The Basel Problem Part 2: Euler's Proof and the Riemann Hypothesis

Bernhard Riemann
12 followers
Hello, and welcome to ZetaMath!
Last time, we told the story of how young Euler, determined to solve the Basel problem,
developed a powerful new method to approximate the value of infinite series.
Using this, he was able to show that the value of this sum
agreed with pi-squared over 6 to at least 17 decimal places.
In today's video, we're going to talk about how Euler went on to prove that the value
is exactly pi-squared over 6. Or did he?
His proof relied on some ultimately correct ideas
that would not be put on a rigorous foundation for a century.
We'll also talk about how Euler extended his solution to the Basel problem,
and in so doing, invented what would become known as the Zeta function.
Finally, we'll talk about how Riemann used ideas very similar to Euler's
to discover the relationship between the Zeta function and the prime numbers.
So now, let's imagine that we're Euler,
and our goal is to prove that this sum has the value pi-squared over 6.
Seeing that the pi is involved is suggestive of two things.
Circles, and trigonometry.
I'll put a link here to a really neat proof that 3Blue1Brown posted that involves circles,
but Euler's proof ultimately comes down to carefully analyzing the function sine of x.
Indeed, Euler's proof ultimately amounts to writing down
two different formulas for sine of x and equating them.
The first formula is this infinite series,
which is known as the Maclaurin series of sine of x,
or if you want, this is the Taylor series of sine of x centered at 0.
This was well known at the time.
If you follow this proof through very carefully,
you'll notice that this 3 factorial here ends up being the 6 in pi-squared over 6.
You should think of this Maclaurin series as being the fingerprint of sine of x.
In other words, this series uniquely identifies the function sine of x,
and no other series of this sort does.
I know that Taylor and Maclaurin series are often mysterious,
and you may have forgotten about them from when you first saw them in your calculus class,
so if you'd like a quick little refresher that's also very entertaining,
I recommend going to watch this short 3Blue1Brown video on the topic.
In addition to this infinite sum,
Euler also sought to write sine of x as an infinite product.
His motivation here came from the idea that the zeros of a product
correspond to the zeros of the terms in the product.
Thus, if we were to write down an infinite product for sine of x,
each zero of sine of x would have to appear as a zero of one of the terms in the product.
This idea probably sounds a little familiar
if you stop thinking about infinite products and just think about factoring.
So for example, if you think about factoring polynomials,
the key result you'll think of is probably something like this,
which we usually call the factor theorem,
which says that some number c is a zero of a polynomial
if and only if x minus c is one of the factors of the polynomial.
If I show you the graph of a polynomial like this one,
we can see that it has three zeros,
one at minus 2, one at 1, and one at 3,
and so we know we can factor out an x plus 2, an x minus 1, and an x minus 3,
meaning that if p of x is this polynomial,
we could write p of x equals x plus 2 times x minus 1 times x minus 3
times some other polynomial q of x.
If we knew a little extra information,
like for example that p of x was a cubic,
or that p of x only had these three zeros,
then we would know that this q of x had to be constant.
Euler's idea to get his infinite product representation
was to take the same approach but to the function sine of x,
even though sine of x is not itself a polynomial.
That is to say, for example,
since sine of pi equals 0, he thought,
we should be able to factor an x minus pi out of sine of x.
We have to stop and think for a moment about what the phrase
factor out means here.
What we want is sine of x is equal to x minus pi times something.
Conveniently, if this equation is going to be true,
we know exactly what the something has to be.
Namely, the something has to be sine of x divided by x minus pi.
But what's special about sine of x divided by x minus pi?
Couldn't I do this for any number?
To answer this question,
let's look at the graph of sine of x divided by x minus pi.
Is this the graph you expected?
What happened?
It might be illuminating if next to this graph,
I also show you the graph of sine of x divided by x minus 2.
You'll notice these two graphs have a very key difference.
Namely, the presence of this vertical asymptote.
This vertical asymptote here occurs because at x equals 2,
the denominator of this function is equal to 0.
So really then, the surprising thing
is that the graph of sine of x divided by x minus pi
does not appear to have an asymptote.
Now, if we're really technical about it,
the graph actually has a hole in it at the point pi comma minus 1.
But if we just go ahead and fill in the point pi comma minus 1 in the graph,
we're left with an honest-to-god continuous function.
This is what we really mean here when we talk about factoring out terms.
The function that we're left with, this something that we talked about before,
it should be a function that is continuous everywhere.
Or at the very least, it should just have a few holes in it
that if we fill them in, we get a function that's continuous everywhere.
To convince you that something sort of innate to the function sine of x is going on here,
I want to look at this from another point of view.
