Talk:Fourier transform

Clarifying the contents

Hi all. I came here looking for a transform pairs table because I'm studying communication systems and ended up wondering why some Fourier transforms differed from the ones that I have tabulated in my textbook. I didn't understand why the 2pi factor was under a square root!! Anyway, I freaked out, googled a lot and finally read the article (it is kinda intimidating) finding out the answer my self (good job!). What I would like to notice is that it takes a while to locate and understand the 3 different conventions thing, and I think that it could be more simple if we give the reader a big picture wise look of that in the "contents". For instance, putting it like:
1 Definitions
1.1 Communications and signal processing (f) convention
1.2 Mathematics angular frequency (2pi) convention
1.3 Mathematics angular frequency (2pi unitary) convention

Finally, like some textbooks do, It would be really simplifying to say "in the rest of this article, we will use convention {whatever} unless stated otherwise". And stick to that in order to preserve consistency.

I would also like to say that I agree with Feraudyh (disgruntled editor) that this article is really hard to read for students like me, it seems really different from what we have on the communications textbooks. I understand that is a mathematical thing and it should be the most formal possible, but still I guess it could be more readable.--Dhcpy 01:41, 1 December 2007 (UTC)

The only way it could take "a while" to locate the 3 conventions is if you read the article haphazardly instead of linearly. I think you have to take some accountability for that yourself. I understand and sympathize with the suggestion to stick to one convention, like a textbook. That would probably serve beginners the best. But non-beginners would rather see their own pet convention presented along with the others, than not see it at all. So it's hard to please everybody.

--Bob K 22:50, 1 December 2007 (UTC)

I also came to this page (a while back) looking for transform pairs, and probably also skipped the first section (Definitions). However, since most of the results are (very helpfully) given for each convention, it seems pretty unambiguous. Is there a key result which is stated in only one version? --catslash 00:01, 2 December 2007 (UTC)

Properties?

Why do we have a section on "# 4 Properties" and a section "# 6 Fourier transform properties". Maybe we should merge #6 into #4. Thoughts? Thenub314 00:28, 22 February 2007 (UTC)

I don't think there is a good reason for the separation. Somebody just one day decided to add "# 6 Fourier transform properties" as a stand-alone section. Maybe they were trying to stay out of the first person's space?

--Bob K 04:30, 22 February 2007 (UTC)

involution claim is wrong

Bo says: "By first conjugating the function, the forward transform equals the reverse transform, a so-called involution". But the forward transform produces ${\overline {X_{3}(f)}}\,$ and the reverse transform operates of $X_{3}(f)\,.$ In order for Bo's claim to be true, the reverse transform would have to operate on the same thing produced by the forward transform. I see no value in this involution idea. It's just smoke and mirrors.

--Bob K 01:25, 8 March 2007 (UTC)

I think what Bo intends to say is, ${\mathcal {F}}\{x\}(\omega )={\overline {{\mathcal {F}}^{-1}\{{\overline {x}}\}(\omega )}}$ but that is not an involution; so I removed the current statement. The correct statement would be something like: "by first conjugating the function and then conjugating the forward transform, the forward transform equals the inverse transform, a so-called involution", which makes my head spin :-) (it is always tough to "do" math in words").

I wonder though, if this property is worth mentioning on the page ... Abecedare 02:08, 8 March 2007 (UTC)

involution claim is correct

Bo, Ignore the above comments : both Bob and I realized too late what you meant. :-) Abecedare 02:20, 8 March 2007 (UTC)

Right. This transform:

$\int _{-\infty }^{\infty }[\cdot ]^{*}\ e^{i2\pi ft}\ df$

is the inverse of this one:

$\int _{-\infty }^{\infty }[\cdot ]^{*}\ e^{i2\pi ft}\ dt$

--Bob K 02:24, 8 March 2007 (UTC)

Exactly! Sorry for all my recent reversions on the article and talk page. I have tried to rewrite the word statement to make it clear that the Fourier transform can be deined in a form that is an involution. Please check if the point is now clear. Also I reverted to using the "overline" convention for conjugation, since it is used in the rest of the page and more importantly * is used for the adjoint operator (and AFAIK, there is no other "standard" notation for the latter). Regards. Abecedare 02:28, 8 March 2007 (UTC)

involution claim is original research

I've removed the "involution" definition, which is non-standard (not to mention non-linear). Bo, if you think this form should be included, please provide a reference to a published source that uses this definition. The question for Wikipedia is not whether the transform can be defined in such a fashion, it is whether it is defined in that fashion. Wikipedia is not the appropriate venue for notational reforms. Steven G. Johnson 05:00, 8 March 2007 (UTC)

No mathematical definition is standard, as no standardization organizations standardize mathematical definitions. The transformation is obviously a conjugate-linear map. That it is also an involution was just checked by editors Abecedare and Bob K. This is sufficient documentation for correctness. Steven G. Johnson's removal of the definition is a violation of Wikipedia:Resolving disputes # Avoidance. The value of the definition is that it avoids the risk of confusing forward and reverse transformation. That's why I want to share it. I'll leave it to Abecedare and Bob K to reinstall it if they like it too. Bo Jacoby 06:53, 8 March 2007 (UTC).

