12 Grassmann algebra and de Rham cohomology

12.1 Differential forms

Definition. Let M be a smooth manifold. A (differential) n-form on M is a (0, n) smooth
tensor field which is totally antisymmetric, i.e.

(X1 , . . . , Xn ) = sgn() (X(1) , . . . , X(n) ),

for any Sn , with Xi (T M ).

Alternatively, we can define a differential form as a smooth section of the appropriate

bundle on M , i.e. as a map assigning to each p M an n-form on the vector space Tp M .
Example 12.1. a) A manifold M is said to be orientable if it admits an oriented atlas,
i.e. an atlas in which all chart transition maps, which are maps between open subsets
of Rdim M , have a positive determinant.
If M is orientable, then there exists a nowhere vanishing top form (n = dim M ) on
M providing the volume.

b) The electromagnetic field strength F is a differential 2-form built from the electric
and magnetic fields, which are also taken to be forms. We will define these later in
some detail.

c) In classical mechanics, if Q is a smooth manifold describing the possible system con-

figurations, then the phase space is T Q. There exists a canonically defined 2-form
on T Q known as a symplectic form, which we will define later.
If is an n-form, then n is said to be the degree of . We denote by n (M ) the
set of all differential n-forms on M , which then becomes a C (M )-module by defining the
addition and multiplication operations pointwise.
Example 12.2. Of course, we have 0 (M ) C (M ) and 1 (M ) (T10 M ) (T M ).
Similarly to the case of forms on vector spaces, we have n (M ) = {0} for n > dim M ,
and otherwise dim n (M ) = dimn M , as a C (M )-module.

We can specialise the pull-back of tensors to differential forms.

Definition. Let : M N be a smooth map and let n (N ). Then we define the

pull-back () n (M ) of as

() : M T M
p 7 ()(p),

where
()(p)(X1 , . . . , Xn ) := ((p)) (X1 ), . . . , (Xn ) ,

for Xi Tp M .

1
 The map : n (N ) n (M ) is R-linear, and its action on 0 (M ) is simply

: 0 (M ) 0 (M )
f 7 (f ) := f .

This works for any smooth map , and it leads to a slight modification of our mantra:

Vectors are pushed forward,

forms are pulled back.

The tensor product does not interact well with forms, since the tensor product of two
forms is not necessarily a form. Hence, we define the following.

Definition. Let M be a smooth manifold. We define the wedge (or exterior ) product of
forms as the map

: n (M ) m (M ) n+m (M )
(, ) 7 ,

where
1 X
( )(X1 , . . . , Xn+m ) := sgn()( )(X(1) , . . . , X(n+m) )
n! m!
Sn+m

and X1 , . . . , Xn+m (T M ). By convention, for any f, g 0 (M ) and n (M ), we set

f g := f g and f = f = f .

Example 12.3. Suppose that , 1 (M ). Then, for any X, Y (T M )

( )(X, Y ) = ( )(X, Y ) ( )(Y, X)

= ( )(X, Y ) (Y )(X)
= ( )(X, Y ) ( )(X, Y )
= ( )(X, Y ).

Hence
= .
The wedge product is bilinear over C (M ), that is

(f 1 + 2 ) = f 1 + 2 ,

for all f C (M ), 1 , 2 n (M ) and m (M ), and similarly for the second argument.

Remark 12.4. If (U, x) is a chart on M , then every n-form n (U ) can be expressed
locally on U as
= a1 an dxa1 dxan ,
where a1 an C (U ) and 1 a1 < < an dim M . The dxai appearing above are
the covector fields (1-forms)
dxai : p 7 dp xai .

2
 The pull-back distributes over the wedge product.

Theorem 12.5. Let : M N be smooth, n (N ) and m (N ). Then, we have

( ) = () ().

Proof. Let p M and X1 , . . . , Xn+m Tp M . Then we have

() () (p)(X1 , . . . , Xn+m )

1 X
sgn() () () (p)(X(1) , . . . , X(n+m) )


=
n! m!
Sn+m
1 X
= sgn() ()(p)(X(1) , . . . , X(n) )
n! m!
Sn+m
()(p)(X(n+1) , . . . , X(n+m) )
1 X

= sgn()((p)) (X(1) ), . . . , (X(n) )
n! m!
Sn+m

((p)) (X(n+1) ), . . . , (X(n+m) )
1 X

= sgn()( )((p)) (X(1) ), . . . , (X(n+m) )
n! m!
Sn+m


= ( )((p)) (X1 ), . . . , (Xn+m )
= ( )(p)(X1 , . . . , Xn+m ).

