Novel results in the semi-Riemannian theory of

conjugate and focal points

D. Szeghy
Eötvös University, Hungary, Pázmány Péter stny 1/C
November 16, 2010

Abstract
In the theory of conjugate and focal points there is a lot of di¤er-
ences between the Riemannian and the semi-Riemannian case. As Helfer
showed, there can be an entire segment in the set of focal points along
a geodesic. In this paper a simple condition on the curvature tensor is
given which yields, that the above case cannot happen, however we don’t
know, whether the focal points are isolated or not under this condition.
Moreover an example is given, where the focal points are unstable under
a geodesic variation, and this is also di¤erent from the Riemannian case,
where the focal points are stable under a geodesic variation.

MSC 5350, Keywords: semi-Riemannian, focal point, conjugat point

1 Introduction
The theory of Jacobi …elds, conjugate and focal points is a fundamental part of
Riemannian geometry on which an essential part of global results is based. As
recent results show the extension of this theory to the semi-Riemannian geom-
etry presents stunning phenomena. For example the conjugate points are not
necessarily isolated along a geodesic, or there can be even an entire segment
of a geodesic which consists of conjugate points as Helfer showed [H]. P. Pic-
cione and D. V. Tausk showed in [P-T] that for any closed H (0; 1) we can
construct a semi-Riemannian manifold (M; g) and a geodesic : (0; 1) ! M
in it, such that any point (t) ; t 2 H is conjugated to (0). In this paper it
will be shown, that the focal points in general are not stable under a geodesic
variation. These facts point to the di¤erence between the Riemannian case and
the semi-Riemannian one in the conjugate and focal point theory.
First we recall the de…nition of the conjugate and of focal points by means of
Jacobi …elds and of F -Jacobi …elds, in case of a semi-Riemannian submanifold
F . Then using a classical method, which connects the Jacobi equation (see
de…nition 1), with the so called Morse-Sturm equation (see de…nition 6), we
present the example due to Helfer where an entire segment is contained in the

1
set of conjugate points along a geodesic. The above classical method was given
in [Wh] for the Riemannian case and extended later to the semi-Riemannian one
in [H]. After Helfer’s example di¤erent conditions will be given concerning the
curvature tensor which guarantee that there are only isolated conjugate points,
or that there is no open set in the set of focal points along a geodesic. Then a
construction will be given for a Lorentzian manifold M and a semi-Riemannian
submanifold F in it, such that there is a 1-parameter family of geodesics s (t)
starting orthogonally to F where the following hold: There are focal points
on the geodesic 0 (t) ; t 2 R, but there are no focal points on the variational
geodesics s (t) ; t 2 R for every s 2 (0; 1].

2 Jacobi-…elds and Helfer’s example

Before we start let us recall the de…nitions of the Jacobi and F -Jacobi …elds,
moreover those of the conjugate and focal points, (see e,g [B-E-Ea] p. 328 and
[B-E-Ea] p. 447).

De…nition 1 Let (M; g) be a semi-Riemannian manifold, R its curvature ten-

sor, : [0; 1] ! M a geodesic segment and X (t) 2 T (t) M; t 2 [0; 1] a smooth
vector …eld along (t). The vector …eld X (t) is called a Jacobi …eld if it
satis…es the Jacobi-equation:
0 0
R (X (t) ; (t)) (t) + X 00 (t) = 0; for every t 2 [0; 1] , (1)

where 0 (t) = r 0 (t) 0 (t) and X 00 (t) = r2 0 (t) X (t) are the covariant derivatives.
Now if there exists a non-trivial Jacobi …eld X (t) along (t) such that X (0) =
X (1) = 0 then the points (0) and (1) are said to be conjugate to each other.

De…nition 2 Let (M; g) be a semi-Riemannian manifold, F M a semi-

Riemannian submanifold and : [0; 1] ! M a geodesic segment, which starts
orthogonally to F , thus (0) 2 F; 0 (0) ? T (0) F with respect to the metric
tensor g. Let 0 (0) : T (0) F ! T (0) F be the Weingarten endomorphism of

T (0) F . Now if X (t) 2 T (t) M; t 2 [0; 1] is a Jacobi …eld which satis…es also
the initial conditions

1 X (0) 2 T (0) F; (2)

0
2 X (0) + 0 (0) (X (0)) 2 N (0) F;

then X (t) is called an F -Jacobi …eld, where

def
N (0) F = v2T (0) M j g (v; w) = 0 for every w 2 T (0) F

is the normal space of F at (0). If there is a nontrivial F -Jacobi …led X (t)

along (t) such that X (1) = 0 then the point (1) is called a focal point of
F along the geodesic (t).

2
Remark 3 In case when F = fxg then the above de…nition of F -Jacobi …elds
reduces to that of the Jacobi …elds with initial condition X (0) = 0 and the
de…nition of focal points to the de…nition of conjugate points.

In order to present the constructions of the above mentioned examples, let

us recall some basic facts.

Remark 4 Every F -Jacobi …eld is induced by a geodesic variation where the

variational geodesics are orthogonal to F , i.e. if : I J 3 (s; t) 7! (s; t) 2 M
def
is a smooth map where I; J R and s (t) = (s; t) is a geodesic for every
…xed parameter s; which starts orthogonally to F; then T (0;t) @s 2 T 0 (t) M is
an F -Jacobi …eld along the geodesic 0 (t). Moreover every F -Jacobi …eld can
be obtained this way.

See e.g. [B-E-Ea]. Moreover note that:

The Jacobi …elds de…ned along a geodesic segment form a 2n-dimensional

vector space, where n = dim M .
The Jacobi …elds X (t) de…ned along a geodesic segment (t) with the
initial condition X (0) = 0, form an n-dimensional vector space
The F -Jacobi …elds de…ned along a geodesic segment also form an n-
dimensional vector space.

Let (M; g) be a semi-Riemannian manifold, F M a semi-Riemannian

submanifold. If x 2 F and v 2 Nx F then let v : I ! M denote the geodesic
for which v (0) = x; v0 (0) = v, i.e. the geodesic starting from x with velocity
vector v. Then for any two Jacobi …eld X (t) ; Y (t) along v (t) the following
holds:

g (X 0 (t) ; Y (t)) g (X (t) ; Y 0 (t)) const:; (3)

where X 0 (t) = r v0 (t) X (t) ; Y 0 (t) = r v0 (t) Y (t) with respect to the Levi-Cività
connection, (see the closely analogous Riemannian case [B-C] p. 228). Let
Y (t) = v0 (t) in the above equation (3), since v0 (t) is a Jacobi …eld and
r v0 (t) v0 (t) = 0, then the above equation reduces to:

g (X 0 (t) ; 0
v (t)) const:
0
for any Jacobi …eld X (t) along v (t). Moreover r 0
v (t) v (t) = 0 implies that
d 0 0 0
dt g (X (t) ; v (t)) = g (X (t) ; v (t)) const; thus
0
g (X (t) ; v (t)) = c1 t + c2 ; with c1 ; c2 2 R.

