« first day (4021 days earlier)      last day (1297 days later) » 

00:09
someone said the word 'affirmative action' and I thought there is another math terminology on algebra or something
Is this a good place to ask about gambling paradoxes?
and gambling strategy
askaway
So suppose I am in a game where I start with a certain number of points, say 100, and the goal is to reach 100000 points before any of the other players do the same and without going bankrupt.
The way to do so is by gambling in some game with a positive expected value over and over.
What I need ultimately need to do is find a balance between gaining points fast enough and having a low probability of bankruptcy.
This is a very dynamic problem that's hard to solve outright, but suppose I pick as a constraint that I want to have, i.e., no more than 1% probability of going bankrupt before reaching 100000 points. Then I can find the maximum wager size subject to that constraint pretty easily, either mathematically or by simulation.
In other words, the 1% is arbitrary, but would be "whatever the maximum probability of bankruptcy I'm comfortable with" is.
Now, say I start playing with my strategy I develop, and get to 1000 points (or 101 for that matter).
Now my probability of failure is lower than 1%, so I should want to increase the percentage of my points I bet each time to retarget a 1% failure rate.
However, changing my strategy in this way means my original strategy didn't really have only a 1% failure rate, because I could foresee that I would want to adjust the strategy as I go.
So extending this argument, it seems like the only rational strategy would be one where at any number of points I have, my probability of failure (reaching 0 before 100000) is exactly 1%.
The problem is that when I try to think about this using backward induction, that seems like an impossibility.
If my strategy at 9990 - 9999 points each yield a 1% chance of failure, then my strategy at 9989 points must be an expectation over the latter 10 outcomes, with some additional uncertainty.
Why does it seem like maintaining a constant probability of failure as I play is both required by rationality (because once I actually get to 1000, I shouldn't care that my decisions now mess up my earlier odds) and impossible to construct?
00:48
why not change the failure rate so that it sums to 1% or 2%? you know 1% the first round, 1/2% the second round, 1/4% the third round, etc?
If that leads to an acceptable amount of winnings.
Otherwise, the cumulative probability of failure could get quite large. It all depends on the average win and the standard deviation of the win.
The reason I had in mind was that when I get to round 2 in your example trajectory, I will still have a riskofbankruptcy-speed tradeoff preference where I am willing to accept 1% risk, so I will prefer to continue with a faster, higher stakes strategy from that round.
Whatever my risk tolerance, 1% in the example, it seems like it needs to be a "fresh start property."
If I start the game with 100, I still prefer optimizing for speed at anything lower than 1% risk of failure. If I start with 25000, I still prefer optimizing for speed at anything lower than 1% risk of failure.
There should be an optimal substructure here I think. If you take a full game and chop some rounds off the beginning, it seems like it's still the same game with the same risk tolerance applying.
Remember that I am competing against other players to get 100000 before them, so that's why 1/4% from any state forward is bad.
 
1 hour later…
02:08
Hello, given two growth functions $\zeta(n), k\times g(n)$, where $k>0$ is a constant. Why $1 \lt \ \frac{\zeta(n)}{k\times g(n)}\lt 1$ is false? If you think this question is too much, I understand. But I am asking logically here.
Huh?
Why is $1<1$ false?
I am going crazy
Evidently.
Pray for me
please
I don’t pray.
02:32
Pray you don't pray
:)
@TedShifrin that is topology related
Can you work in combinatorial math (discrete math) topologically?
03:02
@Avra There might be a typo here. $1$ cannot be strictly lesser than and strictly greater than anything.
03:30
Hi, can someone please explain how the following is true:
"For any $c>0$ the relations
$$(x,y)\sim (x, \,y+ k c)\quad(k\in{\mathbb Z}), \qquad (x,y)\sim \bigl(x+\ell,\,(-1)^\ell y\bigr) \quad(\ell\in{\mathbb Z})$$
define a Klein bottle $K_c$ of "length" $1$ and "width" $c$ as a quotient of the $(x,y)$-plane , and with a rectangle $[0,1]\times[0,c]$ as fundamental domain."

