Академический Документы
Профессиональный Документы
Культура Документы
1
2 3 INTRODUCTION
1 Denition 3 Introduction
See also: Fourier analysis
The Fourier transform of the function f is traditionally One motivation for the Fourier transform comes from
denoted by adding a circumex: f . There are several
common conventions for dening the Fourier transform
of an integrable function f : R C.[1][2] Here we will
use the following denition:
f() = f (x) e2ix dx,
In the rst frames of the animation, a function f is resolved
f (x) = f() e2ix d, into Fourier series: a linear combination of sines and cosines
(in blue). The component frequencies of these sines and cosines
spread across the frequency spectrum, are represented as peaks
for any real number x. in the frequency domain (actually Dirac delta functions, shown
in the last frames of the animation). The frequency domain rep-
The statement that f can be reconstructed from f is resentation of the function, f, is the collection of these peaks at
known as the Fourier inversion theorem, and was rst in- the frequencies that appear in this resolution of the function.
troduced in Fouriers Analytical Theory of Heat,[3][4] al-
though what would be considered a proof by modern stan-
the study of Fourier series. In the study of Fourier se-
dards was not given until much later.[5][6] The functions ries, complicated but periodic functions are written as the
f and f often are referred to as a Fourier integral pair or
sum of simple waves mathematically represented by sines
Fourier transform pair.[7] and cosines. The Fourier transform is an extension of
For other common conventions and notations, including the Fourier series that results when the period of the rep-
using the angular frequency instead of the frequency , resented function is lengthened and allowed to approach
see Other conventions and Other notations below. The innity.[10]
Fourier transform on Euclidean space is treated sepa- Due to the properties of sine and cosine, it is possible
rately, in which the variable x often represents position to recover the amplitude of each wave in a Fourier se-
and momentum. The conventions chosen in this article ries using an integral. In many cases it is desirable to
are those of harmonic analysis, and are characterized as use Eulers formula, which states that e2i = cos(2) +
the unique conventions such that the Fourier transform is i sin(2), to write Fourier series in terms of the basic
both unitary on L2 and an algebra homomorphism from waves e2i . This has the advantage of simplifying many
L1 to L , without normalizing the Lebesgue measure.[8] of the formulas involved, and provides a formulation for
Many other characterizations of the Fourier transform ex- Fourier series that more closely resembles the denition
ist. For example, one uses the Stonevon Neumann the- followed in this article. Re-writing sines and cosines as
orem: the Fourier transform is the unique unitary inter- complex exponentials makes it necessary for the Fourier
twiner for the symplectic and Euclidean Schrdinger rep- coecients to be complex valued. The usual interpre-
resentations of the Heisenberg group. tation of this complex number is that it gives both the
3
0.8
cn e2i( T )x =
n
f (x) = f(n ) e2in x , 0.6
n= n=
0.4
where the last sum is simply the rst sum rewritten using
the denitions n = n/T, and = n + 1/T n/T = 1/T. 0.2
1
0.5
0.4
0.8
0.4
0.2
0.6 0.3
0.4 0.2
-0.2
0.1
0.2
-0.4
0
0 -6 -4 -2 0 2 4 6
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 1 1.5 2
-2 -1.5 -1 -0.5 0 0.5
Fourier transform with 3 and 5 Hz labeled.
Real and imaginary parts of integrand for Fourier
transform at 3 Hz
|f (x)| dx < .
1
0.8 5.1 Basic properties
0.8
0.6 The Fourier transform has the following basic
0.6 properties:[12]
0.4
0.4
0.2 Linearity
0.2
0
For any complex numbers a and b, if h(x) = af
0
(x) + bg(x), then () = a f () + b ().
-0.2
-0.2
-0.4 Translation / time shifting
-0.4
-0.6
-0.6
For any real number x0 , if h(x) = f (x x0 ),
-0.8 then () = e2ix0 f ().
-0.8
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Modulation / frequency shifting
Real and imaginary parts of integrand for Fourier
transform at 5 Hz For any real number 0 if h(x) = e2ix0 f (x),
then () = f ( 0 ).
5.3 Units and duality 5
Time scaling
F 0 = Id, F 1 = F, F 2 = P, F 4 = Id
For a non-zero real number a, if h(x) = f (ax),
then F 3 = F 1 = P F = F P
( )
1 These equalities of operators require careful denition
h() = f .
|a| a of the space of functions in question, dening equality
of functions (equality at every point? equality almost ev-
The case a = 1 leads to the time-reversal prop- erywhere?) and dening equality of operators that is,
erty, which states: if h(x) = f (x), then () = dening the topology on the function space and operator
f (). space in question. These are not true for all functions, but
are true under various conditions, which are the content
Conjugation of the various forms of the Fourier inversion theorem.
This fourfold periodicity of the Fourier transform is sim-
If h(x) = f (x), then
ilar to a rotation of the plane by 90, particularly as the
two-fold iteration yields a reversal, and in fact this anal-
h() = f().
ogy can be made precise. While the Fourier transform
In particular, if f is real, then one has the reality can simply be interpreted as switching the time domain
condition and the frequency domain, with the inverse Fourier trans-
form switching them back, more geometrically it can be
f() = f(), interpreted as a rotation by 90 in the timefrequency do-
main (considering time as the x-axis and frequency as the
that is, f is a Hermitian function. And if f is y-axis), and the Fourier transform can be generalized to
purely imaginary, then the fractional Fourier transform, which involves rotations
by other angles. This can be further generalized to linear
f() = f(). canonical transformations, which can be visualized as the
action of the special linear group SL2 (R) on the time
Integration frequency plane, with the preserved symplectic form cor-
responding to the uncertainty principle, below. This ap-
Substituting = 0 in the denition, we obtain proach is particularly studied in signal processing, under
timefrequency analysis.
f (0) = f (x) dx.
