, ,

1.1 Affine manifolds

We say that

A : = (M, \nabla)

is an affine manifold if

\nabla

is a torsion free connection on the tangent bundle

T M

of a smooth

m

-dimensional manifold

M

. Let

ℛ (X, Y) : = \nabla_{X} \nabla_{Y} - \nabla_{Y} \nabla_{X} - \nabla_{[X, Y]}

be the associated curvature operator. We say that

A

is locally affine homogeneous if given any points

P, Q \in M

, there is a diffeomorphism

Φ_{P, Q}

from a neighborhood of

P

to a neighborhood of

Q

with

Φ_{P, Q} (P) = Q

so that

Φ_{P, Q}^{*} \nabla = \nabla

. We say that

A

is locally

k

-affine curvature homogeneous if given any points

P, Q \in M

, there is a linear isomorphism

φ_{P, Q}

from

T_{P} M

T_{Q} M

so that

φ_{P, Q}^{*} \nabla^{i} ℛ_{Q} = \nabla^{i} ℛ_{P}

for

0 \leq i \leq k

By taking

φ_{P, Q} = (Φ_{P, Q})_{*}

, it is clear that any locally affine homogeneous manifold is locally

k

-affine curvature homogeneous for all

k

. What is perhaps somewhat surprising is that given

k

, there exists a

k

-affine curvature homogeneous manifold

A_{k}

of dimension

a (k)

which is not locally affine homogeneous, see, for example, the discussion in [5, 6, 8, 9] ; one has that

a (k) \to \infty

k \to \infty

1.2 Pseudo-Riemannian manifolds

There are similar notions in the metric context. Let

ℳ : = (M, g)

be a pseudo-Riemannian manifold of signature

(p, q)

. We take

\nabla

to be the Levi-Civita connection and let

R \in \otimes^{4} T^{*} M

be the associated curvature tensor:

R (X, Y, Z, W) : = g (ℛ (X, Y) Z, W) .

We say that

ℳ

is locally homogeneous if given any points

P, Q \in M

, there is an isometry

Φ_{P, Q}

from a neighborhood of

P

to a neighborhood of

Q

with

Φ_{P, Q} (P) = Q

We say that

ℳ

k

-curvature homogeneous if given any two points

P, Q \in M

, there is an isometry

φ_{P, Q}

from

T_{P} M

T_{Q} M

so that

φ_{P, Q}^{*} \nabla^{i} R_{Q} = \nabla^{i} R_{P}

for

0 \leq i \leq k

(g, R)

determines

ℛ

, locally homogeneous (resp.

k

-curvature homogeneous) manifolds are locally affine homogeneous (resp.

k

-affine curvature homogeneous).

We refer to the discussion in [1] for a review of some of the literature in this subject.

Given

k

, there is a pseudo-Riemannian manifold

ℳ_{k}

of dimension

m (k)

which is

k

-curvature homogeneous (and hence

k

-affine curvature homogeneous) but not locally affine homogeneous (and hence not locally homogeneous) [5] ; one has that

m (k) \to \infty

k \to \infty

If, however,

m

is bounded, one has the following result due to Singer [12] in the Riemannian (

p = 0

) setting and to Podesta and Spiro [10] in the general setting:

Theorem 1.1. There exists an integer

k_{p, q}

so that if

ℳ

is a geodesically complete simply connected pseudo-Riemannian manifold of signature

(p, q)

which is

k_{p, q}

-curvature homogeneous, then

ℳ

is homogeneous.

We refer to Opozoda [9] for a similar result in the affine setting; there is an additional technical hypothesis which must be imposed.

1.3 Vanishing scalar invariants

Adopt the Einstein convention and sum over repeated indices. We can construct scalar invariants by contracting indices. For example, the scalar curvature

τ

, the norm

| ρ |^{2}

of the Ricci tensor, and the norm

| R |^{2}

of the full curvature tensor are scalar invariants defined by:

\begin{matrix} τ : = g^{i_{1} j_{1}} g^{i_{2} j_{2}} R_{i_{1} i_{2} j_{2} j_{1}}, \end{matrix}

\begin{matrix} | ρ |^{2} : = g^{i_{1} j_{1}} g^{i_{2} j_{2}} g^{i_{3} j_{3}} g^{i_{4} j_{4}} R_{i_{1} i_{2} j_{2} i_{3}} R_{j_{1} i_{4} j_{4} j_{3}}, \end{matrix}

\begin{matrix} | R |^{2} : = g^{i_{1} j_{1}} g^{i_{2} j_{2}} g^{i_{3} j_{3}} g^{i_{4} j_{4}} R_{i_{1} i_{2} i_{3} i_{4}} R_{j_{1} j_{2} j_{3} j_{4}} . \end{matrix}

By Weyl's theorem [13] , all universal polynomial scalar invariants of the covariant derivatives of the curvature tensor arise in this way; thus such invariants are called Weyl scalar invariants. We say that a pseudo-Riemannian manifold is VSI if all the scalar Weyl invariants vanish. This is not possible for non-flat manifolds in the Riemannian setting but is possible in the higher signature setting, see, for example, the discussion in [7, 11] .

1.4 Lorentzian manifolds

In this paper, we shall deal with the

3

-dimensional Lorentzian setting – i.e.

signature

(1, 2)

. We shall be discussing a number of tensors. For the sake of brevity, we shall only give the non-zero components up to the usual symmetries. Let

{x, y, \tilde{x}}

be coordinates on

R^{3}

. Let

f = f (y)

be a smooth function on

R

and let

ℳ_{f} : = (R^{3}, g_{f})

where

g_{f}

is the Lorentz metric on

R^{3}

given by:

g_{f} (\partial_{x}, \partial_{x}) = - 2 f (y) and g_{f} (\partial_{x}, \partial_{\tilde{x}}) = g_{f} (\partial_{y}, \partial_{y}) = 1 .

Let

S_{ɛ}

be defined by

f_{ɛ} (y) : = \frac{1}{2} ɛ y^{2}

for

ɛ = \pm 1

Theorem 1.2.

