DOI 10.

1007/s10958-017-3653-4
Journal of Mathematical Sciences, Vol. 228, No. 6, February, 2018

TRACE THEOREM AND APPLICATIONS

Yu. A. Dubinskii
National Research University “Moscow Power Engineering Institute”
14, Krasnokazarmennaya St., Moscow 111250, Russia
julii [email protected] UDC 517.936.225

We characterize three-dimensional vector fields on the basis of the trace of a certain

combination of normal derivatives, curl, and divergence. We clarify an unconditional
connection between the values of a vector-valued function and the values of gradient,
curl, and divergence on the boundary, which makes it possible to consider boundary
value problems with boundary conditions that involve the basic first order differential
operations of field theory. Bibliography: 3 titles.

1 Trace Theorem
Let G ⊂ R3 be a domain with piecewise Lipschitz boundary Γ. We denote by W2m (G) the
Sobolev spaces of vector-valued functions u(x) = (u1 (x), u2 (x), u3 (x)) equipped with the norm
3

||u||m = ||uj ||m ,
j=1

where
m

||uj ||m = ||∇k uj ||0
k=0
or, which is equivalent, 
||uj ||m = ||D α uj ||0 ;
|α|m

here, m  0 is an integer and || · ||0 is the L2 (G)-norm. We also use the spaces W21 (G), W22 (G),
1/2 −1/2 1/2
the trace space W2 (Γ) of functions in W21 (G), and the dual W2 (Γ) of W2 (Γ).
−1/2
Theorem 1.1. There exists a linear bounded operator DΓ : W21 (G) → W2 (Γ) such that
1) DΓ is symmetric, i.e.,
DΓ u, v|Γ  = u|Γ , DΓ v
for any u(x) = (u1 (x), u2 (x), u3 (x)) and v(x) = (v1 (x), v2 (x), v3 (x)) in W21 (G), where DΓ u, v|Γ 
denotes the value of the functional DΓ u on the boundary value v|Γ of v(x) ∈ W21 (G),

Translated from Problemy Matematicheskogo Analiza 90, January 2018, pp. 49-54.
1072-3374/18/2286-0655 
c 2018 Springer Science+Business Media New York

655
 2) for any u ∈ W22 (G)
 
∂u 
DΓ u = − [cirl u, n] − div u · n 
∂n Γ

in the sense of traces of functions in Sobolev spaces. Here, n = n(γ), γ ∈ Γ, is the unit
normal vector to the boundary Γ and ∂u/∂n = (∂u1 /∂n, ∂u2 /∂n, ∂u3 /∂n) is the vector of normal
derivatives of components of u(x). Moreover,
   
(∇u, ∇v)dx = (cirl u, cirl v)dx + div u · div vdx + (DΓ u, v)dγ
G G G Γ

for any u(x) ∈ W22 (G) and v(x) ∈ W22 (G).

First of all, we formulate some consequences of this theorem. From the symmetry property
of the operator DΓ we obtain the following assertion.

Corollary 1.1. If DΓ u = 0 for some u(x) ∈ W21 (G), then for any v(x) ∈ W21 (G)
  
(∇u, ∇v)dx = (cirl u, cirl v)dx + div u · div vdx.
G G G

Conversely, if this integral identity holds, then DΓ u = 0.

Thus, the operator DΓ is the deviator of the operators

 2 
∂ u ∂2u ∂2u
−Δu ≡ − + + , δu ≡ cirl2 u − ∇ div u
∂x21 ∂x22 ∂x23

that act within the framework of duality (W21 (G), W2−1 (G)) according to the usual rules of
integration by parts.

Corollary 1.2. Let u(x) ∈ W21 (G). Then

  
|∇u| dx = | cirl u| dx + (div u)2 dx
2 2

G G G

if and only if DΓ u, u|Γ  = 0.

Proof of Theorem 1.1. We consider the bilinear form


L(u, v) ≡ [(∇u, ∇v) − (cirl u, cirl v) − div u · div v]dx,
G

where u(x) ∈ W21 (G), v(x) ∈ W21 (G), and show that for any fixed function u(x) this bilinear
form determines a linear bounded functional over the space of traces of functions v(x) ∈ W21 (G),
1/2
i.e., over the space W2 (Γ).
Lemma 1.1. If v(x) ∈ W21 (G) and v|Γ = 0, then L(u, v) = 0 for any u(x) ∈ W21 (G).

