Poloidal-Toroidal Decomposition of Solenoidal Vector Fields in The Ball

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

ISSN 1990-4789, Journal of Applied and Industrial Mathematics, 2019, Vol. 13, No. 3, pp. 480–499.

c Pleiades Publishing, Ltd., 2019.


Russian Text 
c The Author(s), 2019, published in Sibirskii Zhurnal Industrial’noi Matematiki, 2019, Vol. XXII, No. 3, pp. 74–95.

Poloidal-Toroidal Decomposition
of Solenoidal Vector Fields in the Ball
S. G. Kazantsev1* and V. B. Kardakov2
1
Sobolev Institute of Mathematics, pr. Akad. Koptyuga 4, Novosibirsk, 630090 Russia
2
Novosibirsk State University of Architecture and Civil Engineering,
ul. Leningradskaya 113, Novosibirsk, 630113 Russia
Received April 8, 2019; in final form, April 8, 2019; accepted June 13, 2019

Abstract—Under study is the polynomial orthogonal basis system of vector fields in the ball
which corresponds to the Helmholtz decomposition and is divided into the three parts: potential,
harmonic, and solenoidal. It is shown that the decomposition of a solenoidal vector field with
respect to this basis is a poloidal-toroidal decomposition (the Mie representation). In this case, the
toroidal potentials are Zernike polynomials, whereas the poloidal potentials are generalized Zernike
polynomials. The polynomial system of toroidal and poloidal vector fields in a ball can be used for
solving practical problems, in particular, to represent the geomagnetic field in the Earth’s core.

DOI: 10.1134/S1990478919030098
Keywords: solenoidal, toroidal and poloidal vector fields, Mie representation, vector spherical
harmonic, Zernike polynomial

INTRODUCTION

We study the orthogonal polynomial basis of the basic space L2 (B 3 ) constructed earlier in [1] which
corresponds to the Helmholtz decomposition and is divided into the three parts: potential, harmonic, and
solenoidal. It is shown that the decomposition of a solenoidal vector field with respect to this basis is the
poloidal-toroidal decomposition. This presentation is also called the Mie decomposition or represen-
(N +k)
tation (see [2, 3]). In this case, the Zernike polynomials ZN  turn out to be the toroidal potentials,
[1](N +k)
while the generalized Zernike polynomials ZN  , the poloidal potentials. Also the construction of
other poloidal-toroidal bases is presented that is connected with using various spectral problems for
the vector Laplacian [4–6]. The polynomial system of toroidal and poloidal fields in a ball can be used,
in particular, to represent the geomagnetic field in the Earth’s core [7]. The decomposition of a solenoidal
vector field in spherical geometry onto poloidal and toroidal components is used in fluid mechanics and
magnetohydrodynamics (see the system of MHD equations) for the scalarization of the problem.

The following section contains the basic concepts and notation. The needed facts of functional
analysis are reminded as well. The definitions are given of the differential operations on the sphere,
the scalar and vector spherical harmonics. In Section 2, the polynomial bases of the basic scalar and
vector spaces L2 (B 3 ) and L2 (B 3 ) are considered. The last section is devoted to the poloidal-toroidal
decomposition of solenoidal vector field in the ball. The Appendix contains the basic properties of the
polynomial basis vector fields.
*
E-mail: kazan@math.nsc.ru

480
POLOIDAL-TOROIDAL DECOMPOSITION 481

1. PRELIMINARY INFORMATION AND NOTATION


Let Ox1 x2 x3 be a Cartesian coordinate system in the Euclidean space R3 with the standard basis
i, j, and k. We use the bold type to designate all vector quantities. The inner product of the vectors
x, y ∈ R3 with the coordinates x = (x1 , x2 , x3 ) and y = (y1 , y2 , y3 ) is defined in the usual manner:
x · y = x1 y1 + x2 y2 + x3 y3 . By B 3 = {x : |x| < 1} and S 2 = {ξ : |ξ| = 1} we denote the unit ball and
the unit sphere in R3 respectively. Here, |x| and |ξ| denote the Euclidean norms of vectors.
Let us recall the definition of the basic differential operations, ∇v, div , and rot . The operator ∇ or
the gradient calculates the gradient (potential) vector field
⎛ ∂v ⎞
⎜ ∂x1 ⎟
⎜ ∂v ⎟
⎜ ⎟
∇v := ⎜ ⎟,
⎜ ∂x2 ⎟
⎝ ∂v ⎠
∂x3
whereas the scalar function v is called the scalar potential. A smooth vector field a(x) = (a1 , a2 , a3 ) is
called solenoidal if its divergence equals zero (div a = 0), where the divergence operator div is defined
as
⎛ ⎞
⎜ a1 ⎟
⎜ ⎟ ∂a1 ∂a2 ∂a3
div a ≡ ∇ · a = div ⎜ a2 ⎟ := + + .
⎝ ⎠ ∂x1 ∂x2 ∂x3
a3
Another differential operator rot (the rotor operator) is defined in a short way via the symbolic
determinant
 
 
 i j k 
 ∂ ∂ 
 ∂
rot a := ∇ × a =  ,
 ∂x1 ∂x2 ∂x3 
 
 a a2 a 
1 3

here × means the vector product. If a vector field is potential then the rotor of such a field is equal to zero:
rot ∇v = 0.
The vector Laplace operator Δ is defined by the following formula via the component-wise application
of scalar Laplacian Delta
⎛ ⎞
⎜ Δa 1 ⎟
⎜ ⎟
Δa = (Δa1 , Δa2 , Δa3 ) = ⎜ Δa2 ⎟ .
⎝ ⎠
Δa3
The vector Laplace operator satisfies the vector identity
Δa = ∇ div a − rot rot a.
We now present the formulas for calculating ∇, div , rot , and Δ for a special class of scalar and
vector functions that will often be used below. Let a smooth scalar function f (s) of one variable be given
together with two constant vectors a and b. The functions of the form f (x · a) are called functions of
the plane wave type or ridge functions. Note the formulas

df (s) 
∇x f (x · a) = a = f  (x · a)a,
ds  s=(x·a)

div x f (x · a)b = f (x · a)(a · b),
rot x f (x · a)b = f  (x · a)(a × b),
Δx f (x · a) = f  (x · a)|a|2 .

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


482 KAZANTSEV, KARDAKOV

1.1. Sobolev Function Spaces in the Ball B 3


The paper uses the standard notation for the Sobolev spaces L2 (B 3 ), H 1 (B 3 ), and H01 (B 3 ) in the

3
scalar case and L2 (B 3 ) ≡ (L2 (B 3 ))3 , H1 (B 3 ) ≡ (H1 (B 3 ))3 , H10 (B 3 ) ≡ H10 (B 3 ) in the vector case.
The Lebesgue space L2 (B 3 ) is the Hilbert space of functions on the ball B 3 with the following inner
product and finite norm:

(u, v)L2 (B 3 ) = u(ξ) · v(ξ) dξ, u2L2 (B 3 ) = (u, u)L2 (B 3 ) . (1)
B3

Let
H 1 (B 3 ) ≡ H(∇) = {v ∈ L2 (B 3 ) : ∇v ∈ L2 (B 3 )}
be the Sobolev space of functions on B 3 . It is a complex Hilbert space with the standard inner product
and norm

(u, v)H 1 (B 3 ) = uv̄ dx + ∇u · ∇v dx, u2H 1 (B 3 ) = (u, u)H 1 (B 3 ) . (2)
B3 B3

Here, ∇u is the gradient of u, whereas the corresponding norm is denoted by u2H 1 (B 3 ) .


