DSPRelated.com
Forums
Online · Live + on demand
February 9-11, 2027
Price goes up Jan 1: register now to lock in today's rate
Register Now →

Dirac delta

Started by Andor ●August 21, 2003
Andor wrote:
> > Jerry Avins wrote: > ...
...
> > > > > "Ah", you say! "No object can be size one, > > Wrong again. I can easily define the size of an object as being "1", with > the apropriate unit. We do it all the time, that's what units are all about.
Agreed. Every object has size one when the measurement unit is "self". That's what normalizing is all about.
> I'm not quite sure what this has to do with the Dirac delta though :). >
It has to do with the existence of intangibles, a class that includes the Dirac delta. :-) Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
In article 3f489793_2@news.tiscalinet.ch, Andor at an2or@nospam.com wrote on
08/24/2003 06:51:

> Jerry Avins wrote: > ... >> Of course a delta doesn't exist; nobody can produce one. It's a handy > > Hey waitaminute! I think I have to get the definition of "non-existence" > straight: the Dirac delta does not exist as a limit of integrable functions, > because any limit of integrable functions does not have the properties of > the Dirac delta. So, by contradiction, it _cannot_ exist.
that's Fatou's lemma in Real Analysis, right. the integral of a function that is zero almost everywhere is zero. the thing that bothers me (with this lemma) is that the limit of the integral of these integrable functions (not the limit of the functions themselves) *does* exist and is 1 (for the unit impulse). In article 3f48a838$1_1@news.tiscalinet.ch, Andor at an2or@nospam.com wrote on 08/24/2003 08:01:
> robert bristow-johnson wrote: > >>> r b-j wrote some time ago: >>> >>>> +inf >>>> integral{g(t) * d(t) dt} = g(0) ? >>>> -inf >>>> >>>> (where d(t) is our nasty Dirac impulse function, g(t) >>>> is any reasonably well-behaved function of t and "*" >>>> means "multiplication") >> >> BTW, i only wrote it (some time ago) because it is the main property of the >> Dirac delta "function" (or whatever mathematicians wanna call it) that both >> engineers/physicists and pure mathematicians can agree on. > > We actually agree on most points. All I wanted to state was the fact that if > the Dirac delta were defined as D g = g(0) > > (or if this equation were true: >>>> +inf >>>> integral{g(t) * d(t) dt} = g(0) >>>> -inf > ) > > then this defintion would be useful for continuous functions only.
the functions, g(t), don't have to be continuous, just *defined* at 0, for the latter definition. there could be a step discontinuity at 0 but if g(0) is defined (not at some undefined meanpoint between g(-epsilon) and g(+epsilon). what if: { 1 for t >= 0 g(t) = { ( 0 for t < 0 ? doesn't the functional definition work in that case? for the definition regarding the limit of the rectangular or whatever function becoming an impulse, then maybe g(t) has to be a continuous function. i'm not sure about that.
> I wanted > to make it clear that the integral was needed to "average" the value of g > around 0 (read my reply to Brandon) to extend D on the class of integrable > functions, so the integral is needed in the definition of the Dirac delta > functional. I think that this is also the reason for all the confusion.
but in the limit, there isn't much around 0 that g() has to be averaged.
> The use of the Dirac delta outside of integrals can be justified as a measure > (delta(A) = 1 if 0 is element of A, A a subset of the real numbers).
but that isn't an expression of d(t) as a function of t. which is fine. it's just that i don't understand where your functional "delta()" comes into play.
> With > this delta measure, you can well, measure sets and, more importantly, > define integrals over delta-measurable functions. > > Both constructions result in the same thing, but one defines a functional on > the space of integrable functions, and one defines a measure on subsets of > the real numbers. This measure (like all normalized measures) can then be > used to define a distribution for a random variable with a point mass at 0, > thus the term "distribution".
in other words, a random variable that is a constant (not very random) so it's p.d.f. is an impulse, right?
> Regards,
and also to you. r b-j
robert bristow-johnson wrote:

> >> Of course a delta doesn't exist; nobody can produce one. It's a handy > > > > Hey waitaminute! I think I have to get the definition of "non-existence" > > straight: the Dirac delta does not exist as a limit of integrable functions, > > because any limit of integrable functions does not have the properties of > > the Dirac delta. So, by contradiction, it _cannot_ exist. > > that's Fatou's lemma in Real Analysis, right. the integral of a function > that is zero almost everywhere is zero. the thing that bothers me (with > this lemma) is that the limit of the integral of these integrable functions > (not the limit of the functions themselves) *does* exist and is 1 (for the > unit impulse).
The two limits do not coincide because the series of integrable functions is not monotone increasing over the interval where we integrate. This condition is necessary to take the limit out of the integral (this is the monotone convergence theorem).
> >>>> +inf > >>>> integral{g(t) * d(t) dt} = g(0) > >>>> -inf > > > > the functions, g(t), don't have to be continuous, just *defined* at 0, for > the latter definition.
Look at this example: f(x) = sin(x), x in ]0,Pi] and f(0) = 1. Since the functions f and sin are in the same L^p class (p >= 1) on the interval [0, Pi], the functional D extended to L^p gives the same results when applied to f and sin, namely D f = D sin = sin(0) = 0 =\= f(0). That's what I mean by "averaging" the value of f around 0.
> > The use of the Dirac delta outside of integrals can be justified as a measure > > (delta(A) = 1 if 0 is element of A, A a subset of the real numbers). > > but that isn't an expression of d(t) as a function of t. which is fine. > it's just that i don't understand where your functional "delta()" comes into > play.
d(t) := delta( {t} ) (the argument to delta has to be a subset of the reals) ...
> in other words, a random variable that is a constant (not very random) so > it's p.d.f. is an impulse, right?
Correct. And the distribution of this random variable X is the delta measure: P[ X <= t ] = delta( ]-infinity, t] ) Andor.
robert bristow-johnson wrote:
> > In article 3f489793_2@news.tiscalinet.ch, Andor at an2or@nospam.com wrote on > 08/24/2003 06:51: > > > Jerry Avins wrote: > > ... > >> Of course a delta doesn't exist; nobody can produce one. It's a handy > > > > Hey waitaminute! I think I have to get the definition of "non-existence" > > straight: the Dirac delta does not exist as a limit of integrable functions, > > because any limit of integrable functions does not have the properties of > > the Dirac delta. So, by contradiction, it _cannot_ exist. > > that's Fatou's lemma in Real Analysis, right. the integral of a function > that is zero almost everywhere is zero. the thing that bothers me (with > this lemma) is that the limit of the integral of these integrable functions > (not the limit of the functions themselves) *does* exist and is 1 (for the > unit impulse). >
... I'm too ignorant to be bothered. It makes perfect sense to me that the integral of a function that is zero a.e. be zero AS LONG AS THE EXCEPTIONAL POINTS ARE FINITE. So it doesn't seem contradictory to me that the integral of a Dirac delta is a step. Jerry -- Engineering is the art of making what you want from things you can get. &#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;
Jerry Avins wrote:
...
> I'm too ignorant to be bothered. It makes perfect sense to me that the > integral of a function that is zero a.e. be zero AS LONG AS THE > EXCEPTIONAL POINTS ARE FINITE. So it doesn't seem contradictory to me > that the integral of a Dirac delta is a step.
Jerry, we talked about this before :). The integral is zero if the function is zero a.e. The exceptional points can be infinite, and there can be uncountably many infinite points (ie. as many as there are real numbers) in the interval where you integrate, as long as the measure of the set of these uncountably many points is zero.
I wrote:

> > > The use of the Dirac delta outside of integrals can be justified as a measure > > > (delta(A) = 1 if 0 is element of A, A a subset of the real numbers). > > > > but that isn't an expression of d(t) as a function of t. which is fine. > > it's just that i don't understand where your functional "delta()" comes into > > play. > > d(t) := delta( {t} ) (the argument to delta has to be a subset of > the reals) > > ... > > > in other words, a random variable that is a constant (not very random) so > > it's p.d.f. is an impulse, right? > > Correct. And the distribution of this random variable X is the delta > measure: > > P[ X <= t ] = delta( ]-infinity, t] )
On second thoughts, this doesn't get us any further. If the distribution of this rv is the delta measure, we want that the density is the Dirac delta. But we can't get such a density, because the delta measure is not absolute continuous w.r.t the Lebesgue measure. This just means that you cannot get the value of the point mass back in an integral with the Lebesgue measure - ie. there is no density function. So the probability measure has to be split up into a sum of an integral over an absolute continuous function (the density) and a signular measure (delta measure) which returns the discrete steps (like for example all discrete random variables have no density function). Just one more proof that the Dirac delta does not exist as a function :). So the name "distribution" for the Dirac delta doesn't come from probability theory (because the Dirac delta is a "density", so to speak). I looked it up in a physics book - they define a distribution as a functional. Since this is a German book, and the use the term "Distribution" and not "Verteilung" which would be the correct translation of the probability distribution, I think the two have nothing in common, except their English names. Regards, Andor
Andor wrote:
> > Jerry Avins wrote: > ... > > I'm too ignorant to be bothered. It makes perfect sense to me that the > > integral of a function that is zero a.e. be zero AS LONG AS THE > > EXCEPTIONAL POINTS ARE FINITE. So it doesn't seem contradictory to me > > that the integral of a Dirac delta is a step. > > Jerry, we talked about this before :). The integral is zero if the > function is zero a.e. The exceptional points can be infinite, and > there can be uncountably many infinite points (ie. as many as there > are real numbers) in the interval where you integrate, as long as the > measure of the set of these uncountably many points is zero.
I know we did. But I can't pound into my head that while the derivative of a unit step is a delta, the integral of a delta isn't a unit step. You can see why I so admire Heavyside. I don't believe everything math tells me. I know I won't ever be able to dice up a gold sphere, then reassemble the pieces in a way that would make me rich. Who shaves the barber? :-) Jerry -- Engineering is the art of making what you want from things you can get. &#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;
Jerry Avins wrote:
...
> I know we did. But I can't pound into my head that while the derivative > of a unit step is a delta, the integral of a delta isn't a unit step. > You can see why I so admire Heavyside. I don't believe everything math > tells me. I know I won't ever be able to dice up a gold sphere, then > reassemble the pieces in a way that would make me rich. Who shaves the > barber? :-)
Perhaps alchemists should have studied more measure theory and paradox decompositions instead of fiddling with Toad's Skin and Bat's Ears ... :)