# Harmonic analysis Harmonic analysis is the branch of [[Mathematical_analysis|mathematical analysis]] that represents functions and signals as superpositions of basic waves and studies what survives the translation between a [[Time_domain|time-domain]] description and its [[Frequency_domain|frequency-domain]] shadow. It generalizes [[Fourier_analysis|Fourier analysis]] far beyond periodic functions — to groups, [[Manifold|manifolds]], and graphs — and it underwrites an astonishing share of the vault: every [[Signal_processing|signal-processing]] chain, every spectral method for a [[Partial_differential_equation|partial differential equation]], every [[Sine_wave|sinusoidal]] decomposition of an [[Oscillation|oscillation]], and the [[Uncertainty_principle|uncertainty principle]] itself are theorems of this field wearing work clothes. Its central discovery is that "frequency" is not a physics metaphor but an algebraic structure, which is why the same mathematics prices vibrations, compresses images, and diagonalizes the [[Harmonic_oscillator|harmonic oscillator]]. ## Fourier's scandalous claim The founding act was Joseph Fourier's assertion (memoir 1807; *Théorie analytique de la chaleur*, 1822) that essentially arbitrary functions — even ones with corners — expand in sines and cosines. He needed it to solve [[Heat_transfer|heat conduction]]: on a rod, each mode sin(nπx/L) decays independently at rate proportional to n², so expanding the initial temperature in modes solves the equation term by term. The same separation tames the [[Wave_equation|wave equation]], whose standing modes at [[Angular_frequency|angular frequencies]] ω_n are literally the harmonics of a string — the etymology of the field. Making Fourier's claim rigorous took a century and forced the invention of much of modern analysis: Riemann's and Lebesgue's integrals, Cantor's [[Set_theory|set theory]] (born from studying sets where trigonometric series fail), and eventually the L² theory in which the Plancherel theorem (1910) says the transform preserves energy — ∫|f|² dx = ∫|f̂|² dξ — making it a rotation of an infinite-dimensional space, the home turf of [[Functional_analysis|functional analysis]]. ## The transform zoo and one theorem that runs it Applications meet the field through a family of [[Integral_transform|integral transforms]]. The Fourier transform handles signals on the whole line; Fourier series handle periodic ones; the [[Laplace_transform|Laplace transform]] extends to growing signals and turns an [[Ordinary_differential_equation|ordinary differential equation]] into algebra, which is why [[Control_theory|control theory]] lives in the s-plane; the [[Z-transform|Z-transform]] is its [[Discrete_time_and_continuous_time|discrete-time]] twin, ruling digital filters; the [[Discrete_cosine_transform|discrete cosine transform]] packs image energy into few coefficients, the engine of [[Image_compression|image compression]]; the [[Discrete_wavelet_transform|discrete wavelet transform]] trades pure frequency for localization. One theorem powers them all: transforms convert [[Convolution|convolution]] into multiplication. Since every linear time-invariant system acts by convolution with its [[Impulse_response|impulse response]], the transform diagonalizes the system — the [[Transfer_function|transfer function]] and [[Frequency_response|frequency response]] of a [[Linear_time-invariant_system|linear time-invariant system]] are Fourier data, and complex exponentials e^{iωt} are its eigenfunctions. Computationally the whole edifice became cheap when the [[Fast_Fourier_transform|fast Fourier transform]] (Cooley–Tukey, 1965) cut the cost to O(N log N). ## Localization and the uncertainty principle The transform's power has a price: a function and its transform cannot both be sharply concentrated. Quantitatively, σ_t·σ_ω ≥ ½, with equality only for Gaussians — the mathematical fact that becomes the Heisenberg [[Uncertainty_principle|uncertainty principle]] when [[Quantum_mechanics|quantum mechanics]] identifies momentum with the frequency variable of the [[Schrödinger_equation|Schrödinger equation]]'s wavefunction. The same inequality is an engineering constraint: a filter cannot have both a narrow band and a short ring, and a musical note cannot be simultaneously brief and pitch-certain. Wavelets (Morlet and Grossmann, 1980s; Daubechies' compactly supported families, 1988) answer by tiling the time–frequency plane adaptively — fine in time at high frequency, fine in frequency at low — which is exactly what transient-rich signals like seismograms and heartbeats need. Sampling sits under the same roof: the [[Nyquist–Shannon_sampling_theorem|Nyquist–Shannon sampling theorem]] says a band-limited signal is fully determined by [[Sampling_(signal_processing)|samples]] taken at twice its bandwidth, the bridge from analysis to [[Information_theory|information theory]] that [[Claude_Shannon|Claude Shannon]] crossed in 1948–49. ## Groups underneath the sines The deep reason sines keep appearing: e^{iωt} are the characters of the translation group. Twentieth-century harmonic analysis rebuilt Fourier theory on any locally compact group equipped with Haar's invariant measure (1933): Fourier series live on the circle, the Fourier transform on the real line, the discrete transform on ℤ/Nℤ, and Pontryagin duality (1930s) organizes all the commutative cases. For noncommutative symmetry — rotations, the [[Lie_group|Lie groups]] of physics — characters become matrix-valued representations, and the Peter–Weyl theorem (1927) decomposes functions on the group; spherical harmonics, the atomic-orbital shapes of the [[Hydrogen_atom|hydrogen atom]], are this machinery on the sphere. The upshot for [[Group_theory|group theory]] and [[Complex_analysis|complex analysis]] alike is a slogan with teeth: harmonic analysis is the spectral theory of symmetry, and "frequency" is whatever a group's representations say it is — a viewpoint that now extends to [[Graph_theory|graphs]], where eigenvectors of the [[Adjacency_matrix|adjacency matrix]] play the role of Fourier modes for [[Network_theory|networks]]. ## Where the vault hears it Downstream dependencies are everywhere. [[Acoustics|Acoustics]] and [[Sound|sound]] engineering decompose pressure waves; [[Digital_signal_processing|digital signal processing]] implements the theory in silicon; [[Autocorrelation|autocorrelation]] and [[Least-squares_spectral_analysis|least-squares spectral analysis]] extract periodicities from noisy [[Time_series|time series]]; [[Norbert_Wiener|Norbert Wiener]]'s generalized harmonic analysis (1930) founded the spectral view of random signals that yields the [[Wiener_filter|Wiener filter]]; [[Filter_design|filter design]] and [[Speech_coding|speech coding]] are applied inequalities from this field; [[Magnetic_resonance_imaging|magnetic resonance imaging]] reconstructs bodies from Fourier samples. Even number theory leans on it — but the vault's readers will meet harmonic analysis most often as the reason a [[Complex_system|complex system]]'s response can be read one frequency at a time. **On the spine:** [[Fourier_analysis]] · [[Functional_analysis]] · [[Signal_processing]] · [[Uncertainty_principle]] · [[Integral_transform]]. ## Wikipedia : Wikitube **Strict pair:** [Wikipedia](https://en.wikipedia.org/wiki/Harmonic_analysis) : [Wikitube](https://en.wikitube.io/wiki/Harmonic_analysis) ## Previous hub tags Hubs: `Systems`. Portals: [[PORTAL_Graph_theory]], [[PORTAL_Decision_theory]], [[PORTAL_Information_theory]], [[PORTAL_Control_theory]], [[PORTAL_Operations_research]]. --- *Repopulated 2026-08-12 · redlink fill · 0 deletions.*