[applause and cheering]
[applause] Thank you very much. It is good to be here. I don't know if you people realize what a beautiful campus you have and how you basically just
study in heaven. Today I want to talk with you about what I think is one of the most underappreciated formulas. Not because those who know it don't
appreciate it, but because not enough people know about this and more importantly, not enough people
understand where it comes from. And this should be one of the gems. This should be the e to the pi eye of the mathematical community. But I really
want you to come away knowing not just what it is uh but why it's true and what it represents. Now before I dive straight into it, I think the scene
is best set if we start with two different puzzles. Okay. The first puzzle is just going to give us a
sense of the meaning of what we're about to do because it runs the risk of feeling pretty abstract. And then the second puzzle is going to be a kind
of forewarning uh not to trust our intuitions. So, let's start with the first puzzle. This is our warm-up for today. We're going to start by imagining
that there's a random number. I'm going to call it X. It's going to sit between negative 1, positive
1, chosen uniformly. X is going to have a friend. Another random number Y chosen between negative 1 and positive 1 uniformly. And we're going to ask a
question, a probability question about both of these, which is what's the probability that if you square X and then you add it to the square of Y, you
end up with something which is less than or equal to one. So, let me throw this out to the
audience. Does anyone want to raise their hand and uh throw out probability that you choose this random X, choose this random Y, you add their
squares, it's going to be less than one. Let's see. So, I see a blue shirt up here. Yeah. Do you want to pass it out? >> Um, I mean x^2 + y^2 makes
like a unit circle. So, my guess would be it be related to the area of a unit circle over the area of like a
unit square. >> Amazing. So I asked this question and immediately in your mind there's an image that pops up. And the point I want to make is this is
an analytical question. This is just a question about two numbers. If you're savvy with your probability, you could set up some integrals not thinking
about geometry and answer the question, but that would feel a little bit like a crime. The question
is kind of screaming out at you that it wants to be a picture. The natural way that this wants to be answered is to think of that pair of numbers x
and y as a point in a two-dimensional space. And that way when you're thinking of a random choice for x independent from a random choice of y, you're
choosing a random point inside. Well, in this case, it's a square that goes from x=1 to positive 1.
Same with y. So it's a 2x2 square. Now the reason it's screaming out to be geometric is because this constraint of x^2 + y^2 being less than equal to
1 is the meaning of being inside a circle, right? It's saying if x^2 + y^2 um is equal to 1, that would actually land you on the boundary. This
follows straight from the Pythagorean theorem. You can draw a nice little right triangle. X gives you one
leg. Y gives you the other leg. And you end up answering in this case area of that circle divided by area of the square. All you need to know is the
area of a circle. P R^ square. This is one of those formulas we learn in high school. So you can write down pi / 4. All well and good. And we could
bump it up. I could say, what if instead of choosing two random numbers X and Y, they had a third
friend Z. Same deal. Uniform between negative 1 and 1. You square it. You want to ask the probability that the sum of all those numbers is less than
or equal to one. In your mind's eye, it's screaming to you, hey, I want to be a three-dimensional picture. Now, it doesn't have to be. You could
answer it purely analytically, but it would somehow feel wrong not to turn this into a threedimensional
picture because in this case, that constraint of x^2 + y^2 + z^2 being less than 1 means sitting inside of a unit sphere. And so in this case, to
answer the probability question, you just need to know what's the volume of that unit sphere as it compares to the volume of this 2x2x2 cube. This is
another one of those formulas we might have learned in middle school or high school. Volume of a sphere
ends up being 4/3 pi * its radius squared. Radius in this case is one. You work it all out and you get, I guess, a little above 50%. Now, there's no
reason that we'd have to stop there, right? This is three numbers, but what if I asked about four numbers? Now, at this point, your mind's eye is
going to start kind of squinking a little bit because it's like, I know what I want to be seeing right
now. In the same way that the first problem screamed out that it wants you to be thinking about a circle, this one wants you to be thinking about
whatever the four-dimensional equivalent of circles and spheres should be. But again, it's a perfectly reasonable question. It's a very empirical
question. It's very real. And I don't have to stop there. We could say what if you had a hundred different
numbers you add up all their squares and I want to know the probability that they add up to be something less than one. To ask that question is to
kind of force yourself to want to know the volume of a 100dimensional unit ball. Now that might seem kind of weird like if you just heard a
mathematician off the street saying that they found some formula for the volumes of highdimensional balls. That
risks sounding like something which is just nonsense, right? It's so divorced from reality because we're used to living in three dimensions. But the
point I want to make is that higher dimensional geometry is real. Just because it's not describing a literal physical space doesn't mean that it
doesn't have a kind of utility. Now it probably doesn't need to be emphasized nowadays because um machine
learning is a wash in high dimensional geometry. Essentially anytime that you're representing something with a long list of numbers, you could choose
to interpret that as a point in a highdimensional space. And one of the big lessons of geometry is that's a fruitful choice to make. It helps your
problem solving. It makes certain things that would have been really hard, like setting up a hundred
different integrals for this probability question, turn into something easier. I don't think I need to emphasize this to you because these days we
have things like large language models. And if you dig under the hood and you say, "What's really going on?" When you interact with, you know, chatbt
or claude or Gemini or whichever one your favorite one is, the under the hood scheme basically looks
like breaking up all of your text into little chunks. And each of those chunks gets turned into a long list of numbers. And you don't have to think
about that as a point in high dimensional space. You could just say, "Oh, it's this big long list of numbers." But to try to understand what these
models are doing, it ends up being really fruitful to think about those long lists of numbers that the
words turn into as being points in a space. When people do research to interpret these things, all of their language is centered around talking about
points in some kind of highdimensional space. That's all to say high dimensions are useful. But this is a little bit of a problem because
higherdimensional geometry can get very counterintuitive. And the second puzzle here is just an absolute
classic. This is one of those things where I think everybody sometime, you know, if they're entering a STEM field going into college, they should see
this example at some point as a kind of like ASOPs fable of something you might want to avoid. So here's going to be the setup. We're again going to
have a 2x2 square centered at the origin. Goes from x=1 to y= pos1. I'm going to put a circle at each
one of its corners. So the corners are sitting at 1 one negative 1 one negative 1 negative 1 on and on. I'm going to put a unit circle on each one of
those. And so by definition those corner circles have a radius of one. And the question I'm going to ask here is to say what if I was to give them
another friend, a circle that sits on the inside that gets as big as it can to be tangent to all of
them on the outside. What's the radius of that inner circle? Just a geometry question. Just a little puzzle. We're still in the warm-up phase. Again,
anyone want to raise their hand and kind of toss out a proposed answer here? All right, we got a raised hand over here. >> You can take a look at the
diagonal, calculate the length Pythagoras, and then subtract from that the R twice. >> Love it. So,
we're going to look at the diagonal from the origin to the corner of that cube. And that looks like square root of well, in this case, the legs are 1
and 1. So, square root of 1 2 + 1 2 that distance is square of two. And then we're going to subtract off the radius of that outer circle. Right?
