Zhen Liu and Jinbiao Wu
Zhen Liu and Jinbiao Wu
Abstract
The vector potential is a fundamental concept widely applied across various fields. This paper
arXiv:2405.05228v1 [math-ph] 8 May 2024
presents an existence theorem of a vector potential for divergence-free functions in W m,p (RN , T)
with general m, p, N . Based on this theorem, one can establish the space decomposition theorem
for functions in W0m,p (curl; Ω, RN ) and the trace theorem for functions in W m,p (Ω) within the
Lipschitz domain Ω ⊂ RN . The methods of proof employed in this paper are straightforward,
natural, and consistent.
Keywords: Generalized Vector Potential Theorem; Trace Theorem; Decomposition Theorem;
Lipschitz Domain
1. Introduction
The vector potential plays a crucial role in physics and engineering, with applications
spanning electromagnetic phenomena [17,23], fluid dynamics [6], and quantum mechanics
[2,24]. Its versatility and importance arise from its ability to concisely represent intricate
vector fields, thereby enabling the analysis of various physical systems.
Given a divergence-free vector function v, it is cumbersome to construct a vector
potential w satisfying curl w = v. Additionally, some researchers are concerned not
only with the existence of the vector potential but also with its regularity. The following
theorem shows the existence of regular vector potentials for some divergence-free Sobolev
functions with p = 2 in R3 . There are many analyses and proofs of theorems related to
vector potential based on this result, see [4, 19].
Theorem 1.1. For any integer m ≥ 0, there exists a continuous mapping
L : H(div 0; R3 , R3 ) ∩ H m (R3 , R3 ) → Hloc
m+1
(R3 , R3 ),
such that curl Lv = v and div Lv = 0.
This paper mainly extends Theorem 1.1 to Theorem 3.3, demonstrating the existence
of a vector potential for functions in W m,p (RN , T) with m ≥ 0, 1 < p < ∞, and N ≥ 2.
This generalization is not straightforward. One challenge arises from the complexities
introduced in defining operators in any N dimensions. For instance, the action of the
curl operator on the vector will yield an anti-symmetric tensor, as defined in Definition
2.2. Another challenge is that certain properties of Hilbert spaces when p = 2 may not
extend to Sobolev spaces with general values of p.
Based on the results in Theorem 3.3, a decomposition result of W0m,p (curl; Ω, RN )
in Theorem 1.2 can be proved. An essential step involves in the proof extending the
function defined on Ω to RN and then utilizing Theorem 3.3 to determine a vector
potential. By following similarly procedures, this paper gives decomposition results for
Lp (curl; Ω, RN ) and W 1,p (curl; Ω, R3 ).
1
2 ZHEN LIU AND JINBIAO WU
2.1. Notations. In this paper, unless otherwise specified, lowercase English letters and
Greek letters denote scalars and scalar functions, respectively. Bold lowercase English
letters denote vectors or vector functions, while bold uppercase English letters denote
tensors or tensor functions of at least second order.
For 1 ≤ p < ∞ and m integer W m,p (D, X) denotes the Sobolev space of functions
within domain D, taking values in space X, of Lp (D, X) whose distributional deriva-
tives up to the order m also belong to Lp (D, X). Clearly, W 0,p (D, X) = Lp (D, X). The
corresponding norm and semi-norm are denoted as k·kW m,p (D,X) and |·|W m,p (D,X) , respec-
tively. Similarly, let C m (D, X) denote the space
T∞of m-times continuously differentiable
functions, taking values in X. Let C (D, X) = m=0 C m (D, X) and C0∞ (D, X) consists
∞
of all those functions in C ∞ (D, X) that has compact support in D. Let W0m,p (D, X)
be the closure of C0∞ (D, X) in the space W m,p (D, X). In this paper, X could be R, RN
or T, where T denotes the set of all skew-symmetry RN ×N matrices. If X = R, then
W m,p (D) abbreviates W m,p (D, X), similarly for C m (D). When p = 2, W m,2 (D, X)
becomes Hilbert spaces and denoted as H m (D, X). Define
H(div; RN , RN ) := {v ∈ L2 (RN , RN ); div v ∈ L2 (RN )},
H(div 0; RN , RN ) := {v ∈ H(div; RN , RN ); div v = 0},
which are used in Theorem 1.1.
