Front. Math.

China 2016, 11(5): 1345–1362

DOI 10.1007/s11464-016-0575-2

On combinatorial Gauss-Bonnet Theorem

for general Euclidean simplicial complexes
Stephan KLAUS
Scientific Administrator of the MFO and Adjunct Professor at Mainz University,
Mathematisches Forschungsinstitut Oberwolfach gGmbH (MFO), Schwarzwaldstrasse 9-11,
D-77709 Oberwolfach-Walke, Germany


c Higher Education Press and Springer-Verlag Berlin Heidelberg 2016

Abstract For a finitely triangulated closed surface M 2 , let αx be the sum

of angles at a vertex x. By the well-known  combinatorial version of the 2-
dimensional Gauss-Bonnet Theorem, it holds x (2π − αx ) = 2πχ(M 2 ), where
χ denotes the Euler characteristic of M 2 , αx denotes the sum of angles at
the vertex x, and the sum is over all vertices of the triangulation. We give
here an elementary proof of a straightforward higher-dimensional generaliza-
tion to Euclidean simplicial complexes K without assuming any combinatorial
manifold condition. First, we recall some facts on simplicial complexes, the
Euler characteristics and its local version at a vertex. Then we define δ(τ )
as the normed dihedral angle defect around a simplex τ. Our main result
is dim(τ ) δ(τ ) = χ(K), where the sum is over all simplices τ of the
τ (−1)
triangulation. Then we give a definition of curvature κ(x) at a vertex and
we prove the vertex-version x∈K0 κ(x) = χ(K) of this result. It also possible
to prove Morse-type inequalities. Moreover, we can apply this result to
combinatorial (n + 1)-manifolds W with boundary B, where we prove that
the difference of Euler characteristics is given by the sum of curvatures over the
interior of W plus a contribution
 from the normal curvature along the boundary
B : χ(W ) − 2 χ(B) = τ ∈W −B (−1)
1 dim(τ ) δ(τ ) + τ ∈B (−1)dim(τ ) ρ(τ ).
Keywords Curvature, dihedral angle, Euclidean simplex, triangulation, Euler
characteristic, Euler manifold, combinatorial manifold, pseudo manifold
MSC 51M20, 52B70, 53A55, 53C20, 53C23, 55U05, 55U10, 57Q05, 57Q15,
57R05, 57R20

1 Introduction

The generalized Gauss-Bonnet Theorem is one of the most interesting

Received April 8, 2016; accepted June 30, 2016
E-mail: [email protected]
1346 Stephan KLAUS

theorems connecting topology (Euler characteristic) and geometry (curvature)

of manifolds: 
Pf n (Ω) = (2π)n χ(M 2n ),
M

where χ denotes the Euler characteristic of a closed smooth manifold M 2n of

even dimension, and Pf n (Ω) denotes the Pfaffian of the curvature 2-form matrix
of a Riemannian metric g on M.
There is also a combinatorial version for surfaces. For a finitely triangulated
closed surface M 2 , let αx be the sum of angles at a vertex x. Then αx is a
combinatorial measure of the curvature concentrated in x : if αx = 2π, then
M 2 is flat in x; if αx < 2π, then M 2 has positive curvature in x; and if αx > 2π,
then M 2 has negative curvature in x. By the well-known combinatorial version
of the Gauss-Bonnet Theorem [11], it holds

(2π − αx ) = 2πχ(M 2 ),
x

where the sum is over all vertices x of the triangulation.

Allendoerfer and Weil [1] and Chern [9] gave the first proofs of the gen-
eralized Gauss-Bonnet Theorem for closed Riemannian manifolds in arbitrary
(even) dimensions. A very conceptual proof can be found in [15, Appendix C].
We would like to refer to [20] as a concise survey on the Gauss-Bonnet Theorem
and its further generalizations and developments.
Banchoff [2–4] has proved a generalized Gauss-Bonnet Theorem for
combinatorial (triangulated or polyhedral) manifolds using also analytical
methods from the classical Gauss-Bonnet Theorem.
Here, we give a straightforward generalization to arbitrary n-dimensional
Euclidean simplicial complexes where we do not assume any manifold
condition or pureness condition concerning the underlying simplicial complex.
In our proof, we use only simple geometric and combinatorial reasoning. Thus,
this result can be applied not only to combinatorial n-manifolds, but also to
manifolds with boundary, corners, or more general singular structures.
Bloch [6–8] obtained similar, but more complicated results concerning a
stratified Euler characteristics which is not a homotopy invariant in general.
Our results concern the usual Euler characteristics and work with a slightly
different construction which also makes the proofs simpler.

2 A local combinatorial formula for Euler characteristic

We recall some basic definitions and results from combinatorial topology (see
[17], for example).
Let K be a simplicial complex (always assumed to be finite), where we de-
note the set of r-simplices by Kr and its number by cr . The Euler characteristic
Combinatorial Gauss-Bonnet Theorem for general Euclidean simplicial complexes 1347

can be given in this context as


χ(K) = c0 − c1 + c2 − c3 ± · · · = (−1)k ck .
k0

This is an invariant of the homotopy type of K.

