You are on page 1of 13

Tina Memo No.

2003-003
Internal Report

Products and Convolutions of Gaussian Probability Density


Functions
P.A. Bromiley
Last updated
9 / 4 / 2014

Imaging Sciences Research Group, Institute of Population Health,


School of Medicine, University of Manchester,
Stopford Building, Oxford Road,
Manchester, M13 9PT.

Products and Convolutions of Gaussian Probability


Density Functions
P. A. Bromiley
Imaging Sciences Research Group, Institute of Population Health,
School of Medicine, University of Manchester,
Manchester, M13 9PT, UK
paul.bromiley@manchester.ac.uk
Abstract

It is well known that the product and the convolution of Gaussian probability density functions (PDFs)
are also Gaussian functions. This document provides proofs of this for several cases; the product of
two univariate Gaussian PDFs, the product of an arbitrary number of univariate Gaussian PDFs, the
product of an arbitrary number of multivariate Gaussian PDFs, and the convolution of two univariate Gaussian PDFs. These results are useful in calculating the effects of smoothing applied as an
intermediate step in various algorithms.

The Product of Two Univariate Gaussian PDFs

Let f (x) and g(x) be Gaussian PDFs with arbitrary means f and g and standard deviations f and g

1
f (x) =
e
2f

(xf )2
2 2
f

g(x) =

and

Their product is

1
e
f (x)g(x) =
2f g

Examine the term in the exponent


=

(xf )2
2 2
f

(x )
22g
1
g
e
2g

(xg )2
2
2g

(x g )2
(x f )2
+
2
2f
2g2

Expanding the two quadratics and collecting terms in powers of x gives


=

(f2 + g2 )x2 2(f g2 + g f2 )x + 2f g2 + 2g f2


2f2 g2

Dividing through by the coefficient of x2 gives

x2 2

f g2 +g f2
x
f2 +g2

2f g2 +2g f2
f2 +g2

2 2

2 2f+g2
f

This is again a quadratic in x, and so Eq. 2 is a Gaussian function. Compare the terms in Eq. 5 to a the usual
Gaussian form
(x)2
(x2 2x+2 )
1
1
2 2
e 22 =
e
P (x) =
2
2
Since a term that is independent of x can be added to complete the square in , this is sufficent to complete the
proof in cases where the normalisation can be ignored. The product of two Gaussian PDFs is proportional to a
Gaussian PDF with a mean that is half the coefficient of x in Eq. 5 and a standard deviation that is the square
root of half of the denominator i.e.
s
f2 g2
f g2 + g f2
and

=
f g =
f
g
f2 + g2
f2 + g2

i.e. the variance f2 g is twice the harmonic mean of the individual variances f2 and g2 , and the mean f g is the
sum of the individual means f and g weighted by their variances. In general, the product is not itself a PDF
as, due to the presence of the scaling factor, it will not have the correct normalisation.
The product f (x)g(x) can now be written in the usual Gaussian form directly, with an unknown scaling constant
(this may be sufficient in cases where renormalisation can be applied). Alternatively, proceeding from Eq. 5,
suppose that is the term required to complete the square in i.e.
2 
2

f g2 +g f2
f g2 +g f2

2
2
2
2
f +g
f +g
=0
=
2 2 2
f

(f2 +g2 )

Adding this term to gives

x2 2x

f g2 +g f2
f2 +g2

f g2 +g f2
f2 +g2

2f2 g2
(f2 +g2 )

2

2f g2 +2g f2
f2 +g2

f g2 +g f2
f2 +g2

2f2 g2
(f2 +g2 )

2

After some manipulation, this reduces to


2

f 2 +g 2
2
x g2 +2 f
(x f g )
(f g )2
(f g )2
g
f
=
=
+
+
2
2

2(f2 + g2 )
2f2 g
2(f2 + g2 )
2 2f+g2
g

Substituting back into Eq. 2 gives


#
"
#
"
(x f g )2
(f g )2
1
exp
exp
f (x)g(x) =
2f g
2f2 g
2(f2 + g2 )
Multiplying by f g /f g and rearranging gives
#
#
"
"
1
1
(f g )2
(x f g )2
q
=
exp
exp
2f2 g
2(f2 + g2 )
2f g
2(f2 + g2 )
Therefore, the product of two Gaussians PDFs f (x) and g(x) is a scaled Gaussian PDF
"
#
Sf g
(x f g )2
f (x)g(x) =
exp
2f2 g
2f g
where
f g =

f2 g2
f2 + g2

and

f g =

f g2 + g f2
f2 + g2

(1)

and the scaling factor S is itself a Gaussian PDF on both f and g with standard deviation
Sf g

