Computer methods

in applied
mechanics and
engineering
ELSEVlER Comput. Methods Appl. Mech. Engrg. 169 (1999) 391-405

Edge element formulations of eddy current problems

Oszk~ Bfr6
IGTE, Technical University of Graz, Kopernikusgasse 24, A-8IO Graz, Austria

Received 15 December 1997

Abstract

Various formulations of eddy current problems in terms of scalar and vector potentials are reviewed in the paper. The vector potentials are
approximated by edge finite elements and the scalar potentials by nodal ones. The formulations are ungauged, leading, in most cases, to
singular finite element equations systems. Special attention is paid to ensuring that the right-hand sides of the equations systems are
consistent. The resulting numerical schemes prove to be as robust as the corresponding Coulomb gauged approaches realized by nodal
elements. 1999 Elsevier Science S.A. All rights reserved.

1. Introduction

The numerical analysis of three-dimensional eddy current problems with the aid of the finite element method
has been one of the main directions of research in computational electromagnetics during the past decade.
Applications in mainstream use vector and scalar potential functions to describe the field quantities. Finite
element techniques using nodal based functions to approximate both scalar and vector potentials (nodal finite
elements) were first to emerge [1-5]. Various possible formulations were proposed by the present author in terms
of different potential functions. The most outstanding feature of these approaches is the incorporation of the
Coulomb gauge on the vector potentials in the governing equations [6-8]. This technique is consistent with
nodal finite elements and leads to robust numerical realizations.
The use of nodal finite elements to approximate vector potentials leads to difficulties when applied to two
specific types of problems. One of these involves highly permeable conducting regions (iron parts) surrounded
by nonmagnetic and nonconducting domains (e.g. air) with a magnetic vector potential A used to describe the
field in iron. The use of a magnetic scalar potential qb in air in the vicinity of the iron/air interface is not
desirable because of the weak coupling between the two potentials [9]. The remaining option of employing A in
air, too, leads to considerable inaccuracies at the interface due to the continuity of the normal component A n of
the magnetic vector potential resulting in a natural interface condition which requires a j u m p in the divergence
of A [9,10]. The problem can be alleviated by relaxing the continuity of An; this is, however, not a natural
approach in a nodal finite element context. A second problem not easily treated with the aid of nodal finite
elements arises when singularities are to be approximated at sharp corners along which the tangential
components of a vector potential vanish and this vector potential describes the field on both sides of the
interface constituted by the corner. Indeed, the continuity of the normal component renders the field obtained as
the curl of the vector potential to be zero at the corner instead of an expected high value. A typical example is
the modeling of thin flaws in a conducting piece, in which the eddy currents are derived from a current vector
potential T, by setting the tangential components of T to zero.
The above problems are avoided if the vector potentials are approximated by edge (tangential vector) finite
elements [11-13]. The reason is that these enforce the physically necessary continuity of the tangential
components of the vector potentials only, but not that of the normal component. However, A or T cannot be
gauged in the same way as in formulations corresponding to nodal finite element approaches since the lack of

0045-7825/99]$ - see front matter 1999 Elsevier Science S.A. All rights reserved.
PII: S0045-7825 (98)00165-0
392 O. B{r6 / Comput. Methods Appl. Mech. Engrg. 169 (1999) 391-405

continuity of their normal components implies that their divergence cannot be defined at element interfaces. One
of the most outstanding properties of edge finite elements is that the curls of the vector basis functions are not
linearly independent although the basis functions themselves are. This holds both for the element shape
functions, each of which is defined in a single finite element, and also for the global basis functions associated
with the edges of the global finite element mesh, which are defined to coincide with the shape functions in an
assembly of finite elements containing the relevant edge. The maximal set of linearly independent curls of the
basis functions is constituted by those associated with the edges in a cotree of the graph defined by the finite
element mesh. This fact can be used to gauge the vector potentials by setting the degrees of freedom
corresponding to the tree edges to an arbitrary value (e.g. to zero) thus eliminating the linearly dependent curls
from among the basis functions [14]. An appropriate algebraic reduction of the basis functions has been shown
in [15] to lead to the Coulomb gauge, although this procedure destroys the sparsity of the finite element
matrices.
The interdependence of the curls of the edge basis functions means that an ungauged formulation leads to a
singular finite element curl-curl matrix. It has been shown in a static context [16] that iterative methods like the
Incomplete Cholesky Conjugate Gradient (ICCG) technique lend themselves better to the solution of such
singular (but positive semi definite) systems than to that of the nonsingular system arising if the tree-cotree
gauging method of [14] is employed. Singular systems can obviously be only solved if the right-hand side is
consistent. Even minor inconsistencies lead to the failure of the iterative procedure [16].
The aim of this paper is to review all possible formulations of eddy current problems in terms of scalar and
vector potentials with the vector potentials approximated by edge finite elements and the scalar potentials by
nodal ones. The formulations are ungauged, leading in most cases to singular finite element equations systems.
Special attention will be paid to ensuring that the right-hand sides of the equations are consistent. If this is the
case, the numerical stability of the formulations turns out to be comparable to that of the corresponding
Coulomb gauged approaches realized by nodal elements.