In this paper, let Ω ⊂ RN denote a Lipschitz connected domain as defined in Definition
2.1, with an oriented compact boundary denoted as Γ.
3
2.2. Preliminaries. This subsection extends the definition of the curl operator to higher
dimensions and provides several identities and estimates of Newton potential functions
for subsequent use.
The following definitions and lemmas are adapted from [5, 16].
Definition 2.2 (curl operator in N dimensions). Define operator curl : C 1 Ω, RN →
C 0 (Ω, T) for a vector function f as anti-symmetric matrix function A with i-j-element
defined as follows
1 ∂fj ∂fi
Aij = [curl f ]ij := − , 1 ≤ i, j ≤ N.
2 ∂xi ∂xj
Unless otherwise specified, the Einstein summation convention is used in this paper.
Define the trace operator γt (f ) on Γ with i-j-element as 12 (fj ni − fi nj ). Following
Definition 2.2, introduce the spaces
Given two tensor functions A and B, in order to avoid unnecessary coefficient in the
following derivation, specially, let
Z
(2.2) (A, B)Ω := 2 Aij Bij dx.
Ω
This paper will frequently employ solutions of the Laplace equation to construct po-
tential functions satisfying the requirements. Consequently, the definition of the kernel
function in N -dimensional space is presented.
Definition 2.6 (Kernel function in N dimensions). Define the fundamental solutions of
the Laplace equation as follows
(
1
− 2π (log |x − y|) N = 2,
(2.3) λ(x − y) = 1 2−N
N (N −2)VN |x − y| otherwise,
For function φ(x) with compact support, define the Newton potential function using
the kernel function in (2.3) as
Z
ϕ(x) = φ(y)λ(x − y)dy.
RN
It is easy to verify that −△ϕ(x) = φ(x) and the following estimates hold.
2,p
Lemma 2.7 ( [15, Theorem 9.11]). Let Ω̃ be an open set in RN and ψ ∈ Wloc (Ω̃) ∩
p
L (Ω̃), 1 < p < ∞, be a strong solution of the equation −∆ϕ = φ in Ω̃. Then for any
domain Ω ⊂⊂ Ω̃, where Ω ⊂⊂ Ω̃ if the closure of Ω denoted as Ω̄ ⊂ Ω̃ and Ω̄ is a
compact subset of RN , it holds that
kϕkW 2,p (Ω) ≤ C(N, p, Ω, Ω̃) kϕkLp (Ω̃) + kφkLp (Ω̃) .
Lemma 2.8 (Calderon-Zygmund inequality [9]). Let φ ∈ Lp (RN ), 1 < p < ∞, and let
ϕ be the Newtonian potential of φ. Then there exists a constant C(N, p) such that
2
∂ij ϕ Lp (RN )
≤ C(N, p)kφkLp (RN ) , i, j = 1, 2, · · · , N.
Lemma 2.9. Let φ ∈ Lp (RN ), 1 < p < ∞, and let ϕ be the Newtonian potential of φ.
Then for any bounded domain Ω, there exists a constant C(N, p, Ω) such that
kϕkW 1,p (Ω) ≤ C(N, p, Ω)kφkLp (RN ) .
Theorem 3.1 (Weyl’s lemma [20, corollary 1.2.1]). Let ϕ : Ω → R be measurable and
locally integrable in Ω. Suppose that for all ψ ∈ C0∞ (Ω),
Z
ϕ(x)∆ψ(x)dx = 0.
Ω
Then ϕ is harmonic and, in particular, smooth.