"

(f g )2
=q
exp
2(f2 + g2 )
2(f2 + g2 )
1

f2 + g2

These can be written more conveniently as


1
1
1
= 2 + 2
f2 g
f
g

f g =

g
f
+ 2
f2
g

f2 g

and

Sf g = r

1
2 2
2 f2 g
fg

"

1 (f g ) 2
f g
exp
2 f2 g2

(2)

It is much easier to generate a proof by induction for the scaling factor of products of larger numbers of Gaussians
if it is written in the form of a sum of terms, each of which involves a single subscript i.e. the parameters of a
single Gaussian PDF. Appendix A provides the necessary proof, giving
"
!#
2f g
2g
1 2f
1
exp
+ 2 2
(3)
Sf g = r
2 f2
g
f g
2 2
2 f2 g
fg

The Product of n Univariate Gaussian PDFs

Let N (, ) represent a Gaussian PDF with mean and standard deviation . Let subscript i refer to an
individual Gaussian PDF in a product of n univariate Gaussian PDFs. Furthermore, let the subscript i = 1...n
refer to the parameters of the distribution that is the product n individual Gaussian PDFs and subscripts of the
form i = (1...n 1)n refer to the parameters of a distribution that is the product of two Gaussian PDFs, one of
which is itself the product of n 1 Gaussian PDFs. Therefore, the results from Section 1 can be applied to the
first two Gaussian PDFs in the product of n Gaussian PDFs to produce a Gaussian PDF and a scaling factor. The
remaining n 2 PDFs can then be introduced iteratively using the same expressions i.e.
n
Y

N (i , i ) = Si=1...2 N (i=1...2 , i=1...2 )

n
Y

N (i , i )

i=3

i=1

= Si=1...2 S(i=1...2)3 N ((i=1...2)3 , (i=1...2)3 )

n
Y

N (i , i ) = ...

i=4

= Si=1...2 ...S(...((i=1...2)3)...n) N ((...((i=1...2)3)...n) , (...((i=1...2)3)...n) ) = Si=1...n N (i=1...n , i=1...n )


Applying the expression for the standard deviation from Eq. 2 iteratively gives
1
2
i=1...n

1
2
i=1...n1

X 1
1
1
1
1
=
=
...
=
+
+
2
2
n2
i=1...n2
n1
n2
2
i=1 i

(4)

Similarly, the mean is given by






 2
i=1...n1
i=1...n2
n 2
n1 i=1...n1
n 2
i=1...n =
+ 2 i=1...n =
+ 2
+ 2 i=1...n
2
2
2
i=1...n1
n
i=1...n2
n1 i=1...n1
n
"
#


n
X
n1
n 2
i
i=1...n2
2

=
...
=
+
+
=
i=1...n
2
2
2 i=1...n
i=1...n2
n1
n2

i=1 i

(5)

By inspection of Eq. 3, state the form


1

Si=1...n =

(2)(n1)/2

"
2
i=1...n
1
Qn
2 exp 2

i=1 i

n
X
2

i
2

i=1 i

2
i=1...n
2
i=1...n

!#

(6)

for the scaling factor. Similarly, using Eq. 4 to manipulate some of the standard deviation terms,
s
"
!#
2
2(i=1...n)(n+1)
i=1...n+1
2n+1
1 2i=1...n
1
exp
+ 2 2
S(i=1...n)(n+1) =
2
2
2
n+1
2 i=1...n
n+1
(i=1...n)(n+1)
(2)(1/2) i=1...n
The scaling factor is the product of individual scaling factors for each pairwise multiplication, so
Si=1...n+1 = Si=1...n S(i=1...n)(n+1) =
1
(2)n/2

