Roe's Partitioned Manifold Index Theorem

• The theorem converts a global index problem on a complete manifold into a boundary index problem on a partitioning hypersurface using Roe algebras and Mayer–Vietoris sequences.
• It establishes an equality of K-theory classes rather than merely numerical indices, thus applying to Dirac-type and broader elliptic pseudodifferential operators.
• Subsequent generalizations extend the result to uniform coarse settings, relative index problems, and quantitative applications in positive scalar curvature and band-width estimates.

Searching arXiv for recent and foundational papers on Roe's Partitioned Manifold Index Theorem and related generalizations. arXiv search query: all:"partitioned manifold index theorem Roe" Roe’s Partitioned Manifold Index Theorem identifies the boundary of the coarse or analytic index class of a Dirac-type operator on a complete manifold cut by a hypersurface with the index class of the induced operator on that hypersurface. In its standard coarse-geometric form, the theorem converts a global index problem on a noncompact manifold into a boundary index problem on the partitioning hypersurface, and it does so at the level of $K$-theory classes in Roe-type algebras rather than merely numerical indices. Subsequent work places the theorem in Mayer–Vietoris, $KK$-theoretic, uniform coarse, localization-algebra, relative, multi-partitioned, and quantitative frameworks, and extends it from Dirac operators to substantially larger classes of elliptic pseudodifferential operators (Siegel, 2012, Engel, 2014).

1. Classical geometric and analytic content

A partitioned manifold is a complete Riemannian manifold $M$ together with a closed, two-sided hypersurface $N$ such that $M=M_+\cup M_-$ and $M_+\cap M_-=N$, with product structure near $N$. In the form recalled in later treatments, there is a collar neighborhood $U\cong (-\varepsilon,\varepsilon)\times N$ in which the metric is a product and a compatible Dirac-type operator decomposes as

$D=\sigma(\nu)(\partial_t+B)$

or, in multigraded form,

$$D_M = D_{(-1,1)} \,\widehat{\otimes}\, 1 \;+\; 1 \,\widehat{\otimes}\, D_N,$$

where $KK$0 or $KK$1 is the induced tangential Dirac operator on $KK$2 (Engel, 2018, Siegel, 2012).

The theorem then states that the Mayer–Vietoris boundary of the index class of $KK$3 is the index class of the induced operator on $KK$4. One standard formulation is

$KK$5

with the degree lowered by one under the boundary map (Siegel, 2012). In the version expressed directly in the six-term exact sequence for the partition, one writes

$KK$6

or equivalently

$KK$7

depending on whether one formulates the theorem through classes on the halves or through the global coarse index class (Engel, 2018).

The parity convention is part of the statement. If the operator is odd and ungraded, the index lies in $KK$8; if it is even and graded, it lies in $KK$9; the boundary map lowers degree by one (Engel, 2014). This grading shift is not a secondary bookkeeping issue: it is the formal expression of the fact that passing from $M$0 to a hypersurface $M$1 lowers dimension by one.

A common simplification is to describe the theorem as equality of Fredholm indices. That description is too narrow. The basic content is an equality of $M$2-theory classes in Roe-type algebras, from which numerical pairings are extracted only after applying traces, cyclic cocycles, or other functionals.

2. Roe algebras, localization near the hypersurface, and Mayer–Vietoris

The theorem belongs to coarse index theory, so its natural receptacles are Roe algebras and related quotients. For a complete manifold $M$3, the Roe algebra $M$4 is the $M$5-closure of locally compact finite-propagation operators, while the larger pseudolocal algebra $M$6 contains $M$7 as an ideal (Siegel, 2012, Seto, 2017). In bounded-geometry settings one also works with the uniform Roe algebra $M$8, defined as the closure of finite propagation operators with uniformly bounded coefficients on a uniformly discrete quasi-lattice (Engel, 2014).

For a partition $M$9 with interface $N$0, there is an ideal $N$1 of operators supported near $N$2. This localized Roe algebra is canonically identified in $N$3-theory with the intrinsic Roe algebra of $N$4 under the usual coarse-equivalence hypotheses (Hochs et al., 22 Jul 2025). The partition gives a coarse excisive decomposition and hence a Mayer–Vietoris exact sequence in $N$5-theory. In one of its standard forms,

$N$6

(Engel, 2018). In the localized formulation, one uses the extension

$N$7

whose boundary map defines the partitioned index class (Hochs et al., 22 Jul 2025).

