Optimal partial regularity of second-order parabolic systems under natural growth condition

  • Shuhong Chen1 and

    Affiliated with

    • Zhong Tan2Email author

      Affiliated with

      Boundary Value Problems20132013:152

      DOI: 10.1186/1687-2770-2013-152

      Received: 17 August 2012

      Accepted: 10 May 2013

      Published: 26 June 2013

      Abstract

      We consider the regularity for weak solutions of second-order nonlinear parabolic systems under a natural growth condition when m > 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq1_HTML.gif, and obtain a general criterion for a weak solution to be regular in the neighborhood of a given point. In particular, we get the optimal regularity by the method of A-caloric approximation introduced by Duzaar and Mingione.

      Keywords

      nonlinear parabolic systems natural growth condition A-caloric approximation optimal partial regularity

      1 Introduction

      Electrorheological fluids are special viscous liquids, that are characterized by their ability to undergo significant changes in their mechanical properties when an electric field is applied. This property can be exploited in technological applications, e.g., actuators, clutches, shock absorbers, and rehabilitation equipment to name a few [1].

      A model was developed for these liquids within the framework of rational mechanics [2, 3]; it takes into account the complex interactions between the electro-magnetic fields and the moving liquid. If the fluid is assumed to be incompressible, it turns out that the relevant equations of the model are the system
      div ( E + P ) = 0 , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ1_HTML.gif
      (1.1)
      curl E = 0 , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ2_HTML.gif
      (1.2)
      ρ 0 v t div S + ρ 0 [ v ] v + ϕ = ρ 0 f + [ E ] P , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ3_HTML.gif
      (1.3)
      div v = 0 , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ4_HTML.gif
      (1.4)
      where E is the electric field, P is the polarization, ρ 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq2_HTML.gif is the density, v is the velocity, S is the extra stress, ϕ is the pressure, and f is the mechanical force. In fact, in a model capable of explaining many of the observed phenomena, the extra stress has the form
      S = α 21 ( ( 1 + | D | 2 ) p 1 2 1 ) E E + ( α 31 + α 33 | E | 2 ) ( 1 + | D | 2 ) p 2 2 D + α 51 ( 1 + | D | 2 ) p 2 2 ( D E E + E D E ) , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ5_HTML.gif
      (1.5)
      where α i j http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq3_HTML.gif are material constants, and where the material function p depends on the strength of the electric field | E | 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq4_HTML.gif and satisfies
      1 < p p ( | E | 2 ) p 0 < . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ6_HTML.gif
      (1.6)
      Since the material function p, which essentially determines S, depends on the magnitude of the electric field | E | 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq4_HTML.gif, we have to deal with an elliptic or parabolic system of partial differential equations with the so-called non-standard growth conditions, i.e., the elliptic operator S satisfies
      S ( D , E ) D c 0 ( 1 + | E | 2 ) ( 1 + | D | 2 ) p 2 2 | D | 2 , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ7_HTML.gif
      (1.7)
      | S ( D , E ) | c 1 ( 1 + | D | 2 ) p 0 1 2 | E | 2 . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ8_HTML.gif
      (1.8)
      Equality (1.5) of electrorheological fluids with the conditions (1.7) and (1.8) encouraged us to considered the partial regularity of a more simple and standard model as the following:
      u t i α = 1 n D α A i α ( z , u , D u ) = B i ( z , u , D u ) , i = 1 , 2 , , N , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ9_HTML.gif
      (1.9)

      where Ω R n http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq5_HTML.gif is a bounded domain and T > 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq6_HTML.gif, z = ( x , t ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq7_HTML.gif with x Ω http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq8_HTML.gif, 0 < t T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq9_HTML.gif, denote a point in Q T = Ω × ( T , 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq10_HTML.gif. Let u ( z ) = ( u 1 ( z ) , u 2 ( z ) , , u N ( z ) ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq11_HTML.gif be a vector-valued function defined in Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq12_HTML.gif. Denote by Du the gradient of u, i.e., D u = { D α u i } i = 1 , , N ; α = 1 , , n http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq13_HTML.gif. m > 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq1_HTML.gif is a real number.

      In order to define the weak solution of (1.9), one needs to impose some regularity conditions and constructer conditions to A i α http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq14_HTML.gif and B i http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq15_HTML.gif. For a vector field A i α : Q T × R N × R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq16_HTML.gif, we shall denote the coefficients by A i α ( z , u , p ) = A i α ( x , t , u , p ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq17_HTML.gif if z = ( x , t ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq7_HTML.gif, u R N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq18_HTML.gif and p R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq19_HTML.gif. We assume that the functions ( z , u , p ) A i α ( z , u , p ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq20_HTML.gif; ( z , u , p ) A i α p β j ( z , u , p ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq21_HTML.gif are continuous in Q T × R N × R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq22_HTML.gif and that the following growth and ellipticity conditions are satisfied:

      (H1) There exists a constant L such that
      | A i α ( z , u , p ) | L ( 1 + | p | ) m 2 for all  z Q T , u R n  and  p R n N . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equa_HTML.gif
      (H2) A i α ( z , u , p ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq23_HTML.gif are differentiable functions in p and there exists a constant L such that
      | A i α p β i ( z , u , p ) | L ( 1 + | p | 2 ) m 2 2 for all  z Q T , u R n  and  p R n N . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equb_HTML.gif
      (H3) A i α http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq14_HTML.gif is uniformly strongly elliptic, that is, for some λ > 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq24_HTML.gif, we have
      ( A i α p β i ( z , u , p ) p ˜ α i ) p ˜ β j λ | p ˜ | 2 ( 1 + | p | 2 ) m 2 2 for all  z Q T , u R n  and  p , p ˜ R n N , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equc_HTML.gif

      where λ > 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq25_HTML.gif and 1 L < http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq26_HTML.gif. Now we shall specify the regularity assumptions on A i α ( z , u , p ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq23_HTML.gif with respect to the ‘coefficient’ ( z , u ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq27_HTML.gif and assume that the function ( z , u ) A i α ( z , u , p ) 1 + | p | http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq28_HTML.gif is Hölder continuous with respect to the parabolic metric | x x 0 | 2 + | t t 0 | http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq29_HTML.gif with Hölder exponent β ( 0 , 1 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq30_HTML.gif but not necessarily uniformly Hölder continuous; namely we shall assume that:

      (H4) There exists a constant L such that
      | A i α ( z , u , p ) A i α ( z 0 , u 0 , p ) | L θ ( | u | + | u 0 | , | x x 0 | + | t t 0 | + | u u 0 | ) ( 1 + | p | ) m 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equd_HTML.gif

      for any z = ( x , t ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq7_HTML.gif and z 0 = ( x 0 , t 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq31_HTML.gif in Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq12_HTML.gif. u and u 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq32_HTML.gif in R n http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq33_HTML.gif and for all p R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq19_HTML.gif, where θ ( y , s ) = min { 1 , K ˜ ( y ) s β } http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq34_HTML.gif, K ˜ : [ 0 , ) ( 1 , ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq35_HTML.gif is a given non-decreasing function. Note that θ is concave in the argument. This is the standard way to prescribe (non-uniform) Hölder continuity of the function A i α ( z , u , p ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq23_HTML.gif. We find it a bit difficult to handle, therefore, in many points of the paper, we shall use:

      (H4′) For β ( 0 , 1 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq36_HTML.gif and K : [ 0 , ) [ L , ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq37_HTML.gif monotone nondecreasing such that
      | A i α ( z , u , p ) A i α ( z 0 , u 0 , p ) | K ( | u | ) ( | x x 0 | + | t t 0 | + | u u 0 | ) β ( 1 + | p | ) m 2 , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Eque_HTML.gif

      valid for any z = ( x , t ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq38_HTML.gif and z 0 = ( x 0 , t 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq31_HTML.gif in Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq12_HTML.gif, u and u 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq32_HTML.gif in R n http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq33_HTML.gif and p R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq19_HTML.gif.

