A Geometric Definition of the Integral and Applications

Joshua Lackman

Abstract
The standard definition of integration of differential forms is based on local coordinates and
arXiv:2601.05228v1 [[Link]] 8 Jan 2026

partitions of unity. This definition is mostly a formality and not used used in explicit computations
or approximation schemes. We present a definition of the integral that uses triangulations instead.
Our definition is a coordinate–free version of the standard definition of the Riemann integral on
Rn and we argue that it is the natural definition in the contexts of Lie algebroids, stochastic
integration and quantum field theory, where path integrals are defined using lattices. In particular,
our definition naturally incorporates the different stochastic integrals, which involve integration
over Hölder continuous paths. Furthermore, our definition is well–adapted to establishing integral
identities from their combinatorial counterparts. Our construction is based on the observation
that, in great generality, the things that are integrated are determined by cochains on the pair
groupoid. Abstractly, our definition uses the van Est map to lift a differential form to the pair
groupoid. Our construction suggests a generalization of the fundamental theorem of calculus which
we prove: the singular cohomology and de Rham cohomology cap products of a cocycle with the
fundamental class are equal.

Contents
0 Introduction 2
0.1 Outline of Paper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1 Examples 4
1.1 Riemann-Stieltjes Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Brownian Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Itô and Stratonovich Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.2 Definition of the Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Borel Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Euler Characteristic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5 Gauss-Bonnet Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Cochains 8
2.1 The Nerve of a Groupoid and Cochains . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Simplicial Maps and the Groupoid Differential . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Local Lie Groupids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 The van Est Map 12

3.1 Lie Algebroid Differential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4 The Riemann Sums 13

5 Main Theorem: Convergence and the FTC on Manifolds 16

5.1 Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1
6 Functional Integrals on a Lattice 19
6.1 Feynman’s Path Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6.2 General Functional Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

Appendix 20

A Basic Theory of Lie Groupoids and Lie Algebroids 20

A.1 Lie Groupoids and Lie Algebroids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
A.2 The van Est Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
A.3 Generalized Riemann–Stieltjes Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
A.4 Index of Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

References 24

0 Introduction
In this paper, we present a simple geometric definition of the integral of a differential form, one that
is sometimes better suited for explicit computation and approximation than the standard definition,
and which is better suited for integration over non–differentiable maps (eg. stochastic integration) and
Lie algebroid morphisms. Its original motivation is rooted in lattice constructions of path integrals in
quantum field theory. In particular, our formulation of integration is coordinate–free and partition of
unity free, and more–or–less provides an answer to the question: what is the most general thing that
we can integrate?

The idea is as follows: let ω be a top form on an oriented, compact n–dimensional manifold M
and let ∆M be a triangulation. There should be a Riemann–like definition of the integral given as
Z X
ω = lim Ω(∆) (0.0.1)
M |∆M |→0
∆∈∆M

where Z
Ω(∆) ≈ ω. (0.0.2)
∆
In particular, such a definition would be well–adapted to rigorously establishing integral identities
from their combinatorial counterparts, eg. Stokes’ theorem and the Gauss–Bonnet theorem.

In the elementary case where M = [a, b] and ω = f dx, a triangulation is given by a choice of
points a = x0 < x1 < · · · < xk = b and the standard choices for the approximations are
Ω(xi , xi+1 ) = f (xi )(xi+1 − xi ) , f (xi+1 )(xi+1 − xi ) , (0.0.3)
which result in the left and right Riemann sums, respectively.

In order to generalize Riemann’s construction to manifolds, we make the following observation: the ob-
jects that we can canonically assign a Riemann–like sum to for any triangulation are exactly cochains
on the pair groupoid that are invariant under even permutations. Therefore, in order to obtain a
Riemann–like sum as in 0.0.1, we need to lift ω to a cochain on the local pair groupoid, via the van
Est map VE.

In more basic terms: up to even permutation, an orientation of M determines an orientation of the

(n + 1) vertices of any n–simplex1 ∆ ,→ M. Therefore, we need to choose some function
Ω : M n+1 → R (0.0.4)
1 An orientation of a set of points is an ordering that is defined up to even permutations.

2
that is invariant under even permutations, in which case the Riemann–like sum is obtained by eval-
uating Ω on the oriented vertices of each simplex in ∆M . In order for this Riemann–like sum to
converge to the correct integral, it is required that for each x ∈ M the leading order term of the
Taylor expansion of
Ωx : M n → R , Ωx (x1 , . . . , xn ) = Ω(x, x1 , . . . , xn ) (0.0.5)
at (x, x, . . . , x) is equal to ωx . For stochastic integrals, higher order information is required to integrate.
We show that eq. (0.0.1) holds and prove the following generalization of the fundamental theorem of
calculus:
Theorem 0.0.1. Let (ΩM , Ω∂M ) ∈ C ∞ (M n+1 , R) × C ∞ (∂M n , R) be locally closed2 and antisymmet-
ric. Then Z Z
X X
ΩM (∆) − Ω∂M (∆) = VE(ΩM ) − VE(Ω∂M ) . (0.0.6)
∆∈∆M ∆∈∆∂M M ∂M

The left side is exactly the pairing between the fundamental class and the induced simplicial cocycle.

If Ω∂M = 0 then this result tells us that

X Z
ΩM (∆) = VE(ΩM ) , (0.0.7)
∆∈∆M M

from which we recover the fundamental theorem of calculus by choosing the trivial triangulation of
M = [a, b] and letting3
Ω[a,b] (x, y) = f (y) − f (x) . (0.0.8)
If instead we consider a cohomologically trivial cocycle (ΩM , Ω∂M ), then a combinatorial argument
shows that the left side is zero and we recover Stokes’ theorem.

To get a sense for why this construction of the integral is natural in the context of path integrals
and stochastic integrals, we have the following corollary of theorem 0.0.1, where for a map f : M → N

f∆M : ∆M → N n+1 (0.0.9)

is the induced map, ie. f∆M maps a simplex with vertices (x0 , . . . , xn ) to (f (x0 ), . . . , f (xn )).
Corollary 0.0.2. If Ω is an n–cocycle on Pair Nloc , then for any closed n–dimensional manifold M
(or closed interval) and for any map f : M → N and triangulation ∆M ,4
X Z
∗
f∆ M
Ω(∆) = f ∗ VE(Ω) . (0.0.10)
∆∈∆M M

Therefore, given a closed n–form ω on N, computing a VE–antiderivative for ω reduces the problem
of integrating ω over arbitrary maps f to the problem of determining a triangulation and computing
a finite sum. Note that, the left side makes sense even for non–differentiable f.

This corollary generalizes to the case that TN is replaced by a Lie algebroid. Cocycles and the
van Est map are well–known to those who study Lie groupoids and Lie algebroids.
2 We haven’t defined the differential yet, but (ΩM , Ω∂M ) defines a relative cocycle on the local pair groupoid of M.
3 That is, the triangulation given by x0 = a, x1 = b.
4 The triangulation needs to be small enough so that the image of f
∆M is contained in Pair Nloc .

3
0.1 Outline of Paper
After the introduction, we will give a wide set of examples of naturally appearing cochains and cocycles.
In particular, we will discuss how stochastic integrals fit into this paradigm. We will then move on to
the main body of the text, defining cochains, the van Est map and proving the main results. Finally,
we will discuss applications to lattice constructions of functional integrals. We have included an index
of notation in section A.4.

Acknowledgements
I’d like to thank David Pechersky for discussions about Brownian motion.

1 Examples
The most natural objects to integrate are germs of cochains at the identity of the pair groupoid that
are invariant under even permutations. In other words, these are objects in the set

{Ω : M n+1 → R : Ω is invariant under even permutations}/ ∼ , (1.0.1)

where ∼ identifies two such functions if they agree on a neighborhood of the diagonal.5 In particular,
thinking of the integral this way allows us to integrate differential forms over non–differentiable maps
by first lifting the forms to cochains. We will see such examples in this section.

The reader may want to return to this section after reading more of the text.

1.1 Riemann-Stieltjes Integral

The Riemann-Stieltjes integral
Z 1 n−1
X
f dg := lim f (xi )(g(xi+1 ) − g(xi )) (1.1.1)
0 n→∞
i=0

nicely fits into this paradigm: while dg doesn’t necessarily make sense as a differential form,

(x, y) 7→ f (x)(g(y) − g(x)) (1.1.2)

is a 1-cochain on the pair groupoid.

