ISSN 0001-4346, Mathematical Notes, 2008, Vol. 83, No. 2, pp. 145151. c Pleiades Publishing, Ltd., 2008.

Original Russian Text c S. V. Astashkin, 2008, published in Matematicheskie Zametki, 2008, Vol. 83, No. 2, pp. 163169.

LiebThirring Inequality for Lp Norms

S. V. Astashkin*
Samara State University
Received May 31, 2007; in nal form, August 28, 2007

AbstractIn this paper, we obtain the LiebThirring inequality for Lp -norms. The proof uses only the standard apparatus of the theory of orthogonal series. DOI: 10.1134/S0001434608010173 Key words: LiebThirring inequality, Lp -norm, orthonormal system, orthogonal series, Marcinkiewicz theorem, Fourier multiplier, Rademacher function.

In 1976, Lieb and Thirring obtained the following result [1]. Theorem A. If = {j }N L2 (Rd ) is an arbitrary orthonormal system, j=1 max 1, and
N

d 2

<p1+

d , 2

:=
j=1

2 , j

(1)

then
p/(p1) 2(p1)/d N

dx

Rd

Cp,d
j=1

2 L2 (Rd ) ,

(2)

where, as is customary, = (/x1 , . . . , /xd ). More recently, inequalities of type (2) for nite orthonormal systems were established by dierent authors (they were of interest mainly because of their applications to the theory of partial dierential equations). The well-known methods of proof based on nontrivial results from spectral theory used by these authors were quite complicated (for details, see [2] and [3]). Recently, Kashin [4] proposed a new approach to the proof of inequalities of LiebThirring type, in which he used only the standard apparatus of the theory of orthogonal series: Lp -inequalities for vectorvalued Rademacher series and the classical LittlewoodPaley theorem. Let us recall some denitions and notation from [4]. Suppose that = {j }N L2 (S 1 ), j=1 where j 1, 1 j N. (3)

N Let us dene the operator P : l2 L2 (S 1 ) as follows: N

P ({cj }N ) j=1
*

=
j=1

cj j .

E-mail: astashkn@ssu.samara.ru .

145

146

ASTASHKIN

Besides, suppose that m : L2 (S 1 ) T2m is the orthogonal projection onto the space T2m , m = 0, 1, 2, . . . , of trigonometric polynomials t(z) of the form t(z) =
2m |k|<2m+1

ak z k

and, nally, m () = m P
N l2 T2m .

(4)

Note that, for the special case in which is an orthonormal (or suborthonormal) system, we have m () 1, m = 0, 1, . . . . (5)

Recall that a system {j }N is said to be suborthonormal if it satises the inequality j=1

N N

ci cj
i,j=1 S1

i j d
j=1

c2 j

for all real c1 , c2 , . . . , cN . In [4], the following theorem was proved. Theorem B. There exists an absolute constant C > 0 such that, for an arbitrary system of the form (3), the following inequality holds:
S1 N

2 d

C
m=0

m ()
j=1 2m |i|<2m+1

|j (i)|2 ,

(6)

where the function is dened by (1), and

m

m () :=
k=0

2k 2 (), k

m = 0, 1, 2, . . . .

In particular, if is an orthonormal (or suborthonormal) system, then, in view of (5), inequality (6) can be rewritten as
N S1

2 d C1
j=1

(1/2) 2 L2 (S 1 ) ,

(7)

where f (1/2) denotes the one-half derivative of a function f . Recall that, for an arbitrary > 0 and a function f (z) =
j=0

f (j)z j L2 (S 1 ),

the derivative of order is dened by the relation f () (z) :=

j=0

|j| f (j)z j .