5.3 Units and duality
That is, the evaluation of the Fourier transform
at the origin ( = 0) equals the integral of f over In mathematics, one often does not think of any units as
all its domain. being attached to the two variables t and . But in physi-
cal applications, must have inverse units to the units of
t. For example, if t is measured in seconds, should be
5.2 Invertibility and periodicity in cycles per second for the formulas here to be valid. If
the scale of t is changed and t is measured in units of 2
Further information: Fourier inversion theorem and
seconds, then either must be in the so-called "angular
Fractional Fourier transform frequency", or one must insert some constant scale factor
into some of the formulas. If t is measured in units of
Under suitable conditions on the function f, it can be re- length, then must be in inverse length, e.g., wavenum-
covered from its Fourier transform f. Indeed, denoting bers. That is to say, there are two copies of the real line:
the Fourier transform operator by F, so F( f ) := f, then forone measured in one set of units, where t ranges, and the
suitable functions, applying the Fourier transform twice other in inverse units to the units of t, and which is the
simply ips the function: F2 ( f )(x) = f (x), which can range of . So these are two distinct copies of the real
be interpreted as reversing time. Since reversing time line, and cannot be identied with each other. Therefore,
is two-periodic, applying this twice yields F4 ( f ) = f, the Fourier transform goes from one space of functions
so the Fourier transform operator is four-periodic, and to a dierent space of functions: functions which have a
similarly the inverse Fourier transform can be obtained dierent domain of denition.
by applying the Fourier transform three times: F3 ( f ) = In general, must always be taken to be a linear form on
f. In particular the Fourier transform is invertible (under the space of ts, which is to say that the second real line
suitable conditions). is the dual space of the rst real line. See the article on
More precisely, dening the parity operator P that in- linear algebra for a more formal explanation and for more
verts time, P[ f ] : t f (t): details. This point of view becomes essential in gener-
6 5 PROPERTIES OF THE FOURIER TRANSFORM
1.0
1
x(t) = 1 ()eit d.
x
2
0.5
0.8
1
x(t) = 2 ()eit d.
x 0.6
2
0.4
Furthermore, there is no way to x which square root of
0.2
negative one will be meant by the symbol i (it makes no
sense to speak of the positive square root since only real x
-6 -4 -2 2 4 6
numbers can be positive, similarly it makes no sense to say
-0.2
rotation counter-clockwise, because until i is chosen,
there is no xed way to draw the complex plane), and
hence one occasionally sees the Fourier transform written The sinc function, which is the Fourier transform of the rect-
with i in the exponent instead of i, and vice versa for the angular function, is bounded and continuous, but not Lebesgue
inversion formula, a convention that is equally valid as the integrable.
one chosen in this article, which is the more usual one.
For example, in probability theory, the characteristic The Fourier transform may be dened in some cases for
function of the probability density function f of a ran- non-integrable functions, but the Fourier transforms of
dom variable X of continuous type is dened without a integrable functions have several strong properties.
negative sign in the exponential, and since the units of x The Fourier transform f of any integrable function f is
are ignored, there is no 2 either: uniformly continuous and[13]
5.6 Poisson summation formula 7
However, f need not be integrable. For example, the 5.6 Poisson summation formula
Fourier transform of the rectangular function, which is in-
tegrable, is the sinc function, which is not Lebesgue inte- Main article: Poisson summation formula
grable, because its improper integrals behave analogously
to the alternating harmonic series, in converging to a sum
without being absolutely convergent. The Poisson summation formula (PSF) is an equation that
relates the Fourier series coecients of the periodic sum-
It is not generally possible to write the inverse transform mation of a function to values of the functions continuous
as a Lebesgue integral. However, when both f and f are Fourier transform. The Poisson summation formula says
integrable, the inverse equality that for suciently regular functions f,
f (x) = f()e2ix d f(n) = f (n).
n n
holds almost everywhere. That is, the Fourier transform It has a variety of useful forms that are derived from the
is injective on L1 (R). (But if f is continuous, then equality basic one by application of the Fourier transforms scal-
holds for every x.) ing and time-shifting properties. The formula has appli-
cations in engineering, physics, and number theory. The
frequency-domain dual of the standard Poisson summa-
5.5 Plancherel theorem and Parsevals the- tion formula is also called the discrete-time Fourier trans-
orem form.
Poisson summation is generally associated with the
Let f (x) and g(x) be integrable, and let f () and () be physics of periodic media, such as heat conduction on a
their Fourier transforms. If f (x) and g(x) are also square- circle. The fundamental solution of the heat equation on
integrable, then we have Parsevals Formula:[15] a circle is called a theta function. It is used in number the-
ory to prove the transformation properties of theta func-
tions, which turn out to be a type of modular form, and it
is connected more generally to the theory of automorphic
f (x)g(x) dx = g () d,
f ()
forms where it appears on one side of the Selberg trace
formula.
where the bar denotes complex conjugation.
The Plancherel theorem, which follows from the above,
states that[16]
5.7 Dierentiation
Suppose f (x) is a dierentiable function, and both f and
its derivative f are integrable. Then the Fourier trans-
2 2
|f (x)| dx = f () d. form of the derivative is given by
h() = f()f() = |f()|2 .
