Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
1. Introduction
Let K be a star body in Rn , then the centroid body of K is a convex
body K defined by its support function:
Z
1
hK () =
|(x, )|dx, Rn .
vol(K) K
Let K and L be two origin-symmetric star bodies in Rn such that K
L, what can be said about the volumes of K and L? Lutwak [L] proved
that, if L is a polar projection body then vol(K) vol(L). On the other
hand, if K is not a polar projection body, then there is a body L, so that
K L, but vol(K) > vol(L). Since in R2 every convex body is a polar
projection body [S], the results of Lutwak imply the following:
Suppose that K and L are two origin-symmetric convex bodies in Rn such
that K L. If n = 2, then we necessarily have vol(K) vol(L), while
this is no longer true if n 3.
Let K be a star body in Rn and p 1, then the p-centroid body of K is
the body p K defined by:
hp K () =
1
vol(K)
|(x, )| dx
1/p
,
Rn .
(1)
1
vol(K)
1/p
|(x, )| dx
,
Rn .
(3)
1
vol(K)
ln |(x, )|dx ,
Rn .
(4)
Now we can ask the question discussed above for all p > 1. Namely,
suppose that
p L p K,
(5)
Clearly the integral in (3) diverges if p 1, but still we can make sense
of this integral considering fractional derivatives. Indeed, if 1 < p < 0
Z
Z
Z
1
1
|(x, )|p dx =
|z|p
(kxkK )dx dz
vol(K) K
vol(K)
(x,)=z
Z
1
=
|z|p AK, (z)dz
vol(K)
2(p + 1) (p1)
=
A
(0),
vol(K) K,
(p1)
where AK, (z) is the parallel section function of K, and AK, (0) is its
fractional derivative at zero. (For details on fractional derivatives, see e.g.
[K5, Section 2.6]). So, in such terms our problem can be written as follows:
Suppose K and L are two origin-symmetric star bodies, so that for all
S n1 :
(p1)
AK,
(0)
(p1)
AL,
(0)
.
vol(K)
vol(L)
Do we necessarily have an inequality for the volumes of K and L?
Note that Koldobsky already considered such inequalities (see e.g. [K4])
without dividing by volumes. So, for 1 < p < 0 the positive part of our
results can also be obtained from the results of Koldobsky, but we give our
own proof. The case p = 1 leads to the following modification of the
Busemann-Petty problem. Let K and L be two convex origin-symmetric
bodies in Rn such that
voln1 (K )
voln1 (L )
.
vol(K)
vol(L)
Does this imply an inequality for the volumes of K and L?
It is easy to show that in dimensions n 4 we have vol(L) vol(K). The
proof is almost identical to that of the original solution of the BusemannPetty problem from [GKS]. The counterexamples in dimensions n 5 from
[GKS] also work in this situation.
In view of all these remarks one can consider our results as a certain bridge
between the results of Lutwak-Grinberg-Zhang about p-centroid bodies and
the results of Busemann-Petty type obtained by Koldobsky.
2. Centroid inequalities for 1 < p < 1, p 6= 0.
The Minkowski functional of a star-shaped origin-symmetric body K
is defined as
kxkK = min{a 0 : x aK}.
Rn
(7)
(8)
Integrating both sides of the last inequality over S n1 with the measure
K , we get
Z
Z
Z
Z
1
1
|(x, )|p dx dK ()
|(x, )|p dx dK ().
vol(L) S n1 L
vol(K) S n1 K
Applying Fubinis Theorem,
Z
Z
1
1
kxkpK dx
kxkpK dx.
vol(L) L
vol(K) K
(9)
Note that
Z
K
kxkpK dx =
S n1
1
n+p
kk1
K
Z
S n1
!
krkpK rn1 dr d
kkn
K d =
n
vol(K).
n+p
(10)
kxkK dx
,
n+p
vol(L) L
n + p vol(K)
therefore vol(L) vol(K), which proves the theorem for 0 < p < 1.
Now consider 1 < p < 0. In this case (7) is equivalent to
Z
Z
1
1
p
|(x, )| dx
|(x, )|p dx.
vol(L) L
vol(K) K
(11)
Integrating both sides of (11) over S n1 with the measure K and using
the same argument as in the first part of the proof, we get
Z
n
1
kxkpK dx
.