"
2
2
i=1...n+1
1
i=1...n
Qn
exp
2
2
2
2
i=1 i i=1...n n+1

n
X
2

i
2

i
i=1

2(i=1...n)(n+1)
2
+ n+1

2
2
n+1
(i=1...n)(n+1)

!#

This gives two terms to deal with: First, the standard devation term
Qn
Qn
Qn+1 2
2
2
2
2 2
n+1
i=1 i
i=1 i i=1...n n+1
i=1 i
=
=
2
2
2
2
i=1...n
i=1...n+1
i=1...n+1
i=1...n+1
Second, the term in the exponent; using Eq. 5 gives
2(i=1...n)(n+1)
2
(i=1...n)(n+1)

Therefore

n
X
2

i
2

i
i=1

n
n+1
X
X 2
2i=1...n+1
2n+1
2n+1
2i
2i=1...n
i
=
=
+
+
=
2
2
2
2
i=1...n
n+1
2
n+1
2
i=1...n+1
i=1 i
i=1 i

n+1
X 2
2(i=1...n)(n+1)
2(i=1...n+1)
2n+1
i

=
2
2
2
n+1
(i=1...n)(n+1)
2
(i=1...n+1)
i=1 i

So
1
=
(2)n/2

Si=1...n+1

n+1
X

"
2
i=1...n+1
1
Qn+1 2 exp
2

i=1 i

i=1

2
2i
i=1...n+1
2
2
i
i=1...n+1

!#

which, together with Eq. 3, constitutes a proof by induction of Eq. 6. As with the product of two univariate
Gaussian PDFs, the scaling factor is a Gaussian function. However, it is not a PDF, as it does not have the correct
normalisation.

The Product of n Multivariate Gaussian PDFs

The multivariate Gaussian PDF can be written as


p(x) =

1
(2)d/2

|V |

exp

1
(x )T V 1 (x )
2

where is the mean vector, V 1 the covariance matrix, and d is the dimensionality of x; adopt the standard
notation using bold face symbols to represent vectors and matrices. The Gaussian PDF can also be written in
canonical notation as


1 T
T
(7)
p(x) = exp + x x x
2
where
= V 1

= V 1

and

So the product of n Gaussian PDFs i = 1...n is

n
Y
pi (x) = exp i=1...n +
i=1

where

i=1...n
So
n
Y

i=1

n
X

1
i =
=
2
i=1


1
d log 2 log|| + T 1
2

n
X

i=1

nd log 2

pi (x) = exp i=1...n + n n +

!T

1
x xT
2

n
X
i=1

n
X

log |i | +

i=1

!T

n
X
i=1

n
X

!
i x

iT 1
i i

i=1

1
x xT
2

1
= exp(i=1...n n )exp n + nT x xT n x
2

where
n =

n
X

n
X

i=1

!
i x
(8)

i=1

i=1

and

n =

n
X


1
d log 2 log|n | + nT 1
(9)
n n
2
Comparing Eqs. 7, 9 and 8 shows that the result is, as in the previous sections, a scaled Gaussian PDF over x
with a mean vector and covariance matrix given by
n =

Vn1 =

n
X

Vi1

Vn1 n =

and

n
X
i=1

i=1

The scaling factor is again a Gaussian function.

Vi1 i

The Convolution of Two Univariate Gaussian PDFs

We wish to find the convolution of two Gaussian PDFs


f (x) =

1
e
2f

(xf )2
2 2
f

g(x) =

and

(x )
22g
1
g
e
2g

in the most general case i.e. non-identical means. The convolution of two functions f (t) and g(t) over a finite
range1 is defined as
Z x
f (x )g( )d = f g
0

However, the usual approach is to use the convolution theorem [2],


F 1 [F (f (x))F (g(x))] = f (x) g(x)
where F is the Fourier transform
F (f (x)) =

f (x)e2ikx dx

and F 1 is the inverse Fourier transform


F 1 (F (k)) =

F (k)e2ikx dk

Using the transformation


x = x f
the Fourier transform of f (x) is given by
F (f (x)) =

1
2f

x2
2 2
f

e2ikf
e2ik(x f ) dx =
2f

x2
2 2
f

e2ikx dx

Using Eulers formula [2],


ei = cos i sin

we can split the term in ex to give


e2ikf
F (f (x)) =
2f

x2
2 2
f

[cos(2kx ) i sin(2kx )] dx

The term in sin (x ) is odd and so its integral over all space will be zero, leaving
e2ikf
F (f (x)) =
2f

x2
2 2
f

cos(2kx ) dx

This integral is given in standard form in [1]


