last day (14 days later) » 

5:03 PM
hello friend
@Semiclassical
 
For reference, you can do that by clicking on my icon in chat and using the "start new room with this user" option
handy for this kind of thing
(there might be some rep requirements as far as what users can create rooms. I dunno)
 
I have good rep
 
neat.
 
okay so first
 
So let's take on these cases in turn.
sure
 
5:05 PM
how do I know when$(2-x)\chi_{[1,2]}$ vanishes
 
Not sure what you mean by that.
It vanishes outside of $[1,2]$ because the indicator function isn't supported outside this interval.
 
we said that it's $0$ when $x > 3$
ohh
$x$ to $x+1$
 
Actually, I think it should be $x-1$ to $x$ ?
Lemme check, though
We've got $$(f*\chi_{[0,1]})(x)=\int_{-\infty}^\infty f(x')\chi_{[0,1]}(x-x')\,dx'$$
The indicator function vanishes unless $0<x-x'<1$ i.e. $x'>x-1$ and $x'<x$
So that's $x'\in(x-1,x)$, no?
 
the variable of interest is $x$
 
5:10 PM
so the bounds should be from ... oh
 
you'll note that I had to write this out to be sure :P
so $(f * \chi_{[0,1]})(x)=\int_{x-1}^x f(x')\,dx'$. Agreed?
 
yeah
 
mmkay.
So let's take this case by case.
First, let's take the case where the right endpoint of the interval is in $[0,2]$ but not the left one.
so $x-1<0<x<2$. That amounts to $0<x<1$.
In that case, we can split up the integral as $\int_{x-1}^0+\int_0^x$.
What will happen to the integrand over the first part of the integral, though?
 
it is zero
 
Right. So we only need worry about the rest.
Hence we've got $\int_0^x f(x')\,dx'$ where $0<x<1$. But if $f(x)$ is our triangle map we know how $f(x)$ behaves over this interval.
 
5:16 PM
so it's just $\frac {x^2}2$
 
Right.
Let's skip to the other case of this form, where it's in on the left and out on the right.
What range of $x$ does this require?
 
something to 3
 
yeah.
 
2 to 3
 
right. doing it from 1 to 2 will end up being complicated so I'm passing it over for now.
 
5:18 PM
$3-\frac {x^2}2$
 
Lemme check that.
You've got $f(x')=2-x'$ for $1<x<2$.
And you should end up with the integral being $\int_{x}^1 f(x')\,dx'$.
That doesn't match yours.
One sanity check: The integral should vanish when $x=2$.
 
why is the integral only over $[0..1]$?
 
Yours doesn't, so that's not right.
 
if anything, $[0..2]$
 
b/c I am being silly.
 
5:21 PM
haha good one Semiclassic
 
yeah, I'm wrong.
 
you're so funny
:P
 
no, wait, I'm just testing you. you correctly caught my totally intentional mistake.
 
nice
 
that's my story and I'm sticking to it
But yeah. It should really be $\int_{x-1}^{2}f(x')\,dx'$ with $2<x<3$
 
5:22 PM
so $\int_0^y + \cdots + \int_y^2$?
no, that doesnt make sense
 
ugh, more silly mistakes. should be right now, though.
 
the first one we had $\int_0^y + \int_y^1$, right?
 
No.
you had $\int_{x-1}^x = \int_{x-1}^0+\int_0^x=\int_{x-1}^1+\int_1^x$
 
right
RIGHT
ahah
 
The first decomposition was useful when $0\in(x-1,x)$ i.e. $0<x<1$ and the second when $1\in (x-1,x)$ i.e. $1<x<2$.
 