In the homogeneous space H01 (B 3 ) = {v ∈ H 1 (B 3 ) : v(ξ) = 0, ξ ∈ S 2 } the inner product (the
Dirichlet integral) and norm have the form

(u, v)H01 (B 3 ) = ∇u · ∇v̄ dx, uH 1 (B 3 ) = (u, u)H01 (B 3 ) = |∇u|2 dx.
2
(3)
0
B3 B3

This norm in H01 (B 3 ),


as a subspace of H 1 (B 3 ),
is equivalent to the norm of the basic space H 1 (B 3 ).
In the vector case, L2 (B 3 ) is the Hilbert space of vector functions defined in the ball B 3 , with the
inner product and finite norm

(u, v)L2 (B 3 ) = u(x) · v(x) dx, u2L2 (B 3 ) = (u, u)L2 (B 3 ) . (4)
B3

Denote by L2 (S 2 ) ≡ L2 (S 2 ; C) the space of the square-integrable functions on the sphere S 2 with


the inner product and finite norm

(u, v)L2 (S 2 ) = u(ξ)v(ξ) dξ, uL2 (S 2 ) = (u, u)L2 (S 2 ) ,
S2

where dξ is the Lebesgue measure on S 2 and S 2 dξ = 4π. For the vector fields on the sphere,
we consider the Hilbert space L2 (S 2 ) ≡ (L2 (S 2 ))3 with the inner product (·, ·)L2 (S 2 ) and the norm
 · L2 (S 2 ) :

(a, b)L2 (S 2 ) := a(ξ) · b(ξ) dξ, a2L2 (S 2 ) := (a, a)L2 (S 2 ) .
S2

1.2. The Helmholtz Decomposition


The important role in vector analysis is played by the harmonic vector fields h that satisfy the
homogeneous system
div h(x) = 0, rot h(x) = 0
and are both solenoidal and potential fields. From this system it follows that the harmonic vector field
satisfies the equation Δh = 0 with the vector Laplacian Δ, and each component h is an inner harmonic
function. The converse statement is not true. A harmonic vector field is a potential field with the

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


POLOIDAL-TOROIDAL DECOMPOSITION 483

harmonic potential h = ∇h, Δh = 0. We denote by ∇ Harm(B 


3 ) = {h = ∇h : h ∈ H 1 (B 3 ), Δh = 0}

the subspace of harmonic vector fields, and by ∇H01 (B 3 ) = p = ∇v : v ∈ H01 (B 3 ) , the subspace of
potential fields having only the normal component on the boundary. Then the orthogonal sum gives the
entire set of potential fields:
L2
∇H 1 (B 3 ) = {u = ∇v : v ∈ H 1 (B 3 )} = ∇H01 (B 3 ) ⊕ ∇ Harm(B 3 ).
Also let H0 (div = 0) be the subspace of solenoidal fields having only the tangential component on S 2 :
H0 (div = 0) := {v ∈ (L2 (B 3 ))3 : div v = 0, ξ · v(ξ) = 0 on S 2 }.
In terms of these subspaces, the Helmholtz decomposition means that (L2 (B 3 ))3 is the orthogonal
sum of the three subspaces [8–12]:
L2 L2
(L2 (B 3 ))3 = ∇H01 (B 3 ) ⊕ ∇ Harm(B 3 ) ⊕ H0 (div = 0).
Thus, each vector field f ∈ (L2 (B 3 ))3 can be uniquely represented as the sum of potential p, harmonic

h, and solenoidal J fields with only the tangential component on S 2 :

f = p + h + J. (5)
On the other hand, for each potential vector field p, there is a scalar function v, a scalar potential such
that p = ∇v. Similarly, for a solenoidal fields J there is a vector field u, called a vector potential, satisfying
the equation rot u = J. It is clear that the vector potential u for J is defined nonuniquely and, in some
cases, we can take as a vector potential a solenoidal vector field or impose some boundary conditions
on this potential. Therefore, (5) can be written in terms of the scalar and vector potentials. For example,
there exist a harmonic potential h ∈ Harm(B 3 ), a scalar potential v ∈ H01 (B 3 ), and a vector potential

3
u ∈ H01 (B 3 ) such that

h = ∇h, p = ∇v, J = rot u.
In this case, the potentials u, v, and h are uniquely determined.
An arbitrary solenoidal field
L2
J ∈ ∇ Harm(B 3 ) ⊕ H0 (div = 0)
can be expanded into the sum of toroidal and poloidal fields T and P:
J(x) = T(x) + P(x) = rot (xt(x)) + rot rot (xp(x)),
where t and p are suitable scalar functions. Thus, the toroidal-poloidal decomposition represents
a solenoidal vector field via two scalar potentials t and p. Further information about the toroidal-poloidal
decomposition is presented in Section 3.

1.3. Scalar and Vector Spherical Harmonics


Let ξ be a unit vector on a sphere. Then
⎛ ⎞
⎜ cos ϕ sin θ ⎟
⎜ ⎟
ξ = i sin θ cos ϕ + j sin θ sin ϕ + k cos θ = ⎜ sin ϕ sin θ ⎟ ,
⎝ ⎠
cos θ
where (ϕ, θ) is a polar system of coordinates on the sphere, 0 < ϕ < 2π is longitude, while 0 < θ < π is
latitude. The polar coordinate system (ϕ, θ) generates unit vector fields e1 (ξ) and e2 (ξ) on the sphere S 2 :
⎛ ⎞ ⎛ ⎞
⎜ − sin ϕ ⎟ ⎜ cos ϕ cos θ ⎟
1 ∂ ⎜ ⎟ ∂ ⎜ ⎟
e1 := ξ = ⎜ cos ϕ ⎟ , e2 := ξ = ⎜ sin ϕ cos θ ⎟ ,
sin θ ∂ϕ ⎝ ⎠ ∂θ ⎝ ⎠
0 − sin θ

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


484 KAZANTSEV, KARDAKOV

which form a local orthonormal basis in the tangent (tangential) plane drawn at the point ξ and
orthogonal to the vector ξ. The complex-valued spherical harmonics YN l are defined in the polar
coordinate system with help of the associated Legendre functions
YN l (ξ) = (−1)l kN l eilϕ PNl (cos θ), |l| ≤ N.
The external factor (−1)l here is called the Condon–Shortley phase factor, whereas kN l is the normal-
izing coefficient:

2N + 1 (N − l)!
kN l = .
4π (N + l)!

Under complex conjugation of a spherical harmonic, the rule YN l (ξ) = (−1)l YN,−l (ξ) holds as well as
the rule of parity YN l (−ξ) = (−1)N YN l (ξ). The lowest spherical harmonics (the angular momentum
N = 0, 1) have the form

1 3
Y00 (ξ) = √ , Y1,−1 (ξ) = (ξ1 − iξ2 ),
4π 8π
 
3 3
Y10 (ξ) = ξ3 , Y11 (ξ) = − (ξ1 + iξ2 ).
4π 8π
This shows that the point (vector) ξ ∈ S 2 can be written in terms of these spherical harmonics:
⎛ ⎞ ⎛ 1

√ Y1,−1 (ξ) − Y1,1 (ξ)
⎜cos ϕ sin θ⎟  4π ⎜ ⎜ i2

⎜ ⎟ ⎜

ξ = ⎜ sin ϕ sin θ ⎟ = √ Y1,−1 (ξ) + Y1,1 (ξ) ⎟ ⎟.
⎝ ⎠ 3 ⎜⎝ 2 ⎠
cos θ Y (ξ) 1,0

It is known that the subspace span{YN l , |l| ≤ N } of all spherical harmonics of degree N is an
eigensubspace of the Beltrami–Laplace operator Δξ with eigenvalue
−λ2N = −N (N + 1), ΔξYN l (ξ) = −N (N + 1)YN l (ξ).
The dimension of this subspace is 2N + 1, and so we can choose an orthonormal basis for it in various
ways. The scalar complex-valued spherical harmonics {YN l , |l| ≤ N }N ≥0 form an orthonormal basis
for L2 (S 2 ),

N 
YN l (ξ)YN  l (ξ) dξ = δN  δ  ,

S2

where δji is the Kronecker symbol.