This suggests the following higher–dimensional generalization: let M be an oriented n–dimensional

manifold and let Ω be a completely antisymmetric (n − 1)–cochain on Pair M, ie. an antisymmetric
map M n → R. Then we can define
Z X
f dΩ := lim f (π1 (∆))δ ∗ Ω(∆) . (1.1.3)
M ∆M
∆∈∆M

Here, πi : M n+1 → M n is the projection that forgets the ith factor and
n+1
X
δ ∗ Ω : M n+1 → R , δ ∗ Ω = (−1)i πi∗ Ω . (1.1.4)
i=1

5 Moregenerally, given a germ of an invariant cochain at the identity of a Lie groupoid, we can integrate it over a
Lie algebroid morphism whose domain is a tangent bundle.

4
Proposition 1.1.1. Suppose that f is continuous and that δ ∗ Ω has bounded variation. Then 1.1.3
exists.
We give a proof of this in section A.3. More generally, we can replace δ ∗ Ω with a cocycle on the local
pair groupoid.

1.2 Brownian Motion

Classically, only the lowest order term of the Taylor expansion contributes to the integral. However,
for stochastic integrals (and path integrals in general) there are higher order terms that contribute.
This is the reason that the Itô and Stratonovich integrals are distinct ([5]). In other words, the things
we can integrate over Wiener paths γ : [0, 1] → R are not one–forms, but are instead things of the
form
f (x) dx + g(x) dx2 ∈ Γ(T∗ R ⊕ T∗ R ⊗ T∗ R) . (1.2.1)
There is a natural action of S2 on this space of sections, given by

f (x) dx + g(x) dx2 → −f (x) dx + (g(x) − f ′ (x)) dx2 , (1.2.2)

and the antisymmetric sections are given by

1
f (x) dx + f ′ (x) dx2 . (1.2.3)
2
The antisymmetric sections satisfy the fundamental theorem of calculus and give the Stratonovich
integral.

1.2.1 Itô and Stratonovich Integrals

To understand the previous discussion, consider a continuous path

γ : [0, 1] → R (1.2.4)

starting at the origin. Suppose that we want to compute

Z 1
γ ∗ (f dx) . (1.2.5)
0

The following approximations correspond to left and right sums for the corresponding Riemann–Stieltjes
integral over [0, 1]:
n−1
X n−1
X
f (γ(ti ))(γ(ti+1 ) − γ(ti )) , f (γ(ti+1 ))(γ(ti+1 ) − γ(ti )) . (1.2.6)
i=0 i=0

These have corresponding 1–cochains6 on Pair R, given by

ΩL (x, y) = f (x)(y − x) , ΩR (x, y) = f (y)(y − x) , (1.2.7)

and the summands of 1.2.6 are 1–cochains on Pair [0, 1] that are obtained by pulling back ΩL , ΩR via
the induced morphism
(γ, γ) : Pair [0, 1] → Pair R . (1.2.8)
If γ is smooth then both of the sums of 1.2.6 converge to 1.2.5. However, the sums in 1.2.6 have
different limits in L2 , with respect to the Wiener measure on paths beginning at the origin; these
6 In the context of the theory of rough paths, functions of simplices were considered in [13].

5
paths are generically only Hölder continuous.7 The reason for the difference is that, for fixed x, the
Taylor expansions at y = x of the 1–cochains
(x, y) 7→ f (x)(y − x) , (x, y) 7→ f (y)(y − x) ∼ f (x) (y − x) + f ′ (x) (y − x)2 + · · · (1.2.9)
differ at order 2, which we could suggestively right as
f (x) dx , f (x) dx + f ′ (x) dx2 . (1.2.10)
In order to integrate 1.2.1 over paths in Wiener space, we just need to choose a 1–cochain whose Taylor
expansion to order two is 1.2.1 and then take an L2 –limit of the corresponding Riemann sums.8 A
special class of such objects are the antisymmetric ones.
Remark 1.2.1. The antisymmetric ones are used in Feynman’s path integral in the presence of a
magnetic potential ([6]), where in a sense we are about to describe the action is really given by
Z 1
1 2 1
S[γ] = γ̇ + V (γ) dt + γ ∗ (A(x) dx + A′ (x) dx2 ) , (1.2.11)
0 2 2
where γ : [0, 1] → R. Note the extra dx2 term that is classically irrelevant.

1.2.2 Definition of the Integrals

We will define the integral of f (x) dx + g(x) dx2 as a random variable, but first we need the following:

Proposition 1.2.2. Let Ω1 , Ω2 : R × R → R be smooth and such that for all x ∈ R the Taylor
expansions of
y 7→ Ω1 (x, y) , y 7→ Ω2 (x, y) (1.2.12)
centered at y = x agree to order 2. Then for j = 1, 2, the random variables
n−1
X
{γ ∈ C([0, 1], R) : γ(0) = 0} → R , γ 7→ lim Ωj (γ(ti ), γ(ti+1 )) (1.2.13)
n→∞
i=0

are equal.9
Proof. This follows from Taylor’s theorem and the fact that the Wiener process has finite quadratic
variation.
Due to this result we can make the following definition:
Definition 1.2.3. Let f (x), g(x) be smooth. We define the random variable
Z 1
γ 7→ γ ∗ (f (x) dx + g(x) dx2 ) (1.2.14)
0

to be equal to
n−1
X
γ 7→ lim Ω(γ(ti ), γ(ti+1 )) , (1.2.15)
n→∞
i=0
where Ω : R × R → R is smooth and such that
1 2
Ω(x, x) = 0 , ∂y Ω(x, y)|y=x = f (x) , ∂ Ω(x, y)|y=x = g(x) , (1.2.16)
2 y
ie. for all x, the second order Taylor expanson of y 7→ Ω(x, y) at y = x is f (x) dx + g(x) dx2 .10
7 With Hölder exponent 1/3 < α < 1/2. The left and right sums do converge to the same result when α > 1/2.
8 This same phenomenon is present in Feynman’s path integral, which consist of the same paths as Wiener space.
9 This is a limit in L2 , with respect to the Wiener measure.
10 We can use a metric to define Taylor expansions on manifolds by splitting the short exact sequence of jet bundles

0 → Symn T ∗ M → J n (M ) → J n−1 (M ) → 0 ([14]).

6
In particular, the Itô and Stratonovich integrals are given by, respectively,
Z 1 Z 1
1
γ ∗ (f dx) , γ ∗ (f dx + f ′ (x) dx2 ) . (1.2.17)
0 0 2

Remark 1.2.4. We could describe this in more invariant language: let ∆ : R ,→ R×R be the diagonal
embedding and
I := sheaf of functions on R × R that vanish on ∆(R) . (1.2.18)
Then ∆∗ (I/I 2 ) is identified with the cotangent sheaf, and we can integrate its sections over smooth
paths. However, over paths in Wiener space we can only integrate sections of ∆∗ (I/I 3 ). In other
words, given
[Ω] ∈ H 0 (∆∗ I/I 3 ) , (1.2.19)
we can integrate it by lifting it to some
Ω ∈ H 0 (I) (1.2.20)
and taking an L2 –limit of Riemann–like sums. The action of S2 is induced by (x, y) → (y, x).

1.3 Borel Measures

Let M be a compact n–dimensional manifold. We can construct a completely symmetric n–cochain
from any finite Borel measure µ: choose a triangulation ∆M of M and consider the set

U ⊂ Pair(n) M = M n+1 (1.3.1)

consisting of (n + 1)-tuples of points in M that are contained in a common simplex of ∆M .11 We can
define

Ω : U → R , Ω(x0 , . . . , xn ) = µ({interior of the convex hull of (x0 , . . . , xn )} . (1.3.2)

We can extend Ω by zero to a small open set containing U.

The cochain we constructed isn’t continuous for a general measure, but it is if the measure is de-
termined by a density.

