Home Asymptotic analysis of fundamental solutions of hypoelliptic operators
Article Open Access

Asymptotic analysis of fundamental solutions of hypoelliptic operators

  • George Chkadua EMAIL logo and Eugene Shargorodsky
Published/Copyright: October 28, 2023
Become an author with De Gruyter Brill

Abstract

Asymptotic behavior at infinity is investigated for fundamental solutions of a hypoelliptic partial differential operator

𝐏 ⁢ ( i ⁢ ∂ x ) = ( P 1 ⁢ ( i ⁢ ∂ x ) ) m 1 ⁢ ⋯ ⁢ ( P l ⁢ ( i ⁢ ∂ x ) ) m l

with the characteristic polynomial that has real multiple zeros. Based on asymptotic expansions of fundamental solutions, asymptotic classes of functions are introduced and existence and uniqueness of solutions in those classes are established for the equation 𝐏 ⁢ ( i ⁢ ∂ x ) ⁢ u = f in ℝ n . The obtained results imply, in particular, a new uniqueness theorem for the classical Helmholtz equation.

1 Introduction

The famous radiation conditions

(1.1) ∂ r ⁡ u ± i ⁢ k ⁢ u = O ⁢ ( r 1 - n 2 - 1 )   as  ⁢ r = | x | → ∞ ,

derived by Sommerfeld [8] for the Helmholtz equation

( Δ + k 2 ) ⁢ u = 0   in  ⁢ ℝ n ,

were generalized in [11, 6] to the polymetaharmonic equation

(1.2) ( Δ + k 1 2 ) m 1 ⁢ ⋯ ⁢ ( Δ + k l 2 ) m l ⁢ u = 0   in  ⁢ ℝ n ,

where k 1 , … , k l ∈ ℝ are constants. Uniqueness of solutions for the polymetaharmonic equation is studied in [6] in the case where the characteristic polynomial has real multiple zeros.

A different generalization of (1.1) is developed in [9, 10] where radiation conditions for the hypoelliptic differential equation

(1.3) P ⁢ ( i ⁢ ∂ x ) ⁢ u = f   in  ⁢ ℝ n ,

are obtained in the case where the characteristic polynomial P ⁢ ( ξ ) has real simple zeros: ∇ ξ ⁡ P ⁢ ( ξ ) ≠ 0 at the real zeros of P ⁢ ( ξ ) . The results obtained in [9, 10] are applicable to a wide range of operators and remove the restriction of spherical symmetry that played an important role in works devoted to the Helmholtz and polymetaharmonic equations.

The main aim of this paper is to extend the above results to the hypoelliptic differential equation

(1.4) ( P 1 ⁢ ( i ⁢ ∂ x ) ) m 1 ⁢ ⋯ ⁢ ( P l ⁢ ( i ⁢ ∂ x ) ) m l ⁢ u = f   in  ⁢ ℝ n ,

which combines the presence of multiple zeros of the characteristic polynomial (cf. (1.2)) with the general geometry of the zero set similar to that of (1.3).

In Section 2, asymptotic behavior at infinity is investigated for fundamental solutions of (1.4). Based on the asymptotic expansions of fundamental solutions derived in this section, in Section 4 we introduce asymptotic classes of functions M m 1 , … , m l σ ⁢ ( 𝐏 ) and prove existence and uniqueness of solutions for (1.4) in these classes. This is preceded in Section 3 by a treatment of an important model case l = 1 :

( P ⁢ ( i ⁢ ∂ x ) ) m ⁢ u = 0   in  ⁢ ℝ n .

The leading term in the asymptotic expansion of a function in M m 1 , … , m l σ ⁢ ( 𝐏 ) behaves at infinity like O ⁢ ( r 1 - n 2 + m - 1 ) , where m = max ⁡ { m 1 , … , m l } (see Definition 4.1). So, the decay at infinity is r m - 1 times slower than in the case of simple zeros of the characteristic polynomial treated in [9, 10], i.e., in the case m = 1 (see (1.3)). Because of this, we need more detailed asymptotic information than in [9, 10] in order to be able to prove uniqueness of solutions of (1.4), and some of our results are new even for the case of simple zeros of the characteristic polynomial. The more detailed asymptotic information we need is captured by the definition of the class M m 1 , … , m l σ ⁢ ( 𝐏 ) . Our main uniqueness result for (1.4) implies a new uniqueness theorem for the classical Helmholtz equation in the asymptotic classes M m σ for all m (see Definition 3.1).

Section A contains several results from [10] that are used throughout the paper.

2 Asymptotic expansions of fundamental solutions of hypoelliptic differential equations

2.1 Notation

The set of natural numbers { 1 , 2 , 3 , … } is denoted by ℕ .

For an open set U ⊂ ℝ n , let C 0 ∞ ⁢ ( U ) denote the space of infinitely differentiable functions with compact support in U. Let 𝒮 ⁢ ( ℝ n ) be the Schwartz space of all rapidly decaying smooth functions and let 𝒮 ′ ⁢ ( ℝ n ) be the space of tempered distributions.

Let the direct and the inverse Fourier transforms ℱ ± 1 : 𝒮 ⁢ ( ℝ n ) → 𝒮 ⁢ ( ℝ n ) be defined as follows:

( ℱ ± 1 ⁢ ψ ) ⁢ ( ξ ) = 1 ( 2 ⁢ π ) n / 2 ⁢ ∫ ℝ n e ± i ⁢ ( ξ , x ) ⁢ ψ ⁢ ( x ) ⁢ 𝑑 x , ( ξ , x ) = ξ 1 ⁢ x 1 + ⋯ + ξ n ⁢ x n .

They are extended to 𝒮 ′ ⁢ ( ℝ n ) in the standard way:

〈 ℱ ± 1 ⁢ u , ψ 〉 := 〈 u , ℱ ± 1 ⁢ ψ 〉 , u ∈ 𝒮 ′ ⁢ ( ℝ n ) , ψ ∈ 𝒮 ⁢ ( ℝ n ) ,

where 〈 ⋅ , ⋅ 〉 denotes the duality between the spaces 𝒮 ′ ⁢ ( ℝ n ) and 𝒮 ⁢ ( ℝ n ) .

We denote the unit sphere in ℝ n by 𝕊 n - 1 .

Throughout the paper, we use the symbol “ ∼ ” to denote asymptotic expansion that can be differentiated infinitely many times.

2.2 Fundamental solutions

An operator 𝐏 ⁢ ( i ⁢ ∂ x ) is said to be hypoelliptic if any distributional solution of the equation 𝐏 ⁢ ( i ⁢ ∂ x ) ⁢ u = 0 in ℝ n is an infinitely differentiable function (see, e.g., [2, Chapter 4, Section 4.1] or [3, Chapter 11, Section 11.1]). The operator 𝐏 ⁢ ( i ⁢ ∂ x ) is hypoelliptic if and only if the complex roots z = ξ + i ⁢ τ of the characteristic polynomial 𝐏 ⁢ ( z ) satisfy the inequality

(2.1) | τ | > c 1 ⁢ | ξ | γ - c 2

with certain constants c 1 > 0 , c 2 > 0 , and 0 < γ ≤ 1 . We consider the following hypoelliptic differential equation:

(2.2) 𝐏 ⁢ ( i ⁢ ∂ x ) ⁢ u = f   in  ⁢ ℝ n ,

where

(2.3) 𝐏 ⁢ ( i ⁢ ∂ x ) := ( P 1 ⁢ ( i ⁢ ∂ x ) ) m 1 ⁢ ⋯ ⁢ ( P l ⁢ ( i ⁢ ∂ x ) ) m l ,

P j ⁢ ( i ⁢ ∂ x ) , j = 1 , … , l , are hypoelliptic differential operators with constant coefficients, and m j ∈ ℕ .

The characteristic polynomial of the hypoelliptic differential operator 𝐏 ⁢ ( i ⁢ ∂ x ) has the following form:

(2.4) 𝐏 ⁢ ( ξ ) = ( P 1 ⁢ ( ξ ) ) m 1 ⁢ ⋯ ⁢ ( P l ⁢ ( ξ ) ) m l ,

where P j ⁢ ( ξ ) , j = 1 , … , l , are the characteristic polynomials of the differential operators P j ⁢ ( i ⁢ ∂ x ) , j = 1 , … , l .

We assume that the following conditions are satisfied:

  1. ∇ ξ ⁢ ∏ j = 1 l P j ⁢ ( ξ ) ≠ 0 at the real zeros of P j ⁢ ( ξ ) , j = 1 , … , l .

  2. The set of real zeros of P j ⁢ ( ξ ) consists of infinitely differentiable connected surfaces S j ( 1 ) , … , S j ( ϰ j ) , ϰ j ∈ ℕ , of dimension n - 1 .

  3. The total curvature of the surfaces S j ( 1 ) , … , S j ( ϰ j ) , j = 1 , … , l , is not zero at any point.

It follows from (2.1) that the surfaces

S j ( 1 ) , … , S j ( ϰ j ) , j = 1 , … , l ,

are compact, and condition (i) implies that they have no boundary or common points. Condition (iii) implies that the surfaces S j ( 1 ) , … , S j ( ϰ j ) , j = 1 , … , l , are strictly convex (see [4, Chapter 7, Section 5, Theorem 5.6]). A class of such hypoelliptic differential operators with m j = 1 , j = 1 , … , l , was considered in [9, 10].

We will need the following result from [10, Chapter 7, Section 2].

Lemma 2.1.

If condition (i) is satisfied, then condition (ii) is equivalent to the following one:

  1. Each polynomial P j ⁢ ( Ξ ) can be represented in the form

    P j ⁢ ( Ξ ) = T j ⁢ ( Ξ ) ⁢ Q j ⁢ ( Ξ ) ,

    where T j ⁢ ( ξ ) is a hypoelliptic polynomial with real coefficients and ∇ ξ ⁢ ∏ j = 1 l T j ⁢ ( ξ ) ≠ 0 at the real zeros of T j ⁢ ( ξ ) , and Q j ⁢ ( ξ ) is a hypoelliptic polynomial that does not have real zeros.

Note that T j ⁢ ( Ξ ) = 0 defines the surfaces S j , where

S j := ⋃ ν = 1 ϰ j S j ( ν ) , j = 1 , … , l .

Below, “ p . v . ” will denote integrals in the sense of the Cauchy principal value. In particular,

(2.5) p . v . ∫ ℝ n ψ ⁢ ( ξ ) T j ⁢ ( ξ ) ⁢ 𝑑 ξ := lim ε → 0 + ⁡ ∫ | T j ⁢ ( ξ ) | > ε ψ ⁢ ( ξ ) T j ⁢ ( ξ ) ⁢ 𝑑 ξ , ψ ∈ 𝒮 ⁢ ( ℝ n ) .

Lemma 2.2.

Let Ό ∈ S ⁢ ( R ) and m ∈ N . Then

lim ε → 0 + ⁡ ∫ ℝ Φ ⁢ ( t ) ( t ± i ⁢ ε ) m ⁢ 𝑑 t = 1 ( m - 1 ) ! ⁢ p . v . ∫ ℝ D m - 1 ⁢ Φ ⁢ ( t ) t ⁢ 𝑑 t ∓ i ⁢ π ( m - 1 ) ! ⁢ D m - 1 ⁢ Φ ⁢ ( 0 ) ,

where

D m - 1 ⁢ Ό ⁢ ( t ) = d m - 1 d ⁢ t m - 1 ⁢ Ό ⁢ ( t ) .

Proof.

Integrating by parts, one gets

∫ ℝ Φ ⁢ ( t ) ( t ± i ⁢ ε ) m ⁢ 𝑑 t = 1 m - 1 ⁢ ∫ ℝ D ⁢ Φ ⁢ ( t ) ( t ± i ⁢ ε ) m - 1 ⁢ 𝑑 t
= 1 ( m - 1 ) ⁢ ( m - 2 ) ⁢ ∫ ℝ D 2 ⁢ Φ ⁢ ( t ) ( t ± i ⁢ ε ) m - 2 ⁢ 𝑑 t
⋮
= 1 ( m - 1 ) ! ⁢ ∫ ℝ D m - 1 ⁢ Φ ⁢ ( t ) t ± i ⁢ ε ⁢ 𝑑 t .

Then (see, e.g., [1, Chapter 1, Section 5.1])

lim ε → 0 + ⁡ ∫ ℝ Φ ⁢ ( t ) ( t ± i ⁢ ε ) m ⁢ 𝑑 t = lim ε → 0 + ⁡ 1 ( m - 1 ) ! ⁢ ∫ ℝ D m - 1 ⁢ Φ ⁢ ( t ) t ± i ⁢ ε ⁢ 𝑑 t
= 1 ( m - 1 ) ! ⁢ p . v . ∫ ℝ D m - 1 ⁢ Φ ⁢ ( t ) t ⁢ 𝑑 t ∓ i ⁢ π ( m - 1 ) ! ⁢ D m - 1 ⁢ Φ ⁢ ( 0 ) .

This concludes the proof. ∎

Let α j ( ν ) ∈ C 0 ∞ ⁢ ( ℝ n ) be such that α j ( ν ) = 1 on S j ( ν ) and supp ⁡ α j ( ν ) ∩ supp ⁡ α k ( υ ) = ∅ if j ≠ k or ν ≠ υ , j , k = 1 , … , l , ν = 1 , … , ϰ j , υ = 1 , … , ϰ k .

Let

σ = ( σ 1 ( 1 ) , … , σ 1 ( ϰ 1 ) , … , σ l ( 1 ) , … , σ l ( ϰ l ) )

be a vector whose components are equal to 1 or -1, i.e., σ j ( ν ) = ± 1 , j = 1 , … , l , ν = 1 , … , ϰ j .

Consider the following distribution:

(2.6)

〈 E m 1 , … , m I σ , ψ 〉 := lim ε → 0 + ⁡ ∫ ℝ n ψ ⁢ ( ξ ) ⁢ d ⁢ ξ ∏ j = 1 l ( T j ⁢ ( ξ ) + i ⁢ ε ⁢ ∑ ν = 1 ϰ j σ j ( ν ) ⁢ α j ( ν ) ⁢ ( ξ ) ) m j ⁢ ( Q j ⁢ ( ξ ) ) m j
= ∑ j = 1 l ∑ ν = 1 ϰ j lim ε → 0 + ⁡ ∫ ℝ n α j ( ν ) ⁢ ( ξ ) ⁢ ψ ⁢ ( ξ ) ⁢ d ⁢ ξ ∏ j = 1 l ( T j ⁢ ( ξ ) + i ⁢ ε ⁢ ∑ ν = 1 ϰ j σ j ( ν ) ⁢ α j ( ν ) ⁢ ( ξ ) ) m j ⁢ ( Q j ⁢ ( ξ ) ) m j
+ lim ε → 0 + ⁡ ∫ ℝ n ( 1 - ∑ j = 1 l ∑ ν = 1 ϰ j α j ( ν ) ⁢ ( ξ ) ) ⁢ ψ ⁢ ( ξ ) ⁢ d ⁢ ξ ∏ j = 1 l ( T j ⁢ ( ξ ) + i ⁢ ε ⁢ ∑ ν = 1 ϰ j σ j ( ν ) ⁢ α j ( ν ) ⁢ ( ξ ) ) m j ⁢ ( Q j ⁢ ( ξ ) ) m j

for all ψ ∈ 𝒮 ⁢ ( ℝ n ) .

