sigma07-114. 240KB Jun 04 2011 12:10:10 AM

Symmetry, Integrability and Geometry: Methods and Applications

SIGMA 3 (2007), 114, 10 pages

Some Sharp L2 Inequalities for Dirac Type Operators⋆
Alexander BALINSKY

†

and John RYAN

‡

†

Cardif f School of Mathematics, Cardif f University,
Senghennydd Road, Cardif f, CF 24 4AG, UK
E-mail: BalinskyA@cardiff.ac.uk
URL: http://www.cf.ac.uk/maths/people/balinsky.html

‡

Department of Mathematics, University of Arkansas, Fayetteville, AR 72701, USA
E-mail: jryan@uark.edu
URL: http://comp.uark.edu/∼jryan/

Received August 31, 2007, in final form November 14, 2007; Published online November 25, 2007
Original article is available at http://www.emis.de/journals/SIGMA/2007/114/
Abstract. We use the spectra of Dirac type operators on the sphere S n to produce sharp L2
inequalities on the sphere. These operators include the Dirac operator on S n , the conformal
Laplacian and Paenitz operator. We use the Cayley transform, or stereographic projection,
to obtain similar inequalities for powers of the Dirac operator and their inverses in Rn .
Key words: Dirac operator; Clifford algebra; conformal Laplacian; Paenitz operator
2000 Mathematics Subject Classification: 15A66; 26D10; 34L40

This paper is dedicated to the memory of Tom Branson

1

Introduction

Sobolev and Hardy type inequalities play an important role in many areas of mathematics and
mathematical physics. They have become standard tools in existence and regularity theories for
solutions to partial differential equations, in calculus of variations, in geometric measure theory
and in stability of matter. In analysis a number of inequalities like the Hardy–Littlewood–
Sobolev inequality in Rn are obtained by first obtaining these inequalities on the compact
manifold S n and then using stereographic projections to Rn to obtain the analogous sharp
inequality in that setting. See for instance [10]. This technique is also used in mathematical
physics to obtain zero modes of Dirac equations in R3 (see [9]).
In fact the stereographic projection corresponds to the Cayley transformation from S n minus
the north pole to Euclidean space. Here we shall use this Cayley transformation to obtain some
sharp L2 inequalities on the sphere for a family of Dirac type operators. The main trick here
is to employ a lowest eigenvalue for these operators and then use intertwining operators for the
Dirac type operators to obtain analogous sharp inequalities in Rn .
Our eventual hope is to extend the results presented here to obtain suitable Lp inequalities
for the Dirac type operators appearing here, particularly the Dirac operator on Rn .

2

Preliminaries

We shall consider Rn as embedded in the real, 2n dimensional Clifford algebra Cln so that for
each x ∈ Rn we have x2 = −kxk2 . Consequently if e1 , . . . , en is an orthonormal basis for Rn
⋆

This paper is a contribution to the Proceedings of the 2007 Midwest Geometry Conference in honor of
Thomas P. Branson. The full collection is available at http://www.emis.de/journals/SIGMA/MGC2007.html

2

A. Balinsky and J. Ryan

then
ei ej + ej ei = −2δij
and
1, e1 , . . . , en , e1 e2 , . . . , en−1 en , . . . , ej1 , . . . , ejr , . . . , e1 , . . . , en
is an orthonormal basis for Cln , with 1 ≤ r ≤ n and j1 < · · · < jr .
−x
Note that for each x ∈ Rn \{0} we have that x is invertible, with multiplicative inverse kxk
2.
−1

Here, up to a sign, x is the Kelvin inverse of x. It follows that {A ∈ Cln : A = x1 · · · xm with
m ∈ N and x1 , . . . , xm ∈ Rn \{0}} is a subgroup of Cln . We shall denote this group by GP in(n).
We shall need the following anti-automorphisms on Cln :
∼: Cln → Cln : ej1 · · · ejr → ejr · · · ej1
and
− : Cln → Cln : ej1 · · · ejr → (−1)r ejr · · · ej1 .
For A ∈ Cln we denote ∼ (A) by A˜ and we denote −(A) by A. Note that for A = a0 + · · · +
a1...n e1 · · · en the scalar part of AA is a20 + · · · + a21...n := kAk2 .
Lemma 1. If A ∈ GP in(n) and B ∈ Cln then kABk = kAkkBk.
Proof . ABAB = B AAB = BkAk2 B = kAk2 BB. Therefore Sc(ABAB) = kAk2 Sc(BB) =
kAk2 kBk2 , where Sc(C) is the scalar part of C for any C ∈ Cln . The result follows.