I cannot see how this is the same as the definition I learned which is $(x,0)\sim (x,1)$ and $(0,y)\sim (c,1-y)$. Thank you!
03:46
hello, could i get help with this question? given an $\mathbb{R}$-algebra $\mathcal{F}$ we define $|\mathcal{F}|$ to be the set of homomorphisms onto $\mathbb{R}$. we say $\mathcal{F}$ is geometric when $\cap_{p\in|\mathcal{F}}\ker p=\{0\}$. let $V$ be a finite dim real vector space and $G:V\rightarrow V$ a linear map. we define $\mathcal{F}_G$ to be the algebra generated by $G^k$, $k=0,1,\dots$. the question asks to characterize the $G$ for which $\mathcal{F}_G$ is geometric
it seems if $P$ is the minimal polynomial of $G$ then $\mathcal{F}_G$ is $\mathbb{R}[x]/P$? i think this is right
in this case it seems like for e.g. nilpotent $G$, $\mathcal{F}_G$ is not geometric
or in general when $P$ is not irreducible?
for some reason i am very hesitant about these claims
sorry, i have a small typo in the definition of geometric, it should say $\cap_{p\in|\mathcal{F}|}\ker p=\{0\}$
i rather suspect if the claim $\mathcal{F}_G\cong\mathbb{R}[x]/P$ is true then $\mathcal{F}_G$ is geometric if and only if $P$ is irreducible, but i am not sure how to show it
however maybe we need only $P$ to be squarefree in the sense that no irreducible polynomial divides it more than once (surely if $\mathcal{F}_G$ has nilpotent elements they must all map to $0$ under any $p\in|\mathcal{F}_G|$, now we have nonzero elements always in $\ker p$, so $\mathcal{F}_G$ is not geometric)
04:54
okay, i am pretty sure $\mathcal{F}_G$ is $\mathbb{R}[x]/P$. We can let $\phi\mathbb{R}[x]\rightarrow\mathbb{R}[G]$ be the evaluation map, so by the first isomorphism theorem $\mathbb{R}[G]$ is really $\mathbb{R}[x]/\ker\phi$. if $\ker\phi$ is generated by $P$ (since $\mathbb{R}[x]$ is a PID) then $P(G)=0$. but if $Q$ has degree less than $P$ and $Q(G)=0$ then $Q\in\ker\phi$ but is not in $\langle P\rangle$. so $P$ has minimal degree
sorry, that should be $\phi:\mathbb{R}[x]\rightarroww\mathbb{R}[G]$
argghhhhh
$\phi:\mathbb{R}[x]\rightarrow\mathbb{R}[G]$
 
2 hours later…
07:24
My dog is a beagle alien, and I'm an eagle alien
3
 
2 hours later…
09:47
where can i find the english version to this article https://de.wikipedia.org/wiki/Dirac-Identit%C3%A4t#Anwendungen
when i search for diracs identity i do not seem to find anything on wikipedia
I am trying to find examples for this so i can understand it better, thank you
09:59
Take a sequence $a_n\to 1$, such that $0 < a_n < 1$ for every $n$. There exists $x_n$ such that $d(x_n,M) \ge a_n$ for all $n$. Since $$d(x_n,M) = \inf_{m\in M} \|x_n - m\|$$
$d(x_n,M) \ge a_n$ implies $\|x_n - m\| \ge a_n$ for all $n$. $S$ is compact, so $x_n$ has a convergent subsequence $x_{n_k} \to x \in S$. $$\|x_{n_k} - m\| \ge a_{n_k} \quad \forall m\in M, \forall k\in \mathbb N$$
Taking $k\to\infty$, $$\|x - m\| \ge 1 \quad \forall m\in M$$
Taking $\inf$ over all $m\in M$, $d(x,M) \ge 1$.
Above is my attempt for the following problem:
Could I please get help/hints in completing my solution?