An n-dimensional simplicial complex is called pure if its maximal simplices
(with respect to inclusion) all have dimension n.
For a simplex σ, its star St(σ) is the subcomplex spanned by all simplices
that contain σ. The link Lk(σ) of σ is the subcomplex of St(σ) formed by
simplices that do not intersect with σ. The star is the join of σ with its link,

St(σ) = σ ∗ Lk(σ).

Let x ∈ K0 be a vertex and denote by ck (x) the number of k-simplices

in K which contain x. Hence, c0 (x) = 1 and ck (x) is equal to the number of
(k − 1)-simplices of the link Lk(x) of x. More generally, for any simplex σ, we
define cm (σ) as the number of m-simplices which contain σ, and hence,

cm (σ) = cm−k−1 (Lk(σ))

for m > k, where k denotes the dimension of σ. Note that ck (σ) = 1 and
cm (σ) = 0 for m < k.
Definition The characteristic polynomial of a simplicial complex K and its
relative version with respect to a simplex σ are defined as
 
c(t) := ck tk , c(σ; t) := ck (σ)tk .
k0 k0

Note that c(σ; t) starts with tk , where k is the dimension of σ.

The following elementary observation serves as a guide to our geometric
result. It can be found (at least implicitely, but without the naming) in
[12–14,19].
Definition The local Euler density λ(x) ∈ Q of K at the vertex x is defined
as
1 1 1  (−1)k
λ(x) := 1 − c1 (x) + c2 (x) − c3 (x) ± · · · = ck (x).
2 3 4 k+1
k0

Theorem 2.1 For any simplicial complex K, it holds that


λ(x) = χ(K).
x∈K0

Proof As every k-simplex has exactly k + 1 vertices, the sum x ck (x) is equal
to (k + 1)ck , which proves the result. 
1348 Stephan KLAUS

In some sense, this result can be considered as the most elementary and
purely combinatorial version of a generalized Gauss-Bonnet Theorem.
As an example, let K be a pure 1-dimensional simplicial complex. Then
1
λ(x) = 1 − c1 (x),
2
where c1 (x) just gives the number of adjacent vertices. In particular, for a
combinatorial 1-manifold M 1 , it holds that 1  c1 (x)  2 with inner points x
corresponding to c1 (x) = 2 and boundary points corresponding to c1 (x) = 1.
Hence, we get the well-known result that χ(M 1 ) is equal to half the number
of boundary points (which, as a corollary, has to be even) and vanishes for a
closed 1-manifold.
As a more interesting example, let M 2 be a simplicial complex which is a
triangulated closed 2-manifold. Then the link of x is an n(x)-gon with n(x) ∈ N,
and hence,
c1 (x) = n(x) = c2 (x),
which yields
n(x)
λ(x) = 1 − .
6
Hence, vertices with a hexagon as their link do not contribute to the Euler
characteristic whereas the contribution is positive for n-gons with n < 6 (i.e.,
for pentagons, quadrangles, and triangles) and negative for n-gons with n > 6.
See also [11, Chapter 1], where a similar result is stated. Note that the result
is a special case of the combinatorial Gauss-Bonnet Theorem for triangulated
surfaces, where all triangles have the same edge length.
As a well-known application of Poincaré duality, the Euler characteristic of
a closed odd-dimensional manifold vanishes, χ(M 2n+1 ) = 0. It is well known,
[12,19], that this is also true for Euler manifolds, which are defined as follows.
Definition [12] A (closed) Euler-n-manifold M is a pure n-dimensional
simplicial complex such that the link Lk(σ) of each simplex σ has the same
Euler characteristic as the sphere S n−k−1 , where k denotes the dimension of σ,
i.e.,
χ(Lk(σ)) = 1 + (−1)n−k−1 .
The condition of a Euler-manifold is equivalent to

c(σ, −1) = 1 − χ(S n−k−1 ) = (−1)n−k .

M is called a (closed) combinatorial n-manifold if the link of each simplex is

homeomorphic to a sphere, i.e.,

Lk(σ) ≈ S n−k−1 .

Note that pseudo-manifolds are defined differently. In the standard

reference, [18, Chapter 3.8], a closed pseudo-n-manifold is a pure n-dimensional
Combinatorial Gauss-Bonnet Theorem for general Euclidean simplicial complexes 1349