2. Problem definition

An eddy current problem arises if a time varying magnetic field is excited within a body made of conducting
material. The displacement current density is assumed to be negligible, i.e. the differential equations of
quasi-static fields hold. This leads to a static but time varying magnetic field in the nonconducting regions O n
surrounding the eddy current carrying conductors constituting the region ~(2~. In $2, the current density
distribution is unknown, an eddy current field is present here.
The excitation can be of three types. The first possibility called coil excitation is realized by a coil with given
current density Jo situated in ~2. If the total current through a conductor is prescribed but the current density
distribution herein is unknown, a skin effect problem with current excitation is obtained. If, conversely, the
voltage of the conductor is given, a skin effect problem with voltage excitation is arrived at. Most attention will
be paid to the case of coil excitation, skin effect problems are conveniently treated by different potential
formulations (see Section 5).
The time variation will be assumed to be sinusoidal with an angular frequency of to, complex notation will be
hence used throughout the paper. The equations are thus valid for linear material characteristics only. The
generalization to nonsinusoidal time variation and to nonlinear problems is, however, straightforward.
The differential equations to be solved are, hence, the following:

curlH=J0 ing2, (1)

divB=0 ing2 n, (2)

curlH = J in g2~, (3)

curl E = - j w B in/2 C, (4)

divJ=0 in~, (5)

divB=0 inOc (6)

 O. B{r6 / Comput. Methods Appl. Mech. Engrg. 169 (1999) 391-405 393

where H is the magnetic field intensity, B is the magnetic flux density, E is the electric field intensity and J is
the current density. Eqs. (5) and (6) obviously follow from (3) and (4). The field quantities satisfy the following
constitutive equations:

B =/zH, H = t,B in f2,, and in ~(2,, (7)

J = crE , E = pJ inO (8)

where # is the permeability, u the reluctivity, o- the conductivity and p the resistivity.
The effect of the given current density of the coils will be assumed to be described by an impressed field
quantity T O satisfying

curl T O= J0- (9)

Its introduction eliminates the necessity of modeling the geometry of the coils in the finite element mesh. It is
always possible to make the choice T O= H s where H s is the field of the coils in free space which can be
computed from the Biot-Savart law. For simple coil geometries it is possible to construct a function T O
satisfying (9) without having to resort to Biot-Savart integration [17]. Frequently it is advantageous to use an
interpolation of T O with edge element shape functions instead of its exact value. This places T O in the same
function space as the gradients of the nodal shape functions and avoids cancellation errors in highly permeable
nonconducting regions [17].
The boundary conditions of the problem can be given at artificial far boundaries or along symmetry planes.
For simplicity, homogeneous boundary conditions will be assumed along them (n is the outer normal unit
vector), i.e. on F,, the outer boundary of ~n, and on Fc, the outer boundary of ~2,. we have:

HXn=0 or B.n=O onF~, (10)

HXn=O or EXn=0 onF~. (11)

In some cases the equations will be simplified if these are replaced by

HXn=ToXn or B.n=#To.n onF,, (10a)

HXn=ToXn or EXn=O on~. (lla)

Since the function T O must satisfy the symmetry conditions on symmetry planes, (10) and (lOa) as well as (11)
and (1 la) are equivalent there. Far boundaries are assumed to be placed at such distances from the excitations
that the differences between (10) and (10a) as well as between (11) and (1 la) do not affect the results of the
analysis appreciably.
The continuity conditions on the interface F~c between the conducting and nonconducting regions are

H X n and B . n are continuous on F,c. (12)

These interface conditions will provide the coupling between the formulations in the nonconducting eddy
current free region ~n and the conducting eddy current region .(2c.
The solution of the differential equations (1)-(4) with the boundary and interface conditions (10)-(12) is
unique [8].
The potential formulations and the ways of their coupling will be reviewed in Sections 4, 5 as well as 6 by
introducing the relevant potential functions, writing the differential equations, boundary and interface conditions
with these and then formulating the finite element Galerkin equations.

3. Edge and nodal basis functions

The vector potentials will be approximated by edge basis functions N i (i = 1, 2 . . . . . ne) and the scalar
potentials by nodal basis functions N~ (i = 1,2 . . . . . nn) where n e and n n are the numbers of edges and nodes in
the finite element mesh, respectively. Both the rte edge and n n nodal basis functions are linearly independent, but
394 O. Bir6 / Comput, Methods Appl. Mech. Engrg. 169 (1999) .7;91-405

there are linear interdependencies among the gradients of the nodal basis functions and among the curls of the
edge basis functions.
In fact, since the sum of all nodal basis functions is 1:
nn

~] N , = 1, (13)
i=1

the following is obtained after taking the gradient of (13):

nn

grad N, = 0. (14)
i=1

This means that the maximal number of linearly independent gradients of the nodal basis functions is n, - 1, i.e.
the number of tree edges in the graph defined by the finite element mesh. It follows from the linear
independence of the nodal basis functions that n, - 1 of their gradients are in fact linearly independent.
On the other hand, the gradients of the nodal basis functions are in the function space spanned by the edge
basis functions, i.e. we have the following n n - 1 linearly independent relations:
n e

gradN, = ~] ci~Nk, i = 1,2 . . . . . n, - 1 (15)

k=l

where
n e

y 2c i k > 0 , i=1,2 ..... n,-1. (16)

k=l

Taking the curl of each of the equations in (15) results in

n e

ci~curiNk=O, i = 1 , 2 . . . . . nn - 1 . (17)
k=l

Together with (16) and with the linear independence of the equations in (15), this implies that the maximal
number of linearly independent curls of the edge basis functions is n~ - (n,, - 1), i.e. the number of cotree edges
in the graph of the finite element mesh. Since there are no more linearly independent gradients in the space
spanned by the edge basis functions than n , , - 1, not less than n e - ( n , - 1) of the curls of the edge basis
functions are linearly independent.

4. F o r m u l a t i o n s in e d d y c u r r e n t f r e e r e g i o n s

There are two ways to describe the static magnetic field in the nonconducting eddy current free region /2.'
either by a magnetic scalar potential or by a magnetic vector potential. In writing these formulations, the
boundary conditions (10) or (10a) will be assumed to hold, and the interface F,c to the eddy current region S2
will be disregarded. The latter will be treated in Section 6.