In [1] it is shown that if y = M (x) is a M¨
obius transformation then M (x) = (ax+b)(cx+d)−1
where a, b, c and d ∈ Cln and satisfy the conditions
(i) a, b, c, d ∈ GP in(n).
˜ ˜ba ∈ Rn
(ii) a˜
c, c˜d, db,
(iii) ad˜ − c˜

c ∈ R\{0}.
In particular if we regard Rn as embedded in Rn+1 in the usual way, then y = (en+1 x +
1)(x + en+1 )−1 is the Cayley transformation from Rn to the unit sphere S n in Rn+1 . This map
corresponds to the stereographicPprojection of Rn onto S n \{en+1 }.
The Dirac operator in Rn is nj=1 ej ∂x∂ j . Note that D2 = −△n , where △n is the Laplacian
in Rn , and D4 is the bi-Laplacian △2n .

3

Eigenvectors of the Dirac–Beltrami operator on S n

P
∂
n+1 . For each point in x ∈ Rn+1 \{0}
We start with the Dirac operator Dn+1 = n+1
j=1 ej ∂xj in R
this operator can be rewritten as x−1 xDn+1 . Now xDn+1 = x∧Dn+1 −x·D
n+1 . Now x∧Dn+1 =
Pn+1
P

∂
∂
∂
∂
e
e
(x
−
x
)
and
x
·
D
is
the
Euler
operator
n+1
k ∂xj

1≤j 0, and
(k)
DS



k
:= DS (DS − w)(DS − w) · · · DS − w
2



for k even and k > 0.
When k = 1 we obtain DS , when k = 2 we obtain △S and when k = 4 the operator
(4)
DS = △S (DS − w)(DS − 2w). Moreover
(DS − w)(DS − 2w) = DS2 − wDS − 2DS w − 2 = DS2 + wDS − 2 = −△S − 2.
(4)

Consequently DS = −△S (△S + 2). When n = 4 this operator becomes −△S (△S + 2) is the

Paenitz operator on S 4 described in [2] and elsewhere. As 2 ∈ σ(DS ) when n = 4 it may be
seen that 0 is in the spectrum of DS − 2w. Consequently when n = 4 zero is in the spectrum of
the Paenitz operator and so this operator is not invertible on L2 (S 4 ). It is easy to see that it is
invertible in all other dimensions.

Some Sharp L2 Inequalities for Dirac Type Operators

5

5

Some Sharp L2 inequalities on S n and Rn

Theorem 1. Suppose that φ : S n → Cln+1 is a C 1 function. Then
kDS φkL2 ≥

n
kφkL2 .
2

Proof . As φ ∈ C 1 (S n ) then φ ∈ L2 (S n ). It follows that
φ=

∞
X
X

pm +

m=0 pm ∈Pm

−∞ X
X

qm ,

m=0 qm ∈Qm

where pm and qm are eigenvectors of Γw . Further the eigenvectors pm can be chosen so that
within Pm they are mutually orthogonal. The same can be done for the eigenvectors qm . Moreover as φ ∈ C 1 then DS φ ∈ C 0 (S n ) and so DS φ ∈ L2 (S n ). Consequently



∞ 
∞ 


X
X
X
X
n
n
m+
DS φ = w 
qm  .
pm +
− −m
2
2
m=0

m=0

pm ∈Pm

qm ∈Qm

But wpm (w) ∈ Qm and wqm (w) ∈ Pm . Consequently
DS φ =

∞ 
X

m+

m=0

∞ 
 X
X
n X
n
pm .
qm +
− −m
2
2
qm ∈Qm

m=0

pm ∈Pm

It follows that
kDS φkL2

∞ 
2 X
X
n
n 2 X
2
kpm k2L2
kqm kL2 +
− −m
=
m+
2
2
m=0
m=0
pm ∈Pm
qm ∈Qm


∞
−∞ X
 n 2 X
X
X

kpm k2L2 +
kqm k2L2 
≥
2
∞ 
X

m=0 pm ∈Pm

m=0 qm ∈Qm

as ± n2 are the smallest eigenvalues of Γw + n2 . That is ± n2 are the eigenvalues closest to zero.
Therefore
 n 2
kφk2L2 .
kDS φk2L2 ≥
2

The result follows.



