a

r
X
i
v
:
c
o
n
d
-
m
a
t
/
0
2
0
7
7
4
4
v
1


[
c
o
n
d
-
m
a
t
.
s
t
a
t
-
m
e
c
h
]


3
1

J
u
l

2
0
0
2
Tunneling Rate for Superparamagnetic
Particles by the Instanton Method
E. H. Martins Ferreira
a
, M. C. Nemes
a,b,1
a
Departamento de Fsica, Universidade Federal de Minas Gerais,
CP 702, 30161-970 Belo Horizonte, MG, Brazil
b
Instituto de Fsica, Universidade de S ao Paulo,
CP 66318, 05315-970 S ao Paulo, SP, Brazil
Abstract
We derive the tunneling rate for paramagnetic molecules in the context of a collective
spin model. By means of path integral methods an analytical expression is derived.
Given the very large spins in question (s 3000 h), the observation of magnetization
changes due to pure unitary tunnel eects is unlikely.
1 Introduction
The spin tunneling in magnetic materials has atracted much attention since
experimental evidences of this phenomenon were observed. Although tunnel-
ing is a purely quantum eect, it is interesting to note how it still mani-
fests in superparamagnetic particles of macroscopic size (of the order of tenths
of nanometers). These particles are generally magnetic grains or clusters,
whose spins are highly correlated, and they behave like they were a single
large spin particle[1,2]. The Macroscopic Quantum Tunneling (MQT) is the
tunneling of a single macroscopic variable (in the present case, the magneti-
zation) through a barrier, between two minima of the eective potential of
a macroscopic system. For small magnetic clusters, these minima correspond
to two states of opposite magnetization (easy direction), and the barrier is
proportional to the exchange anisotropy. When there is a repeated coherent
tunneling betwenn the two wells, then we have a case of Macroscopic Quantum
Coherence (MQC), and all spins have their relative orientation unchanged[3].
1
permanent address: a
Preprint submitted to Elsevier Preprint 31 October 2013
The existence of tunneling removes the degenerescence of the energy levels
of the system, and the tunneling rate is given by the energy dierence be-
tween these levels. A useful and well known technique for performing these
calculations is the instanton[4], which is based on the Path Integral formalism
developed by Feynman[5]. In fact, the application of the method, together
with the Villain transformation to the tunneling of spins has been introduced
by Enz and Schilling in ref [6] (see also the review by Schilling in [7]). This is
a non-pertubative method that provides the energy of the fundamental states
with great precision. The proposal of this work is to show how to apply the
instanton technique to the problem of the spin tunneling of a superparamag-
netic particle with a detailed discussion of all the necessary steps to nd the
energy splitting of the lowest levels.
2 The Hamiltonian of the system in terms of canonical operators
We start here with a general Hamiltonian of a single large spin system with
easy direction on the z-axis which has been used in the context of superpara-
magnetic particles[8], given by
H = AS
2
z
+B(S
2
+
+S
2

), (1)
with the anisotropy constants A and B satisfying A B > 0. S
z
and S

=
S
x
iS
y
are the standard spin operators. In the basis of the eigenvectors of
S
2
and S
z
, these spin operators satisfy
S
z
|s, m = m|s, m (2a)
S

|s, m =
_
(s m)(s m + 1)|s, m1 (2b)
To solve this problem, we shall use the Villain transform[9], which consists of
dening the operators S
+
and S

in terms of S
z
and a new operator ,
S
+
= e
i
_
s(s + 1) S
z
(S
z
+ 1) (3a)
S

=
_
s(s + 1) S
z
(S
z
+ 1) e
i
(3b)
with
[ , S
z
] = i. (4)
The eigenvectors of , {|}, are such that
|m =
e
im

2
. (5)
2
To expand S
+
and S

we need that ||S

z
|| s. This would be true, were the
systems easy direction not on the z-axis, but on the y or x-axis. This direction
is clearly arbitrary, so we choose it on the y-axis, and rewrite the Hamiltonian
as
H = AS
2
y
+ 2B(S
2
z
S
2
x
), (6)
or equivalently
H =
A
4
(S
2
+
+S
2

