Riemannian Simplicial Complexes

• Riemannian simplicial complexes are defined via intrinsic barycentric maps using Karcher means to triangulate manifolds under precise curvature and size conditions.
• They ensure non-degeneracy through thickness and affine-independence criteria while incorporating discrete analogues of Dirac, Hodge, and connection structures.
• They bridge discrete and continuous geometry by linking metric realizations, spectral convergence, and higher-dimensional Cheeger inequalities in manifold approximation.

Searching arXiv for the cited papers to ground the article in current records. arXiv search query: "Riemannian simplices and triangulations" Riemannian simplicial complexes arise in several closely related senses in the recent literature. In one sense, they are simplicial complexes whose vertices lie in a Riemannian manifold and whose simplices are realized intrinsically by barycentric coordinate maps defined through Karcher means; under explicit curvature, size, and quality conditions, such complexes triangulate the manifold (Dyer et al., 2014). In a second sense, simplicial complexes are equipped with discrete analogues of Riemannian data—Dirac operators, Hodge Laplacians, connection Laplacians, curvature-like quantities, and Gauss–Bonnet-, Poincaré–Hopf-, and Lefschetz-type theorems—which behave functorially under algebraic constructions such as products and disjoint unions (Knill, 2017). Related developments treat simplicial complexes as metric objects built from random variables (Marin, 2017), or as proximity complexes derived from the geodesic metric of a Riemannian manifold, notably Vietoris–Rips and Čech complexes (Adamaszek et al., 2015). A broader survey places these constructions within a discrete-to-continuous program linking simplicial complexes, graphs, triangulations, and Riemannian manifolds through Hodge theory, Morse theory, Laplace-type operators, and Cheeger inequalities (Eidi et al., 4 Dec 2025).

1. Intrinsic geometric simplices in a Riemannian manifold

Let $(M,g)$ be an $n$-dimensional Riemannian manifold, and let $d(\cdot,\cdot)$ denote the Riemannian distance. Given vertices $\{p_0,\dots,p_j\}\subset M$ contained in an open geodesic ball $B_\rho(c)$ whose closure is convex, and weights $\lambda_i\ge 0$ with $\sum_i \lambda_i=1$, the Karcher functional is

$$F_\lambda(x)=\frac12\sum_{i=0}^j \lambda_i\, d(x,p_i)^2.$$

Its minimizer $x_\lambda$, when unique, is the Karcher mean. The gradient formula is

$$\operatorname{grad}F_\lambda(x)=-\sum_{i=0}^j \lambda_i \exp_x^{-1}(p_i),$$

so critical points satisfy

$n$0

Under the radius condition

$n$1

with sectional curvatures bounded above by $n$2, Karcher’s theorem yields strict convexity of $n$3 on $n$4, hence existence and uniqueness of the minimizer (Dyer et al., 2014).

Using barycentric coordinates on the standard Euclidean simplex

$n$5

one obtains the barycentric coordinate map

$n$6

A Riemannian $n$7-simplex $n$8 is defined as the image $n$9. Faces are the images of faces of $d(\cdot,\cdot)$0, and edges are minimizing geodesic segments between the corresponding vertices (Dyer et al., 2014).

Non-degeneracy is the decisive geometric condition. The simplex is non-degenerate if $d(\cdot,\cdot)$1 is a smooth embedding; otherwise it is degenerate. This formulation is intrinsic: the simplex is not defined by ambient Euclidean interpolation, but by a variational center-of-mass construction using the manifold’s own Riemannian distance (Dyer et al., 2014).

2. Non-degeneracy, thickness, and curvature-scale control

The differential structure of the barycentric coordinate map is governed by the vector field

$d(\cdot,\cdot)$2

with Karcher equation $d(\cdot,\cdot)$3. Differentiation gives

$d(\cdot,\cdot)$4

Karcher’s convexity estimate implies that $d(\cdot,\cdot)$5 is positive definite and invertible. The rank of $d(\cdot,\cdot)$6 is therefore controlled by the rank of the matrix of lifted edge vectors $d(\cdot,\cdot)$7, leading to the affine-independence criterion: a Riemannian simplex $d(\cdot,\cdot)$8 is non-degenerate if and only if, for every $d(\cdot,\cdot)$9, the lifted Euclidean simplex

$\{p_0,\dots,p_j\}\subset M$0

is a non-degenerate Euclidean simplex (Dyer et al., 2014).

