a

r
X
i
v
:
1
2
0
1
.
1
9
7
5
v
1


[
m
a
t
h
.
C
A
]


1
0

J
a
n

2
0
1
2
When physics helps mathematics: calculation of the sophisticated
multiple integral
A. L. Kholodenko
375 H.L.Hunter Laboratories, Clemson University, Clemson, SC 29634-0973, USA

Z. K. Silagadze
Budker Institute of Nuclear Physics and Novosibirsk
State University, 630 090, Novosibirsk, Russia

Abstract
There exists a remarkable connection between the quantum mechanical Landau-Zener problem
and purely classical-mechanical problem of a ball rolling on a Cornu spiral. This correspondence
allows us to calculate a complicated multiple integral, a kind of multi-dimensional generalization of
Fresnel integrals. A direct method of calculation is also considered but found to be successful only
in some low-dimensional cases. As a byproduct of this direct method, an interesting new integral
representation for (2) is obtained.
1
INTRODUCTION
According to Vladimir Arnold [1], mathematics can be considered as some branch of
physics. In writing this note we have no intentions to advocate such point of view. Never-
theless, in our opinion the following calculus problem is a hard nut to crack if only mathe-
matical considerations are being used. The problem lies in exactly calculating the multiple
integral of the type
I
n
=

ds
1
s
1
_

ds
2

s
2n1
_

ds
2n
cos (s
2
1
s
2
2
) cos (s
2
2n1
s
2
2n
). (1)
Because of the s
1
s
2
symmetry, the n = 1 case is simple
I
1
=
1
2!

ds
1

ds
2
_
cos s
2
1
cos s
2
2
+ sin s
2
1
sin s
2
2
_
=

2
, (2)
since its calculation involves known Fresnel integrals

ds cos s
2
=

ds sin s
2
=
_

2
.
However, already for n 2 the above symmetry is lost and things quickly become messy.
The n = 2 and n = 3 cases lay at the borderline. They can be done with some eorts even
though the calculations become noticeably more involved. Unfortunately, they do not admit
an apparent generalization by using the induction method. Already for n = 4 the attempt
to use the same methods meets diculties. Nevertheless, we found a way to calculate such
type of integrals by invoking some physical arguments.
IF YOU CANNOT SOLVE A PROBLEM, THEN THERE COULD BE AN EASIER
PROBLEM YOU CAN SOLVE
Trying to follow George P olyas advice, let us consider the following system of ordinary
dierential equations
d
ds
_
_
_
_
_
_
x
y
z
_
_
_
_
_
_
=
1
R
_
_
_
_
_
_
0 0 sin
as
2
2
0 0 cos
as
2
2
sin
as
2
2
cos
as
2
2
0
_
_
_
_
_
_
_
_
_
_
_
_
x
y
z
_
_
_
_
_
_
. (3)
2
Such system of equations emerges naturally when one tries to describe a motion of a sphere
S
2
rolling on the at surface R
2
without slippage. More accurately, it describes the rolling
of a sphere of radius R along the Cornu spiral on R
2
whose curvature = as is proportional
to the arc-length s [3, 4]. But what is the relation of this problem to our integral (1)?
As the matrices
M(s) =
_
_
_
_
_
_
0 0 sin
as
2
2
0 0 cos
as
2
2
sin
as
2
2
cos
as
2
2
0
_
_
_
_
_
_
do not commute for dierent values of s, the solution of (3) is given by the time ordered
exponential
U(s, s
0
) = T exp
_
_
s
_
s
0
M(s
1
) ds
1
_
_
= 1 +
s
_
s
0
M(s
1
) ds
1
+
s
_
s
0
ds
1
s
1
_
s
0
ds
2
M(s
1
) M(s
2
) +
s
_
s
0
ds
1
s
1
_
s
0
ds
2
s
2
_
s
0
ds
3
M(s
1
) M(s
2
) M(s
3
) + (4)
so that
_
_
_
_
_
_
x(s)
y(s)
z(s)
_
_
_
_
_
_
= U(s, s
0
)
_
_
_
_
_
_
x(s
0
)
y(s
0
)
z(s
0
)
_
_
_
_
_
_
. (5)
Let us take a closer look at the individual terms of the above innite series (4). Writing the
matrix M(s
i
) in the block-form
M
i
M(s
i
) =
_
_
_
0
i