It should be noted from the proof of Theorem 1 that this inequality is sharp.
In the proof of Theorem 1 it is noted that the operator DS takes Pm to Qm and it takes Qm
to Pm . This is also true of the operator DS + αw for any α ∈ C. As △S = DS (DS + w) it now
follows that the spectrum, σ(△S ), of the conformal Laplacian, △S , is {−( n2 + m)( n2 + m + 1),
−( n2 + m)( n2 + m − 1) : m ∈ N ∪ {0}}. So the smallest eigenvalue is n(2−n)
. We therefore have
4
the following sharp inequality:
Theorem 2. Suppose φ : S n → Cln+1 is a C 2 function. Then
k△S φkL2 ≥

n(n − 2)
kφkL2 .
4
(k)

We now proceed to generalize Theorems 1 and 2 for all operators DS . We begin with:

6

A. Balinsky and J. Ryan

Lemma 2. (i) For k even the smallest eigenvalue of DSk is
n(2 − n) · · · (n + k − 2)(k − n)
2k
and
(ii) for k odd
n(n + 2)(2 − n) · · · (n + k − 1)(k − 1 − n)
.
2k
Proof . Let us first assume that k even. As DS + αw : Pm → Qm and DS + αw : Qm → Pm for
any α ∈ R then
(n + 2m)(2 − n − 2m) · · · (n + k − 2 + 2m)(k − n − 2m)
2k
and
(2m − n)(n + 2 + 2m) · · · (k − 2 − n − 2m)(n + k + 2m)
2k
(k)

are eigenvalues of DS for m = 0, 1, 2, . . .. But for any positive even integer l the term (n + l −
2 + 2m)(l − n − 2m) is closer to zero than (l − 2 − n − 2m)(n + l + 2m). The result follows for k
even. The case k is odd is proved similarly.

(k)

It should be noted that when n is even and k ≥ n then 0 is an eigenvalue of DS . Consequently
(k)
in these cases DS is not an invertible operator on L2 (S n ).
From Lemma 2 we have:
Theorem 3. Suppose φ : S n → Cln+1 is a C k function. Then for k even
(k)

kDS φkL2 ≥

|n(2 − n) · · · (n + k − 2)(k − n)|
kφkL2
2k

and for k odd
|n(n + 2)(2 − n) · · · (n + k − 1)(k − 1 − n)|
kφkL2 .
2k
Again these inequalities are sharp.
When n is odd then of course n2 is not an integer. It follows that in odd dimensions zero is
(k)

kDS φkL2 ≥

(k)

not an eigenvalue for the operator DS . In the cases n even and k ≥ n the smallest eigenvalue is
zero so for those cases the inequality in Theorem 3 is trivial. This includes the Paenitz operator
on S 4 . It follows that none of these operators have fundamental solutions. The fundamental
(k)
solutions for DS for all k when n is odd and for 1 ≤ k < n when n is even are given in [11].
We shall denote them by Ck (w, y).
(k)
is the smallest eigenvalue for DS for k odd then
As n(n+2)(2−n)···(n+k−1)(nk−1−n)
2k
−2k
n(n + 2)(2 − n) · · · (n + k − 1)(k − 1 + n)
(k)−1

is the largest eigenvalue of DS
for n odd or for 1 ≤ k ≤ n − 1 when n is even.
Similarly for k even and n odd and k even with 1 < k < n − 1 for n even
2k
n(2 − n) · · · (n + k − 2)(k − n)
(k)−1

is the largest eigenvalue of DS
.
Similarly to Theorem 3 we now have the following sharp inequality:

Some Sharp L2 Inequalities for Dirac Type Operators

7

Theorem 4. Suppose φ : S n → Cln+1 is a continuous function. Then for n odd and k even
and for n even and k even with 1 < k < n
kCk (w, y) ⋆ φ(w)kL2 ≤

