20.04.2013 Views

Singular metrics and associated conformal groups underlying ...

Singular metrics and associated conformal groups underlying ...

Singular metrics and associated conformal groups underlying ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Annali di matematica pura ed applicata manuscript No.<br />

(will be inserted by the editor)<br />

Kevin R. Payne<br />

<strong>Singular</strong> <strong>metrics</strong> <strong>and</strong> <strong>associated</strong><br />

<strong>conformal</strong> <strong>groups</strong> <strong>underlying</strong><br />

differential operators of mixed <strong>and</strong><br />

degenerate types<br />

Received: date<br />

Abstract For partial differential equations of mixed elliptic-hyperbolic <strong>and</strong><br />

degenerate types which are the Euler-Lagrange equations for an <strong>associated</strong><br />

Lagrangian, we examine an <strong>associated</strong> metric structure which becomes singular<br />

on the hypersurface where the operator degenerates. In particular, we<br />

show that the “non-trivial part” of the complete symmetry group for the<br />

differential operator (calculated in a previous paper [15]) corresponds to a<br />

group of local <strong>conformal</strong> transformations with respect to the metric away<br />

from the metric singularity <strong>and</strong> that the group extends smoothly across the<br />

singular surface. In this way, we define <strong>and</strong> calculate the <strong>conformal</strong> group for<br />

these operators as well as for lower order singular perturbations which are<br />

defined naturally by the singular metric.<br />

Keywords Symmetry <strong>groups</strong> · Mixed type equations · <strong>Singular</strong> Riemannian-<br />

Lorentzian <strong>metrics</strong> · Conformal transformations<br />

Mathematics Subject Classification (2000) 35M10 · 58J70 · 53A30<br />

1 Introduction<br />

The purpose of this note is to formulate precisely <strong>and</strong> to prove various claims<br />

made in a recent paper [15] concerning the association of a group of <strong>conformal</strong><br />

transformations to a class of differential operators of mixed-elliptic hyperbolic<br />

or degenerate type. We consider the operator<br />

Work supported by MIUR, Project “Metodi Variazionali ed Equazioni Differenziali<br />

Non Lineari” <strong>and</strong> MIUR, Project “Metodi Variazionali e Topologici nello Studio di<br />

Fenomeni Non Lineari”.<br />

K. R. Payne<br />

Dipartimento di Matematica “F. Enriques”, Università di Milano, Via Saldini 50,<br />

20133 Milano, Italy<br />

E-mail: payne@mat.unimi.it


2 K. R. Payne<br />

L = K(y)∆x + ∂ 2 y<br />

(1.1)<br />

where (x, y) ∈ R N × R, ∆x is the Laplace operator on R N with N ≥ 1, <strong>and</strong><br />

the coefficient K ∈ C 0 (R) satisfies<br />

K(0) = 0 <strong>and</strong> K(y) = 0 for y = 0. (1.2)<br />

The equation degenerates along the hypersurface y = 0 <strong>and</strong> our main interest<br />

will be cases in which K yields a change of type; that is, K also satisfies<br />

yK(y) > 0 for y = 0 (1.3)<br />

so that the operator (1.1) is of mixed type (elliptic for y > 0 <strong>and</strong> hyperbolic<br />

for y < 0). However, much of what will be discussed depends only on the<br />

form of the degeneracy in (1.1) − (1.2).<br />

In the paper [15], we classified the symmetry <strong>groups</strong> <strong>and</strong> calculated the<br />

<strong>associated</strong> conservation laws for the equation Lu = 0, which is the Euler-<br />

Lagrange equation for the Lagrangian<br />

L(y, u, ∇u) = 1 <br />

K(y)|∇xu|<br />

2<br />

2 + u 2 y . (1.4)<br />

In fact, the class defined by (1.1) − (1.3) represents the simplest examples<br />

of second order equations of mixed type <strong>associated</strong> to a Lagrangian with<br />

degeneracy on a hypersurface. As is to be expected, the largest possible<br />

symmetry <strong>groups</strong> occur when K takes a pure power form<br />

K(y) = y|y| m−1 , m > 0 (1.5)<br />

in the mixed type case, or ±|y| m in the purely elliptic/hyperbolic but degenerate<br />

cases. The operator in (1.1) with (1.5) is known as the Gellerstedt<br />

operator <strong>and</strong> gives the Tricomi operator when K(y) = y, while the choice<br />

K(y) = y 2 yields the degenerate elliptic Grushin operator. Such operators<br />

arise in many physical <strong>and</strong> geometrical problems with a particular structure,<br />

such as: transonic fluid flow [3] [16], quantum cosmology [11], <strong>and</strong> the imbedding<br />

of manifolds with curvature that changes sign [13]. See also section 6 of<br />

[15] for a brief discussion.<br />

If one takes the limiting case, m = 0 in (1.5), one arrives at the Laurentiev-<br />

Bitsadze operator which glues the Laplacian for y > 0 to the D’Alembertian<br />

for y < 0. It is well known that the symmetry <strong>groups</strong> for these classical<br />

operators are given by the group of <strong>conformal</strong> transformations with respect<br />

to the Euclidian <strong>and</strong> Minkowski <strong>metrics</strong> respectively (cf. [17] <strong>and</strong> [5]). We will<br />

show that this also holds in a suitably interpreted sense for the mixed type<br />

operators satisfying (1.1) <strong>and</strong> (1.5) with respect to a suitable singular metric<br />

of mixed Riemannian-Lorentzian signature as announced in [15]. Moreover,<br />

the analogous result holds in the purely elliptic/hyperbolic but degenerate<br />

setting of (1.1) with K(y) = ±|y| m , m > 0.