I partially agree with both of you. (1) the "definition" is antilinear but correct in that if the Fourier transform was so defined it would be involution, (2) it is non-standard, i.e. it is not covered in mainstream Fourier transform texts that I have seen, although obviously the property ${\mathcal {F}}\{x\}(\omega )={\overline {{\mathcal {F}}^{-1}\{{\overline {x}}\}(\omega )}}$ on which the "definition" is based is both well-known and trivially derived.

So Bo Jacoby, can you please provide some reference(s) in which the $\int _{-\infty }^{\infty }[\cdot ]^{*}\ e^{i2\pi ft}\ dt$ is offered as a definition of the Fourier transform? ... else including it would be a violation of wikipedia's guidelines against original research. Regards. Abecedare 07:23, 8 March 2007 (UTC)

Thank you, Abecedare. It is not the definition, but it is a useful variant form. I agree that the nice properties are well-known and trivially derived, so there is no question about reliability, which is the reason behind the general ban against original research in wikipedia. It may be considered original research on my part, but not on yours, so I cannot reinstall it, but you can. Just refer to me. Bo Jacoby 08:05, 8 March 2007 (UTC).

Sorry Bo Jacoby, can't do that, since as per WP:RS we need well-regarded published sources not only to vouchsafe the correctness of the "definition" (which is not really the issue here), but also its mainstreamness (to invent a word, since this is only talk-space). And given that 100's of books and 1000s of papers (very conservative estimate) have been written on Fourier transforms, we will need not just isolated references, but sufficiently "gold-plated" ones (example, text-books used in top-ranked university courses, or review articles on Fourier transforms written in prominent journals) before we can add this as a definition. Unless those are forthcoming, there is nothing I or any other editor can do about this. Regards. Abecedare 08:24, 8 March 2007 (UTC)

I am grateful that my humble observation is honored by the name of original research. I approve Bob K's formulation: "a non-standard type of Fourier transform that is an involution, which means that the inverse transform is the same operation as the forward transform". This does not make any promises of gold-plated mainstream references but gave a straightforward and clear explanation. The corresponding FFT is foolproof because only one procedure is needed for transforming forwards and backwards. For real x(t) it coincides with one of the other definitions. Too bad that only very few people got the chance to read before removal. I can do no more about it. Any of you are welcome to undo the edit "04:56, 8 March 2007 Stevenj" when you realize that our readers deserve the best. Bo Jacoby 13:53, 8 March 2007 (UTC).

There is a standard involution on the Banach algebra L¹(G, Haar) for any locally compact group G. For G abelian the involution is f*(x) = complex congugate(f(-x)), quite different from your definition. The operation you define isn't even an involution operation on the Banach space L¹ (G) for non-finite G. It will be very confusing to overload the term "involution" by calling your transformation an involution also.--CSTAR 15:06, 8 March 2007 (UTC)

Bo, I don't know if your transform is an involution or not. I agree that it is kind of cool and interesting, but I don't see how it furthers the cause of this article, which is to help people understand the standard Fourier transform(s). In fact, I think it is counterproductive. I wish there were fewer forms, not more. If indeed yours is an involution, then why not add your transform to that article?

--Bob K 16:00, 8 March 2007 (UTC)

See the definition in involution is simply: "a function that is its own inverse, so that f(f(x)) = x for all x in the domain of f". Bob K checked that the variant 4 of the fourier transform satisfies this condition, just above. The forward and backward transformation is the same. There are many involutions: the identity, the sign change, the reciprocal, the complex conjugation, &c. Bo Jacoby 16:33, 8 March 2007 (UTC).

Well no, I did not say that $f\,$ and $t\,$ are in the same domain.

And $\int _{-\infty }^{\infty }[\cdot ]^{*}\ e^{i2\pi ft}\ dt$ is not a function of a variable. Rather it is a transform of a function.

And if that's not enough, the transform is not its own inverse, because the inverse substitutes $df\,$ for $dt\,$.

--Bob K 18:12, 8 March 2007 (UTC)

The variables f and t are not in the same domain, but the functions X and x are in the same domain. The fourier transformation transforms x into X. The transform maps the function x of time to the function that maps the frequency f to the integral $\int _{-\infty }^{\infty }{\overline {x(t)}}\ e^{i2\pi ft}dt$

$x\mapsto \left(f\mapsto \int _{-\infty }^{\infty }{\overline {x(t)}}\cdot \ e^{i\cdot 2\cdot \pi \cdot f\cdot t}\cdot dt\right)$

In this expression both t and f are dummy variables. Other variable names can be substituted for t and f. They may even be swapped:

$x\mapsto \left(t\mapsto \int _{-\infty }^{\infty }{\overline {x(f)}}\cdot \ e^{i\cdot 2\cdot \pi \cdot t\cdot f}\cdot df\right)$

Using $f\cdot t=t\cdot f$ you get

$x\mapsto \left(t\mapsto \int _{-\infty }^{\infty }{\overline {x(f)}}\cdot \ e^{i\cdot 2\cdot \pi \cdot f\cdot t}\cdot df\right)$

So the transformation that transforms a function of t into a function of f also transforms a function of f into a function of t. Bo Jacoby 22:38, 8 March 2007 (UTC).