In this paper, we show that, following the approach used in [4], one can generalize inequalities (6) and (7) to the case of Lp -norms. In the proof, we shall use the Marcinkiewicz theorem on Fourier multipliers. We begin by stating the results and making a few remarks.
MATHEMATICAL NOTES Vol. 83 No. 2 2008

LIEBTHIRRING INEQUALITY FOR Lp NORMS

147

Theorem 1. For any k = 2, 3, . . . , there exists a constant C = C(k) such that, for an arbitrary system of the form (3), the following inequality holds:
S1 N

k d C
m=0

m ()
j=1 2m |i|<2m+1

|j (i)|2 ,

(8)

where, for m = 0, 1, 2, . . . ,
m mk1 m2

m () :=
mk1 =0

2mk1 2 k1 () m
mk2 =0

2mk2 2 k2 () m
m1 =0

2m1 2 1 (), m

while the numbers m () are dened by relation (4). Remark 1. For k = 1, in view of Parsevals equality, we can write
N S1 N

d =
S1 j=1

2 j

=
m=0 j=1 2m |i|<2m+1

|j (i)|2 .

Remark 2. Inequality (8) generalizes relation (6). If, additionally, is an orthonormal (or suborthonormal) system, then, in view of (5),
m mk1 m2

m ()
mk1 =0

mk1 mk2 =0

mk2

m1 =0

2m1 2k(k1)/2 2(k1)m ,

and we obtain the inequality

N S1

k d

C1
j=1

((k1)/2) 2 L2 (S 1 ) .

(9)

Note that, in the case k = 2m + 1, m N, inequality (9) was obtained by another method in [5] (for a simpler proof, see [6]). A more general relation than (7) and (9) is established in the following theorem. Theorem 2. Suppose that k, l N, where k 2 and k is divisible by l. Then there exists a constant C = C(k) such that, for an arbitrary orthonormal (or suborthonormal) system of the form (3), the following inequality holds:
N S1

d C
j=1

((l1)/2) 2 2k/l

k/l

(10)

Let us now pass to the proof of these results. Proof of Theorem 1. Following [4], we can express the function f as

f=
m=0

m (f ),

where m (f ) :=
2m |i|<2m+1

f (i)z i .

Denote
N N

m (t, z) := m
j=1

rj (t)j (z)
2008

=
j=1

rj (t)m (j )(z),

MATHEMATICAL NOTES

Vol. 83

No. 2

148

ASTASHKIN

where the rj (t) are Rademacher functions on [0, 1]. Then, in view of the vector-valued version of Khinchines inequality [7, p. 52] and the LittlewoodPaley inequality [8, p. 63, inequalities (9)], we nd
N k

Q :=
S1

k
1 0

d =
S1 N j=1 2k

2 (z) j

d
1 0 S1 2 m (t, z) m=0 k

C1

rj (t)j
j=1 L2k

dt C2

d dt,

where the constants depend only on k. Hence Q C2 m

mk =0 0

1 S1

2 2 2 m1 (t, z)m2 (t, z) mk (t, z) d dt.

(11)

1 =0

For each i = 1, 2, . . . , k, we have

N 2 mi (t, z) 2 N

=
j=1

rj (t)mi (j )(z)

=
j,q=1

rj (t)rq (t)mi (j )(z)mi (q )(z)

and, therefore, for all t [0, 1],

N S1 2 2 2 m1 (t, z)m2 (t, z) mk (t, z) d N N N

=
j1 =1 q1 =1

jk =1 qk =1

rj1 (t)rq1 (t)rj2 (t)rq2 (t) rjk (t)rqk (t)

Note that the integral

1 0

S1

m1 (j1 )m1 (q1 ) mk (jk )mk (qk ) d.

rj1 (t)rq1 (t)rj2 (t)rq2 (t) rjk (t)rqk (t) dt

is nonzero (is equal to 1) if and only if, in the collection of the natural numbers (j1 , q1 , . . . , jk , qk ), each number occurs an even number of times. Therefore, the desired quantity
1