656
 Proof. Since smooth compactly supported functions are everywhere dense in the space
0
1 1
W 2 (G) = {v(x) ∈ W2 (G) : v|Γ = 0},

it suffices to consider v(x) ∈ D(G). Integrating by parts, we find


L(u, v) = (u, (−Δv − cirl 2 v + ∇ div v))dx ≡ 0
G

by the classical formula −Δv ≡ cirl 2 v − ∇ div v of field theory, which is required.
1/2
By Lemma 1.1, we can regard L(u, v) as a functional over the space W2 (Γ). Namely, taking
1/2
ϕ ∈ W2 (Γ), we can consider the form Lu (ϕ) = L(u, v), where v(x) ∈ W21 (G) is an arbitrary
1/2
extension of ϕ ∈ W2 (Γ) into the interior of the domain G.
1/2
We consider a particular extension v(x) = lG ϕ, where lG : W2 (Γ) → W21 (G) is a linear
bounded operator extension. We have

|Lu ϕ|  M1 ||u||1 ||v||1 = M1 ||u||1 ||lG ϕ||1  M2 ||u||1 ||ϕ||1/2 , (1.1)

1/2
where M1 and M2 are constants and ||ϕ||1/2 is the W2 (Γ)-norm of ϕ. Consequently, for every
fixed element u(x) ∈ W21 (G) the mapping ϕ → Lu (ϕ) is a linear continuous functional over the
1/2
space W2 (Γ). This functional is denoted by DΓ u, i.e.,

Lu (ϕ) = DΓ u, ϕΓ , (1.2)

1/2 −1/2
where ·, ·Γ means the duality of the pair (W2 (Γ), W2 (Γ)). By (1.1), the linear mapping
1 −1/2
u → DΓ u is bounded. Thereby DΓ u : W2 (G) → W2 (Γ) is the desired operator. In the above
arguments, the roles of u(x) and v(x) can be exchanged, which means that DΓ u is symmetric.
Thus, we have proved the first part of the theorem.
1/2
Let us prove the second part. We consider u(x) ∈ W22 (G) and ϕ ∈ W2 (Γ). Integrating by
parts, for any v(x) ∈ W21 (G) such that v|Γ = ϕ we have

Lu (ϕ) = [(∇u, ∇v) − (cirl u, cirl v) − div u div v]dx
G
   
2 ∂u
= (−Δu − cirl u + ∇ div u, v)dx + − [cirl u, n] − div u · n, ϕ dγ
∂n
G Γ
  
∂u
= − [cirl u, n] − div u · n, ϕ dγ,
∂n
Γ

which, in accordance with the definition of DΓ u (cf. (1.2)), means that

 
∂u
DΓ u = − [cirl u, n] − div u · n .
∂n Γ

The theorem is proved.

657
 We note that the results and methods of [1, 2] were used in the proof.
Remark 1.1. Since the operator DΓ u is symmetric, for any pair of vector-valued functions
u(x) ∈ W22 (G) and v(x) ∈ W22 (G)
     
∂u ∂v
− [cirl u, n] − div u · n, v dγ = u, − [cirl v, n] − div v · n dγ. (1.3)
∂n ∂n
Γ Γ

It is obvious that the identity (1.3) connects the boundary values of an arbitrary pair of
vector-valued functions of 3-d field theory and the boundary values of their gradients, curls, and
divergences. This unconditional connection “suggests” statements of boundary value problems
in field theory that are well posed in the sense of Hadamard–Petrovski or are normally solvable
in the sense of Hausdorff. Thereby the identity (1.3) can be regarded as a navigator for obtaining
statements of well-posed problems in 3-D field theory.