But transforms are not functions. So your transform cannot be "a function that is its own inverse". --Bob K 23:50, 8 March 2007 (UTC)

According to the article Function, a transform is also a function. It is a very general concept. The domain need not be a set of numbers, but can be a set of other functions too. Bo Jacoby 00:12, 9 March 2007 (UTC).

A transform is a function on the (appropriate) space of functions. For instance ${\mathcal {F}}$: L²(R) → L²(R). However all the discussion about the correctness of the involution "definition" is moot because of the WP:OR concerns (the policy specifically talks about original research that "defines or introduces new terms (neologisms), or provides new definitions of existing terms"). This can be settled only by references (see WP:ATT) and no amount of debating on the talk page can overcome that. Regards. Abecedare 02:37, 9 March 2007 (UTC)

And besides, a brand new definition does not help people understand "Fourier transform". It's bad enough that we already have three to worry about. A fourth would be counterproductive. Why not use the involution article to showcase your involution? Why here? This article is not about involutions.

--Bob K 04:01, 9 March 2007 (UTC)

Bob, please note that no Wikipedia article should be used by any editors to "showcase" their X, for any X that is not found in reputable sources. Let's not foist our problems onto other unsuspecting articles. —Steven G. Johnson 15:11, 9 March 2007 (UTC)

Right. Point taken. --Bob K 15:26, 9 March 2007 (UTC)

smoother introduction to the 3 different conventions

I propose the following section to replace sections Fourier_transform#Definition and Fourier_transform#Normalization_factors_and_alternative_forms. One thing that bothers me about the current version is the abrupt appearance of the ${\sqrt {2\pi }}\,$ factors. I think it can be done in a more natural looking progression.

Another thing that bothers me is the paragraph:

Then the inverse transform can be written:

$x(t)={\frac {1}{\sqrt {2\pi }}}\int _{-\infty }^{\infty }A(\omega )e^{i(\omega t+\phi (\omega ))}\,d\omega$

which is a recombination of all the frequency components of x(t).  Each component is a complex sinusoid of the form e^iωt whose amplitude is proportional to A(ω) and whose initial phase angle (at t = 0) is φ(ω).

Wouldn't it be more logical if the components being recombined had amplitudes equal to A(·)?   Well this way they do:

--Bob K 00:46, 11 April 2007 (UTC)

Definitions

There are several common conventions for defining the Fourier transform of a complex-valued Lebesgue integrable function, $x.\,$  In communications and signal processing, for instance, it is often the function:

$X(f)=\int _{-\infty }^{\infty }x(t)\ e^{-i2\pi ft}\,dt$,   for every real number $f.\,$

When the independent variable $t\,$ represents time (with SI unit of seconds), the transform variable $f\,$ represents ordinary frequency (in hertz).  The complex-valued function, $X,\,$ is said to represent $x\,$ in the frequency domain.   I.e., if $x\,$ is a sufficiently smooth function, then it can be reconstructed from $X\,$ by the inverse transform:

$x(t)=\int _{-\infty }^{\infty }X(f)\ e^{i2\pi ft}\,df$,   for every real number $t.\,$

Other notations for $X(f)\,$ are:  ${\hat {x}}(f)\,$  and  ${\mathcal {F}}\{x\}(f)\,$.

The interpretation of $X\,$ is aided by expressing it in polar coordinate form:  $X(f)=A(f)\ e^{i(2\pi ft+\phi (f))}\,$, where:

$A(f)=|X(f)|,\,$   the amplitude

$\phi (f)=\angle X(f),\,$   the phase

Then the inverse transform can be written:

$x(t)=\int _{-\infty }^{\infty }A(f)\ e^{i(2\pi ft+\phi (f))}\,df$

which is a recombination of all the frequency components of $x(t).\,$   Each component is a complex sinusoid of the form e^{i 2 π f t} whose amplitude is A(f) and whose initial phase angle (at t = 0) is φ(f).

In mathematics, the Fourier transform is commonly written in terms of angular frequency:  $\omega =2\pi f,\,$  whose units are radians per second.

The substitution $f={\frac {\omega }{2\pi }}\,$ into the formulas above produces this convention:

$X(\omega )=\int _{-\infty }^{\infty }x(t)\ e^{-i\omega t}\,dt$

$x(t)={\frac {1}{2\pi }}\int _{-\infty }^{\infty }X(\omega )\ e^{i\omega t}\,d\omega ,$

which also happens to be a bilateral Laplace transform evaluated at s = i ω.