(12)
vol(L) L
n+p
Passing to spherical coordinates and applying Holders inequality
!
Z
Z
Z kk1
L
p
p
n+p1
kxkK dx =
r
kkK dr d
S n1
1
n+p
Z
S n1
(n+p)/n Z
p/n
1
n
n
kkL d
kkK d
n+p
S n1
S n1
n
(vol(L))(n+p)/n (vol(K))p/n .
n+p
Z
kknp
kkpK d
L
np
kxkK
() =
|(, )|p kknp
d.
K
2(p + 1) sin (p/2) S n1
We will use this formula in the following form:
Z
(n + p)
np
kxkK
() =
|(x, )|p dx.
2(p + 1) sin (p/2) K
Also we can write this formula in terms of fractional derivatives of the
parallel section function of K. Recall that the parallel section function of a
an origin-symmetric star body K is defined by
Z
AK, (z) =
(kxkK )dx.
(x,)=z
For 1 < q < 0 the fractional derivative of this function at zero is defined
by
Z
Z
1
1
(q)
1q
AK, (0) =
|z|
AK, (z)dz =
|(x, )|1q dx.
2(q)
2(q) K
In fact one can see that this is analytically extendable to q < 1. Therefore
Lemma 2.3 can be reformulated as follows. Let 1 < p < 1, p 6= 0, then
(n + p) (p1)
kxknp
() =
A
(0).
K
sin(p/2) K,
Note, that for 1 < p < 0 this formula was proved in [GKS].
Now recall a version of Parsevals formula on the sphere proved by Koldobsky [K3].
Lemma 2.4. If K and L are origin-symmetric infinitely smooth bodies in
n+p
S n1
Remark 2.5. A proof of this formula via spherical harmonics was given in
[K4]. Repeating this proof word by word and using the above definition of
the fractional derivative of order q < 1, one can easily extend this result
to 1 < p < 0.
Now we prove a negative counterpart of Theorem 2.1.
Theorem 2.6. Let L be an infinitely smooth origin-symmetric strictly convex body in Rn , for which (Rn , k kL ) does not embed in Lp , 1 < p < 1,
p 6= 0. Then there exists an origin-symmetric convex body K in Rn such
that
p K p L.
but
vol(L) > vol(K).
Proof. First consider 0 < p < 1. Since (Rn , k kL ) does not embed in Lp ,
there exists a S n1 such that kxkpL () is positive; for more details
see [K1]. Because kxkpL () is a continuous function on S n1 , there exists
a neighborhood of where it is positive. Define
= { S n1 : kxkpL () > 0}.
Choose a non-positive infinitely-smooth even function v supported in .
Extend v to a homogeneous function |x|np
v(x/|x|2 ) of degree n p on
2
n
R . By [K5, Lemma 3.16], the Fourier transform of |x|np
v(x/|x|2 ) is equal
2
to |x|p2 g(x/|x|2 ) for some infinitely smooth function g on S n1 .
Define a body K by
np
g(x/|x|2 )
+ |x|np
kxknp
= kxkL
2
K
for some small so that the body K is convex (see e.g. the perturbation
argument from [K5, p.96]). Applying the Fourier transform to both sides
we get
kxknp
K
() =
kxknp
L
we have
|(x, )| dx <
L
(13)
()d
<
kxkpL () kxknp
L
n1
S
Z
n
= (2)
kxkpL kxknp
dx = (2)n nvol(L).
(14)
L
S n1
Here we used a version of Parsevals formula (Lemma 2.4 and Remark 2.5)
and the fact that v is negative on .
On the other hand, again using Parsevals formula and (10)
Z
Z
p
np
np
n
kxkL () kxkK
()d = (2)
kxkpL kxkK
dx
S n1
S n1
= (2) (n + p)
K
kxkpL dx
(2) nvol(K)
vol(L)
vol(L)
p/n
.
(15)
(16)
Again using the formula from Lemma 2.3 and the fact that v() is nonpositive, we have
Z
Z
1
1
|(x, )|p dx <
|(x, )|p dx,
vol(K) K
vol(L) L
which is the same as
p K p L,
since 1 < p < 0.
Consider the integral
Z
1
kxkpL () kxknp
()d
K
vol(K) S n1
Z
Z
1
p
np
n
=
kxkL () kxkL
()d+(2)
kxkpL ()v()d
vol(L) S n1
S n1
Z
1
np
>
kxkpL () kxkL
()d = (2)n n.