Im1 ,m2 ,...,mk :=

S1

2 2 m1 (t, z) mk (t, z) d dt =

1 S1 0 2 2 m1 (t, z) . . . mk (t, z) dt d

is the sum of integrals of the form

S1

m1 (j1 )m(1) (j1 )m2 (j2 )m(2) (j2 ) mk (jk )m(k) (jk ) d,

(12)

where is a permutation of the set {1, 2, . . . , k}. Conversely, expressing any such permutation as the product of cycles, we can readily verify that the corresponding integral (12) is one of the summands making up Im1 ,m2 ,...,mk . Thus,
N N N

Im1 ,m2 ,...,mk =

j1 =1 j2 =1

jk =1

S1

m1 (j1 )m(1) (j1 )m2 (j2 )

m(2) (j2 ) mk (jk )m(k) (jk ) d

N

=
S 1 j =1 1 N

m1 (j1 )m(1) (j1 )

N

j2 =1

m2 (j2 )m(2) (j2 )

jk =1

mk (jk )m(k) (jk ) d.

MATHEMATICAL NOTES

Vol. 83 No. 2 2008

LIEBTHIRRING INEQUALITY FOR Lp NORMS

149

Hence, by Cauchys inequality, we have

N 1/2 2 m1 (j1 ) j1 =1 N 2 m(k) (jk ) jk =1 N N 2 m1 (j ) j=1 N 2 m1 (j ) j=1 2 mk (j ) d. 2 mk (j ) d j1 =1 1/2 N 2 m(1) (j1 ) 1/2 N 1/2 2 mk (jk ) jk =1

Im1 ,m2 ,...,mk

S1

=
S 1 j=1 N

= k!
S 1 j=1

Thus, Im1 ,m2 ,...,mk k! Jm1 ,m2 ,...,mk , where

N N 2 m1 (j ) j=1 j=1 2 mk (j ) d.

Jm1 ,m2 ,...,mk :=

S1

Since, for any permutation : {1, 2, . . . , k} {1, 2, . . . , k}, Jm1 ,m2 ,...,mk = Jm(1) ,m(2) ,...,m(k) , in view of (11), we can write Q k! C2 m

mk =0

Jm1 ,m2 ,...,mk = (k!)2

0m1 mk N N 2 m1 (j ) L (S 1 )

Jm1 ,m2 ,...,mk

j=1 2 mk1 (j )

1 =0

(k!)2
0m1 mk j=1 N

L (S 1 )

By a lemma from [4], we have

S1

2 mk (j ) d. j=1

N 2 i (j ) j=1 L (S 1 )

2i+1 2 () i

(13)

for each i = 0, 1, 2, . . . . Therefore, Q (k!)2 C2

mk

2mk1 +1 2 k1 () m
mk =0 mk1 =0 mk1 m2 N

mk2 =0

2mk2 +1 2 k2 () m
m1 =0 N

2m1 +1 2 1 () m

S 1 j=1

2 mk (j ) d

= (k!)2 2k1
m=0

m ()
j=1 2m |i|<2m+1

|j (i)|2 ,

and inequality (8) holds with constant C = C2 2k1 (k!)2 .

MATHEMATICAL NOTES Vol. 83 No. 2 2008

150

ASTASHKIN

Proof of Theorem 2. Preserving the notation from the proof of Theorem 1, we can write Q k! C2 where, as before,
N N 2 m1 (j ) j=1 2 mk (j ) d.

m1 =0 mk =0

Jm1 ,m2 ,...,mk ,

Jm1 ,m2 ,...,mk :=

S 1 j=1

Suppose that n = k/l. Then again, in view of (13) and (5) (see also Remark 2), we have Q k!(l!)n C2
l1 0m1 ml i=1 N 2 mi (j ) j=1 l1 N 2 m(n1)l+i (j ) j=1 N 2 m2l (j ) j=1 j=1 2 mk (j ) d L (S 1 ) l1 L (S 1 ) 0ml+1 m2l i=1 N 2 ml+i (j ) j=1 L (S 1 )

0ml(n1)+1 mk i=1 N N 2 ml (j ) j=1

S1

k!(l!)n 2(l1)(n+k/2)
ml =0 N

2(l1)ml
m2l =0 N

2(l1)m2l
mk =0 N 2 mk (j ) d j=1 n 2 2(l1)m m (j )

2(l1)mk

S1

2 ml (j ) j=1 j=1 N

2 m2l (j )

= k!(l!)n 2(l1)(n+k/2)
S1 j=1 m=0

d.