This algebraic structure explains why the theorem is natural. The partition supplies a geometric excision datum, the Roe algebra packages large-scale elliptic information, and the boundary homomorphism is the functorial mechanism that transfers the index class from $N$8 to $N$9. In the uniform coarse setting the same pattern persists, with $M=M_+\cup M_-$0, uniform $M=M_+\cup M_-$1-homology, and the uniform coarse assembly map replacing their non-uniform counterparts (Engel, 2014).

The role of a partitioning function is also structural. In Roe’s approach one uses a characteristic or Heaviside-type function, denoted $M=M_+\cup M_-$2, $M=M_+\cup M_-$3, or $M=M_+\cup M_-$4, to compress operators to the two halves and to control commutators near the interface (Seto, 2017, Engel, 2014). The theorem is therefore simultaneously geometric and operator-theoretic: the hypersurface $M=M_+\cup M_-$5 enters both through the collar geometry and through these compression operators.

3. Boundary-of-Dirac mechanisms

The product collar near $M=M_+\cup M_-$6 is the decisive analytic input. Because the Dirac operator splits as normal plus tangential part, the boundary class in $M=M_+\cup M_-$7-homology is represented by the tangential Dirac operator. In the explicit Mayer–Vietoris framework, if $M=M_+\cup M_-$8 and $M=M_+\cup M_-$9 are the characteristic-function multipliers of the two halves, then the boundary map may be written at operator level by compressing a representative $M_+\cap M_-=N$0 or $M_+\cap M_-=N$1 of the Dirac class:

$M_+\cap M_-=N$2

This compression formula makes the slogan “boundary of Dirac is Dirac” into an operator identity inside the quotient algebra (Siegel, 2012).

Later work reframes the same principle in $M_+\cap M_-=N$3-theory. In a non-equivariant and equivariant setting, one constructs an extension class $M_+\cap M_-=N$4 for a collar compactification and proves

$M_+\cap M_-=N$5

Under Paschke duality and coarse assembly, this $M_+\cap M_-=N$6-identity becomes the partitioned-manifold index identity in the Roe algebra (Abdolmaleki et al., 2023). This formulation shows that PMIT is not merely a consequence of a specific functional-calculus trick; it is a manifestation of functoriality of boundary maps in analytic $M_+\cap M_-=N$7-homology.

A third viewpoint uses cyclic cohomology. Roe’s cyclic $M_+\cap M_-=N$8-cocycle on a partitioned manifold is

$M_+\cap M_-=N$9

defined on a dense subalgebra on which the commutators are trace class. Connes’ pairing then gives

$N$0

so the partition is detected by a Fredholm index of a compressed operator (Seto, 2017). In the absolute Dirac case this reproduces Roe’s identity relating the pairing of the coarse Dirac class to the index of the boundary Dirac operator. The cyclic-cocycle picture is therefore a numerical shadow of the $N$1-theoretic boundary identity.

These proof paradigms are complementary. Explicit compression formulas clarify how the Mayer–Vietoris boundary is implemented. $N$2-theoretic proofs isolate the categorical mechanism. Cyclic-cocycle proofs explain how numerical boundary indices arise from the same structure.

4. Uniform coarse generalization to elliptic pseudodifferential operators

A major extension replaces Dirac-type operators by elliptic pseudodifferential operators on manifolds of bounded geometry. In this setting one works with uniform pseudodifferential operators whose local symbols satisfy chart-independent estimates

$N$3

together with quasilocality conditions for remainders and functional calculus (Engel, 2014). For symmetric elliptic operators of positive order, Schwartz functional calculus produces quasilocal smoothing operators, order-zero operators become uniformly pseudolocal, and negative-order quasilocal operators become uniformly locally compact. These properties are exactly what is needed to produce classes in uniform $N$4-homology (Engel, 2014).

The resulting theorem states that if $N$5 is a symmetric, elliptic, odd, multigraded pseudodifferential operator of product type near the partitioning hypersurface $N$6, then its uniform analytic index class satisfies

$N$7

Here $N$8 is defined through the uniform coarse assembly map $N$9 (Engel, 2014).

