Normal Pseudomanifolds: Structure & Classification

• Normal pseudomanifolds are pure d-dimensional simplicial complexes where every (d-1)-face lies in one or two facets and links of codimension ≥2 are connected, ensuring controlled local singularities.
• They leverage invariants like g2 and rigidity inequalities to relate face numbers and vertex link topologies, providing quantitative tools for classification.
• Constructive operations such as connected sums, edge contractions, foldings, and vertex suspensions underpin modern classifications, especially in three-dimensional cases.

Normal pseudomanifolds are pseudomanifolds whose local singular behavior is constrained by connected-link conditions. In the simplicial boundaryless convention, a normal $d$-pseudomanifold is a pure $d$-dimensional simplicial complex in which every $(d-1)$-face lies in exactly two facets and the link of every face of codimension at least $2$ is connected; boundary versions replace “exactly two” by “one or two” and regard the codimension-$1$ faces contained in exactly one facet as the boundary. This places normal pseudomanifolds strictly beyond manifolds, because vertex links need not be spheres, but excludes disconnected local pinches. In dimension $3$, vertex links are connected closed surfaces, and recent work organizes the subject around face-number invariants such as $g_2$, rigidity inequalities, and constructive classifications by local operations (Basak et al., 2022, Basak et al., 2018).

1. Definition and local link structure

For a simplicial complex $K$ and a face $\sigma\in K$, the link is

$$\operatorname{link}_K(\sigma)=\{\tau\in K:\tau\cap \sigma=\emptyset,\ \tau\cup \sigma\in K\}.$$

A normal $d$0-pseudomanifold is a pure simplicial complex satisfying the two-facet incidence condition on ridges together with connectedness of links in codimension at least $d$1. In dimension $d$2, this implies that the link of every vertex is a connected compact surface; a vertex is nonsingular if its link is $d$3, and singular otherwise. The multiset of nonspherical vertex links records the singularity data of the complex (Basak et al., 2018).

Normality is recursive. A connected pure $d$4-complex is a normal $d$5-pseudomanifold if and only if the link of every vertex is a normal $d$6-pseudomanifold. In the flag setting, this recursive viewpoint is especially effective because links are again flag normal pseudomanifolds, and it underlies graph-theoretic reformulations of the subject (Basak et al., 29 Jun 2026).

In the PL-topological framework of pseudomanifold bordism, normality is exactly the local condition that all links are connected. More precisely, for a classical PL stratified pseudomanifold $d$7, normality is equivalent to the requirement that for every singular point $d$8, the link $d$9 satisfies $(d-1)$0. This characterization treats normal pseudomanifolds as one instance of a larger class of spaces defined by local link properties (Friedman, 2013).

2. Face numbers, $(d-1)$1-vectors, and rigidity

The principal numerical invariant in the recent combinatorial theory is $(d-1)$2. For a $(d-1)$3-dimensional simplicial complex,

$(d-1)$4

In dimension $(d-1)$5,

$(d-1)$6

This is the form used throughout the classification theory of normal $(d-1)$7-pseudomanifolds. For a triangulated connected compact surface $(d-1)$8, one has

$(d-1)$9

with $2$0 taken over $2$1; thus the $2$2-value of a vertex link in a normal $2$3-pseudomanifold depends only on the topology of that surface (Basak et al., 2018).

Rigidity theory gives the fundamental lower bounds. For every normal $2$4-pseudomanifold with $2$5,

$2$6

for every vertex $2$7, and more generally for every face of codimension at least $2$8. In dimension $2$9, this inequality compares the global excess of edges with the topology of each vertex link. A complex is said to have relatively minimal $1$0 if equality holds for the link of some nonempty face of codimension at least $1$1; in dimension $1$2, this means equality for some vertex link (Basak et al., 2018).

The same rigidity package identifies $1$3 with the dimension of the generic stress space and supports local-to-global estimates. For normal $1$4-pseudomanifolds, a useful refinement states that if $1$5 are edges whose endpoints lie in $1$6 but which are not edges of $1$7, then

$1$8

This estimate is central in the proofs that small $1$9 forces very few singular vertices (Basak et al., 2022).

A second identity links $3$0-numbers to local Euler characteristics. For a $3$1-dimensional normal pseudomanifold,

$3$2

In particular, singular vertices with $3$3 links contribute $3$4, and parity constraints on the sum of Euler characteristics force the number of singular vertices to be even in the small-$3$5 regime (Basak et al., 2022).

3. Constructive operations

The modern classification theory is operation-theoretic. Connected sum of normal pseudomanifolds preserves normality, and in dimension $3$6 it is additive on $3$7: $3$8 Handle addition is combinatorially similar but, in dimension $3$9, increases $g_2$0 by $g_2$1; this is why handle additions are excluded in small-$g_2$2 classifications (Basak et al., 2022).