Subtract off the one that was by definition radius 1. So the answer square of 2 minus one. You work it
out. It's about 0.4. You look at the picture. You say, "I don't know. Seems reasonable. Why is this an interesting question?" I say, "Wait, wait,
wait. You'll see. You'll see." So then you go to three dimensions, and we're going to do the same setup. We're going to have a unit ball on each one
of the corners. So in this case, we're going to have not four, but eight different unit spheres. Each one
of them has a radius one. They're sitting on all of the corners. And we're going to play the same game. I say, introduce a new sphere centered at the
origin that is as big as it can be, meaning it gets as big as it needs to be to be tangent to those eight outer spheres. Same question. What is the
radius of that inner sphere? In this case, anybody want to raise their hands, toss out a proposed
answer to the radius of this one up close. >> Roo3 - 1. >> Roo3 minus one. Okay. So, so you say root3 because you understand Pythagorean theorem in
three dimensions. I might belabor the point here. Uh, forgive me if this is telling you something I already know. I really want to emphasize why the
distance rule is what it is because the result I'm about to get to is so counterintuitive that when
I've put it online before, there have been comments to the effect of maybe distance in higher dimensions just doesn't work the way that we think it
does. Um, what I want to emphasize is everything's just stemming from the rule in two dimensions. So if you have a unit cube, you can play the same
two-dimensional Pythagorean theorem game because having that diagonal of one square, that could be one
leg of a new right triangle where another leg is this zaxis. And then from there you do the normal two-dimensional Pythagorean theorem. So we're not
imbuing 3D space with a new rule. It's more like it's inheriting the rule from all its two-dimensional subspaces. You work it out, you end up with
roo3. So in this case, that means the final answer roo3 minus 1. Again, you do the calculation. Looks to
be about 0.7. You say, "Seems reasonable. Why is this interesting?" But let's just keep upping it up in four dimensions. I don't know how to [clears
throat] draw a four-dimensional cube. This is my best attempt. It's something that would look like a cube, but it's got another cube as the top cap
and some lines connecting the vertices. But we know how to calculate the distance. The exact same logic
is going to let us go up one more where you would be drawing a two-dimensional right triangle where one leg is that root three. That's the diagonal of
a unit cube. and then you just step up in some fourth direction. Now, this one's kind of cute because the distance to the corner is um square root of
four, which means it's a whole number. It's two. And in this case, whatever that diagram should
look like in four dimensions. It's got 16 different corners, 16 different unit balls. One thing that's really nice about four dimensions is that a
unit ball fits nice and snugly right in the middle of them. So, this gives you a little instinct for what sphere packing in four dimensions can look
like. Now, where this gets more interesting is if we think in generality. If we say okay, not just for
three or four dimensions, but for n dimensions, what would this look like? The distance to the corner of that cube, it's going to look like roo of n.
We're always subtracting off the radius of that corner sphere, which is by definition 1. So, for example, up in 10 dimensions, it looks like 10 - 1,
which is about 2.16. Now, if you just look at that and you kind of nod along with the formulas, you
might say, okay, seems reasonable. I guess that's what things would be. But it gets really weird when you try to interpret it geometrically and
physically because here's what I'm going to do. I'm going to draw a new box outside of the entire diagram that we had. So before we introduce that
center circle, I'm going to draw this 4x4x4 bounding box outside of everything. And if in two dimensions I
said, "Notice that inner circle is inside all of those." Or in three dimensions the inner sphere is inside of that whole bounding box, you'd say,
"Yeah, obviously." Like it seems like by definition it must be inside of it. But the weird part is the distance from your origin to the edge of that
bounding box because it's 4x4 ends up being two to get to that edge. So if we're saying up in 10
dimensions, whatever this diagram should look like with its 2 to the 10th different corners, whole bunch of little unit balls up there expanding some
sort of inner sphere, that inner sphere is actually dramatically bigger than all your corner spheres. It's even bigger than that bounding box. It
pokes outside the bounding box. Now, that's just bizarre. And it's so bizarre. Um there's this common
phrase that I've heard from other people who have described this problem where they say uh this suggests that higher dimensional spheres are very
spiky. I don't know if anyone's heard this intuition would say you know high dimensional spheres are spiky. I hate this because like they're not
they're they're very round. They're round because they are defined to be round. What is a sphere in a high
number of dimensions? It's all of the points that are a given distance from the origin. So it is as round as round could be. Spheres are not the
problem. This is counterintuitive, but spheres, they're fine. Do you know who's the problem? It's the cubes. Cubes are messed up in higher dimensions.
At the very least, they defy our intuition. Cuz what's really going on with this example is we've got
this very highdimensional cube where to get from the origin to the edge of it, that point that a highdimensional sphere on the inside could bounce
outside of, let's say this is a 2x two box, it's just going to be a distance one to get to the edge of that 2x2 box. But to get to the corner, that's
actually a lot longer because up in 10 dimensions, to get to that corner, you've got to walk in your
first dimension, then your second dimension, then your third dimension, then your fourth dimension, then your sixth, on and on and on 10 times. Now,
even if you're not taking perpendicular steps and you take the shortcut to just go straight across, you know, the shortcut can only get you so much.
Um, so it gets you a square root speed up, but not anything better than that. So the very loose
schematic picture I have in my head for highdimensional cubes and take this with a grain of salt is it's something like this where your corners are
just way way farther away than the edge. Okay, again grain of salt because it's not like their uh sides are curved but just cram this intuition into
two dimensions that's what we have to do. So for this example, if we're thinking what's going on when
we put those unit spheres on the corners, those unit spheres are just way far away. They're just way off there. Um, so when we have that inside one
that's growing as big as it needs to be to be tangent to all of those, it's got to get massive. Now, this picture is also still not quite right
because I'm cramming it into two dimensions and I'm only showing four corners, but there's many, many, many
corners. There's two to the end different corners. So, the like new loose schematic that you might have for this is something where rather than just
those four corners, you have a whole bunch of corners. So, very vaguely speaking, like this is the picture to have in your mind for what's going on
with that example. So that the inner sphere does bust out um but in some sense it's still contained by
all of those corners. So this is all to say highdimensional geometry for all of its utility is unfortunately very counterintuitive. And I think maybe
the better way of framing is that it risks being counterintuitive because I think if you find the right ways to think about these shapes they actually
can become your friends. And this is what brings us to the real me. The real reason we're all here.
