You are on page 1of 18

General Relativity with Torsion

:
Extending Wald’s Chapter on Curvature
Steuard Jensen

Enrico Fermi Institute and Department of Physics
University of Chicago
5640 S. Ellis Avenue, Chicago IL 60637, USA
November 16, 2005
Abstract
Most applications of differential geometry, including general rela-
tivity, assume that the connection is “torsion free”: that vectors do
not rotate during parallel transport. Because some extensions of GR
(such as string theory) do include torsion, it is useful to see how torsion
appears in standard geometrical definitions and formulas in modern
language. In this review article, I step through chapter 3, “Curva-
ture”, of Robert Wald’s textbook General Relativity and show what
changes when the torsion-free condition is relaxed.
Contents
1 Introduction 2
2 Defining Torsion 2
3 Curvature (with Torsion) 3
3.1 Derivative Operators and Parallel Transport . . . . . . . . . . . . . . . . . . 3
3.2 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.3 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.4 Methods for Computing Curvature . . . . . . . . . . . . . . . . . . . . . . . 12
3.4a Coordinate Component Method . . . . . . . . . . . . . . . . . . . . . 12
3.4b Orthonormal Basis (Tetrad) Methods . . . . . . . . . . . . . . . . . 14
A B-Fields and Non-symmetric Metrics 15
B Differential Forms, et cetera 16
B.1 Differential Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

Electronic address: sjensen@theory.uchicago.edu
1
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 2
1 Introduction
Robert Wald’s textbook General Relativity [1], like most work on differential
geometry, includes the assumption that derivative operators ∇
a
are “torsion
free”: for all smooth functions f, ∇
a

b
f − ∇
b

a
f = 0. This property
corresponds intuitively to the condition that vectors not be rotated by parallel
transport. Such a condition is natural to impose, and the theory of general
relativity itself includes this assumption.
However, differential geometry is equally well defined with torsion as with-
out, and some extensions of general relativity include torsion terms. The first
of these was “Einstein-Cartan theory”, as introduced by Cartan in 1922 [2]
(translated to English with commentary as an appendix to [3]). One review
of work in this area is [4]. Of greater current interest, the low energy limit
of string theory includes a massless 2-form field whose field strength plays
the role of torsion. While torsion can always be treated as an independent
tensor field rather than as part of the geometry, the latter approach can be
more efficient and may potentially give greater insight into the theory.
I have written this document primarily for my own reference, but I am
happy to share it with others. It is written with the assumption that the
reader has a copy of Wald’s book close at hand; I have not attempted to
make it stand on its own. (Readers with a decent knowledge of GR may
be able to follow most of it unaided.) I have tried to provide appropriate
generalizations of every numbered equation in chapter 3 (“Curvature”) that
changes in the presence of torsion; in the handful of cases where I have simply
described the change in the text, the equation number is printed in bold. (I
have also done this for appendix B.1 on differential forms.) Any equation not
shown with a correction here remains unchanged in the presence of torsion.
I have done my best to avoid mistakes in either presentation or results, but
I will be grateful for any corrections or feedback on what follows.
2 Defining Torsion
As with most geometric concepts, there are several ways to define torsion.
Following Wald’s presentation, I will define it in terms of the commutator
of derivative operators. (The formulas are provided here for reference; I will
explain them in more detail as they arise over the course of the chapter.) As
explained in the footnote on Wald’s p. (31), the torsion of a connection is
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 3
characterized by the torsion tensor T
c
ab
, which is defined by

a

b
f −∇
b

a
f = −T
c
ab

c
f
for any smooth function f. It is also directly related to commutators of
vector fields,
[v, w]
c
= v
a

a
w
c
−w
a

a
v
c
−T
c
ab
v
a
w
b
,
and to an antisymmetric component in the Christoffel symbols,

c
[ab]
= Γ
c
ab
−Γ
c
ba
= T
c
ab
.
In string theory, the field strength of the massless 2-form field B
ab
is the
negative of the torsion as defined above:
3∇
[a
B
bc]
= H
abc
= −T
abc
.
(In the language of differential forms, H = dB.) This form is considerably
more constrained than general torsion (which need not even be antisymmetric
on all three covariant indices), and this will not be our definition in most of
what follows. A few further comments on this are given in appendix A.
3 Curvature (with Torsion)
The section and equation numbers here are set to match those in Wald’s
book. When those equations are not altered by the presence of torsion, they
will not be given here, so the numbering may jump occasionally.
3.1 Derivative Operators and Parallel Transport
The first change to Wald’s presentation must be to omit the torsion free con-
dition (his condition 5) when defining a derivative operator. Problem 3.1.a
in his book asks the reader to show the existence of the torsion tensor (and
gives a hint on how to do so, by echoing the derivation of the “change of
derivative” tensor C
c
ab
). The proof is not difficult, so to avoid doing people’s
homework for them I will simply cite the result here:

a

b
f −∇
b

a
f = −T
c
ab

c
f .
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 4
This formula comes straight from the footnote to condition 5. The torsion
is antisymmetric in its second and third indices, but in general no symmetry
involving the first index is required at all.
As noted in Problem 3.1.b, the formula for the commutator of two vector
fields changes in the presence of torsion.
[v, w](f) = v
a

a
(w
b

b
f) −w
a

a
(v
b

b
f)
= (v
a

a
w
b
−w
a

a
v
b
)∇
b
f −v
a
w
b
T
c
ab

c
f , (3.1.1)
where we have simply applied the Leibnitz rule. This leads to the modified
expression
[v, w]
c
= v
a

a
w
c
−w
a

a
v
c
−T
c
ab
v
a
w
b
. (3.1.2)
Wald’s derivation of the tensor C
c
ab
relating two derivative operators is
unchanged in the presence of torsion, but the symmetry of that tensor is lost.
Taking the commutator of Eq. (3.1.8) under exchange of a and b yields:
(∇
a

b
−∇
b

a
)f = (
˜

a
˜

b

˜

b
˜

a
)f −(C
c
ab
−C
c
ba
)∇
c
f
(where of course ∇
c
and
˜

c
are equal when applied to f in the final term).
Substituting the definition of torsion in the first two terms shows that the
commutator is
C
c
ab
−C
c
ba
= T
c
ab

˜
T
c
ab
(≡ ∆T
c
ab
) . (3.1.9)
If the two derivative operators have equal torsion, then these coefficients will
have the usual symmetry.
The expressions showing how C
c
ab
relates derivative operators when ap-
plied to general tensors are unchanged by the presence of torsion. And with
the torsion-free condition relaxed, any C
c
ab
will define a new derivative opera-
tor, regardless of its symmetry. In particular, the definition of the Christoffel
symbol Γ
c
ab
will now also incorporate torsion.
The definition of parallel transport is not changed in the presence of tor-
sion, and it still defines a connection on the manifold. But when seeking
a derivative operator compatible with the metric g
ab
, torsion remains un-
constrained. Requiring that parallel transport leave inner products g
ab
v
a
w
b
invariant essentially means that the vectors’ lengths and angles relative to
one another must be unchanged from point to point. But this does not spec-
ify anything about “global” rotations of the tangent space during parallel
transport: that is the physical meaning of torsion.
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 5
When we allow derivative operators with torsion, the statement of The-
orem 3.1.1 must be modified as explained in Problem 3.1.c:
Theorem 3.1.1 Let g
ab
be a metric and T
c
ab
be a torsion. Then there exists
a unique derivative operator ∇
a
with this torsion satisfying ∇
a
g
bc
= 0.
Wald’s proof holds as written up through Eq. (3.1.26). We have seen that
torsion changes the symmetry rule in Eq. (3.1.9), so the next step in the
proof becomes
2C
cab
=
˜

a
g
cb
+
˜

b
g
ac

˜

c
g
ab
−∆T
abc
−∆T
bac
+ ∆T
cab
. (3.1.27)
We have already seen that the difference in torsion is responsible for the
antisymmetric part C
c[ab]
, but this expression shows that it can contribute to
the symmetric part C
c(ab)
as well. The torsion contribution to this symmetric
part is zero if and only if the difference in torsion is totally antisymmetric in
its three indices, ∆T
cab
= ∆T
[cab]
.
It is clear from this that Eq. (3.1.28) becomes
C
c
ab
=
1
2
g
cd
_
˜

a
g
db
+
˜

b
g
ad

˜

d
g
ab
−∆T
abd
−∆T
bad
+ ∆T
dab
_
. (3.1.28)
That in turn leads to modified expressions for the Christoffel symbols:
Γ
c
ab
=
1
2
g
cd
(∂
a
g
db
+ ∂
b
g
ad
−∂
d
g
ab
−T
abd
−T
bad
+ T
dab
) , (3.1.29)
or in components,
Γ
ρ
µν
=
1
2

σ
g
ρσ
_
∂g
σν
∂x
µ
+
∂g
µσ
∂x
ν

∂g
µν
∂x
σ
−T
µνσ
−T
νµσ
+ T
σµν
_
. (3.1.30)
With this expression for the Christoffel symbols in hand, intuition for the
effect of torsion can be developed by considering Wald’s Eq. (3.1.19) on a
manifold with a flat Cartesian metric but non-zero torsion for various choices
of curve (with tangent vector t
a
) and vector to be parallel transported v
a
. In
the particularly simple case where the torsion is totally antisymmetric, that
equation becomes
dv
ν
dt
+
1
2