And I want here to look specifically at a different 0 of sine of x.
Namely, I want to look at the 0 x equals 0.
And so, what polynomial factor should that correspond to?
That should correspond to x minus 0, which is just x.
Which is to say that sine of x over x should be in some sense a continuous function
even at 0, even though it certainly has this 0 in the denominator at x equals 0.
But if I replace this sine of x now with its Maclaurin series,
then we'll see that in fact an x goes into every term.
And if I now just cancel this x with every term in the numerator,
then I'm left with this power series for a function that has no apparent problem at 0.
In fact, it's pretty clear that at 0, this function should just take the value 1.
So what are all the 0s of sine of x?
Well, there's this one at 0, this one at pi, this one at minus pi, 2 pi, minus 2 pi, and so on.
All the 0s of sine of x are exactly of the form some integer n times pi.
Hopefully at this point, I've got you at least a little bit convinced
that I can always factor x minus n pi out of sine of x.
If you'd like to think a little bit more about what this claim is actually saying,
what I recommend you do is think about what the limit as x goes to n pi
of sine of x divided by x minus n pi is.
You can show this limit exists using L'Hopital's rule,
and the fact that this limit exists tells you that even though the function isn't defined here,
the discontinuity is just a hole in the graph,
which we can fill in exactly as we did in the cases above.
We don't just want to factor out the term x minus n pi for one 0 of sine of x.
We actually eventually want to do it for all the 0s of sine of x.
As a stepping stone towards that,
let's try and factor out the terms corresponding to these three 0s.
So, we want to factor x times x minus pi times x plus pi out of sine of x,
and we want to check that this function,
sine of x divided by x times x minus pi times x plus pi, still seems reasonable.
And looking at its graph now, we see that we were successful,
in the sense that this graph is still continuous with these particular holes at these three points.
Looking closely at this graph on a larger range,
we might key into a small problem with our plan to factor out all of the 0s.
Namely, this cubic denominator, its value is huge.
And what that results in is the fact that the value of this quotient,
sine of x over this polynomial, is very, very small for even modest values of x.
In particular, remember that Euler's goal here is to write a product representation of sine of x,
meaning that what we want to have left over,
once we factor out terms corresponding to all of the 0s, is just 1.
But this graph here, it does not look like it is going to be anything like the graph of 1.
The trick to overcome this is to represent the 0s by a different linear function.
We happen to choose the function x minus n pi, because it looks the prettiest.
But there are lots of linear functions that have n pi as a 0.
In particular, we can multiply this x minus n pi by any constant that we choose,
and we'll still have a linear function which is 0 at n pi.
In this case, the constant that I want to multiply by is minus 1 over n pi.
When I perform that multiplication, what I'm left with is the linear function
1 minus x divided by n pi.
So why is this linear function better than x minus n pi?
To answer this, I want to think about what happens when n is really big relative to x.
In x minus n pi, if n is really big relative to x, that makes this quantity really big.
However, in our new linear function, if n is very big relative to x,
that makes this term here close to 0,
which makes this whole linear function have a value that is very close to 1.
A lot of terms like this x minus n pi will result in the product blowing up,
or if you want to use fancy terms, it'll result in the product not converging.
However, a lot of terms like this results in multiplying by, well,
a bunch of numbers really close to 1.
And multiplying by a bunch of numbers close to 1
is going to result in a function that nicely converges.
So let's now factor x times 1 minus x over pi times 1 plus x over pi out of sine of x.
That is, our new linear functions corresponding to the zeros 0, pi, and negative pi.
So let's look at a graph of this quotient.
I'll overlay the graph of sine of x on top of it,
so that you can see that we've actually factored out all three of these zeros,
and our function now is not crashing towards the axis quite as quickly,
which is what we wanted to avoid.
Let's take this a step further,
and now let's factor out the linear term corresponding to the zeros at 2 pi and minus 2 pi,
which means that I'm now also dividing by 1 minus x over 2 pi and 1 plus x over 2 pi.
And again, in the graph, you can see that we've successfully plucked off
two more zeros here and here.
So, deep breath.
Now let's think what's going to happen if we actually factor out
the linear terms corresponding to all the zeros of sine of x at once.
I'm not going to think too carefully about what this means,
but this is going to mean that we're going to have sine of x is equal to,
and this product, again, having a term for all the infinitely many zeros of sine of x,
times, and then sine of x divided by that same product.
Of course, what we hope to find is that this quotient at the end here
is just literally the function 1,
because that would mean that sine of x is really equal to this product,
and then we'd be done.
Here's an animation where I'm including progressively more and more terms into the denominator.
