JOURNAL OF COMPUTATIONAL PHYSICS 129, 190200 (1996)

ARTICLE NO. 0243

The Bean Model in Superconductivity: Variational Formulation
and Numerical Solution
Leonid Prigozhin*
Mathematical Institute, OCIAM, University of Oxford, Oxford, United Kingdom
Received February 21, 1996
wells equations, the formulated rules provide a macro-
scopic model of magnetic eld penetration into a type-
The Bean critical-state model describes the penetration of mag-
netic eld into type-II superconductors. Mathematically, this is a
II superconductor. Solution of the arising mathematical
free boundary problem and its solution is of interest in applied
problem is, however, difcult and the known exact analyti-
superconductivity.We derive a variational formulation for the Bean
cal solutions are mostly restricted to idealized geometries,
model and use it to solve two-dimensional and axially symmetric
such as an innitely long cylinder in a parallel eld, a half-
critical-state problems numerically. 1996 Academic Press, Inc.
space in a rotating parallel eld, an innitely thin strip or
disk in a perpendicular eld, etc. [38]. Substitutes of exact
Beans model relations were sometimes employed in nu- 1. INTRODUCTION
merical procedures to make the calculations feasible (see,
Into type-II superconductors, the magnetic eld pene-
e.g., [9]).
trates in the form of superconductive electron current vor-
The main complication is the presence of an unknown
tices. Each of these vortices carries the same amount, one
(free) boundary dividing the regions of subcritical (usually
quantum, of magnetic ux; hence the magnetization is pro-
zero) currents and of critical currents. To solve the problem
portional to the vortex density. At a close to zero tempera-
numerically, several front-tracking algorithms have
ture, the distribution of magnetic vortices in type-II
been developed [1012]. Such algorithms can be very accu-
superconductors is determined by the balance of electro-
rate and also efcient for problems with simple free bound-
magnetic driving forces and forces pinning vortices to ma-
aries. However, the implementation of such methods be-
terial inhomogeneities. Whenever the external magnetic
comes difcult if the free boundary topology is
eld is changed, magnetic vortices start to enter or leave
complicated, changes in time, or is not known a priori.
the superconductor through its boundary. There appears
In this work we derive a variational formulation for two-
a region where driving forces overcome pinning and the
dimensional and axially symmetric critical-state problems.
system of vortices rearranges itself into another metastable
This formulation makes unnecessary separate consider-
state such that all vortices are pinned again and the equilib-
ation and different treatment of critical and subcritical
rium with the external eld at the boundary is re-estab-
regions, allows one to avoid front-tracking, and thus facili-
lished. Since the unpinned vortices move rapidly, the sys-
tates signicantly the numerical solution of the free bound-
tem quickly adjusts itself to the changing external
ary problem. There is, however, another difculty, typical
conditions, and a quasistationary model of equilibrium can
of problems in electromagnetism: the magnetic eld has
be justied.
to be found in an unbounded space. We further reformu-
These are the basic and somewhat simplied assump-
late the problem in terms of current density, which is un-
tions of the phenomenological critical-state models of hys-
known only in the region occupied by a superconductor.
teretic magnetization in type-II superconductors [1, 2]. In
This latter variational formulation serves as a basis for an
terms of macroscopic quantities, these assumptions may
efcient numerical algorithm described in this work. We
be formulated as follows: current density never exceeds
present the results of numerical simulation and compare
some critical value determined by the density of pinning
them with the known solutions.
forces and, as long as this threshold is not reached, the
A different variational formulation for the generalized
magnetic induction remains unchanged. Jointly with Max-
Bean model has been recently proposed by Bossavit [13].
This formulation must also be restricted to two-dimen-
sional and axially symmetric problems (because of the as-
* Present address: Department of Applied Mathematics and Computer
sumption that the electric eld and current density are
Sciences, The Weizmann Institute of Science, Rehovot 76100, Israel.
E-mail: leon@wisdom.weizmann.ac.il. parallel; see the next section for a discussion). The difculty
190
0021-9991/96 $18.00
Copyright 1996 by Academic Press, Inc.
All rights of reproduction in any form reserved.
BEAN CRITICAL-STATE PROBLEMS 191
electric eld are collinear. Only this case will be considered
in our work. Hence
E J in , (4)
where
(x, t) 0 (5)
FIGURE 1
is an unknown nonnegative function.
Equation (4) may be regarded as Ohms law with an
effective resistivity . However, since the resistivity is an
of determining the magnetic eld in an innite space has
auxiliary unknown, this relation, for a given current den-
not been resolved in [13], and no numerical results have
sity, xes only the possible direction of the electric eld
been reported.
but not its magnitude.
According to the critical state model, the current density
2. MODEL OF THE CRITICAL STATE
in superconductor cannot exceed some critical value, J
c
.
In the Bean model of the critical state, J
c
is a constant
Let a superconductor occupy a three-dimensional spatial
determined by the properties of superconductive material.
domain with the boundary , and let be the exterior
However, Kim et al. [2] found that generally the critical
space (Fig. 1).
current density depends on the magnetic eld and various
Maxwell equations with the displacement current omit-
relations of the type J
c
J
c
(H) have been proposed (see,
ted read
e.g., [14, 15]). The constraint on the current density may
be written as
B
t
curl E 0, (1)
curl H J
c
(H) in . (6)
J curl H, (2)
In the regions, where the current density is less than critical,
and we assume B
0
H, where
0
is the permeability of
the vortices are pinned. Hence, there is no dissipation of
the vacuum. The current density in the exterior space is
energy and the current is purely superconductive:
assumed to be known:
curl H J
c
(H) 0. (7)
curl HJ
e
in . (3)
To complete the model, the initial and boundary conditions
Here J
e
(x, t) is the density of external current, which should must be specied. Let
satisfy the condition div J
e
0.
In the presence of electrical current owing through
B
t0
B
0
(x) (8)
the superconductor, the magnetic vortices respond to the
action of the Lorentz force which we average into a body
with div B
0
0 (together with (1), the last condition en-
force with density
sures div B 0). On the boundary dividing the two media,
the tangential component of electric eld E is continuous,
F
L
J B.
[E