Now if X (t) is an F -Jacobi …eld then equation (2) yields g (X (0) ; v0 (0)) = 0.
Hence c2 = 0 holds in the above equation, i.e. for any F -Jacobi …eld X (t) the
following holds
g (X (t) ; v0 (t)) = c1 t: (4)

3
Now if X (t0 ) = 0 for some t0 6= 0 then equation (4) yields that c1 = 0; thus
X (t) is orthogonal to the geodesic v (t) at any point of the geodesic. This
shows that only those F -Jacobi …elds can give focal points, i.e. can vanish at
some parameter t0 6= 0, which are orthogonal to the geodesic.

Remark 5 If X (t) is an F -Jacobi …eld along v for which X (t0 ) = 0; t0 6= 0,

then g (X 0 (t) ; v0 (t)) = 0 for every t 2 I.

Proof. As it is shown above this remark, the assumptions on X (t) yield

that g (X (t) ; v0 (t)) = 0 for every t 2 I. Then by a di¤erentiation
d 0
0= g (X (t) ; v (t)) = g (X 0 (t) ; 0
v (t)) :
dt

Now the method used in [H] for the construction of the above mentioned
example is reproduced.

De…nition 6 Let be a non-degenerate symmetric bilinear form on Rn and

Q : I Rn ! Rn ; t 2 I R a 1-parameter family of endomorphisms such that
def
Qt (:) = Q (t; :) is -symmetric, i.e.

(Qt (v) ; w) = (v; Qt (w)) , for every v; w 2 Rn .

Moreover let P Rn be a subspace such that the restriction of to P is non-

degenerate, and let S : P ! P be an -symmetric endomorphism. Furthermore
let P ? denote the orthogonal complement of P with respect to and let X : I !
Rn be a smooth map. Then the di¤ erential equation

d2
X (t) + Qt X (t) = 0; t 2 I (5)
dt2
withe initial conditions
0
X (0) 2 P; X (0) + S X (0) 2 P ? (6)

is called the Morse-Sturm system corresponding to the quadruple ( ; Q; P; S).

Since the solutions of (5) build a 2n-dimensional vector space, the solutions
with initial conditions (6) build an n-dimensional vector space.
It will be shown that to a Jacobi equation (1) along a geodesic a Morse-Sturm
system corresponds in a natural way.
Let (M; g) be a semi-Riemannian manifold, F M a semi-Riemannian sub-
manifold and v : I ! M; v 2 Nx F; x 2 F a geodesic starting orthogonally to
F at x, where I R is a segment. Now let e1 ; : : : ; en 2 Tx M be an orthonormal
base and extend these vectors to parallel vector …elds E1 ; : : : ; En : I ! M along
v . Then any vector …eld X (t) along v (t) can be given in the following form
X
X (t) = fi (t) Ei (t) ;
i

4
with suitable functions fi : I ! R; i = 1; : : : ; n. There is an isomorphism
between the vector space of smooth curves of Rn and the vector space of smooth
vector …elds along the geodesic v (t). This isomorphism will be given by the
1-parameter family At : Rn ! T v (t) M of isomorphisms de…ned as follows.
Let e1 ; : : : ; en be a base of Rn , and for a t 2 I let
n
X n
X
At : wi ei 7! wi Ei (t) :
i=1 i=1
P
So a curve I 3 t 7! hi (t) ei 2 Rn can be identi…ed with the vector …eld
i
!
X X
At hi (t) ei = hi (t) Ei (t)
i i

de…ned along the geodesic v.

Since At is an isomorphism its inverse
X X
1
At : T v (t) M 3 wi Ei (t) 7! wi ei 2 Rn
i i

yields for each vector …eld de…ned along v a curve in Rn :

!
X X
1
At hi (t) Ei (t) = hi (t) ei :
i i

By the above de…nition for a vector w 2 Rn , as a constant curve in Rn , t 7!

At (w) is a vector …eld along v . Now let the quadruple ( ; Q; P; S) on Rn for a
Morse-Sturm system de…ned by the following way:
def
(a) (v; w) = g (A0 (v) ; A0 (w)) ; v; w 2 Rn
def
(b) Qt (v) = At 1 fR (At (v) ; 0
v (t)) 0
v (t)g ; v 2 Rn ; t 2 I
def
(c) P = A0 1 T v (0)
F ;
def
(d) S (v) = A0 1 0
v (0)
(A0 (v)) :
Roughly speaking the above de…ned quantities take the images of the vectors
by At , then calculate the values of the metric tensor, the curvature tensor and
the Weingarten endomorphism for these images, and then they "pull back" the
result by At 1 .
In the quadruple ( ; Q; P; S) the elements Qt and S must be -symmetric.
As Ei ; i = 1; : : : ; n are parallel vector …elds,
g (Ei (t) ; Ej (t)) = (ei ; ej ) = for every t 2 I; ij
P
where
P ij is the Kronecker delta symbol. So if X (t) = i fi (t) Ei (t) ; Y (t) =
j hj (t) Ej (t) then:
0 1
X X
g (X (t) ; Y (t)) = g @ fi (t) Ei (t) ; hj (t) Ej (t)A =
i j

5
 X X
fi (t) hi (t) g (Ei (t) ; Ei (t)) = fi (t) hi (t) g (Ei (0) ; Ei (0)) =
i i

0 1
X X
g@ fi (t) Ei (0) ; hj (t) Ej (0)A =
i j
0 ! 0 11
X X
g @ A0 fi (t) ei ; A0 @ hj (t) ej AA =
i j

g A0 At 1 (X (t)) ; A0 At 1 (Y (t)) : (7)

Moreover since the curvature tensor is g symmetric, the above de…ned Qt is
symmetric, because
(a) (b)
(Qt (v) ; w) = g (A0 (Qt (v)) ; A0 (w)) =
eq. (7)
g A0 At 1 fR (At (v) ; 0
v (t)) 0
v (t)g ; A0 At 1 (At (w)) =
0 0
g (R (At (v) ; v (t)) v (t) ; At (w)) ; (8)
thus
eq (8) 0 0
(Qt (v) ; w) = g (R (At (v) ; v (t)) v (t) ; At (w)) =
0 0 eq. (8)
g (At (v) ; R (At (w) ; v (t)) v (t)) = (v; Qt (w))

Similarly by the g-symmetry of the Weingarten endomorphism 0

v (0)
the -
symmetry of S follows, because
(a) (d)
(S (v) ; w) = g A0 A0 1 0
v (0)
(A0 (v)) ; A0 (w) =
g 0
v (0)
(A0 (v)) ; A0 (w) = g A0 (v) ; 0
v (0)
(A0 (w)) = (v; S (w)) :

Thus the above de…ned ( ; Q; P; S) yields a Morse-Sturm system.

