The discrete-time Fourier transform represents a sequence indexed by integer sample number using a continuous, periodic frequency variable. In cycles per sample, the transform pair isFor an absolutely summable sequence, the sum converges uniformly to a continuous function, and the inverse formula follows by integrating the exponentials term by term. Their integral over one frequency period is one when the integer exponent is zero and zero otherwise. The primary text gives the same pair in angular frequency , where the inverse integral carries a factor .[ref]
Because , . Discrete time therefore does not imply discrete frequency: the DTFT is defined at every frequency in one period. With sampling frequency , a physical frequency corresponds to ; frequencies separated by integer multiples of give identical sampled complex exponentials. This is the basic frequency ambiguity behind aliasing.
An original two-sample example takes and all other entries zero. ThenAt zero frequency its value is one; at its magnitude is with phase ; at it vanishes. As the impulse response of a two-tap moving average, it preserves a constant input and removes an alternating-sign input. SciPy’s frequency-response definition evaluates precisely such sums when its denominator is one; its frequency units depend on the fs parameter.[ref]
For a finite record extended by zero, the unnormalized discrete Fourier transform gives samples of the same continuous function:Zero padding the record to length instead evaluates on a denser grid. It does not change or restore unobserved time samples. The rectangular window and Hann window demonstrate how the selected finite record shapes that continuous response.
Convergence assumptions matter. For when , zero otherwise, and , the geometric sum givesAt , that ordinary convergent-sum argument fails. Square-summable sequences admit an interpretation, while perpetual tones require a generalized interpretation; neither should be treated as an absolutely convergent version of the same series without stating the interpretation.
A useful check of numerical transform conventions is to evaluate a finite sum at arbitrary frequencies, compare its DFT-grid values with the FFT output, and repeat after zero-padding. This tests the exponent sign, normalization and frequency units independently. In a physical acquisition, one must also state sampling rate, record duration and the response before sampling; dense numerical frequency points alone do not validate those experimental inputs.
The Fourier support of a function or distribution is the support of a function of its Fourier transform, with distributional support used when necessary. For an ordinary transform, it is the closure of the set of frequencies where is nonzero. Thus a bandlimited function has Fourier support in a bounded frequency region.
Under the convention , multiplication by translates Fourier support by . Translation of in physical space instead multiplies its transform by a phase, leaving Fourier support unchanged. This distinguishes where a signal occurs from which frequencies it contains.
For example, the distribution has Fourier support . A nonzero Gaussian function has transform nonzero at every real frequency, so it has no compact Fourier support despite rapid frequency decay. Small coefficients and exact vanishing describe different properties.
The short-time Fourier transform analyzes the local frequency content of a signal by translating a window and then taking a Fourier transform. For a window and suitable signal , one angular-frequency convention isThe time coordinate selects a neighborhood through the window function, while tests a local oscillation. The complex conjugation and Fourier normalization are part of the convention.
For and an integrable window, substitution gives . Its magnitude is centered on the tone frequency according to the window transform. A narrower time window improves temporal localization but spreads its Fourier transform, reflecting the uncentred Fourier uncertainty principle. The squared magnitude gives a spectrogram, which displays changing frequencies rather than only one whole-record spectrum.
A half-line Fourier transform separates the contributions supported on the positive and negative real axes. With the displayed sign convention, a bound as gives a holomorphic in ; a bound as gives a holomorphic in . The estimates hold uniformly on compact subsets of these half-planes and remain integrable after multiplication by any fixed power of , which justifies complex differentiation under the integral. The overlap strip is the natural domain for a Wiener-Hopf equation. Holomorphic continuation of a transform beyond its convergence region is a separate question from convergence of the defining integral.
For integrable functions on the real line, equality of their Fourier transforms implies equality almost everywhere. To prove it, subtract and assume . Convolve with a Gaussian heat kernel . The convolution theorem gives , which is integrable, so the Fourier inversion theorem gives almost everywhere. Finally is an approximate identity: . This follows from translation continuity in Lp with exponent one and the vanishing Gaussian tails. Hence in . If the two original functions are continuous, equality almost everywhere strengthens to equality everywhere, since a nonzero continuous difference has magnitude bounded below by a positive constant on some interval.
For , use the Fourier transform convention . Splitting the integral at zero gives . By the Fourier inversion theorem, the dual identity isThis pair connects an exponential function with the kernel of a Cauchy distribution and produces the Poisson kernel for the upper half-plane by inversion of .
A time-dependent response can be represented by its Fourier transform in angular frequency. With , a time derivative transforms to when boundary terms vanish. A time-translation-invariant response convolution becomes multiplication by a frequency-dependent response coefficient. This is why a dispersive electrical conductivity is naturally specified in the frequency domain. The inverse-transform modes use .
For , put . The regular distribution of is a tempered distribution, since the function is bounded. Its Fourier transform is the integrable functionwith arbitrary finite value at zero when the displayed integral diverges there. To prove the identity, the gamma function gives . Pair this with a Schwartz function, use the Fourier transform of a Gaussian, and interchange integrals by Fubini's theorem. The absolute bound for the transformed pairing follows from Tonelli's theorem:The positive kernel and dominated convergence give positivity and continuity away from zero. It has a continuous representative at zero exactly when , in which case . For , Fatou's lemma gives as , so no change on a set of zero Lebesgue measure can give continuity.
With the Fourier transform convention , the inverse Fourier transform is the displayed linear operator. The integral is defined directly for an integrable function . The Fourier inversion theorem ensures under its hypotheses, in particular for Schwartz functions, and for integrable and at continuity points of . The operation also extends to tempered distributions; the normalization and sign depend on the chosen Fourier transform convention.