] 0 on ,
If the vortices become unpinned, they move in the direction
of this force, and so their velocity v is parallel to F
L
. The
where [] denotes the jump across the boundary. We ne-
movement of vortices induces the electric eld
glect the surface current, which is much less than the total
current in most applications of type-II superconductors
[16]. The tangential component of magnetic eld H on this E B v,
boundary is thus assumed to be continuous too:
which is thus parallel to B (J B). If B is perpendicular
[H

] 0 on . (9)
to J, as is always the case for two-dimensional problems
and also for some three-dimensional problems, e.g., those
with axial symmetry, the vectors of current density and We also suppose that H 0 as x .
192 LEONID PRIGOZHIN
The mathematical model obtained contains a system The model now consists of Eqs. (3), (5)(10) and the
boundary condition at innity. The electric eld has been of equations and inequalities which is difcult to attack
directly. Furthermore, in accordance with the postulates excluded and there remain only two unknowns: the
magnetic eld and the effective resistivity. We will now of the Bean model, the effective resistivity is not dened
explicitly but only implicitly determined by (5)(7). How- also exclude from the model the resistivity, an auxiliary
variable introduced to specify the possible direction of ever, it can be shown that this model is well posed mathe-
matically and a much more convenient variational formula- electric eld.
Let us dene the external magnetic eld H
e
as a quasista- tion of this model can be derived. Below we present an
outline of transition to such formulation; see [17] for the tionary magnetic eld induced by the external current in
the absence of the superconductor, i.e., as a solution of mathematical details.
the problem
3. VARIATIONAL FORMULATIONS
curl H
e
J
e
, div H
e
0,
(11) 3.1. Magnetic Field Formulation
H
e
0 as x .
Let us dene a space of test vector functions whose
curl vanishes in the exterior and whose tangential compo-
Since div J
e
0, the problem has a unique solution and
nents are continuous across the boundary ,
it is the curl of convolution [18],
V (x, t) curl 0 in , [

