Why The Sum of Everything Is Almost Negative Nothing

Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

The Euler-Maclaurin formula, Bernoulli

numbers, the zeta function, and real-variable


analytic continuation
The Riemann zeta function
series

is defined in the region

Thus, for instance, it is known that

by the absolutely convergent

, and thus

For
, the series on the right-hand side of (1) is no longer absolutely convergent, or
even conditionally convergent. Nevertheless, the function can be extended to this region (with
a pole at
) by analytic continuation. For instance, it can be shown that after analytic
continuation, one has
,
, and
, and more generally

for
, where
are the Bernoulli numbers. If one formally applies(1) at these values
of , one obtains the somewhat bizarre formulae

and

Clearly, these formulae do not make sense if one stays within the traditional way to evaluate
infinite series, and so it seems that one is forced to use the somewhat unintuitive analytic
continuation interpretation of such sums to make these formulae rigorous. But as it stands, the
formulae look wrong for several reasons. Most obviously, the summands on the left are all

positive, but the right-hand sides can be zero or negative. A little more subtly, the identities do
not appear to be consistent with each other. For instance, if one adds (4)to (5), one obtains

whereas if one subtracts from (5) one obtains instead

and the two equations seem inconsistent with each other.


However, it is possible to interpret (4), (5), (6) by purely real-variable methods, without recourse
to complex analysis methods such as analytic continuation, thus giving an elementary
interpretation of these sums that only requires undergraduate calculus; we will later also explain
how this interpretation deals with the apparent inconsistencies pointed out above.
To see this, let us first consider a convergent sum such as (2). The classical interpretation of this
formula is the assertion that the partial sums

converge to

as

, or in other words that

where
denotes a quantity that goes to zero as
test estimate

. Actually, by using the integral

we have the sharper result

Thus we can view


of
.

as the leading coefficient of the asymptotic expansion of the partial sums

One can then try to inspect the partial sums of the expressions in (4), (5), (6), but the coefficients
bear no obvious relationship to the right-hand sides:

For (7), the classical Faulhaber formula (or Bernoulli formula) gives

for
clear.

, which has a vague resemblance to (7), but again the connection is not particularly

The problem here is the discrete nature of the partial sum

which (if
is viewed as a real number) has jump discontinuities at each positive integer value
of . These discontinuities yield various artefacts when trying to approximate this sum by a
polynomial in . (These artefacts also occur in (2), but happen in that case to be obscured in the
error term
; but for the divergent sums (4), (5), (6), (7), they are large enough to cause
real trouble.)
However, these issues can be resolved by replacing the abruptly truncated partial
sums
with smoothed sums
, where
is a cutoff function,
or more precisely a compactly supported bounded function that equals at . The case when is
the indicator function
then corresponds to the traditional partial sums, with all the attendant
discretisation artefacts; but if one chooses a smoother cutoff, then these artefacts begin to
disappear (or at least become lower order), and the true asymptotic expansion becomes more
manifest.
Note that smoothing does not affect the asymptotic value of sums that were already absolutely
convergent, thanks to the dominated convergence theorem. For instance, we have

whenever is a cutoff function (since


pointwise as
and is uniformly
bounded). If is equal to on a neighbourhood of the origin, then the integral test argument then
recovers the
decay rate:

However, smoothing can greatly improve the convergence properties of a divergent sum. The
simplest example is Grandis series

The partial sums

oscillate between and , and so this series is not conditionally convergent (and certainly not
absolutely convergent). However, if one performs analytic continuation on the series

and sets
, one obtains a formal value of
for this series. This value can also be obtained
by smooth summation. Indeed, for any cutoff function , we can regroup

If is twice continuously differentiable (i.e.


), then from Taylor expansion we see that
the summand has size
, and also (from the compact support of ) is only non-zero
when
. This leads to the asymptotic

and so we recover the value of

as the leading term of the asymptotic expansion.

Exercise 1 Show that if is merely once continuously differentiable (i.e.


), then we have a similar asymptotic, but with an error term of
instead
of
. This is an instance of a more general principle that smoother cutoffs
lead to better error terms, though the improvement sometimes stops after some
degree of regularity.
Remark 1 The most famous instance of smoothed summation is Cesro
summation, which corresponds to the cutoff function
.
Unsurprisingly, when Cesro summation is applied to Grandis series, one again
recovers the value of .
If we now revisit the divergent series (4), (5), (6), (7) with smooth summation in mind, we
finally begin to see the origin of the right-hand sides. Indeed, for any fixed smooth cutoff
function , we will shortly show that

and more generally

for any fixed

where

is the Archimedean factor

(which is also essentially the Mellin transform of ). Thus we see that the
values (4), (5), (6), (7) obtained by analytic continuation are nothing more than the constant
terms of the asymptotic expansion of the smoothed partial sums. This is not a coincidence; we
will explain the equivalence of these two interpretations of such sums (in the model case when
the analytic continuation has only finitely many poles and does not grow too fast at infinity)
below the fold.
This interpretation clears up the apparent inconsistencies alluded to earlier. For instance, the
sum
consists only of non-negative terms, as does its smoothed

partial sums
(if
forces the highest-order term

is non-negative). Comparing this with (13), we see that this


to be non-negative (as indeed it is), but does not prohibit

the lower-order constant term

from being negative (which of course it is).

