Chapter 7
Fourier Series and Transforms
Video Lecture 20, Part 1: One Note, Two Instruments (0:00)
7.1Even and Odd Functions
Video Lecture 20, Part 2: Even and Odd Functions (4:39)
A function is said to be even if
(7.1)A function is said to be odd if
(7.2)
Properties of Integrals over Symmetric Limits:

In general, any function can be expressed as the sum of an even and an odd function:
The first term is the even part of , and the second term is the odd part.
7.2Periodic Functions and Fourier Series
Video Lecture 20, Part 3: Periodic Functions and the Waves That Fit (9:13)
A function is said to be periodic with period if
The smallest positive satisfying this relation is called the fundamental period.

Common periodic functions include
all sharing the same fundamental period .
For a periodic function of period , the average value (or mean value) of the function over one period is defined as
7.2.1Fundamental Periodic Basis
For the exponential function to have period ,
Since for integers , we have

Thus, the functions
form a complete orthogonal set over one period.

For the product turns a whole number of times in one period, so its real and imaginary parts have as much area above the axis as below; for it is , with area .

7.2.2Complex Fourier Series
Video Lecture 20, Part 4: The Complex Fourier Series (13:57)
A periodic function of period can be represented as
where the coefficients are
Each coefficient gives the contribution of the wave to ; the list of all of them is the spectrum of .
Starting from , multiply by and integrate:
by orthogonality. Hence,
For a real-valued function , the complex coefficients satisfy
7.2.3Real Form of the Fourier Series
Video Lecture 20, Part 5: The Real Form and Its Coefficients (19:17)
Starting from the complex Fourier series.
We begin with the complex form of the Fourier series for a periodic function of period :
We separate the term and rearrange positive and negative indices:
Using , we have
Expressing in terms of sine and cosine.
Let
where and are real numbers.
Then
Hence the series becomes
Thus the final real form of the Fourier series is
: average (DC component) of ,
: amplitude of the cosine component,
: amplitude of the sine component,
frequency of the -th harmonic:
For of period :
Multiply both sides of the Fourier expansion by and integrate over one period:
Using the orthogonality relations:
all terms vanish except the one where :
Hence,
Similarly, multiplying the Fourier series by and integrating gives
Finally, integrating both sides of the Fourier expansion over one period. Since the integrals of all sine and cosine terms vanish over a full period, only the constant term survives:
Hence,
consistent with the general formula for at . Note that it is , not , that equals the average value (mean) of over one period; writing the constant term as is precisely what lets the single formula cover the case .
7.2.4Dirichlet Conditions and Convergence of Fourier Series
Video Lecture 20, Part 6: When It Works: Dirichlet (25:11)
A periodic function of period can be represented by a Fourier series (without modification) on the interval if it satisfies the following conditions:
is single-valued and piecewise continuous within one period.
has a finite number of finite discontinuities.
has a finite number of maxima and minima within one period.
The integral of its absolute value over one period is finite:
(7.33)
![A function that satisfies the Dirichlet conditions on one period [-,]: a finite number of finite jumps and a finite number of maxima and minima.](assets/img/ch07_fig16_dirichlet.png)
If satisfies the Dirichlet conditions and has period , then its Fourier series
converges as follows:
At every point where is continuous, the Fourier series converges to .
At each point of discontinuity , the Fourier series converges to the midpoint of the jump:
(7.35)
7.2.5Meaning and Interpretation of the Fourier Series
Video Lecture 20, Part 7: What the Harmonics Mean (27:43)
The Fourier series allows us to express any sufficiently smooth periodic function (with period ) as a sum of simple oscillatory components — sine and cosine waves. Mathematically,
Each term in the sum represents a harmonic component (or mode) of the signal.