µ,λ
t
µ
T
ν
µλ
v
λ
= 0 (when g
ab
= η
ab
and T
cab
= T
[cab]
) .
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 6
So for instance, if T
z
xy
> 0, parallel transport along the x direction will cause
v to rotate about the x-axis in a left-handed manner.
In this special case, it is clear from this expression that a vector v
a
tangent
to the curve (i.e. parallel to t
a
) will not be affected by torsion. This is not
the case for more general choices of torsion (which can contribute to the
symmetric part of Γ
c
ba
). General torsion can lead to significant changes
when we consider geodesics.
3.2 Curvature
Our definition of curvature must also be generalized in the presence of torsion.
Wald’s approach is still valid, but the first term on the right hand side of
Eq. (3.2.1) no longer cancels in the next step. After subtracting ∇
b

a
(fω
c
)
from Eq. (3.1.1), we find
(∇
a

b
−∇
b

a
)(fω
c
) = (−T
d
ab

d
f)ω
c
+ f(∇
a

b
−∇
b

a

c
.
Adding an appropriate derivative of ω
c
to both sides then gives
(∇
a

b
−∇
b

a
+ T
d
ab

d
)(fω
c
) = f(∇
a

b
−∇
b

a
+ T
d
ab

d

c
. (3.2.2)
Thus, by the same reasoning as in the torsion-free case, the expression in
parentheses is a tensor:

a

b
ω
c
−∇
b

a
ω
c
+ T
d
ab

d
ω
c
= R
abc
d
ω
d
. (3.2.3)
This defines the Riemann curvature tensor in the presence of torsion.
This simple change in definition,
(∇
a

b
−∇
b

a
) → (∇
a

b
−∇
b

a
+ T
d
ab

d
) ,
is the only modification necessary for some time. This substitution arises
in the middle line of Eq. (3.2.8) due to the change in Eq. (3.1.2) for the
commutator of vector fields, but all of the results in that section are the
same: the new Riemann tensor correctly measures the path dependence of
parallel transport. And the generalized definition is used in Eq. (3.2.10),
Eq. (3.2.11), and Eq. (3.2.12), which show that it is still valid for arbitrary
tensor fields. (There is also an extra term in the second line of Eq. (3.2.10),
but it is straightforward to find and has no broad significance.)
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 7
The generalizations of the “four key properties of the Riemann tensor”
are more interesting. Property (1) (R
abc
d
= −R
bac
d
) still follows directly
from the definition, because T
c
ab
is antisymmetric in a and b. Property (2)
is much less attractive:
R
[abc]
d
= −∇
[a
T
d
bc]
+ T
e
[ab
T
d
c]e
. (3.2.14)
Property (3) (R
abcd
= −R
abdc
) still holds. And the Bianchi identity, property
(4), is modified:

[a
R
bc]d
e
= T
f
[ab
R
c]fd
e
. (3.2.16)
In the proof of property (2), the starting point is modified:
2∇
[a

b
ω
c]
+ T
d
[ab|

d
ω
|c]
= −∇
[a
T
d
bc]
ω
d
+ T
e
[ab
T
d
c]e
ω
d
. (3.2.17)
In deriving this equation, we have used Eq. (B.1.7) for ∇
2
ω. This provides
the appropriate form for Eq. (3.2.18):
R
[abc]
d
ω
d
= ∇
[a

b
ω
c]
−∇
[b

a
ω
c]
+ T
d
[ab|

d
ω
|c]
= 2∇
[a

b
ω
c]
+ T
d
[ab|

d
ω
|c]
=
_
−∇
[a
T
d
bc]
+ T
e
[ab
T
d
c]e
_
ω
d
. (3.2.18)
The proof of property (3) in Eq. (3.2.19) is essentially unchanged; the
only change in the equation is the usual substitution of definitions. However,
the change in property (2) means that the Riemann tensor with torsion is no
longer symmetric under exchange of the first pair of indices with the second.
In the absence of torsion, we were able to write:
2R
cdab
= R
cdab
¸ .. ¸
−R
dacb
¸ .. ¸
−R
acdb
= R
dcba
¸ .. ¸
−R
abdc
. ¸¸ .
−R
bdac
¸ .. ¸
−R
bacd
. ¸¸ .
−R
cbad
= 2R
abcd
. ¸¸ .
But when we use our modification of property (2), this becomes much more
complicated. All four uses of Eq. (3.2.14) add different torsion factors:
2R
cdab
= 2R
abcd
+ 3
_

[b|
T
a|cd]
−∇
[a|
T
b|cd]
−∇
[d|
T
c|ab]
+∇
[c|
T
d|ab]
+ T
ae[b
T
e
cd]
−T
be[a
T
e
cd]
−T
ce[d
T
e
ab]
+ T
de[c
T
e
ab]
_
.
(3.2.20)
If the torsion is totally antisymmetric, T
cab
= T
[cab]
, this expression simpli-
fies enormously. The torsion-squared terms cancel out completely, and the
derivatives of torsion combine and simplify to
R
abcd
+∇
[a
T
b]cd
= R
cdab
+∇
[c
T
d]ab
(when T
cab
= T
[cab]
). (3.2.20a)
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 8
In the special case where dT = 4∂
[a
T
bcd]
= 0, an even more elegant relation
holds:
R
cdab
= R
abcd
|
T→−T
(when T
cab
= T
[cab]
and ∂
[a
T
bcd]
= 0). (3.2.20b)
The proof of this relation is given in section 3.4a, once we have found an
explicit expression for the Riemann tensor.
Finally, we come to the Bianchi identity, property (4). We will simply
apply our usual change in definition to the basic formulas used by Wald:
(∇
a

b
−∇
b

a
+ T
e
ab

e
)∇
c
ω
d
= R
abc
e

e
ω
d
+ R
abd
e

c
ω
e
(3.2.21)
and

a
[(∇
b

c
−∇
c

b
+ T
e
bc

e

d
] = ∇
a
[R
bcd
e
ω
e
] = ω
e

a
R
bcd
e
+ R
bcd
e

a
ω
e
.
(3.2.22)
Without torsion, antisymmetrizing over a, b, and c makes the left hand sides
of these equations equal. But the torsion term in the second equation adds
considerable complication:

[a
_
T
e
bc]

e
ω
d
_
= (∇
[a
T
e
bc]
)∇
e
ω
d
+ T
e
[bc

a]

e
ω
d
= (∇
[a
T
e
bc]
)∇
e
ω
d
+ T
e
[bc
R
a]ed
f
ω
f
+ T
e
[bc|

e

|a]
ω
d
−T
e
[bc
T
f
a]e

f
ω
d
.
The third term on the right hand side is finally of the proper form to match
the torsion term in Eq. (3.2.21), so after antisymmetrization, the second equa-
tion is equal to the first plus the first, second, and fourth terms immediately
above.
Meanwhile, the antisymmetrized first term on the right hand side of
Eq. (3.2.21) is also non-trivial once torsion is included, as we must use
Eq. (3.2.14):
R
[abc]
e

e
ω
d
= (−∇
[a
T
e
bc]
+ T
f
[ab
T
e
c]f
)∇
e
ω
d
.
These terms will cancel with the first and fourth extra terms from the pre-
vious correction. Showing terms from the second equation first:
ω
e

[a
R
bc]d
e
+ R
[bc|d
e

a]
ω
e
= R
[abc]
e

e
ω
d
+ R
[ab|d
e

|c]
ω
e
+ T
e
[bc
R
a]ed
f
ω
f
+ (∇
[a
T
e
bc]
−T
f
[bc
T
e
a]f
)∇
e
ω
d
= R
[ab|d
e

|c]
ω
e
+ T
e
[bc
R
a]ed
f
ω
f
. (3.2.23)
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 9
After a final cancellation of terms between the two sides, this leaves
ω
e