(1) All scalar Weyl invariants of $ℳ_{f}$ vanish.
(2) $S_{ɛ}$ is an indecomposable local symmetric space.
(3) If $f^{''} (y) \neq 0$ , then $ℳ_{f}$ is $0$ -curvature modeled on $S_{ɛ}$ for $ɛ = sign (f^{''})$ .
(4) Assume that f ′ ′ ( y ) ≠ 0 and that f ′ ′ ′ ( y ) ≠ 0 for all y ∈ R .
- (a) $ℳ_{f}$ is $1$ -affine curvature homogeneous.
- (b) $ℳ_{f}$ is $1$ -curvature homogeneous if and only if $f^{''} = a e^{b y}$ .
- (c) The following assertions are equivalent:
  - (i) $ℳ_{f}$ is locally homogeneous.
  - (ii) $ℳ_{f}$ is $k$ -curvature homogeneous for all $k$ .
  - (iii) $ℳ_{f}$ is $2$ -curvature homogeneous.
  - (iv) $ℳ_{f}$ is $2$ -affine curvature homogeneous
  - (v) $f^{'} = a e^{b y}$ .

1.5 Completeness

Let

{exp}_{P} : T_{P} M \to M

be the exponential map. We say that an affine manifold

A

is geodesically complete if all geodesics extend for infinite time.

Theorem 1.3.

(1) The manifolds $S_{\pm}$ are geodesically complete.
(2) The map ${exp}_{P}$ for $S_{+}$ is not surjective for any point $P \in R^{3}$ .
(3) The map ${exp}_{P}$ for $S_{-}$ is a global diffeomorphism from $T_{P} R^{3}$ to $R^{3}$ $\forall P \in R^{3}$ .

\nabla

is a torsion free connection, the Jacobi operator

J_{\nabla}

and Ricci form

ρ_{\nabla}

are:

J_{\nabla} (x) : ξ \to ℛ (ξ, x) x and ρ_{\nabla} (x, x) : = Tr (J_{\nabla} (x)) .

An affine manifold

A

is said to Ricci explode if there exists a geodesic

γ

A

which is defined for

t \in [0, T)

where

T < \infty

{lim}_{t \to T} | ρ (\dot{γ}, \dot{γ}) (t) | = \infty

. Such a manifold is necessarily geodesically incomplete. Furthermore, such a manifold can not be embedded as an open subset of a geodesically complete affine manifold.

Assume

f^{'}

never vanishes; by replacing

y

- y

, we may assume

f^{'} > 0

. The growth of

f^{'}

- \infty

is crucial.

Theorem 1.4. Assume that

f^{'} (y) > 0

for all

y \in R

(1) If $\exists$ $C > 0$ so $f^{'} (y) \leq C | y |$ for $y \leq - 1$ , $ℳ_{f}$ is geodesically complete.
(2) If $\exists$ $ε, δ > 0$ so $f^{'} (y) \geq ε | y |^{1 + δ}$ for $y \leq - 1$ , $ℳ_{f}$ Ricci explodes.

The remainder of this paper is devoted to the proof of these results. In Section 2 , we determine the curvature of the manifolds

ℳ_{f}

and establish Theorem 1.2 .

In Section 3 , we establish Theorem 1.3 by solving the geodesic equations on

S_{\pm}

quite explicitly. In Section 4 , we use results from the theory of ordinary differential equations to establish two slightly more general results from which Theorem 1.4 will follow.

Various properties of certain of the manifolds in this family have been studied by many authors [2, 3, 4, 7, 11] . For example the existence of

1

-curvature homogeneous

3

-dimensional Lorentzian manifolds which are not locally homogeneous follows from the discussion in [2] and the existence of

3

-dimensional VSI Lorentzian manifolds is established in [11] . In this paper, we present a unified treatment of a number of results concerning this family; we discuss some previously known results but also present some new results in affine geometry and deal with questions of geodesic completeness. We feel this family provides a rich family of examples. In particular, one has:

Example 1.5. For

1 \leq i \leq 3

, let

N_{i, \pm} : = ℳ_{f_{i, \pm}}

where

\begin{matrix} f_{1, -} (y) = - e^{- y}, & f_{2, -} (y) = - e^{- y} + y, & f_{3, -} (y) = - e^{- y} - e^{- 2 y} \end{matrix}

\begin{matrix} f_{1, +} (y) = e^{y}, & f_{2, +} (y) = e^{y} + y, & f_{3, +} (y) = e^{y} + e^{2 y} . \end{matrix}

We have

f_{i}^{'} (y) > 0

f_{i}^{''} (y) \neq 0

, and

f_{i}^{'''} \neq 0

for all

y

. We apply the results of Theorems 1.1 , 1.2 , 1.3 , and 1.4 to see:

(1) $S_{-}$ is a geodesically complete indecomposible symmetric space.
(2) $N_{1, -}$ is $0$ -curvature modeled on $S_{-}$ , locally homogeneous, and Ricci explodes.
(3) $N_{2, -}$ is $0$ -curvature modeled on $S_{-}$ , $1$ -curvature modeled on $N_{1, -}$ , not $2$ -curvature homogeneous, and Ricci explodes.
(4) $N_{3, -}$ is $0$ -curvature modeled on $S_{-}$ , not $1$ -curvature homogeneous, $1$ -affine curvature modeled on $N_{1}$ , and Ricci explodes.
(5) $S_{+}$ is a geodesically complete indecomposible symmetric space.
(6) $N_{1, +}$ is $0$ -curvature modeled on $S_{+}$ , geodesically complete, and homogeneous.
(7) $N_{2, +}$ is $0$ -curvature modeled on $S_{+}$ , $1$ -curvature modeled on $N_{1, +}$ , not $2$ -curvature homogeneous, and geodesically complete.
(8) $N_{3, +}$ is $0$ -curvature modeled on $S_{+}$ , not $1$ -curvature homogeneous, $1$ -affine curvature modeled on $N_{1, +}$ , and geodesically complete.