4.1. The @-formulation

The magnetic scalar potential @ is defined by

H = T O - grad q5 (18)
where T o obeys (9). If ferromagnetic materials are present within the problem region, T o is represented with the
aid of the edge basis functions as
n e

T o = ~, tkNk (19)
k=l

where t k are the line integrals of T o along the edges [17].

 O. Bfr6 / Comput. Methods Appl. Mech. Engrg. 169 (1999) 391-405 395

The assumption (18) results in the exact satisfaction of (1), and writing (2) with the first relationship in (7)
used leads to the differential equation

div(/z grad 4)) = div(/zT0) in ,(2,. (20)

Using the first boundary condition in (10a) and the second one in (10) we have the following Dirichlet or
Neumann boundary conditions on the scalar potential:
q0=0 or txgradqb.n=IzTo.n on/~,,. (21)

The scalar potential is approximated by nodal basis functions as

n n

q) ~ qO~"~= ~ qOkN~ (22)

k-I

where qOk are the nodal values of q0~').

Setting (22) into (20) as well as into the Neumann boundary condition in (21) and using the basis functions N,
as weighting functions, the following Galerkin equations are obtained:

-fa gradN"/zgradq~'")dS2=-fn gradNi'/xTod/2' i = 1 , 2 . . . . . n. (23)

n n

where use has been made of the fact that the basis functions N~ satisfy the homogeneous Dirichlet boundary
conditions in (2 l).

4.2. The A-formulation

The magnetic vector potential A is defined by
B = curl A . (24)
This results in the exact satisfaction of (2), and writing (1) with the second relationship in (7) used leads to the
differential equation
curl0, c u r l A ) = curl T o in/2~ (25)

where T o has been introduced as in (9).

Using the second boundary condition in (10) and the first one in (10a) we have the following Dirichlet or
Neumann boundary conditions on the vector potential:
An=0 or pcurlAn=T on onF,. (26)

The vector potential is approximated by edge basis functions as

t/e

a ~ a ")= ~ akNk (27)

k-I

where a k are the line integrals of A along the edges.

Setting (27) into (25) as well as into the Neumann boundary condition in (26) and using the basis functions
N i as weighting functions, the following Galerkin equations are obtained:

(_ curlNi.vcurlA'n'df2= I~, c u r i N g . T o d d , i = 1 , 2 . . . . . n~ (28)

n n

where use has been made of the fact that the basis functions/V i satisfy the homogeneous Dirichlet boundary
conditions in (26).
As shown in Section 3, the functions curl N i (i = 1, 2 . . . . . ne) are
not linearly independent and consequently,
the matrix of the equations system (28) is singular. The right-hand side is, however, obviously consistent since
the same linear interdependence among its elements is present as among the rows of the left-hand side. Note that
this would not be the case if the Galerkin equations were set up in the traditional way by using homogeneous
396 O. Bird / Comput. Methods Appl. Mech. Engrg. 169 (1999) 391-40.5

Neumann boundary conditions in (26) (i.e. assuming the first boundary condition in (10) instead of in (10a)) and
T O were not introduced. Indeed, this would lead to the Galerkin equations

f , , curl N i . u curiA (") dO = (~ N i 'J0 d O , i = 1,2 . . . . . n~ (28a)

n n

where the consistence of the right-hand side is not ensured. The situation can still be salvaged if the edge
element representation (19) of T O is used, resulting in

curlNi.pcurlA~n) dO = N ~ - c u r l ~ tkNkdl~, i = 1 , 2 . . . . . n~ (28b)

n n k=l

where the integrands on the right-hand side are seen to be linear combinations of the functions curl N~
(i = 1, 2 . . . . . ne) and hence consistence is again given [ 16]. The form (28), however, needs no assumption about
T o, it can also be chosen to be the Biot-Savart field H s.

4.3. The A - f o r m u l a t i o n

The representation of the coil current density Jo in the A-formulation is not satisfactory unless the coils are
carefully modeled by the finite element mesh. This is evident from the form (28a) of the Galerkin equations
which is equivalent to (28) provided the divergence of Jo is zero. Indeed, (28a) shows that only the finite
elements with J0 # 0 contribute to the integrals on the right-hand side. An exact modeling of the coils can be
avoided if the magnetic flux density is sought as the sum of the impressed Biot-Savart field in free space, tzoHs,
and of the curl of a reduced vector potential A r [9]:
B = ixoHs + c u r l A r . (29)
Note that the impressed field must be divergence free and must have continuous normal component on finite
element interfaces, so the option of using any other function T O than H s is not allowed here.
The assumption (29) results again in the exact satisfaction of (2), and writing (1) with the second relationship
in (7) used leads to the differential equation

curl(v curlAr) = curl H s - curl v I ~ H s in O n (30)

where the satisfaction of (9) by H s has been made use of.

Using both the second and the first boundary conditions in (10a) (with T O replaced by H s and assuming that
the far boundaries lie in regions with /x =/&)) we have the following homogeneous Dirichlet or Neumann
boundary conditions on the reduced vector potential:
ArXn=O or vcurlArXn=O onF~. (31)

The reduced vector potential is approximated by edge basis functions as

n e

A r - A ~"~
r = ~_~ a~N k (32)
k=l

where a k are the line integrals of A r along the edges.