Quantitative control is expressed through thickness. For a Euclidean $\{p_0,\dots,p_j\}\subset M$1-simplex $\{p_0,\dots,p_j\}\subset M$2 with longest edge $\{p_0,\dots,p_j\}\subset M$3 and altitudes $\{p_0,\dots,p_j\}\subset M$4, the thickness is

$\{p_0,\dots,p_j\}\subset M$5

A simplex is non-degenerate iff $\{p_0,\dots,p_j\}\subset M$6. The paper adapts this quality measure to Riemannian simplices by examining the thickness of liftings in tangent spaces (Dyer et al., 2014).

A common misconception is that positive thickness alone suffices in all dimensions. The surface case behaves this way: in dimension $\{p_0,\dots,p_j\}\subset M$7, a Riemannian triangle is degenerate precisely when its three vertices lie on a common geodesic, and positive thickness suffices as in the Euclidean case. In dimension $\{p_0,\dots,p_j\}\subset M$8, however, non-degeneracy can be guaranteed only when the quality exceeds a positive bound depending on simplex size and curvature. In the simplified criterion stated in the paper, if $\{p_0,\dots,p_j\}\subset M$9, $B_\rho(c)$0 with

$B_\rho(c)$1

and there exists $B_\rho(c)$2 such that

$B_\rho(c)$3

then $B_\rho(c)$4 is non-degenerate (Dyer et al., 2014).

These bounds are supported by Rauch comparison and parallel transport estimates. For $B_\rho(c)$5 with $B_\rho(c)$6,

$B_\rho(c)$7

and for transition maps between tangent spaces one has

$B_\rho(c)$8

for $B_\rho(c)$9, $\lambda_i\ge 0$0. These estimates quantify how curvature distorts lifted simplices and underwrite the non-degeneracy criterion (Dyer et al., 2014).

3. Assembly into simplicial complexes and triangulations

A Riemannian simplicial complex in the intrinsic geometric sense is an abstract simplicial complex $\lambda_i\ge 0$1 with vertex set $\lambda_i\ge 0$2, whose simplices are realized as Riemannian simplices via barycentric coordinate maps. For a vertex $\lambda_i\ge 0$3, its star $\lambda_i\ge 0$4 is the subcomplex consisting of all simplices containing $\lambda_i\ge 0$5, together with all their faces. For triangulation, each $\lambda_i\ge 0$6 is required to be a full star: $\lambda_i\ge 0$7 is a closed topological $\lambda_i\ge 0$8-ball, $\lambda_i\ge 0$9 lies in its interior, and $\sum_i \lambda_i=1$0 has no simplices of dimension larger than $\sum_i \lambda_i=1$1 (Dyer et al., 2014).

The local Euclidean model is obtained by the secant map in normal coordinates. For each $\sum_i \lambda_i=1$2, $\sum_i \lambda_i=1$3 maps the vertices of $\sum_i \lambda_i=1$4 into $\sum_i \lambda_i=1$5, and the paper assumes this realizes $\sum_i \lambda_i=1$6 as a piecewise linear Euclidean complex whose $\sum_i \lambda_i=1$7-simplices have thickness at least $\sum_i \lambda_i=1$8. Together with the size condition

$\sum_i \lambda_i=1$9

and the requirement that all vertices of $$F_\lambda(x)=\frac12\sum_{i=0}^j \lambda_i\, d(x,p_i)^2.$$0 lie in $$F_\lambda(x)=\frac12\sum_{i=0}^j \lambda_i\, d(x,p_i)^2.$$1, one obtains a global triangulation theorem (Dyer et al., 2014).

Under the paper’s assumptions—coverage of $$F_\lambda(x)=\frac12\sum_{i=0}^j \lambda_i\, d(x,p_i)^2.$$2 by the balls $$F_\lambda(x)=\frac12\sum_{i=0}^j \lambda_i\, d(x,p_i)^2.$$3, local Euclidean full-star realizations, and thickness bounds—the barycentric coordinate maps on simplices fit together to define a global smooth map

$$F_\lambda(x)=\frac12\sum_{i=0}^j \lambda_i\, d(x,p_i)^2.$$4

that is a homeomorphism. In particular, all Riemannian simplices defined by the faces of $$F_\lambda(x)=\frac12\sum_{i=0}^j \lambda_i\, d(x,p_i)^2.$$5 are non-degenerate, and $$F_\lambda(x)=\frac12\sum_{i=0}^j \lambda_i\, d(x,p_i)^2.$$6 triangulates $$F_\lambda(x)=\frac12\sum_{i=0}^j \lambda_i\, d(x,p_i)^2.$$7 intrinsically (Dyer et al., 2014).