 
2 hours later…
12:22
$\psi: X \times \Bbb R^*_+ \to X$ such that $\psi(x,1)=x$ and $\psi(\psi(x,e^t)),e^s)=\psi(x,e^{s+t}).$
this is a group action on the multplictative group of real numbers from what I understand
it looks almost exactly like the definition of a flow. Do I need more conditions to show that the above group action is isomorphic to a flow?
Definition of a flow: $\varphi: X \times \Bbb R \to X$ such that $\varphi(x,0)=x$ and $\varphi(\varphi(x,t),s)=\varphi(x,t+s)$
 
2 hours later…
14:37
You may restrict the equivalence relation to the fundamental domain, so the first space you describe is $[0,1]\times[0,c]$ with the identifications $(x,0)\sim(x,c)$ and $(0,y)\sim(1,c-y)$, whereas the second space you describe is $[0,c]\times[0,1]$ with the identifications $(x,0)\sim(x,1)$ and $(0,y)\sim(c,1-y)$. I suggest drawing pictures.
Both are rectangles with one side of length $1$ and one side of length $c$, where one pair of edges gets identified with the same orientation and the other pair of edges gets identified with the opposite orientation. However, which pair is the short one
@P-addict yes, this is correct
 
2 hours later…
16:21
I have a simple question regarding topology. Let $\tau$ be a topology on $\Bbb{R}^2$ such that all sequences satisfy
$$(x_n,y_n)\xrightarrow{\tau} (x,y) \iff x_n \to x, y_n\to y$$
where the latter convergence is in $\Bbb{R}$. Is $\tau$ necessarily the Euclidean topology?
I am almost certain that the answer is no and that the counterexample had something to do with discrete topologies but I simply cannot find the right keyword combination to search for it
I don't have an immediate answer, but such a topology would need to be non-sequential, which strikes me as rather peculiar
17:15
Certainly the discrete topology has that property.
convergence is boring though
I might be missing something, but a sequence converges in the discrete topology iff it is eventually constant. It should hold $$\left(\frac1n,\frac1n\right) \to (0,0)$$
since $\frac1n \to 0$ in $\Bbb{R}$, but the discrete topology doesn't satisfy this.
To clarify, the the equivalence above $x_n \to x$ and $y_n \to y$ is understood to be w.r.t. the standard topology on $\Bbb{R}$.
@mechanodroid you need to say that
I mentioned "where the latter convergence is in $\Bbb{R}$"
By convergence in $\Bbb{R}$ I meant convergence w.r.t. the Euclidean topology, no other topology on $\Bbb{R}$ is mentioned
Yes, you're right.
17:26
well you changed the topology on the left side on $\mathbb{R}$, and from the responses, it seems that people were assuming you'd done the same on the right
So you're saying that the sequence converges in the usual product topology, which is equivalent to the Euclidean topology.
@robjohn He had it written correctly in the question. I screwed up.
The "discrete counterexample" I may be remembering might actually be the one you mentioned, which means that it wasn't the same problem
@TedShifrin okay
Yeah, I think I just explained it. Your criterion says usual convergence in the product topology. So we're done.
@TedShifrin Yes, $\tau$ is basically sequential product topology
But I'm worried about the convergence of nets
17:28
But you have the usual topology in each coordinate.
Yes, but if $x_j \to x$ and $y_j \to y$ are convergent nets in $\Bbb{R}$, why would $(x_j,y_j) \xrightarrow{\tau} (x,y)$? We know this only for sequences
that doesn't tell us much about nets, does it
$\tau$ might not be sequential
@Thorgott Yes, exactly
Do we care? You're defining the product topology based on a topology that is given to be the usual topology.
the topology is not being defined
it's just assumed to have that property
17:31
OK, fair enough. Have fun :P
@Thorgott Exactly
 
1 hour later…
18:33
From Wiki: locally, each covering map is 'isomorphic' to a projection in the sense that there is a homeomorphism, $h$, from the pre-image $p^{-1}(U)$, of an evenly covered neighborhood $U$, onto $U\times F$, where $F$ is the fiber, satisfying the local trivialization condition, which states the following: if $\pi \colon U\times F\to U$ is the projection onto the first factor, then the composition $\pi \circ h:p^{-1}(U)\to U$ equals $p$ locally (within $p^{-1}(U)$).
