Brezis-Gallouet-Wainger tye inequality with critical fractional Sobolev sace and BMO Nguyen-Anh Dao, Quoc-Hung Nguyen arxiv:1805.06672v1 [math.ap] 17 May 2018 May 18, 2018 Abstract. In this aer, we rove the Brezis-Gallouet-Wainger tye inequality involving the BMO norm, the fractional Sobolev norm, and the logarithmic norm of C η, for η 0,1. 1 Introduction and main results The main urose of this aer is to established L -bound by means of the BMO norm, or the critical fractional Sobolev norm with the logarithm of C η norm. Such a L -estimate of this tye is known as the Brezis-Gallouet-Waigner BGW tye inequality. Let us remind that Brezis-Gallouet [3], and Brezis-Wainger [4] considered the relation between L, W k,r, and W s,, and roved that there holds f L C 1+log r 1 r 1+ f W s,, s > n 1.1 rovided f W k,r 1, for kr = n. Its alication is to rove the existence of solutions of the nonlinear Schrödinger equations, see details in [3]. We also note that an alternative roof of 1.1 was given by H. Engler [5] for any bounded set in with the cone condition. Similar embedding for vector functions u with divu = 0 was investigated by Beale-Kato-Majda: u L C1+ rotu L 1+log1+ u W s+1,+ rotu L 2, 1.2 for s > n, see [1] see also [10] for an imrovement of 1.2 in a bounded domain. An alication of 1.2 is to rove the breakdown of smooth solutions for the 3-D Euler equations. After that, estimate 1.2 was enhanced by Kozono and Taniuchi [6] in that rotu L can be relaxed to rotu BMO : u L C1+ rotu BMO 1+log1+ u W s+1,. 1.3 To obtain 1.3, Kozono-Taniuchi [6] roved a logarithmic Sobolev inequality in terms of BMO norm and Sobolev norm that for any 1 < <, and for s > n/, then there is a constant C = Cn,,s such that the estimate f L C 1+ f BMO 1+log + f W s, 1.4 Alied Analysis Research Grou, Faculty of Mathematics and Statistics, Ton Duc Thang University, Ho Chi Minh City, Vietnam. Email address: daonguyenanh@tdt.edu.vn Scuola Normale Sueriore, Centro Ennio de Giorgi, Piazza dei Cavalieri 3, I-56100 Pisa, Italy. Email address: quoc-hung.nguyen@sns.it 1
holds for all f W s,. Obviously, 1.4 is a generalization of 1.1. Besides, it is interesting to note that Gagliardo-Nirenberg tye inequality with critical Sobolev sace directly yields BGW tye inequality. For examle, H. Kozono, and H. Wadade [8] roved the Gagliardo-Nirenberg tye inequalities for the critical case and the limiting case of Sobolev sace as follows: u L q C n r q 1 r q u L 2ru n 1 q L, 1.5 r holds for all u L Ḣ n r,r with 1 <, 1 < r <, and for all q with q < see also Ozawa [11]. And q u L q C n q u L u 1 q BMO, 1.6 holds for all u L BMO with 1 <, and for all q with q <. As a result, 1.5 imlies u L C 1+ u L + 2ru n L r log1+ 2u s 1 r L q, 1.7 for every 1 <, 1 < r <, 1 < q < and n/q < s <. While 1.6 yields u L C 1+ u L + u BMO log1+ s 2u L q, 1.8 for every 1 <, 1 < q <, and n/q < s <. Thus, 1.7 and 1.8 may be regarded as a generalization of BGW inequality. Note that in 1.7 and 1.8, the logarithm term only contains the semi-norm u Ẇs,. Furthermore, Kozono, Ogawa, Taniuchi [7] roved the logarithmic Sobolev inequalities in Besov sace, generalizing the BGW inequality and the Beale-Kato-Majda inequality. Motivated by these above results, in this aer, we study BGW tye inequality by means of the BMO norm, the fractional Sobolev norm and the C η norm, for η 0,1. Then, our first result is as follows: Theorem 1.1 Let η 0,1, and α 0,n. Then, there exists a constant C = Cn,η > 0 such that the estimate [ ] f L C +C f BMO 1+log + fy su z z y +1 αdy + f C 1.9 η holds for all f C η BMO. We accet the notation log + s = logs if s 1, and log + s = 0 if s 0,1. fy Remark 1.2 It is clear that su z z y +1 αdy is finite if f L 1. On the other hand, if f L r, r > 1, then for any α n r,n, we have su z where the constant C is indeendent of f. fy z y +1 αdy C f L r, 2