A related local-index-theoretic treatment formulates the extension for multigraded, symmetric, elliptic uniform pseudodifferential operators and expresses the same partition phenomenon in terms of the induced tangential operator $U\cong (-\varepsilon,\varepsilon)\times N$0. Near the hypersurface, the principal symbol has product form

$U\cong (-\varepsilon,\varepsilon)\times N$1

and the theorem becomes

$U\cong (-\varepsilon,\varepsilon)\times N$2

In amenable situations, traces on the uniform Roe algebra then give numerical equalities

$U\cong (-\varepsilon,\varepsilon)\times N$3

(Engel, 2018).

This extension changes the scope of PMIT in two ways. First, the operator class is no longer restricted to generalized Dirac operators. Second, the theorem is embedded in a larger uniform coarse package: uniform $U\cong (-\varepsilon,\varepsilon)\times N$4-homology, uniform $U\cong (-\varepsilon,\varepsilon)\times N$5-theory, uniform Chern characters, Poincaré duality on bounded-geometry manifolds, and Mayer–Vietoris sequences adapted to uniformity (Engel, 2014). A plausible implication is that the partitioned-manifold theorem is best understood not as an isolated boundary formula, but as one manifestation of a uniform local-to-global index formalism.

5. Relative, Toeplitz, multipartition, noncompact, and quantitative variants

Several later developments preserve the partitioned-manifold principle while altering the index class, the boundary geometry, or the ambient coarse category.

Variant	Representative statement	Source
Relative partitioned theorem	$U\cong (-\varepsilon,\varepsilon)\times N$6	(Seto, 2017)
Even-dimensional Toeplitz form	$U\cong (-\varepsilon,\varepsilon)\times N$7	(Seto, 2014)
Multi-partitioned version	$U\cong (-\varepsilon,\varepsilon)\times N$8	(Schick et al., 2013)
Noncompact hypersurfaces	$U\cong (-\varepsilon,\varepsilon)\times N$9 in $D=\sigma(\nu)(\partial_t+B)$0	(Hochs et al., 22 Jul 2025)
Quantitative form	$D=\sigma(\nu)(\partial_t+B)$1 at controlled scale	(Hochs et al., 6 Feb 2026)
Secondary/localization-algebra form	$D=\sigma(\nu)(\partial_t+B)$2 and corresponding secondary boundary formulas	(Zeidler, 2014)

The relative theory extends Roe’s cyclic cocycle to ideals supported near a closed subset $D=\sigma(\nu)(\partial_t+B)$3 and proves partitioned relative index formulas after compactification of the boundary pieces. In this setting the theorem compares differences of coarse index classes rather than a single global class (Seto, 2017). This is useful when two manifolds agree outside controlled subsets.

The Toeplitz extension addresses an even-dimensional deficiency of the classical Dirac pairing. When $D=\sigma(\nu)(\partial_t+B)$4 is even, the boundary Dirac index can vanish for parity reasons, so one replaces the odd coarse Dirac class by a $D=\sigma(\nu)(\partial_t+B)$5-class built from a function $D=\sigma(\nu)(\partial_t+B)$6. The pairing with Roe’s cocycle then detects the Toeplitz index on the boundary hypersurface rather than the Dirac index (Seto, 2014). A two-dimensional analogue on partitioned Riemannian surfaces gives the same kind of equality for $D=\sigma(\nu)(\partial_t+B)$7 and Toeplitz operators on $D=\sigma(\nu)(\partial_t+B)$8 (Seto, 2014).

The multi-partitioned theorem iterates the boundary construction. If $D=\sigma(\nu)(\partial_t+B)$9 is cut by $$D_M = D_{(-1,1)} \,\widehat{\otimes}\, 1 \;+\; 1 \,\widehat{\otimes}\, D_N,$$0 pairwise coarsely transversal hypersurfaces with compact transverse intersection $$D_M = D_{(-1,1)} \,\widehat{\otimes}\, 1 \;+\; 1 \,\widehat{\otimes}\, D_N,$$1, then the $$D_M = D_{(-1,1)} \,\widehat{\otimes}\, 1 \;+\; 1 \,\widehat{\otimes}\, D_N,$$2-fold Mayer–Vietoris boundary of the coarse index on $$D_M = D_{(-1,1)} \,\widehat{\otimes}\, 1 \;+\; 1 \,\widehat{\otimes}\, D_N,$$3 is the Fredholm index of the induced Dirac operator on $$D_M = D_{(-1,1)} \,\widehat{\otimes}\, 1 \;+\; 1 \,\widehat{\otimes}\, D_N,$$4 (Schick et al., 2013). An equivariant and higher-index version turns this into a map from the higher index of the ambient manifold to the Rosenberg index of the fiber or submanifold, with applications to fiber bundles over aspherical bases of finite asymptotic dimension (Zeidler, 2015).