1.4 Euler Characteristic

As an example of a quantity that isn’t an integral in the standard sense but makes sense as an
integral in our sense, let M be an n-dimensional compact manifold (with boundary) and consider the
symmetric n–cochain

Ω : M n+1 → R , Ω(x0 , . . . , xn ) = (−1)|{x0 ,...,xn }|+1 . (1.4.1)

Then for any triangulation ∆M of M,

X
Ω(∆) = χ(M ) .12 (1.4.2)
∆∈∆M

Taking the limit, it therefore makes sense to say that

Z
Ω = χ(M ) . (1.4.3)
M
11 That is, (x0 , . . . , xn ) ∈ U if there is a simplex containing all of x0 , . . . , xn .
12 Ω is nonzero on degenerate simplices, so the sum includes those too.

7
1.5 Gauss-Bonnet Theorem
The proof of the Gauss-Bonnet theorem involves finding a natural cocycle:

Let (M, g) be an oriented Riemannanian surface (with boundary). We get a degree 2-cocycle (ΩM , Ω∂M )
as follows: let ΩM ∈ Λ2 Pair Mloc be given by

ΩM (x0 , x1 , x2 ) = ±(sum of internal angles of the corresponding geodesic triangle − π) , 13 (1.5.1)

where the sign is chosen according to whether the vectors determined by (x0 , x1 ), (x0 , x2 ) are oriented
or not.

As for Ω∂M ∈ Λ1 Pair ∂Mloc , it is given by

Ω∂M (x0 , x1 ) = ∓(sum of internal angles of the corresponding semicircle) , (1.5.2)

where the semicircle is determined by the arc on the boundary and the geodesic connecting x0 , x1 ,
and the sign is chosen according to whether the vector on ∂M determined by (x0 , x1 ) is oriented or
not (it’s minus if oriented)

The standard counting arguments (eg. at each interior vertex the adjacent angles add up o 2π) show
that the relative Riemann sum is equal to

2π(number of vertices − number of edges + number of faces) (1.5.3)

and theorem 0.0.1 shows that this is exactly equal to

Z Z
VE(ΩM ) − VE(Ω∂M ) . (1.5.4)
M ∂M

This is the Gauss-Bonnet theorem.

2 Cochains
In this section we will define the nerve of a groupoid, followed by completely symmetric/antisymmetric
cochains on groupoids and algebroids, and finally local groupoids. Our description of the nerve of a
groupoid will differ from the traditional one, eg. in [4]. We will describe the basic theory of Lie
groupoids in section .0.

We have included an index of notation at the end of this paper, see A.4.
Remark 2.0.1. The convention we use for wedge products is that dx1 ∧ · · · ∧ dxn equals the antisym-
metrization of dx1 ⊗ · · · ⊗ dxn , eg.
1
dx ∧ dy = (dx ⊗ dy − dy ⊗ dx) . (2.0.1)
2
We do this because to define the integral we partition manifolds into simplices rather than paral-
lelpipeds, and with this definition dx1 ∧ · · · ∧ dxn gives the volume of the standard n-simplex when
evaluated on (∂x1 , . . . , ∂xn ), as opposed to the volume of the parallelpiped.
13 Ω is antisymmetric and defined in a neighborhood of the diagonal in M 3 . This is how the curvature is defined on
M
a simplicial complex in Regge calculus, [12].

8
2.1 The Nerve of a Groupoid and Cochains
We are going to give a definition of the nerve of a groupoid that is slightly different, but equivalent
to the standard definition. Our definition makes the relationships between Lie groupoids and Lie
algebroids clearer, and as a result it makes the van Est map easier to define.
Definition 2.1.1. The nerve of a Lie groupoid G ⇒ M , denoted G(•) , is a simplicial manifold14
which in degree n ≥ 1 is given by the following fiber product:

G(n) = G s ×s G s ×s · · · s ×s G . (2.1.1)
| {z }
n times

We set G(0) = M.

In the context of n-cochains (soon to be defined), we will frequently identify x ∈ M with (id(x), . . . , id(x)) ∈
G(n) ; we call the image of M in G(n) the identity.

The traditional definition of the nerve sets

G(n) = G t ×s G t ×s · · · t ×s G , (2.1.2)
| {z }
n times

which is the space consisting of n arrows which are sequentially composable. These two definitions are
equivalent, with the isomorphism given by

G t ×s · · · t ×s G → G s ×s · · · s ×s G , (g1 , g2 , . . . , gn ) 7→ (g1 , g1 g2 , . . . , g1 g2 · · · gn ) . (2.1.3)

Our definition of G(•) makes it clear that the symmetric group Sn 15 acts on G(n) by permutations.
A little less obvious is that Sn+1 acts on G(n) , which can be seen from the fact that G(n) is naturally
identified the space of morphisms from the standard n–simplex into G. Using our definition of the
nerve, we can explicitly write out this action:
Definition 2.1.2. For σ ∈ Sn+1 and for (g1 , . . . , gn ) ∈ G(n) , we let

σ · (g1 , . . . , gn ) := (gσ−1 −1
−1 (0) gσ(1) , . . . , gσ −1 (0) gσ(n) ) , (2.1.4)

where g0 := id(s(g1 )). The result is a point in G(n) whose common source is t(gσ−1 (0) ).
With this definition, if σ fixes 0 then we get the obvious permutation action of Sn .
Example 2.1.3. Consider the pair groupoid Pair M ⇒ M (defined in A.1.4). We have that

M (n) ∼
= M n+1 = {(x0 , x1 , . . . , xn ) : xi ∈ M } . (2.1.5)

The n arrows are (x0 , x1 ), (x0 , x2 ), . . . , (x0 , xn ); Sn acts by permuting the factors x1 , . . . , xn while
Sn+1 acts by permuting all factors.
The following definition is standard, eg. see [15]:

Definition 2.1.4. An n-cochain is a function Ω : G(n) → R on a groupoid G ⇒ M. It is said to be

normalized if Ω(g1 , . . . , gn ) = 0 whenever gi is an identity for some 1 ≤ i ≤ n. By convention, we
consider 0-cochains to be normalized without any further condition. We denote the sets of n-cochains
and normalized n-cochains by C n (G), C0n (G), respectively.
14 We describe the face and degeneracy maps in the next section.
15 We define Sn to be permutations of the set {0, 1, . . . , n − 1}.

9
Since we have an action of Sn+1 on G(n) , we get an action of Sn+1 on n-cochains Ω by duality, ie.

(σ · Ω)(g1 , . . . , gn ) = Ω(σ −1 · (g1 , . . . , gn )) . (2.1.6)

Definition 2.1.5. A n-cochain Ω is antisymmetric (symmetric) if it is antisymmetric (symmetric)

with respect to the action of Sn .
Normalized antisymmetric cochains behave much like n-forms on the corresponding Lie algebroid (soon
to be defined). However, the following cochains are really the right analogue due to their behaviour
with respect to the groupoid differential and their connection to Stokes’ theorem. In particular, they
are automatically normalized:
Definition 2.1.6. A cochain Ω : G(n) → R is completely antisymmetric if it is antisymmetric with
respect to the action of Sn+1 . We denote the set of completely antisymmetric cochains by Λ(G).
There is a natural graded product on antisymmetric cochains.

Completely antisymmetric cochains vanish on degenerate points in G(n) , ie. points (g1 , . . . , gn ) for
which some gi is an identity or if there are gi , gj such that i ̸= j but gi = gj . This is also true for
normalized cochains which are symmetric (instead of antisymmetric) under Sn+1 , but for cochains
which aren’t normalized we will add an additional condition:
Definition 2.1.7. A cochain Ω : G(n) → R is completely symmetric if it is symmetric with respect to
the action of Sn+1 and if
{s(g1 ), g1 , . . . , gn } = {s(g1′ ), g1′ , . . . , gn′ } =⇒ Ω(g1 , . . . , gn ) = Ω(g1′ , . . . , gn′ ) . (2.1.7)
We denote completely symmetric and normalized completely symmetric n–cochains by S n (G), S0n (G),
respectively.
An n–cochain satisfying eq. (2.1.7) naturally defines an Sk+1 –invariant k-cochain for any k ≤ n, ie.
Ω(g1 , . . . , gk ) := Ω(g1 , . . . , gk , s(g1 ), . . . , s(g1 )), (2.1.8)
where s(g1 ) is repeated (n−k) times. This is important for defining the integral of symmetric cochains.