Conformal <strong>groups</strong> for operators of mixed <strong>and</strong> degenerate types 3<br />

2 Symmetry <strong>groups</strong> for the operator L<br />

We recall briefly the results in [15] on the symmetry <strong>groups</strong> for the equation<br />

Lu = 0 (2.1)<br />

with L of the form (1.1) − (1.2) <strong>and</strong> with power type degeneration (1.5) (or<br />

K(y) = ±|y| m ). One has, apart from certain trivial symmetries for these<br />

linear <strong>and</strong> homogeneous equations, symmetries coming from: 1) translations<br />

in the “space variables” x; 2) rotations in the space variables; 3) certain<br />

anisotropic dilations; 4) inversion with respect to a well chosen hypersurface<br />

(cf. Theorem 2.5 of [15]).<br />

More precisely, the trivial one parameter symmetry <strong>groups</strong> arise from the<br />

fact that if u solves (2.1) then so does u + ɛβ with β any solution of (2.1)<br />

<strong>and</strong> ɛ ∈ R. The non trivial symmetries are represented by the fact that if u<br />

solves (2.1) with K(y) = y|y| m−1 , m > 0, then so do<br />

<strong>and</strong><br />

uk;ɛ(x, y) = Tk;ɛu(x, y) = u(x − ɛek, y) (2.2)<br />

uj,k;ɛ(x, y) = Rj,k;ɛu(x, y) = u(Aj,k;ɛx, y) (2.3)<br />

uλ(x, y) = Sλu(x, y) = λ −p(m,N) <br />

u λ −(m+2) x, λ −2 <br />

y<br />

uk;ɛ(x, y) = Ik;ɛu(x, y) = D −q(N,m)<br />

k,ɛ u<br />

<br />

x + ɛdek<br />

,<br />

Dk,ɛ<br />

where ɛ ∈ R (<strong>and</strong> |ɛ| is small in (2.5)), λ > 0, {ek} N k=1<br />

of R N , Aj,k;ɛ is the ɛ-rotation in the xj − xk plane,<br />

where<br />

<strong>and</strong><br />

p(m, N) =<br />

N(m + 2) − 2<br />

2<br />

y<br />

D 2/(m+2)<br />

k,ɛ<br />

<br />

(2.4)<br />

(2.5)<br />

is the st<strong>and</strong>ard basis<br />

> 0, (2.6)<br />

Dk,ɛ(x, y) = 1 + 2ɛxk + ɛ 2 d(x, y), (2.7)<br />

d(x, y) = |x| 2 +<br />

q(m, N) =<br />

4<br />

(m + 2) 2 y|y|m+1 , (2.8)<br />

N(m + 2) − 2<br />

2(m + 2)<br />

> 0. (2.9)<br />

The same result holds for the degenerate elliptic/hyperbolic cases where<br />

K(y) = ±|y| m where it is enough to replace y|y| m+1 with ±|y| m+2 in (2.8).<br />

The symmetries (2.2) − (2.4) are variational in the sense that they leave<br />

invariant the integral of the Lagrangian (1.4) while the symmetry (2.5) is<br />

a divergence symmetry <strong>and</strong> is only locally well defined. All yield <strong>associated</strong>


4 K. R. Payne<br />

conservation laws. The infinitesimal generators of the nontrivial symmetries<br />

are given by the vector fields<br />

v T k = ∂<br />

, k = 1, . . . , N (2.10)<br />

∂xk<br />

v D = (m + 2)x · ∇x + 2y ∂<br />

∂y<br />

v R jk = xk<br />

v I k = −d(x, y) ∂<br />

N(m + 2) − 2<br />

− u<br />

2<br />

∂<br />

, (2.11)<br />

∂u<br />

∂ ∂<br />

− xj , 1 ≤ j < k ≤ N (2.12)<br />

∂xj ∂xk<br />

+ 2xkx · ∇x +<br />

∂xk<br />

4 ∂<br />

xky<br />

m + 2 ∂y<br />

N(m + 2) − 2<br />

− xku<br />

m + 2<br />

∂<br />

,<br />

∂u<br />

k = 1, . . . , N (2.13)<br />

<strong>and</strong> together with the trivial symmetries generate the complete symmetry<br />

group (cf. Theorem 2.5 of [15]). In particular, we recall that the proof exploits<br />

well known infinitesimal techniques for classifying the infinitesimal generators<br />

v =<br />

N<br />

ξ i (x, y, u) ∂<br />

i=1<br />

∂xi<br />

+ η(x, y, u) ∂<br />

∂<br />

+ ϕ(x, y, u)<br />

∂y ∂u<br />

(2.14)<br />

of symmetries, where v is thought of as a vector field which acts an open<br />

subset M of the 0-jet space, R N+1 × U (0) R N+1 × R (the space of values<br />

for independent <strong>and</strong> dependent variables) together with the action of their<br />

prolongations onto higher order jet spaces (which includes the values of higher<br />

order derivatives of u). In the smooth coefficient case for L where K(y) =<br />

y m , which is degenerate elliptic for m even <strong>and</strong> of mixed type for m odd,<br />

the complete theory can be applied globally on all of R N+1 . Tedious but<br />

elementary calculations show that the coefficients of each generator v in (2.14)<br />

satisfies: ξ i = ξ i (x, y) <strong>and</strong> η = η(x, y) are independent of u; ϕ = α(x, y)u +<br />