And strangely enough, as we watch this animation,
it actually really does appear that the graph of this function
really is converging to the constant function 1.
If we believe this, then we arrive at Euler's product formula for sine of x.
Namely, sine of x is this product across one linear term,
for each zero of sine of x.
We can tidy this up just a bit by combining the terms for the opposite zeros.
That is, we're going to combine the term for plus pi and minus pi,
the term for plus 2 pi and minus 2 pi, and so on down the line.
And when I do that, thanks to the difference of squares formula,
I get the following expression for sine of x.
Namely, that it's x times this product as n goes from 1 to infinity,
of 1 minus x squared over n squared times pi squared.
Here we might pause and come up for oxygen a bit and remember what it is that we're doing.
Namely, we're trying to calculate the value of this sum.
And nothing we've done in the past however many minutes sort of obviously connects to that.
But now, looking at this product,
you can see the 1 over n squareds that we're going to use to actually prove this result.
Now you could rightly claim I haven't proved this product formula.
Euler didn't either.
But if you want to be convinced that it's really definitely true,
one way we can do that is to look at the graphs of the partial products here,
next to the graph of sine of x.
So the k here in this graph indicates how many terms of Euler's product we're using.
And you can see that as k increases,
the graph gets closer and closer and closer to the graph of sine of x.
We can think of the Maclaurin series of sine of x
as a sequence of polynomials that approximate our function arbitrarily well.
Or if you prefer, we could say as a sequence of polynomials that converge to it.
Just like we obtained this sequence of polynomials by taking progressively more terms of the sum,
we can also obtain a sequence of polynomials by taking progressively more terms of this product.
So, for example, my first partial product here will be x times 1 minus x squared over pi squared,
which if I write it out, that's just x minus x cubed over pi squared.
Similarly, our second partial product will be this, and our third this, and so on.
In fact, I can line up this list of partial products
next to our list of partial sums of the Maclaurin series here in this table.
The crucial observation here is that both of these sequences of polynomials converge to sine of x.
And so what that means is that the coefficients in the right-hand column
must converge to the same thing as the coefficients in the left-hand column.
For now, let's look just at the x-cubed terms in both columns.
On the left, as soon as we get past the zeroth line,
this term is always minus x cubed over 6.
On the right-hand side, in the first line, we get minus x cubed over pi squared.
Then, in the second line, we get minus 5x cubed over 4 pi squared.
In the third, minus 49x cubed over 36 pi squared.
And so on.
To illuminate this, let me note that we can write this term in the first line
as minus x cubed over pi squared times 1.
Similarly, this term in the second line we can write as minus x cubed over pi squared
times 1 plus 1 over 2 squared.
And the third line we can write as minus x cubed over pi squared
times 1 plus 1 over 2 squared plus 1 over 3 squared.
Indeed, if you think about where this is coming from,
I think you can convince yourself that the coefficient of x cubed on the k-th line
will be exactly negative the sum as n goes from 1 to k of 1 over n squared times pi squared.
Which, just to make the connection of this to the Basel problem clear,
is the same thing as minus 1 over pi squared times the sum as n goes from 1 to k of 1 over n squared.
Why is this happening?
Well, to see why exactly, I want to write out just what the k-th partial product is,
and I want to think about how I would go about getting an x cubed term out of this.
Really, when I foil out this whole big mess,
what that really amounts to doing is choosing for each of these binomials
which of the two terms I want to include in my product.
So, for example, I could choose this x, and then I could choose the first 1 in each of these
binomials, and I could multiply all of those together, and that would give me an x.
Or, I could choose the last term in each of these binomials, and then that would give me
x to the 2k plus 1 divided by k factorial squared times pi to the 2k.
How can I get an x cubed term?
Well, I'm stuck with this x, so I need two more x's to get up to an x cubed,
and all of my x's come in pairs after this original x.
So the only way I can get an x cubed term is to choose this x, which isn't much of a choice,
and then choose exactly one of these remaining x squareds.
So if I take this x, this x squared, and the remaining 1's, then I get minus x cubed over
pi squared. If instead, I chose this x, this minus x squared over 4 pi squared, and the rest 1's,
then I get minus x cubed over 4 pi squared.
Indeed, we're always going to get the minus x cubed over n squared pi squared term by taking
the x, the minus x squared over n squared pi squared, and all of the remaining 1's.
That shows us why we get this sum that's closely related to the Basel problem
as the coefficient of x cubed.
But now, we're almost done.
Because if I look at the limit here, the coefficient of x cubed in the left-hand column
is just going to be minus a sixth, because it's always minus a sixth.