      (H5) There exist constants a and b such that
      | B i ( z , u , p ) | a | p | m + b , sup Q T | u | = V , 2 a V < λ . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equf_HTML.gif
      Finally, we remark a trial consequence of the continuity of A i α p β j http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq39_HTML.gif; this implies the existence of a function ω : [ 0 , ) × [ 0 , ) [ 0 , ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq40_HTML.gif with ω ( t , 0 ) = 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq41_HTML.gif for all t such that t ω ( t , s ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq42_HTML.gif is nondecreasing for fixed s, s ω m ( t , s ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq43_HTML.gif is concave and nondecreasing for fixed t, and such that
      ( H 6 ) | A i α p β j ( x , t , u , p ) A i α p β j ( x 0 , t 0 , u 0 , p 0 ) | L ( 1 + | p | 2 + | p 0 | 2 ) m 2 2 × ω ( | u | + | p | , | x x 0 | 2 + | t t 0 | + | u u 0 | 2 + | p p 0 | 2 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equg_HTML.gif

      for any z = ( x , t ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq7_HTML.gif and z 0 = ( x 0 , t 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq31_HTML.gif in Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq12_HTML.gif, any u, u 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq32_HTML.gif in R n http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq33_HTML.gif and p , p 0 R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq44_HTML.gif whenever | u | + | p | + | u u 0 | + | p p 0 | M http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq45_HTML.gif.

      From (H2) and (H3) we immediately deduce the following:
      | A i α ( z , u , p ) A i α ( z , u , q ) | L ( 1 + | p | 2 + | q | 2 ) m 2 2 | p q | , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ10_HTML.gif
      (1.10)
      ( A i α ( z , u , p ) A i α ( z , u , q ) ) ( p q ) λ ( 1 + | p | 2 + | q | 2 ) m 2 2 | p q | 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ11_HTML.gif
      (1.11)

      for all z Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq46_HTML.gif, u R N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq18_HTML.gif and p , q R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq47_HTML.gif.

      Definition 1.1 By a weak solution of (1.9) under the assumptions (H1)-(H5), we mean a vector-valued function u L m ( T , 0 ; W 1 , m ( Ω , R N ) ) L ( Q T ; R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq48_HTML.gif such that
      Q T ( A i α ( z , u , D u ) D α φ i u i φ t i ) d z = Q T B i ( z , u , D u ) φ i d z http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ12_HTML.gif
      (1.12)

      for all φ C 0 ( Q T , R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq49_HTML.gif.

      In [4] Duzaar and Mingione considered the partial regularity of homogeneous systems of (1.9) with m 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq50_HTML.gif under the natural growth condition. In this paper, we extend their results to the case of m > 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq1_HTML.gif. We have to overcome the difficulty of m > 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq1_HTML.gif. Motivated by the works of Duzaar [4, 5], Chen and Tan [69] and Tan [10], we use the technique of ‘A-caloric approximation’ to establish the optimal partial regularity of nonlinear parabolic systems (1.9). In fact, the use of the ‘A-caloric approximation lemma’ allows optimal regularity, without the use of Reverse-Hölder inequalities and (parabolic) Gehring’s lemma. The method is based on an approximation result that we called the ‘A-caloric approximation lemma’. This is the parabolic analogue of the classical harmonic approximation lemma of De Giorgi [11, 12] and allows to approximate functions with solutions to parabolic systems with constant coefficients in a similar way as the classical harmonic approximation lemma does with harmonic functions. And we can obtain the following theorem.

      Theorem 1.1 Let u L m ( T , 0 ; W 1 , m ( Ω , R N ) ) L ( Q T ; R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq51_HTML.gif be a weak solution to system (1.9) under the assumptions (H1)-(H4) and the natural growth condition (H5) and denote by Q 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq52_HTML.gif the set of regularity points of u in Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq12_HTML.gif:
      Q 0 = { z Q T : D u C β , β / 2 ( O , R n N ) , O Q T is a neighborhood of z } . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equh_HTML.gif
      Then Q 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq52_HTML.gif is an open subset with full measure, and therefore
      D u C β , β / 2 ( Q 0 , R n N ) , | Q T Q 0 | = 0 . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equi_HTML.gif

      At the end of the section, we summarize some notions which we will be used in this paper. For x 0 R n http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq53_HTML.gif, t 0 R http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq54_HTML.gif, we denote B ( x 0 , R ) = { x R n : | x x 0 | < R } http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq55_HTML.gif, Q ( ( x 0 , t 0 ) , R ) = B ( x 0 , R ) × ( t 0 R 2 , t 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq56_HTML.gif. If v is an integrable function in Q ( z 0 , ρ ) = Q ρ ( z 0 ) = B ρ ( x 0 ) × ( t 0 ρ 2 , t 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq57_HTML.gif, z 0 = ( x 0 , t 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq58_HTML.gif, we will denote its average by http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq59_HTML.gif , where α n http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq60_HTML.gif denotes the volume of the unit ball in R n http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq33_HTML.gif. We remark that in the following, when not crucial, the ‘center’ of the cylinder will be often unspecified, e.g., Q ρ ( z 0 ) = Q ρ http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq61_HTML.gif; the same convention will be adopted for balls in R n http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq33_HTML.gif therefore denoting B ( x 0 , ρ ) = B ρ ( x 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq62_HTML.gif. Finally, in the rest of the paper, the symbol C will denote a positive, finite constant that may vary from line to line; the relevant dependencies will be specified.

      2 The A-caloric approximation technique and preliminaries

      In this section we introduce the A-caloric approximation lemma [4] and some preliminaries. Recall a strongly elliptic bilinear form A i α http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq14_HTML.gif on R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq63_HTML.gif with an ellipticity constant λ > 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq25_HTML.gif, and upper bound Λ > 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq64_HTML.gif means that λ | p ˜ | 2 A i α ( p ˜ , p ˜ ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq65_HTML.gif, A i α ( p , p ˜ ) Λ | p | | p ˜ | http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq66_HTML.gif, p , p ˜ R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq67_HTML.gif, we define A-caloric approximation function.

      Definition 2.1 We shall say that a function h L 2 ( 1 , 0 ; W 1 , 2 ( B ρ , R N ) ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq68_HTML.gif is A-caloric on Q ρ http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq69_HTML.gif if it satisfies
      Q ρ ( h i φ t i A i α ( D h , D α φ i ) ) d z = 0 for all  φ C 0 ( Q ρ , R N ) . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equj_HTML.gif

      Remark 2.2 Obviously, when A ( p ˜ , p ˜ ) | p ˜ | 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq70_HTML.gif for every p ˜ R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq71_HTML.gif, then an A-caloric function is just a caloric function h t h 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq72_HTML.gif.