Most of the paper will emphasize the following cochains, which we will use as the space of primi-
tives of differential forms:
Definition 2.1.8. Let An0 G denote smooth, normalized n-cochains which are invariant under An+1
(even permutations).
Lemma 2.1.9. We have the following decomposition:
An0 G = S0n G ⊕ Λn G . (2.1.9)
Proof. A short computation shows that any Ω ∈ An0 G is a sum of its symmetrization and antisym-
metrization.
The direct sum
An G = S n G ⊕ Λn G (2.1.10)
also plays an important role. The first summand is like the space of signed measures, these don’t need
an orientation to be integrated. The second summand is like the space of top forms, these require an
orientation to be integrated.
Definition 2.1.10. For a Lie algebroid g, we denote pointwise multilinear maps g⊕n → R by C n (g),
and we denote the antisymmetric ones by Γ(Λn g∗ ).

10
2.2 Simplicial Maps and the Groupoid Differential
We complete the construction of the nerve by writing out the face and degeneracy maps as well as
the groupoid differential.

Let G ⇒ M be a Lie groupoid. There are n + 2 face maps

δ0 , . . . , δn+1 : G(n+1) → G(n) , n ≥ 0. (2.2.1)
For n = 0 we have that δ0 = t, δ1 = s. For n ≥ 1 and for k ≥ 1 we have that δk is the projection which
drops the kth arrow, and
δ0 (g1 , . . . , gn+1 ) = (g1−1 g2 , . . . , g1−1 gn+1 ) . (2.2.2)
There are also n + 1 degeneracy maps
σ0 , . . . , σn : G(n) → G(n+1) , n ≥ 0. (2.2.3)
For n = 0 we have σ0 (x) = id(x). For n ≥ 1 we have σ0 (g1 , . . . , gn ) = (s(g1 ), g1 , . . . , gn ),
σk (g1 , . . . , gn ) = (g1 , . . . , gk−1 , s(g1 ), gk , . . . , gn ) , k ≥ 1 . (2.2.4)
The following definition is standard:
Definition 2.2.1. For all n ≥ 0, there is a differential δ ∗ : C n (G) → C n+1 (G) which is given by the
alternating sum of the pullbacks of the face maps, ie.
n
X
δ∗ Ω = (−1)i δi∗ Ω . (2.2.5)
i=0

This differential satisfies δ ∗ ◦ δ ∗ = 0. It restricts to a differential δ ∗ |Λn (G) : Λn (G) → Λn+1 (G) since it
commutes with antisymmetrization of cochains.

2.3 Local Lie Groupids

We will find it useful to make use of local Lie groupoids (see eg. [3]), which is what we get when we
restrict the space of arrows of a groupoid to a neighborhood of the identity bisection. As a result,
not all arrows which are composable in the groupoid are composable in the local groupoid, but ones
which are close enough to the identity bisection are.
Definition 2.3.1. Let G ⇒ M be a Lie groupoid. Let U be a neighborhood of M ⊂ G(1) which is
closed under inversion. We call this a local Lie groupoid and denote it by Gloc ⇒ M.
For the constructions we want to make there isn’t much of a difference between a local Lie groupoid
and a groupoid. We can think of a morphism of local Lie groupoids f : Hloc → Gloc as one which
satisfies f (h1 · h2 ) = f (h1 ) · f (h2 ) whenever the composition on the left side makes sense in Hloc . More
precisely:
Definition 2.3.2. We have a simplicial manifold G•loc given by the largest sub-simplicial manifold of
G• such that G0loc = M, G1loc = U.
By convention, when we speak of the local groupoid we will assume that the fibers of s : U → M are
contractible.

Using this definition of the nerve, all of the structures defined in the previous sections naturally
carry over to the local groupoid. Furthermore, we will be leaving U implicit when we talk about local
groupoids and thus we won’t clearly distinguish between different local Lie groupoids (since all we
really care about is the germ near the identity bisection).
Example 2.3.3. An open cover {Vi }i of M determines a local groupoid Pair Mloc ⇒ M, where
Pair(n) Mloc = {(x0 , . . . , xn ) ∈ M n+1 : {x0 , . . . , xn } ⊂ Vi for some Vi } . (2.3.1)

11
3 The van Est Map
In this section we will define the van Est map VE. Our definition is different but equivalent to the
standard one (up to a constant), and it has the following advantage: the van Est map takes an
n–cochain Ω on G to an n–form VE(Ω) on g, but the standard definition defines

VE(Ω)(X1 , . . . , Xn ) , X1 , . . . Xn ∈ gx (3.0.1)

by first extending X1 , . . . , Xn to sections of g. Our definition doesn’t require these extensions because
of our definition of the nerve. See section A.2 for the standard definition and proof of equivalence.

Recall that a vector ξ ∈ g at a point x ∈ M is a vector tangent to the source fiber of G at x.

Now, given an n-cochain Ω and a point x ∈ M, we can restrict

Ω : G s ×s · · · s ×s G → R (3.0.2)
| {z }
n times

to a map
Ωx : s−1 (x)× · · · ×s−1 (x) → R , (3.0.3)
| {z }
n times

so it makes sense to independently differentiate Ω in each of the n components.

Definition 3.0.1. Let G ⇒ M be a Lie groupoid and g → M its corresponding Lie algebroid. For
each n ≥ 1 we define a (surjective) map

VE : An0 (G) → Γ(Λn g∗ ) , Ω 7→ VE(Ω) (3.0.4)

as follows: for X1 , . . . , Xn ∈ g over x ∈ M, we let

VE(Ω)(X1 , . . . , Xn ) = Xn · · · X1 Ωx , (3.0.5)

where Xi differentiates Ωx in its ith component. If Ω is a 0-cochain we define VE(Ω) = Ω.

Remark 3.0.2. This definition makes sense on Gloc as well, without modification.
In the case that G = Pair M and g = TM, An0 (G) is the set of functions Ω on M n+1 that are invariant
under even permutations and that vanish whenever any two of its arguments are equal. In this case,
given X1 , . . . , Xn ∈ Tx M, VE(Ω)(X1 , . . . , Xn ) is obtained by simultaneously differentiating Ω with
respect to Xi in the ith component, for all 1 ≤ i ≤ n.
Example 3.0.3. Consider the n-cochain on Pair Rn given by
1
Ω(x0 , x1 , . . . , xn ) = det [x1 − x0 , . . . , xn − x0 ] , (3.0.6)
n!
where xi = (x1i , . . . , xni ). Then VE(Ω) = dx10 ∧ · · · ∧ dxn0 (see remark 2.0.1 for our convention for wedge
products).
In any coordinate system on s−1 (x), the nth-order Taylor expansion of Ωx at the identity is deter-
mined by VE(Ω)x . In other words, VE(Ω)x determines the n-jet of Ωx . In particular, the following
lemma is needed to justify our construction of the integral in theorem 5.0.1. First, we set up some
notation:

For a Lie groupoid G ⇒ M, let y = (y 1 , . . . , y n ) be coordinates on s−1 (x) in a neighborhood of

some x ∈ s−1 (x), whose coordinate is x = (x1 , . . . , xn ). This determines a product coordinate sys-
tem on the n-fold product s−1 (x) × · · · × s−1 (x) in a neighborhood of x, written (y1 , . . . , yn ), where
yi = (yi1 , . . . , yin ), 1 ≤ i ≤ n.

12
Lemma 3.0.4. Suppose that the source fibers of G ⇒ M are n-dimensional and that Ω ∈ An0 G. Then
in the product coordinate system (defined in the previous paragraph), we can evaluate at (y1 , . . . , yn )
the nth-order Taylor expansion of Ωx centered at x. The result is

VE(Ω)x (∂y1 , . . . , ∂yn ) det [y1 − x, . . . , yn − x] . (3.0.7)

Here, ∂yj is the coordinate vector field at the point x.

Proof. This follows from the fact that Ω is normalized and invariant under even permutations, which
implies that all lower order terms in the Taylor expansion vanish.
Due to this lemma, Ω evaluated on a tiny linear simplex (in any coordinate system) is approximately
equal to VE(Ω) evaluated on the corresponding simplex.
Remark 3.0.5. The previous lemma suggests that we define a map Jn such that for each x ∈ M
and Ω ∈ Cn (G), Jn (Ω)x is equal to the n–jet of Ωx at x. This is required to fully understand the
construction of the Feynman/Wiener path integral.