simplicial complex, which has only to satisfy the condition that each (n − 1)-
simplex bounds exactly two n-simplices. This is equivalent to χ(Lk(σ)) = 2
for all (n − 1)-simplices σ. There is also an additional condition for pseudo
manifolds concerning connectivity, which we do not consider here. Hence, our
definition is stronger than the usual definition in the literature. We remark that
odd-dimensional pseudo-manifolds can have non-vanishing Euler characteristic.
As an example, take a combinatorial 3-manifold and identify two points which
are sufficiently distant such that the quotient space still is a simplicial complex.
Then the quotient space is a pseudo-3-manifold with χ = −1.
The definition of a Euler-manifold is still much weaker than the definition
of a combinatorial manifold as the links of simplices are allowed to be homology
spheres, for examples. As a more singular example of a Euler-2-manifold,
consider a triangulated surface and identify some vertices (with the
restriction that the quotient space still has to be a simplicial complex). Then
the link of such a point is a disjoint union of circles and has vanishing Euler
characteristic. Also the suspension of the Poincaré homology 3-sphere (which
exists as combinatorial 3-manifold) is a Euler-4-sphere.
Definition A Euler-n-sphere S is a Euler-n-manifold which satisfies
χ(S) = 1 + (−1)n .
Note that in a Euler-n-manifold M, the link Lk(σ) of any k-simplex σ is a
Euler-(n − k − 1)-sphere. This follows from
Lk (τ ) = Lk(τ ∗ σ),
where τ is a simplex in Lk(σ) and Lk (τ ) denotes the link of τ within Lk(σ).
Moreover, it is straightforward to see that the suspension of a Euler-n-sphere
is a Euler-(n + 1)-sphere.
For a Euler-manifold, there are fundamental relations for the coefficients of
its characteristic polynomial, see [12–14,19].
Theorem 2.2 For a Euler-manifold of even dimension 2m, the following 2m
equations hold:
 
2m + 1
2c1 − 3c2 + 4c3 − 5c4 ± · · · − c2m = 0,
1
 
2m + 1
3c2 − 6c3 + 10c4 ± · · · + c2m = 2c1 ,
2
 
2m + 1
4c3 − 10c4 ± · · · − c2m = 0,
3
 
2m + 1
5c4 ± · · · + c2m = 2c3 ,
4
······ ,
   
2m 2m + 1
c2m−1 − c2m = 0,
2m − 1 2m − 1
1350 Stephan KLAUS
 
2m + 1
c2m = 2c2m−1 .
2m
For a Euler-manifold of odd dimension 2m + 1, the following 2m + 1 equations
hold:  
2m + 2
2c1 − 3c2 + 4c3 − 5c4 ± · · · + c2m+1 = 2c0 ,
1
 
2m + 2
3c2 − 6c3 + 10c4 ± · · · − c2m+1 = 0,
2
 
2m + 2
4c3 − 10c4 ± · · · + c2m+1 = 2c2 ,
3
 
2m + 2
5c4 ± · · · − c2m+1 = 0,
4
······ ,
   
2m + 1 2m + 2
c2m − c2m+1 = 0,
2m 2m
 
2m + 2
c2m+1 = 2c2m .
2m + 1
Proof For the convenience of the reader, we recall the easy proof which can be
found in different forms (sometimes implicitely) in [12], [13], [14], or [19]. We
have
  
i+1
ci (σ) = ci ,
k+1
σ∈Mk

where the sum is over allk-dimensional simplices σ. The reason is that each i-
i+1 
simplex contains exactly k+1 subsimplices of dimension k. Then the equations
follow from

ck (σ) − ck+1 (σ) ± · · · + (−1)n−k−1 cn (σ) = 1 + (−1)n−k−1

and summation over all k-simplices σ. 

We remark that only half of these equations are independent, which can
be seen in low-dimensional cases by inspection, but we will not give here a
proof of this statement. As an immediate consequence, we obtain the following
well-known result which is also stated in [14].
Theorem 2.3 The Euler characteristic of a Euler-manifold of odd dimension
vanishes:
χ(M 2m+1 ) = 0.
Proof (See also [14].) Take the sum of all 2m + 1 fundamental relations. This
gives
2c1 + 2c3 + 2c5 + · · · = 2c0 + 2c2 + 2c4 + · · ·
 
because the alternating sum of all binomial coefficients nk over 0  k  n
vanishes and our sum misses the first and last coefficients 1 and (−1)n . 
Combinatorial Gauss-Bonnet Theorem for general Euclidean simplicial complexes 1351

There is a second proof using the observation that the characteristic

polynomials satisfy
 tk dk c(t) tk+1 dk+1 c(t)
c(σ, t) = + .
k! dtk (k + 1)! dtk+1
σ∈Mk

Then performing a telescopic sum also proves that χ(M 2m+1 ) = 0.