      Lemma 2.3 (A-caloric approximation lemma)

      There exists a positive function δ ( n , N , λ , Λ , ε ) 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq73_HTML.gif with the following property: Whenever A is a bilinear form on R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq63_HTML.gif, which is strongly ellipticity constant λ > 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq25_HTML.gif and upper bound Λ, ε is a positive number, and u L 2 ( 1 , 0 ; W 1 , 2 ( B , R N ) ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq74_HTML.gif with
      Q ( | u | 2 + | D u | 2 ) d z 1 , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ13_HTML.gif
      (2.1)
      is approximatively A-caloric in the sense that
      | Q ( u φ t A ( D u , D φ ) ) d z | δ sup Q | D φ | for all φ C 0 ( Q , R N ) , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ14_HTML.gif
      (2.2)
      then there exists an A-caloric function h such that
      Q ( | h | 2 + | D h | 2 ) d z 1 , and Q | u h | 2 d z ε . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ15_HTML.gif
      (2.3)

      Actually, we could have directly applied Theorem 5 of [13] with the choice X = W 1 , 2 ( B , R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq75_HTML.gif, B = L 2 ( B , R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq76_HTML.gif, R = W l , 2 ( B , R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq77_HTML.gif, F = ( v k ) k N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq78_HTML.gif, p = 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq79_HTML.gif to conclude that ( v k ) k N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq80_HTML.gif is relatively compact in L 2 ( Q T , R N ) = L 2 ( 1 , 0 ; L 2 ( B , R N ) ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq81_HTML.gif.

      Lemma 2.4 There exists a positive function δ ( n , N , λ , Λ , ε ) 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq82_HTML.gif with the following property: Whenever A is a bilinear form on R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq63_HTML.gif which is strongly ellipticity constant λ > 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq25_HTML.gif and upper bound Λ, ε is a positive number, and u L 2 ( t 0 ρ 2 , t 0 ; W 1 , 2 ( B ρ ( x 0 ) , R N ) ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq83_HTML.gif with
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ16_HTML.gif
      (2.4)
      is approximatively A-caloric in the sense that
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ17_HTML.gif
      (2.5)
      then there exists h L 2 ( t 0 ρ 2 , t 0 ; W 1 , 2 ( B ρ ( x 0 ) , R N ) ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq84_HTML.gif A-caloric on Q ρ ( z 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq85_HTML.gif such that
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ18_HTML.gif
      (2.6)

      For u L 2 ( Q ρ ( z 0 ) , R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq86_HTML.gif we denote by l z 0 , ρ http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq87_HTML.gif the unique affine function (in space) l ( z ) = l ( x ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq88_HTML.gif minimizing http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq89_HTML.gif , amongst all affine functions a ( z ) = a ( x ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq90_HTML.gif which are independent of t. To get an explicit formula for l z 0 , ρ http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq87_HTML.gif, we note that such a unique minimum point exists and takes the form l z 0 , ρ ( x ) = ξ z 0 , ρ + ν z 0 , ρ ( x x 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq91_HTML.gif, where ν z 0 , ρ R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq92_HTML.gif. A straightforward computation yields that http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq93_HTML.gif , for any affine function a ( x ) = ξ + ν ( x x 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq94_HTML.gif with ξ R N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq95_HTML.gif and ν R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq96_HTML.gif. This implies in particular that http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq97_HTML.gif and http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq98_HTML.gif .

      For convenience we recall from [14] the following.

      Lemma 2.5 Let u L 2 ( Q ρ ( z 0 ) , R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq86_HTML.gif, 0 < θ < 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq99_HTML.gif, and l z 0 , ρ http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq87_HTML.gif respectively l z 0 , θ ρ http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq100_HTML.gif the unique affine functions minimizing http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq89_HTML.gif respectively http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq101_HTML.gif . Then there holds
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equk_HTML.gif
      Moreover, if D u L 2 ( Q ρ ( z 0 ) , R n N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq102_HTML.gif, we have
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equl_HTML.gif

      3 Caccioppoli second inequality

      In this section we prove Caccioppoli’s second inequality.

      Theorem 3.1 (Caccioppoli second inequality)

      Let u L m ( T , 0 ; W 1 , m ( Ω , R N ) ) L ( Q T ; R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq103_HTML.gif be a weak solution to (1.9) under the assumptions (H1)-(H4) and the natural growth condition (H5). Then, for any M > 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq104_HTML.gif, any affine function l ( z ) = l ( x ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq88_HTML.gif independent of t and satisfying | l ( z 0 ) | + | D l | M http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq105_HTML.gif, and any Q ρ ( z 0 ) Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq106_HTML.gif with 0 < ρ < R 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq107_HTML.gif, we have
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equm_HTML.gif
      Proof We take the test function φ = η 2 ξ 2 ( u l ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq108_HTML.gif, where η ( x ) C 0 1 ( B R ( x 0 ) ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq109_HTML.gif is a cut-off function in space such that 0 η 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq110_HTML.gif, η 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq111_HTML.gif in B ρ ( x 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq112_HTML.gif, | D η | 1 ( R ρ ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq113_HTML.gif. While ξ C 1 ( R ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq114_HTML.gif is a cut-off function in time such that, with 0 < σ < ρ http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq115_HTML.gif being arbitrary,
      { ξ 1 , on  ( t 0 ρ 2 , t 0 σ 2 ) , ξ 0 , on  ( , t 0 R 2 ) ( t 0 , ) , 0 ξ 1 , on  R , ξ t 0 , on  ( t 0 ρ 2 , ) , | ξ t | 1 | R ρ | 2 , on  ( t 0 R 2 , t 0 ρ 2 ) . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equn_HTML.gif
      Thus, we obtain
      Q R ( z 0 ) A i α ( z , u , D u ) D ( u l ) i ξ 2 η 2 d z = 2 Q R ( z 0 ) A i α ( z , u , D u ) ξ 2 η η ( u l ) i d z + Q R ( z 0 ) u i t φ i d z + Q R ( z 0 ) B i ( z , u , D u ) φ i d z . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equo_HTML.gif
      We further have
      Q R ( z 0 ) A i α ( z , u , D l ) D α ( u l ) i ξ 2 η 2 d z = 2 Q R ( z 0 ) A i α ( z , u , D l ) ξ 2 η η ( u l ) i d z Q R ( z 0 ) A i α ( z , u , D l ) D α φ i d z http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equp_HTML.gif
      and
      0 = Q R ( z 0 ) A i α ( z 0 , l ( z 0 ) , D l ) D α φ i d z . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equq_HTML.gif
      Adding these equations and using l t 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq116_HTML.gif, we deduce
      Q R ( z 0 ) [ A i α ( z , u , D u ) A i α ( z , u , D l ) ] D ( u l ) ξ 2 η 2 d z = 2 Q R ( z 0 ) [ A i α ( z , u , D u ) A i α ( z , u , D l ) ] ξ 2 η η ( u l ) d z Q R ( z 0 ) [ A i α ( z , u , D l ) A i α ( z , l , D l ) ] D α φ i d z Q R ( z 0 ) [ A i α ( z , l , D l ) A i α ( z 0 , l ( z 0 ) , D l ) ] D α φ i d z + Q R ( z 0 ) ( u l ) i t φ i d z + Q R ( z 0 ) B i ( z , u , D u ) φ i d z I + II + III + IV + V . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ19_HTML.gif
      (3.1)
      By (1.10) and Young’s inequality, we have
      I ε Q R ( z 0 ) ( 1 + | D l | 2 ) m 2 2 | D u D l | 2 ξ 2 η 2 d z + ε ( m 1 ) C m Q R ( z 0 ) ξ m | η | m | u l | m d z + C 2 ε Q R ( z 0 ) ( 1 + | D l | 2 ) m 2 2 ξ 2 | η | 2 | u l | 2 d z + ε Q R ( z 0 ) | D u D l | m ξ m m 1 η m m 1 d z . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ20_HTML.gif
      (3.2)
      By the condition (H4′) and Young’s inequality, we can get
      II ε Q R ( z 0 ) ξ 2 η 2 | D u D l | 2 d z + ( 1 ε + 1 ) 1 | R ρ | 2 Q R ( z 0 ) ξ 2 η | u l | 2 d z + ( 1 ε + 2 2 1 β ) [ K ( | l | ) ( 1 + | D l | ) m 2 ] 2 1 β α n R n + 2 + 2 β 1 β . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ21_HTML.gif
      (3.3)
      Similarly, we can estimate III as follows:
      III ε Q R ( z 0 ) ξ 2 η 2 | D u D l | 2 d z + Q R ( z 0 ) ξ 2 | η | 2 | u l | 2 d z + ( 1 ε + 4 ) [ K ( | l | ) ( 1 + | D l | ) m 2 + β ] 2 α n R n + 2 + 2 β . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ22_HTML.gif
      (3.4)
      Using the fact that ξ 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq117_HTML.gif on ( , t 0 R 2 ) ( t 0 , ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq118_HTML.gif, taking into account that ξ ξ t 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq119_HTML.gif for t > t 0 ρ 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq120_HTML.gif and | ξ t | 1 | R ρ | 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq121_HTML.gif, we infer
      IV = Q R ( z 0 ) ( u l ) i t φ i d z = Q R ( z 0 ) | u l | 2 η 2 t ( ξ 2 ) d z + 1 2 Q R ( z 0 ) ξ 2 η 2 t | u l | 2 d z = 1 2 Q R ( z 0 ) | u l | 2 η 2 t ( ξ 2 ) d z = Q R ( z 0 ) | u l | 2 η 2 ξ ξ t d z 1 | R ρ | 2 Q R ( z 0 ) | u l | 2 d z , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ23_HTML.gif
      (3.5)
      and for μ positive to be fixed later, we have
      V = Q R ( z 0 ) a | D u | m ξ 2 η 2 | u l | d z + Q R ( z 0 ) ( | u l | R ρ ξ η ) ( ξ η b ( R ρ ) ) d z Q R ( z 0 ) a [ ( 1 + μ ) | D u D l | m + ( 1 + 1 μ ) | D l | m ] ξ 2 η 2 | u l | d z + 1 2 ε Q R ( z 0 ) ( | u l | R ρ ξ η ) 2 d z + ε 2 Q R ( z 0 ) ξ 2 η 2 b 2 R 2 d z a V ( 1 + μ ) Q R ( z 0 ) ξ 2 η 2 | D u D l | m d z + 1 ε Q R ( z 0 ) ( | u l | R ρ ξ η ) 2 d z + ε 2 [ a 2 ( 1 + 1 μ ) 2 | D l | 2 m + b 2 ] α n R n + 4 . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ24_HTML.gif
      (3.6)
      By (1.11) we have
      Q R ( z 0 ) [ A i α ( z , u , D u ) A i α ( z , u , D l ) ] D α ( u l ) i ξ 2 η 2 d z λ Q R ( z 0 ) ( 1 + | D u | 2 + | D l | 2 ) m 2 2 | D u D l | 2 ξ 2 η 2 d z λ Q R ( z 0 ) [ ( 1 + | D l | 2 ) m 2 2 | D u D l | 2 ξ 2 η 2 + | D u D l | m ξ 2 η 2 ] d z . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ25_HTML.gif
      (3.7)
      Combining (3.2)-(3.7) in (3.1) and noting that R 2 β 1 β R 2 β http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq122_HTML.gif ( R 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq123_HTML.gif), that 2 2 1 β > 4 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq124_HTML.gif, that [ K ( | l | ) ( 1 + | D l | ) m 2 + β ] 2 [ K ( | l | ) ( 1 + | D l | ) m 2 ] 2 1 β http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq125_HTML.gif (for K 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq126_HTML.gif), choosing ε sufficiently small and taking into account that 2 a V λ http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq127_HTML.gif, that ξ 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq128_HTML.gif for t [ t 0 ρ 2 , t 0 σ 2 ] http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq129_HTML.gif, that η 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq111_HTML.gif on B ρ ( x 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq112_HTML.gif, we infer that
      t 0 ρ 2 t 0 σ 2 B ρ ( x 0 ) [ ( 1 + | D l | 2 ) m 2 2 | D u D l | 2 + | D u D l | m ] d x d t C 1 [ ( 1 + | D l | 2 ) m 2 2 Q R ( z 0 ) 1 | R ρ | 2 | u l | 2 d z + Q R ( z 0 ) 1 | R ρ | m | u l | m d z ] + C 2 [ K ( | l | ) ( 1 + | D l | ) m 2 ] 2 1 β α n R n + 2 + 2 β + C 3 [ a 2 ( 1 + 1 μ ) 2 | D l | 2 m + b 2 ] α n R n + 4 . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equr_HTML.gif