Setting (32) into (30) as well as into the Neumann boundary condition in (31) and using the basis functions
N i as weighting functions, the following Galerkin equations are obtained:

f~ c u r l N i - u c u r l A ~ ' d O = I., c u r l N i ' ( H s - U l X H s ) d 1 2 ' i = 1,2 . . . . . n e (33)

n n

where use has been made of

N,n=0 or (Hs-u~oHs)n=O onF~. (34)

The consistence of the right-hand side in (33) is obvious.

 O. B[r6 / Comput. Methods Appl. Mech. Engrg. 169 (1999) 391-405 397

5. Formulations in eddy current regions

Two vector potential functions can be used to describe the eddy current field in the conducting region f2c:
either a magnetic vector potential or a current vector potential. The magnetic vector potential can be employed
with or without an electric scalar potential, whereas the current vector potential description must be augmented
by a magnetic scalar potential. The resulting formulations are reviewed in this section with the boundary
conditions (11) or (1 la) assumed to hold on the entire boundary of ~2c. The treatment of the interface conditions
on F,c serving to couple these formulations to those in f 2 follows in Section 6.

5.1. The A *-formulation

The simplest way to describe the eddy current field is by means of a modified vector potential A* [18] as
follows:

B = curiA*, (35)

E = -jo.~A*. (36)

Basically, this is equivalent to the use of the electric field intensity as the underlying vector variable. Faraday's
law (4) is, hence, exactly satisfied and using the first relationship in (8) as well as the second one in (7), (3)
leads to the differential equation

curl u curiA* + j w o ' A * = 0 in/2 c . (37)

It is not necessary to use (5), since it follows from (3).

Using the boundary conditions (1 la), the following Dirichlet and Neumann boundary conditions on the
modified vector potential result:

A*n=0 or ucurlA*n=ToXn onF~. (38)

The modified vector potential is approximated by edge basis functions as

tl e

A* ~ A *~n~= ~ a k Nk (39)
k=l

where a t are the line integrals of A* along the edges.

Setting (39) into (37) as well as into the Neumann boundary condition in (38) and using the basis functions
N as weighting functions, the following Galerkin equations are obtained:

f a curl N/' ~, curl A * '' d O + f a Jw'Ni "A*n' d1"2 = f a curl N i " T O d r 2 , i = 1, 2 . . . . . n~ (40)
c c c

where use has been made of the fact that the basis functions N~ satisfy the homogeneous Dirichlet boundary
conditions in (38) as well as of

curl T o = J o = 0 in/2, (41)

i.e. of the current density of the coils being zero in the eddy current region.
The equations (40) are linearly independent, so this formulation leads to a nonsingular system. Still, the
numerical stability is rather inferior. This is similar to the case encountered when an edge element E-
formulation is applied to high frequency problems [19].
Note that in contrast to the similar formulation with nodal finite elements, A* can be used even if the
conductivity is discontinuous. Indeed, the normal continuity of the current density follows weakly from (37) and
edge elements do not enforce the normal continuity of the modified vector potential and hence of the electric
field intensity.
398 O. B{r6 / Comput. Methods Appl. Mech. Engrg. 169 (1999) 391-405

5.2. The A, V-formulation

In addition to its poor numerical stability, the A*-formulation has the further disadvantage that it does not
lend itself well to skin effect problems. Neither the total current nor the voltage of a conductor can be easily
prescribed. The additional introduction of an electric scalar potential V in the equation defining the electric field
intensity provides a simple way of taking a given voltage into account, and hence facilitates the treatment of
skin effect problems with voltage excitation.
The field quantities are then derived from the potentials as
B = curl A , (42)

E = -jo.vt - j o ) grad V. (43)

The time derivative in the second term of (43) is used to ensure the symmetry of the Galerkin equations [6].
Faraday's law (4) follows from (42)-(43), too, and the same manipulations as at the A*-formulation lead to the
differential equation
curl vcurlA +jwo'A +joker gradV= 0 in ~ , . (44)

Although (5) follows from (3), it is also written now in order to arrive at the necessary number of equations:

-div(jo~rA +floo- grad V) = 0 in $2c . (45)

Obviously, (45) is not independent of (44). This fact will be reflected in the Galerkin equations system written
below being singular.
Using the boundary conditions (1 la), the following Dirichlet and Neumann boundary conditions on the vector
and scalar potentials result:
An=0 or vcurlAXn=T on one, (46)

V=Vo=constant or n.(-jwo'A-floo-gradV)=O on~. (47)

The first boundary condition in (47) permits the prescription of the voltage of a conductor. The second one
states that the normal component of the current density is zero. This follows from the boundary condition on the
tangential components of H, but has to be included here because of the presence of the differential equation
(45).
The vector potential is approximated by edge basis functions as
n e

A "-~'A" = ~ akNk (48)

k=l

where a~ are the line integrals of A along the edges, and the electric scalar potential by nodal basis functions as
n n

V-~ V (n~= ~, VkN~ (49)

k-1

where Vk are the nodal values of V("~.

Setting (48)-(49) into (44)-(45) as well as into the Neumann boundary conditions in (46)-(47) and using
the basis functions N i in (44) and Ni in (45) as weighting functions, the following Galerkin equations are
obtained:

f~curlNi.vcurlA("~d~2+faj~ooN,'A~")dl2+f~jwo'N~'gradV'"'dI2=fecurlN~'Todl2,
c c c c

i = 1, 2 . . . . . n e , (50)

I.. JJ'gradN~'A~"~dO+
c
I.. jo)o'gradN.gradV'"'d/2=0,
c
i= 1,2 . . . . . n. (51)

where use has been made of (41) and of the fact that the basis functions N i satisfy the homogeneous Dirichlet
boundary condition in (46) and N, the homogeneous counterpart of the Dirichlet boundary condition in (47).
 O. B{r6 / Comput. Methods Appl. Mech. Engrg. 169 (1999) 391-405 399

The linear interdependence of Eqs. (50) and (51) is apparent in view of (15) and (17). In fact, building the
linear combinations of the equations in (50) with the coefficients cik defined in (15) results in the equations in
(51) once (17) has been taken into account. The consistence of the right-hand side is obvious.
The numerical stability of the equations system (50)-(51) is better than that of the system (40). This is
consistent with the experience reported in [19] in the context of high frequency problems.