Edge contraction and its inverse, edge expansion, are the main topology-preserving local moves. If $g_2$3 satisfies the usual link condition and $g_2$4 is an $g_2$5-cycle, then in dimension $g_2$6 contraction changes $g_2$7 by

$g_2$8

while the inverse expansion changes it by $g_2$9. When one endpoint is nonsingular, contraction preserves the underlying topology $K$0 (Basak et al., 2022).

Bistellar moves provide the unit changes in $K$1. In a $K$2-dimensional normal pseudomanifold, a bistellar $K$3-move decreases $K$4 by $K$5, and the inverse bistellar $K$6-move increases $K$7 by $K$8. Two-facets insertion and contraction play a parallel role in reduction arguments for manifold cases with $K$9 (Basak et al., 2022).

Foldings are the characteristic singularity-producing operations. In dimension $\sigma\in K$0, vertex folding increases $\sigma\in K$1 by $\sigma\in K$2, and edge folding increases $\sigma\in K$3 by $\sigma\in K$4. Edge folding preserves normality and creates exactly two singular vertices with $\sigma\in K$5 links. These operations are detected combinatorially by missing tetrahedra together with separating or Möbius-strip behavior in vertex links, and they are indispensable in constructive descriptions of nonspherical normal $\sigma\in K$6-pseudomanifolds (Basak et al., 2021, Basak et al., 2022).

Two auxiliary constructions recur throughout the literature. Facet subdivision preserves $\sigma\in K$7, so it does not alter relative minimality. One-vertex suspension $\sigma\in K$8 satisfies

$\sigma\in K$9

where $$\operatorname{link}_K(\sigma)=\{\tau\in K:\tau\cap \sigma=\emptyset,\ \tau\cup \sigma\in K\}.$$0 is the number of vertices of $$\operatorname{link}_K(\sigma)=\{\tau\in K:\tau\cap \sigma=\emptyset,\ \tau\cup \sigma\in K\}.$$1 not adjacent to $$\operatorname{link}_K(\sigma)=\{\tau\in K:\tau\cap \sigma=\emptyset,\ \tau\cup \sigma\in K\}.$$2; if $$\operatorname{link}_K(\sigma)=\{\tau\in K:\tau\cap \sigma=\emptyset,\ \tau\cup \sigma\in K\}.$$3 is a graph cone point, then $$\operatorname{link}_K(\sigma)=\{\tau\in K:\tau\cap \sigma=\emptyset,\ \tau\cup \sigma\in K\}.$$4 is unchanged. This move is fundamental in realizing suspensions of surfaces and in the classification of relatively minimal $$\operatorname{link}_K(\sigma)=\{\tau\in K:\tau\cap \sigma=\emptyset,\ \tau\cup \sigma\in K\}.$$5 examples (Basak et al., 2018).

4. Three-dimensional classification

The sharpest current results concern normal $$\operatorname{link}_K(\sigma)=\{\tau\in K:\tau\cap \sigma=\emptyset,\ \tau\cup \sigma\in K\}.$$6-pseudomanifolds with small $$\operatorname{link}_K(\sigma)=\{\tau\in K:\tau\cap \sigma=\emptyset,\ \tau\cup \sigma\in K\}.$$7. The case $$\operatorname{link}_K(\sigma)=\{\tau\in K:\tau\cap \sigma=\emptyset,\ \tau\cup \sigma\in K\}.$$8 is the classical stacked-sphere case: the complex is a connected sum of boundaries of $$\operatorname{link}_K(\sigma)=\{\tau\in K:\tau\cap \sigma=\emptyset,\ \tau\cup \sigma\in K\}.$$9-simplices. For $d$00, normal $d$01-pseudomanifolds have no singular vertices and are $d$02-spheres obtained from boundaries of $d$03-simplices by connected sums and edge expansions. For $d$04, nonspherical examples have exactly two singular vertices, each with link $d$05. For $d$06, a normal $d$07-pseudomanifold has at most two singular vertices; if it has none, then it is a $d$08-sphere, while every nonspherical example is obtained from boundaries of $d$09-simplices by connected sums, edge expansions, and an edge folding (Basak et al., 2022).

The relatively minimal $d$10 problem yields a complementary classification. In dimension $d$11, every normal pseudomanifold with relatively minimal $d$12 is, up to facet subdivision, built from a complex whose entire graph equals the graph of the star of a single vertex. With at most two singularities, this forces pseudocompression body topology. In particular, if $d$13 and $d$14 is not a triangulation of $d$15, then $d$16 is obtained by taking the one-vertex suspension of a triangulation of $d$17 at a graph cone point and subdividing facets; equivalently, the only nonspherical topology occurring with $d$18 in dimension $d$19 is $d$20 (Basak et al., 2018).

The range of singularity patterns has also been extended beyond two singular vertices under $d$21-minimality. If a $d$22-minimal normal $d$23-pseudomanifold has three singular vertices including one $d$24-singularity, or four singular vertices including two $d$25-singularities, then it is obtained from a one-vertex suspension of a surface and some boundary complexes of $d$26-simplices by connected sums, vertex foldings, and edge foldings. These results reduce the three- and four-singularity cases to the two-singularity classification by reversing one or two edge foldings (Basak et al., 2022).

