He writes that $C^\infty(M_1\times\cdots\times M_k)$ and $C^\infty(M_1)\oplus\cdots\oplus C^\infty(M_k)$ are isomorphic by the "universal property of the direct sum"
My proof uses the finiteness of their dimension as a crucial element @TedShifrin
Is it still true in the infinite-dimensional case (technically not called manifolds anymore but we can generalize)?
(I found a map from left to right called $f$ and a map from right to left called $g$ and show $f\circ g$ is the identity, which combined with the fact that they're linear maps between spaces of the same dimension is enough)
If we have charts to vector spaces $V$ and $W$ I think we can do $T(M_1\times M_2)\cong T(V\times W)\cong V\times W\cong T(V)\times T(W)\cong T(M_1)\times T(M_2)$
I suppose it specifies at most one of the Ms is a manifold with boundary because the product of two of those has corners and that's not a valid sm.man.w/bd
There are three other instances of this (words being "miswritten" but pronounced correctly) in that chapter
@Ajay We can prove mathematically that, if a mathematically ideal computer ran that code and produced that output, then the theorem is true
We cannot prove mathematically that any given physical computer behaves like a mathematically ideal one
but it certainly inspires confidence
There's a more thorny philosophical issue with "probabilistic proofs". There's a certain primality testing algorithm,
that works like this: every composite number has certain "witness numbers" to its compositeness
For a composite number C and a witness W between 1 and C, there's some function you can compute that if it gives a certain answer means C is a composite
Primes have no witnesses, while 3/4 of the numbers between 1 and a composite number are witnesses
So here's the algorithm: select a whole bunch of candidate witnesses at random and check them in turn. If none of them work, then the number is probably prime
@Ajay It seems that it would depend upon whether you're confident that whatever axiom set the system is using corresponds with the one(s) people use, and that whatever inference rules the system is using are valid, and that the system is sufficiently error-free.
You can make that probability as high as you want by choosing more candidates, but you can only get certainty by checking literally all numbers between 1 and P
@Ajay If I run this test 100 times, the odds of it falsely claiming a composite number is prime is (3/4)^100 which is like 3*10^-13
My understanding is that the current formal verification state-of-the-art is still a decent clip away from being able to formally verify a lot of things, and certainly a long way away from producing its own theorems, but there also appears to be quick progress in the former regard, at least.
(And here, what I'm talking about is distinct from the sort of "lots of case-checking" proofs that are the typical examples of computer-aided proofs, like the one for the four-color theorem. That's mostly just an issue of "do people trust the people who gave the computer the inputs and the algorithm to have done that properly")
At least to my eyes, there's a difference between "formally verifying Perelman's proof of the geometrization conjecture in Lean" and the original four-color theorem computer-assisted step.
Ah, I see - the former was all comprehensible to humans (or at least one human) but we'd like to automatically check for errors, whereas the latter was a massive search
Yeah, that's basically what I mean. It would be helpful to be able to take some kind of research preprint and slap that into Agda or Coq or Lean or whatever, but as I understand it, the state of the art is such that that's only possible in narrow cases, usually ones that are already relatively easy to "algorithmize" (for lack of a better word).
Four-Color: "How can you call it a proof when humans can't understand it?" Perelman: "How can you call it a proof when a computer hasn't checked it?"
Both are valid questions but in very different directions
@Fargle There have been some high-profile formal verifications. A year and a half ago, they formally verified in Lean the proof that the Continuum Hypothesis is independent from ZFC
(The type theory of Lean is as strong as ZFC+countably infinitely many universes, I think?)
Here's a neat puzzle: Given that computer proof verification exists, and given that it's possible to write computer programs that have access to their own code (it's a neat puzzle to show that you can do this, by the way),
part 1: Suppose we write a program R that searches for proofs in Lean that R doesn't halt, and if it finds one it halts. What does R do?
part 2: Suppose we write a program F that searches for proofs in Lean that F does halt, and if it finds one it halts. What does F do?
@Fargle Yeah I can't give you a timeline but I feel like it's on its way
Basically we need to (a) fill Lean up with enough libraries that it has the standard definitions of nearly every field built-in and (b) teach undergrads Lean
We teach undergrads coding already, (b) isn't too difficult
(a) is a legit multi-year research project but it's happening super quickly I think
Another question: How would I find the number of solutions to the equation $$4^x = y^2 + 15$$ By plugging in values i've found that there are 2 pairs of solutions, but that doesn't seem so efficient.
math.stackexchange.com/questions/420642/… @robjohn I understand that the radius of convergence is greater than $1$. But how that implies $|\sum_{j=0}^n\binom{n}{j}(-1)^{n-j}z_0^{n-j}a_{k-j}|\leq cr^k$ for some $r$?
