6.1. MATHEMATICAL PRELIMINARIES
I will now provide an intuitive description of vectors in curved spaces and their generalization to the infinite-dimensional (also curved) spaces. The important concept of the vector and multivector, or polyvector, derivative will also be explained. The usual functional derivative is just a component description of the vector derivative in an infinite-dimensional space. My aim is to introduce the readers into those very elegant mathematical concepts and give them a feeling about their practical usefulness. To those who seek a more complete mathematical rigour I advise consulting the literature [22]. In the case in which the concepts discussed here are not found in the existing literature the interested reader is invited to undertake the work and to develop the ideas initiated here further in order to put them into a more rigorous mathematical envelope. Such a development is beyond the scope of this book which aims to point out how various pieces of physics and mathematics are starting to merge before our eyes into a beautiful coherent picture.
The relationγµ·γβ =δµβ implies
∂αγµ·γβ+γµ·∂αγβ = 0, (6.5) from which it follows that
Γµαβ =γµ·∂αγβ. (6.6) Multiplying the latter expression byγµ, summing overµand using
γµ(γµ·a) =a, (6.7)
which holds for any vectora, we obtain
∂αγβ = Γµαβγµ. (6.8) In general the connection isnot symmetric. In particular, when it is sym-metric, we have
γµ·(∂αγβ−∂βγα) = 0 (6.9) and also
∂αγβ −∂βγα= 0 (6.10)
in view of the fact that (6.9) holds for anyγµ. For a symmetric connection, after using (6.1), (6.2), (6.10) we find
Γµαβ = 12gµν(gνα,β+gνβ,α−gαβ,ν). (6.11) Performing the second derivative we have
∂β∂αγµ=−∂βΓµασγσ−Γµασ∂βγσ =−∂βΓµασγσ+ ΓµασΓσβργρ (6.12) and
[∂α, ∂β]γµ=Rµναβγν , (6.13) where
Rµναβ =∂βΓµνα−∂αΓµνβ+ ΓµβρΓραν−ΓµαρΓρβν (6.14) isthe curvature tensor. In general the latter tensor does not vanish and we have a curved space.
SOME ILLUSTRATIONS
Derivative of a vector. Letabe an arbitrary position-dependent vector, expanded according to
a=aµγµ. (6.15)
Taking the derivative with respect to coordinatesxµ we have
∂νa=∂νaµγµ+aµ∂νγµ. (6.16) Using (6.8) and renaming the indices we obtain
∂νa=³∂νaµ+ Γµνρaρ´γµ, (6.17) or
γµ·∂νa=∂νaµ+ Γµνρaρ≡Dνaµ, (6.18) which is the well known covariant derivative. The latter derivative is the projection of∂νaonto one of the basis vectors.
Locally inertial frame. At each point of a spaceVN we can define a set ofN linearly independent vectorsγa,a= 1,2, ..., N, satisfying
γa·γb =ηab, (6.19)
where ηab is the Minkowski tensor. The set of vector fields γa(x) will be called theinertial or Lorentz (orthonormal) frame field. .
A coordinate basis vector can be expanded in terms of local basis vector
γµ=eµaγa, (6.20)
where the expansion coefficients eµa form the so called fielbein field (in 4-dimensions “fielbein” becomes “vierbein” or “tetrad”).
eµa=γµ·γa. (6.21)
Also
γa=eµaγµ, (6.22)
and analogous relations for the inverse vectorsγµ and γa satisfying
γµ·γν =δµν, (6.23)
γa·γb =δab, (6.24)
From the latter relations we find
γµ·γν = (eµaγa)·(eνbγb) =eµaeνa=gµν, (6.25) γa·γb = (eµaγµ)·(eνbγν) =eµaeµb =ηab. (6.26)
A vectoracan be expanded either in terms ofγµ orγa:
a=aµγµ=aµeµaγa=aaγa, aa=aµeµa. (6.27) Differentiation gives
∂µγa=ωabµγb, (6.28) whereωabµ is the connection for the orthonormal frame field γa. Inserting (6.20) into the relation (6.3) we obtain
∂νγµ=∂ν(eµaγa) =∂νeµaγa+eµa∂νγa=−Γµνσγσ, (6.29) which, in view of (6.28), becomes
∂νeµa+ Γµνσeσa+ωabνeµb = 0. (6.30) Because of (6.25), (6.26) we have
∂νeµa−Γσνµeσa+ωabνeµb = 0. (6.31) These are the well known relations for differentiation of the fielbein field.