It follows from the definition of ι j ( ν ) that

1 - ∑ j = 1 l ∑ ν = 1 ϰ j α j ( ν ) ⁢ ( ξ ) = ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) .

Therefore, (2.6) can be rewritten in the form

〈 E m 1 , … , m I σ , ψ 〉 = ∑ j = 1 l ∑ ν = 1 ϰ j lim ε → 0 + ⁡ ∫ ℝ n α j ( ν ) ⁢ ( ξ ) ⁢ ψ ⁢ ( ξ ) ⁢ d ⁢ ξ ∏ j = 1 l ( T j ⁢ ( ξ ) + i ⁢ ε ⁢ ∑ ν = 1 ϰ j σ j ( ν ) ⁢ α j ( ν ) ⁢ ( ξ ) ) m j ⁢ ( Q j ⁢ ( ξ ) ) m j
+ lim ε → 0 + ⁡ ∫ ℝ n ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) ⁢ ψ ⁢ ( ξ ) ⁢ d ⁢ ξ ∏ j = 1 l ( T j ⁢ ( ξ ) + i ⁢ ε ⁢ ∑ ν = 1 ϰ j σ j ( ν ) ⁢ α j ( ν ) ⁢ ( ξ ) ) m j ⁢ ( Q j ⁢ ( ξ ) ) m j .

Taking into account the definition of ι j ( ν ) , we obtain

〈 E m 1 , … , m I σ , ψ 〉 = ∑ j = 1 l ∑ ν = 1 ϰ j lim ε → 0 + ⁡ ∫ ℝ n α j ( ν ) ⁢ ( ξ ) ⁢ ψ ⁢ ( ξ ) ⁢ d ⁢ ξ ∏ j = 1 l ( T j ⁢ ( ξ ) + i ⁢ ε ⁢ σ j ( ν ) ⁢ α j ( ν ) ⁢ ( ξ ) ) m j ⁢ ( Q j ⁢ ( ξ ) ) m j + ∫ ℝ n ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) ⁢ ψ ⁢ ( ξ ) 𝐏 ⁢ ( ξ ) ⁢ d ⁢ ξ .

Let β j , 𝔦 ( ν ) , 𝔦 = 1 , … , N j ( ν ) , be a partition of unity in a neighborhood of the surface S j ( ν ) , i.e.,

β j , 𝔦 ( ν ) ∈ C 0 ∞ ⁢ ( U j , 𝔦 ( ν ) ) , 0 ≤ β j , 𝔦 ( ν ) ≤ 1 , ∑ 𝔦 = 1 N j ( ν ) β j , 𝔦 ( ν ) ⁢ ( ξ ) = 1 , ξ ∈ S j ( ν ) ,

where { U j , 𝔦 ( ν ) } 𝔦 = 1 N j ( ν ) is an open cover of the surface S j ( ν ) . Then we get

〈 E m 1 , … , m I σ , ψ 〉 = ∑ j = 1 l ∑ ν = 1 ϰ j ∑ 𝔦 = 1 N j ( ν ) lim ε → 0 + ⁡ ∫ ℝ n φ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ d ⁢ ξ ( T j ⁢ ( ξ ) + i ⁢ ε ⁢ σ j ( ν ) ⁢ α j ( ν ) ⁢ ( ξ ) ) m j
+ ∫ ℝ n ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) ⁢ ψ ⁢ ( ξ ) 𝐏 ⁢ ( ξ ) ⁢ d ⁢ ξ   for all  ⁢ ψ ∈ 𝒮 ⁢ ( ℝ n ) ,

where φ j , 𝔦 ( ν ) ⁢ ( ξ ) := θ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ ψ ⁢ ( ξ ) , φ j , 𝔦 ( ν ) ∈ C 0 ∞ ⁢ ( ℝ n ) , 𝔦 = 1 , … , N j ( ν ) , ν = 1 , … , ϰ j , j = 1 , … , l , and

(2.7) θ j , 𝔦 ( ν ) ⁢ ( ξ ) := β j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ α j ( ν ) ⁢ ( ξ ) ( P 1 ⁢ ( ξ ) ) m 1 ⁢ ⋯ ⁢ ( P j - 1 ⁢ ( ξ ) ) m j - 1 ⁢ ( P j + 1 ⁢ ( ξ ) ) m j + 1 ⁢ ⋯ ⁢ ( P l ⁢ ( ξ ) ) m l ⁢ ( Q j ⁢ ( ξ ) ) m j .

We choose a partition of unity such that, if supp ⁡ β j , 𝔦 ( ν ) ∩ S j ( ν ) ≠ ∅ , then ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) ≠ 0 on supp ⁡ β j , 𝔦 ( ν ) for some n 𝔦 . Then one can introduce the following local coordinates in a neighborhood of supp ⁡ β j , 𝔦 ( ν ) :

(2.8) t = T j ⁢ ( ξ ) , ξ n 𝔦 ′ = ξ n 𝔦 ′ ,

where

ξ n 𝔦 ′ := ( ξ 1 , … , ξ n 𝔦 - 1 , ξ n 𝔦 + 1 , … , ξ n ) ∈ ℝ n - 1 .

It follows from the implicit function theorem that (2.8) has a C ∞ smooth inverse

(2.9) ξ n 𝔦 = h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) , ξ n 𝔦 ′ = ξ n 𝔦 ′ .

Let

g j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) := ∏ q = 1 s ( ∂ t q ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) q ! ) k q ⁢ ∂ t m j - s ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ,
(2.10) f j , 𝔦 , k 1 , … , k s ( ν ) ( ξ ) := g j , 𝔦 , k 1 , … , k s ( ν ) ( ξ n 𝔦 ′ , t ) | t = T j ⁢ ( ξ ) .

Lemma 2.3.

The following equality holds for all ψ ∈ S ⁢ ( R n ) :

〈 E m 1 , … , m I σ , ψ 〉 = lim ε → 0 + ⁡ ∫ ℝ n ψ ⁢ ( ξ ) ⁢ d ⁢ ξ ∏ j = 1 l ( T j ⁢ ( ξ ) + i ⁢ ε ⁢ ∑ ν = 1 ϰ j σ j ( ν ) ⁢ α j ( ν ) ⁢ ( ξ ) ) m j ⁢ ( Q j ⁢ ( ξ ) ) m j
= ∑ j = 1 l 1 ( m j - 1 ) ! ⁢ ∑ ν = 1 ϰ j ∑ 𝔦 = 1 N j ( ν ) ∑ s = 0 m j - 1 ∑ k 1 , … , k s ( m j - 1 s ) ⁢ s ! k 1 ! ⁢ ⋯ ⁢ k s !
× [ p . v . ∫ ℝ n ∂ ξ n 𝔦 k ⁡ ( θ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ ψ ⁢ ( ξ ) ) ⁡ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) T j ⁢ ( ξ ) d ξ
- i π σ j ( ν ) ∫ S j ( ν ) ∂ ξ n 𝔦 k ⁡ ( θ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ ψ ⁢ ( ξ ) ) ⁡ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) | ∇ ⁡ T j ⁢ ( ξ ) | d S ] + ∫ ℝ n ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ( ξ ) ) ψ ⁢ ( ξ ) 𝐏 ⁢ ( ξ ) d ξ ,

where k = k 1 + ⋯ + k s and the sum is over all s-tuples of nonnegative integers ( k 1 , … , k s ) satisfying the constraint k 1 + 2 ⁢ k 2 + ⋯ + s ⁢ k s = s , and

( m j - 1 s ) = ( m j - 1 ) ! s ! ⁢ ( m j - 1 - s ) ! .

Proof.

Making the change of variable (2.8), one gets

lim ε → 0 + ⁡ ∫ ℝ n φ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ d ⁢ ξ ( T j ⁢ ( ξ ) + i ⁢ ε ⁢ σ j ( ν ) ⁢ α j ( ν ) ⁢ ( ξ ) ) m j = lim ε → 0 + ⁡ ∫ ℝ n φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ) ⁢ | ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) | ( t + i ⁢ ε ⁢ σ j ( ν ) ) m j ⁢ 𝑑 t ⁢ 𝑑 ξ n 𝔦 ′
= lim ε → 0 + ⁡ ∫ ℝ n - 1 { ∫ ℝ φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ) ⁢ | ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) | ( t + i ⁢ ε ⁢ σ j ( ν ) ) m j ⁢ 𝑑 t } ⁢ 𝑑 ξ n 𝔦 ′ .

Let Φ j , 𝔦 ( ν ) ∈ C 0 ∞ ⁢ ( ℝ n ) be defined as follows:

Φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) := φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ) ⁢ | ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) | .

It follows from Lemmas 2.2, A.1 and the dominated convergence theorem that

lim ε → 0 + ⁡ ∫ ℝ n φ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ d ⁢ ξ ( T j ⁢ ( ξ ) + i ⁢ ε ⁢ σ j ( ν ) ⁢ α j ( ν ) ⁢ ( ξ ) ) m j
= lim ε → 0 + ⁡ ∫ ℝ n - 1 ∫ ℝ Φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ( t + i ⁢ ε ⁢ σ j ( ν ) ) m j ⁢ 𝑑 t ⁢ 𝑑 ξ n 𝔦 ′
= 1 ( m j - 1 ) ! ⁢ ∫ ℝ n - 1 p . v . ∫ ℝ ∂ t m j - 1 ⁡ Φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) t ⁢ 𝑑 t ⁢ 𝑑 ξ n 𝔦 ′ - i ⁢ π ⁢ σ j ( ν ) ( m j - 1 ) ! ⁢ ∫ ℝ n - 1 ∂ t m j - 1 ⁡ Φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , 0 ) ⁢ 𝑑 ξ n 𝔦 ′ .

Further,

∂ t m j - 1 ⁡ Φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) = ∂ t m j - 1 ⁡ ( φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ) ⁢ | ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) | )
= ∑ s = 0 m j - 1 ( m j - 1 s ) ⁢ ∂ t s ⁡ φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ) ⁢ ∂ t m j - 1 - s ⁡ | ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) |
= ∑ s = 0 m j - 1 ( m j - 1 s ) ⁢ ∂ t s ⁡ φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ) ⁢ | ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) | ⁢ ∂ t m j - s ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) .

Using the FaĂ  di Bruno formula (see, e.g., [5, Theorem 1.3.2]), one gets

∂ t m j - 1 ⁡ Φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) = ∑ s = 0 m j - 1 ( m j - 1 s ) ⁢ { ∑ k 1 , … , k s s ! k 1 ! ⁢ ⋯ ⁢ k s ! ⁢ ∂ ξ n 𝔦 k ⁡ φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ) ⁢ ∏ q = 1 s ( ∂ t q ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) q ! ) k q }
× | ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) | ⁢ ∂ t m j - s ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ,

where k = k 1 + ⋯ + k s and the sum is over all s-tuples of nonnegative integers ( k 1 , … , k s ) satisfying the constraint k 1 + 2 ⁢ k 2 + ⋯ + s ⁢ k s = s . Hence,

(2.11)

lim ε → 0 + ⁡ ∫ ℝ n φ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ d ⁢ ξ ( T j ⁢ ( ξ ) + i ⁢ ε ⁢ σ j ( ν ) ⁢ α j ( ν ) ⁢ ( ξ ) ) m j
= 1 ( m j - 1 ) ! ⁢ ∑ s = 0 m j - 1 ∑ k 1 , … , k s ( m j - 1 s ) ⁢ s ! k 1 ! ⁢ ⋯ ⁢ k s !
× [ ∫ ℝ n - 1 p . v . ∫ ℝ 1 t ∂ ξ n 𝔦 k φ j , 𝔦 ( ν ) ( ξ n 𝔦 ′ , h j , 𝔦 ( ν ) ( ξ n 𝔦 ′ , t ) ) ∏ q = 1 s ( ∂ t q ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) q ! ) k q
× | ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) | ⁢ ∂ t m j - s ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ⁢ d ⁢ t ⁢ d ⁢ ξ n 𝔦 ′ - i ⁢ π ⁢ σ j ( ν ) ⁢ ∫ ℝ n - 1 ∂ ξ n 𝔦 k ⁡ φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , 0 ) )
× ∏ q = 1 s ( ∂ t q ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , 0 ) q ! ) k q | ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , 0 ) | ⁢ ∂ t m j - s ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , 0 ) ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , 0 ) d ξ n 𝔦 ′ ] .

Let dS be the surface area element of the surface T j ⁢ ( ξ ) = 0 , j = 1 , … , l :

d ⁢ S = 1 + | ∇ ξ n 𝔦 ′ ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , 0 ) | 2 ⁢ d ⁢ ξ n 𝔦 ′ .

Since

T j ⁢ ( ξ n 𝔦 ′ , h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , t ) ) ≡ t ,

we have

∂ ξ n 𝔦 ⁡ T j ⁢ ∂ t ⁡ h j , 𝔦 ( ν ) = 1 , ∂ ξ k ⁡ T j + ∂ ξ n 𝔦 ⁡ T j ⁢ ∂ ξ k ⁡ h j , 𝔦 ( ν ) = 0   for  ⁢ k ≠ n 𝔦 .

So,

d ⁢ S = 1 + ( ∂ ξ n 𝔦 ⁡ T j ) - 2 ⁢ | ∇ ξ n 𝔦 ′ ⁡ T j | 2 ⁢ d ⁢ ξ n 𝔦 ′ = | ∇ ξ ⁡ T j | ⁢ | ∂ t ⁡ h j , 𝔦 ( ν ) | ⁢ d ⁢ ξ n 𝔦 ′ ,
(2.12) d ⁢ ξ n 𝔦 ′ = d ⁢ S | ∇ ξ ⁡ T j | ⁢ | ∂ t ⁡ h j , 𝔦 ( ν ) | .

