e is transcendental

Robert Hines
November 6, 2015

e is irrational and transcendental numbers exist

The irrationality of e is straightforward to prove, and has been known since at least Euler
(who first called e, “e”).
Theorem. e is irrational.
n
X xk
Proof. Let Hn (x) = , Hn = Hn (1). Then
k=0
k!
∞
X 1
e − Hn =
k=n+1
k!
1  1 1 
= 1+ + + ...
(n + 1)! n + 2 (n + 2)(n + 3)
1  1 1 
< 1+ + + . . .
(n + 1)! n + 1 (n + 1)2
1 1 1 n+1 1 1
= 1 = = .
(n + 1)! 1 − n+1 (n + 1)! n n n!

If e = pq , then q!(e − Hq ) ∈ Z, but

1
0 < q!(e − Hq ) < ,
q
a contradiction.
[Fun fact: Hn (x) ∈ Q[x] is irreducible for all n].

The moral of the proof is that Hn (a rational number) approximates e too well.
Consider the following
Lemma (Liouville, 1844). If ξ is a real algebraic number of degree n > 1, then there is
a constant A > 0 (depending on ξ) such that
h A
ξ− > n.
k k

1
Proof. Suppose p(x) ∈ Z[x] is irreducible of degree n with p(ξ) = 0. Then

p(ξ) − p(h/k) = (ξ − h/k)p0 (α)

for some α between ξ and h/k by the mean value theorem. The left hand side is a non-
zero rational number (p(ξ) = 0 and p is irreducible so h/k is not a root) with denominator
less than k n so that we get
1 h
n
≤ ξ − sup{p0 (x) : x ∈ (ξ − 1, ξ + 1)}.
k k

The above result can be improved to

Theorem (Thue-Siegel-Roth). For all  > 0, there are only finitely many rational solu-
tions to
h 1
ξ− < 2+
k k
if ξ is algebraic and irrational.
The 2 +  exponent is the best possible since we have the following
Proposition. If ξ ∈ R is irrational then there are infinitely many rationals p/q such
that
|ξ − p/q| < 1/q 2 .
Proof. This is an application of the pigeonhole principle. Two of the n + 1 numbers
1, {kξ} (the fractional part of kξ) for 1 ≤ k ≤ n must lie in one of the n subintervals
(i/n, (i + 1)/n], 0 ≤ i ≤ n − 1 of (0, 1]. Hence there is a p and 1 ≤ q ≤ n such that

|qξ − p| < 1/n, i.e |ξ − p/q| < 1/nq ≤ 1/q 2 .

Infinitely many of these p/q must be distinct, else |ξ − p/q| takes on a minimum value,
say larger than 1/n for some n, and the above construction gives a contradiction.
Transcendental numbers exist (by cardinality arguments - thanks Cantor!), but let’s
exhibit one explicitly (as Liouville did).
Proposition. ξ = ∞ −n!
P
n=0 10 is transcendental.
Proof. Let kj = 10j! , hj = 10j! jn=0 10−n! . Then (hj , kj ) = 1 (as hj ≡ 1 mod 10) and
P

∞ ∞
hj X X
ξ− = 10−n! < 10−n
kj n=j+1 n=(j+1)!
1 10
= 10−(j+1)! = (10j! )−j
1 − 1/10 9 · 10j!
< Aj kj−j ,

where A(j) → 0 as j → ∞ so that ξ is transcendental by the lemma above.

e is transcendental
We now begin the proof that e is transcendental (Hermite, 1873). We have to be able
to simultaneously approximate ex at different values to obtain a contradiction similar to
that given above for the irrationality P
of e.
For a polynomial f (x), let F (x) = ∞ (i)
i=0 f (x). Integrating by parts a bajillion times,
we get Z x
e x
f (t)e−t dt = F (0)ex − F (x) (the “Hermite identity”).
0
Consider the specific polynomial
xp−1 (x − 1)p · · · · · (x − n)p
f (x) = .
(p − 1)!
(n will be the degree of the fictitious minimal polynomial for e over Q and p will be a
large prime).
We have the following estimate for 0 ≤ k ≤ n:
Z k
k
|e F (0) − F (k)| = e k
f (t)e−t dt
0
n
≤ ne sup {f (t)}
t∈[0,n]

n (np )n
p−1
=
(p − 1)!