The compatibility mechanism is canonical. If two adjacent simplices share a face, then the barycentric coordinate constructions agree on the shared face because the vertex sets and barycentric restrictions agree there. This prevents mismatched gluings and provides a simplicial realization of $$F_\lambda(x)=\frac12\sum_{i=0}^j \lambda_i\, d(x,p_i)^2.$$8 in the manifold (Dyer et al., 2014).

The resulting triangulation is quantitatively controlled. The paper states that

$$F_\lambda(x)=\frac12\sum_{i=0}^j \lambda_i\, d(x,p_i)^2.$$9

and for the induced piecewise flat metric $x_\lambda$0 on $x_\lambda$1,

$x_\lambda$2

This suggests a bi-Lipschitz regime when simplices are sufficiently small and well-shaped (Dyer et al., 2014).

The motivation stated in the paper includes mesh generation on manifolds, geometric modeling, and concrete sampling criteria ensuring that a simplicial complex built from sampled points is homeomorphic to the manifold (Dyer et al., 2014).

4. Discrete Dirac, Hodge, and connection structures

A distinct but related notion of Riemannian simplicial structure is developed in the strong ring $x_\lambda$3 generated by finite abstract simplicial complexes. Addition is disjoint union, $x_\lambda$4 with disjoint vertex sets, and multiplication is the set-theoretic Cartesian product

$x_\lambda$5

This product is not a simplicial complex in general, but it still carries a connection graph and supports Dirac-, Hodge-, and connection-Laplacian constructions. The paper states that the strong ring is precisely the largest algebraic setting in which these analytic and topological structures remain well behaved (Knill, 2017).

For a simplicial complex $x_\lambda$6, let $x_\lambda$7 be the space of real functions on $x_\lambda$8-simplices and

$x_\lambda$9

With the exterior derivative $$\operatorname{grad}F_\lambda(x)=-\sum_{i=0}^j \lambda_i \exp_x^{-1}(p_i),$$0 induced by the signed boundary operator, one defines the Dirac operator

$$\operatorname{grad}F_\lambda(x)=-\sum_{i=0}^j \lambda_i \exp_x^{-1}(p_i),$$1

and the Hodge Laplacian

$$\operatorname{grad}F_\lambda(x)=-\sum_{i=0}^j \lambda_i \exp_x^{-1}(p_i),$$2

The discrete Hodge theorem takes the form

$$\operatorname{grad}F_\lambda(x)=-\sum_{i=0}^j \lambda_i \exp_x^{-1}(p_i),$$3

For products, the exterior derivative satisfies

$$\operatorname{grad}F_\lambda(x)=-\sum_{i=0}^j \lambda_i \exp_x^{-1}(p_i),$$4

which yields a discrete Künneth formula and the ring-homomorphic behavior of the Poincaré polynomial $$\operatorname{grad}F_\lambda(x)=-\sum_{i=0}^j \lambda_i \exp_x^{-1}(p_i),$$5 (Knill, 2017).

The connection graph $$\operatorname{grad}F_\lambda(x)=-\sum_{i=0}^j \lambda_i \exp_x^{-1}(p_i),$$6 has simplices of $$\operatorname{grad}F_\lambda(x)=-\sum_{i=0}^j \lambda_i \exp_x^{-1}(p_i),$$7 as vertices, with edges between intersecting simplices. If $$\operatorname{grad}F_\lambda(x)=-\sum_{i=0}^j \lambda_i \exp_x^{-1}(p_i),$$8 is its adjacency matrix, the connection Laplacian is

$$\operatorname{grad}F_\lambda(x)=-\sum_{i=0}^j \lambda_i \exp_x^{-1}(p_i),$$9

The unimodularity theorem states

$n$00

so $n$01 is invertible over $n$02. Its inverse $n$03 plays the role of a Green function, and the energy theorem gives

$n$04

This realizes Euler characteristic as the total potential of the connection Green function (Knill, 2017).