      Then the desired result follows by taking the limit σ 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq130_HTML.gif. □

      4 The proof of the main theorem

      The next lemma is a prerequisite for applying the A-caloric approximation technique.

      Lemma 4.1 Let u L m ( T , 0 ; W 1 , m ( Ω , R N ) ) L ( Q T ; R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq51_HTML.gif be a weak solution to (1.9) under the assumptions (H1)-(H6). Then for any M > 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq104_HTML.gif, we have
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equs_HTML.gif
      for any Q ρ ( z 0 ) Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq106_HTML.gif and φ C 0 ( Q ρ ( z 0 ) , R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq131_HTML.gif with ρ 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq132_HTML.gif and any affine function l ( z ) = l ( x ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq88_HTML.gif independent of time, satisfying | l ( z 0 ) | + | D l | M http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq105_HTML.gif. Here C Eu = C Eu ( M , L , m ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq133_HTML.gif and we write
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equt_HTML.gif
      Proof Without loss of generality, we can assume that sup Q ρ ( z 0 ) | D φ | 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq134_HTML.gif. From (1.12) and the fact that http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq135_HTML.gif and http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq136_HTML.gif , we deduce
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equu_HTML.gif
      In turn, we split the first integral as follows:
      I = 1 | Q ρ ( z 0 ) | s 1 ( ) d z + 1 | Q ρ ( z 0 ) | s 2 ( ) d z = I 1 + I 2 , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equv_HTML.gif

      and s 1 = Q ρ ( z 0 ) { z : | D u D l | 1 } http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq137_HTML.gif, s 2 = Q ρ ( z 0 ) { z : | D u D l | > 1 } http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq138_HTML.gif.

      We proceed estimating the two resulting pieces. As for I 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq139_HTML.gif, using (H6), the fact that s ω m ( t , s ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq43_HTML.gif is concave and Jensen’s inequality (note that m 1 m > 1 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq140_HTML.gif), we get
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equw_HTML.gif
      To estimate I 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq141_HTML.gif, we preliminarily observe that, using Hölder inequality,
      | s 2 | s 2 | D u D l | d z ( s 2 d z ) 1 2 ( s 2 | D u D l | 2 d z ) 1 2 | s 2 | ( Q ρ | D u D l | 2 d z ) 1 2 , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equx_HTML.gif
      and therefore
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equy_HTML.gif
      Similarly, we also have
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equz_HTML.gif
      Using (H1), (H2) and the previous inequality, we then conclude the estimate of I 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq141_HTML.gif as follows:
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equaa_HTML.gif
      Combining the estimates found for I 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq139_HTML.gif and I 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq141_HTML.gif, we have
      | I | L ( 1 + M 2 ) m 2 2 ω ( M + 1 , Φ ) Φ + 2 L ( 1 + M ) m 2 Φ . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equab_HTML.gif
      For the remaining pieces, using (H4′), we deduce
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equac_HTML.gif
      Here we have used that K 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq126_HTML.gif and the assumption that ρ 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq142_HTML.gif. Using again (H4′) and Young’s inequality, we estimate
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equad_HTML.gif
      and
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equae_HTML.gif
      Noting the definition of H and combining the estimates just found for I, II, III and IV, we obtain
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equaf_HTML.gif

      A simple scaling argument yields the result for general φ. □

      The next lemma is a standard estimate for weak solutions to linear parabolic systems with constant coefficients [15], Lemma 5.1.