And on the other hand, the coefficient of x cubed in the right-hand column
will be minus 1 over pi squared times the sum as n goes from 1 to infinity of 1 over n squared,
which is exactly Euler's sum.
And so, if we set these two things equal to each other and solve a not very difficult equation,
we get precisely that the value of Euler's sum is pi squared over 6.
One thing you might notice is that there's a lot more information in this table than we've used.
We so far only used the x cubed terms in order to calculate the value of this particular sum.
But we can use the information in this table to evaluate a lot of other sums as well.
As an example, I encourage you to try and use the information in this table
to prove that the sum as n goes from 1 to infinity of 1 over n to the fourth
actually has the exact value pi to the fourth over 90.
I'll prove this result for you later in the video in a different way,
I think it's very instructive to do it just from the information that you have before you right now.
Euler realized that it was worth studying all of these series together.
That is, all the series that looked like the sum as n goes from 1 to infinity
of 1 over n to some power.
He defined a new function that's now known as the zeta function
that's given by zeta of s is equal to the sum as n goes from 1 to infinity of 1 over n to the s.
Just so that we're sure we understand what this is saying,
zeta of 2 is pi squared over 6,
and that's exactly saying that this sum that Euler started with is pi squared over 6.
The exercise I gave you a moment ago to calculate this sum
was exactly to show that zeta of 4 has the value pi to the fourth over 90.
We're going to try now and extract more information out of this product formula for sine of x.
Infinite products are notoriously difficult to work with,
but there's a standard trick to turn them into infinite sums.
Namely, if you take the logarithm of an infinite product, it turns it into an infinite sum.
If we do that here, then we end up with, on the left-hand side, the log of the sine of x,
and on the right-hand side, the log of x plus the sum as n goes from 1 to infinity
of the log of 1 minus x squared over n squared pi squared.
This, though, is ugly in a different way.
Namely, there's these natural logarithms everywhere.
But, luckily, there's a standard trick for that, too,
which is to just take the derivative of everything.
And that's going to get rid of the logs,
because the derivative of natural log of x is just 1 over x.
And so, being mindful not to forget the chain rule,
once we take the derivative of both sides, we get the following equation.
The cotangent of x is 1 over x plus the sum as n goes from 1 to infinity
of minus 2x over n squared pi squared minus x squared.
Let's multiply both sides of this equation by x to clear out this denominator.
And now, let's factor a minus 2x squared over n squared pi squared
out of each term in the sum on the right.
We did this so we can apply the geometric series formula to the fraction at the right,
replacing it with 1 plus x squared over n squared pi squared
plus x to the fourth over n to the fourth pi to the fourth plus, and so on.
Let's multiply this back through.
Now let's distribute the summation.
And now, actually, we have one summation for the x squared terms,
one for the x to the fourth terms, one for the x to the sixth terms, and so on.
So now I'm going to pull out of each sum everything that doesn't involve n.
Now, remembering our definition of zeta of s,
we see that this sum here is exactly zeta of 2, this sum is exactly zeta of 4,
this one exactly zeta of 6, and so on.
If we condense this whole formula down to a single summation,
then what we're left with is x times the cotangent of x is equal to 1 plus the sum
of minus 2 x to the 2k power divided by pi to the 2k power times zeta of 2k.
This formula lets us relate the values of the zeta function at each even integer
to the function x cotangent of x.
Indeed, what we have here is really an explicit formula for
the Maclaurin series of x cotangent of x in terms of these zeta values.
What this means is that we can now compute zeta of 2k for any k we like
just by calculating the Maclaurin series of x times the cotangent of x.
A simple query in your favorite computer algebra system
will tell you that the Maclaurin series of x cotangent of x begins 1 minus x squared over 3
minus x to the fourth over 45 minus 2 x to the sixth over 945 plus and so on.
Equating that term by term with our previous formula,
we can see we get almost immediately that zeta of 2 is pi squared over 6,
but then we also get that zeta of 4 is pi to the fourth over 90,
and zeta of 6 is pi to the sixth over 945.
And we can keep going as long as we care to have our computer algebra system
keep churning through the Maclaurin series of x cotangent x.
In light of all of this, you might ask if there's a standard formula
for the Maclaurin series of x times the cotangent of x.
As it would happen, there is, but it's not usually taught in calculus courses.
Why? Because it involves the Bernoulli numbers that we discussed in the previous video,
which are not typically part of the calculus curriculum.
The Maclaurin series of x times the cotangent of x is given by 1 plus the sum as k goes from 1
to infinity of minus 1 to the kth power times the 2kth Bernoulli number times 2x to the 2k power
divided by 2k factorial.
As an aside, you actually might have noticed when you took calculus that it was a little weird
that you got these nice formulas for the Taylor series of sine and cosine, but never of tangent.