Since p M was arbitrary, the statement follows.

12.2 The Grassmann algebra

Note that the wedge product takes two differential forms and produces a differential form
of a different type. It would be much nicer to have a space which is closed under the action
of . In fact, such a space exists and it is called the Grassmann algebra of M .

Definition. Let M be a smooth manifold. Define the C (M )-module

dim
MM
Gr(M ) (M ) := n (M ).
n=0

The Grassmann algebra on M is the algebra ((M ), +, , ), where

: (M ) (M ) (M )

is the linear continuation of the previously defined : n (M ) m (M ) n+m (M ).

Recall that the direct sum of modules has the Cartesian product of the modules as
underlying set and module operations defined componentwise. Also, note that by algebra
here we really mean algebra over a module.
Example 12.6. Let = + , where 1 (M ) and 3 (M ). Of course, this + is
neither the addition on 1 (M ) nor the one on 3 (M ), but rather that on (M ) and, in
fact, (M ).

3
 Let n (M ), for some n. Then

= ( + ) = + ,

where n+1 (M ), n+3 (M ), and (M ).

Example 12.7. There is a lot of talk about Grassmann numbers, particularly in supersym-
metry. One often hears that these are numbers that do not commute, but anticommute.
Of course, objects cannot be commutative or anticommutative by themselves. These qual-
ifiers only apply to operations on the objects. In fact, the Grassmann numbers are just the
elements of a Grassmann algebra.
The following result is about the anticommutative behaviour of .

Theorem 12.8. Let n (M ) and m (M ). Then

= (1)nm .

We say that is graded commutative, that is, it satisfies a version of anticommutativity

which depends on the degrees of the forms.

Proof. First note that if , 1 (M ), then

= = .

Recall that is n (M ) and m (M ), then locally on a chart (U, x) we can write

= a1 an dxa1 dxan
= b1 bm dxb1 dxbm

with 1 a1 < < an dim M and similarly for the bi . The coefficients a1 an and
b1 bm are smooth functions in C (U ). Since dxai , dxbj 1 (M ), we have

= a1 an b1 bm dxa1 dxan dxb1 dxbm

= (1)n a1 an b1 bm dxb1 dxa1 dxan dxb2 dxbm
= (1)2n a1 an b1 bm dxb1 dxb2 dxa1 dxan dxb3 dxbm
..
.
= (1)nm a1 an b1 bm dxb1 dxbm dxa1 dxan
= (1)nm

since we have swapped 1-forms nm-many times.

Remark 12.9. We should stress that this is only true when and are pure degree forms,
rather than linear combinations of forms of different degrees. Indeed, if , (M ), a
formula like
=
does not make sense in principle, because the different parts of and can have different
commutation behaviours.

4
12.3 The exterior derivative
Recall the definition of the gradient operator at a point p M . We can extend that
definition to define the (R-linear) operator:

d : C (M )
(T M )
f 7 df

where, of course, df : p 7 dp f . Alternatively, we can think of df as the R-linear map

C (M )
df : (T M )
X 7 df (X) = X(f ).

Remark 12.10. Locally on some chart (U, x) on M , the covector field (or 1-form) df can
be expressed as
df = a dxa
for some smooth functions i C (U ). To determine what they are, we simply apply both
sides to the vector fields induced by the chart. We have
 

df b
= (f ) = b f
x xb

and  
a
a dx = a (xa ) = a ba = b .
xb xb
Hence, the local expression of df on (U, x) is

df = a f dxa .

Note that the operator d satisfies the Leibniz rule

d(f g) = g df + f dg.

We can also understand this as an operator that takes in 0-forms and outputs 1-forms

d : 0 (M )
1 (M ).

This can then be extended to an operator which acts on any n-form. We will need the
following definition.

Definition. Let M be a smooth manifold and let X, Y (T M ). The commutator (or

Lie bracket) of X and Y is defined as

[X, Y ] : C (M )
C (M )
f 7 [X, Y ](f ) := X(Y (f )) Y (X(f )),

where we are using the definition of vector fields as R-linear maps C (M )
C (M ).