With Fourier transform convention , differentiation under the integral gives , proving the displayed identity. For example it applies when all for are absolutely integrable, or in the tempered distribution sense. It is the dual counterpart of the Fourier transform of a derivative.
For the Fourier transform convention , Fourier inversion at gives . Thus transforming twice reflects the argument and multiplies by . The identity holds directly on the Schwartz space, and extends to the L2 space and to tempered distributions. It lets an already known transform pair be used in reverse without another integration.
For on and zero outside, the angular-frequency Fourier transform follows by writing the cosine as two complex exponentials. Combining fractions gives . The apparent singularities at are removable, with value . If , jumps at the endpoints give an oscillatory envelope. If , the function is continuous at the endpoints but its first derivative jumps, giving an envelope. This is an explicit instance of Fourier decay from a derivative jump.
For real , integrate around a rectangle of height . Its top-edge value is times the bottom-edge value, and the only enclosed simple pole is , with residue . The vertical-edge integrals vanish as their real parts tend to positive and negative infinity. The residue theorem gives , proving the stated Fourier transform identity. This is an example of a function reproducing its shape under a Fourier transform, up to rescaling.
If is a real-valued function and an integrable function, its Fourier transform has Hermitian symmetry: . The value at negative frequency is the complex conjugate of the value at positive frequency.
A window function is a function that vanishes outside a chosen finite interval; multiplying an input by it selects and weights that part of the history before a Fourier transform. The rectangular window is the indicator function of . Applied to the weighted causal extension by zero of an input, it produces the finite-history integral used to approximate a Laplace transform.
Scalloping loss is the underestimate of an isolated sinusoidal amplitude caused by measuring it only at the nearest sampled discrete Fourier transform frequency, after correcting the window’s coherent amplitude scale. An off-grid tone is evaluated away from the maximum of the window response. This differs from the coherent gain of a window, which describes the response at the correctly centered frequency. The instrument manual treats amplitude scaling and off-bin errors separately.[ref]
Let be a finite real window with positive sum, and let , where is the nearest unpadded DFT bin and . Write . The coherent-gain-corrected bin magnitude relative to isHere is a positive loss when . The half-bin amplitude response of a window evaluates this ratio at the largest nearest-bin offset. Other tones, noise and the negative-frequency image of a real sinusoid can also affect a measured peak; the derivation isolates one complex tone.
For a rectangular window, its geometric-sum transform givesFor , the half-bin ratio is about and the loss about . An amplitude-one tone exactly centered on a bin is recovered as one, while the same tone shifted by half a bin gives only about at either nearest bin. Dividing by coherent gain cannot remove this offset dependence.
For the periodic Hann window, usewhere is the rectangular response. At , this gives a half-bin ratio about , or loss. The smaller nearest-bin error comes with a wider main lobe and a larger equivalent noise bandwidth; window choice affects several measurement properties at once.
Zero padding reduces the nearest-grid offset without changing the measured record. Padding to points makes the nearest offset at most in original-bin units. With and , the rectangular response at that worst offset is about , a loss of only . This is denser sampling of the same discrete-time Fourier transform, not extra physical frequency resolution. A fitted peak can reduce a grid error too, but its signal model and noise sensitivity must be stated.
A proposed analyzer test holds tone amplitude fixed while sweeping frequency through one bin interval, retaining record length, sampling rate and window. Divide out coherent gain, plot the measured peak versus fractional-bin offset, and compare with the exact response above. Include phase and nearby-tone checks for a real signal. Summing spectral powers instead answers an energy question and requires its own bandwidth normalization; it should not silently replace this single-bin amplitude definition.
The half-bin amplitude response of a window is the magnitude response to an isolated tone halfway between neighboring discrete Fourier transform frequencies, divided by the response to a bin-centered tone of the same amplitude. It quantifies a particular off-grid amplitude error after correcting the coherent gain of a window. For coefficients with nonzero sum, defineA loss convention reports in decibels. The primary instrument manual tabulates approximately for rectangular weighting and for Hanning weighting.[ref]
To derive the ratio, take a complex tone . Evaluating the windowed transform at bin gives . For real window coefficients, the responses at opposite offsets have equal magnitude. Dividing by removes the on-bin scaling and leaves . This calculation isolates the frequency-grid effect from noise and other spectral components.
For a rectangular window, summing the geometric series yieldsFor , the ratio is about . The large- value is about , explaining the familiar scalloping loss rather than an exact finite-record constant.
For the periodic Hann window, . Its finite response iswhere is the rectangular response and for . Substitution gives about at . A useful limiting derivation replaces the normalized sum by its integral:with continuous limiting values at . At its magnitude is . For a short periodic window, the coefficients instead give . The finite definition avoids treating the asymptotic value as exact.
The symmetric Hann window uses denominator , not ; it has a different finite response and coefficient sum. SciPy explicitly distinguishes the symmetric filter-design convention from the periodic spectral convention.[ref] The exact half-bin ratio should therefore use the coefficients actually applied.
For an original acquisition example, a -sample record at has bin spacing. A complex tone at lies halfway between bins 10 and 11. An amplitude-one tone gives a corrected nearest-bin magnitude about with rectangular weighting or with periodic Hann weighting. Zero padding changes the sampled grid of the same discrete-time Fourier transform; if the padded grid contains , its value there has no half-bin error relative to that tone.