Theorem 1 (the Funk–Hecke formula). Let f (t) ∈ L1 (−1, 1) be an integrable function of one
variable. Then for each spherical harmonic YN l of degree N we have
1
f (ξ · η)YN l (ξ) dξ = 2πYN l (η) f (t)PN (t) dt, (6)
S2 −1

where ξ · η is the inner product of unit vectors, whereas PN (·) is the Legendre polynomial of
degree N .
The Funk–Hecke formula is useful for calculating integrals over S 2 and plays an important role
in the theory of spherical harmonics. More detailed information about the Funk–Hecke formula,
as well as a general overview of spherical harmonics and the related problems, can be found in the
monographs [13–16]. Together with the scalar spherical harmonics YN l we will use the vector spherical

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


POLOIDAL-TOROIDAL DECOMPOSITION 485

harmonics, for whose definition we need the surface gradient operator ∇ξ and the rotated surface
gradient operator ∇⊥
ξ [17].

Definition 1. The surface gradient and the rotated surface gradient for smooth functions u(ξ) and
v(ξ) on S 2 are defined as follows:
∂u 1 ∂u
∇ξ u := e1 (ξ) + e2 (ξ), (7)
∂θ sin θ ∂ϕ
1 ∂v ∂v
∇⊥
ξ v = ξ × ∇ξ v := − e1 (ξ) + e2 (ξ), (8)
sin θ ∂ϕ ∂θ
where ξ = i sin θ cos ϕ + j sin θ sin ϕ + k cos θ.
Obviously,
ξ · ∇ξ u(ξ) = 0, ξ · ∇⊥
ξ u(ξ) = 0, ∇u · ∇⊥ u = 0.
Therefore, ∇u and ∇⊥ u are tangent (tangential) vector fields on S 2 . The vector field ∇⊥ v on the sphere
is obtained by rotation of ∇v by the angle π/2 in the tangent plane. The field ∇ξu is called poloidal,
electric, or potential; while ∇⊥ξ v, toroidal, magnetic, or solenoidal. The scalar functions u and v are
called the velocity potential and the flow function respectively. If we use the formulas for the components
of the spatial gradient ∇u in the polar coordinate system
∂u 1 ∂u 1 ∂u
(∇u)r = , (∇u)θ = , (∇u)ϕ = ,
∂r r ∂θ r sin θ ∂ϕ
then it is possible to decompose the scalar Laplace operator into the radial and angular parts:
 
ξ∂ ∇ξ
∇v = ∇x v = + v(x), x = rξ; (9)
∂r r
i.e., the surface gradient ∇ξ is the angular part of the spatial gradient ∇. For example,
 
ξ∂ ∇ξ N

∇(r YN l (ξ)) =
N
+ r YN l (ξ) = r N −1 N ξYN l (ξ) + ∇ξYN l (ξ) ,
∂r r (10)
x = rξ,
 
3 3
rY1,−1 (ξ) = (x1 − ix2 ), rY10 (ξ) = ξ3 ,
8π 4π

3
rY11 (x) = − (x1 + ix2 ),

 
3  3
∇rY1,−1 (ξ) = (1, −i, 0) , ∇rY10 (ξ) = (0, 0, 1) ,
8π 4π

3
∇rY11 (x) = − (1, i, 0) .

Definition 2. In the canonical coordinates, the surface divergence div ξ of a vector function v(ξ) =
v1 e 2 3 2
1 (ξ) + v e2 (ξ) + v ξ on the sphere S is defined as follows:
 
1 ∂ 1 ∂ 2
div ξ v := (v sin θ) + v + 2v 3 . (11)
sin θ ∂θ ∂ϕ

Finally, we define the Beltrami operator Δ ≡ Δξ , which is also called the Beltrami–Laplace operator:
1 ∂ ∂u 1 ∂2u
Δξ u(ξ) = div ξ ∇ξu(ξ) := sin θ + , (12)
sin θ ∂θ ∂θ sin2 θ ∂ϕ2

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


486 KAZANTSEV, KARDAKOV

i.e., the divergence of gradient will be the Laplace operator on the sphere S 2 . It is easy to check that
div ξ ∇⊥ ⊥
ξ u(ξ) = 0. Thus, we say that ∇ξ u is a potential field, while ∇ξ u is a solenoidal field on the
sphere S 2 . Additional information on the surface differential operators can be found, for example, in the
monographs [15–17].
Let us proceed to the definition of vector spherical harmonics (see [1, 15–17]).
Definition 3. Given ξ ∈ S 2 , put
(1) √
y00 (ξ) := ξY00 (ξ) = ξ/ 4π. (13)
(j)
For all N ∈ N and |l| ≤ N we define the vector spherical harmonics yN l (ξ) of the three types j = 1, 2, 3
on S 2 :
(1)
yN l (ξ) := ξYN l (ξ), (14)
(2)
yN l (ξ) := ∇ξ YN l (ξ), (15)

yN l (ξ) := ∇⊥
(3) (2)
ξ YN l (ξ) = ξ × yN l (ξ), (16)
where ∇ξ and ∇⊥
ξ are the surface gradient operators (7) and (8).
From the definition of vector spherical harmonics (13)–(16) it follows that
(1) (2) (3)
ξ × yN l (ξ) = 0, ξ · yN l (ξ) = 0, ξ · yN l (ξ) = 0.
(1) (2) (3)
Thus, yN l are the radial vector fields, whereas yN l and yN l are the tangent (tangential) vector fields on
the sphere S 2 . The system (13)–(16) forms an orthogonal basis in the function space L2 (S 2 ):
(1) (1)
N  l
yN l , yN  l L2 (S 2 ) = δN δl ,
(2) (2)
(3) (3)
N  l
yN l , yN  l L2 (S 2 )
= yN l , yN  l L2 (S 2 ) = N (N + 1)δN δl .

Vector spherical harmonics arise naturally. Thus, (10) can be written as


∇(r N YN l (ξ)) = r N −1 N yN l (ξ) + yN l (ξ) ,


(1) (2)
x = rξ,
and also

∇(r −N −1 YN l (ξ)) = r −N −2 − (N + 1)yN l (ξ) + yN l (ξ) ,


(1) (2)
x = rξ.
Put
(i) (1) (2) (e) (1) (2)
hN l (ξ) := N yN l (ξ) + yN l (ξ), hN l (ξ) := −(N + 1)yN l (ξ) + yN l (ξ).
(i) (e)
We see that the vector spherical harmonics hN l and hN l are related to harmonic vector fields, namely,
(i) (e)
hN l gives the boundary values of the field ∇(r N YN l (ξ)) inside the ball, whereas hN l gives the boundary
values of the field ∇(r −N −1YN l (ξ)) harmonic outside the ball. The new system of spherical harmonics
 (1) (i) (e) (3) 
y00 , hN l , hN l , yN l also forms an orthogonal basis in the function space L2 (S 2 ).
For vector spherical harmonics, the vector Funk–Hecke formulas [1, 15–16] hold. The vector Funk–
Hecke formulas will also be used to calculate spherical integrals, in particular, to calculate vector fields
in the spherical coordinates and their boundary values on the sphere S 2 .
Theorem 2 (The Funk–Hecke Formulas for Vector Spherical Harmonics). Let F (t) ∈ L1 (−1, 1) be
an integrable function of one variable on the interval (−1, 1). If N = 0 then
1
(1) (1)
y00 (η)F (ξ · η) dη = 2πy00 (ξ) sF (s) ds.
S2 −1