] 0 on
H
e
curl (G J
e
),
(the functions and their curls are assumed to be square
where G 1/4 x is the Green function of Laplace equa- integrable). Multiplying (1) by an arbitrary function from
tion (G (1/2) ln (1/x) for two-dimensional problems). V and integrating in time and over , we obtain
We can now introduce a new variable, h H H
e
, satis-
fying
0
T
0

B
t
curl E

curl h 0 in , (12)

0

T
0

H
t

T
0

[E curl div(E )]
curl h J
c
(h H
e
) in , (13)
[h

] 0 on . (14)

0

T
0

H
t

T
0

(E ) n,
Let us dene the set of functions
since curl 0 in . (The normal n is directed towards
K (h) V curl J
c
(h H
e
) in .
the domain .) Similarly, in (1), (2), and (4) yield
This set depends on h and, due to (12)(14), h itself belongs
0
0

T
0

H
t

T
0

curl H curl
to K (h). Furthermore, since curl h curl H in , the
relations (6), (7) yield

T
0

(E ) n.

T
0

curl h
2

T
0

J
2
c
(h H
e
).
Here

and

mean that the boundary values are taken

from the sides of and correspondingly.
Using the last relation and Eq. (10), taking into account
Adding these two equations and taking into account that
that h V for any function from K (h), we obtain
the tangential components of E and are continuous on
, we obtain the variational relation

0

T
0

R
3
h H
e

t
( h)

0

T
0

R
3
H
t

T
0

curl H curl 0 (10)

T
0

curl h curl( h)
which is valid for all test functions from V (R
3
denotes

T
0

(J
2
c
(h H
e
) curl curl h) 0.
the entire space).
BEAN CRITICAL-STATE PROBLEMS 193
This proves that h is a solution to the problem Variational reformulations of free boundary problems
are very convenient for numerical solution because the
free boundary does not appear in such formulations
nd function h such that
explicitly. The same numerical algorithm can be applied
everywhere, e.g., in the critical and subcritical regions
h K (h),
(15)
of the superconductor, and the front-tracking becomes
(h H
e
/t, h) 0 for any K (h),
unnecessary (see [23]). However, the solution of (16)
must be calculated in an innite domain, and this is an
h
t0
h
0
,
additional complication. To avoid this difculty, we now
derive an equivalent variational formulation for the cur-
where (u, w)
T
0

R
3 u w is the scalar product of two
rent density.
vector functions and h
0
B
0
/
0
H
e

t0
.
A problem of the form (15) is called a quasivariational
3.2. Current Density Formulation
inequality. More familiar are variational formulations
Let us rst note that instead of the set of admissible
where the solution is sought as an extremal point of some
functions K in the variational inequality (16) we can use
variational functional. If the problem contains a unilateral
the subset
constraint, like condition (6) in our case, the extremum is
to be found on the set of functions satisfying this constraint.
The variational (if the admissible set does not depend K
0
K div 0.
on the unknown solution) or quasivariational inequalities
express then an optimality condition for the constrained
Indeed, if h is a solution to (16), there exists a function
optimization problems [19]. Although for the nonstation-
such that the pair H, , where H h H
e
, is a solution
ary problems, like the critical-state problem under consid-
of the critical state problem. Since div H 0, we have
eration, no appropriate variational functional exists [20], a
div h 0 and h belongs to K
0
. Thus h is a solution to
formulation in the form of a variational or quasivariational
(16) with the set K
0
instead of K. It can be proved that
inequality may, as we saw, be still available.
this new variational inequality has also only one solution,
The problem (15) is equivalent to the Bean model: as
and so it is equivalent to (16).
proved in [17], the function h(x, t) is a solution of the
Furthermore, for any function from K
0
, the function
quasivariational inequality (15) if and only if there exists
curl belongs to the set
(x, t) such that the pair H, , where H h H
e
, is a
solution of the critical-state problem (3), (5)(10). It is also
shown in [17] that the effective resistivity is a Lagrange
multiplier related to the current density constraint in the
K
1