2 Curvature

The following Lemma is immediate from the definition:

Lemma 2.1. One has for the manifold

ℳ_{f}

that:

(1) Christoffel symbols:
- (a) $\nabla_{\partial_{x}} \partial_{x} = f^{'} \partial_{y}$ .
- (b) $\nabla_{\partial_{x}} \partial_{y} = \nabla_{\partial_{y}} \partial_{x} = - f^{'} \partial_{\tilde{x}}$ .
(2) Components of R and of ℛ :
- (a) $ℛ (\partial_{x}, \partial_{y}) \partial_{y} = f^{''} \partial_{\tilde{x}}$ .
- (b) $ℛ (\partial_{x}, \partial_{y}) \partial_{x} = - f^{''} \partial_{y}$ .
- (c) $R (\partial_{x}, \partial_{y}, \partial_{y}, \partial_{x}) = f^{''}$ .
(3) Components of the Ricci tensor ρ :
- (a) $ρ (\partial_{x}, \partial_{x}) = f^{''}$ .
(4) Components of ∇ R and of ∇ ℛ :
- (a) $\nabla_{\partial_{y}} ℛ (\partial_{x}, \partial_{y}) \partial_{y} = f^{'''} \partial_{\tilde{x}}$ .
- (b) $\nabla_{\partial_{y}} ℛ (\partial_{x}, \partial_{y}) \partial_{x} = - f^{'''} \partial_{y}$ .
- (c) $\nabla R (\partial_{x}, \partial_{y}, \partial_{y}, \partial_{x}; \partial_{y}) = f^{'''}$ .
(5) Components of ∇ 2 R and of ∇ 2 ℛ :
- (a) $\nabla_{\partial y} \nabla_{\partial y} ℛ (\partial_{x}, \partial_{y}) \partial_{y} = f^{''''} \partial_{\tilde{x}}$ .
- (b) $\nabla_{\partial y} \nabla_{\partial y} ℛ (\partial_{x}, \partial_{y}) \partial_{x} = - f^{''''} \partial_{y}$ .
- (c) $\nabla_{\partial x} \nabla_{\partial x} ℛ (\partial_{x}, \partial_{y}) \partial_{y} = f^{'} f^{'''} \partial_{\tilde{x}}$ .
- (d) $\nabla_{\partial x} \nabla_{\partial x} ℛ (\partial_{x}, \partial_{y}) \partial_{x} = - f^{'} f^{'''} \partial_{y}$ .
- (e) $\nabla^{2} R (\partial_{x}, \partial_{y}, \partial_{y}, \partial_{x}; \partial_{y}, \partial_{y}) = f^{''''}$ .
- (f ) $\nabla^{2} R (\partial_{x}, \partial_{y}, \partial_{y}, \partial_{x}; \partial_{x}, \partial_{x}) = f^{'} f^{'''}$ .

We shall need a technical lemma related to the structure of

\nabla^{k} R

when

f

is a pure exponential. Let

\nabla^{k} R (\vec{ξ}) : = \nabla^{k} R (ξ_{1}, ξ_{2}, ξ_{3}, ξ_{4}; ξ_{5}, . . ., ξ_{4 + k})

for

\vec{ξ} = (ξ_{1}, . . ., ξ_{4 + k})

. We suppose

ξ_{i} = \partial_{x}

ξ_{i} = \partial_{y}

; there is no need to take

ξ_{i} = \partial_{\tilde{x}}

since

\nabla^{k} R (\vec{ξ})

vanishes if any

ξ_{i} = \partial_{\tilde{x}}

. Let

α (\vec{ξ})

denote the number of times that

ξ_{i} = \partial_{x}

Lemma 2.2. If

f = a e^{b y}

, then

\nabla^{k} R (\vec{ξ}) = γ_{\vec{ξ}} (a, b) e^{\frac{1}{2} α (\vec{ξ}) b y}

Proof. We proceed by induction on

k

. Lemma 2.2 follows from Lemma 2.1 when

k = 0, 1, 2

;

γ_{\vec{ξ}}

is zero if

α (\vec{ξ})

is odd. We have by definition that:

\begin{matrix} \nabla^{k} R (\vec{ξ}) & = & ξ_{k + 4} \nabla^{k - 1} R (ξ_{1}, . . ., ξ_{k + 3}) \end{matrix}

(2.a)

\begin{matrix} - & \sum_{1 \leq i \leq k + 3} \nabla^{k - 1} R (ξ_{1}, . . ., ξ_{i - 1}, \nabla_{ξ_{k + 4}} ξ_{i}, ξ_{i + 1}, . . ., ξ_{k + 3}) . \end{matrix}

(2.b)

Suppose

ξ_{k + 4} = \partial_{y}

. Let

\vec{η} : = (ξ_{1}, . . ., ξ_{k + 3})

;

α (\vec{ξ}) = α (\vec{η})

. Since

\nabla_{\partial_{y}} ξ_{i}

is a multiple of

\partial_{\tilde{x}}

, the terms in ( 2.b ) vanish and only the term in ( 2.a ) enters. Thus:

\begin{matrix} \nabla^{k} R (\vec{ξ}) = \partial_{y} \nabla^{k - 1} R (\vec{η}) = \partial_{y} γ_{\vec{η}} (a, b) e^{\frac{1}{2} α (\vec{η}) b y} \end{matrix}

\begin{matrix} = \frac{1}{2} α (\vec{η}) b γ_{\vec{η}} (a, b) e^{\frac{1}{2} α (\vec{η}) b y} = γ_{\vec{ξ}} (a, b) e^{\frac{1}{2} α (\vec{ξ}) b y} for \end{matrix}

\begin{matrix} γ_{\vec{ξ}} (a, b) : = \frac{1}{2} α (\vec{η}) b γ_{\vec{η}} (a, b) . \end{matrix}

Suppose that

ξ_{k + 4} = \partial_{x}

. Since

ξ_{k + 4} \nabla^{k - 1} R (ξ_{1}, . . ., ξ_{k + 3}) = 0

, the term in ( 2.a ) vanishes. Since

\nabla_{\partial_{x}} \partial_{y}

is a multiple of

\partial_{\tilde{x}}

, we can ignore terms where

ξ_{i} = \partial_{y}

. Set

{\vec{η}}_{i} : = (ξ_{1}, . . ., ξ_{i - 1}, \partial_{y}, ξ_{i + 1}, . . ., ξ_{k + 3}) .