3.1 Lie Algebroid Differential

For completeness we define the Lie algebroid differential. There is a standard definition of this, but
keeping in the spirit of emphasizing the groupoid over the algebroid:
Definition 3.1.1. Let G ⇒ M be a Lie groupoid. There is a differential

d : Γ(Λ• g∗ ) → Γ(Λ•+1 g∗ ) (3.1.1)

that is uniquely defined by the property that on Λ• G

VE δ ∗ = d VE . (3.1.2)

Example 3.1.2. If g = T M then d is the exterior derivative.

We will now state a simple version of the van Est isomorphism theorem, proved by Crainic [4]. We
will state a more general one in section 5.
Theorem 3.1.3. (see [4]) Assume the source fibers of G ⇒ M are n-connected. Then VE descends
to an isomorphism between the cohomology of G and the cohomology of g up to degree n, and it is
injective in degree n + 1.
If the Lie algebroid is TM then VE defines an isomorphism between the cohomology of Pair Mloc and
the de Rham cohomology of M. See section 2.2.

4 The Riemann Sums

The construction of the functional integral we would like to make suggests a generalized notion of
Riemann sums on manifolds.16 Recall that Riemann sums on manifolds are only defined locally, not
globally. We will give a global generalization. Before doing so, we will motivate the definition, but first
we recall some basic algebraic topology:
Definition 4.0.1. A smooth triangulation of a compact manifold (with boundary) M is given by a
simplicial complex ∆M , together with a homeomorphism |∆M | → M from the geometric realization to
M such that the restriction of this map to each simplex is a smooth embedding. We denote the set of
smooth triangulations of M by TM .17
16 There are other things called generalized Riemann sums and generalized Riemann integrals in the literature, but

these are different from what we define.

17 We will abuse notation and identify a triangulation with the corresponding simplicial set.

13
For some of the constructions we do, the actual map |∆M | → M doesn’t matter except for on vertices,
and thus we will often leave this map implicit and refer to ∆M as a triangulation. Associated to every
triangulation ∆M is a simplicial set, obtained by picking a total ordering of the vertices. If M is
oriented we should choose an ordering which is compatible with the orientation. We won’t distinguish
∆M from this simplicial set.
Remark 4.0.2. An orientation of M that is given by an orientation of Tx M for all x ∈ M determines
an ordering of the vertices of any top–dimensional simplex ∆ ,→ M, up to even permutation, in the
following way: a choice of vertex x ∈ ∆ determines a basis for Tx M, given by the vectors at x tangent
to the one–dimensional faces of ∆. From the orientation we get an orientation of this basis up to even
permutation, and since every one–dimensional face connects two vertices we also get an ordering of
the vertices up to even permutation.
We will assume standard facts about triangulations of compact manifolds (with boundary), eg. they
always exist and triangulations of the boundary can be extended to the entire manifold. See [11] for
details.

Let M be a manifold and consider an n-cochain Ω on Pair M. Suppose we have an embedding of

the standard n-simplex |∆n | inside M and we want to assign a number to it — we could do this by
choosing an ordering of the vertices (v0 , . . . , vn ) and assigning it the value Ω(v0 , . . . , vn ). However, in
order for this quantity to be independent of the ordering we need Ω to be completely symmetric.

Assuming Ω is completely symmetric, we can assign a value to a given triangulation ∆M by forming

the Riemann–like sum X
Ω(∆) . (4.0.1)
∆∈∆M

Here, we sum over all simplices, making sure to sum over degenerate simplices only once (in the case
that Ω isn’t normalized). We can “integrate” Ω by taking a direct limit over triangulations.

To be precise, we take the limit in the sense of nets, over an equivalence class of triangulations.
We can do this because equivalence classes of triangulations form directed sets:
Definition 4.0.3. Two triangulations are equivalent if they have a common linear subdivision18 (eg.
barycentric subdivisions of a given triangulation). The triangulations in an equivalence class form a
directed set ordered by linear subdivision.
In great generality these limits are independent of the equivalence class chosen, as we will see in the
next section.

On the other hand, if Ω is not completely symmetric but is only invariant under An+1 , then in order
to assign a value to an n-simplex we need a choice of ordering of the vertices, up to even permutation.
An orientation of M induces such an ordering of the vertices of each n-simplex of the triangulation
of M,19 and therefore we can still use 4.0.1.
Definition 4.0.4. Let Ω ∈ An Pair Mloc ,20 where M is compact n-dimensional (with boundary, and
oriented if necessary). We define
Z X
Ω = lim Ω(∆) (4.0.2)
M ∆M ∈TM
∆∈∆M

18 That is, subdivisions of a simplex |∆| ,−

→ |∆M | are induced by linear subdivisions of |∆|.
19 Forexample, by picking a vertex and looking at the orientation of the tangent vectors at that vertex corresponding
to the 1-dimensional faces.
20 Recall the equality An G = S n G ⊕ Λn G.

14
if the limit of 4.0.1 exists over each equivalence class of triangulations and is independent of the
equivalence class.21
In the elementary case that M = [a, b] and ω = f dx, the left and right Riemann sums are obtained,
respectively, by letting
Ω(x, y) = f (x)(y − x) , f (y)(y − x) . (4.0.3)
In either case, VE(Ω) = f dx and therefore we can use these Ω to define the integral of f dx . Similarly,
there is another cochain we can use, as in the following:
Example 4.0.5. Let f : [a, b] → R. Let Ω be the 1-cochain on Pair [a, b] given by

Ω(x, y) = f (y) − f (x) . (4.0.4)

Then for any triangulation a = x0 < x1 < · · · < xn = b we have that

n−1
X
Ω(xi , xi+1 ) = f (b) − f (a) . (4.0.5)
i=0

Therefore, the proof of the FTC is trivial when starting from our definition of the integral of a differ-
ential form, since VE(Ω) = df.
Example 4.0.6. Let M be an n-dimensional manifold and let m ∈ M. We have an n-cochain Ω on
Pair M given by Ω(m0 , . . . , mn ) = 1 if mi = m for all 0 ≤ i ≤ n, and 0 otherwise. Let f : M → R
and let s denote the common source map Pair(n) M → M. We have that, for any triangulation ∆M
having the point m as a vertex, X
s∗ f Ω(∆) = f (m) . (4.0.6)
∆∈∆M

Therefore, the limit is f (m) . Thus, Ω is an n-cochain representing the Dirac measure concentrated at
m.
We have been focusing on cochains that are invariant under even permutations because that is the
minimal property needed to define the integral. However, ones that are actually antisymmetric are
special, due to the following:
Lemma 4.0.7. (Stokes’ Theorem) Let M be a compact oriented (n + 1)-manifold with boundary
and let Ω be a completely antisymmetric n-cochain. Let ∆M be a triangulation of M with induced
triangulation ∆∂M of the boundary. Then
X X
Ω(∆n ) = δ ∗ Ω(∆n+1 ) , (4.0.7)
∆∈∆∂M ∆∈∆M

where δ ∗ is the groupoid differential on Pair M.

Proof. Since Ω is completely antisymmetric, this follows from the fact that an n-dimensional face in
the interior appears as a face exactly twice, with opposite orientations.

Together with the result in the next section, this makes rigorous the standard multivariable calculus
textbook proof of Stokes’ theorem.
21 Recall that equivalence classes of triangulations are directed sets and the limit is taken in the sense of nets. Due to

barycentric subdivision, the size of the simlpices shrinks to zero.

15
5 Main Theorem: Convergence and the FTC on Manifolds
Let ω be an n-form on an oriented n-manifold M (with boundary). Given the discussion in the previous
section, we can assign a Riemann sum to ω by triangulating M and assigning to ω a normalized cochain
that is invariant under even permutations, denoted Ω. The defining property of Ω is that VE(Ω) = ω..
We state a generalization of the fundamental theorem of calculus; we include a part 0. The proofs will
follow.
Theorem 5.0.1. (Part 0) Let M be an oriented compact n-dimensional manifold (with boundary),
let ω be an n-form on M and let
Ω ∈ An0 Pair Mloc (5.0.1)
satisfy VE(Ω) = ω. Then Z
X
lim Ω(∆) = ω, (5.0.2)
∆M ∈TM M
∆∈∆M

where the limit is taken in the sense of 4.0.4.