T
i
0
_
_
_,
where

i
=
1
R
_
_
_
sin
as
2
i
2
cos
as
2
i
2
_
_
_,
it is easy to prove by induction that
M
1
M
2
M
2n+1
=
_
_
_
0 (1)
n

T
2

2n1

T
2n

2n+1
(1)
n+1

T
1

2

T
2n1

2n

T
2n+1
0
_
_
_,
and
M
1
M
2
M
2n
=
_
_
_
(1)
n

T
2

2n1

T
2n
0
0 (1)
n

T
1

2

T
2n1

2n
_
_
_.
3
But

T
i

i+1
=
1
R
2
cos
a
2
(s
2
i
s
2
i+1
).
Besides, then our calculations is supplemented by the following initial and boundary condi-
tions
s
0
= , s = , x() = y() = 0, z() = 1, (6)
in view of these relations and taking into account (5) and (4), we nally obtain
z() = 1 +

n=1
(1)
n
R
2n

ds
1
s
1
_

ds
2

s
2n1
_

ds
2n
f
n
(s
1
, s
2
, . . . , s
2n
), (7)
where
f
n
(s
1
, s
2
, . . . , s
2n
) = cos
a
2
(s
2
1
s
2
2
) cos
a
2
(s
2
3
s
2
4
) cos
a
2
(s
2
2n1
s
2
2n
).
After rescaling
s
i

2
a
s
i
, (8)
our original integral (1) indeed shows up in the series (7):
z() = 1 +

n=1
(1)
n
_
2
aR
2
_
n
I
n
. (9)
Thus, if we can nd z(), we can nd I
n
as well! But how can we nd z()?
SPINORIZATION AND THE HOPF MAP
Now we shall follow the advice of Jacques Hadamard: The shortest path between two
islands of truths in the real domain passes through the complex plane. By adopting his
suggestion to our case it is convenient at this point to use two complex variables a and b
instead of three real variables x, y and z through relations [5]
x = ab

+ ba

, y = i(ab

ba

), z = aa

bb

. (10)
Notice: introduction of two complex variables is equivalent to looking at solution for our
problem in C
2
! Furthermore, in view of the initial conditions and (3), the variables x,y and z
are constrained to unit sphere S
2
. This causes the variables a and b to be constrained to S
3
,
that is to obey the equation |a|
2
+|b|
2
= 1. See Ref.[5] for more details. Interestingly enough,
4
under these conditions Eq.10 describes the Hopf map S
3
S
2
[6]. From Ref.[5] it can be
seen that the complex variables a and b must satisfy the following system of dierential
equations
i
d
ds
_
_
_
a
b
_
_
_ =
1
2R
_
_
_
0 e
ias
2
/2
e
ias
2
/2
0
_
_
_
_
_
_
a
b
_
_
_, (11)
if the real variables x, y and z to satisfy (3). Now we can again formally solve (11) by using
the time-ordered exponential series. This time, however, the solution is known and it was
obtained by Rojo in [2]. We just shortly repeat it to ensure the continuity of our exposition.
The solution is formally given by
_
_
_
a(s)
b(s)
_
_
_ = U(s, s
0
)
_
_
_
a(s
0
)
b(s
0
)
_
_
_, (12)
where
U(s, s
0
) = 1 i
s
_
s
0
H(s
1
) ds
1
+ (i)
2
s
_
s
0
ds
1
s
1
_
s
0
ds
2
H(s
1
) H(s
2
) + (13)
with
H(s) =
1
2R
_
_
_
0 e
ias
2
/2
e
ias
2
/2
0
_
_
_.
The imposed initial conditions (6) are now translated into
a() = 1, b() = 0. (14)
Since the unitary evolution (12) conserves the norm aa

+bb

, from (14) we obtain back the

equation for 3-sphere, that is aa

+bb

= 1, valid for any time s. This result implies that

z() = |a()|
2
|b()|
2
= 2|a()|
2
1. (15)
Evidently, we need only to calculate a() to obtain z().
By examining the product of H(s
i
) matrices, we observe that for the odd number of
multipliers the matrix product does not have non-zero diagonal terms and, hence, does not
contribute to a() thanks to the initial conditions (14). The remaining terms with even
number of multipliers have easily calculable non-zero diagonal elements so that we get
a() = 1 +
_
i
2R
_
2

