Lecture 09 - Differential Structures: The Pivotal Concept of Tangent Vector Spaces (Schuller's Geometric Anatomy of Theoretical Physics)
Lecture 09 - Differential Structures: The Pivotal Concept of Tangent Vector Spaces (Schuller's Geometric Anatomy of Theoretical Physics)
Lecture 09 - Differential Structures: The Pivotal Concept of Tangent Vector Spaces (Schuller's Geometric Anatomy of Theoretical Physics)
A routine check shows that this is indeed a vector space. We can similarly define
C (U ), with U an open subset of M .
Note that f is a map R R, hence we can calculate the usual derivative and
evaluate it at 0.
Remark 9.1. In differential geometry, X,p is called the tangent vector to the curve at
the point p M . Intuitively, X,p is the velocity at p. Consider the curve (t) := (2t),
which is the same curve parametrised twice as fast. We have, for any f C (M ):
by using the chain rule. Hence X,p scales like a velocity should.
1
addition
: Tp M Tp M Tp M
(X,p , X,p ) 7 X,p X,p ,
: R Tp M Tp M
(, X,p ) 7 X,p ,
Note that the outputs of these operations do not look like elements in Tp M , because
they are not of the form X,p for some curve . Hence, we need to show that the above
operations are, in fact, well-defined.
Proposition 9.2. Let X,p , X,p Tp M and R. Then, we have X,p X,p Tp M
and X,p Tp M .
Since the derivative is a local concept, it is only the behaviour of curves near p that
matters. In particular, if two curves and agree on a neighbourhood of p, then X,p and
X,p are the same element of Tp M . Hence, we can work locally by using a chart on M .
X,p (f ) := (f )0 (0)
= [f x1 ((x ) + (x ) x(p))]0 (0)
2
where xa , with 1 a d, are the component functions of x, and since the derivative
is linear, we get
ii) The second part is straightforward. Define (t) := (t). This is again a smooth
curve through p and we have:
X,p (f ) := (f )0 (0)
= f 0 ((0)) 0 (0)
= f 0 ((0)) 0 (0)
= (f )0 (0)
:= ( X,p )(f )
Remark 9.3. We now give a slightly different (but equivalent) definition of Tp M . Consider
the set of smooth curves
: (x )0 (0) = (x )0 (0)
for some (and hence every) chart (U, x) containing p. Then, we can define
Tp M := S/ .
: C (M ) C (M ) C (M )
(f, g) 7 f g,
3
The usual qualifiers apply to algebras as well.
i) associative if v, w, z A : v (w z) = (v w) z;
ii) unital if 1 A : v V : 1 v = v 1 = v;
Definition. A Lie algebra A is an algebra whose product [, ], usually called Lie bracket,
satisfies
i) antisymmetry: v A : [v, v] = 0;
ii) the Jacobi identity: v, w, z A : [v, [w, z]] + [w, [z, v]] + [z, [v, w]] = 0.
Note that the zeros above represent the additive identity element in A, not the zero scalar
[v, w] := v w w v,
Remark 9.7. The definition of derivation can be extended to include maps A B, with suit-
able structures. The obvious first attempt would be to consider two algebras (A, +A , A , A ),
(B, +B , B , B ), and require D : A
B to satisfy
4
However, this is meaningless as it stands since B : B B B, but on the right hand side
B acts on elements from A too. In order for this to work, B needs to be a equipped with
a product by elements of A, both from the left and from the right. The structure we are
looking for is called a bimodule over A, and we will meet this later on.
Example 9.8. The usual derivative operator is a derivation on C (R), the algebra of smooth
real functions, since it is linear and satisfies the Leibniz rule.
The second derivative operator, however, is not a derivation on C (R), since it does
not satisfy the Leibniz rule. This shows that the composition of derivations need not be a
derivation.
Example 9.9. Consider again the Lie algebra (End(V ), +, , [, ]) and fix End(V ). If
we define
D := [, ] : End(V )
End(V )
7 [, ],
D ([, ]) := [, [, ]]
= [, [, ]] [, [, ]] (by the Jacobi identity)
= [[, ], ] + [, [, ]] (by antisymmetry)
=: [D (), ] + [, D ()].
= (D1 (D2 (v)) D2 (D1 (v))) w + v (D1 (D2 (w) D2 (D1 (w)))
= [D1 , D2 ](v) w + v [D1 , D2 ](w)
5
If we have a manifold, we can define the related notion of derivation on an open subset
at a point.
Definition. Let M be a manifold and let p U M , where U is open. A derivation on
U at p is an R-linear map D : C (U )
R satisfying the Leibniz rule
Tp M := Derp (U ),
for some open U containing p. One can show that this does not depend on which neigh-
bourhood U of p we pick.
dim Tp M = dim M.
Remark 9.13. Note carefully that, despite us using the same symbol, the two dimensions
appearing in the statement of the theorem are entirely unrelated. Indeed, recall that dim M
is defined in terms of charts (U, x), with x : U x(U ) Rdim M , while dim Tp M = |B|,
where B is a Hamel basis for the vector space Tp M . The idea behind the proof is to
construct a basis of Tp M from a chart on M .