The key insight is that $4^x$ is a perfect square, so we can transform the problem from being about some unknown power to a quadratic. And then see if the solutions to that quadratic give us any $z$ which are powers of 2.
@copper.hat no sorry my mistake there was nothing abigous.
I made a thinking error according to the union since I did not take the union over $n$ but I wanted to take the union over $x_n$ in $\Bbb{Q}$ therefore I got confused with the cardinality but now it is clear. Sorry for the circumstances:(
@onepotatotwopotato $\sum\limits_{j=0}^n(-1)^{n-j}\binom{n}{j}a_{k-j}z_0^{n-j}$ is the coefficient of $z^k$ in $f(z)(z-z_0)^n$, and since the radius of convergence of $f(z)(z-z_0)^n$ is greater than $1$, the root test shows that there is an $r\lt1$ so that $\left|\sum\limits_{j=0}^n(-1)^{n-j}\binom{n}{j}a_{k-j}z_0^{n-j}\right|\le cr^k$.
@anak Technically, it is a 35 minute talk, with five minutes of transition time.
(Technically, it is two 15 minute talks, back-to-back, with five minutes between talks. Which is why it is 35 minutes, which is a slightly weird amount of time.)
Anyway a sentence like $\exists A_1,\ldots,\exists A_4$ such that $\bigcup A_i$ is everything and $\forall B_1,\ldots,\forall B_4$ if $\{B_i\}$ is a cover refining $\{A_i\}$ there are $i_1,i_2,i_3\in\{1,2,3,4\}$ with $B_{i_1}\cap B_{i_2}\cap B_{i_3}\neq\varnothing$ can be written in your language, holds in $\Bbb R^2$ and fails in $\Bbb R$
Or in words you can cover $\Bbb R^2$ with four sets, such that any refinement of that cover also consisting of $4$ sets is such that three of them have nonempty intersection (so witnessing that $\dim\Bbb R^2\geq 2$)
@XanderHenderson What kind of stuff gets talked about by community college math folk usually? I am quite ignorant of how the community college system works, and I didn't even realize they would have conferences for that.
Now if you can figure out the smallest size of a cover of $\Bbb R^3$, such that any subcover of the same size witnesses that $\dim\Bbb R^3\geq 3$, you can distinguish $\Bbb R^3$ and $\Bbb R^2$ @Akiva
But that is a comment made out of ignorance. I genuinely don't know the statistics.
@anak I'm not convinced that active learning is much of an improvement.
I am convinced that it is a useful technique for those people for who it works (which is tautological), but it is not very scalable, works poorly in remote / hybrid environments, and requires a lot of buy-in (both on the part of the instructor and on the part of the students).
So I guess you have seen all the studies that suggest it increases performance, but have concluded this is probably because the surrounding factors were just right for it to happen?
@anak In part. It is also worth noting that the studies are conducted in environments where the instructors have a lot more autonomy with respect to course content and practices than a lot of us have.
There are certain topics which I must cover (because of our articulation agreements with the universities). If I don't cover those topics, the classes don't transfer. But active learning techniques usually take more time to cover the same material.
@anak Depends.
But there is also a common effect in studies regarding education (and other social programs): the people who are engaged in the study are highly motivated. They want their techniques to be successful, hence they are likely to put a lot of time and effort into making the program successful.
The kinds of effect reported in such studies tend to shrink significantly when you try to scale them, or when you ask others to implement the same ideas.
And, to be fair, I have been interested in "active learning" since it was called "the Moore method". There are things to be taken from it. But it is not a panacea.
I've found both large and small classes to be effective for active learning. I do agree about your online sentiment, though, most students taking online courses don't want to be engaged.
a lot of K-12 stuff fails because there are lot of really good general ed research notions that founder when the instructor doesn't know math, and the instructor generally doesn't know math. it's like trying to teach a language that you can't speak. the latest and greatest techniques will not help
I am strictly adhering to the rule that I should not cancel any terms in a trig equation, but I just looked up a solution where terms were cancelled though there was a possibility of them being zero
community college is kind of a middle ground where the subject matter knowledge is there but often the students lack motivation, maybe more than high school students do
i couldn't explain why math X is a prerequisite for so many things in college, it shouldn't be
thunder i think often there is an unstated rule that a trig identity is an identity if it holds at all but a discrete set of points. once tan or sec get in there for example you are going to have those points.