which goes to zero as p → ∞ for a fixed n. [Mildly interesting: this proof requires the
existence of infinitely many primes.] We now show that such an estimate is impossible if
e is algebraic by showing that nk=1 F (0)ek − F (k) is an integer between 0 and 1.
P

1. F (0) ∈ Z\pZ for p > n:

We have f (x) = a(x)b(x) where
xp−1
a(x) = , b(x) = (x − 1) · · · · · (x − n)p ,
(p − 1)!
so that
N  
(N )
X
(i) (N −i) N
f (x) = a (x)b (x) .
i=0
i
Note that a (0) = 0 unless i = p − 1 in which case a(p−1) (0) = 1. Hence
(i)

∞ ∞ X N  
X
(N )
X
(i) (N −i) N
F (0) = f (0) = a (0)b (0)
N =0 N =0 i=0
i
∞  
X N
= b(N −(p−1)) (0)
N =p−1
p−1
2
= b(0) + p(. . . ) = (−1)p n!p + p(. . . ) ∈ Z\pZ (remember that p > n).
 2. F (k) ∈ pZ for 1 ≤ k ≤ n:
We have f (x) = c(x)d(x) where
(x − k)p xp−1 (x − 1)p · · · · · (x − n)p
c(x) = , d(x) = .
(p − 1)! (x − k)p
Note that c(i) (k) = 0 unless i = p in which case c(p) (k) = p. Hence
∞ X N  
X
(i) (N −i) N
F (k) = c (k)d (k)
N =0 i=0
i
∞  
X N
=p d(N −p) (k) ∈ pZ.
N =p
p
Pn k
Now, if e were algebraic, say k=0 ck e = 0, ck ∈ Z, c0 6= 0, and p > |c0 |, then
n
X
1≤ ck F (k) (because c0 , F (0) ∈ Z\pZ, F (k) ∈ pZ)
k=0
n n
!
X X
= ck F (k) − ck ek F (0)
k=0 k=0
n
X
ck F (k) − ek F (0)


=
k=0
n
X
≤M |F (k) − ek F (0)| (where M = max{|ck |})
k
k=0

which is less than 1 for p large as shown above. Hence e is transcendental.

What about π?
Here is a proof that π is irrational in the
P∞ spirit of Hermite.
For a polynomial f (x), let F (x) = n=0 (−1)n f (2n) (x) (this mimics sin x in the way
we mimicked ex before). We have
Z π
d 0
(F (x) sin x − F (x) cos x) = f (x) sin x, f (x) sin xdx = F (0) + F (π).
dx 0

If π = a/b were rational, consider the polynomial

bn n 1
f (x) = x (π − x)n = xn (a − bx)n ∈ Q[x].
n! n!
We have bounds
Z π
bn π 2n π 2(π 2 b)n
Z
0< f (x) sin xdx ≤ sin xdx = → 0 as n → ∞.
0 n! 0 n!
Note that f (k) (0) = 0 for 0 ≤ k < n as f has a zero of order n at zero. We also have
f (k) (0) ∈ Z for k ≥ n by the following (easy) lemma.
Lemma. For p(x) ∈ Z[x], k! divides all the coefficients of p(k) (x).
dk n n

n−k
Proof. dx k x = k! k x for k ≤ n and higher derivatives are zero.

Hence f (k) (0) ∈ Z for all k. Finally note that

f (x) = f (π − x), f (k) (x) = (−1)k f (k) (π − x)

so that f (k) (π) = (−1)k f (k) (0) ∈ Z for all k. Therefore F (0) + F (π) ∈ Z, a contradiction,
and π is irrational.
We can prove that π is transcendental using the methods we used for e, although
the details are slightly more tedious. We start with the identity eπi + 1 = 0. If πi were
algebraic (degree n), we would have
n
Y X P m
X
0= (1 + eγi ) = e i i γi = a + e αi
i=1 i ∈{0,1} i=1

where the γi are the galois conjugates of πi, a = 2n − m are the number of zero exponents
in the first sum (note that a ≥ 1), and the αi are the non-zero exponents in the first sum.
Thinking about symmetric functions for a while (details omitted), we see that
n
!
Y X
φ(x) = x− i γi ∈ Q[x].
i ∈{0,1} i=1

