Physics 326: Quantum Mechanics I Prof. Michael S.

Vogeley The Fourier Transform and Free Particle Wave Functions

1
1.1

The Fourier Transform

A function f (x) that is periodic with period 2L, f (x) = f (x + 2L) can be expanded in a Fourier Series over the interval (L, L),
nx nx Bn sin + An cos f ( x) = L L n=0 n=0

Noting that the coecients may be complex (we never said f (x) was real), and recalling that ei = cos + i sin , the series may be written as

f ( x) =
n=

an einx/L

The an are the Fourier coecients. Each an species the contribution to f (x) of a wave exp(inx/L) with wavenumber k = n/L. The Fourier modes are orthonormal, 1 2L
L L

dx einx/L

imx/L

1 2L

L L

dx einx/L eimx/L = mn

where mn is the Kronecker delta function. To obtain the values of the Fourier coecients an , integrate both sides of the Fourier series expansion above 1 2L
L L

dx einx/L f (x) =

1 2L

L L

dx einx/L
m=

am eimx/L

and use the orthonormality of the plane waves (note that only terms with m = n survive on the right hand side) to obtain the Fourier coecients an = 1 2L
L L

dx f (x)einx/L

1.2

Fourier integral

To proceed to the Fourier transform integral, rst note that we can rewrite the Fourier series above as f (x) =
n=

an einx/L n

where n = 1 is the spacing between successive integers. If we dene k = and A(k ) = then the Fourier series may be written as f (x) =
k

n L 2Lan

A(k ) inx/L e k 2

Now take the limit L , so that k becomes innitesimal. Now the discrete sum becomes an integral, 1 f (x) = A(k )eikx dk 2 Following the discussion above that yielded the coecients an , the coecients A(k ) of the continuous transform are obtained by 1 A(k ) = 2

f (x)eikx dk

These two equations dene the Fourier transform relations. The latter is usually referred to as a forward transform, decomposing the spatial function f (x) into Fourier modes, represented by their coecients A(k ), while the former is the inverse transform, which reconstructs the spatial function. (Note that here Ive put factors of 1/ 2 in front of both integrals; some conventions leave out the factor in front of the forward transform and put 1/(2 ) in front of the inverse transform, or vice versa.) The functions f (x) and A(k ) are a Fourier transform pair. Complete knowledge of one of the pair yields the other through an analytic transform, thus they have identical information content. They are just dierent representations of the same function. Note carefully that both f (x) and A(k ) are, in general, complex. In the case where f (x) is a real function, then A(k ) must be Hermitian, A (k ) = A(k ). The common pairs of transform spaces includes position-wavenumber and time-frequency.

1.3

Dirac Delta Function

If we combine the two equations above for f (x) and A(k ), plugging the second into the rst, we obtain 1 dy f (y )eiky dk eikx f ( x) = 2 Now interchange the order of integration, 1 dk eikx eiky ] f (x) = dy f (y )[ 2 The term in brackets is the Dirac Delta function, 1 (x y ) = dk eik(xy) 2 This function obviously has the property that

f (y ) (x y ) = f (x)

1.4

Fourier transform pairs

If f (x) is very narrow, then its Fourier transform A(k ) is a very broad function and vice versa. The Dirac delta function provides the most extreme example of this property. If the Fourier transform is a constant, say A(k ) = 1/ 2 , then the spatial function is exactly the function f (x) = (x). In this case the spatial function is precisely localized and its Fourier transform is completely delocalized. If this sounds like the uncertainty relation to you, youre on the right track. To lead into discussion of another Fourier transform pair, let us begin with another form of the Dirac delta function. Any properly normalized (meaning that the integral over all space is unity) peaked function approaches a delta function in the limit of innitesimal width, so a Gaussian will work, 2 x (x) = lim e x 0 The Fourier transform of this delta function is obviously a constant function, which is the same as a Gaussian of innite width. We can see this by considering the Fourier transform of a 1-d Gaussian 2 x f (x) = e x which can be shown to be another Gaussian 1 2 2 A(k ) = ek /(4 ) 2 Now we clearly see the relationship between the dispersion in the spatial function and its Fourier transform. Narrowing the extent of the function in one domain broadens its extent in the other domain. For a Gaussian, the dispersions are in the relation x k = 1. 3

The Schr odinger Equation and Free Particle Wave Functions

2.1

The wave-particle duality problem can be somewhat reconciled by thinking about particles as localized wave packets. Waves of dierent frequencies are superposed so that they interfere completely (or nearly so) outside of a small spatial region. Clearly, both the amplitudes (magnitude of waves of dierent frequencies) and phases (relative shifting of the waves) are required to achieve such a superposition. A plane wave, which varies only in the x but not in y or z , has the form eikxit where k is the wavenumber k = 2/ and is the angular frequency of the wave. In general, = (k ). A superposition of such waves can be used to represent the wave packet dk g (k )eikxi(k)t f (x, t) =

If the wave packet is quite localized in k space, with g (k ) narrowly centered at k = k0 , such as 2 g (k ) e(kk0 ) with corresponding = (k0 ), then we can associate this wave packet with a particle that has energy E=h and momentum p=h k This suggests that, in general, we can rewrite any wave packet as a superposition of plane waves that have a distribution of momenta p, (x, t) = 1 2h
h dp (p)ei(pxEt)/

2.2

Solutions to the Schr odinger equation

The wave packet (x, t) above is a general solution to the dierential equation ih (x, t) h 2 2 (x, t) = t 2m x2 4

This is the time-dependent Schr odinger equation for a free particle, i.e., where V = 0, written in the position representation. This shows that plane waves exp(ikx it) are eigenfunctions of the free particle Hamiltonian. The probability of nding a particle at position x is proportional to the squared modulus | (x, t)|2 . Likewise, the probability of nding this same particle with momentum p is proportional to |(p)|2 . These two representations are related by the Fourier transform relation above. The Fourier transform relation between the position and momentum representations immediately suggests the Heisenberg uncertainty relation. We showed above that the dispersions of a spatial Gaussian and its Fourier transform are in the relation x k = 1. The product of dispersions is minimized by a Gaussian, thus in general xk 1. Using p=h k , this suggests that the uncertainty relation must be of order xp h

References

More about Fourier transforms can be found in the classic text The Fourier Transform and Its Application, R.N. Bracewell (McGraw Hill). The discussion above closely follows the development in Quantum Physics, S. Gasiorowicz (Wiley).