Hence applying the Minkowski and LittlewoodPaley inequalities, we nd Q k!(l!)n 2(l1)(n+k/2) C2

N j=1 N S1 2 2(l1)m m (j ) m=0 n 2 2n n 1/n n

k!(l!)n 2(l1)(n+k/2) C3
j=1

(14)

where C3 > 0 is a constant depending only on k and

j (z) =
m=0

2m(l1)/2
2m |i|<2m+1

j (i)z i ,

j = 1, 2, . . . , N.

Dene the sequence (i )iZ : 0 = 0 and i = 2(l1)m/2 |i|(1l)/2 if 2m |i| < 2m+1 , m = 0, 1, 2, . . . . Since |i | 1, i Z, and the variation of this sequence on each binary interval, i.e., on the interval [2m , 2m+1 ] or [2m+1 , 2m ], m = 0, 1, 2, . . . , is at most 1 2(1l)/2 , it follows by the Marcinkiewicz theorem (see [8, pp. 6769] or [9, Chap. 4, Theorem 6]), that the sequence (i )iZ denes a Fourier multiplier in the space Lp (S 1 ) for each p (1, ). Thus, for all f Lp (S 1 ),
p 1/p

2m(l1)/2
S1 m=0 2m |i|<2m+1

f (i)z i

C4 f ((l1)/2)

Lp (S 1 ) ,

MATHEMATICAL NOTES

Vol. 83 No. 2 2008

LIEBTHIRRING INEQUALITY FOR Lp NORMS

151

where C4 > 0 is a constant depending only on p. Combining this with inequality (14), we obtain (10). The theorem is proved. ACKNOWLEDGMENTS This work was supported by the Russian Foundation for Basic Research (grant no. 07-01-96603). REFERENCES
1. E. Lieb and W. Thirring, Inequalities for the moments of the eigenvalues of the Schrodinger Hamiltonian and their relation to Sobolev inequalities, in Studies in Mathematical Physics (Princeton Univ. Press, Princeton, NJ, 1976), pp. 269303. 2. R. Temam, Innite-Dimensional Dynamical Systems in Mechanics and Physics, in Applied Mathematical Sciences (Springer-Verlag, New York, 1997), Vol. 68. 3. A. A. Il in, Lieb-Thirring integral inequalities and their applications to attractors of Navier-Stokes equations, Mat. Sb. 196 (1), 3366 (2005) [Russian Acad. Sci. Sb. Math. 196 (1), 2961 (2005)]. 4. B. S. Kashin, On a class of inequalities for orthonormal systems, Mat. Zametki 80 (2), 204208 (2006) [Math. Notes 80 (12), 199203 (2006)]. 5. J.-M. Ghidaglia, M. Marion, and R. Temam, Generalization of the SobolevLiebThirring inequalities and applications to the dimension of attractors, Dierential Integral Equations 1 (1), 121 (1988). 6. A. Eden and C. Foias, A simple proof of the generalized LiebThirring inequalities in one-space dimension, J. Math. Anal. Appl. 162 (1), 250254 (1991). 7. B. S. Kashin and A. A. Saakyan, Orthogonal Series (AFTs, Moscow, 1999) [in Russian]. 8. S. M. Nikol skii, Approximation of Functions of Several Variables and Embedding Theorems (Nauka, Moscow, 1969) [in Russian]. 9. I. Stein, Singular Integrals and Dierentiability Properties of Functions (Princeton University Press, Princeton, NJ, 1970; Mir, Moscow, 1973).

MATHEMATICAL NOTES

Vol. 83

No. 2

2008