This is the first of two papers dealing with Killing-Yano tensors. In this
paper, K illing-Y ano tensors are classified according to their valence. The
cases of valence 1, 3 and 4 are trivial or almost trivial. Canonical line
elements of metrics adm itting Killing-Yano tensors of valence 2 and
rank 4 are derived w ithout any further restrictions. (The case of rank 2
will be treated in p art II of this series.) Furtherm ore, we give a relativistic
analogy of the classical Kepler motion in a plane if there is no force com
ponent orthogonal to this plane. This analogy demonstrates the way in
which a K illing-Y ano tensor may be considered as being related to the
angular momentum of a particle.
1. I n t r o d u c t i o n
The purpose of this and a further paper (referred to as II) is to examine the
properties of K illing-Yano tensors which were first introduced by Yano (1952) from
a purely m athem atical point of view. We classify Killing-Yano tensors, derive all
line elements adm itting a K illing-Yano tensor and present physical interpretations
for situations described by these tensors. There are several motivations for per
forming an investigation of this kind.
(a) Nearly all interesting and most striking properties of the K err (vacuum) space-
time are closely related to the existence of the Killing tensor first investigated by
Penrose & W alker (1970), whereas the significance of the Killing tensor itself is
still unclear. This may be a reason why ‘there are aspects of the K err metric which
need understanding’, as Chandrasekhar (1979) formulated. A remarkable property
of the Killing tensor of the K err metric is th a t it proves to be a product of a skew-
symmetric tensor with itself which satisfies the Killing-Yano tensor equation, as
pointed out by Floyd (1973) (see Penrose (1973)), Rudiger (1973) and Collinson
(1976). Moreover, the existence of a Killing-Yano tensor characterizes the Kerr
space-time in a certain sense.
(b) The standard application of symmetries (especially isometries) in general
relativity seems to have reached a stagnation point, as Collins (1979) pointed out.
For this reason, we choose the Killing-Yano tensor, which is not related to
symmetries of the space-time, as the starting point for our discussion, and, indeed,
we find metrics w ithout any symmetry adm itting a Killing-Yano tensor. On the
other hand, a class of Killing-Yano tensors leads to the existence of Killing vector
[ 361 ]
Downloaded from http://rspa.royalsocietypublishing.org/ on May 17, 2018
The following two generalizations of the Killing vector equation have become of
interest in general relativity and have been discussed extensively since the
discovery of the fourth constant of motion in the K err space-tim e by Carter (1968).
In the following, let Mbe a smooth manifold of dimension n w
metric gab.
(a) An r-form field (1 ^ r ^ n )fai....ars a^ to be a Killing
valence riff
^ (a 1f a 2)a3...ar+1 = (®^ - 1 )
where Va denotes the metric connection. This definition goes back to Yano (1952)*
According to (2.1), the ( r —l)-form field
iai...ar- 1' f
is symmetric and proves to be a Killing tensor called the associated Killing tensor.
Therefore
= ( 2 . 2)
^ (efd) cba
then f abcd is proportional to the alternating pseudotensor 7jabcd, and the Killing-Yano
tensor equation ensures th a t the proportionality factor is a constant. The associated
Killing tensor equals the space-time metric up to a constant. Thus, the space-tim e
is not restricted b y / a&crf.
The remaining two cases are discussed in the following two sections.
3. K il l in g - Y ano t e n s o r s of v a l e n c e 3
L e t/abc be a Killing-Yano tensor of valence 3:
dfc)ba6 .
We may dualize f abc according to
6/*° = P r ^ f m n r -
This enables us to give an alternative form to the Killing—Yano tensor equation
K a b = t t f b * - f Z f * m9ab
K ab = - e S 2(gab- e t atb(3.2)
and possesses the eigenvalue zero to the eigendirection /*® and the three-fold
degenerate eigenvalue — eS2.
Conversely, a Killing tensor adm itting a system of eigenvalues of this kind can be
characterized by the dual vector/*® of the Killing-Yano tensor itself as follows:
A symmetric tensor possessing the eigenvalue zero to the non-null eigenvector
ta and the three-fold degenerate eigenvalue — with is a Killing tensor
iff the vector field / * “ = Sta is a confirmed Killing vector and a gradient field.