5:26 PM
ahah moment
 
so $\int_{x-1}^0 + \int_0^y + \int_y^3$
 
no.
the upper endpoint of $3$ is for $x$.
it's not for the integration interval.
it's still $x$ for the integration interval.
 
oh right, I mixed bound and unbound variables
or whatever they call them
 
dummy/free
is the one i know
but, let's go back to the case I said we should focus on
namely, 1<x<2
 
5:29 PM
"The term "dummy variable" is also sometimes used for a bound variable"
 
right.
 
bound and free
 
dummy is indicative of the fact that its name doesn't matter
i could relabel $x'$ to $t$ without changing the output
 
as my Calc II professor said
it's not meant to insult or denegrate the variable in any way :P
 
lol
anyways.
if 1<x<2, then in what way does the integration interval [x-1,x] lie in relation to the support [0,2] of f(x') ?
what's the intersection, is what I'm really after
 
5:32 PM
the first is a subset of the latter
 
yeah.
which means I'm being silly.
 
so the subset is the intersection
 
I keep doing 1<x<2 when I mean 2<x<3.
I want to start with the 2<x<3 case first.
 
only $[1..2]$
err
yeah
 
careful. if 2<x<3, then x-1>1.
so the left-endpoint of the integration interval lies to the right of 1.
so, is 1 within the integration interval?
 
5:34 PM
I mean, whether it's $[1..2]$ or $(1..2]$ doesnt really matter
in this case
 
i wondered if you'd raise that objection. let me state it more clearly
suppose x=2.5
then the integration interval is [1.5,2.5] and the support is [0,2].
what's the intersection?
 
$[1.5..2]$
 
right. not [1..2]
So what's the intersection for a generic $x\in(2,3)$?
 
$[x..3]$
err
 
that'd be [2.5,3]
 
5:36 PM
$[x..2]$
$[x-1..2]$
 
and that would be [2.5,2]
right.
 
tricky
 
lower endpoint is provided by the integration interval, the upper by the support
So our integral comes out as $\int_{x-1}^2 f(x')\,dx'$.
 
got it
 
What's $f(x')$ over this region? Note that $x-1>1$, so this is a subset of $[1,2]$.
 
5:38 PM
$1-x$
 
2-x, actually.
 
oops
 
That's what we got earlier.
As a check, 2-x=x at x=1 as it should, and also 2-x=0 at x=2.
so it's a continuous function
wouldn't work that way if it were 1-x over that region
 
you have a typo at $2-x=1$ but I got it
 
do I? hmm.
So our integral is $\int_{x-1}^ 2(2-x')\,dx'=\ldots ?$
 
5:40 PM
I think so
okay so just FTOC
 
right
 
so we're done?
$\blacksquare$?
 
Well, no.
 
theres a middle case
 
For one, you haven't actually told me the result of the integral. For another, yes, the middle case.
 
5:44 PM
$(2 - 2(x-1) - \frac 1 2 (x-1)^2)\chi_{[2,3]}$
 
as a check, if $x=3$ then this gives $2-2(2)-1/2(4) = -4$ not zero.
So you've got a small error
basically, note that the lower endpoint of integration gives $-(2(x-1)-(x-1)^2/2)$. What happens with the minus signs?
 
booo, sign error
 
yep
so it's really +1/2 not -1/2
 
no mathematician is immune
 
Moreover, if you expand that and simplify, you can show that the result is of the form $\frac{1}{2}(x-3)^2$.
So it increases quadratically to the left of $x=3$ and is zero at this endpoint.
That's the mirror of what we got earlier for $x\in(0,1)$ where the result was $x^2/2$
 
5:48 PM
replace $\mu = x-1$ and then complete the square or something?
 
probably $\mu=2-x'$.
 
$x$ you mean?
 
No, $x'$. I have in mind using the substitution rule in the integral.
 
in the integral??
oh
 
5:50 PM
oh, that's easier
 
Yeah. Anyways, you see how that works.
Now for the middle case.
 
im going to have to go over all of this and write it up because now it's just pages of scribbles
by the way the original question was
$X, Y, Z$ are i.i.d $\sim U(0,1)$
 
it was $(\chi * \chi * \chi)(x)$, wasn't it ?
 
find the pdf of $X + Y + Z$
 
5:51 PM
but theres a theorem that if X, Y, are independent
then the pdf of $X+Y$ is $f_X * f_Y$
not sure how else to do it
 
Something something Fourier transform, probably.
But yeah.
 
we didn't learn that
 
right.
Anyways
 
I know Laplace transforms but eh
That's more for functions with infinite support
 
One thing to note is that we've already shown that $(\chi * \chi * \chi)(x)=x^2/2$ for $x\in(0,1)$ and $=(x-3)^2/2$ for $x\in(2,3)$.
I'm dropping the interval [0,1] in $\chi$ because ugh so tedious to write.
 
