Nuclear Physics B 387 (1992) 215—235 215

North-Holland

Axion hair and dynamical torsion from anomalies

M.J. Duncan ‘, N. Kaloper 2 and K.A. Olive ~
School of Physics and Astronomy, University of Minnesota, Minneapolis, MN 55455, USA

Received 23 March 1992

(Revised 17 July 1992)
Accepted for publication 17 July 1992

Recent studies have shown that black holes can indeed have hair in gravitational theories
beyond minimal Einstein gravity. The hair is associated with the axion and dilaton of string
theory, whose gauge and gravitational structure is richer than the conventional, minimally
coupled theory. The axion in this case also has a dual role as a dynamical degree of freedom of
the torsion. In this paper we suggest that there is a large class of models for which a torsion may
be transmuted to an axion, and black holes develop hair. Indeed we show that even a model as
simple as QED on a torsionful background space-time gives rise to axionic hair. We discuss this
phenomenon and its implications.

1. Introduction

For many years it was thought that black holes could have no hair other than
that dictated by electromagnetic and gravitational gauge invariance [1,21.The only
fields that a black hole could support were those whose source was the electric and
magnetic charges, and the mass and spin of the black hole itself. No other sources
gave rise to stable hair. However, the “no-hair” theorem was purely in the context
of Einstein gravity. Recent work motivated by string theory [31has shown that
quite a few examples do in fact exist. This is because string theory introduces extra
couplings of the fields which are “topological” in origin. Hairy solutions exist for
the axions and dilatons of string theory. Static classical axion hair has been
demonstrated for Kerr (spinning) [4,5] and dyonic Reissner—Nordström [5—7]black
holes. There is also quantum axionic hair which only manifests itself in an
interference experiment similar to that of the Aharonov—Bohm effect [8]. Dilaton
hair can be supported by static spherically symmetric holes [5,7,9]. There are
probably many more solutions yet to be found. In this paper we shall focus on the
axionic hair.
E-mail: duncan@umnhep
2 E-mail: kaloper@umnhep
E-mail: fvs6308@umnacvx

0550-3213/92/$05.OO © 1992 — Elsevier Science Publishers B.V. All rights reserved

216 Mi. Duncan et al. / Axion hair

Classical axionic hair needs a pseudoscalar source. It should therefore come as

no surprise that it is found for black holes which carry odd parity quantities such as
spin [4,5,10] or dyons [5—7].The couplings arises as follows: String theory contains
a three-form gauge field strength H~Awhose only dynamical content turns out to
be an axion [11]. The Bianchi identity for the three-form, which must be imposed
for anomaly cancellation, links it to the Hirzebruch density for the gravity and the
Chern class for the gauge fields. These then become the sources for the axion. The
point we wish to explore in this paper derives from the fact that the three-form can
be regarded as torsion, and so we might say that in a sense the black hole has a
stringy torsion hair. It is then quite reasonable to ask if torsion in a general model,
i.e. one not motivated by string theory, is in any way related to an axion hair.
The hair so obtained need to be distinguished from the standard Maxwell hair
of the Kerr—Newman black hole family. Its appearance is provoked entirely by the
axion-anomaly non-minimal coupling and mathematically comes from the need
that the solutions cancel the non-homogeneous source terms in the equations of
motion. It should immediately be observed that in the solution there are no new
integration constants, corresponding to new charges. Indeed, all the integration
constants which would normally appear in the external field solution have been set
equal to zero since they are incompatible with a regular horizon, when the matter
fields are (pseudo) scalars, as the standard no-hair theorem of GR tells us. As a
consequence of this, the charges associated with the axion hair (and likewise the
dilaton in the context of string theory motivated models with non-metric connec-
tion) is completely determined by the mass, angular momentum and gauge charges.
The parameter space uniquely determining a black hole solution with such hairs is
the same as in the standard GR except now the solutions follow different orbits in
it. It is in this spirit that these hairs are dubbed “secondary” [12] as opposed to the
Maxwell hairs which are “primary” since they do introduce new physical parame-
ters in the solutions.
The role of torsion in general relativity has always been something of an enigma
[13]. In the Riemann—Cartan approach to space-time torsion is as much part of the
geometry as the metric. If we insist that geometry is the basis for a theory of
gravitation then we must confront the consequences of torsion. These are at first
sight peculiar. First, when coupled to matter in a minimal way it is only sensitive to
the spin density of fermions. Secondly, it does not propagate and is therefore
inexorably tied to the matter. The equivalence principle, which is really the basis
for gravitation, is invalidated by the presence of a background torsion. Classical
torsion then seems physically extraneous. String theory, on the other hand,
circumvents these impotencies as its richer structure provides non-minimal cou-
plings, whose origin is the afore-mentioned anomaly cancellation. The net effect is
that torsion is identified with an axion.
In sect. 2 we briefly review how the three-form of string theory can be viewed as
a torsion, and how it is also equivalent to an axion. In sect. 3 we shall demonstrate
 M.J. Duncan et al. / Axion hair 217

that a simple model, QED on a background space-time with torsion, can also
transmute the torsion to an axion. The key to the phenomenon in this case is the
recognition that the model possesses the familiar Adler—Bell—Jackiw anomaly for
the axial current. In order to maintain the classical nature of the torsion we have
to constrain it by decoupling it from the anomaly. This then results in an axion. In
sect. 4 we discuss axionic hair from this model. It has recently been shown that at
the one-loop level of curved space electromagnetism there is an anomaly in the
chiral current of the photon. It then contributes a second-order effect to our
model. In sect. 5 we examine this and also discuss the corrections to photon
propagation. Finally we pull the results together and discuss some implications of
our work.