Geodesic equation in VN. Let p be the momentum vector satisfying the equation of motion
dp
dτ = 0. (6.32)
Expandingp=pµγµ, wherepµ=mX˙µ, we have
˙
pµ+pµγ˙µ= ˙pµγµ+pµ∂νγµX˙ν. (6.33) Using (6.8) we obtain, after suitably renaming the indices,
( ˙pµ+ ΓµαβpαX˙β)γµ= 0, (6.34) which isthe geodesic equationin component notation. The equation of mo-tion (6.32) says that the vectorp does not change during the motion. This means that vectors p(τ) for all values of the parameter τ remain paral-lel amongst themselves (and, of course, retain the same magnitude square p2). After using the expansion p = pµγµ we find that the change of the componentspµ is compensated by the change of basis vectorsγµ.
Geometry in a submanifoldVn. In the previous example we considered a geodesic equation in spacetime VN. Suppose now that a submanifold
— a surface — Vn, parametrized by ξa, is embedded in VN. Let1 ea, a= 1,2, ..., n, be a set of tangent vectors to Vn. They can be expanded in
1Notice that the indexahas now a different meaning from that in the case of a locally inertial frame considered before.
terms of basis vectors ofVN:
ea=∂aXµγµ, (6.35)
where
∂aXµ=ea·γµ (6.36)
are derivatives of the embedding functions ofVn. They satisfy
∂aXµ∂bXµ= (ea·γµ)(eb·γµ) =ea·eb =γab, (6.37) which is the expression for the induced metric ofVn. Differentiation ofea gives
∂bea=∂b∂aXµγµ+∂aXµ∂b.γµ (6.38) Using
∂bγµ=∂νγµ∂bXν (6.39) and the relation (6.8) we obtain from (6.38), after performing the inner product withec,
(ec·∂bea) =∂a∂bXµ∂cXµ+ Γσµν∂aXµ∂bXν∂cXσ. (6.40) On the other hand, the left hand side of eq. (6.38) involves the connection ofVn:
ec·∂bea= Γcba, (6.41) and so we see that eq. (6.40) is a relation between the connection ofVnand VN. Covariant derivative in the submanifold Vn is defined in terms of Γdba.
An arbitrary vectorP inVN can be expanded in terms ofγµ:
P =Pµγµ. (6.42)
It can be projected onto a tangent vectorea:
P ·ea=Pµγµ·ea=Pµ∂aXµ≡Pa (6.43) In particular, a vector ofVN can be itself a tangent vector of a subspace Vn. Let p be such a tangent vector. It can be expanded either in terms of γµ orea:
p=pµγµ=paea, (6.44) where
pa=pµγµ·ea=pµ∂aXµ,
pµ=paea·γµ=pa∂aXµ. (6.45) Such symmetric relations betweenpµandpahold only for a vector pwhich is tangent to Vn.
Suppose now that p is tangent to a geodesic of Vn. Its derivative with respect to an invariant parameterτ along the geodesic is
d
dτ(paea) = d
dτ (pa∂aXµγµ)
= ( ˙pa∂aXµ+∂a∂bXµpaξ˙b+pa∂aXαΓµαβX˙β)γµ (6.46) where we have used eq. (6.8) and
d
dτ∂aXµ=∂a∂bXµξ˙b.
The above derivative, in general, does not vanish: a vector pof VN that is tangent to a geodesic in a subspaceVn changes withτ.
Making the inner product of the left and the right side of eq. (6.46) with ec we obtain
dp
dτ ·ec = ˙pc+ Γcabpaξ˙b= 0. (6.47) Here Γcab is given by eq. (6.40). For a vector p tangent to a geodesic of Vn the right hand side of eq. (6.47) vanish.