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


POLOIDAL-TOROIDAL DECOMPOSITION 487

For N ≥ 1 the following equalities hold for vector spherical harmonics (13)–(16):

(1) (1) (2)
yN l (η)F (ξ · η) dη = α11 2πyN l (ξ) + α12 2πyN l (ξ),
S2

(2) (1) (2)
yN l (η)F (ξ · η) dη = α21 2πyN l (ξ) + α22 2πyN l (ξ),
S2

(3) (3)
yN l (η)F (ξ · η) dη = α33 2πyN l (ξ),
S2
where the coefficients are calculated by the formulas
 1 1  1
1
α11 = N F (s)PN −1 (s) ds + (N + 1) F (s)PN +1 (s) ds = F (s)sPN (s) ds,
2N + 1
−1 −1 −1

 1 1  1
1 [1]
α12 = F (s)PN −1 (s) ds − F (s)PN +1 (s) ds =− F (s)PN +1 (s) ds,
2N + 1
−1 −1 −1
1
α21 = N (N + 1)α12 , α22 = α11 + α12 , α33 = F (s)PN (s) ds.
−1

Here
s
[1] PN +1 (s) − PN −1 (s)
PN +1 (s) := PN (t) dt = ,
2N + 1 (17)
1
[1] [1]
PN +1 (−1) = PN +1 (1) = 0, N ≥ 1.
(i) (e)
For the vector spherical harmonics hN l and hN l the following symmetric formulas hold:
1
(i) (i)
hN l (η)F (ξ · η) dη = 2πhN l (ξ) F (s)PN −1 (s) ds,
S2 −1
1
(e) (e)
hN l (η)F (ξ · η) dη = 2πhN l (ξ) F (s)PN +1 (s) ds.
S2 −1

2. POLYNOMIAL BASES FOR THE BASIC SPACES L2 (B 3 ) AND L2 (B 3 )


(N +2k) [1](N +2k)
2.1. Zernike Polynomials ZN l and Generalized Zernike Polynomials ZN l
Let us remind the definition of Zernike polynomials on B 3 (see [1]).
(N +2k)
Definition 4. The Zernike polynomials ZN l and the generalized Zernike polynomials
[1](N +2k)
ZN l are defined by the formulas

(N +2k) (3/2)
ZN l (x) := YN l (η)CN +2k (x · η) dη, N, k ≥ 0, |l| ≤ N, (18)
S2

[1](N +2k)
ZN l (x) := YN l (η)PN +2k (x · η) dη, N, k ≥ 0, |l| ≤ N, (19)
S2

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


488 KAZANTSEV, KARDAKOV
(3/2)
where Cm (·) are the Gegenbauer polynomials of degree m and type 3/2, whereas Pm (·) are the
Legendre polynomials. The degree of polynomials (18) and (19) are determined by the superscript
N + 2k.
(N +2k) [1](N +2k)
Using the Funk–Hecke formula (6), we can write ZN l and ZN l in the polar coordinates
x = rξ as
1
(N +2k) (3/2)
ZN l (x) = 2πYN l (ξ) PN (s)CN +2k (rs) ds, (20)
−1
1
[1](N +2k)
ZN l (x) = 2πYN l (ξ) PN (s)PN +2k (rs) ds (21)
−1

and calculate the boundary values for r = 1:


(N +2k)
ZN l (ξ) = 4πYN l (ξ),

⎨ 4π Y (ξ), k = 0, (22)
[1](N +2k) Nl
ZN l (ξ) = 2N + 1 ξ ∈ S2.

0, k ≥ 1,
The norm of Zernike polynomials depends on the superscript, the degree of the polynomial [1]:
 (N +2k)  4π
Z  =√ ,
Nl L2 (B 3 )
2N + 4k + 3
whereas, the boundary values (22), on the subscript.
Owing to the following property of the Legendre and Gegenbauer polynomials
(3/2) (3/2)
Cm (s) − Cm−2 (s)
Pm (s) =
2m + 1
[1](N +2k) (N +2k)
we can express the generalized polynomials ZN l in terms of the Zernike polynomials ZN l .
We have
(N ) (N +2k) (N +2k−2)
[1](N ) ZN l [1](N +2k) ZN l − ZN l
ZN l = , ZN l = , k ≥ 1. (23)
2N + 1 2N + 4k + 1
 (N +2k) 
The system of ZN l , |l| ≤ N N,k≥0 forms an orthogonal basis for the space L2 (B 3 ). The
 [1](N +2k) 
system ZN l , |l| ≤ N N,k≥0 will also be a basis for L2 (B 3 ). Owing to (23), the Gram–Schmidt
orthogonalization process for this system leads us to an orthogonal basis formed by the Zernike
(N +2k) (N )
polynomials ZN l . If k = 0 then ZN l are homogeneous harmonic polynomials (the Laplace spherical
(N ) (N )
functions), ΔZN l = 0 and ZN l (x) = 4π|x|N YN l (ξ), x = |x|ξ,, and form an orthogonal basis for the
subspace of harmonic functions
 (N ) 
Harm(B 3 ) = span ZN l , |l| ≤ N N ≥0 ,
the Bergman space. The harmonic Zernike polynomials of the first degree have the form
 
(1) 3 (1) 3
Z1,−1 (x) = 4π (x1 − ix2 ), Z10 (x) = 4π x3 ,
8π 4π

(1) 3
Z11 (x) = −4π (x1 + ix2 ).

(N )
[1](N ) ZN l [1](N )
For k = 0 we derive from (23) ZN l = , i.e., ZN l are harmonic functions.
2N + 1

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


POLOIDAL-TOROIDAL DECOMPOSITION 489
[1](N )
Eliminating the harmonic polynomials ZN l , we see that the system
 [1](N +2k) 
ZN l , |l| ≤ N N ≥0, k≥1

is a basis for the homogeneous space H01 (B 3 ) orthogonal with respect to (3). The orthogonality follows
from the equality

[1](N +2k) (1) (3/2) (N +2k−1)
∇ZN l (x) = yN l (η)CN +2k−1 (x · η) dη = AN l (x)
S2
(N +2k−1)
since the potential vector fields AN l are orthogonal in L2 (B 3 ) [1]:

 [1](N +2k)  2(2N + 4k + 1)
Z  = 4π ,
Nl L2 (2N + 4k − 1)(2N + 4k + 3)
 [1](N +2k)    4π
Z  1 = A(N +2k−1)  = √ .
Nl H Nl L2
0 2N + 4k + 1

2.2. Spectral Basis of the Spaces L2 (B 3 ) and H01 (B 3 )


(1) (2) (m) (m)
Let λN < λN < · · · < λN < . . . be the positive roots of the equation jN (λ) = 0, jN (λN ) = 0. It
 (m) 
is known (see [18]) that the system of eigenfunctions ψN l : |l| ≤ N N ≥0, m≥1 , where
(m) (m) (m) (m) (m)
−ΔψN l = λN ψN l , ψN l (x) = 4πiN jN (λN |x|)YN l (ξ) ∈ H01 (B 3 ),
forms a basis for L2 (B 3 ) and H01 (B 3 ) which is orthogonal both in L2 (B 3 ) and in the homogeneous
 space
H01 (B 3 ) in the sense of the inner products (1) and (3) respectively. Here jN (r) = π/(2r)JN +1/2 (r)
is the spherical Bessel function of the first kind, where Jν (z) are the Bessel functions of the first kind:

 (−1)k z N +2k+1/2
JN +1/2 (z) = .
2N +2k+1/2 k!Γ(N + k + 3/2)
k=0

The following formula gives the decomposition of the harmonic function (spherical polynomial) with
respect to the spectral basis:
∞ (m)

(N )
 YN l (ξ)jN λN r
ZN l (x) = 8π (m) (m)
, 0 ≤ r < 1. (24)
λ
m=1 N jN +1 λN

2.3. Solenoidal and Potential Polynomial Vector Fields in a Ball


In [1], for the space L2 (B 3 ) of vector fields in a unit ball, an orthogonal basis formed by polynomial
vector functions in accordance with the Helmholtz decomposition is constructed; i.e., with the division
of the basis into the three parts: potential, harmonic, and solenoidal. Namely,
 (N −1) 
HN l , |l| ≤ N N ≥1 is an orthogonal basis for the space of harmonic fields ∇ Harm(B 3 ),
 (N +2k+1) 
AN l , |l| ≤ N N,k≥0 is an orthogonal basis for the space of potential fields ∇H01 (B 3 ),
 (N +2k+1) (N +2k) 
BN l , CN l , |l| ≤ N N ≥1,k≥0 is an orthogonal basis for the space of solenoidal vector fields
H0 (div = 0).
Thus, in the basic space L2 (B 3 ) we have an orthogonal polynomial basis and for each vector function
f ∈ L2 (B 3 ) has a unique expansion (into a Fourier series) is true:

f (x) = p + h + J,

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


490 KAZANTSEV, KARDAKOV

where p is the potential part, h is the harmonic part, and J is the solenoidal part:
 ∞ 
∞  ∞ 

(N +2k+1)A (N +2k+1) (N −1)H (N −1)
p = ∇v = fN l AN l (x), h= fN l HN l (x),
N =0 k=0 l N =1 l

◦  ∞ 
∞  ∞ 
 ∞ 
(N +2k+1)B (N +2k+1) (N +2k)C (N +2k)
J= fN l BN l (x) + fN l CN l (x).
N =1 k=0 l N =1 k=0 l

Definition 5. Following [1], we define the vector fields on B 3 :



(N +2k−1) (1) (3/2)
AN l (x) := yN l (η)CN +2k−1 (x · η) dη, N, k ≥ 0, (25)
S2

(N +2k−1) (2) (3/2)
BN l (x) := yN l (η)CN +2k−1 (x · η) dη, N ≥ 1, k ≥ 0, (26)
S2

(N +2k) (3) (3/2)
CN l (x) := yN l (η)CN +2k (x · η) dη, N ≥ 1, k ≥ 0, (27)
S2
(−1)
where for convenience we assume A00 (x) ≡ 0.
Note that this definition takes into account all variants of integer variables N and k for which these
integrals exist and are not equal to zero. Similar to the Zernike polynomials, the superscript in the round
brackets indicates the degree of a vector polynomial of the variables x1 , x2 , and x3 . From the definition
of vector polynomials, we can get their explicit form, however, in this work, the explicit form is not used.
The integral definitions (25)–(27) allow us to study the properties of vector polynomials without using
their explicit form. The system of vector fields (25)–(27) allows another indexing system (n, k, l) under
which the vector polynomials of the same degree n are grouped:
(n)
An−1−2k,l , k = 0, . . . , [(n − 1)/2], |l| ≤ n − 1 − 2k, n = 1, 2, . . . ,
(n)
Bn+1−2k,l , k = 0, . . . , [n/2], |l| ≤ n + 1 − 2k, n = 0, 1, 2, . . . ,
(n)
Cn−2k,l , k = 0, . . . , [(n − 1)/2], |l| ≤ n − 2k, n = 1, 2, . . . .
Indexing in (25)–(27) is convenient by that it is indifferent to the parity of the polynomial degree.
It can be said that (25)–(27) are a generalization of (20) for the vector case. In the construction
of basis vector fields, the vector spherical harmonics and the Gegenbauer polynomials are used. It is
obvious that many properties of vector polynomials, like the Zernike polynomials, are connected with
(N +2k−1)
the properties of the Gegenbauer polynomials. For example, the polynomial vectors rot BN l and
(N +2k)
rot CN l can be calculated in the form of expansion with respect to the basis

(N +2k−1)

k−1
(N +2s)
rot BN l = (2N + 4s + 3)CN l ,
s=0
(28)
(N +2k)

k
(N +2s−1)
rot CN l =− (2N + 4s + 1)BN l ,
s=0

where

d (3/2) (5/2)

[(n−1)/2]
(3/2)
Cn (s) = 3Cn−1 (s) = (2n − 4s + 1)Cn−1−2s (s).
ds
s=0

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


POLOIDAL-TOROIDAL DECOMPOSITION 491

In particular, we have
(N +1) (N ) (N ) (N −1)
rot BN l = (2N + 3)CN l , rot CN l = −(2N + 1)BN l ,
(N +2) (N −1) (N +1)
rot CN l = −(2N + 1)BN l − (2N + 5)BN l .
(N +2k+1) (N +2k+1) (N +2k)
Under the restriction x = ξ ∈ S 2 we see that the vector fields AN l , BN l , and CN l for
k ≥ 0 coincide with the vector spherical harmonics (14)–(16):
(N +2k+1) (1)
AN l (ξ) = 4πyN l (ξ), (29)
(N +2k+1) (2)
BN l (ξ) = 4πyN l (ξ), (30)
(N +2k) (3)
CN l (ξ) = 4πyN,l (ξ). (31)

The first vector polynomials in (25) and (26) give us the harmonic vector fields

(N −1) (1) (3/2) (N −1) (2) (3/2)
AN l (x) = yN l (η)CN −1 (x · η) dη, BN l (x) = yN l (η)CN −1 (x · η) dη,
S2 S2

which assume the following values on the boundary:



(N −1) (1) (3/2) 4π (1) (2)
4π (i)
AN l (ξ) = yN l (η)CN −1 (ξ · η) dη = N yN l (ξ) + yN l (ξ) = h (ξ), (32)
2N + 1 2N + 1 N l
S2


(N −1) (2) (3/2)
BN l (ξ) = yN l (η)CN −1 (ξ · η) dη
S2
4π(N + 1) (1) (2)
4π(N + 1) (i)
= N yN l (ξ) + yN l (ξ) = hN l (ξ). (33)
2N + 1 2N + 1
This leads to the equality
(N −1) 1 (N −1)
AN l (x) = B (x).
N + 1 Nl
(N −1)
Therefore, when forming the basis, we should take into account only one of the following: AN l or
(N −1)
BN l .
(N −1)
Definition 6. As a standard harmonic vector field we take HN l :

(N −1) (1) (2)
(3/2) (i) (3/2)
HN l (x) := N yN l (η) + yN l (η) CN −1 (x · η) dη = hN l (η)CN −1 (x · η) dη. (34)
S2 S2

Then
(N −1) (N −1) (N −1)
HN l (x) = N AN l (x) + BN l (x).
On the boundary, the following equality holds:
(N −1) (N −1) (N −1) (1) (2)
(i)
HN l (ξ) = N AN l (ξ) + BN l (ξ) = 4π N yN l (ξ) + yN l (ξ) = 4πhN l (ξ).
Hence, we can write
(N −1) 1 (N −1)
AN l (x) = H (x),
2N + 1 N l
(N −1) N + 1 (N −1) (N −1) (N )
BN l (x) = H (x) ⇒ HN l (x) = ∇ZN l (x),
2N + 1 N l

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


492 KAZANTSEV, KARDAKOV
(N ) (N −1)
where ZN l is a harmonic Zernike polynomial. Let us calculate the norm HN l . We have