S
+
S

S
+
) + 2B(S
2
z

1
4
(S
2
+
+S
2

+S
+
S

+S

S
+
)).
(7)
Besides the commutation relation (4), we need another one, which can be
derived from it and proves to be much useful, that is
[S
z
, e
im
] = m e
im
. (8)
One implication of the above equation is
f(S
z
) e
i
= e
i
f(S
z
1), (9)
where f(S
z
) is any function of S
z
. With these relations, we can write the
Hamiltonian, in the normal order, in terms of these new operators, that is
H = A[(s(s + 1) S
2
z
)(cos
2
1) +iS
z
sin 2 ] +
2B[S
2
z
(s(s + 1) S
2
z
) cos
2
iS
z
sin 2 ].
(10)
3 Transition amplitude
We have to compute the following matrix element

f
, t
f
|
i
, t
i
=
f
| e
iH(t
f
t
i
)
|
i
. (11)
This is achieved in a standard fashion by discretizing the time in intervals
= (t
f
t
i
)/N and inserting dierent projection operators like
_
d
k
|
k

k
|,
between each time interval. We get

f
, t
f
|
i
, t
i
= lim
N
_
_
N1

k=1
d
k
_

f
| e
iH
|
N1
. . .
1
| e
iH
|
i
(12)
3
Each one of the inner products above is given by

k
| e
iH
|
k1
=
s

m
k
=s

k
|m
k
m
k
| e
iH
|
k1

=
s

m
k
=s
e
im
k
(
k

k1
)
2
e
iH(m
k
,
k1
)
(13)
with
H(m
k
,
k1
)

k
|H|
k1

k
|
k1

=
= A[(s(s + 1) m
2
k
)(cos
2

k1
1) +im
k
sin 2
k1
] +
2B[m
2
k
(s(s + 1) m
2
k
) cos
2

k1
im
k
sin 2
k1
]
(14)
Grouping the m
k
-terms and making the substitution
m
k
+i
(A2B) sin2
k1
2[A+ 2B (A2B) cos
2

k1
]
p
k
(15)
we get, after transforming the summation into an integral and taking the limit
s (see Eq. (2.13) of Johnson[10]),

k
| e
iH
|
k1
=
_
dp
k
2
e
i(p
k
+)(
k

k1
)
e
iH(p
k
,
k1
)
(16)
where
H(p
k
,
k1
)
p
2
k
2M(
k1
)
+V (
k1
)
M(
k1
)
1
2[A+ 2B (A2B) cos
2

k1
]
V (
k1
) (A2B)s(s + 1) cos
2

k1
As(s + 1)
and = 0 if s is integer or =
1
2
if s is half-integer. Now we use Eq. (16)
back into Eq. (12) and take the limit N to nd

f
, t
f
|
i
, t
i
= lim
N
_
_
N1

k=1
d
k
__
N

k=1
dp
k
2
_
exp
_
i
N

k=1
(p
k
+ )
_

k1

_
H(p
k
,
k1
)
_
=
_
D[]D[p] exp
_
i
_
dt
_
(p + )


p
2
2M()
V ()
__
.
(17)
4
Completing square for p
k
, and computing the gaussian integral, we get

f
, t
f
|
i
, t
i
=
_
D[(t)] exp
_
i
_
dt
_

+
M()

2
2
V ()
__
. (18)
Making a Wick rotation, we then have

f
| e
H(
f

i
)
|
i
=
_
D[()] exp
_

_
d
_
i

+
M()

2
2
+V ()
__
=
_
D[()] e
S
E
(19)
from where we nd the classical Hamiltonian and Lagrangian of the system
H(p, ) = [(2B A) cos
2
+ 2B +A]p
2
+ (A2B)s(s + 1) cos
2
As(s + 1)
(20a)
L(,

) =

2
4[(2B A) cos
2
+ 2B +A]
(A2B)s(s + 1) cos
2
+As(s + 1)
(20b)
4 Expansion of the action with variable mass
Let us take s integer, so that = 0, as it is appropriate for superparamagnetic
particles. The case with =
1
2
leads to a topological selection rule, already
discussed in the literature[11]. The Euclidean action is then
S
E
=
_
T/2
T/2
d
1
2
M()

2
+V (). (21)
The instanton is the solution which minimizes the Euclidean action. In the
semi-classical limit, we make an expansion of the action around the instanton

, taking
=

+c
i

i
(22)
(repeated indices indicate summation).
i
are eigenfunctions of an operator
D
op
, and form an orthonormal basis with the boundary condition
i
(T/2) =
5
0. Expanding each one of the terms above, we obtain
M() = M(

+c
i

i
) = M(

) +
M

c
i

i
+
1
2

2
M

c
i

i
c
j

j
(23a)