2. The axion and torsion in string theory

We will begin our discussion by describing how the three-form of string theory
has a dual role as torsion and as an axion. The gravitational multiplet in string
theory consists of a metric tensor g~,an antisymmetric two-form with components
B~and a scalar dilaton field. Anomaly cancellation [3] requires that Lorentz and
gauge Chern—Simons terms be added to the field strength of B so that we may
define a three-form H dB + (a’/8K)(flL
= Q~)which then implies the Bianchi
—

identity
a
dH=—Tr(RAR-FAF) (1)
where K2 =

It is well known that the three-form H of string theory can be identified with
torsion to at least o(a’) in the string tension. The gauge-invariant kinetic term for
this field is ~f H A * H. Further details on three-form quantization are to be
—

found in ref. [14]. It was recognized very early by Scherk and Schwarz [15] that the
o(a’°) action for string theory is equivalent to ordinary gravity with torsion To ~‘.

zeroth order in the string tension the string gravitational action can be written as

= f{~R — ~ dDx, (2)

where we are employing the conventions of Gross and Sloan [16]. We can redefine
the connection to absorb the three-form. It is readily seen that the connection
F~,,+ (K2/3)’/2g~H~~produces a curvature scalar R(F)
= R(F) =

* Actually, full string gravity would necessitate non-metricity of the connection, ~ —. g~,AB~o’ to
account for the dilaton 0~.We are ignoring the dilaton in this paper although everything we say can
be generalized to the case where it is re-introduced.
218 M.J. Duncan eta!. / Axion hair

~K2H,IVAH~~~A plus a total divergence. The action (2) is therefore equivalent to

—
the Einstein—Hilbert action for a torsionful connection as the divergence is only a
surface term. Thus the (con)torsion is fully specified by the field HILVA.
Interestingly enough, this association of torsion with the three-form also holds
to the first order in a’ [17]. Neglecting the dilaton and gauge contributions these
higher terms can be written most generally as

S = ~ dDX{~(RabcdRQbCd + A 2)
1R abRab + A2R

— + A
3Rat~HacdHb~+ A4RHabcH)

+ ~ ( HabcHa~HbdfHcef+ ASHaCdHbCdHbmnHamn + A6( H~Ha6c)

2)

~ VHaCdVHb
7a b cd

In the case of the heterotic string the Lorentz Chern—Simons terms in the
definition of Habe are also of first order in a’, and would contribute the additional
terms

12~ R dHHf — ~ K2HUbCHadCH

6JHCef.

The coefficients A1,.. , A7 can be left unspecified due to the field redefinition
.

ambiguities present in the theory [17,18]. For example, the theory is invariant
under the following transformation:
8gab a’(aR ab + bRg~~ =+ CHacdHbCd +

= a’( eV~H’~
6~). (4)

As a concrete example we now restrict ourselves to the action of Gross and Sloan
[16] where the fields are defined in such a way that the coefficients are A1 —4, =

A2 1, A3 —2, A4 ~, A5 —3, A6 ~, A7 0. In this representation we have a

= = = = = =

manifestly unitary background field action which matches string amplitudes.

On the other hand, the most general action in terms of curvature squared terms
in the torsionful theory is [17,19]

S = -~ fv’~~
dDx (~){ ~lkabcd a~ + A2RabcdR

+AiRabcdR + A4Ra6R + A5RabR + A6R}. (5)

 Mi. Duncan et a!. / Axion hair 219

No choice of A1,. A6 yield directly the action (3). However, upon a suitable field
. .,

redefinition (a suitable choice of the parameters a,. ., e above) what we can show .

is that the action (5) becomes that of (3), with the above-listed values of the
couplings A1,.. . , A7. This occurs when the values are chosen to be

- 9+2V~ - 18+V~
A1=— 18 ‘ A2=

- 9+V~ - 2%/i—27+108e
A3= 18 ‘ A4= 18
2
A
- 99 + 10v~+ 108e - 684 + 40V’~ 216D
— — 4DV~+ 27D
5=— 18 A~= 144

27+4y~ 162+4V~—27D
a= 72 ‘ b=— 576

2~—27 54+4V~—27D
= 216 ‘ d = — 1728. (6)

Thus we see that even to o(a’) string gravity is equivalent to torsionful gravity
containing curvature squared terms. Details of this calculation follow closely ref.
[19], with the exception that our result points out that even the manifestly unitary
amplitude matching action after a field redefinition can in fact be torsionised.
When we account for the dilation we would also have to add suitable non-metric
pieces.
It is also well known that the three-form H in four dimensions contains only
one propagating degree of freedom (an axion). To see this and the couplings of the
axion to gravity and the gauge sector, it is useful to define the dual of H, Y * H: =

Y~=~ (7)

The Bianchi identity (1) becomes

a’ ~ ~
1~”
+F~”F~’). (8)
~(Ra action
Using the constraint (7) we can write the o(a’) ~R’~”(2) in terms of the dynamical
— ~— ~ ~‘

axion mode as

1 1
= J{~R — 2 — 8K ~ +F~FPA)}

xf~d~x (9)

where ~P = —3~ 31)4.

220 M.i Duncan et a!. / Axion hair

Thus we can see clearly the relationship between the contorsion tensor and the
three-form H and the axion and its coupling to the Hirzebruch and Pontryagin
densities R A R and F A F.

3. QED on curved space with torsion

We have seen that in string theory the torsion can be viewed as an axion. The
point we wish to emphasize is that this relationship is not unique to string theory,
bu can occur in even the most simple of models. To demonstrate this we consider
QED on a curved space-time background. This is a very useful model to discuss as
it is simple enough to be examined in detail. We shall adopt the most minimal
scheme for the gauge and gravitational couplings, the torsion then being included
in the manner detailed in the appendix (which contains the relevant formulae we
use throughout the paper). Our goal is to clarify the rOle of torsion once we take
into account the one-loop corrections to the theory. We also wish to determine the
effective action to leading order in the Planck mass M~ ~ Our starting
=

point is the action for a massless, charged Dirac fermion interacting with the
electromagnetic field and the torsionful gravity,