5.8 Convolution theorem
Main article: Convolution theorem
5.10 Eigenfunctions
The Fourier transform translates between convolution One important choice of an orthonormal basis for L2 (R)
and multiplication of functions. If f (x) and g(x) are in- is given by the Hermite functions
tegrable functions with Fourier transforms f () and ()
respectively, then the Fourier transform of the convolu-
tion is given by the product of the Fourier transforms f
4
2
n (x) = ex Hen (2x ),
2
( )n
x2 d x2
complex valued functions f on the real line, generated by o the real axis at all, or it might converge to a complex
the translations (Ty f )(x) = f (x + y) and multiplication analytic function for all values of = + i, or something
by e2ix , (M f )(x) = e2ix f (x). These operators do not in between.[21]
commute, as their (group) commutator is The PaleyWiener theorem says that f is smooth (i.e., n-
times dierentiable for all positive integers n) and com-
( ) pactly supported if and only if f ( + i) is a holomorphic
M1 Ty1 M Ty f (x) = e2iy f (x) function for which there exists a constant a > 0 such that
for any integer n 0,
which is multiplication by the constant (independent of x)
e2iy U(1) (the circle group of unit modulus complex
numbers). As an abstract group, the Heisenberg group is n a| |
the three-dimensional Lie group of triples (x, , z) R2 f () Ce
U(1), with the group law
for some constant C. (In this case, f is supported on [a,
a].) This can be expressed by saying that f is an entire
( function which ) is rapidly decreasing in (for xed ) and
(x1 , 1 , t1 )(x2 , 2 , t2 ) = x1 + x2 , 1 + 2 , t1 t2 e2i(x1 1 +x2 2 +x1 2 ) .
of exponential growth in (uniformly in ).[22]
Denote the Heisenberg group by H 1 . The above proce- (If f is not smooth, but only L2 , the statement still
dure describes not only the group structure, but also a holds provided n = 0.[23] ) The space of such functions
standard unitary representation of H 1 on a Hilbert space, of a complex variable is called the PaleyWiener space.
which we denote by : H 1 B(L2 (R)). Dene the linear This theorem has been generalised to semisimple Lie
automorphism of R2 by groups.[24]
If f is supported on the half-line t 0, then f is said to
( ) ( ) be causal because the impulse response function of a
x
J = physically realisable lter must have this property, as no
x
eect can precede its cause. Paley and Wiener showed
so that J 2 = I. This J can be extended to a unique auto- that then f extends to a holomorphic function on the com-
morphism of H 1 : plex lower half-plane < 0 which tends to zero as goes
to innity.[25] The converse is false and it is not known
how to characterise the Fourier transform of a causal
( 2ix
) function.[26]
j(x, , t) = , x, te .
The integral for the Fourier transform for some constants C, a 0, then[28]
f() = e2it f (t) dt f(i ) = e2 t f (t) dt,
can be studied for complex values of its argument . De- convergent for all 2 < a, is the two-sided Laplace
pending on the properties of f, this might not converge transform of f.
10 7 FOURIER TRANSFORM ON EUCLIDEAN SPACE
The more usual version (one-sided) of the Laplace case, there are many conventions. For an integrable func-
transform is tion f (x), this article takes the denition:
st
F (s) = f (t)e dt. f() = F(f )() = f (x)e2ix dx
0 Rn
If f is also causal, then where x and are n-dimensional vectors, and x is the
dot product of the vectors. The dot product is sometimes
written as x, .
f(i ) = F (2 ).
All of the basic properties listed above hold for the n-
Thus, extending the Fourier transform to the complex do- dimensional Fourier transform, as do Plancherels and
main means it includes the Laplace transform as a special Parsevals theorem. When the function is integrable, the
casethe case of causal functionsbut with the change Fourier transform is still uniformly continuous and the
of variable s = 2i. RiemannLebesgue lemma holds.[14]
Generally speaking, the more concentrated f (x) is, the
f( + ia)e2it d = f( + ib)e2it d
more spread out its Fourier transform f () must be. In
particular, the scaling property of the Fourier transform
by Cauchys integral theorem. Therefore, the Fourier in-
may be seen as saying: if we squeeze a function in x, its
version formula can use integration along dierent lines,
Fourier transform stretches out in . It is not possible
parallel to the real axis.[29]
to arbitrarily concentrate both a function and its Fourier
Theorem: If f (t) = 0 for t < 0, and | f (t) | < Cea| t | for transform.
some constants C, a > 0, then
The trade-o between the compaction of a function and
its Fourier transform can be formalized in the form of
an uncertainty principle by viewing a function and its
f (t) = f( + i )e2it d, Fourier transform as conjugate variables with respect to
the symplectic form on the timefrequency domain: from
for any < a/2. the point of view of the linear canonical transformation,
This theorem implies the Mellin inversion formula for the the Fourier transform is rotation by 90 in the time
Laplace transformation,[28] frequency domain, and preserves the symplectic form.
Suppose f (x) is an integrable and square-integrable func-
b+i tion. Without loss of generality, assume that f (x) is nor-
1
f (t) = F (s)est ds malized:
2i bi
( )
1 ( )
D0 (f )D0 f f (t) = a() cos(2t) + b() sin(2t) d.