Could anyone give me a hint of how to prove this/show $h$ is a homeomorphism?
19:01
I don't get what you're trying to show, this is a definition
also not sure if you saw it, but I responded to your Klein bottle question earlier
Oh yes I saw it, thank you!
I'm trying to show that a covering map has this local trivialization condition
but that's the definition of a covering map
Presumably you have an actual example of spaces and a map?
The definition of covering map I have in mind is where the preimage of admissible open neighborhood U is a disjoint union of open sets each which map homeomorphically onto U.
So from that description, write down $h$.
19:10
$h: p^{-1}(V(a))\to V(a)\times F$ ?
I don't know where $V(a)$ came from, but yes.
That just what it says from the wiki
I changed U to V sorry
You are told you have a disjoint union of sets each homeomorphic to $U$.
Yes, that's from my definition
So how do you write down $h$ from that?
19:15
From my definition?
You are also confusing $\pi$ and $p$, apparently. Do you have an explicit picture in front of you?
Yes I tried to draw it, so $\pi: V(a)\times F\to V(a)$ and $p$ acts on V(a)
Oh, I see, right, $\pi$ and $p$ really are both here. The point is that $p$ looks like $\pi$ once you use the homeomorphism $h$, yes. No, $p$ does not act on $V(a)$.
$p$ acts on $p^{-1}(V(a))$?
Right.
19:23
i think have simplified my earlier question down to the following: suppose $P\in\mathbb{R}[x]$ is squarefree and reducible, i.e. the product of distinct irreducibles $P_1P_2\dots P_n$, $n\geq2$. for any $Q\in\mathbb{R}[x]/P$, does there exist an algebra homomorphism $p:\mathbb{R}[x]/P\rightarrow\mathbb{R}$ such that $p(Q)\neq0$? i'm not really sure where to go from here. hopefully i am not spamming my question too much
Is it true that if $p:V_\alpha\to V(a)$ is a homeomorphism, then $h:\bigsqcup V_\alpha\to V(a)\times\{x_\alpha\}_\alpha$ is?
You mean $p|_{V_\alpha}$, of course. You still haven't defined $h$. You have to define it.
@P-addict You really need to write $\Bbb R[x]/(P)$ or $\langle P\rangle$, or something!
ah sorry, you're right!
And $Q$ is not right. You mean its equivalence class.
Is $h:(V_\alpha)\mapsto (V(a),x_\alpha)$ valid?
19:27
Does that make sense, @pritchard?
It seems right to me and also a homeomorphism but I wouldn't know how to prove it
Do a concrete example, @P-addict. I don't understand what you know about $Q$.
@P-addict you can explicitly write down all algebra homomorphism $\mathbb{R}[x]/(P)\rightarrow\mathbb{R}$
How does $V_\alpha$ turn into $V(a)$ magically, @pritchard?
do so by explicitly writing down all algebra homomorphism $\mathbb{R}[x]\rightarrow\mathbb{R}$ and checking which of these factor through the quotient
19:29
By the map $p$?
Well, then you need to write that in your definition of $h$, don't you?
@TedShifrin sorry, i may have been unclear. the question was asking for which $P$ does every $[Q]\in\mathbb{R}[x]/\langle P\rangle$ have this non-vanishing property? so we need $Q$ to be arbitrary
What if $[Q]=[0]$?
@TedShifrin I'm really not quite sure. Do you mean I should write h as a composition of maps?
oops! every nonzero $[Q]\in\mathbb{R}[x]/\langle P\rangle$, my mistake
19:38
Define $h$ on points in its domain, @pritchard, the way you always define functions.