Theorem 3.2 (The Stein extension theorem [1, Theorem 5.24]). Let Ω ⊂ RN be a
bounded Lipschitz domain and m ≥ 1 be an integer, 1 ≤ p ≤ ∞. Then Ω has a bounded
extension operator E : W m,p (Ω) → W m,p (RN ),i.e., for φ ∈ W m,p (Ω), there exists a
constant C such that
Eφ(x) = φ(x) a.e. in Ω, and kEφkW m,p (RN ) ≤ C(m, p)kφkW m,p (Ω) .
It is now appropriate to present the main theorem.
6 ZHEN LIU AND JINBIAO WU
Theorem 3.3 (Generalized vector potential theorem). Assume that vector function v
has compact support set in RN , v ∈ Lp (RN , RN ), and curl v ∈ W m,p (RN , T), then there
m+1,p
exists w ∈ Wloc (RN , RN ) such that curl w = curl v and div w = 0. Additionally, the
following estimate holds for every bounded domain Ω,
kwkW m+1,p (Ω,RN ) ≤ C(N, p, Ω) kvkLp (RN ,RN ) + k curl vkW m,p (RN ,T) .
Proof. The results of the above theorem will be demonstrated constructively. The con-
struction approach remains generally consistent across various values of m. However,
for m = 0, the overall lower regularity complicates the construction process. Hence, a
detailed description is provided specifically for the case of m = 0.
Step 1 constructs w ∈ Lp (RN , RN ) such that curl w = curl v and div w = 0. Define
Z
(3.1) η(x) = div λ(x − y)v(y)dy = div g(x).
RN
Let w = v + ∇η = v + ∇ div g. It follows by Lemma 2.8 that
(3.2) kwkLp (RN ,RN ) ≤ kvkLp (RN ,RN ) + C(N, p)kvkLp (RN ,RN ) ≤ C(N, p)kvkLp (RN ,RN ) .
Thus w ∈ Lp (RN , RN ). It is evident that curl ω = curl v because of curl(∇ div g) = 0.
By the definition of g(x) in (3.1), it follows that −△gi = vi , i = 1, 2, · · · , N . Thus,
∂ ∂ ∂ψ
(3.3) (∇gi , ∇ψ) = (∇gi , ∇ ψ) = (vi , ), i = 1, 2, · · · , N, ∀ψ ∈ C0∞ (RN ),
∂xi ∂xi ∂xi
which implies
∂ ∂ψ
(3.4) −(∇ gi , ∇ψ) = (vi , ), i = 1, 2, · · · , N, ∀ψ ∈ C0∞ (RN ).
∂xi ∂xi
It is important to note that the notation used in (3.3) and (3.4) do not employ Einstein
summation convention. Summing up (3.4) for i = 1, 2, · · · , N , it follows that
−(∇(div g(x)), ∇ψ) = (v, ∇ψ), ∀ψ ∈ C0∞ (RN ),
which proves that div w = 0.
1,p
Step 2 constructs w1 ∈ Wloc (RN , RN ) satisfies
(3.5) (∇w1 , ∇q) = (curl v, curl q), ∀q ∈ C0∞ (RN , RN ).
Let Z
w1 = curl∗ λ(x − y) curl v(y)dy = curl∗ H(x).
RN
Then for any bounded domain Ω, it follows by Lemma 2.8 and 2.9 that
kw1 kW 1,p (Ω,RN ) = kw1 kLp (Ω,RN ) + |w1 |W 1,p (Ω,RN )
≤ |H|W 1,p (Ω,T) + |H|W 2,p (Ω,T) ≤ C(N, p, Ω)k curl vkLp (RN ,RN ) .
1,p
Thus w1 ∈ Wloc (RN , RN ). Following similar procedures as Step 1, one can verify that
∗
−△w1 = curl curl v in weak sense, i.e.,
(∇w1 , ∇q) = (∇ curl∗ H, ∇q) = (div ∇ H, curl q) = (curl v, curl q), ∀q ∈ C0∞ (RN , RN ).
1,p
Step 3 proves w − w1 ∈ Wloc (RN , RN ). It follows from Step1 that
(3.6) (curl w, curl q)T + (div w, div q) = (curl v, curl q), ∀q ∈ C0∞ (RN , RN ).