= - ~ d4x, (10)

where the full covariant derivative is D,~ ieA,~with e2 4~~aQED. The

= — =

gravitational covariant derivative is that of (A.14). Immediately we can separate

out the torsion dependence. The spin connection splits into two pieces wab,.~ =

Wabji + Kahp. The former is the connection for a torsionless space, related to the

familiar Christoffel symbol, and the latter is the contorsion containing the other
degrees of freedom of the full Riemann—Cartan space-time. The action (10) can
therefore be rewritten,

sb = - (D~y~)~ d4x

+efAy~~ d4x+ ~f~{yc ~ab}~K 4x, (11)

6~ d

with the gravitational covariant derivative on a torsionless space. To this we

add the Maxwell action for the electromagnetic field. We choose to define the field
strength using the usual exterior derivative to maintain conventional U(1) gauge
invariance. The field strength is F dA implying the Bianchi identity dF 0. The
= =
 M.J. Duncan et al. / Axion hair 221

field strength is thus independent of the torsion, and the action for the Maxwell
field is just the standard:

d4x= _~fFA *F.

Sem= ~

This is a minimal description of electromagnetism. It is possible to construct a

non-minimal description for which the field strength does depend on torsion at the
classical level [20]. To do so would require a modification to the gauge transforma-
tion properties of A and unless the torsion were in addition given a “mass”, this
would be in conflict with experimental results [21]. However, this takes us off our
minimal approach.
As we can see, with our minimal prescription, the only appearance of the
contorsion in QED occurs in the last term of (11). Moreover, there is an identity
for the Dirac matrices {yC cab) 2E~~dydyS.
= This implies that it is really only the
totally antisymmetric part of the contorsion that couples to matter. This is one of
the two irreducible parts of the contorsion. You can easily check from Young
tableaux that the full twenty-four components of the contorsion reduce as 6 ~ 4 =

4 e 20, the former being totally antisymmetric in three indices. The torsion can be
similarly reduced, and from (A.8) it is seen that both the totally antisymmetric
parts are related by T[abc] = ~2K[abc]. Once we project out this part we can use it
to define a three-form T (l/3!)Ta
=

6ce’~A e’~’ A ec. The c-symbol from the Dirac

matrix identity suggests that it would be neater to work with the dual one-form
S * T whose components are Sd (1/3!)e~dTQbC.Taking all this into account,
= =

we find that the last term of (11) is more elegantly written

4x= _~fsA *j5. (12)

~ d

The interaction of torsion and matter is seen to be a vector field coupling to the
fermion’s axial current. At present the vector field’s dynamical properties are not
known to us. We could, of course, force it to be a gauge potential (an axiphoton)
[22] although nothing in the theory suggests this. What we intend to show is that its
dynamical properties are dictated by the anomaly in the axial current.
The minimal coupling of the torsion to the background space-time is given in
(A.12). The form of this is dictated by the Riemann—Cartan theory. As we
discussed above, we should decompose the contorsion into its irreducible parts
Kabc = ~EabCdSd + Kabc, these being the 4 and 20 representations, respectively.
The torsion part of the gravitational action turns out to be

=
222 Mi. Duncan et a!. / Axion hair

and S is the combination (A.13) using the “~abc~ Therefore the complete gravita-
tional action has three distinct pieces:

5G ~f~(R+S) d4x+ ~fSA *S. (13)

The first is the standard Einstein—Hilbert action for the classical background
metric. The only place the 20 appears in the full theory is in the second term. As
the K are then simply algebraic with no dynamics their equations of motion are
trivial. As they appear quadratically in (13), they may be integrated out of the path
integral. In what follows we ignore them. It is the field that is the object of
interest and whose dynamical character we wish to clarify.
At this point it is expedient to examine the classical equations of motion. We
treat the fields A~, ~t, ~li and S~as independent degrees of freedom. The Maxwell
equations for the electromagnetic field are unchanged as it does not couple to
torsion classically, dF 0 and d * F * j, where the fermion current is j~
= = =

e~/iy’~’
if’. The fermion satisfies the modified Dirac equation,

— ~S~
iy~Y~~ifi 5ifi0,
= (14)
1y~’y
where = — ieA,~,and we find that classically the torsion acts as an axial
source. From variation of the metric we find the Einstein equations, ~ K2T~~, =

for the background space-time. The total stress—energy tensor has contributions
from the electromagnetic, Dirac and torsion fields, T~ T~+ T~ + T~.The first
=

is the same as in the torsionless theory by minimal, i.e. no, coupling. The second is
the symmetric stress—energy tensor for an on-shell fermion and does have a
contribution from the torsion,

T~ = F~AF~A
—

T~= — ~ ~ — ~ + ~5(y~)y5~.

The third is the correction to the Einstein equations from (13) due to the torsion,
here written in the form of a stress tensor,

T~= ~ (15)

Classically, the equation of motion for the torsion is S ~K2j5. This would result
=

in an algebraic substitution [13] of the torsion by the axial current in eqs. (14) and
(15) above. Additionally, an immediate consequence of (14) is that the axial
current is classically conserved. This would then imply that d * S 0 for the =
 Mi. Duncan et a!. / Axion hair 223

torsion. Equivalently we might at first think that we could just integrate S,~out of
the path integral to leave an effective four-fermion interaction —afj5 A j5 with *

strength a (3K2/16). This is a dimension-six operator suppressed by two powers

=

of the Planck mass, and would yield the same classical equations of motion.
This approach will lead to confusion when we come to construct the quantum
theory. The source of this is obvious. The path integral is invariant under the
global symmetry i/i exp(iJ3y5)i/i. However, the Noether current j5 is not con-
—+