Divide by xa and clear denominators to get a polynomial

m
X
ψ(x) = bi xi ∈ Z[x], bm > 0, b0 6= 0
i=0

whose roots are exactly the αi . Furthermore, assume bm αi is an algebraic integer for all
i.
Once again we apply the “Hermite identity,” this time to the polynomial
(m−1)p m
bm bmp
m
Y
f (x) = xp−1 ψ p (x) = xp−1 (x − αi )p .
(p − 1)! (p − 1)! i=1

Plug in x = αi and sum over i to get

m
X m
X Z αi
−aF (0) − F (αi ) = αi
e f (t)e−t dt.
i=1 i=1 0

Our goal, as before, is to show that the LHS is a non-zero integer but that the RHS can
be made arbitrarily small. We have
!p
Y
F (0) = (−1)mp bmp
m αi ∈ Z\pZ
i
for large p. We also have
m
X X Y
F (αi ) = pbmp
m αip−1 (αi − αj )p ∈ pZ
i=1 i j6=i

for large p because it is symmetric in αi and the denominator is cleared by bmp m .

We now estimate the integral on the RHS:
Z αi

p
eαi
f (t)e−t dt ≤ |αi ||bm−1
m ||ψ|(|α i |) /(p − 1)! → 0
0
as p → ∞.

Generalizations
Theorem (Lindemann-Weierstrass, 1885). If α1 , . . . , αk are distinct algebraic numbers,
then eα1 , . . . , eαk are linearly independent over Q.
We also have the solution of Hilbert’s seventh problem
Theorem (Gelfond-Schneider, 1934). For algebraic a 6∈ {0, 1} and irrational algebraic
b, ab is transcendental.
√
So numbers such as 2 2 , ii are transcendental.
Another generalization due to Lang (an axiomatization of Schneider’s methods) is
Theorem. Suppose K is a number field, {fi }ni=1 meromorphic functions of order ≤ ρ
such that K({fi }i ) has transcendence degree ≥ 2 over K and K[{fi }i ] is closed under
differentiation. If {wj }m
j=1 are distinct complex numbers such that fi (wj ) ∈ K for all i, j
then m ≤ 20ρ[K : Q].
Theorem (Hermite-Lindemann). eα is transcendental for all α ∈ Q\{0}.
Proof. The proof that π is transcendental directly generalizes to this. Or, take the
meromorphic functions in the theorem above to be z, ez and K to be Q(α, eα ). Theses
function take values in K for z any integer multiple of α.
Theorem (Schneider). If ℘ is a Weierstrass a function with g2 , g3 algebraic, then ℘(α)
is transcendental for all Q\{0}.
Sketch. First the relevant definitions. If Λ ⊆ C is a rank two lattice, define
1 X 1 1
℘(z; Λ) = 2 + 2
− 2.
z (z − w) w
w∈Λ\{0}

Then ℘ satisfies the algebraic differential equation

℘02 = 4℘3 − g2 ℘ − g3
where X 1 X 1
g2 = 60 , g3 = 140 .
w4 w6
w∈Λ\{0} w∈Λ\{0}

Addition formula, etc.?????

 A far-reaching generalization of the theorem of Gelfond-Schneider is

Theorem (Baker, 1966). If {αi }ni=1 and {βi }ni=0 are algebraic (and αi 6= 0) of degree at
most d and with heights at most A, B (for {αi }ni=1 and {βi }ni=0 respectively) then

Λ := β0 + β1 log α1 + · · · + βn log αn

is either zero or |Λ| > B −C for an effectively computable constant C depending onlyP on
n, d, A, and {log αi }i . [The height of an algebraic number γ is maxi {|ci |} where i ci xi
is the minimal polynomial of γ over Z.]

For related results and applications, such as the class number one problem for imag-
inary quadratic fields:
√
Q( −d) with d > 0 has class number one iff d ∈ {1, 2, 3, 7, 11, 19, 43, 67, 163},

Baker was awarded a Fields medal in 1970.

References
[1] Baker, Alan, Transcendental Number Theory

[2] Niven, Ivan, Irrational Numbers

[3] http://math.stanford.edu/~ksound/TransNotes.pdf