Remark 1.3 If su f B R, then 1.9 imlies f L C +C f BMO 1+log + [ α+η + f C η ]. 1.10 Remark 1.4 Note that if f W s, with s > n, then 1.9 is stronger than 1.4 since W s, C 0,η C η, with η = s n. Concerning the BGW tye inequality involving the fractional Sobolev sace, we have the following result: Theorem 1.5 Let s > 0, 1 be such that s = n. Let α > 0, η 0,1. Then, there exists a constant C = Cn,s,,η,α > 0 such that the estimate f L C +C f Ẇs, 1+ 1 log + fy su z z y +1 αdy + f C η 1.11 holds for all f C η Ẇs,, where Ẇs, is the homogeneous fractional Sobolev sace, see its definition below. Remark 1.6 As Remark 1.4, we can see that 1.11 is stronger then 1.1. Furthermore, if su f B R, then 1.9 imlies f L C +C f Ẇs, 1+ log +[ α+η ] 1 + f C η. 1.12 Remark 1.7 We consider f δ x = log x +δψ x, where ψ Cc 1[0,, 0 ψ 1, ψ x = 1 if x 1 4, and δ > 0 is small enough. It is not hard to see that for any δ > 0 small enough and f δ L R d logδ, f δ BMOR d 1, f δ Ẇ n, logδ 1, su z f δ y z y +1 αdy 1, f δ C η δ 1. Therefore, the ower 1 and 1 of the term log + fy 2 su z z y +1 αdy + f in 1.9 and 1.11 resectively are shar that there are no such estimates of the form: f 1 C +C f 1 BMO 1+ log + 2 and f 2 L C +C f 2 Ẇ n, 1+ su z log + su z C η f 1 y γ z y +1 αdy + f 1 C, η f 2 y γ z y +1 αdy + f 2 C η hold for all f 1 BMO C η, f 2 C η Ẇs,, for some γ 0,1. 1, 3
Before closing this section, let us introduce some functional saces that we use through this aer. First of all, we recall C η, η 0,1, as the homogeneous Holder continuous of order η, endowed with the semi-norm: fx fy f C = su η x y x y η. Next, if s 0,1, then we recall Ẇ s, the homogeneous fractional Sobolev sace, endowed with the semi-norm: f Ẇs, = Rn fx fy 1 x y n+s dxdy. When s > 1, and s is not an integer, we denote Ẇs, as the homogeneous fractional Sobolev sace endowed with the semi-norm: f Ẇs, = D σ f Ẇs [s],. If s is an integer, then σ =[s] f Ẇs, = σ =[s] D σ f L. We refer to [9] for details on the fractional Sobolev sace. After that, we accet the notation f Ω := f = 1 fxdx for any Borel set Ω Ω Ω Ω. Finally, C is always denoted as a constant which can change from line to line. And Ck,n,l means that this constant merely deends on k,n,l. 2 Proof of the Theorems We first rove Theorem 1.1. Proof of Theorem 1.1. It is enough to rove that f0 C +C f BMO 1+log + fy 2 y +1 αdy + f C, 2.1 η Let m 0 N, set B ρ := B ρ 0, we have, m f0 = f0 f + B 2 m 0 B 2 m 0 f f0 + j= m 0 m B 2 j j= m 0 B 2 j f f + f B 2 j+1 B 2 m 0 f f B2 j+1 +C2 m 0n α y η f C dy +2m η 0 f BMO +C2 m 0n α B 2 m 0 C2 m 0min{n α,η} B 2 m 0 B 2 m 0 fy y +1 αdy fy y +1 αdy + f C η +Cm 0 f BMO. 4 fy y +1 αdy
Choosing log + 2 m 0 = fy y +1 αdy + f min{n α,η} C η +1, we get 2.1. The roof is comlete. Next, we rove Theorem 1.5. Proof of Theorem 1.5. To rove it, we need the following lemmas: Lemma 2.1 Let a 0 = 1, and let a 0,a 1,...,a R k+2, for any k 1, be a unique solution of the following system: Then, we have 2 jl = 0, l = 0,...,k. 2.2 a := k k j +1 0. 2.3 Moreover, for any m 1, and for b,b l R, l = m,...,m, we have k+m l+m b j+l = b l + b l b+ab. 2.4 As a result, we obtain b 1 a j=l m+1 b l b + 1 a b j+l + 1 a Proof. First of all, we note that 0, for j = 0,...,k +1. Set Then, Qx = x j. Q 1 = j. On the other hand, by 2.2, we have Q2 l = 0, for l = 0,...,k. Thus, j=1 k Qx = a x 2 l, and Q 1 = l=0 k 1 2 l. l=1 k+m b l. 2.5 5