I don't want to talk about how it's real and useful. I don't even want to talk about how it's counterintuitive. I want to talk about something that's
just completely beautiful about higher dimensional spheres specifically. And this is that formula that I referenced. What it's going to describe is
the volume of a ball up in a higher number of dimensions. You can focus on a ball with radius one, but
um just add an r to the end there if you want. So, we're going to build up to this. We're going to talk about where it comes from. I want it to feel
like it's deep in your bones, something that you really felt like you could have discovered it yourself. Um, and in order to get up there, I think
it's worth just spending some time with the familiar friends that we have, which are the relevant
formulas you might have come across already in school. And so there's four main ones here. You've got the circumference of a circle. So this is kind
of measuring the boundary in two dimensions. And this is 2 pi time the radius. This is basically the definition of pi. So after here, everything is a
derivation, but this one you could call kind of a definition. Um, this is this is what the constant
pi even means. Now below it I'm going to draw p r^ squ that's the measuring the interior of the square. So still in two dimensions interior our good
old friend p r^ squ up in three dimensions the boundary that surface area looks like 4 pi r^ 2 very fun very suspicious because it's exactly four
times the circle so that might be worth revisiting and then the interior uh the volume of a sphere is 4/3
pi r cubed. So these are the ones we might have seen them inside a chart and our goal today is basically to fill out this chart. So the way I'm I'm
orchestrating this, I'm going to label that bottom row BN, meaning a ball in N dimensions. So for example, a ball in two dimensions, that's a filledin
circle. A ball in three dimensions, that's the thing we would usually call a ball, but we're going to
use this as a very general term across all the dimensions. And then above that, I'm going to label things with uh this little dell symbol that you
often use to mean partial. So this is just a bit of notation. In this case, it doesn't mean partial, it actually means the boundary. So the boundary
of a ball in n dimensions. So the boundary of a ball in two dimensions, that's the circle. In three
dimensions, it's the shell of the sphere. Now that notation that kind of looks like a little d kind of gives some calculus vibes, I actually think is
helpful for an important rule about this chart. I want you to think a little bit because I know all of you have probably had some kind of exposure to
calculus in your background. I want you to think about the relationship between the bottom row and
the top row. And again, I'm going to just throw this out. Anyone want to raise their hands and kind of say what is the relationship? So, right over
here. Yeah. >> It's a derivative, right? How many of you like thought this was very fun the first time? I don't know, maybe this is the first time
you've seen it. The first time I saw this, I remember thinking that's weird. Why is that true? I was but
a little calculus student who maybe had didn't have a deep enough intuition for what the whole field was. But there's a very good reason this is true
in terms of the meaning of a derivative. So if you think of what would it mean to take the derivative of the area formula, you're kind of saying,
well, suppose you were to increase that radius of a circle by a little amount, some little dr. How much
has that area changed? What is the resulting change to that area that we might call da? And what this is saying is approximately it's about the
circumference of that circle times that little dr. Now, this is wrong if you interpret dr and da as literal amounts, uh, some kind of actual tiny
nudge like a physicist. But the idea is that it becomes less wrong if you let that VR get smaller and smaller.
I'm a fan of the physicist perspective. I don't mean that in a derogatory way. Actually thinks the way we should all teach it. But uh the the
technical meaning of this would be this is not actually an equation. It's something that becomes less and less wrong as you let that little dr um
approach something close to zero or approach zero rather. And then same intuition the other way around is
basically saying you could think of the interior of the circle as kind of integrating together a whole bunch of boundaries. And when you integrate all
of those uh boundaries that grow proportionally to r, taking that integral looks like turning the r into r squ and dividing by two. That division is
going to be very important for us. So remember it. Same deal in three dimensions. Just to just to
state it all, you've got a ball. If you were looking at its volume and you want to know what's the rate of change of its volume. A really nice way to
think about that is that it is based on the surface area of that sphere. You're saying what is the tiny change in volume caused by a tiny change to
the radius? Well, it's about the area of that shell times the thickness that you've given it. Now,
that's wrong as stated with a literal number that you plug in for little dr, but it come becomes less wrong if you let dr shrink and shrink and shrink
closer to zero. And then on the flip side too, you could think of it by integrating together a bunch of shells and integrating the r 2 turns into 1/3
r cubed. And that 1/3 is important. That idea that integration gives us that division is going to
matter a lot. So we know how to go up and down in this diagram. That's the important part. If you know something on the bottom row, you can step up.
Now our real question is how do we go left and right? Right? And before we go in the interesting direction to all the higher dimensions, let's give a
little love to the first dimension because we do a lot of two-dimensional geometry. We do a lot of
threedimensional geometry. One-dimensional geometry always kind of gets the short drift. And maybe it's because it's a little less fun. But take a
moment and think, what should the volume of the interior of a one-dimensional ball be? Anyone want to want to hazard an answer? What should the
formula be for onedimensional ball? Yeah. Here. >> Would it just be a constant because it's kind of like a
line? Well, it's going to >> it [clears throat] is a line, but it's not going to be constant because we want to think of uh with respect to a radius.
So, as you increase that radius, >> 2 * r. >> Yeah, exactly. 2 * r. So, we say what is a ball in one dimension? Well, it's all the points with in a
distance r from some center point that ends up looking like a line. And in the same way that in two
dimensions, the way that we're measuring like volume is with area. In three dimensions, it's our normal meaning of volume. down here. What I mean by
the volume of a one-dimensional ball using this word volume in a general sense across all the dimensions would just be the length. And the length is 2
R. And what's fun here is if we just try to blindly follow the rule of taking a derivative to get up
top, we end up with the idea that the volume of the boundary of a one-dimensional ball is two, which at first seems weird because what's the boundary
of a one-dimensional ball? But you think about it, you say, well, it's just two different points. And what should zero dimensional volume be? I guess
just counting those points seems as reasonable as anything, but it follows the pattern. So that
gives us a little reassurance and that'll be a general theme as we move forward here. Something that seems a little suspicious at first, but that
follows the pattern kind of gives us a comfortable feeling in our hearts. Now, of course, I know you want to see how we move to the right, how do we
get into those higher dimensions? But to get a running start at it, I want to linger a little bit on
where exactly those three-dimensional formulas are coming from. because the logic that gets us an understanding of, for example, the surface area of a
sphere is going to actually carry us even farther if we think about it the right way. So, I alluded to the fact that a sphere is kind of suspicious
because it's got this area that's exactly four times the area of a circle. That may make you wonder
like where does it come from? Why do we get such a nice formula? And there's a really lovely way of thinking about this surface area that dates a long
time back. It goes all the way back to Archimedes. And he had this very clever idea which was to say we can think about the area of the surface of a
sphere by imagining taking all these little patches on it and projecting them out onto a cylinder
that encloses that sphere. Now you might think this act of projecting and kind of warping it is going to be such a a violent affair that it changes
the area. Of course it should change the area. But Archimedes very cleverly pointed out that if you think of one little rectangle of area on that
sphere and you imagine what happens as you project it out onto that cylinder, there are two competing
effects. It doesn't end up the same, but what happens is it gets stretched out in one direction, but then because it's kind of casting a shadow of
something at an angle, it effectively gets squished down in another direction. And those two effects actually cancel each other out perfectly. And
it's not too hard to reason about why either. So let's think about one of them. We're going to slice open
the sphere. We're going to put some names to things. Let's say the radius of the whole sphere is capital R. And then the distance between the Z-axis
and our specific choice of a little rectangle. I'm going to call that little D. How much does it get stretched out, say on the bottom in terms of
capital R and little D. And here you can draw two different similar triangles. We could think of the
triangle that's isoclesles and has a tip on that Z-axis point and its base is the original base of that little rectangle. And then you say, think of a
new triangle whose base is the projected version of that little rectangle that was on the sphere. And when you project it out, you're going to get a
similar triangle. And because they're similar, the ratios of all of their sides should be the same.