Z

eat cos (2xt) dt =


0

1
2

and so
F (f (x)) = e2ikf e2

x2
e a
a

f2 k2

The second term in this expression is a Gaussian PDF in k: the Fourier transform of a Gaussian PDF is another
Gaussian PDF. The first term is a phase term accounting for the mean of f (x) i.e. its offset from zero. The Fourier
transform of g(x) will give a similar expression, and so
F (f (x))F (g(x)) = e2ikf e2
1 In

f2 k2 2ikg 2 2 g2 k2

practice, convolutions are more often performed over an infinite range


Z
f (x )g( )d = f g

= e2ik(f +g ) e2

(f2 +g2 )k2

Comparing Eq. 25 to Eq. 24, we can see that it is the Fourier transform of a Gaussian PDF with mean and
standard deviation
q
f g = f + g and f g = f2 + g2
and therefore, since the Fourier transform is invertible,

Pf g (x) = F 1 [F (f (x))F (g(x))] = q

1
2(f2 + g2 )

(x(f +g ))2
2)
2( 2 +g
f

It may be worth noting a general result at this point; the area under a convolution is equal to the product of the
areas under the factors

Z Z
Z
f (u)g(t u)du dt
(f g)dt =

f (u)

Z

Z

f (u)du

g(t u)dt du

 Z

g(t)dt

Therefore, the preservation of the normalisation when convolving PDFs i.e. the fact that the convolution is also a
PDF, normalised such that the area under the function is equal to unity, is a special case rather than being true
in general.

Summary

It is well known that the product and the convolution of a pair of Gaussian PDFs are also Gaussian. In the case
of the product of two univariate Gaussian PDFs N (f , f ) and N (g , g ), the result is a scaled Gaussian PDF
where the scaling factor is itself a Gaussian PDF on both f and g
"
!
#
Sf g
(x f g )2
1
1
g
f
1
N (f , f )N (g , g ) =
exp
= 2 + 2 , f g =
+ 2 f2 g
where
2f2 g
f2 g
f
g
f2
g
2f g
and

Sf g = r

1
f2 g2
f2 g

"

1 (f g ) 2
f g
exp
2 f2 g2

It should be noted that this result is not the PDF of the product of two Gaussian random variates; in that case,
the product normal distribution applies.
The product of n univariate Gaussian PDFs is given by
n
Y

(x i=1...n )2
N (i , i ) = p
exp

2
2
2i=1...n
2i=1...n
i=1


Si=1...n

and

Si=1...n =

(2)(n1)/2

where

1
2
i=1...n

n
X
1
=
2
i=1 i
n
X
2

"
2
i=1...n
1
Qn
2 exp 2

i=1 i

i
2

i=1 i

i.e. is a Gaussian PDF scaled by a Gaussian function.

i=1...n

2
i=1...n
2
i=1...n

The product of n multivariate Gaussian PDFs is given by


n
Y



1
N (i , Vi1 ) = exp(i=1...n n )exp n + nT x xT n x
2
i=1
where
i = Vi1

i = Vi1 i

n =

n
X

i=1

n =


1
d log 2 log|n | + nT 1
n n
2
7

n =

n
X
i=1

and

!#

#
n
X
i
2
=
2 i=1...n

i
i=1
"

i=1...n

n
X

1
i =
=
2
i=1

nd log 2

i.e. a Gaussian PDF scaled by a Gaussian function.

n
X
i=1

log |i | +

n
X

iT 1
i i

i=1

The convolution of two Gaussian PDFs is a Gaussian PDF with mean and standard deviation
q
f g = f + g and f g = f2 + g2
These results can be useful in a number of applications; for example, the convolution of Gaussian distributions
freqently occurs in smoothing applied as an intermadiate step in various machine vision algorithms. Products of
Gaussian PDFs may occur during the application of Bayes theorem, and in some problems related to Gaussian
processes.