2 Applications
We introduce the space of three-dimensional vector fields
1
W2,0 = {u(x) ∈ W21 (G) : DΓ u = 0}

or, which is the same,

    
1
W2,0 = u(x) ∈ W21 (G) : (∇u, ∇v)dx = (cirl u, cirl v)dx + div u · div vdx ∀v ∈ W21 (G) .
G G G
1 ⊆ W 1 is a closed subspace.
It is obvious that W2,0 2
We note that from the symmetry property of DΓ u, i.e., from the identity

DΓ u, v|Γ  = u|Γ , DΓ v ∀v ∈ W21 (G),

it follows that the condition u|Γ = 0 implies that the operator DΓ u is trivial. In other words,
◦ ◦
{u ∈ W 12 } implies {DΓ u = 0}. Thus, W 12 ⊂ W2,0
1 ⊂ W 1 , i.e., W 1 is an intermediate space
2 2,0
◦
between W 12 and W21 . This space will be basic in the following problem.
1. A weak solution to the boundary value problem. In the classical setting, it is required to
find a vector-valued function u(x) = (u1 (x), u2 (x), u3 (x)) ∈ W22 (G) that simultaneously solves
the following two problems:
⎧ ⎧
⎨−Δu(x) = h(x), ⎨δu(x) = h(x),
 (∗)
⎩ ∂u  = ϕ(γ), ⎩([cirl u, n] + div u · n)Γ = ϕ(γ).
∂n Γ
We find a weak solution in W2,0 1 (G).

Definition 2.1. A vector-valued function u0 (x) ∈ W2,01 (G) is a weak solution to the problem
1 (G) the following integral identity holds:
(∗) if for any test function v0 ∈ W2,0
   
(∇u0 , ∇v0 )dx− ϕ, v0 |Γ  = (cirl u0 , cirl v0 )dx+ div u0 · div v0 dx− ϕ, v0 |Γ  = (h, v0 )dx.
G G G G

658
 −1/2
Here, h(x) ∈ L2 (G), ϕ, v0 |Γ  is the value of the functional ϕ ∈ W2,0 (Γ) at the “trace” of
1 (G) on the boundary Γ.
v0 ∈ W2,0
−1/2
Theorem 2.1. For any function h(x) ∈ L2 (G) and any functional ϕ(γ) ∈ W2,0 (Γ) such
that 
(h, c)dx + ϕ, cΓ = 0,
G

where c = (c1 , c2 , c3 ) is a constant vector-valued function, there exists a unique weak solution.
Moreover,

||u0 ||2L2 + ||∇u0 ||2L2 = ||u0 ||2L2 + || cirl u0 ||2L2 + || div u0 ||2L2  M (||h||L2 + ||ϕ||−1/2 ),

where M > 0 is a constant.

The proof of this theorem can be found in [3].

2. Connection with the Neumann problem. Let u(x) ∈ W21 (G) be a weak solution to the
Neumann problem
∂u 
−Δu(x) = h(x),  = 0,
∂n Γ
within the framework of the classical duality (W21 (G), W2−1 (G)). This means that u(x) satisfies
the identity  
(∇u, ∇v)dx = (h, v)dx (2.1)
G G

for any v(x) ∈ W21 (G).

Since 1 (G)
is a closed subspace of W21 (G), equipped with the inner
W2,0
1
product in W2 (G) generating the norm, we have the orthogonal decomposition

W21 (G) = W2,0

1 1
(G) ⊕ (W2,0 (G))⊥ .

Thus, any u(x) ∈ W21 (G) is represented as the orthogonal sum

u(x) = u0 (x) + u⊥ (x);

moreover, DΓ u0 (x) = 0 and, consequently, DΓ u⊥0 = DΓ u(x). In this case, the identity (2.1) takes
the form    
(∇u0 , ∇v0 )dx + (∇u⊥ 0 , ∇v ⊥
0 )dx = (h, v 0 )dx + (h, v0⊥ )dx
G G G G
or, which is equivalent,
 
1
(∇u0 , ∇v0 )dx = (h, v0 )dx ∀ v0 ∈ W2,0 (G), (2.2)
G G
 
(∇u⊥ ⊥
0 , ∇v0 )dx = (h, v0⊥ )dx, ∇v0⊥ ∈ (W2,0
1
(G))⊥ . (2.3)
G G

1 (G) is a weak solution to the problem (∗), i.e., a solution

By (2.2), the function u0 (x) ∈ W2,0
to the Neumann problem within the framework of duality (W2,0 1 (G), W −1 (G)). Similarly, the
2,0