2
= (

+c
i

i
)
2
=

2
+ 2

c
i

i
+c
i

i
c
j

j
(23b)
V () = V (

) +
V

c
i

i
+
1
2

2
V

c
i

i
c
j

j
(23c)
Substituting the expressions above into Eq. (21) and integrating by parts, the
action becomes
S[] = S[

] +
1
2
_
T/2
T/2
d

i,j
c
i
c
j

i
_
M(

)
2

1
2

2
M

+

2
V

2
_

j
= S
0
+
1
2
_
T/2
T/2
d

i,j
c
i
c
j

i
D
op

j
(24)
The transition amplitude is then

f
| e
H(
f

i
|
i
= e
S
0
[det(D
op
)]
1/2
. (25)
A chain of instantons and anti-instantons is also a solution which minimizes
the action. When we take these solutions into account(see [4]), we get nally

f
| e
HT
|
i
=
_
m

_
1/2
e
T/2

n odd
(K e
S
0
T)
n
n!
, (26)
where T
f

i
and
K = lim
T
_
det(D
op
)
det(
2

+
2
)
_
1/2
. (27)
5 The instanton
To nd the instanton, we write the Euclidean action as
S
E
=
_
d
_
ip

+H(p, )
_
. (28)
Using simple variational calculus, we have that the p coecient must vanish,
and then
i

= 2[(2B A) cos
2

+ 2B +A] p. (29)
6
The condition that the instanton is the solution of minimal energy gives
H( p,

) = As(s + 1)
p
2
=
(2B A)s(s + 1) cos
2

(2B A) cos
2
+ 2B +A
, (30)
and so
i

= 2

2B A
_
s(s + 1)| cos

|
_
(2B A) cos
2
+ 2B +A (31)
We make k
2BA
2B+A
, such that 1 < k < 1, and then

= 2

A
2
4B
2
_
s(s + 1)| cos

|
_
1 +k cos
2
(32)
Now, integrating the equation above, we have
_
d

cos

_
1 +k cos
2

=
_
2
_
s(s + 1)

A
2
4B
2
d. (33)
Making the substitution x = sin

, dx = cos

d

, we can readily solve the

integral to nd
_
dx
(1 x
2
)
_
1 +k(1 x
2
)
=
1
2
ln
_
1 +k kx
2
+x

1 +k kx
2
x
_
(34)
It is easy to check that
_
V

(
0
)/M(
0
) is equal to 2
_
s(s + 1)

A
2
4B
2
(
0
is the initial value of () equal to () = /2), and therefore the
solution is given by

= asin
_
_
_

1 +k sinh[(
0
)]
_
1 + (1 +k) sinh
2
[(
0
)]
_
_
_
(35)
p() =

k
1 +k
2
_
s(s + 1)
cosh[(
0
)]
(36)
There are thus two instanton-like solutions, with positive (I
+
) and negative
(I

) signal of Eq. (36), and the corresponding anti-instantons (

I
+
and

I

),
given by the positive or negative signal of the argument of the function sinh in
Eq. (35). Figure 1 shows the dierent types of instantons and anti-instantons.
They are necessary to the calculation of the Berrys phase later (section 6).
Making the identication p = i
_
s(s + 1) cos(

), we obtain the classical angle

7

() = acos
_
_
_
2
_
k/(1 +k)
cosh[(
0
)]
_
_
_
. (37)
Figure 2 shows the positive graphs of these functions. We can see clearly the
expected result. For example, for the instanton represented in this picture
(I
+
), we have = /2 and = /2 when , and = /2 and
= /2 when , which are the states of minimal energy. The classical
action can now be readily computed
S
0
= i
_
T/2
T/2
d

p = i
_
/2
/2
d

p
S
0
=
_
s(s + 1) ln
_
1 +

k
1

k
_
(38)
5.1 Eingenvalue zero
If we apply the operator D
op
to the function

D
op

= M(

)
2

1
2

2
M

+

2
V

_
M(

1
2

2
M

+
V

_
(39)
we see the result is zero because the term in parantheses is simply the equation
of motion of the system. Then we may conclude that

is the eigenfunction of
D
op
with eigenvalue zero, and we must treat this eigenvalue separately.
If

centered at
0
is an instanton solution, then

centered at
0
+ a must
also be. Thus

0
+a
() =

0
( a)

0
() a

0
() +c
0

0
, (40)
with the normalization condition imposed by
_
T/2
T/2
d
2
0
= N
_
T/2
T/2
d

