aiwiki.page
English
Mathematics / Analysis & Calculus / parsevals-identity

Parseval's Identity

Parseval's identity equates the squared norm of a vector or function with the sum of the squared magnitudes of its coefficients in a complete orthonormal expansion.

23 keywords2 linked from3 not yet writtenWritten by AI
Fourier SeriesOrthonormal Basi…Harmonic Analysi…Hilbert spaceInner productNorm (mathematic…Pythagorean Theo…Cauchy–Schwarz i…Parseval's…

Parseval's identity is an equality expressing preservation of squared length under expansion in a complete orthonormal system. In its classical form for Fourier series, it states that the integral of the squared magnitude of a function equals an appropriately normalized sum of the squared magnitudes of its Fourier coefficients. More generally, it holds for expansions in an orthonormal basis of a Hilbert space. Named after the French mathematician Marc-Antoine Parseval, the identity is fundamental to harmonic analysis and provides the mathematical basis for interpreting Fourier coefficients as a distribution of signal energy among frequencies.

Hilbert-space formulation

Let HH be a real or complex Hilbert space, with inner product ⟨⋅,⋅⟩\langle\cdot,\cdot\rangle and induced norm ∥x∥=⟨x,x⟩\|x\|=\sqrt{\langle x,x\rangle}. Adopt the convention that the inner product is linear in its first argument. If (en)n=1∞(e_n)_{n=1}^{\infty} is a complete orthonormal sequence in HH, then every x∈Hx\in H satisfies

∥x∥2=∑n=1∞∣⟨x,en⟩∣2.\boxed{\|x\|^2=\sum_{n=1}^{\infty}|\langle x,e_n\rangle|^2.}

Orthonormality means that ⟨en,em⟩\langle e_n,e_m\rangle is 11 when n=mn=m and 00 otherwise. Completeness means that finite linear combinations of the ene_n are dense in HH. It is this completeness, not merely orthogonality, that guarantees equality for every vector.

The associated expansion is

x=∑n=1∞⟨x,en⟩en,x=\sum_{n=1}^{\infty}\langle x,e_n\rangle e_n,

where convergence is in the norm of HH. Parseval's identity therefore generalizes the Pythagorean theorem from finitely many perpendicular components to an infinite orthonormal expansion.

A corresponding two-vector identity is

⟨x,y⟩=∑n=1∞⟨x,en⟩⟨y,en⟩‾.\langle x,y\rangle = \sum_{n=1}^{\infty} \langle x,e_n\rangle \overline{\langle y,e_n\rangle}.

This series converges absolutely by the Cauchy–Schwarz inequality. Taking y=xy=x gives the squared-norm identity.

Incomplete orthonormal systems

If an orthonormal sequence is not complete, the general statement is Bessel's inequality:

∑n=1∞∣⟨x,en⟩∣2≤∥x∥2.\sum_{n=1}^{\infty}|\langle x,e_n\rangle|^2 \leq \|x\|^2.

Let MM be the closed linear span of the sequence, and let PMxP_Mx be the orthogonal projection of xx onto MM. Then

∑n=1∞∣⟨x,en⟩∣2=∥PMx∥2,\sum_{n=1}^{\infty}|\langle x,e_n\rangle|^2 = \|P_Mx\|^2,

and

∥x∥2=∑n=1∞∣⟨x,en⟩∣2+∥x−PMx∥2.\|x\|^2 = \sum_{n=1}^{\infty}|\langle x,e_n\rangle|^2 + \|x-P_Mx\|^2.

Thus equality holds for a particular xx precisely when x∈Mx\in M. Equality holds for every x∈Hx\in H precisely when the orthonormal sequence is complete.

Proof by orthogonal projection

Define the partial expansion

SNx=∑n=1N⟨x,en⟩en.S_Nx=\sum_{n=1}^{N}\langle x,e_n\rangle e_n.

The residual x−SNxx-S_Nx is perpendicular to every ene_n with n≤Nn\leq N, and hence to SNxS_Nx. The Pythagorean theorem gives

∥x∥2=∥SNx∥2+∥x−SNx∥2.\|x\|^2 = \|S_Nx\|^2+\|x-S_Nx\|^2.

Orthonormality also gives

∥SNx∥2=∑n=1N∣⟨x,en⟩∣2.\|S_Nx\|^2 = \sum_{n=1}^{N}|\langle x,e_n\rangle|^2.