16 2 0
The equality is attained only in the case This is called an expansion as a trigonometric integral, or
a Fourier integral expansion. The coecient functions a
x2
and b can be found by using variants of the Fourier cosine
f (x) = C1 e 2 transform and the Fourier sine transform (the normalisa-
tions are, again, not standardised):
f() = C1 e
2 2
2 n
where the logarithms may be in any base that is consistent. f (). Let the set Hk be the closure in L (R ) of linear
The equality is attained for a Gaussian, as in the previous combinations of functions of the form f (| x |)P(x) where
case. P(x) is in Ak. The space L2 (Rn ) is then a direct sum of
the spaces Hk and the Fourier transform maps each space
Hk to itself and is possible to characterize the action of
7.2 Sine and cosine transforms the Fourier transform on each space Hk.[14]
Let f (x) = f 0 (| x |)P(x) (with P(x) in Ak), then
Main article: Sine and cosine transforms
Fouriers original formulation of the transform did not use f() = F (||)P ()
0
complex numbers, but rather sines and cosines. Statisti-
cians and others still use this form. An absolutely inte- where
grable function f for which Fourier inversion holds good
can be expanded in terms of genuine frequencies (avoid-
ing negative frequencies, which are sometimes consid- k n+2k2 n+2k
[34]
ered hard to interpret physically ) by F 0 (r) = 2i r 2 f0 (s)J n+2k2 (2rs)s 2 ds.
2
0
12 8 FOURIER TRANSFORM ON FUNCTION SPACES
Here Jn k / denotes the Bessel function of the rst 8 Fourier transform on function
kind with order n + 2k 2/2. When k = 0 this gives
a useful formula for the Fourier transform of a radial
spaces
function.[38] Note that this is essentially the Hankel trans-
form. Moreover, there is a simple recursion relating the 8.1 On Lp spaces
cases n + 2 and n[39] allowing to compute, e.g., the three-
dimensional Fourier transform of a radial function from 8.1.1 On L1
the one-dimensional one.
The denition of the Fourier transform by the integral
formula
striction problems for the Fourier transform. The Fourier is valid for Lebesgue integrable functions f; that is, f
transform of an integrable function is continuous and the L1 (Rn ).
restriction of this function to any set is dened. But for
The Fourier transform F : L1 (Rn ) L (Rn ) is a bounded
a square-integrable function the Fourier transform could
operator. This follows from the observation that
be a general class of square integrable functions. As such,
the restriction of the Fourier transform of an L2 (Rn ) func-
tion cannot be dened on sets of measure 0. It is still an
active area of study to understand restriction problems in f () |f (x)| dx,
Rn
Lp for 1 < p < 2. Surprisingly, it is possible in some cases
to dene the restriction of a Fourier transform to a set which shows that its operator norm is bounded by 1. In-
S, provided S has non-zero curvature. The case when S deed, it equals 1, which can be seen, for example, from
is the unit sphere in Rn is of particular interest. In this the transform of the rect function. The image of L1 is a
case the TomasStein restriction theorem states that the subset of the space C 0 (Rn ) of continuous functions that
restriction of the Fourier transform to the unit sphere in tend to zero at innity (the RiemannLebesgue lemma),
Rn is a bounded operator on Lp provided 1 p 2n + although it is not the entire space. Indeed, there is no
2/n + 3. simple characterization of the image.
One notable dierence between the Fourier transform in
1 dimension versus higher dimensions concerns the par- 8.1.2 On L2
tial sum operator. Consider an increasing collection of
measurable sets ER indexed by R (0,): such as balls Since compactly supported smooth functions are inte-
of radius R centered at the origin, or cubes of side 2R. grable and dense in L2 (Rn ), the Plancherel theorem al-
For a given integrable function f, consider the function lows us to extend the denition of the Fourier transform
fR dened by: to general functions in L2 (Rn ) by continuity arguments.
The Fourier transform in L2 (Rn ) is no longer given by
an ordinary Lebesgue integral, although it can be com-
puted by an improper integral, here meaning that for an
fR (x) = f()e2ix d, x Rn . L2 function f,
ER
Suppose in addition that f Lp (Rn ). For n = 1 and 1
f() = lim f (x)e2ix dx
< p < , if one takes ER = (R, R), then fR converges R |x|R
to f in Lp as R tends to innity, by the boundedness of
the Hilbert transform. Naively one may hope the same where the limit is taken in the L2 sense. Many of the
holds true for n > 1. In the case that ER is taken to be properties of the Fourier transform in L1 carry over to
a cube with side length R, then convergence still holds. L2 , by a suitable limiting argument.
Another natural candidate is the Euclidean ball ER = { Furthermore, F : L2 (Rn ) L2 (Rn ) is a unitary opera-
: | | < R}. In order for this partial sum operator to con- tor.[40] For an operator to be unitary it is sucient to show
verge, it is necessary that the multiplier for the unit ball that it is bijective and preserves the inner product, so in
be bounded in Lp (Rn ). For n 2 it is a celebrated theo- this case these follow from the Fourier inversion theorem
rem of Charles Feerman that the multiplier for the unit combined with the fact that for any f, g L2 (Rn ) we have
ball is never bounded unless p = 2.[17] In fact, when p
2, this shows that not only may fR fail to converge to f in
Lp , but for some functions f Lp (Rn ), fR is not even an
element of Lp . f (x)Fg(x) dx = Ff (x)g(x) dx.