Corollary 2.4 Any Euler-manifold of odd dimension is a Euler-sphere. The
suspension of any odd-dimensional Euler-manifold is a Euler-sphere of even
dimension.
The following result is a remarkable refinement of the vanishing of the Euler
characteristic of an odd-dimensional Euler-manifold. It means that in odd
dimensions, also the local combinatorial curvature vanishes.
Theorem 2.5 The local Euler density of a Euler-manifold of odd dimension
n = 2m + 1 vanishes identically, i.e., λ(x) = 0 at all vertices x.
Proof We show that for any Euler-sphere M 2m of even dimension,
1 1 1 1
c0 − c1 + c2 ∓ · · · + c2m = 1. (2.1)
2 3 4 2m + 2
Then the statement follows as the link of x is a Euler-sphere of even dimension
and λ(x) is given as 1 minus the above series (2.1).
In order to prove (2.1), we will prove a decomposition of formal series
1 1 1 1
c0 − c1 + c2 − c3 ± · · ·
2 3 4 5
= a1 · ( c0 − c1 + c2 − c3 ± · · · )
+ a2 · ( 2 c1 − 3 c2 + 4 c3 ∓ · · · )
+ a3 · ( 3 c2 − 6 c3 ± · · · )
+ a4 · ( 4 c3 ∓ · · · )
+··· (2.2)
with coefficients ai ∈ Q. Clearly, this decomposition exists and is unique because
this is an upper triangular system of equations with non-zero coefficients on
the diagonal. The important point to show is that a1 = 1/2 and that the
higher odd-order coefficients a3 , a5 , . . . vanish. Then the series with coefficient
a1 yields the value 12 χ(M 2m ) = 1 whereas the series with even-order coefficients
a2k do not give a contribution because of the fundamental relations for even-
dimensional Euler-manifolds. Note that the decomposition (2.2) is equivalent
to the following system of equations for the coefficients of ck , k  0 :
k
  
(−1)k k−i k+1
= (−1) ak+1 .
k+2 i
i=0
1352 Stephan KLAUS

Now, we show that there is a (unique) solution for the coefficients ak which
involves Bernoulli numbers and has all necessary properties (i.e., a1 = 1/2 and
higher odd-order coefficients vanish). Recall [10] that the Bernoulli numbers
are given by the exponential generating power series
x  Bn
= xn .
ex −1 n!
n0

In particular,
1 1 1 1
B0 = 1, B1 = − , B2 = , B4 = , B6 = , ...,
2 6 30 42
and odd-order Bernoulli numbers vanish from B3 on. See also [15, Appendix
B] but note that there are different conventions for enumeration and signs of
the Bi . Multiplication of the generating power series with ex − 1 yields the
well-known recursion
k  
k+1
Bi = δ(k)
i
i=0

with δ(0) := 1 and δ(k) = 0 for k > 0. If we separate the first summand to the
left-hand side (i = 0, k > 0) and divide by k + 1, then we obtain with
   
k+1 k+1 k
=
i i i−1
that
 Bi k  
1 k
− = .
k+1 i i−1
i=1
With the index shifts i → i + 1 and k → k + 1, this proves that
Bk
ak = (−1)k
k
solves the decomposition (2.2) and hence proves (2.1). 
The rational number λ(x) can be considered as a kind of local Euler class or
combinatorial curvature measure of K in x. We will give a refinement of λ(x)
with a more geometric interpretation in Section 5.

3 Euclidean simplicial complexes: higher triangle inequalities and

higher angle sums

Definition A Euclidean simplicial complex K is defined as a simplicial

complex with a metric which is a Euclidean (flat) metric on every simplex.
Such a metric is fixed by specifying the edge lengths of all simplices.
Combinatorial Gauss-Bonnet Theorem for general Euclidean simplicial complexes 1353

Thus, we are given an abstract n-dimensional simplicial complex K and

additionally, the edges τ ∈ K1 have a length d(τ ) ∈ R+ .
Recall that the existence of a Euclidean m-simplex σ with prescribed edge
lengths dij , 0  i < j  m, can be decided by certain non-linear inequalities
for the dij , which can be considered as higher triangle inequalities. These
inequalities are given by the Cayley-Menger determinants Γ, see [5, Chapter
9.7]. We recall that Γ with respect to the dij above is defined as the following
determinant:
0 1 1 1 ··· 1

1 0 d201 d202 · · · d20m

1 d210 0 d212 · · · d21m

Γ := 1 d2 d2 0 · · · d22m .
20 21
. . .. .. ..
.. .. . . .

1 d2 d2 d2 · · · 0
m0 m1 m2

Theorem 3.1 (Cayley-Menger, see [5]) Assume that edge lengths dij on an
abstract m-simplex σ are given. Then they are realizable by a unique Euclidean
structure on σ if and only if

(−1)dim(τ ) Γτ > 0

for every subsimplex τ ⊂ σ.

Sketch of Proof For the convenience of the reader, we recall the idea of the
proof from [5]. Let the first vertex v0 be the origin of a Euclidean coordinate
system, and let vi ∈ Rn , 1  i  n, be the coordinate vectors of the remaining
vertices which together form a square matrix V. Then the positive definite Gram
matrix G := V T V has entries vi · vi = d20i and

1 2
vi · vj = (d + d20j − d2ij ).
2 0i
In order to make G symmetric also with respect to the first vertex v0 , one
applies elementary matrix transformations and this yields the Cayley-Menger
determinant:
(−1)n+1
det(G) = Γ.
2n
The reverse statement is similarly proved by an induction on dimension. 
Hence, the function d : K1 → R+ has to satisfy the Cayley-Menger
determinant condition for every simplex τ ∈ K. We call the pair (K, d) of an
abstract simplicial complex with such an edge length function realizable. Thus,
a pair (K, d) defines a Euclidean simplicial complexes, which is then uniquely
defined as a metric space up to isometry, if and only if it is realizable.
Remark Note that this does not imply that such a Euclidean simplicial
complex can be simplex-wise linearly and isometrically embedded into a
1354 Stephan KLAUS