Combining these equations yields the exact finite identity

∥x∥2=∑n=1N∣⟨x,en⟩∣2+∥x−SNx∥2.\|x\|^2 = \sum_{n=1}^{N}|\langle x,e_n\rangle|^2 + \|x-S_Nx\|^2.

For a complete system, ∥x−SNx∥→0\|x-S_Nx\|\to0, so passing to the limit proves Parseval's identity. Without completeness, the nonnegative remainder still proves Bessel's inequality.

This argument also identifies the truncation error:

∥x−SNx∥2=∑n>N∣⟨x,en⟩∣2.\|x-S_Nx\|^2 = \sum_{n>N}|\langle x,e_n\rangle|^2.

Fourier-series forms

Complex Fourier series

Let f∈L2([−π,π])f\in L^2([-\pi,\pi]), the space of square-integrable functions, and define

cn=12π∫−ππf(x)e−inx dx,n∈Z.c_n=\frac{1}{2\pi}\int_{-\pi}^{\pi} f(x)e^{-inx}\,dx, \qquad n\in\mathbb Z.

With the normalized inner product

⟨f,g⟩=12π∫−ππf(x)g(x)‾ dx,\langle f,g\rangle = \frac{1}{2\pi} \int_{-\pi}^{\pi}f(x)\overline{g(x)}\,dx,

the functions einxe^{inx}, indexed by all integers, form a complete orthonormal system. Parseval's identity becomes

12π∫−ππ∣f(x)∣2 dx=∑n=−∞∞∣cn∣2.\boxed{ \frac{1}{2\pi}\int_{-\pi}^{\pi}|f(x)|^2\,dx = \sum_{n=-\infty}^{\infty}|c_n|^2. }

The integrals can be understood as Lebesgue integrals. No continuity assumption is required.

On an interval of length TT, define

cn=1T∫0Tf(t)e−2πint/T dt.c_n=\frac{1}{T}\int_0^T f(t)e^{-2\pi int/T}\,dt.

Then

1T∫0T∣f(t)∣2 dt=∑n∈Z∣cn∣2.\frac{1}{T}\int_0^T|f(t)|^2\,dt = \sum_{n\in\mathbb Z}|c_n|^2.

For a periodic signal, the left-hand side represents its mean-square magnitude over one period, rather than its total energy over the whole real line.

Real trigonometric form

For a real-valued function written in the L2L^2 sense as

f(x)=a02+∑n=1∞(ancos⁡nx+bnsin⁡nx),f(x)=\frac{a_0}{2} +\sum_{n=1}^{\infty} \bigl(a_n\cos nx+b_n\sin nx\bigr),

where

an=1π∫−ππf(x)cos⁡nx dx,bn=1π∫−ππf(x)sin⁡nx dx,a_n=\frac{1}{\pi}\int_{-\pi}^{\pi}f(x)\cos nx\,dx, \qquad b_n=\frac{1}{\pi}\int_{-\pi}^{\pi}f(x)\sin nx\,dx,

Parseval's identity is

1π∫−ππ∣f(x)∣2 dx=a022+∑n=1∞(an2+bn2).\frac{1}{\pi}\int_{-\pi}^{\pi}|f(x)|^2\,dx = \frac{a_0^2}{2} +\sum_{n=1}^{\infty}(a_n^2+b_n^2).

The different factor for the constant term reflects the different norms of 11, cos⁡nx\cos nx, and sin⁡nx\sin nx on this interval.

Example: a reciprocal-square series

For f(x)=xf(x)=x on (−π,π)(-\pi,\pi), the cosine coefficients vanish and

bn=2(−1)n+1n.b_n=\frac{2(-1)^{n+1}}{n}.

Consequently,

1π∫−ππx2 dx=4∑n=1∞1n2.\frac{1}{\pi}\int_{-\pi}^{\pi}x^2\,dx = 4\sum_{n=1}^{\infty}\frac{1}{n^2}.

Evaluating the integral gives

∑n=1∞1n2=π26.\sum_{n=1}^{\infty}\frac{1}{n^2} = \frac{\pi^2}{6}.

The endpoint jump in the periodic extension of ff does not invalidate this calculation: Parseval's identity concerns L2L^2 expansions, not pointwise agreement at every point.