[a
R
bc]d
e
= ω
e
T
f
[bc
R
a]fd
e
, (3.2.24)
which leads to the Bianchi identity as stated above.
In the orthonormal tetrad method of section 3.4b, the Riemann tensor is
treated as a differential 2-form in its first two indices, R
abµ
ν
≡ R
µ
ν
. In the
notation introduced in my appendix B.1, the left hand side of the Bianchi
identity with torsion is written ∇R
µ
ν
. As shown below Eq. (B.1.6), ∇ can
be related to the usual derivative d by
(∇R
µ
ν
)
abc
= (dR
µ
ν
)
abc
+ 3T
d
[ab
R
c]dµ
ν
_
= (dR
µ
ν
)
abc
+ T
σ
∧ R
(1)
σµ
ν
_
.
Comparing this with our result for the Bianchi identity yields the result
dR
µ
ν
= 0, just as in the torsion-free case. (On the other hand, a direct
form-based proof of that identity would make this an alternate derivation of
our Bianchi identity.)
We can still define the Ricci tensor as the trace of the Riemann tensor,
and its symmetry can be found by contracting Eq. (3.2.20) with g
bd
:
R
ac
= R
ca
−3∇
[a
T
b
bc]
+ T
b
be
T
e
ac
. (3.2.26)
If the torsion is totally antisymmetric, the final term vanishes and the equa-
tion reduces to R
ac
= R
ca
+ ∇
b
T
b
ac
; if dT = 0, Eq. (3.2.20b) indicates that
R
ac
= R
ca
|
T→−T
. The Ricci scalar’s definition is entirely unchanged.
The Weyl tensor C
abcd
can still be defined as the trace free part of the Rie-
mann tensor. Wald’s expression assumes that the Ricci tensor is symmetric,
but only minor re-ordering is required:
R
abcd
= C
abcd
+
2
n −2
(R
a[c
g
d]b
−R
b[c
g
d]a
)−
2
(n −1)(n −2)
Rg
a[c
g
d]b
. (3.2.28)
The Weyl tensor still satisfies properties (1) and (3), and its analog of con-
dition (2) can be computed from C
[abc]d
= R
[abc]d
+ 2/(n − 2) R
[ab
g
c]d
using
results above. (I have not checked whether C
abc
d
remains invariant under
conformal transformations.)
The modified Bianchi identity remains complicated after contraction:

a
R
cbd
a
+∇
b
R
cd
−∇
c
R
bd
= 2T
e
a[b
R
c]ed
a
−T
e
bc
R
ed
. (3.2.29)
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 10
Contracting again with g
bd
then gives
2∇
a
R
c
a
−∇
c
R = −T
eab
R
ceab
−2T
ea
c
R
ea
. (3.2.30)
Because no derivatives are explicit on the right hand side, there is no obvious
generalization of the Einstein tensor for arbitrary torsion.
In the special case where T
cab
= T
[cab]
, the antisymmetry allows us to
use identities related to property (2) to reduce the right hand side of this
expression to (T
eab

c
T
eab
− T
eab

e
T
cab
− 2T
cab

e
T
eab
)/2. This is not quite
a total (covariant) derivative, so even in this case no obvious generalization
of the Einstein tensor exists: there is no clear analog of Eq. (3.2.31) or
Eq. (3.2.32). In Einstein-Cartan theory, the stress-energy tensor is modified
by terms related to spin and has similarly non-vanishing divergence.
3.3 Geodesics
The concept of a geodesic can be formulated in two ways. One is the defini-
tion given by Wald in Eq. (3.3.1): a “straightest possible line” whose tangent
vector is parallel propagated along itself. The other is the source of the name
(as I understand it): a “shortest possible path” between any two of its points
(or more generally, an “extremal length path” between them). In the pres-
ence of torsion, these two concepts need no longer be equivalent. For the
sake of consistency with Wald’s definition, we will take the term “geodesic”
to imply the first meaning but not necessarily the second.
Most of Wald’s discussion in this section requires no modification at all.
As mentioned at the end of section 3.1, the geodesic equation depends only
on the symmetric part of the Christoffel symbols, so if T
cab
= T
[cab]
, torsion
will not affect the equation at all.
The first statement that may be changed by the presence of torsion is the
assertion that in Gaussian normal coordinates, geodesics remain orthogonal
to the hypersurfaces S
t
. Because the definition of the commutator of vector
fields has changed, we find
n
b

b
(n
a
X
a
) = n
a
n
b

b
X
a
= n
a
X
b

b
n
a
+ n
a
T
a
bc
n
b
X
c
=
1
2
X
b

b
(n
a
n
a
) + T
abc
n
a
n
b
X
c
= T
abc
n
a
n
b
X
c
. (3.3.6)
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 11
Thus, orthogonality is maintained if and only if the torsion is totally anti-
symmetric.
The next change, as stated earlier, comes in the proof that the shortest
path between two points is a geodesic. The essence of the change is that
while the geodesic equation may depend on torsion, distances depend only
on the metric (so even on a manifold with torsion, the “shortest paths” must
correspond to torsion-free geodesics).
Formally, all of the mathematics leading up to Eq. (3.3.13) remain un-
changed, but the result may no longer match our generalized expression for
the Christoffel symbols in equation (3.1.30). The antisymmetric part of the
Christoffel symbols will not contribute here in any case, but the torsion will
change the symmetric part (and thus invalidate the conclusion that extremal
length paths are geodesics) if and only if T
cab
= T
[cab]
. I do not believe that a
geodesic equation including torsion can be obtained from a “point particle”
Lagrangian as in Eq. (3.3.14), although that Lagrangian can still be used to
find the part of the Christoffel symbols that depends only on the metric.
Finally, in the discussion of the geodesic deviation equation, the first
change is directly related to that of Eq. (3.3.6):
T
b

b
X
a
= X
b

b
T
a
+ T
a
bc
T
b
X
c
; (3.3.16)
as in Eq. (3.3.6), X
a
T
a
need no longer be constant along the geodesics.
The acceleration a
a
= T
c

c
v
a
is then given by:
a
a
= T
c

c
(T
b

b
X
a
)
= T
c

c
(X
b

b
T
a
+ T
a
de
T
d
X
e
)
= (T
c

c
X
b
)(∇
b
T
a
) + X
b
T
c

c

b
T
a
+ T
c

c
(T
a
de
T
d
X
e
)
= (X
c

c
T
b
)(∇
b
T
a
) + (T
b
cd
T
c
X
d
)(∇
b
T
a
) + X
b
T
c

b

c
T
a
−X
b
T
c
T
d
cb

d
T
a
−R
cbd
a
X
b
T
c
T
d
+ T
c

c
(T
a
de
T
d
X
e
)
= X
c

c
(T
b

b
T
a
) −R
cbd
a
X
b
T
c
T
d
+ T
c

c
(T
a
de
T
d
X
e
)
= R
bcd
a
X
b
T
c
T
d
+ T
c

c
(T
a
de
T
d
X
e
) . (3.3.18)
This result may be more intuitive when written explicitly in terms of the
relative velocity: T
c

c
(v
a
− T
a
de
T
d
X
e
) = R
bcd
a
X
b
T
c
T
d
. The left hand side
shows that the relative velocity of two geodesics naturally changes during
parallel transport due to the effects of torsion. Only the deviations from
that direct torsion contribution are due to the manifold’s curvature.
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 12
3.4 Methods for Computing Curvature
3.4a Coordinate Component Method
The given expressions for derivatives of a dual vector field still hold for non-
zero torsion. When constructing the Riemann tensor, the second line of
Eq. (3.4.2) no longer vanishes after a and b are antisymmetrized, but its
contribution precisely cancels out the explicit torsion term in our modified
definition of the Riemann tensor. Thus, Wald’s expression for the Riemann
tensor is essentially correct, except that it assumes the symmetry of the
Christoffel symbol in one place. The proper index order is
R
abc
d
ω
d
=
_
−2∂
[a
Γ
d
b]c
+ 2Γ
e
[a|c
Γ
d
|b]e
¸
ω
d
. (3.4.3)
In components, this yields the following expression for the Riemann tensor:
R
µνρ
σ
=

∂x
ν
Γ
σ
µρ


∂x
µ
Γ
σ
νρ
+

α

α
µρ
Γ
σ
να
−Γ
α
νρ
Γ
σ
µα
) . (3.4.4)
It may at times be useful to decompose the Christoffel symbols into a
sum of metric- and torsion-derived parts:
Γ
c
ab

ˆ
Γ
c
ab
+
1
2
(T
c
ab
−T
ab
c
−T
ba
c
) .
Here,
ˆ
Γ
c
ab
is the part of the Christoffel symbol that is independent of tor-
sion. (As usual, the final two torsion terms cancel if the torsion is totally
antisymmetric.) We can use this to expand out the torsion contribution to
the Riemann tensor, writing the torsion independent part as
ˆ
R
abc
d
:
R
abc
d
=
ˆ
R
abc
d
−∂
[a
T
d
b]c
+ ∂
[a
T
b]c
d
+ ∂
[a|
T
c|b]
d
+
ˆ
Γ
e
[a|c
_
T
d
|b]e
−T
b]e
d
−T
e|b]
d
_
+
_
T
e
[a|c
−T
[a|c
e
−T
c[a
e
_
ˆ
Γ
d
|b]e
+
1
2
_
T
e
[a|c
−T
[a|c
e
−T
c[a
e
_ _
T
d
|b]e
−T
b]e
d
−T
e|b]
d
_
.
For totally antisymmetric torsion, this simplifies substantially:
R
abc
d
=
ˆ
R
abc
d
−∂
[a
T
d
b]c
+
ˆ
Γ
e
[a|c
T
d
|b]e
+ T
e
[a|c
ˆ
Γ
d
|b]e
+
1
2
T
e
[a|c
T
d
|b]e
=
ˆ
R
abc
d
−∇
[a
T
b]c
d
+
1
2
_
T
e
[a|c
T
e|b]
d
−T
e
ab
T
ec
d
_
.
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 13
To compare this with Eq. (3.2.20a) for totally antisymmetric torsion, note
that if the d index is lowered,
ˆ
R
abcd
and both terms in parentheses on the
second line are invariant under {a, b} ↔ {c, d}.
We can now derive the elegant expression R
cdab
= R
abcd
|
T→−T
that holds
in the special case dT = 4∂
[a
T
bcd]
= 0 (as stated in Eq. (3.2.20b) above).
In the final line above, all of the terms are manifestly invariant under this
transformation except −∇
[a
T
b]c
d
. Thus, our goal is to show that ∇
[a
T
b]cd
=
˜