Rn Rn
13
In particular, the image of L2 (Rn ) is itself under the for all Schwartz functions . So it makes sense to dene
Fourier transform. Fourier transform Tf of Tf by
8.1.3 On other Lp
Tf () = Tf ()
The denition of the Fourier transform can be extended
to functions in Lp (Rn ) for 1 p 2 by decomposing such for all Schwartz functions . Extending this to all tem-
functions into a fat tail part in L2 plus a fat body part in pered distributions T gives the general denition of the
L1 . In each of these spaces, the Fourier transform of a Fourier transform.
function in Lp (Rn ) is in Lq (Rn ), where q = p/p 1 is the Distributions can be dierentiated and the above-
Hlder conjugate of p (by the HausdorYoung inequal- mentioned compatibility of the Fourier transform with
ity). However, except for p = 2, the image is not easily dierentiation and convolution remains true for tem-
characterized. Further extensions become more techni- pered distributions.
cal. The Fourier transform of functions in Lp for the range
2 < p < requires the study of distributions.[13] In fact, it
can be shown that there are functions in Lp with p > 2 so
that the Fourier transform is not dened as a function.[14]
9 Generalizations
The Fourier transform may be generalized to any locally
Tf () = f (x)(x) dx compact abelian group. A locally compact abelian group
Rn
14 9 GENERALIZATIONS
is an abelian group that is at the same time a locally com- 9.4 Compact non-abelian groups
pact Hausdor topological space so that the group op-
eration is continuous. If G is a locally compact abelian The Fourier transform can also be dened for func-
group, it has a translation invariant measure , called tions on a non-abelian group, provided that the group is
Haar measure. For a locally compact abelian group G, compact. Removing the assumption that the underlying
the set of irreducible, i.e. one-dimensional, unitary repre- group is abelian, irreducible unitary representations need
sentations are called its characters. With its natural group not always be one-dimensional. This means the Fourier
structure and the topology of pointwise convergence, the transform on a non-abelian group takes values as Hilbert
set of characters is itself a locally compact abelian space operators.[42] The Fourier transform on compact
group, called the Pontryagin dual of G. For a function groups is a major tool in representation theory[43] and
f in L1 (G), its Fourier transform is dened by[13] non-commutative harmonic analysis.
Let G be a compact Hausdor topological group. Let
denote the collection of all isomorphism classes of nite-
f () = (x)f (x) d any for G.
dimensional irreducible unitary representations, along
G with a denite choice of representation U () on the
Hilbert space H of nite dimension d for each . If
The RiemannLebesgue lemma holds in this case; f () is a nite Borel measure on G, then the FourierStieltjes
is a function vanishing at innity on . transform of is the operator on H dened by
()
9.3 Gelfand transform
, H = U g , d(g)
G
f (g) = f (g 1 ). 7
Taking the completion with respect to the largest possi- denes an isomorphism between the Banach space M(G)
bly C*-norm gives its enveloping C*-algebra, called the of nite Borel measures (see rca space) and a closed sub-
group C*-algebra C*(G) of G. (Any C*-norm on L1 (G) space of the Banach space C() consisting of all se-
is bounded by the L1 norm, therefore their supremum ex- quences E = (E) indexed by of (bounded) linear oper-
ists.) ators E : H H for which the norm
Given any abelian C*-algebra A, the Gelfand transform
gives an isomorphism between A and C 0 (A^), where
E = sup E
A^ is the multiplicative linear functionals, i.e. one-
dimensional representations, on A with the weak-* topol-
ogy. The map is simply given by is nite. The "convolution theorem" asserts that, further-
more, this isomorphism of Banach spaces is in fact an
isometric isomorphism of C* algebras into a subspace of
( ) C(). Multiplication on M(G) is given by convolution
a 7 7 (a) of measures and the involution * dened by
( )
f (g) = d tr f()Ug() Some problems, such as certain dierential equations, become
easier to solve when the Fourier transform is applied. In that
case the solution to the original problem is recovered using the
where the summation is understood as convergent in the inverse Fourier transform.
L2 sense.
The generalization of the Fourier transform to the non-
the equations of the mathematical physics of the nine-
commutative situation has also in part contributed to the
teenth century can be treated this way. Fourier studied
development of noncommutative geometry. In this con-
the heat equation, which in one dimension and in dimen-
text, a categorical generalization of the Fourier trans-
sionless units is
form to noncommutative groups is TannakaKrein dual-
ity, which replaces the group of characters with the cate-
gory of representations. However, this loses the connec- 2 y(x, t) y(x, t)
tion with harmonic functions. = .
2x t
The example we will give, a slightly more dicult one, is
10 Alternatives the wave equation in one dimension,
( ) ( )
cos 2(x t) or sin 2(x t)
11.1 Analysis of dierential equations
satises the wave equation. These are called the elemen-
Perhaps the most important use of the Fourier transfor- tary solutions.
mation is to solve partial dierential equations. Many of Second, note that therefore any integral
16 11 APPLICATIONS
2 y(, f ) = f 2 y(, f ).
2 y(x, 0) cos(2x) dx = a+ + a
This is equivalent to requiring (, f ) = 0 unless = f.
and Right away, this explains why the choice of elementary
solutions we made earlier worked so well: obviously f =
( f ) will be solutions. Applying Fourier inversion to
these delta functions, we obtain the elementary solutions
2 y(x, 0) sin(2x) dx = b+ + b . we picked earlier. But from the higher point of view, one
does not pick elementary solutions, but rather considers
Similarly, taking the derivative of y with respect to t and the space of all distributions which are supported on the
then applying the Fourier sine and cosine transformations (degenerate) conic 2 f 2 = 0.
yields We may as well consider the distributions supported on
the conic that are given by distributions of one variable
on the line = f plus distributions on the line = f as
y(u, 0)
2 sin(2x) dx = (2) (a+ + a ) follows: if is any test function,
t
and
y(, f ) d df = s+ (, ) d+ s (, ) d,
y(u, 0) where s, and s, are distributions of one variable.