This imlies Next, we observe that k j = 1 2 l 0. 2.6 j=1 j=1 0 = k +1 = a+ The last equation and 2.6 yield a = j = 0. j=1 k 1 2 l 0. j=1 Now, we rove 2.4. We denote LHS res. RHS is the left hand side res. the right hand side side of 2.4. It is not difficult to verify that l+m b = ab. Then, a direct comutation shows RHS = a 0 b m +a 0 +a 1 b 1 m +...+a 0 +...+a k b +a 1 +...+a b m +a 2 +...+a b m+1 +...+a b k+m = a 0 +a 1 l=1 m Note that b l + m b l l= m RHS = RHS + = = = k+m k+m = LHS. j+ b l l=j m j=l m+1 j=l m+1 +...+a k b l = 0. Thus, l= m b l + b l + b l l= m l= l+m m+k 1 b + b l + b l b l m+k +a l+m Or, we get 2.4. Finally, 2.5 follows from 2.4 by using the triangle inequality. In other words, we b l b l b l. 6
get Lemma 2.1. Next, we have Lemma 2.2 Assume a 0,a 1,...,a as in Lemma 2.1. Let Ω j = B 2 j+1\b 2 j, where B ρ := B ρ 0 for any ρ > 0. Then, there holds f Ω j C Dk fy D k f B 2 k+3\b B dy. 2.7 2 1 2 k+3\b 2 1 For any l R, we set = B 2 k+l+3\b 2 l 1. As a consequence of 2.7, we obtain f Ω j+l C2kl D k fy D k fy dydy. 2.8 Moreover, by the triangle inequality we get from 2.8 f Ω j+l C 2kl D k fy dy. 2.9 Proof. Assume a contradiction that 2.7 is not true. There exists then a sequence f m m 1 W k,1 B 2 k+3\b 2 1 such that D k f m y D k f m dy 1 B 2 k+3\b 2 1 m, 2.10 and Let us ut B 2 k+3\b 2 1 f m = 1, m 1. Ω j f m x = f m x P k,m x, with P k,m x = k l=0 α 1 +...+α n=l c l,k,m α 1,...,α n x α 1 1 xα 2 2...xαn n, and c l,k,m α 1,...,α n is a constant such that D l fm = 0, l = 0,...,k. 2.11 B 2 k+3\b 2 1 By a sake of brief, we denote c l,m = c l,k,m α 1,...,α n. Since P k,m is a olynomial of at most k-degree, then D k P k,m = const. This fact, 2.10, and 2.11 imly B 2 k+3\b 2 1 D k fm y dy = B 2 k+3\b 2 1 D k f m y D k f m dy 1 B 2 k+3\b 2 1 m. 7
It follows form the comact embeddings that there exists a subsequence of f m m 1 still denoted as f m m 1 such that that f m f strongly in L 1 B 2 k+3\b 2 1, and D k f = 0, in B2 k+3\b 2 1. This imlies that f is a olynomial of at most k 1-degree, i.e: fx = k 1 l=0 α 1 +...+α n=l c l,k α 1,...,α n x α 1 1 xα 2 2...xαn n, x B 2 k+3\b 2 1. On the other hand, we observe that for any l = 0,...,k Ω j α 1 +...+α n=l = = Ω 1 α 1 +...+α n=l Ω 1 α 1 +...+α n=l cα 1,...,α n x α 1 1 xα 2 2...xαn n dx 1 dx 2...dx n cα 1,...,α n 2 j x 1 α 1 2 j x 2 α 2...2 j x n αn dx 1 dx 2...dx n cα 1,...,α n 2 jl x α 1 1 xα 2 2...xαn n dx 1dx 2...dx n = 0, by 2.2. This imlies and f = 0, 2.12 Ω j fm Ω j = f m Ω j = 1. Remind that f m f strongly in L 1 B 2 k+3\b 2 1, then we have f Ω j = 1. Or, we comlete the roof of 2.7. The roof of 2.8 res. 2.9 is trivial then we leave it to the reader. This uts an end to the roof of Lemma 2.2. Now, we are ready to rove Theorem 1.5. It is enough to show that f0 C +C f Ẇs, 1+log + 2 1 fy y +1 αdy + f C. 2.13 η Set s 1 = s k, s 1 [0,1. Then, we divide our study into the two cases: 8