      Lemma 4.2 Let h L 2 ( t 0 ρ 2 , t 0 ; W 1 , 2 ( B ρ ( x 0 ) , R N ) ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq143_HTML.gif be a weak solution in Q ρ ( z 0 ) = B ρ ( x 0 ) × ( t 0 ρ 2 , t 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq144_HTML.gif of the following linear parabolic system with constant coefficients:
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equag_HTML.gif
      where the coefficients A i α http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq14_HTML.gif satisfy A i α ( p , p ) λ | p | 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq145_HTML.gif, A i α ( p , p ˜ ) L | p | | p ˜ | http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq146_HTML.gif for any p , p ˜ R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq147_HTML.gif. Then h is smooth in Q ρ ( z 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq85_HTML.gif and there exists a constant C pa = C pa ( n , N , L / λ ) 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq148_HTML.gif such that
      ψ ˜ ( z 0 , θ ρ ) C pa θ 2 ψ ˜ ( z 0 , ρ ) , 0 < θ < 1 . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equah_HTML.gif
      Here we write
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equai_HTML.gif

      In the following we consider a weak solution u of the nonlinear parabolic system (1.9) on a fixed sub-cylinder Q ρ ( z 0 ) Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq149_HTML.gif and ρ 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq132_HTML.gif.

      Lemma 4.3 Given M > 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq104_HTML.gif and 0 < β < α < 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq150_HTML.gif, there exist θ ( 0 , 1 2 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq151_HTML.gif and δ ( 0 , 1 ] http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq152_HTML.gif depending only on n, N, λ, L, β, α and m such that if
      ω ( M + 1 , Ψ ˜ ( z 0 , ρ , l z 0 , ρ ) ) + Ψ ˜ ( z 0 , ρ , l z 0 , ρ ) δ 2 , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equaj_HTML.gif
      on Q ρ ( z 0 ) Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq149_HTML.gif for some 0 < ρ 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq153_HTML.gif and such if
      | l z 0 , ρ ( z 0 ) | + | D l z 0 , ρ | M , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equak_HTML.gif
      then
      Ψ ˜ ( z 0 , θ ρ , l z 0 , θ ρ ) θ 2 α Ψ ˜ ( z 0 , ρ , l z 0 , ρ ) + C 6 ρ 2 β H 2 ( M ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equal_HTML.gif
      for
      Ψ ˜ ( z 0 , ρ , l z 0 , ρ ) = Ψ ( z 0 , ρ , l z 0 , ρ ) + H 2 ( M ) ρ 2 β . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equam_HTML.gif
      Proof Given M > 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq154_HTML.gif. And we shall always consider ρ 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq132_HTML.gif. We first want to apply Lemma 4.1 on Q ρ / 2 ( z 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq155_HTML.gif to u l http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq156_HTML.gif, where l ( z ) = l ( x ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq88_HTML.gif is an affine function independent of t satisfying | l ( z 0 ) | + | D l | M http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq105_HTML.gif. We observe that Ψ has the following property:
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ26_HTML.gif
      (4.1)
      From Caccioppoli’s second inequality, we infer
      Φ ( z 0 , ρ / 2 , l ) C cac [ 2 m Ψ ( z 0 , ρ , l ) + 2 H ( M ) ρ 2 β ] = C ˜ cac Ψ ˜ ( z 0 , ρ , l ) . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ27_HTML.gif
      (4.2)
      From Lemma 4.1 we therefore get, for any φ C 0 ( Q ρ / 2 ( z 0 ) , R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq157_HTML.gif, that
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ28_HTML.gif
      (4.3)

      where C ˜ Eu = C ˜ Eu ( L , M , m ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq158_HTML.gif.

      For given ε > 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq159_HTML.gif to be specified later, we let δ = δ ( n , N , λ , L , ε ) ( 0 , 1 ] http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq160_HTML.gif to be constant from Lemma 2.3. Define γ = C ˜ Eu Ψ ( z 0 , ρ ) + 4 δ 2 H 2 ( M ) ρ 2 β http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq161_HTML.gif and w = γ 1 ( u l ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq162_HTML.gif.

      Then from (4.3) we deduce that, for all φ C 0 ( Q ρ / 2 ( z 0 ) , R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq157_HTML.gif, the following holds:
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ29_HTML.gif
      (4.4)
      Moreover, we estimate, using Caccioppoli’s second inequality, (4.1) and (4.2),
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ30_HTML.gif
      (4.5)

      provided we have chosen C ˜ Eu 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq163_HTML.gif large enough.

      Assuming the smallness condition,
      ω ( M + 1 , Ψ ˜ ( z 0 , ρ , l z 0 , ρ ) ) + Ψ ˜ ( z 0 , ρ , l z 0 , ρ ) δ 2 , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ31_HTML.gif
      (4.6)
      satisfied. Then (4.4) and (4.5) allow us to apply Lemma 2.4, i.e., they yield the existence of h L 2 ( t 0 ρ 2 , t 0 ; W 1 , 2 ( B ρ ( x 0 ) , R N ) ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq84_HTML.gif solving the A i α p β j http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq39_HTML.gif-heat equation on Q ρ / 2 ( z 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq155_HTML.gif and satisfying
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ32_HTML.gif
      (4.7)
      and
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ33_HTML.gif
      (4.8)
      From Lemma 4.2 we recall that h satisfies, for any 0 < θ < 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq99_HTML.gif, the a priori estimate (note that C pa = C pa ( n , N , λ , L ) 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq164_HTML.gif)
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equan_HTML.gif
      Here we have used that http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq165_HTML.gif , and http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq166_HTML.gif and (4.7). Combining the previous estimate with (4.8), we deduce
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ34_HTML.gif
      (4.9)
      Recalling back ( u l ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq167_HTML.gif via w = u l γ http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq168_HTML.gif, we arrive at
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ35_HTML.gif
      (4.10)
      Next we use the minimizing property of l z 0 , θ ρ / 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq169_HTML.gif
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ36_HTML.gif
      (4.11)
      At the same time, from (4.11), we can see that: For 2 m n + 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq170_HTML.gif ( n 3 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq171_HTML.gif), we have 2 < m < m http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq172_HTML.gif, where
      m = { m ( n + 2 ) n m + 2 if  n + 2 > m , m > m if  m = n + 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equao_HTML.gif

      with 1 m < 1 m < 1 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq173_HTML.gif. Therefore we can find s [ 0 , 1 ] http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq174_HTML.gif such that 1 m = 1 s 2 + s m http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq175_HTML.gif.

      Using Sobolev’s, Caccioppoli’s and Young’s inequalities together with (4.11), we have
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ37_HTML.gif
      (4.12)
      Using Lemma 2.5, Caccioppoli’s inequality, (4.4), (4.6), (4.12) and Young’s inequality, we obtain
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ38_HTML.gif
      (4.13)
      From (4.12) and (4.13), we conclude
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ39_HTML.gif
      (4.14)
      provided γ 2 ( m m ) / m θ 2 + ( n + m ) m / m http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq176_HTML.gif and we fixed ε = θ n + 6 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq177_HTML.gif. That it is to say,
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ40_HTML.gif
      (4.15)
      Combining (4.11) and (4.15) yields the desired estimate
      Ψ ( z 0 , θ ρ / 2 , l z 0 , θ ρ / 2 ) C 5 θ 2 ( Ψ ( z 0 , ρ , l ) + 4 δ 2 H 2 ( M ) ρ 2 β ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ41_HTML.gif
      (4.16)

      for C 5 = C 4 + 12 C pa http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq178_HTML.gif. Given β < α < 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq179_HTML.gif, we choose 0 < θ < 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq99_HTML.gif such that 2 2 α C 5 θ 2 θ 2 α http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq180_HTML.gif with θ = θ ( n , m , N , λ , L , α , β ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq181_HTML.gif. This also fixes the constants ε = ε ( n , m , N , λ , L , α , β ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq182_HTML.gif and δ = δ ( n , m , N , λ , L , α , β ) ( 0 , 1 ] http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq183_HTML.gif. Thus we have shown Lemma 4.3. □