Do you mean write $h$ as a function of $V_\alpha$?
@Thorgott ahh, i was thinking about this but i failed to characterize every such map. i suppose it's something like this: restricting the morphism to the constants we get a morphism of algebras $\mathbb{R}\rightarrow\mathbb{R}$. in particular this map is linear so it is multiplication by a constant $a$, but if $f(x)f(y)=f(xy)$ for all $x,y$ then $a^2=a$, so $a$ is $0$ or $1$. if it is $0$ then the whole map (on $\mathbb{R}[x]$) is trivial, else it is evaluation at wherever $x$ maps to
so we are concerned with maps evaluating at roots of $P$
precisely
though the fact that the morphism of algebras $\mathbb{R}\rightarrow\mathbb{R}$ is the identity is immediate from the definition of a morphism of $\mathbb{R}$-algebras
(and the $0$ map is not a morphism of $\mathbb{R}$-algebras)
oh dear, that was a silly mistake, you're right
okay! so if i am understanding this right, $\mathbb{R}[x]/\langle P\rangle$ doesn't have the property i want when there is some nonzero $[Q]\in\mathbb{R}[x]/\langle P\rangle$ with the following property: when $r$ is a root of $P$, $Q(r)=0$
19:54
indeed
write down the intersection of all kernels explicitly and this translates to a condition on $P$
this means $P$ must be the "smallest" polynomial which has the roots it does: it must be $(x-r_1)\dots(x-r_n)$ for distinct $r_i\in\mathbb{R}$, else it is this polynomial times another with no new roots, and then $[(x-r_1)\dots(x-r_n)]\neq[0]$ will always vanish at evaluation at roots of $P$
yup
now if you remember that your $P$ was the minimal polynomial of some linear operator on a finite-dimensional vector space, this condition translates into a condition on that operator
20:11
oh. and we also know these are all the eigenvalues, right? since if $\lambda$ is an eigenvalue with eigenvector $v$ not one of the $r_i$, then $(G-r_iI)(cv)=\lambda cv-r_icv=c'v$, $c,c'$ nonzero constants, so sending $v$ through the composition of the $G-r_iI$ won't spit out $0$, so $P(G)$ is not the $0$ map, contradiction
i mean $\lambda$ is not one of the $r_i$, that's a bit unclear
20:31
yes, the roots are precisely the Eigenvalues
but the minimal polynomial splitting into distinct linear factors is equivalent to another well-known condition
20:48
i'm not entirely sure. if we extend a set of independent eigenvectors to a basis the other vectors in this basis can be chosen so they map to linear combinations of each other. so we can write $G$ wrt to this basis with the eigenvalues along the diagonal for the first few columns (however many eigenvectors $G$ has), $0$s elsewhere in these rows and columns, then the remaining lower right square matrix must be $0$ under the same polynomial $P$? i think
oh but then it must have those eigenvalues, so we can do the "same thing" for that matrix and so on
OH $G$ must be diagonalizable?
*it must have eigenvalues among the $r_i$ (but not necessarily all of them will be)
because we can write $G$ as $\begin{pmatrix}D&0\\0&A\end{pmatrix}$ for some diagonal matrix $D$ (with the eigenvalues), so since $G-\lambda I$ will be of the same form, since $I$ is diagonal, and since $\begin{pmatrix}D_1&0\\0&A\end{pmatrix}\begin{pmatrix}D_2&0\\0&B\end{pmatrix}=\begin{pmatrix}D_1D_2&0\\0&AB\end{pmatrix}$ (i think) we necessarily have $P(A)=0$, so $A$ can be written in the same way just described but due to finiteness this cannot repeat indefinitely
i think this proves $G$ is diagonalizable but i'm sure there's a nicer way to conclude the proof than what i just did
i guess just induction on the dimension of $V$ (where $G:V\rightarrow V$) should finish this proof in a cleaner way
21:20
I do not believe this arguments works as given
it is not clear a priori that such a basis exists
stable subspaces don't always have stable complements
but the statement is true: the minimal polynomial splits into distinct linear factors iff the operator is diagonalizable
oh shoot, i assumed $G$ had only nonzero eigenvalues, that was my mistake
21:33
@robjohn , is there any material regarding little-o that you can recommend? I appreciate any suggestion.