5
Definition. The exterior derivative on M is the R-linear operator

d : n (M )
n+1 (M )
7 d

with d being defined as

n+1
X
(1)i+1 Xi (X1 , . . . , X


d(X1 , . . . , Xn+1 ) := ci , . . . , Xn+1 )
i=1
X
(1)i+j [Xi , Xj ], X1 , . . . , X


+ ci , . . . , X
cj , . . . , Xn+1 ,
i<j

where Xi (T M ) and the hat denotes omissions.

Remark 12.11. Note that the operator d is only well-defined when it acts on forms. In order
to define a derivative operator on general tensors we will need to add extra structure to our
differentiable manifold.
Example 12.12. In the case n = 1, the form d 2 (M ) is given by

d(X, Y ) = X((Y )) Y ((X)) ([X, Y ]).

Let us check that this is indeed a 2-form, i.e. an antisymmetric, C (M )-multilinear map

d : (T M ) (T M ) C (M ).

By using the antisymmetry of the Lie bracket, we immediately get

d(X, Y ) = d(Y, X).

Moreover, thanks to this identity, it suffices to check C (M )-linearity in the first argument
only. Additivity is easily checked

d(X1 + X2 , Y ) = (X1 + X2 )((Y )) Y ((X1 + X2 )) ([X1 + X2 , Y ])

= X1 ((Y )) + X2 ((Y )) Y ((X1 ) + (X2 )) ([X1 , Y ] + [X2 , Y ])
= X1 ((Y )) + X2 ((Y )) Y ((X1 )) Y ((X2 )) ([X1 , Y ]) ([X2 , Y ])
= d(X1 , Y ) + d(X2 , Y ).

For C (M )-scaling, first we calculate [f X, Y ]. Let g C (M ). Then

[f X, Y ](g) = f X(Y (g)) Y (f X(g))

= f X(Y (g)) f Y (X(g)) Y (f )X(g)
= f (X(Y (g)) Y (X(g))) Y (f )X(g)
= f [X, Y ](g) Y (f )X(g)
= (f [X, Y ] Y (f )X)(g).

6
Therefore
[f X, Y ] = f [X, Y ] Y (f )X.
Hence, we can calculate

d(f X, Y ) = f X((Y )) Y ((f X)) ([f X, Y ])

= f X((Y )) Y (f (X)) (f [X, Y ] Y (f )X)
= f X((Y )) f Y ((X)) Y (f )(X) f ([X, Y ]) + (Y (f )X)
 
= f X((Y )) f Y ((X))   f ([X, Y ]) + Y (f 
 
Y (f
 )(X)  )(X)


= f d(X, Y ),

which is what we wanted.

The exterior derivative satisfies a graded version of the Leibniz rule with respect to the
wedge product.

Theorem 12.13. Let n (M ) and m (M ). Then

d( ) = d + (1)n d.

Proof. We will work in local coordinates. Let (U, x) be a chart on M and write

= a1 an dxa1 dxan =: A dxA

= b1 bm dxb1 dxbm =: B dxB .

Locally, the exterior derivative operator d acts as

d = dA dxA .

Hence

d( ) = d(A B dxA dxB )

= d(A B ) dxA dxB
= (B dA + A dB ) dxA dxB
= B dA dxA dxB + A dB dxA dxB
= B dA dxA dxB + (1)n A dxA dB dxB
= B d dxB + (1)n A dxA d
= d + (1)n d

since we have anticommuted the 1-form dB through the n-form dxA , picking up n minus
signs in the process.

An important property of the exterior derivative is the following.

Theorem 12.14. Let : M N be smooth. For any n (N ), we have

(d) = d( ()).

7
Proof (sketch). We first show that this holds for 0-forms (i.e. smooth functions).
Let f C (N ), p M and X Tp M . Then

(df )(p)(X) = df ((p))( (X)) (definition of )

= (X)(f ) (definition of df )
= X(f ) (definition of )
= d(f )(p)(X) (definition of d(f ))
= d( (f ))(p)(X) (definition of ),

so that we have (df ) = d( (f )).