And the reason for that is because that formula also involves the Bernoulli numbers.
We now have two formulas for the Maclaurin series of x cotangent of x,
one of which involves these zeta values and the other involving the Bernoulli numbers.
We can equate the x to the 2k power coefficients on both sides.
We can take the absolute value of both sides so that we don't have to think about the
sines of the Bernoulli numbers, which remember alternate or whatever,
and solve for zeta of 2k, which we know is positive anyway.
And then once we do that, we obtain Euler's remarkable theorem that zeta of 2k is given
exactly by the absolute value of the 2kth Bernoulli number times 2 pi to the 2kth power
divided by 2 times 2k factorial.
Using this, we can check, for example, that zeta of 2 is 1 sixth times 4 pi squared over 4,
which is pi squared over 6.
And so another way to think about the 6 in the denominator of pi squared over 6
is that it's the 6 in the denominator of the fact that the second Bernoulli number is 1 sixth.
It must be noted how striking it is that this strange sequence of constants,
the Bernoulli numbers, appeared both in the way that Euler approximated zeta of 2,
as we discussed in the previous video, but also in his eventual formula for the exact value.
Euler's formula allows us to compute zeta of 2, zeta of 4, zeta of 6, and so on.
Zeta of 1 corresponds to this series, which is the harmonic series,
which diverges, so we don't need to think about that.
But that leaves the obvious next question.
What is zeta of 3?
Which is just to say, I know the sum of 1 over 1 squared plus 1 over 2 squared plus 1 over 3
squared plus dot dot dot, but what do I get if I take the sum of 1 over 1 cubed plus 1 over 2
cubed plus 1 over 3 cubed, and so on? As it would happen, Euler solved the Basel problem and did
everything we've talked about in these two videos very early in his life. Over the rest of his long
life, he returned many times to this question, but he had little success. The pattern of these
values for the even integers would certainly suggest that the value of this series should be
some rational multiple of pi cubed, but at least empirically, this seems not to be the case.
Even today, almost nothing is known about the exact value of zeta of 3.
Indeed, almost nothing is known about the value of zeta at any odd positive integer.
The question about the value of the sum of 1 over n squared began as an idle curiosity.
It was, in some sense, the simplest series whose value was unknown, so it was a natural
target for study. Over time, though, it has become clear that the zeta function is a central object
in mathematics, owing in part to its intimate connection with the prime numbers. Euler happened
upon this connection when studying the question, what is the sum across the primes p of 1 over p?
You can think of this question as being related, in a sense, to the density of the primes,
that is, how many primes there are. Here are a few relevant comparisons.
If instead of summing over just the reciprocals of the primes, we summed over the reciprocals of
all the positive integers, well that would give us what we called zeta of 1, and that we decided
is infinity. If instead I summed 1 over each power of 2, then I would get 1. Since 1 is much
smaller than infinity, one way you can think about this is as saying that at least asymptotically
there's fewer powers of 2 than there are integers. Of course, the Basel problem that we've talked
about is the sum of 1 over the squares. That has value zeta of 2, which we decided was pi squared
over 6. Again, this value is finite instead of infinite, which is suggesting that in some sense
there's fundamentally fewer squares than there are positive integers. What if we instead looked
at the reciprocals of the odd positive integers? Well, it's not so hard to convince yourself that
this sum is also infinite. In some sense, there are so many odd positive integers that we don't
get a refined sense of how many of them there are by summing their reciprocals.
As a last example, we can take something really rare, like the factorials, and you probably know
that the sum of the reciprocals of the factorials is exactly equal to e, owing to the Taylor series
of e to the x. Now, you shouldn't take this sense of a set's size too seriously in terms of comparing,
for example, the fact that this sum was e and this sum was pi squared over 6. But what we can
take seriously is that we've really split subsets of the positive integers into two groups here.
One group of these subsets of the positive integers are those where the sum of their
reciprocals is infinite, and one where the sum of their reciprocals is finite.
So Euler's real question here, more so than the exact value of this sum,
is whether this sum itself is finite or infinite. In other words, are the primes
rare, like the squares or the factorials or the powers of two? Or are they common,
like the odd integers or the integers themselves? We can attempt to get some intuition for this
by looking at a table of the partial sums. So this first line, for example, is telling you
that if you sum the reciprocals of just the primes that are less than a thousand,
then you get 2.198. If you go up to 2000, you get 2.292, and so on. Unfortunately,
it's not crystal clear from this data set whether we think this number should be going to infinity
or if it's really going to some fixed number, like maybe it's going to 2.5, for example.