On the other hand, starting fromp=pµγµ and using (6.8) the left hand side of eq. (6.47) gives
dp dτ ·ec=
µdpµ
dτ + ΓµαβpαX˙β
¶
∂cXµ= 0. (6.48) Eqs.(6.47), (6.48) explicitly show that in general a geodesic of Vn is not a geodesic ofVN.
A warning is necessary. We have treated tangent vectorseato a subspace Vn as vectors in the embedding space VN. As such they do not form a complete set of linearly independent vectors inVN. An arbitrary vector of VN, of course, cannot be expanded in terms of ea; only a tangent vector to Vn can be expanded so. Therefore, the object ³dp
dτ ·ec´ec should be distinguished from the object ³dp
dτ ·γµ´γµ = dp/dτ. The vanishing of the former object does not imply the vanishing of the latter object.
DERIVATIVE WITH RESPECT TO A VECTOR
So far we have considered derivatives of position-dependent vectors with respect to (scalar) coordinates. We shall now consider the derivative with respect to a vector. LetF(a) be a polyvector-valued function of a vector valued argumentawhich belongs to ann-dimensional vector spaceAn. For an arbitrary vector e inAn the derivative of F in the direction e is given
by µ
e· ∂
∂a
¶
F(a) = lim
τ→0
F(a+eτ)−F(a)
τ = ∂F(a+eτ)
∂τ
¯¯
¯¯
¯τ=0
. (6.49) Forewe may choose one of the basis vectors. Expandinga=aνeν, we have
µ eµ· ∂
∂a
¶
F(a)≡ ∂F
∂aµ = lim
τ→0
F(aνeν +eµτ)−F(aνeν) τ
= lim
τ→0
F((aν +δνµτ)eν)−F(aνeν)
τ . (6.50)
The above derivation holds for an arbitrary function F(a). For instance, forF(a) =a=aνeν eq. (6.50) gives
∂F
∂aµ = ∂
∂aµ(aνeν) =eµ=δµνeν (6.51) For the componentsF ·eα =aα it is
∂aα
∂aµ =δµα (6.52)
The derivative in the directioneµ, as derived in (6.50), is the partial deriva-tive with respect to the componentaµ of the vector argumenta. (See Box 6.1 for some other examples.)
In eq. (6.50) we have derived the operator eµ· ∂
∂a ≡ ∂
∂aµ. (6.53)
For a running indexµthese are components (or projections) of the operator
∂
∂a =eµ µ
eµ· ∂
∂a
¶
=eµ ∂
∂aµ (6.54)
which isthe derivative with respect to a vectora.
The above definitions (6.49)–(6.53) hold for any vectoraofAn. Suppose now that all those vectors are defined at a point a of an n-dimensional manifold Vn. They are said to be tangent to a point x in Vn [22]. If we allow the pointx to vary we have thus a vector fielda(x). In components it is
a(x) =aµ(x)eµ(x), (6.55) where aµ(x) are arbitrary functions of x. A point x is parametrized by a set ofncoordinatesxµ, henceaµ(x) are functions ofxµ. In principleaµ(x) are arbitrary functions ofxµ. In particular, we may choose
aµ(x) =xµ. (6.56)
Box 6.1: Examples of differentiation by a vector 1)Vector valued function:
F =x=xνeν ; ∂F
∂xµ =eµ=δµνeν ; ∂F
∂x =eµ∂F
∂xµ =eµeµ=n.
2)Scalar valued function F =x2 =xνxν ; ∂F
∂xµ = 2xµ; ∂F
∂x =eµ2xµ= 2x.
3)Bivector valued function
F =b∧x= (bαeα)∧(xβeβ) =bαxβeα∧eβ,
∂F
∂xµ = lim
τ→0
bαeα∧(xβeβ+eµτ)−bαeα∧xβeβ
τ =bαeα∧eµ,
∂F
∂x =eµ∂F
∂xµ =bαeµ(eα∧eµ) =bαeµ·(eα∧eµ) +bαeµ∧eα∧eµ,
=bα(δαµeµ−δµµeα) =bαeα(1−n) =b(1−n).