Euclidean space RN of large dimension without subdivision. A counterexample

is given by the abstract simplicial complex ∂Δ2 (i.e., the boundary of a
triangle) with edge lengths 1, 1, and 3. A PL metric embedding of this Euclidean
simplicial complex (which, as a metric space, is just a circle of circumference 5)
is possible only by performing a subdivision.
Definition We follow [11] for the definition of higher dihedral angles. Let
σ ⊂ Rm be a Euclidean m-simplex, and let τ be a k-dimensional subsimplex.
Then the normed dihedral angle at τ with respect to σ is defined as the ratio

vol(Dεm ∩ σ)
θ(τ ⊂ σ) := ∈ [0, 1],
vol(Dεm )

where Dεm denotes a small disk (i.e., with radius ε ≈ 0) around an inner point of
τ and vol denotes the volume. Clearly, this ratio does not depend on the chosen
inner point of τ and on the radius of Dεm if ε is small enough. By definition, it
holds that θ(σ ⊂ σ) = 1 and

1
θ(τ ⊂ σ) =
2
for all codimension-1 subsimplices τ, i.e., (m − 1)-faces of σ. Note that the usual
(non-normed) dihedral angle α is given by multiplication of θ with the volume
vol(S m−k−1 ) of the unit (m − k − 1)-sphere.
Now, there holds the following fundamental theorem of Poincaré which can
be considered as the generalization of the angle sum in a triangle, see [11,
Chapter 1].
Theorem 3.2 (Poincaré, see [11]) Let σ be a Euclidean m-simplex. Then it
holds that 
(−1)dim(τ ) θ(τ ⊂ σ) = 0,
τ

where the sum is over all subsimplices τ of σ.

Sketch of Proof For the convenience of the reader, we recall the idea of the
proof from [11]. Let the first vertex v0 be the origin of an affine coordinate
system x1 , x2 , . . . , xn such that the other vertices are given bythe standard
basis. Then the simplex is given by the equations xi  0 and xi  1. Let
Hi , 1  i  n, be the  corresponding half-spaces of xi  0, and let H0 be the
half-space given by xi  0. Denote by fi , 0  i  n, their characteristic
functions which take the value 1 on the half-space and 0 otherwise. Then
the dihedral angle at the r-subsimplex with vertex indices i0 , i1 , . . . , ir is given
by the normalized integral of the product function fi0 fi1 · · · fir over a small
disc. The statement then follows from the trivial observation that the products
f0 f1 · · · fn and (1 − f0 )(1 − f1 ) · · · (1 − fn ) give zero integrals, and finally by
expanding the last product. 
Combinatorial Gauss-Bonnet Theorem for general Euclidean simplicial complexes 1355

Example In a Euclidean triangle with vertices x0 , x1 , x2 , it holds that

1
θ0 + θ1 + θ2 − 3 ∗ + 1 = 0,
2

which is, after multiplication with vol(S 1 ) = 2π, just the classical angle sum in
a triangle
α0 + α1 + α2 = π.
Moreover, in a Euclidean tetrahedron with vertices x0 , x1 , x2 , x3 , it holds that

1
θ0 + θ1 + θ2 + θ3 − (θ01 + θ02 + · · · + θ23 ) + 4 ∗ − 1 = 0,
2

which is, after multiplication with vol(S 2 ) = 4π, a formula of the French
mathematician De Gua from the 18th century ([11, Section 13]),

α0 + α1 + α2 + α3 − 2(α01 + α02 + · · · + α23 ) + 4π = 0.

Remark We end this section by a remark on the connection between the

edge lengths dij of a Euclidean n-simplex σ and its dihedral angles θ(τ ⊂ σ).
Although a realizable set of dij fixes the geometry of σ, and hence, the θ can
be expressed as functions of the dij , this functional dependence is analytically
very complicated. The range of definition is a semi-algebraic set given
by the Cayley-Menger determinant conditions, but the dihedral angles are
transcendental functions (e.g., involving the dilogarithm already in dimension
4, see [16]) and there are many open questions.

4 Generalized Gauss-Bonnet Theorem for a Euclidean simplicial complex

Definition Now, we consider a Euclidean (i.e., realizable) simplicial complex

(K, d). For any simplex τ of K, we define the normed dihedral angle defect
around τ by

δ(τ ) := 1 − θ(τ ⊂ σ),
σ max

where the sum is over all maximal simplices σ of K which contain τ. By

definition, δ(τ ) is a real number. If τ is maximal, then δ(τ ) = 0 by
definition. Note that we do not assume any manifold condition or pureness
condition concerning the simplicial complex K.
For example, if (K, d) is a triangulation of some bounded region of Rn ,
then, clearly, δ(τ ) = 0 for all inner simplices τ of K. Hence, δ(τ ) ∈ R can be
considered as a rough measure of curvature of (K, d) around τ with δ(τ ) = 0
meaning flatness at τ.
Now, we can formulate our main result.
1356 Stephan KLAUS

Theorem 4.1 For any Euclidean simplicial complex (K, d), it holds that

(−1)dim(τ ) δ(τ ) = χ(K),
τ

where the sum is over all simplices τ of K.