2k
kφkL2
|n(2 − n) · · · (n + k − 2)(k − n)|

and for n odd and k odd and n even and k odd with 1 ≤ k ≤ n − 1
kCk (w, y) ⋆ φ(w)kL2 ≤

2k
kφkL2 .
|n(n + 2)(2 − n)) · · · (n + k − 1)(k − 1 − n)|

Let us now turn to Rn and retranslate Theorems 3 and 4 in this context. In [11] the Cayley
transformation C(x) = (en+1 + 1)(x + en+1 )−1 is used to show that
(k)

DS = J−k (C, x)−1 Dk Jk (C, x),
where Jk (C, x) =

2

n−k
2

(x+en+1 )

n−k+1
(k1+kxk2 ) 2

(1)

when k is odd and Jk (C, x) =

2

n−k
2

(1+kxk2 )

n−k
2

when k is even.

Note that Jk (C, x) ∈ GP in(n + 1). By applying Lemma 1 we now see that on Rn the Cayley
transformation can be applied to Theorem 3 to give:
Theorem 5. Suppose φ : Rn → Cln+1 is a C k function with compact support. Then for each
k ∈ N for n odd and for k = 1, . . . , n − 1 for n even
1
Z
2
k
2
2 k
n
kD φ(x)k (1 + kxk ) dx
Rn

≥ |n(n + 2) · · · (n + k − 1)(k − 1 − n)|

Z

Rn

kφk2 2k
dxn
(1 + kxk2 )k

for k odd, and
1
Z
k
2
2
2 k
n
2
k△n φ(x)k (1 + kxk ) dx

 12

Rn

≥ |n(2 − n) · · · (n + k − 2)(k − n)|

Z

Rn

kφ(x)k2 2k
dxn
(1 + kxk2 )k

 21

for k even.
Proof . For any M¨obius transformation M (x) = (ax + b)(cx + d)−1 the associated Jacobian
2n
n → Cl
k
over a domain in Rn is kcx+dk
n+1 a C function the integral
2n . Consequently for ψ : S
R
(k)
2
S n kDS ψ(w)k dσ(w) by equation (1) becomes
Z
2n dxn
.
kJ−k (C, x)−1 Dk Jk (C, x)ψ(C(x))k2
(1 + kxk2 )n
Rn
By Lemma 1 this expression becomes
Z
1
(1 + kxk2 )k kDk Jk (C, x)ψ(C(x))k2 dxn .
2k Rn
Further
Z

2

kψk dσ(x) =

Sn

=

Z

ZR

kψ(C(x))k2
n

Rn

2n dxn
(1 + kxk2 )n

kJk (C, x)−1 Jk (C, x)ψ(C(x))k

2n dxn
.
(1 + kxk2 )n

8

A. Balinsky and J. Ryan

By Lemma 1 this last expression becomes
Z
k
2
kJk (C, x)ψ(x)k2 (1 + kxk2 )−k dxn .
Rn

On placing Jk (C, x)ψ(C(x)) = φ(x) Theorem 3 now gives the result.



In [11] it is shown that the kernel Ck (w, y) is conformally equivalent to the kernel Gk (x − y)
x−y
Ck
1
k
in Rn , where Gk (x − y) = C
ωn kx−ykn+1−k when k is odd and Gk (x − y) = ωn kx−ykn−k when k is
even. Here Ck is a real constant chosen so that DGk = Gk−1 for k > 1 and with C1 = 1.
(k)−1
(k)
J−k (C, x). Consequently:
As J−k (C, x)−1 DS Jk (C, x) = Dk then D−k = Jk (C, x)−1 DS
Theorem 6. Suppose h : Rn → Cln+1 is a continuous function with compact support. Then for
n odd and k odd and for n even and any odd integer k satisfying 1 ≤ k < n
Z

R

Z



n

Rn

2

n
Gk (x − y)h(x)dx )

1
dy n
(1 + kyk2 )k
Z

1
≤
|n(n + 2) · · · (n + k − 1)(k − 1 − n)|

!1

2

2

2 k

n

kh(x)k (1 + kxk ) dx

Rn

1
2

and for n odd and k even and for n even and k an even integer satisfying 1 < k < n
Z