21:50
Let $P(X)$ be the set of 1-element and 2-element subsets of $X = \left\{1,2,3\right\}$, i.e. $P(X) = \left\{\left\{1\right\},\left\{2\right\},\left\{3\right\},\left\{1,2\right\},\left\{1,3\right\},\left\{2,3\right\}\right\}$. Let the group $S_3$ act on $X$. How many subsets of $X$ does the permutation $(12)$ fix? Is it correct to say it would fix $\left\{3\right\}$ and $\left\{1,2\right\}$ so $2$?
@Joey, yes.
Thanks, Ted.
Are you looking at orbits/stabilizers?
if $f,g:V\rightarrow V$ are linear, $V$ finite-dim, $g\circ f=f\circ g$, and $\ker f\cap\ker g=\{0\}$, can we express $\ker(g\circ f)$ using $\ker f$ and $\ker g$? maybe their direct sum
Is $\ker(g)$ really a subspace of $\ker(g\circ f)$?
21:54
should be? since $g\circ f=f\circ g$ and $\ker g$ is a subspace of the latter
*subspace of the kernel of the latter
Ah, right.
Is there a parametric form available for the Fernandez-Guasti squircle?
@TedShifrin Yes. Characters/representations.
Ah, cool.
22:11
Actually, better question. Suppose I have the solution set of $x^2 + y^2 - s^2 x^2 y^2 = s^2$. If I have a point in the solution set as $(u, v)$, then what is $s$?
I want to use $s$ as a linear interpolant for a Gaussian kernel to create a cool glowing edge effect programmatically.
To be more specific, if a point $(u,v)$ on a plane for $u,v \in [0.0\text{f}, 1.0\text{f}]$ (and suffix f denotes an IEEE 754 floating-point value) belongs to the squircle of radius and squircle parameter $s$, apply a brightness based on $s$ directly or as the input to a 2D Gaussian function.
22:31
@schn I've never seen anything dedicated to that. It is usually just a definition of the error term.
22:57
@P-addict $g\circ f=f\circ g$ implies that $f$ restricts to a map $\ker(g)\rightarrow\ker(g)$ and $\ker(f)\cap\ker(g)=\{0\}$ implies this is injective, whence surjective by finite-dimensionality, i.e. $\ker(g)\cap\mathrm{im}(f)=\ker(g)$. This implies that $\ker(g\circ f)=f^{-1}(\ker(g))$ has dimension $\dim\ker(f)+\dim\ker(g)$, forcing $\ker(f)\oplus\ker(g)\subseteq\ker(g\circ f)$ to be an equality.
there's probably a cleaner argument, but sometimes brute force suffices
I feel dumb. I managed to leave in the $s^2$ coefficient on the LHS even though it's supposed to cancel out.
ahh, it seems so straightforward now. thank you so much! this allows us to finish the earlier argument, i believe: the $(G-r_iI)$ commute and have disjoint kernels, so associativity should allow us to extend $\ker(f)\oplus\ker(g)=\ker(g\circ f)$ to $n$ functions, i think? then the kernel of the composition of the $(G-r_iI)$, which is $V$, is also the direct sum of the kernels, which implies $G$ is diagonal! is this right?
this is correct
you actually don't need the full strength of this observation, though
you always have $\dim\ker(g\circ f)\le\dim\ker(f)+\dim\ker(g)$ for any $f,g$, which suffices for this argument
oh
my goodness
LOL i'm so dumb, that's so much easier
i think that solves that problem. thank you so much, Thorgott, Ted! sorry for dragging it on so long, i really appreciate the help. :-)

« first day (4021 days earlier)      last day (1297 days later) »