So in particular, we know the ratio of the long sides of those isles triangles is going to be capital R to D. So that should also be the ratio of that
little side. Meaning that's the factor by which it gets stretched out in that direction when we project. Now what about the other way? This competing
factor. What's going on there is you kind of think about casting a shadow. So the height of that
rectangle looks like this diagonal line here and then it gets projected down onto something that is perpendicular to the direction of projection. So
how much how much does that get scaled? What's that factor? Well, you can do a nice little um angle chasing argument to point out that this little
blue right triangle that I've drawn showing us that projection factor is actually similar to this brown
right triangle which again has the side lengths little d and then a hypotenuse r. So finding that it's just kind of chase some angles, see what has to
be equal, and you end up deducing that these have to be the same. And what that ends up telling you is the factor by which it gets squished down in
the other direction is that same r / d. So when you do this to all of your little rectangles, each
one of them is getting stretched in one direction, squished in another. But the idea is that it's area preserving. And because it's area preserving,
we just have to ask the area of the cylinder, which is a lot easier because with a cylinder, you can just unwrap it. You can basically imagine, you
know, slicing it like a piece of paper, unwrapping it, and you get a rectangle where one edge of that
rectangle came from the circumference of that sphere, which would be 2 pi r. And then the other edge just came from the height on that z-axis, the 2 *
r. When you multiply those together, you get yourself that 4 pi r 2. So that's where that formula comes from, or at least one way of seeing where that
formula comes from. And then, like I said, if you're at one row of this diagram, you can just take
derivatives and integrals to get yourself up and down. So knowing the surface area is enough to give you the boundary on the inside. Now here's the
key question. This is a very clever thing that Archimedes did. I don't know if Archimedes ever took a moment to think about four-dimensional balls or
fivedimensional balls. But if he ever did, I have to imagine he might have wondered, is there some way
I could generalize that really nice argument to talk about the boundary of a three-dimensional ball and how to measure it? How do I generalize that?
And I want you to think about this for a moment because this is really the crux of everything we're going to talk about today is taking this clever
Archimedian idea but somehow using it to let us take rightward steps in our diagram. And what I want to
convince you of is that kind of the right way to think about what Archimedes was really doing is that he was making a night's move in this might sound
a little bit weird. Let me explain. So, it's a nice move where you might think of it combined with that 2 pi r factor because if we think of that
unwrapped cylinder and we say where did those two side lengths come from? We were multiplying two
amounts. Where did they come from? One of them at the top came from the boundary of a two-dimensional ball. That is a circle. That was your 2 pi r.
And even though initially it might just look like, oh, that other side it was a line because it started off as a line. I want to try to convince you
the right way to think about that other side is that it's a one-dimensional ball. It's the interior of
a one-dimensional ball. And here's what I mean. So we're thinking about a threedimensional sphere, right? What is that? What does it mean to be on the
boundary of a threedimensional sphere? Well, really in think in terms of coordinates, it means you take three numbers x, y, and z. And as long as the
sum of the squares of those numbers is one, that's what it means to be on the boundary of a sphere.
Now, I could say, suppose I ask you, hey, I would like you to give me a point on the boundary of a sphere. One way you could do that, it's kind of a
weird way, but one way you could do it is to say, first I'm going to choose my z value, and as long as z^2 is less than or equal to 1, I'm going to be
safe. That's going to give me enough room to find an x^2 and a y^2. And then after choosing a value
for z, uh, you have a circle worth of options for x^2 and y^2. It's going to be everything um, such that the sum of those squares is 1 minus whatever
z^ squ was. Now, usually you couldn't use this kind of breakdown to make a nice uh surface area calculation because there's this dependence where the
size of that value Z is influencing the size of the circle of options that remains. So, you can't
just take, you know, the length of that line on the Z-axis and then multiply it by the size of that circle because that circle is changing size. And
it's actually really not obvious, even if you're savvy with multivariable calculus and you want to set this all up in sign way. It's really not
obvious how you can get those two numbers to relate to each other. But the core point of the Archimedian
idea was to basically say you don't have to do that. Now before I draw that, actually I I want to take the same diagram. We're kind of moving the Z
squ um and we're letting that influence X and Y, but I want to show the same animation in a way that's going to generalize to the four-dimensional
case. So it's not going to rely on me drawing that threedimensional picture. Basically, as you move Z
farther away from zero, it constricts that circle and you have this dependence between what Z you happened to choose and the circles worth of options
left over. Okay, so what did our communities do? Basically, he said, well, it's actually going to be the same as if there was no dependence as if you
took those circles worth of options and you just projected it out. So at all different values of Z,
it happened to be a unit circle. Um that's what we mean by a cylinder. If you are comfortable with the language of set theory, you could say that
cylinder is a cartisian product between the line segment and then the circle. And the really nice thing about taking a product like that where what's
happening on one of those sets is just completely independent from the other one is if you want to
calculate the area in this case, you can just multiply those two numbers because of that independence. So that's what I mean by taking a knight's move
here is that if we're taking that interior of a one-dimensional ball and each one is kind of getting turned into a circle projected out and because of
this Archimedian idea, it's independent enough that we're safe to just multiply those numbers. Now
what's this going to look like in four dimensions? Okay, so we're going to give some notation here. We could say the boundary of a threedimensional
ball looks like the interior of a one-dimensional ball [snorts] times the boundary of a two-dimensional ball, just a circle. Okay, I take that. Here's
what it looks like in four dimensions. We're going to say the boundary of a four-dimensional ball is
going to look like multiplying the area of the interior of a two-dimensional ball by a circle again. And again, let's think in terms of the
coordinates. Why does this make sense? What do we mean by the boundary of a four-dimensional ball? Well, we mean all the points XYZ W got four
coordinates now where you add up their squares and you equal one. One way you could think to make a choice for such a
point on that uh boundary is to say first I'm going to choose a z and a w and if they're going to give me a valid option all I need is that the sum of
their squares is less than equal to one. If that's the case I'll be safe. I'll have some choice for my x and y. And then after that leaves you with a
circles worth of options. Some x^2 + y^2 that would have to equal this thing that depends on z and
w. And initially that would mean you can't just naively multiply these because of that dependence. So the equivalent of the animation we had earlier
would be basically saying if the choice of Z and W was farther away from the origin that constricts the circles worth of options you have left over.