Similarly, if we add together (12) and (11) we obtain

while if we subtract from (12) we obtain

These two asymptotics are not inconsistent with each other; indeed, if we shift the index of
summation in (17), we can write

and so we now see that the discrepancy between the two sums in (8), (9)come from the shifting
of the cutoff
, which is invisible in the formal expressions in (8), (9) but become
manifestly present in the smoothed sum formulation.

Exercise 2 By Taylor expanding


and using (11), (18)show
that (16) and (17) are indeed consistent with each other, and in particular one
can deduce the latter from the former.
1. Smoothed asymptotics
We now prove (11), (12), (13), (14). We will prove the first few asymptotics byad hoc methods,
but then switch to the systematic method of the Euler-Maclaurin formula to establish the general
case.
For sake of argument we shall assume that the smooth cutoff
is supported in the
interval
(the general case is similar, and can also be deduced from this case by redefining
the parameter). Thus the sum
is now only non-trivial in the range
.
To establish (11), we shall exploit the trapezoidal rule. For any smooth function
on an interval
, we see from Taylor expansion that

for any

. In particular we have

, and

and

eliminating

, we conclude that

Summing in , we conclude the trapezoidal rule

We apply this with


conclude that

, which has a

norm of

But from (15) and a change of variables, the left-hand side is just

from the chain rule, and

. This gives (11).

The same argument does not quite work with (12); one would like to now set
,
but the
norm is now too large (
instead of
). To get around this we have to
refine the trapezoidal rule by performing the more precise Taylor expansion

where

. Now we have

and

We cannot simultaneously eliminate both


expansion

and

. However, using the additional Taylor

one obtains

and thus on summing in , and assuming that


(by telescoping series)

We apply this with

vanishes to second order at

, one has

. After a few applications of the chain rule and product

rule, we see that


gives (12).

; also,

, and

. This

The proof of (13) is similar. With a fourth order Taylor expansion, the above arguments give

and

Here we have a minor miracle (equivalent to the vanishing of the third Bernoulli number
the
term is automatically eliminated when we eliminate the
term, yielding

) that

and thus

With
vanish, and the

, the left-hand side is


norm is
, giving (13).

, the first two terms on the right-hand side

Now we do the general case (14). We define the Bernoulli numbers


formula

for all

, or equivalently

recursively by the

The first few values of

can then be computed:

From (19) we see that

for any polynomial


identity with

(with
being the -fold derivative of ); indeed, (19)is precisely this
, and the general case then follows by linearity.

As (20) holds for all polynomials, it also holds for all formal power series (if we ignore
convergence issues). If we then replace by the formal power series

we conclude the formal power series (in ) identity

leading to the familiar generating function

for the Bernoulli numbers.


If we apply (20) with

equal to the antiderivative of another polynomial

, we conclude that

which we rearrange as the identity

which can be viewed as a precise version of the trapezoidal rule in the polynomial case. Note that
if has degree , the only the summands with
can be non-vanishing.
Now let

be a smooth function. We have a Taylor expansion

for

and some polynomial

for

and

of degree at most

. We conclude that

Translating this by an arbitrary integer

(which does not affect the

Summing the telescoping series, and assuming that


we conclude the Euler-Maclaurin formula

We apply this with

; also

vanishes to a sufficiently high order at

. The left-hand side is

vanish except for the


term, which is
product rule and chain rule (or by viewing
function
) we see that

norm), we obtain

. All the terms in the sum

. Finally, from many applications of the


where is the smooth
, and the claim (14) follows.

Remark 2 By using a higher regularity norm than the


the error term
can in fact be improved to
if is sufficiently smooth.

norm, we see that


for any fixed
,

Exercise 3 Use (21) to derive Faulhabers formula (10). Note how the presence
of boundary terms at cause the right-hand side of (10) to be quite different
from the right-hand side of (14); thus we see how non-smooth partial
summation creates artefacts that can completely obscure the smoothed
asymptotics.
2. Connection with analytic continuation
Now we connect the interpretation of divergent series as the constant term of smoothed partial
sum asymptotics, with the more traditional interpretation via analytic continuation. For sake of

concreteness we shall just discuss the situation with the Riemann zeta function series
though the connection extends to far more general series than just this one.

In the previous section, we have computed asymptotics for the partial sums

when is a negative integer. A key point (which was somewhat glossed over in the above
analysis) was that the function
was smooth, even at the origin; this was implicitly used
to bound various
norms in the error terms.
Now suppose that is a complex number with
, which is not necessarily a negative
integer. Then
becomes singular at the origin, and the above asymptotic analysis is not
directly applicable. However, if one instead considers the telescoped partial sum

with equal to near the origin, then by applying (22) to the


function
(which vanishes near the origin, and is now
smooth everywhere), we soon obtain the asymptotic

Applying this with


concludes that

equal to a power of two and summing the telescoping series, one

for some complex number


which is basically the sum of the various
terms
appearing in (23). By modifying the above arguments, it is not difficult to extend this asymptotic
to other numbers than powers of two, and to show that
is independent of the choice of
cutoff .
From (24) we have

which can be viewed as a definition of in the region


. For instance, from (14), we
have now proven (3) with this definition of
. However it is difficult to compute
exactly
for most other values of .