Theorem 5.0.2. (Part 1) If in addition ∂M = ∅, Ω ∈ Λn Pair M and Ω is closed under δ ∗ , then the
Riemann–like sum approximation is exact for any triangulation, ie.
X Z
Ω(∆) = ω. (5.0.3)
∆∈∆M M

More generally, if (ΩM , Ω∂M ) ∈ Λn Pair (M, ∂M )loc is closed, then

X X Z Z
ΩM (∆) − Ω∂M (∆) = VE(ΩM ) − VE(Ω∂M ) . (5.0.4)
∆∈∆M ∆∈∆∂M M ∂M

Remark 5.0.3. Here, (ΩM , Ω∂M ) is a relative groupoid cocycle, defined analogously to relative de
Rham cocycles: a k-cochain is a pair
(ΩM , Ω∂M ) ∈ Λk Pair Mloc ⊕ Λk−1 Pair ∂Mloc , (5.0.5)
and the differential is given by
(ΩM , Ω∂M ) 7→ (δ ∗ ΩM , i∗ ΩM − δ ∗ Ω∂M ) , (5.0.6)

where i : Pair(k) ∂Mloc → Pair(k) Mloc is the map induced by the inclusion ∂M ,−
→ M.
Part 2 is most generally stated for a Lie groupoid G ⇒ M. It is the van Est isomorphism theorem for
G-modules and is stated and proved in [8]. An even stronger statement involving double groupoids
can be found in [9]. We will state a simple version of it here:
Theorem 5.0.4. (Part 2) Let G be a Lie groupoid and let V be a (real or complex) vector space.
Then VE induces an isomorphism
H ∗ (Gloc , V ) ∼
= H ∗ (g, V ) . (5.0.7)
The following gives a way of computing the integral of a Lie algebroid form over a morphism, and may
be useful for computing some functional integrals. Note that, by Lie’s second theorem a Lie algebroid
morphism x : T M → g integrates to a unique morphism X : Pair Mloc → Gloc .
Corollary 5.0.5. Let M be either a closed manifold or a closed interval, with n = dim M. Let ω be
a closed n–form on g and let Ω ∈ Λn G be closed and satisfy VE(Ω) = ω. Then for any x : T M → g,
X Z
X∗ Ω(∆) = x∗ ω , (5.0.8)
∆∈∆M M

where X : Pair Mloc → Gloc integrates x.

16
Proof. This follows from theorem 5.0.2 and fact that VE, δ ∗ are natural with respect to pullbacks by
morphisms.
Remark 5.0.6. With regards to 5.0.2, 5.0.5, if Ω is a cocycle and ∂M ̸= ∅, then the Riemann–like
sum of Ω won’t depend on the triangulation of the interior, just on the triangulation of the boundary.
Example 5.0.7. We can recover the Poincaré lemma on Rm : we get a primitive for ω by trivializing
the cocycle Ω in the proof of part 2. Explicitly, we define
Z
n−1 m
Ω0 ∈ Λ Pair R , Ω0 (x1 , . . . , xn ) = ω, (5.0.9)
C(0,x1 ,...,xn )

where C(0, x1 , . . . , xn ) is the convex hull of 0, x1 , . . . , xn ∈ Rm . We then have that

d VE(Ω0 ) = ω , (5.0.10)

this follows from the fact that d VE = VE δ ∗ .

Note that, a short computation shows that for any M and any m ∈ M, the map

An Pair M → An−1 Pair M , Ωm (m1 , . . . , mn ) = Ω(m, m1 , . . . , mn ) (5.0.11)

trivializes any cocycle Ω .

5.1 Proofs
We will proves parts 0,1,2.
Proof. (part 0) The result is true if and only if it’s true locally, so we can assume M = [0, 1]n . For
exposition purposes, we will first prove the one–dimensional case: Let f dx be a 1-form and let Ω be
a normalized cochain on Pair [0, 1] such that VE(Ω) = f dx. We then have that the cochain

(x, y) 7→ Ω(x, y) − f (x)(y − x) (5.1.1)

maps to 0 under VE. For a triangulation 0 = x0 < x1 , . . . < xn = 1 we can write

n−1
X n−1
X n−1
X  
Ω(xi , xi+1 ) = f (xi )(xi+1 − xi ) + Ω(xi , xi+1 ) − f (xi )(xi+1 − xi ) . (5.1.2)
i=0 i=0 i=0
R1
The first term on the right converges to 0 f dx, so we just need to show that the second term converges
to 0. From Taylor’s theorem and the fact that Ω(x, x) = 0, we have that for y > x

∂ 
Ω(x, y ′ ) − f (x)(y ′ − x) |y′ =ξx,y

Ω(x, y) − f (x)(y − x) = (y − x) ′
(5.1.3)
∂y

for some ξx,y ∈ (x, y). The condition on VE implies that

∂   y′ →0
′
Ω(x, y ′ ) − f (x)(y ′ − x) −−−→ 0 (5.1.4)
∂y
uniformly, and the result follows.

The general case is essentially the same. Consider a smooth function f : [0, 1]n → R , where [0, 1]n
has coordinates given by (x1 , . . . , xn ) . Let ∆[0,1]n be a triangulation of [0, 1]n . We want to con-
struct Riemann sums associated to the n-form f dx1 ∧ · · · ∧ dxn . First, we antidifferentiate this to

17
Pair[0, 1]n ⇒ [0, 1]n , with the n-cochain given by

Ω(x10 , . . . , xn0 , . . . , x1n , . . . xnn ) = f (x10 , . . . , xn0 ) Vol∆ (x10 , . . . , xn0 , . . . , x1n , . . . xnn ) . (5.1.5)

This cochain is normalized and invariant under even permutations. Taking the limit over all triangu-
lations (using this cochain) give us the desired integral.

Any other normalized cochain Ω′ that is invariant under even permutations and is such that

V E(Ω′ ) = f dx1 ∧ · · · ∧ dxn (5.1.6)

differs from Ω by some normalized cochain Ω0 that is invariant under even permutations and is such
that V E(Ω0 ) = 0 . Let Ω0 be such a cochain. The only thing we need to verify is that
X
lim Ω0 (∆n ) = 0 , (5.1.7)
∆[0,1]n
∆∈∆[0,1]n

which follows by lemma 3.0.4, as in the 1–dimensional case.

Proof. (part 1) This follows by picking any two triangulations of M, pulling back (ΩM , Ω∂M ) to

Pair Mloc × Pair [0, 1] (5.1.8)

via the projection onto Pair Mloc (the pullback of (ΩM , Ω∂M ) will still be closed), extending the two
triangulations of M to a triangulation of

M × [0, 1], (5.1.9)

applying lemma 4.0.7 to deduce that the left side of 5.0.4 is independent of the triangulation and
finally applying theorem 5.0.1.
Proof. (part 2)22 We generalize the construction of the antiderivative in the fundamental theorem
of calculus. First, we choose an identification of

Gloc g
(5.1.10)

M M

which is the identity on M and for which the derivative restricts to the identity map on g ⊂ TG.23

Let ω be a closed n-form on g. For g1 , . . . , gn ∈ Gloc with source x ∈ M, let C(g1 ,...,gn ) be the
convex hull of x, g1 , . . . gn , defined using [Link] This space is naturally oriented by the vectors
(g1 − x, . . . , gn − x), if they are linearly independent. The following is a completely antisymmetric
n-cocycle which maps to ω under VE:
Z
Ω(g1 , . . . , gn ) = ω. (5.1.11)
C(g1 ,...,gn )

Here, we have identified ω with its left translation to Gloc .

22 This construction is not so different from the one found in [3].
23 We’re using the natural identification of V with the tangent space at the origin for a vector space V. Here, V is a
fiber of g. This is called a tubular structure in [3].
24 That is, convex hulls make sense in a vector space.

18
6 Functional Integrals on a Lattice
We’ll first interpret Feynman’s construction of the path integral using cochains and morphisms of
simlpicial sets, and then generalize it.