The half-bin response is not automatically a universal worst-case bound for every conceivable window. Nor does the isolated-complex-tone calculation cover every real-sinusoid measurement: near zero frequency or the sampling limit, the mirrored spectral component can overlap and make the observed magnitude phase-dependent. A proposed test sweeps a constant-amplitude tone through a bin interval, varies phase for real signals, and compares the complete finite-sum response before quoting a single off-bin accuracy number.
The equivalent noise bandwidth of a finite window is the width of an ideal rectangular frequency response that passes the same white-noise power when both responses have the same peak gain. It quantifies noise collected by a spectral amplitude estimate, rather than the separation needed to distinguish two tones. For real coefficients with nonzero sum and sampling frequency , is in hertz; is in bins of an -sample record. The primary software documentation gives as the exact conversion factor between its squared spectrum and power-spectral-density scalings.[ref]
An original variance derivation explains the formula. Let independent zero-mean sample noise have variance , and form the complex amplitude estimate at frequency ,Cross terms vanish by the independence and zero means, soThe flat two-sided input noise spectral density is . Integrating it over the peak-normalized window response gives the same result by Parseval identity. A one-sided real-signal convention folds negative-frequency power into positive frequencies; factors of two must be handled consistently, rather than changing the window’s bandwidth ratio.
For a rectangular window, and , giving . For the periodic Hann window with , the cosine sums give and , hence . These agree with the instrument manual’s tabulated values.[ref]
For an original numerical comparison, take and , so the original frequency grid spacing is . Rectangular and periodic Hann bandwidths are respectively and . With independent sample noise of standard deviation , the RMS magnitude of the zero-mean complex noise estimate is for rectangular weighting and about for Hann weighting. The latter variance is 1.5 times greater, despite Hann’s lower distant sidelobes. It is the RMS complex-error magnitude, not automatically the standard deviation of a positive spectral-magnitude display.
Zero padding leaves , the actual window coefficients and unchanged. If a 256-sample Hann record is padded to 1024 points, the display spacing becomes and the same bandwidth spans six displayed bins. Replacing by the padded FFT length in a formula while leaving the coefficient sum unchanged confuses these two bin units.
The noise covariance between windowed Fourier bins also matters when combining several spectral values: a taper can correlate neighboring estimates. Bandwidth alone does not specify their joint statistics, and colored noise requires weighting its actual spectral density. A proposed check repeats the same estimator on independent synthetic noise records with known variance, then on measured noise with a characterized acquisition response, comparing empirical mean squared complex error with the formula. Report window convention, detrending, one-sided/two-sided scaling and record duration so the test can be reproduced.
Noise covariance between windowed Fourier bins describes the statistical dependence created when several spectral estimates use the same noisy record. A window changes not only each estimate’s equivalent noise bandwidth, but also the overlap of their frequency responses. Uncorrelated time samples therefore need not produce uncorrelated spectral bins after tapering.
Let independent zero-mean sample errors satisfy . Use real coefficients , nonzero sum , and coherent-amplitude-normalized estimates on the original -point grid,Expanding their complex covariance and removing the vanishing cross terms givesThus the covariance is the Fourier transform of the squared window, evaluated at the bin difference. Its diagonal agrees with the equivalent noise bandwidth variance formula. SciPy’s spectrum/density scaling likewise uses the coefficient-square sum separately from the squared coherent sum.[ref]
For a rectangular window, the exponential orthogonality sum gives on this unpadded grid. This is a statement about covariance; independence additionally follows for an appropriate Gaussian noise model, not for every possible input distribution.
For the periodic Hann window, with ,Its original-grid covariance has only five nonzero cyclic diagonals:with bin indices modulo . Normalizing by the common diagonal gives adjacent-bin covariance ratio and second-neighbor ratio . The negative sign follows from this window’s placement in the record; changing the time placement can change covariance phases without changing their magnitudes. The periodic and symmetric coefficient conventions must not be interchanged in this finite calculation.[ref]
SciPy 1.16.2 documentation. scipy.signal.windows.hann, sym parameter and defining raised-cosine coefficients
Noise covariance between windowed Fourier bins for rectangular and periodic Hann windows
. Exact original N=16 grid covariance matrices for independent equal-variance sample errors. Each matrix is divided by its own diagonal variance, so the plot compares correlation structure rather than absolute noise level. The corner bands arise from cyclic bin indexing. No sampled experimental data or random simulation is used. Details.Correlation of neighboring periodogram powers is different from the sign of this complex covariance. Suppose each input sample has independent Gaussian real and imaginary parts, each of variance , with all components independent across time. The nonconjugated second moments vanish. Gaussian fourth-moment pairing, as in Isserlis's theorem, then givesConsequently neighboring Hann-bin powers are positively correlated despite the negative covariance of their complex amplitudes. If , their individual power variances are , and the variance of their arithmetic mean isTreating them as independent would incorrectly give . For real Gaussian input, a further term enters the power covariance; the pseudo-covariance of a complex random variable and conjugate-frequency symmetry therefore need attention near paired bins and the real-valued endpoint bins.
Zero padding creates more closely spaced evaluations of the same record, not more independent noise realizations. Even a rectangular record generally has correlated estimates on that denser grid. Detrending, colored input noise and overlapping records change the covariance calculation too. SciPy exposes detrending and FFT-length choices explicitly; its default constant detrending must be disabled when testing the unmodified independent-sample model above.[ref]
A proposed numerical check generates many independent records with the stated noise distribution, computes the estimates with fixed coefficients, and compares empirical complex covariance and power covariance separately with the formulas. A physical noise test additionally needs a characterized acquisition bandwidth and an assessment of time-sample correlation. Averaging several correlated spectral values does not supply the same precision gain as averaging independent records.