The 2π factor can be split evenly between the Fourier transform and the inverse, which leads to another popular convention:

$X(\omega )={\frac {1}{\sqrt {2\pi }}}\int _{-\infty }^{\infty }x(t)\ e^{-i\omega t}\,dt$

$x(t)={\frac {1}{\sqrt {2\pi }}}\int _{-\infty }^{\infty }X(\omega )\ e^{i\omega t}\,d\omega .$

A transform whose inverse has the same multiplicative factor is called a unitary transform.

Variations of all three conventions can be created by conjugating the complex-exponential kernel of both the forward and the reverse transform. The signs must be opposites. Other than that, the choice is (again) a matter of convention.

Comments on Bob's version

I like your version and support replacing the "Definitions" and "Normalization Factors and alternative forms" by the one above. Some comments:

• We can replace "sufficiently smooth" by continuous, since that is the condition for the stated point-wise convergence ("for every real number $t.\,$").
• The description "A transform whose inverse has the same multiplicative factor is called a unitary transform." is not clear to me, since I don't know what "multiplicative factor" of a transform means. IMO perhaps we can just state that the third form of the transform is unitary, and leave the explanation (F*F = FF* = I where F* is adjoint operator) to the "Completeness section".

Aside: I just noticed that properties section was missing the $1/{\sqrt {2\pi }}$ factor without which the all the completeness, parseval's convolution etc results in the section were incorrect! I have corrected that now. Abecedare 01:36, 11 April 2007 (UTC)

Proof of inverse fourier transform

Why is the inverse fourier transform true?

$x(t)=\int _{-\infty }^{\infty }X(f)\ e^{j2\pi ft}\,df$

$x(t)=\int _{-\infty }^{\infty }\left(\int _{-\infty }^{\infty }x(\tau )\ e^{-j2\pi f\tau }\,d\tau \right)\ e^{j2\pi ft}\,df$

now what can I do? How can I reduce the right side of the equation so that it equals the left? — Preceding unsigned comment added by Iownatv (talk • contribs)

The simple derivation, often found in engineering texts, proceeds as follows:

$\int _{-\infty }^{\infty }x(\tau )\left(\int _{-\infty }^{\infty }\ e^{-j2\pi f(\tau -t)}\,df\right)\,d\tau =\int _{-\infty }^{\infty }x(\tau )\delta (t-\tau )\,d\tau =x(t)$

A more rigorous derivation (needed to justify the above steps) requires careful consideration of the functional space to which $x(t)$ belongs and the convergence properties of the various integrals. For this reason the article does not delve into deriving the various Fourier transform results and properties. Such analysis, at varying level of rigor, can be found for example in Mallat's "A wavelet tour of signal processing"; Bracewell's "The Fourier Transform and Its Applications" or Stein and Weiss's "Introduction to Fourier Analysis on Euclidean Spaces". Abecedare 05:28, 3 May 2007 (UTC)

Thanks Iownatv 20:20, 5 May 2007 (UTC)

Heuristic "Derivation"?

Would it be appropriate to add a heuristic "derivation" of the fourier integral and inverse as a limiting process of the exponential Fourier series of an arbitrary periodic function to aperiodic by letting the period tend to infinity? This at least makes it somewhat intuitive and plausible, before one goes on to the convergence and existence conditions and all of that. DivisionByZer0 20:35, 14 June 2007 (UTC)

Range and domain

What about the range and domain of the Fourier transform? The article does not say anything about which functions (or generalized functions) can be transformed or what is the range of the transform. --KYN 20:52, 3 August 2007 (UTC)

The real case

Many, particularly engineers and physicists, may be interested in only the real case and may be confused by the use of complex exponentials instead of trigonometric functions. The MoS guideline Wikipedia:Make technical articles accessible says that articles should begin with easier sections the reader can relate to. I suggest that a section using sines be added to the beginning before going on to complex exponentials. Loom91 20:15, 13 August 2007 (UTC)

Engineers and physicists, really? I don't recall ever having seen the Fourier transform taught without using complex exponentials in some form. The Fourier series, yes, and we do have the trig expressions in that article, but by the time anyone gets to Fourier transform on the real line they are usually talking about complex exponentials.

The problem is, even if you write it in terms of sine and cosine, you still almost have to deal with relatively advanced concepts such as delta functions just to show that the inverse works. Using trig just makes these concepts more complicated to express, but doesn't really allow someone without the math background to "understand" the transform. (The case of Fourier series is quite different, in that it is elementary to show that the inverse formulas work. Proving convergence is more difficult, although L2 convergence is pretty easy for the series case.)

On the other hand, the introductory paragraph is way too vague: "the Fourier transform is ... a certain linear operator that maps functions to other functions."? Someplace in the introduction it should say that the Fourier transform expresses a function an terms of a linear combination of sines and cosines or sinusoids or something of that sort.

