Lie Derivatives and (Conformal) Killing Vectors

0. Motivations for Lie derivatives

On some manifold, M, or at least in some neighborhood, U ⊆ M, we are concerned

e that are important for some
with a congruence of curves, all with tangent vectors given by ξ,

particular problem, for instance the motions of a physical system over time, beginning at

various different, nearby initial positions that lie in U . The Lie derivative is then a method

of making a comparison of the value that some geometric object has at some point along the

curve relative to the value that it would have had if it had been “dragged” along that curve

to the same point, where we will carefully define below what we mean by the “dragging” of

a geometric object. The Lie derivative, along ξe of some object T , will be defined (below) as

the first-order change of this type, i.e., the first non-trivial term in the (infinite) Taylor series

describing the changes that occur when moved along the curve with this tangent vector, and

will be denoted by Lξe(T ).

Especially important for relativity theory is the behavior of the metric when moved along
e as a vector field on a
curves on a manifold. We will define a conformal Killing vector, ξ,

manifold such that when the metric is dragged along the curves generated by ξe its Lie derivative

is proportional to itself:

Lgµν = 2χ gµν , (0.1)

e
ξ

for some scalar field χ.

The physical understanding of this requirement is that when the metric is dragged along

some congruence of curves it remains itself modulo some scale factor, χ, which may, per-

haps, vary from place to place on the manifold.

In the case that χ is zero, we refer to this as a true Killing vector, and, clearly the metric is

left completely invariant as it is dragged along the curves with that true Killing vector as their

tangent vector. An obvious example is dragging a spherically symmetric metric along a path

of constant latitude on a sphere.

 On the other hand, when the field χ is constant, but not zero, the associated Killing vector

is said to be homothetic, and the metric is being changed by a (constant) scale factor as it

moves along. This too is quite interesting from first principles; a simple example is a dilata-

tion, where objects are, for instance, enlarged but otherwise unchanged as you proceed in some

direction—for instance, toward (spacelike) infinity.

Lastly, when χ is actually a function on the manifold, then this is an actual conformal Killing

vector, rather than the more special cases just considered above. The physical meaning of this

case is not quite as obvious, but will be shown below to be related to certain invariances not

of the metric but, rather, of the curvature of the manifold. [Killing vectors are named for a

Norwegian mathematician named W. Killing, who first described these notions in 1892.]

In order to create an appropriate definition of the Lie derivative, with respect to some

vector field, we must first back up quite a bit, to create enough “new” differential geometry

to do this properly. As well, after doing that we will want to proceed forward from that

definition, to discover the integrability conditions that the existence of a Killing vector puts

on the connection and curvature of a manifold.

1. Maps between manifolds, and their associated pullbacks and pushforwards

On some manifold, M, we first choose a point P ∈ U ⊆ M and a map, ϕ, from U to some

other manifold, N , which could be just a different region of our original manifold:

ϕ:U ⊆M → W ⊆N , Q ≡ ϕ(P ) ∈ W ⊆ N (1.1)

Since there are various different tensor spaces attached to the manifold in this neighborhood,

and in particular at the point P , it is reasonable to suppose that there should be associated

with this map of the manifolds a method of correlating tensors over P ∈ M to tensors over

Q ≡ ϕ(P ) ∈ N . We intend to explain how this happens in some detail, but will find it
[ ]
convenient to begin with the simplest sort of tensors, i.e., functions which are tensors of 00 .

Therefore, we now define the pullback of functions defined over N to those defined over M.

2
 Let f be a function defined over the neighborhood W ⊆ N , i.e., f : W ⊆ N → R. Then we

may define an associated function ϕ∗ f : U ⊆ M → R, the pullback of f via ϕ as the following:


pullback  for ϕ : U ⊆ M → W ⊆ N and f : N → R ,
: (1.1)
of functions  ϕ∗ f : U ⊆ M → R so that ∀S ∈ U ⊆ M, (ϕ∗ f )(S) ≡ f (ϕ(S)) .