 (N −1)  2 2    (N −1)  √
B  = 16π N (N + 1) = N + 1 H(N −1)  ⇒ H  = 4π N .
Nl L2 (2N + 1)2 2N + 1 Nl L2 Nl L2

As an example, we consider the harmonic vector fields that are constant, N − 1 = 0. There should be
three of them: l = −1, 0, 1. Then
 
3  3
(0, 0, 1) ,
(0) (1) (0) (1)
H1,−1 = ∇Z1,−1 = 4π (1, −i, 0) , H10 = ∇Z10 = 4π
8π 4π

3
(1, i, 0) .
(0) (1)
H11 = ∇Z11 = −4π

3. POLOIDAL-TOROIDAL DECOMPOSITION OF A SOLENOIDAL VECTOR FIELD


(N +2k−1) (N +2k)
In this section, for solenoidal basis fields, we show that BN l are toroidal, while CN l are
poloidal vector fields, and, in result of expansion of a solenoidal vector field in the ball with respect to this
basis, we obtain the poloidal-toroidal decomposition.
Let t and p be scalar functions. Then the differential operators (7), (8), and (12) naturally arise when
calculating rot (xt(x)) and rot rot (xp(x)) using the spherical coordinates x = rξ (see formula (2.2)
in [19]). Namely, there hold the equalities
1 ∂t ∂t
rot (xt(x)) = −x × ∇t(x) = e1 − e2 = −∇⊥
ξ t, (35)
sin θ ∂ϕ ∂θ
 
ξ 1 ∂ ∂p 1 ∂p ξ 1 ∂
rot rot (xp(x)) = − Δξp + r e1 + e2 = − Δ ξ p + r∇ξp(rξ). (36)
r r ∂r ∂θ sin θ ∂ϕ r r ∂r
Note that x × ∇ is the angular momentum operator (the momentum operator) from quantum mechanics
and, under repeated application of it, we obtain the angular part of the Laplace operator:
1 ∂ ∂ 1 ∂2
Δξ = (x × ∇)2 = sin θ + .
sin θ ∂θ ∂θ sin2 θ ∂ϕ2

Definition 7. The solenoidal fields


T(x) = rot (xt(x)), P(x) = rot rot (xp(x))
are called toroidal and poloidal fields, respectively. The scalar functions t and p in these representations
are referred to as scalar potentials.
Taking (35) and (36) into account, we have
ξ 1 ∂
T(x) = −∇⊥
ξ t(rξ), P(x) = − Δξp + r∇ξp(rξ), x = rξ, 0 ≤ r ≤ 1. (37)
r r ∂r
The vector field P will be poloidal if it is the rotor of a toroidal field; in other words, if there exists a scalar
field p such that P(x) = rot rot (xp(x)). Thus, the rotor of a toroidal field is poloidal and, conversely, the
rotor of a poloidal field is a toroidal field. Each toroidal field is solenoidal since the divergence of a rotor
equals zero.
It can be seen from (37) that x · T(x) = 0. Therefore, the lines of force (current lines) of a toroidal
vector field lie on the spherical surfaces r = const ≤ 1, and this is used for visualization of such vector
fields. A solenoidal vector field T will be toroidal if and only if it is tangential in all spheres centered at
the origin, i.e., does not have a radial component.
When the vector Laplacian acts on toroidal and poloidal fields, the following relations hold [5, 20]:
If a field T is toroidal then ΔT = rot (xΔt(x)), and if a field P is poloidal then ΔP = rot rot (xΔp(x)).

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


POLOIDAL-TOROIDAL DECOMPOSITION 493

Definition 7 is symmetrical in the sense that the rotor of a toroidal field is a poloidal field and,
conversely, the rotor of a poloidal field is a toroidal one. Indeed,
rot rot rot (xp(x)) = −Δ rot (xp(x)) = − rot (xΔp(x)). (38)
Recall that every solenoidal field J can be decomposed into the sum of a toroidal T and a poloidal P
fields:
J(x) = T(x) + P(x) = rot (xt(x)) + rot rot (xp(x)),
where t and p are some suitable scalar functions. The toroidal-poloidal decomposition is unique if it is
required that the average values of scalar fields t and p turn to zero on every sphere of radius 0 < r ≤ 1.
Let us present the available method for solving the decomposition problem (see [4]; formula (2.39)
in [20]). Using the representations (37) in the sought-for decomposition, we obtain the equality
ξ 1 ∂
J(x) = T(x) + P(x) = − Δξp + ∇ξ p(rξ) + ∇ξp(rξ) − ∇⊥ ξ t(rξ), x = rξ.
r r ∂r
After inner multiplication by x, we have x · J(x) = −Δξp(x). Hence, the scalar function p can be
determined by the inversion of the Beltrami–Laplace operator. If we further consider the rotor of the
vector field J
rot J(x) = rot rot (xt(x)) + rot rot rot (xp(x))
ξ
= rot rot (xt(x)) − rot (xΔp(x)) = − Δξ t + . . . ,
r
then, in much the same way as above, we obtain the equation x · rot J(x) = −Δξ t(x). The scalar
toroidal potential t is also determined by the inversion of Δξ . Since the inverse Beltrami–Laplace
operator should be calculated for all values of r, 0 < r ≤ 1; therefore, the condition arises for the
decomposition uniqueness: The average values of the scalar fields t and p vanish on each sphere of
radius r:

t(rξ) dξ = p(rξ) dξ = 0, and also t(0) = p(0) = 0.
S2 S2

Theorem 3. Let functions f and F be given on [−1, 1] and f = F  . Then the vector field of the
(3)
form yN l (η)f (x · η) dη will be toroidal with the potential
S2

t(x) = − YN l (η)f (x · η) dη,
S2
(2)
whereas the vector field yN l (η)f (x · η) dη will be poloidal with the potential
S2

p(x) = YN l (η)F (x · η) dη :
S2
 
(3)
rot x YN l (η)f (x · η) dη = − yN l (η)f (x · η) dη, (39)
S2 S2
  
=t(x)
 
(2)
rot rot x YN l (η)F (x · η) dη = yN l (η)f (x · η) dη. (40)
S2 S2
  
=p(x)

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


494 KAZANTSEV, KARDAKOV

Proof. Let us prove (39). We have


 
rot x YN l (η)f (x · η) dη = − YN l (η)x × ∇x f (x · η) dη
S2 S2

=− YN l (η)x × ηf  (x · η) dη = YN l (η)η × xf  (x · η) dη
S2 S2


= YN l (η)η × (x − (x · η)η)f (x · η) dη = YN l (η)η × ∇η f (x · η) dη
  
S2 =∇η f S2

= YN l (η)∇⊥
η f (x · η) dη = − ∇⊥
η YN l (η)f (x · η) dη
S2 S2

(3)
=− yN l (η)f (x · η) dη.
S2

Here we used the formula of integration by parts on the sphere for the operator ∇⊥
ξ

u(ξ)∇⊥ξ v(ξ) dξ = − v(ξ)∇⊥ ξ u(ξ) dξ.
S2 S2

Note that, the formula of integration by parts has the form



u(ξ)∇ξ v(ξ) dξ = − v(ξ)∇ξ u(ξ) dξ + 2 ξu(ξ)v(ξ) dξ.
S2 S2 S2

In proving (40) we take into account the previous calculations, and infer that
 
(3)
rot rot x YN l (η)F (x · η) dη = − rot yN l (η)F (x · η) dη
S2 S2

(3)  (2)
= yN l (η) × ηF (x · η) dη = yN l (η)f (x · η) dη.
S2
S2

Theorem 3 is proved.