Proof. A symmetric tensor field with the assumed system of eigenvalues is given
by (3.2) which implies
Vfcifw = - 2eSg„bVa)S + 2f* V J * ,. (3.3)
First, assume th a t K ab is a Killing tensor. Because the eigenvalues of Killing tensors
are constant along the associated non-null eigendirections, we obtain
V«£ = e(tmVmS )ta(3.4)
Equation (3.4) inserted into (3.3) and contracted by eaebec with arbitrary ea ensures
that/*® is a conformal Killing vector with conformal factor tmVrnS. Moreover, from
(3.4), we deduce th a t/* is a gradient field such th a t
V J t=
Conversely, we assume (3.5) to be valid. Then, because of (3.4), / * “ is geodesic.
Equations (3.4) and (3.5) inserted into (3.3) ensure th a t K ub is a Killing tensor.
Moreover, f abc is a Killing-Yano tensor if the dual vector satisfies (3.5). If we
insert the definition o f/ * “ into (3.5) and contract with yabcd, we obtain / 2 V
= rjdcba an(l consequently the Killing-Yano tensor equation fo r /a6c. We collect our
results.
Downloaded from http://rspa.royalsocietypublishing.org/ on May 17, 2018
ds 2 = e(dx4)2— S 2(x*) y
This may be considered as being a generalization of the first integral of the geodesic
equation in the Friedm ann cosmos discussed by von Laue (1931) and later Honl
(1949) which can be understood as being the angular momentum of a particle moving
freely on the surface of an expanding sphere with radius S. This interpretation can
be made more precise in the following way. The position-vector field ra in a flat
space-time can uniquely be characterized (apart from a rescaling and from a
translation of the origin) by the fact th a t it is a conformal Killing vector field and a
gradient field. In this sense, / * “, which possesses these properties because of (3.5),
may be considered as being a four-dimensional position-vector field. Then, the first
integral —S 2 | | p x j | 2 is the square of the angular momentum lab = 2
4. K i l l i n g - Y a n o t e n s o r s of v a l e n c e 2
This section deals with Killing-Yano tensors of valence 2 denoted by f * b in
accordance with Dietz & Rudiger (1980). In the following, we first briefly enumerate
some definitions concerned with bivectors which we frequently need in this section.
The dual of a bivector Vab given by V*b = Fmn satisfies F*6* = - V ab. Vab is
said to be (anti-) self-dual iff Vab = ( ± ) i V*b. A scalar product for bivectors is defined
by 2 V- W := Vmn Wrn w hich implies, for self-dual bivectors, 2 VamVbm = V 28b.
Furthermore, a real bivector Vub is of rank 4 iff ^ 0, otherwise it is of rank 2.
Downloaded from http://rspa.royalsocietypublishing.org/ on May 17, 2018
V J & = 0, (4.1a)
which is equivalent to
Vc/*6 = Vabcm£m> (4.16)
Za ’ =
The K illing-Yano tensor equation for f*b implies an analogous equation for f ab:
v j* , = 2 (4.2)
Sometimes, it seems to be of advantage to combine these two real equations for
f ab and fihnto
i the single complex equation
Vc 0 ab = 2 {-lbga]c+ iyabc
for the self-dual bivector field 0 :ab= f ab -I- if*b. The associat
the form
= & / * ”» = + i (** + **)9a>, (**)
where the bar denotes complex conjugation.
For later use, we give the integrability conditions for the K illing-Yano tensor
equation in two equivalent forms: I f we differentiate (4.1) again, we obtain, by
virtue of the Ricci identity, after a short calculation (see Tachibana 1968),
+ = 0 (4.5)
in terms of 0b.a
We decompose (4.5) into the following system of equations:
+ 0 V W * - 0, (4.6o)
2(V [J6l + iV,J f , ) - lCabmn 0™“ - \R4>ab - 0, (4.66)
2V(o£w- = 0, (4.6c)
V , J m = 0, (4.6d)
which is equivalent to (4.5).
Downloaded from http://rspa.royalsocietypublishing.org/ on May 17, 2018
fa b = Im
®ab = t ( D ab + iD*b).(4.8)
This tetrad is unique up to transform ations of the form
Consequently, the associated Killing tensor possesses the two two-fold degenerate
eigenvalues (Im i/r)2 and —(Re \Jr)2.