Then
a(x) =xµeµ(x). (6.57) Under a passive coordinate transformation the components aµ(x) change according to
a0µ(x0) = ∂x0µ
∂xν aν(x). (6.58)
This has to be accompanied by the corresponding (active) change of basis vectors,
e0µ(x0) = ∂xν
∂x0µeν(x), (6.59)
in order for a vector a(x) to remain unchanged. In the case in which the components fieldsaµ(x) are just coordinates themselves, the transformation (6.58) reads
a0µ(x0) = ∂x0µ
∂xν xν. (6.60)
In new coordinates x0µ the components a0µ of a vector a(x) = xµeµ are, of course, not equal to the new coordinates x0µ. The reader can check by performing some explicit transformations (e.g., from the Cartesian to spherical coordinates) that an object as defined in (6.56), (6.57) is quite
a legitimate geometrical object and has, indeed, the required properties of a vector field, even in a curved space. The set of points of a curved space can then, at least locally2, be considered as a vector field, such that its components in a certain basis are coordinates. In a given space, there are infinitely many fields with such a property, one field for every possible choice of coordinates. As an illustration I provide the examples of two such fields, denotedX and X0:
a(x) =xµeµ(x) =a0µ(x0)e0µ(x0) =X,
b(x) =bµ(x)eµ(x) =x0µe0µ(x0) =X0. (6.61) Returning to the differential operator (6.49) we can consideraas a vector field a(x) and the definition (6.49) is still valid at every point x ofVn. In particular we can choose
a(x) =x=xνeν. (6.62) Then (6.50) reads
µ eµ· ∂
∂x
¶
F(x)≡ ∂F
∂xµ = lim
τ→0
F((xν+δµντ)eν)−F(xνeν)
τ . (6.63)
This is the partial derivative of a multivector valued functionF(x) of posi-tionx. The derivative with respect to the polyvectorx is
∂
∂x =eµ µ
e0µ· ∂
∂x
¶
=eµ ∂
∂xµ (6.64)
Although we have denoted the derivative as∂/∂xor∂/∂a, this notation should not be understood as implying that∂/∂acan be defined as the limit of a difference quotient. The partial derivative (6.50) can be so defined, but not the derivative with respect to a vector.
DERIVATIVE WITH RESEPCT TO A POLYVECTOR
The derivative with respect to a vector can be generalized to polyvectors.
Definition (6.49) is then replaced by µ
E∗ ∂
∂A
¶
F(A) = lim
τ→0
F(A+Eτ)−F(A)
τ = ∂F(A+Eτ)
∂τ . (6.65)
2Globally this canot be true in general, since a single coordinate system cannot cover all the space.
Here F(A) is a polyvector-valued function of a polyvector A, and E is an arbitrary polyvector . The star “ * ” denotes the scalar product
A∗B=hABi0 (6.66)
of two polyvectorsA andB, wherehABi0 is the scalar part of the Clifford product AB. Let eJ be a complete set of basis vector of Clifford algebra satisfying3
eJ∗eK =δJK, (6.67)
so that any polyvector can be expanded asA=AJeJ. ForE in eq. (6.65) we may choose one of the basis vectors. Then
µ
eK∗ ∂
∂A
¶
F(A)≡ ∂F
∂AK = lim
τ→0
F(AJeJ+eKτ)−F(AJeJ)
τ . (6.68)
This is the partial derivative ofF with respect to the multivector compo-nentsAK. The derivative with respect to a polyvector Ais the sum
∂F
∂A =eJ µ
eJ∗ ∂
∂A
¶
F =eJ ∂F
∂AJ. (6.69)
The polyvectorAcan be a polyvector field A(X) defined over the position polyvector fieldX which is a generalizatin of the position vector field x defined in (6.61). In particular, the field A(X) can be A(X) = X. Then (6.68), (6.69) read
µ
eK∗ ∂
∂X
¶
F(X)≡ ∂F
∂XK = lim
τ→0
F(XJeJ+eKτ)−F(XJeJ)
τ , (6.70)
∂F
∂X =eJ µ
eJ∗ ∂
∂X
¶
F =eJ ∂F
∂XJ (6.71)
which generalizes eqs. (6.63),6.64).