a discussion on “wiener – khinchin theorem and its applications”
DESCRIPTION
A discussion by Harsh Purwar, Student, Indian Institute of Science Education and Research, Kolkata.TRANSCRIPT
Indian Institute of Science Education and Research, Kolkata 14th December 2010
1 | Term Paper on ‘Wiener – Khinchin Theorem & its applications’ by Harsh Purwar (07MS – 76)
A Discussion on “Wiener – Khinchin Theorem and its applications”
Harsh Purwar (07MS – 76) Student, Non-Equilibrium Statistical Mechanics (ID – 419)
Indian Institute of Science Education and Research, Kolkata
Power Spectral Density
Power spectral density or power spectrum in short, of a stationary random process is an important quantitative characterizer of the random process. Consider a general stationary random process, ( ). In general, the plot of ( ) versus may be expected to be a highly irregular curve – the random process is, after all, a ‘noise’ in a sense. We may ask: perhaps a Fourier transform of the function ( ) would give us some insight into its time variation, by decomposing into its individual harmonics? But ( ), regarded as a
function of , need not be integrable, in general. To be precise ∫ | ( )|
may not be finite. Therefore
the Fourier transform may not exist. However, a much regular function of can be associated with the process ( ): namely its autocorrelation function. Basically power spectrum of a stationary process is a measure of how much of the intensity of the fluctuating signal ( ) lies in an infinitesimal frequency window centered at the frequency . Suppose we monitor the process (or signal) ( ) over a very long interval of time, say from up to . The power spectral density ( ) is then defined as,
( )
|∫
( )|
( )
Note: It is evident from the definition above that ( ) is real and positive.
Wiener – Khinchin Theorem
Observe; above definition do not have any average over all realizations of the random process. An averaging of this sort is not necessary because the process is supposed to be ergodic: given a sufficient amount of time (ensured by passing to the limit ), ( ) will take on all values in its sample space. Consequently, we may expect ( ) to be expressible in terms of a statistical average. Indeed, this is the
content of the Wiener – Khinchin theorem. Statement: The power spectral density ( ) of a stationary random process ( ) is equal to the Fourier
transform of its autocorrelation function. That is,
( )
∫
⟨ ( ) ( )⟩ ( )
Proof: We have,
|∫
( )|
∫
∫
( ) ( ) ( )
Indian Institute of Science Education and Research, Kolkata 14th December 2010
2 | Term Paper on ‘Wiener – Khinchin Theorem & its applications’ by Harsh Purwar (07MS – 76)
∫
∫
( ) ( ) ( )
As left hand side is a real quantity, the imaginary part, ( ) of ( ) must vanish. This actually is true because ( ) is an odd function of and vanishes when integrated. Now introducing function we have,
|∫
( )|
∫
∫
( ) ( ) ( ) ( ( ) ( ))
∫
∫
( ) ( ) ( ) ( ) ∫
∫
( ) ( ) ( ) ( )
∫
{
∫
( ) ( ) ( ) ( )⏟
∫
( ) ( ) ( ) ( )⏟
}
∫
{
∫
( ) ( ) ( ) ( )⏟
∫
( ) ( ) ( ) ( )⏟
}
The second integral above has as can be seen from the limits of integration but in this regime ( ) vanishes. Similarly the fourth term also vanishes and we are left with,
|∫
( )|
∫
∫
( ) ( ) ( ) ∫
∫
( ) ( ) ( )
function has been replaced by 1. Now interchanging the dummy variable and in the second term we get,
|∫
( )|
∫
∫
( ) ( ) ( ) ∫
∫
( ) ( ) ( )
∫
∫
( ) ( ) ( )
Changing variable from to . Implies and we get,
|∫
( )|
∫
∫
( ) ( )
∫
∫
( ) ( )
Interchanging the order of integration we get,
|∫
( )|
∫
∫
( ) ( )
Now again change variable to , implying . We get,
|∫
( )|
∫
∫
( ) ( )
∫
∫
( ) ( )
Substituting this back in the definition of ( ) we have,
( )
∫
∫
( ) ( )
Indian Institute of Science Education and Research, Kolkata 14th December 2010
3 | Term Paper on ‘Wiener – Khinchin Theorem & its applications’ by Harsh Purwar (07MS – 76)
∫
∫
( ) ( )
Due to the ergodicity of the process ( ) we can write,
∫
( ) ( ) ⟨ ( ) ( )⟩
Now using the fact that the autocorrelation function of a stationary random process is a function only of the difference of the two time arguments involved, and not of the two arguments separately, we have,
⟨ ( ) ( )⟩ ⟨ ( ) ( )⟩ ⟨ ( ) ( )⟩ Hence,
( )
∫
⟨ ( ) ( )⟩ ( )
Considering ( ) being a classical variable i.e. to say, ⟨ ( ) ( )⟩ ⟨ ( ) ( )⟩ ⟨ ( ) ( )⟩
And as , we have
( )
∫
⟨ ( ) ( )⟩
∫
⟨ ( ) ( )⟩
Applications of WKT:
Position autocorrelation of the Brownian oscillator: Starting from the result obtained by my friend Anish Bhardwaj in his mid-semester presentation on Brownian oscillator, its velocity autocorrelation function,
⟨ ( ) ( )⟩
⁄ ( ( )
( )) ( )
where is given by,
(
)
⁄
I would try to find the autocorrelation function of the position of the oscillator. The Fourier transform of the autocorrelation function in W-K Theorem can be inverted to get,
⟨ ( ) ( )⟩ ∫
( ) ( ) ( )
Differentiating both sides with respect to and in succession we get,
⟨ ( ) ( )⟩ ∫
( ) ( )
Now setting and we get,
⟨ ( ) ( )⟩ ∫
( ) ( )
But from the inverted W-K theorem equation ( ) for we also have,
⟨ ( ) ( )⟩ ∫
( ) ( )
Comparing ( ) and ( ) we have, ( ) ( ) ( )
I intend to apply this equation to the Brownian oscillator taking ( ) ( ) and hence ( ) ( ). Using result ( ) for the velocity auto correlation function of the oscillator in the formula ( ) it can be shown that its power spectral density (PSD) is given by,
Indian Institute of Science Education and Research, Kolkata 14th December 2010
4 | Term Paper on ‘Wiener – Khinchin Theorem & its applications’ by Harsh Purwar (07MS – 76)
( )
{
( )
}
So the PSD of the position using ( ) is given by,
( )
{
( )
}
Using the inverted form of the Wiener Khinchin theorem, equation ( ) we have,
⟨ ( ) ( )⟩ ∫
( )
Using the results of contour integration the above integration was evaluated to give,
⟨ ( ) ( )⟩
{
⁄ (
)
⁄ (
)
Observe that it is indeed an even function of , as required of the autocorrelation function of a (one component) stationary random process.
Works Cited
1. Balakrishnan, V. Elements of Nonequilibrium Statistical Mechanics. Madras : Ane Books Pvt. Ltd., 2008. pp. 223-228, 237-238, 281-282. 9789380156255.