A related sharp theorem replaces the condition of relative minimality by an explicit gap to a vertex link. If

$d$27

for some vertex $d$28, and $d$29 has only one singularity or has two singularities with at least one $d$30-singularity, then $d$31 is obtained from boundary complexes of $d$32-simplices by connected sums, bistellar $d$33-moves, edge contractions, edge expansions, vertex foldings, and edge foldings. In the one-singularity case, $d$34 is a handlebody with its boundary coned off (Basak et al., 2021).

Another quantitative formulation uses the average edge order

$d$35

For normal $d$36-pseudomanifolds with singularities, $d$37, with equality if and only if $d$38 is the one-vertex suspension of a triangulation of $d$39 with seven vertices. Moreover, if $d$40, then $d$41 again admits a construction from boundary complexes of $d$42-simplices by connected sums, bistellar $d$43-moves, edge contractions, edge expansions, vertex folding, and edge folding (Basak et al., 2024).

5. Higher-dimensional structure and reconstruction

In dimension $d$44, recent work studies normal $d$45-pseudomanifolds that are $d$46- and $d$47-optimal at a singular vertex $d$48, meaning

$d$49

Under the hypothesis of at most two singularities, one can pass, after facet subdivisions, to an auxiliary normal $d$50-pseudomanifold $d$51 satisfying

$d$52

This relative $d$53-skeleton condition forces missing $d$54-simplices of the form $d$55, which in turn drive decomposition by connected sums and foldings. If there is exactly one singular vertex, then $d$56 is obtained from boundary complexes of $d$57-simplices by vertex foldings and connected sums. If there are exactly two singularities and optimality holds at one of them, then two constructive descriptions are available: either from boundary complexes of $d$58-simplices by one-vertex suspensions, vertex foldings, and connected sums, or from boundary complexes of $d$59-simplices by vertex foldings, edge foldings, and connected sums (Basak et al., 6 May 2025).

Normal pseudomanifolds also appear in reconstruction theory from partial incidence data. For $d$60 and

$d$61

a normal simplicial $d$62-pseudomanifold in which the link of each $d$63-face is a homology $d$64-manifold is determined by the incidences of its $d$65- and $d$66-faces. This extends the corresponding determinability theorem for homology manifolds. The contrast is equally important: for every $d$67, there exist normal, orientable $d$68-pseudomanifolds with isomorphic $d$69-skeleta but nonisomorphic full face posets, so normal pseudomanifolds are not determined by the $d$70-skeleton in general (Hinman, 20 May 2025).

6. Bordism, discrete models, and related classes

Normal pseudomanifolds fit naturally into bordism theories defined by local link conditions. If

$d$71

then the associated class $d$72 is precisely the class of stratified normal pseudomanifolds. In this setting, the oriented stratified and unstratified bordism groups agree,

$d$73

and the corresponding bordism homology theories are naturally isomorphic on all pairs. Normality is stable under cones, products with manifolds, joins, and gluing along collared boundaries. The same framework shows that normality is independent of the Witt and IP conditions: neither implies the other (Friedman, 2013).

Discrete topology supplies a second reinterpretation. For simplicial complexes, a complex is a discrete $d$74-surface if and only if it is a normal $d$75-pseudomanifold without boundary, and it is an $d$76-PCM with non-empty border if and only if it is a normal $d$77-pseudomanifold with boundary. This equivalence makes the connected-link condition the combinatorial core of “no pinches.” It also shows that normality is weaker than smooth or combinatorial manifold conditions in the paper’s sense: a pinched sphere is a pseudomanifold but not a normal pseudomanifold, while a pinched simplicial box is a normal pseudomanifold with boundary that is not smooth (Boutry, 3 Aug 2025).

In the flag category, normal pseudomanifolds admit a purely graph-theoretic reformulation. A graph is a discrete $d$78-pseudomanifold if and only if it is the edge graph of a flag normal $d$79-pseudomanifold. This equivalence leads to sharp small-vertex classifications. Every flag normal $d$80-pseudomanifold with at most $d$81 vertices is a simplicial $d$82-sphere, and for $d$83 every flag normal $d$84-pseudomanifold with at most $d$85 vertices is either a simplicial $d$86-sphere or a flag triangulation of $d$87. The suspended $d$88 family shows that the sphere characterization at $d$89 is optimal (Basak et al., 29 Jun 2026).

These developments collectively delineate the modern role of normal pseudomanifolds. They are rigid enough for sharp inequalities, constructive classifications, and bordism theories driven by local links, yet flexible enough to admit genuine singularities, nonmanifold topologies, and failures of naive reconstruction. In dimension $d$90, the theory is particularly explicit: small $d$91, relative minimality, and low average edge order force strong restrictions on the number and type of singular vertices, often reducing the global structure to connected sums, expansions, and foldings built from simplex boundaries (Basak et al., 2022).