Since individual coordinates are just functions, this allows me to use a coordinate system near

Q ∈ N to define a coordinate system near P ∈ M, i.e., to pullback the coordinate chart near

Q ∈ N to create a (very strongly associated) coordinate chart near P ∈ M.

Let {y α }n1 be a coordinate system defined over W ⊆ N ; then {xµ ≡ ϕ∗ (y α )}m

1 is a proper

coordinate system defined over U ⊆ M.

Do note that it is not necessary that the dimensions of the two manifolds be the same,

and we will shortly give examples where this is in fact true. On the other hand, all of our

usages of these ideas to create the Lie derivative will in fact relate to the case when the two

dimensions are in fact the same, and, in fact, to the case where the two manifolds are the same,

even though the two neighborhoods will be different.

Now, since tangent vectors act as operators on functions, this allows us to define the

pushforward of tangent vector fields:


pushforward  for ϕ : U ⊆ M → W ⊆ N and ve ∈ T|U , ve : F|U → F|U
: (1.2)
of vectors  (ϕ ve) ∈ T so that ∀f ∈ F , (ϕ ve)(f ) ≡ ve(ϕ∗ f ) .
∗ |W |W ∗

To see that there actually is something to this definition, let us calculate the relationship

between the components of the two vector fields, relative to their appropriate coordinate sys-

tems, {y µ } and {xµ }, as defined above, noting that in coordinates, we may think of f as a

function of the coordinates y µ , while ϕ∗ f is a function of the coordinates xµ :

α
∂ ν ∂ ∗ ν ∂y ∂
(ϕ∗ ve)(f ) = (ϕ∗ ve)α α
(f ) = (e
v ) ν
(ϕ f ) = e
v (f )
∂y ∂x ∂x ∂y α
ν

∂y α
=⇒ (ϕ∗ ve)α = v ν . (1.3)
∂xν

3
It is worth remarking that this appears to be exactly the same as the transformation that
happens to the components of a vector when one performs a coordinate transformation at a
single point on the manifold. However, it is really quite different philosophically because we
are comparing components of vectors acting on two different manifolds. There is nonetheless
a reason for this apparent equivalence having to do with the difference between thinking of a
coordinate transformation in a passive mode or in an active mode.
a). The passive mode thinks of transformations of the coordinate charts at each point of
some neighborhood, leaving of course the actual points themselves unchanged.
b). The active mode thinks of transformations of the points themselves of a manifold,
and our method created above for mapping charts into other charts sort of drags the
coordinate charts along.

Continuing, now having a method to move tangent vectors between the two manifolds, and
also functions, we may easily perform the reverse sort of mapping for 1-forms, remembering

∼ ∈ Λ|W is denoted
1
that 1-forms map tangent vectors to real numbers: the pullback of a 1-form, α
by ϕ∗ α
∼ ∈ Λ|U :
1


 for ϕ : U ⊆ M → W ⊆ N and α ∼ : T|W → R
∼ ∈ Λ|W , α
1
pullback
: (1.4)
of 1-forms  ϕ∗ α : Λ1 → R so that ∀ ve ∈ T , (ϕ∗ α)(e
∼ |U |U ∼ v) ≡ α e) .
∼(ϕ∗ v

We now want the relationship between the two sets of components, which we acquire by
following along the same line of reasoning as was used to determine the relationship between
the two sets of components of the vector fields:

∂y β
(ϕ∗ α µ ∗
∼)µ v = (ϕ α v) ≡ α
∼)(e e) = αβ (ϕ∗ ve)β = αβ v µ
∼ (ϕ∗ v
∂xµ (1.5)
∂y β
=⇒ (ϕ∗ α
∼)µ = αβ µ .
∂x
We see that it is exactly the same (Jacobian) matrix that is involved in the transformation
as for tangent vectors. However, the difference in the two behaviors is the way the matrix is
multiplied, on the left in the one case and on the right in the other, which is again analogous
to the way in which contravariant and covariant objects in general are transformed.