Product behavior sharply separates Hodge and connection geometry. The paper proves

$n$05

in the sense that Hodge eigenvalues add, while

$n$06

so connection spectra multiply. Inductive dimension is also additive under products, and the paper presents this as a discrete analogue of product behavior in Riemannian geometry (Knill, 2017).

Curvature-like quantities appear in several forms. On barycentric refinements $n$07, the sign $n$08 acts as simplicial curvature, and for Whitney complex graphs the vertex curvature is

$n$09

Alternatively, the row sums $n$10 provide a curvature on simplices, again summing to $n$11. Within the strong ring, Gauss–Bonnet, Poincaré–Hopf, and Brouwer–Lefschetz extend to products and signed sums, and Wu characteristics define further ring homomorphisms (Knill, 2017).

A common misunderstanding in this algebraic setting is to identify the strong ring with the full Stanley–Reisner ring. The paper explicitly distinguishes them: many elements of the full Stanley–Reisner ring are not geometric, and for them unimodularity of $n$12 and cohomology via $n$13 can fail. The strong ring is the subring generated by simplicial complexes for which these structures persist (Knill, 2017).

5. Metric realizations by simplicial random variables

Another metric model of simplicial complexes replaces piecewise Euclidean geometry by an $n$14-type space of random variables. Let $n$15 be the vertex set of an abstract simplicial complex $n$16, let $n$17 be a nonatomic standard probability space, and endow $n$18 with the discrete metric of diameter $n$19. The space $n$20 of measurable maps $n$21, modulo equality almost everywhere, carries the metric

$n$22

The simplicial subspace $n$23 consists of those $n$24 whose essential image

$n$25

is a simplex of $n$26 (Marin, 2017).

This realization is generally not complete, so the paper studies its closure $n$27. A point belongs to the closure if and only if every nonempty finite subset of its essential image belongs to $n$28. The inclusion

$n$29

is a weak homotopy equivalence (Marin, 2017).

The comparison with the usual geometric realization is mediated by the probability law map

$n$30

where $n$31 is the usual geometric realization equipped with the $n$32-metric. The paper proves that $n$33 is uniformly continuous and actually $n$34-Lipschitz, extends continuously to completions, is a Serre fibration, admits a continuous global section, and is a weak homotopy equivalence. Consequently, $n$35, $n$36, $n$37, $n$38, and $n$39 all have the same weak homotopy type (Marin, 2017).

The metric geometry here is not Riemannian in the inner-product sense. The paper explicitly characterizes it as an $n$40-type, Finsler-like metric rather than a Hilbert metric. Nonetheless, it provides a canonical metric enrichment of an abstract simplicial complex, with controlled Lipschitz homotopies and fibration properties (Marin, 2017).

This suggests a broadened viewpoint on Riemannian simplicial complexes: metric structure need not come only from Euclidean simplices or triangulated manifolds. A simplicial complex can also be realized as a complete metric space of random variables whose law map recovers the classical barycentric simplex of probability weights (Marin, 2017).

6. Vietoris–Rips and Čech complexes from Riemannian distance

For a metric space $n$41 and threshold $n$42, the Vietoris–Rips complex $n$43 has simplices given by finite subsets of diameter less than, or at most, $n$44; the Čech complex is the nerve of radius-$n$45 metric balls. When $n$46 is a Riemannian manifold equipped with its geodesic metric, these are distance-built simplicial complexes derived directly from Riemannian geometry (Adamaszek et al., 2015).

Hausmann’s theorem states that if $n$47 is a closed Riemannian manifold and $n$48 is sufficiently small, then $n$49. Latschev’s theorem extends this stability to finite metric spaces Gromov–Hausdorff close to $n$50, again in the small-$n$51 regime. For the circle $n$52 with arc-length metric scaled to circumference $n$53, the paper analyzes the full range $n$54, far beyond the regime addressed by Hausmann’s theorem (Adamaszek et al., 2015).

For dense $n$55,

$n$56

For the closed-threshold complex on the full circle,

$n$57

Thus the homotopy types proceed through

$n$58

with wedge singularities at critical values, and eventually become contractible as $n$59 (Adamaszek et al., 2015).

The ambient Čech complexes on $n$60 display the same sequence of odd spheres and wedge singularities, but at shifted thresholds

$n$61

The paper gives an explicit transformation $n$62 and a homotopy equivalence relating Čech complexes to Vietoris–Rips complexes on transformed subsets (Adamaszek et al., 2015).