β(x, y) is linear in u; <strong>and</strong><br />

Lα = Lβ = 0 (2.15)<br />

Lξ i = 2y m αxi , i = 1, . . . , N (2.16)<br />

Lη = 2αy<br />

(2.17)<br />

2y m ξ i xi − 2ym ηy − my m−1 η = 0, i = 1, . . . , N (2.18)<br />

<br />

m<br />

y ξ i xj + ξj <br />

xi<br />

= 0, 1 ≤ i < j ≤ N (2.19)<br />

y m ηxi + ξ i y = 0, i = 1, . . . , N. (2.20)<br />

For m ∈ R + \ N, the lack of regularity in the coefficients creates no essential<br />

difficulty as each group action preserves the half-spaces R N+1<br />

± . See the<br />

Appendix of [15] for details.


Conformal <strong>groups</strong> for operators of mixed <strong>and</strong> degenerate types 5<br />

3 A singular Riemannian-Lorentzian structure<br />

As we pointed out in [15], there is a natural geometric structure <strong>associated</strong><br />

to the operator (1.1); namely the linear operator in (1.1) is <strong>associated</strong> to<br />

a singular geometry on R N+1 via the metric g which in global coordinates<br />

(x, y) on R N+1 is given by the matrix<br />

[gij] =<br />

<br />

−1 K(y) IN×N 0<br />

. (3.1)<br />

0 1<br />

For K smooth <strong>and</strong> satisfying (1.2)−(1.3), g is a smoothly varying symmetric<br />

<strong>and</strong> non degenerate bilinear form on the tangent bundles T R N+1<br />

± <strong>and</strong> hence<br />

is a smooth Riemannian/Lorentzian metric on the upper/lower half spaces<br />

which becomes singular on the hyperplane y = 0. Notice that the inverse<br />

tensor [gij ] to (3.1) on the cotangent bundles T ∗R N+1<br />

± extends continuously<br />

across the metric singularity <strong>and</strong> is smooth if m is an integer. If one computes<br />

the <strong>associated</strong> Laplace-Beltrami operator/D’Alembertian in their respective<br />

, one arrives at the geometrically natural operator<br />

half spaces R N+1<br />

±<br />

Lg = divg ◦ grad g = K(y)∆x + ∂ 2 y − NK′ (y)<br />

2K(y) ∂y<br />

(3.2)<br />

as will be shown below. Hence the linear operator in (1.1) is the principal<br />

<strong>and</strong> everywhere smooth part of this “geometric operator”. Mixed signature<br />

<strong>metrics</strong> have received some attention in the context of quantum cosmology,<br />

beginning with the Hawking-Hartle no boundary hypothesis which if true<br />

implies that space-time evolved from a Riemannian past into a Lorentzian<br />

present, as well as models for tunnelling effects in quantum gravity (cf. Chapters<br />

3 <strong>and</strong> 5 of [11]). More recently, <strong>and</strong> in the spirit of the present observations,<br />

the description of harmonic forms in certain projective surfaces has<br />

been studied via a Hodge system with respect to a singular metric [19].<br />

As stated in the introduction, we would like to establish precisely the connection<br />

between the symmetry <strong>groups</strong> of the operator (1.1) or (3.2) <strong>and</strong> the<br />

group of <strong>conformal</strong> transformations with respect to this <strong>underlying</strong> singular<br />

geometric structure (3.1). The analysis proceeds along classical lines away<br />

from the metric singularity on the hypersurface y = 0 <strong>and</strong> the simple form of<br />

the metric allows for a continuation across the singularity in an explicit way.<br />

To obtain the richest possible classes of symmetries <strong>and</strong> <strong>conformal</strong> transformations<br />

we will restrict our attention to type change functions of power<br />

type where three possible situations can be treated in the same way; namely<br />

operators of degenerate elliptic, degenerate hyperbolic, or mixed type:<br />

(DE) K(y) = |y| m , m > 0<br />

(DH) K(y) = −|y| m , m > 0<br />

(MT) K(y) = y|y| m−1 , m > 0<br />

To simply notation, we will treat explicitly only the case<br />

K(y) = y m with m ∈ N (3.3)


6 K. R. Payne<br />

which is of the type (DE)/(MT) for m even/odd. Obvious modifications can<br />

be made for the other cases.<br />

In the rest of this section, we recall briefly the main geometric notions<br />

needed to calculate a “<strong>conformal</strong> group” <strong>associated</strong> to the singular metric<br />

(3.1); these <strong>conformal</strong> structures will be analyzed in section 4. In all cases,<br />

for y > 0 <strong>and</strong> y < 0 the metric g is a smoothly varying symmetric (diagonal)<br />

non degenerate bilinear form of constant index (0 in the elliptic regions (Riemannian<br />

metric) <strong>and</strong> N in the hyperbolic regions (Lorentzian metric)) <strong>and</strong><br />

hence gives a semi-Riemannian metric on each half space (cf. the monograph<br />

of O’Neill [18]). We recall that the index equals the maximal dimension of<br />

a subspace on which the restricted form is negative definite. One has then<br />

the st<strong>and</strong>ard notions of a Levi-Civita connection, covariant derivatives of<br />

tensor fields, Christoffel symbols, divergence <strong>and</strong> gradient (see [18] as well<br />

as Chapter 2 of the monograph of Taylor [23] whose notation we will often<br />

follow). To simplify notation we will denote the coordinates on RN+1 by<br />

(x1, . . . , xN, xN+1) with xN+1 in the place of y <strong>and</strong> denote the corresponding<br />