4
 Now we have sufficiently many definitions to explain why it is that one can pullback

differential forms (including functions, which are after all just 0-forms) but must pushforward

tangent vectors. If we remember that we have created the coordinates on M, {xµ }m

1 , as

dependent on the mapping ϕ and the coordinates {y α }n1 on N , this suggests that, for any

particular actual function performing the mapping we should be able to write out explicitly

the xµ = xµ (y α ). Therefore, in particular then the chain rule of multi-variable calculus allows

us to write the following relationships:

∂ ∂xµ ∂ µ ∂xµ α
= , dx = dy , (1.6)
∂y α ∂y α xµ ∂y α

where of course the chain rule treats these two sorts of objects quite differently. Comparing

with the transformations of components for tangent vectors—which have basis vectors as partial

derivatives with respect to the coordinates—and the components for 1-forms—which have basis

vectors as differentials of the coordinates—we see that they transform in exactly the opposite

ways, in both cases, so that those transformation relationships are actually just rather advanced

formulations of the chain rule, on objects defined over manifolds.

2. Definitions for the Lie derivative of tensor fields

The Lie derivative is a method to determine how vector fields are changing in vector spaces

over nearby points. It is different from the covariant derivative, which relies on the specification

of an affine connection, the function of which is to describe, locally, how the choice of a basis

set for vectors changes from point to point over a manifold. Instead, the Lie derivative relies

on the behavior of the vector field which determines a congruence of (non-intersecting) curves

defined in the neighborhood of a point.

To describe this derivative, we continue the discussion in §1, but now specialize to the
e
case that the manifold N is indeed the same as M, and that there is a given vector field, ξ,

defined over some neighborhood that includes both U and W , which might overlap, at least

to some extent. Taking a parameter λ along each of a family of curves with tangent vector

5
e we may create curves which begin at some arbitrary point R near P ∈ U , which we may
ξ,
then denote by Γ(λ; R) = eλξeR. We can think of the set of these curves, through points in U ,
as constituting an example of the map ϕ between manifolds, as described above, mapping, for
instance P into Q, which we will then note by the symbol ϕλ , since there is a different such
map for each value of λ. In particular, ϕ0 is then the identity map, and ϕ−λ = (ϕλ )−1 , i.e.,
the inverse map to ϕλ , exists and it is just e−λξe. In general, this says that we may think of a
set of curves defined over some neighborhood either as
i.) a congruence of maps from λ ∈ R to the manifold, parameterized by the (nearby)
point on the manifold at which they begin, in the passive mode as described
above, or
ii.) a family of maps from the manifold to itself, parameterized by the real number λ,
which is then an active mode view of the same set of functions. In this case they
are referred to as a family of flows (of the manifold), and we can think of them in
the usual approach, via Taylor series, as defining the chart of coordinates to be used
along the flow in terms of the one where the flow began:

d µ d2
y µ = xα δαµ + λ ϕλ (P ) + 12 λ2 2 ϕµλ (P ) + . . . ]; , (2.1)
dλ dλ

where evaluating the derivatives at the point P means the same as saying that they
are evaluated at λ = 0.

Since these maps are completely invertible, the use of these maps allows us to pullback
or pushforward whatever tensor we want between, for instance, the tensor spaces over P and
those over Q, because we can always use the inverse of the maps if that is what is needed.
[ ]
For example, let us consider two tensors of 11 , T defined at P and B defined at Q. Then we
may define the pushforward value for T, namely ϕ∗λ T, as a tensor over Q, and, as well, the
pullback value for B, namely ϕ∗λ B, a tensor over P by the following schemes where we note
that one defines a tensor by giving its action on the appropriate numbers of tangent vectors
[ ]
and 1-forms. Since these are 11 tensors, we need to have available over the manifold at the