served as it has an anomaly due to one-loop quantum effects. We have to take this
into consideration before we integrate out the torsion because otherwise the
theory will be inconsistent. The root of this inconsistency lies at the heart of
semiclassical quantum gravity [23]. We do not have a full quantum theory of
gravity, and this is reflected in our use of the (extended) Einstein—Hilbert action
(13). We wish to be able to perform matter loop corrections in the background of
the geometry. These corrections would then influence the gravitational equations
of motion by introducing new terms into the stress tensor (source for the metric)
and the spin density (source for the torsion). Unless we are careful restrictions
from the gravitational side of the equations, such as a divergenceless G~,can be
at odds with the source side. In the case of the stress tensor corrections we are
obliged to add counterterms to the original action in order that the final result
satisfies certain compatibility conditions, such as being divergenceless [24]. This is
similarly true for the torsion. In the semiclassical theory torsion appears alge-
braically and quadratically in (13) and so we should be able to integrate it out.
What would remain are new couplings in an effective theory suppressed by powers
of M~.However, before we do this we should ensure that it is first compatible with
the quantum corrections.
Before we determine the appropriate counterterm for this we shall discuss the
anomaly. As is well known, the axial current, though apparently conserved at the
tree level, is not conserved at the one-loop level [25]. There are contributions to
the anomaly from both the gauge and gravitational fields. Given the fermion action
(10) the anomaly is

e2 1 ——

d * i5 = — ~—~FA F — 96
2tr(R AR) G(A, ~) (16)
6a, etc. Torsion can contribute to
where by “trace”
the anomaly we the
through mean tr(R AR)density
Hirzebruch Rab Afor
R the connection ~i. The anomaly is
=

more commonly expressed in components as

e2 1 —

V~j5 ~ = — l
92ir2R~/3~

where the duals are *F ~%/ ~ = and ~ as =

usual. The Hirzebruch densities for the two connections are related through
224 M.i Duncan eta!. / Axion hair

tr(R A R) tr(R A R) + d tr(K A R + K A DK + 4K A K A K). If the axial current is

=

coupled to a gauge field then the expression for the anomaly can be converted
from G(A, ~) to G(A, w) by adding finite local counter-terms to the action [26].
The gauge variation of the counterterms would exactly compensate for the exact
piece. Torsion is not a gauge field but nevertheless we shall find that our
counterterm exhibits a similar behaviour to leading order in the Planck mass.
At the classical level the axial current is conserved. As the equation of motion
identifies torsion with the axial current we could conclude that d * S 0. This =

results in a conserved “torsion charge” Q5 f * S. However, at the quantum level

=

the axial current is anomalous. As we do not have the means to identify all the
quantum effects related to torsion it would be inconsistent to include only the
contribution from the axial anomaly. Therefore we add a counterterm which
maintains d * S 0 at the quantum level. This is compatible with the classical
=

result and it is gauge invariant. We can then proceed with integrating out the
torsion. We can easily accomplish this by introducing a delta functional into the
path integral. Gathering together the torsion terms from (12) and (13) the relevant
integral is

f~S8(d*S)exP(if{~-~SA *S_~SA*i5j) (17)

and by a standard representation of the delta functional as a path integral can be

written as

f~s~ exp(if [~_~s

A * S — A * ~ + s})
cI~d*

where ‘~Pis essentially a Lagrange multiplier. This is2dP

now integrated
-~K2i5.Weeasily. First
are then
—
of
able
all re-express
to we completethethe squaresinbythe
exponent defining
above 5’
as S 4ic= —

3 1 2K2 3 1
—S’A *5’— — — d’15A *dcj— —K~j5A *j5— —K2d~A~
4K2 2 3 16 2

Next we note that the jacobian of the transformation from S to 5’ is unity. Hence
we can integrate out the field 5’ completely. What remains is the integral over the
field ~ Rescaling this according to D (3/2K2)’/2~ the torsion path integral (17)
=

is

f~exP(if1_~d~A
*d~- ~d~A *j5— ~j5A ~ (18)
 MJ. Duncan eta!. / Axion hair 225

This is recognized as the kinetic term for a scalar and the standard couplings of a
axionic pseudoscalar to fermions. The axion decay constant is f~, (3K2/8Y 1/2 In =

addition to the dimension-six four-fermion interaction the constraint has intro-

duced a dimension-five operator. We can also redefine the fermion fields by an
axial transformation to rewrite the effective torsion action as

1 1 1
exp if[_ ~ A *d~+
1~G(A, cv) - A * (19)

to leading order in the Planck mass. We conclude that QED on a torsionful space
is equivalent to QED on a torsionless space coupled to an axion. In this sense the
torsion has become dynamical due to anomalies.
At this point we would like to compare this with some of the non-minimal
mechanisms for dynamical torsion to be found in the literature. As mentioned
above dynamical torsion does not occur in the purely classical theory [13] and all
we get is a four-fermion interaction. Previous studies have made torsion dynamical
by force. The authors of ref. [21] introduced into the action a counterterm
proportional to f(S AA A F). This was prompted by their calculation of the
photon polarization tensor in the background of the torsion. For electromagnetic
gauge invariance they then had to impose S d4. However this direct substitution
results in the wrong sign for the scalar kinetic term in (13), and a negative energy
density in (15). Another way to impose dynamics is to treat S~as if it were a true
axial vector field [22]. In this case the anomalies would violate gauge invariance
[25] and it requires the introduction of Wess—Zumino terms. Moreover this
requires a different gravitational action than the minimal (13) in order to have a
standard kinetic term for the massless vector. Even more exotic [22] is treating the
dual of S as the field strength of a two-form field F dB, or even making T the
=

potential for a four-form field strength [27] M dT. The former is superficially
=

similar to the very special form of torsion in string theories and the latter is really
more relevant for considerations of the cosmological constant problem [28]. Again,
both require departure from the minimal model. Finally we mention that some
authors [20] have adopted a non-minimal, i.e. non-vanishing, coupling of torsion to
electromagnetism at the classical level. As before, this needs a priori restrictions
on the form of the torsion such as an exact 5, or even T” ea A d~for the 20, in
=

order to ensure gauge invariance. Other models of this general type can be found
in ref. [29]. We summarize by saying that the common feature of these proposals is
the neglect of the anomaly in the chiral current, and the imposition rather than
derivation of the correct kinetic energy terms.
We have seen that when we take QED on a background space-time and
minimally introduce torsion that the effective theory below the Planck scale
exchanges the torsion for an axion. This is due to the constraint on the torsion
226 MJ. Duncan et a!. / Axion hair

arising from the chiral anomaly. Let us now examine the consequences. The new
equations of motion for the axion and the Maxwell field are

e 2 1
f~,(d*d4)=—~FAF+
2tr(RAR),
4ir
2
d*F=*i————~d4AF.
1 e (20)
f,~,2ir

The fermion vector current is conserved as is required for gauge invariance. The
axial current is still anomalous, d * i5 G(A, w). However, from the axion equa-
=

tion of motion we see that there is now another conserved current present,
P =j5 +f
4 d~whose origin is the constraint.