2 cos(2x) dx = (2) (b+ b ) .
t
Then Fourier inversion gives, for the boundary condi-
These are four linear equations for the four unknowns a tions, something very similar to what we had more con-
and b, in terms of the Fourier sine and cosine transforms cretely above (put (, f ) = e2i(x+tf ) , which is clearly of
of the boundary conditions, which are easily solved by polynomial growth):
elementary algebra, provided that these transforms can
be found.
{ }
In summary, we chose a set of elementary solutions, y(x, 0) = s+ () + s () e2ix+0 d
parametrised by , of which the general solution would
be a (continuous) linear combination in the form of an and
integral over the parameter . But this integral was in
the form of a Fourier integral. The next step was to ex-
press the boundary conditions in terms of these integrals, y(x, 0) { }
and set them equal to the given functions f and g. But = s+ () s () 2ie2ix+0 d.
t
these expressions also took the form of a Fourier integral
because of the properties of the Fourier transform of a Now, as before, applying the one-variable Fourier trans-
derivative. The last step was to exploit Fourier inversion formation in the variable x to these functions of x yields
11.3 Quantum mechanics 17
two equations in the two unknown distributions s (which half the dimension, for example, the q-axis alone, but
can be taken to be ordinary functions if the boundary con- instead of considering only points, takes the set of all
ditions are L1 or L2 ). complex-valued wave functions on this axis. Neverthe-
From a calculational point of view, the drawback of less, choosing the p-axis is an equally valid polarisation,
course is that one must rst calculate the Fourier trans- yielding a dierent representation of the set of possible
forms of the boundary conditions, then assemble the so- physical states of the particle which is related to the rst
lution from these, and then calculate an inverse Fourier representation by the Fourier transformation
transform. Closed form formulas are rare, except when
there is some geometric symmetry that can be exploited,
2i pq
and the numerical calculations are dicult because of the (p) = (q)e h dq.
oscillatory nature of the integrals, which makes conver-
gence slow and hard to estimate. For practical calcula- Physically realisable states are L2 , and so by the
tions, other methods are often used. Plancherel theorem, their Fourier transforms are also L2 .
(Note that since q is in units of distance and p is in units
The twentieth century has seen the extension of these
of momentum, the presence of Plancks constant in the
methods to all linear partial dierential equations with
exponent makes the exponent dimensionless, as it should
polynomial coecients, and by extending the notion of
be.)
Fourier transformation to include Fourier integral opera-
tors, some non-linear equations as well. Therefore, the Fourier transform can be used to pass from
one way of representing the state of the particle, by a wave
function of position, to another way of representing the
11.2 Fourier transform spectroscopy state of the particle: by a wave function of momentum.
Innitely many dierent polarisations are possible, and all
Main article: Fourier transform spectroscopy are equally valid. Being able to transform states from one
representation to another is sometimes convenient.
The Fourier transform is also used in nuclear magnetic The other use of the Fourier transform in both quantum
resonance (NMR) and in other kinds of spectroscopy, e.g. mechanics and quantum eld theory is to solve the ap-
infrared (FTIR). In NMR an exponentially shaped free in- plicable wave equation. In non-relativistic quantum me-
duction decay (FID) signal is acquired in the time domain chanics, Schrdingers equation for a time-varying wave
and Fourier-transformed to a Lorentzian line-shape in the function in one-dimension, not subject to external forces,
frequency domain. The Fourier transform is also used in is
magnetic resonance imaging (MRI) and mass spectrom-
etry.
2 h
(x, t) = i (x, t).
x2 2 t
11.3 Quantum mechanics This is the same as the heat equation except for the pres-
ence of the imaginary unit i. Fourier methods can be used
The Fourier transform is useful in quantum mechanics in to solve this equation.
two dierent ways. To begin with, the basic conceptual
structure of Quantum Mechanics postulates the existence In the presence of a potential, given by the potential en-
of pairs of complementary variables, connected by the ergy function V(x), the equation becomes
Heisenberg uncertainty principle. For example, in one
dimension, the spatial variable q of, say, a particle, can
2 h
only be measured by the quantum mechanical position (x, t) + V (x)(x, t) = i (x, t).
x 2 2 t
operator at the cost of losing information about the mo-
mentum p of the particle. Therefore, the physical state of The elementary solutions, as we referred to them above,
the particle can either be described by a function, called are the so-called stationary states of the particle, and
the wave function, of q or by a function of p but not Fouriers algorithm, as described above, can still be used
by a function of both variables. The variable p is called to solve the boundary value problem of the future evo-
the conjugate variable to q. In Classical Mechanics, the lution of given its values for t = 0. Neither of these
physical state of a particle (existing in one dimension, for approaches is of much practical use in quantum mechan-
simplicity of exposition) would be given by assigning def- ics. Boundary value problems and the time-evolution of
inite values to both p and q simultaneously. Thus, the set the wave function is not of much practical interest: it is
of all possible physical states is the two-dimensional real the stationary states that are most important.
vector space with a p-axis and a q-axis called the phase In relativistic quantum mechanics, Schrdingers equa-
space. tion becomes a wave equation as was usual in classical
In contrast, quantum mechanics chooses a polarisation of physics, except that complex-valued waves are consid-
this space in the sense that it picks a subspace of one- ered. A simple example, in the absence of interactions
18 12 OTHER NOTATIONS
It possesses a Fourier transform, in terms of the two real functions A() and () where:
19
13 Other conventions
A() = f() ,
The Fourier transform can also be written in terms of
is the amplitude and angular frequency:
( )
= 2,
() = arg f() ,
whose units are radians per second.
is the phase (see arg function).