R

!1
2
2

1
n
dy
Gk (x − y)h(x)dxn

(1 + kyk2 )k
Rn
Z
1
2
1
2
2 k
n
≤
kh(x)k (1 + kxk ) dx
.
|n(n + 2)(2 − n) · · · (n + k − 2)(k − n)|
Rn

Z



n

Dirac type operators in Rn

6

In this section we demonstrate a somewhat alternative approach to obtained Theorems 5 and 6.
We have previously seen that DS pm = (m + n2 )pm for pm ∈ Pm , that DS qm = (− n2 − m)qm
−1
for qm ∈ Qm and J−1
(C, x)DJ1 (C, x) = DS . Consequently
DJ1 (C, x)pm (C(x)) =


2
n
J1 (C, x)pm (C(x))
m
+
1 + kxk2
2

DJ1 (C, x)qm (C(x)) =


 n
2
−
m
J1 (C, x)qm (C(x)).
−
1 + kxk2
2

and

Further:
Proposition 1. ψ(w) ∈ L2 (S n ) if and only if
ψ ′ (x) =

1
1

(1+kxk2 ) 2

J1 (C, x)ψ(C(x)) and φ′ (x) =

then
Z

φ(w)ψ(w)dσ(w) =

Sn

This leads us to:

Z

Rn

′

φ (x)ψ ′ (x)dxn .

1
1

(1+kxk2 ) 2
1
1

(1+kxk2 ) 2

J1 (C, x)ψ(C(x)) ∈ L2 (Rn ). Further if
J1 (C, x)φ(C(x)) for ψ and φ ∈ L2 (S n )

Some Sharp L2 Inequalities for Dirac Type Operators

9

Theorem 7. Suppose h : Rn → Cln+1 is a smooth function with compact support. Then
Z

2

2

kDh(x)k (1 + kxk )dx

n

Rn

1

2

≥n

kh(x)k2 n
dx
1 + kxk2

Z

Rn

 12

.

In [11] it is shown that J−2 (C, x)−1 △S J2 (C, x) = △n . Proposition 1 can easily be adapted
2
4
replacing J1 (C, x) by J2 (C, x) and 1+kxk
2 by (1+kxk2 )2 . From Theorem 2 we now have:
Theorem 8. Suppose that h is as in Theorem 7. Then
Z

2

2 2

n

k△n h(x)k (1 + kxk ) dx

Rn

1
2

≥ n(n − 2)

Z

Rn

kh(x)k2
dxn
(1 + kxk2 )2

 21

.

Using Lemma 2 we also have
Theorem 9. Suppose h is as in Theorem 7. Then for n odd and k even and for n even and k
an even integer belonging to {1, . . . , n − 1}
Z

k
2

2

2 k

k△n h(x)k (1 + kxk ) dx

n

Rn

1
2

≥ |n(2 − n) · · · (n + k − 2)(k − n)|

Z

Rn

kh(x)k2
dxn
(1 + kxk2 )k

 21

and for n odd and k odd and for n even and k belonging to {1, . . . , n − 1}
Z

k

2

2 k

n

kD h(x)k (1 + kxk ) )dx

Rn

1
2

≥ |n(n + 2)(2 − n) · · · (n + k − 1)(k − 1 − n)|

Z

Rn

kh(x)k2
dxn
(1 + kxk2 )k

 12

.

Theorem 10. Suppose h : Rn → Cln+1 is a continuous function with compact support. Then
for k odd and n odd and for n even and k odd and satisfying 1 ≤ k ≤ n − 1
Z

Rn

kGk ⋆ h(x)k2 n
dx
(1 + kxk2 )k

 21

1
≤
|n(n + 2)(2 − n) · · · (n + k − 1)(k − 1 − n)|

Z

2

2 k

kh(x)k (1 + kxk ) dx

Rn

n

1
2

and for n odd and k even and for n even and k even and satisfying 1 < k < n
Z

Rn

kGk ⋆ h(x)k2 n
dx
(1 + kxk2 )k

 21

1
≤
|n(n + 2)(2 − n) · · · (n + k − 2)(k − n)|

7

Z

Rn

2

2 k

n

kh(x)k (1 + kxk ) dx

1
2

.