Therefore, substituting (2.12) in (2.11) and taking into account (2.10), we obtain in the original coordinate system

lim ε → 0 + ⁡ ∫ ℝ n φ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ d ⁢ ξ ( T j ⁢ ( ξ ) + i ⁢ ε ⁢ σ j ( ν ) ⁢ α j ( ν ) ⁢ ( ξ ) ) m j
= 1 ( m j - 1 ) ! ⁢ ∑ s = 0 m j - 1 ∑ k 1 , … , k s ( m j - 1 s ) ⁢ s ! k 1 ! ⁢ ⋯ ⁢ k s ! ⁢ p . v . ∫ ℝ n ∂ ξ n 𝔦 k ⁡ ( φ j , 𝔦 ( ν ) ⁢ ( ξ ) ) ⁡ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) T j ⁢ ( ξ ) ⁢ 𝑑 ξ
- i ⁢ π ⁢ σ j ( ν ) ( m j - 1 ) ! ⁢ ∑ s = 0 m j - 1 ∑ k 1 , … , k s ( m j - 1 s ) ⁢ s ! k 1 ! ⁢ ⋯ ⁢ k s !
× ∫ ℝ n - 1 ∂ ξ n 𝔦 k ⁡ ( φ j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , 0 ) ) ) ⁡ g j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ n 𝔦 ′ , 0 ) ⁢ | ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , 0 ) | ∂ t ⁡ h j , 𝔦 ( ν ) ⁢ ( ξ n 𝔦 ′ , 0 ) d ξ n 𝔦 ′
= 1 ( m j - 1 ) ! ⁢ ∑ s = 0 m j - 1 ∑ k 1 , … , k s ( m j - 1 s ) ⁢ s ! k 1 ! ⁢ ⋯ ⁢ k s ! ⁢ p . v . ∫ ℝ n ∂ ξ n 𝔦 k ⁡ ( φ j , 𝔦 ( ν ) ⁢ ( ξ ) ) ⁡ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) T j ⁢ ( ξ ) ⁢ 𝑑 ξ
- i ⁢ π ⁢ σ j ( ν ) ( m j - 1 ) ! ⁢ ∑ s = 0 m j - 1 ∑ k 1 , … , k s ( m j - 1 s ) ⁢ s ! k 1 ! ⁢ ⋯ ⁢ k s ! ⁢ ∫ S j ( ν ) ∂ ξ n 𝔦 k ⁡ ( φ j , 𝔦 ( ν ) ⁢ ( ξ ) ) ⁡ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) | ∇ ξ ⁡ T j ⁢ ( ξ ) | ⁢ 𝑑 S .

Hence,

〈 E m 1 , … , m I σ , ψ 〉 = ∑ j = 1 l 1 ( m j - 1 ) ! ⁢ ∑ ν = 1 ϰ j ∑ 𝔦 = 1 N j ( ν ) ∑ s = 0 m j - 1 ∑ k 1 , … , k s ( m j - 1 s ) ⁢ s ! k 1 ! ⁢ ⋯ ⁢ k s !
× [ p . v . ∫ ℝ n ∂ ξ n 𝔦 k ⁡ ( θ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ ψ ⁢ ( ξ ) ) ⁡ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) T j ⁢ ( ξ ) d ξ
- i π σ j ( ν ) ∫ S j ( ν ) ∂ ξ n 𝔦 k ⁡ ( θ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ ψ ⁢ ( ξ ) ) ⁡ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) | ∇ ξ ⁡ T j ⁢ ( ξ ) | d S ]
+ ∫ ℝ n ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) ⁢ ψ ⁢ ( ξ ) 𝐏 ⁢ ( ξ ) ⁢ d ⁢ ξ   for all  ⁢ ψ ∈ 𝒮 ⁢ ( ℝ n ) .

This concludes the proof. ∎

Theorem 2.4.

Let

𝐏 ⁢ ( i ⁢ ∂ x ) = ( P 1 ⁢ ( i ⁢ ∂ x ) ) m 1 ⁢ ⋯ ⁢ ( P l ⁢ ( i ⁢ ∂ x ) ) m l

be a hypoelliptic differential operator satisfying conditions (i)–(iii) above. Then

ℰ m 1 , … , m I σ = ( 2 ⁢ π ) - n / 2 ⁢ ℱ - 1 ⁢ ( E m 1 , … , m I σ ) ∈ 𝒮 ′ ⁢ ( ℝ n )

is a fundamental solution of the operator P ⁢ ( i ⁢ ∂ x ) , where F - 1 is the inverse Fourier transform.

Proof.

Using the definition of E m 1 , … , m I σ and Lebesgue’s dominated convergence theorem, one gets

〈 𝐏 ⁢ ( i ⁢ ∂ x ) ⁢ ℰ m 1 , … , m I σ , φ 〉 = 〈 ( P 1 ⁢ ( i ⁢ ∂ x ) ) m 1 ⁢ ⋯ ⁢ ( P l ⁢ ( i ⁢ ∂ x ) ) m l ⁢ ℰ m 1 , … , m I σ , φ 〉
= 1 ( 2 ⁢ π ) n / 2 ⁢ 〈 ( P 1 ⁢ ( ξ ) ) m 1 ⁢ ⋯ ⁢ ( P l ⁢ ( ξ ) ) m l ⁢ E m 1 , … , m I σ , ℱ - 1 ⁢ ( φ ) 〉
= 1 ( 2 ⁢ π ) n / 2 ⁢ 〈 E m 1 , … , m I σ , ( P 1 ⁢ ( ξ ) ) m 1 ⁢ ⋯ ⁢ ( P l ⁢ ( ξ ) ) m l ⁢ ℱ - 1 ⁢ ( φ ) 〉
= 1 ( 2 ⁢ π ) n / 2 ⁢ lim ε → 0 + ⁡ ∫ ℝ n ( P 1 ⁢ ( ξ ) ) m 1 ⁢ ⋯ ⁢ ( P l ⁢ ( ξ ) ) m l ⁢ ℱ - 1 ⁢ ( φ ) ⁢ ( ξ ) ⁢ d ⁢ ξ ∏ j = 1 l ( T j ⁢ ( ξ ) + i ⁢ ε ⁢ ∑ ν = 1 ϰ j σ j ( ν ) ⁢ α j ( ν ) ⁢ ( ξ ) ) m j ⁢ ( Q j ⁢ ( ξ ) ) m j
= 1 ( 2 ⁢ π ) n / 2 ⁢ ∫ ℝ n ℱ - 1 ⁢ ( φ ) ⁢ ( ξ ) ⁢ 𝑑 ξ
= φ ⁢ ( 0 )
= 〈 δ , φ 〉   for all  ⁢ φ ∈ 𝒮 ⁢ ( ℝ n ) .

This concludes the proof. ∎

Now we investigate asymptotic behavior at infinity of the above fundamental solutions of the operator 𝐏 ⁢ ( i ⁢ ∂ x ) . There are as many of those fundamental solutions as there are distinct vectors σ, that is, 2 ∑ j = 1 l ϰ j .

From conditions (i)–(iii), we have that S j ( ν ) , ν = 1 , … , ϰ j , j = 1 , … , l , are C ∞ smooth, compact ( n - 1 ) -dimen- sional surfaces in ℝ n without boundary. We specify an orientation on the surfaces S j ( ν ) by choosing the normal vector

n ⁢ ( ξ j ( ν ) ) = σ j ( ν ) ⁢ ∇ ⁡ T j ⁢ ( ξ j ( ν ) ) / | ∇ ⁡ T j ⁢ ( ξ j ( ν ) ) | , ξ j ( ν ) ∈ S j ( ν ) , ν = 1 , … , ϰ j , j = 1 , … , l .

It follows from the above conditions (i)–(iii) that for an arbitrary unit vector ω = x / r , r = | x | , there exists exactly one point ξ j ( ν ) = ξ j ( ν ) ⁢ ( ω ) ∈ S j ( ν ) , ν = 1 , … , ϰ j , j = 1 , … , l , such that the unit normal vector n ⁢ ( ξ j ( ν ) ) to S j ( ν ) at the point ξ j ( ν ) has the same direction as ω and exactly one point ξ j ( ν + ϰ j ) ⁢ ( ω ) ∈ S j ( ν ) , ν = 1 , … , ϰ j , at which these vectors have opposite directions (see [4, Chapter 7, Section 5, Theorem 5.6]).

Let

μ j ( ν ) ⁢ ( ω ) := ( ξ j ( ν ) ⁢ ( ω ) , ω ) ,

where ( ⋅ , ⋅ ) denotes the scalar product of two vectors.

Let 𝒦 j ( ν ) ⁢ ( ω ) denote the total curvature of the surface S j ( ν ) at the point ξ j ( ν ) ⁢ ( ω ) and suppose that δ j ( ν ) = ± 1 depending on whether the directions of the vector n ⁢ ( ξ j ( ν ) ) and the outward normal to S j ( ν ) are the same or opposite, ν = 1 , … , ϰ j , j = 1 , … , l .

The following theorem is our main result on the asymptotic behavior at infinity of fundamental solutions of 𝐏 ⁢ ( i ⁢ ∂ x ) .

Theorem 2.5.

Let the hypoelliptic differential operator

( P 1 ⁢ ( i ⁢ ∂ x ) ) m 1 ⁢ ⋯ ⁢ ( P l ⁢ ( i ⁢ ∂ x ) ) m l

satisfy conditions (i)–(iii). Then its fundamental solution constructed in Theorem 2.4 has the following uniform in ω ∈ S n - 1 asymptotic expansion as r = | x | → ∞ :

(2.13) ℰ m 1 , … , m I σ ⁢ ( x ) ∼ ∑ j = 1 l ∑ ν = 1 ϰ j e - i ⁢ μ j ( ν ) ⁢ ( ω ) ⁢ r ⁢ ∑ s = 0 m j - 1 ∑ k 1 , … , k s ∑ q = 0 k ∑ p = 0 ∞ b k 1 , … , k s ( ν ) , j , q , p ⁢ ( ω ) ⁢ r 1 - n 2 + k - p - q ,

where

b k 1 , … , k s ( ν ) , j , q , p ∈ C ∞ ⁢ ( 𝕊 n - 1 ) , k = k 1 + ⋯ + k s ,

and the sum is over all s-tuples of nonnegative integers ( k 1 , … , k s ) satisfying the constraint

k 1 + 2 ⁢ k 2 + ⋯ + s ⁢ k s = s ,

and

b k 1 , … , k s ( ν ) , j , q , 0 ⁢ ( ω ) = - i ⁢ σ j ( ν ) ⁢ ( 2 ⁢ π ) ( 1 - n ) / 2 ( m j - 1 - s ) ! ⁢ k 1 ! ⁢ ⋯ ⁢ k s ! ⁢ | 𝒦 j ( ν ) ⁢ ( ω ) | ⁢ ∑ 𝔦 = 1 N j ( ν ) ( k q )
× ( - i ⁢ ω n i ) k - q ⁢ ∂ ξ n 𝔦 q ⁡ θ j , 𝔦 ( ν ) ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) ⁢ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) | ∇ ⁡ T j ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) | ⁢ e i ⁢ δ j ( ν ) ⁢ ( n - 1 ) ⁢ π 4 .

Expansion (2.13) admits term by term differentiation any number of times with respect to x.

Proof.

Clearly, if the hypoelliptic operator 𝐏 ⁢ ( i ⁢ ∂ x ) satisfies conditions (i)–(iii), then, for sufficiently small δ > 0 and | t | ≤ δ , the operator

∏ j = 1 l ( T j ⁢ ( i ⁢ ∂ x ) - t ) m j ⁢ ( Q j ⁢ ( i ⁢ ∂ x ) ) m j

also satisfies these conditions. If S j ( ν ) ⁢ ( t ) , ν = 1 , … , ϰ j , are connected surfaces on which T j ⁢ ( ξ ) = t and ξ j ( ν ) ⁢ ( ω , t ) are points on S j ( ν ) ⁢ ( t ) , where the vector σ j ( ν ) ⁢ ∇ ⁡ T j ⁢ ( ξ ) has the same direction as ω for ν = 1 , … , ϰ j and the opposite direction for ν = ϰ j + 1 , … , 2 ⁢ ϰ j , then the points ξ j ( ν ) ⁢ ( ω , t ) , ν = 1 , … , 2 ⁢ ϰ j , depend smoothly on their arguments. Let μ j ( ν ) ⁢ ( ω , t ) = ( ξ j ( ν ) ⁢ ( ω , t ) , ω ) .

Let U j ( ν ) be the neighborhood of the surface S j ( ν ) and let

U j = ⋃ ν = 1 ϰ j U j ( ν )

be the neighborhood of the surface

S j = ⋃ ν = 1 ϰ j S j ( ν )

swept out by the surface

S j ⁢ ( t ) = ⋃ ν = 1 ϰ j S j ( ν ) ⁢ ( t ) , t ∈ ( - δ , δ ) .

Suppose that the α j ( ν ) introduced above satisfy the following conditions: α j ( ν ) ⁢ ( ξ ) = 0 for ξ ∉ U j , and α j ( ν ) ⁢ ( ξ ) = 1 if | T j ⁢ ( ξ ) | < δ / 2 . Then, for the fundamental solution of the differential operator (2.3) constructed in Lemma 2.3 and Theorem 2.4, the following equalities hold for all φ ∈ 𝒮 ⁢ ( ℝ n ) :

〈 ℰ m 1 , … , m I σ , φ 〉 = ( 2 ⁢ π ) - n / 2 ⁢ 〈 ℱ ξ → x - 1 ⁢ ( E m 1 , … , m I σ ) , φ 〉
= ( 2 ⁢ π ) - n / 2 ⁢ 〈 E m 1 , … , m I σ , ℱ ξ → x - 1 ⁢ ( φ ) 〉
= 1 ( 2 ⁢ π ) n / 2 ⁢ ∑ j = 1 l 1 ( m j - 1 ) ! ⁢ ∑ ν = 1 ϰ j ∑ 𝔦 = 1 N j ( ν ) ∑ s = 0 m j - 1 ∑ k 1 , … , k s ( m j - 1 s ) ⁢ s ! k 1 ! ⁢ ⋯ ⁢ k s !
× [ p . v . ∫ ℝ n ∂ ξ n 𝔦 k ⁡ ( θ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ ℱ ξ → x - 1 ⁢ ( φ ) ⁢ ( ξ ) ) ⁡ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) T j ⁢ ( ξ ) d ξ
- i π σ j ( ν ) ∫ S j ( ν ) ∂ ξ n 𝔦 k ⁡ ( θ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ ℱ ξ → x - 1 ⁢ ( φ ) ⁢ ( ξ ) ) ⁡ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) | ∇ ξ ⁡ T j ⁢ ( ξ ) | d S ]
+ 1 ( 2 ⁢ π ) n / 2 ⁢ ∫ ℝ n ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) ⁢ ℱ ξ → x - 1 ⁢ ( φ ) ⁢ ( ξ ) 𝐏 ⁢ ( ξ ) ⁢ d ⁢ ξ
= 1 ( 2 ⁢ π ) n ⁢ ∑ j = 1 l 1 ( m j - 1 ) ! ⁢ ∑ ν = 1 ϰ j ∑ 𝔦 = 1 N j ( ν ) ∑ s = 0 m j - 1 ∑ k 1 , … , k s ( m j - 1 s ) ⁢ s ! k 1 ! ⁢ ⋯ ⁢ k s !
× ∑ q = 0 k ( k q ) ∫ ℝ n ( - i x n 𝔦 ) k - q [ p . v . ∫ U j ( ∂ ξ n 𝔦 q ⁡ θ j , 𝔦 ( ν ) ⁢ ( ξ ) ) ⁢ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) T j ⁢ ( ξ ) e - i ⁢ ( x , ξ ) d ξ
- i π σ j ( ν ) ∫ S j ( ν ) ( ∂ ξ n 𝔦 q ⁡ θ j , 𝔦 ( ν ) ⁢ ( ξ ) ) ⁢ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) | ∇ ⁡ T j ⁢ ( ξ ) | e - i ⁢ ( x , ξ ) d S ] φ ( x ) d x
(2.14) + 1 ( 2 ⁢ π ) n ⁢ ∫ ℝ n [ ∫ ℝ n ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) ⁢ 1 𝐏 ⁢ ( ξ ) ⁢ e - i ⁢ ( x , ξ ) ⁢ d ⁢ ξ ] ⁢ φ ⁢ ( x ) ⁢ 𝑑 x .