Acknoweldgements

Thanks are due to David Kirchheimer, University of Bristol, for pointing out the importance of the product of an
arbitrary number of Gaussian PDFs and providing Matlab code for numerical testing of the results.

References
[1] M Abramowitz and I A Stegun. Handbook of Mathematical Functions. National Bureau of Standards, Washington DC, 1972.
[2] M L Boas. Mathematical Methods in the Physical Sciences. John Wiley and Sons Ltd., 1983.

Rewriting the Scaling Factor

Using Eq. 4 and 5


2i=1...n
=
4
i=1...n
So

n
X
i
2
i=1 i

!2

n
X
2

i
4

i
i=1

+2

n
X
i j
2 2
i=1 j=i+1 i j

n1
X

n
n
X
i j
2i=1...n X 2i
2
= 4

2 2
i=1...n i=1 i4
i=1 j=i+1 i j
n1
X

The terms in the exponent of the scaling factor for the product of univariate Gaussians take the form
!
n
n1
n
n1
X X
X X
2j
2i
2i j
(i j )2 2
2
i=1...n =
i=1...n
2 2
2 2 2 2 + 2 2

i j
i j
i j
i j
i=1 j=i+1
i=1 j=i+1
which, substituting the above expression for the cross term,
!
n
n1
n
X X
X
2j 2
2i 2
2i=1...n 2
2i 2
=

i=1...n
+

+
4
i2 j2 i=1...n i2 j2 i=1...n
4 i=1...n i=1...n
i=1 j=i+1
i=1 i
=

n1
X

n
X

i=1 j=i+1

2i

2
2
i=1...n
i=1...n
+
i2 j2
i4

X
2
=
i=1...n
2
i=1...n
i=1

2i
i2

2i=1...n
2
i=1...n

Note: Appendices B and C are an older version of the derivation for the product of n univariate Gaussian PDFs;
they do not use the manipulation given in Appendix A and so are considerably more complicated. The aim here
was to illustrate the derivation using the proof for three Gaussian PDFs, and then to replicate each step with
n Gaussian PDFs. However, these derivations are made redundant by the simpler versions given in the main
document.

The Product of Three Univariate Gaussian PDFs

Since the product of two Gaussian PDFs is a scaled Gaussian PDF, the above proof can be extended to give the
product of larger numbers of Gaussian PDFs. We adopt the following notation: N (, ) denotes a Gausian PDF
with mean and standard deviation ; subscripts f , g, h etc. indicate the parameters of individual Gaussian
PDFs in the product; subscripts e.g. f g indicate the parameters of the products of those distributions; subscripts
e.g. (f g)h indicate the parameters of the product of the distribution h with a distribution that is itself the product
of the distributions f and g. Therefore, the product of three Gaussian PDFs is
N (f , f )N (g , g )N (h , h )
= Sf g N (f g , f g )N (h , h )
= Sf g S(f g)h N ((f g)h , (f g)h )
Defining
Sf gh N (f gh , f gh ) = Sf g S(f g)h N ((f g)h , (f g)h )

(10)

we have
Sf gh = Sf g S(f g)h

f gh = (f g)h

and

f gh = (f g)h

Since the expressions for the mean and standard deviation in Equation 2 are expressed as the sums over individual
terms that feature only the parameters of a single distribution f , g, they can be extended to multiple distributions
easily
1
1
1
1
1
1
1
= 2
(11)
= 2 + 2 = 2 + 2+ 2
f2 gh
(f g)h
f g
h
f
g
h
and
f gh = (f g)h =

h
f g
+ 2
f2 g
h

2
(f
g)h

h
g
f
+ 2+ 2
f2
g
h

f2 gh

The scaling factor is given by


Sf gh = Sf g S(f g)h

"
#
#
(f g )2
(f g h )2
1
q
exp
exp
=q
2(f2 + g2 )
2(f2 g + h2 )
2(f2 + g2 )
2(f2 g + h2 )
1

"

"