Proof By definition,
    
(−1)dim(τ ) δ(τ ) = (−1)dim(τ ) 1 − θ(τ ⊂ σ)
τ τ σ max
  
= (−1)dim(τ ) − (−1)dim(τ ) θ(τ ⊂ σ),
τ σ max τ

where the second summation is over all maximal σ and we can interchange the
summations on τ and σ by defining θ(τ ⊂ σ) := 0 if τ is not a subsimplex
of σ. The right summand vanishes because of Poincaré’s Theorem and the left
summand just gives χ(K). 
This generalized Gauss-Bonnet Theorem for a Euclidean simplicial complex
expresses the topological invariant χ(K) as a sum of local invariants defined by
geometry. Note that it also makes sense if K has odd dimension, whereas the
classical Gauss-Bonnet-Chern Theorem

Pf n (Ω) = (2π)n χ(M 2n )
M

only makes sense in even dimensions because otherwise the Pfaffian cannot be
defined.

5 Curvature at a vertex

Our result computes the Euler characteristic as a sum of contributions for

each subsimplex τ, hence we call the formula above the simplex-version of the
generalized Gauss-Bonnet Theorem. We give now a localized variant of our
result by the same method that we used in the first section for the local Euler
density λ(x).
Definition We define the curvature at a vertex x ∈ K0 as
1  1  1 
κ(x) := δ(x) − δ(σ 1 ) + δ(σ 2 ) − δ(σ 3 ) ± · · ·
2 1
3 2
4 3
x∈σ x∈σ x∈σ
 (−1)k 
= δ(σ).
k+1
k0 σ∈Kk , x∈σ

This leads to the vertex-version of our generalized Gauss-Bonnet Theorem.

Combinatorial Gauss-Bonnet Theorem for general Euclidean simplicial complexes 1357

Theorem 5.1 For any Euclidean simplicial complex (K, d), it holds that

κ(x) = χ(K),
x∈K0

i.e., the sum over all vertex curvatures gives the Euler characteristic.
Proof This is an immediate consequence of our main result as every k-simplex
τ has (k + 1) vertices which explains that the contribution of (−1)k δ(τ ) to κ(x)
comes with a factor of 1/(k + 1). 

6 Morse-type inequalities

We recall that Morse-type inequalities hold for any chain complex C∗ of finite
type with coefficients over a fixed field. Denote the boundaries by B∗ , the cycles
by Z∗ , and the homology groups by H∗ . Then the short exact sequences

Zm → Cm → Bm−1 , Bm → Zm → Hm ,

yield
cm = hm + bm + bm−1
for the ranks of the corresponding groups. In case of the chain complex of
a simplicial complex, note that the hm are the Betti numbers with respect
to the fixed field of coefficients. Taking alternating sums leads to the Morse
inequalities (m  0) :

hm − hm−1 + hm−2 ∓ · · · + (−1)m h0  cm − cm−1 + cm−2 ∓ · · · + (−1)m c0 .

The difference between both sides is just given by bm .

Theorem 6.1 For any Euclidean simplicial complex (K, d), the following
Morse-type inequalities hold for all m  0 :

hm − hm−1 + hm−2 ∓ · · · + (−1)m h0  δm − δm−1 + δm−2 ∓ · · · + (−1)m δ0 ,

(m) (m) (m) (m)

where
(m)

δi := δ(m) (σ)
σ : dim(σ)=i

is the sum of all normed dihedral angle defects in dimension i and the angle
defects δ(m) (σ) are taken with respect to the m-skeleton Skm (K).
Proof This follows from the Morse inequalities, from the fact that

cm − cm−1 + cm−2 ∓ · · · + (−1)m c0 = (−1)m χ(Skm (K)),

and from our main result applied to Skm (K). 

1358 Stephan KLAUS

7 Application to combinatorial manifolds and pseudo-manifolds

If M is a pure simplicial complex of dimension n, then δ(σ) = 0 for every

n-dimensional simplex σ of M. If M is a closed pseudo-n-manifold, which we
defined in the first chapter as a pure n-complex such that each (n − 1)-
dimensional simplex bounds exactly two n-dimensional simplices, then we also
have δ(σ) = 0 for every (n − 1)-dimensional simplex σ of M. This gives a quick
proof of the following result, well-known in the case of combinatorial manifolds
[2–4].
Corollary 7.1 For a Euclidean simplicial complex (M, d) which is a pseudo-
n-manifold, it holds that

(−1)dim(τ ) δ(τ ) = χ(M n ),
τ : dim(τ )n−2

where the sum is over all simplices τ in the (n − 2)-skeleton of M.

