bands of $P_j$42 (Jorgensen et al., 2015).

6. Ultrapowers, definability, and limits of the concept

In modular theory of von Neumann algebras, glocal spectral subspaces mark a failure of local spectral constraints to commute with a global ultrapower. For a $P_j$43-probability space $P_j$44 with modular flow $P_j$45, the spectral subspace for closed $P_j$46 is

$P_j$47

Ando–Goldbring prove that if $P_j$48 is a type $P_j$49 factor, $P_j$50 is nonempty, proper, and closed, and $P_j$51 is a nonprincipal ultrafilter, then

$P_j$52

The forward inclusion always holds, but equality fails except in the trivial cases $P_j$53 or $P_j$54; it also fails to produce a definable set in the model-theoretic sense. The mechanism is asymptotic annihilation of off-$P_j$55 spectral leakage in the ultralimit, enabled by type $P_j$56 modular dynamics and the structure of the ultraproduct. The paper further notes that principal ultrafilters produce equality, and that the type $P_j$57 tracial case has trivial modular flow, so no glocal gap appears (Ando et al., 21 May 2026).

A common misconception is that “glocal” always means an exact local-to-global assembly of a projector. The surveyed literature shows three distinct regimes. In automorphic analysis, shift-modulation fiberization, and Cauchy-factorized graph Fourier analysis, the assembly is exact (Blomer et al., 2024, Cabrelli et al., 2011, Fernández-Menduiña et al., 21 Feb 2026). In microlocal elliptic theory, the decomposition is only valid modulo $P_j$58 or $P_j$59 (Capoferri et al., 2021). In ultrapower modular theory, the local and global notions do not commute at all, and the global object is strictly larger than the coordinatewise one (Ando et al., 21 May 2026).

A second misconception is that “spectral subspace” has a fixed technical meaning across these papers. It may mean an $P_j$60-range of a pseudodifferential projection, a spectral band $P_j$61, an Arveson-type modular subspace, a Banach local spectral subspace defined by analytic resolvent solvability, or an invariant direct-integral subspace selected by a measurable range function. The term “glocal” therefore names a recurring structural principle rather than a single formal definition. This suggests that the most stable cross-disciplinary content of the notion is the coupling of local spectral control with a global spectral realization, whether by tensor products, pseudodifferential quantization, fiberization, functional calculus, or ultraproduct limits.