For each fixed , it is not hard to see that the expression


is complex
analytic in . Also, by a closer inspection of the error terms in the Euler-Maclaurin formula
analysis, it is not difficult to show that for in any compact region of
,
these expressions converge uniformly as
. Applying Moreras theorem, we conclude
that our definition of
is complex analytic in the region
.
We still have to connect this definition with the traditional definition (1) of the zeta function on
the other half of the complex plane. To do this, we observe that

for

for

large enough. Thus we have

. The point of doing this is that this definition also makes sense in the

region
(due to the absolute convergence of the sum
and integral
.
By using the trapezoidal rule, one also sees that this definition makes sense in the
region
, with locally uniform convergence there also. So we in fact have a globally
complex analytic definition of
, and thus a meromorphic definition of
complex plane. Note also that this definition gives the asymptotic

near

, where

on the

is Eulers constant.

We have thus seen that asymptotics on smoothed partial sums of gives rise to the familiar
meromorphic properties of the Riemann zeta function
. It turns out that by combining the
tools of Fourier analysis and complex analysis, one can reverse this procedure and deduce the
asymptotics of from the meromorphic properties of the zeta function.
Lets see how. Fix a complex number with
, and a smooth cutoff
function
which equals one near the origin, and consider the expression

where

is a large number. We let

be a large number, and rewrite this as

where

The function
representation

is in the Schwartz class. By the Fourier inversion formula, it has a Fourier

where

and so (26) can be rewritten as

The function
is also Schwartz. If
sum and use (1) to rewrite (26) as

is large enough, we may then interchange the integral and

Now we have

integrating by parts (which is justified when

where

We can thus write (26) as a contour integral

is large enough) we have

Note that is compactly supported away from zero, which makes


an entire function
of
, which is uniformly bounded whenever is bounded. Furthermore, from repeated
integration by parts we see that
is rapidly decreasing as
, uniformly for in a
compact set. Meanwhile, standard estimates show that
is of polynomial growth in
for

in a compact set. Finally, the meromorphic function

at
(with residue
theorem, we can write (26) as

) and at

(with residue

for any
. Using the various bounds on and
integration by parts we have
and

has a simple pole


). Applying the residue

, we see that the integral is

. From

and thus we have

for any

, which is (14) (with the refined error term indicated in Remark2).

The above argument reveals that the simple pole of


at
is directly connected to
the
term in the asymptotics of the smoothed partial sums. More generally, if a
Dirichlet series

has a meromorphic continuation to the entire complex plane, and does not grow too fast at
infinity, then one (heuristically at least) has the asymptotic

where ranges over the poles of


the famous explicit formula

, and

are the residues at those poles. For instance, one has

where is the von Mangoldt function, are the non-trivial zeroes of the Riemann zeta function
(counting multiplicity, if any), and
is an error term (basically arising from the trivial zeroes of
zeta); this ultimately reflects the fact that the Dirichlet series

has a simple pole at


(with residue
) and simple poles at every zero of the zeta function
with residue
(weighted again by multiplicity, though it is not believed that multiple zeroes
actually exist).
The link between poles of the zeta function (and its relatives) and asymptotics of (smoothed)
partial sums of arithmetical functions can be used to compare elementary methods in analytic
number theory with complex methods. Roughly speaking, elementary methods are based on
leading term asymptotics of partial sums of arithmetical functions, and are mostly based on
exploiting the simple pole of at
(and the lack of a simple zero of Dirichlet -functions
at
); in contrast, complex methods also take full advantage of the zeroes of and
Dirichlet -functions (or the lack thereof) in the entire complex plane, as well as the functional
equation (which, in terms of smoothed partial sums, manifests itself through the Poisson
summation formula). Indeed, using the above correspondences it is not hard to see that the prime
number theorem (for instance) is equivalent to the lack of zeroes of the Riemann zeta function on
the line
.
With this dictionary between elementary methods and complex methods, the Dirichlet hyperbola
method in elementary analytic number theory corresponds to analysing the behaviour of poles
and residues when multiplying together two Dirichlet series. For instance, by using the
formula (11) and the hyperbola method, together with the asymptotic

which can be obtained from the trapezoidal rule and the definition of , one can obtain the
asymptotic

where
is the divisor function (and in fact one can improve the
bound
substantially by being more careful); this corresponds to the fact that the Dirichlet series

has a double pole at

with expansion

and no other poles, which of course follows by multiplying (25) with itself.

Remark 3 In the literature, elementary methods in analytic number theorem


often use sharply truncated sums rather than smoothed sums. However, as
indicated earlier, the error terms tend to be slightly better when working with
smoothed sums (although not much gain is obtained in this manner when
dealing with sums of functions that are sensitive to the primes, such as , as the
terms arising from the zeroes of the zeta function tend to dominate any saving in
this regard).
Share this:

You might also like