(17)
vol(L) S n1
Here we used Parsevals formula and the fact that v is negative on .
On the other hand, again using Parsevals formula and Holders inequality
Z
Z
p
np
n
()d = (2)
kxkpL kxknp
dx
kxkL () kxkK
K
S n1
S n1
Z
(2)
S n1
n
p/n Z
n
kxkL dx
p/n
= (2) n (vol(L))
S n1
(n+p)/n
(vol(K))
(n+p)/n
kxkn
K dx
.
(18)
10
By the formula
(q)
AL, (0) =
cos q
2
kxkn+q+1
()
L
(n q 1)
4
(1 x2 N x4 ).
3
We have
Z
4
=
3
dz =
1
1
1
N a1+p
N +
(1+p)
1+p
(3 + p)a3+p
(1 + p)aN
N
!
.
11
(ln kxkK ) () = an
n2
z
A (z) A (0) A00 (0) z2 ... An2 (z) (n2)!
zn
dz,
(21)
S n1
for any infinitely smooth origin-symmetric star body K (see [KKYY, Remark 3.2]).
In our next Lemma we prove that a representation similar to (19) holds
for all infinitely smooth bodies, with being a signed measure.
Lemma 3.3. Let K be an infinitely smooth origin-symmetric star body in
Rn . Then
Z
1
ln kxkK =
ln |(x, )| (ln kxkK ) ()d + CK ,
(22)
(2)n S n1
where CK is the constant from (20).
Proof. Since the body K is infinitely smooth, by Theorem 3.2, (ln kxkK ) ()
is a continuous homogeneous function of degree n on Rn \ {0}.
Let be an even test function supported outside of the origin, then
Z
,
S n1
Z
=
S n1
Z
=
Rn
S n1
12
Z
ln |(x, )|(x)dx
(ln kxkK ) ()d
=
S n1
Rn
Now compute the inner integral using Fubinis theorem and the connection between the Radon and Fourier transforms (see e.g. [K5, Lemma 2.11]):
Z
Z
Z
ln |(x, )|(x)dx =
ln |t|
(x)dxdt
Rn
1
=
2
(x,)=t
Z
1
(x)dx
Here we used the formula for the Fourier transform of ln |t| (see [GS,
p.362])
(ln |z|) (t) = |t|1
(23)
It follows that
Z
as distributions
outside of the origin. Hence, the functions (2)n ln kxkK
R
and S n1 ln |(x, )| (ln kxkK ) ()d may differ only by a polynomial. But
Z
1
ln |(x, )| (ln kxkK ) ()d + ln kxkK
(2)n S n1
is a homogeneous function of degree zero, therefore this polynomial is some
constant C, which is exactly the constant from Definition 3.1, as computed
in [KKYY].
Now we need a version of Parsevals formula for L0 . How does the formula
of Lemma 2.4 look if we pass to the limit as p 0? The answer to this
question is given in our next Lemma. Even though in the proof we use
an argument based on Lemma 3.3, one can obtain the following Lemma by
taking the limit in Parsevals formula.
13
(2)n S n1 L
L
Proof. By Lemma 3.3 we have
Z
1
ln |(x, )| (ln kxkK ) ()d = ln kxkK CK .
(2)n S n1
Integrating this equality over the body L we get the statement of the Lemma.
Now we prove the main result of this section.
Theorem 3.5. Let K and L be two origin-symmetric star bodies in Rn such
that (Rn , k kK ) embeds in L0 and
0 K 0 L
(24)
Z
ln kxkK dx
L
1
vol(K)
1
ln kxkK dx = ,
n
K
14
Therefore
vol(L) vol(K).
Remark 3.6. Since every three dimensional normed space embeds in L0 (see
[KKYY, Corollary 4.3]), the previous theorem holds for all convex bodies in
R3 .
To prove our next Theorem we need the following Lemma.
Lemma 3.7. Let K be an origin-symmetric star body in Rn , then the
n
Fourier transform of kxkn
K is a continuous function on R \ {0} and equals
Z
n
(kxkK ) () = n
ln |(x, )|dx +
K
Z
0
+ (n (1) 1)vol(K)
kkn
K ln kkK d.