(x, t)
|
J
c
in ,
0 in ,
div 0 in R
3

,
Bean model.
It may be noted that the variational inequalities arise
also in models of some other electromagnetic phenomena
[21]. Also, the similarity of magnetic eld accumulation in
which can be regarded a set of possible current densities
type-II superconductors to the growth of a sandpile has
in the superconductor. On the other hand, for any function
been mentioned by several authors [16, 3]. Indeed, the
which belongs to K
1
, there exists a unique solution to
inequality (15) is similar to a scalar variational inequality
the problem
arising in the model of pile growth [22].
If, as was assumed by Bean, the critical current density
curl , div 0,
does not depend on the magnetic eld, the set of admissible
functions is xed: K (h) K. The inequality (15) becomes
0 as x .
a variational inequality:
This solution, which we denote by R , belongs to K
0
and
nd h K such that
may be written as a curl of a convolution,
(h H
e
/t, h) 0 for any K, (16)
R curl(G ).
h
t0
h
0
.
We have already considered such a problem above
and dened H
e
as R J
e
. It is easy to see that h R J The existence and uniqueness of solution to this problem
were proved in [17]. and we can rewrite inequality (16) with the admissible set
194 LEONID PRIGOZHIN
K
0
as eld. Such a uniform eld H
e
(t) in the neighborhood of a
superconductor can be generated by placing the supercon-
ductor inside a long solenoid. For two-dimensional cong- nd J K
1
such that
urations, i.e., a long cylinder in perpendicular eld, the
(R J J
e
/t, R R J) 0 for any K
1
,
uniform external eld could be generated by two parallel
innite sheets of current. Since H
e
curl A
e
, we can easily
R J
t0
h
0
,
nd the vector potential corresponding to the uniform eld
H
e
up to a constant of integration. For two-dimensional
or, equivalently,
problems, the potential of the current which induces the
eld H
e
H
e,1
e
1
H
e,2
e
2
is A
e
(x
2
H
e,1
x
1
H
e,2
C)e
3
nd J K
1
such that
in the space between the two plates. The potential inside
(R *R J J
e
/t, J) 0 for any K
1
,
the solenoid, where H
e
H
e
e
z
, is A
e
(rH
e
/2 C/r)e

in cylindrical coordinates r, , z. A somewhat laborious

J
t0
curl h
0
,
direct calculation of potentials using (18) shows that the
constant of integration C is zero in both cases.
where R * is the operator adjoint to R.
Since J J(x
1
, x
2
, t)e
3
for two-dimensional, and J J(r,
The product R *Rcan be calculated explicitly as follows.
z, t)e

for axially symmetric problems, the zero-divergence

Let , be two functions which disappear at innity
condition in the denition of K
1
is satised automatically.
rapidly enough and such that div div 0. Using
All functions from this set are zero in the exterior and
Greens theorem we obtain
may be regarded simply as the functions dened in . For
two-dimensional problems, the inequality (17) is scalar and
(R *R , ) (R , R )
can be written as

T
0

R
3
curl(G ) curl(G )
nd J K
2
such that

T
0

R
3
G curl curl(G )
(G J/t x
2
H
e,1
/t x
1
H
e,2
/t, J) 0
for any K
2
,

T
0

R
3
G [graddiv(G ) (G )]
J
t0
J
0
, (19)

T
0

R
3
G [G (graddiv ) (G ) ]
where J
0
is the only nonzero component of curl h
0
and

T
0

R
3
G ,
K
2
(x, t) J
c
J
c

since div 0 and G is the delta function. Thus

R *R G , which is the magnetic vector potential
is the set of functions dened in .
of current .
A similar scalar formulation of (17) can be obtained for
We thus arrive at the variational inequality
axially symmetric problems. The only nonzero component
of the convolution, (G J/t)