6
 ∼ ∈ Λ|Q and v
e ∈ T|Q , and then the same sort
1
point Q a 1-form field and a tangent vector field, α

β ∈ Λ1|P and w
of objects over the point P , namely ∼ e ∈ T|P . This allows us then to define the

pushforward of T and the pullback of B as follows:

( )( β )
−1 ∂y µ ∂x
(ϕ∗λ T)(α e) ≡
∼, v T(ϕ∗λ α
∼, (ϕλ )∗ ve) =⇒ (ϕ∗λ T) α =
µ
ν α
Tν β ;
at Q acting at P ∂x ∂y
( ν )( α ) (2.2)
∗ −1 ∗ ∗ ∂x ∂y
(ϕλ B)(∼ e ≡ B((ϕλ ) ∼
σ , w) e
σ , ϕ∗λ w) ν
=⇒ (ϕλ B) β = µ β
Bµ α .
at P acting at Q ∂y ∂x

Now we consider the case where our tensor is defined over an entire neighborhood including

both P and Q, and the curve with tangent vector ξe joining them, and Y is a tensor defined

over this entire neighborhood. We may now take this tensor as it is defined at Q and pull it

back to the tensor space over the point P . Will this pullback be equal to the original tensor

field Y defined in the tensor space over the point P . The answer is “probably not!” See the

figure below to visualize this better:

I apologize that the figure, unfortu-

nately, uses the symbol t for the pa-

rameter I am calling λ. Also P is

denoted by x, and Q by ϕt (x)

7
 Both the tensors ϕ∗−λ (Y|ϕλ (P ) ) and Y|P lie in the same tensor space, defined over P ;
therefore, we may certainly ask for their difference, which depends on λ, and is shown in the
figure as the vector which is the difference of these two quantities, both in the tensor space
over P , which we denote as

∆λ Y|P ≡ ϕ∗−λ (Y|ϕλ (P ) ) − Y|P .

e what is the first-order change of this sort, which we

We may then ask, for the given curve ξ,
e More precisely, we define the following tensor in
will call the Lie derivative in the direction ξ.
the (appropriate) tensor space at P :

1{ ∗ }
L Y ≡ lim [ϕ−λ (Y|ϕλ (P ) )]|P − Y|P . (2.3)
e |P λ→0 λ
ξ

We shall see below that this is indeed a tensor of the same type as Y.

To evaluate this geometrically-based idea in terms of components, we first take the coor-
e
dinates of P as xµ , and then recall that Q = ϕλ (P ) lies along the path with tangent vector ξ,
so that
d µ d2
y µ ≡ ϕµλ [x(P )] = xµ (P ) + λ ϕλ (P ) + 12 λ2 2 ϕµλ (P ) + . . .
dλ dλ (2.1′ )
d
= xµ (P ) + λξ µ (λ) P + 12 λ2 ξ µ (λ)|P + . . . ,
dλ
This allows us to think of the tensor fields in terms of a functional dependence on the coordi-
nates of the points at which they exist, so that we can represent YP as Y[x(P )]. Since we are
comparing two different vectors in the same vector space, with respect to the same basis set,
it is simplest if we look at the problem in terms of its components. Therefore, let us suppose,
[ ]
again, the special case that Y is a 11 tensor, so that its components are Y ν β , which gives us

Y ν β |Q −→ e ν β )(x) + O(λ2 ) .
= Y ν β (x + λξ) = Y ν β (x) + λ ξ(Y (2.4a)

As well we have to effect the tensor transformations described above in Eq. (2.2). Since {y µ }
e we have
are coordinates at Q, along the curve with tangent vector ξ,

∂y µ ∂xν
= δνµ + λ ξ,ν
µ
+ O(λ2 ) , = δµν − λ ξ,µ
ν
+ O(λ2 ) . (2.4b)
∂xν ∂y µ

8
Applying the second line of Eqs. (2.2), for ϕ∗−λ , to the first term in Eq. (2.4a), and bringing
along the second term there, and inserting this into the definition given in Eq. (2.3), taking
the limit gives us