_

ds
1
e
ias
2
1
/2
s
1
_

ds
2
e
ias
2
2
/2
+ .
5
After rescaling (8) this expression acquires the form
a() = 1 +

n=1
(1)
n
_
1
2aR
2
_
n
J
n
, (16)
where
J
n
=

ds
1
e
is
2
1
s
1
_

ds
2
e
is
2
2
s
2
_

ds
3
e
is
2
3

s
2n1
_

ds
2n
e
is
2
2n
. (17)
In contrast to I
n
, the multiple integral J
n
is doable. It can be calculated as follows [2]. First,
we write
J
n
=

ds
1

s
2n1
_

ds
2n
(s
1
s
2
) (s
2n1
s
2n
) e
n
(s
2
1
, . . . , s
2
2n
),
with
e
n
(s
2
1
, . . . , s
2
2n
) = exp {i(s
2
1
s
2
2
+ s
2
3
s
2
4
+ + s
2
2n1
s
2
2n
)}.
Then, we use the integral representation for the Heaviside step function
(s) =
1
2i

d
e
is
i
=
_

_
1, if s > 0,
1
2
, if s = 0,
0, if s < 0.
(18)
Using the result, we can perform the integrals over ds
i
. For this purpose we sequentially
complete the squares, e.g.
i
_
s
2
(
2

1
) + s
2
2
_
= i
_
_
s
2
+

2

1
2
_
2

(
2

1
)
2
4
_
,
and then evaluate the Gaussian integrals

e
is
2
ds =

(cos s
2
i sin s
2
) ds =
_

2
(1 i).
As a result, we nally obtain
J
n
=
n

d
1
2i

d
2n1
2i
exp
_
i
2
[
2
(
1

3
) +
4
(
3

5
) + +
2n2
(
2n3

2n1
)]
_
(
1
i)(
2
i) (
2n1
i)
. (19)
As it can be seen, the terms quadratic in
i
are all canceled thanks to the alternating signs
in exponents in (17). It is this feature that distinguishes, as we shall demonstrate in the next
6
section, the calculation of J
n
from the calculation of I
n
and makes the integral J
n
solvable.
For this purpose we rescale even-index variables
2i
2
2i
in (19) and perform integrals in
these variables taking into account (18). The result is:
J
n
=
n

d
1
2i

d
3
2i

d
2n1
2i
(
1

3
)(
3

5
) (
2n3

2n1
)
(
1
i)(
3
i) (
2n1
i)
,
or
J
n
=

d
1
2i

1
_

d
3
2i

2n3
_

d
2n1
2i

n
(
1
i)(
3
i) (
2n1
i)
. (20)
It is the symmetry of the integrand in (20) that makes the calculation of (20) as easy as the
calculation of I
1
from which our story had begun. In the present case we obtain
J
n
=
n
1
n!
_
_

d
1
2i
1

1
i
_
_
n
=

n
n!
[(0)]
n
=
1
n!
_

2
_
n
. (21)
Then (16) shows that
a() = exp
_

4aR
2
_
, (22)
and from (15) we get
z() = 2 exp
_

2aR
2
_
1 = 1 +

n=1
(1)
n
2
n!
_

2aR
2
_
n
.
Comparing with (9), we nally obtain the desired expression for the integral I
n
:
I
n
=
2
n!
_

4
_
n
. (23)
DIRECT CALCULATION OF I
2
If you are still unhappy by our use of indirect methods of calculation of deceptively simply
looking integral (1), here we discuss some features of the direct method. Unfortunately, as
far as we can see, it works well only for small values of n.
Indeed, let us write
I
2
=

ds
1

ds
2

ds
3

ds
4

4
(s
1
, s
2
, s
3
, s
4
) cos (s
2
1
s
2
2
) cos (s
2
3
s
2
4
), (24)
7
where

4
(s
1
, s
2
, s
3
, s
4
) = (s
1
s
2
) (s
2
s
3
) (s
3
s
4
). (25)
Then, as in the previous section, we use the integral representation for the step function and
notice that

ds
1

ds
2
e
i
1
s
1
e
is
2
(
1

2
)
cos (s
2
1
s
2
2
) =
1
2

ds
1

ds
2
_
exp
_
i

2
1
4
+ i
_
s
1
+

1
2
_
2
+ i
(
1

2
)
2
4
i
_
s
2
+

1

2
2
_
2
_
+
exp
_
i

2
1
4
i
_
s
1

1
2
_
2
i
(
1

2
)
2
4
+ i
_
s
2

2
2
_
2
__
=
cos
_
1
4

2
(2
1

2
)
_
. (26)
Similar calculations are done for integrals over ds
3
and ds
4
. As a result, after rescaling