i Case: s 1 0,1: We aly Lemma 2.1 with b = f0, b j = f. Then, for any m 0 1, there is a Ω j constant C = Ck > 0 such that 0 m0 1 f0 C f f0 + k+m 0 f Ωl Ω j+l + f. 2.14 Ωl 0 Concerning the first term on the right hand side of 2.14, we have Thus, 0 f f0 Ωl f f0 Ωl Ω l f f0 Ω l x η f C η dx. 2 l+1η f C η Cη,k2 mη f C η. 2.15 Next, we use 2.8 in Lemma 2.2 to obtain f Ω j+l C 2 kl D k fy D k fz dydz, 2.16 where = B 2 k+l+3\b 2 l 1. It follows from Hölder s inequality m 2 kl m 2 kl 2 D k fy D k fz dydz D El k fy D k fz 1 1 y z n+s dydz y z n+s1 1 dydz. 1 Since y,z, we have y z y + z 2 k+l+4. Thus, the right hand side of the indicated inequality is less than Cn,,k2 kl+ln+s 1 2 m Note that n = s = k +s 1, and 2 Cn,,k2 2ln. Then, there is a constant C = Ck,s,n > 0 such that m 2 kl D k fy D k fz dydz C Thanks to the inequality m 1 c j= m 0 D El k fy D k fz 1 y z n+s dydz 1. m j 2m 0 1 m j= m 0 c j D El k fy D k fz dydz 1. y z n+s 1 1 2.17, 2.18 9
we have m 2m 0 1 D El k fy D k fz 1 y z n+s dydz 1 m D El k fy D k fz y z n+s dydz 1 1. 2.19 Moreover, we observe that Thus, m + l= χ El y 1,y 2 k +4, for all y 1,y 2. D El k fy D k fz Rn D k fy D k fz y z n+s dydz k +4 1 y z n+s dydz. 1 Combining 2.17, 2.19 and 2.20 yields m 2 kl 2.20 D k fy D k 1 fz dydz Ck,s,nm0 f Ẇs,. 2.21 It remains to treat the last term. Then, it is not difficult to see that for any α > 0 k+m 0 f Ck,n2 m 0n f Ωl B 0 2 k+m 0 2.22 Ck,n,α2 m 0n α fx dx B x +1 α. 2 k+m 0 Inserting 2.15, 2.21, and 2.22 into 2.14 yields f0 C2 m 1 0min{n α,η} fy y +1 αdy + f C +Cm η 0 f Ẇs,. 2.23 By choosing we obtain 2.13. log + 2 m 0 = fy y +1 αdy + f min{n α,η} C η +1, ii Case: s 1 = 0 s = k: The roof is similar to the one of the case s 1 0,1. There is just a difference of estimating the second term on the right hand side of 2.14 as follows: Use 2.9, we get m f Ω j+l C m 2 kl D k f. 2.24 10
Alying Hölder s inequality, we have m 2 kl We utilize the fact l= m m D k f 2 kl 1/ D k f 0 Cn,k 1 Cm0 m m 1/ D k f D k f 1/ 1/ χ El y k+4, y again in order to get D k f 1/. 2.25 1/ k+4 D k f. 2.26 From 2.26, 2.25, and 2.24, we get m f Ω j+l Ck,n f Ẇ. 2.27 s, Thus, we obtain another version of 2.23 as follows: f0 C2 m 1 0min{n α,η} fy y +1 αdy + f C +Cm η 0 f Ẇs,. 2.28 By the same argument as above after 2.23, we get the roof of the case s 1 = 0. This comletes the roof of Theorem 1.5. References [1] J. T. Beale, T. Kato, A. Majda, Remarks on the breakdown of smooth solutions for the 3-D Euler equations, Com. Math. Phys., 94 1984, 61-66. [2] Brezis H., Functional analysis, Sobolev saces and artial differential equations, Sringer, 2010. [3] Brezis H., Gallouet, T., Nonlinear Schrodinger evolution equations, Nonlinear Anal. TMA 4 1980, 677-681. [4] Brezis H., Wainger, S., A note on limiting cases of Sobolev embeddings and convolution inequalities, Comm. Partial Differential Equations 5 1980, 773-789. [5] H. Engler, An alternative roof of the Brezis-Wainger inequality, Comm. Partial Differential Equations 14 1989, 541-544. 11
[6] H. Kozono, Y. Taniuchi, Limiting case of the Sobolev inequality in BM O with alication to the Euler equations, Com. Math. Phys. 214 2000, 191-200. [7] H. Kozono, T. Ogawa, Y. Taniuchi, The critical Sobolev inequalities in Besov saces and regularity criterion to some semi-linear evolution equations, Math. Z., 242 2002, 251-278. [8] H. Kozono, and H. Wadade, Remarks on GagliardoNirenberg tye inequality with critical Sobolev sace and BMO, Math. Z. 295 2008, 935-950. [9] Eleonora Di Nezza, Giamiero Palatucci, Enrico Valdinoci, Hitchhikers guide to the fractional Sobolev saces, Bulletin des Sciences Mathmatiques, 136 2012, 521-573. [10] T. Ogawa, Y. Taniuchi, On blow-u criteria of smooth solutions to the 3-D Euler equations in a bounded domain, J. Differential Equations, 190 2003, 39-63. [11] T. Ozawa, On critical cases of Sobolev s inequalities, Journal of Functional Analysis, 127 1995, 259-269. 12