Correlation of neighboring periodogram powers describes dependence between squared spectral magnitudes estimated from the same random record. Their dependence is not determined by the sign of amplitude covariance. A negative correlation between complex Fourier amplitudes can produce a positive correlation between their powers, changing the precision obtained by averaging nearby spectral values.
Let be zero-mean jointly Gaussian complex Fourier estimates. Define their complex covariance and nonconjugated second moment byand put . Applying Isserlis's theorem to the four factors gives three pairings:Subtracting the product of the means therefore yieldsIn particular, Gaussian power covariance is nonnegative even when a complex amplitude covariance is negative. Also ; ignoring the pseudo-covariance of a complex random variable changes the answer when that term is present.
Consider independent sample noise with independent Gaussian real and imaginary components of equal variance. Its nonconjugated moments vanish, so after any deterministic complex linear transform. If the normalized bin variances are all , thenFor the periodic Hann window with on the original -point Fourier grid, noise covariance between windowed Fourier bins gives amplitude covariance ratios at separation one and at separation two. The corresponding power correlations are and . More distant nonconjugate bins have zero covariance in that particular model.
An original calculation averages consecutive powers without wrapping around the Fourier grid. Assume so that the block does not acquire additional cyclic neighbor pairs. Counting the adjacent pairs and second-neighbor pairs givesFor , the variance is , rather than the independent-value result . For , it is , rather than . A variance-equivalent independent-power count is the reciprocal of this normalized variance: about for those ten bins. For , the bracket simplifies to , approaching roughly independent powers for large blocks in sufficiently long records.
Correlated periodic Hann window powers give less variance reduction than independent spectral powers
. Exact finite-block variance and variance-equivalent independent-power count for Gaussian complex white sample noise on the original N=64 grid. Blocks contain one to 32 consecutive bins, so the plotted calculation has no extra wraparound pairs. Each variance is normalized by its model’s single-power variance. The lines are analytical predictions, not experimental observations. Details.Real-valued Gaussian input has conjugate Fourier symmetry and can have nonzero , especially for mirrored frequencies and endpoint bins. A deterministic spectral component also invalidates the zero-mean calculation unless its additional mean-dependent terms are included. Colored noise, taper convention and detrending change and ; the formulas above do not make all measured spectra follow one correlation pattern.
A periodogram may be expressed as a squared spectrum or a power spectral density. SciPy documents their window-dependent conversion, one-sided folding and optional detrending explicitly.[ref]
SciPy 1.16.2 documentation. scipy.signal.periodogram, scaling, return_onesided and detrend parameters; Notes
A common positive scaling does not change the correlation coefficient, but combining unequal scalings or folding conjugate values requires carrying those transformations through the covariance.A proposed check generates independent records under the stated noise model, calculates bin powers with the same window, and compares empirical pair covariances and block-average variance with the predictions. Independent records provide a different averaging resource from additional bins or zero-padded frequency points in one record. Report which values were averaged and their covariance before assigning an independent-sample precision gain.
A rectangular window keeps all samples in a selected finite record with equal weight and sets the signal outside it to zero. For discrete samples, for and zero otherwise. Calling this no taper does not remove the effect of having a finite record. SciPy's primary implementation documentation identifies the boxcar and rectangular window as this same all-ones choice.[ref]
Its discrete-time Fourier transform at frequency in cycles per sample is a finite geometric sum:At integer , the limiting value is . The first zeros surrounding the central maximum are at , so the central zero-to-zero width is . The phase factor depends on the record's time origin; the magnitude response does not.
For a continuous centered record of duration , on has transformwhere the normalized sinc function is . Multiplication by the finite window spreads a sharp spectral component according to this response, producing spectral leakage.
An original finite-record example quantifies an off-bin amplitude error. A complex tone of amplitude lying half a DFT bin from the nearest sampled frequency has normalized responseAt the ratio is about , or relative to the coherent peak. The coherent gain of a window is still one; it does not correct this scalloping loss. Zero-padding samples the same frequency response more densely rather than lengthening the measured record.
For independent equal-variance sample noise, rectangular weighting has equivalent noise bandwidth of one DFT bin under the usual coherent-amplitude normalization. A Hann window suppresses distant sidelobes but changes the central width and noise scaling. The primary analysis manual reports these separate window properties rather than one universal best window.[ref]
A proposed test records a known sinusoidal input with an integer and then a non-integer number of cycles in the same record length, holding the sampling rate and amplitude fixed. Compare the raw DFT bins and the continuous window response. This separates finite-record leakage and bin placement from aliasing, which involves a different sampling ambiguity.
A Hann window tapers a record with a raised cosine. For the endpoint-symmetric convention with ,It vanishes at the two stored endpoints. A periodic convention instead uses denominator , corresponding to sampling one cosine period without storing its repeated endpoint. SciPy's primary documentation distinguishes these with
sym=True and sym=False; the default is symmetric.[ref] The adjacent Hanning window alias also supports the terminology used in older instrument documentation.An original short-array example makes the difference concrete. At , the symmetric coefficients are , while the periodic coefficients are . Their means are and . A symmetric even-length array need not contain a coefficient equal to one, despite the nominal raised-cosine peak.
The coherent gain of a window is . It equals for the periodic Hann convention with , and for the symmetric convention with . A coherent complex tone gives windowed DFT magnitude at its exact bin, so amplitude recovery must use the chosen coefficient sum.
For the periodic convention, its discrete-time Fourier transform is obtained from the rectangular window transform by modulation:The first central zeros lie at for the usual case, twice as far from the center as the rectangular window's first zeros. Smooth tapering reduces distant sidelobes while widening the central feature. Nearby frequencies and a weak component far from a strong one can therefore favor different window choices.