The -th term in the Fourier series,
is called the -th harmonic; the two waves of the complex series are its two halves. It makes whole turns in one period. If the variable is a position , the period is the wavelength , and
if the variable is a time , the period is , and
Thus,
Each harmonic contributes a specific frequency, amplitude, and phase to the total waveform. The original function is reconstructed by summing all harmonic components:
As more harmonics are included, the approximation of becomes more accurate. Lower harmonics () describe the overall shape, while higher harmonics capture finer details and sharp transitions.
Applications in Signal Processing
A low-pass filter removes high-frequency (large ) components from a signal, keeping only low-frequency terms:
where sets the cutoff frequency.
Signal processing: Decompose and reconstruct signals by frequency content.
Noise reduction: Remove unwanted high-frequency components using low-pass filtering.
Data compression: Approximate a signal using only the dominant harmonics.
Physics: Analyze normal modes in vibrating strings, membranes, and quantum systems.
Video Lecture 20, Part 8: The Square Step: the Verdict (30:37)
Find the Fourier series of

Fourier form on . With period , the series is
where, from the coefficient formulas with so that ,
The unknowns are , and : once they are found, the expansion is known, each coefficient giving the contribution of its harmonic.
Solution.
Step 1: . The constant term of the series is , so we expect it to equal the average of , namely .
Step 2: (all ).
Step 3: (all ).
Hence
Result.
(i.e., the sum runs over odd ). At the jump points , the series converges to the average (Dirichlet’s theorem).
Video Lecture 20, Part 9: Two Ramps (35:28)

Fourier form on (period ).
Solution.
Step 2: Cosine coefficients ().
Step 3: Sine coefficients ().
Final series.
or, separating even/odd indices,
At the jump points , the series converges to the midpoint value .
Let
We find its real Fourier series

The variable is now time: , the period , and the wave number becomes the angular frequency . The unknowns are , and .
Solution.
Hence the DC term is .
Step 2: Cosine coefficients.
Integration by parts with , gives
Therefore
Using and ,
Hence
Step 3: Sine coefficients.
With , ,
Thus
Result (real Fourier series).
vanish for even ; cosine terms appear only for odd harmonics and decay as .
decay as with alternating sign, reflecting the sharp corner at (Gibbs effect upon truncation).
The average (DC) level over a period is .
Video Lecture 20, Part 10: The Sawtooth Is the Violin (42:13)
Consider a spatially periodic function
which repeats every interval of length . This represents a periodic ramp waveform, analogous to a sawtooth pattern in signal processing.
![The sawtooth: f(x)=x on [-L/2,L/2], repeated with period L.](assets/img/ch07_fig20_sawtooth.png)
We want its series in complex form, ; the unknowns are and .
Solution.
For a periodic function of period , the complex Fourier coefficients are
Substitute :
Step 2. Integration by parts.
Let
Then
Step 3. Simplify the second term.
The exponential integral gives
since for all integer . Hence, only the boundary term contributes.
Step 4. Evaluate the boundary term.
Using , we have
Step 5. DC (zero-frequency) coefficient.
since is an odd function and its average over one period is zero.
Step 6. Complete complex Fourier series.
Interpretation.
The amplitude of each harmonic decreases as , meaning higher-frequency components contribute less to the total waveform.
Because is an odd function, the coefficients are purely imaginary (), corresponding to a sine-only spectrum.
Truncating the series to a few terms yields a smooth approximation of the ramp—analogous to applying a low-pass filter that removes sharp discontinuities.
7.3Fourier Transform
Video Lecture 21, Part 1: From Series to Transform (0:00)
The Fourier transform can be viewed as the limiting case of the complex Fourier series when the period . In this limit, the function becomes nonperiodic, and the discrete spectrum of wave numbers becomes continuous.
Transition from Discrete to Continuous Spectrum
Consider a periodic function with period :
When , the spacing between successive wave numbers becomes infinitesimal:
The discrete index is replaced by a continuous variable :
Let
Then becomes a continuous function that represents the spectral amplitude density of .
Substituting into the series and taking the limit :
: function in the spatial (time) domain.
: its representation in the frequency (wave number) domain.
The transform pair is symmetric up to the normalization factor .
Video Lecture 21, Part 2: The Pulse and the Sinc (4:32)
Let
This is a nonperiodic rectangular pulse centered at the origin with total width and unit height. The unknown is its Fourier transform : which waves the pulse contains, and how much of each.