The substitution = /2 into the formulas above pro-
Then the inverse transform can be written: duces this convention:
( )
f (x) = A() ei 2x+() d, f() = f (x)eix dx.
Rn
which is a recombination of all the frequency compo- Under this convention, the inverse transform becomes:
nents of f (x). Each component is a complex sinusoid of
the form e2ix whose amplitude is A() and whose initial
1
phase angle (at x = 0) is (). f (x) = f()eix d.
(2)n Rn
The Fourier transform may be thought of as a mapping
on function spaces. This mapping is here denoted F and Unlike the convention followed in this article, when the
F( f ) is used to denote the Fourier transform of the func- Fourier transform is dened 2
this way, it is no longer a
n
tion f. This mapping is linear, which means that F can unitary transformation on L (R ). There is also less sym-
also be seen as a linear transformation on the function metry between the formulas for the Fourier transform and
space and implies that the standard notation in linear al- its inverse.
gebra of applying a linear transformation to a vector (here Another convention is to split the factor of (2)n evenly
the function f ) can be used to write F f instead of F( f between the Fourier transform and its inverse, which
). Since the result of applying the Fourier transform is leads to denitions:
again a function, we can be interested in the value of this
function evaluated at the value for its variable, and this
is denoted either as F f () or as ( F f )(). Notice that in f() = 1 n f (x)eix dx,
the former case, it is implicitly understood that F is ap- (2)
R
2 n
( ) ( )
( )
F f (x + x0 ) = F f (x) e2ix0 E eitX = eitx dX (x).
is used to express the shift property of the Fourier trans- As in the case of the non-unitary angular frequency
form. convention above, there is no factor of 2 appearing in
Notice, that the last example is only correct under the as- either of the integral, or in the exponential. Unlike any
sumption that the transformed function is a function of x, of the conventions appearing above, this convention takes
not of x0 . the opposite sign in the exponential.
20 15 TABLES OF IMPORTANT FOURIER TRANSFORMS
14.2 Numerical integration of a series of The following tables record some closed-form Fourier
ordered pairs transforms. For functions f (x), g(x) and h(x) denote their
Fourier transforms by f, , and respectively. Only the
If the input function is a series of ordered pairs (for ex- three most common conventions are included. It may be
ample, a time series from measuring an output variable useful to notice that entry 105 gives a relationship be-
repeatedly over a time interval) then the output function tween the Fourier transform of a function and the orig-
must also be a series of ordered pairs (for example, a inal function, which can be seen as relating the Fourier
complex number vs. frequency over a specied domain transform and its inverse.
of frequencies), unless certain assumptions and approxi-
mations are made allowing the output function to be ap-
proximated by a closed-form expression. In the general
case where the available input series of ordered pairs are 15.1 Functional relationships
assumed be samples representing a continuous function
over an interval (amplitude vs. time, for example), the The Fourier transforms in this table may be found in
series of ordered pairs representing the desired output Erdlyi (1954) or Kammler (2000, appendix).
21
The Fourier transforms in this table may be found in Space-time Fourier transform
Campbell & Foster (1948), Erdlyi (1954), or Kammler Spectral density
(2000, appendix).
Spectral density estimation
[23] Kirillov & Gvishiani 1982. Bochner, S.; Chandrasekharan, K. (1949), Fourier
Transforms, Princeton University Press.
[24] Clozel & Delorme 1985, pp. 331333.
[25] de Groot & Mazur 1984, p. 146. Bracewell, R. N. (2000), The Fourier Transform and
Its Applications (3rd ed.), Boston: McGraw-Hill,
[26] Champeney 1987, p. 80.
ISBN 0-07-116043-4.
[27] Chateld 2004, p. 297.
Campbell, George; Foster, Ronald (1948), Fourier
[28] Kolmogorov & Fomin 1999.
Integrals for Practical Applications, New York: D.
[29] Wiener 1949. Van Nostrand Company, Inc..
[30] Champeney 1987, p. 63.
Champeney, D.C. (1987), A Handbook of Fourier
[31] Widder & Wiener 1938, p. 537. Theorems, Cambridge University Press.
[32] Pinsky 2002, p. 131.
Chateld, Chris (2004), The Analysis of Time Se-
[33] Stein & Shakarchi 2003, p. 158. ries: An Introduction, Texts in Statistical Science
[34] Chateld 2004, p. 113. (6th ed.), London: Chapman & Hall/CRC.
[42] Hewitt & Ross 1970, Chapter 8. de Groot, Sybren R.; Mazur, Peter (1984), Non-
[43] Knapp 2001.
Equilibrium Thermodynamics (2nd ed.), New York:
Dover.
[44] Gradshteyn et al. 2015.
Feller, William (1971), An Introduction to Probabil- Jordan, Camille (1883), Cours d'Analyse de l'cole
ity Theory and Its Applications, Vol. II (2nd ed.), Polytechnique, Vol. II, Calcul Intgral: Intgrales
New York: Wiley, MR 0270403. dnies et indnies (2nd ed.), Paris.
Folland, Gerald (1989), Harmonic analysis in phase Kaiser, Gerald (1994), A Friendly Guide to Wavelets,
space, Princeton University Press. Birkhuser, ISBN 0-8176-3711-7.