Conversely, a real symmetric tensor K ab with such a system of eigenvalues can be
w ritten as a product of an algebraically general real bivector with itself (see, for
example, Dietz & Rudiger 1980).
(ii) An algebraically special bivector can be cast into the form
i fa b = 2 k a ( m b ) - ^ b ] ) ’
and correspondingly
®ab = 2h a m bV (4 -10)
If we insert (4.8) and (4.10) into the integrability condition (4.6a), we immediately
obtain the following result.
The Weyl tensor of a space-time adm itting an algebraically general (special)
Killing-Yano tensor belongs to the Petrov-Pirani type D (type N).
I t is possible to obtain a further consequence of the integrability conditions, if we
assum e/^, to be of rank 4 such t h a t/- /* ^ 0. We split up equation (4.6c) into its
real and imaginary parts. If we contract the real p art w ith /* 6d and insert R cmf * dm
from the imaginary part, we obtain / * m6V(a£TO) = 0. I t follows th a t
V(a£&) = 0
because o f/-/* ^ 0. Therefore, £a is a real Killing vector field, and the integrability
condition (4.6c) simplifies to Rm(a0 b)m = 0 which can be solved for :
K ab = - 2 (^ )-X 0 a m ® mb + i E 9ab>
Downloaded from http://rspa.royalsocietypublishing.org/ on May 17, 2018
r - = K “mijr
is a second real Killing vector and th a t £ a nd 7ja gener
acting transitively orthogonally.
The vector field
Sa- = ®am£m (4-11)
is a gradient field, since 2 S a= Vai/r2, which is orthogonal to the
£a and 7]a.V[a$ 6) = 0 and equation (4.3) imply
bW ym^ab u 0 u= 0 = L
= 9 = LgKab.
So, the definition of 7]a leads to [£, 7j]a = 0. Furtherm ore, rja obeys the Killing vector
equation because
^(aVb) = \dj^Kab —£,mVm= 0,
where we have used the Killing tensor equation for K ab. Moreover, (4.3) and the self
duality of 0 ab ensure th a t 0 ab and therefore K ab remain invariant under the action
of 7ja.T he isometry group generated by £a and 7ja acts orthogonally transitively if a
family of 2-surfaces exists which simultaneously are orthogonal to the independent
Killing vectors £“ and 7ja. Now, by an argument due to Sommers (1973), it is easy to
prove th a t and 7ja are independent Killing vectors if and only if Re x]r and Im xjr
are independent functions such th a t d(R e \Jr) A d (I m ^ ) # 0.
For this case, a family of 2-surfaces is determined by the real and imaginary parts
of the (complex) gradient field S a defined by (4.11). These 2-surfaces are, in fact,
orthogonal to £a and 7ja because —0 = S m7im is valid.
(5.1 a)
II
r:
0
v
q
*
The next step is to gauge the chosen tetrad field with the free transform ations (4.9)
which imply
(p, p)-> (pexipd, p e x $ ( - 6 )),(5.3a)
(r, 7 t)-> (rex p (i^), 77exp ( —i<j>)),
for the Newman-Penrose coefficients p,
equation (5.1 6 )into real and imaginary parts and obtain
d (R e ^ ) = ^Jr{pk —pn
), di ( I m ^ ) = }Jr(nm —Tm),
where we have used (5.1c).
Suppose th a t d( Re \Jr)^ 0 ^ d (Im ^ ). Then a suitable choice of ^ in (5.3 6)ensures
r = — 7T. The 1-form d(Re^") may be spacelike, null or timelike according to the
value 1, 0, or —1 of e= sgn(pp). For e ^
p = ep\ e= 0 is equivalent to pp —0. For the latte
means of the inversion (4.9c), and the choice of 6 remains free. So, for all values of e,
we have found p =ep. If d (Re \jr) or d (Im) \jf vanishes identically, the
and r = —itsi again possible but the transform ations (5.3a) or (5.3 6 ) are
available, respectively.