, can in this case be ex-

nd J K
1
such that
pressed in terms of complete elliptic integrals of the rst
(G J/t A
e
/t, J) 0 for any K
1
, (17)
and second kind, K and E, and does not depend on .
This leads to the variational inequality
J
t0
curl h
0
,
where nd J K
2
such that
(L J/t r
2
H
e
/t, J) 0 for any K
2
, (20)
A
e
G J
e
(18)
J
t0
J
0
,
is the vector potential of the external current. The inequal-
ity (17) contains a nonlocal operator of convolution. How-
with the linear operator
ever, this is a reasonable price for reducing the problems
formulation to a bounded domain.
Let us now consider the magnetization of superconduc-
L (r, z) a(r, z; r, z)(r, z)dr dz, (21)
tors in a temporally varying uniform external magnetic
BEAN CRITICAL-STATE PROBLEMS 195
where
a(r, z; r, z) (rr)
1/2
k

2
k
2
K(k) E(k) K(k)

,
k 2

rr
(r r)
2
(z z)
2

1/2
.
The current J and all functions K
2
are, in this case,
dened in the half cross section of the superconductor,
FIG. 2. Long superconductor in perpendicular eld and nite ele-
which is a body of rotation.
ment discretization.
4. NUMERICAL SOLUTION
be placed into a perpendicular magnetic eld (Fig. 2). We
The rst step in the numerical solution of (19) and (20)
dene a regular nite element net with square elements
is the nite difference approximation in time. This leads
in a rectangle, covering domain , and denote by I

the
to stationary variational inequalities at each time layer,
set of elements which lie mostly inside this domain. At
each time layer, the current density is approximated by a
nd J K
3
such that
(22)
constant, J
i
, within each nite element e
i
. The discretized
optimization problem (23) can be written as
(L J f, J) 0 for any K
3
,
where
min
J
1
,..., J
N

J
m
J
c
if nI

J
m
0 if nI

1
2

N
i, j1
J
i
M
i, j
J
j

N
i1
f
i
J
i
(24)
K
3
(x) J
c
J
c
.
For axially symmetric problems, the operator L is dened Here N is the number of nite elements,
by (21), f L J

r
2
(H
e
H

e
), and

means that the

value is taken fromthe previous time layer. For two-dimen-
M
i, j

e
i

e
j
G (x x) dx
1
dx
2
dx
1
dx
2
,
sional problems, L is the convolution with G (1/2) ln(1/
x) and f L J

x
2
(H
e,1
H

e,1
) x
1
(H
e,2
H

e,2
).
It can be shown that these stationary variational inequal-
f
i

e
i
f(x)

N
j1
M
i, j
J

j
(H
e,1
H

e,1
)
e
i
x
2
ities are equivalent to constrained optimization problems
(H
e,2
H

e,2
)
e
i
x
1
.
min
JK
3

1
2
(L J, J) ( f, J)

. (23)
For axially symmetric problems we approximate half the
cross section in a similar way and obtain (24) with the coef-
cients
Indeed, let J

be a solution to (23) and let be an arbitrary

function from K
3
. For any [0, 1], the function J

M
i, j

e
i

e
j
a(r, z; r, z) drdzdrdz,
(1 )J

also belongs to K
3
. Since J

is the point
of minimum,
f
i

N
j1
M
i, j
J
j
(H
e
H

e
)
e
i
r
2
drdz.
0
d
d

1
2
(L J

, J

) ( f, J

0
(L J f, J),
Below we compare the numerical and analytical solutions
also for innitely thin strips and disks in perpendicular
elds. The cross sections are then intervals; to nd the and J also solves the variational inequality (22). The con-
verse is also true. numerical solutions we solve (24) with the appropriate co-
efcients. To proceed with the solution, we now discretize the
problem in space and perform the optimization numeri- Some of the integrals M
i, j
are singular and their evalua-
tion needs special consideration (Appendix). Provided the cally. Let an innitely long cylinder with a cross section
196 LEONID PRIGOZHIN
The zero eld core shrinks with the growth of external
eld and completely disappears when this eld becomes
sufciently strong (Fig. 4). Note that the possible direction
of current density is known and the magnitude can take
only the values J
c
or zero. This fact can be used in the
numerical procedure; instead of performing the minimiza-
tion in (24) with high accuracy, one can simply round the
values of current density, obtained at some stage of minimi-
zation, to the closest of these three possible values. By a
different method, a similar result was recently obtained by
Brandt [24].
Results of numerically simulating the penetration of per-
pendicular magnetic eld into cylindrical superconductors
of various cross sections are shown in Fig. 5.
In [4], Bean presented an asymptotic solution for the
magnetization of a superconducting half space in a rotating
FIG. 3. Thin strip in a perpendicular eld. The distribution of current
density: , analytical; , numerical solutions. Numbers indicate
the ratio H
e
/H
c
.
coefcients are found, solving the optimization problem is
not difcult. We used the method of point underrelaxation
with projection [23], which in this case was more efcient
than the relaxation or overrelaxation algorithms. The com-
putation of a typical example (see below) needed about
20 min of IBM RS6000/370 time for the nite-element grid
80 80.
5. EXAMPLES
In all examples below we assume the superconductor is
initially in the virgin (zero eld) state. Let us consider rst
a long strip in a perpendicular external eld. If the strip
thickness is much smaller than the strip width 2L, the
current density integrated across the strip thickness,
I(x
1
, t)