Noise normalization is another distinction. For , the periodic Hann coefficients obey and , giving equivalent noise bandwidthin DFT bins. The manufacturer's table lists both coherent gain and noise bandwidth.[ref] Correcting a coherent amplitude does not make noise behave as though the rectangular weights had been used.
The original comparison below evaluates the finite 64-sample periodic Hann and rectangular coefficients at a dense frequency grid, normalizing each response by its own coefficient sum. The extra frequency samples display the same finite-record transform more finely; they do not create a longer observation.
Original 64-sample rectangular and periodic Hann windows: lower distant sidelobes accompany a wider central lobe
. Each spectral curve is normalized to its own coherent peak. The horizontal frequency axis is measured in DFT-bin offsets. The first positive zeros are at one bin for the rectangular window and two bins for the periodic Hann window. Details.The coherent gain of a discrete window function is its mean coefficient:It describes the amplitude scaling of a coherent complex tone under the chosen DFT normalization. Window coefficients and their normalization must be specified; a window's name alone does not fix every finite-length convention.[ref]
LabVIEW Analysis Concepts by National Instruments. Printed pages 5-20–5-24, periodic/symmetric windows and coherent-gain scaling
For , multiply by and evaluate the unnormalized discrete Fourier transform at the exact tone bin:Thus recovers when . This result is exact for this complex-tone model. A real cosine also has a negative-frequency component; the usual doubled one-sided amplitude requires that its contribution to the chosen positive-frequency bin be absent or separately accounted for.
A rectangular window has . The periodic Hann window has for . Its coherent bin therefore has half the unwindowed amplitude before correction. The symmetric endpoint convention , for , instead gives . At this is , so blindly multiplying by two is not the exact correction for that convention.
Coherent gain does not correct every amplitude error. An off-bin tone samples the window's frequency response away from its maximum, and a DC-removing window with zero sum has no usable gain correction at all. Coherent sampling, leakage and window normalization are distinct conditions.
Noise uses a different scaling. For independent equal-variance samples, a weighted DFT bin's noise variance is proportional to , not . This motivates equivalent noise bandwidth in bins,which is for a periodic Hann window with . The manufacturer's table lists both coherent gain and noise bandwidth rather than treating one as a substitute for the other.[ref]
An image mask is a function on image coordinates specifying which pixels or regions are retained or weighted. A binary mask selects a region, while a soft mask takes intermediate weights. Applying the mask pointwise gives for image intensity .
For example, a mask equal to one inside a disk and zero outside retains only that disk of an image. An edge with fractional weights can smooth the transition. The mask describes location or weighting, rather than assigning a new color to every retained pixel.
In Fourier analysis, multiplication by a window function in the image plane corresponds to convolution in frequency, with the constant depending on the Fourier convention. A sharp binary boundary can therefore introduce broad high-frequency contributions. This explains why cropping and spectral filtering have different mathematical effects.
Multiplying a measured signal by a finite-duration window function convolves its Fourier transform with the transform of the window. A spectrally sharp component therefore contributes to nearby frequencies rather than remaining a single ideal line. Smooth tapering can reduce distant sidelobes at the cost of widening the main spectral feature. This finite-record effect is distinct from aliasing, which superposes copies of the spectrum because of time sampling.
For nonzero with , integration by parts gives . The Cauchy-Schwarz inequality and Plancherel theorem yield the moment product bound with the angular-frequency Fourier convention. Equality occurs exactly for with real and . Translation, modulation or a quadratic phase increase one of these uncentred moments; the equality family is narrower than that for centred variances.
For the displayed integral defines a bounded continuous function on the Pontryagin dual group. It takes convolution to pointwise multiplication and left translation to multiplication by . The transform depends on the scale of Haar measure on ; the compatible dual Haar measure scales reciprocally.
For the dual Haar measure, the formula first follows for continuous compactly supported convolution squares from the Bochner theorem. More generally, if and , the right-hand side is a continuous representative of ; equality holds everywhere if is continuous, and otherwise at every point where the chosen representative is continuous. Convolution with normalized small-support positive convolution squares proves the extension: the spatial convolutions approach in , while their transforms approach in by equicontinuity of compact families of characters and an integrable-tail split.
For a piecewise smooth decaying function, a jump in its th derivative at contributes the displayed algebraic term to its high-frequency Fourier transform. Distributionally, contains , and integration by parts or the derivative-transform identity supplies the factor . Adequate integrability of the subsequent derivatives controls the remainder. Smooth localized polynomial terms give no algebraic cusp contribution.
For this formula is interpreted away from as a tempered distribution or by exponential regularization. It also supplies local high-frequency cusp terms after multiplying by a smooth cutoff. Even nonnegative integer give smooth polynomials and only contact terms at , so their localized forms have no algebraic tail. For example, the cusp expansion of gives .
Multiplication by a unit-modulus linear phase shifts the Fourier transform. Substituting into the defining integral gives the displayed identity. For a surface measure this permits independent translations of cap transforms without changing their input norms or angular supports.
The angular-frequency Fourier transform of a finite measure, or a complex measure of finite total variation, is defined by the displayed integral. It is bounded by total variation and continuous by dominated convergence. An integrable density recovers the ordinary Fourier transform of a function. A measure supported on a curved surface instead leads to a Fourier extension operator.
For a finite complex measure on the additive circle group , its Fourier coefficients are integrals against the group characters , . They obey , where the norm is the total variation norm of a measure. A probability measure has zeroth coefficient one. The normalized Haar measure has all nonzero-frequency coefficients zero. A bounded family of finite Radon measures converges in the weak-star topology exactly when its coefficients converge to those of the proposed limit: extend linear combinations to all continuous tests using the density of trigonometric polynomials and the common variation bound.