The general result follows from the linearity of and the fact that the pull-back
distributes over the wedge product.

Remark 12.15. Informally, we can write this result as d = d , and say that the exterior
derivative commutes with the pull-back.
However, you should bear in mind that the two ds appearing in the statement are
two different operators. On the left hand side, it is d : n (N ) n+1 (N ), while it is
d : n (M ) n+1 (M ) on the right hand side.
Remark 12.16. Of course, we could also combine the operators d into a single operator
acting on the Grassmann algebra on M

d : (M ) (M )

by linear continuation.
Example 12.17. In the modern formulation of Maxwells electrodynamics, the electric and
magnetic fields E and B are taken to be a 1-form and a 2-form on R3 , respectively:

E := Ex dx + Ey dy + Ez dz
B := Bx dy dz + By dz dx + Bz dx dy.

The electromagnetic field strength F is then defined as the 2-form on R4

F := B + E dt.

In components, we can write

F = F dx dx ,
where (dx0 , dx1 , dx2 , dx3 ) (dt, dx, dy, dz) and

0 Ex Ey Ez
E
x 0 Bz By
F =

Ey Bz 0

Bx
Ez By Bx 0

The field strength satisfies the equation

dF = 0.

8
This is called the homogeneous Maxwells equation and it is, in fact, equivalent to the two
homogeneous Maxwells (vectorial) equations

B=0
B
E+ = 0.
t
In order to cast the remaining Maxwells equations into the language of differential forms,
we need a further operation on forms, called the Hodge star operator.
Recall from the standard theory of electrodynamics that the two equations above imply
the existence of the electric and vector potentials and A = (Ax , Ay , Az ), satisfying

B=A
A
E = .
t
Similarly, the equation dF = 0 on R4 implies the existence of an electromagnetic 4-potential
(or gauge potential) form A 1 (R4 ) such that

F = dA.

Indeed, we can take

A := dt + Ax dx + Ay dy + Az dz.

Definition. Let M be a smooth manifold. A 2-form 2 (M ) is said to be a symplectic

form on M if d = 0 and if it is non-degenerate, i.e.

( Y (T M ) : (X, Y ) = 0) X = 0.

A manifold equipped with a symplectic form is called a symplectic manifold.

Example 12.18. In the Hamiltonian formulation of classical mechanics one is especially

interested in the cotangent bundle T Q of some configuration space Q. Similarly to what
we did when we introduced the tangent bundle, we can define (at least locally) a system of
coordinates on T Q by
(q 1 , . . . , q dim Q , p1 , . . . , pdim Q ),
where the pi s are the generalised momenta on Q and the q i s are the generalised coordinates
on Q (recall that dim T Q = 2 dim Q). We can then define a 1-form 1 (T Q) by

:= pi dq i

called the symplectic potential. If we further define

:= d 2 (T Q),

then we can calculate that

d = d(d) = = 0.
Moreover, is non-degenerate and hence it is a symplectic form on T Q.

9
12.4 de Rham cohomology
The last two examples suggest two possible implications. In the electrodynamics example,
we saw that
(dF = 0) ( A : F = dA),
while in the Hamiltonian mechanics example we saw that

( : = d) (d = 0).

Definition. Let M be a smooth manifold and let n (M ). We say that is

closed if d = 0;

exact if n1 (M ) : = d.

The question of whether every closed form is exact and vice versa, i.e. whether the
implications
(d = 0) ( : = d)
hold in general, belongs to the branch of mathematics called cohomology theory, to which
we will now provide an introduction.
The answer for the direction is affirmative thanks to the following result.

Theorem 12.19. Let M be a smooth manifold. The operator

d2 d d : n (M ) n+2 (M )

is identically zero, i.e. d2 = 0.

For the proof, we will need the following concepts.