5.3. The A t , V-formulation

A natural extension of the At-formulation presented in Section 4.3 for eddy current free regions is the A,,
V-formulation in eddy current domains:

B = #oils + curiA,, (52)

E = - j w A s -jo~A r - j w grad V. (53)

The impressed vector potential A s can be obtained from the current density J0 as

P,o
roe d12e
As(P )=~
f a Q Jo(Q) (54)

where 120 is the part of the region / 2 where J0 is nonzero. Hence,

curiA s = izoHs (55)
is satisfied. Therefore, Faraday's law (4) follows from (52)-(53), too and the differential equations

curiO, curl A , ) + jwo'A, + joJo" grad V= - j w o ' A s - curl ~,lxoHs in 12c, (56)

-div(jog~rA, +jogo" grad V) = div(joJcrAs) in 1"2c (57)

are obtained. Similarly to the A, V formulation, the fact that (57) follows from (56) will result in the Galerkin
equations system written below being singular.
Using the boundary conditions (11), the following Dirichlet and Neumann boundary conditions on the vector
and scalar potentials result:
A rn=-A sn or vcurlA r x n = - v / ~ H sXn one, (58)

V= Vo = constant or n (-jo~rA r - jw~r grad V) = n . (jwcrA s) on 4 . (59)

The reduced vector potential is approximated by edge basis functions as

n e

A r ~A~'= ~ akN~ (60)

k=l

where a k are the line integrals of A, along the edges, and the electric scalar potential by nodal basis functions as
n n

V ~ V ~")= ~ VkNk (61)

k=l

where Vk are the nodal values of V ("~.

Setting (60)-(61) into (56)-(57) as well as into the Neumann boundary conditions in (58)-(59) and using
the basis functions N i in (60) and ~ in (61) as weighting functions, the following Galerkin equations are
obtained:

f,,ccurlNi" v c u r l A ~ d 1 2 + fo 'A''
c
j oTvi , d12+ f. c
jo~o'Ni'gradV~md12

= - [ . curlN,'~,~//~d12- [.. joJ~N~'AsdO, i= 1,2..... ~, (62)

c c
400 O. B{r6 / Comput.Methods Appl. Mech. Engrg. 169 (1999) 391-405

f a Jc'gradNl'A(r~)dg2+ fa J~-gradNl'gradV'n) d n = - f a Jw'gradNi'Asdg2'

c c c

i = 1, 2 . . . . . n n (63)

where use has been made of the fact that the basis functions N~ and N~ satisfy the homogeneous counterparts of
the Dirichlet boundary conditions in (58) and (59).
By the same reasoning as put forward for Eqs. (50), (51), Eqs. (62) and (63) are linearly interdependent.

5.4. The T, q)-formulation

In skin effect problems with current excitation, the total current of a conductor cannot be easily prescribed if
the magnetic vector potential is used in the eddy current region. A current vector potential, however, excellently
suits these types of problems [20]. The use of a current vector potential facilitates the modeling of thin flaws in
conducting pieces by setting its tangential components to zero over the surface of such defects. This results in
enforcing the normal component of the current density to vanish. Furthermore, the edge element representation
of the current vector potential permits the incorporation of an eddy current magnetic field as impressed field in a
static analysis [21].
The field quantities are derived from the potentials as

H = T O+ T - grad ~ , (64)

J = curl T o + curl T . (65)

The curl of T o is a current density yielding the prescribed total current in ,(2 in case of a skin effect problem
with current excitation [20]. Ampere's law (3) follows from (64)-(65), and using the second relationship in (8)
as well as the first one in (7), Faraday's law (4) leads to the differential equation

curl p curl T +jwlxT -joo# grad/2 = - c u r l p curl T O-jtol~T o in ,(2. (66)

Although (6) follows from (4), its time derivative is also written in order to arrive at the necessary number of
equations:
jto d i v ( # T - # grad ,(2) = -rio div(#To) in/2c . (67)

The Galerkin equations system written below will be singular since (67) is a consequence of (66).
Using the boundary conditions (1 la), the following Dirichlet and Neumann boundary conditions on the vector
and scalar potentials result:
TXn=0 or pcurlTXn=-pcurlT on one, (68)

q~ = q~o = constant or n (/xT - / ~ grad 6b) = - n # T 0 on F . (69)

The second boundary condition in (69) states that the normal component of the flux density is zero. This follows
from the boundary condition on the tangential components of E, but has to be included here because of the
presence of the differential equation (67).
The current vector potential is approximated by edge basis functions as
n e

T -~ T ~n) = ~ tkNk (70)

k:l

where t k are the line integrals of T along the edges, and the magnetic scalar potential by nodal basis functions as
n n

q, = ,~n' = ~ 'rkN~ (71)

k--I

where 6bk are the nodal values of qCnk

Setting (70)-(71) into (66)-(67) as well as into the Neumann boundary conditions in (68)-(69) and using
 O. Bfr6 / Comput. Methods AppL Mech. Engrg. 169 (1999) 391-405 401

the basis functions Ni in (66) and Ni in (67) as weighting functions, the following Galerkin equations are
obtained:

facurlNi.pcurlT<n'd~2+fs ~ jw#N~.TU"dd2-fa jw/zN~-grad qO<")dO

c c c

=-ff~ curlNi.pcurlTodO-fa jw/zNi. TodO, i=1,2 ..... %, (72)

c

- fa Jwtx gradN" T(n~dd2 + f~ J~/z gradNi "grad CrpU')dO = fe J~ggradN"TdO'

c c ~c

i = 1, 2 . . . . . n~ (73)

where use has been made of the fact that the basis functions N~ satisfy the homogeneous Dirichlet boundary
condition in (68) and N, the homogeneous counterpart of the Dirichlet boundary condition in (69).
The linear interdependence of Eqs. (72) and (73) is of a nature similar to that between (50) and (51). The
consistence of the right-hand side is also easy to see.