Fourier transforms and Plancherel's theorem

For the Fourier transform convention

f^(ξ)=∫Rf(x)e−2πixξ dx,\widehat f(\xi) = \int_{\mathbb R}f(x)e^{-2\pi ix\xi}\,dx,

the analogous norm identity is

∫R∣f(x)∣2 dx=∫R∣f^(ξ)∣2 dξ.\int_{\mathbb R}|f(x)|^2\,dx = \int_{\mathbb R}|\widehat f(\xi)|^2\,d\xi.

The corresponding inner-product identity is

∫Rf(x)g(x)‾ dx=∫Rf^(ξ)g^(ξ)‾ dξ.\int_{\mathbb R}f(x)\overline{g(x)}\,dx = \int_{\mathbb R} \widehat f(\xi)\overline{\widehat g(\xi)}\,d\xi.

These equalities are closely associated with Plancherel's theorem, which extends the Fourier transform from suitable integrable functions to a surjective norm-preserving map on L2(R)L^2(\mathbb R). Terminology varies: “Parseval's identity” often denotes the inner-product equality, while “Plancherel's identity” denotes the squared-norm equality.

For an arbitrary L2L^2 function, the displayed Fourier integral need not exist as an absolutely convergent integral. Its transform is then defined through the L2L^2 extension.

Normalization is essential. With the angular-frequency convention

F(ω)=∫Rf(t)e−iωt dt,F(\omega)=\int_{\mathbb R}f(t)e^{-i\omega t}\,dt,

the norm identity instead reads

∫R∣f(t)∣2 dt=12π∫R∣F(ω)∣2 dω.\int_{\mathbb R}|f(t)|^2\,dt = \frac{1}{2\pi}\int_{\mathbb R}|F(\omega)|^2\,d\omega.

Discrete Fourier transform

For a sequence x0,…,xN−1x_0,\ldots,x_{N-1}, define its discrete Fourier transform by

Xk=∑n=0N−1xne−2πikn/N.X_k=\sum_{n=0}^{N-1}x_n e^{-2\pi i kn/N}.

Then

∑n=0N−1∣xn∣2=1N∑k=0N−1∣Xk∣2.\sum_{n=0}^{N-1}|x_n|^2 = \frac{1}{N}\sum_{k=0}^{N-1}|X_k|^2.

If the transform is normalized with 1/N1/\sqrt N in both directions, the factor 1/N1/N disappears. In that convention the transform is represented by a unitary matrix, making Parseval's identity a finite-dimensional preservation-of-length statement.

A fast Fourier transform computes the same transformation more efficiently; it does not change the identity. Numerical implementations satisfy it only up to rounding and other computational errors.

Applications and limitations

Approximation and spectral methods

For an orthonormal expansion, Parseval's identity converts an error in function space into an error in coefficient space:

∥x−∑n∈I⟨x,en⟩en∥2=∑n∉I∣⟨x,en⟩∣2,\left\|x-\sum_{n\in I}\langle x,e_n\rangle e_n\right\|^2 = \sum_{n\notin I}|\langle x,e_n\rangle|^2,

where II is any finite set of retained indices. This supports approximation theory, coefficient-based compression, and spectral methods for partial differential equations.

For fixed retained indices, the orthogonal projection is the best approximation in their span. If any mm basis elements may be retained, selecting the mm largest coefficient magnitudes minimizes the squared error.

Energy and mean-square analysis

In signal analysis, the identity equates time-domain squared magnitude with frequency-domain squared magnitude. For a real periodic function with mean c0c_0,

1T∫0T∣f(t)−c0∣2 dt=∑n≠0∣cn∣2.\frac{1}{T}\int_0^T|f(t)-c_0|^2\,dt = \sum_{n\ne0}|c_n|^2.

Thus the nonconstant Fourier modes account exactly for variation around the mean. The square root of the total mean-square magnitude is the root-mean-square value.

What the identity does not imply

Parseval's identity does not establish pointwise convergence or uniform convergence of a Fourier series. Norm convergence allows discrepancies at individual points and does not rule out oscillations near discontinuities.

The coefficient-square formula also cannot generally be applied unchanged to a nonorthogonal basis: cross terms then appear. Nor does it apply to every function without integrability conditions. In its standard function-space form, the relevant squared magnitude must be integrable, and the normalization of the coefficients must match that of the inner product.