basis elements in T RN+1 by Dk = ∂/∂xk, while ∂kf represents the partial<br />

derivative of a scalar valued function f.<br />

Denoting by X (R N+1<br />

± ) the space of vector fields X with smooth coefficients<br />

X = N+1 k=1 XkDk, one has the covariant derivative of X with respect<br />

to Dj given by the vector field<br />

∇Dj X =<br />

N+1 <br />

k=1<br />

<br />

X k ;jDk where X k ;j := ∂jX k N+1<br />

+<br />

l=1<br />

Γ k<br />

ljX l<br />

(3.4)<br />

<strong>and</strong> Γ l jk are the Christoffel symbols or connection coefficients with respect to<br />

the Levi-Civita connection ∇ : X (R N+1<br />

± ) × X (R N+1<br />

± ) → X (R N+1<br />

± ). For our<br />

diagonal <strong>metrics</strong> one has<br />

Γ l jk = 1<br />

2 gll<br />

<br />

∂gjl<br />

+<br />

∂xk<br />

∂gkl<br />

−<br />

∂xj<br />

∂gjk<br />

<br />

, j, k, l = 1, . . . N + 1 (3.5)<br />

∂xl<br />

which for (3.1) with (3.3) yields<br />

Γ l ⎧<br />

⎪⎨<br />

−m/(2xN+1) k = N + 1, l = j = N + 1<br />

−m/(2xN+1) j = N + 1, l = k = N + 1<br />

jk =<br />

⎪⎩<br />

m/(2x m+1<br />

N+1 ) l = N + 1, j = k = N + 1<br />

0 otherwise<br />

(3.6)<br />

Making use of (3.5) <strong>and</strong> (3.6) one defines the divergence of a vector field by<br />

divgX :=<br />

=<br />

N+1 <br />

j=1<br />

N+1 <br />

j=1<br />

X j<br />

;j =<br />

N+1<br />

<br />

∂jX j N<br />

+<br />

j=1<br />

j=1<br />

∂jX j − mN<br />

X<br />

2xN+1<br />

N+1<br />

Γ j<br />

N+1,j XN+1<br />

<strong>and</strong> the gradient of a scalar valued function u as the vector field<br />

(3.7)


Conformal <strong>groups</strong> for operators of mixed <strong>and</strong> degenerate types 7<br />

grad gu =<br />

N+1 <br />

j=1<br />

X j Dj where X j :=<br />

N+1 <br />

k=1<br />

g jk ∂ku (3.8)<br />

<strong>and</strong> hence the coefficients of grad gu are identifiable with the vector valued<br />

function<br />

∇gu = (x m N+1∂1u, . . . , x m N+1∂Nu, ∂N+1u) (3.9)<br />

from which the claim (3.2) follows.<br />

For these simple <strong>metrics</strong>, one calculates easily their <strong>associated</strong> curvatures<br />

which become infinite as one approaches the singular hypersurface. In particular<br />

the scalar curvature S blows up like CN,mx −2<br />

N+1 . In fact, one easily<br />

computes (again using the fact that g is diagonal)<br />

S :=<br />

N+1 <br />

j=1<br />

g jj (Ric)jj =<br />

−m(m + 2)N<br />

2x 2 N+1<br />

where the Ricci curvature Ric is given in the usual way in terms of Γ l jk<br />

(Ric)jk =<br />

N+1 <br />

l=1<br />

<br />

∂lΓ l kj − ∂kΓ l<br />

lj +<br />

N+1 <br />

i=1<br />

(3.10)<br />

l<br />

ΓliΓ i kl − Γ l kiΓ i <br />

lj<br />

<br />

. (3.11)<br />

Additional comments concerning the notion of singular geometric structures<br />

<strong>and</strong> relations to other singular metric structures will be given in section 5.<br />

4 Associated <strong>conformal</strong> transformations<br />

In terms of the geometric data presented in section 3, one can use the classical<br />

infinitesimal techniques in order to describe the isometries <strong>and</strong> <strong>conformal</strong><br />

transformations on the half-spaces R N+1<br />

± . We recall that a vector field X =<br />

N+1 j=1 Xj Dj generates an isometry (its time-t flow F t X preserves g; that is<br />

[F t X ]∗g = g) if <strong>and</strong> only if it is a Killing field; that is, its coefficients satisfy<br />

the system<br />

(2DefX)jk :=<br />

N+1 <br />

l=1<br />

<br />

gklX l ;j + gjlX l <br />

;k = 0, j, k = 1, . . . N + 1 (4.1)<br />

where DefX is the deformation tensor of X, which for a diagonal g reduces<br />

to<br />

(2DefX)jk := gkkX k ;j + gjjX j<br />

;k = 0, j, k = 1, . . . N + 1. (4.2)<br />

Moreover, X generates a <strong>conformal</strong> transformation ([F t X ]∗ g = α(t, x)g for<br />

some scalar valued function α) if <strong>and</strong> only if X is a <strong>conformal</strong> Killing field;<br />

that is


8 K. R. Payne<br />

DefX = 1<br />

N + 1 (divgX) g (4.3)<br />

In preparation for the results which follow we introduce the following definition<br />

suited to our limited aims.<br />

Definition 4.1. Let g be a semi-Riemannian metric on a half-space R N+1<br />

± .<br />

The <strong>conformal</strong> algebra <strong>associated</strong> to g is the Lie algebra Kg(R N+1<br />