Euler's idea to answer this question was to relate the sum over the primes of 1 over p
to this sum we already know, the sum over all the integers of 1 over n.
Now let's rewrite each of these denominators in terms of its prime factorization. In this way,
we see that each term in our series is really a product of reciprocals of prime powers.
For example, if I look at 1 over 60, that's 1 over 2 squared times 1 over 3 times 1 over 5.
In other words, to specify a term in this sum, we just have to specify which prime powers
we're multiplying the reciprocals of. This leads to Euler's powerful observation
that we can actually factor this harmonic series here as a product where the terms in the product
correspond to the primes. So the first one here will be 1 plus a half plus a fourth plus an eighth
plus dot dot dot, and we're accounting here for all the denominators that are powers of 2.
But then we multiply that by 1 plus a third plus a ninth plus a 27th and so on,
taking care of the prime 3. 4 is not prime, so we're going to just skip that entirely,
and 5 is, so we'll multiply by 1 plus a fifth plus a 25th plus dot dot dot, and so on. We're
going to include one of these terms, one of these infinite series, for every prime.
If you multiply out all of these infinite sums, then what you'll get in the denominators is every
possible combination of prime powers. That is to say, you'll get every positive integer exactly
once. And so what we're saying is if you multiply this product of these infinite sums, one for each
prime, you get exactly this series sum across all positive integers of 1 over n. Applying the
infinite sums more concisely, namely 1 plus 1 over p plus 1 over p squared plus dot dot dot,
that's the same thing as 1 over 1 minus 1 over p. Similarly, it's useful to remember
that the sum as n goes from 1 to infinity of 1 over n, that is zeta of 1, that's infinity.
So the key step here is that we have again an infinite product like we did earlier,
and we want to work with it, and the way we can make our infinite product a little bit nicer
is to take logarithms so that we can turn it into an infinite sum. So I'm going to take the
logarithm of both sides of this equality. On the right-hand side, I'm going to get the log of
infinity, which I'm just going to say is infinity and not talk about it anymore. And on the left-hand
side, I'm going to get minus the sum across the primes of the log of 1 minus 1 over p.
When we encountered this situation earlier, our next step was to take the derivative of both sides
to get rid of the logarithms. Here though, unfortunately, both sides of this equation
are constants, and so if we take the derivatives we're just going to get zero, and so we have to
use a different trick to get rid of the logarithms. Namely, we're going to remember the Maclaurin
series of the logarithm of 1 minus x. In fact, this Maclaurin series is just minus x minus x
squared over 2 minus x cubed over 3 minus x to the fourth over 4, and so on. Since the terms in
our sum look like the log of 1 minus p to the minus 1, we can rewrite those terms in our sum
now as infinite sums themselves. Namely, the log of 1 minus p to the minus 1 is the same thing as
minus p to the minus 1 minus p to the minus 2 over 2 minus p to the minus 3 over 3, and so on.
And so our conclusion here is that if we now take the sum over the primes p of the sum as k
goes from 1 to infinity of p to the minus k over k, that sum is infinite. Now, the first term in
each of these inside sums is exactly 1 over p, and so what I'm going to do here is separate out the
1 over p piece from all of the other pieces. So, namely, I'm going to rewrite this as the sum over
p of 1 over p plus the sum as k goes from 2 to infinity of p to the minus k over k. Now, if I
distribute this sum, on the left I get the sum over the primes of 1 over p, that's the sum Euler
is interested in here, plus then the sum over the primes of the sum as k goes from 2 to infinity
of p to the minus k over k. And the sum of these sums, we know that sum is infinite. And what that
tells us is at least one of this sum on the left and this sum on the right has to be infinite,
because if they were both finite, their sum would be finite. Let's take a closer look at the inner
sum in the right-hand sum. So, let's write this out, not in sigma notation. So, this is p to the
minus 2 over 2 plus p to the minus 3 over 3 plus p to the minus 4 over 4 and so on. Now, I'm going
to round this up considerably. Namely, I'm going to round this up to p to the minus 2 plus p to
the minus 3 plus p to the minus 4 and so on. And I want to note that in doing that, I've made every
term bigger, so I've only made the sum itself bigger. The reason I've done this is that this
sum, I now know how to calculate, because it's now a geometric series. The first term of the
geometric series is p to the minus 2, and the common ratio is 1 over p, and so the value of
this upper bound for the thing that was inside the sigma to start with is p to the minus 2 divided
by 1 minus 1 over p. If you do a bit of algebra, that simplifies to 1 over p squared minus p,
and you can now say, okay, well, I'm only making the denominator smaller if I replace this p by a
p squared over 2. This is a bit technical. All we're really doing here is trying to get this
back to a sum that we already know how to evaluate or compare to something.