About 15 years ago, we contracted with Cisco to create "connected classrooms", which are classrooms with a lot of cameras and screens, which enable remote (but not "online") courses.
@ThunderGlove An identity is an equation.
Not all equations are identities, but all identities are equations.
But it would be helpful to have an example which exemplifies your confusion.
@anak In an "online" course, students are expected to attend courses remotely, from whatever location they choose, using their own equipment, over the interwebs.
"Remote" just means that students and instructors are not necessarily in the same room.
All online classes are remote, but not all remote classes are online.
Again, "online" doesn't mean "it uses the internet". It has a specific meaning of every student connecting to a course one their own, possibly asynchronously.
For the remote classes that I teach, I am in one classroom with a group of students, and am simultaneously being broadcast to other classrooms with their own groups of students.
@XanderHenderson "online course", in the most common context I hear it, means a course where classroom content (assignments, lectures, etc.) is distributed and accessed primarily (if not solely) online.
@XanderHenderson This makes sense now, though, thanks!
If I were writing that solution, I would have started with something like "The equation is only meaningful if... . Therefore, assume that these conditions hold. The equation becomes...".
Amongst mathematicians, reading math books is considered the best way to learn, right? (I mean the best medium of course, there's no substitute for practice)
Besides, different people learn in very different ways. Some people absorb lecture material and interact in class in a lively fashion; others just find the classroom experience overwhelming and would rather read on their own.
I always liked learning from lectures, getting insights from the prof (assuming a good prof). The occasional disastrous lecturer (such as Irving Segal teaching out of his own book for graduate real analysis) was not a good experience.
@ILikeMathematics Generally speaking, you can pack more details in a book without boring learners than a lecture, since readers can skip details, while lecturees fail to be able to escape the narrow corridor of time.
Once one gets to advanced mathematics, some texts are so dense and sophisticated that one basically has to know the material before one can read the text.
@leslie Yes. Dare I tell the OP about the geometric series? Yikes.
@leslietownes There is a professor at UCR is generally considered to be a great lecturer. I went to a lot of his talks, and always felt like I was really understanding everything as he went along. And then I would go home and try to consolidate my notes. At that point, I would learn that I had understood nothing, that most of the lecture was entirely vacuous, and that I had several hours of reading ahead of me. :/
But the lectures were energizing! And you felt like you were learning something!
I had that same experience with Atiyah and with Bott when I heard them lecture. Both stupendously clear and provided great intuition. I think the moral of the story is that one needs to take notes in math lectures; saying "oh, I'll never forget a word of this" is hopeless.
Please explain the first comment, I don't understand and the only thing I've done is reversing the sin(z) expansion? Cant see that I've broken any rules? — zzz__5 mins ago
That's the thing that makes math so challenging, I think. It's also the reason that I think that linear algebra and algebra are easier than baby analysis — fewer quantifiers!
I wonder also if writing sentences like that OP using symbols is partly the problem. Writing out for all and there exists makes one understand a little better what is being said.
Fair point. There's the tendency to not worry about simple stuff when you're doing advanced stuff. OTOH, you can't build a solid advanced structure if your foundations aren't solid.
Solving quadratic Diophantines can get a bit messy. But it is kinda satisfying. I can't remember all the tricks. But I know where to find them when I need them. :)
In general relativity, a congruence (more properly, a congruence of curves) is the set of integral curves of a (nowhere vanishing) vector field in a four-dimensional Lorentzian manifold which is interpreted physically as a model of spacetime. Often this manifold will be taken to be an exact or approximate solution to the Einstein field equation.
== Types of congruences ==
Congruences generated by nowhere vanishing timelike, null, or spacelike vector fields are called timelike, null, or spacelike respectively.
A congruence is called a geodesic congruence if it admits a tangent vector field...
@MoreAnonymous I don't do relativity, so let's forget about that. Just do the question in $\Bbb R^3$ to start with. You're starting with two skew lines and asking how you fill up space with lines? This has nothing to do with differential geometry.