A complex spectral Fourier transform parametrizes the two-dimensional Fourier transform by complex coordinates. Put , andSince , the real frequency vector is . Its Jacobian is four, givingBoth constants follow from the declared angular-frequency convention.
For smooth compactly supported , a specified Cauchy-kernel solution of isThe identity verifies the equation in the distribution sense. Its local singularity is integrable. Differentiation in cancels and yieldsThis is the link with a spectral dbar equation. It uses the displayed normalized Cauchy transform solution; the spatial differential equation alone leaves homogeneous solutions and does not force this spectral derivative for an arbitrary choice of .
A bandlimited function has its Fourier transform supported in a bounded frequency region. For with support of contained in , the transform is bounded and hence integrable. Fourier inversion supplies the bounded continuous representativeThe same formula for complex defines an entire function: derivatives can pass under the finite-interval integral, uniformly on compact subsets. It obeys .
For , the exact scaling law isThus has bandwidth , while the stretched function has bandwidth . This fixes the otherwise ambiguous phrase that time and bandwidth scale inversely. Products add frequency-support intervals through the convolution theorem.
Band limitation differs from compact spatial support. A nonzero entire representative cannot vanish on an open real interval by the identity theorem. Consequently a nonzero integrable bandlimited function cannot also have compact support. This matters when interpreting a sharply started signal or a wave-packet front.
With the convention, a right-supported decaying function has a transform analytic in the upper half-plane and a left-supported one in the lower half-plane. Incident modes continued with small imaginary frequency can have poles outside the initial convergence strip; these poles must be tracked separately from outgoing scattered waves.
For , the Half-range Fourier transform is holomorphic in the lower half-plane. Let , and , with their boundaries oriented counterclockwise. The rotated arguments on and on remain in the lower half-plane. For and transform-admissible smooth decay, the Cauchy integral theorem and Jordan lemma give the displayed null integrals. Arbitrary multiples can therefore be added to Fourier inversion without changing the recovered function on the positive half-line.
// Target: integrable-systems.bigb
For , let be the upper branch , oriented left to right. Above , lies in the lower half-plane, where a decaying half-line Fourier transform is analytic. For , closing upward proves that the displayed integral vanishes. An additional multiplier analytic above with at most polynomial growth is allowed, using exponential decay and, if needed, regularized contours. This supplies a null term in a Fokas method inversion formula without changing the function in the interior.
The Fourier transform with negative exponential and inverse factor is a continuous isomorphism of the Schwartz space. Weighted frequency derivatives are transforms of derivatives of weighted input functions, so their suprema are controlled by finitely many Schwartz seminorms. Fourier inversion gives , where . Transposition gives a continuous isomorphism on tempered distributions as well.
For the angular-frequency Fourier transform and Schwartz seminorms , every output seminorm is bounded by finitely many input seminorms. To prove this, differentiation under the integral sign and integration by parts giveThe generalized Leibniz rule, , and implyThis proves continuity of the Fourier transform on the Schwartz space and rapid decay of every transformed derivative.
For sufficiently regular decaying data, this half-line transform has inverse . To derive it, extend evenly to the real line. Its Fourier transform with negative exponential is , and its odd sine contribution vanishes. The Fourier inversion theorem then gives the stated factor . If the forward transform is instead defined as , the inverse factor is .
With boundary terms at infinity vanishing, two integrations by parts give the cosine transform of as . Thus a homogeneous Neumann boundary condition makes it a multiplier , paralleling the Fourier sine transform with a homogeneous Dirichlet boundary condition.
The Fourier sine transform is a half-line transform adapted to a homogeneous Dirichlet boundary condition. With the displayed normalization, inversion is . It is equivalent to the Fourier transform of an odd extension. If and boundary terms vanish, two integrations by parts give the transform of as .
For sufficiently regular integrable functions on a half-line, Fourier sine transform inversion recovers the odd extension at continuity points. A zero endpoint value is required when demanding continuity at the endpoint.
The unnormalized sinc function is , extended by value one at zero. The normalized sinc function is , with the same removable value. These arise, with convention-dependent scaling, as inverse Fourier transforms of frequency-interval indicator functions, so they describe ideal low-pass filters and the Shannon scaling function.
The normalized sinc function is away from zero and one at zero. The value at zero removes the apparent singularity, giving an entire function. Its nonzero integer zeros make it useful for Sampling Theory.
With the cycles-per-unit Fourier transform convention, the inverse transform of the unit-width rectangle isThe factor of is therefore tied to a frequency convention. The unnormalized sinc function uses instead, so the two names must be distinguished when comparing formulas.
The sinc function for , extended by value one at zero. The displayed power series has infinite radius of convergence, so the extension is entire. Its removable value follows from the first coefficient or from the sine limit. This convention differs by the argument scaling from the normalized sinc convention.
The triangular function on , zero elsewhere, has Fourier transform in the angular-frequency convention. Its squared L2 norm is . Parseval's identity therefore givesThis proves the displayed value. The removable value at zero is one, and the fourth-power decay bounds the tail. The sinc-squared integral follows similarly from a rectangular function rather than a triangle.
Fourier space labels a field by the frequency or wavevector variables of its Fourier transform, rather than by its original coordinates. Spatial gradients act by multiplication by , and the negative Laplacian acts by .
If and for , then its Fourier transform is times continuously differentiable and the displayed formula holds. The dominated convergence theorem permits each derivative under the integral. Thus vanishing moments give a zero Taylor polynomial through degree at the origin. Unweighted membership only guarantees a continuous function as the Fourier transform.