2
= 1. (41)
The integration over c
0
was already performed while computing S
0
and there-
fore must be taken o the determinant. However, we must correct the fact that
the integration was over
0
, and consequently we need a correct normalization
8
factor. A variation d
0
in the center of the instanton causes a variation
[()] =

d
0
.
Since () =

() +c
i

i
, a variation in c
0
associated to
0
, gives
[()] = dc
0

0
.
As
0
= 1/

, then dc
0
=

N d
0
, and
K =

N
2
lim
T
_
det

(D
op
)
det(
2

+
2
)
_
1/2
. (42)
The prime indicates that the eigenvalue zero must not be computed in the de-
terminant and the factor 1/

2 comes from the denition of D[] =

n
dc
n
/

2.
5.2 Calculation of K
From the theory of dierential equations, the following relation
2
can be de-
rived for the eigenvalue = 0,
det(D
1
op
)
det(D
2
op
)
=

1
(T/2)

2
(T/2)
, (43)
where D
1,2
op
are second order dierential operators in the interval [T/2, T/2]
and
1,2
are eigenfunctions of these operatores, such that

1,2
(T/2) = 0 e

1,2
(T/2) = 1. (44)
We multiply and divide Eq. (43) by
0
(T), which is the smallest eigenvalue
of D
op
and vanishes in the limit T .

() =
1

sinh[( + T/2)] is an
eignefunction of the operator
2

+
2
which obeys the conditions above and
has the form lim
T

(T/2) = e
T
/2 in the limit T . So, in this limit,
Eq. (43) becomes
det(D
op
)
det(
2

+
2
)
1

0
(T)
=
2(T/2)

0
(T) e
T
. (45)
5.3 Calculation of ()
As already mentioned,

is an eigenfunction of the operator D

op
. Now we
calculate this function explicitly starting from the instanton solution (Eq.
2
A prove of this relation may be found in Coleman[4], Appendix 1
9
(35)). Taking the derivative with respect to , we see that

() =

1 +k cosh()
1 + (1 +k) sinh
2
()
.
In the limit , which is the limit of interest, this equation is simply
lim

() =
2

1 +k
e
||
and after normalization, we have

N
() =
1

N
2

1 +k
e
||
C e
||
= x
1
. (46)
This function however doesnt fulll the conditions imposed by Eq. (44), be-
cause

N
( = ) = 0. We then seek a solution y
1
, linearly independent
of x
1
, such that = My
1
+ Nx
1
satisfy the boundary conditions. We write
D
op
=
2
+I()+J

(). If x
1
and y
1
are eigenfunctions of D
op
with eigenvalue
zero, so
_
_
_
y
1
[
2
x
1
+I()x
1
+J

()x
1
] = 0
x
1
[
2
y
1
+I()y
1
+J

()y
1
] = 0
(47)
Subtracting both equations
(y
1
x
1
x
1
y
1
) +I()[y
1
x
1
x
1
y
1
] = 0
and denig W(x
1
, y
1
) = y
1
x
1
x
1
y
1
, the Wronskian of the system and
F/F = I(), we see that
W(x
1
, y
1
) +
F
F
W(x
1
, y
1
) = 0
(FW) = 0 FW = constant.
(48)
As D
op
= M
2

M)

+J(), we can identify F with M(

), and then
M(

)W(x
1
, y
1
) = constant.
We normalize W(x
1
, y
1
) such that M(

)W(x
1
, y
1
) = 2M
0
C
2
, where M
0
is the
limiting value of M() when . Thus, in this limit, W(x
1
, y
1
) = 2C
2
.
Using the Wronskian, we get, in the limit ,
y
1
= C e

,
or equivalently
y
1
=

||
C e
||
. (49)
10
We need now to nd the coecients M and N, in order that satisfy the
boundary conditions. They are
M =
e
T/2
2C
and N =
e
T/2
2C
.
From the above, we nally have
() =
1
2
_
e
T/2
e
||
+

||
e
T/2
e
||
_
(50)
and therefore we nd the value of (T/2) = 1/.
5.4 Calculation of
0
We start with the eigenvalue equation of the operator D
op
,
[M
2

M)

+J()]

() =

()
We write

() in an integral form like

() = () D
_

T/2
ds[y
1
()x
1
(s) x
1
()y
1
(s)]