S n1
Proof. Let be an even test function. Using the definition of the action of
a homogeneous function of degree n (see [GS, p.303]) we get
n
h(kxkn
K ) , i = hkxkK , (x)i
Z
Z
(0))dx
+
=
kxkn ((x)
K
B1 (0)
r
S n1
Z
=
S n1
Rn \B
kkn
K
kkn
K ((r)
Z
1 (0)
(0))drd
+
r1 ((r)
(0))dr
+
kxkn
K (x)dx
Z
r1 kkn
K (r)drd
(r)dr d
S n1 1
Z
1
Z
1
1
=
kkn
K h|r| , (r)id
2 S n1
Z
Z
1
n
0
kkK h2 (1) 2 ln |t|,
()did
=
2 S n1
(,)=t
Z
0
= h
kkn
K (1) ln |(, )| d, ()i.
S n1
Here we used the formula for the Fourier transform of |r|1 from [GS, p.361]:
(|r|1 ) (t) = 20 (1) 2 ln |t|.
Thus we have proved that
(kxkn
K ) () =
Z
S n1
0
kkn
(1)
ln
|(,
)|
d.
K
(26)
15
kk1
K
r
S n1
1
=
n
n1
Z
S n1
ln |(, )|
ln rdrd +
S n1
kkn
K ln kkK
S n1
rn1 drd
1
1
+ kkn
K d +
n
n
Therefore
Z
kkn
K ln |(, )|d =
n1
S
Z
Z
=n
ln |(x, )|dx +
K
kk1
K
Z
Z
S n1
kkn
K ln kkK +
kkn
K ln |(, )|d.
1
kkn
K d.
n
(kxkn
)
()
=
n
ln |(x, )|dx +
K
K
Z
0
+ (n (1) 1)vol(K)
kkn
K ln kkK d.
S n1
Theorem 3.8. There are convex bodies K and L in Rn , n 4 such that
0 K 0 L
for every S n1 , but
vol(K) < vol(L).
Proof. Let L be a strictly convex infinitely smooth body in Rn , n 4, for
which (ln kxkL ) is not positive everywhere. (See [KKYY, Theorem 4.4]
for an explicit construction of such a body.)
Let S n1 be such that (ln kxkL ) () < 0. By continuity of the
function (ln kxkL ) () on the sphere there is a neighborhood of where this
function is negative. Let
= { S n1 : (ln kxkL ) () < 0}.
Choose an infinitely smooth body D whose Minkowski norm kxkD is equal
to 1 outside of and kxkD < 1 for x . Let v be a homogeneous function
of degree 0 on Rn \ {0}, defined as follows:
v(x) = ln kxkD ln |x|2 .
Clearly v(x) < 0 if x and v(x) = 0 if x S n1 \ .
In view of Theorem 3.2, the Fourier transforms of ln kxkD and ln |x|2
outside of the origin are some homogeneous functions of degree n, therefore
16
we have
S n1
Z
g()d = 0.
(27)
S n1
(28)
Note that formula (27) validates this definition, since integrating the last
equality over the unit sphere we get the same quantity in both sides. Also,
since L is strictly convex, there is an small enough, so that K is also convex
(see e.g. the perturbation argument from [K5, p.96]). From now on we fix
such an .
Now we will show that K together with L constructed above satisfy the
assumptions of the theorem. Apply the Fourier transform to both sides of
n
(28). Note, that the Fourier transform of |x|n
2 g(x/|x|2 ) is equal to (2) v
on test functions, whose Fourier transform is supported outside of the origin.
Such distributions can differ only by a polynomial, which must be a constant
in this case, since both functions cannot grow faster than a logarithm (see
Lemma 3.7). So
= (2)n (v + ),
|x|n
2 g(x/|x|2 )
for some constant whose value has no significance for us. Hence, by Lemma
3.7, the Fourier transform of (28) looks as follows:
n
K ln |(x, )|dx
=
vol(K)
L ln |(x, )|dx
vol(L)
+ n v() + C,
(29)
+ n
n vol(K)
vol(L)
R
R
n
kkn
L ln kkL ddx
K kkK [ln + ln kkK ] ddx
=
L
+ n
vol(K)
vol(L)
R
R
n
kkn
L ln kkL ddx
K kkK ln kkK ddx
= n ln +
L
+ n .
vol(K)
vol(L)
17
(30)
18
[GKS]