Definition. Given an object which carries some indices, say Ta1 ,...,an , we define the anti-
symmetrization of Ta1 ,...,an as

1 X
T[a1 an ] := sgn() T(a1 )(an ) .
n!
Sn

Similarly, the symmetrization of Ta1 ,...,an is defined as

1 X
T(a1 an ) := T(a1 )(an ) .
n!
Sn

Some special cases are

1 1
T[ab] = (Tab Tba ), T(ab) = (Tab + Tba )
2 2
1
T[abc] = (Tabc + Tbca + Tcab Tbac Tcba Tacb )
6
1
T(abc) = (Tabc + Tbca + Tcab + Tbac + Tcba + Tacb )
6

10
Of course, we can (anti)symmetrize only some of the indices
1
T ab[cd]e = (T abcde T abdce ).
2
It is easy to check that in a contraction (i.e. a sum), we have

Ta1 an S a1 [ai aj ]an = Ta1 [ai aj ]an S a1 an

and
Ta1 (ai aj )an S a1 [ai aj ]an = 0.

Proof. This can be shown directly using the definition of d. Here, we will instead show it
by working in local coordinates.
Recall that, locally on a chart (U, x), we can write any form n (M ) as

= a1 an dxa1 dxan .

Then, we have

d = da1 an dxa1 dxan

= b a1 an dxb dxa1 dxan ,

and hence
d2 = c b a1 an dxc dxb dxa1 dxan .
Since dxc dxb = dxb dxc , we have

dxc dxb = dx[c dxb] .

Moreover, by Schwarzs theorem, we have c b a1 an = b c a1 an and hence

c b a1 an = (c b) a1 an .

Thus

d2 = c b a1 an dxc dxb dxa1 dxan

= (c b) a1 an dx[c dxb] dxa1 dxan
= 0.

Since this holds for any , we have d2 = 0.

Corollary 12.20. Every exact form is closed.

We can extend the action of d to the zero vector space 0 := {0} by mapping the zero
in 0 to the zero function in 0 (M ). In this way, we obtain the chain of R-linear maps
d d d d d d d d
0 0 (M ) 1 (M ) n (M ) n+1 (M ) dim M (M ) 0,

11
where we now think of the spaces n (M ) as R-vector spaces. Recall from linear algebra
that, given a linear map : V W , one can define the subspace of V

ker() := {v V | (v) = 0},

called the kernel of , and the subspace of W

im() := {(v) | v V },

called the image of .

Going back to our chain of maps, the equation d2 = 0 is equivalent to

im(d : n (M ) n+1 (M )) ker(d : n+1 (M ) n+2 (M ))

for all 0 n dim M 2. Moreover, we have

n (M ) is closed ker(d : n (M ) n+1 (M ))

n (M ) is exact im(d : n1 (M ) n (M )).

The traditional notation for the spaces on the right hand side above is

Z n := ker(d : n (M ) n+1 (M )),

B n := im(d : n1 (M ) n (M )),

so that Z n is the space of closed n-forms and B n is the space of exact n-forms.
Our original question can be restated as: does Z n = B n for all n? We have already
seen that d2 = 0 implies that B n Z n for all n (B n is, in fact, a vector subspace of Z n ).
Unfortunately the equality does not hold in general, but we do have the following result.

Lemma 12.21 (Poincar). Let M Rd be a simply connected domain. Then

Z n = Bn , n > 0.

In the cases where Z n 6= B n , we would like to quantify by how much the closed n-forms
fail to be exact. The answer is provided by the cohomology group.

Definition. Let M be a smooth manifold. The n-th de Rham cohomology group on M is

the quotient R-vector space
H n (M ) := Z n /B n .

You can think of the above quotient as Z n / , where is the equivalence relation

: B n .

The answer to our question as it is addressed in cohomology theory is: every exact n-form
on M is also closed and vice versa if, only if,

H n (M )
=vec 0.

12
Of course, rather than an actual answer, this is yet another restatement of the question.
However, if we are able to determine the spaces H n (M ), then we do get an answer.
A crucial theorem by de Rham states (in more technical terms) that H n (M ) only
depends on the global topology of M . In other words, the cohomology groups are topological
invariants. This is remarkable because H n (M ) is defined in terms of exterior derivatives,
which have everything to do with the local differentiable structure of M , and a given
topological space can be equipped with several inequivalent differentiable structures.
Example 12.22. Let M be any smooth manifold. We have

H 0 (M )
=vec R(# of connected components of M )

since the closed 0-forms are just the locally constant smooth functions on M . As an
immediate consequence, we have

H 0 (R)
=vec H 0 (S 1 )
=vec R.

Example 12.23. By Poincar lemma, we have

H n (M )
=vec 0

for any simply connected M Rd .

13