2
2
2
, we end up with the result:
I
2
=
2

d
1
2i

d
2
2i

d
3
2i
cos [
2
(
1

2
)] cos [
2
(
3

2
)]
(
1
i)(
2
i)(
3
i)
. (27)
The integrals over d
1
and d
3
can be easily calculated. Indeed,

d
1
2i
cos (
1

2
2
)

1
i
=
1
2

d
1
2i
_
e
i
1

1
i
e
i
2
2
+
e
i
1

1
i
e
i
2
2
_
=
1
2
_
(
2
)e
i
2
2
+ (
2
)e
i
2
2
_
.
Therefore,
I
2
=

2
4

d
2
2i
_
(
2
)e
i
2
2
+ (
2
)e
i
2
2
_
2

2
i
. (28)
Now we use the well known result
1
i
= P
1

+ i() (29)
to split the integral (28) into the principal value and the -function parts:
I
2
=

2
4
(I
2P
+ I
2
).
Of course, the -function part is obtained instantly
I
2
=
1
2
.
8
As for the principal value part, we have
I
2P
=
1
2i
lim
0
_
_

e
2i
2
2

2
d
2
+

e
2i
2
2

2
d
2
_
_
=
1
2i
lim
0
_
_

e
2i
2
2

2
d
2

e
2i
2
2

2
d
2
_
_
=

_
0
sin (2
2
2
)

2
d
2
=
1
2

_
0
sin [(

2
2
)
2
]
(

2
2
)
2
d[(

2
2
)
2
] =
1
4
.
In making the last step we have used the Dirichlet integral

_
0
sin

d =

2
.
Putting all terms together, we obtain nally
I
2
=

2
16
. (30)
Which is a special case of (23), as expected.
Can we think about the general case (n > 2) by acting in the manner just described? By
repeating the above steps when n > 2, we obtain
I
n
=
_

2
_
n

_

d
2
2i

d
4
2i

d
2n2
2i
f
n
(
2
,
4
, . . . ,
2n2
)
(
2
i)(
4
i) (
2n2
i)
, (31)
where
f
n
(
2
,
4
, . . . ,
2n2
) = (0,
2
)(
2
,
4
) (
2n4
,
2n2
)(
2n2
, 0), (32)
with
(
1
,
2
) = (
1

2
) e
i(
2
1

2
2
)
+ (
2

1
) e
i(
2
2

2
1
)
. (33)
Evidently, for n > 2 things begin to look rather inconclusive and the above direct method
needs some fresh input in order to be brought to completion.
DIRECT CALCULATION OF I
3
Now let us calculate
I
3
=
_

2
_
3
I,
9
where
I =

dx
2i
(0, x)f(x)
x i
, (34)
and
f(x) =

dy
2i
(x, y)(y, 0)
y i
. (35)
Using the relation (29) in (34), we get
I =
1
2
(0, 0)f(0) + lim
0

dx
2i
(0, x)f(x) (0, x)f(x)
x
.
However, (0, 0) = 1, while
f(0) =

dy
2i
(0, y)(y, 0)
y i
=
1
2

2
(0, 0) + lim
0

dy
2i

2
(0, y)
2
(0, y)
y
,
and since

2
(0, y)
2
(0, y) = e
2iy
2
e
2iy
2
,
when y > 0, we get
f(0) =
1
2

1

_
0
sin (2y
2
)
y
dy =
1
4
,
and, therefore,
I =
1
8
+ K, (36)
where
K = lim
0

dx
2i
e
ix
2
f(x) e
ix
2
f(x)
x
. (37)
Now, using again (29), we have
f(x) =
1
2
(x, 0)(0, 0) + lim
0

dy
2i
(x, y)(y, 0) (x, y)(y, 0)
y
and since in (37) x > 0, we get
f(x) =
1
2
e
ix
2
+ lim
0

dy
2i
(x y)e
ix
2
+ (y x)e
i(2y
2
x
2
)
e
i(x
2
2y
2
)
y
. (38)
Analogously,
f(x) =
1
2
e
ix
2
+ lim
0

dy
2i
e
i(2y
2
x
2
)
(x y)e
ix
2
(y x)e
i(x
2
2y
2
)
y
. (39)
10
In light of (37), (38) and (39),
K = K
1
+ K
2
,
where
K
1
=
1
2