[c
˜
T
d]ab
, where
˜

a
is a new derivative operator that differs from ∇
a
only in
its torsion:
˜
T
c
ab
= −T
c
ab
.
Using results from appendix B.1,
2∇
[a
T
bcd]
= 2∂
[a
T
bcd]
−3T
e
[ab
T
cd]e
= −3T
e
[ab
T
cd]e
.
Therefore,

[a
T
b]cd
= 2∇
[a
T
bcd]
−∇
[c
T
d]ab
= −3T
e
[ab
T
cd]e
−∇
[c
T
d]ab
.
We must next convert to the new derivative operator
˜

a
, which is related to

a
by the tensor C
c
ab
= T
c
ab
. We find that

[c
T
d]ab
=
˜

[c
T
d]ab
−T
e
cd
T
eab
−T
e
[c|a
T
|d]eb
−T
e
[c|b
T
|d]ae
=
˜

[c
T
d]ab
−T
e
cd
T
eab
+ 2T
e
[c|a
T
e|d]b
.
Putting this all together, the result is

[a
T
b]cd
= −3T
e
[ab|
T
e|cd]

˜

[c
T
d]ab
+ T
e
cd
T
eab
−2T
e
[c|a
T
e|d]b
=
˜

[c
˜
T
d]ab
,
as the explicit torsion squared terms cancel out when the antisymmetrizations
are expanded. This proves the desired relation.
Getting back to the flow of Wald’s presentation, the next step is to correct
his expression for the Ricci tensor. Again, it is essentially correct apart from
some index ordering:
R
µρ
=

ν

∂x
ν
Γ
ν
µρ


∂x
µ

ν
Γ
ν
νρ
+

α,ν

α
µρ
Γ
ν
να
−Γ
α
νρ
Γ
ν
µα
) . (3.4.5)
The formula for the contracted Christoffel symbol requires the addition
of a torsion piece,
Γ
a

=

ν
Γ
ν
νµ
=
1
2

ν,α
g
να
∂g
να
∂x
µ
+

ν
T
ν
νµ
. (3.4.7)
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 14
As far as I know, there is no further simplification to be found for this
torsion contribution in general (for totally antisymmetric torsion, it vanishes
entirely). Thus, the final simple formula for the contracted Christoffel symbol
is
Γ
a

=
1
2
1
g
∂g
∂x
µ
+

ν
T
ν
νµ
=

∂x
µ
ln
_
|g| +

ν
T
ν
νµ
. (3.4.9)
And following from this, the divergence of a vector field is

a
T
a
= ∂
a
T
a
+ Γ
a
ab
T
b
=

µ
1
_
|g|

∂x
µ
(
_
|g| T
µ
) +

µ,α
T
α
αµ
T
µ
. (3.4.10)
3.4b Orthonormal Basis (Tetrad) Methods
Wald names the torsion free condition “ingredient (2)” in determining the
curvature, so that term is a signal indicating that changes are required. Most
of the basic definitions used in this approach remain unchanged, so the first
change is simply to substitute the modified definition in the expression for
the components of the Riemann tensor:
R
ρσµν
= R
abcd
(e
ρ
)
a
(e
σ
)
b
(e
µ
)
c
(e
ν
)
d
= (e
ρ
)
a
(e
σ
)
b
(e
µ
)
c
(∇
a

b
−∇
b

a
+ T
d
ab

d
)(e
ν
)
c
. (3.4.17)
The extra term can be expressed very simply using the definition of the
connection 1-forms, leading to the result
R
ρσµν
= (e
ρ
)
a
(e
σ
)
b
_

a
ω
bµν
−∇
b
ω
aµν
+ T
d
ab
ω
dµν

α,β
η
αβ

aβµ
ω
bαν
−ω
bβµ
ω
aαν
]
_
. (3.4.20)
This leads directly to a corresponding change in the expression written in
terms of the Ricci rotation coefficients:
R
ρσµν
= (e
ρ
)
a

a
ω
σµν
−(e
σ
)
a

a
ω
ρµν
+

λ
T
λ
ρσ
ω
λµν
(3.4.21)

α,β
η
αβ

ρβµ
ω
σαν
−ω
σβµ
ω
ραν
+ ω
ρβσ
ω
αµν
−ω
σβρ
ω
αµν
} .
We next come to the the discussion of “ingredient (2)”, and the first
change is due to the corrected formula for the commutator of vector fields:
(e
σ
)
a
[e
µ
, e
ν
]
a
= (e
σ
)
a
_
(e
µ
)
b

b
(e
ν
)
a
−(e
ν
)
b

b
(e
µ
)
a
−T
a
bc
(e
µ
)
b
(e
ν
)
c
_
= ω
µσν
−ω
νσµ
−T
σµν
. (3.4.23)
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 15
Eq. (3.4.24) for the antisymmetrized derivative of the connection 1-forms
still holds, and Eq. (B.1.6) shows precisely how to write it in terms of the
ordinary derivative:

[a
(e
σ
)
b]
= ∇
[a
(e
σ
)
b]
+
1
2
T
c
ab
(e
σ
)
c
=

µ,ν
η
µν
(e
µ
)
[a
ω
b]σν
+
1
2
T
c
ab
(e
σ
)
c
. (3.4.25)
Finally, we can translate these results into the language of differential
forms. The torsion becomes a collection of 2-forms, (T
σ
)
ab
= (e
σ
)
c
T
c
ab
.
Then Eq. (3.4.25) can be written
de
σ
=

µ
e
µ
∧ ω
σ
µ
+T
σ
, (3.4.27)
or, using the notation introduced in appendix B.1 for a derivative with tor-
sion, ∇e
σ
=

µ
e
µ
∧ ω
σ
µ
. Similarly, we can write Eq. (3.4.20) as
R
µ
ν
= dω
µ
ν
+

α
ω
µ
α
∧ ω
α
ν
. (3.4.28)
This equation is identical to the torsion free result, but that fact is sufficiently
surprising (and the equation is sufficiently important) that I have chosen
to duplicate it here anyway. In terms of the derivative with torsion, this
equation is less elegant R
µ
ν
= ∇ω
µ
ν
+

α

µ
α
∧ω
α
ν
+T
α
ω
αµ
ν
). In general,
it seems that using the ordinary derivative d is the simplest approach: the
torsion appears in the equations of structure exactly once, and in a very
straightforward way.
A B-Fields and Non-symmetric Metrics
[Note that this appendix has nothing to do with Wald’s appendix A, which for
its part has nothing to do with torsion.]
In string theory (and several related theories), the metric g
ab
is accompanied
by a 2-form field B
ab
whose field strength H
abc
= 3∂
[a
B
cd]
. The two sometimes
appear in the combination G
ab
≡ g
ab
+ B
ab
, which can in some ways be
thought of as a generalized metric that is not necessarily symmetric (but
remains nondegenerate).
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 16
However, this perspective should not be taken too seriously. Tensor in-
dices are still raised and lowered with g
ab
and g
ab
alone (otherwise, the map
between the tangent space and its dual would be more difficult to define, as
G
ab
v
b
= G
ba
v
b
). And the derivative operator is still chosen so as to satisfy

a
g
bc
= 0. (If the derivative operator instead obeyed ∇
a
G
bc
= 0, this would
imply both ∇
a
g
bc
= 0 and ∇
a
B
bc
= 0. The latter leads to a constraint on the
torsion T
d
[ab
B
c]d
= −∂
[a
B
bc]
. An appropriate torsion could probably be cho-
sen to satisfy this constraint, which would make Eq. (3.1.27) hold as written
above, but this is not the approach that is relevant here.)
The actual effect of the B-field in string theory is to introduce a torsion
directly. Eq. (3.1.29) becomes
Γ
c
ab
=
1
2
g
cd
(∂
a
G
db
+ ∂
b
G
ad
−∂
d
G
ab
)
=
1
2
g
cd
_

a
g
db
+ ∂
b
g
ad
−∂
d
g
ab
−3∂
[a
B
bd]
_
.
This corresponds to a torsion T
abc
= −H
abc
. (The relative sign here is purely
conventional: if we had reversed the order of indices on the three G
ab
terms,
the sign of H would have been positive.) In components, this reads
Γ
ρ
µν
=
1
2
g
ρσ
_
∂g
σν
∂x
µ
+
∂g
µσ
∂x
ν

∂g
µν
∂x
σ

∂B
νσ
∂x
µ

∂B
σµ
∂x
ν

∂B
µν
∂x
σ
_
.
B Differential Forms, et cetera
B.1 Differential Forms
Perhaps the greatest value of differential forms as they are usually presented
is that their properties are independent of the choice of derivative operator.
In the presence of torsion that independence is partly broken, although they
are still independent of the metric.
Naturally, the results change only when derivatives are involved. This first
appears when relating the antisymmetrized derivatives for different derivative
operators:

[b
ω
a
1
···ap]