LY
µ
ν = ξ η ∂η Y µ ν − ξ,η
µ
Y η ν + ξ,ν
η
Y µη . (2.4c)
e
ξ

More or less, simply as a mnemonic for remembering how this formulation works, we compare
e which
its form to that of the covariant derivative of a tensor, Y, along the same direction, ξ,
is
∇ξeY µ ν = Y µ ν,η ξ η + (Γµ ρη ξ η )Y ρ ν − (Γρ νη ξ η )Y µ ρ .

We see that the purely-algebraic format is to replace the contraction of the direction with the
µ e µ η
connection, i.e., Γ
∼ ρ (ξ) = Γ ρη ξ , by the negative of the derivative of the direction in which

the Lie derivative is being taken, i.e., by −ξ µ ,ρ .

It is then useful to go ahead and write down some specific, important examples, and
also to present them in both a form that is independent of choice of basis and one involving
coordinates explicitly. Therefore we begin with just the action on a function:

for f ∈ F , e ) = ξ η ∂η f = ξ α f,α = ∇ f ,
L f = ξ(f (2.5)
e
ξ
e
ξ

so that we see—as is hopefully logical—that there are no differences between the Lie derivative,
the ordinary derivative operator, and the covariant derivative when acting on just a function.

We can, however, now go further and look again at the action on a vector field, say ve, and
also on a 1-form, say α
∼:

for ve ∈ T , L ve = eeµ (ξ η v µ ,η − ξ µ ,η v η ) = [ξe, ve] ;

e
ξ
(2.6)
∼ ∈ Λ , Lα ω ν {ξ µ αν,µ + ξ µ ,ν αµ } .
1
for α ∼ =∼
e
ξ

It is quite convenient that the form for the Lie derivative of a tangent vector simply takes the
form of the commutator of the two vector fields, which is completely coordinate- and basis-
independent. It would be very good if we also had a form like that for the Lie derivative of a

9
1-form, which we now want to obtain. However, to obtain that form it is convenient to first
define the action of an arbitrary p-form on a single vector. We know that a p-forms needs p
vectors in order to give a number; therefore, if it is given just 1 vector it remains a p − 1-form,
i.e., it wants that many more vectors yet. In principle, one also needs to tell such a product
just where, in the list of p different desired vectors, this given vector will be placed. There are
two reasonably common places where it might be put, namely all the way to the left, or the
right, in the list—the only ones available for a 2-form. Therefore the so-called interior product,
or step product of a vector and a p-form has been created for this purpose:

ve ⌋ ∼
β ≡∼ v , ·, . . . , ·) ,
β (e β ⌊ ve ≡ ∼
∼
β (·, . . . , ·, ve) , (2.7)

where in each case the entries of · in the list of allowed objects for the p-form to act on has
simply not yet been given.
With that information in hand, now let’s consider the following calculations, using a coordinate
basis set:
d[α v )] = d(αµ v µ ) = (αµ,ρ v µ + αµ v µ ,ρ )dxρ ,
∼(e

ve ⌋ dα
∼ = dα v , ·) = v µ (αρ,µ − αµ,ρ )dxρ ,
∼(e (2.8)
=⇒ d[α v )] + α
∼(e v , ·) = (v µ αρ,µ + αµ v µ ,ρ )dxρ = Lα
∼(e ∼ ,
e
v

so that the last line of the equation just above provides us with a coordinate- and basis-free
way of calculating the Lie derivative of a 1-form, if and when that may be needed.
In particular this form allows us to prove a very useful theorem, namely that
The Lie derivative and the exterior derivative commute when
acting on p-forms.