And someplace it should explicitly say that sines and cosines, for mathematical convenience, are written in terms of complex exponentials via Euler's identity. And way too much of the introductory sections is devoted to the trivial topic of normalization conventions, which sheds almost no light on the fundamental concepts of Fourier transforms.

In short, I agree that some rewriting would be very helpful to make this article more accessible. I'm not inclined to think that the major problem lies with trig. vs. complex exponentials, though. —Steven G. Johnson 21:17, 13 August 2007 (UTC)

regarding "linear combination":

Do you disagree with [1] ??

--Bob K 00:01, 14 August 2007 (UTC)

Yes, I disagree with that. While there may be a formal context in which you might shy from using the term "linear combination" to refer to an infinite series or continuous integral, colloquial usage has no such restriction. And while each "individual" frequency has measure zero in the integral unless it is weighted by a delta function, conceptually this is the simplest way for a nontechnical person to think of the Fourier transform (essentially, as a limit of a Fourier series). And after all, the current text talks about "frequency components", which has much the same problem except that it is more vague about what the "components" are. (And it is not as if the concept of a "sinusoidal component" cannot be made rigorous, by integrating against a test function with any prescribed narrow bandwidth around a given frequency.) The topic paragraph should be handwavy and conceptual, should be informal for the benefit of the nontechnical reader, and should try to convey a picture of what is going on without worrying about distributions and measurable sets. All of those precise distinctions can be made later in the article. Calling it "a certain linear operator that maps functions to other functions" conveys next to nothing (this description applies equally well to multiplying functions by 2). —Steven G. Johnson 18:09, 14 August 2007 (UTC)

regarding normalizations:

Still, I think it is an important topic, because it is the source of much confusion and disagreement. There aren't many places a person can go and find all three conventions harmoniously and impartially explained, including a reasonably logical progression from one to the next and the next. And formulas for converting any convention to any other.

--Bob K 00:13, 14 August 2007 (UTC)

I'm not saying that the article should omit this information. Just that it is not central to understanding Fourier transforms, and shouldn't occupy the dominant position that it does now in the introductory section. The introductory section should pick whatever convention is used in most of the article (unitary in ω, it looks like), and mention that other conventions are discussed in the section (e.g.) "Conventions and Normalizations" below. —Steven G. Johnson 18:09, 14 August 2007 (UTC)

Loom91, were you referring to Sine and cosine transforms, which can avoid complex functions altogether (albeit at the cost of requiring even/odd symmetry of the transformed function) ? If so, those are linked from the pages "See Also" section and in the List of Fourier-related transforms. Abecedare 21:26, 13 August 2007 (UTC)

The lead section already says Loosely speaking, the Fourier transform decomposes a function into a continuous spectrum of its frequency components, and the inverse transform synthesizes a function from its spectrum of frequency components, and makes no mention of normalization. This could be elaborated in an Introduction inserted before the Definition section - but a precise definition (including normalization) is pretty central, and ought to follow immediately after. I agree with Bob K, that the handling of the (possibly contentious) normalization issue is now very good. And by the way, even us engineers were taught complex algebra at school --catslash 21:26, 14 August 2007 (UTC)

fourier transform of an integral

hi! does somebody know where i can find a proof of $\int _{-\infty }^{t}f(u)\,du\quad {\stackrel {\mathcal {F}}{\Longleftrightarrow }}\quad {\frac {1}{i\omega }}F(\omega )+\pi F(0)\delta (\omega )\,$

thanks! --217.117.224.177 09:57, 21 September 2007 (UTC)

${\begin{aligned}\int _{-\infty }^{t}f(\tau )\,d\tau &=\int _{-\infty }^{\infty }u(t-\tau )\cdot f(\tau )\,d\tau \\&=u(t)*f(t)\quad \longleftarrow \quad {\mbox{convolution}}\end{aligned}}$

Therefore:

${\begin{aligned}{\mathcal {F}}\left\{\int _{-\infty }^{t}f(\tau )\,d\tau \right\}&={\sqrt {2\pi }}\cdot {\mathcal {F}}\{u(t)\}\cdot {\mathcal {F}}\{f(t)\}\\&={\sqrt {2\pi }}\cdot {\sqrt {\frac {\pi }{2}}}\left({\frac {1}{i\pi \omega }}+\delta (\omega )\right)\cdot F(\omega )\\&=\left({\frac {1}{i\omega }}+\pi \delta (\omega )\right)\cdot F(\omega )\\&={\frac {1}{i\omega }}F(\omega )+\pi \delta (\omega )F(\omega )\\&={\frac {1}{i\omega }}F(\omega )+\pi \delta (\omega )F(0)\end{aligned}}$

I assumed the unitary convention, but if $F(\omega )\,$ is the non-unitary transform, this expression is also a non-unitary transform, because the non-unitary transform of $u(t)\,$ is ${\sqrt {2\pi }}\cdot {\sqrt {\frac {\pi }{2}}}\left({\frac {1}{i\pi \omega }}+\delta (\omega )\right).\,$  I.e., the ${\sqrt {2\pi }}\,$ factor comes from $U(\omega )\,$ instead of from the transform property.