7
There exists an open set Ω̃ such that Ω ⊂⊂ Ω̃ due to Ω is bounded. Owing to −△w2 = 0,
the function w2 has the smoothness enough for estimating. By Lemma 2.7,
(3.8) |w2 |W 1,p (Ω,RN ) ≤ kw2 kW 2,p (Ω̃,RN ) ≤ C(N, p, Ω, Ω̃)kw2 kLp (Ω̃,RN ) ,
(3.9) ≤ C(N, p)k curl vkLp (RN ,RN ) + C(N, p, Ω)kw2 kLp (Ω̃,RN )
≤ C(N, p, Ω) k curl vkLp (RN ,RN ) + kw1 kLp (Ω̃,RN ) + kwkLp (Ω̃,RN ) .
Employing the above generalized vector potential theorem, one can derive the general
decomposition result in Theorem 1.2. For clarity, the details of the case m = 1 are
provided therein.
Theorem 3.4. The following decomposition holds
W01,p (curl; Ω, RN ) = W02,p (Ω, RN ) + ∇(W02,p (Ω)),
where Ω ⊂ RN is a bounded Lipschitz domain.
8 ZHEN LIU AND JINBIAO WU
Then ṽ ∈ W 1,p (RN , RN ) and curl ṽ ∈ W 1,p (RN , RN ). By Theorem 3.3, there exists a
2,p
function w̃ ∈ Wloc (RN , RN ) such that
(3.12) curl w̃ = curl ṽ, div ω̃ = 0.
It follows from (3.12) that curl(ṽ − w̃) = 0 in RN . According to Assumption (A1), there
2,p
exists a function η ∈ Wloc (RN ) such that ∇η = ṽ − w̃. By (3.11), it follows that
v = w̃ + ∇η, in Ω,
(3.13)
0 = w̃ + ∇η, in Ωc .
2,p 3,p
Due to the restriction of w̃ to Ωc belongs to Wloc (Ωc , RN ) in (3.13), then η ∈ Wloc (Ωc ).
3,p
It follows by Theorem 3.2 that there exists a η̃ ∈ Wloc (RN ) which is the extension of η
from Ωc to RN . Note that η may not equal to η̃ in Ω though η = η̃ in Ωc . However, the
following relationship holds
(3.14) v = w̃ + ∇η = w̃ + ∇η̃ + ∇(η − η̃), in Ω.
2,p
Since w̃ + ∇η̃ ∈ Wloc (RN , RN ) and w̃ + ∇η̃ = 0 in Ωc , then w̃ + ∇η̃ ∈ W02,p (Ω, RN ) .
Similarly, η − η̃ ∈ W02,p (Ω) due to η ∈ W 2,p (Ω), η̃ ∈ Wloc
3,p
(R3 ) and η − η̃ = 0 in Ωc . Thus
the proof is completed by (3.14).
By employing procedures analogous to those detailed above, the proof of Theorem 1.2
can be established. It is crucial to recognize that a fundamental step in this proof is the
extension of a function from Ω to RN . Subsequent lemmas can be derived using similar
methods.
Theorem 3.5. For the bounded Lipschitz domain Ω ⊂ RN . The following decomposition
holds
Lp (curl; Ω, RN ) = W 1,p (Ω, RN ) + ∇(W 1,p (Ω)),
(3.15)
Lp0 (curl; Ω, RN ) = W01,p (Ω, RN ) + ∇(W01,p (Ω)).
Theorem 3.6. For the bounded Lipschitz domain Ω ⊂ R3 . The following decomposition
holds
(3.16) W 1,p (curl; Ω, R3 ) = W 2,p (Ω, R3 ) + ∇(W 2,p (Ω)).
Extending functions with zero boundary conditions is straightforward and leads di-
rectly to the result in the second line of (3.15). For the first row of (3.15), the ex-
tension of functions in Lp (curl; Ω, RN ) adheres to similar procedures as those detailed
in [10, Lemma 2.2]. With respect to the decomposition result (3.16), the methodologies
described in [19] are applicable to functions in W 1,p (curl; Ω, R3 ).