Hence, the full curvature information is singularly concentrated to the
(n − 2)-skeleton of M, which also fits to the fact that the complement of the
(n − 2)-skeleton in a pseudo-manifold, i.e., M − Skn−2 (M ), is a (smooth!)
non-compact flat manifold.
Now, we make the formula explicit in the special case of a Euclidean
combinatorial (pseudo-)manifold M n of low dimension n  5. We obtain the
following formulas, where we recall that χ(M n ) = 0 in odd dimensions n.
Corollary 7.2 For a pseudo-2-manifold, the classical Gauss-Bonnet Theorem
holds: 
δ(x) = χ(M 2 ).
x∈M0

For a χ-pseudo-3-manifold, it holds that

 
δ(x) = δ(τ ).
x∈M0 τ ∈M1

For a pseudo-4-manifold, it holds that

  
δ(x) + δ(τ ) = δ(τ ) + χ(M 4 ).
x∈M0 τ ∈M2 τ ∈M1

For a χ-pseudo-5-manifold, it holds that

   
δ(x) + δ(τ ) = δ(τ ) + δ(τ ).
x∈M0 τ ∈M2 τ ∈M1 τ ∈M3

The general structure in every dimension is clear. In dimension 3, this

means that the sum of dihedral angle defects in dimension 0 equals the sum of
dihedral angle defects in dimension 1. In particular, if one side is positive, the
Combinatorial Gauss-Bonnet Theorem for general Euclidean simplicial complexes 1359

other side also has to be positive. It would be very interesting if this global
result can also be refined to a local result, i.e., if it is possible to find a metric
d such that all dihedral angle defects in dimension 0 and 1 are positive.
In dimension 3, there is the following refinement of the vanishing of χ similar
to the vanishing of the local Euler density λ(x).
Theorem 7.3 Let M 3 be a Euclidean simplicial complex which is a
combinatorial 3-manifold. Then the curvature κ(x) at all vertices vanishes
identically.
Proof We have
      
1
κ(x) = 1 − θ(x ∈ σ) − 1− θ(τ ⊂ σ)
2
σ : x∈σ∈M3 τ : x∈τ ∈M1 σ : τ ⊂σ∈M3
c1 (x)  1 
= 1− − θ(x ∈ σ) + θ(τ ⊂ σ).
2 2
σ : x∈σ∈M3 τ,σ : x∈τ ⊂σ∈M3

Now, we use the fundamental result AΔ = α + β + γ − π for spherical triangles

([5]) in its normalized form:
1 1
θ(x ∈ σ) = (θ(τ1 ⊂ σ) + θ(τ2 ⊂ σ) + θ(τ3 ⊂ σ)) − ,
2 4

 edges in σ ∈ M3 with vertex x. If we insert this

where τi denote the three
formula for the term σ : x∈σ∈M3 θ(x ∈ σ), then the two dihedral angle sums
cancel and we obtain
c1 (x) c3 (x)
κ(x) = 1 − + .
2 4
As Lk(x) is a combinatorial 2-sphere, we have

c1 (x) − c2 (x) + c3 (x) = 2, 2c2 (x) = 3c3 (x),

which shows that

1
c1 (x) − c3 (x) = 2,
2
and hence, κ(x) = 0. 
It would be interesting if this result could be generalized to higher odd
dimensions.

8 Application to combinatorial manifolds with boundary

Now, we consider the case of a Euclidean combinatorial manifold W n+1 with

boundary B n := ∂W n+1 and similar spaces.
A standard example is given by the cylinder W = M × I, where M is a
Euclidean simplicial complex and the interval has a constant length l. Here, the
1360 Stephan KLAUS

prisms σ × I have to be sub-divided into simplices in order to get the structure

of a simplicial complex on the whole cylinder. Note that if we give the vertices
of M an order, then there is a canonical subdivision of prisms. We assume that
this subdivision also respects the metric product structure of the prisms, which
means that the edge length of all subsimplices of M × I are well-defined. Note
that the boundary of the cylinder is

B = ∂(M × I) = M
M,

and thus, has even Euler characteristic.

The same is true in the more general case of a combinatorial manifold W n+1
with boundary B n . This well-known fact can be seen using the double of W :

:= W ∪B W,
W

which gives

) = 2χ(W ) − χ(B).
χ(W
Now, if n is odd, χ(B) = 0, and if n is even, χ(B) = 2χ(W ).
Definition For a Euclidean simplicial complex which is a combinatorial
(pseudo-)manifold W n+1 with boundary B n , we define the normal curvature
at a boundary simplex τ of B as
     
n+1 1 n
ρ(τ ) := 1 − θ(τ ⊂ σ ) − 1− θ(τ ⊂ σ )
2 n
n+1
σ ∈Wn+1 σ ∈Bn
1 1  
= − + θ(τ ⊂ σ n ) − θ(τ ⊂ σ n+1 ),
2 2 n
σ ∈Bn σn+1 ∈Wn+1