We use the transform pair
Solution.
Step 2. Evaluate the integral.
Step 3. Simplify and express the result.
At , we take the limit
Key features.
is real, even, and has zeros at for .
The main lobe width — increasing (wider pulse) narrows the spectrum.
The transform pair summarizes the width–bandwidth tradeoff: a narrow function in -space corresponds to a wide function in -space and vice versa.

Video Lecture 21, Part 3: The Fading Note (8:37)
Let
This function decays exponentially on both sides of :

We use the Fourier transform convention:
Solution.
Step 2. Evaluate each integral.
Step 3. Combine the results:
Thus, the Fourier transform of is
which is known as the Lorentzian function.
Shape of :

As increases, becomes broader and shorter; as decreases, it becomes narrower and taller.
7.4A Quantum Particle in Fourier Space
Video Lecture 21, Part 4: A Quantum Particle in Fourier Space (12:36)
The momentum-space wave function
In quantum mechanics, the state of a particle is described by a wave function ; the quantity is the probability of finding the particle between and . According to de Broglie, a plane wave has the definite momentum , where J s. Expanding in plane waves,
therefore expresses the state in terms of states of definite momentum. The Fourier transform is the momentum-space wave function, and is proportional to the probability that the particle has momentum .
A free particle (one on which no force acts) localized near the origin is described by a Gaussian wave packet,
Its position probability density, , is a normal distribution centred at .
Find the momentum-space wave function of the Gaussian wave packet
We use the transform pair
The unknown is : how much of each momentum the packet contains.
Solution.
Step 1. Write the transform.
Step 2. Remove the odd part. Write . The Gaussian is even and is odd, so their product integrates to zero over :
Step 3. Find a differential equation for . Differentiate under the integral sign with respect to :
Since , we may write and integrate by parts:
The boundary term is zero because the Gaussian vanishes at , and the last integral is :
Step 4. Solve the differential equation. Separate the variables and integrate from to :
Step 5. Find . Substitute , , and use the Gaussian integral :
Result.
Key features.
The Fourier transform of a Gaussian is again a Gaussian, centred at .
is in the denominator of the exponent of and in the numerator of the exponent of : a packet that is narrower in position (smaller ) has a broader distribution of momenta (Figure 7.17).

The uncertainty relation
Video Lecture 21, Part 5: Uncertainty, and the Spreading Electron (20:24)
The width of a probability density is measured by its standard deviation :
For a normal distribution , by symmetry, and integration by parts gives , so and
The standard deviation of a Gaussian can therefore be read off its exponent. Another common measure of width is the full width at half maximum (FWHM), the distance between the two points at which falls to half its peak value. For the normal distribution, gives , so .
In quantum mechanics, is the standard deviation of , and the momentum uncertainty is , where is the standard deviation of . The Heisenberg uncertainty principle states that for every wave function.
Find , and the product for the Gaussian wave packet, whose position and momentum wave functions are
We use . The unknowns are and , the standard deviations of and .
Solution.
Step 1. Position. Squaring doubles the exponent:
This is the normal form with , so .
Step 2. Momentum. Squaring ,
Match the exponent to the normal form, , which gives . Hence and .
Step 3. Product.
Result.
Key features.
The product does not depend on : squeezing the packet in position broadens it in momentum by the same factor.
The Gaussian wave packet attains the minimum allowed by the uncertainty principle.
This is the width–bandwidth relation of the rectangular pulse, multiplied by .
Time evolution of a free particle
The time evolution of a free particle of mass is governed by the Schrödinger equation
The equation is linear, so any superposition of solutions is again a solution.
A free particle of mass is the Gaussian wave packet at .
- (a)
Show that the plane wave solves the Schrödinger equation if .
- (b)
Find and the width of .
- (c)
An electron ( kg) starts with Å. When has its width doubled?
We use the momentum-space wave function of the packet, . The unknowns are and .
Solution.
(a) Step 1. Differentiate. For ,
(a) Step 2. Substitute. The left side is , and the right side is . They are equal if
which is the free-particle energy .
(b) Step 1. Superpose plane waves. At , . By (a), each plane wave evolves by acquiring the factor , and by linearity so does the sum:
In Fourier space, the partial differential equation has become one multiplication for each .
(b) Step 2. Insert and . The two exponents combine:
(b) Step 3. Evaluate the integral. Steps 2–5 of the previous Gaussian example give, for any , . Rename and set :
(The result also holds for complex with ; we use it here without proof.) Therefore
(b) Step 4. Square the magnitude. With ,
(b) Step 5. Read off the width. Compare with :
(c) Doubling time. when , that is, , with . For the electron,
Result.
Key features.
: the momentum distribution never changes, as expected with no force acting.
The packet stays Gaussian but spreads, because its faster components move ahead of its slower ones (Figure 7.18).
The spreading time grows with : a heavy or wide packet spreads slowly, an electron confined to an atom spreads in about s.