The interchange of the inverse Fourier transform with the Cauchy principal value integral above can be easily justified with the help of Fubini’s theorem, Lemma A.1 and the Lebesgue dominated convergence theorem. It follows from (2.14) that the fundamental solution of the differential operator (2.3) can be written in the form

(2.15)

ℰ m 1 , … , m I σ ⁢ ( x ) = 1 ( 2 ⁢ π ) n ⁢ ∑ j = 1 l 1 ( m j - 1 ) ! ⁢ ∑ ν = 1 ϰ j ∑ 𝔦 = 1 N j ( ν ) ∑ s = 0 m j - 1 ∑ k 1 , … , k s ( m j - 1 s ) ⁢ s ! k 1 ! ⁢ ⋯ ⁢ k s !
× ∑ q = 0 k ( k q ) ( - i x n 𝔦 ) k - q [ p . v . ∫ U j ∂ ξ n 𝔦 q ⁡ θ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) T j ⁢ ( ξ ) e - i ⁢ ( x , ξ ) d ξ
- i π σ j ( ν ) ∫ S j ( ν ) ∂ ξ n 𝔦 q ⁡ θ j , 𝔦 ( ν ) ⁢ ( ξ ) ⁢ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) | ∇ ⁡ T j ⁢ ( ξ ) | e - i ⁢ ( x , ξ ) d S ]
+ 1 ( 2 ⁢ π ) n / 2 ⁢ ℱ ξ → x - 1 ⁢ [ ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) ⁢ 1 𝐏 ⁢ ( ξ ) ] .

We write the integrals over U j as repeated integrals, integrating first over the surface S j ⁢ ( t ) and then with respect to t (see (2.8) and (2.5)). From (2.9) and (2.12), we get

d ⁢ ξ = | ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) | ⁢ d ⁢ S ⁢ d ⁢ t | ∇ ⁡ T j ⁢ ( ξ ) | ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) .

Let

Ψ j , k 1 , … , k s ( ν ) , 𝔦 , q ⁢ ( t , ω , r ) := - i ⁢ ( 2 ⁢ π ) 1 - n ⁢ σ j ( ν ) ⁢ ∫ S j ( ν ) ⁢ ( t ) ( ∂ ξ n 𝔦 q ⁡ θ j , 𝔦 ( ν ) ⁢ ( ξ ) ) ⁢ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ ) | ∇ ⁡ T j ⁢ ( ξ ) | ⁢ e - i ⁢ ( x , ξ ) ⁢ 𝑑 S .

Then (2.15) can be rewritten in the following form:

(2.16)

ℰ m 1 , … , m I σ ⁢ ( x ) = ∑ j = 1 l 1 2 ⁢ ( m j - 1 ) ! ⁢ ∑ ν = 1 ϰ j ∑ 𝔦 = 1 N j ( ν ) ∑ s = 0 m j - 1 ∑ k 1 , … , k s ( m j - 1 s ) ⁢ s ! k 1 ! ⁢ ⋯ ⁢ k s !
× ∑ q = 0 k ( k q ) ( - i x n 𝔦 ) k - q [ - σ j ( ν ) i ⁢ π p . v . ∫ - δ δ Ψ j , k 1 , … , k s ( ν ) , 𝔦 , q ⁢ ( t , ω , r ) t d t + Ψ j , k 1 , … , k s ( ν ) , 𝔦 , q ( 0 , ω , r ) ]
+ 1 ( 2 ⁢ π ) n / 2 ⁢ ℱ ξ → x - 1 ⁢ [ ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) ⁢ 1 𝐏 ⁢ ( ξ ) ] .

The function Ψ j , k 1 , … , k s ( ν ) , 𝔦 , q ⁢ ( t , ω , r ) admits the following asymptotic expansion as r → ∞ (see Theorem A.3):

(2.17) Ψ j , k 1 , … , k s ( ν ) , 𝔦 , q ⁢ ( t , ω , r ) ∼ e - i ⁢ μ j ( ν ) ⁢ ( ω , t ) ⁢ r ⁢ ∑ p = 0 ∞ a k 1 , … , k s ( ν ) , 𝔦 , j , q , p ⁢ ( ω , t ) ⁢ r 1 - n 2 - p + e - i ⁢ μ j ( ν + ϰ j ) ⁢ ( ω , t ) ⁢ r ⁢ ∑ p = 0 ∞ a k 1 , … , k s ( ν + ϰ j ) , 𝔦 , j , q , p ⁢ ( ω , t ) ⁢ r 1 - n 2 - p ,

where a k 1 , … , k s ( ν ) , 𝔦 , j , q , p , a k 1 , … , k s ( ν + ϰ j ) , 𝔦 , j , q , p are infinitely differentiable functions of ω and t.

We have

a k 1 , … , k s ( ν ) , 𝔦 , j , q , 0 ⁢ ( ω , 0 ) = - i ⁢ σ j ( ν ) ⁢ ( 2 ⁢ π ) 1 - n 2 | 𝒦 j ( ν ) ⁢ ( ω ) | ⁢ ( ∂ ξ n 𝔦 q ⁡ θ j , 𝔦 ( ν ) ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) ) ⁢ f j , 𝔦 , k 1 , … , k s ( ν ) ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) | ∇ ⁡ T j ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) | ⁢ e - i ⁢ γ j ( ν ) ⁢ π 4 ,

where γ j ( ν ) is the difference between the numbers of positive and negative principal curvatures of the surface S j ( ν ) at the point ξ j ( ν ) ⁢ ( ω ) , ν = 1 , … , ϰ j , j = 1 , … , l . Since the surface S j ( ν ) is strictly convex, the principal curvatures have the same sign, and γ j ( ν ) = - δ j ( ν ) ⁢ ( n - 1 ) .

Substituting (2.17) into the right-hand side of (2.16) and applying Lemma A.2 and Lemma A.1, one gets

(2.18)

ℰ m 1 , … , m I σ ⁢ ( x ) - 1 ( 2 ⁢ π ) n / 2 ⁢ ℱ ξ → x - 1 ⁢ [ ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) ⁢ 1 𝐏 ⁢ ( ξ ) ]
∼ ∑ j = 1 l ∑ ν = 1 2 ⁢ ϰ j d j ( ν ) ⁢ e - i ⁢ μ j ( ν ) ⁢ ( ω , 0 ) ⁢ r ⁢ ∑ s = 0 m j - 1 ∑ k 1 , … , k s ∑ q = 0 k ∑ p = 0 ∞ b k 1 , … , k s ( ν ) , j , q , p ⁢ ( ω , 0 ) ⁢ r 1 - n 2 + k - p - q ,

where

b k 1 , … , k s ( ν ) , j , q , p ⁢ ( ω , 0 ) = 1 ( m j - 1 ) ! ⁢ ( m j - 1 s ) ⁢ s ! k 1 ! ⁢ ⋯ ⁢ k s ! ⁢ ∑ 𝔦 = 1 N j ( ν ) ( k q ) ⁢ ( - i ⁢ ω n 𝔦 ) k - q ⁢ a k 1 , … , k s ( ν ) , 𝔦 , j , q , p ⁢ ( ω , 0 ) ,
d j ( ν ) = 1 2 ⁢ ( 1 + σ ~ j ( ν ) ⁢ sign ⁢ ∂ t ⁡ μ j ( ν ) ⁢ ( ω , 0 ) ) .

Here, σ ~ j ( ν ) = σ j ( ν ) if ν = 1 , … , ϰ j , and σ ~ j ( ν ) = σ j ( ν - ϰ j ) if ν = ϰ j + 1 , … , 2 ⁢ ϰ j .

It is easy to see that d j ( ν ) = 1 if ν = 1 , … , ϰ j , and d j ( ν ) = 0 if ν = ϰ j + 1 , … , 2 ⁢ ϰ j . Indeed, since T j ⁢ ( ξ j ( ν ) ⁢ ( ω , t ) ) ≡ t , differentiating this identity with respect to t, one gets

(2.19) ( ∇ ⁡ T j ⁢ ( ξ j ( ν ) ⁢ ( ω , t ) ) , ∂ t ⁡ ξ j ( ν ) ⁢ ( ω , t ) ) = 1 .

The first factor in this scalar product is proportional to ω. More precisely, it follows from the definition of the points ξ j ( ν ) ⁢ ( ω , t ) that

(2.20) σ ~ j ( ν ) ⁢ ∇ ⁡ T j ⁢ ( ξ j ( ν ) ⁢ ( ω , t ) ) = c j ( ν ) ⁢ ( t ) ⁢ ω , ν = 1 , … , 2 ⁢ ϰ j ,

where c j ( ν ) ⁢ ( t ) > 0 if ν = 1 , … , ϰ j , and c j ( ν ) ⁢ ( t ) < 0 if ν = ϰ j + 1 , … , 2 ⁢ ϰ j .

Substituting equality (2.20) into (2.19), one obtains

(2.21) ( ∂ t ⁡ ξ j ( ν ) ⁢ ( ω , t ) , ω ) = σ ~ j ( ν ) ⁢ ( c j ( ν ) ⁢ ( t ) ) - 1 .

Since the left-hand side of (2.21) is equal to ∂ t ⁡ μ j ( ν ) ⁢ ( ω , t ) , one gets

d j ( ν ) = 1   if  ⁢ ν = 1 , … , ϰ j ,

and

d j ( ν ) = 0   if  ⁢ ν = ϰ j + 1 , … , 2 ⁢ ϰ j .

The assertion of Theorem 2.5 follows from (2.18) if we show that the second term on the left-hand side of (2.18) and all its derivatives decay faster than r - M for any M as r → ∞ .

Now, let

K m 1 , … , m I σ ⁢ ( x ) := ℱ ξ → x - 1 ⁢ [ ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) ⁢ 1 𝐏 ⁢ ( ξ ) ] .

Then

ℱ ξ → x - 1 ⁢ [ ∂ ξ β ⁡ ( ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) ⁢ 1 𝐏 ⁢ ( ξ ) ) ] = ( i ⁢ x ) β ⁢ K m 1 , … , m I σ ⁢ ( x )

for all multi-indices β = ( β 1 , … , β n ) . Since

| ∂ ξ β ⁡ ( ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) ⁢ 1 𝐏 ⁢ ( ξ ) ) | ≤ c β ( 1 + | ξ | ) 2 ⁢ m + | β | , c β > 0 ,

the function

∂ β ⁡ ( ∏ j = 1 l ∏ ν = 1 ϰ j ( 1 - α j ( ν ) ⁢ ( ξ ) ) ⁢ 1 𝐏 ⁢ ( ξ ) )

is summable for sufficiently large | β | > n - 2 ⁢ m . Therefore, its inverse Fourier transform is bounded and

| K m 1 , … , m I σ ⁢ ( x ) | ≤ c r M   for all  ⁢ M > n - 2 ⁢ m .

The derivatives of K m 1 , … , m I σ can be estimated similarly. ∎

Remark 2.6.

The asymptotic expansion of the fundamental solution ℰ m 1 , … , m I σ ⁢ ( x ) can be rewritten in the following form:

(2.22) ℰ m 1 , … , m I σ ⁢ ( x ) ∼ ∑ j = 1 l ∑ ν = 1 ϰ j e - i ⁢ μ j ( ν ) ⁢ ( ω ) ⁢ r ⁢ { a 0 , m j ( ν ) , j ⁢ ( ω ) ⁢ r 1 - n 2 + m j - 1 + ∑ q = 1 ∞ a q , m j ( ν ) , j ⁢ ( ω ) ⁢ r 1 - n 2 + m j - 1 - q }   as  ⁢ r → ∞ ,

where a q , m j ( ν ) , j ∈ C ∞ , j = 1 , … , l , ν = 1 , … , ϰ j , q = 0 , 1 , … ,

a 0 , m j ( ν ) , j ⁢ ( ω ) = - i ⁢ σ j ( ν ) ⁢ ( 2 ⁢ π ) ( 1 - n ) / 2 ( m j - 1 ) ! ⁢ | 𝒦 j ( ν ) ⁢ ( ω ) | ⁢ ∑ 𝔦 = 1 N j ( ν ) ( - i ⁢ ω n 𝔦 ) m j - 1 ⁢ θ j , 𝔦 ( ν ) ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) ⁢ f j , 𝔦 ( ν ) ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) ⁢ ∂ ξ n 𝔦 ⁡ T j ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) | ∇ ⁡ T j ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) | ⁢ e i ⁢ δ j ( ν ) ⁢ ( n - 1 ) ⁢ π 4 ,

θ j , 𝔦 ( ν ) is defined by (2.7), and (see (2.9))

f j , 𝔦 ( ν ) ( ξ ) = ( ∂ t h j , 𝔦 ( ν ) ( ξ n 𝔦 ′ , t ) ) m j | t = T j ⁢ ( ξ ) .

Expansion (2.22) admits term by term differentiation with respect to x any number of times.

3 Existence and uniqueness of solutions for a model hypoelliptic equation

Let us consider the following hypoelliptic differential equation:

(3.1) ( P ⁢ ( i ⁢ ∂ x ) ) m ⁢ u = f   in  ⁢ ℝ n ,

where m ∈ ℕ . We assume that the characteristic polynomial P ⁢ ( ξ ) of the differential operator P ⁢ ( i ⁢ ∂ x ) satisfies the following conditions (see [10, Chapter 7]):

  1. ∇ ξ ⁡ P ⁢ ( ξ ) ≠ 0 at the real zeros of P ⁢ ( ξ ) .

  2. The set of real zeros of P ⁢ ( ξ ) consists of infinitely differentiable connected surfaces S ( ν ) , ν = 1 , … , ϰ , of dimension n - 1 .