˜

[b
ω
a
1
···ap]
= −
p

j=1
C
d
[ba
j
ω
a
1
···|d|···ap]
= (−1)
p
p
2
∆T
d
[ba
1
ω
a
2
···ap]d
.
(B.1.6)
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 17
(Note that this corrects a sign error in Wald’s equation, which had no effect
when torsion was zero.) Thus, this map is only unique when the manifold’s
torsion is specified, but it still does not require a preferred metric. I will
denote this map by ∇, and I will continue to use d to refer to the torsion free
case. (This notation is almost certainly not standard, but it seems sensible.)
In particular,
(∇ω)
ba
1
···ap
= (dω)
ba
1
···ap
+ (−1)
p
p(p + 1)
2
T
d
[ba
1
ω
a
2
···ap]d
.
The final term is clearly a differential form as well, but it is not clear (to me,
at least) how to express it in pure form language. It is tempting to write it
as (−1)
p
T
d
∧ (ω
(p−1)
)
d
, simply treating the contracted indices as labels on
a set of forms much like µ in the tetrad (e
µ
)
a
. In fact, I am fairly confident
that this approach would at least work in the orthonormal tetrad context:
there, the torsion is treated as a set of 2-forms T
σ
, and I see no danger in
decomposing a p-form ω into a set of (p −1)-forms (ω
(p−1)
)
σ
.
It is clear that for general torsion, the Poincar´e lemma will not hold:

2
= ∇ ◦ ∇ = 0. In particular, for a scalar field f this is simply the
definition of torsion: (∇
2
f)
ab
= −T
c
ab

c
f. For more general forms, the
formula becomes
2(∇
2
ω)
bca
1
···ap
(p + 2)(p + 1)
= 2∇
[b

c
ω
a
1
···ap]
=
p

j=1
R
[bca
j
d
ω
a
1
···|d|···ap]
−T
d
[bc|

d
ω
|a
1
···ap]
= −p(−1)
p
R
[bca
1
d
ω
a
2
···ap]d
−T
d
[bc|

d
ω
|a
1
···ap]
(B.1.7)
= p(−1)
p
(∇
[b
T
d
ca
1
|
−T
e
[bc
T
d
a
1
|e

|a
2
···ap]d
−T
d
[bc|

d
ω
|a
1
···ap]
In the final line, we have substituted for the antisymmetrized Riemann tensor
using Eq. (3.2.14). Because this expression is much less elegant (and much
less useful) than d
2
= 0, most equations involving differential forms will still
be best expressed in terms of the torsion-free derivative.
References
[1] Robert M. Wald. General Relativity. University of Chicago Press, 1984.
Typeset November 16, 2005; CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 18
[2]
´
Elie Cartan. Sur une g´en´eralisation de la notion de courbure de Riemann
et les espaces `a torsion. C. R. Acad. Sci. (Paris), 174:593–595, 1922.
[3] Peter G. Bergmann and Venzo de Sabbata, editors. Cosmology and Grav-
itation: Spin, Torsion, Rotation, and Supergravity, volume 58 of NATO
Advanced Study Institutes Series: Series B, Physics. Plenum (New York),
1980.
[4] F. W. Hehl, P. Von Der Heyde, G. D. Kerlick, and J. M. Nester. General
relativity with spin and torsion: Foundations and prospects. Rev. Mod.
Phys., 48:393–416, 1976.

Typeset November 16.) As explained in the footnote on Wald’s p. This property corresponds intuitively to the condition that vectors not be rotated by parallel transport. differential geometry is equally well defined with torsion as without. Following Wald’s presentation.1 on differential forms. It is written with the assumption that the reader has a copy of Wald’s book close at hand. there are several ways to define torsion. like most work on differential geometry. in the handful of cases where I have simply described the change in the text.) Any equation not shown with a correction here remains unchanged in the presence of torsion. I have written this document primarily for my own reference. and the theory of general relativity itself includes this assumption. 2005. the latter approach can be more efficient and may potentially give greater insight into the theory. as introduced by Cartan in 1922 [2] (translated to English with commentary as an appendix to [3]). the low energy limit of string theory includes a massless 2-form field whose field strength plays the role of torsion. (The formulas are provided here for reference. a b f − b a f = 0. but I am happy to share it with others. I have done my best to avoid mistakes in either presentation or results. (I have also done this for appendix B. Such a condition is natural to impose. (Readers with a decent knowledge of GR may be able to follow most of it unaided. Of greater current interest. I have not attempted to make it stand on its own.35 $Date: 2005/11/16 22:24:09 GMT 2 1 Introduction Robert Wald’s textbook General Relativity [1]. 2 Defining Torsion As with most geometric concepts. but I will be grateful for any corrections or feedback on what follows. (31).) I have tried to provide appropriate generalizations of every numbered equation in chapter 3 (“Curvature”) that changes in the presence of torsion. However. CVS $Revision: 1. the torsion of a connection is . includes the assumption that derivative operators a are “torsion free”: for all smooth functions f . and some extensions of general relativity include torsion terms. the equation number is printed in bold. While torsion can always be treated as an independent tensor field rather than as part of the geometry. One review of work in this area is [4]. I will explain them in more detail as they arise over the course of the chapter. The first of these was “Einstein-Cartan theory”. I will define it in terms of the commutator of derivative operators.

2005. Problem 3.35 $Date: 2005/11/16 22:24:09 GMT 3 characterized by the torsion tensor T c ab . so the numbering may jump occasionally. It is also directly related to commutators of vector fields.1 Derivative Operators and Parallel Transport The first change to Wald’s presentation must be to omit the torsion free condition (his condition 5) when defining a derivative operator. 3 Curvature (with Torsion) The section and equation numbers here are set to match those in Wald’s book.Typeset November 16. 2Γc [ab] = Γc ab − Γc ba = T c ab . The proof is not difficult. CVS $Revision: 1. the field strength of the massless 2-form field Bab is the negative of the torsion as defined above: 3 [a Bbc] = Habc = −Tabc . by echoing the derivation of the “change of c derivative” tensor Cab ). .) This form is considerably more constrained than general torsion (which need not even be antisymmetric on all three covariant indices). H = dB. they will not be given here.a in his book asks the reader to show the existence of the torsion tensor (and gives a hint on how to do so. and this will not be our definition in most of what follows. [v. which is defined by a bf − b af = −T c ab cf for any smooth function f . 3. In string theory. A few further comments on this are given in appendix A. and to an antisymmetric component in the Christoffel symbols.1. When those equations are not altered by the presence of torsion. so to avoid doing people’s homework for them I will simply cite the result here: a bf − b af = −T c ab cf . (In the language of differential forms. w]c = v a a wc − wa a v c − T c ab v a wb .

This leads to the modified expression [v.Typeset November 16. 2005.2) Wald’s derivation of the tensor C c ab relating two derivative operators is unchanged in the presence of torsion. As noted in Problem 3. but in general no symmetry involving the first index is required at all. But when seeking a derivative operator compatible with the metric gab . w](f ) = v a a (wb b f ) − wa a (v b b f ) = (v a a wb − wa a v b ) b f − v a wb T c ab cf .1. Substituting the definition of torsion in the first two terms shows that the commutator is ˜ C c ab − C c ba = T c ab − T c ab (≡ ∆T c ab ) . In particular. CVS $Revision: 1.1. w]c = v a a wc − wa a v c − T c ab v a wb . (3. (3. but the symmetry of that tensor is lost. [v. And with the torsion-free condition relaxed. . The expressions showing how C c ab relates derivative operators when applied to general tensors are unchanged by the presence of torsion. The definition of parallel transport is not changed in the presence of torsion. (3. The torsion is antisymmetric in its second and third indices.1) where we have simply applied the Leibnitz rule. the definition of the Christoffel symbol Γc ab will now also incorporate torsion. regardless of its symmetry. and it still defines a connection on the manifold.9) If the two derivative operators have equal torsion. But this does not specify anything about “global” rotations of the tangent space during parallel transport: that is the physical meaning of torsion.1.8) under exchange of a and b yields: ( a b − b a )f = ( ˜ a ˜ b − ˜ b ˜ a )f − (C c ab − C c ba ) cf (where of course c and ˜ c are equal when applied to f in the final term). Requiring that parallel transport leave inner products gab v a wb invariant essentially means that the vectors’ lengths and angles relative to one another must be unchanged from point to point. torsion remains unconstrained. then these coefficients will have the usual symmetry. Taking the commutator of Eq. the formula for the commutator of two vector fields changes in the presence of torsion. any C c ab will define a new derivative operator.1.1.b. (3.35 $Date: 2005/11/16 22:24:09 GMT 4 This formula comes straight from the footnote to condition 5.