0
J(x
1
, x
2
, t) dx
2
,
can be found analytically [5, 6]. For a growing external
magnetic eld H
e
H
e
(t)e
2
,
I
I
c

arctg[cx
1
/(b
2
x
2
1
)
1/2
], x
1
b,
x
1
/x
1
, b x
1
L,
where I
c
J
c
, c tgh(H
e
/H
c
), b L(1 c
2
)
1/2
, and
H
c
I
c
/. The numerical solution of (19) approximates
the analytical one very well (Fig. 3).
The real strips have a nite thickness, and the magnetic
FIG. 4. Thick strip in a perpendicular eld. The shrinking zero eld
eld rst penetrates into their surface region. This is the
core (left) and current density (right). Numbers indicate H
e
/LJ
c
, where
2L is the strip width. region of shielding current which has the critical density.
BEAN CRITICAL-STATE PROBLEMS 197
and H
c
I
c
/2. The numerical and analytical solutions are
presented in Fig. 7.
The problem of magnetic eld penetration into a super-
conducting ball has been solved asymptotically for weak
external elds H
e
in [3],
() R
3H
e
2J
c
sin ,
where is the azimuth angle, R is the ball radius, and ()
is the boundary of the zero eld core. The asymptotic
and numerical solutions are close (Fig. 8). The further
magnetization of a superconducting ball and some other
bodies of rotation in a growing uniform eld has also been
simulated numerically (Fig. 9).
6. CONCLUSION
The Bean critical-state model is a quasistationary macro-
scopic model of equilibrium of the system of magnetic
FIG. 5. The penetration of a perpendicular magnetic eld into cylin-
drical superconductors of different cross sections. Numbers indicate H
e
/
RJ
c
, R is the radius of circular cross section.
parallel external eld. As shown in his work, a stationary
rotating solution develops as t . We simulated the
magnetization of long cylinders with square and circular
cross sections in the rotating perpendicular external eld
(Fig. 6). In these two examples, the external eld rst grows
up to a given magnitude and then starts to rotate. The zero
eld core may only shrink and, as is apparent from these
examples, has almost reached its nal shape even before
the rst turn of external eld is completed.
The magnetization of a thin disk in a perpendicular eld
has been recently considered in [7], and the distribution
of integrated current density I(r, t) was found analytically.
In an increasing magnetic eld,
I
I
c