Thus, for all cases, the gauge
p= ep, t = —77, (5.4)
can be attained, the transformations (5.3a) or (5.36) still being available if one of
the quantities p ,r or e vanishes. The Killing-Yano tensor equations (5.1)
k = (T ~ 0 — v — A, (5.5
dip = ip[p(ek —n) + T(m + m)],(5.56)
pip= pip}= (5.5c)
We proceed to evaluate the integrability conditions where we insert the gauge (5.4)
and make use of the Killing-Yano tensor equation (5.5). Moreover, the only non
vanishing curvature terms are ¥/2, 0 n and —4/1 as we pointed out in § 4. Then,
equation (4.66) is equivalent to
D ( ^ r ) + 5(^p) = ip p [2T+((5.6a)
oc + ft
13 Vol. 375. A
Downloaded from http://rspa.royalsocietypublishing.org/ on May 17, 2018
where for j = 1, 2£(1)a = and £(2)a = rja, and the explicit form of the coefficients
may be read off if we compare (5.16) with (5.17). Then the Killing equations may be
w ritten as
2 9y(a 0 6)aii) + 2 9v(a
The xy- and ^-com ponents of this equation yield a homogeneous system of linear
equations for dxa(-1) and dxb(j) whose determ inant is MQ —N P which cannot vanish
because the metric determ inant is proportional to MQ —N P. Therefore, and 60)
are independent of x.For e# 0, the uy- and av-eom
are also independent of u,which may be shown in the same way.
ww-component of the Killing equation and the relation between and b{i) give us
(after applying du and after using the equations (5.5 and (5.14 6,^)) again a
homogeneous system of linear equations for du, a(j) and with non-vanishing
determ inant MQ —N P . Therefore, aw and 6(;) depend only on y and v. So, we are able
to use the transform ation (5.15 6) which is still available so th a t the coordinates y
an d /o r v are adapted to the Killing vectors.
W ith these preparations, it is easy to derive the final form of the line elements.
We tre a t four separate cases.
(i) p#0 j=. t .Re ijr and Im \Jr are then independent functions, and £a, y
independent Killing vectors adapted to the coordinates v and y, respectively.
Moreover, we find the relations
N = -\i/r\~ 2B, M
(2 —e2) Q — |^ |~ 2 A,
if we compare (5.16) with (5.17). Finally, we fix the transform ation (5.15a) by
u' = Re \Jr and x' — Im \Jr. We insert these relations into (5.13) and obtain
du\ 2 / dx \ 2
ds2 = — (ule
A{u)J + \R 0r)/
— (u 2 + x2)~x {[B(x) (dv —u 2 dy)](dv + d
Ci 11
—2
( 1 —e2) - ■A ( u ()dv + x 2dy).
For this line element, the null-basis forms and the coordinate forms are related by
1 . u
P= A(u), fi =
(u2+
1 u —ix
T —— B(X), 7TT,
(
So, the Killing-Yano tensor equations (5.1) are satisfied for arbitrary functions A
and B.
(ii) p#0 = r. This implies B = 0 so th a t d ifr = E A
are functions of u. The Killing vectors become parallel and the coordinate v may be
adapted to £a. Then, by comparison with (5.176), we find
N = 0, (2 —e2) Q=
and
F = F(x, y), M = M (x,y), = C
A 2
d.s2= —|^r|2 (dv + Cdy)
W2
(\n /I
+ (F d x )2+ (M d y )2+ 2( 1(dv
- +C ^
+( u2
+ l 2)-1e[A(u) (dv -f 2lC(x, y) dy)]2
c| 7/
- 2(1 - g2) A (u)(dv+ 2 y) dy)
Downloaded from http://rspa.royalsocietypublishing.org/ on May 17, 2018
+ i dv[D{x,y){dx C{x,y))%]\, / ? = - a .
Therefore, the Killing-Yano tensor equation (5.1) is satisfied for arbitrary functions
A, C and D.
(iii) p= 0 # r. For this case, we find A=
Im ^r are functions of x. Again, the Killing vectors are parallel and 0/0y may be
adapted to £a.
If we proceed in the same way as before, we obtain the line element
r = - J_ B(x) 7T = — T
V2 1 V + x2)F
(du C(u,v))* 1
e = -2 7 2 (^2+ ^2)_i (0mU(w, v))-1J0m - « v) (0„ C(u, #))*] j ,
D(u, v)
{du C(u,v))Y
7 = 2^2 (^2+ x2)- 1 (0wC'(w,v))-1Je0. + 0„[i)(«.») a G(u, »))*]!;
D(u, v)
k —ix \
OC= - 272 (k2+ x2)-i (x) - i B(x) , /? = —a.
k 2 + x2j
Downloaded from http://rspa.royalsocietypublishing.org/ on May 17, 2018
376 W. Dietz and R. Rudiger
Conversely, the K illing-Y ano tensor equation (5.1) is satisfied for arbitrary
functions B, C and D.