6. Coupling of formulations

The various formulations in f 2 and J'2 presented in Sections 4 and 5 can be coupled by satisfying the
interface conditions (12). Basically, all four potential pairs in the eddy current region can be used
simultaneously either with a magnetic vector or a magnetic scalar potential in the nonconducting domain. This
results in the following formulations: A* - A; A* - qb; A, V - A; A, V - ~b; Ar, V - Ar; A r, V - ~ ; T, ~/' - A; T,
q b - @. Four further possibilities arise through the use of A in one part of O n and of q0 in another part. The
resulting formulations: A * - A - d~; A, V - A - c19; At, V - A r - ~; T, c 1 9 - A - ~ are necessary if the
exclusive use of qb in the nonconducting region is prevented by the multiply connectedness of the eddy current
region but it is not desired to use A everywhere in O,,.
The coupling of magnetic vector potential formulations in both regions (A* - A ; A, V-A; A r, V - A r) is
straightforward. The continuity of the tangential component of the vector potential on Q, is explicitly satisfied
through the use of the edge basis functions. This enforces the continuity of B n. The continuity of H X n is
accounted for in the weak sense by the surface integrals of the form

fr, Ni" (1t X n) dF (74)

arising in the Galerkin equations both in ,(2,, a n d / 2 and thus canceling. Therefore, the equations of the A* - A
formulation are (28), (40), those of the A, V - A formulation are (28), (50), (51) and those of the A r, V - A r
formulation are (33), (62), (63). The A* - A formulation is an option which is not open if nodal elements are
used, with edge elements it is, however, valid. The reason is that the normal component of the magnetic vector
potential along the interface F~c is determined by the Coulomb gauge in a nodal formulation and this contradicts
the fact that the vanishing of the normal component of the current density here sets A- n to zero (see (36)). On
the other hand, the use of edge basis functions allows for a discontinuity in the normal component of A, making
the A* - A formulation a viable one. Its numerical stability is, however, rather inferior as pointed out in Section
5.1.
The coupling of T, qb in f2 to qb in f2,, (T, ~ - formulation) is similarly forthright. The continuity of
H x n is enforced by the continuity of the magnetic scalar potential and by setting the tangential component of
the current vector potential to zero on the interface:
Txn=0 onF~c. (75)
The normal component of the flux density is satisfied in the weak sense through the surface integrals of the form

fr N,B" n dF (76)
c
402 O. B:r6 / Comput. Methods Appl. Mech. Engrg. 169 (1999) 391-40.5

canceling. Therefore, the Galerkin equations of the T, 4 - 4 formulation are (23), (72), (73).
As regards the A* - 4, A, V - 4 and A~, V - 4 formulations, the coupling of A or A~ to 4) on the interface
is required in them. The same problem arises in the formulations with the simultaneous use of A or Ar and 4 in
f2n. The coupling of the magnetic vector and scalar potentials by means of the interface conditions (12) can be
carried out in the classic way of [2] or [3]. This will be presented in Sections 6.1 and 6.2 for the simplest case of
coupling A or A~ to 4 along an interface in ,(2~.
The remaining T, 4 - A and T, 4 - A - 4 formulations necessitate the coupling of T, 4 in/'2,, to A in ~,,.
The way to do this has been presented in [7]. The specialties arising in the edge element context will be treated
in Section 6.3.

6. I. Coupling A to 4
Let S2A and f2~ denote two subregions of 12n separated by an interface FA,. The magnetic field is derived in
f2a from the magnetic vector potential in accordance with (24) and in 12~ from the magnetic scalar potential as
in (18). The differential equations to be solved are (25) in OA and (20) in O~. The corresponding boundary
conditions on the outer boundary of O n are (26) and (21) and the potentials are approximated as in (27) and
(22). The interface conditions stating the continuity of the tangential components of H and of the normal
component of B are

vcurlAXn=Ton-grad4Xn OnFA~, (77)

/xgrad4).n=/xT 0.n-n'curlA OnFa, (78)

where n is the unit normal pointing from Oa towards g2,. Following [2] or [3], the resulting Galerkin equations
are

fe~AcurlNi'vcurlA~n) d~2+ fra, Ni'(grad4(n)n)dF=f~A curlNi" T d O ' i=1,2 ..... ne , (79)

-fs~ eb gradNi'tzgrad4(")d~2+ frAq) A'n"(grad~n)dF=- f~ q~ g r a d N " ' / z T d $ 2 '

i = 1, 2 . . . . . n n . (80)

Eqs. (79) are still linearly interdependent. This will be shown here for the case when FA, is a closed surface,
but, taking account of the boundary conditions (27) and (22), it can be proved for the general case, too.
If FA, is closed, the surface integral in (79) can be rewritten using Gauss' theorem as