where the sum is over the maximal simplices σ n of B and σ n+1 of W,

respectively. Then the normal curvature measures the deviation of the
metric at the boundary from the product metric on the cylinder B × I.
Theorem 8.1 Let B be a Euclidean simplicial complex. In the cylinder B × I
with the product metric, the normal curvature vanishes for all simplices in the
boundary: ρ(τ ) = 0.
Proof Let σ × I be a prism. We have to compare the normed dihedral angle
at the bottom τ ⊂ σ with the dihedral angle with respect to the whole prism
τ ⊂ σ × I. Obviously, the dihedral angle with respect to the whole prism is
half of the dihedral angle at the bottom as it is formed by stabilization (i.e.,
introduction of a further coordinate for I) and intersection with a half-space
concerning this additional coordinate. By definition, ρ(τ ) vanishes. 
Now, we get the generalized Gauss-Bonnet-Chern Theorem for
combinatorial pseudo-manifolds with boundary also being a pseudo-manifold.
Combinatorial Gauss-Bonnet Theorem for general Euclidean simplicial complexes 1361

Theorem 8.2
1  
χ(W ) − χ(B) = (−1)dim(τ ) δ(τ ) + (−1)dim(τ ) ρ(τ ).
2
τ ∈W −B τ ∈B

Proof This is a straightforward computation:

1
χ(W ) − χ(B)
2  
 
= (−1)dim(τ )
1− θ(τ ⊂ σ)
τ ∈W σ∈Wn+1
 
1 
− (−1)dim(τ )
1− θ(τ ⊂ σ)
2
τ ∈B σ∈Bn
   
= (−1) dim(τ )
1− θ(τ ⊂ σ)
τ ∈W −B σ∈Wn+1
  
1 1  
+ (−1)dim(τ ) − + θ(τ ⊂ σ) − θ(τ ⊂ σ) . 
2 2
τ ∈B σ∈Bn σ∈Wn+1

For a cylinder, we have

1
χ(M × I) − χ(∂(M × I)) = 0.
2
Again, the Gauss-Bonnet-Chern formula expresses the topological term χ(W )−
1
2 χ(B) as a sum of local terms defined by geometry, i.e., as the sum of curva-
tures over the interior of W plus a contribution from the normal curvature along
the boundary.

Acknowledgements The author would like to thank Anand Dessai, Fernando Galaz-
Garcia, Ruth Kellerhals, Enrico Leuzinger, and Wilderich Tuschmann for helpful discussion
and comments on this subject.

References
1. Allendoerfer C B, Weil A. The Gauss-Bonnet Theorem for Riemannian polyhedra.
Trans Amer Math Soc, 1943, 53: 101–129
2. Banchoff T F. Critical points and curvature for embedded polyhedra. J Differential
Geom, 1967, 1: 245–256
3. Banchoff T F. Critical points and curvature for embedded polyhedral surfaces. Amer
Math Monthly, 1970, 77: 475–485
4. Banchoff T F. Critical points and curvature for embedded polyhedra. II. In: Differential
Geometry (College Park, Md, 1981/1982). Progr Math, Vol 32. Boston: Birkhäuser,
1983, 34–55
5. Berger M. Geometry I. Berlin: Springer, 1987
6. Bloch E D. The angle defect for arbitrary polyhedra. Beiträge Algebra Geom, 1998,
39: 379–393
1362 Stephan KLAUS

7. Bloch E D. Critical points and the angle defect. Geom Dedicata, 2004, 109: 121–137
8. Bloch E D. The angle defect for odd-dimensional simplicial manifolds. Discrete
Comput Geom, 2006, 35(2): 311–328
9. Chern S S. A simple intrinsic proof of the Gauss-Bonnet formula for closed Riemannian
manifolds, Ann of Math, 1944, 45: 747–752
10. Conway J H, Guy R K. The Book of Numbers. Berlin: Springer, 1996, 107–109
11. Hopf H. Differential Geometry in the Large. Lecture Notes in Math, Vol 1000. Berlin:
Springer, 1989
12. Klee V. A combinatorial analogue of Poincaré’s duality theorem. Canad J Math, 1964,
16: 517–531
13. Levitt N. The Euler characteristic is the unique locally determined numerical homotopy
invariant of finite complexes. Discrete Comput Geom, 1992, 7(1): 59–67
14. MacLaurin C, Robertson G. Euler characteristic in odd dimensions. Austral Math Soc
Gaz, 2003, 30(4): 195–199
15. Milnor J W, Stasheff J D. Characteristic Classes. Ann of Math Stud, Vol 76.
Princeton: Princeton Univ Press, 1974
16. Murakami J. Volume formulas for a spherical tetrahedron. Proc Amer Math Soc, 2012,
140(9): 3289–3295
17. Rourke C, Sanderson B. Introduction to Piecewise-Linear Topology. Berlin: Springer,
1982
18. Spanier E H. Algebraic Topology. Berlin: Springer, 1966
19. Wall C T C. Arithmetic invariants of subdivision of complexes. Canad J Math, 1966,
18: 92–96
20. Wu H -H. Historical development of the Gauss-Bonnet Theorem. Sci China Ser A,
2008, 51(4): 777–784