4. Axion hair on black holes

We can pursue this minimal QED model with torsionful gravity and analyse the
expected structure for black holes. Due to the axion’s coupling to the Hirzebruch
density in (20) the black hole geometry can act as a source for a classical axion
background, as can any electromagnetic field of the black hole through the other
density. It is straightforward to derive the coupled equations of motion for the
axion, electromagnetic and gravitational fields,
2
V2’I~=——G=
“~- f1 in’ 1 2ç R a13p.~’*R~f3~L~~ e 2~ F~ *F~
IT’~~T iT

(V1*F~
2 iT 2ç ~. i*’t’)

R~ ~g~R
— = K2(T~ + T~+ T~), (21)

where the stress tensors are the usual ones for a scalar and the electromagnetic
field, with the addition of a term from the axion coupling to the Hirzebruch
density (the gauge density does not contribute). This latter stress tensor is

1
T~~?=
pP ‘,n 2~ V~’~IVP,~*R
‘~‘ p~iAv
‘IT
 M.J. Duncan et a!. / Axion hair 227

In the limit that the axial sources G are small, and i~f is similarly small, the
first-order solution to the metric is the Kerr—Newman geometry. For this geometry
we can evaluate the density directly [5],

—A2 cos2O)2 4r2A2 cos29

~2iT2
— 2 3(r2 +A2 cos2O)4—
G — QeQm(1

+ 2(Q~—Q~)rA(r2—A2cos2O) cos 0
3(r2+A2 cos20)4

1 8G~M2rAcos 0(3r2 —A2 cos2O)(r2 — 3A2 cos2O)

+ (22)
16iT2 (r2 +A2 cos2O)6

where A is the angular momentum per unit mass and Q is the electric or magnetic
charge of the black hole. Thus the limit of a small axial source corresponds to
small charge or angular momentum. In this case (22) can be expanded to give [5]

G 2iT2
_~_JQ~m(~ \ 3r6
10A cos2O) cos
+ 2(Q~—Q~~)A 3r5 ü

1 (24G~M2A cos 0
+
16ir2 r7

It can now be seen that (21) implies the presence of axion hair for all rotating or
dyonic (both Qe and Qm non-vanishing) black holes, in any torsionful theory of
gravity. The solution for the axion hair is, to the first few terms in the multipole
expansion,

q~(r, 0) = ~{[4i~1~1r (i_ ~ + ...)IP

0(cos 0)
2 3
—
A
192iT2MGNr2(2(1+
5 2GNM
r +
18G~,M
Sr2 )_~~))
+ . . .IPI(cos 0)
QeQm~42

— 6iT2GNMr3 + ... P
2(cos 0) + (23)
228 M.i. Duncan et a!. / Axion hair

5. Second-order effects

In sect. 4 we have seen that there exist classical solutions for the axion, gauge
and gravitational fields which exhibit axion hair under a variety of circumstances.
We now wish to discuss two effects that arise to next order as corrections to these
solutions. The first is the correction due to the bosonic chiral anomaly and the
second is the rotation of the plane of polarization of a photon.
The bosonic chiral anomaly [30] arises for electromagnetism on a curved space
at the one-loop level. The current which is anomalous is the chiral current for the
photon,

P~ = — ~ EpPAPAF,

or P * (A A F). This is not a Noether current, unlike the chiral current of the
=

fermion of sect. 2. It does not directly arise from an invariance of the action under
a transformation of the vector potential, but is instead related to the duality
transformation of the field strength. Moreover this current is not gauge invariant,
although the charge obtained by integrating the zeroth component over space is.
This charge equals the difference between the number of right- and left-polarized
photons. The anomaly in this current is

1 — —

d*P=—~—~tr(RAR) (24)

to be compared with (16). However, by definition d* P F A F. =

The torsion only couples to the fermion classically. Therefore the arguments of
sect. 3 are uneffected. The torsion is still transmuted to an axion. Consider the
case of a chargeless Kerr black hole. The gravitation itself induces a gauge density
in the exterior [31]. Therefore the axion source equation is replaced by

f
4(d*d~) = +
1 2
96iT
( —
2aQED
iT ) tr(RAR)

from the first of equation (20). This may also be regarded as the two-loop
contribution to the anomaly. In general, when there is also a classical exterior field
such as in the Kerr—Newman case, all that this anomaly does is to multiply the
Hirzebruch density term by the factor (1 2aQED/iT). —

Now let us look at the second of equation (20). A physical consequence of the
axionic coupling to photons is the rotation of the plane of polarization of a beam
of light. This can be seen from our eq. (20),
= 2A(V~4)~ (25)
 Mi Duncan et a!. / Axion hair 229