7.5The Fourier Transform in Two and Three Dimensions
Video Lecture 21, Part 6: The Fourier Transform in 2D (25:59)
The functions considered so far depend on one variable. An image is a function of two variables, its brightness , and a field in space is a function of three. The Fourier transform extends to these cases directly; the new feature is that a wave in more than one dimension has a direction.
Plane waves and the wave vector
In one dimension, the wave is characterized by a single wave number . In two dimensions, a plane wave is characterized by its wave numbers along and along , which form the wave vector . With , the exponent is a dot product:
Wavefronts. The phase is constant on lines perpendicular to . The wavefronts are therefore parallel lines perpendicular to , and the wave propagates along .
Wavelength. , as in one dimension; a longer wave vector means a shorter wavelength.
Three dimensions. and . The wavefronts are planes perpendicular to , hence the name plane wave.
The one-dimensional transform is applied along each axis, with a factor for each:
where runs over all of space and .
Reading a two-dimensional transform
The transform can itself be displayed as an image on the -plane. Each point represents one plane wave, and at that point is its amplitude in . Points near the origin correspond to long wavelengths, which carry the large-scale structure of an image; points far from the origin correspond to short wavelengths, which carry its fine detail.
The Dirac delta function. The transform of a single plane wave is concentrated at one point of the -plane. To describe it we need the Dirac delta function , which is zero for every and has unit area:
It is the limit of a peak that becomes ever taller and narrower while keeping unit area. An example is the transform of the rectangular pulse, , as : its height grows, its width shrinks, and its area stays . The inverse transform of is , so the transform of a single plane wave is a delta function:
An image of parallel stripes has brightness , where is a fixed wave vector. The brightness is greatest where is a multiple of , and the stripes are perpendicular to .
Find the two-dimensional Fourier transform of the stripe pattern
We use
The unknown is : which plane waves the image contains.
Solution.
Step 1. Write the transform.
Step 2. Write the cosine in exponential form. With ,
Step 3. Separate the double integrals. Each exponential is a product of a function of and a function of , so each double integral is a product of two single integrals. For the first term,
Step 4. Evaluate each single integral. By the delta-function result, the brackets are and . The second term is the same with replaced by .
Result.
Key features.
is zero everywhere except at the two points and (Figure 7.19): the image contains exactly two plane waves.
Rotating the stripes rotates the two points about the origin.
Closer stripes (shorter wavelength ) move the two points farther from the origin.

Scanners and electrical interference can superimpose a periodic stripe pattern on an image. Figure 7.20 (left) shows a portrait of Fourier corrupted in this way. Its brightness is the face plus one stripe pattern; the face varies slowly across the image.
The brightness of a corrupted portrait is
Use the two-dimensional Fourier transform to remove the stripes while leaving the face unchanged. We use the transform pair of the definition and the transform of the stripe pattern from the previous example. The unknown is a modified transform whose inverse is .
Solution.
Step 1. Transform the image. The transform is linear, so the transform of is the sum of the transforms of its two terms. By the previous example,
where .
Step 2. Locate the two contributions. The face varies slowly, so is concentrated near the origin of the -plane. The stripes appear as two isolated peaks at , away from the face (Figure 7.20, middle).
Step 3. Filter. Set to zero in a small disk around each of the points . This removes the two delta functions, and changes only where it is already negligible:
Step 4. Transform back.
Result. The stripes are removed and the face is unchanged (Figure 7.20, right).
Key features.
The method works because the noise and the image occupy different regions of the -plane.
Image-processing software removes periodic noise in the same way.