The Riemann-Lebesgue lemma and Fourier inversion give a bounded injective map with dense image. It is not onto: inverse transforms of have unbounded norms while the targets have uniform norm , contradicting the bounded inverse that the open mapping theorem would give.
The order- Hankel transform replaces the exponential kernel of a Fourier transform by a Bessel function:Under suitable integrability and inversion hypotheses, . The measures and belong to this convention and cannot be discarded.
For a radial function on the plane, angular integration gives when the Fourier transform kernel is . Thus rotational symmetry reduces a two-dimensional transform to a one-dimensional radial integral.
For the Fourier transform convention , angular integration of an isotropic power spectrum gives the displayed Hankel transform. A purely radial Hankel transform without the prefactor corresponds to defining its spectrum as .
For the Cartesian convention , an isotropic function has . Its inverse is . The self-reciprocal Hankel transform instead uses . In particular . Mixing these spectral conventions changes attenuation coefficients by .
The Wigner distribution of two fields is the Fourier transform of a field's two-point product in its separation variable:Its inverse reconstructs . For vector fields, specify whether the product is an outer product or a scalar contraction; the latter can encode an average over receiver channels.
For , with , normalize the phase-space function by . A coherent state of amplitude has . Quadrature marginals are physical probability distributions, but the full function can be negative. Cross terms of an optical cat state give interference fringes, even when low-order moments approach those of a mixture.
For , differentiate the two-point product . The difference of its two Laplacians is , so Fourier transformation in gives the displayed free-transport equation for the Wigner transform. Its solution is .
For receiver-dependent Green functions and aperture weight , define . Its inverse separation Fourier transform is the kernel of the time reversal operator on the source plane.
The Japanese bracket is a smooth positive frequency weight, comparable to . It defines Sobolev space norms and the growth estimates of a symbol class without singularities at zero frequency.
For real and vectors , the displayed bound follows from , applied in the opposite direction when . One can take . Combined with Young's convolution inequality, it proves Sobolev multiplication by a smooth cutoff for all real orders: the rapidly decreasing transform of the cutoff absorbs the extra weight. The negative-order case is particularly useful when a distribution is not initially a function.
For the angular-frequency convention,
The Fresnel integral is an oscillatory Gaussian integral. In the angular-frequency convention, its distributional form isfor nonzero real .
For the optical convention of Fresnel integrals, and . Both functions are odd and tend to as . If the unscaled convention uses phase , multiply its argument by and its value by to obtain this convention. An aperture factor with phase uses . Stating the convention prevents missing factors of or in diffraction calculations.
In the unscaled convention, the finite Fresnel integrals are and . Their Taylor series follow by termwise integration. At infinity both tend to , and their oscillatory tails have asymptotic expansions obtained by integration by parts. This convention differs from the convention with phase .
For , define real tail functions by . Then and . Repeated integration by parts giveswhere . These are asymptotic expansions, rather than convergent infinite sums at a fixed argument.
For , polar coordinates reduce the integral to . The result is . Its damped limit for real is , which fixes the sign of the oscillatory complex Gaussian integral. The Fresnel integral convention can be compared with NIST definitions.
An additive character of a locally compact field is a continuous group homomorphism from to the complex unit circle.
For a nontrivial additive character on a local field, the character annihilator of an additive subgroup is . If , then . This is a character pairing construction, distinct from the annihilator of a module.
For a local field , a nontrivial additive character , and a Haar measure , the Fourier transform isIt maps the Schwartz-Bruhat space to itself.
For a compact open additive subgroup of a non-Archimedean local field, translation and additive character orthogonality give the displayed transform of a coset indicator. With the self-dual Haar measure, transforming again gives . Every Schwartz-Bruhat function is a finite linear combination of these indicators.
Relative to a nontrivial additive character, the self-dual additive Haar measure makes the twice-applied Fourier transform over a local field equal to reflection. If the annihilator of the valuation ring is and its residue field has elements, the normalization is . Volumes of a compact open subgroup and its character annihilator then multiply to one.
For an additive character of a non-Archimedean local field ,For the character induced from the standard character of , this annihilator is the inverse different.
A Fourier multiplier is a function on frequency space used to define an operator by multiplication after taking a Fourier transform.
For a Fourier multiplier , define using a unitary Fourier transform on L2 space. Its operator norm is exactly the essential supremum of :The unitary normalization matters when writing the intermediate integrals. For example, has this normalization. The Plancherel theorem then turns the bound into multiplication of functions:The required extension of the Fourier transform to all of L2 space is the unitary extension in the Plancherel theorem.[ref]
To prove sharpness, write . For every , the measurable set where has positive Lebesgue measure. Choose a subset of finite positive Lebesgue measure and prescribe , using an indicator function. The inverse Fourier transform gives and . Letting tend to zero proves the equality of operator norms. The case is immediate.
For example, is an ideal low-pass filter: it is an orthogonal projection onto the signals with Fourier transform supported in that interval, and has operator norm one. In contrast, also has operator norm one on , but no nonzero signal attains equality: except on a set of Lebesgue measure zero. Changing a Fourier multiplier at one frequency cannot change this operator norm.
Suppose every continuous periodic function has a unique continuous periodic with Fourier coefficients . Uniqueness first makes linear. If and uniformly, continuity of each coefficient functional gives for every fixed , hence . Thus has a closed graph and the closed graph theorem makes it bounded. Applying it to normalized exponentials gives .