      In the following, we want to iterate Lemma 4.3. That is,

      Lemma 4.4 For M > 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq154_HTML.gif and Q ρ ( z 0 ) Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq106_HTML.gif, suppose that the conditions
      ( i ) | l z 0 , ρ | + | ( D l ) z 0 , ρ | M ; ( ii ) ρ ρ 0 ( M ) ; ( iii ) Ψ ˜ ( ρ ) Ψ ˜ 0 ( M ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equap_HTML.gif
      are satisfied. Then, for every j N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq184_HTML.gif, we have
      Ψ ˜ ( z 0 , θ j ρ , l z 0 , θ ρ ) θ 2 α j Ψ ˜ ( z 0 , ρ , l z 0 , ρ ) + C 6 ( M ) ( θ j ρ ) 2 β H 2 ( M ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equaq_HTML.gif
      and
      | l z 0 , θ j ρ | + | ( D l ) z 0 , θ j ρ | 2 M . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equar_HTML.gif
      Moreover, the limit
      Γ z 0 = lim j ( D u ) z 0 , θ j ρ / 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equas_HTML.gif
      exists, and the estimate
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equat_HTML.gif

      is valid for a constant C = C ( n , N , λ , α , L , β , M , m ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq185_HTML.gif.

      Proof For fixed z 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq186_HTML.gif we shall denote l z 0 , ρ l ρ http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq187_HTML.gif. For given M > 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq154_HTML.gif (and β < α < 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq179_HTML.gif), we determine δ = δ ( 2 M ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq188_HTML.gif, θ = θ ( 2 M ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq189_HTML.gif and C 6 = C 6 ( 2 M ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq190_HTML.gif according to Lemma 4.3. Then we can find Ψ ˜ 0 ( M ) > 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq191_HTML.gif sufficiently small such that
      ω ( M + 1 , 2 Ψ ˜ 0 ( M ) ) + Ψ ˜ 0 ( M ) δ 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ42_HTML.gif
      (4.17)
      and
      Ψ ˜ 0 ( M ) M 2 θ n + 4 ( 1 θ α ) 2 4 ( n + 2 ) 2 . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ43_HTML.gif
      (4.18)
      Given this, we can also find ρ 0 ( M ) ( 0 , 1 ] http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq192_HTML.gif so small that, writing
      C 7 ( M ) = C 6 ( 2 M ) θ 2 β θ 2 α , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equau_HTML.gif
      we have
      C 7 ( M ) ρ 0 ( M ) 2 β H 2 ( M ) min { δ 2 16 , Ψ ˜ 0 ( M ) , M 2 θ n + 4 ( 1 θ β ) 2 4 ( n + 2 ) 2 } . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ44_HTML.gif
      (4.19)
      Now, suppose that the conditions (i), (ii) and (iii) are satisfied on Q ρ ( z 0 ) Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq149_HTML.gif. Then, for j = 1 , 2 , 3 , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq193_HTML.gif , we shall show
      ( I ) j Ψ ˜ ( z 0 , θ j ρ , l z 0 , θ ρ ) θ 2 α j Ψ ˜ ( z 0 , ρ , l z 0 , ρ ) + C 7 ( M ) ( θ j ρ ) 2 β H 2 ( M ) , ( II ) j | l z 0 , θ j ρ ( z 0 ) | + | ( D l ) z 0 , θ j ρ | 2 M . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equav_HTML.gif
      Note first that ( I ) j http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq194_HTML.gif combined with (ii), (iii) and (4.19) yields
      ( I ) j Ψ ˜ ( θ j ρ ) 2 Ψ ˜ 0 ( M ) . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equaw_HTML.gif
      Moreover, we have ρ ρ 0 ( M ) 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq195_HTML.gif and | l z 0 , ρ | + | ( D l ) z 0 , ρ | M http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq196_HTML.gif. There we can apply Lemma 4.3 to conclude that ( I ) 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq197_HTML.gif holds. Furthermore, using Lemma 2.5, (iii) and (4.18), we deduce
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equax_HTML.gif
      i.e., ( II ) 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq198_HTML.gif holds. We now assume that ( I ) ι http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq199_HTML.gif and ( II ) ι http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq200_HTML.gif for ι = 1 , 2 , , j 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq201_HTML.gif hold. We can apply Lemma 4.3 to calculate
      Ψ ˜ ( θ j ρ ) θ 2 α j Ψ ˜ ( ρ ) + C 6 ( 2 M ) ( θ j ρ ) 2 β θ 2 β ι = 0 j 1 θ 2 ( α β ) ι θ 2 α j Ψ ˜ ( ρ ) + C 6 ( 2 M ) θ 2 β θ 2 α ( θ j ρ ) 2 β = θ 2 α j Ψ ˜ ( ρ ) + C 7 ( 2 M ) ( θ j ρ ) 2 β , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equay_HTML.gif
      showing ( I ) j http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq194_HTML.gif. To show ( II ) j http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq202_HTML.gif we estimate
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equaz_HTML.gif

      Here we have used in turn Lemma 2.5, the definition of Ψ ( θ ι 1 ρ ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq203_HTML.gif and ( I ) ι http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq199_HTML.gif for ι = 1 , 2 , , j 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq204_HTML.gif.

      Since | D l θ j ρ | 2 M http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq205_HTML.gif. We are in a position to apply Theorem 3.1. We obtain
      Φ ( θ j ρ / 2 , ( D u ) θ j ρ / 2 ) Φ ( θ j ρ , ( D u ) θ j ρ ) C cac ( 2 M ) Ψ ˜ ( θ j ρ ) C cac ( 2 M ) ( θ 2 α j Ψ ˜ ( ρ ) + C 7 ( 2 M ) ( θ j ρ ) 2 β ) . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ45_HTML.gif
      (4.20)
      We now consider 0 < r ρ / 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq206_HTML.gif. We fix k N { 0 } http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq207_HTML.gif with θ k + 1 ρ / 2 < r θ k ρ / 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq208_HTML.gif. Then the previous estimate implies
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equba_HTML.gif
      Next, we show that ( ( D u ) θ j ρ / 2 ) j N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq209_HTML.gif is a Cauchy sequence in R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq63_HTML.gif. For K > j http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq210_HTML.gif we deduce
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbb_HTML.gif
      This proves the claim. Therefore the limit Γ z 0 = lim j ( D u ) θ j ρ / 2 R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq211_HTML.gif exists and from the previous estimate, we infer (taking the limit k http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq212_HTML.gif)
      | ( D u ) θ j ρ / 2 Γ z 0 | C 8 ( M ) θ 2 α j Ψ ˜ ( ρ ) + ( θ j ρ ) 2 β . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbc_HTML.gif
      Combining this with (4.20), we arrive at
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbd_HTML.gif
      For 0 < r ρ / 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq206_HTML.gif, we find k N { 0 } http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq207_HTML.gif with θ k + 1 ρ / 2 < r < θ k ρ / 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq213_HTML.gif. Then the previous estimate implies
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Eqube_HTML.gif

      This proves the assertion of the lemma. □

      An immediate consequence of the previous lemma and of isomorphism theorem of Campanato-Da Prato [16] is the following result.