Fourier, J.B. Joseph (1878) [1822], The Analytical Katznelson, Yitzhak (1976), An Introduction to Har-
Theory of Heat, translated by Alexander Freeman, monic Analysis, Dover, ISBN 0-486-63331-4.
The University Press (translated from French).
Kirillov, Alexandre; Gvishiani, Alexei D. (1982)
Gradshteyn, Izrail Solomonovich; Ryzhik, Iosif [1979], Theorems and Problems in Functional Anal-
Moiseevich; Geronimus, Yuri Veniaminovich; ysis, Springer (translated from Russian).
Tseytlin, Michail Yulyevich; Jerey, Alan (2015),
Zwillinger, Daniel; Moll, Victor Hugo, eds., Table Knapp, Anthony W. (2001), Representation Theory
of Integrals, Series, and Products, translated by of Semisimple Groups: An Overview Based on Exam-
Scripta Technica, Inc. (8th ed.), Academic Press, ples, Princeton University Press, ISBN 978-0-691-
ISBN 978-0-12-384933-5. 09089-4.
James, J.F. (2011), A Students Guide to Fourier Polyanin, A. D.; Manzhirov, A. V. (1998), Hand-
Transforms (3rd ed.), Cambridge University Press, book of Integral Equations, Boca Raton: CRC Press,
ISBN 978-0-521-17683-5. ISBN 0-8493-2876-4.
24 20 EXTERNAL LINKS
Press, William H.; Flannery, Brian P.; Teukolsky, Wilson, R. G. (1995), Fourier Series and Optical
Saul A.; Vetterling, William T. (1992), Numerical Transform Techniques in Contemporary Optics, New
Recipes in C: The Art of Scientic Computing, Second York: Wiley, ISBN 0-471-30357-7.
Edition (2nd ed.), Cambridge University Press.
Yosida, K. (1968), Functional Analysis, Springer,
Rahman, Matiur (2011), Applications of Fourier ISBN 3-540-58654-7.
Transforms to Generalized Functions, WIT Press,
ISBN 978-1-84564-564-9.
20 External links
Rudin, Walter (1987), Real and Complex Analysis
(3rd ed.), Singapore: McGraw Hill, ISBN 0-07- Encyclopedia of Mathematics
100276-6.
Weisstein, Eric W. Fourier Transform.
Simonen, P.; Olkkonen, H. (1985), Fast method for MathWorld.
computing the Fourier integral transform via Simp-
sons numerical integration, Journal of biomedi-
cal engineering, 7 (4): 337340, doi:10.1016/0141-
5425(85)90067-6.
21.2 Images
File:Commutative_diagram_illustrating_problem_solving_via_the_Fourier_transform.svg Source: https://upload.wikimedia.org/
wikipedia/commons/f/fd/Commutative_diagram_illustrating_problem_solving_via_the_Fourier_transform.svg License: CC BY-SA 3.0
Contributors: Own work Original artist: Quietbritishjim
File:Fourier_transform_of_oscillating_function.svg Source: https://upload.wikimedia.org/wikipedia/commons/4/40/Fourier_
transform_of_oscillating_function.svg License: CC BY-SA 3.0 Contributors: Own work Original artist: Thenub314
File:Fourier_transform_time_and_frequency_domains_(small).gif Source: https://upload.wikimedia.org/wikipedia/commons/7/72/
Fourier_transform_time_and_frequency_domains_%28small%29.gif License: Public domain Contributors: Own work Original artist:
Lucas V. Barbosa
File:Fourier_unit_pulse.svg Source: https://upload.wikimedia.org/wikipedia/commons/5/51/Fourier_unit_pulse.svg License: CC BY-
SA 4.0 Contributors: Own work Original artist: Sawomir_Biay, IkamusumeFan
File:Function_ocsillating_at_3_hertz.svg Source: https://upload.wikimedia.org/wikipedia/commons/c/c3/Function_ocsillating_at_3_
hertz.svg License: CC BY-SA 3.0 Contributors: Own work Original artist: Thenub314
File:Offfreq.svg Source: https://upload.wikimedia.org/wikipedia/commons/e/ef/Offfreq.svg License: CC BY-SA 3.0 Contributors: Own
work Original artist: Thenub314
File:Onfreq.svg Source: https://upload.wikimedia.org/wikipedia/commons/f/f6/Onfreq.svg License: CC BY-SA 3.0 Contributors: http:
//en.wikipedia.org/wiki/Image:Onfreq.png Original artist: Original: Nicholas Longo, SVG conversion: DX-MON (Richard Mant)
File:Phase_shift.svg Source: https://upload.wikimedia.org/wikipedia/commons/5/55/Phase_shift.svg License: CC BY-SA 3.0 Contribu-
tors: Own work Original artist: Peppergrower
File:Rectangular_function.svg Source: https://upload.wikimedia.org/wikipedia/commons/1/11/Rectangular_function.svg License: CC
BY-SA 3.0 Contributors: ? Original artist: ?
File:Sinc_function_(normalized).svg Source: https://upload.wikimedia.org/wikipedia/commons/d/d4/Sinc_function_
%28normalized%29.svg License: CC BY-SA 3.0 Contributors: Own work Original artist: Omegatron
File:Sine_voltage.svg Source: https://upload.wikimedia.org/wikipedia/commons/8/8a/Sine_voltage.svg License: Public domain Contrib-
utors: SVG von de:Bild:Sinus-spannung.gif Original artist: Matthias Krger
26 21 TEXT AND IMAGE SOURCES, CONTRIBUTORS, AND LICENSES