The half-plane Fourier projection is , where is the directional Hilbert transform. A disk of radius , modulated to put a boundary point at the origin, has symbol , which converges almost everywhere to this half-plane symbol. The Plancherel theorem makes this an multiplier limit.
Using the cycles-per-unit inverse convention , the inverse Fourier transform of in three dimensions, with , is . Its leading oscillatory tail is not absolutely integrable against the radial volume element . Multiplying by a Gaussian Fourier transform retains a nonzero boundary-derivative tail, so the radial cutoff need not map even Schwartz functions into . A tensor product of one-dimensional triangular cutoffs is a valid approximate identity, and must not be confused with this Euclidean radial cutoff.
A Fourier multiplier operator is defined by . On , a bounded symbol gives a bounded operator by the Plancherel theorem, while an unbounded symbol defines a closed linear operator on the functions for which remains square-integrable.
Suppose is locally integrable andThen has finite limits at zero from both sides. With the convention , the positive-frequency limit minus the negative-frequency limit isSubtracting suitable reflected copies of leaves an function, whose Fourier transform is continuous; the jump then follows from the Dirichlet integral.
If a nontrivial uniform estimateholds for all integrable functions on , dilation by every forcesThus must be the conjugate exponent of .
For the angular-frequency convention , the inverse-distance tempered distribution on has the exact transformBoth sides are locally integrable in three dimensions and have at most polynomial growth, so they define distributions even though the original function is not integrable at infinity.
One direct proof introduces , with . Angular integration of the radial Fourier transform gives, for ,The damped sine integral follows by taking the imaginary part of . Letting against a Schwartz function is justified by the locally integrable majorants and together with rapid decay. This proves the distributional identity with its constant.
The result agrees with : the Fourier transform of a derivative multiplies the transformed potential by . Changing to a cycles-per-unit convention changes the numerical coefficient, so the transform convention is part of the statement.
With the sign conventionthe Hilbert transform is a singular integral operator. For a Schwartz function, the Cauchy principal value removes equal small intervals on both sides of . Subtracting locally makes the remaining quotient bounded there, while rapid decay controls infinity. Reversing the denominator reverses the operator sign.
For the angular-frequency Fourier transform, its Fourier multiplier is , as derived under Hilbert-transform Fourier multiplier. Plancherel theorem therefore extends it to an isometry on , and on that space because the squared multiplier is minus one almost everywhere. This frequency-dependent phase shift connects real and imaginary boundary parts of suitable holomorphic functions; Hilbert transform of the sinc function supplies an explicit example.
The transform need not preserve boundedness. For the indicator function of , direct integration givesIt diverges logarithmically near the endpoints and has tail . This example also explains why a compactly supported input can produce a long-range response. The finite Hilbert transform instead restricts the integration interval and has its own endpoint and null-space issues.
The finite Hilbert transform is a singular integral operator whose integration is restricted to a finite interval. The displayed convention uses the denominator ; reversing that denominator reverses the sign. The Cauchy principal value deletes equal intervals on the two sides of the interior point . Endpoint behavior matters: the integrable function belongs to the null space for this interval. This nontrivial null space is one reason that an inversion formula needs a normalization or a stated endpoint class.
For , the displayed expression solves , where and is the finite Hilbert transform with denominator . To derive it, put at infinity, take to be the Cauchy transform of , and set . The Sokhotski–Plemelj formula gives and the displayed jump . The endpoint behavior is at most inverse square root, and at infinity. Cauchy reconstruction from a jump across a finite interval then gives and hence . Adding any constant multiple of gives another solution in the integrable endpoint class.
Multiplying an auxiliary sign-changing factor by a growing analytic factor can invalidate this reconstruction. For example, using instead of with produces the constant function and zero jump. The boundary sum alone is not enough: the entire remainder in reconstructing from its jump must also be controlled.
With convention , the analytic function has removable singularity at zero and decays on large upper semicircles. Its boundary real part is and imaginary part is . The upper-half-plane contour indented above gives . Taking real parts after reversing the denominator proves the displayed transform, with value zero at by continuous extension.
For a unit vector , the directional Hilbert transform integrates along lines parallel to . Its Fourier multiplier is under the usual transform convention. Applied to a rectangle whose long axis is , it reduces to the one-dimensional logarithmic tail of an interval indicator function, separately on each transverse coordinate.
For a Schwartz function , the Hilbert transform satisfies at both infinities. Splitting the inverse Fourier transform at the sign multiplier's jump and integrating each half-line by parts extracts the zero-frequency value ; the remainder vanishes by the Riemann-Lebesgue lemma. A nonzero zeroth moment therefore obstructs Schwartz decay.
For and , the Fourier transform multiplier is . The Heaviside function transform and the double-transform reflection identity give the sign. Plancherel theorem proves the isometry, and multiplication by proves commutation with differentiation. The map need not preserve the Schwartz space.
Articles by others on the same topic
The Fourier transform is a mathematical operation that transforms a function of time (or space) into a function of frequency. It is a fundamental tool in both applied mathematics and engineering, primarily used for analyzing and processing signals.
Continuous version of the Fourier series.
Can be used to represent functions that are not periodic: math.stackexchange.com/questions/221137/what-is-the-difference-between-fourier-series-and-fourier-transformation while the Fourier series is only for periodic functions.
Therefore, the Fourier transform can be seen as a generalization of the Fourier series that can also decompose functions defined on the entire real line.
As a more concrete example, just like the Fourier series is how you solve the heat equation on a line segment with Dirichlet boundary conditions as shown at: Section "Solving partial differential equations with the Fourier series", the Fourier transform is what you need to solve the problem when the domain is the entire real line.