      Theorem 4.1 (Description of regularity points)

      Let u L m ( T , 0 ; W 1 , m ( Ω , R N ) ) L ( Q T ; R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq103_HTML.gif be a weak solution to the system (1.9) under the assumptions (H1)-(H3) and (H4′), (H5), and denote by Σ the singular set of u. Then Σ Σ 0 Σ 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq214_HTML.gif, where
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbf_HTML.gif
      and
      Σ 2 = { z 0 Q T : lim sup ρ 0 ( | ( u ) z 0 , ρ | + | ( D u ) z 0 , ρ | ) = } . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbg_HTML.gif

      At last, we have the following.

      Theorem 4.2 (Almost everywhere regularity)

      Let u L m ( T , 0 ; W 1 , m ( Ω , R N ) ) L ( Q T ; R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq103_HTML.gif be a weak solution to the system (1.9) under the assumptions (H1)-(H3) and (H4), (H5), and denote by Σ the singular set of u. Then Σ Σ 1 Σ 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq215_HTML.gif, where Σ 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq216_HTML.gif is as in Theorem  4.1 and
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbh_HTML.gif
      Proof We start taking a point z 0 ( x 0 , t 0 ) Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq217_HTML.gif such that
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ46_HTML.gif
      (4.21)
      and
      sup ρ > 0 | ( u ) z 0 , ρ | + sup ρ > 0 | ( D u ) z 0 , ρ | M < . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ47_HTML.gif
      (4.22)

      The proof is complete if we show that such points are regularity points.

      Step 1: a comparison estimate. Consider the unique weak solution v L m ( t 0 4 ρ 2 , t 0 ; W 1 , m ( B ρ ( x 0 ) , R N ) ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq218_HTML.gif of the initial boundary value problem
      { Q 2 ρ ( z 0 ) ( v i φ t i A i α ( z 0 , ( u ) z 0 , ρ , D v ) D α φ i ) d z = 0 , φ C 0 ( Q 2 ρ ( z 0 ) , R N ) , v = u , on  B 2 ρ ( x 0 ) × { t 0 4 ρ 2 } B 2 ρ ( x 0 ) × ( t 0 4 ρ 2 , t 0 ) . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbi_HTML.gif
      Then the difference u v http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq219_HTML.gif satisfies
      Q 2 ρ ( z 0 ) [ ( u v ) i φ t i ( A i α ( z , u , D u ) A i α ( z 0 , ( u ) z 0 , ρ , D v ) ) D α φ i ] d z + Q 2 ρ ( z 0 ) B i ( z , u , D u ) φ i d z = 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbj_HTML.gif
      for every φ C 0 ( Q 2 ρ ( z 0 ) , R N ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq220_HTML.gif. We now choose φ = χ ( t ) ( u v ) i http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq221_HTML.gif with χ 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq222_HTML.gif for ( , s ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq223_HTML.gif, χ 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq224_HTML.gif on ( s + ε , ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq225_HTML.gif, and χ ( t ) = ( s + ε t ) / ε http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq226_HTML.gif for s t s + ε http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq227_HTML.gif, where [ s , s + ε ] ( t 0 4 ρ 2 , t 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq228_HTML.gif. Then
      1 2 Q 2 ρ ( z 0 ) t ( | u v | 2 χ ) d z + 1 2 Q 2 ρ ( z 0 ) | u v | 2 t χ d z Q 2 ρ ( z 0 ) ( A i α ( z , u , D u ) A i α ( z 0 , ( u ) z 0 , ρ , D v ) ) ( D u i D v i ) χ d z + Q 2 ρ ( z 0 ) B i ( z , u , D u ) χ ( t ) ( u v ) i d z = 0 . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbk_HTML.gif
      Letting ε 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq229_HTML.gif, we easily obtain that for a.e. s ( t 0 4 ρ 2 , t 0 ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq230_HTML.gif
      1 2 u ( , s ) v ( , s ) L 2 ( B 2 ρ ( x 0 ) ) 2 + B 2 ρ ( x 0 ) × ( t 0 4 ρ 2 , t 0 ) ( A i α ( z 0 , ( u ) z 0 , ρ , D u ) A i α ( z 0 , ( u ) z 0 , ρ , D v ) ) D ( u v ) i χ d z = B 2 ρ ( x 0 ) × ( t 0 4 ρ 2 , s ) ( A i α ( z 0 , ( u ) z 0 , ρ , D u ) A i α ( z , u , D u ) ) D ( u v ) i d z + B 2 ρ ( x 0 ) × ( t 0 4 ρ 2 , s ) B i ( z , u , D u ) ( u v ) i d z = 0 . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbl_HTML.gif
      The second term of the left-hand side of the previous equation can be estimated by the use of monotonicity, i.e., (H3). We therefore obtain
      1 2 u ( , s ) v ( , s ) L 2 ( B 2 ρ ( x 0 ) ) 2 + λ ( 1 + | D v | 2 + | D u | 2 ) m 2 2 B 2 ρ ( x 0 ) × ( t 0 4 ρ 2 , s ) | D u D v | 2 d z B 2 ρ ( x 0 ) × ( t 0 4 ρ 2 , s ) ( A i α ( z 0 , ( u ) z 0 , ρ , D u ) A i α ( z , u , D u ) ) D ( u v ) i d z + B 2 ρ ( x 0 ) × ( t 0 4 ρ 2 , s ) B i ( z , u , D u ) ( u v ) i d z = I + II . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ48_HTML.gif
      (4.23)
      To estimate the right-hand side, we use (H4) which easily yields
      | A i α ( z 0 , ( u ) z 0 , ρ , D u ) A i α ( z , u , D u ) | L θ ( 2 | ( u ) z 0 , 2 ρ | + | u ( u ) z 0 , 2 ρ | , 4 ρ + | u ( u ) z 0 , 2 ρ | ) ( 1 + | D u | ) m 2 . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbm_HTML.gif
      Using the previous estimate, Young’s inequality and the fact that θ 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq231_HTML.gif, we have
      | I | λ 2 ( 1 + | D v | 2 + | D u | 2 ) m 2 2 B 2 ρ ( x 0 ) × ( t 0 4 ρ 2 , s ) | D u D v | 2 d z + 2 L 2 λ Q 2 ρ ( z 0 ) θ ( 2 | ( u ) z 0 , 2 ρ | + | u ( u ) z 0 , 2 ρ | , 4 ρ + | u ( u ) z 0 , 2 ρ | ) ( 1 + | D u | m ) d z . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbn_HTML.gif
      Having combined the previous estimate with (4.23), we arrive at
      1 2 u ( , s ) v ( , s ) L 2 ( B 2 ρ ( x 0 ) ) 2 + λ 2 ( 1 + | D v | 2 + | D u | 2 ) m 2 2 B 2 ρ ( x 0 ) × ( t 0 4 ρ 2 , s ) | D u D v | 2 d z 2 L 2 λ Q 2 ρ ( z 0 ) θ ( 2 | ( u ) z 0 , 2 ρ | + | u ( u ) z 0 , 2 ρ | , 4 ρ + | ( u ) z 0 , 2 ρ | ) ( 1 + | D u | m ) d z + II = 2 L 2 λ III + II . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ49_HTML.gif
      (4.24)

      We shall provide on estimate for III. We denote http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq232_HTML.gif , σ > 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq233_HTML.gif.