  3. The total curvature of the surfaces S ( ν ) , ν = 1 , … , ϰ , is not zero at any point.

Let σ = ( σ 1 , … , σ ϰ ) be a vector with components equal to 1 or -1, i.e., σ ν = ± 1 , ν = 1 , … , ϰ . It follows from Remark 2.6 that the fundamental solution of the operator ( P ⁢ ( i ⁢ ∂ x ) ) m has the following uniform in ω ∈ 𝕊 n - 1 asymptotic expansion as r → ∞ :

(3.2) ℰ m σ ⁢ ( x ) ∼ ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ { a 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 + ∑ q = 1 ∞ a q , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 - q } ,

where a q , m ( ν ) ∈ C ∞ , ν = 1 , … , ϰ , q = 0 , 1 , … .

Definition 3.1.

We say that a distribution u belongs to the class M m σ ⁢ ( P ) for m ∈ ℕ if it has the following uniform in ω = x / | x | asymptotic expansion as r = | x | → ∞ :

(3.3) u ∼ ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ { d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 + ∑ q = 1 ∞ d q , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 - q } ,

where d q , m ( ν ) ∈ C ∞ , ν = 1 , … , ϰ , q = 0 , 1 , … , and this expansion admits term by term differentiation any number of times with respect to x.

Definition 3.2.

A distribution u belongs to the Sommerfeld type radiation class Som σ ⁥ ( P ) if it can be represented as a sum

u ⁢ ( x ) = ∑ ν = 1 ϰ u ν ⁢ ( x ) ,

where

u ν ⁢ ( x ) = O ⁢ ( r 1 - n 2 ) as  ⁢ r → ∞ ,
∂ r ⁡ u ν ⁢ ( x ) + i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ u ν ⁢ ( x ) = O ⁢ ( r 1 - n 2 - 1 ) as  ⁢ r → ∞ .

It is clear that M 1 σ ⁢ ( P ) ⊂ Som σ ⁡ ( P ) .

Lemma 3.3.

Let Q be any polynomial. Then

Q ( i ∂ x ) ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ) = Q ( ξ ( ν ) ( ω ) ) e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + i 2 ∑ l , q = 1 n Q ( l , q ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) q e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + O ( r - 2 )   as  r → ∞ ,

where

Q ( l , q ) ⁢ ( ξ ( ν ) ⁢ ( ω ) ) = ∂ ξ l ⁡ ∂ ξ q ⁡ Q ⁢ ( ξ ( ν ) ⁢ ( ω ) ) .

Proof.

By definition, μ ( ν ) ⁢ ( ω ) = ( ξ ( ν ) ⁢ ( ω ) , ω ) . Hence,

μ ( ν ) ⁢ ( ω ) ⁢ r = ( ξ ( ν ) ⁢ ( ω ) , ω ) ⁢ r = ( ξ ( ν ) ⁢ ( ω ) , x )

and

∂ x l ⁡ ( μ ( ν ) ⁢ ( ω ) ⁢ r ) = ( ξ ( ν ) ⁢ ( ω ) ) l + ( ∂ x l ⁡ ξ ( ν ) ⁢ ( ω ) , x ) .

The vector ∂ x l ⁡ ξ ( ν ) ⁢ ( ω ) is tangent to S ( ν ) at a point where the normal vector to S ( ν ) has the same direction as ω, i.e., as x. Therefore,

∂ x l ⁡ ( μ ( ν ) ⁢ ( ω ) ⁢ r ) = ( ξ ( ν ) ⁢ ( ω ) ) l

and

∂ x q ( ξ ( ν ) ( ω ) ) l = ∂ x q ∂ x l ( μ ( ν ) ( ω ) r ) = ∂ x l ∂ x q ( μ ( ν ) ( ω ) r ) = ∂ x l ( ξ ( ν ) ( ω ) ) q .

The rest of the proof is by induction on deg ⁡ Q ⁢ ( ξ ) . It is clear that the lemma holds for deg ⁡ Q ⁢ ( ξ ) ≤ 2 . Suppose now it holds for any polynomial Q ⁢ ( ξ ) with deg ⁡ Q ⁢ ( ξ ) ≤ N and prove it for deg ⁡ Q ⁢ ( ξ ) = N + 1 .

Any polynomial Q ⁢ ( Ξ ) of degree N + 1 can be represented in the form

Q ⁢ ( ξ ) = Q 0 ⁢ ( ξ ) + ∑ p = 1 n Q p ⁢ ( ξ ) ⁢ ξ p ,

where deg ⁡ Q 0 ⁢ ( ξ ) = N , deg ⁡ Q p ⁢ ( ξ ) ≤ N , p = 0 , 1 , … , n . Then, using the induction assumption, one gets

Q ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ) = Q 0 ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ) + ∑ p = 1 n Q p ⁢ ( i ⁢ ∂ x ) ⁢ i ⁢ ∂ x p ⁡ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r
= Q 0 ( ξ ( ν ) ( ω ) ) e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + i 2 ∑ l , q = 1 n Q 0 ( l , q ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) q e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + O ( r - 2 )
+ ∑ p = 1 n Q p ⁢ ( i ⁢ ∂ x ) ⁢ ( ( ξ ( ν ) ⁢ ( ω ) ) p ⁢ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r )   as  ⁢ r → ∞ .

Hence, applying the formula (see, e.g., [7, Chapter 1, Section 1.7])

(3.4) Q p ⁢ ( i ⁢ ∂ x ) ⁢ ( u ⁢ v ) = ∑ | α | ≤ N 1 α ! ⁢ Q p ( α ) ⁢ ( i ⁢ ∂ x ) ⁢ u ⁢ ( i ⁢ ∂ x ) α ⁢ v ,

where Q p ( α ) ⁢ ( i ⁢ ∂ x ) is the differential operator of order N - | α | with the symbol

Q p ( α ) ⁢ ( ξ ) = ∂ ξ α ⁡ Q p ⁢ ( ξ ) ,

one obtains

Q p ⁢ ( i ⁢ ∂ x ) ⁢ ( ( ξ ( ν ) ⁢ ( ω ) ) p ⁢ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r )
= ( ξ ( ν ) ( ω ) ) p ( Q p ( ξ ( ν ) ( ω ) ) e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + i 2 ∑ l , q = 1 n Q p ( l , q ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) q e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r )
+ i ∑ l = 1 n ( Q p ( l ) ( i ∂ x ) e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ) ∂ x l ( ξ ( ν ) ( ω ) ) p + O ( r - 2 )
= ( ξ ( ν ) ( ω ) ) p Q p ( ξ ( ν ) ( ω ) ) e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + i 2 ∑ l , q = 1 n Q p ( l , q ) ( ξ ( ν ) ( ω ) ) ( ξ ( ν ) ( ω ) ) p ∂ x l ( ξ ( ν ) ( ω ) ) q e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r
+ i ∑ l = 1 n Q p ( l ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) p e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + O ( r - 2 )   as  r → ∞ ,

and

Q ( l ) ⁢ ( ξ ( ν ) ⁢ ( ω ) ) := ∂ ξ l ⁡ Q ⁢ ( ξ ( ν ) ⁢ ( ω ) ) .

Hence,

Q ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ) = ( Q 0 ⁢ ( ξ ( ν ) ⁢ ( ω ) ) + ∑ p = 1 n Q p ⁢ ( ξ ( ν ) ⁢ ( ω ) ) ⁢ ( ξ ( ν ) ⁢ ( ω ) ) p ) ⁢ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r
+ i 2 ∑ l , q = 1 n Q 0 ( l , q ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) q e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r
+ i 2 ∑ l , q , p = 1 n Q p ( l , q ) ( ξ ( ν ) ( ω ) ) ( ξ ( ν ) ( ω ) ) p ∂ x l ( ξ ( ν ) ( ω ) ) q e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r
+ i ∑ l , p = 1 n Q p ( l ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) p e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + O ( r - 2 )
= Q ( ξ ( ν ) ( ω ) ) e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + i 2 ∑ l , q = 1 n Q ( l , q ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) q e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + O ( r - 2 )

as r → ∞ . ∎

Theorem 3.4.

Let Q be any polynomial and let u ∈ M m σ ⁢ ( P ) . Then

Q ⁢ ( i ⁢ ∂ x ) ⁢ u ∼ ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ Q ⁢ ( ξ ( ν ) ⁢ ( ω ) ) ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 + ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ ∑ q = 1 ∞ b q , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 - q .

Proof.

From Lemma 3.3, we have

Q ( i ∂ x ) ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ) = Q ( ξ ( ν ) ( ω ) ) e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + i 2 ∑ l , q = 1 n Q ( l , q ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) q e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + O ( r - 2 )   as  r → ∞ .

Applying formula (3.4) to Q ⁢ ( i ⁢ ∂ x ) , one gets

Q ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 )
= Q ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ) ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 + ∑ 1 ≤ | α | ≤ m i | α | α ! ⁢ ∂ x α ⁡ ( d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 ) ⁡ Q ( α ) ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r )
= e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r Q ( ξ ( ν ) ( ω ) ) d 0 , m ( ν ) ( ω ) r 1 - n 2 + m - 1 + e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r i 2 ∑ l , q = 1 n Q ( l , q ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) q d 0 , m ( ν ) ( ω ) r 1 - n 2 + m - 1
+ i ⁢ ∑ l = 1 n ∂ x l ⁡ ( d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 ) ⁡ Q ( l ) ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ) + O ⁢ ( r 1 - n 2 + m - 3 )
= e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ Q ⁢ ( ξ ( ν ) ⁢ ( ω ) ) ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 + O ⁢ ( r 1 - n 2 + m - 2 )   as  ⁢ r → ∞ .

Thus,

Q ⁢ ( i ⁢ ∂ x ) ⁢ u ∼ ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ Q ⁢ ( ξ ( ν ) ⁢ ( ω ) ) ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 + ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ ∑ q = 1 ∞ b q , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 - q . ∎

Theorem 3.5.

Let u ∈ M m σ ⁢ ( P ) , i.e.,

u ∼ ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 + ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ ∑ q = 1 ∞ d q , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 - q .

Then

P ⁢ ( i ⁢ ∂ x ) ⁢ u ∼ ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ i ⁢ σ ν ⁢ ( m - 1 ) ⁢ | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 2
+ ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ ∑ q = 1 ∞ d ~ q , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 2 - q ,

i.e., P ⁢ ( i ⁢ ∂ x ) ⁢ u ∈ M m - 1 σ ⁢ ( P ) .

Proof.

From Lemma 3.3, we have

P ( i ∂ x ) ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ) = P ( ξ ( ν ) ( ω ) ) e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + i 2 ∑ l , q = 1 n P ( l , q ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) q e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + O ( r - 2 )   as  r → ∞ .

Since ξ ( ν ) ⁢ ( ω ) ∈ S ( ν ) implies P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) = 0 , one gets

P ( i ∂ x ) ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ) = i 2 ∑ l , q = 1 n P ( l , q ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) q e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + O ( r - 2 )   as  r → ∞ ,

where

P ( l , q ) ⁢ ( ξ ( ν ) ⁢ ( ω ) ) = ∂ ξ l ⁡ ∂ ξ q ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) .

Hence applying formula (3.4) to the operator P ⁢ ( i ⁢ ∂ x ) , one obtains

P ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 )
= P ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ) ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 + ∑ 1 ≤ | α | ≤ m i | α | α ! ⁢ ∂ x α ⁡ ( d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 ) ⁡ P ( α ) ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r )
= e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r i 2 ∑ l , q = 1 n P ( l , q ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) q d 0 , m ( ν ) ( ω ) r 1 - n 2 + m - 1
+ i ⁢ ∑ l = 1 n ∂ x l ⁡ ( d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 ) ⁡ P ( l ) ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ) + O ⁢ ( r 1 - n 2 + m - 3 )   as  ⁢ r → ∞ ,

where P ( l ) ⁢ ( i ⁢ ∂ x ) is the differential operator with the symbol P ( l ) ⁢ ( ξ ) = ∂ ξ l ⁡ P ⁢ ( ξ ) .

It follows from Lemma 3.3 that

i ⁢ ∑ l = 1 n ∂ x l ⁡ ( d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 ) ⁡ P ( l ) ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r )
= i ⁢ ∑ l = 1 n ∂ x l ⁡ ( d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 ) ⁡ P ( l ) ⁢ ( ξ ( ν ) ⁢ ( ω ) ) ⁢ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r
- 1 2 ∑ l = 1 n ∂ x l ( d 0 , m ( ν ) ( ω ) r 1 - n 2 + m - 1 ) ( ∑ p , q = 1 n P ( l , p , q ) ( ξ ( ν ) ( ω ) ) ∂ x p ( ξ ( ν ) ( ω ) ) q e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + O ( r - 2 ) )
= i ⁢ r 1 - n 2 + m - 1 ⁢ ∑ l = 1 n ( ∂ x l ⁡ d 0 , m ( ν ) ⁢ ( ω ) ) ⁢ P ( l ) ⁢ ( ξ ( ν ) ⁢ ( ω ) ) ⁢ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r
+ i ⁢ ( 1 - n 2 + m - 1 ) ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 2 ⁢ ∑ l = 1 n ω l ⁢ P ( l ) ⁢ ( ξ ( ν ) ⁢ ( ω ) ) ⁢ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r
+ O ⁢ ( r 1 - n 2 + m - 3 )   as  ⁢ r → ∞ ,

where

P ( l , p , q ) ⁢ ( ξ ( ν ) ⁢ ( ω ) ) = ∂ ξ l ⁡ ∂ ξ p ⁡ ∂ ξ q ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) .

Using Euler’s formula for homogeneous functions, we get

(3.5) ∑ l = 1 n x l ⁢ ∂ x l ⁡ d 0 , m ( ν ) ⁢ ( ω ) = 0 .

Since ξ ( ν ) ⁢ ( ω ) belongs to the zero surface of P ⁢ ( ξ ) , the vector ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) / | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | is normal to this surface, and hence parallel to x due to the definition of ξ ( ν ) ⁢ ( ω ) . Now, equation (3.5) implies that

∑ l = 1 n ∂ x l ⁡ d 0 , m ( ν ) ⁢ ( ω ) ⁢ P ( l ) ⁢ ( ξ ( ν ) ⁢ ( ω ) ) = 0 .

One also has

∑ l = 1 n ω l ⁢ P ( l ) ⁢ ( ξ ( ν ) ⁢ ( ω ) ) = σ ν ⁢ | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | .

Thus,

P ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 )
= e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r i 2 ∑ l , q = 1 n P ( l , q ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) q d 0 , m ( ν ) ( ω ) r 1 - n 2 + m - 1
+ i ⁢ σ ν ⁢ ( 1 - n 2 + m - 1 ) ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | ⁢ r 1 - n 2 + m - 2 ⁢ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r
+ O ⁢ ( r 1 - n 2 + m - 3 )   as  ⁢ r → ∞ .