We have

\nabla_{\partial_{x}} \partial_{x} = f^{'} \partial_{y}

. If

ξ_{i} = \partial_{x}

, then

α ({\vec{η}}_{i}) = α (\vec{ξ}) - 2

. We compute:

\begin{matrix} \nabla^{k} R (\vec{ξ}) = - a b e^{b y} \sum_{i : ξ_{i} = \partial_{x}} \nabla^{k - 1} R ({\vec{η}}_{i}) = - a b e^{b y} \sum_{i : ξ_{i} = \partial_{x}} γ_{{\vec{η}}_{i}} (a, b) e^{\frac{1}{2} α ({\vec{η}}_{i}) b y} \end{matrix}

\begin{matrix} = γ_{\vec{ξ}} (a, b) e^{\frac{1}{2} α (\vec{ξ}) b y} for \end{matrix}

\begin{matrix} γ_{\vec{ξ}} (a, b) : = - a b \sum_{i : ξ_{i} = \partial_{x}} γ_{{\vec{η}}_{i}} (a, b) . \end{matrix}

The Lemma now follows from these two special cases. □

Proof of Theorem 1.2 . We show that all the scalar Weyl invariants of

ℳ_{f}

vanish as follows. Consider the orthonormal basis

e_{1}^{+} : = \partial_{x} + \frac{1}{2} \partial_{\tilde{x}}, e_{2}^{-} : = \partial_{x} - \frac{1}{2} \partial_{\tilde{x}}, e_{3}^{+} : = \partial_{y} .

We form scalar Weyl invariants by contracting indices in pairs and then summing over repeated indices. Since

\nabla^{k} R (. . ., e_{1}^{+}, . . .) = \nabla^{k} R (. . ., e_{2}^{-}, . . .) = \nabla^{k} R (. . ., \partial_{x}, . . .)

, since

g^{11} = + 1

, and since

g^{22} = - 1

, terms where

e_{i} = e_{1}^{+}

(i.e.

i = 1

) and terms where

e_{i} = e_{2}^{-}

(i.e.

i = 2

) appear with opposite signs in any Weyl summation and cancel;

\nabla^{k} R (. . ., e_{3}^{+}, . . .) = 0

. Assertion (1) now follows.

f

is quadratic, then

\nabla R = 0

. Thus

S_{+}

and

S_{-}

are local symmetric spaces.

The curvature tensor and metric are indecomposible; they are not irreducible as

Span {\partial_{\tilde{x}}}

is invariant under the isotropy representation. Assertion (2) follows.

Set

ɛ_{f} : = sign (f^{''})

. We say that a basis

ℬ = {X, Y, \tilde{X}}

is normalized if we have:

\begin{matrix} \begin{matrix} 1) the non-zero components of g are g (X, \tilde{X}) = g (Y, Y) = 1, \\ 2) the non-zero components of R are R (X, Y, Y, X) = ɛ_{f}, \\ 3) we have \nabla R (ξ_{1}, ξ_{2}, ξ_{3}, ξ_{4}; X) = 0 \forall ξ_{1}, ξ_{2}, ξ_{3}, ξ_{4} . \end{matrix} \end{matrix}

(2.c)

We may define a normalized basis, and thereby establish Assertion (3), by setting

\begin{matrix} X : = | f^{''} |^{- 1 / 2} {\partial_{x} + f \partial_{\tilde{x}}}, Y : = \partial_{y}, \tilde{X} : = | f^{''} |^{1 / 2} \partial_{\tilde{x}} . \end{matrix}

(2.d)

We say

ℬ

is affine normalized if the non-zero components of

ℛ

and

\nabla ℛ

are

\begin{matrix} ℛ (X, Y) Y = ɛ_{f} \tilde{X}, & ℛ (X, Y) X = - ɛ_{f} Y, \end{matrix}

\begin{matrix} \nabla_{Y} ℛ (X, Y) Y = \tilde{X}, & \nabla_{Y} ℛ (X, Y) X = - Y . \end{matrix}

We construct an affine normalized basis by rescaling the coordinate frame. Let

a_{1}

a_{2}

, and

a_{3}

be constants to be determined. By Lemma 2.1 ,

\begin{matrix} ℛ (a_{1} \partial_{x}, a_{2} \partial_{y}) a_{2} \partial_{y} = a_{1} a_{2}^{2} a_{3}^{- 1} f^{''} a_{3} \partial_{\tilde{x}}, \end{matrix}

\begin{matrix} ℛ (a_{1} \partial_{x}, a_{2} \partial_{y}) a_{1} \partial_{x} = - a_{1}^{2} f^{''} a_{2} \partial_{y}, \end{matrix}

\begin{matrix} \nabla_{a_{2} \partial_{y}} ℛ (a_{1} \partial_{x}, a_{2} \partial_{y}) a_{2} \partial_{y} = a_{1} a_{2}^{3} a_{3}^{- 1} f^{'''} a_{3} \partial_{\tilde{x}}, \end{matrix}

\begin{matrix} \nabla_{a_{2} \partial_{y}} ℛ (a_{1} \partial_{x}, a_{2} \partial_{y}) a_{1} \partial_{x} = - a_{1}^{2} a_{2} f^{'''} a_{2} \partial_{y} . \end{matrix}

Assume that

f^{''} (y)

and

f^{'''} (y)

never vanish. We define an affine normalized basis and prove Assertion (4a) by setting

\begin{matrix} \begin{matrix} X : = a_{1} \partial_{x}, & Y : = a_{2} \partial_{y}, & \tilde{X} : = a_{3} \partial_{\tilde{x}} & where \\ a_{1} : = {| f^{''} |}^{- 1 / 2}, & a_{2} : = | f^{''} | {f^{'''}}^{- 1}, & a_{3} : = a_{1} a_{2}^{2} | f^{''} | . \end{matrix} \end{matrix}

(2.e)

We note for future reference that

\begin{matrix} \begin{matrix} \nabla_{X} \nabla_{X} ℛ (X, Y) Y = a_{1}^{3} a_{2}^{2} \nabla_{\partial_{x}} \nabla_{\partial_{x}} ℛ (\partial_{x}, \partial_{y}) \partial_{y} = a_{1}^{3} a_{2}^{2} f^{'} f^{'''} \partial_{\tilde{x}} \\ = f^{'} f^{'''} {f^{''}}^{- 2} \tilde{X} . \end{matrix} \end{matrix}

(2.f )

We study the relevant symmetry group to construct additional invariants of the

1

-model. Let

ℬ = {X, Y, \tilde{X}}

be the normalized basis defined in Equation ( 2.d ).