The use of (39) and (40) will be demonstrated by derivation of formula (38) which was mentioned
earlier. Let us apply the rotor operator to (40):
 
rot rot rot x YN l (η)F (x · η) dη = rot yN l (η)F  (x · η) dη
(2)

S2 S2
  
=p

yN l (η)F  (x · yN l (η)F  (x · η) dη
(2) (3)
= η× η) dη =
S2 S2
   
(39) 
= − rot x YN l (η)F (x · η) dη = − rot xΔ YN l (η)F (x · η) dη .
S2 S2

We now formulate the main result of this paper:

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


POLOIDAL-TOROIDAL DECOMPOSITION 495
(N +2k−1) (N +2k)
Theorem 4. The solenoidal vector fields BN l and CN l of the main basis (26) and (27)
(N +2k−1)
form an orthogonal system of poloidal and toroidal vector fields in L2 (B 3 ). Namely, BN l
is poloidal:
(N +2k−1) [1](N +2k)

BN l (x) = rot rot xZN l (x) , (41)


(N +2k)
whereas CN l is toroidal:
(N +2k) (N +2k)
(N +2k)
CN l (x) = − rot xZN l (x) = x × ∇ ZN l (x). (42)

(N +2k) [1](N +2k)


The average values of the scalar potentials −ZN l and ZN l on each sphere of radius
0 < r ≤ 1 are equal to zero:

[1](N +2k) (N +2k)
ZN l (rξ) dξ = ZN l (rξ) dξ = 0, N = 1, 2, . . . .
S2 S2

The expansion of the solenoidal vector field J ∈ L2 (B 3 ) with respect to the main basis is the
poloidal-toroidal decomposition
  ∞ 
∞  
(N +2k−1)B [1](N +2k)
J(x) = P(x) + T(x) = rot rot x JN l ZN l (x)
N =1 k=0 l
 ∞ 
 ∞  
(N +2k)C (N +2k)
+ rot −x JN l ZN l (x) ,
N =1 k=0 l

where P is the poloidal part and T is the toroidal part:


 ∞ 
∞  ∞ 
 ∞ 
(N +2k−1)B (N +2k−1) (N +2k)C (N +2k)
P= JN l BN l (x), T= JN l CN l (x).
N =1 k=0 l N =1 k=0 l

The expressions in parentheses define the poloidal and toroidal potentials.

(3/2)
Proof. We use Theorem 3 on putting f (s) = CN +2k (s) in (39) and F (s) = PN +2k (s) in (40). Now

(N +2k) (3/2)
ZN l (x) = YN l (η)CN +2k (x · η) dη,
S2

[1](N +2k)
ZN l (x) = YN l (η)PN +2k (x · η) dη.
S2

Theorem 4 is proved.

The general properties of poloidal and toroidal fields are now transferred to the vector polynomials
(N +2k) (N +2k−1)
CN l and BN l , for example,

(x) = ∇⊥
(N +2k) (N +2k)
CN l ξ ZN l (rξ),

(N +2k−1) ξ [1](N +2k) 1 ∂ [1](N +2k)


BN l (x) = − Δξ ZN l (rξ) + r∇ξZN l (rξ),
r r ∂r
(N +2k) (N +2k−1) [1](N +2k)
x · CN l (x) = 0, x · BN l (x) = −ΔξZN l (x),
which can also be obtained directly using the vector Funk–Hecke theorem.

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


496 KAZANTSEV, KARDAKOV

Suppose now that f (s) = eiλs and F (s) = eiλs /(iλ). Then, by Theorem 3, we have
 
(3)
rot x YN l (η)e iλx·η
dη = − yN l (η)eiλx·η dη,
S2 S2
 
(2)
rot rot x YN l (η)eiλx·η dη = iλ yN l (η)eiλx·η dη.
S2 S2

Introduce the notation



(2) (2) (3) (3)
JN l (x, λ) = yN l (η)eiλx·η dη, JN l (x, λ) = yN l (η)eiλx·η dη.
S2 S2

In result, we obtain the toroidal and poloidal fields depending on the parameter λ. From the Funk–Hecke
formulas it follows that
(3) (3)
JN l (x, λ) = 4πiN jN (λr)yN l (ξ),

(2) (1) (2)


JN l (x, λ) = α21 2πyN l (ξ) + α22 2πyN l (ξ)
(e)
(N + 1)jN −1 (λr) (i) h (ξ)
= 4πiN −1 hN l (ξ) + 4πiN +1 N jN +1 (λr) N l ,
2N + 1 2N + 1

YN l (η)eiλx·η dη = 4πiN jN (λ)YN l (ξ).
S2

In deriving these relations, we used the formula


1
PN (s)eiλrs ds = 2iN jN (λr).
−1

In what follows, the construction of a poloidal-toroidal basis is connected with the method of
choosing λ; i.e., it involves the application of spectral problems for the vector Laplacian.
1. Consider the case when the toroidal and poloidal fields vanish on the boundary of the sphere
(see [4, 5]). To do this, we need to determine λ. In the toroidal case, we have

(3) (3) (3)
JN l (x, λ) = yN l (η)eiλx·η dη = 4πiN jN (λr)yN l (ξ).
S2
(3)
Since we need JN l (x, λ) ∈ H10 (B 3 ), we obtain the condition
(m) (3) (m)
jN (λ) = 0 ⇒ λ = λN ⇒ JN l (ξ, λN ) = 0.
The toroidal potentials have the form
(m)
(m)

tN l x, λN = −4πiN jN λN r YN l (ξ). (43)

In the poloidal case, we have


(2) 4πiN −1 (N + 1) (i) 4πiN +1 N (e)
JN l (x, λ) = jN −1 (λr)hN l (ξ) + jN +1 (λr)hN l (ξ). (44)
2N + 1 2N + 1
First, we nullify the second term on the boundary, putting
(m) (m)

λ = λN +1 , jN +1 λN +1 = 0.

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


POLOIDAL-TOROIDAL DECOMPOSITION 497

We have
(m) (m)
(m) (2) (m)
(N + 1)λN +1 jN −1 (λN +1 ) (i)
iλN +1 JN l ξ, λN +1 = 4πiN hN l (ξ).
2N + 1
We see that this is the trace of a harmonic field since
(N −1)4π(N + 1) (i)
BN l (ξ) = h (ξ),
2N + 1 N l
(N −1) [1](N )
1 (N )

BN l (x) = rot rot xZN l (x) = rot rot xZN l (x) .


2N + 1
Hence the desired poloidal vector field
(2) (m)
(m) (2) (m)
(m) (N −1)
J̃N l x, λN +1 := iλN +1 JN l x, λN +1 − iN λN +1 BN l (x)
(2) (m)
(2) (m)

has the required property J̃N l ξ, λN +1 = 0 or J̃N l x, λN +1 ∈ H10 (B 3 ). In this case, the poloidal
potential has the form
(m)
(m)
(m) (m)
[1](N )
pN l x, λN +1 = 4πiN YN l (ξ)jN λN +1 r − iN λN +1 jN −1 λN +1 ZN l (x)
 (m) (m) 
(m) λ jN −1 (λN +1 ) N
= 4πiN YN l (ξ) jN (λN +1 r) − N +1 r
2N + 1
(m)
(m)

= 4πiN YN l (ξ) jN λN +1 r − jN λN +1 r N

since

(m)
λ(m) jN −1 (λN
(m)
+1 )
jN λN +1 = N +1 .
2N + 1

2. The next case was considered in [6]. As above, the toroidal potentials were given by (43), whereas,
for the poloidal fields, the condition was required of their continuous extension as harmonic fields outside
(m)
the sphere. To this end, we need to make vanish the first summand in (44) putting λ = λN −1 and
(m)

jN −1 λN −1 = 0. In result, we have

(2) (m)
4πiN +1 N (m)
(e)
JN l x, λN −1 = jN +1 λN −1 r hN l (ξ),
2N + 1
(m)
(m)
N −1 jN (λN −1 r)
pN l x, λN −1 = 4πi YN l (ξ) (m)
.
λN −1

CONCLUSION

The polynomial system of vector fields (25)–(27) is a complete orthogonal system in L2 (B 3 ),


and the Fourier expansion of a vector field with respect to this basis is performed by the Helmholtz
decomposition (5).