The noncompact-hypersurface theory removes the compactness restriction on the partitioning hypersurface, under hypotheses such as a uniform tubular neighborhood and uniform bilipschitz comparison between intrinsic and ambient metrics near $$D_M = D_{(-1,1)} \,\widehat{\otimes}\, 1 \;+\; 1 \,\widehat{\otimes}\, D_N,$$5. The theorem then identifies the partitioned index localized to $$D_M = D_{(-1,1)} \,\widehat{\otimes}\, 1 \;+\; 1 \,\widehat{\otimes}\, D_N,$$6 with the coarse index of the induced Dirac operator on $$D_M = D_{(-1,1)} \,\widehat{\otimes}\, 1 \;+\; 1 \,\widehat{\otimes}\, D_N,$$7 in $$D_M = D_{(-1,1)} \,\widehat{\otimes}\, 1 \;+\; 1 \,\widehat{\otimes}\, D_N,$$8-theory of a localized Roe algebra (Hochs et al., 22 Jul 2025). The quantitative version goes further by producing a controlled index class $$D_M = D_{(-1,1)} \,\widehat{\otimes}\, 1 \;+\; 1 \,\widehat{\otimes}\, D_N,$$9 supported in a neighborhood of size $KK$00 of the hypersurface, where $KK$01 is the amplitude of a defining function $KK$02 (Hochs et al., 6 Feb 2026).

These variants clarify two recurrent misconceptions. PMIT is not restricted to compact or cocompact interfaces in modern formulations, and it is not restricted to ordinary Dirac operators once one moves to the uniform pseudodifferential framework.

6. Applications and conceptual role

The most persistent application is to positive scalar curvature. If the ambient coarse index vanishes under uniformly positive scalar curvature, PMIT shows that a nonvanishing boundary index on the partitioning hypersurface obstructs such a metric on the ambient manifold. This mechanism appears in the classical codimension-one setting, in multi-partitioned manifolds, and in noncompact hypersurface theorems (Engel, 2018, Schick et al., 2013, Hochs et al., 22 Jul 2025). In the multi-partitioned setting, positivity on a single quadrant already forces vanishing of the multi-partitioned index, so a nonzero index on the compact transverse intersection obstructs uniformly positive scalar curvature on the whole manifold (Schick et al., 2013).

The theorem also yields cobordism-type consequences. The noncompact hypersurface theory explicitly derives cobordism invariance of the coarse index from the partitioned-manifold identity (Hochs et al., 22 Jul 2025). The analytic-structure-group approach likewise uses explicit Mayer–Vietoris formulas to reprove and generalize partitioned-manifold index theorems in equivariant and higher-codimension settings (Siegel, 2012).

A second application is to secondary invariants. In localization algebras, one has

$KK$03

and when the metric has uniformly positive scalar curvature outside a subset, the same formalism yields boundary formulas for partial secondary invariants and for the higher relative index of two such metrics (Zeidler, 2014). This makes PMIT part of a broader delocalized APS-type theory rather than a statement confined to primary coarse indices.

A third application is quantitative geometry. The quantitative partitioned index theorem produces controlled representatives of the partitioned index near noncompact hypersurfaces and is used to prove noncompact band-width estimates. Under a nonvanishing coarse index assumption on the cross-section, the scalar curvature of a regular band of width $KK$04 satisfies

$KK$05

which is the sharp Gromov-type band-width bound obtained in that framework (Hochs et al., 6 Feb 2026).

Conceptually, Roe’s theorem is a large-scale boundary principle. It says that the index class of an elliptic object on a partitioned space is not lost in the coarse Roe algebra of the ambient manifold; it survives as a boundary class on the hypersurface. The theorem’s later generalizations show that this principle is robust under changes of operator class, coefficient algebra, equivariance, partition complexity, and hypersurface compactness. In current usage, PMIT functions less as a single theorem than as a unifying template for boundary transfer in coarse index theory.