6.1 Feynman’s Path Integral

To see more of the relationship with path integrals, given a Lagrangian
1
L(x, ẋ) = mẋ2 − V (x) , (6.1.1)
2
the quantum mechanical amplitude of a particle initially located at position xi at time ti to be
measured at position xf at time tf is given by
Z R tf
i
⟨xf , tf |xi , ti ⟩ := Dx e ℏ ti L dt . (6.1.2)
{x:[ti ,tf ]→R: x(ti )=xi ,x(tf )=xf }

Given an initial wave function ψi = ψi (x) at time 0, we can use this amplitude to determine that the
wave function at time t is given by
Z
′
i
Rt
ψ(x, t) = Dx e ℏ 0 L dt ψi (x(t′ )) . (6.1.3)
{x:[0,t]→R: x(0)=x}

Feynman constructed eq. (6.1.3) by triangulating the interval, constructing an approximation to the
path integral, and taking the limit as the spacing goes to zero. These approximations involve sums
which are formally similar to Riemann sums. The result is (we let t = 1)25 :
Z ∞ N N
 
Y 1 i Xm 2 2
ψ(x, 1) = lim CN dxn exp N (xn − xn−1 ) − V (xn ) ψi (xN ) , (6.1.4)
N →∞ −∞ n=1 N ℏ n=1 2

where x0 = x and all integration variables are integrated over (−∞, ∞). By letting ℏ = −i, this
constructs Brownian motion ([1],[5]). The sum is only formally similar to a Riemann sum. However,
the terms of the sum can be constructed from 1–cochains on Pair R, Pair [0, 1], with the condition
that they have the correct Taylor expansionz (or jets). In particular, the term (xn − xn−1 )2 is deter-
mined by the 1–cochain on Pair R given by Ω(x, y) = (y − x)2 . The 3–jet along the source fiber at the
diagonal agrees with dx2 (see footnote 10) and this turns out to be the only important property Ω has.

We make the following important observation, which we will use in the next subsection: there is
a natural identification

{(x0 , x1 , . . . , xN ) : x0 , x1 , . . . , xN ∈ R} ∼
= Hom(∆[0,1] , Pair R) . (6.1.5)

On the left side is the domain of integration in 6.1.4 and on the right side is the set of morphisms
between our triangulation of [0, 1] and the pair groupoid of R. That is, we are approximating the
domain of integration of the path integral by morphisms of simplicial sets. Heuristically,

C([0, 1], R) = lim Hom(∆[0,1] , Pair R) . (6.1.6)

|∆|→0
25 C are constants diverging to ∞ as N → ∞.
N

19
6.2 General Functional Integrals
The framework we develop can be applied to put on a lattice those functional integrals arising from
any classical field theory valued in a tangent bundle or Lie algebroid g — this includes any functional
integral whose domain of integration is a space of maps between manifolds. That is, we are considering
functional integrals whose domain of integration is of the form Hom(TM, g),26 and whose action
functional is given by integration. To put such a functional integral on a lattice, we can:
1. triangulate the domain: M → ∆M
2. integrate g (eg. the tangent bundle) to G (eg. the local pair groupoid): g → G
3. approximate the domain of integration: Hom(TM, g) → Hom(∆M , G).
4. integrate the cochain data to the groupoid, (eg. differential forms to cocycles),
5. form the Riemann–like sums,
6. define a measure on Hom(∆M , G) by using available data (eg. a Riemannian metric, symplectic
form, Haar measure).
This construction produces Feynman’s path integral and Brownian motion in their respective con-
texts, and it can be used to put the Poisson sigma model on a lattice, at least in the cases where the
space of maps can be replaced with the space of Lie algebroid morphisms ([2]). In this case, there
is overlap with the data required to construct a geometric quantization of the Poisson manifold ([7],
[15]); both require a cocycle on the symplectic groupoid.

For the Wiener path integral, the limit of the lattice approximations over triangulations is inde-
pendent of the choice of cochain integrations if one is careful — this is related due to the discussion
in section 1.2. One can generalize the van Est map so that higher order information is recorded via
jets, see section 6.1, remark 3.0.5, footnote 10.
Remark 6.2.1. There are two degenerate cases to consider: one is that g is the zero vector bundle
over a point ∗, and the second is that M = ∗ and g is a tangent bundle. Our construction of the
integral of a differential form is a special case of the former, since
Z Z Z
ω= ω (6.2.1)
M Hom(T M,T ∗) M

and in this case steps 1–6 reduce to steps 1, 4, 5. The integral of a measure is a special case of the
latter since Hom(T ∗, T M ) = M and therefore
Z Z
dµ = dµ . (6.2.2)
M Hom(T ∗,T M )

In this case, steps 1–6 reduce to steps 2, 3, 6.

Appendix
A Basic Theory of Lie Groupoids and Lie Algebroids
In this section we will begin by describing Lie groupoids and Lie algebroids and we will give some
important examples. See [10] for a textbook account.
26 Hom(TM, TN ) is naturally identified with the space of maps M → N.

20
A.1 Lie Groupoids and Lie Algebroids
Definition A.1.1. A groupoid is a category G ⇒ M for which the objects M and arrows G are sets
and for which every morphism is invertible. Notationally, we have two sets M, G with structure maps
of the following form:

s, t : G → M ,
id : M → G ,
· : G t ×s G → G ,
−1
: G → G.

Here s, t are the source and target maps, id is the identity bisection (ie. M can be thought of as
the set of identity arrows inside G), · is the multiplication, denoted (g1 , g2 ) 7→ g1 · g2 , and −1 is the
inversion map. We will frequently identify a point x ∈ M with its image in G under id and write x ∈ G.

A Lie groupoid is a groupoid G ⇒ M such that G, M are smooth manifolds, such that all struc-
ture maps are smooth and such that the source and target maps submersions.
For brevity, we will sometimes denote a (Lie) groupoid G ⇒ M exclusively by its space of arrows G.
Definition A.1.2. A morphism of groupoids G → H is a functor between them, ie. a function which
is compatible with the multiplcations. A morphism of Lie groupoids is a functor which is smooth.
Example A.1.3. Any Lie group G is a Lie groupoid G ⇒ {e} over the manifold containing only the
identity e ∈ G.
The following example is the one most relevant to Brownian motion:
Example A.1.4. Let M be a manifold. The pair groupoid, denoted

Pair(M ) ⇒ M ,

is the Lie groupoid whose objects are the points in M and whose arrows are the points in M × M. An
arrow (x, y) has source and target x, y, respectively. Composition is given by (x, y) · (y, z) = (x, z), the
identity bisection is id(x) = (x, x) and the inversion is (x, y)−1 = (y, x).
The infinitesimal counterpart of a Lie groupoid is a Lie algebroid.
Definition A.1.5. A Lie algebroid is a triple (g, [·, ·], α) consisting of
1. A vector bundle π : g → M ,
2. A vector bundle map (called the anchor map) α : g → T M ,
3. A Lie bracket [·, ·] on the space of sections Γ(g) of π : g → M,
such that for all smooth functions f and all ξ1 , ξ2 ∈ Γ(g) the following Leibniz rule holds: [ξ1 , f ξ2 ] =
(α(ξ1 )f )ξ2 + f [ξ1 , ξ2 ] .
Example A.1.6. Any Lie algebra g is a Lie algebroid g → {0} over the manifold containing only
0 ∈ g.
The following example is the one most relevant to this text.
Example A.1.7. Let M be a manifold. The tangent bundle T M → M is a Lie algebroid, where the
anchor map α is the identity. Sections in Γ(T M ) are just vector fields and the Lie bracket is Lie
bracket of vector fields.

21
A.2 The van Est Map
We now state the standard definition of the van Est map, given by Weinstein–Mu in [16]. The de-
scription of the nerve that they use is

G(n) = G t ×s G t ×s · · · t ×s G . (A.2.1)
| {z }
n times

Let G ⇒ M be a Lie groupoid. Given X ∈ Γ(g) , we can left translate it to a vector field LX on G(1) .
Now suppose that we have an n–cochain Ω, n ≥ 1 . We get an (n − 1)–cochain LX Ω by defining

LX Ω(g1 , . . . , gn−1 ) = LX Ω(g1 , . . . , gn−1 , ·)|t(gn−1 ) , (A.2.2)

ie. we differentiate it in the last component and evaluate it at the identity t(gn−1 ) .

The following definition uses sections of g → M, so it needs to be checked that it is well–defined:

Definition A.2.1. Let Ω be a normalized n-cochain and let X1 , . . . , Xn ∈ gx . We define

X
V E(Ω)(X1 , . . . , Xn ) = sgn(σ)LX̃σ(1) · · · LX̃σ(n) Ω , (A.2.3)
σ∈Sn

where X̃i ∈ Γ(g) is an extension of Xi , 1 ≤ i ≤ n.