--Bob K 21:03, 25 September 2007 (UTC)

Archiving parts of this talk page

This talk page is now so long that I believe its practical use is reduced. Therefore I propose to archive the earlier parts of this talk page, starting from 2004 until 2005, leaving Talk:Fourier transform#Alternative forms as the first posting that remains in the talk page. I propose to use the "Subpage archive method/cut and paste procedure" described here. In principle I would like to archive a larger chunk, let's say including all 2006 postings, to make the page more manageable but there appears to be some discussions which were initiated in these early postings but still have attracted answers or comments in 2007. Any optinions? --KYN 15:21, 23 September 2007 (UTC)

You can archive as much as you want. If it is needed you may post here a summary of the most interesting/active part of the archive. At any point anyone can restart any part of the archived discussion. (Igny 03:05, 24 September 2007 (UTC))

All discussions started 2006 or earlier have now been moved to "Archive 2" which, together with "Archive 1", now is reachable from the Archive icon at the top of this page. If there are active threads which have been removed, please reinsert them on this page. --KYN 09:56, 1 October 2007 (UTC)

Distribution 312

According to my signals and systems book (Signals and Systems by Oppenheim, Willsky and Nawab) the fourier transform for the signal $x(t)=\sum _{n=-\infty }^{\infty }\delta (t-nT)\,$ is $X(j\omega )={\frac {2\pi }{T}}\sum _{n=-\infty }^{+\infty }\delta \left(\omega -{\frac {2\pi n}{T}}\right)$. However the wikipedia article says that the fraction in front of the sum is ${\frac {\sqrt {2\pi }}{T}}$. I changed this a few days ago, but it was changed back, so I thought I'd start a discussion about it. There is a pretty good example in my book that shows the derivation of this (p. 299) so I'm pretty sure that square root shouldn't be there, but obviously someone thinks it should be there. —Preceding unsigned comment added by Fkp1 (talk • contribs) 21:31, 31 October 2007 (UTC)

Does your book use the same normalization convention (Fourier transform#Definitions) as is used in the article? The article table uses the unitary angular frequency convention... check that they match. That's usually the source of factor-of-${\sqrt {2\pi }}$ mismatches... ǝɹʎℲxoɯ (contrib) 21:53, 31 October 2007 (UTC)

PS- That sounds like a book oriented towards engineers, which makes it quite likely that they use the non-unitary convention (which is asymmetrical, but saves slightly on computational effort). ǝɹʎℲxoɯ (contrib) 21:55, 31 October 2007 (UTC)

I was the one who changed it. I checked it directly by numerically integrating on a computer, so I'm quite confident in having reverted it. You should check the conventions defining Fourier transforms in your book, there's likely a factor of (2*pi)^(1/2) differing between the definition used here and the one in your book.

By the way, if it was incorrect, it would be a good service to change it in both the ω and the f columns; after changing just one, they were inconsistent, which was why I was checking in the first place. --Steve 02:04, 1 November 2007 (UTC)

Ahh, that explains it. Sorry for the trouble! Fkp1 16:23, 1 November 2007 (UTC)

Why is distribution 311 a distribution?

Unless I am mistaken, $e^{-at}u(t)$ is a square-integrable function when $a>0$. Why then is this Fourier transform pair in the distribution section?

Butala 13:46, 1 November 2007 (UTC)

Removed a paragraph of disgruntled editor

I removed this from the "Definitions" section and reproduce it here because that is where it belongs

Wikipedia is supposed to be for everyone. This is only useful to mathematicians. Mathematicians have many books containing this material. This is an opportunity to make Fourier transforms understandable to lay people. This is what Wikipedia used to be like. Now the mathematicians have subsumed Wikipedia into their own domain, and use it as a place to lay down the same formalities that are available in their many books---books that are the reason that people need Wikipedia to remain a place for the non-specialist to go for understanding. After all, when deciding what should be here, what non-mathematician could know better than mathematicians? And how many non-mathematicians will show up here to participate in the decision? Below, you can see the result. Wikipedia used to be useful. It used to be by the people for the people. This page is an example of how Wikipedia pages about math are only useful to mathematicians. If you edit this out, then replace it with something that will provide an understanding of Fourier transforms to somebody who doesn't already understand them. Doing that elegantly is a greater challenge than creating the formalisms below. Wikipedia is not a repository for enshrining formalities. Use your books for that. It is for people to learn from. Nobody who comes here to find out what a Fourier transform is will understand this. Shame on the mathematicians for their lack of consideration and focus on their own interests.