The article shows that the expansion of a solenoidal vector field with respect to this basis also
yields the poloidal-toroidal decomposition (the Mie decomposition). Thus, in the article, a polynomial
orthogonal system of toroidal and poloidal vector fields in a ball is proposed.

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


498 KAZANTSEV, KARDAKOV

APPENDIX: Basic Properties of Polynomial Vector Fields (25)–(27)


The harmonic and poloidal fields for N ≥ 1:
(N )
(N −1) [1](N ) ∇ZN l (N −1) (N −1)
AN l = ∇ZN l = , div AN l = 0, rot AN l = 0,
2N + 1
(2)

(N −1)   (N −1) 
(1)
4π N yN l + yN l   4π N
AN l S2
= , AN l L2
= .
2N + 1 2N + 1
The potential fields for N ≥ 0 and k ≥ 1:

(N +2k−1) [1](N +2k) (N +2k−1)



k−1
(N +2s)
AN l = ∇ZN l , div AN l = (2N + 4s + 3)ZN l ,
s=0

(N +2k−1) (N +2k−1)  (1)  (N +2k−1) 


A  4π
rot AN l = 0, AN l S2
= 4πyN l , Nl L2
=√ .
2N + 4k + 1
The harmonic and poloidal fields for N ≥ 1:
(N −1) [1](N ) (N −1) (N −1)
BN l = (N + 1)∇ZN l ,
div BN l = 0, rot BN l = 0,

(N −1) 
(1) (2)
4π(N + 1) N yN l + yN l
BN l S2
= ,
2N + 1

 (N −1)  4π(N + 1) N (N −1) [1](N )

B  = , BN l (x) = rot rot xZN l (x) .


Nl L2 2N + 1
The solenoidal and poloidal fields for N ≥ 1 and k ≥ 1:

(N +2k−1) (3) (N +2k−1)
BN l = − rot yN l (η)PN +2k (x · η) dη, div BN l = 0,
S2

(N +2k−1)

k−1
(N +2s) (N +2k−1)  (2)
rot BN l = (2N + 4s + 3)CN l , BN l S2
= 4πyN l ,
s=0

 (N +2k−1)  N (N + 1) [1](N +2k)

B  = 4π ,
(N +2k−1)
BN l (x) = rot rot xZN l (x) .
Nl L2 2N + 4k + 1
The solenoidal and toroidal fields for N ≥ 1 and k ≥ 0:

(N +2k) (2) (N +2k)
CN l = rot yN l (η)PN +2k+1 (x · η) dη, div CN l = 0,
S2

(N +2k)

k
(N +2s−1) (N +2k)  (3)
rot CN l =− (2N + 4s + 1)BN l , CN l S2
= 4πyN l ,
s=0

 (N +2k)  N (N + 1)
C  = 4π ,
Nl L2
2N + 4k + 3
(N +2k) (N +2k)
(N +2k)
CN l (x) = − rot xZN l (x) = x × ∇ZN l (x).
The harmonic and poloidal fields for N ≥ 1:
(N −1) (N ) (N −1) (N −1) (N −1)  (i)
HN l = ∇ZN l , div HN l = 0, rot HN l = 0, HN l S2
= 4πhN l ,
 (N −1)  √ (N −1) 1 (N )

H  = 4π N , HN l (x) = rot rot xZN l (x) .


Nl L2 N +1

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019


POLOIDAL-TOROIDAL DECOMPOSITION 499

REFERENCES
1. E. Yu. Derevtsov, S. G. Kazantsev, and Th. Schuster, “Polynomial Bases for Subspaces of Vector Fields in
the Unit Ball. Method of Ridge Functions,” J. Inverse Ill-Posed Problem 15 (1), 19–55 (2007).
2. G. E. Backus, “Poloidal and Toroidal Fields in Geomagnetic Field Modeling,” Rev. Geophys. 24, 75–109
(1986).
3. G. Backus, R. Parker, and C. Constable, Foundations of Geomagnetism (Univ. Press, Cambridge, 1996).
4. V. M. Bykov, “The Stokes Flows in a Ball,” Prikl. Mekh. Tekhn. Fiz. No. 2, 65–70 (1980).
5. B. Rummler, “The Eigenfunctions of the Stokes Operator in the Open Unit Ball and in the Open Spherical
Annulus,” in 8th Asian Computational Fluid Dynamics Conference, Hong Kong, 10–14 January, 2010,
ACFD0163-T001-A-001-3.
6. G. M. Vodinchar and L. K. Krut’eva, “Basis Systems for a Geomagnetic Field,” Vestnik KRAUNTs. Fiz.-
Mat. Nauki No. 1(1), 24–30 (2010).
7. P. H. Roberts and E. King, “On the Genesis of the Earth’s Magnetism,” Reports on Progress in Physics
76 (9), 096801 (2013).
8. G. Veil’, Method of Orthogonal Projection in Theory of Potential. Ser. Math. Theor. Phys. (Nauka,
Moscow, 1984) [in Russian].
9. G. Auchmuty, “Orthogonal Decompositions and Bases for three-Dimensional Vector Fields,” Numer. Funct.
Anal. Optim. 15 (5–6), 455–488 (1994).
10. H. Kozono and T. Yanagisawa, “Lr -Variational Inequality for Vector Fields and the Helmholtz–Weyl
Decomposition in Bounded Domains,” Indiana Univ. Math. J. 58 (4), 1853–1920 (2009).
11. Ch. J. Amick, “Decomposition Theorems for Solenoidal Vector Fields,” J. Lond. Math. Soc. No. 2, 288–296
(1977).
12. V. Girault and P.-A. Raviart, Finite Element Methods for Navier–Stokes Equations. Theory and
Algorithms (Springer, Berlin, 1986).
13. K. Atkinson and H. Han, Spherical Harmonics and Approximations on the Unit Sphere: An Introduc-
tion, Vol. 2044 (Springer, Berlin, 2012).
14. C. Muller, Spherical Harmonics (Springer, Berlin, 1966).
15. W. Freeden and M. Schreiner, Spherical Functions of Mathematical Geosciences. A Scalar, Vectorial,
and Tensorial Setup (Springer, Berlin, 2009).
16. D. A. Varshalovich, A. N. Moskalev, and V. K. Khersonskii, Quantum Theory of Angular Momentum
(Nauka, Leningrad, 1975) [in Russian].
17. J.-C. Nedelec, Acoustic and Electromagnetic Equations: Integral Representations for Harmonic Prob-
lems (Springer, New York, 2001).
18. O. A. Ladyzhenskaya, Boundary Value Problems of Mathematical Physics (Nauka, Moscow, 1973) [in
Russian].
19. D. J. Ivers, “Kinematic Dynamos in Spheroidal Geometries,” Proc. Roy. Soc. A473. 20170432 (2017);
http://dx.doi.org/10.1098/rspa.2017.0432
20. G. Moffat, Magnetic Field Excitation in a Conducting Medium (Mir, Moscow, 1980) [in Russian].

JOURNAL OF APPLIED AND INDUSTRIAL MATHEMATICS Vol. 13 No. 3 2019

You might also like