1
q
exp
=
2
2 (f2 + g2 )(f2 g + h2 )
1

(f g h )2
(f g )2
+
f2 + g2
f2 g + h2

!#

This can be dealt with as two separate terms; first

(f2 + g2 )(f2 g + h2 ) = (f2 + g2 )(


Second

f2 g2
+ h2 ) = f2 g2 + f2 h2 + g2 h2
+ g2

f2

(f g )2 (f2 g + h2 ) + (f g h )2 (f2 + g2 )
(f g h )2
(f g )2
+
=
2
2
2
f + g2
f g + h
(f2 + g2 )(f2 g + h2 )

The denominator is the same as the first term, above, so only the numerator need be dealt with
M = (f g )2 (f2 g + h2 ) + (f g h )2 (f2 + g2 )
Substituting the expressions for f g and f g from Eq. 1,
M = (f

g )2 h2

#2
"
f2 g2
f g2 + g f2
+ (f g ) 2
+
h (f2 + g2 )
f + g2
f2 + g2
2

(12)

One approach is expand the expression fully and then pair all of the terms to give an overall factor of f2 + g2 .
However, this is impractical as a route to a proof for the product of arbitrary numbers of Gaussians. Instead,
M = (f g )2 h2 + (f

f2 g2
g )2 2
f + g2

i2
h
f g2 + g f2 h (f2 + g2 )
(f2 + g2 )

i2
h
2
2
2 2
(

)
+
(

)
f
h
g
h

g
f
f g
= (f g )2 h2 + (f g )2 2
+
f + g2
f2 + g2
= (f g )2 h2 + (f g )2

f2 g2
(g h )2 f4
f2 g2
(f h )2 g4
+
+
2(

)(

)
+
f
h
g
h
f2 + g2
f2 + g2
f2 + g2
f2 + g2

Now, observe that


(A B)2 + 2(A C)(B C) = A2 2AB + B 2 + 2AB 2AC 2BC + 2C 2
= A2 2AC + C 2 + B 2 2BC + C 2 = (A C)2 + (B C)2

(13)

Therefore
(f g )2

f2 g2
f2 g2
f2 g2
f2 g2
2
2
+
2(

)(

)
=
(

)
+
(

)
f
h
g
h
f
h
g
h
f2 + g2
f2 + g2
f2 + g2
f2 + g2

and so
M = (f g )2 h2 +

(g h )2 f4
f2 g2
f2 g2
(f h )2 g4
2
2
+
+
(

)
+
(

)
f
h
g
h
f2 + g2
f2 + g2
f2 + g2
f2 + g2

= (f g )2 h2 + (f h )2

4 + f2 g2
g4 + f2 g2
2 f
+
(

)
g
h
f2 + g2
f2 + g2

= (f g )2 h2 + (f h )2 g2 + (g h )2 f2
Collecting terms, this gives
Sf gh

"

1 (f g )2 h2 + (f h )2 g2 + (g h )2 f2
q
exp
=
2
f2 g2 + f2 h2 + g2 h2
2 f2 g2 + f2 h2 + g2 h2
1

which can be written more conveniently as


"
1
1
r
exp
Sf gh =
2
2
2
2
f g h
2
2

(f g )2
(f h )2
(g h )2
+
+
f2 g2
f2 h2
g2 h2

f gh

Therefore, the product of three Gaussian PDFs is a scaled Gaussian PDF


"
#
(x f gh )2
Sf gh
exp
f (x)g(x)h(x) =
2f2 gh
2f gh
where f gh , f gh and Sf gh are given by Eqs. 11, 12 and 14 respectively.
As in Section 1, the scaling factor can be rewritten using Appendix A to give
"
!#
2f gh
2g
1 2f
1
2h
r
exp
Sf gh =
+ 2+ 2 2
2
2 f2
g
h
f gh
f2 g2 h
2
2
f gh

10

f2 gh

(14)