but this expression shows that it can contribute to the symmetric part Cc(ab) as well. It is clear from this that Eq. (3. (3.1. (3. Then there exists a unique derivative operator a with this torsion satisfying a gbc = 0.19) on a manifold with a flat Cartesian metric but non-zero torsion for various choices of curve (with tangent vector ta ) and vector to be parallel transported v a . intuition for the effect of torsion can be developed by considering Wald’s Eq. . ∆Tcab = ∆T[cab] .1. In the particularly simple case where the torsion is totally antisymmetric. (3.29) ∂gσν ∂gµσ ∂gµν + − − Tµνσ − Tνµσ + Tσµν ∂xµ ∂xν ∂xσ . We have seen that torsion changes the symmetry rule in Eq. 2 or in components.35 $Date: 2005/11/16 22:24:09 GMT 5 When we allow derivative operators with torsion.1 must be modified as explained in Problem 3. (3.Typeset November 16.28) becomes 1 C c ab = g cd ˜ a gdb + ˜ b gad − ˜ d gab − ∆Tabd − ∆Tbad + ∆Tdab 2 .1. The torsion contribution to this symmetric part is zero if and only if the difference in torsion is totally antisymmetric in its three indices. (3.1 Let gab be a metric and T c ab be a torsion.1.1.27) We have already seen that the difference in torsion is responsible for the antisymmetric part Cc[ab] . that equation becomes dv ν 1 + dt 2 tµ T ν µλ v λ = 0 µ.1. Wald’s proof holds as written up through Eq. CVS $Revision: 1.26).1.1.1. Γρ µν = 1 2 g ρσ σ (3. so the next step in the proof becomes 2Ccab = ˜ a gcb + ˜ b gac − ˜ c gab − ∆Tabc − ∆Tbac + ∆Tcab .c: Theorem 3.28) That in turn leads to modified expressions for the Christoffel symbols: 1 Γc ab = g cd (∂a gdb + ∂b gad − ∂d gab − Tabd − Tbad + Tdab ) .1. 2005.30) With this expression for the Christoffel symbols in hand. (3.λ (when gab = ηab and Tcab = T[cab] ) .1.9). the statement of Theorem 3.

2. (3.10). Eq. we find ( a b − b a )(f ωc ) = (−T d ab d f )ωc + f( a b − b a )ωc .1). (There is also an extra term in the second line of Eq. ( a b − b a) →( a b − b a + T d ab d) . General torsion can lead to significant changes when we consider geodesics.1) no longer cancels in the next step.2 Curvature Our definition of curvature must also be generalized in the presence of torsion. 3. but the first term on the right hand side of Eq. 2005. This substitution arises in the middle line of Eq. Adding an appropriate derivative of ωc to both sides then gives ( a b − b a + T d ab d )(f ωc ) = f( a b − b a + T d ab d )ωc . After subtracting b a (f ωc ) from Eq. parallel to ta ) will not be affected by torsion. but it is straightforward to find and has no broad significance. Wald’s approach is still valid.) .e.2. This simple change in definition. by the same reasoning as in the torsion-free case. And the generalized definition is used in Eq. it is clear from this expression that a vector v a tangent to the curve (i. the expression in parentheses is a tensor: a b ωc − b a ωc + T d ab d ωc = Rabc d ωd .35 $Date: 2005/11/16 22:24:09 GMT 6 So for instance. is the only modification necessary for some time.2) Thus. CVS $Revision: 1. (3. which show that it is still valid for arbitrary tensor fields.1. (3.11). (3. This is not the case for more general choices of torsion (which can contribute to the symmetric part of Γc ba ).12). and Eq. (3.2.2. In this special case.3) This defines the Riemann curvature tensor in the presence of torsion.2.2.2. (3. but all of the results in that section are the same: the new Riemann tensor correctly measures the path dependence of parallel transport.2.1.Typeset November 16. (3.2) for the commutator of vector fields. (3.8) due to the change in Eq. (3. if T z xy > 0.10). parallel transport along the x direction will cause v to rotate about the x-axis in a left-handed manner. (3.

Tcab = T[cab] .20a) . (3.Typeset November 16. the change in property (2) means that the Riemann tensor with torsion is no longer symmetric under exchange of the first pair of indices with the second.18) The proof of property (3) in Eq.19) is essentially unchanged. This provides d ω|c] (3. because T c ab is antisymmetric in a and b.14) Property (3) (Rabcd = −Rabdc ) still holds. this expression simplifies enormously. property (4). (3. (3. The torsion-squared terms cancel out completely.16) [a Rbc]d = T [ab Rc]f d . 2 (3.17) In deriving this equation. we have used Eq. (3.2. (B.7) for the appropriate form for Eq. the only change in the equation is the usual substitution of definitions. (3.14) add different torsion factors: [b| Ta|cd] − [a| Tb|cd] − [d| Tc|ab] + [c| Td|ab] + Tae[b T e cd] − Tbe[a T e cd] − Tce[d T e ab] + Tde[c T e ab] .35 $Date: 2005/11/16 22:24:09 GMT 7 The generalizations of the “four key properties of the Riemann tensor” are more interesting.2.2. (3.1.2. Property (2) is much less attractive: R[abc] d = − d [a T bc] + T e [ab T d c]e . this becomes much more complicated.2. we were able to write: 2Rcdab = Rcdab −Rdacb −Racdb = Rdcba −Rabdc −Rbdac −Rbacd −Rcbad = 2Rabcd But when we use our modification of property (2).2. 2005. the starting point is modified: 2 [a b ωc] + T d [ab| d ω|c] =− d [a T bc] ωd + T e [ab T d c]e ωd .2. CVS $Revision: 1. and the derivatives of torsion combine and simplify to 2Rcdab = 2Rabcd + 3 Rabcd + [a Tb]cd = Rcdab + [c Td]ab (when Tcab = T[cab] ). All four uses of Eq.18): d R[abc] d ωd = [a b ωc] − [b a ωc] + T [ab| = 2 [a b ωc] + T d [ab| d ω|c] = − [a T d bc] + T e [ab T d c]e ωd . ω.2. In the proof of property (2).2. In the absence of torsion. However. And the Bianchi identity. Property (1) (Rabc d = −Rbac d ) still follows directly from the definition. is modified: e f e (3.20) If the torsion is totally antisymmetric.

2.2.21) is also non-trivial once torsion is included. once we have found an explicit expression for the Riemann tensor.2. (3. These terms will cancel with the first and fourth extra terms from the previous correction. But the torsion term in the second equation adds considerable complication: a a b − b a + T e ab e) c ωd = Rabc e e ωd + Rabd e c ωe (3. (3. and c makes the left hand sides of these equations equal.23) .2. antisymmetrizing over a.22) Without torsion. we come to the Bianchi identity.2. the antisymmetrized first term on the right hand side of Eq.Typeset November 16. the second equation is equal to the first plus the first. The third term on the right hand side is finally of the proper form to match the torsion term in Eq. b.35 $Date: 2005/11/16 22:24:09 GMT 8 In the special case where dT = 4∂[a Tbcd] = 0. (3.20b) The proof of this relation is given in section 3.4a. an even more elegant relation holds: Rcdab = Rabcd |T →−T (when Tcab = T[cab] and ∂[a Tbcd] = 0). We will simply apply our usual change in definition to the basic formulas used by Wald: ( and + Rbcd e a ωe .2. (3. (3. Meanwhile. property (4). Showing terms from the second equation first: ωe [a Rbc]d e + R[bc|d e a] ωe = R[abc] e e ωd + R[ab|d e |c] ωe + T e [bc Ra]ed f ωf + ( [a T e bc] − T f [bc T e a]f ) e ωd = R[ab|d e |c] ωe + T e [bc Ra]ed f ωf . so after antisymmetrization. (3.21).14): R[abc] e e ωd = (− e [a T bc] + T f [ab T e c]f ) e ωd . CVS $Revision: 1. second.2. Finally. 2005. as we must use Eq. and fourth terms immediately above.21) [( b c − c b + T e bc e )ωd ] = a [Rbcd e ωe ] = ωe a Rbcd e [a T e bc] e ωd = ( = ( + e e [a T bc] ) e ωd + T [bc a] e ωd e e f [a T bc] ) e ωd + T [bc Ra]ed ωf T e [bc| e |a] ωd − T e [bc T f a]e f ωd .

1. and its analog of condition (2) can be computed from C[abc]d = R[abc]d + 2/(n − 2) R[ab gc]d using results above. this leaves ωe [a Rbc]d e = ωe T f [bc Ra]f d e . just as in the torsion-free case. Rabµ ν ≡ Rµ ν . (3.) We can still define the Ricci tensor as the trace of the Riemann tensor. (3. the final term vanishes and the equation reduces to Rac = Rca + b T b ac .29) . (3. Wald’s expression assumes that the Ricci tensor is symmetric. (On the other hand.24) which leads to the Bianchi identity as stated above.Typeset November 16.4b. (3.2. Comparing this with our result for the Bianchi identity yields the result dRµ ν = 0. The Ricci scalar’s definition is entirely unchanged.26) If the torsion is totally antisymmetric.28) n−2 (n − 1)(n − 2) The Weyl tensor still satisfies properties (1) and (3). the left hand side of the Bianchi identity with torsion is written Rµ ν .20) with g bd : Rac = Rca − 3 b [a T bc] + T b be T e ac . can be related to the usual derivative d by ( Rµ ν )abc = (dRµ ν )abc + 3T d [ab Rc]dµ ν = (dRµ ν )abc + T σ ∧ R(1) σµ ν .1. the Riemann tensor is treated as a differential 2-form in its first two indices.20b) indicates that Rac = Rca |T →−T .) The modified Bianchi identity remains complicated after contraction: a Rcbd a + b Rcd − c Rbd = 2T e a[b Rc]ed a − T e bc Red . In the orthonormal tetrad method of section 3. As shown below Eq. (3.35 $Date: 2005/11/16 22:24:09 GMT 9 After a final cancellation of terms between the two sides. Eq.2. (I have not checked whether Cabc d remains invariant under conformal transformations. but only minor re-ordering is required: Rabcd = Cabcd + 2 2 (Ra[c gd]b − Rb[c gd]a ) − Rga[c gd]b . and its symmetry can be found by contracting Eq.2.6). (3.2. (B. 2005. if dT = 0.2. The Weyl tensor Cabcd can still be defined as the trace free part of the Riemann tensor.2. a direct form-based proof of that identity would make this an alternate derivation of our Bianchi identity. CVS $Revision: 1. In the notation introduced in my appendix B.