The principal combinatorial invariant is the winding fraction of the directed Vietoris–Rips graph. For a cyclic graph $n$63,

$n$64

The clique complex of a finite cyclic graph is classified by this invariant: if

$n$65

then $n$66; at the singular values $n$67, wedge decompositions by even spheres appear after dismantling to an appropriate circulant graph (Adamaszek et al., 2015).

This analysis corrects a possible overextension of small-scale intuition. For Riemannian manifolds, the statement “Vietoris–Rips complexes recover the manifold” is only a sufficiently-small-scale result. On the circle, larger scales produce homotopy types of higher odd spheres and singular wedge phases before contractibility. The paper notes that, in topological data analysis, this means higher-dimensional persistence for data sampled from a circle is not automatically noise; it can be systematic and inherent to the construction (Adamaszek et al., 2015).

7. Discrete-to-continuous approximation, Hodge theory, and higher Cheeger geometry

A broader synthesis treats simplicial complexes and graphs as discrete counterparts of Riemannian manifolds. On a Riemannian manifold, the Laplace–Beltrami operator on functions is

$n$68

while on $n$69-forms the Hodge Laplacian is

$n$70

The Hodge theorem gives

$n$71

For a simplicial complex $n$72, the cochain coboundary $n$73 and its adjoint $n$74 define the Eckmann Laplacians

$n$75

with the discrete Hodge theorem

$n$76

The survey presents this as a structural parallel between continuous Hodge theory and its simplicial counterpart (Eidi et al., 4 Dec 2025).

The approximation problem then becomes concrete. In triangulations $n$77 of a Riemannian manifold, with mesh size $n$78, one asks which scalar products on cochains best approximate Riemannian inner products on differential forms, and how eigenvalues and eigenvectors of $n$79 converge to those of $n$80. This is associated in the survey with the Dodziuk–Patodi program. For random geometric graphs built from samples of a manifold, Belkin–Niyogi-type results give convergence of normalized graph Laplacians to the Laplace–Beltrami operator on functions, and Γ-convergence results for graph-based BV functionals approximate manifold BV and Cheeger cuts. The survey emphasizes that higher-order spectral convergence for $n$81 remains largely open (Eidi et al., 4 Dec 2025).

Cheeger theory supplies the main variational bridge. On a compact Riemannian manifold,

$n$82

On graphs, the normalized Laplacian satisfies

$n$83

hence $n$84. The survey extends this landscape to higher-dimensional simplicial complexes by defining several equivalent Cheeger constants $n$85, including chain, cochain, $n$86-Laplacian, and filling-profile formulations, and proving their equivalence (Eidi et al., 4 Dec 2025).

For the normalized up-Laplacian $n$87, the survey states the higher-dimensional Cheeger inequality

$n$88

It also gives dual Cheeger inequalities controlling the top of the spectrum. In particular, for a simplicial complex of top dimension $n$89, disorientability is characterized by the spectral extremum of the up-Laplacian: a simplicial complex is disorientable if and only if $n$90 achieves the maximal possible eigenvalue $n$91 (Eidi et al., 4 Dec 2025).

The survey also places discrete Morse theory within this framework. Forman’s discrete Morse functions on simplicial complexes satisfy inequalities analogous to smooth Morse inequalities, and Forman’s discrete Witten deformation parallels the analytic Witten Laplacian. A further bridge is provided by the Lovász extension on the order complex $n$92: for an injective discrete Morse function $n$93, a simplex $n$94 is critical of index $n$95 if and only if the corresponding vertex $n$96 is a critical point of the Lovász extension $n$97 of index $n$98 in metric, topological, and PL Morse theory. The survey presents this as an exact correspondence between discrete and continuous Morse vectors on a barycentric subdivision-type geometric model (Eidi et al., 4 Dec 2025).

Taken together, these developments treat simplicial complexes as carriers of analytic, spectral, variational, and Morse-theoretic structures that mirror those of Riemannian manifolds. The dominant open direction identified in the survey is to extend the currently strongest convergence theory—from graph Laplacians and $n$99 Cheeger functionals—to higher-order Hodge and Eckmann Laplacians on triangulations and random simplicial complexes approximating a manifold (Eidi et al., 4 Dec 2025).