      If we let A t σ A t = { z Q 2 ρ ( z 0 ) : | u ( u ) z 0 , 2 ρ | t } http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq234_HTML.gif, then
      | A t | 1 t Q 2 ρ ( z 0 ) | u ( u ) z 0 , 2 ρ | d z | Q 2 ρ | t ε 2 ρ . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ50_HTML.gif
      (4.25)
      We now split III
      III = A t ( ) d z + Q 2 ρ ( z 0 ) A t ( ) d z = IV + V http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbo_HTML.gif
      and estimate IV and V. We have, using that θ 1 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq231_HTML.gif, (4.25) and (4.22)
      IV 2 m 1 Q 2 ρ ( z 0 ) | D u ( D u ) z 0 , 2 ρ | m d z + ( 1 + 2 m 1 | ( D u ) z 0 , 2 ρ | m ) | A t | 2 m 1 Q 2 ρ ( z 0 ) | D u ( D u ) z 0 , 2 ρ | m d z + 2 m 1 ( 1 + 2 M 2 ) | Q 2 ρ | t ε 2 ρ . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbp_HTML.gif
      From the definition of θ, we have
      V 4 K ( 2 M + t ) ( ρ + t ) β Q 2 ρ ( z 0 ) ( 1 + | D u | m ) d z . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbq_HTML.gif
      Noting that sup Q T u = V http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq235_HTML.gif, we have
      II 2 m 1 a V Q 2 ρ ( z 0 ) | D u ( D u ) z 0 , 2 ρ | m d z + Q 2 ρ ( z 0 ) | u v | 2 ρ 2 d z + ( 2 m 1 a M m + b ) α n ρ n + 2 ρ 2 . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbr_HTML.gif
      We now choose the parameter t carefully, i.e., t = ε 2 ρ http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq236_HTML.gif and let ε suitably small. Then connecting the previous estimates for II, III, IV and V to (4.24), we easily have the estimate we were interested in, that is,
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ51_HTML.gif
      (4.26)
      In particular, we see that
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ52_HTML.gif
      (4.27)
      We observe that, as a consequence of (4.21) and (4.22), we have that
      lim inf ρ 0 S ( ρ ) = 0 . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ53_HTML.gif
      (4.28)
      Step 2: A Poincare-type inequality. Let us define
      v ˜ = v ( D u ) z 0 , 2 ρ ( x x 0 ) . http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbs_HTML.gif
      Therefore v ˜ http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq237_HTML.gif solves
      Q ρ ( z 0 ) ( v ˜ i φ t i A ˜ i α ( D v ˜ ) D α φ i ) d z = 0 , φ C 0 ( Q 2 ρ ( z 0 ) , R N ) , http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbt_HTML.gif
      where A ˜ i α ( p ) = A i α ( z 0 , ( u ) z 0 , 2 ρ , ( D v ) z 0 , 2 ρ + p ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq238_HTML.gif for every p R n N http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq19_HTML.gif. From [17], Theorem 3.1, we conclude that v ˜ W 1 , 2 ( t 0 ρ 2 , t 0 ; W 1 , 2 ( B 2 ρ ( x 0 ) , R N ) ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq239_HTML.gif and that
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbu_HTML.gif
      In view of the previous estimate, using the Poincare inequality for v and (4.26), we find
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbv_HTML.gif

      where C = C ( n , λ , L ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq240_HTML.gif.

      Finally, by comparison, we get the Poincare inequality for u via (4.26) and the previous estimate
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equ54_HTML.gif
      (4.29)

      for a constant C = C ( n , λ , L ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq240_HTML.gif.

      Step 3: Conclusion. From the previous estimate and (4.28), the assertion readily follows. Indeed if z 0 Q T http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq241_HTML.gif satisfies (4.21) and (4.22), then we have
      http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_Equbw_HTML.gif

      therefore z 0 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq186_HTML.gif is a regular point in view of Theorem 4.1. □

      Declarations

      Acknowledgements

      Supported by the National Natural Science Foundation of China (Nos: 11201415, 11271305), the Natural Science Foundation of Fujian Province (2012J01027) and the Training Programme Foundation for Excellent Youth Researching Talents of Fujian’s Universities (JA12205).

      Authors’ Affiliations

      (1)
      School of Mathematics and Statistics, Minnan Normal University
      (2)
      School of Mathematical Science, Xiamen University

      References

      1. Wittle M: Computer simulations of an electrorheological fluid. J. Non-Newton. Fluid Mech. 1993, 37: 233–263.View Article
      2. Rajagopal KR, Ruzicka M: On the modeling of electrorheological materials. Mech. Res. Commun. 1996, 23: 401–407. 10.1016/0093-6413(96)00038-9View Article
      3. Rajagopal KR, Ruzicka M: Mathematical modeling of electrorheological materials. Contin. Mech. Thermodyn. 2001, 13: 59–78. 10.1007/s001610100034View Article
      4. Duzaar F, Mingione G: Second order parabolic systems, optimal regularity, and singular sets of solutions. Ann. Inst. Henri Poincaré, Anal. Non Linéaire 2005, 22: 705–751. 10.1016/j.anihpc.2004.10.011MathSciNetView Article
      5. Duzaar F, Grotowski JF: Optimal interior partial regularity for nonlinear elliptic systems: the method of A -harmonic approximation. Manuscr. Math. 2000, 103: 267–298. 10.1007/s002290070007MathSciNetView Article
      6. Chen S, Tan Z: The method of p -harmonic approximation and optimal interior partial regularity for energy minimizing p -harmonic maps under the controllable growth condition. Sci. China Ser. A 2007, 50(1):105–115.MathSciNetView Article
      7. Chen S, Tan Z: The method of A -harmonic approximation and optimal interior partial regularity for nonlinear elliptic systems under the controllable growth condition. J. Math. Anal. Appl. 2007, 335: 20–42. 10.1016/j.jmaa.2007.01.042MathSciNetView Article
      8. Chen S, Tan Z: Optimal interior partial regularity for nonlinear elliptic systems. Discrete Contin. Dyn. Syst., Ser. A 2010, 27(3):981–993.MathSciNetView Article
      9. Chen S, Tan Z:Optimal interior partial regularity for nonlinear elliptic systems for the case 1 < m < 2 http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq242_HTML.gif under natural growth condition. J. Inequal. Appl. 2010., 2010: Article ID 680714
      10. Zhong T: C 1 , α http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq243_HTML.gif-partial regularity of nonlinear parabolic systems. J. Partial Differ. Equ. 1992, 5: 23–34.
      11. De Giorgi, E: Frontiere orientate di misura minima, Sem. Scuola Normale Superiore Pisa (1960–1961)
      12. De Giorgi E: Un esempio di estremali discontinue per un problema variazionale di tipo ellittico. Boll. Unione Mat. Ital. 1968, 1(4):135–137.MathSciNet
      13. Simon L:Compact sets in the space L p ( 0 , T ; B ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq244_HTML.gif. Ann. Mat. Pura Appl. 1987, 146(4):65–96.MathSciNet
      14. Kronz M: Partial regularity results for minimizers of quasiconvex functionals of higher order. Ann. Inst. Henri Poincaré, Anal. Non Linéaire 2002, 19: 81–112. 10.1016/S0294-1449(01)00072-5MathSciNetView Article
      15. Campanato S:Equazioni paraboliche del secondo ordine e spazi L 2 , θ ( Ω , δ ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq245_HTML.gif. Ann. Mat. Pura Appl. 1996, 73(4):55–102.MathSciNet
      16. Da Prato G:Spazi L 2 , θ ( Ω , δ ) http://static-content.springer.com/image/art%3A10.1186%2F1687-2770-2013-152/MediaObjects/13661_2012_Article_416_IEq246_HTML.gif e loro proprieta. Ann. Mat. Pura Appl. 1965, 69(4):383–392.MathSciNetView Article
      17. Campanato S: On the nonlinear parabolic systems in divergence form Hölder continuity and partial Hölder continuity of the solutions. Ann. Mat. Pura Appl. 1984, 137(4):83–122.MathSciNetView Article

      Copyright

      © Chen and Tan; licensee Springer. 2013

      This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://​creativecommons.​org/​licenses/​by/​2.​0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.