The Product of n Univariate Gaussian PDFs

Let subscript i refer to an individual Gaussian PDF in a product of n univariate Gaussian PDFs. Based on
the derivations in Sections 1 and B, it is clear that the product is also a Gaussian PDF, multiplied by a scaling
factor. The notation used in Section B is extended, so that the subscript i = 1...n refers to the parameters of the
distribution that is the product n individual Gaussian PDFs and subscript i = (1...n 1)n refers to the parameters
of a distribution that is the product of two Gaussian PDFs, one of which is itself the product of n 1 Gaussian
PDFs. In addition, define

n =

n
n
X

i2

i=1

and

n =

n
Y

i2

i=1

j=1
j6=i

By inspection of the results for the products of two and three Gaussian PDFs, state
n
Y

i=1

where
1
2
i=1...n

and

N (i , i ) = p

n
X
1
=
2

i=1 i

Si=1...n

or

2
i=1...n

Si=1...n
2
2i=1...n

n
=
n

(xi=1...n )2
2 2
i=1...n

i=1...n

#
n
X
i
2
=
2 i=1...n

i
i=1
"

n
n1
2
2
X X
(

i
j

exp i=1...n
=p
2 2
2

(2)n1 n
i
j
i=1 j=i+1

(15)

The above expressions can be proved by observing that, following Eq. 10,

Si=1...n N (i=1...n , i=1...n ) = Si=1...n1 N (i=1...n1 , i=1...n1 )N (n , n )


= Si=1...n1 S(i=1...n1)n N (i=1...n , i=1...n )

(16)

Therefore, using Eq. 2


1
2
i=1...n

1
2
i=1...n1

1
n2

and

i=1...n =

n
i=1...n1
+ 2
2
i=1...n1
n

2
i=1...n

2
2
2
Eq. 2 can then be substituted to expand i=1...n1
into i=1...n2
and n1
, and i=1...n1 into i=1...n2 and
n1 ; repeating this gives

1
2
i=1...n

i=1...n

1
2
i=1...n1

X 1
1
1
1
1
= 2
+ 2 + 2 = ... =
2
n
i=1...n2
n1
n
2
i=1 i

Q.E.D.




 2
n 2
i=1...n2
n1 i=1...n1
n 2
i=1...n1
=
+ 2 i=1...n =
+ 2
+ 2 i=1...n
2
2
2
i=1...n1
n
i=1...n2
n1 i=1...n1
n
" n
#


X i
i=1...n2
n1
n 2
2
=

=
...
=
+
+
i=1...n
2
2
2 i=1...n Q.E.D.
i=1...n2
n1
n2

i
i=1


Eq. 15 can be written using as

Si=1...n

n
n
Y
X
1
1
1

2
exp
k2
=p
(i j )

n1
2
n
n
(2)
i=1 j=i+1
k=1

n1
X

k6=i,j

Similarly, Eq. 2 gives the scaling factor for the product of N (i=1...n , i=1...n ) and N (n , n ) as
"
#
2
1 (i=1...n n+1 )
1
exp
S(i=1...n)n+1 = q
2
2
2 i=1...n
+ n+1
2( 2
+ 2 )
i1...n

n+1

11

Therefore, the aim here is to show that

Si=1...n+1 = q

1
(2)n

2
n (i1...n

2
+ n+1

The standard deviation term is

n
n
2
Y
X
(i=1...n n+1 )
1
1

2
k2
+
exp
(i j )

2
2
2

i=1...n
+ n+1
n
)
i=1 j=i+1

n1
X

k=1
k6=i,j

2
2
2
2
2
n (i1...n
+ n+1
) = n i1...n
+ n n+1
= n + n n+1
= n+1

The exponential term, ignoring the 1/2, is

n
n
n1
2
Y
(i=1...n n+1 )
1
X X
2
k2
+
(i j )

2
2
n
i=1...n
+ n+1
i=1 j=i+1
k=1
k6=i,j


Pn1 Pn
i=1

j=i+1

(i j )2

Qn

k2



k=1
k6=i,j
2
(i=1...n

2
2
) + (i=1...n n+1 ) n
(i=1...n
+ n+1

2
+ n+1
)n

The denominator is, as expected, the standard deviation term that was dealt with above; ignore this, and let the
numerator be called M

n
n
n1
Y
X
X
2

2
2
2
+ n+1
) + (i=1...n n+1 ) n
k2 (i=1...n
M =
(i j )
i=1 j=i+1

n1
X

n
X

i=1 j=i+1

2
(i j )