Proposition 7 Let (M; g) be a semi-Riemannian manifold, F M a semi-

Riemannian submanifold and v : I ! M; v 2 Nx F; x 2 F a geodesic starting
orthogonally to F at x. Let ( ; Q; P; S) and At de…ned as above, then At 1
yields an isomorphism, between the vector space of the F -Jacobi …elds along v
and the vector …eld of the solutions of the Morse-Sturm system corresponding to
( ; Q; P; S). Moreover for any F -Jacobi …eld X, the equality X (t0 ) = 0 holds if
and only if At01 (X (t0 )) = 0.
P
Proof. The identity X 0 (t) = i f 0 (t) Ei (t) holds, because Ei0 (t) 0 for
every i. Thus @t2 At 1 (X (t)) = At 1 (X 00 (t)). Applying At 1 to the condition
(1) and using X (t) = At At 1 (X (t)) the equality
1
At R At At 1 (X (t)) ; 0
(t) 0
(t) + At 1 (X 00 (t)) = 0

6
 def
follows. Let X (t) = At 1 (X (t)) then by (a) condition (5) holds for X (t).
Moreover applying A0 1 to the initial conditions (2), and using that X (0) =
A0 A0 1 (X (0)) ; then as above by (c) and (d) condition (6) holds. So if
def
X (t) is an F -Jacobi …eld, then X (t) = At 1 (X (t)) is a solution of the Morse-
Sturm system corresponding to ( ; Q; P; S). From the form of At it is clear that
X (t0 ) = 0 if and only if At01 (X (t0 )) = 0, moreover At 1 (X (t)) = At 1 (Y (t))
if and only if X (t) = Y (t) thus At is injective. Since the solutions form an
n-dimensional vector space in both cases, At 1 is an isomorphism between the
solutions.
Actually, the converse of the above proposition is helpful in creating the
examples mentioned.
Let be a semi-euclidean inner product on Rn and consider the product
manifold Rn R with the semi-euclidean metric + induced by and the
canonical euclidean metric of R. Let e1 ; : : : ; en+1 be an orthonormal base
such that e1 ; : : : ; en is a base in Rn f0g and en+1 in f0g R. Consider
the semi-Riemannian metric ge on Rn R induced by + and by the inverse
of canonical isomorphism Bx : Rn+1 ! Tx Rn+1 ; x 2 Rn+1 , i.e. ge (v; w) =
def
+ Bx 1 (v) ; Bx 1 (w) for any v; w 2 Tx Rn+1 and x 2 Rn+1 . Put (t) =
t en+1 ; t 2 R and let ! : Rn+1 ! R be a smooth function such that ! ( (t)) 0
and @i ! ( (t)) 0 holds, where @i is the partial derivative in the direction of
ei . Put
def
g = e! ge;
which de…nes a new semi-Riemannian metric on Rn+1 . Then by simple but long
calculations, see e.g. the appendix of [Sz1], the following can be proved:

1. the line (t) ; t 2 R is a geodesic in Rn+1 ; g ;

2. the constant vector …elds Ei : t 7! B (t) (ei ) ; i = 1; : : : ; n + 1 along
(t) ; t 2 R will be parallel vector …elds in Rn+1 ; g along (t) ; t 2 R.

PnNow for every vector …eld X (t) along (t) which is orthogonal to , X (t) =
i=1 xi (t) Ei (t) holds with some functions xi : R ! R. Moreover let X (t) =
T
(x1 (t) ; : : : ; xn (t)) be the column vector form of X (t). Note that here i goes
from 1 to n and not to n + 1; as it was stated preceding de…nition 6 only
the Jacobi …elds which are orthogonal to may give focal or conjugate points.
For such orthogonal vector …elds the Jacobi equation (1) along will have the
following form, see e.g. the appendix of [Sz1]:
Pn
A, i=1 xi (t) @i @n+1 ! (( (t))) = 0;
B, X 00 (t) 21 ( (t)) X (t) = 0; where as a matrix, (x) 2 Rn n ; and the
elements of the matrix are ij (x) = @i @j ! (x) (ei ; ei ) ; where i means the rows
T
and j the columns, moreover X (t) = (x1 (t) ; : : : ; xn (t)) :

7
 Now
h consider
i the Morse-Sturm system corresponding to ( ; Q; P; S) and let
ij
Qt = Qt be the matrix representation of Qt : Put
ij=1;:::;n

n
X
! (x) = 2 xi xj Qij
xn+1 (ei ; ei ) , where x = (x1 ; : : : ; xn+1 ) 2 R
n+1
:
i;j=1

In this case A will be true since @i @n+1 ! (( (t))) 0 and B will be equal to
equation (5). Now according to lemma 2.3.2 of [M-P-T] there is a submanifold
F M such that (0) 2 F; T (0) F = P and the Weingarten endomorphism
0 (0) corresponding to F is S.

Proposition 8 Consider the Morse-Sturm system corresponding to a quadruple

( ; Q; P; S), then for the above constructed manifold Rn+1 ; g and submanifold
F there is an isomorphism A et : X (t) ! A et (X (t)) between the vector space of
the solutions of the Morse-Sturm system corresponding to ( ; Q; P; S) and the
vector space of the F -Jacobi …elds along which are orthogonal to . Moreover
Aet (X (t0 )) = 0 if and only if X (t0 ) = 0:

Proof. Let X (t) = (x1 (t) ; : : : ; xn (t)) be a solution of the Morse-Sturm sys-
et (X (t)) def Pn
tem, and put A = i=1 xi (t) Ei (t), where Ei (t) are the constant vec-
tor …elds de…ned above in point 2. Then as it was stated preceding this propo-
sition, B is equal to equation (5) and A is automatically true, thus A et (X (t))
is an F -Jacobi …eld along . Moreover from the de…nition of A et it is clear that
Aet (X (t0 )) = 0 if and only if X (t0 ) = 0, i.e. A ft is injective. Since the solu-
tions of the Morse-Sturm system and the solutions of the F -Jacobi …elds along
which are orthogonal to form both an n-dimensional vector …eld, the map
Aet must be an isomorphism, between the vector spaces of the solutions.
The above proposition is important, because it is much more easier to cal-
culate with a Morse-Sturm system than to calculate with the objects of a semi-
Riemannian manifold. In the above construction is spacelike. If at the begin-
ning of the construction where + was de…ned on Rn R, the negative of the
canonical metric is taken on the R part, then will be a timelike geodesic.
It is important to note that if Qt is symmetric, i.e. QTt = Qt for every t;
then the canonical euclidean metric < :; : > can be taken on Rn instead of ,
because the symmetric Qt will be also symmetric with respect to < :; : >.
Since the solutions are independent of , the system (< :; : >; Q; P; S) can be
also considered in proposition 8, but then the manifold Rn+1 ; g corresponding
to the Jacobi di¤erential system will be Riemannian. Thus the solution of
equation (5) and (6) can be identi…ed with the F -Jacobi …elds along a geodesic
in a Riemannian manifold.
Let us present the interesting example given by Helfer.

Example 9 There is a Morse-Sturm system ( ; Q; P; S) on R2 such that for the

corresponding semi-Riemannian manifold and geodesic , given by proposition
8, the points (t), t 1 are conjugate to (0).

8
 Proof. See [H]. The example is based on the following observation. Consider
d2
the Jacobi di¤erential equation dz 2 X (t) + Qt X (t) = 0 with initial condition

X (0) = 0: Let X1 (t) ; X2 (t) be a base in the vector space of the solutions,
i.e. every solution can be expressed as a linear combination X (t) = X1 (t) +
X2 (t). If on the segment (0; t0 ) there are no conjugate points, then the vectors
X1 (t) and X2 (t) are independent for every t 2 (0; t0 ). Thus if X (t) is the 2 2
matrix which has the columns X1 (t) ; X2 (t), then

X00 (t) + Qt X (t) = 0 (9)

yields
Qt = X00 (t) X 1
(t) ; t 2 (0; t0 ) : (10)
So if …rst the solutions are given, i.e. X (t), then the above equation can be
used to …nd the corresponding Qt if it exists. Let

a (t) b (t)
X (t) = , then
c (t) d (t)

Qt = X00 (t) X 1
(t) =
00 00 1
a (t) b (t) d (t) b (t)
:
c00 (t) d00 (t) c (t) a (t) a (t) d (t) b (t) c (t)

I.e.

Qt =
c (t) b00 (t) d (t) a00 (t) a (t) b00 (t) + b (t) a00 (t) 1
:
c (t) d00 (t) d (t) c00 (t) a (t) d00 (t) + b (t) c00 (t) a (t) d (t) b (t) c (t)

0 1
Put [ ] = as the matrix of the inner product on R2 ; which is a non-
1 0
degenerate inde…nite metric. I.e. R2 ; is Lorentzian and the base of is
lightlike. Since the matrix Qt must be symmetric, the following matrix is
symmetric

[ ] Qt =
d (t) c00 (t) c (t) d00 (t) a (t) d00 (t) b (t) c00 (t) 1
;
d (t) a00 (t) c (t) b00 (t) a (t) b00 (t) b (t) a00 (t) a (t) d (t) b (t) c (t)

i.e.
a (t) d00 (t) b (t) c00 (t) = d (t) a00 (t) c (t) b00 (t) (11)
or equivalently

a00 (t) d (t) a (t) d00 (t) = b00 (t) c (t) b (t) c00 (t)

9
By integration a0 (t) d (t) a (t) d0 (t) = b0 (t) c (t) b (t) c0 (t) + const: Since at
t = 0 the solutions, the Jacobi …elds, vanish, the above equation is reduced to
0 = 0 + const, i.e. const = 0. So if a (t) ; b (t) ; c (t) is set then
Z t
d (t) = a (t) a 2 (t) (b (s) c0 (s) b0 (s) c (s)) ds + 1 (12)
0

will satisfy the di¤erential equation (11). Now let

a (t) = t on 0; 41 , a (t) = 1 on 1
2; 1 and interpolate smoothly and
monotonically on 14 ; 12 ;
b (t) = 0 on 0; 12 , b (t) = t 1 on 3
4; 1 and on 1 1
4; 2 interpolate
smoothly by a negative function;
c (t) = 0 on 0; 34 [ [1; 1) and let c (t) have a smooth positive bump of
area 21 on 34 ; 1 .
Then by equality (12):
d (t) = a (t) on 0; 34 ;
Since b (t) = 1 on 12 ; 1 and c (t) = 0 on 0; 43 ;the following equality
holds: R1 R1
d (1) 1 = 3 (b (s) c0 (s) b0 (s) c (s)) ds = b (t) c (t) j13 2 3 b0 (s) c (s) ds =
4 4 4
1
0 2 2 = 1, i.e. d (1) = 0;
Moreover d (t) = 0 on [1; 1).
Thus the above properties of the functions a (t) ; b (t) ; c (t) ; d (t) yield that
1 t 1
X (t) = ; for all t 1;
0 0
thus the two base solutions, the columns, are not linearly independent, thus
every (t) ; t 1 is a conjugate point. Now the corresponding Qt must be
constructed. Since a (t) d (t) b (t) c (t) > 0 on t 2 (0; 1) ; the matrix X (t) is
invertible, thus Qt exists on (0; 1) by equation (10). Since Qt is the zero matrix
on 0; 41 , Qt can be extended to the zero matrix on ( 1; 0] smoothly. On
3
4 ; 1 the following holds:

Qt =
0 0 1
:
c (t) d00 (t) d (t) c00 (t) a (t) d00 (t) + b (t) c00 (t) a (t) d (t) b (t) c (t)
Since by equation (11) on 43 ; 1 the equalities a (t) d00 (t) b (t) c00 (t) = 0 and
d00 (t) = b (t) c00 (t) hold, moreover c (t) d00 (t) d (t) c00 (t) = c (t) b (t) c00 (t)
d (t) c00 (t) = (c (t) b (t) a (t) d (t)) c00 (t) is true:
0 0
Qt = .
:c00 (t) 0

10
But this can be extended smoothly to [1; 1) as the zero matrix. Since equation
(9) holds on (0; 1) ; by the construction of Qt ; and also on ( 1:0][[1; 1), which
can be easily veri…ed, the desired example is obtained.