arctg(r/R)[(R
2
b
2
)/(b
2
r
2
)]
1/2
, 0 r b,
1, b r R,
FIG. 6. Cylindrical superconductors in a rotating eld: the evolution
of zero-eld core. The external eld (right column) rst reaches the value
H
e
/RJ
c
0.2 and then starts to rotate. where I
c
J
c
, R is the disk radius, b R/cosh(H
e
/H
c
),
198 LEONID PRIGOZHIN
FIG. 7. Thin disk in a perpendicular eld. Current density: ,
analytical; , numerical solutions. Numbers indicate H
e
/H
c
.
vortices in type-II superconductors. In this model, the ef-
fective resistivity characterizing the energy dissipation due
to the movement of vortices is determined by the external
conditions and the systems state in a nonlocal way. Physi-
FIG. 9. Magnetization of bodies of rotation: ball, hollow ball, two
cally, this results from the assumption that when the equi-
adjacent cylinders, cone (half cross sections are shown). Numbers are
librium is violated the time needed for the system of vorti-
H
e
/RJ
c
, R is the ball radius.
ces to reach a new metastable state is negligible on the
time scale of observations. Mathematically, the resistivity
in the critical-state model is a Lagrange multiplier and it
variational or quasivariational inequality describing pene-
is excluded as the model is formulated as an evolutionary
tration of the magnetic eld.
In our work such a formulation has been proposed for
two-dimensional and axially symmetric congurations. We
further reformulated the model and derived an equivalent
evolutionary variational inequality in terms of the current
density, which is unknown only in a bounded domain.
This variational inequality served as a basis for an efcient
numerical solution of critical-state problems with the eld-
independent critical current density. The numerical results
were tested by comparison with the known analytical solu-
tions for simple geometries and good accuracy was demon-
strated.
Provided the current distribution in the superconductor
is found, the magnetic eld in any area of interest can be
calculated via the BiotSavart law. This allows us to ac-
count also for any given dependence of critical current
density on the magnetic eld. In this case the inequality
is quasivariational and its numerical solution would need
an additional loop of iterations.
The main advantage of the proposed approach is its
universality; the method allows one to completely avoid
FIG. 8. Ball in a weak external eld. The zero-eld core boundary:
the tracking of the free boundary, does not need any a
, numerical, ---, asymptotic solutions. H
e
/J
c
R 0.1, half cross
section is shown. priori information on the free boundary behavior, and is
BEAN CRITICAL-STATE PROBLEMS 199
therefore applicable without modications even if the free A3. Axially Symmetric Problems
boundary topology is complicated and changes in time.
Transforming each nite element in the cross section
into the unit square by a change of variables, we obtain
APPENDIX: EVALUATION OF INTEGRALS
M
i
1
i
2
, j
1
j
2

e
i
1
i
2

e
j
1
j
2
a(r, z; r, z) d
5
I
i
1
, j
1
,i
2
j
2

,
A1. Innitely Thin Strip in a Perpendicular Field
The nite elements e
i
are intervals (x
i1
, x
i
), x
i
x
i1

d. In this case
where
I
i, j,m

1
0

1
0

1
0

1
0
[(r i)(r j)]
1/2
M
i, j

1
2

e
i

e
j
ln
1
x x
dx dx
d
2
2
ln(d) I
ij
,
k

2
k
2
K(k) E(k) K(k)

,
where the integrals I
m

1
0

1
0
ln(x x m) are evalu-
ated analytically: I
0
, I
1
2 ln(2), and for
m 1
k 2

(r i)(r j)
(r r i j)
2
(z z m)
2

1/2
.
I
m
(m1)
2
ln(m1) m
2
ln(m)
(m1)
2
ln(m1).
The integrand is singular at the points where r i
r j and z z m 0 because at these points k 1
A2. Cylinder in a Perpendicular Field
and the complete elliptic integral of the rst type, K(k),
becomes innite. If there are no such points in the region
The cross section of a cylinder is approximated by a
of integration, a numerical quadrature with a few knots
regular nite element mesh with the same step d in x
1
and
can be used to estimate the integral. Otherwise, one can
x
2
. Denote by e
mn
the element with the left lower corner
use the asymptotic expansion [25]
at the node with coordinates (x
1,m
, x
2,n
). Changing the
variables to transform each element into the unit square,
we obtain
K(k) ln
4
k

1
4

ln
4
k
1

(k)
2
...,
M
i
1
i
2
, j
1
j
2

e
i
1
i
2

e
j
1
j
2
G (x x)
where k 1 k
2
, to single out the nonregular part of
the integrand. In the neighborhood of a singular point,

d
4
4
2 ln(d) I
i
1
j
1
,i
2
j
2

,
k

2
k
2
K(k) E(k) K(k)