Since ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) is parallel to x, the equality

(3.6) ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) = ρ ⁢ ( x ) ⁢ x

holds with some scalar function ρ. The left-hand side of (3.6) is homogeneous of degree 0 with respect x. Hence ρ is homogeneous of degree -1.

Let us write the divergence of the left- and the right-hand sides of (3.6):

∑ l = 1 n ∂ x l P ( l ) ( ξ ( ν ) ( ω ) ) = ∑ l , q = 1 n P ( l , q ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) q ,
∑ l = 1 n ∂ x l ⁡ ( ρ ⁢ ( x ) ⁢ x l ) = ∑ l = 1 n ( ∂ x l ⁡ ρ ⁢ ( x ) ) ⁢ x l + n ⁢ ρ ⁢ ( x ) = - ρ ⁢ ( x ) + n ⁢ ρ ⁢ ( x ) = ( n - 1 ) ⁢ ρ ⁢ ( x ) = σ ν ⁢ ( n - 1 ) ⁢ | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | r .

Therefore,

∑ l , q = 1 n P ( l , q ) ( ξ ( ν ) ( ω ) ) ∂ x l ( ξ ( ν ) ( ω ) ) q = σ ν ⁢ ( n - 1 ) ⁢ | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | r .

Hence,

P ⁢ ( i ⁢ ∂ x ) ⁢ ( e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 )
= i ⁢ σ ν ⁢ ( m - 1 ) ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | ⁢ r 1 - n 2 + m - 2 ⁢ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r + O ⁢ ( r 1 - n 2 + m - 3 )   as  ⁢ r → ∞ .

Thus,

P ⁢ ( i ⁢ ∂ x ) ⁢ u ∼ ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ i ⁢ σ ν ⁢ ( m - 1 ) ⁢ | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 2
+ ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ ∑ q = 1 ∞ d ~ q , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 2 - q .

This concludes the proof. ∎

Corollary 3.6.

Let u ∈ M m σ ⁢ ( P ) . Then ( P ⁢ ( i ⁢ ∂ x ) ) j ⁢ u ∈ M m - j σ ⁢ ( P ) , j = 1 , … , m .

Theorem 3.7.

Let u ∈ M m σ ⁢ ( P ) for m = 2 , 3 , … . Then

u - r ⁢ [ ∑ j = 0 ϰ - 1 A j ⁢ ( ω ) ⁢ ∂ r j ] ⁢ P ⁢ ( i ⁢ ∂ x ) ⁢ u = O ⁢ ( r 1 - n 2 + m - 2 )   as  ⁢ r → ∞ ,

where ( A 0 ⁢ ( ω ) , … , A ϰ - 1 ⁢ ( ω ) ) is the solution of the following almost everywhere solvable system:

∑ j = 0 ϰ - 1 ( - i ⁢ μ ( ν ) ⁢ ( ω ) ) j ⁢ A j ⁢ ( ω ) = ( i ⁢ σ ν ⁢ ( m - 1 ) ⁢ | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | ) - 1 , ν = 1 , … , ϰ .

Proof.

It follows from Theorem 3.5 that

(3.7)

r ⁢ [ ∑ j = 0 ϰ - 1 A j ⁢ ( ω ) ⁢ ∂ r j ] ⁢ P ⁢ ( i ⁢ ∂ x ) ⁢ u
= ∑ ν = 1 ϰ ∑ j = 0 ϰ - 1 e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ A j ⁢ ( ω ) ⁢ ( - i ⁢ μ ( ν ) ⁢ ( ω ) ) j ⁢ i ⁢ σ ν ⁢ ( m - 1 ) ⁢ | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1
+ O ⁢ ( r 1 - n 2 + m - 2 )   as  ⁢ r → ∞ .

The first ϰ coefficients of the asymptotic expansion (3.7) are equal to the first ϰ coefficients of the asymptotic expansion of u if the following equalities hold:

∑ j = 0 ϰ - 1 A j ⁢ ( ω ) ⁢ ( - i ⁢ μ ( ν ) ⁢ ( ω ) ) j ⁢ i ⁢ σ ν ⁢ ( m - 1 ) ⁢ | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | ⁢ d 0 , m ( ν ) ⁢ ( ω ) = d 0 , m ( ν ) ⁢ ( ω )

for ν = 1 , … , ϰ , i.e., if

(3.8) ∑ j = 0 ϰ - 1 ( - i ⁢ μ ( ν ) ⁢ ( ω ) ) j ⁢ A j ⁢ ( ω ) = ( i ⁢ σ ν ⁢ ( m - 1 ) ⁢ | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | ) - 1 , ν = 1 , … , ϰ .

The determinant of the coefficient matrix of system (3.8) is the Vandermonde determinant

det ⁡ ( 1 - i ⁢ μ ( 1 ) ⁢ ( ω ) ⋯ ( - i ⁢ μ ( 1 ) ⁢ ( ω ) ) ϰ - 1 1 - i ⁢ μ ( 2 ) ⁢ ( ω ) ⋯ ( - i ⁢ μ ( 2 ) ⁢ ( ω ) ) ϰ - 1 ⋮ ⋮ ⋱ ⋮ 1 - i ⁢ μ ( ϰ ) ⁢ ( ω ) ⋯ ( - i ⁢ μ ( ϰ ) ⁢ ( ω ) ) ϰ - 1 ) = ∏ k > j ( - i ) ⁢ ( μ ( k ) ⁢ ( ω ) - μ ( j ) ⁢ ( ω ) ) .

Now, μ ( k ) ⁢ ( ω ) - μ ( j ) ⁢ ( ω ) ≠ 0 for k ≠ j almost everywhere. Indeed, μ ( ν ) ⁢ ( ω ) = ( ξ ( ν ) ⁢ ( ω ) , ω ) , and any ω derivative of ξ ( ν ) ⁢ ( ω ) is a vector tangent to S ( ν ) at a point where the normal vector to S ( ν ) has the same direction as ω. Hence,

{ ω : ∇ ω ⁡ ( μ ( k ) ⁢ ( ω ) - μ ( j ) ⁢ ( ω ) ) = 0 } = ⋂ θ ⊥ ω { ω : ( ξ ( k ) ⁢ ( ω ) - ξ ( j ) ⁢ ( ω ) , θ ) = 0 } ,

where the intersection is taken over all θ perpendicular to the vector ω. Then

{ ω : μ ( k ) ⁢ ( ω ) - μ ( j ) ⁢ ( ω ) = 0 } ∩ { ω : ∇ ω ⁡ ( μ ( k ) ⁢ ( ω ) - μ ( j ) ⁢ ( ω ) ) = 0 }
= { ω : ( ξ ( k ) ⁢ ( ω ) - ξ ( j ) ⁢ ( ω ) , ω ) = 0 } ∩ ⋂ θ ⊥ ω { ω : ( ξ ( k ) ⁢ ( ω ) - ξ ( j ) ⁢ ( ω ) , θ ) = 0 }
= { ω : ξ ( k ) ⁢ ( ω ) - ξ ( j ) ⁢ ( ω ) = 0 }
= ∅ .

So, if the set

{ ω ∈ 𝕊 n - 1 : μ ( k ) ⁢ ( ω ) - μ ( j ) ⁢ ( ω ) = 0 } , k ≠ j ,

is nonempty, then it is an ( n - 2 ) -dimensional submanifold of 𝕊 n - 1 . Therefore, (3.8) is solvable for almost all ω ∈ 𝕊 n - 1 . Taking its solution ( A 0 ⁢ ( ω ) , … , A ϰ - 1 ⁢ ( ω ) ) , one gets

u - r ⁢ [ ∑ j = 0 ϰ - 1 A j ⁢ ( ω ) ⁢ ∂ r j ] ⁢ P ⁢ ( i ⁢ ∂ x ) ⁢ u = O ⁢ ( r 1 - n 2 + m - 2 )   as  ⁢ r → ∞ . ∎

Remark 3.8.

Suppose that the strictly convex surfaces S ( ν ) , ν = 1 , … , ϰ , are nested so that, for any k ≠ j , the surface S ( k ) either encloses S ( j ) or is enclosed by it. Then the tangent hyperplanes to S ( k ) at ξ ( k ) ⁢ ( ω ) and to S ( j ) at ξ ( j ) ⁢ ( ω ) are parallel to each other as they are both orthogonal to ω, but do not coincide. Hence they intersect the line joining the origin with ω at distinct points, and

μ ( k ) ⁢ ( ω ) = ( ξ ( k ) ⁢ ( ω ) , ω ) ≠ ( ξ ( j ) ⁢ ( ω ) , ω ) = μ ( j ) ⁢ ( ω ) .

So, μ ( k ) ⁢ ( ω ) - μ ( j ) ⁢ ( ω ) ≠ 0 for all ω ∈ 𝕊 n - 1 , k ≠ j , k , j = 1 , … , ϰ . This means that (3.8) is solvable for all ω ∈ 𝕊 n - 1 , and its solution ( A 0 ⁢ ( ω ) , … , A ϰ - 1 ⁢ ( ω ) ) is a C ∞ smooth function of ω. Then it follows from the above that the following representation formula holds for any u ∈ M m σ ⁢ ( P ) , m = 2 , 3 , … :

u = u 1 + r ⁢ [ ∑ j = 0 ϰ - 1 A j ⁢ ( ω ) ⁢ ∂ r j ] ⁢ P ⁢ ( i ⁢ ∂ x ) ⁢ u ,

where u 1 ∈ M m - 1 σ ⁢ ( P ) . A similar representation formula was obtained by Vekua in [11] in the case of spherical symmetry.

Theorem 3.9.

Let E m σ ∈ M m σ ⁢ ( P ) , m ∈ N , be a fundamental solution of the differential operator ( P ⁢ ( i ⁢ ∂ x ) ) m and let f ∈ L 1 , c ⁢ o ⁢ m ⁢ p ⁢ ( R n ) be an integrable function with a compact support. Then E m σ ∗ f ∈ M m σ ⁢ ( P ) .

Proof.

Since ℰ m σ ∈ M m σ ⁢ ( P ) (see Theorem 2.5 and Remark 2.6), we have

(3.9) ℰ m σ ⁢ ( x ) ∼ ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ { a 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 + ∑ q = 1 ∞ a q , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 - q }   as  ⁢ r → ∞ .

Let N be any positive integer number. Then

(3.10)

( ℰ m σ ∗ f ) ⁢ ( x ) = ∫ ℝ n ℰ m ⁢ ( x - y ) ⁢ f ⁢ ( y ) ⁢ 𝑑 y
= ∑ ν = 1 ϰ ∑ q = 0 N ∫ K e - i ⁢ μ ( ν ) ⁢ ( ω ~ ) ⁢ r ~ ⁢ a q , m ( ν ) ⁢ ( ω ~ ) ⁢ r ~ 1 - n 2 + m - 1 - q ⁢ f ⁢ ( y ) ⁢ 𝑑 y + O ⁢ ( r 1 - n 2 + m - 1 - ( N + 1 ) )   as  ⁢ r → ∞ ,

where

ω ~ = x - y | x - y | , r ~ = | x - y | , K = supp ⁡ f .

Let

B q , m ( ν ) ⁢ ( ω ~ , r ~ ) := a q , m ( ν ) ⁢ ( ω ~ ) ⁢ r ~ 1 - n 2 + m - 1 - q .

Taylor’s formula for B q , m ( ν ) ⁢ ( ω ~ , r ~ ) with respect to the variable y at the point y = 0 reads

(3.11) B q , m ( ν ) ⁢ ( ω ~ , r ~ ) = ∑ | α | ≤ N 1 α ! ⁢ ∂ x α ⁡ B q , m ( ν ) ⁢ ( ω , r ) ⁢ ( - y ) α + O ⁢ ( r 1 - n 2 + m - 1 - q - ( N + 1 ) )   as  ⁢ r → ∞ ,

uniformly in y ∈ K . Then it follows from (3.10) that

(3.12)

( ℰ m σ ∗ f ) ⁢ ( x ) = ∑ ν = 1 ϰ ∑ q = 0 N ∫ K e - i ⁢ μ ( ν ) ⁢ ( ω ~ ) ⁢ r ~ ⁢ B q , m ( ν ) ⁢ ( ω ~ , r ~ ) ⁢ f ⁢ ( y ) ⁢ 𝑑 y + O ⁢ ( r 1 - n 2 + m - 1 - ( N + 1 ) )
= ∑ ν = 1 ϰ ∑ q = 0 N ∑ | α | ≤ N ∫ K e - i ⁢ μ ( ν ) ⁢ ( ω ~ ) ⁢ r ~ ⁢ 1 α ! ⁢ ∂ x α ⁡ B q , m ( ν ) ⁢ ( ω , r ) ⁢ ( - y ) α ⁢ f ⁢ ( y ) ⁢ 𝑑 y + O ⁢ ( r 1 - n 2 + m - 1 - ( N + 1 ) )
= ∑ ν = 1 ϰ ∑ q = 0 N ∑ | α | ≤ N 1 α ! ⁢ ∂ x α ⁡ B q , m ( ν ) ⁢ ( ω , r ) ⁢ ∫ K e - i ⁢ μ ( ν ) ⁢ ( ω ~ ) ⁢ r ~ ⁢ ( - y ) α ⁢ f ⁢ ( y ) ⁢ 𝑑 y
+ O ⁢ ( r 1 - n 2 + m - 1 - ( N + 1 ) )   as  ⁢ r → ∞ .

It is clear that

g ν ⁢ ( x ) := μ ( ν ) ⁢ ( ω ) ⁢ r

is a homogeneous of degree 1 function of x.

Let us write Taylor’s formula for g ν ⁢ ( x - y ) with respect to the variable y at the point y = 0 :

(3.13)

g ν ⁢ ( x - y ) - g ν ⁢ ( x ) = ∑ 1 ≤ | γ | ≤ N + 1 1 γ ! ⁢ ∂ x γ ⁡ g ν ⁢ ( x ) ⁢ ( - y ) γ + R N + 2 ( ν ) ⁢ ( x , y )
= ∑ 1 ≤ | γ | ≤ N + 1 [ ( - y ) γ γ ! ⁢ ∂ x γ ⁡ g ν ⁢ ( ω ) ] ⁢ r 1 - | γ | + R N + 2 ( ν ) ⁢ ( x , y ) ,

where

R N + 2 ( ν ) ⁢ ( x , y ) = ∑ | γ | = N + 2 1 γ ! ⁢ ∂ x γ ⁡ g ν ⁢ ( x - t ⁢ y ) ⁢ ( - y ) γ = ∑ | γ | = N + 2 ( - y ) γ γ ! ⁢ ∂ x γ ⁡ g ν ⁢ ( ω ~ t ) ⁢ r ~ t - ( N + 1 )

with

ω ~ t = x - t ⁢ y | x - t ⁢ y | , r ~ t = | x - t ⁢ y | , 0 < t < 1 ,

and

R N + 2 ( ν ) ⁢ ( x , y ) = O ⁢ ( r - ( N + 1 ) )   as  ⁢ r → ∞ ,

uniformly in y ∈ K . Further,

(3.14)

∫ K e - i ⁢ μ ( ν ) ⁢ ( ω ~ ) ⁢ r ~ ⁢ ( - y ) α ⁢ f ⁢ ( y ) ⁢ 𝑑 y = e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ ∫ K e - i ⁢ [ μ ( ν ) ⁢ ( ω ~ ) ⁢ r ~ - μ ( ν ) ⁢ ( ω ) ⁢ r ] ⁢ ( - y ) α ⁢ f ⁢ ( y ) ⁢ 𝑑 y
= e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ ∫ K e - i ⁢ [ g ν ⁢ ( x - y ) - g ν ⁢ ( x ) ] ⁢ ( - y ) α ⁢ f ⁢ ( y ) ⁢ 𝑑 y .