And in general it's very non-trivial to use this kind of relationship to somehow say how am I going to
figure out the volume of the boundary of this four-dimensional ball using this breakdown. But the clever thing is that same Archimedian idea where you
take uh various points with a constant ZV valueue and you project them out onto a cylinder, the same logic ends up holding. But in this case, instead
of a constant ZV value, you'd be choosing constant Z and W, but you still take that XY's worth of
options and project them out onto something with radius one. You'll still have the same effect where that stretches things in one direction, but it
squishes things in another. And the ratios work out that those perfectly cancel each other out. And in terms of what we have on screen here, what that
basically means is we can pretend like there is no dependence on Z and W as far as a volume
calculation is concerned. We can just take that interior of the circle and then multiply it by the boundary of the circle. And this is very bizarre
for me to think about because the product of the interior of a circle and a circle like this is going to be a solid donut. So with whereas Archimedes
was projecting onto a cylinder, somehow the four-dimensional version of Archimedes would be imagining
projecting his hypersphere onto a donut that encloses the sphere. I can't visualize that either, but we can do the analysis to say this must be what's
going on. So in our diagram, we're making another night's move, but this time we're shifting everything over. What is the volume of the boundary of a
four-dimensional sphere? Well, we're going to take that p r^ 2. We multiply it by the 2 pi or 2 pi
r rather. And there we go. Evidently, that volume looks like 2i^2 * r cubed. This is kind of fun because we picked up another factor of pi. Maybe you
weren't expecting that. And again, once you have something in one column of this diagram, you just take derivatives and integrals to go up and down.
So if we want to know the volume, or rather the four-dimensional equivalent of volume of a
hypersphere, you just take an integral to go down. In this case, integrating r cubed looks like turning it into an r 4th and then dividing by 4. So,
as far as the constant is concerned, what we've done here is basically divided by four. Again, that division is going to be very important. And so, we
end up with this conclusion where evidently the volume of a four-dimensional ball looks like 2 / 2 *
r 4. Now, that's very pretty. It's kind of a nice formula. And maybe you don't believe me, right? Maybe you think, I don't know, Grant, you're pulling
the wool over my eyes in some way here. Like I don't know if that Archimedian trick really necessarily works. Maybe you're skeptical about the
integration step. But I want to emphasize again this is not just we claim it and we hand it down from on
high. This is an empirical question. If you want to test whether this is true, whether you've done your math right, this is wouldn't be a proof, but
you could just go and run a simulation. You can ask that probability question we had at the start. Go type up some whatever your favorite programming
language is. You have a choose four random numbers between negative 1 and one. you add up their
squares and you just sample when do and don't the sum of those squares end up you know bigger or smaller than one and what you'll find is the
numerical answer to that ends up lining up very closely the bigger your simulation is with the formula that I just gave you that nice beautiful pi^2 /
two so that's fun and it's also generalizable we could just keep going we can make that same night's move
if we want to go up to the fifth dimension and just to spell it all out what this is going to end up looking like I won't do this for every single one
of them because boy are we stretching what we can even draw in pictures on the screen now. But we're going to say the boundary of a five-dimensional
ball is in some sense uh thought of as a product between the interior of one and three dimensions, a
solid three-dimensional ball and a circle. And what I actually mean by that is if you choose a point on that boundary, well, what we mean is choosing
five numbers where the sum of the squares of those numbers is equal to one. And you could think of that as saying first I'm going to choose three of
those numbers and I'm safe to do that as long as the sum of their squares is less than one meaning
point inside a three-dimensional ball and I have a circle's worth of options left. And this is a a tongue-in-cheek equation. It's not the case that
the boundary of the ball is literally like a product a cartisian product between your interior of a ball and a circle. But what I'm saying is there's
a a volume preserving move that takes you from that boundary of a five-dimensional ball onto the thing
that literally is that cartisian product. So we might as well think of it as being that as far as volume calculations are concerned, which is very
fun. Now in terms of our diagram, all that means is to make this knight's move, you do the same thing. You multiply by 2 pi r. And sometimes people
will ask like why are you staying constant at that 2 pi r? Everything else is moving over. Why does that
one stay fixed? And the basic idea is that when you're making this Archimedian move where you want to um project out, it's important that you have two
distinct actions that are canceling each other out. One that's stretching in one direction and the other one which is casting a shadow. So if you
tried to break down your choice of a point on the boundary of some highdimensional sphere, not as a a
bunch of things and then two more, but a bunch of things and you know, three more or four more, uh you wouldn't run into that same really nice
cancellation. So every time you make this nice move in the diagram and then you integrate to go down it gives us one more column and again that
integration as far as the constant is concerned up in five dimensions it just means dividing by five that r
fourth turns into r 5th and you have that integration factor dividing by five and we can just keep this going on autopilot. Every time you make one of
these moves um getting into that nice move looks like multiplying by 2 pi r. So you kind of have this 2 pi jump up and to the right and then an
integration factor to get down and we just fill in the whole diagram. And in some sense we're done,
right? In some sense this gives you the rule that you could use to decide on the volume of any dimensional sphere that you want. But of course we want
a nice formula for this, right? We want to come away with formula. We want to analyze the formula. We want to think about what it means. So let's take
a moment to see if we can write down what that should be. So the way I'm going to do this uh each
one of these volumes in higher dimensions looks like some constant times r to that number of dimensions. Right? So for example in one dimension that
constant is two. In two dimensions that constant is pi. In three dimensions 4/3 pi on and on and on. What we really want to know is what is this
constant? And the key rule that we have is a recursive one. It's saying if you want to know that constant
in some dimension, look two dimensions earlier and it should be the same as taking that but multiplying by 2 pi and dividing by n. And that's the
formulaic way of basically capturing this night's move and integrate idea. You take a night's move to get up and to the right that multiplies you by 2
pi and then you integrate to get down. That's going to divide you by n. So if you have this recurrence
relation, it kind of tells you everything you need to know. For example, let's say uh couple months from now, you find yourself in one of your exams.