3 Conditions for the curvature tensor

As the above example shows the set of focal points along a geodesic can include
a non-empty open set in fact, a segment. Of course some strict conditions can
be given which imply, that the conjugate points are isolated along a geodesic. If
(M; g) and F are analytic then the focal points are isolated along any geodesic
starting orthogonally to F , see e.g [M-P-T] or [Sz1]. For example such condition
is given in remark 11, where the curvature tensor splits along the geodesic with
respect to a positive and negative distribution, see the following de…nition.

De…nition 10 Let (M; g) be a semi-Riemannian manifold, : I ! M a geo-

desic and E1 (t) ; : : : ; En (t) parallel vector …elds along , for which E1 (t) ; : : : ; En (t)
is an orthonormal base of T (t) M , and g (Ei (t) ; Ei (t)) = 1 for i = 1; : : : ; k and
g (Ei (t) ; Ei (t)) = 1 for i = k + 1; : : : ; n. Assume that
0 0
g (R (Ei (t) ; (t)) (t) ; Ej (t)) = 0

for every t 2 [0; ] if i 2 f1; : : : ; kg ; j 2 fk + 1; : : : ; ng. Then we say that

the curvature tensor splits along with respect to the positive and
negative distributions spanned by E1 ; : : : ; Ek and Ek+1 ; : : : ; En .

Remark 11 Let (M; g), : I ! M and E1 (t) ; : : : ; En (t) as in the above de…n-
ition. Assume that the curvature tensor splits along with respect to the positive
and negative distributions spanned by E1 ; : : : ; Ek and Ek+1 ; : : : ; En . Then the
conjugate points are isolated along .
00
Proof. Consider the Morse-Sturm equation X (t) + Qt X (t) = 0 corre-
sponding to the Jacobi equation with respect to the base …elds Ei (t), as given
by proposition 7. By the assumption of the remark and equality (a) above, the
matrix of Qt and in the base E1 (t) ; : : : ; Ek (t) ; Ek+1 (t) ; : : : ; En (t) has the
following form:
Q+t 0 Id 0
Qt = ; = :
0 Qt 0 Id
00
This gives that X (t) = X + (t) ; X (t) is a Jacobi …eld if and only if X + (t) +
00
Q+ +
t X + (t) = 0 and X (t) + Qt X (t) = 0. Since Qt ; Qt are symmetric,
by our remark above example 9, we have that X + (t) (or X (t)) corresponds
to a Jacobi …eld along a geodesic + (t) (or (t)) in a Riemannian manifold
(M+ ; g+ ) (or (M ; g )). Thus the set of conjugate points of is the union of
the sets of conjugate points along + and . Since in the Riemannian case the
conjugate points are isolated, thus the conjugate points along + and are
isolated. So the conjugate points along are also isolated.

11
 Examples to the above remark can be easily given by proposition 8. Now a
not so strict condition is given which yields in the Lorentzian case, that there
is no open set, i.e. a segment, in the set of focal points along a geodesic. First
some earlier results will be recalled.
Let F be a semi-Riemannian submanifold and v : I ! M; v 2 Nx F; x 2 F
a geodesic starting orthogonally to F at x for which v0 (0) = v. To the geodesic
v (t), a curve ev (t) in a Lagrangian Grassmannian can be associated such
that the following holds. There is an algebraic subvariety in , which is called
the Maslov cycle, such that ev (t0 ) 2 if and only if v (t0 ) is a focal point on
v . If ev intersects the Maslov cycle transversally at ev (t0 ) then v (t0 ) is an
isolated focal point. The above result was proved in [M-P-T]. This is also proved
in [Sz] for conjugate points, and the extension of this latter result to focal points
yields another proof for the above fact. In the above mentioned papers there is
condition, which is equivalent to the assumption that ev intersects the Maslov
cycle transversally. For this condition the following de…nition is needed.

De…nition 12 Let J (F; v) denote the vector space of the F -Jacobi …elds along
v (t). Put

def
VF;v (t0 ) = fY (t0 ) j Y 2 J (F; v )g T v (t0 )
M
def
DF;v (t0 ) = fY 0 (t0 ) j Y 2 J (F; v) ; Y (t0 ) = 0g T v (t0 )
M;

where Y 0 (t0 ) = r 0
v (t0 )
Y (t0 ) with respect to the Levi-Cività connection.

Remark 13 The subspaces VF;v (t0 ) and DF;v (t0 ) are orthogonal with respect
to g, moreover dim VF;v (t0 ) + dim DF;v (t0 ) = dim M

Proof. See [Sz], where the proof is the following. If in equation (3) X; Y
are F -Jacobi …elds then g (X 0 (t) ; Y (t)) g (X (t) ; Y 0 (t)) 0, because for t = 0
the initial conditions of equation (2) with the g symmetry of the Weingarten
endomorphism yield that the constant in equation (3) is 0. As X (t0 ) = 0 we
have that g (X 0 (t0 ) ; Y (t0 )) = 0. Since X 0 (t0 ) 2 DF;v (t0 ) ; Y (t0 ) 2 VF;v (t0 )
can be arbitrary elements of the subspaces we have proved the remark.
Now some equivalent conditions are given in the following lemma.

Lemma 14 The following are equivalent:

1. gjDF;v (t0 ) is non-degenerate on DF;v (t0 );

2. gjVF;v (t0 ) is non-degenerate on VF;v (t0 );
3. DF;v (t0 ) \ VF;v (t0 ) = f0g;

4. DF;v(t0 ) VF;v(t0 ) = T v (t0 )

M;
5. the associated curve ev intersects the Maslov cycle transversally at
ev (t0 ) and (t0 ) is an isolated focal point.

12
 Proof. See [Sz] or [M-P-T].
In the next proposition a simple condition is given which yields that there
is no open set in the set of focal points along a geodesic.

Proposition 15 Let (M; g) be a Lorentzian manifold, F M a semi-Riemann-

ian submanifold for which dim F < dim M and v : I ! M; v 2 Nx F; x 2 F a
geodesic starting orthogonally to F at x. If for every value t 2 I
0 0
g (R (w; v (t)) v (t) ; w) < 0

holds in case of each lightlike vector w 2 T v (t) M such that g (w; v0 (t)) = 0,
then there is no open segment contained in the set of focal points along v .