9
Sufficient proof for Theorem 4.2 is provided for N = 2 in [3]. This paper establishes
the sufficiency of Theorem 4.2 for arbitrary N . To demonstrate this sufficiency, the
following lemma is essential.
Lemma 4.3. For any i = 1, 2, · · · , N , let
φ0 = s · ei ,
N −1
(4.1) X
φ1 = (∂τj s · n)τj · ei + φ2 n · ei ,
j=1
then there exists a ωi ∈ W 2,p (Ω) such that γ0 (ωi ) = φ0 and γ1 (ωi ) = φ1 .
Proof. Only the lemma needs to be proved for the case i = 1. Recall the definition of s
and S in Theorem 4.2, it follows from (4.1) that
∇Γ φ0 + φ1 n = ∇Γ s · e1 + γ1 (ω1 )n
N
X −1
= ∇Γ s + ∂τj (s)n (n ⊗ τj ) + φ2 (n ⊗ n) · e1
j=1
= S · e1 ∈ W 1−1/p,p (Γ, RN ).
10 ZHEN LIU AND JINBIAO WU
By Theorem 4.1, which provides a sufficient condition for functions in W 2,p (Ω), the proof
is completed.
The demonstration of the sufficiency of Theorem 4.2 proceeds as follows.
Sufficient proof of Theorem 4.2. The proof can be divided into three steps.
Step 1 constructs v = (ω1 , ω2 , · · · , ωN ) − ∇θ ∈ W 1,p (Ω, RN ). By Theorem 4.1, there
exists a function θ ∈ W 2,p (Ω) satisfies γ0 (θ) = f0 and γ1 (θ) = f1 . Additionally, one can
construct (ω1 , ω2 , · · · , ωN ) ∈ W 2,p (Ω, RN ) using the trace function s due to Lemma 4.3.
Step 2 decomposes v = w + ∇η. It can be verified that
γ0 (vi ) = γ0 (ωi ) − γ0 (∇θ) · ei = s · ei − (∇Γ φ0 + φ1 n) · ei = 0.
Thus v ∈ W01,p (Ω, RN ). Additionally, one can verify that
[curl v]ij = [curl(ω1 , ω2 , · · · , ωN ) − curl(∇θ)]ij
1 ∂ωj ∂ωi
= − ∈ W 1,p (Ω, T), 1 ≤ i, j ≤ N.
2 ∂xi ∂xj
1
[γ0 (curl v)]ij = (γ0 (∇ωj ) · ei − γ0 (∇ωi ) · ej )
2
= S · ej · ei − S · ei · ej = 0, 1 ≤ i, j ≤ N.
It follows from S ∈ S that curl v ∈ W01,p (Ω, T). Theorem 3.3 implies that there exist
w ∈ W02,p (Ω, RN ) and η ∈ W02,p (Ω) satisfy
(4.2) v = w + ∇η, in Ω.
Combining (4.2) with v = (ω1 , ω2 , · · · , ωN ) − ∇θ yields
(4.3) ∇(η + θ) = (ω1 , ω2 , · · · , ωN ) − w, in Ω.
Step 3 verifies that φ = η + θ satisfies the following condition
γ0 (φ) = γ0 (η) + γ0 (θ) = 0 + φ0 = φ0 , on Γ.
γ1 (φ) = γ1 (η) + γ1 (θ) = 0 + φ1 = φ1 , on Γ.
(4.4) γ2 (φ) = ∇(∇φ) · n · n
= ∇(ω1 , ω2 , · · · , ωN ) · n · n − ∇w · n
= S · n · n = φ2 , on Γ.
The proof is completed by noting that φ ∈ W 3,p (Ω) since (4.3) and (4.4).
Let q > 0 be any positive integer. Introduce notations as follows
q
q
z }| {
(n⊗ ) := n ⊗ n ⊗ · · · ⊗ n,
q
z }| {
(n·q ) := n · n · · · · · n,
N
X −1
∇Γ (·) := τi ⊗ ∂τi (·).
i=1
11
References
[1] Robert A Adams and John JF Fournier. Sobolev spaces. Elsevier, 2003.