The geometrical optics approximation is where we take the wavelength of the light
to be much smaller than the typical distance over which the space-time curves [32].
The light has (complex) polarization vector fa(x) as measured in the tetrad basis
at point x, and is normalized to unit, faf” 1. In the gauge V,~A~’ 0 we find that
= =

in this approximation the polarization satisfies

(k.V)f”= fa_ _Ac~~Cd(V~,b)kcfd

where 1 is the affine parameter of the light’s path and k is the wavevector. The
polarization is parallel propagated along its path if the axion is absent. If the axion
is present this is no longer so. For the case of the Kerr black hole with small spin
we can treat the space-time as Schwarzschild to leading order [4]. Suppose the
light is travelling on a radial null geodesic (r, t(r)) of fixed angle 0 and ~. Then
k~(r) w(r) (1, n) and f”(r) (0, f(r)) where n is the unit vector in the radial
= =

direction of propagation, (~.f) 0, and w(r) is the frequency measured by an

=

observer at radius r. As the axion is static the above reduces to an equation for the
polarization along the path of propagation,

d
—f(r)=—fl(r)xf(r)
dr

where the rotation vector is 11 A(af’/ar)n. (A similar effect in a cosmological

=

context, the polarization of the light from distant galaxies, has been discussed in
ref. [33].) In general this is an incredibly small effect due to the weakness of the
hair. However, it may be appreciable for primordial black holes. We hope to
return to this in detail in a later publication.

6. Discussion

In this paper we have discussed the relationship between axion and torsion in
string theory and demonstrated that a similar relationship can be established in
standard General Relativity in presence of fermions with anomalous axial current.
As a result, the effective theory describing dynamical degrees of freedom in these
models contains black hole solutions with exterior pseudoscalar hair, generated by
the non-minimal couplings of the axion through the anomaly. We also pointed out
one way that this hair manifests itself, by affecting the polarization of photons
emitted from the vicinity of a black hole. Though it is tempting to associate the
presence of the axion with torsion, this is not true in general. For example there
may be several axions present in the theory (there is always at least two in string
theory). One linear combination of these axions will determine the propagating
degree of freedom of torsion. However this is the linear combination which gives
230 M.i Duncan et a!. / Axion hair

rise to black hole hair only if the ratio of gauge anomaly coupling to gravitational
anomaly coupling is the same for all axions. Hence the real source of the external
secondary hairs lies in the appearance of the nonminimal couplings.

We would like to thank A. Vainshtein and I. Khriplovich for valuable conversa-

tions. M.J.D. would also like to thank G. Segrè for many discussions on the nature
of torsion. This work was supported in part by DOE grant DE-ACO2-83ER-40105.
K.A.O. was also supported in part by a Presidential Young Investigator Award,
and N.K. received assistance from a Doctoral Dissertation Fellowship of the
University of Minnesota.

Appendix A

GRAVITY WITH TORSION

The most efficient way to incorporate torsion into general relativity and mani-
fest its behavior is to consider the Cartan equations. At each point on the
space-time manifold we construct an orthonormal basis of four vectors, and their
one-form duals, for the tangent space. The use of such “free-fall” bases greatly
expedite calculations. We denote the tetrad of vectors by Ea and their dual
one-forms by e’~ e~”dx~’. The riemannian metric is
= ~ or equivalently =

gPV flabEP.Ebv in terms on their components where flab is the (flat) tangent space
=

metric ~. We can choose the “direction” of this basis independently at each

neighborhood on the manifold. Choosing a different orthonormal basis at some
point is equivalent to a local Lorentz rotation e”(x) Lab(x)eb(x). To construct a —*

covariant exterior derivative we must introduce the spin connection one-form

which transforms as ii~(x)~L”(x)w~(x)L ~(x) (dL~(x))L’~(x).The covari- —

ant derivative D is then defined. Acting on any q-form with tangent space
(Lorentz) indices it is

DQ”b =dQ”’b +WaCAQCb + ... ~ A~— ... (A.1)

Covariant constancy of the tangent space metric, D’Ilab 0, implies that the spin =

connection is antisymmetric in its indices ~ab ~‘~ba~ Similarly we can show that
=

the Levi-Civita symbol is also covariantly constant, De~b~d0. The spin connection =

one-form’s components on the coordinate manifold or the tangent space are given
by w~ w~dx~ wg~e’~.
= We raise and lower indices with g~ and flab on the
=

coordinate manifold and tangent space respectively. Using the components of the
tetrad e~and E we can convert from one basis to the other.
* Throughout the paper we use the conventions of Misner et al. [321.On the tangent space the metric is
= diag(— 1, + 1, + 1, + 1) and the Levi-Civita symbol, ~*bcd’ is such that ~O123 = + ~
 Mi. Duncan et a!. / Axion hair 231

The bars over the connection and derivative defined above denote that so far
we are considering the most general case. Ordinarily, one sets the torsion to zero a
priori and restricts the theory. However, as we have argued in this paper, such a
restriction can ignore some interesting physics. In the general case the torsion and
curvature two-forms are defined by

= de” + i~i~
A eb,

= di~+ i~A ~, (A.2)

both of which transform covariantly under local Lorentz rotations by construction.

From the definition of the covariant derivative (A.1) there immediately follows the
two Cartan equations and the Bianchi identity for these geometrical objects,

De” =

DT’~= A

=0. (A.3)

The spin connection is more fundamentally defined [32] with respect to the
familiar tensor covariant derivative acting on the basis vectors by VE~ 0 Eb. =

When expressed in terms of the tetrad of one-forms this is

a.....~
~ a
— A a_—a b
— ~e* — ~

From this and (A.2) we find that the components of the torsion, T” = ~T~dx~ A
dxv, are related to the affine connection through

~ —2e~”F[~V]. (A.5)

This is the identification of the torsion as the antisymmetric part of the affine
connection that we usually meet in text books (recall that it is only this part of the
affine connection that behaves as a tensor under general coordinate transforma-
tions).
As the torsion equation (A.2) is linear in the connection we can separate the
connection into a part totally absent of torsion and a part wholly containing it,
= w~+ Kb” respectively. The torsionless part of the connection is defined by

t’ De” = 0. (A.6)
de’~+ w~A e

The general operation of this “torsionless” covariant derivative is as in (A.1) with

the bars removed. Once again, covariant constancy of flab under the action of D
232 M.i Duncan et a!. / Axion hair

implies antisymmetry of Web, and implies the covariant constancy of the Levi-Civita
symbol. Eq. (A.6) can be used to express the torsionless connection solely in terms
of the tetrad and its derivatives,