But after you make that substitution, then it falls out that this giant sigma that we had to
start with, it's less than 2 over p squared. Thus, this double sum is less than or equal to the single
sum across primes of 2 over p squared. But we actually know the sum across all positive integers
of 2 over n squared, because that's twice the sum that Euler calculated when he solved the
Basel problem. So namely, our conclusion here is that this double sum, what we've shown is that
its value is less than or equal to pi squared over 3. In particular, this double summation
is always finite. Remembering that we showed that one of these two sums has to be infinite,
and we've now shown the second one is finite, well, we now know the first one is infinite.
So that tells us that if you sum the reciprocals of the primes,
what you get is infinity. This immediately tells you that there's infinitely many primes,
but this would be the most laborious possible way to show that there are infinitely many primes.
But it actually says a lot more than just that there's infinitely many primes. Namely,
it says that there's lots of primes. Because if you remember, for example, the squares,
the powers of 2, the factorials, there's infinitely many of all of those,
but the sum of the reciprocals of each of those sets was some finite number.
After Euler showed that the sum of the reciprocals of the primes was in fact infinite,
Euler realized that this key step where he factored the original sum 1 over n
into this product of things across the primes would work equally well not just for zeta of 1,
but in fact for any zeta value. And so what Euler realized was that if he took zeta of s,
that is, by definition, the sum over n of 1 over n to the s, I can rewrite that as the product
over the primes of this geometric series, 1 plus 1 over p to the s plus 1 over p to the 2s,
and so on, which I'll use the geometric series formula to rewrite in this fashion.
Just as we did before, we can now take the logarithm of both sides,
leaving us with the logarithm of zeta of s is equal to the sum across the primes
of minus the logarithm of 1 minus p to the minus s.
Now, when we got to this step before, we noted that we kind of wanted to take the derivative,
but we couldn't because there wasn't a variable. But now that the left-hand side,
instead of having the singular value zeta of 1, has instead the function the logarithm of zeta of s,
I can now actually take the derivative of both sides.
And if you do that, what you end up with is negative zeta prime of s over zeta of s
is equal to the sum across the primes of the log of p over 1 minus p to the minus s.
If we use the geometric series formula to expand out the denominator here,
as we did earlier in this video, then we get that this is equal to the sum across the prime powers,
p to the k, of the logarithm of p divided by p to the k s.
I'm going to leave the derivation of this equality as an exercise for you.
If you expand this sum, it now looks like log of 2 over 2 to the s plus log of 3 over 3 to the s
plus log of 2 over 4 to the s. Notice it's a 2 and not a 4 because we're only taking the prime
in the logarithm in the numerator, and so on down the line. Notice we skip 6 because 6 isn't a prime
power, and so we'll have one term here for each prime power. Here's the key observation.
On the left-hand side of this equality, I have this thing about the zeta function,
and on the right-hand side, I have a sum that depends purely on the prime powers.
Note that the more prime powers there are, the more terms there are in the right-hand sum,
and so in particular, the values of zeta prime of s over zeta of s encode information
about the distribution of the prime powers. This can lead to a fantasy that somehow we could
disentangle that information. That is, we could take the analytic function here on the left
and somehow shake it a bit and make information about the primes fall out. And a hundred years
or so later, that's exactly what Riemann would go on to do. The Riemann hypothesis is perhaps the
important unsolved problem in mathematics, and it is rightly named after Riemann. However,
Riemann's total published writings on the zeta function and the Riemann hypothesis and all of
these things consist of one single 8-page article. The whole paper hinges on Riemann's observation
that just like Euler did with sine of x, the zeta function can be written as a product
in terms of its zeros. Namely, Riemann's product looks like this.
Zeta of s is equal to e to some power divided by s minus 1 times the product across all the
zeros alpha of the zeta function of 1 minus s over alpha times some power of e. Now,
these powers of e here, these are here for technical reasons that involve making sure
this infinite product converges, so I'm just going to blur those out so that you don't think
about them. And instead, I want to focus on the fact that this product across the zeros of zeta
of 1 minus s over alpha is exactly like Euler's product formula for sine of x that we encountered
earlier. In the first video of this series, I showed how we could derive information about
the distribution of the primes by approximating the log of n factorial in two different ways.
One of those ways involved the prime numbers, and one of them didn't. And by equating these two
approximations, we were able to approximate the sum across the primes of the log of p over p.