Proof. Assume that v (t0 ) is a focal point. If DF;v (t0 ) \ VF;v (t0 ) = f0g
holds, then by the above lemma the associated curve ev intersects the Maslov
cycle transversally at ev (t0 ) and the focal point v (t0 ) is isolated. Thus the
proposition holds at v (t0 ).
So DF;v (t0 ) \ VF;v (t0 ) 6= f0g can be assumed. As DF;v (t0 ) ; VF;v (t0 ) are
orthogonal by remark 13, the subspace DF;v (t0 ) \ VF;v (t0 ) is a 1-dimensional
lightlike subspace. Let X (t) be an F -Jacobi …eld for which X (t0 ) = 0; X 0 (t0 ) 2
DF;v (t0 ) \ VF;v (t0 ). Then X 0 (t0 ) is a lightlike vector which is orthogonal to
0
v (t0 ) by remark 5. Now the derivatives of the energy function

def
f (t) = g (X (t) ; X (t)) ;

will be calculated at t = t0 .

f (t0 ) = g (X (t0 ) ; X (t0 )) = 0

f 0 (t0 ) = 2g (X 0 (t0 ) ; X (t0 )) = 0
f 00 (t0 ) = 2g (X 00 (t0 ) ; X (t0 )) + 2g (X 0 (t0 ) ; X 0 (t0 )) = 0:

For the third derivative f 000 (t0 )the following is true:

d d
f 000 (t0 ) = ff 00 (t)g jt=t0 = f2g (X 00 (t) ; X (t)) + 2g (X 0 (t) ; X 0 (t))g jt=t0 =
dt dt
d 0 0
f2g (R (X (t) ; v (t)) v; X (t)) + 2g (X 0 (t) ; X 0 (t))g jt=t0 :
dt
Since
d
f2g (R (X (t) ; v0 (t)) v0 ; X (t))g =
dt
2g r v0 (t) R (X (t) ; v0 (t)) v0 (t) + R (X 0 (t) ; v0 (t)) 0
v (t) ; X (t) +
0 0
2g (R (X (t) ; v (t)) v (t) ; X 0 (t)) ;
d 0 0
at t = t0 the equality dt (2g (R (X (t) ; v (t)) v; X (t))) jt=t0 = 0 holds. The
equality
d
(2g (X 0 (t) ; X 0 (t))) = 4g (X 00 (t) ; X 0 (t)) = 4g (R (X (t) ; 0
v (t)) 0
v (t) ; X 0 (t))
dt

13
 d
yields at t = t0 that dt (2g (X 0 (t) ; X 0 (t))) jt=t0 = 0. Thus

f 000 (t0 ) = 0:

Finally the …rst non-vanishing derivative f 0000 (t0 ) will be calculated. Note that

X 00 (t0 ) = R (X (t0 ) ; 0
v (t0 )) 0
v (t0 ) = 0 and
000 0 0
X (t) = r 0
v (t)
(R (X (t) ; v (t)) v (t)) =
0 0
r 0
v (t)
R (X (t) ; v (t)) v (t) + R (X 0 (t) ; 0
v (t)) 0
v (t)

as above. So

f 0000 (t0 ) = 2g (X 0000 (t0 ) ; X (t0 )) + 8g (X 000 (t0 ) ; X 0 (t0 )) +

6g (X 00 (t0 ) ; X 00 (t0 )) = 0+
0 0
8g r 0
v (t0 )
R (X (t0 ) ; v (t0 )) v (t0 ) + R (X 0 (t0 ) ; 0
v (t0 )) 0
v (t0 ) ; X 0 (t0 ) + 0:

As X 0 (t0 ) is a lightlike vector which is orthogonal to 0

v (t0 ), by the assumption
of the proposition:

f 0000 (t0 ) = g (R (X 0 (t0 ) ; 0

v (t0 )) 0
v (t0 ) ; X 0 (t0 )) < 0:

Since
f (t0 ) = f 0 (t0 ) = f 00 (t0 ) = f 000 (t0 ) = 0; f 0000 (t0 ) < 0
the function f has a strict local maximum at t = t0 . As f (t0 ) = 0 there is an
" > 0 such that f (t) = g (X (t) ; X (t)) < 0 if t 2 (t0 "; t0 + ") ; t 6= t0 . I.e.
the F -Jacobi …eld X (t) is timelike before and after t0 if t is in a suitably small
neighbourhood of t0 .
Now by summing up the above results the following is obtained:
If v (t0 ) is a focal point then either

DF;v (t0 ) \ VF;v (t0 ) = f0g thus v (t0 ) is an isolated focal point, or;
there is an F -Jacobi …eld X (t) which has the following properties:

1. X (t0 ) = 0;
2. X (t) is timelike for every t 2 (t0 "; t0 + ") if " > 0 is small enough,
except at t = t0 ;
3. X 0 (t0 ) 2 DF;v (t0 ) \ VF;v (t0 ) and it is a lightlike vector.

Assume on the contrary that there is an open segment U in the set of focal
points along the geodesic v (t). Let v (t0 ) 2 U . Then by the above observation
DF;v (t0 ) \ VF;v (t0 ) 6= f0g holds and there is an F -Jacobi …eld X1 (t) with the
above properties 1-3. Let t1 be so close to t0 that v (t1 ) 2 U and X1 (t1 ) is
timelike. Since X1 (t1 ) 2 VF;v (t1 ) is timelike, DF;v (t1 ) \ VF;v (t1 ) = f0g must
hold. Thus by lemma 14 the associated curve ev (t) intersects the Maslov cycle
either orthogonal at t = t1 or it does not intersect the Maslov cycle. I.e. either

14
 v (t1 ) is an isolated focal point, or it is not a focal point. But this contradicts
the fact that v (t1 ) 2 U .
As the following trivial example shows the condition of the above proposition
can be satis…ed.

Example 16 The elements of a quadruple ( ; Q; P; S) on R2 will be de…ned as

follows.

def 1 0
= ;
0 1
def a (t) b (t)
Qt = , which is symmetric with any
b (t) c (t)
smooth functions a (t) ; b (t) ; c (t) ;
def
P = f0g 2 R2 ;
S:P ! P is the trivial map.