Diffraction
Video Lecture 21, Part 7: Light Through an Opening (31:44)
Diffraction is the spreading of a wave after it passes through an aperture. Light passing through a wide aperture continues almost in a straight line; an aperture comparable in size to the wavelength spreads it over a wide range of angles.
Let light of wavelength pass through an aperture in an opaque screen and fall on a distant screen. By Huygens' principle, each point of the aperture acts as a source of waves. In a direction at angle to the forward direction, the wave from a point at height travels an additional distance and therefore acquires the phase , where . Summing these contributions over the aperture gives the amplitude on the distant screen:
where the aperture function is inside the aperture and outside, and , are the horizontal and vertical angles. The amplitude is therefore the two-dimensional Fourier transform of the aperture; the observed intensity is its squared magnitude.
Find the diffraction pattern of a square aperture of side centred at the origin,
and the angles at which the intensity is zero. How does the pattern change as decreases? We use the diffraction formula above. The unknown is the intensity as a function of and .
Solution.
Step 1. Restrict the integral to the aperture. inside the square and outside, so
Step 2. Separate the double integral.
Step 3. Evaluate each integral. As for the rectangular pulse,
and the same with in place of .
Step 4. Square to obtain the intensity.
Each factor takes its largest value, , at (straight ahead).
Step 5. Find the zeros. The intensity vanishes when with , that is, with With , this gives , and similarly for .
Result.
Key features.
Each factor is large only when its own is near zero, so the light is concentrated along the two axes: a cross-shaped pattern (Figure 7.21).
Reducing moves the zeros to larger angles: a smaller aperture produces a wider pattern, in accordance with the width–bandwidth relation.

A circular aperture of diameter has circular symmetry, so its diffraction pattern is a central bright disk surrounded by concentric rings (the Airy pattern, Figure 7.21). Its transform involves a Bessel function and is not derived here; the first dark ring occurs at . This limits the angular resolution of a telescope of diameter to about .
Three dimensions
Video Lecture 21, Part 8: The Fourier Transform in 3D (36:33)
If , the exponential factors in the same way, , and the triple integral reduces to a product of three one-dimensional transforms:
For example, the three-dimensional Gaussian (Section 7.4) has the transform , which is again spherically symmetric.
The wave function of an atom is spherically symmetric but not separable in Cartesian coordinates. For a spherically symmetric function, the transform reduces to a single radial integral.
If depends only on , its Fourier transform depends only on and is
Since cannot depend on the direction of , choose along the axis, so that . In spherical coordinates (Chapter 5), , and
The integral gives . With , the integral is . Combining these factors gives the result.
In the ground state of the hydrogen atom, the electron's wave function is
where Å is the Bohr radius. As in one dimension, the momentum-space wave function is the (three-dimensional) Fourier transform , and . Since will not be a Gaussian, we measure its width by the half width at half maximum (HWHM): the value of at which falls to half its peak value, that is, half the FWHM.
Find the momentum-space wave function of the hydrogen ground state and its half width at half maximum . Estimate the speed of the electron from ( kg). We use the transform of a spherically symmetric function,
The unknowns are , and the speed .
Solution.
Step 1. Write the radial integral.
Step 2. Use a complex exponential. Write . Then with , and
Step 3. Integrate by parts. Since , the boundary terms vanish at :
Step 4. Take the imaginary part. Multiply numerator and denominator by , with and :
Step 5. Assemble .
Step 6. Find the half width. The peak is at . Half the peak value is reached when
Step 7. Estimate the speed. With ,
Result.
Key features.
The width in is of order : the smaller the atom, the broader its momentum distribution.
The root-mean-square momentum computed from is exactly (the integral is not evaluated here), which gives m/s . Both estimates give the same order of magnitude.
Figure 7.22 shows the electron's distribution in both spaces: of size about in position space and about in momentum space.