For some unbeknownsted reason, your exam asks, "What is the volume of an eight-dimensional ball?" And you're like, "Grant told me this, but I forgot."
Don't worry, it's okay if you forget, cuz all you need to know if you want to know the volume of an
eight-dimensional ball is the volume of a sixdimensional ball. You take pi divided by half the dimensions and then multiply it by that. And you say,
"Ah, crap. I also forget the volume of a sixdimensional ball. Not a worry. Not a worry at all. All you have to remember is the volume of a
four-dimensional ball. You take pi divided by half the dimension. So in this case, half of six would be three.
Pi divided by half the dimension. Multiply by that. And you're like, well, now this is embarrassing because I also forget the volume of a
four-dimensional ball. But not a worry. You go down one more. You multiply by pi / the dimension. Now, at this point, presumably you do know the
volume of a two-dimensional ball, which is to say the area of a circle. But if you're a programmer at heart and
you're looking at this, right, and you're thinking, I'm defining a recursive function. If you ever wrote a recursive function, your base case was at n
equals 2. It would be very embarrassing to submit that poll request. So looking at this diagram, it kind of screams out to you, hang on a second,
maybe we should actually be venturing farther back, even though I know the area of a circle. Like
should I be going one further? Should I have a base case at zero? [laughter] And I mentioned one dimension was unloved, but like poor zero dimensions.
We didn't even think about zero dimensional geometry and say, okay, what would this mean? Well, in terms of the formulas, we know what it wants us to
be saying if we want this recurrence relation to be true. The volume of a uh a zero dimensional
unit ball should be one. And we're like, hang on, where are we? Zero dimensions. Can't move left. Can't move up. Can't I can't move anywhere. Just a
point. So, a unit ball is everything, I guess. And the volume of everything is one. Seems reasonable. What's the boundary of a unit ball? You're like,
the boundary of everything, I guess, would be nothing. So, zero, which I guess makes sense. And I I
actually once gave this talk at Stanford and you know how at the end of talks like people come up for Q&As's and there's always the annoying person
who doesn't understand the Q part of Q&A and they're like I have a thing to say. So someone comes up he's like I have three points I want to say. It
was okay though cuz this person was Donald Ko and he was like it was like in the zero dimensional case
you can still write the program to verify the answer because you just choose zero numbers and ask the probability that the sum of their squares is
less than one and it's probability one. I was like, "Ah, touché, Donald can. Thanks." So, point is, we've got this uh answer to our eight dimensional
question, but it's not hard to see how it generalizes. We were playing this little hopscotching game
going back, and you're basically multiplying pi in the numerator by half the dimensions number of times. And then that denominator, because we were
always dividing by that half dimension, it's just a factorial. So, in this specific example, it's four factorial, but more generally, it's half the
number of dimensions factorial. And so we can clean things up. We take that constant, like I was saying,
it's multiplied by an R to the n. And I can tell you, wonderful, we can use this. It lets you kind of hopscotch two dimensions up if you think of it
as a recurrence. But you can also just zero shot your way into whatever you want. If we're going to ask this for, say, 100 dimensions. And I say, what
a beautiful formula. As promised, we have a beautiful formula. And you might say, well, hang on a
second. You're not done. Because there's something kind of awkward about this formula, which is what's going on at all the odd numbers. because the
denominator is asking us to take something like half factorial or three halves factorial or five halves factorial which feels kind of weird but we
shouldn't have a rule that's different for the odd number dimensions because in terms of all the logic
that we were applying with this night's move and integrate nothing about that logic was telling us hey only do this for an even number of dimensions.
So we kind of want in our souls for this to be something that applies to everything because the recurrence rule applies to everything, right? Like if
you forgot the volume of a seven-dimensional ball and it's a crucial fact that comes up on your
homework for some godforsaken reason, you could still find your way up to it, right? You play this game where you keep multiplying by uh pi / half the
dimension. And actually in this case, I think it's kind of nicer to put that two back up in the numerator. So if you wanted to think about what is it
that takes us from say a one-dimensional ball up to a three-dimensional ball, you're multiplying
that 2 pi and then you're dividing by three. And so like maybe next time you look at the volume of a sphere, you can think about it as saying, hm,
actually it is a one-dimensional ball, but we did that nice move which was 2 pi and divided by 3 and that's what gets you that 4/3 pi factor. But you
could keep going from there to get up to five. You multiply by another 2 pi and divide by 5. Multiply
by another 2 pi from there, divide by 7. And if you had been looking at those oddnumbered cases and thinking that the numbers were very ugly, like for
example, we have this, you know, 945 in one of the denominators and initially you see that you're like, where did that number come from? Here it's
actually nice to think about. It's like a factorial, but just where you do all the odd numbers. That's
9 * 7 * 5 * 3. And then the numerator always looks like a power of 2 * a power of pi. But if we wanted this to all be true for our general formula,
right, we don't want to have one rule that we write for the even case, one rule that we write for the odd case, like I don't know, Thanksgiving with
divorced parents or something where you have to make a choice. It's nice if our parents get back
together, right? Like let's uh let's think about what would be required for this to make sense. And in this case, the marriage counselor just suggests
looking at the n=1 case and asking what has to be true there. So what this formula would say for n= 1 would be that the volume of a one-dimensional
ball which again is just a line segment looks like pi half pi to the 1/2 and raising to the 1/2 is
the same as taking a square root / 1/2 factorial whatever the heck that should be times r. But we do a little arrangement and we say well this is kind
of telling us what 1/2 factorial would have to be if you know mom and dad are getting back together. it would have to be square of pi / two. And once
you have that, you actually have the rest of them because if you think about factorials again in a
recursive way, what should it mean? Like n factorial recursively is defined to be n * whatever n minus one factorial is. So what is three halves
factorial? Well, it's going to be three halves times whatever 1/2 factorial is, which we're saying should be of pi / 2. What's 5 half factorials?