K(k) ln
4
k
where the integrals

1
2
ln((r r i j)
2
(z z m)
2
),
I
m,n

1
0

1
0

1
0

1
0
ln((x
1
x
1
m)
2
(x
2
x
2
n)
2
)
and so the singular part of a is
can be evaluated analytically (we used the program for
symbolic computations Maple to nd some of them). How-
a
s
[(r i)(r j)]
1/2
ln((r r i j)
2
ever, the resulting expressions are complicated and it is
(z z m)
2
).
much easier to estimate the nonsingular integrals by means
of a simple (with a few knots) quadrature. The values of
The regular part, a
r
a a
s
, can be integrated numerically,
the singular integrals are given in Table I.
while the singular part can be rst integrated analytically
in z and z, which removes the singularity, and then numer-
ically in r and r. For the innitely thin disk in a perpendicu-
lar eld, the singular part of the integral can be evalu-
TABLE I
ated analytically.
Singular Integrals I
m,n
ACKNOWLEDGMENTS
I
0,0
I
0,1
I
1,0
I
1,1
The author is grateful to S. J. Chapman, J. R. Ockendon, and J. F.
1.610173 0.0130569 0.6885433
Rodrigues for helpful comments on this work. The hospitality of the
200 LEONID PRIGOZHIN
Oxford Centre for Industrial and Applied Mathematics and Centro de 12. K. L. Telschow and L. S. Koo, Phys. Rev. B 50, 6923 (1994).
Matema` tica e Aplicaco

es Fundamentais (Lisbon), where this work was

13. A. Bossavit, IEEE Trans. Magn. 30, 3363 (1994).
done, is highly appreciated. The research was supported by the European
14. J. Wang and M. Sayer, Physica C 212, 395 (1993).
Science Foundation (FBP programme).
15. J. L. Chen and T. J. Yang, Physica C 224, 345 (1994).
16. P. G. de Gennes, Superconductivity of Metals and Alloys (Benjamin,
REFERENCES
New York, 1966).
17. L. Prigozhin, Europ. J. Appl. Math. 7, 237 (1996).
1. C. P. Bean, Phys. Rev. Lett. 8, 250 (1962); Rev. Mod. Phys. 36, 31
18. J. A. Stratton, Electromagnetic Theory (McGrawHill, New York,
(1964).
1941).
2. Y. B. Kim, C. F. Hempstead, and A. R. Strnad, Phys. Rev. Lett. 9,
19. I. Ekeland and R. Temam, Convex Analysis and Variational Problems
306 (1962); Phys. Rev. 129, 528 (1963).
(North-Holland, Amsterdam, 1976).
3. A. M. Campbell and J. E. Evetts, Adv. Phys. 21, 199 (1972).
20. V. L. Berdichevsky, Variational Principles in Continuum Mechanics
4. C. P. Bean, J. Appl. Phys. 41, 2482 (1970).
(Nauka, Moscow, 1983). [Russian]
5. E. H. Brandt, M. V. Indenbom, and A. Forkl, Europhys. Lett. 22,
21. G. Duvaut and J. L. Lions, Inequalities in Mechanics and Physics
735 (1993).
(Springer-Verlag, Berlin, 1976).
6. E. Zeldov et al., Phys. Rev. B 49, 9802 (1994).
22. L. Prigozhin, Phys. Rev. E 49, 1161 (1994).
7. P. N. Mikheenko and Yu. E. Kuzovlev, Physica C 204, 229 (1993).
23. R. Glowinski, J. L. Lions, and R. Tre molie res, Numerical Analysis
8. J. Zhu et al., Physica C 212, 216 (1993).
of Variational Inequalities (North-Holland, Amsterdam, 1981).
9. H. Hashizume et al., IEEE Trans. Magn. 28, 1332 (1992).
24. E. H. Brandt Phys. Rev. Lett. 76, 4030 (1996).
10. M. Ashkin, J. Appl. Phys. 50, 7060 (1979).
25. I. S. Gradstein and I. M. Ryzhik, Table of Integrals, Series, and
Products (Academic Press, New York, 1965). 11. R. Navarro and L. J. Campbell, Phys. Rev. B 44, 10146 (1991).