(s) (51)
where D is such that the solution above satisfy the eigenvalue equation. Taking
the rst and second order derivatives of

(), we have

() =

() D
_

T/2
ds[y

1
()x
1
(s) x

1
()y
1
(s)]

(s)
D[y
1
()x
1
() x
1
()y
1
()]
. .
=0

()

() =

() D
_

T/2
ds[y

1
()x
1
(s) x

1
()y
1
(s)]

(s)
D[y

1
()x
1
() x

1
()y
1
()]
. .
=W(x
1
,y
1
)

().
Now, we substite it back into the eigenvalue equation, and use the equation
of motion, M

() (

M)

() +J()() = 0, to get
[ MDW(x
1
, y
1
)]

() = D
_

T/2
ds
_
M[y

1
()x
1
(s) x

1
()y
1
(s)]
+ (

M)[y

1
()x
1
(s) x

1
()y
1
(s)] J()[y
1
()x
1
(s) x
1
()y
1
(s)]
_

(s)
(52)
11
The term on right is equivalent to
D
_

T/2
ds
_
[M
2

M)

+J()][y
1
()x
1
(s) x
1
()y
1
(s)]
_

(s) (53)
As x
1
() and y
1
() are eigenfunctions of D
op
with eigenvalue zero, then the
right side is null. Thus
[ MDW(x
1
, y
1
)]

() = 0
D =

MW(x
1
, y
1
)
T

2M
0
C
2
(54)
Iterating Eq. (51), we have

() = () D
_
ds[ ](s) +D
2
_
ds[ ]
_
ds

[ ](s

) + (55)
Since 0, then D
2
D, and thus

()
0
()

2M
0
C
2
_

T/2
ds[y
1
()x
1
(s) x
1
()y
1
(s)]
0
(s).
Using the results for x
1
and y
1
, we have, for = T/2

(T/2) =
1

4M
0

2
C
2
_
T/2
T/2
ds[e
T
x
2
1
(s) e
T
y
2
1
(s)]. (56)
This integral can now be readly solved and gives a result equal to C
2
/ e
T
.
Therefore
1

0
C
2
/ e
T
4M
0

2
C
2
= 0

0
= 4M
0

2
e
T
(57)
5.5 Value of K
With the results obtained here so far, the value of K may be readily computed
K =

N
2
lim
T
_
det

(D
op
)
det(
2

+
2
)
_
1/2
=

N
2
lim
T
_
2(T/2)

0
(T) e
T
_
1/2
=
=

N
2
_
2M
0

2
.
M
0
is given by lim
/2
M() = [2(A + 2B)]
1
. N is given by the nor-
malization condition of x
1
, and may be solved to give N = 4/(1 + k).
12
1 + k = 4B/(2B + A) and = 2
_
s(s + 1)

A
2
4B
2
. Hence the nal re-
sult for K is
K =

3/2

2B
. (58)
6 Berrys phase
As seen in section 5, the model has two instanton-like solutions and their two
corresponding anti-instanton. Then, there are two possible paths to go from
one minimum state to the other and two others to return. The action along
any of the paths is the same up to a geometrical phase (Berrys phase) which
is now computed. This phase is given by

B
= exp
__
d is

_
= exp
__
d is
_
For the instanton I
+
and the anti-instanton

I

, the Berry phase is

B
(I
+
) =
B
(

) = exp
_
_
/2
/2
d is
_
= e
is
(59a)
and to the instanton I

and the anti-instanton

I
+
,

B
(I

) =
B
(

I
+
) = exp
_
_
/2
/2
d is
_
= e
is
(59b)
We must introduce these phases in the calculation of Eq. (26), because the
contribution of each instanton is now multiplied by this phase. Taking account
of the four possibilities of a 2-link chain (back and forth), the summation in
Eq. (26) becomes
(K e
S
0

B
(I
+
)T)(K e
S
0

B
(

I
+
)T)
2
+
(K e
S
0

B
(I

)T)(K e
S
0

B
(

)T)
2
+
(K e
S
0

B
(I
+
)T)(K e
S
0

B
(

)T)
2
+
(K e
S
0

B
(I

)T)(K e
S
0

B
(

I
+
)T)
2
=
(K e
S
0
T)
2
(
B
(I
+
) +
B
(I

))(
B
(

I
+
) +
B
(

))
2
(60)
With the results obtianed in (59a) and (59b), we see that
(
B
(I
+
) +
B
(I