Feraudyh 10:40, 6 November 2007 (UTC)

Hello. I am an engineer, not a mathematician. And I have done a lot of work recently to try to mitigate this very problem. For one thing, the very first sentence contains a link to Fourier analysis, a more remedial article, with the recently-added section: Fourier Analysis#How it works (a basic explanation). I considered adding a statement to the effect that it was "prerequisite" information, but that does not seem to be a Wikipedia tradition. So I am wondering if "disgruntled editor" has taken Fourier analysis into account.

I have also tried to improve the consistency in terms of using SI units (hertz) and unitary normalizations. And I have tried to emphasize the similarities and relationships between the continuous FT and its "special cases".

In theory, the Wikipedia strategy is to start the article at the lowest level and end at the highest level, with smooth transitions. But that is challenging at best and impractical at worst, especially at these pay levels :-). I appreciate a lot of the things the mathematicians have contributed, and I appreciate that I can find them online, because if I had to go to a college library, I wouldn't bother. My inclination, if I made the "rules", would be to encourage multiple levels of the same article: Fourier transform 101, Fourier transform 201, etc. No plan is without drawbacks, and a drawback of that plan is the same inconsistencies one sees between textbooks (conventions, definitions, terminology, symbols, notation). But hey, that is reality, and Wikipedia's charter (for better or worse) is to document reality, not "fix" it.

--Bob K 12:00, 6 November 2007 (UTC)

Did you mean Fourier tranform for dummies? Containing myths, misconceptions, trivia, etc? It is hard for me to picture more than two levels of difficulty for most math articles. Someone should start dummipedia or something where each article (mathematical and others) should start with In a kingdom far far away... (Igny 03:57, 7 November 2007 (UTC))

• Dummipedia started! — PM Poon (talk) 16:02, 22 February 2008 (UTC)

No matter, because that is not how Wikipedia works, and I don't expect it will change anytime soon. But FYI, I was assuming that what we have now, minus a few expendable pieces, would be the 101 level. I don't understand why "disgruntled" says: "This page is an example of how Wikipedia pages about math are only useful to mathematicians." That is why I asked for more specifics.

--Bob K 05:44, 7 November 2007 (UTC)

Dear Feraudyh. I completely agree with you. The article should be readabel to you. I regret that it is not. I have tried to mend it, but I am opposed by another WP-editor who believes that he can do better and that WP should contain 'standard' mathematics even if that is incomprehensible mathematics. See: User:Bob_K#two_articles_for_discrete_fourier_transform_.3F. I am sorry. Keep complaining. It helps. Bo Jacoby 13:49, 6 November 2007 (UTC).

Can we get more specific? I don't actually see a lot of high falootin' stuff in the article. It looks pretty basic to me. What would you remove? "Generalization" perhaps? It's kind of interesting, but admittedly not very consequential. What about "Completeness"? I always skip over that one, but maybe that's just me. I don't seem to have much use for "Localization property" either, but it looks well-written. I think the article would look less intimitating without all the square root operators, which result from using radian frequency instead of ordinary frequency. But lots of books do prefer radian frequency, so we're probably stuck with it.

--Bob K 17:52, 6 November 2007 (UTC)

I'm afraid I have to side this time with the non-mathematicians complaining about the (lack of) accessibility of this article. It is true that there is still a huge scope to improve the presentation – and not only for the benefit of non-mathematicians but mathematicians as well. Without any attempt at completeness, I would like to make the following observations and make a few suggestions:

• Terminology: My modest proposal would be to use systematically Fourier transformation for the operation (function) of forming the fourier transform of a function (or distribution...). And then mention that many authors use transform for both.
• Title of the article: Fourier transform is likely to be the first article someone interested in the topic looks at. Fourier transform(ation)s are however much more general than the one for functions on the real line. Therefore this article should be renamed something like "Fourier transform on R" or "Fourier transform on the real line". This is particularly the case if "Fourier analysis" currently contains both the general discussion and the "down to earth" discussion. Thus that article appears a better candidate to bear the title of this one.
• The lead does not provide practically any motivation for having the Fourier transformation. General statements such as "transforms certain hard problems about the original function to more tractable ones about its transform", "provides a way represent a complicated function by means of simpler ones" and so forth would be a good start, to be followed by slightly more technical comments. Moreover, the "time domain / frequency domain" terminology in the lead is hardly helpful to people without physics/engineering background. This includes mathematicians: an aspiring number theorist looking forward to a brief introduction to Fourier transformations would have hard time with time and frequency, I'm afraid.
• Definition is immedetiately in terms of an explicit formula with no discussion, background etc. Reference is immediately made to Lebesgue-integrable functions, and then for some unexplicable reason the discussion moves into the realm of engineering with the sudden introduction of SI units (I'm again thinking about my poor aspiring number theorist...). More generally, my humble opinion is that the context provided by harmonic analysis (dual groups,...) in fact helps both mathematicians (larger dose of detail and formalities) and typical non-mathematicians (less detail) to understand the Fourier transformation. The case may be different for engineers and some physicists, who typically have a practical reason to use the transformation, and may be happy to get quickly to the mechanics of applying it. Renaming this article and addressing the more general set-up (first less, then more technically) in a new Fourier transformation article should help here.
• Unexplained formulas fill most of the article. Too little explanation, too much reliance on collection of formulas. Currently the article seems to aim at being a collection of facts to refresh memory than an aid to understanding.
• Too many tables fill the article. While it's great that effort has been put into providing an online reference to the formulas, their place is hardly in the article itself. They are more "list-like" material, that it would be good to move to a separate article and reference here.