Well, five halves times three halves times whatever 1/2 factorial should be. So basically, if we take this
one suggestion from the mathematical universe that 1/2 factorial should be of pi / 2, our whole formula works out. And what's reassuring here, some of
you might actually know that there is a way to generalize the factorial. Um, it doesn't just generalize to half integers. It's based on something
called the gamma function. It generalizes it to real numbers, uh, to complex numbers. And if this was a
much longer lecture, we could have a whole other discussion about the gamma function and how there's a completely parallel path that you could use to
derive the volumes of higher dimensional spheres that comes from basically studying Gaussian distributions in high dimensions where if you want to
integrate them and understand what it looks like to integrate a Gaussian function, there's a certain
trick that you can apply and that trick will give you a formula for the higher dimensional spheres in terms of the gamma function. And what that ends
up telling us is the way that we want to extend our factorial definition, which is kind of like forcing this formula to be true, it actually does
align with the way that it gets generalized even more broadly in a different corner of math. Now, in
addition to just having a beautiful formula to gawk at symbolically, I want to take a moment to look at the actual numbers behind it and plug in some
amounts and see if we can interpret what's going on because something very strange actually happens when you do that. So, let's think about this. I'm
going to draw the volume of a unit ball in all of our dimensions. So, starting with zero dimensions,
no longer does zero dimension get no love. We're going to give it the love it deserves. That constant is one. In one dimension, the constant the
volume of a unit ball is two. In two dimensions, uh it's pi, just area of a of a unit circle. In three dimensions, it's 4/3 pi. And all these numbers
are going up. You might be like, yeah, of course they're going up because each sphere in one dimension
like contains the smaller one. So, it's getting bigger. uh they should be getting bigger. Up in four dimensions, it gets bigger still. We end up
around 4.93. Five dimensions is bigger still, but it's kind of slowing down, which might give us pause. Then it gets very worrying because at six
dimensions it's smaller. Like that's weird. Seven is smaller, eight is smaller, 9 is smaller. Still, it
starts decreasing very rapidly. We're like, hang on a second. Up in, for example, 10 dimensions, that's actually smaller than the two-dimensional
ball. And that's that's a little strange to interpret to say the least. And so if you draw the curve for this where we're going to fill it in not just
for integers but for everything using the gamma function, here's what it looks like. And let's think
about why, right? Like why does it turn around? Why does it turn around at five? That's kind of weird. What's special about five dimensions? Um and
then what happens as we go out towards an even higher number of dimensions? And the nice thing about the fact that you and I have landed here not
because I just gave you the formula and said trust me that it's true but because you see where it comes
from is we can think about these questions in terms of where it came from. Ultimately it's it's rooted in this recurrence relation where understanding
what happens in one dimension you look two dimensions earlier and then you do that nice move that takes you 2 pi and then you do an integration which
gives you that division by n. So every time we're stepping two units to the right in this diagram,
you're multiplying 2 pi / n, where n is kind of the x coordinate of where you land. So for example, jumping from 1 up to 3, that's 2 pi/3. 2 pi is
bigger than three, so that's why you're growing. From 2 up to 4, you multiply by 2 pi over 4. 2 pi is bigger than four, so you're still growing. 2 pi
is bigger than five, not by as much, so you're not growing as much. From four to six, 2 pi is bigger
than six, but now just barely. And then it's at that point that it turns around where as you go from five up to 7, you're taking 2 pi / 7, but now the
denominator is starting to win out. And in fact, it starts to win out by more and more and more because as you get into a higher dimension, that
integration factor is big playing a bigger and bigger role and it starts to dominate. So if you wanted
to ask why was five the biggest basically because at that value of n where the numerator denominator are as close as they can be when it's equal to
six that's straddling this number five and so that's the point at which the change on this graph um is going to be around zero. So if we keep going
though, what ends up being very surprising and I think this is one of those just genuinely baffling
facts about higher dimensions is that your unit balls, they don't just get small, they get downright infantal. For example, let's take a
100dimensional ball using our formula. It looks like pi to the 50. Cool. That's big. Nice big number. But divided by 50 factorial, which is way
bigger, right? because all of those factors of pi in the numerator on average are much smaller than the terms that
you're multiplying in the denominator. Almost all of them are much bigger than pi. Numerically this is around 2.37 * 10 -40th. So a ball in 100
dimensions it's nothing. It's just absolutely nothing. And this again is an empirical fact. That question I asked at the very beginning I said imagine
you choose a 100 different random numbers each between negative 1 and one. You square them. You want to
know the probability that the sum of those squares is less than one. Now, on the one hand, you could say, well, it's it's very expected that this
should be a small number, right? Because think about the coincidence that's required here. All 100 of your numbers have to happen to be so small that
when you add up all those squares, you end up less than one. And I'd say, I agree, that's surprising. I
would have thought that smallalness is taken into account by that 2 to the 100 in the denominator. I would look at it and be like, well, there's the
source of your smallness. The thing that's surprising here is that it's not just the denominator that makes it small, but the numerator is also
working against you. That numerator is a puny number as well. This volume of higher dimensional spheres
being very small and downright negligible. It actually comes up a lot. You know, people who study machine learning, they have to think about quirks in
higher dimensions. This is one of them. Your balls are just puny.
Cryptographers, same deal. Well, you don't clap. you can't clap for that. There's a time for clapping and that's not it. Um, uh, if you study
cryptography, this fact comes up a lot as well. Quantum mechanics comes up as well. And I think it's worth just a little moment to divert all of your
attention away from the concept of puny balls and instead on the question of like why where did that come
from? And really what's happening here is the integration factor, right? when you're dividing by that n, for example, when you're thinking about the
volume of a threedimensional ball, you're kind of adding together all of these shells, you're integrating together all these things that are
proportional to r 2 and we're saying that effectively gives you this 1/3 factor. It's very similar actually if
you have, let's say, like a square pyramid and you think of it as being built up from a whole bunch of squares whose side lengths are increasing
linearly. You know how with a pyramid it looks like 1/3 time its base uh times the height. Basically the same thing is going on here where that's
where the 1/3 comes from. And in general when you're doing this in a higher number of dimensions that factor
is getting bigger and bigger. And this actually relates to a distinct fact about higher dimensional spheres which is also relevant in all of these
fields if you want to get a little bit of an intuition for how distances behave. This has to do with not just what the volume of a sphere is but where
that volume is. And it's basically almost all right next to the boundary. And it's not too hard to
explain why. So let's say we're in two dimensions and I want to understand how much of the area of that circle, let's say it's a unit circle, is
within a distance of like 0.01 of the boundary. It's kind of within 1% of the boundary. Well, you could calculate that is you could say, all right,
the area of the entire circle is p r^ squ. And then if we chop out everything other than that boundary,
that would be another circle, but whose radius was scaled down by 99%. And if you scale that down by 99%. The area gets scaled down by its square
which is approximately 0.98 in this case 0.9801. But the idea is that 2% of the of the area of that circle is close to the boundary. But circles are
actually very unusual in this respect in the scope of all of the possible spheres across all the