n
Y

k=1
k6=i,j

k=1
k6=i,j

2
+
k2 i=1...n

n1
X

n
X

i=1 j=i+1

Focussing on the last of these three terms

"

n
X
i=1

i
i2

=
However,

2
i=1...n
n+1

n
X

#2

n
Y

= n

"

n
X
i=1

n
X

i
i2

2
(i j )

n
Y

k=1
k6=i,j

2
i=1...n
n+1
2
i=1...n

2
+
k2 n+1

#2

n
Y

n
X
i
2
i=1 i

"

n
X

n
Y

2
i=1...n

n
X

j6=i

n
Y

j6=i

n1
X

n
X

1
2
2

n+1
i=1...n
j

n i=1 j=1

n
X

j6=i

n
Y

j6=i

(i n+1 )2

j4
(i n+1 )(j n+1 )
+2

n i=1
i=1 j=i+1
j=1
j6=i

Qn

n
Y

k=1

k2

n
Y

l=1
l6=i,j

l2

k2

2
= i=1...n
n
so, recombining this with the first two terms of M gives

n
n
n1
n
n
n1
Y
X
X
Y
X X
2

2
2
2
+
k2 i=1...n
k2 n+1
M=
(i j )
+
(i j )
k=1

i=1 j=i+1

i=1 j=i+1

k=1
k6=i,j

k=1
k6=i,j

n
n
n1
n
n
Y
X
X
Y
X

1
2

(i n+1 )2
k2 i=1...n
j4
(i n+1 )(j n+1 )
+2
n i=1
i=1 j=i+1
j=1
k=1
k6=i,j

j6=i

12

1
1
2
2
2
n+1
i
(i n+1 )

j
j
j

n i=1 j=1 i=1


n i=1
j=1
j=1

n+1

#2

Applying Eq. 13 to the second and fourth terms gives

n
X

n1
X

i=1 j=i+1

2
(i j )
n
X

n1
X

n
X

i=1 j=i+1

k=1
k6=i,j

n
n
Y
X
1

2
2
k2 n+1
+
(i j )
n
i=1 j=i+1

n1
X

k=1
k6=i,j

2
k2 n+1

k=1
k6=i,j

j=1
j6=i

j4

2
k2 i=1...n


n 
2
1 X
2 n
+
(i n+1 ) 4
n i=1
i

n 
X

(i n+1 )

i=1

n
Y

n
Y

n
X

n
2
i2 j2 i=1...n

j=1
j6=i


2

n
i4

n
n
X
X

n
(i n+1 )2
2

2 2 i=1...n
j=1
j6=i

i=1

n
n
n
n
2
Y
X
X
2

n i=1...n
X

2
2
(i n+1 )2 n +

k2 n+1
+
(i j )
2 2
n 4

i
i
j
i=1
i=1 j=i+1
j=1

n1
X

Now, examine

(i n+1 )2


(i n+1 )2 + (j n+1 )2

k=1
k6=i,j

n
X

j6=i

n
n
n
2 X
2
Y
n2
n i=1...n
1
n2
1 X 1
i=1...n
1
2
1 +
= n
j2
+
=
=

=
=
i=1...n n 2
n i4 j=1 i2 j2
n i4 j=1 i2 j2 n j=1 i2 j2
i j=1 j2
i2
j=1
j6=i

So,

n
Y

i=1 j=i+1

i=1

2
2
(i n+1 ) + (j n+1 )

2
(i j )

n
X

n1
X

k=1
k6=i,j

2
k2 n+1
+
n

n
X

i=1 j=i+1

n1
X

n
Y

M=

n1
X

n
X

i=1 j=i+1

n
X
j6=i

2
(i j )

n
Y

k=1
k6=i,j

j6=i

2
k2 n+1
+

Collecting terms
Si=1...n S(i=1...n)n+1 = p

1
(2)n n+1

n
X
i=1

(i n+1 )2

n
Y

j=1
j6=i

j2
=

n n+1
X
X

i=1 j=i+1

2
(i j )

n n+1
2
X (i j )2
X
i=1...n+1
= Si=1...n+1

exp
2
i2 j2
i=1 j=i+1

13

n+1
Y

k=1
k6=i,j

k2

Q.E.D.

You might also like