Riemann's situation here is much better. He has two exact formulas for zeta of s,
one of which involves the primes, and one of which doesn't. By equating these two formulas,
taking logarithms, and differentiating both sides, as we've already discussed,
Riemann derives the formula. If you take the sum across the prime powers p to the k
less than some fixed number x of the logarithm of that p, the base of the p to the k,
then what you get is x minus the logarithm of 2 pi minus the sum across the zeros alpha of the
zeta function of x to the alpha over alpha. I want to stress here, this is not an approximation.
This equal sign is not a mistake. This equation that involves the primes on the one side and has
no primes on the other side is exactly correct. This function on the left-hand side is usually
called the psi function. That is, psi of x is defined to be the sum across the prime powers
p to the k less than or equal to some given real number x of the logarithm of that p.
In the first video, I showed you a number of graphs of the closely related function
theta of x, which is the sum across the primes less than or equal to x of log of p.
And the punchline of those graphs and the work we did in that video
was that this theta of x is approximately equal to x. Here's a graph that includes theta of x
in blue and psi of x in red, where we can see that psi of x is always just a tiny bit bigger
than theta of x. Drawing in the graph of x on top of this, we can see that psi of x is
actually even better approximated by x than theta of x was. Our arguments in the earlier video
were all about what should happen for very large x. But even for small x, you can see the
approximation psi of x is approximately x is quite accurate. Here I'm only showing the graph up to x
equals 20. But remember, Riemann's formula here is not an approximation. This x in the formula
is the dominant term, but these two remaining terms account for all of the error.
To make this clearest, I'm going to remove everything here but the graph of psi of x,
and then I'm going to fill in the value of psi of x at every real number,
instead of simply at select integer values as I was doing before.
Riemann's formula for psi of x is correct at all of these points. Here again is the approximation
psi of x is approximately equal to x, obtained by using just the first term of Riemann's sum.
I'll go ahead and add in the log of 2 pi term, which shifts the line down a little bit,
and you can see how that already significantly improves the approximation.
Now I'll draw in the graph using just the first handful of zeros of zeta of s. Notice how
significantly even the small number of zeros has improved the accuracy of our approximation.
Watch as we progressively add more zeros and see how quickly the graph snaps to the graph of our
function psi of x. This is the heart and soul of what's going on here. The graph of psi of x
zigzags back and forth over the line x minus log of 2 pi, and the extent and the direction
of that zigging and zagging is dictated entirely by the zeros of the zeta function.
There are a great many technicalities here, but what you should take away from this is that
Riemann's work on the zeta function followed the same structure as Euler's solution to the
Basel problem. Find a new product formula for a function in terms of its zeros,
set it equal to a known formula, isolate the thing you're interested in, and solve.
One of the technicalities that arises here stems from the fact that Euler's zeta function,
whose graph I'm showing you here, has no zeros. Riemann's big contribution to this story
was to think of the zeta function not as graphed here as a real valued function,
but instead as a function defined on all of the complex numbers. In other words,
Riemann wanted to be okay with saying things like, if you take zeta of 3 minus 2i, well,
that's of course going to be the sum as n goes from 1 to infinity of 1 over n to the 3 minus 2i
power, and whatever on earth that means, you can I guess approximate it, and get that hey,
this is some number that's approximately 0.97 plus 0.15i. Shown here is a portion of the complex
plane, with each zero of the zeta function in this region marked with a red dot. Like sine of x,
this function has infinitely many zeros. Those on the real line are known as the trivial zeros,
and they are very well behaved. They occur exactly at each negative even integer.
These non-trivial zeros, though, are less well behaved. However, this graph definitely suggests
that they all lay on the same vertical line. Riemann was able to show that the real part of
this zero was exactly one half, and Riemann wrote in his paper that he believed it was likely
that all non-trivial zeros of the zeta function had real part exactly equal to one half. That is,
all the non-trivial zeros of the zeta function lay exactly on this vertical line.
This innocent claim has become known as the Riemann hypothesis, and 150 years later,
despite tremendous effort by many great mathematicians, the problem remains unsolved.
Thanks to modern computing, though, we've been able to verify that the Riemann hypothesis is true
for at least the first 12 trillion zeros. If you've not thought much about complex functions
before, this discussion probably raises a lot of new questions. For example, why does this picture
say that zeta of minus 2 is zero? If we look back to our definition, zeta of minus 2 should be the
sum across the positive integers of 1 over n to the minus 2. But if we remember our properties
of exponents, that's just the sum across the positive integers of n squared. And certainly,
it doesn't seem reasonable to claim that if you sum the positive squares, that you're going to get
zero. We'll address this next time, through the magic of what's known as analytic continuation.
But for now, we at the channel would again like to thank Keith Welker for our theme music,
and we wish you all the best. If you'd like more content like this,
remember to like, comment, and subscribe.