The above may look harsh, but it is definitely not meant as critique of the editors that have worked on the article. However, in order to move the article (rather the topic) to a new level I'm afraid that something like the above should be addressed. In other words, work remains to be done. Stca74 14:24, 7 November 2007 (UTC)

Your insights are appreciated. Thanks for taking the time. I frankly doubt that a single article can perfectly accommodate both the number-theorists and the EEs, or even do it "well". Because of the inherent instability of Wikipedia, a compromise strategy will lead to endless thrashing. I suspect it already has. The whole structure is just wrong. Math books don't look much like EE books, and there are insurmountable reasons for that. So we're just wasting our time here trying to fit a square peg in a round hole. Can we split off into separate articles? Or will someone find a policy/rule violation and force everyone back into the same foxhole? If we split, and the two articles really do belong together, that will become apparent over time, and someone can merge them. But in the meantime, separate groups can work independently, unconstrained by the need to compromise with each other.

--Bob K 15:18, 7 November 2007 (UTC)

OK, that went well. So nobody really wants to deal with the whole interdisciplinary problem. The only hope for an in-depth treatment is to divide and conquer, which is also what universities do, by department. The time/frequency view is sufficient for EEs, so that is what they teach, along with real-world units such as hertz. Wikipedia's charter is to reflect that reality (and others), not to merge it all together into an abstract generality.

So instead of working on the actual article, how about we step back and work on the bigger picture first? Apparently there needs to be an overview article about "Fourier transformations" that would itself be linked to from even higher-level articles (on transformations, representations, analysis, etc). The overview article would try to explain that specialized treatments of the same basic subject have evolved to best suit the needs of various practitioners and theorists. Then it would provide links to the appropriate, but separate, articles. We might take a stab at predicting what those articles will turn out to be, but it will probably evolve on its own over time as articles tend to split and merge. Perhaps just two will be sufficient (one for mathematical formalities, and another focussed on the time/frequency domain implications). But perhaps not.

Of course, the same points, such as the duality of convolution and multiplication, will end up being made in multiple places, but each time it will be in terms of the conventions familiar to the interested reader. The page-count will be non-minimal, but each reader only needs to read the thread that applies to himself.

--Bob K 16:21, 9 November 2007 (UTC)

Categories of Fourier transforms

This new section is important, but somewhat confusing. The dash in 'aperiodic-discrete' is confusing. It perhaps means that the function is aperiodic and discrete. Why not write that? As to "The Fourier Transform for this (Aperiodic-Continuous) type of signal is simply called the Fourier Transform". No, it is called the fourier integral. The reader wants to know a reason for this trouble; why not a single fourier transform to cover these cases? The problem is that a 'discrete function' as the limit of continuous functions does not exist, and the solution in terms of distributions is too advanced. Bo Jacoby (talk) 12:08, 20 January 2008 (UTC).

Useless

I've cited this article, chosen pretty much at random as many others are similar, as an example of one that I consider useless in that it is written at too high a level to be generally accessible, in an a discussion I have sought to provoke about how Wikipedia's mathematical articles are structure in the Village pump/Policy area in the Community Portal.

David Colver (talk) 15:15, 2 February 2008 (UTC)

Invertibility condition

The text says "If x is Hölder continuous, then it can be reconstructed from by the inverse transform". The linked page on Hölder continuity tells me that anything bounded is Hölder continuous with exponent 0. But this includes bounded discontinuous functions, which cannot be reconstructed from the inverse transform. It appears to me that one of the two pages (and I don't know which) must be in error. —Preceding unsigned comment added by 80.177.154.109 (talk) 12:33, 7 February 2008 (UTC)

Article is target for vandalism

Folks from the popular 4chan boards and other sites are targeting the LaTeX source in higher math articles for minor vandalism, I could see this hurting homework grades and exam prep for quite a few students, something to keep an eye out for.

search google for "Vandalize_Every_Equation" to see a link to a wiki on another popular site with details (it's blacklisted on wikipedia's spam filters so I can't link to it directly)

140.247.10.17 (talk) 06:04, 16 February 2008 (UTC)anonymous

Bessel function

Sorry, but I'm not sure that Bessel function are square integrable (the norm L2 of its transform diverge). If is not so, please tell me why? (Sorry if I don't sign this message) --penaz