659
 ⊥
function u⊥ 1
0 (x) ∈ W2,0 (G) is a weak solution to the Neumann problem within the framework
of duality ((W2,0 (G)) , (W2,0 (G))∗ ). It is obvious that the equality
1 ⊥ 1

(h, v0⊥ )dx = 0
G

for any v0⊥implies that u0 (x) ≡ u(x) up to an additive constant. The following assertion is also
true: If the right-hand side h(x) ∈ L2 (G) is orthogonal to the subspace (W2,0 1 (G))⊥ , then the

function u0 (x) found in Theorem 2.1 is a weak solution to the problem (∗) not only within the
framework of duality (W2,0 1 , W −1 ), but also in the sense of the classical duality of Sobolev spaces
2,0
(W21 , W2−1 ).
−1
3. Homogenous case. We consider a particular case of Theorem 1.1, where ϕ(γ) ∈ W2,0 (Γ) is
1/2
the restriction of some functional over the space W2 (Γ). Namely, we consider a weak solution
to the problem (∗) with
∂u 
 = DΓ w, ([cirl u, n] + div u · n)Γ = DΓ w, (2.4)
∂n Γ
where w ∈ W2! (G) is arbitrary. It is obvious that for any test function v0 (x) ∈ W2,0
1 (G)

DΓ w, v0  = w, DΓ v0  = 0.
Thus, for any function w(x) ∈ W21 (G) we obtain the same weak solution corresponding to the
case w(x) ≡ 0, i.e., the problem (∗) with the conditions
∂u0 
 = DΓ u0 , ([cirl u0 , n] + div u0 · n)Γ = DΓ u0 . (2.5)
∂n Γ
Therefore, if a weak solution u0 (x) ∈ W2,0 1 (G) satisfies the integral identity for any v (x) ∈
0
W21 (G), then the conditions (2.4) and (2.5) should lead to the same solution. Thus, unlike
1 (G), W −1 (G)), the study of the problem (∗) within the framework of duality
the duality (W2,0 2,0
(W21 (G), W2−1 (G)) leads to the nonuniqueness of a solution.
Remark 2.1. If a weak solution u0 (x) ∈ W2,0 1 (G) belongs to W 2 (G), then the conditions
2
(2.4) are equivalent to the conditions
∂u0 
 = 0, [cirl u0 , n]Γ = 0, (div u0 )|Γ = 0.
∂n Γ

4. The case of a polyhedron. Let G ⊂ R3 be a polyhedron, and let u(x) ∈ W2,0

1 (G) ∩ W 2 (G).
2
A direct calculation shows that
{DΓ u = 0} ⇔ {div τ1 ,τ2 utang = 0, ∇τ1 ,τ2 unorm = 0} (2.6)
on each face, where τ1 , τ2 are the Cartesian coordinates in the plane, utang is the tangential
component and unorm is the normal component of a vector-valued function u. Thus, the space
1 (G) is the closure in W 1 (G) of the subspace of vector-valued functions u ∈ W 2 (G) satisfying
W2,0 2 2
the conditions (2.6) on the faces of G. The corresponding problem takes the form
−Δu(x) = h(x), δu(x) = h(x)
with the conditions
∂u
|Γ = 0, (div utang )Γ = 0, (∇unorm )Γ = 0,
∂n
where div and ∇ are “complete” operations with respect to the three variables τ1 , τ2 , and n.

660
 Acknowledgment
The work is supported by the Russian Scientific Foundation (project No. 14-11-00306).

References

1. J. L. Lions and E. Magenes, Non-Homogeneous Boundary Value Problems and Applications,

Springer, Berlin etc. (1972).
2. R. Temam, Navier–Stokes Equations. Theory and Numerical Analysis, Am. Math. Soc.,
Providence, RI (2001).
3. Yu. A. Dubinskii, “On some formula in 3-D field theory and corresponding boundary value
problem,” J. Math. Sci., New York 219, No. 6, 959–966 (2016).

Submitted on October 4, 2017

661