± ) of all<br />

<strong>conformal</strong> Killing fields X on R N+1<br />

± ; that is, fields with smooth coefficients<br />

that satisfy (4.3).<br />

The fact that Kg(R N+1<br />

± ) forms a Lie algebra for any g with constant index on<br />

a half space follows the st<strong>and</strong>ard theory [18], while the possibility of gluing<br />

together two such algebras along the singular hypersurface will depend on<br />

the particular form of g. We begin with a lemma for the high dimensional<br />

cases (N ≥ 2).<br />

Lemma 4.2. Let K be a pure power type change function of the form (3.3)<br />

on a half-space R N+1<br />

± . If N ≥ 2, then the <strong>conformal</strong> algebra Kg(R N+1<br />

± ) is<br />

(N + 1)(N + 2)/2 dimensional <strong>and</strong> spanned by the vector fields<br />

Ik = −dDk + 2xk<br />

where<br />

S = (m + 2)<br />

Tk = Dk, k = 1, . . . , N (4.4)<br />

N<br />

(xjDj) + 2xN+1DN+1<br />

j=1<br />

(4.5)<br />

Rjk = xkDj − xjDk, 1 ≤ j < k ≤ N (4.6)<br />

N<br />

(xjDj) + 4<br />

m + 2 xkxN+1DN+1, k = 1, . . . , N (4.7)<br />

j=1<br />

d = d(x1, . . . xN, xN+1) =<br />

N<br />

j=1<br />

x 2 j +<br />

4<br />

xm+2<br />

(m + 2) 2 N+1<br />

(4.8)<br />

Proof: Elementary calculation shows that the defining system (4.3) for the<br />

coefficients {X j } N+1<br />

j=1 of an arbitrary <strong>conformal</strong> Killing field is, for xN+1 = 0,<br />

equivalent to the system<br />

2xN+1∂jX j = mX N+1 + 2<br />

N + 1 xN+1divgX, j = 1, . . . N (4.9)<br />

2xN+1∂N+1X N+1 = 2<br />

N + 1 xN+1divgX (4.10)<br />

x m N+1∂jX N+1 + ∂N+1X j = 0, j = 1, . . . N (4.11)<br />

∂jX k + ∂kX j = 0, 1 ≤ j < k ≤ N (4.12)<br />

Combining (4.9) <strong>and</strong> (4.10) <strong>and</strong> setting y = xN+1, X N+1 = η, <strong>and</strong> X i = ξ i<br />

for 1 ≤ i ≤ N shows that each solution to (4.9) − (4.12) is a solution of


Conformal <strong>groups</strong> for operators of mixed <strong>and</strong> degenerate types 9<br />

(2.18) − (2.20). For N ≥ 2, this latter system was shown to have a finite<br />

dimensional solution space with the basis corresponding to (4.4) − (4.7).<br />

Hence, any <strong>conformal</strong> Killing field gives rise to a one parameter symmetry<br />

group for L.<br />

To finish the proof, consider any regular solution of the full system (2.15)−<br />

(2.20) describing the symmetry group, with a fixed (ξ 1 , . . . ξ N , η) equal to<br />

(X 1 , . . . , X N , X N+1 ) satisfying (2.18) − (2.20). For this solution, the additional<br />

constraints (2.15) − (2.17) force the stronger form (4.9) − (4.10) of<br />

(2.18) to hold as well. Hence each generator of a one dimensional symmetry<br />

group projects onto a <strong>conformal</strong> Killing field.<br />

The explicit representation of the <strong>conformal</strong> algebras away from the metric<br />

singularity allows for a trivial extension across the singularity by smoothness.<br />

We formalize this observation in the following Theorem whose proof is<br />

a direct consequence of Lemma 4.2.<br />

Theorem 4.3. Let K be a pure power type change function of the form<br />

(3.3) on RN+1 with N ≥ 2. Then the generators (4.4) − (4.7) of the <strong>conformal</strong><br />

algebras Kg(R N+1<br />

± ) are all tangential to the singular hypersurface<br />

Σ = {xN+1 = 0} in the sense that limxN+1→0 XN+1 = 0 <strong>and</strong> have a smooth<br />

extension across Σ. The resulting Lie algebra Kg(RN+1 ) agrees with the natural<br />

projection onto T RN+1 of the Lie algebra spanned by (2.10) − (2.13)<br />

which generates the non-trivial part of the symmetry group for the operator<br />

(1.1).<br />

In this way, one could define the <strong>conformal</strong> group <strong>associated</strong> to the singular<br />

metric g as the group of local diffeomorphisms generated by Kg(R N+1 )<br />

which by Theorem 4.3 is seen to be isomorphic to the non-trivial part of<br />

the symmetry group for the operator (1.1). We remark that the same result<br />

holds for all of the cases (DE), (DH), <strong>and</strong> (MT) with any m > 0 where one<br />

only needs to replace x m+2<br />

N+1 in (4.8) with ±|xN+1| m+2 or xN+1|xN+1| m+1 respectively<br />

<strong>and</strong> to accept limited smoothness in the coefficients of the Killing<br />

fields across Σ.<br />

In the planar case (N = 1), the situation is more complicated as one<br />

might expect if one thinks of the limiting case m = 0. One expects a <strong>conformal</strong><br />

group of infinite dimension which turns out to be the case. However, the<br />

symmetry group for the operator (2.1) has been shown to be finite dimensional;<br />

hence there is only a one way correspondence between the <strong>conformal</strong><br />

transformations <strong>and</strong> the symmetries for (2.1). One recovers the correspondence<br />

for the geometric operator (3.2). We begin with a description of the<br />

<strong>conformal</strong> algebras away from the metric singularity.<br />

Lemma 4.4. Let K be a pure power type change function of the form (3.3)<br />

on a half-space R 1+1<br />

± . Then the <strong>conformal</strong> algebra Kg(R 1+1<br />

± ) is infinite dimensional<br />

<strong>and</strong> each <strong>conformal</strong> Killing field is tangential to singular line<br />