_
0
dx
2i
e
2ix
2
e
2ix
2
x
=
1
2

_
0
sin (2x
2
)
x
dx =
1
8
and
K
2
=

_
0
dx
x

_
0
dy
y
(x y) cos (2x
2
) + (y x) cos (2y
2
) cos 2(x
2
y
2
)
2
2
.
After rescaling
x
x

2
, y
y

2
,
and using
dx
x
=
1
2
d(x
2
)
x
2
, (x y) = (x
2
y
2
), if x > 0 and y > 0,
as well as
cos (x y) = [(x y) + (y x)] cos (x y),
we end up with the result
I = K
2
=
1
4
2

_
0
dx
x
x
_
0
dy
y
[cos (x y) cos x]. (40)
To calculate the integral

I =

_
0
dx
x
x
_
0
dy
y
[cos (x y) cos x],
we introduce a parametric integral related to it:

I() =

_
0
dx
x
x
_
0
dy
y
[cos (x y) cos x].
Note that

I(0) = 0 and

I(1) =

I, so that

I =
1
_
0
d

I()
d
d.
However,
d

I()
d
=

_
0
dx
x
x
_
0
dy sin(x y) =
1

lim
0

dx
x
[cos (1 )x cos x] ,
11
which is the same as
d

I()
d
=
1

lim
0
_

_
(1)
dx
x
cos x

dx
x
cos x
_

_ = lim
0
Ci() Ci((1 ))

,
where
Ci(x) =

_
x
dx
x
cos x
stands for the integral cosine function. Using the well known series representation for this
function
Ci(x) = + ln x +

k=1
(x
2
)
k
2k(2k)!
,
0.5772 being the Euler constant, we get
d

I()
d
=
1

lim
0
[ln ln (1 )] =
ln (1 )

,
and, therefore,

I =
1
_
0
ln (1 )

d = (2) =

2
6
.
We have just proved an interesting identity which seems to be a new integral representation
for (2):
(2) =

_
0
dx
x
x
_
0
dy
y
[cos (x y) cos x]. (41)
Collecting all pieces together, we get nally I = 1/24 and
I
3
=

3
8
1
24
=
2
3!
_

4
_
3
,
in agreement with (23).
CONCLUDING REMARKS
This problem had originally aroused in the context of a remarkable correspondence be-
tween the quantum mechanical Landau-Zener problem (known in the context of molecular
scattering) and purely classical problem of a ball (that is 2-sphere) rolling on a Cornu spiral
(that is on the curve in R
2
, known as Cornu spiral) recently established by Bloch and Rojo
in [3, 4]. In fact, the main ingredient of this connection - the application of the Hopf map
- goes back to Feynman, Vernon and Hellwarth [5] who showed that the quantum evolution
12
of any two-level system is determined by the classical evolution (precession) of the magnetic
dipole moment of unit strength in an eective external magnetic eld.
It is remarkable that physics helps to calculate a complicated integral (1). However, we
suspect that there should be a direct method of calculation. For low values of n, we have
provided some examples of the direct method. It is after the readers to tackle the case of
general n.

Electronic address: string@clemson.edu

Electronic address: silagadze@inp.nsk.su

[1] V. I. Arnold, On teaching mathematics, Russ. Math. Surveys 53, 229 (1998).
[2] A. G. Rojo, Matrix exponential solution of the Landau-Zener problem, arXiv:1004.2914v1
[quant-ph].
[3] A. M. Bloch and A. G. Rojo, Kinematics of the Rolling Sphere and Quantum Spin, Commun.
Inf. Syst. 10, 221-238 (2010).
[4] A. M. Bloch and A. G. Rojo, The rolling sphere, the quantum spin, and a simple view of the
Landau-Zener problem, Am. J. Phys. 78, 1014-1022 (2011).
[5] R. P. Feynman, F. L. Vernon and R. W. Hellwarth, Geometrical Representation of the
Schrodinger Equation for Solving Maser Problems, J. Appl. Phys. 28, 49 (1957).
[6] H. K. Urbantke, The Hopf bration-seven times in physics, J. Geom. Phys. 46, 125 (2003).
13