Wabc = ~(Cabc + Cbca — Ccab),

where ~ = 2E~E~ a11,e~”1. We already know that the affine connection for a
torsionless space is the familiar Christoffel symbol. From the definition of the
torsionless spin connection ~ —wg~.e~
= we therefore find from the analogue of
(A.4):

a — aj~uf A a
w6~—e* b\p~~J b ~

The last two expressions are also seen to be equivalent by explicit expansion of the
Christoffel symbol in terms of the tetrad.
The part of the connection containing the torsion is the “contorsion” one-form
K~ KbaCec. It is antisymmetric in the first pair of indices, Kab —K6~.Clearly
= =

from (A.2) and (A.6) we have T” K~A eb, from which we can express one in
=

terms of the other,

T~ = —2K~~1,Kabc = 4(Tcab — Tabe — Tbca). (A.8)

Both have twenty-four independent components in principle. We can now separate

out the torsion and the usual Riemann connection in any expression. From (A.4)
and (A.7) we see that the whole affine connection is, [~ {~}+ K~. Antisymme-
=

try in the lower two indices reproduces (A.5) in conjunction with (A.8). However it
is not the case that symmetry in the lower two indices only yields the Christoffel
symbol, as there is also a contribution from the (con)torsion.
Just as we defined the torsionless covariant derivative (A.6) and the connection
(A.7) we can use them to define the torsionless (Riemann) curvature two-form
Rb” =dw~+ o~A w~. By construction this satisfies the Bianchi identity DRg 0. =

The torsion curvature can be expanded:

~ (A.9)

from (A.2). The advantage of doing so is that the contorsion dependence is made
explicit. The components of the curvatures are 4R~CdecA e’~ ~RgpPdx1*A
= =

dx” and similarly for R~.Expressed fully in a coordinate basis they give the
familiar curvature tensors in terms of the affine connections,

= E~’e,~RgpP~ = + FQ~F~ (i* — ~

 M.i Duncan et a!. / Axion hair 233

and similarly for the torsionless curvature R~pPwith the Christoffel symbol. The
components are antisymmetric in the first and last pair of indices. Thus there are
really only two contractions we can perform on the curvatures. The first gives the
Ricci tensors ~~‘pP and
= A~ R~Av• The second gives the curvature scalars
=

= and ~ g~T!’pP. The only remaining piece of information to be

=

extracted from the Cartan equations (A.4) is the identity for the torsionless
curvature R~A e~’ 0. Translated to components it is the familiar cyclic identity
=

R~~d} 0 or equivalently R(~AP] 0 for the Riemann tensor. This implies that the
= =

torsionless Ricci tensor is symmetric. The torsionfull quantities do not possess

these symmetries.
We now come to the action for gravity including torsion. We choose to define
this in terms of the torsionful curvature as

SG~J~v~d4x (A.10)

where K2 8iTGN. An alternative expression for this in the notation of forms

= * is
found to be

50 -__~fR,~i,A *(e”Ae”)

+ DKac +Kac A K~)A *(e” Aeb), (All)

where we have used the properties of the Hodge dual. The second term of the
latter is just a total derivative as the tetrad and the Levi-Civita symbol are
covariantly constant under D,

f(DKabA *(eaAeb))=fd(KabA *(eaAeb))=f (KabA *(eaAeb)),

a surface term by Stokes’ theorem and thus irrelevant to the equations of motion.
The first term in (A.11) is the usual Einstein—Hilbert action for a torsionless space.
* In our notation the Hodge dual for a p-form derives from

= ~

This then leads to the useful identity

(~~Aeb) A * (~Aeb) =

where the canonical volume form is = A A e2 A e3 = ,/i~ d4x. The b is the permutation
symbol; in particular ~ah~d =
234 Mi Duncan et a!. / Axion hair

The last term contains all the torsion. The gravitational action can therefore be
written

5G= ~f~(~+~) d4x; (A.12)

where the torsion part ~ is

~1=K~VK”~ _Kl~~,~K~
= 5A
~ — 4T~AT~+ 4’~~’5T, (A.l3)

which is not very illuminating at first glance. However, as we explain in the body of
the text, dynamical symmetries of the (con)torsion greatly simplify this. As there
are no derivatives acting on the torsion in this action we might conclude at first
that it does not propagate.
Finally we mention that the gravitational covariant derivatives acting on spinors
and their conjugates are

Dif~ d~
= — ah(T~~bif1,

iJ~=dif~+~Wabtfr, (A.14)
1~ ~i[y’, ~b] The
=

and similarly for the torsionless case where i~r” 7a are a

representation of the Dirac matrices and are covariantly constant.

References

[11 W. Israel, Phys. Rev. 164 (1967) 1776;

B. Carter, Phys. Rev. Lett. 26 (1971) 331;
J.D. Beckenstein, Phys. Rev. D5 (1972) 1239;
R. Price, Phys. Rev. D5 (1972) 2419, 2439;
D.C. Robinson, Phys. Rev. Lett. 34 (1975) 905
[21B. Carter, in General relativity, an Einstein centennary survey, ed. SW. Hawking and W. Israel
(Cambridge Univ. Press, Cambridge, 1979);
W. Israel, Commun. Math. Phys. 8 (1968) 245;
R.M. Wald, Phys. Rev. Lett. 26 (1971) 1653;
ED. Fackerell and JR. Ipser, Phys. Rev. OS (1972) 2455
[3] M. Green, J. Schwarz and E. Witten, Superstring theory (Cambridge Univ. Press, Cambridge, 1987;
D. LOst and S. Theisen, Lectures on string theory (Springer, Berlin, 1989)
[4] BA. Campbell, M.J. Duncan, N. Kaloper and K.A. Olive, Phys. Lett. B251 (1990) 34
[5] BA. Campbell, N. Kaloper and K.A. Olive, Phys. Lett. B285 (1992) 199
[6] BA. Campbell, N. Kaloper and K.A. Olive, Phys. Lett. B263 (1991) 364;
K. Lee and E. Weinberg, Phys. Rev. D44 (1991) 3159
[7] A. Shapere, S. Trivedi and F. Wilczek, Mod. Phys. Lett. A6 (1991) 2677
[81 M.J. Bowick et al., Phys. Rev. Lett. 61(1988) 2823
[9] G.W. Gibbons, NucI. Phys. B207 (1982) 337;
G.W. Gibbons and K. Maeda, Nuci. Phys. B298 (1988) 741;
 Mi Duncan eta!. / Axion hair 235