The following shows that on normalized cochains that are invariant under even permutations, the
definition of the van Est map we’ve given in the text agrees with the standard definition, up to a
constant.
Proposition A.2.2. Let

f : G t ×s G t ×s · · · t ×s G → G s ×s G s ×s · · · s ×s G ,
| {z } | {z }
n times n times
f (g1 , g2 , . . . , gn ) = (g1 , g1 g2 , . . . , g1 · · · gn ) . (A.2.4)

Then for a normalized cochain

Ω : G s ×s G s ×s · · · s ×s G → R , (A.2.5)
| {z }
n times

we have that X
V E(f ∗ Ω)(X1 , . . . , Xn ) = sgn(σ)Xσ(1) · · · Xσ(n) Ω . (A.2.6)
σ∈Sn

Proof. We only need to extend the vectors to local sections within the corresponding orbit, so we may
assume that the groupoid is transitive. Since the computation is local, we may assume the groupoid
is of the form Pair(X) × H ⇒ X , where H is a Lie group.27 The source and target of (x, y, h) are
given by x, y, respectively, and the composition is given by

(x, y, h) · (y, z, h′ ) = (x, z, hh′ ) . (A.2.7)

The result then follows quickly by working in local coordinates, applying the chain rule and using the
fact that Ω is normalized.
27 Transitive groupoids (ie. groupoids where all objects are isomorphic) are Atiyah groupoids of principal bundles, and

the local triviality of principal bundles implies that transitive groupoids are locally of the aforementioned form. See eg.
[10].

22
A.3 Generalized Riemann–Stieltjes Integral
Here we prove proposition 1.1.1.
Definition A.3.1. Let Ω ∈ An Pair Mloc where M is n-dimensional (and oriented if necessary). We
define its total variation to be28 X
lim sup |Ω| . (A.3.1)
∆M ∈TM
∆∈∆M

Proposition A.3.2. Suppose that f is continuous and that δ ∗ Ω has bounded variation. Then 1.1.3
exists (for the given equivalence class of triangulations used in the total variation).
Proof. The proof is an adaptation of the standard proof for the case of an interval. The result is true
if and only if it’s true on each top dimensional face of our geometric triangulation, so we will assume
M = |∆n |. First, we need to show that if |∆M | ≤ |∆′M |, then
X X
|δ ∗ Ω| ≤ |δ ∗ Ω| . (A.3.2)
∆∈∆M ∆∈∆′M

This follows from lemma 4.0.7 (the triangulated Stokes’ theorem) and the triangle inequality. Next, for
(x0 , . . . , xn ) ∈ Pair(n) M, define V(x0 ,...,xn ) (δ ∗ Ω) to be the total variation of δ ∗ Ω over linear subdivisions
of the convex hull of x0 , . . . , xn ∈ M. We then get a completely symmetric n-cochain V• (δ ∗ Ω) given
by
V• (δ ∗ Ω)(x0 , . . . , xn ) = V(x0 ,...,xn ) (δ ∗ Ω) . (A.3.3)
From A.3.2, it follows that V• (δ ∗ Ω) is additive in the sense that if M is subdivided by the top-
dimensional faces ∆1 , . . . , ∆k , then
k
X
V• (δ ∗ Ω)(M ) = V• (δ ∗ Ω)(∆i ) . (A.3.4)
i=1

The same is true for V• (δ ∗ Ω) − δ ∗ f. We can write the right side of 1.1.3 as
X X
lim (s∗ f )V∆ (δ ∗ Ω) − (s∗ f )(V∆ (δ ∗ Ω) − δ ∗ Ω(∆)) . (A.3.5)
∆M
∆∈∆M ∆∈∆M

Therefore, to show convergence it is enough to show convergence of both terms. This follows by
observing that the following terms go to 0 as we take the limit over triangulations:
X
(sup f − inf f )V∆ (δ ∗ Ω) , (A.3.6)
|∆|
∆∈∆M |∆|
X
(sup f − inf f )(V∆ (δ ∗ Ω) − δ ∗ Ω(∆)) , (A.3.7)
|∆|
∆∈∆M |∆|

where the infimum and supremum are taken over all points in |∆|.

A.4 Index of Notation

1. C n (G), C0n (G) are the subspaces of (normalized) smooth n-cochains on G, 2.1.4
2. Λn G is the subspace of completely antisymmetric n-cochains, 2.1.6
3. S0n G is the subspace of normalized, completely antisymmetric n-cochains, 2.1.7
28 Depending on the cochain, one may have to fix an equivalence class of triangulations in order to talk about its total
variation.

23
 4. An0 G = S0n G⊕Λn G is the subspace of normalized n-cochains invariant under even permutations,
2.1.8
5. Gloc is a local Lie groupoid, 2.3

6. Pair Mloc is the local pair groupoid, 2.3.3, A.1.4

7. δ ∗ , d are the Lie groupoid and Lie algebroid differentials, respectively, 2.2.1, 3.1.1
8. TM is the set of all smooth triangulaions of M and ∆M is a triangulation, 4.0.1.

There is no conflict of interest to report. There is no associated data.

References
[1] Lars-Erik Andersson and Bruce K. Driver. Finite dimensional approximations to Wiener measure
and path integral formulas on manifolds. Journal of Functional Analysis 165 (1998): 430-498.
[2] F. Bonechi, A. S. Cattaneo and M. Zabzine, Geometric quantization and non-perturbative Poisson
sigma model. Adv. Theor. Math. Phys. 10 (2006) 683 [arxiv:math/0507223].
[3] Alejandro Cabrera, Ioan Mǎrcuţ and Marı́a Amelia Salazar. On local integration of Lie brackets.
Journal für die reine und angewandte Mathematik (Crelles Journal), vol. 2020, no. 760, 2020, pp.
267-293. [Link]
[4] Marius Crainic. Differentiable and Algebroid Cohomology, van Est Isomorphisms, and Character-
istic Classes. Commentarii Mathematici Helvetici, Vol.78, (2003) pp. 681-721.
[5] Rick Durrett. Probability: Theory and Examples. (2019).
[6] B. Gaveau, E. Mihóková, M. Roncadelli and L. S. Schulman. Path integral in a magnetic field using
the Trotter product formula. Am. J. Phys. 72, 385–388 (2004). [Link]

[7] Eli Hawkins. A Groupoid Approach to Quantization. J. Symplectic Geom. 6 (2008), no. 1, 61-125.
[8] Joshua [Link] of Lie Groupoid Modules and the Generalized van Est Map. Inter-
national Mathematics Research Notices, rnab027, (2021), [Link]
[9] Joshua Lackman. The van Est Map on Geometric Stacks. arxiv:2205.02109 (2022).

[10] Mackenzie, K. (2005). The Transitive Theory. In General Theory of Lie Groupoids and Lie Alge-
broids,(London Mathematical Society Lecture Note Series, pp. 179-180). Cambridge: Cambridge
University Press.
[11] J. R. Munkres. Elementary differential topology. Lectures given at Massachusetts Institute of
Technology, Fall, 1961. Revised ed, Annals of Mathematics Studies. 54. Princeton, N.J.: Princeton
University Press. MI, 112 p. (1966). ZBL0161.20201.
[12] Tullio Regge. General Relativity Without Coordinates. Il Nuovo Cimento MIM, 558-571, 1961.
[13] E. Stepanov and D. Trevisan. Towards Geometric Integration of Rough Differential Forms. J
Geom Anal 31, 2766–2828 (2021). [Link]

[14] Ravi Vakil. A Beginner’s Guide to Jet Bundles from the Point of View of Algebraic Geometry.
(1998).

24
[15] Alan Weinstein. Symplectic Groupoids, Geometric Quantization, and Irrational Rotation Alge-
bras. In: Dazord, P., Weinstein, A. (eds) Symplectic Geometry, Groupoids, and Integrable Systems.
Mathematical Sciences Research Institute Publications, vol 20, (1991). Springer, New York, NY.

[16] Alan Weinstein and Ping Xu. Extensions of symplectic groupoids and quantization. Journal für
die reine und angewandte Mathematik. Vol. 417, (1991) pp. 159-190.

25