5:54 PM
I'll allow it :P
 
In particular, by continuity we have $(\chi * \chi * \chi)(1) = 1/2 = (\chi * \chi * \chi)(2)$.
 
??
oh
 
pretty sure the convolution is always continuous.
 
how do we know it's continuous at $x = 1,2$?
let me check
continuous if they have compact support
which these do
 
I think it's continuous so long as it's a continuous pdf
Right.
You'd run into problems if you had a point mass somewhere
 
5:56 PM
wait no
well
$\chi$ is not continuous
at $0$ and $1$
 
hmm, no.
 
More generally, if either function (say f) is compactly supported and the other is locally integrable, then the convolution f∗g is well-defined and continuous.
okay, it's locally integrable
 
Right.
So we can use that as a criterion without issue.
In addition, we have a special case. Suppose $x=3/2$. Then the convolution integral gives $(\chi * \chi * \chi)(3/2)=\int_{1/2}^{3/2} f(x')\,dx'$.
Blah, laptop just ran out of charge
 
:(
 
Have to say anything else via phone
 
5:59 PM
can I use the fact that I know the integral over $\chi * \chi * \chi$ is $1$?
 
Anyways. Note that $f(x')$ is symmetric about x'=1.
Let's not do that yet.
 
k
i have a feeling the professor underestimated the length of this problem
for students who arent familiar with convolution
 
Where I'm going is that we can use symmetry to restrict to 1/2 < x' < 1, at a cost of an overall factor of 2
So the integral becomes $2\int_{1/2}^1 f(x')\,dx'=?$
 
which is just $1$
 
Is it? Hmm
 
6:03 PM
$2 \cdot (1 - \frac 1 2 )$
 
f isn't 1 on that interval
 
oh $f = \chi * \chi$?
 
So the anti derivative is x^2/2
Yeah. Sorry, that was poorly stated
 
np
 
But this isn't doing what I wanted it to :(
 
6:06 PM
I guess I'll go ask the prof before class
youve been amazing help though
 
Well, here's my guess
 
you're a physics student?
 
I think that in the middle it'll be a quadratic fumction
 
reasonable guess
 
And, moreover, it'll join up with the other two functions to be C^1 smooth rather than just C^0.
The reason I say that is that our triangle map was C^0 with compact support
I think that ensures that the convolution is C^1
 
6:09 PM
so find three points to make a parabola
maybe
 
sure. Or, match the slope at the endpoint as well
If it's C^1 then the derivative at x=1 has to match as well
And at x=2
 
ugh I need a break, Im not doing that now
 
That'd give enough info to uniquely determine a parabola
Heh, ok. Quick summary though
 
<3
 
With the first convolution, you got a piecewise linear function which was C^0 with compact support
I think with the second convolution you get a piecewise quadratic function which is C^1 with compact support
 
6:13 PM
there's no way he should expect us to know that about the convolution though
 
probably just underestimated the difficulty
or we're missing some trick
 
This is more a way to avoid the tedium of doing the middle casr
For the middle case, the entire integration interval is in the support
 
right
 
And you have to consider the part with x'<1 separately from x'>1
So you end up with two integrals not one =S
 
6:16 PM
;_;
 
If I'm reasoning right, though, you should find that the resulting function matches up smoothly with the rest of it
So that's a nice way to check your answer
Another is to verify that it's properly normalized
Anyhow, I'm a bit sick of this so I'll leave you to it :)
 

last day (14 days later) »