7.6The Fourier Series as an Eigenvalue Problem
Video Lecture 21, Part 9: Beyond Sines and Cosines (43:35)
Why can sines and cosines represent essentially any function? The answer is the same as for symmetric matrices in Chapter 3: a symmetric matrix has orthogonal eigenvectors that form a basis. Here the matrix is replaced by a differential operator, and the eigenvectors by eigenfunctions.
The differential equation as an eigenvalue problem
Recall from Chapter 3 that a symmetric matrix has eigenvectors . Eigenvectors belonging to different eigenvalues are orthogonal, and together they form a basis: any vector can be expanded as with .
Now consider a string of length fixed at both ends and the operator ( denotes the operator and the length; the minus sign makes the eigenvalues positive). Its eigenvalue problem is a differential equation with a boundary condition at each end:
Its solutions are
Applying to returns multiplied by a constant, just as . Thus is an eigenfunction of with eigenvalue . The correspondence is summarized below.
| Linear algebra (Chapter 3) | Functions (this chapter) |
| vector , with components | function , with a value at each |
| dot product | inner product |
| symmetric matrix | operator with boundary conditions |
| eigenvector, | eigenfunction, |
If and vanish at both ends, two integrations by parts give : the operator is symmetric, like the matrix . Applied to two eigenfunctions, this gives
so : eigenfunctions with different eigenvalues are orthogonal.
There is one eigenfunction for each , and together they form a complete set: any reasonable function on can be expanded as with , which is the Fourier sine series. With the periodic condition in place of fixed ends, the same operator has the eigenfunctions , and the expansion is the complex Fourier series of this chapter (for complex functions the inner product is ).
Sturm–Liouville problems
The argument used only one property of , its symmetry, and many equations of physics share this property. A string whose mass per unit length varies along its length vibrates according to ; heat conduction in a nonuniform bar, the problem studied by Sturm and Liouville in the 1830s, leads to an equation of the same type.
Consider the operator
with , , and a boundary condition at each end (for example ). Then:
the eigenvalues are real;
eigenfunctions with different eigenvalues are orthogonal with respect to the weight function : for ;
the eigenfunctions form a complete set: any with finite can be expanded as
(7.168)
For piecewise smooth the series converges to wherever is continuous, and to the average of the one-sided limits at a jump, as in the Dirichlet theorem.
The Fourier series is the special case , . In general the weight defines a weighted inner product, .
Consider a string of unit length, fixed at both ends and held at tension , whose mass per unit length increases along the string as . In a normal mode every point oscillates as , and Newton's second law gives . The modes cannot be found in closed form; they are computed numerically and normalized so that .
The modes of the string satisfy
- (a)
Identify , and the weight function in the Sturm–Liouville form.
- (b)
The first two modes give (each chosen positive near ). Which integral of vanishes, and why?
- (c)
Find the coefficients in the expansion of an initial shape.
We use the Sturm–Liouville form . The unknowns are , , , the vanishing integral, and .
Solution.
(a) Compare with the Sturm–Liouville form. The equation is , so
(b) Step 1. Write the equations of the two modes.
(b) Step 2. Multiply and integrate. Multiply the second equation by and the first by , and integrate over :
(b) Step 3. Integrate by parts. Since and vanish at both ends,
and in the same way . The two left sides of Step 2 are equal.
(b) Step 4. Subtract.
The eigenvalues are different, so the weighted integral vanishes.
(c) Project onto one mode. Multiply by and integrate. By orthogonality, only the term survives, and by the normalization its integral is :
Result.
Key features.
The weight is essential: the unweighted integral, , does not vanish.
The modes are not sine functions; their oscillations are concentrated toward the heavier end (Figure 7.23).
The computed eigenvalues, , are close to those of a uniform string with the average density , .