~t N"(grad 4'") n)dF= f~ div(Ni grad 4'"')df2= faACUrlN~'grad 4'") dg2 (81)
,'4P 2A

where grad qb ~') in OA is an arbitrary gradient (i.e. curl free) function with its tangential components on FA~
equal to the tangential components of grad 4 <") on _FA~defined in 12~. The presence of curl N~ in the integrand
of the last integral in (81) accounts for the linear interdependence of Eqs. (79).
The surface integrals in (79) and (80) couple the two potentials in a symmetric way and they augment the
Galerkin equations in (28) and (23), which had been written without an interface. Therefore, the Galerkin
equations of the A* - 4, A, V - 4, A* - A - 4 and A, V - A - 4 formulations are obtained by writing
(40)-(23), ( 5 0 ) - ( 5 1 ) - ( 2 3 ) , ( 4 0 ) - ( 7 9 ) - ( 8 0 ) and ( 5 0 ) - ( 5 1 ) - ( 7 9 ) - ( 8 0 ) with the surface integral in (79) added
to the left-hand sides of (40) and (50) and the surface integral of (80) written over Fn, augmenting the left-hand
side of (23).

6.2. Coupling A r to tlb

In case of the A r, V - 4 and A r, V-Ar - 4 formulations it is necessary to couple the reduced magnetic
vector and scalar potentials over an interface FA~. In this case the Biot-Savart field H s must be used in the
 o. Bird / CompurMethods Appl. Mech. Engrg. 169 (1999) 391-405 403

O-region, too. This is only feasible if there are no ferromagnetic materials present, so it will be assumed that
/x =/.to, v = vo : 1/txo everywhere.
Again, let ~2a and ~2~ denote two subregions of f2n separated by an interface FA~. The magnetic field is
derived i n / 2A from the reduced magnetic vector potential in accordance with (29) and in ~ from the magnetic
scalar potential as in (18) (T o =Hs). The differential equations to be solved are (33) in [2z and (20) in ~ (in
both cases, the right-hand sides are zero since Uo/Xo = 1 and d i v ( / . t o H s ) : 0). The corresponding boundary
conditions on the outer boundary of ~,, are (34) and (21) and the potentials are approximated as in (32) and
(22) (T O = Hs). The interface conditions stating the continuity of the tangential components of H and of the
normal component of B are

~curlA~n=-gradOn onFa~, (82)

~ grad O . n = - n - c u r l A ~ on Fa~ (83)

where n again points from /2A towards /2~. The resulting Galerkin equations are

I curlN~'pocurlA~")d[2+ I~ N~'(gradO(~'n)dF=O, i = 1 , 2 . . . . . n~, (84)

o

- fao gradNi'p'gradO(")dl2+ frA. A~)'(gradNiXn)dF

=-f, gradNi'tzoHsd~2-fl Ni'tzoHs.ndF i=1,2 ..... n,. (85)

The linear interdependence of Eqs. (84) can be shown as for Eqs. (79). The surface integrals in (84) and (85)
couple the two potentials in a symmetric way and they augment the Galerkin equations in (33) and (23), which
had been written without an interface. Therefore, the Galerkin equations of the A~, V - and A r, V - A ~ -
formulations are obtained by writing ( 6 2 ) - ( 6 3 ) - ( 2 3 ) , and ( 6 2 ) - ( 6 3 ) - ( 8 4 ) - ( 8 5 ) with the surface integral in
(84) added to the left-hand side of (62) and the surface integrals of (85) written over F,, added to the left- and
right-hand sides of (23).

6.3. Coupling A to T, 0

If the T, O-formulation is used in the eddy current region [2,. and the A-formulation in at least a part ~2a of the
eddy current free region, let FAT denote the interface between the two. The magnetic and electric fields are
derived in ~,. from the current vector potential and magnetic scalar potential in accordance with (64), (65) and
the magnetic field in OA from the magnetic vector potential as in (24). The differential equations to be solved
are (66), (67) in ~2 and (28) i n / 2A. The corresponding boundary conditions on the outer boundary of ~2, and O R
are (68), (69) and (26). The vector potentials are sought in the forms
n Te

T ~ T <n)= ~ tkNrk (86)

k=l

nAe

A ~ A ~"~= ~ a~Nak (87)

k=l

and the scalar potential is approximated as in (71). nTe, nae denote the numbers of the edges and Nrk, Nak the
edge basis functions in Oc and ~2a, respectively.
The interface conditions stating the continuity of the tangential components of H and of the normal
component of B are

~,curlA n = T O n + T n - grad n on FAr, (88)

- / . , T . n + / z grad O . n = / . i T O n - n curl A on Far (89)

404 O. Bir6 I Comput. Methods Appl. Mech. Engrg. 169 (1999) 391-405

where n is the unit normal pointing from f2A towards {2,.. Furthermore, following [7], the continuity of the
tangential component of E on FAr can be written as

p curl T X n = - p curl T O n - j o a 4 n. (90)

T h e r e s u l t i n g G a l e r k i n e q u a t i o n s w i t h the t i m e d e r i v a t i v e o f ( 2 8 ) used are

f, curlNr, . p curl T~"' d.O + fe jwlzNr, " T(") df2 - fz~ jwtxNT, " grad ~("' d$2

-f[ jw(NriA(n))'ndF=-f, curlN,'pcurlT0dS2-j, jwtzNi'TodO, i = 1 , 2 . . . . . n,,.,

"AT 2 ]~

(91)

- f~ Jto/z gradN~" T(") d~2 + f~ jwlz gradN~.grad ~n~ d l 2 - f, jo~").(gradNg n ) d F

c" c "AT

= ( . j~o/zgrad N / T Od O , i = 1,2 . . . . . n . , (92)

c

- f~ jwcurlNAi" ~'curlA(n)dO- f& jO~NAi.(gradC19(n)n)dF- fr a jto(T(n)NAi)'n)dF

~A r r
= - ~
d~~a
j w curl Nai" T O d O i = 1, 2 . . . . . time. (93)

T h e l i n e a r i n t e r d e p e n d e n c e o f t h e s e e q u a t i o n s s y s t e m c a n b e seen s i m i l a r l y to t h a t o f (72), ( 7 3 ) a n d (79).