Σ = {x2 = 0} <strong>and</strong> extends smoothly across Σ.<br />

Proof: The defining equations (4.3) for the coefficients of the <strong>conformal</strong><br />

Killing fields reduce to the pair of equations


10 K. R. Payne<br />

x2∂1X 1 − x2∂2X 2 − m<br />

2 X2 = 0 (4.13)<br />

x m 2 ∂1X 2 + ∂2X 1 = 0 (4.14)<br />

where we have used also divgX = ∂1X 1 + ∂2X 2 − mX 2 /(2x2). Hence the<br />

system is equivalent to the corresponding (ξ, η) = (X 1 , X 2 ) part of the system<br />

(2.18) <strong>and</strong> (2.20) describing the symmetry group. These two equations in two<br />

unknowns have an infinite number of independent solutions including those<br />

corresponding to the translation, dilation <strong>and</strong> inversion; namely<br />

(X 1 , X 2 ) (1) = (1, 0), (4.15)<br />

(X 1 , X 2 ) (2) = ((m + 2)x1, 2x2), (4.16)<br />

(X 1 , X 2 ) (3) = ((m + 2) 2 x 2 1 − 4x m+2<br />

2 , 4(m + 2)x1x2) (4.17)<br />

as well as a remarkable sequence of independent solutions defined inductively<br />

by (X 1 , X 2 ) (k) = (ξ (k), η (k)) with<br />

ξ (k+1) = ξ 2 (k) − xm 2 η 2 (k) , <strong>and</strong> η (k+1) = 2ξ (k)η (k), k ≥ 2 (4.18)<br />

as a simple calculation shows. Hence the <strong>conformal</strong> algebras have infinite<br />

dimension.<br />

All of the solutions above extend smoothly across the singular line x2 = 0<br />

where the component X 2 tends to zero as x2 → 0; hence in this sense each X<br />

is tangential to the singular line. Moreover, the equation (4.13) shows that<br />

any C 1 continuation of X 2 must vanish as well.<br />

As in the higher dimensional cases, we can now define the <strong>conformal</strong> algebra<br />

Kg(R1+1 ) as the Lie algebra of smooth extensions across Σ of Kg(R 1+1<br />

± )<br />

<strong>and</strong> the <strong>conformal</strong> group <strong>associated</strong> to the singular metric g as the group of<br />

local diffeomorphisms generated by Kg(R1+1 ) which is infinite dimensional.<br />

On the other h<strong>and</strong>, the non-trivial part of the symmetry group <strong>associated</strong> to<br />

the operator (1.1) is only three dimensional <strong>and</strong> corresponds to the generators<br />

(4.15) − (4.17). The point being that the remainder of the system describing<br />

the symmetry group (2.15) − (2.17) forces the pair (ξ, η) = (X1 , X2 ) to be<br />

a linear combination of the generators (4.15) − (4.17) as has been shown in<br />

[15]. On the other h<strong>and</strong>, if one considers the operator (3.2) one recovers the<br />

complete <strong>conformal</strong> group as the non-trivial part of its symmetry group.<br />

Theorem 4.5. Let K be a pure power type change function on R 1+1 . Then<br />

the <strong>conformal</strong> algebra Kg(R 1+1 ) agrees with the natural projection onto T R 1+1<br />

of the Lie algebra A(R 1+1 ) of generators of a one parameter symmetry group<br />

for the operator<br />

Lg = K(y)∂ 2 x + ∂ 2 y − m<br />

2y ∂y. (4.19)<br />

Proof: As a first step, we need to find the system analogous to (2.15) −<br />

(2.20) which characterizes the generators v = ξ(x, y, u)∂/∂x+η(x, y, u)∂/∂y+<br />

ϕ(x, y, u)∂/∂u of one parameter symmetry <strong>groups</strong> for the operator (4.19). In


Conformal <strong>groups</strong> for operators of mixed <strong>and</strong> degenerate types 11<br />

this case, since the operator has a singular coefficient along Σ, we will use<br />

the classification machinery away from Σ <strong>and</strong> then require smoothness in the<br />

coefficients of v. St<strong>and</strong>ard calculations for y = 0 show that: ξ = ξ(x, y), η =<br />

η(x, y) are independent of u; ϕ = α(x, y)u + β(x, y) is linear in u; <strong>and</strong><br />

Lgα = Lgβ = 0 (4.20)<br />

Lgξ = 2y m αx<br />

Lgη + m<br />

y ηy − m<br />

η = 2αy<br />

(4.22)<br />

2y2 (4.21)<br />

2y m ξx − 2y m ηy − my m−1 η = 0 (4.23)<br />

y m ηx + ξy = 0 (4.24)<br />

Relabelling by (x, y; ξ, η) = (x1, x2; X 1 , X 2 ) we see that (4.23)−(4.24) is precisely<br />

the system (4.13) − (4.14). Hence each generator of a one dimensional<br />

symmetry group must project onto a <strong>conformal</strong> Killing field. Conversely,<br />

given a solution to (4.13) − (4.14), if one selects α = β = 0, one trivially<br />

obtains (4.20) <strong>and</strong> verifies that (4.21)−(4.22) hold as well with αx = αy = 0.<br />

