# The transcendence of π

> An annotated edition of this paper on Fermat's Library, where scientists and readers explain its ideas, equations and history in the margins. Annotated edition: https://fermatslibrary.com/s/the-transcendence-of-pi

- **Authors:** Ivan Niven
- **Topic:** Mathematics
- **Original:** https://s3-us-west-1.amazonaws.com/fermatslibrary/uploads%2F1455947543271-i4v1pv6bxcd6lxr-8b7f7b57a7f9c431bb8a967d88998a62%2Fniven1939.pdf
- **Pages:** 3
- **Annotations:** 25 comments in 16 threads
- **Featured:** Fermat's Library Journal Club
- **Annotated edition:** https://fermatslibrary.com/s/the-transcendence-of-pi

## Annotations

### Annotation 1 (page 1)

Beside the passage: “THE TRANSCENDENCE OF c IVAN NIVEN,* University of Pennsylvania”

**Annotation** by Luís Batalha (2016-02-20):

> A transcendental number is a real or complex number that is not algebraic - that is, it is not a root of a non-zero polynomial equation with rational coefficients.

### Annotation 2 (page 1)

Beside the passage: “THE TRANSCENDENCE OF c IVAN NIVEN,* University of Pennsylvania”

**Annotation** by Luís Batalha (2016-02-20):

> Ivan Niven was a Canadian-American mathematician, specializing in number theory. He is famous for completing the solution of most of Waring's problem in 1944. He also has an Erdős number of 1 because he has coauthored a paper with Paul Erdös.
>
> ![](https://upload.wikimedia.org/wikipedia/commons/6/6e/Ivan_Niven.jpg)

### Annotation 3 (page 1)

Beside the passage: “Among the proofs of the transcendence of e, which are in general variations and simplifications of the original proof of Hermite, perhaps the simplest is that of A. Hurwitz.t His solution of the problem contains an ingenious device which we now employ to obtain a relatively simple proof of the transcendence of 7r. We …”

**Annotation** by Matt Yang (2016-02-23):

> It's interesting that we know $e$ and $\pi$ are transcendental numbers but no one has proved that $\pi+e$ or $e\pi$ are transcendental.

**Reply** by Luís Batalha (2016-02-23):

> In fact it is known that one of those two numbers $e\pi$ or $\pi +e$ is transcendental.

**Reply** by Yunjiang Jiang (2017-11-15):

> Luis. Any reference to the latter fact?

**Reply** by Luís Batalha (2017-11-15):

> Check out this video touching on that topic:
> [![](https://i.ytimg.com/vi/ZxA-xD9mUmM/maxresdefault.jpg)](https://www.youtube.com/watch?v=ZxA-xD9mUmM)

### Annotation 4 (page 1)

Beside the passage: “Among the proofs of the transcendence of e, which are in general variations and simplifications of the original proof of Hermite, perhaps the simplest is that of A. Hurwitz.t His solution of the problem contains an ingenious device which we now employ to obtain a relatively simple proof of the transcendence of 7r. We …”

**Annotation** by Matt Yang (2016-02-22):

> Is $i$ an algebraic number?

**Reply** by Luís Batalha (2016-02-22):

> Yes Mike, for instance if you have the polynomial $f(x)=x^2+1$, the coefficients are rational and the roots are $i$ and $-i$.

### Annotation 5 (page 1)

Beside the passage: “Among the proofs of the transcendence of e, which are in general variations and simplifications of the original proof of Hermite, perhaps the simplest is that of A. Hurwitz.t His solution of the problem contains an ingenious device which we now employ to obtain a relatively simple proof of the transcendence of 7r. We …”

**Annotation** by Luís Batalha (2016-02-22):

> It's actually not obvious that the product (or sum) of 2 algebraic numbers is also an algebraic number. To prove this let's first define the Resultant $R(f,g)$ of 2 polynomials $f(x)=a_0x^n+...+a_n$ and $g(x)=b_0x^m+...+b_m$ with $a_0b_0 \neq 0$ as
>
> $$
> R(f,g) = a_0^mb_0^n \prod_{ij}(\alpha_i-\beta_j)
> $$
>
> where the $\alpha_i$ are the roots of $f(x)$ and $\beta_j$ are the roots of $g(x)$. It's not difficult to notice that if the 2 polynomials f and g have a root in common the resultant will be zero.
>
> To prove that the sum of 2 algebraic numbers is also an algebraic number we need to find a polynomial whose roots are $\alpha_i+\beta_j$. To construct such polynomial we will use the resultant of $f(x)$ and $g(t-x)$
>
> $$R(f(x),g(t-x))$$
>
> Now if we do the transformation of variables $x \rightarrow t-x$ the roots of $g(t-x)$ will be $t-\beta_j$ and so
>
> $$R(f(x),g(t-x))=  a_0^mc_0^n \prod_{ij}(\alpha_i-(t-\beta_j))=\\
> =a_0^mc_0^n \prod_{ij}(\alpha_i+\beta_j-t) $$
>
> Now, it's not difficult to see that the resultant itself will be a polynomial in $t$ of degree $mn$ with roots $\alpha_i+\beta_j$.
>
> To prove that the product of 2 algebraic numbers is also an algebraic number we use a different resultant:
>
> $$
> R(f(x),x^mg(t/x))
> $$
>
> $$
> x^mg(1/x)=b_0+...+b_mx^m
> $$
>
> The roots of $x^mg(1/x)$ are $1/\beta_j$ and so the roots of  $x^mg(t/x)$ are $t/\beta_j$. If we now plug that into the resultant we get a polynomial with degree $mn$ in t whose roots obey the equation:
>
> $$
> \alpha_i-t/\beta_j=0\\
> t=\alpha_i\beta_j
> $$

**Reply** by Yunjiang Jiang (2017-11-15):

> It is not obvious to me why the resultant is a polynomial in t with rational coefficients.