[2] Yakir Aharonov and David Bohm. Significance of electromagnetic potentials in the quantum theory.
Physical review, 115(3):485, 1959.
[3] Aissa Aibèche, Cherif Amrouche, and Bassem Bahouli. Trace operator’s range characterization for
sobolev spaces on lipschitz domains of R2 . Comptes Rendus. Mathématique, 361(G3):587–597, 2023.
[4] Cherif Amrouche, Christine Bernardi, Monique Dauge, and Vivette Girault. Vector potentials in
three-dimensional non-smooth domains. Mathematical Methods in the Applied Sciences, 21(9):823–
864, 1998.
[5] Douglas N Arnold. Finite element exterior calculus. SIAM, 2018.
[6] George Keith Batchelor. An introduction to fluid dynamics. Cambridge university press, 1967.
[7] Annalisa Buffa and Giuseppe Geymonat. On traces of functions in W 2,p (Ω) for lipschitz domains
in R3 . Comptes Rendus de l’Académie des Sciences-Series I-Mathematics, 332(8):699–704, 2001.
[8] Victor I Burenkov. Sobolev spaces on domains, volume 137. Springer, 1998.
[9] Alberto P Calderón and Antoni Zygmund. On the existence of certain singular integrals. In Selected
Papers of Antoni Zygmund, pages 19–73. Springer, 1952.
[10] Zhiming Chen, Qiang Du, and Jun Zou. Finite element methods with matching and nonmatching
meshes for maxwell equations with discontinuous coefficients. SIAM Journal on Numerical Analysis,
37(5):1542–1570, 2000.
[11] Ricardo G Durán and Marı́a Amelia Muschietti. On the traces of W 2,p (Ω) for a lipschitz domain.
Revista matemática complutense, 14(2):371–377, 2001.
[12] Emilio Gagliardo. Caratterizzazioni delle tracce sulla frontiera relative ad alcune classi di funzioni
in n variabili. Rendiconti del seminario matematico della universita di Padova, 27:284–305, 1957.
[13] Giuseppe Geymonat. Trace theorems for sobolev spaces on lipschitz domains. necessary conditions.
In Annales mathématiques Blaise Pascal, pages 187–197, 2007.
[14] Giuseppe Geymonat and Françoise Krasucki. On the existence of the airy function in lipschitz
domains. application to the traces of H 2 . Comptes Rendus de l’Académie des Sciences-Series I-
Mathematics, 330(5):355–360, 2000.
[15] David Gilbarg, Neil S Trudinger, David Gilbarg, and NS Trudinger. Elliptic partial differential
equations of second order. Springer, 1977.
12 ZHEN LIU AND JINBIAO WU
[16] Erhard Glötzl and Oliver Richters. Helmholtz decomposition and potential functions for n-
dimensional analytic vector fields. Journal of Mathematical Analysis and Applications, 525(2):Paper
No. 127138, 19, 2023.
[17] David J Griffiths. Introduction to electrodynamics. Cambridge University Press, 2023.
[18] Pierre Grisvard. Elliptic problems in nonsmooth domains. SIAM, 2011.
[19] Ralf Hiptmair. Finite elements in computational electromagnetism. Acta Numerica, 11:237–339,
2002.
[20] Jürgen Jost. Partial differential equations, volume 214. Springer Science & Business Media, 2012.
[21] Pier Domenico Lamberti and Luigi Provenzano. On trace theorems for sobolev spaces. Le Matem-
atiche, 75(1), 2020.
[22] Vladimir Maz’ya, Marius Mitrea, and Tatyana Shaposhnikova. The dirichlet problem in lipschitz
domains for higher order elliptic systems with rough coefficients. Journal d’Analyse Mathématique,
110(1):167–239, 2010.
[23] Wolfgang KH Panofsky and Melba Phillips. Classical electricity and magnetism. Courier Corpora-
tion, 2005.
[24] Anthony C Phillips. Introduction to quantum mechanics. John Wiley & Sons, 2013.