Hence, each infinitesimal <strong>conformal</strong> transformation gives rise to an infinitesimal<br />

symmetry, which completes the proof.<br />

We close this section by noting that for the higher dimensional cases<br />

N ≥ 2, one has the analogous result for the operator Lg <strong>and</strong> the same<br />

kind of trivial splitting in the Lie algebra of generators A(R N+1 ). Also,<br />

as one expects, the “isometry group” defined infinitesimally as the smooth<br />

extensions of the Killing fields X (solutions of (4.1)) turn out to be generated<br />

by the translations <strong>and</strong> rotations (for N ≥ 2) in the x-variables.<br />

5 Concluding remarks<br />

There are various uses of the term singular Riemannian/Lorenztian metric<br />

in the literature. We have used the term in the sense of Hermann [12] in<br />

which one has the presence of a smoothly varying symmetric bilinear but<br />

degenerate form γ on the cotangent bundle T ∗ M of a smooth manifold M.<br />

The particular nature of the singular geometry is described by the form<br />

of the degeneracy in γ which in our situation involves a blow up in the<br />

metric g at the boundary of open sets on which γ is non-degenerate. A<br />

situation dual to the one we consider concerns not a blow up in the metric<br />

tensor, but rather a degeneration in it. For example, a conical singularity<br />

in the sense of Cheeger includes the relatively simple case of a metric of<br />

the form g = dr 2 + r 2 ˜g on the cone C(N ) = R + × N where (N , ˜g) is an<br />

n-dimensional Riemannian manifold. The Laplace-Beltrami operator in this<br />

case is the (positive) operator<br />

∆ = − ∂2 n<br />

−<br />

∂r2 r<br />

∂<br />

∂r<br />

1<br />

+ ∆˜g<br />

(5.1)<br />

r2


12 K. R. Payne<br />

where ∆˜g is Laplace-Beltrami operator on N (cf. [4] <strong>and</strong> the references<br />

therein). Here the principal part is also singular, as opposed to the smoothness<br />

that (3.2) has.<br />

In a singular geometric situation described by degeneration in the bilinear<br />

form γ on T ∗ M, one can try to study the geodesics on M (constant<br />

velocity locally extremal curves with respect to some length functional) via<br />

a suitable Hamiltonian system (the bicharacteristics of the quadratic form<br />

defined by γ) as opposed to the Lagrangian approach on the tangent bundle<br />

T M, which becomes singular at the degeneracies of γ. In regular (semi-<br />

Riemannian) situations, the two approaches coincide in the sense that up to<br />

reparametrization, the geodesics are projections onto M of the bicharacteristics.<br />

More importantly, under favorable circumstances, the same continues<br />

to hold with respect to a given <strong>underlying</strong> (singular) geometry. This is the<br />

case for classes of sub-Riemannian manifolds (see section 3.4 of [2] <strong>and</strong> the<br />

references therein). Such classes contain some of our degenerate elliptic examples<br />

as will be clarified below. On the other h<strong>and</strong>, for all of the <strong>metrics</strong> we<br />

consider, one can calculate explicitly <strong>and</strong> globally the bicharacteristics which<br />

then describe the geodesics away from the metric singularity. In the singular<br />

Lorentzian <strong>and</strong> Riemannian-Lorentzian cases, one can study global causality<br />

properties such as disprisoning for null geodesics <strong>and</strong> null pseudo-convexity<br />

(cf. [1] ) as we have done for singular mixed signature <strong>metrics</strong> of the form<br />

ds 2 = dy 2 + N<br />

j,k=1 (yA)−1<br />

jk dxjdxk, with A(x, y) smooth <strong>and</strong> positive definite<br />

(cf. [21] <strong>and</strong> [20]).<br />

Finally, we note the connection of our examples to the notion of a sub-<br />

Riemannian geometry on an n dimensional manifold M in which a metric is<br />

<strong>associated</strong> to a family of m ≤ n smooth vector fields. For example, for an<br />

operator L (1.1) of Grushin type with K(y) = y 2k , k ∈ N one can write L<br />

as a sum of squares of vector fields L = N+1<br />

j=1 (Xj ) 2 with X j = y k Dj for<br />

1 ≤ j ≤ N <strong>and</strong> X N+1 = DN+1. These vector fields together with their Lie<br />

brackets generate T R N+1 <strong>and</strong> hence there is a natural Carnot-Carathéodory<br />

metric <strong>associated</strong> to this family of vector fields by using piecewise integral<br />

curves of ±X j (see [2], [10] for example) which defines the sub-Riemannian<br />

structure. Notice that the Grushin system of vector fields is not of constant<br />

rank, as is often assumed, <strong>and</strong> that the geometry is truly sub-Riemannian<br />

only along the degenerate hypersurface Σ (it is Riemannian away from Σ).<br />

This relaxation on the rank is essential for certain applications to geometric<br />

control theory <strong>and</strong> hypoelliptic equations as is pointed out by Bellaïche<br />

[2]. Carnot-Carathéodory <strong>metrics</strong> can often be given a meaning even with<br />

low order regularity in the coefficients which has implications for degenerate<br />

elliptic equations such as Harnack inequalities <strong>and</strong> Hölder regularity results<br />