Concluding remarks

Let us consider the Paenitz operator on S 5 . Via the Cayley transform this operator stereographically projects to the bi-Laplacian, △25 on R5 . If we restrict attention to the equator, S 4 , of S 5 we
see that the restriction of the Paenitz operator in this context stereographically projects to the

10

A. Balinsky and J. Ryan

restriction of △25 to R4 . This operator is the bi-Laplacian △24 in R4 , while the restriction of the
Paenitz operator on S 5 to its equator, S 4 , is the Paenitz operator on S 4 . The Paenitz operator
on S 4 has a zero eigenvalue. Consequently there is no real hope of obtaining inequalities of the
type we have obtained here in Rn for the bi-Laplacian in R4 . This should explain the breakdown
of the Rellich inequality, described in [8], for the bi-Laplacian in R4 . The same rationale also
explains similar breakdowns of inequalities for Dk in Rn for n even and k ≥ n.
It should be clear that similar sharp L2 inequalities can be obtained for the operator DS +αw
α
provided −α is not in the spectrum of wDS . These operators conformally transform to D+ 1+kxk
2
in Rn . When −α is in the spectrum of wDS then we obtain a finite dimensional subspace of the
weighted L2 space L2 (Rn , (1 + kxk2 )−2 ), with weight (1 + kxk2 )−2 , consisting of solutions to the
α
Dirac equation Du + 1+kxk
2 u = 0.
All inequalities obtained here are L2 inequalities. It would be nice to see similar inequalities
for other suitable Lp spaces.

Acknowledgements
The authors are grateful to the Royal Society for support of this work under grant 2007/R1.

References
[1] Ahlfors L.V., Old and new in M¨
obius groups, Ann. Acad. Sci. Fenn. Math. 9 (1984), 93–105.
[2] Beckner W., Sharp Sobolev inequalities on the sphere and the Moser–Trudinger inequality, Ann. of Math.
138 (1993), 213–242.
[3] Bojarski B., Conformally covariant differential operators, in Proceedings of XXth Iranian Math. Congress,
Tehran, 1989.
[4] Brackx F., Delanghe R., Sommen F., Clifford analysis, Pitman, London, 1982.
[5] Branson T., Ørsted B., Spontaneous generators of eigenvalues, J. Geom. Phys. 56 (2006), 2261–2278,
math.DG/0506047.
[6] Calderbank D., Dirac operators and Clifford analysis on manifolds with boundary, Preprint no. 53, Institute
for Mathematics, Syddansk University, 1997, available at http://bib.mathematics.dk/imada/.
[7] Cnops J., Malonek H., An introduction to Clifford analysis, Textos de Matematica, Serie B, Universidade
de Coimbra, Departmento de Matematica, Coimbra, 1995.
[8] Davies E.B., Hinz A., Explicit construction for Rellich inequalities, Math. Z. 227 (1998), 511–523.
[9] Erdos L., Solovej J.P., The kernel of Dirac operators on S 3 and R3 , Rev. Math. Phys. 10 (2001), 1247–1280,
math-ph/0001036.
[10] Lieb E., Loss M., Analysis, Graduate Texts in Mathematics, Vol. 14, American Mathematical Society, Providence, 2001.
[11] Liu H., Ryan J., Clifford analysis techniques for spherical PDE, J. Fourier Anal. Appl. 8 (2002), 535–564.
[12] Ryan J., Iterated Dirac operators in Cn , Z. Angew. Math. Phys. 9 (1990), 385–401.
[13] Ryan J., Dirac operators on spheres and hyperbolae, Bol. Soc. Mat. Mexicana 3 (1996), 255–269.
[14] Sommen F., Spherical monogenic functions and analytic functionals on the unit sphere, Tokyo J. Math. 4
(1981), 427–456.
[15] Sudbery A., Quaternionic analysis, Math. Proc. Cambridge Philos. Soc. (85) (1979), 199–225.
[16] Van Lanker P., Clifford analysis on the sphere, in Clifford Algebras and their Applications in Mathematical
Physics, Editors V. Dietrich et al., Kluwer, Dordrecht, 1998, 201–215.