L∼β = ξe ⌋ d∼
β + d(ξe ⌋ ∼
β) ,
e
ξ
{ } (2.9)
=⇒ Ld∼β = d(ξe ⌋ d∼
β ) = d L∼ β − d(ξe ⌋ ∼
β ) = d L∼β.
e
ξ e
ξ e
ξ

It is now also a desirable additional theorem to see that the definition of the Lie derivative is
unchanged when we exchange the partial derivatives in its definition for covariant derivatives.

10
The proof of this is somewhat complicated; therefore, to simplify the atual algebra, I will
perform the associated calculations in an ordinary (holonomic) coordinate basis for tangent
vectors and 1-forms. Nonetheless, since the final results will use covariant derivatives in a
totally basis-independent format, the final results are valid in any choice of basis whatsoever.
[ ]
We give this particular example for our standard 11 tensor, Y, so that we see the method for
both contravariant and covariant indices:
ξ η Y µ ν;η − ξ µ ;η Y η ν + ξ η ;ν Y µ η = ξ η Y µ ν,η − ξ µ ,η Y η ν + ξ η ,ν Y µ η
( )
+ ξ η Γµ λη Y λ ν − Γλ νη Y µ λ − Γµ λη ξ λ Y η ν + Γλ ην ξ η Y µ λ (2.10)
= ξ η ∂η Y µ ν − ξ µ ,η Y η ν + ξ η ,ν Y µ η = LY µ ν ,
e
ξ

where we can see that all the terms with components of the connection simply cancel out.

As it is a very important special case of a tensor, which defines a Killing vector as was
noted on the first page of these notes, let us write out explicitly the Lie derivative of the metric,
using covariant derivatives:

η η η η η
Lgµν = ξ gµν;η + ξ ;µ gην + ξ ;ν gµη = (ξ gην );µ + (ξ gµη );ν = ξ(µ;ν) , (2.11)
e
ξ

which shows the fairly standard method of performing actual calculations looking for Killing
vectors, where the Lie derivative of the metric is simply proportional, again, to the metric.

Next we should consider the Lie derivative of the components of the connection. Since they
are not tensor indices, we should expect them to have a different form for the Lie derivative:

ϕ∗λ Γµ νλ |P = X µ τ (X −1 )η ν (X −1 )σ λ Γτ ησ |P + X µ τ (X −1 )τ ν,λ |P ,
(2.12a)
=⇒ LΓµ νλ = ξ η Γµ νλ,η − ξ,τ
µ τ τ µ
Γ νλ + ξ,ν τ µ
Γ τ λ + ξ,λ µ
Γ ντ + (ξ,ν ),λ .
e
ξ

Now we rewrite this second derivative in terms of a second covariant derivative plus whatever
extra terms are needed, and insert it into the calculation above:

µ
ξ µ ;νλ = ξ µ ;ν,λ + Γµ σλ ξ σ ;ν − Γσ νλ ξ µ ;σ = ξ,νλ + Γµ σν,λ ξ σ + Γµ σν ξ,λ
σ
+ Γµ σλ ξ σ ;ν − Γσ νλ ξ µ ;σ

(2.12b)

11
 This gives us two pairs of terms which are simply the difference of the covariant derivative and
the ordinary derivative, which we replace by the form involving the connection, which results
in the following quite interesting result, involving the curvature tensor:

{ µ }
LΓ νλ = ξ ;νλ + ξ Γ ν[λ,η] + Γ τ [η Γ νλ] = ξ ;νλ − ξ R νλη .
µ µ η µ τ µ η µ
(2.12c)
e
ξ

Since the curvature tensor has arisen, it is reasonable that we should also ask about its
Lie derivative. Following the rules above for arbitrary tensors, we may immediately write that
down:
τ τ τ
LRµνλη = ξ Rµνλη;τ + 2ξ ;[µ Rτ ν]λη + ξ ;[λ Rµντ η] , (2.13)
e
ξ

so that the equations above give us the form for the Lie derivative of all the quantities of
interest for Killing vectors.