geodesics remain orthogonal to the hypersurfaces St . the stress-energy tensor is modified by terms related to spin and has similarly non-vanishing divergence.2. (3.3.35 $Date: 2005/11/16 22:24:09 GMT 10 Contracting again with g bd then gives 2 a Rc a − cR = −T eab Rceab − 2T ea c Rea . In the presence of torsion. the antisymmetry allows us to use identities related to property (2) to reduce the right hand side of this expression to (T eab c Teab − T eab e Tcab − 2Tcab e T eab )/2.1.6) . This is not quite a total (covariant) derivative. Most of Wald’s discussion in this section requires no modification at all.2. In the special case where Tcab = T[cab] . so even in this case no obvious generalization of the Einstein tensor exists: there is no clear analog of Eq.Typeset November 16. (3. CVS $Revision: 1. The first statement that may be changed by the presence of torsion is the assertion that in Gaussian normal coordinates. The other is the source of the name (as I understand it): a “shortest possible path” between any two of its points (or more generally. (3.2. an “extremal length path” between them). Because the definition of the commutator of vector fields has changed. (3.30) Because no derivatives are explicit on the right hand side. we find nb b (na X a ) = na nb b X a = na X b b na + na T a bc nb X c 1 b X b (na na ) + Tabc na nb X c = 2 = Tabc na nb X c .3. 2005.1): a “straightest possible line” whose tangent vector is parallel propagated along itself. As mentioned at the end of section 3. torsion will not affect the equation at all. For the sake of consistency with Wald’s definition.32). In Einstein-Cartan theory. so if Tcab = T[cab] . these two concepts need no longer be equivalent. 3.31) or Eq. (3. One is the definition given by Wald in Eq.3 Geodesics The concept of a geodesic can be formulated in two ways. the geodesic equation depends only on the symmetric part of the Christoffel symbols. there is no obvious generalization of the Einstein tensor for arbitrary torsion. we will take the term “geodesic” to imply the first meaning but not necessarily the second.

30).16) as in Eq. as stated earlier. The antisymmetric part of the Christoffel symbols will not contribute here in any case. I do not believe that a geodesic equation including torsion can be obtained from a “point particle” Lagrangian as in Eq. X a Ta need no longer be constant along the geodesics. Formally. all of the mathematics leading up to Eq. (3. The left hand side shows that the relative velocity of two geodesics naturally changes during parallel transport due to the effects of torsion. orthogonality is maintained if and only if the torsion is totally antisymmetric. (3. but the torsion will change the symmetric part (and thus invalidate the conclusion that extremal length paths are geodesics) if and only if Tcab = T[cab] . The next change.1. in the discussion of the geodesic deviation equation.3.6): Tb bX a = Xb bT a + T a bc T b X c . comes in the proof that the shortest path between two points is a geodesic.3. Finally. .3.14). 2005.13) remain unchanged. (3.3. the first change is directly related to that of Eq. (3. although that Lagrangian can still be used to find the part of the Christoffel symbols that depends only on the metric. but the result may no longer match our generalized expression for the Christoffel symbols in equation (3. distances depend only on the metric (so even on a manifold with torsion. The essence of the change is that while the geodesic equation may depend on torsion.18) This result may be more intuitive when written explicitly in terms of the relative velocity: T c c (v a − T a de T d X e ) = Rbcd a X b T c T d .35 $Date: 2005/11/16 22:24:09 GMT 11 Thus.6). Only the deviations from that direct torsion contribution are due to the manifold’s curvature. CVS $Revision: 1.3.Typeset November 16. (3. The acceleration aa = T c c v a is then given by: aa = = = = T c c (T b b X a ) T c c (X b b T a + T a de T d X e ) (T c c X b )( b T a ) + X b T c c b T a + T c c (T a de T d X e ) (X c c T b )( b T a ) + (T b cd T c X d )( b T a ) + X b T c b c T a − X b T c T d cb d T a − Rcbd a X b T c T d + T c c (T a de T d X e ) = X c c (T b b T a ) − Rcbd a X b T c T d + T c c (T a de T d X e ) = Rbcd a X b T c T d + T c c (T a de T d X e ) .3. (3. the “shortest paths” must correspond to torsion-free geodesics).

this simplifies substantially: 1 ˆ ˆ ˆ Rabc d = Rabc d − ∂[a T d b]c + Γe [a|c T d |b]e + T e [a|c Γd |b]e + T e [a|c T d |b]e 2 1 e d d d e d ˆ = Rabc − [a Tb]c + T [a|c Te|b] − T ab Tec . (3. (3. 2 ˆ Here.4a Methods for Computing Curvature Coordinate Component Method The given expressions for derivatives of a dual vector field still hold for nonzero torsion.4) It may at times be useful to decompose the Christoffel symbols into a sum of metric. except that it assumes the symmetry of the Christoffel symbol in one place. α (3. 2 . 2005.2) no longer vanishes after a and b are antisymmetrized.4. The proper index order is Rabc d ωd = −2∂[a Γd b]c + 2Γe [a|c Γd |b]e ωd . Γc ab is the part of the Christoffel symbol that is independent of torsion.4. writing the torsion independent part as Rabc d : ˆ Rabc d = Rabc d − ∂[a T d b]c + ∂[a Tb]c d + ∂[a| Tc|b] d ˆ ˆ + Γe [a|c T d |b]e − Tb]e d − Te|b] d + T e [a|c − T[a|c e − Tc[a e Γd |b]e 1 e T [a|c − T[a|c e − Tc[a e T d |b]e − Tb]e d − Te|b] d . + 2 For totally antisymmetric torsion. Thus.3) In components. Wald’s expression for the Riemann tensor is essentially correct. but its contribution precisely cancels out the explicit torsion term in our modified definition of the Riemann tensor. the second line of Eq.Typeset November 16.4. the final two torsion terms cancel if the torsion is totally antisymmetric. When constructing the Riemann tensor. this yields the following expression for the Riemann tensor: Rµνρ σ = ∂ ∂ σ Γ µρ − µ Γσ νρ + ν ∂x ∂x (Γα µρ Γσ να − Γα νρ Γσ µα ) .4 3.and torsion-derived parts: 1 ˆ Γc ab ≡ Γc ab + (T c ab − Tab c − Tba c ) .) We can use this to expand out the torsion contribution to ˆ the Riemann tensor. (As usual. CVS $Revision: 1.35 $Date: 2005/11/16 22:24:09 GMT 12 3.

[a Tb]cd [a Tbcd] = 2∂[a Tbcd] − 3T e [ab Tcd]e = −3T e [ab Tcd]e .20a) for totally antisymmetric torsion.4. (3.5) The formula for the contracted Christoffel symbol requires the addition of a torsion piece. Getting back to the flow of Wald’s presentation.4. In the final line above.Typeset November 16. (3. Using results from appendix B. d}.1. all of the terms are manifestly invariant under this transformation except − [a Tb]c d . =2 [a Tbcd] − [c Td]ab = −3T e [ab Tcd]e − [c Td]ab . our goal is to show that [a Tb]cd = ˜ [c Td]ab .α ∂gνα + ∂xµ T ν νµ . b} ↔ {c. Γa aµ = ν Γν νµ = 1 2 g να ν. it is essentially correct apart from some index ordering: Rµρ = ν ∂ ∂ ν Γ µρ − µ ν ∂x ∂x Γν νρ + ν α. 2005.7) . where ˜ a is a new derivative operator that differs from a only in ˜ ˜ its torsion: T c ab = −T c ab . note ˆ that if the d index is lowered. Thus. Putting this all together. the next step is to correct his expression for the Ricci tensor.2. as the explicit torsion squared terms cancel out when the antisymmetrizations are expanded. ν (3. We can now derive the elegant expression Rcdab = Rabcd |T →−T that holds in the special case dT = 4∂[a Tbcd] = 0 (as stated in Eq. We find that [c Td]ab = = ˜ [c Td]ab − T e cd Teab − T e [c|a T|d]eb − T e [c|b T|d]ae ˜ [c Td]ab − T e cd Teab + 2T e [c|a Te|d]b . (3. the result is [a Tb]cd ˜ = −3T e [ab| Te|cd] − ˜ [c Td]ab + T e cd Teab − 2T e [c|a Te|d]b = ˜ [c Td]ab . which is related to c c a by the tensor C ab = T ab . This proves the desired relation.35 $Date: 2005/11/16 22:24:09 GMT 13 To compare this with Eq.ν (Γα µρ Γν να − Γα νρ Γν µα ) .2.20b) above). CVS $Revision: 1. We must next convert to the new derivative operator ˜ a . Rabcd and both terms in parentheses on the second line are invariant under {a. 2 Therefore. Again.