(iv) p = 0 = t . The functions A and Ba
nd th
Bei/r and Im \jr are non-vanishing constants. Furtherm ore, (5.11) implies th
basis forms are hypersurface orthogonal so th a t in (5 .1 2 )Y = $ 0 = P . Then, the
equations (5.146, c) and (5.14 a, d) e n s u r e
E =E
(u,v);Q= Q(u,v), =
If we absorb the constant \rjr\ into the functions E, F, M and Q, we find the line
element
ds2 = —e[(E(u, v) d u )2—(
- 4 ( 1 - e2) E(u, v) d uQ(u, v) dv - (F(x, y)
The basis forms and coordinate forms are related by
7 ^ (1 + e2) k= Qdv + eEdu, ^ 2-(l +e2)n =
So, the non-vanishing Newm an-Penrose coefficients are given by
e = - 272 (EQ )-1( du )Q
{u,v),
7 = —272 {EQ)*1 (edu Q(u,v) + dv E(u,v)
a = - 2^ ( F M ) - 1( - d xM{x,y) + idvF(x,y)),
A= - 2^ ( F M ) - i( d x M (x,y) + idy F(xj y
The Killing-Yano tensor equation (5.1) is fulfilled for arbitrary functions E, F, M
and Q.
Additionally, we wish to comment on the param eter e introduced to determine
whether d(Re^") is timelike, spacelike or null. The param eter e has a direct geo
metric meaning if d (R e ^ ) 0^ as in the cases (i) and (ii). B
(iv) are characterized by the vanishing of d (Re \jr), e becomes meaningless in these
two cases. However, e appears in these cases again because we treated all cases
simultaneously as far as possible and, therefore, applied the basis transform ation
(5.10) which depends on e. For th a t reason, e also appears in (iii) and (iv), formally
making the direct correspondence of the two cases (ii) and (iii) more transparent. 6
6. G e o m e t r i c a l i n t e r p r e t a t i o n o p K i l l i n g - Y ano
TENSORS OP RANK FOUR
Contrary to the case of a Killing-Yano tensor of rank two, there does not seem to
exist a simple interpretation of Killing-Yano tensors of rank four. In this chapter,
we shall demonstrate that, under conditions to be specified, one property of the
classical Kepler motion may be rediscovered in a metric adm itting a Killing-Yano
tensor of rank four.
Consider a (Newtonian) gravitational field with vanishing force component
orthogonal to the equatorial plane. Then, a test particle started within this plane
Downloaded from http://rspa.royalsocietypublishing.org/ on May 17, 2018
Space-times admitting Killing—Yano tensors. I 377
and tangent to it will stay in the equatorial plane, and the angular momentum
vector will always be orthogonal to this surface. A four-dimensional analogy of this
situation would be th a t there exists a timelike hypersurface which is totally
geodesic (see Schouten 1954) and which has the property th at, for any geodesic in
with tangent field p a, the vector la =is orthogo
prove th a t the following statem ent is true.
Let f *b e a K illing-Y ano tensor of rank four with the following properties.
(а) The function Z : = |/ * /* is not a constant.
(б) There exists a timelike hypersurface of the family of hypersurfaces
Z = const, such th a t/* 5 is simple on ^ i.e.
(c) On 34*,
R m, r J mnf * r‘ = 0,
is valid.
n, 34?is totally geodesic.
These conditions are fulfilled for di?(0) /d x = 0 on the surfaces = 0 in the cases (i)
and (iii) of § 5.
Let na = Z a( —Z mZ m)-i be the unit normal field on the hypersurfaces Z = const.,
where Z a —V aZW
. e prove th a t Vbwa = 0 on From (4.16) and (4.2)
Za =f}mZ™(6-1)
The real p art of the integrability conditions (4.66) and (4.6c) gives
^ d a = - l ( C abmnfmn+ W a
which implies, by using (4.16), th a t
VftVaZ = - * / * n iC ^ sr
From (4.6a), we obtain
<6 -2 >
which merely expresses the fact th a t the Weyl tensor is of Petrov-Pirani type D.