3. Implications from the existence of a Killing Vector

The existence of a Killing vector tells us immediately about the symmetries of the metric.
e = ∂/∂q is a (true) Killing vector for some manifold, then it is always possible
If, for instance, K
to arrange a coordinate system, with q as one of the coordinates, so that the components of
the metric with respect to that coordinate basis do not depend on q. It should be noted that
this can be done for any collection of Killing vectors provided that they commute with one
another; otherwise, one must choose whichever one is desired. Therefore, for instance, in the
static, spherically-symmetric metric, we may choose coordinates so that gµν are independent
of φ and t, but may not also eliminate the dependence on θ.

To think a little bit more about that necessary dependence on θ, we note that if the
commutator of a pair of Killing vectors is not zero then the tangent vector that is defined by
that commutator is also a Killing vector. This is, in principle, a very useful way to generate
not-yet-found Killing vectors for a given metric.

Another very useful feature of Killing vectors is that they may be used to determine
“constants of the motion,” i.e., quantities that will be constant along any given geodesic. To

12
 e and a 4-velocity u
see this, consider a Killing vector K e, tangent to some geodesic motion. Then
e ·u
it is claimed that the scalar quantity K e u
e ≡ g(K, e) is constant along that motion:

u e ·u
e(K e) = uµ ∇µ (K η gηλ uλ ) = K η uµ ∇µ uη + uλ uµ ∇µ Kλ = uλ uµ K(λ;µ) = 0 . (3.1)

4. Further conditions for the existence of a (conformal) Killing Vector

The symmetries that allow the existence of a Killing vector also put constraints on the

connections and curvatures of the manifold; or the values of the connections and curvatures

put additional constraints on the existence of a Killing vector, as one chooses to proceed. We

now proceed, therefore, to resolve the complete set of such constraints, which we think of as

“integrability conditions” for the existence of a Killing vector. For some given metric, the

determination of its Killing vectors is usually a question of the solution of a coupled set of

partial differential equations, it is clear that that set ought, in general, to have integrability

conditions. Whether they are satisfied or not is in fact a set of constraints on the connections

and curvatures.

The curvature tensor of a Riemannian manifold measures the lack of commutativity of

covariant derivatives, and also satisfies the first and second Bianchi identities. Considering ξe

as some possible Killing vector, these statements are simply formulated mathematically as the

three equations below:

ξµ;[νλ] = + Rη µνλ ξη ⇐⇒ ξ µ ;[νλ] = −Rµ ηνλ ξ η ,

Rη µνλ + Rη νλµ + Rη λµν = 0 , (4.1)

Rσλνµ;η + Rησνµ;λ + Rληνµ;σ = 0 .

We may insert into these identities the requirements for a conformal Killing vector, given

13
on the first page, Eqs. (0.1). We begin by considering together the first two of Eqs. (4.1):

ξµ;νλ − ξµ;λν + ξν;λµ − ξν;µλ + ξλ;µν − ξλ;νµ = 0 ,

=⇒ ξµ;νλ + ξν;λν + ξλ;µν = χ,λ gµν + χ,ν gµλ + χ,µ gλν ,

=⇒ ξµ;νλ = − ξν;λµ − ξλ;µν + χ;λ gµν + χ;ν gµλ + χ;µ gλν (4.2)

= ξλ;[νµ] + [−χ,µ gνλ + χ,ν gλµ + χ,λ gµν ]

= Rη λνµ ξη + [−χ,µ gνλ + χ,ν gλµ + χ,λ gµν ] ,

which may be thought of as the first integrability condition for our original system. However,

using Eq. (2.12c), the definition of the Lie derivative for the connection, and multiplying by

g µη to raise the first index, we may rewrite this requirement in the following more geometric

form:
µ
LΓ νλ = g µη [−χ,η gνλ + χ,ν gλη + χ,λ gην ] . (4.3)
ξ

This is an especially nice statement relative to the geometric meaning of the difference between

the cases of conformal Killing vectors, i.e., non-constant χ, and the holomorphic or true Killing

vectors, where χ is a constant, possibly zero.

if χ is constant then the connection is invariant under the dragging along curves

generated by that Killing vector.