The same theorem applies to the special functions of later chapters: Legendre polynomials (Chapter 9), which arise in problems with spherical symmetry ( on ); Bessel functions, which describe vibrating drums and waves in pipes (); and Hermite functions, the eigenfunctions of the quantum harmonic oscillator. Each arises from its own operator, and each forms a complete set with the same coefficient formula; for example, on . Expansion in eigenfunctions is also the method used in Chapter 10 to solve partial differential equations.
7.7Where Fourier Analysis Is Used
The single idea of this chapter — that a function can be rebuilt from sines and cosines, or from complex exponentials — is one of the most widely used in all of physics and engineering. The reason is that the basis functions are the natural language of anything that oscillates, and of the linear equations that govern waves, heat and quantum systems.
Solving partial differential equations. Fourier's original motive was heat flow. Because each mode evolves independently under the heat, wave and Schrödinger equations, expanding the initial state in modes turns a partial differential equation into a set of ordinary ones. This is the method of Chapter 10.
Signal processing. A signal's Fourier coefficients are its frequency content. Filtering, noise removal and compression are all operations on that spectrum, as the low-pass filter of this chapter showed.
Spectroscopy and optics. The rectangular-pulse and Lorentzian transforms computed above are exactly the line shapes measured in spectroscopy: the width of a spectral line and the lifetime of the state that produced it are a Fourier transform pair.
Quantum mechanics. Position and momentum wavefunctions are Fourier transforms of one another. The width–bandwidth tradeoff seen in the pulse example — narrow in means wide in — is, with , the Heisenberg uncertainty principle.
Normal modes. The harmonics of a Fourier series are the standing waves of a vibrating string or air column; their frequencies are why instruments have a well-defined pitch.
The width–bandwidth tradeoff deserves a second mention because it recurs under so many names. A pulse of duration has a spectrum of width . In signal processing this limits how sharply a filter can act; in spectroscopy it is the natural linewidth; in quantum mechanics it is . They are all one statement about Fourier pairs: you cannot make a function narrow in both domains at once.
Summary Table
| Object | Formula | Notes |
| Complex series | ; the compact form. | |
| Complex coefficient | For real , . | |
| Real series | is the average of . | |
| Real coefficients | The gives at . | |
| Even / odd | even ; odd | Halves the work when has symmetry. |
| Dirichlet / jumps | series | Converges to the midpoint at a jump. |
| Fourier transform | The limit of the series. | |
| Inverse transform | Rebuilds from its spectrum. | |
| Width–bandwidth | Narrow in wide in . | |
| Transform in 3D | One transform per axis; a product splits. | |
| Gaussian pair | A bell becomes a bell; . | |
| Eigenfunction expansion | Sturm–Liouville; Fourier is . |
Fourier transforms are defined with several different placements of the : all on the forward transform (as here), all on the inverse transform, split symmetrically as on each, or attached to the frequency variable instead of . Every choice is self-consistent, but formulas from two sources cannot be mixed without checking. Whenever you quote a transform, note which convention it assumes.
For the Interested Reader
Fourier analysis is a subject where a single good picture replaces pages of algebra: once you see a Fourier series as a sum of rotating vectors, the coefficient formulas stop looking arbitrary. The videos below are the best route to that picture. Everything listed is free.
The idea, seen
3Blue1Brown builds the Fourier series and then the transform from the geometry of rotating vectors, motivated — as Fourier himself was — by heat flow:
But what is a Fourier series? From heat flow to drawing with circles — 3Blue1Brown
Watch on YouTube
But what is the Fourier transform? A visual introduction — 3Blue1Brown
Watch on YouTube
A complementary take, arriving at the transform from the physics rather than the geometry:
To Understand the Fourier Transform, Start From Quantum Mechanics — Physics with Elliot
Watch on YouTube
Worked technique and practice
Paul's Online Math Notes. Fourier Series and Fourier Sine Series — careful worked examples of exactly the coefficient integrals in this chapter, including the even/odd shortcuts.
If only one of these fits your time, watch the two 3Blue1Brown videos before rereading the chapter. Most of what makes Fourier analysis feel abstract — why complex exponentials, why the integral extracts a coefficient, what a negative frequency means — becomes obvious once the series is seen as vectors adding tip to tail.
Video lectures
Each section of this chapter links to the part of a lecture that teaches it. The full lectures are below, and all of them are in the course playlist.