dimensions because let's think about what this would be in like some really big like 10,000 dimensions. Okay, so we know the volume. It's fun. We have
a formula for it. It's not really the point here though. If you wanted to play the same game of saying how much of the volume is right next to the
boundary, you could take the entire ball and then you say, let's scale it down by 99%. And then ask
what's the volume of that smaller one that initially you might think is only slightly smaller. But in this case, when you scale it down, it's not just
getting scaled down in two directions. By definition of what we mean being in 10,000 dimensions, it's getting scaled down in 10,000 different
directions. So the factor by which its volume decreases looks like 0.99 raised to the 10,000 that is
effectively zero. It's 2 * 10 - 44. So this is saying essentially all of the volume is sitting right there on that boundary. And this is related to
what we were just looking at where as you're going from the boundary to the interior, you're making this division that becomes more and more
significant. Uh because if you think about it, what does it mean when we're talking about the surface area of a
sphere of a unit sphere is 4 pi, a number that's uh you know around 12 and the interior of it is 4/3 pi. On the one hand, you're comparing different
units. One is an area, one's a volume, but you could say you take that area and you multiply it just by a tiny little amount and it's giving you like
a disproportionate amount of the entire sphere. So that's actually a nice intuition to have just if
you are working with highdimensional data. Essentially everything if you're dealing with uh something inside a ball is just right next to the surface
effectively right on the surface. While we're here there's there's another fact this is kind of tied to where let's say you want to know on that
boundary where is most of that boundary volume for example most of the surface area of a sphere. Now in a
sphere it's nicely distributed across everything but if this picture was being drawn in many many more dimensions what ends up happening is almost all
the masses at the equator and the reason is tied to what we were just talking about where as you scale it a little bit because there's so many
dimensions to scale in it basically disappears. So if you think of the circles that aren't the equator and
I say circle the uh you know the dminus 2dimensional boundaries that aren't the equator but are just a little bit up forward or just a little bit down
from it. those ones end up becoming effectively nothing because that tiny bit of scaling that's required to get just a little up or just a little down
essentially makes their volume disappear. Now the last thing I want to linger on is how like this
idea that the size of your uh of your spheres in higher dimensions is shrinking feels a little bit like it feels like it's not even a sensible
comparison because we're messing with different units, right? Like what does it mean to compare the volume of a one-dimensional ball to that of a
two-dimensional ball to a threedimensional one? because one is a length, one is an area, one is a volume.
Because in some literal sense, like a threedimensional ball is not smaller or bigger than the four-dimensional ball. It's just incomparable. And so
let's try to think about what this number actually means. How can we give kind of a unit-free interpretation of it? And in a literal sense, I guess
what we're saying is we're really comparing it to the volume of a unit cube. You know that pi for the
area of a unit circle. That's basically area of a circle as compared to the area of a of a square with side length one. And then three dimensions.
What do we mean by that 4/3 pi? Well, we're comparing the volume of that sphere to the volume of a unit cube. And so in some sense, the fact that this
is getting smaller is really just telling you that up in many dimensions, the size of a cube compared
to the size of a sphere ends up getting really disproportionate. The cubes are huge and the spheres are small. And that I was thinking actually that
reminds me of that earlier puzzle. And in fact, there's a different visual that you can think of for what this number is really telling you because
let's pull up that second puzzle that we looked at where we had all these uh corner spheres. What if I
wanted to ask how much of this diagram is represented in those corner spheres versus in that cube? And in this case, there's eight distinct corner
spheres and each one of them has radius 1. And then the cube is a 2x2x2 cube. So you'd be taking 8 times the volume of a unit sphere and you're
dividing it by 8. So that's just getting us back to where we are because that factor of 8 cancels out.
Whatever this picture looks like up in 100 dimensions, you've got 2 to the 100 corner spheres and then you've got this big old cube that's volume 2 to
the 100. Those cancel out. And the fact that a 100dimensional sphere is very small could be interpreted as saying whatever this diagram should be up
in 100 dimensions, those corner spheres are basically none of it. they're effectively none of the
volume that you're looking at. And that kind of ties to the visual I was giving earlier where the very loose huristic for this highdimensional cube is
that it's super spiky. It's something where the distance to the edge is much smaller than the distance to the corner. And so all those corner spheres
are basically none of your diagram. And so with all that, I just want to end you with you on a very
small note, which is that there's a lot of beautiful items that you can come across in math, but actually recognizing that beauty sometimes requires
looking at something that was very familiar, whether that's the factorial or area of a circle and then putting it in the context of something more
general. The beauty that underlies factorials and the beauty that underlies the volume formula for a
sphere is actually only visible once you've stepped back to see things general enough. Otherwise, it would not at all be visible that a pi is secretly
hiding inside your factorial calculations or that there's some unifying formula sitting between your area of a circle and your volume of a sphere. So,
that's all I have for you. I want to end it here. Thank you everyone for coming.
[applause]
Hi there, me again, but from the office this time. I wanted to chime in with a quick little end note partly to talk about that very first animation I
showed at the beginning and explain what that actually is, how it relates to highdimensional spheres. But before then, given that you are
demonstrabably the kind of person who watches a lecture about highdimensional spheres, I have a suspicion that
you would be a good fit for the new experimental thing I'm doing this year, which is essentially a virtual career fair. So, take a moment to go to
3b1b.co/talent. And there what you'll find is a set of career opportunities from various different organizations who are interested in people like
you, the kind of people who watch these videos and the curiosity that represents which tends to correlate
with very technically talented people. So for every group that you find there, I've personally taken a chance to chat with the teams there and I can
tell you the people that I've met there are just clearly very smart. They're very curious in the sense of truth seeeking. A lot of them are just high
agency people. They clearly take charge of what they actually do. They don't just wait to follow
instructions. And if that sounds like the kind of person who you want to work alongside, I would highly encourage you to at least check it out. Listen
to some of the interviews I conducted with the people there. And if you're looking for a new career, hopefully this gives you a helpful new batch of
considerations. Since the last video when I very first mentioned this, there have been actual hires
made, which is pretty cool to me. So, who knows? Maybe if you go and explore the page, you'll be the next one. All right. So for that animation at the
very beginning I initially showed this circle that turns into the various lines of latitude on a sphere a sphere in three dimensions and our minds are
very good at seeing that it's a three-dimensional sphere. After that I basically show the same
thing same code but stepping it up in one dimension. So I represent all of the spheres of latitude on a four-dimensional sphere. So they're kind of a
bunch of four tupils of numbers. But if you change the perspective on that before you project down into three dimensions, you get all of these
different views of what those spheres of latitude end up looking like, which as you rotate that hypersphere
ends up looking just very trippy. And you know, I don't think this is explanatory in the sense of you look at this and you see ah p^2 over 2, but it
is very fun. So thought I'd share that with you. Also, if you want to see a fun application of this formula for highdimensional sphere volumes, you
might enjoy a video I did with number file a couple years back that includes a puzzle that incorporates
this into its answer. [music]
[music]