We now inquire as to whether these integrability conditions have integrability conditions

of their own. We do this by considering again Eqs. (4.1), applying the relation between commu-

tators of covariant derivatives and curvature for the third covariant derivatives of our Killing

vector:

ξµ;ν[λη] = Rσ µλη ξσ;ν + Rσ νλη ξµ;σ . (4.4)

We may, however, also calculate the left-hand side of this equation by taking the covariant

derivative of our first integrability equations, Eq. (4.2), which gives:

ξµ;ν[λη] = Rσ [λνµ ξσ;η] + Rσ [λνµ;η] ξσ + [−χ;µ[η gνλ] + χ;ν[η gλ]µ + χ;[λη] gµν ] . (4.5)

14
 Noting that the very last term above vanishes, because χ is a scalar, we may equate the two
expressions for the commutator of the derivatives and obtain the following:
ξ σ (Rσλνµ;η − Rσηνµ;λ )+ ξ σ ;η Rσλνµ − ξ σ ;λ Rσηνµ + ξ σ ;ν Rσµνλ
(4.6)
+ ξµ;σ R σ
νηλ + χ;νη gλν − χ;νλ gηµ − χ;µη gνλ + χ;µλ gνη = 0 .
If we now intervene with the second Bianchi identity, and use the definition of the Lie derivative
of the curvature tensor, Eq. (8), we may rewrite the last equation in the following form:

L Rηλνµ = 2χRηλνµ − χ;η[ν gλµ] − χ;λ[µ gην] , (4.7)

ξ

relating the Lie derivative of the curvature tensor to the second derivatives of the conformal
factor χ.

5. Specialization to Homothetic Killing Vectors

A homothetic symmetry requires that χ be constant, possibly zero of course, in which case
it would actually be a true Killing vector. We therefore now specialize to the constant case.
Specialization of the equations above gives us
L gµν = 2χ gµν ,
e
ξ
µ
(5.1)
LΓ νλ =0 = LRµ νλη .
e
ξ e
ξ

The first requirement on the second line is simply a repeat of Eqs. (4.3) in the current
situation; however, it is not immediately obvious that the relation on the curvature given
in that line is consistent with the previous Eq. (4.7). In order to show that consistency we
consider how the raising, or lowering, of an index—via the metric—affects results for those
cases where the metric is not preserved by the Lie dragging, i.e., the homothetic cases. We
begin by determining the Lie derivative of the inverse metric:
( )
0 = Lδλµ = L (g µν gνλ ) = Lg
µν
gνλ + g µν Lgνλ
e
ξ e
ξ e
ξ e
ξ
( ) ( )
µν
gνλ + g µν 2χ gνλ = µν
gνλ + 2χδλµ , (5.2)
= Lg Lg
e
ξ e
ξ

=⇒ Lg µν = −2χ g µν .
e
ξ

15
We may then use that to raise the index on the Lie derivative for the curvature:
{ } { }
µ µσ µσ
LR νλη = L (g Rσνλη ) = Lg Rσνλη + g µσ LRσνλη
e
ξ e
ξ e
ξ e
ξ (5.3)
= −2χR µ
νλη + 2χ g µσ
Rσνλη = 0 .

Following that line of calculations, it is also then worthwhile to consider various pieces of the

curvature tensor:

LRνη = LR νµη = 0 ,
µ
e
ξ e
ξ

LR = L (g Rνη ) = −2χ g Rνη = −2χ R ,

νη νη
(5.4)
e
ξ e
ξ

and also LΓµνη = L (gµσ Γσ νη ) = 2χ Γµνη .

e
ξ e
ξ

16