The bivector identity WamWb*m = \gabW ' W* applied to (6) gives
f amf * mb= 0, (6.3)
on ^ and (c) implies, by using (6.2) and (6.3), th a t Cmbrsf rs c c f mb on 2^ . So, we
have V6VaZ = 0 on and, consequently, 0 on Therefore, the second
fundamental form on vanishes, which c h a ra c te riz e s^ as being totally geodesic
(Schouten 1954).
Furthermore, the vector la =f*mPm is orthogonal to 34?
This may be read off from the dualized version of
Z lalb]= 0,
on 34? if p mZ m = 0: dualizing equation (6.4) yields
Pl aKmZ™ = 6,
which is, in fact, fulfilled by virtue of (6.1) and (6.3).
Downloaded from http://rspa.royalsocietypublishing.org/ on May 17, 2018
378 W. Dietz and It. Rudiger
Obviously, this property of la supports, a t least for the specialized case of a particle
with worldline in , the interpretation of la as an angular m om entum vector.
The authors wish to thank Tevian D ray for reading the m anuscript.
R eferences
C arter, B. 1968 P hys. Rev. 174 , 1559.
C arter, B. & M cL enaghan, R . G. 1979 P hys. Rev. D 19 , 1093.
C handrasek h ar, S. 1979 A n in tro d u c tio n to th e th e o ry of th e K e rr m etric an d its p e r tu r b a
tions. In General relativity (ed. S. W . H aw k in g & W . Israel). C am bridge U n iv e rsity Press.
Collins, C. B. 1979 Gen. Rel. Grav. 10 , 925.
Collinson, C. D. 1974 Tensor n . s . 28, 173.
Collinson, C. D. 1976 In t. J . theor. P h ys. 15 , 311.
D ietz, W . & R udiger, R . 1980 Gen. Rel. Grav. 12 , 545.
F loyd, R . 1973 T he d ynam ics of K e rr fields. P h .D . thesis, L ondon U n iv ersity .
H au ser, I. & M alhiot, R . J . 1976 J . m ath. P hys. 17 , 1306.
H au ser, I. & M alhiot, R . J . 1978 J . math. P h ys. 19 , 187.
H onl, H . 1949 A n n in P hys. 6, 169.
H u g h sto n , L. P ., Penrose, R ., Som m ers, P . & W alker, M. 1972 Com m uns. m ath. P h ys. 27 , 303.
H u g h sto n , L. P . & Som m ers, P . 1973 C om m uns. m ath. P hys. 32 , 147.
K ashiw ad a, T. 1968 N a t. Sci. R ep. Ochanomizu XJniv. 19 , 67.
von Laue, N. 1931 Die L ich tfo rtp flan zu n g in R au m en m it zeitlich v eran d erlich er Krvim-
m ung n ach der A llgem einen R e la tiv ita tsth e o rie , S . B . preuss. A ka d . W iss., p. 123.
L etelier, P . S. 1979 G
en.Rel. Grav. 11, 367.
N ew m an, E . & Penrose, R . 1962 J. math. P h ys. 3, 566.
Penrose, R . & W alker, M. 1970 C om m uns math. P hys. 18 , 256.
Penrose, R . 1973 A n n . N . Y . A cad. Sci. 224 , 125.
R udiger, R . 1973 ErhaltungsgroySen fu r freie Teilchen in der A llgem einen R e la tiv ita tsth e o rie ,
U n iv ersity of W urzburg.
Schild, A. 1967 L ectures on general re la tiv ity . I n R elativity theory and astrophysics (ed.
J . E hlers). P rovidence, R hode Isla n d : A m erican M athem atical Society.
Schouten, J . A. 1954 Ricci calculus (2nd edn). B erlin: Springer.
Som m ers, P . D. 1973 Thesis, U n iv ersity of T exas a t A ustin.
S tep h a n i, H . 1978 Gen. Rel. Grav. 9, 789.
T achib an a, S. 1968 Tohoku math. J .2 0 , 257.
T achib an a, S. 1969 Tohoku math. J . 21 , 56.
Y ano, K . 1952 A n n . M ath. 55, 328.