Now from the construction in proposition 8, the curvature tensor of the semi-
Riemannian manifold associated to the Morse-Sturm system corresponding to
( ; Q; P; S) is the following

0 0 a (t) b (t)
R (v; (t)) (t) =
b (t) c (t)

with respect to the orthonormal base …elds E1 (t) ; E2 (t) along . Recall that
g (E1 (t) ; E1 (t)) = 1 and g (E2 (t) ; E2 (t)) = 1, by . Thus for the lightlike
def def
vectors V1 (t) = E1 (t) + E2 (t) and V2 = E1 (t) E2 (t) the equality
0 0
g (R (V1 (t) ; (t)) (t) ; V1 (t)) = a (t) + 2b (t) c (t) ;
0 0
g (R (V2 (t) ; (t)) (t) ; V2 (t)) = a (t) 2b (t) c (t) ;

holds. Thus if a (t) + 2b (t) c (t) < 0 and a (t) 2b (t) c (t) < 0 then the
assumptions of the above proposition hold for the semi-Riemannian manifold
associated to the Morse-Sturm system ( ; Q; P; S) :

It is a question whether the above condition implies that the focal points
are isolated or does not; moreover, whether this condition works in the general
semi-Riemannian case or does not. It is not easy to understand the structure
of the set of focal points along a geodesic in the semi-Riemannian case. Even
in case of a stationary Lorentzian manifold, i.e. a Lorentzian manifold with
a nowhere timelike Killing …eld, which is a strong condition, it is not known
whether the set of conjugate points is discrete or not along a geodesic. However
there are some partial results (see e.g. F. Giannoni, A. Masiello, P. Piccione
[?]).
For an easy observation below, the following lemma is needed.

15
Lemma 17 Let (M; g) be a semi-Riemannian manifold, : I ! M a geodesic
segment and X : I ! M a smooth vector …eld along such that X (t0 ) =
0; X 0 (t0 ) 6= 0 and X (t) 6= 0, if t 6= t0 . Then

lim R X (t) = R X 0 (t0 ) ;

t!t0

def def
i.e. ' (t) = R X (t) for t 6= t0 and ' (t0 ) = R X 0 (t0 ) de…nes a continuous
line bundle ' (t) along the geodesic segment (t).

Proof. Let Ei (t), i = 1; : : : ; n be independentPparallel vector …elds along

(t) such that E1 (t0 ) = X 0 (t0 ). Then X (t) = i fi (t) Ei (t)Pwith suitable
smooth functions fi : I ! R; i = 1; : : : ; n. Since X 0 (t) = 0
i fi (t) Ei (t),
the properties of the parallel vector …elds yield that f1 (t0 ) = = fn (t0 ) =
0; f20 (t0 ) = = fn (t0 ) = 0; f10 (t0 ) = 1. The following equality is equivalent
to our lemma
fi (t) 1 if i = 1
lim pP 2 = ;
t!t0 0 if i 6= 1
i fi (t)

thus it is enough to show that

f 2 (t) 1 if i = 1
lim P i 2 = :
t!t0
i fi (t)
0 if i =
6 1

By the L’Hospital law

d
f 1 (t) fi2 (t) 2fi (t) fi0 (t)
lim P i 2 = lim d
dt
P 2
= lim P 0 :
i fi (t) dt ( i fi (t)) i 2fi (t) fi (t)
t!t0 t!t0 t!t0

By the L’Hospital law

2fi (t) fi0 (t) 2 (f 0 (t) f 0 (t) + fi (t) fi00 (t)) 1 if i = 1

lim P 0 = lim P i 0 i 0 00 = :
t!t0
i 2fi (t) fi (t)
t!t0
i 2 (fi (t) fi (t) + fi (t) fi (t))
0 if i 6= 0

Proposition 18 Let (M; g) be a semi-Riemannian manifold, F M a semi-

Riemannian submanifold and v : I ! M; v 2 Nx F; x 2 F a geodesic starting
orthogonally to F . Let k = n (g) be the index of the metric tensor. Then the
focal points of multiplicity > k are isolated along .

Proof. Let v (t0 ) be a focal point of F along v . By lemma 14 if DF;v (t0 ) \

VF;v (t0 ) = f0g then (t0 ) is isolated. So it can be assumed that DF;v (t0 ) \
VF;v (t0 ) 6= f0g. As the index n (g) = k the equality dim (DF;v (t0 ) \ VF;v (t0 ))
k holds, thus F -Jacobi …elds J1 ; : : : ; Jl ; Jl+1 ; : : : ; Jb can be chosen along v
such that the following hold:

J1 (t0 ) = = Jl (t0 ) = 0; and J10 (t0 ) ; : : : ; Jl0 (t0 ) is a base of DF;v (t0 );

16
 Jl+1 (tq ) ; : : : ; Jb (t0 ) is a base of a maximal dimensional space in VF;v (t0 )
which is complementary to DF;v (t0 ) \ VF;v (t0 ).

Proof. Thus J10 (t0 ) ; : : : ; Jl0 (t0 ) ; Jl+1 (tq ) ; : : : ; Jb (t0 ) are linearly indepen-
dent and they span the subspace span hDF;v (t0 ) ; VF;v (t0 )i, i.e. the subspace
generated by DF;v (t0 ) and VF;v (t0 ). Since dim (span hDF;v (t0 ) ; VF;v (t0 )i) =
dim DF;v (t0 )+dim VF;v (t0 ) dim (DF;v (t0 ) \ VF;v (t0 )), by remark 13 the equal-
ity b dim M k holds. It is well-known that if X (t) is a Jacobi …elds for
which X (t0 ) = 0, then X b t 6= 0 if b t 6= t0 is suitably close to t0 . Since
0 0
J1 (t0 ) ; : : : ; Jl (t0 ) ; Jl+1 (tq ) ; : : : ; Jb (t0 ) are linearly independent, using the above
lemma 17 the following is obtained. If b t 6= t0 is suitably close to t0 then the
vectors J1 b t ,Jl+1 b
t ; : : : ; Jl b t ; : : : ; Jb bt are linearly independent. I.e. for
the dimension of VF;v b t at v b t the equality dim VF;v b t b dim M k
holds. Thus in a suitably small neighbourhood of v (t0 ) along v every focal
point has multiplicity k which proves the proposition.

Remark 19 If in the above proposition n (g) = 1 i.e. (M; g) is Lorentzian,

then the focal points of multiplicity 2 are isolated along the geodesic v . More-
over if (M; g) is Riemannian, as it is well-known, the focal points are isolated
along the geodesic v .

4 Example for the instability of the focal points

De…nition 20 Let (R; gR ) ; (M; gM ) be semi-Riemannian manifolds and :
R M ! R+ a smooth function. A semi-Riemannian metric is de…ned on
def
R M by gj(r;m) (X; Y ) = 2 (r; m) gR (XR ; YR ) + gM (XM ; YM ), where
XR ; XM are the components of X 2 T(r;m) (R M ) according to the decom-
position T(r:m) (R M ) = Tr R Tm M . Then

R M = (R M; g)

is called a twisted product.

See e.g. M. Fernández-López et al. [F-G-K-Ü].