Suppose that

ℬ_{1} = {X_{1}, Y_{1}, {\tilde{X}}_{1}}

is another normalized basis. Expand:

\begin{matrix} X_{1} = a_{11} X + a_{12} Y + a_{13} \tilde{X}, \end{matrix}

\begin{matrix} Y_{1} = a_{21} X + a_{22} Y + a_{23} \tilde{X}, \end{matrix}

\begin{matrix} {\tilde{X}}_{1} = a_{31} X + a_{32} Y + a_{33} \tilde{X} . \end{matrix}

Since

R (ξ_{1}, ξ_{2}, ξ_{3}, {\tilde{X}}_{1}) = 0

for any

ξ_{1}, ξ_{2}, ξ_{3}

, we have

a_{31} = a_{32} = 0

. Since

\nabla R (ξ_{1}, ξ_{2}, ξ_{3}, ξ_{4}; X_{1}) = 0

for any

ξ_{1}, ξ_{2}, ξ_{3}, ξ_{4}

a_{12} = 0

. Thus

X_{1} = a_{11} X + a_{13} \tilde{X}, Y_{1} = a_{21} X + a_{22} Y + a_{23} \tilde{X}, {\tilde{X}}_{1} = a_{33} \tilde{X} .

g (X_{1}, {\tilde{X}}_{1}) = 1

a_{33} a_{11} = 1

. As

g (Y_{1}, {\tilde{X}}_{1}) = 0

a_{21} = 0

. As

g (X_{1}, X_{1}) = 0

a_{13} = 0

. Consequently,

X_{1} = a_{11} X, Y_{1} = a_{22} Y + a_{23} \tilde{X}, {\tilde{X}}_{1} = a_{11}^{- 1} \tilde{X} .

g (X_{1}, Y_{1}) = 0

a_{23} = 0

. As

g (Y_{1}, Y_{1}) = 1

a_{22}^{2} = 1

. As

R (X_{1}, Y_{1}, Y_{1}, X_{1}) = ɛ_{f}

a_{11}^{2} a_{22}^{2} = 1

. Thus

X_{1} = a_{11} X, Y_{1} = a_{22} Y, {\tilde{X}}_{1} = a_{11}^{- 1} \tilde{X} where a_{11}^{2} = a_{22}^{2} = 1 .

In particular we may use Equation ( 2.d ) and Lemma 2.1 to see:

| \nabla R (X_{1}, Y_{1}, Y_{1}, X_{1}; Y_{1}) | = | \nabla R (X, Y, Y, X; Y) | = | f^{'''} {f^{''}}^{- 1} |

is an invariant of the

1

-model. This is constant if and only if

f^{'''} = c f^{''}

, i.e.

f^{''} = a e^{b y}

. Assertion (4b) now follows.

We now establish Assertion (4c). The following implications are immediate:

(4c-i) \Rightarrow (4c-ii) \Rightarrow (4c-iii) \Rightarrow (4c-iv) .

Suppose that

ℳ_{f}

2

-affine curvature homogeneous. Let

ℬ : = {X, Y, \tilde{X}}

be the affine normalized basis defined in Equation ( 2.e ). Suppose that

ℬ_{1} : = {X_{1}, Y_{1}, {\tilde{X}}_{1}}

is another affine normalized basis. Let

\begin{matrix} ℐ_{0} : = {Span}_{ξ_{1}, ξ_{2}, ξ_{3}} {ℛ (ξ_{1}, ξ_{2}) ξ_{3}} = Span {Y, \tilde{X}}, \end{matrix}

\begin{matrix} K_{0} : = {η : ℛ (η, ξ_{1}) ξ_{2} = 0 for all ξ_{1}, ξ_{2}} = Span {\tilde{X}}, \end{matrix}

\begin{matrix} K_{1} : = {η : \nabla_{η} ℛ (ξ_{1}, ξ_{2}) ξ_{3} = 0 for all ξ_{1}, ξ_{2}, ξ_{3}} = Span {X, \tilde{X}} . \end{matrix}

Since these spaces are invariantly defined, we may expand

X_{1} = a_{11} X + a_{13} \tilde{X}, Y_{1} = a_{22} Y + a_{23} \tilde{X}, {\tilde{X}}_{1} = a_{33} \tilde{X} .

We have

ℛ (X_{1}, Y_{1}) X_{1} = a_{11}^{2} ℛ (X, Y_{1}) X

is a multiple of

Y

. Since the basis is normalized, it is also a multiple of

Y_{1}

. Thus

a_{23} = 0

. We now compute

\begin{matrix} ℛ (X_{1}, Y_{1}) Y_{1} = a_{11} a_{22}^{2} a_{33}^{- 1} ɛ_{f} {\tilde{X}}_{1}, & ℛ (X_{1}, Y_{1}) X_{1} = - a_{11}^{2} ɛ_{f} Y_{1}, \end{matrix}

\begin{matrix} \nabla_{Y_{1}} ℛ (X_{1}, Y_{1}) Y_{1} = a_{11} a_{22}^{3} a_{33}^{- 1} {\tilde{X}}_{1}, & \nabla_{Y_{1}} ℛ (X_{1}, Y_{1}) X_{1} = - a_{11}^{2} a_{22} Y_{1} . \end{matrix}

As the basis is normalized,

a_{11}^{2} = 1

a_{22} = 1

a_{33} = a_{11}

. Thus by Equation ( 2.f ),

\begin{matrix} \nabla_{X_{1}} \nabla_{X_{1}} ℛ (X_{1}, Y_{1}) Y_{1} = a_{11}^{3} a_{22}^{2} \nabla_{X} \nabla_{X} ℛ (X, Y) Y \end{matrix}

\begin{matrix} = & a_{11}^{3} a_{22}^{2} f^{'} f^{'''} {f^{''}}^{- 2} \tilde{X} = a_{11}^{3} a_{22}^{2} a_{33}^{- 1} f^{'} f^{'''} {f^{''}}^{- 2} {\tilde{X}}_{1} \end{matrix}

\begin{matrix} = & f^{'} f^{'''} {f^{''}}^{- 2} {\tilde{X}}_{1} . \end{matrix}

This shows that

f^{'} f^{'''} {f^{''}}^{- 2}

is an invariant of the affine 2-model. Consequently if

ℳ_{f}

2

-affine curvature homogeneous, then

f^{'} f^{'''} {f^{''}}^{- 2} = c .

c = 1

, then

f^{'} = a e^{b y}

. If

c \neq 1

, then

f^{'} = α (y + β)^{γ}

for some

{α, β, γ}

. This later choice is ruled out as

f^{'''}

and

f^{''}

are assumed to be globally defined and non-zero.