Proof. W.l.o.g., let (U, x) be a chart centred at p, i.e. x(p) = 0 Rdim M . Define (dim M )-
many curves (a) : R U through p by requiring (xb (a) )(t) = ab t, i.e.
(a) (0) := p
(a) (t) := x1 (0, . . . , 0, t, 0, . . . , 0)
where the t is in the ath position, with 1 a dim M . Let us calculate the action of the
tangent vector X(a) ,p Tp M on an arbitrary function f C (U ):
6
We introduce a special notation for this tangent vector:
:= X(a) ,p ,
xa p
X(f ) = X,p (f )
:= (f )0 (0)
= (f x1 x )0 (0)
= [b (f x1 )(x(p))] (xb )0 (0)
b 0
= (x ) (0) (f ).
xb p
for some scalars a . Note that this is an operator equation, and the zero on the right hand
side is the zero operator 0 Tp M .
Recall that, given the chart (U, x), the coordinate maps xb : U R are smooth, i.e.
x C (U ). Thus, we can feed them into the left hand side to obtain
b
a
0= (xb )
xa p
= a a (xb x1 )(x(p))
= a a (projb )(x(p))
= a ab
= b
7
Remark 9.14. While it is possible to define infinite-dimensional manifolds, in this course we
will only consider finite-dimensional ones. Hence dim Tp M = dim M will always be finite
in this course.
Remark 9.15. Note that the basis that we have constructed in the proof is not chart-
independent. Indeed, each different chart will induce a different tangent space basis, and
we distinguish between them by keeping the chart map in the notation for the basis elements.
This is not a cause of concern for our proof however, since every basis of a vector space
must have the same cardinality, and hence it suffices to find one basis to determine the
dimension.
Remark 9.16. While the symbol x a p has nothing to do with the idea of partial differen-
tiation with respect to the variable xa , it is notationally consistent with it, in the following
sense.
Let M = Rd , (U, x) = (Rd , idRd ) and let x a p Tp Rd . If f C (Rd ), then
(f ) = a (f x1 )(x(p)) = a (f )(p),
xa p
then the real numbers X 1 , . . . , X dim M are called the components of X with respect to
the tangent space basis induced by the chart (U, x). The basis { x a p } is also called a
co-ordinate basis.
Proposition 9.17. Let X Tp M and let (U, x) and (V, y) be two charts containing p.
Then we have
b 1
= a (y x )(x(p))
xa p y b p
must have
b
=
xa p y b p
8
for some b . Let us determine what the b are by applying both sides of the equation to
the coordinate maps y c :
(y c ) = a (y c x1 )(x(p));
xa p
b
(y c ) = b b (y c y 1 )(y(p))
y b p
= b b (projc )(y(p))
= b bc
= c .
Hence
c = a (y c x1 )(x(p)).
Substituting this expression for c gives the result.
Corollary 9.18. Let X Tp M and let (U, x) and (V, y) be two charts containing p. Denote
by X a and X e a the coordinates of X with respect to the tangent bases induced by the two
charts, respectively. Then we have:
e a = b (y a x1 )(x(p)) X b .
X
e b = a (y b x1 )(x(p)) X a .
Hence, we read-off X
Remark 9.19. By abusing notation, we can write the previous equations in a more familiar
form. Denote by y b the maps y b x1 : x(U ) Rdim M R; these are real functions of
dim M independent real variables. Since here we are only interested in what happens at
the point p M , we can think of the maps x1 , . . . , xdim M as the independent variables of
each of the y b .
This is a general fact: if {} is a singleton (we let denote its unique element) and
x : {} A, y : A B are maps, then y x is the same as the map y with independent
variable x. Intuitively, x just chooses an element of A.
Hence, we have y b = y b (x1 , . . . , xdim M ) and we can write
y b b
e b = y (x(p)) X a ,
= (x(p)) and X
xa p xa y b p xa
which correspond to our earlier ea = Aba eeb and veb = Aba v a . The function y = y(x) expresses
the new co-ordinates in terms of the old ones, and Aba is the Jacobian matrix of this map,
evaluated at x(p). The inverse transformation, of course, is given by
xb
B ba = (A1 )ba = (y(p)).
y a
9
Remark 9.20. The formula for change of components of vectors under a change of chart
suggests yet another way to define the tangent space to M at p.
Let Ap : {(U, x) A | p U } be the set of charts on M containing p. A tangent vector
v at p is a map
v : Ap Rdim M
satisfying
v((V, y)) = A v((U, x))
where A is the Jacobian matrix of y x1 : Rdim M Rdim M at x(p). In components, we
have
y b
[v((V, y))]b = (x(p)) [v((U, x))]a .
xa
The tangent space Tp M is then defined to be the set of all tangent vectors at p, endowed
with the appropriate vector space structure.
What we have given above is the mathematically rigorous version of the definition of
vector typically found in physics textbooks, i.e. that a vector is a set of numbers v a which,
under a change of coordinates y = y(x), transform as
y b a
veb = v .
xa
For a comparison of the different definitions of Tp M that we have presented and a proof
of their equivalence, refer to Chapter 2 of Vector Analysis, by Klaus Jnich.
10