2005.21) η {ωρβµ ωσαν − ωσβµ ωραν + ωρβσ ωαµν − ωσβρ ωαµν } . (3.4b Orthonormal Basis (Tetrad) Methods Wald names the torsion free condition “ingredient (2)” in determining the curvature. the divergence of a vector field is 1 ∂ a a a b T α αµ T µ .β a ωbµν − b ωaµν + T d ab ωdµν . it vanishes entirely).4. there is no further simplification to be found for this torsion contribution in general (for totally antisymmetric torsion.4. (3.4.4. and the first change is due to the corrected formula for the commutator of vector fields: (eσ )a [eµ . We next come to the the discussion of “ingredient (2)”.4. (3.23) . CVS $Revision: 1.20) η αβ [ωaβµ ωbαν − ωbβµ ωaαν ] This leads directly to a corresponding change in the expression written in terms of the Ricci rotation coefficients: Rρσµν = (eρ )a − α.α 3. Thus. (3.35 $Date: 2005/11/16 22:24:09 GMT 14 As far as I know.4.9) + T ν νµ = ln |g| + µ 2 g ∂x ∂xµ ν ν And following from this. leading to the result Rρσµν = (eρ )a (eσ )b − α. a b (eµ ) − T a bc (eµ )b (eν )c (3. eν ]a = (eσ )a (eµ )b b (eν )a − (eν )b = ωµσν − ωνσµ − Tσµν .10) ( |g| T µ ) + a T = ∂a T + Γ ab T = µ |g| ∂x µ µ. so that term is a signal indicating that changes are required. the final simple formula for the contracted Christoffel symbol is 1 1 ∂g ∂ Γa aµ = T ν νµ .Typeset November 16. Most of the basic definitions used in this approach remain unchanged.β a ωσµν αβ − (eσ )a a ωρµν + λ T λ ρσ ωλµν (3. so the first change is simply to substitute the modified definition in the expression for the components of the Riemann tensor: Rρσµν = Rabcd (eρ )a (eσ )b (eµ )c (eν )d = (eρ )a (eσ )b (eµ )c ( a b − + T d ab d )(eν )c b a .17) The extra term can be expressed very simply using the definition of the connection 1-forms.

(3.27) or.25) can be written deσ = µ eµ ∧ ω σ µ + T σ . Then Eq. using the notation introduced in appendix B. (T σ )ab = (eσ )c T c ab . CVS $Revision: 1. The torsion becomes a collection of 2-forms. Similarly. this equation is less elegant Rµ ν = ω µ ν + α (ω µ α ∧ ω α ν + T α ωαµ ν ).24) for the antisymmetrized derivative of the connection 1-forms still holds. In terms of the derivative with torsion.1. The two sometimes appear in the combination Gab ≡ gab + Bab . 2005. . we can translate these results into the language of differential forms. it seems that using the ordinary derivative d is the simplest approach: the torsion appears in the equations of structure exactly once. (3. (3. we can write Eq. and Eq. (3. which can in some ways be thought of as a generalized metric that is not necessarily symmetric (but remains nondegenerate).Typeset November 16. A B-Fields and Non-symmetric Metrics [Note that this appendix has nothing to do with Wald’s appendix A.1 for a derivative with torsion.4.4.4.4. eσ = µ eµ ∧ ω σ µ .ν 1 η µν (eµ )[a ωb]σν + T c ab (eσ )c .4.4.] In string theory (and several related theories). but that fact is sufficiently surprising (and the equation is sufficiently important) that I have chosen to duplicate it here anyway. which for its part has nothing to do with torsion.6) shows precisely how to write it in terms of the ordinary derivative: ∂[a (eσ )b] = [a (eσ )b] + 1 c T ab (eσ )c = 2 µ. (B. In general.20) as Rµ ν = dω µ ν + α ωµα ∧ ωαν . and in a very straightforward way.28) This equation is identical to the torsion free result. the metric gab is accompanied by a 2-form field Bab whose field strength Habc = 3∂[a Bcd] .25) 2 Finally. (3. (3.35 $Date: 2005/11/16 22:24:09 GMT 15 Eq.

B B. Naturally. this reads 1 Γρ µν = g ρσ 2 ∂Bνσ ∂Bσµ ∂Bµν ∂gσν ∂gµσ ∂gµν + − − − − µ ν σ ∂x ∂x ∂x ∂xµ ∂xν ∂xσ . which would make Eq. The latter leads to a constraint on the torsion T d [ab Bc]d = −∂[a Bbc] .1 Differential Forms.35 $Date: 2005/11/16 22:24:09 GMT 16 However. In the presence of torsion that independence is partly broken. although they are still independent of the metric. An appropriate torsion could probably be chosen to satisfy this constraint.) The actual effect of the B-field in string theory is to introduce a torsion directly.) In components. the map between the tangent space and its dual would be more difficult to define. this would imply both a gbc = 0 and a Bbc = 0. the results change only when derivatives are involved.1.Typeset November 16. et cetera Differential Forms Perhaps the greatest value of differential forms as they are usually presented is that their properties are independent of the choice of derivative operator. (The relative sign here is purely conventional: if we had reversed the order of indices on the three Gab terms. And the derivative operator is still chosen so as to satisfy a gbc = 0. Eq. 2 (B.1. (3. (If the derivative operator instead obeyed a Gbc = 0.29) becomes Γc ab = 1 cd g (∂a Gdb + ∂b Gad − ∂d Gab ) 2 1 cd = g ∂a gdb + ∂b gad − ∂d gab − 3∂[a Bbd] . 2005. as Gab v b = Gba v b ). this perspective should not be taken too seriously. the sign of H would have been positive. (3. CVS $Revision: 1. Tensor indices are still raised and lowered with g ab and gab alone (otherwise. 2 This corresponds to a torsion Tabc = −Habc . This first appears when relating the antisymmetrized derivatives for different derivative operators: p ˜ [b ωa ···ap ] = − [b ωa1 ···ap ] − 1 j=1 C d [baj ωa1 ···|d|···ap ] = (−1)p p ∆T d [ba1 ωa2 ···ap ]d .27) hold as written above.6) . but this is not the approach that is relevant here.1.

7) d [bc| d ω|a1 ···ap ] −T e [bc T a1 |e )ω|a2 ···ap ]d d −T In the final line. 2 The final term is clearly a differential form as well. but it is not clear (to me. ( ω)ba1 ···ap = (dω)ba1 ···ap + (−1)p p(p + 1) d T [ba1 ωa2 ···ap ]d . which had no effect when torsion was zero. simply treating the contracted indices as labels on a set of forms much like µ in the tetrad (eµ )a . but it seems sensible. we have substituted for the antisymmetrized Riemann tensor using Eq. It is tempting to write it as (−1)p T d ∧ (ω (p−1) )d . and I see no danger in decomposing a p-form ω into a set of (p − 1)-forms (ω (p−1) )σ . General Relativity.2. I will denote this map by . I am fairly confident that this approach would at least work in the orthonormal tetrad context: there. (This notation is almost certainly not standard. It is clear that for general torsion. for a scalar field f this is simply the definition of torsion: ( 2 f )ab = −T c ab c f . and I will continue to use d to refer to the torsion free case. In particular.) In particular. at least) how to express it in pure form language.35 $Date: 2005/11/16 22:24:09 GMT 17 (Note that this corrects a sign error in Wald’s equation. this map is only unique when the manifold’s torsion is specified. 1984. . (3.) Thus.14). the formula becomes 2( 2 ω)bca1 ···ap = 2 (p + 2)(p + 1) p [b c ωa1 ···ap ] = j=1 R[bcaj d ωa1 ···|d|···ap ] − T d [bc| d ω|a1 ···ap ] d ω|a1 ···ap ] = −p(−1)p R[bca1 d ωa2 ···ap ]d − T d [bc| = p(−1) ( p [b T ca1 | d (B. the Poincar´ lemma will not hold: e 2 = ◦ = 0. most equations involving differential forms will still be best expressed in terms of the torsion-free derivative. University of Chicago Press. CVS $Revision: 1. Because this expression is much less elegant (and much less useful) than d2 = 0. In fact.1. For more general forms. References [1] Robert M.Typeset November 16. the torsion is treated as a set of 2-forms T σ . but it still does not require a preferred metric. 2005. Wald.

48:393–416. Rev. Acad. M. Mod. General relativity with spin and torsion: Foundations and prospects..Typeset November 16.35 $Date: 2005/11/16 22:24:09 GMT 18 ´ [2] Elie Cartan. R. Rotation. G. a [3] Peter G. Phys. (Paris). and Supergravity. D. Plenum (New York). volume 58 of NATO Advanced Study Institutes Series: Series B. Hehl. 1922. P. Cosmology and Gravitation: Spin. Kerlick. W. CVS $Revision: 1. [4] F. Sci. Physics. Torsion. Nester. 2005. and J. C. editors. 174:593–595. Sur une g´n´ralisation de la notion de courbure de Riemann e e et les espaces ` torsion. Bergmann and Venzo de Sabbata. . Von Der Heyde. 1980. 1976.