Thus we may conclude

f^{'} = a e^{b y}

; this establishes the implication:

(4c-iv) \Rightarrow (4c-v) .

Finally, we suppose that

f^{'} = a e^{b y}

; we take

f = \frac{a}{b} e^{b y}

. Consider the normalized basis

{X, Y, \tilde{X}}

defined in Equation ( 2.d ):

X = | a b |^{- 1 / 2} e^{- b y / 2} {\partial_{x} + \frac{a}{b} e^{b y} \partial_{\tilde{x}}}, Y = \partial_{y}, \tilde{X} = | a b |^{1 / 2} e^{b y / 2} \partial_{\tilde{x}} .

Let

\nabla^{k} R (\vec{η}) = \nabla^{k} R (η_{1}, η_{2}, η_{3}, η_{4}; η_{5}, . . ., η_{4 + k})

for

\vec{η} = (η_{1}, . . ., η_{4 + k})

where

η_{i} = X

η_{i} = Y

for

1 \leq i \leq 4 + k

. Let

\vec{ξ}

be the corresponding string where

X

and

Y

are replaced by

\partial_{x}

and

\partial_{y}

. Let

α (\vec{η}) = α (\vec{ξ})

be the number of times that

X

or equivalently

\partial_{x}

appear. We apply Lemma 2.2 to see:

\begin{matrix} \nabla^{k} R (\vec{η}) & = & | a b |^{- α (η) / 2} e^{- α (\vec{η}) b y / 2} \nabla^{k} R (\vec{ξ}) \end{matrix}

\begin{matrix} = & | a b |^{- α (η) / 2} e^{- α (\vec{η}) b y / 2} γ_{\vec{ξ}} (a, b) e^{α (\vec{ξ}) b y / 2} \end{matrix}

\begin{matrix} = & | a b |^{- α (η) / 2} γ_{\vec{ξ}} (a, b) . \end{matrix}

This shows that

ℳ_{f}

k

-curvature homogeneous for all

k

; a local version of Theorem 1.1 now shows that

ℳ_{f}

is locally homogeneous as desired. Consequently, (4c-v)

\Rightarrow

(4c-i). □

3 Complete manifolds

Let

γ (t) = (x (t), y (t), \tilde{x} (t))

be a path in

ℳ_{f}

. The geodesic equation becomes

x^{''} (t) = 0, y^{''} (t) = - f^{'} (y (t)) x^{'} (t) x^{'} (t), {\tilde{x}}^{''} (t) = 2 f^{'} (y (t)) y^{'} (t) x^{'} (t) .

The first equation yields

x (t) = x_{0} + x_{1} t

. The remaining equations then become

y^{''} (t) = - x_{1}^{2} f^{'} (y (t)), {\tilde{x}}^{''} (t) = 2 x_{1} f^{'} (y (t)) y^{'} (t) .

The equation for

y

is the crucial one; once

y

is determined, one can express

\begin{matrix} \tilde{x} (t) = {\tilde{x}}_{0} + t {\tilde{x}}_{0}^{'} + 2 x_{1} \int_{s = 0}^{t} \int_{u = 0}^{s} f^{'} (y (u)) y^{'} (u) d u d s . \end{matrix}

(3.a)

Proof of Theorem 1.3 . First set $f_{+} (y) = \frac{1}{2} y^{2}$ . We then have to solve $y^{''} (t) = - x_{1}^{2} y (t) .$ We show $S_{+}$ is geodesically complete by solving this equation:
$y (t) = {\begin{matrix} y_{0} + y_{0}^{'} t & if & x_{1} = 0, \\ y_{0} cos (x_{1} t) + \frac{1}{x_{1}} y_{0}^{'} sin (x_{1} t) & if & x_{1} \neq 0 . \end{matrix}$ A geodesic with $x (0) = x_{0}$ and $x (1) = x_{0} + 2 π$ has the form:
$γ (t) = (x_{0} + 2 π t, y_{0} cos (2 π t) + \frac{1}{2 π} y_{0}^{'} sin (2 π t), \tilde{x} (t)) .$ Thus $y (1) = y (0)$ and the exponential map is not surjective. This establishes the Assertions of the Lemma concerning $S_{+}$ .
Next, we study $f_{-} (y) = - \frac{1}{2} y^{2}$ . We then have to solve $y^{''} (t) = x_{1}^{2} y (t) .$ We show $S_{-}$ is geodesically complete by solving this equation:
$y (t) = {\begin{matrix} y_{0} + y_{0}^{'} t & if & x_{1} = 0, \\ \frac{1}{2} y_{0} {e^{x_{1} t} + e^{- x_{1} t}} + \frac{1}{2 x_{1}} y_{0}^{'} {e^{x_{1} t} - e^{- x_{1} t}} & if & x_{1} \neq 0 . \end{matrix}$ We take $P : = (x_{0}, y_{0}, {\tilde{x}}_{0})$ as the initial point. Suppose $Q : = (x_{1}, y_{1}, {\tilde{x}}_{1})$ is given.
The exponential map is given by setting $t = 1$ . Thus $x (t) = x_{0} + t (x_{1} - x_{0})$ . If $x_{1} - x_{0} = 0$ , then set $y (t) = y_{0} + t (y_{1} - y_{0})$ . If $x_{1} - x_{0} \neq 0$ , we determine $y_{0}^{'}$ uniquely by solving the equation:
$y_{1} = \frac{1}{2} y_{0} {e^{x_{1} - x_{0}} + e^{x_{0} - x_{1}}} + \frac{1}{2 (x_{1} - x_{0})} y_{0}^{'} {e^{x_{1} - x_{0}} - e^{x_{0} - x_{1}}} .$ Once $x$ and $y$ have been determined, we then use Equation ( 3.a ) to solve for ${\tilde{x}}_{0}^{'}$ .
This shows that $S_{-}$ is geodesically complete and that the exponential map is a diffeomorphism. □

4 The proof of Theorem 1.4

Following the discussion in Section 3 , to construct geodesics in the manifold

ℳ_{f}

, we must solve the ODE

y^{''} = - x_{1}^{2} f^{'} (y) .