C.G. Callan, R.C. Myers and M.J. Perry, NucI. Phys. B311 (1988) 673;
D. Garfinkle, G.T. Horowitz and A. Strominger, Phys. Rev. D43 (1991) 3140
[10] BA. Campbell, M.J. Duncan, N. Kaloper and K.A. Olive, Nucl. Phys. B351 (1991) 778
[11] S. Giddings and A. Strominger, NucI. Phys. B306 (1988) 890
[12] 5. Coleman, J. Preskill and F. Wilczek, Mod. Phys. Lett. A6 (1991) 2353; Phys. Rev. Lett. 67 (1991)
1975; Princeton preprint IASSNS-HEP-91-64
[13] F.W. Heyl, P. von der Heyde, GD. Kerlick and J.M. Nester, Rev. Mod. Phys. 48 (1976) 393
[14] C. Burgess and A. Kshirsagar, Nucl. Phys. B324 (1989) 157;
J.D. Brown et al., NucI. Phys. B328 (1989) 219
[15] J. Scherk and J.H. Schwarz, Phys. Lett. B52 (1974) 347
[16] D. Gross and J.H. Sloan, NucI. Phys. B291 (1987) 41
[17] R.R. Matsaev and A.A. Tseytlin, NucI. Phys. B293 (1987) 385
[181 A.A. Tseytlin, Phys. Lett. B176 (1986) 92;
R.R. Matsaev and A.A. Tseytlin, Phys. Lett. B191 (1987) 354
[19] Z. Bern, T. Shimada and D. Hochberg, Phys. Lett. B191 (1987) 267;
Z. Bern and T. Shimada, Phys. Lett. B197 (1987) 119
[20] 5. Hojman, M. Rosenbam, M.P. Ryan and L.C. Shepley, Phys. Rev. D17 (1978) 3141;
W.-T. Ni, in Precision measurements and fundamental constants II, ed. B.N. Taylor and W.D.
Phillips (National Bureau of Standards 617, 1984)
[21] V. Dc Sabbata and M. Gasperini, Phys. Lett. A77 (1980) 300; Phys. Rev. D23 (1981) 2116; Phys.
Lett. A83 (1981) 115
[22] A. Aurilia and E. Spallucci, Phys. Rev. D42 (1990) 464
[23] ND. Birrel and P.C.W. Davies, Quantum fields in curved space (Cambridge Univ. Press, Cam-
bridge, 1984)
[24] R.M. Wald, Commun. Math. Phys. 54 (1977) 1
[251A.Yu. Morozov, Soy. Phys. Usp. 29 (1986) 993;
L. Alvarez-Gaumé, in Fundamental problems of gauge field theory, ed. G. Velo and A.S.
Wightman (Plenum, New York, 1986)
[261 W.A. Bardeen and B. Zumino, NucI. Phys. B244 (1984) 421;
CM. Hull, Phys. Lett. B167 (1986) 51
[27] A. Aurelia, H. Nicolai and P. Townsend, NucI. Phys. B176 (1980) 509;
M.J. Duff and P. van Nieuwenhuizen, Phys. Lett. B94 (1980) 179;
P.G.O. Freund and M. Rubin, Phys. Lett. B97 (1980) 233;
M. Henneaux and C. Teitelboim, Phys. Lett. B143 (1984) 415;
E. Witten, in Proc. 1983 Shelter Island Conf. on Quantum field theory and fundamental problems
of physics, ed. R. Jackiw, N. Khuri, S. Weinberg and E. Witten (MIT Press, Cambridge, MA,
1985);
J.D. Brown and C. Teitelboim, Phys. Lett. B195 (1987) 177
[28] SW. Hawking, Phys. Lett. B134 (1984) 403;
M.J. Duff, Phys. Lett. B226 (1989) 36;
M.J. Duncan and L.G. Jensen, NucI. Phys. B336 (1990) 100
[29] PB. Yasskin and W.R. Stoeger, Phys. Rev. D21 (1980) 2081;
H.-J. Lenzen, Gen. Rel. Gray. 17 (1985) 1137;
P. Baekler, M. GOrses, F.W. Heyl and J.M. McCrea, Phys. Lett. A128 (1988) 245;
R.T. Hammond, Gen. Rd. Gray. 23 (1991) 973
[30] AD. Dolgov, VI. Zakharov and lB. Khriplovich, JETP Lett. 45 (1987) 651;
AD. Dolgov, lB. Khriplovich and VI. Zakharov, NucI. Phys. B309 (1988) 591;
M. Reuter, Phys. Rev. D37 (1988) 1456;
AD. Dolgov, lB. Khriplovich, A.I. Vainshtein and VI. Zakharov, NucI. Phys. B315 (1989) 138;
VI. Zakharov, Phys. Rev. D42 (1990) 1208;
AD. Dolgov, S.V. Kostyuk and I.B. Khriplovich, Novosibirsk preprint IYF-91-31 (1991)
[31] M. Reuter, DESY preprint DESY-91-049 (1991)
[32] C.W. Misner, K.S. Thorne and J.A. Wheeler, Gravitation (Freeman, San Francisco, CA, 1973)
[33] SM. Carroll and GB. Field, Harvard—Smithsonian Center for Astrophysics preprint 3212 (1991)