T h e surface i n t e g r a l s in (91), ( 9 2 ) a n d ( 9 3 ) c o u p l e the t h r e e p o t e n t i a l s in a s y m m e t r i c w a y a n d t h e y a u g m e n t
the G a l e r k i n e q u a t i o n s in (72), ( 7 3 ) a n d ( 2 8 ) (this latter m u l t i p l i e d b y -jto), w h i c h h a d b e e n w r i t t e n w i t h o u t a n
interface. T h e G a l e r k i n e q u a t i o n s o f the T, q b - A f o r m u l a t i o n are (91), (92), ( 9 3 ) a n d t h o s e o f the T,
qb - A - qb f o r m u l a t i o n are o b t a i n e d b y w r i t i n g (91), (92), ( 9 3 ) w i t h the first surface i n t e g r a l in ( 9 3 ) e x t e n d e d
o v e r FA~, t o o a n d a d d i n g Eqs. ( 8 0 ) m u l t i p l i e d b y -jo~.

References

[t] C.J. Carpenter, Comparison of alternative formulations of 3-dimensional magnetic-field and eddy-current problems at power
frequencies, Proc. IEE 124( 11) (1977) 1026-1034.
[2] D. Rodger and J.F. Eastham, A formulation for low frequency eddy current solutions, IEEE Trans. Magnetics 19 (1983) 2443-2446.
[3] C.R.I. Emson and J. Simkin, An optimal method for 3-D eddy currents, IEEE Trans. Magnetics 19 (1983) 2450-2452.
[4] P.J. Leonard and D. Rodger, Finite element scheme for transient 3D eddy currents, IEEE Trans. Magnetics 24 (1988) 90-93.
[5] T. Nakata, N. Takahashi, K. Fujiwara and Y. Okada, Improvements of the T - J2 method for 3-D eddy current analysis, IEEE Trans.
Magnetics 24 (1988) 94-97.
[6] O. Bfr6 and K. Preis, On the use of the magnetic vector potential in the finite element analysis of 3-D eddy currents. IEEE Trans.
Magnetics 25 (1989) 3145-3159.
[7] O. Bfr6 and K. Preis, Finite element analysis of 3-D eddy currents, IEEE Trans. Magnetics 26 (1990) 418-423.
[8] O. Bir6 and K.R. Richter, CAD in electromagnetism, in P.W. Hawkes, ed., Advances in Electronics and Electron Physics, Vol. 82
(Academic Press, 1991) 1-96.
[9] K. Preis, I. Bardi, O. Bfrr, C. Magele, W. Renhart, K.R. Richter and G. Vrisk, Numerical analysis of 3D magnetostatic fields, IEEE
Trans. Magnetics 27 (1991) 3798-3803.
[10] O. Bfr6, Solution of TEAM benchmark problem #10 (Steel plates around a coil), ACES J. 8(2) (1993) 203-215.
[11] A. Kameari, Calculation of transient 3D eddy current using edge elements, IEEE Trans. Magnetics 26 (1990) 466-469.
[12] A. Bossavit, Solving Maxwell equations in a closed cavity and the question of 'spurious modes', IEEE Trans. Magnetics 26 (1990)
702-705.
[13] Z.J. Cendes, Vector finite elements for electromagnetic field computation, IEEE Trans. Magnetics 27 (1990) 3958-3966.
[14] R. Albanese and G. Rubinacci, Magnetostatic field computations in terms of two-component vector potentials, Int. J. Numer. Methods
29 (1990) 515-532.
[15] J.B. Manges and Z.J. Cendes, A generalized tree-cotree gauge for magnetic field computation, IEEE Trans. Magnetics 31 (1995)
1342-1347.
 O. Bir6 / Comput. Methods Appl. Mech. Engrg. 169 (1999) 391-405 405

[16] O. Bfr6, K. Preis and K.R. Richter, On the use of the magnetic vector potential in the nodal end edge finite element analysis of 3D
magnetostatic fields, IEEE Trans. Magnetics 32 (1996) 651-654.
[17] O. Bfr6, K. Preis, G. Vrisk, K.R. Richter and I. Ticar, Computation of 3-D magnetostatic fields using a reduced scalar potential, IEEE
Trans. Magnetics 29 (1993) 1329-1332.
[ 18] T. Morisue, Magnetic vector potential and electric scalar potential in three-dimensional eddy current problem, IEEE Trans. Magnetics
18 (1982) 531-535.
[19] R. Dyczij-Edlinger and O. Bir6, A joint vector and scalar potential formulation for driven high frequency problems using hybrid edge
and nodal finite elements, IEEE Trans. Microwave Theory Tech. 44 (1996) 15-23.
[20] O. Bfr6, K. Preis, W. Renhart, G. Vrisk and K.R. Richter, Computation of 3-D current driven skin effect problems using a current vector
potential, IEEE Trans. Magnetics 29 (1993) 1325-1328.
[21] K. Preis, I. B~irdi, O. Bfr6, J. P~.v6, A. Gasparics and I. Ticar, Numerical simulation and design of a fluxset sensor by finite element
method, I lth Conference on the Computation of Electromagnetic Fields, November 2-6, Rio de Janeiro, Brazil, IEEE Trans.
Magnetics 34 (1998) 3475-3478.