[6], [7]. Moreover, with respect to the natural Sobolev spaces defined by the<br />

vector fields, one recovers critical exponent phenomena for degenerate <strong>and</strong><br />

mixed type equations like those well known in the elliptic setting (cf. [8],<br />

[14]). As pointed out by Strichartz [22], it also makes sense to speak about a<br />

sub-Lorentzian geometry <strong>associated</strong> to any non-degenerate quadratic form on<br />

a sub-bundle (constant rank) S of the tangent bundle with the bracket generating<br />

property. Such geometries have been studied recently [9] under the<br />

constant rank assumption. We do not know if a non-constant rank version of


Conformal <strong>groups</strong> for operators of mixed <strong>and</strong> degenerate types 13<br />

a sub-Lorentzian structure can be <strong>associated</strong> to our examples for which the<br />

bicharacteristics project onto geodesics.<br />

References<br />

1. Beem, J.K., Parker, P.E.: Klein-Gordon solvability <strong>and</strong> the geometry of geodesics.<br />

Pacific J. Math. 107, 1–14 (1983)<br />

2. Bellaïche, A.: Tangent space in sub-Riemannian geometry. In: Sub-<br />

Riemannian Geometry. Bellaïche, A., Risler, J-J. (eds.). Progress in Mathematics<br />

144, 1–78. Birkhäuser, Basel 1996.<br />

3. Bers, L.: Mathematical Aspects of Subsonic <strong>and</strong> Transonic Gas Dynamics.<br />

Surveys in Applied Mathematics, Vol. 3, John Wiley & Sons, New York 1958<br />

4. Cheeger, J.: Spectral geometry of singular Riemannian spaces. J. Differential<br />

Geom. 18, 575–657 (1983)<br />

5. Eisenhart, L.P.: Riemannian Geometry. Princeton University Press, Princeton<br />

1926.<br />

6. Franchi, B., Lanconelli, E.: Hölder regularity theorem for a class of linear<br />

nonuniformly elliptic operators with measurable coefficients. Ann. Scuola<br />

Norm. Sup. Pisa Cl. Sci. (4) 10, 523–541 (1983)<br />

7. Franchi, B., Lanconelli, E.: An embedding theorem for Sobolev spaces related<br />

to non-smooth vector fields <strong>and</strong> Harnack inequality. Comm. in Partial<br />

Differential Equations 9, 1237–1264 (1984)<br />

8. Garofalo, N., Lanconelli, E.: Existence <strong>and</strong> nonexistence results for semilinear<br />

equations on the Heisenberg group. Indiana Univ. Math. J. 41, 71–98 (1992)<br />

9. Grochowski, M.: Geodesics in the sub-Lorentzian geometry. Bull. Polish<br />

Acad. Sci. Math. 50, 161–178 (2002)<br />

10. Gromov, M.: Carnot-Carathéodory spaces seen from within. In: Sub-<br />

Riemannian Geometry. Bellaïche, A., Risler, J-J. (eds.). Progress in Mathematics<br />

144, 79–323. Birkhäuser, Basel 1996.<br />

11. Hawking, S., Penrose, R.: The Nature of Space <strong>and</strong> Time. The Issac Newton<br />

Institute Series of Lectures, Princeton University Press, Princeton 1996.<br />

12. Hermann, R.: Geodescis of singular Riemannian <strong>metrics</strong>. Bull. Amer. Math.<br />

Soc. 79, 780–783 (1973)<br />

13. Lin, C.S.: The local isometric embedding in R 3 of two-dimensional Riemannian<br />

manifolds with Gaussian curvature changing sign cleanly. Comm.<br />

Pure Appl. Math. 39, 867–887 (1986)<br />

14. Lupo, D., Payne, K.R.: Critical exponents for semilinear equations of mixed<br />

elliptic-hyperbolic <strong>and</strong> degenerate types. Comm. Pure Appl. Math. 56, 403–<br />

424 (2003)<br />

15. Lupo, D., Payne, K.R.: Conservation laws for equations of mixed elliptichyperbolic<br />

<strong>and</strong> degenerate types. Duke Math. J., to appear.<br />

16. Morawetz, C.S.: Mixed equations <strong>and</strong> transonic flow. J. Hyperbolic Differ.<br />

Equ. 1, 1–26 (2004)<br />

17. Olver, P.J.: Applications of Lie Groups to Differential Equations, 2nd Ed..<br />

Graduate Texts in Mathematics, Vol. 107. Springer-Verlag, New York, 1993.<br />

18. O’Neill, B.: Semi-Riemannian Geometry, With Applications to Relativity.<br />

Pure <strong>and</strong> Applied Mathematics, Vol. 103. Academic Press, New York, 1983.<br />

19. Otway, T.H.: Hodge equations with change of type. Ann. Mat. Pura Appl.<br />

(4), 181, 437–452 (2002)<br />

20. Payne, K.R.: Boundary geometry <strong>and</strong> location of singularities for solutions<br />

to the Dirichlet problem for Tricomi type equations. Houston J. Math. 23,<br />

709–731 (1997)<br />

21. Payne, K.R.: Propogation of singularities phenomena for equations of Tricomi<br />

type. Applicable Analysis 68, 195–206 (1998)<br />

22. Strichartz, R.S.: Sub-Riemannian geometry. J. Differential Geometry 24,<br />

221–263 (1986)<br />

23. Taylor, M.E.: Partial Differential Equations I, Basic Theory. Applied Mathematical<br />

Sciences, Vol. 115. Springer-Verlag, New York, 1996.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!