Also,

(3.15) e - i ⁢ R N + 2 ( ν ) ⁢ ( x , y ) = 1 + O ⁢ ( R N + 2 ( ν ) ⁢ ( x , y ) ) = 1 + O ⁢ ( r - ( N + 1 ) )   as  ⁢ r → ∞ .

Substituting (3.13) into e - i ⁢ [ g ν ⁢ ( x - y ) - g ν ⁢ ( x ) ] and using (3.15), one gets

(3.16)

e - i ⁢ [ g ν ⁢ ( x - y ) - g ν ⁢ ( x ) ] = ∏ | γ | = 1 e i ⁢ y γ ⁢ ( ∂ x γ ⁡ g ν ) ⁢ ( ω ) ⁢ ∏ 2 ≤ | γ | ≤ N + 1 e - i ⁢ ( - y ) γ γ ! ⁢ ( ∂ x γ ⁡ g ν ) ⁢ ( ω ) ⁢ r 1 - | γ | ⁢ e - i ⁢ R N + 2 ( ν ) ⁢ ( x , y )
= ∏ | γ | = 1 e i ⁢ y γ ⁢ ( ∂ x γ ⁡ g ν ) ⁢ ( ω ) ⁢ ∏ 2 ≤ | γ | ≤ N + 1 { ∑ k = 0 N [ - i ⁢ ( - y ) γ γ ! ⁢ ( ∂ x γ ⁡ g ν ) ⁢ ( ω ) ] k k ! ⁢ r k ⁢ ( 1 - | γ | ) + O ⁢ ( r ( N + 1 ) ⁢ ( 1 - | γ | ) ) }
× ( 1 + O ⁢ ( r - ( N + 1 ) ) )
= ∏ | γ | = 1 e i ⁢ y γ ⁢ ( ∂ x γ ⁡ g ν ) ⁢ ( ω ) ⁢ ∑ k = 0 N ∑ | β | = 2 ⁢ k N + k b β , k ( ν ) ⁢ ( ω ) ⁢ ( - y ) β ⁢ r - | β | + k + O ⁢ ( r - ( N + 1 ) )

as r → ∞ , uniformly in y ∈ K , where | β | := k ⁢ | γ | , k ⁢ ( 1 - | γ | ) = - | β | + k , and | β | ≥ 2 ⁢ k .

Substituting (3.16) into (3.14), we obtain

(3.17)

∫ K e - i ⁢ μ ( ν ) ⁢ ( ω ~ ) ⁢ r ~ ⁢ ( - y ) α ⁢ f ⁢ ( y ) ⁢ 𝑑 y
= e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ ∑ k = 0 N ∑ | β | = 2 ⁢ k N + k b β , k ( ν ) ⁢ ( ω ) ⁢ { ∫ K ∏ | γ | = 1 e i ⁢ y γ ⁢ ( ∂ x γ ⁡ g ν ) ⁢ ( ω ) ⁢ ( - y ) α + β ⁢ f ⁢ ( y ) ⁢ d ⁢ y } ⁢ r - | β | + k + O ⁢ ( r - ( N + 1 ) )
= e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ ∑ k = 0 N ∑ | β | = 2 ⁢ k N + k d α , β , k ( ν ) ⁢ ( ω ) ⁢ r - | β | + k + O ⁢ ( r - ( N + 1 ) )   as  ⁢ r → ∞ ,

where

d α , β , k ( ν ) ⁢ ( ω ) = b β , k ( ν ) ⁢ ( ω ) ⁢ ∫ K ∏ | γ | = 1 e i ⁢ y γ ⁢ ( ∂ x γ ⁡ g ν ) ⁢ ( ω ) ⁢ ( - y ) α + β ⁢ f ⁢ ( y ) ⁢ d ⁢ y .

Taking into account (3.17) and the equality

∂ y α B q , m ( ν ) ( ω , r ) = ∂ y α { a q , m ( ν ) ( ω ~ ) r ~ 1 - n 2 + m - 1 - q } | y = 0 = d ~ q , m ( ν ) , α ( ω ) r 1 - n 2 + m - 1 - q - | α | ,

from (3.12) we obtain

( ℰ m σ ∗ f ) ⁢ ( x ) = ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ { ∑ q = 0 N ∑ | α | ≤ N ∑ k = 0 N ∑ | β | = 2 ⁢ k N + k d q , m , k ( ν ) , α , β ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 - q - | α | - ( | β | - k ) + O ⁢ ( r 1 - n 2 + m - 1 - ( N + 1 ) ) }

as r → ∞ , where

d q , m , k ( ν ) , α , β ⁢ ( ω ) = 1 α ! ⁢ d ~ q , m ( ν ) , α ⁢ ( ω ) ⁢ d α , β , k ( ν ) ⁢ ( ω ) .

Hence,

ℰ m σ ∗ f ∼ ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ { d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 + ∑ p = 1 ∞ d p , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 - p }   as  ⁢ r → ∞ ,

where

d 0 , m ( ν ) ⁢ ( ω ) = a 0 , m ( ν ) ⁢ ( ω ) ⁢ ∫ K ∏ | γ | = 1 e i ⁢ y γ ⁢ ( ∂ x γ ⁡ g ν ) ⁢ ( ω ) ⁢ f ⁢ ( y ) ⁢ d ⁢ y ,

and a 0 , m ( ν ) is the first coefficient of the asymptotic expansion of the fundamental solution ℰ m σ . Thus,

ℰ m σ ∗ f ∈ M m σ ⁢ ( P ) , m ∈ ℕ . ∎

Corollary 3.10.

Let f be a distribution with compact support. Then E m σ ∗ f ∈ M m σ ⁢ ( P ) .

Proof.

Since f is a distribution with compact support, it can be represented in the form

f = ∑ | α | ≤ k ∂ α ⁡ f α ,

where f α , | α | ≤ k , are continuous functions with compact support (see, e.g., [1, Theorem 6.4]). Hence,

( ℰ m σ ∗ f ) ⁢ ( x ) = ∑ | α | ≤ k ∂ α ⁡ ℰ m σ ⁢ ( x ) ∗ f α ⁢ ( x ) .

Since ∂ α ⁡ ℰ m σ ∈ M m σ ⁢ ( P ) (see Theorem 2.5), it follows from the proof of Theorem 3.9 that ℰ m σ ∗ f ∈ M m σ ⁢ ( P ) . ∎

Theorem 3.11.

Let P ⁢ ( i ⁢ ∂ x ) be a hypoelliptic differential operator satisfying conditions (i’)–(iii’), and let f be an arbitrary distribution with compact support. Then, for j = 1 , … , m , m = 1 , 2 , … , the differential equation

(3.18) ( P ⁢ ( i ⁢ ∂ x ) ) j ⁢ u = f   in  ⁢ ℝ n ,

has a unique solution u in the class M m σ ⁢ ( P ) , and it can be represented as

u = ℰ j σ ∗ f   in  ⁢ ℝ n ,

where E j σ is the fundamental solution of the operator ( P ⁢ ( i ⁢ ∂ x ) ) j constructed above.

Proof.

We prove this theorem in two steps.

Existence. It follows from Corollary 3.10 that

ℰ j σ ∗ f ∈ M j σ ⁢ ( P ) ⊆ M m σ ⁢ ( P ) .

It is easy to see that u = ℰ j σ ∗ f is a solution of equation (3.18). Indeed,

( P ⁢ ( i ⁢ ∂ x ) ) j ⁢ u = ( P ⁢ ( i ⁢ ∂ x ) ) j ⁢ ( ℰ j σ ∗ f ) = ( P ⁢ ( i ⁢ ∂ x ) ) j ⁢ ℰ j σ ∗ f = δ ∗ f = f .

Uniqueness. Let j = 1 , … , m and let u ∈ M m σ ⁢ ( P ) be a solution of the homogeneous differential equation

(3.19) ( P ⁢ ( i ⁢ ∂ x ) ) j ⁢ u = 0   in  ⁢ ℝ n .

We have to show that equation (3.19) possesses only the trivial solution in the class M m σ ⁢ ( P ) .

Let v := ( P ⁢ ( i ⁢ ∂ x ) ) j - 1 ⁢ u . Then it follows from Corollary 3.6 that v ∈ M m - j + 1 σ ⁢ ( P ) ⊂ M m σ ⁢ ( P ) , i.e.,

v ∼ ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ { d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 + ∑ q = 1 ∞ d q , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 1 - q } ,

and P ⁢ ( i ⁢ ∂ x ) ⁢ v = ( P ⁢ ( i ⁢ ∂ x ) ) j ⁢ u = 0 in ℝ n . Theorem 3.5 implies

P ⁢ ( i ⁢ ∂ x ) ⁢ v = ∑ ν = 1 ϰ i ⁢ σ ν ⁢ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ ( m - 1 ) ⁢ | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | ⁢ d 0 , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 2
+ ∑ ν = 1 ϰ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ ∑ q = 1 ∞ d ~ q , m ( ν ) ⁢ ( ω ) ⁢ r 1 - n 2 + m - 2 - q
= 0 .

Hence,

∑ ν = 1 ϰ i ⁢ σ ν ⁢ e - i ⁢ μ ( ν ) ⁢ ( ω ) ⁢ r ⁢ ( m - 1 ) ⁢ | ∇ ⁡ P ⁢ ( ξ ( ν ) ⁢ ( ω ) ) | ⁢ d 0 , m ( ν ) ⁢ ( ω ) = 0 .

The exponential functions of the variable r constitute a linearly independent system for almost all ω (see the proof of Theorem 3.7). Hence the coefficients satisfy d 0 , m ( ν ) ⁢ ( ω ) = 0 , ν = 1 , … , ϰ , for almost all ω. Since the coefficients d 0 , m ( ν ) ⁢ ( ω ) are smooth, we have d 0 , m ( ν ) ⁢ ( ω ) = 0 , ν = 1 , … , ϰ , for all ω. Thus v ∈ M m - 1 σ ⁢ ( P ) . Repeating this procedure ( m - 2 ) more times, one gets

v ∈ M 1 σ ⁢ ( P ) ⊂ Som σ ⁡ ( P ) .

Since P ⁢ ( i ⁢ ∂ x ) ⁢ v = 0 in ℝ n , it follows from Theorem A.4 that v = 0 in ℝ n , i.e., ( P ⁢ ( i ⁢ ∂ x ) ) j - 1 ⁢ u = 0 in ℝ n . Repeating the above argument, one arrives at P ⁢ ( i ⁢ ∂ x ) ⁢ u = 0 in ℝ n and then shows as above that u ∈ M 1 σ ⁢ ( P ) . Using Theorem A.4 again, one finally gets u = 0 in ℝ n . ∎

4 Existence and uniqueness of solutions for general hypoelliptic equations

Let 𝐏 ⁢ ( i ⁢ ∂ x ) (see (2.3)) be a hypoelliptic operator satisfying conditions (i)–(iii).

Definition 4.1.

We say that a distribution u belongs to the class M m 1 , … , m l σ ⁢ ( 𝐏 ) for m j ∈ ℕ , j = 1 , … , l , if it has the following uniform in ω = x / | x | asymptotic expansion as r = | x | → ∞ :

(4.1) u ∼ ∑ j = 1 l ∑ ν = 1 ϰ j e - i ⁢ μ j ( ν ) ⁢ ( ω ) ⁢ r ⁢ { d 0 , m j ( ν ) , j ⁢ ( ω ) ⁢ r 1 - n 2 + m j - 1 + ∑ q = 1 ∞ d q , m j ( ν ) , j ⁢ ( ω ) ⁢ r 1 - n 2 + m j - 1 - q } ,

where d q , m j ( ν ) , j ∈ C ∞ , j = 1 , … , l , ν = 1 , … , ϰ j , q = 1 , 2 , … , and this expansion admits term by term differentiation with respect to x any number of times.

Theorem 4.2.

Let E m 1 , … , m I σ be a fundamental solution of the hypoelliptic differential operator P ⁢ ( i ⁢ ∂ x ) and let f be a distribution with compact support. Then

ℰ m 1 , … , m I σ ∗ f ∈ M m 1 , … , m l σ ⁢ ( 𝐏 ) .

The proof of this theorem is analogous to the ones of Theorem 3.9 and Corollary 3.10.

Theorem 4.3.

Let P ⁢ ( i ⁢ ∂ x ) be a hypoelliptic differential operator satisfying conditions (i)–(iii), and let f be an arbitrary distribution with compact support. Then for j k = 1 , … , m k , k = 1 , … , l , the differential equation

(4.2) 𝐏 ~ ⁢ ( i ⁢ ∂ x ) ⁢ u = ∏ k = 1 l ( P k ⁢ ( i ⁢ ∂ x ) ) j k ⁢ u = f   in  ⁢ ℝ n ,

has a unique solution in the class M m 1 , … , m l σ ⁢ ( P ) , and this solution can be represented as

u = ℰ j 1 , … , j I σ ∗ f   in  ⁢ ℝ n ,

where E j 1 , … , j I σ is the fundamental solution of the operator P ~ ⁢ ( i ⁢ ∂ x ) constructed in Section 2.

Proof.

We prove this theorem in two steps.

Existence. It follows from Theorem 4.2 that

ℰ j 1 , … , j I σ ∗ f ∈ M j 1 , … , j l σ ⁢ ( 𝐏 ~ ) ⊂ M m 1 , … , m l σ ⁢ ( 𝐏 ) .

It is easy to see that u = ℰ j 1 , … , j I σ ∗ f is a solution of equation (4.2). Indeed,

𝐏 ~ ⁢ ( i ⁢ ∂ x ) ⁢ u = 𝐏 ~ ⁢ ( i ⁢ ∂ x ) ⁢ ( ℰ j 1 , … , j I σ ∗ f ) = 𝐏 ~ ⁢ ( i ⁢ ∂ x ) ⁢ ℰ j 1 , … , j I σ ∗ f = δ ∗ f = f .