We shall suppose

x_{1} \neq 0

and set

h = - x_{1}^{2} f^{'}

. We begin with:

Lemma 4.1. Let

h : R \to (- \infty, 0)

be smooth. Let

[0, T)

be the maximal domain of the solution

y

to the ODE

y^{''} = h (y)

where

y (0) = y_{0}

and

y^{'} (0) = y_{0}^{'}

. If

T < \infty

{lim}_{t \to T} y (t) = {lim}_{t \to T} y^{'} (t) = - \infty and {limsup}_{y \to T} \frac{h (y (t))}{y (t)} = \infty .

Proof. Since $y^{''} < 0$ , $y^{'}$ is monotonically decreasing and $y$ is bounded from above on $[0, T)$ . Suppose first that $y$ is bounded from below on $[0, T)$ . This implies that $y^{''}$ is bounded and hence $y^{'}$ is bounded as well on $[0, T)$ . Let $y_{1} = lim {inf}_{t \to T} y (t) and y_{1}^{'} = {lim}_{t \to T} y^{'} (t) .$ The fundamental theorem of ODE's shows there exists $κ > 0$ so that if $| z_{1} - y_{1} | < κ, | z_{1}^{'} - y_{1}^{'} | < κ, and s \in (T - κ, T)$ then there exists a solution $z$ to the equation $z^{''} = h (z)$ with initial conditions $z (s) = z_{1}$ and $z^{'} (s) = z_{1}^{'}$ which is valid on the interval $[s, s + κ)$ . We choose $s \in (T - \frac{1}{2} κ, T) so that | y (s) - y_{1} | < κ and | y^{'} (s) - y_{1}^{'} | < κ .$ Let $z^{''} = h (z)$ be defined on $[s, s + κ)$ with $z (s) = y (s)$ and $z^{'} (s) = y^{'} (s)$ . Then $z$ extends $y$ to the region $[0, T + \frac{1}{2} κ)$ which contradicts the assumption that $[0, T)$ was a maximal domain.
Thus $y$ is not bounded from below on $[0, T)$ so ${lim}_{t \to T} y^{'} (t) = - \infty$ . Consequently, $y$ is monotonically decreasing for $t$ close to $T$ so ${lim}_{t \to T} y (t) = - \infty$ as well. Suppose ${limsup}_{t \to T} \frac{h (y (t))}{y (t)} < \infty$ i.e. that there exists $C < \infty$ so $| h (y (t)) | \leq C | y (t) |$ on $[t_{0}, T)$ . We then have ${ln | y (t) |}^{''} = {\frac{y^{'} (t)}{y (t)}}^{'} = \frac{y^{''} (t)}{y (t)} - {\frac{y^{'} (t)}{y (t)}}^{2} = \frac{h (y (t))}{y (t)} - {\frac{y^{'} (t)}{y (t)}}^{2} \leq C .$ This implies $ln | y (t) |$ is bounded from above and hence $| y (t) |$ is bounded from above on $[t_{0}, T)$ which is false. This contradiction shows ${limsup}_{t \to T} \frac{h (y (t))}{y (t)} = \infty$ . □

Proof of Theorem 1.4 (1). We suppose that $f^{'} > 0$ and that $f^{'} (y) \leq C | y |$ for $y \leq - 1$ . We set $h = - x_{1}^{2} f^{'}$ . Choose a maximal domain $[0, T)$ for the solution to the ODE $y^{''} = h (y)$ with initial condition $y (0) = y_{0}$ and $y^{'} (0) = y_{0}^{'}$ . If $T < \infty$ , then ${limsup}_{t \to T} \frac{h (y (t))}{y (t)} = \infty$ which is false. Thus $T = \infty$ and $ℳ_{f}$ is geodesically complete. □

Before proving Theorem 1.4 (2), we must establish:

Lemma 4.2.

(1) Let $α > 0$ . Let ${t_{n}}_{n \geq 1}$ be a sequence of real numbers with $t_{1} = 1$ and with $t_{n + 1} - t_{n} \geq n^{α}$ for $n \geq 1$ . Then $t_{n} \geq \frac{n^{1 + α}}{(1 + α) 2^{1 + α}}$ .
(2) Let $ε > 0$ and $δ > 0$ . Suppose that $h (y) < - ε | y |^{1 + δ}$ for $y \leq - 1$ .
Let $[0, T)$ be the maximal domain of definition for the solution $y$ to the ODE $y^{''} = h (y)$ with $y (0) = - 1$ and $y^{'} (0) = - 1$ . Then $T < \infty$ and ${lim}_{t \to T} y (t) = - \infty$ .

Proof. We prove Assertion (1) by induction on

n

; it holds trivially for

n = 1

. We take

n \geq 2

and use the comparison test to compute:

\begin{matrix} t_{n} & > & t_{n} - t_{1} = \sum_{k = 2}^{n} {t_{k} - t_{k - 1}} \geq \int_{1}^{n} (x - 1)^{α} d x \end{matrix}

\begin{matrix} = & \frac{(n - 1)^{1 + α}}{1 + α} = \frac{n^{1 + α}}{(1 + α) (1 + \frac{1}{n - 1})^{1 + α}} \geq \frac{n^{1 + α}}{(1 + α) 2^{1 + α}} . \end{matrix}

To prove Assertion (2), we suppose first

T = \infty

and argue for a contradiction.

Choose

τ

so that

τ ɛ \geq 2^{1 + δ / 2} (1 + δ / 2) and τ \geq 1 .

With our initial conditions,

y^{''} < 0

y^{'}

is monotonically decreasing and

y^{'} \leq - 1

This implies

y

decreases monotonically. Let

Δ_{n} = τ \cdot n^{- 1 - δ / 2}

. Let

s_{1} = 0

and let

s_{n + 1} = s_{n} + Δ_{n}

for

n \geq 2

. As

δ > 0

S : = {lim}_{n \to \infty} s_{n} = \sum_{n = 1}^{\infty} τ n^{- 1 - δ / 2} < \infty .