))(
B
(

I
+
) +
B
(

)) = 4 cos
2
(s) (61)
and therefore in Eq. (26), we must replace e
S
0
/h
by 2 cos(s) e
S
0
/h
13
7 Splitting of the ground state
Now we can nally calculate the energy splitting of the ground state due to
the ressonant tunelling between them. Summing up Eq. (26), we have

f
| e
HT
|
i
=
_
m

_
1/2
e
T/2
1
2
_
e
2K e
S
0 T cos s
+e
2K e
S
0 T cos s
_
(62)
from what we get the energies of the two lowest states to be
E

=
1
2
2K e
S
0
cos s
Then the splitting in energy is
E
inst
= E
+
E

= 4K e
S
0
cos s
E
inst
= 4
_

2B
_
1/2
_
1 +

k
1

k
_

s(s+1)
cos s
(63)
Replacing the values of and k in terms of A, B and s, and using that s 1,
such that
_
s(s + 1) s +
1
2
, the nal result will be
E
inst
=
8

B
[s(s + 1)(A
2
4B
2
)]
3/4
_
A+ 2B +

A2B

A+ 2B

A2B
_
(s+1/2)
cos s
(64)
If we take the limit B/A 1 and s 1, the equation above is equal to
E
inst
=
8As
3/2

_
B
A
_
s
cos s. (65)
This result may be compared with the expression obtained by Hartmann-
Boutron[12] using other method and with the rigorous universal upper bound
for the tunneling rate of large quamtum spins recently obtained[13] and inde-
pendent of the form of the anisotropy, verifying an excellent agreement with
both results.
We can still test the result, comparing it with the exact value obtained via
diagonalization of the Hamiltonian. In Table 1 we contrast these results, and
it can be seen that, although the analytical calculation is valid for s 1, even
for small s, the discrepance between the values is small.
14
8 Conclusion
We have found an analytical expression for the tunneling rate of a superpara-
magnetic particle using the instanton method and compared it with others
results obtained by dierent methods, and also with numerical calculations.
The result proves to be in good agreement with all of them. It is worth noticing
that for a typical superparamagnetic particle, with s 3000 and f 10
2
,
the tunneling rate is practically zero, and then, no tunneling should be ob-
served in practice. However, the change in magnetization direction of such
particles is observed, what suggests that we must take the interaction with
the environment into account. Work along these lines is in progress.
Acknowledgements
This work is supported by Brazilian research agencies CNPq and FAPESP.
References
[1] P. C. E. Stamp, E. M. Chudnovsky, B. Barbara, Int. J. Mod. Phys. B 6 (1992)
1355.
[2] A. Garg, G.-H. Kim, Phys. Rev. Lett. 63 (1989) 2512.
[3] Garca-Pablos, et al., Phys. Rev. B 53 (1996) 741.
[4] S. Coleman, Aspects of symmetry: selected Erice Lectures, Cambridge
University Press, New York, 1985, Ch. 7: The uses of instantons.
[5] R. P. Feynman, A. R. Hibbs, Quantum mechanics and path integrals, McGraw-
Hill, New York, 1965.
[6] M. Enz, R. Schilling, J. Phys. C 19 (1986) L711; ibid 1765.
[7] R. Schilling, in: L. Gunther, B. Barbara (Eds.), Quantum Tunneling of
Magnetization - QTM 94, Kluwer Publishing, Dordrecht, 1995, p. 59.
[8] H.-D. Pfannes, Hyp. Int. 83 (1994) 79.
[9] J. Villain, J. de Physique 35 (1974) 27.
[10] K. Johnson, Ann. of Phys. 192 (1989) 104.
[11] M. N. Leuenberger, D. Loss, e-print cond-mat/0006075.
[12] F. Hartmann-Boutron, J. Phys. (France) 5 (1995) 1281.
[13] G. Scharf, W. F. Wreszinski, J. L. van Hemmen, J. Phys. A 20 (1987) 4309.
15
Fig. 1. All possible instanton and anti-instanton solutions
Fig. 2. Positive instanton solution for the spin tunneling
Table 1
Comparison between the numerical and calculated results for the ground state en-
ergy splitting for dierent values of s. Here we took f = 0, 2.
s E
calc
0
/A E
num
0
/A
5 5, 61125 10
4
5, 02226 10
4
8 1, 08198 10
6
1, 06013 10
6
10 1, 48684 10
8
1, 51767 10
8
12 1, 93224 10
10
2, 04068 10
10
16