Uniqueness. Let u ∈ M m 1 , … , m l σ ⁢ ( 𝐏 ) be a solution of the homogeneous differential equation

(4.3) 𝐏 ~ ⁢ ( i ⁢ ∂ x ) ⁢ u = ∏ k = 1 l ( P k ⁢ ( i ⁢ ∂ x ) ) j k ⁢ u = 0   in  ⁢ ℝ n .

We have to show that equation (4.3) possesses only the trivial solution in the class M m 1 , … , m l σ ⁢ ( 𝐏 ) .

First, let us consider the equation

( P 1 ⁢ ( i ⁢ ∂ x ) ) j 1 ⁢ u ~ = 0   in  ⁢ ℝ n , j 1 = 1 , … , m 1 , u ~ ∈ M m 1 , … , m l σ ⁢ ( 𝐏 ) .

Let v ~ := ( P 1 ⁢ ( i ⁢ ∂ x ) ) j 1 - 1 ⁢ u ~ , v ~ ∈ M m 1 , … , m l σ ⁢ ( 𝐏 ) . Then P 1 ⁢ ( i ⁢ ∂ x ) ⁢ v ~ = 0 in ℝ n . It follows from Theorem 3.4 and Theorem 3.5 that

P 1 ⁢ ( i ⁢ ∂ x ) ⁢ v ~ = ∑ ν = 1 ϰ 1 i ⁢ σ ν ( 1 ) ⁢ e - i ⁢ μ 1 ( ν ) ⁢ ( ω ) ⁢ r ⁢ ( m 1 - 1 ) ⁢ | ∇ ⁡ P 1 ⁢ ( ξ 1 ( ν ) ⁢ ( ω ) ) | ⁢ d 0 , m 1 ( ν ) , 1 ⁢ ( ω ) ⁢ r 1 - n 2 + m 1 - 2
+ ∑ ν = 1 ϰ 1 e - i ⁢ μ 1 ( ν ) ⁢ ( ω ) ⁢ r ⁢ ∑ q = 1 ∞ d ~ q , m 1 ( ν ) , 1 ⁢ ( ω ) ⁢ r 1 - n 2 + m 1 - 2 - q
+ ∑ j = 2 l ∑ ν = 1 ϰ j P 1 ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) ⁢ e - i ⁢ μ j ( ν ) ⁢ ( ω ) ⁢ r ⁢ d 0 , m j ( ν ) , j ⁢ ( ω ) ⁢ r 1 - n 2 + m j - 1
+ ∑ j = 2 l ∑ ν = 1 ϰ j e - i ⁢ μ j ( ν ) ⁢ ( ω ) ⁢ r ⁢ ∑ q = 1 ∞ d ~ q , m j ( ν ) , j ⁢ ( ω ) ⁢ r 1 - n 2 + m j - 1 - q
= 0 .

As in the proof of Theorem 3.11, the exponential functions of the variable r constitute a linearly independent system for almost all ω (see the proof of Theorem 3.7). Hence,

∑ ν = 1 ϰ 1 i ⁢ σ ν ( 1 ) ⁢ e - i ⁢ μ 1 ( ν ) ⁢ ( ω ) ⁢ r ⁢ ( m 1 - 1 ) ⁢ | ∇ ⁡ P 1 ⁢ ( ξ 1 ( ν ) ⁢ ( ω ) ) | ⁢ d 0 , m 1 ( ν ) , 1 ⁢ ( ω ) = 0 ,
∑ ν = 1 ϰ j P 1 ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) ⁢ e - i ⁢ μ j ( ν ) ⁢ ( ω ) ⁢ r ⁢ d 0 , m j ( ν ) , j ⁢ ( ω ) = 0 , j = 2 , … , l ,

and the coefficients satisfy

d 0 , m j ( ν ) , j ⁢ ( ω ) = 0 , ν = 1 , … , ϰ j , j = 2 , … , l ,

for almost all ω, because P 1 ⁢ ( ξ j ( ν ) ⁢ ( ω ) ) ≠ 0 , ν = 1 , … , ϰ j , j = 2 , … , l . Since the coefficients d 0 , m j ( ν ) , j ⁢ ( ω ) are smooth, we have

d 0 , m j ( ν ) , j ⁢ ( ω ) = 0 , ν = 1 , … , ϰ j , j = 2 , … , l ,

for all ω. If m 1 - 1 > 0 , i.e., m 1 > 1 , the same argument gives d 0 , m 1 ( ν ) , 1 ⁢ ( ω ) = 0 , ν = 1 , … , ϰ 1 , for all ω. Thus,

v ~ ∈ M max ⁡ { m 1 - 1 , 1 } , m 2 - 1 ⁢ … , m l - 1 σ ⁢ ( 𝐏 ) .

Repeating this procedure, one can show that

v ~ ∈ M 1 , m ~ 2 , … , m ~ l σ ⁢ ( 𝐏 ) ,

where m ~ j ≤ 0 are any integer numbers j = 2 , … , l . Since

M 1 , m ~ 2 , … , m ~ l σ ⁢ ( 𝐏 ) ⊂ Som σ ⁡ ( P 1 )

for such m ~ j , it follows from P 1 ⁢ ( i ⁢ ∂ x ) ⁢ v ~ = 0 that v ~ = 0 in ℝ n (see Theorem A.4), i.e., that ( P 1 ⁢ ( i ⁢ ∂ x ) ) j 1 - 1 ⁢ u ~ = 0 in ℝ n . Repeating the above argument, one arrives at P 1 ⁢ ( i ⁢ ∂ x ) ⁢ u ~ = 0 in ℝ n , and then shows as above that u ~ ∈ Som σ ⁡ ( P 1 ) . Using Theorem A.4 again, one gets u ~ = 0 in ℝ n .

Now, let us consider equation (4.3). Let

v := ( P 2 ⁢ ( i ⁢ ∂ x ) ) j 2 ⁢ ⋯ ⁢ ( P l ⁢ ( i ⁢ ∂ x ) ) j I ⁢ u .

Then

v ∈ M m 1 , … , m l σ ⁢ ( 𝐏 )   and   ( P 1 ⁢ ( i ⁢ ∂ x ) ) j 1 ⁢ v = ( P 1 ⁢ ( i ⁢ ∂ x ) ) j 1 ⁢ ⋯ ⁢ ( P l ⁢ ( i ⁢ ∂ x ) ) j I ⁢ u = 0

in ℝ n . It follows from the above that v = 0 in ℝ n . Repeating this procedure successively for

( P 2 ⁢ ( i ⁢ ∂ x ) ) j 2 , … , ( P l ⁢ ( i ⁢ ∂ x ) ) j I ,

one finally gets u = 0 in ℝ n . ∎

It is well known that the classical Helmholtz equation has a unique solution in ℝ n satisfying Sommerfeld’s radiation conditions (see [8]). Theorem 4.3 implies the following uniqueness result for the Helmholtz equation, which requires more detailed information on the asymptotic behavior of solutions at infinity, but allows any polynomial growth.

Theorem 4.4.

The homogeneous Helmholtz equation

( Δ + k 2 ) ⁢ u = 0   in  ⁢ ℝ n ,

has only a trivial solution in the class M m σ ⁢ ( Δ + k 2 ) for any m ∈ R .

A Appendix

Lemma A.1 ([10, Chapter 7, Section 2, Lemma 4]).

For any smooth function f,

| p . v . ∫ | t | < δ t - 1 f ( t ) d t | ≤ 2 δ sup | t | ≤ δ | f ′ ( t ) | .

Lemma A.2 ([10, Chapter 7, Section 2, Lemma 5]).

Suppose that φ ⁢ ( ω , t ) , f ⁢ ( ω , t ) ∈ C ∞ ⁢ ( S n - 1 × R ) , Im ⁡ φ = 0 , f = 0 for | t | > δ > 0 , and that φ t ′ ≠ 0 for | t | ≤ δ . Then

p . v . ∫ - ∞ + ∞ t - 1 ⁢ f ⁢ ( ω , t ) ⁢ e i ⁢ φ ⁢ ( ω , t ) ⁢ λ ⁢ 𝑑 t = α ⁢ π ⁢ i ⁢ f ⁢ ( ω , 0 ) ⁢ e i ⁢ φ ⁢ ( ω , 0 ) ⁢ λ + F ⁢ ( ω , λ ) ,

where α = sign ⁡ φ t ′ ⁢ ( ω , 0 ) , and for any j , N < ∞ and any differential operator P ⁢ ( ω , ∂ ω ) on S n - 1 with bounded coefficients, there is a constant C ⁢ ( N , j , P ) < ∞ such that

| P ⁢ ( ω , ∂ ω ) ⁢ d j d ⁢ λ j ⁢ F ⁢ ( ω , λ ) | < C ⁢ ( N , j , P ) ⁢ λ - N , λ > 1 .

Let S be an infinitely differentiable bounded closed ( n - 1 ) -dimensional surface in ℝ n . We denote by ω points of the unit sphere ω = x r , r = | x | . Set

I ⁢ ( x ) := ∫ S f ⁢ ( ξ ) ⁢ e - i ⁢ ( x , ξ ) ⁢ 𝑑 S , f ∈ C ∞ ⁢ ( ℝ n ) .

We denote by ξ j = ξ j ⁢ ( ω ) , 1 ≤ j ≤ m , the points of S at which the normal to S is parallel to the vector ω, and by λ j ⁢ s , 1 ≤ s ≤ n - 1 , the principal curvatures of S at the point ξ j ⁢ ( ω ) . Let 𝒦 j ⁢ ( ω ) be the total curvature of S at the point ξ j ⁢ ( ω ) , that is,

𝒦 j ⁢ ( ω ) = ∏ s = 1 n - 1 λ j ⁢ s ,

and let γ j be the difference between the number of positive and negative principle curvatures at the point ξ j ⁢ ( ω ) . Let μ j ⁢ ( ω ) := ( ξ j ⁢ ( ω ) , ω ) .

Theorem A.3 ([10, Chapter 1, Section 5, Theorem 9]).

Suppose there are m points ξ j on the surface S at which the normal to S is parallel to the vector ω 0 , and suppose K j ⁢ ( ω 0 ) ≠ 0 . Then there exists ε > 0 such that, for | ω - ω 0 | < ε ,

I ⁢ ( x ) ∼ ∑ j = 1 m { e - i ⁢ μ j ⁢ ( ω ) ⁢ r ⁢ ∑ q = 0 ∞ a q , j ⁢ ( ω ) ⁢ r 1 - n 2 - q }   as  ⁢ r → ∞ ,

where

a 0 , j ⁢ ( ω ) = ( 2 ⁢ π ) n - 1 2 ⁢ f ⁢ ( ξ j ⁢ ( ω ) ) ⁢ e - i ⁢ γ j ⁢ π / 4 | 𝒦 j ⁢ ( ω ) | .

This expansion can be differentiated term by term with respect to x any number of times.

Theorem A.4 ([10, Chapter 7, Section 2, Theorem 6]).

Let P ⁢ ( i ⁢ ∂ x ) be a hypoelliptic differential operator satisfying conditions (i’)–(iii’) (see the beginning of Section 3), and let f be an arbitrary distribution with compact support. Then the differential equation

P ⁢ ( i ⁢ ∂ x ) ⁢ u = f   in  ⁢ ℝ n ,

has a unique solution u in the class Som σ ⁥ ( P ) (see Definition 3.2), and this solution can be represented as

u = ℰ 1 σ ∗ f   in  ⁢ ℝ n ,

where E 1 σ is the fundamental solution of the operator P ⁢ ( i ⁢ ∂ x ) and has the following uniform asymptotic expansion:

u ∼ ∑ j = 1 ϰ { e - i ⁢ μ j ⁢ ( ω ) ⁢ r ⁢ ∑ q = 0 ∞ c q , j ⁢ ( ω ) ⁢ r 1 - n 2 - q }   as  ⁢ r → ∞ ,

where c q , j ∈ C ∞ , and this expansion can be differentiated term by term with respect to x any number of times.

Note that the fundamental solution ℰ 1 σ equals ℰ m σ for m = 1 , which is constructed in Lemma 2.3 and Theorem 2.4.

References

[1] G. Eskin, Lectures on Linear Partial Differential Equations, Grad. Stud. Math. 123, American Mathematical Society, Providence, 2011. 10.1090/gsm/123Search in Google Scholar

[2] L. HĂśrmander, Linear Partial Differential Operators, Springer, Berlin, 1976. Search in Google Scholar

[3] L. HĂśrmander, The Analysis of Linear Partial Differential Operators. II, Grundlehren Math. Wiss. 257, Springer, Berlin, 1983. Search in Google Scholar

[4] S. Kobayashi and K. Nomizu, Foundations of Differential Geometry. Vol. II, Intersci. Tracts Pure Appl. Math. 15, Interscience, New York, 1969. Search in Google Scholar

[5] S. G. Krantz and H. R. Parks, A Primer of Real Analytic Functions, 2nd ed., Birkhäuser Adv. Texts Basler Lehrbßcher, Birkhäuser Boston, 2002. Search in Google Scholar

[6] B. P. Paneyah, Existence and uniqueness of the solution of the n-metaharmonic equation on an unbounded space (in Russian), Vestnik Moskov. Univ. Ser. Mat. Meh. Astr. Fiz. Him. 1959 (1959), no. 5, 123–135. Search in Google Scholar

[7] M. Schechter, Modern Methods in Partial Differential Equations, Dover, Mineola, 2014. Search in Google Scholar

[8] A. Sommerfeld, Die Greensche Funktion der Schwingungsgleichung, Deutsche Math.-Ver. 21 (1912), 309–353. Search in Google Scholar

[9] B. R. Vaĭnberg, Principles of radiation, limiting absorption and limiting amplitude in the general theory of partial differential equations, Uspehi Mat. Nauk 21 (1966), no. 3(129), 115–194. 10.1070/RM1966v021n03ABEH004157Search in Google Scholar

[10] B. R. VaÄ­nberg, Asymptotic Methods in Equations of Mathematical Physics, Gordon & Breach Science, New York, 1989. Search in Google Scholar

[11] E. Vecoua, On metaharmonic functions (in Russian), Trav. Inst. Math. Tbilissi [Trudy Tbiliss. Mat. Inst.] 12 (1943), 105–174. Search in Google Scholar

Received: 2022-11-29
Accepted: 2023-05-02
Published Online: 2023-10-28
Published in Print: 2024-04-01

Š 2023 Walter de Gruyter GmbH, Berlin/Boston

This work is licensed under the Creative Commons Attribution 4.0 International License.

Downloaded on 6.9.2025 from https://www.degruyterbrill.com/document/doi/10.1515/gmj-2023-2072/html
Scroll to top button