We wish to show inductively that

(1) $y^{'} (s_{n}) \leq - n^{1 + δ / 2}$ .
(2) $y (s_{n}) \leq - n$ .
(3) $y^{'} (s_{n + 1}) - y^{'} (s_{n}) \leq - 2^{1 + δ / 2} (1 + \frac{1}{2} δ) n^{δ / 2}$ .

The first two statements hold

n = 1

by the choice of our initial conditions. Since

y

and

y^{'}

decrease monontonically, we may estimate

\begin{matrix} y^{''} (s) \leq - ε | y (s) |^{1 + δ} \leq - ε | y (s_{n}) |^{1 + δ} \leq - ε n^{1 + δ} for s \in [s_{n}, s_{n + 1}], \end{matrix}

\begin{matrix} y^{'} (s_{n + 1}) - y^{'} (s_{n}) \leq - Δ_{n} ε n^{1 + δ} = - τ n^{- 1 - δ / 2} ε n^{1 + δ} \leq - 2^{1 + δ / 2} (1 + δ / 2) n^{δ / 2} . \end{matrix}

Thus statements

(1)_{n}

and

(2)_{n}

imply assertion

(3)_{n}

Statements

(3)_{k}

for

1 \leq k \leq n

together with Assertion (1) imply Statement

(1)_{n + 1}

. Finally, we use Statement

(1)_{n}

together with statement

(2)_{n}

to establish Statement

(2)_{n + 1}

by computing:

\begin{matrix} y^{'} (s) \leq y^{'} (s_{n}) \leq - n^{1 + δ / 2} for s \geq s_{n}, \end{matrix}

\begin{matrix} y (s_{n + 1}) \leq y (s_{n}) + Δ_{n} y^{'} (s_{n}) \leq - n - τ n^{- 1 - δ / 2} n^{1 + δ / 2} \leq - n - 1 . \end{matrix}

This establishes the truth of all the 3 statements. Thus,

{lim}_{s \to S} y (s) = - \infty

. This contradicts the assumption that

T = \infty

This shows that

y

must be defined on a maximal domain

[0, T)

for

T < \infty

; the fact that

{lim}_{t \to T} y (t) = - \infty

now follows from Lemma 4.1 . □

Proof of Theorem 1.4 (2). Suppose $f^{'} (y) > 0$ for all $y$ and that $f^{'} (y) \geq ɛ | y |^{1 + δ}$ for $y \leq - 1$ . Choose a geodesic with $x (0) = 0$ , $x^{'} (0) = 1$ , $y (0) = - 1$ , and $y^{'} (0) = - 1$ .
We then have the differential equation $y^{''} = - f^{'} (y) .$ Thus by Lemma 4.2 for some finite time $T$ , we have ${lim}_{t \to T} y (t) = - \infty$ . Thus $ℳ_{f}$ is geodesically incomplete. We have $ρ (\dot{γ}, \dot{γ}) = f^{''} (y (t))$ .
If $| f^{''} (y) | \leq K$ on $(- \infty, 0]$ , then $f^{'} (y) \leq K | y | + f^{'} (y (0))$ on $(- \infty, 0]$ which is false.
Thus $| f^{''} (y) |$ is not bounded on $(- \infty, 0]$ . Since $y (t) \to - \infty$ as $t \to T$ , $f^{''} (y (t))$ is not bounded on $[0, T)$ . This shows, as desired, that $ℳ_{f}$ Ricci explodes. □

Acknowledgments

Research of P. Gilkey partially supported by the Max Planck Institute in the Mathematical Sciences (Leipzig). Research of S. Nikčević partially supported by MM 1646 (Srbija).

References

E. Boeckx, L. Vanhecke, and O. Kowalski, Riemannian manifolds of conullity two, World Scientific (1996).
P. Bueken and M. Djorić, Three-dimensional Lorentz metrics and curvature homogeneity of order one, Ann. Global Anal. Geom. 18 (2000), 85–103.
P. Bueken and L. Vanhecke, Examples of curvature homogeneous Lorentz metrics, Classical Quantum Gravity 14 (1997), L93–L96.
M. Cahen, J. Leroy, M. Parker, F. Tricerri, and L. Vanhecke, Lorentz manifolds modeled on a Lorentz symmetric space, J. Geom. Phys. 7 (1990), 571-581.
P. Gilkey and S. Nikčevíc, Complete $k$ -curvature homogeneous pseudo-Riemannian manifolds, to appear Annals Global Analysis and Geometry, math.DG/0405024.
O. Kowalski, B. Opozda, and Z. Vlášek, Curvature homogeneity of affine connections on two-dimensional manifolds, Colloq. Math. 81 (1999), 123–139.
A. Koutras and C. McIntosh, A metric with no symmetries or invariants, Classical Quantum Gravity 13 (1996), L47-L49.
B. Opozda, On curvature homogeneous and locally homogeneous affine connections, Proc. Amer. Math. Soc. 124 (1996), 1889–1893.
B. Opozda, Affine versions of Singer's theorem on locally homogeneous spaces, Ann. Global Anal. Geom. 15 (1997), 187–199.
F. Podesta and A. Spiro, Introduzione ai Gruppi di Trasformazioni, Volume of the Preprint Series of the Mathematics Department ”V. Volterra” of the University of Ancona, Via delle Brecce Bianche, Ancona, ITALY (1996).
V. Pravda, A. Pravdová, A. Coley, and R. Milson, All spacetimes with vanishing curvature invariants, Classical Quantum Gravity 19 (2002), 6213–6236.
I. M. Singer, Infinitesimally homogeneous spaces, Commun. Pure Appl. Math. 13 (1960), 685–697.
H. Weyl, The Classical Groups Princeton University Press, Princeton, 1946.

PG: Mathematics Department, University of Oregon, Eugene Or 97403 USA. Email: gilkey@darkwing.uoregon.edu SN: Mathematical Institute, SANU, Knez Mihailova 35, p.p. 367, 11001 Belgrade, Serbia and Montenegro. Email: stanan@mi.sanu.ac.yu

P. Gilkey and S. Nikčević