Lemma 21 Let R M be a generalized warped product and (t) a geodesic

of the semi-Riemannian submanifold R fxg ; x 2 M with the restricted metric
on it, i.e. with the metric given by 2 (r; x) gR at the point (r; x) 2 R fxg :
Then (t) fxg is a geodesic of the generalized warped product

Proof. The proof is a straightforward calculation for non-lightlike geodesics,

for lightlike geodesic a simple continuity argument yields the proof, see [Sz1].
Now an example will be given which shows that the focal points are not
necessarily stable under a variation of a geodesic.

17
Example 22 Let R3 ; gM be a Minkowski space and e1 ; e2 ; e3 an orthonormal
base, where g (e1 ; e1 ) = 1; g (e2 ; e2 ) = g (e3 ; e3 ) = 1. Let (R; g1 ) be the real
def 2 2 2 2
line with the standard metric. If ! (x; s) = x1 x2 2sx1 x2 ; where x =
def 1
x1 e1 + x2 e2 + x3 e3 and (x; s) = e 2 !(x;s) then for the generalized warped
product
R3 R
def
the following will be true. The geodesics s (t) = (0; 0; t; s) ; t 2 R 0; s 0
def
are orthogonal to the Riemannian submanifold F = f0g R, moreover on the
geodesic 0 (t) the only focal point of F is 0 (1) and on the variational geodesics
s (t) ; s > 0 there will be no geodesics.

Proof. Proposition 8 and lemma 21 shows that for every s the curve s (t)
will be a geodesic in R3 R, moreover they will be orthogonal to F . Remark
4 yields that @s s (t) js=s0 is an F -Jacobi …eld. The geodesics in R3 fs0 g
with the restricted metric are also geodesics of R3 R by lemma 21. If they
initial point is the origin, then they will be orthogonal to F in R3 R. So by
remark 4 any in…nitesimal variation X (t) of s0 (t) corresponding to a geodesics
variation in R3 fs0 g with …xed initial points, is on the one hand a Jacobi
…eld along s0 (t) in R3 fs0 g corresponding to the restricted metric such
that X (0) = 0, but on the other hand it is also an F -Jacobi …eld in R3 R
along s0 (t). Let J0 (s0 ) denote the 3-dimensional vector space of the Jacobi
…elds X (t) along s0 (t) in R3 fs0 g corresponding to the restricted metric for
which X (0) = 0. Since the F -Jacobi …elds along s0 (t) form a 4-dimensional
vector space and the non-vanishing @s s (t) js=s0 is orthogonal to every X (t) 2
J0 (s0 ) it follows that every F -Jacobi …eld can be given in the form X (t) +
c@s s (t) js=s0 ; c 2 R; X (t) 2 J0 (s0 ). The orthogonality of the non-vanishing
vector …eld @s s (t) js=s0 to the elements of J0 (s0 ) yields, that every non-trivial
F -Jacobi …elds which vanish at some s0 (t) ; t 6= 0 is in J0 (s0 ). So the conjugate
points of s0 (t) in the manifold R3 fs0 g with the restricted metric are the
focal points of F along s0 (t) in R3 R.
By proposition 8 the conjugate points and the Jacobi …elds can be found by the
di¤erential equation
2
s0
X 00 (t) + 2 X (t) = 0: (13)
s0

For the parameter s0 = 0 a base of the solutions with initial condition X (0) = 0
T T
are X1 (t) = 1 sin (t ) ; 0 ; X2 (t) = 0; 1 sin (t ) . This shows, that 0 (1)
is a conjugate point, thus also focal point. For a parameter s 6= 0 it is easier
to solve the equation in an other base, i.e. let T be an invertible 2 2 matrix
2
def def s
and put Y (t) = T 1 X (t) ; thus X (t) = T Y (t). Put Rs = 2 .
s
Then T Y 00 (t) + Rs T Y (t) = 0. Multiplying by T 1 from left yields Y 00 (t) +

18
 1 1
1 def 2 2
T Rs T Y (t) = 0: Now if T = 1 1 then
2i 2i

2 1 1
1 i s
Y 00 (t) + 2
2
1
2
1 Y (t) = 0; which is
1 i s 2i 2i
2
is 0
Y 00 (t) + 2 Y (t) = 0:
0 + is

Thus a base of the solutions with initial condition Y (0) = 0 are Y1 (t) =
p T p T
p 1 sin t 2 si ; 0 ; Y2 (0) = 0; p 21+si sin t 2 + si . From their
2 si

form it can be seen that Y1 (t) and Y2 (t) are linearly independent for every
t 6= 0; s 6= 0. From this it is easy to see that the solution of (13) are non zero
for every t 6= 0; s 6= 0, i.e. there are no conjugate points, and so no focal points,
on s (t) if s 6= 0.

References
[B] Besse A., Manifolds all of whose Geodesics are Closed. Springer-
Verlag (1978)

[B-C] Bishop R. L., Crittenden R. J., Geometry of Manifolds. Academic

Press, New York, (1964)
[B-E-Ea] Beem, Ehrlich, Easly, Global Lorentzian Geometry, Monographs
and Textbooks in Poor and Applied Mathematics 202 (1996)
[F-G-K-Ü] M. Fernández-López, E. Garcia-Rió, D.N. Kupeli, B. Ünal, A
curvature condition for a twisted product to be a warped prod-
uct, Manuscripta Math. 106 (2001) 213-217
[H] Helfer A., Conjugate points on space-like geodesics or pseudo-
self-adjoint Morse-Sturm-Liouville systems. Paci…c J. Math. 164
(1994) pp. 321-350.

[M-P-T] Mercuri F., Piccione P., Tausk D.V., Stability of the conjugate
index, degenerate conjugate points and the Maslov index in semi-
Riemannian geometry, Paci…c J. Math. 206 (2002), 375–400.
[P-T] Piccione P., Tausk D.V, On the distribution of conjugate points
along semi-Riemannian geodesics, Comm. Anal. Geom., 11
(2003) pp. 33-48.
[Sz] Szeghy, D., On the conjugate locus of pseudo-Riemannian man-
ifolds, Indagationes Mathematicae, 19 (2009) pp. 465-480.
[Sz1] Szeghy, D., Exponential mapping of Lorentzian manifolds, Ph.D.
thesis (in Hungarian), Eötvös University, Budapest

19
[W] Warner F., The conjugate locus of a Riemannian manifold, Amer.
J. Math. 87 (1965) pp. 575-604.
[Wh] Whitehead J. H. C., On covering of a complete space by the
geodesics through a point. Ann. of Math. 36 (1935) pp. 679-704.

20