### Annotation 6 (page 2)

Beside the passage: “470 THE TRANSCENDENCE OF 7w [October, By equation (1), the elementary symmetric functions of a,, a2, , *? are rational numbers. Hence the elementary symmetric functions of the quantities (3) are rational numbers. It follows that the quantities (3) are roots of (4) 02(x) = 0, an algebraic equation with integral …”

**Annotation** by Luís Batalha (2016-02-22):

> We showed above that the sum of 2 algebraic numbers is also an algebraic number and so it is possible to create a polynomial $\theta_2(x)$ whose roots are $\alpha_i+\alpha_j$. The same applies to the sum of 3,4,...n algebraic numbers.

**Reply** by Aaron Meurer (2016-02-23):

> That's true, but the argument here is just based on the fact that coefficients of a polynomial are (up to a sign) the elementary symmetric functions on its roots.

### Annotation 7 (page 2)

Beside the passage: “The deletion of zero roots (if any) from equation (7) gives (8) 0(X) = cxr + ClXr-l + + Cr 0, whose roots /i, 02 O * r are the non-vanishing exponents in the expansion of (2), and whose coefficients are integers. Hence (2) may be written in the form (9) e#1 + e#2 + ... + er + k= 0 where k is a positive integer.”

**Annotation** by Luís Batalha (2016-02-22):

> Note that if we expand expression (2) we get
>
> $$
> 1+ ...+e^{\alpha_i+\alpha_j}+e^{\alpha_i+\alpha_j+\alpha_k}+...e^{\overbrace{\alpha_i+...+\alpha_l }^{n}}=0
> $$

### Annotation 8 (page 2)

Beside the passage: “e-xF(x) - eF (0) = f - e-f(t)d .”

**Annotation** by Luís Batalha (2016-02-22):

> $$
> \frac{d(e^{-x}F(x))}{dx}=-e^{-x}F(x)+e^{-x}F'(x)=\\
> =-e^{-x}F(x)+e^{-x}(F(x)-f(x))=-e^{-x}f(x)
> $$

### Annotation 9 (page 2)

Beside the passage: “F(x) - exF(O)= - X e(1-7)x f(rx)dT .”

**Annotation** by Luís Batalha (2016-02-22):

> If we make the substitution $\xi = \tau x $ the integration limits $\xi = 0 \rightarrow \tau =0$ and $\xi = x \rightarrow \tau =1$. If we multiply both sides of the equation by $e^{x}$ we get this expression.

### Annotation 10 (page 3)

Beside the passage: “rr r 1 (12) 1 F + kF(O) = -E: j e (l-T)#if( rfj)dr. j=l j=j”

**Annotation** by Matt Yang (2016-02-22):

> How does he end up with $kF(0)$?

**Reply** by Luís Batalha (2016-02-22):

> If you start with (9)
> $$
> e^{\beta_1}+...+e^{\beta_r}+k=0 \equiv\\
> k=-(e^{\beta_1}+...+e^{\beta_r})\equiv \\
>
> kF(0)= -\sum e^{\beta_i}F(0)
> $$

### Annotation 11 (page 3)

Beside the passage: “Zf(t)@D =0, for 0 < t < p. j=l Also by (10) the polynomial obtained by multiplying f(x) by (p - 1)! has integral coefficients. Since the product of p consecutive positive integers is divisible by p!, the pth and higher derivatives of (p -l1)!f(x) are polynomials in x with integral coefficients divisible by p!. Hence …”

**Annotation** by Luís Batalha (2016-02-22):

> This is not difficult to see because of the $x^{p-1}$ factor and also because $\theta(\beta_j) = 0$.

### Annotation 12 (page 3)

Beside the passage: “Zf(t)@D =0, for 0 < t < p. j=l Also by (10) the polynomial obtained by multiplying f(x) by (p - 1)! has integral coefficients. Since the product of p consecutive positive integers is divisible by p!, the pth and higher derivatives of (p -l1)!f(x) are polynomials in x with integral coefficients divisible by p!. Hence …”

**Annotation** by Luís Batalha (2016-02-22):

> Note that $f(x)$ can be written as
>
> $$
> f(x)=\frac{c^sx^{p-1}(cx^r+...+c_r)^p}{(p-1)!}
> $$
>
> The highest term in $f(x)$ is of the form $x^{p-1}x^{rp}=x^{p+s}$.
>
> $$
> f(x)=\frac{c^s(wx^{p+s}+...+w_rx^{p-1})}{(p-1)!}
> $$
>
> where $w,...,w_r$ are integers that have resulted from expanding the expression $(cx^r+...+c_r)^p$.
>
>
> It's not difficult to see that for any derivative higher than $p$ the coefficients will be divisible by $pc^s$.

### Annotation 13 (page 3)

Beside the passage: “where the kt are integers. It follows that r p+s F F(o j) = p E kt .”

**Annotation** by Luís Batalha (2016-02-23):

> Using expression (11)
>
> $$
> F(\beta_j)=f(\beta_j)+f^{(1)}(\beta_j)+...+f^{(s+p+1)}(\beta_j)
> $$
> and knowing that $\sum^{r}_{j=1}f^{(t)}(\beta_j)=0$ for $0\leq t<p$ we get
> $$
> \sum^{r}_{j=1} F(\beta_j)=\sum^{r}_{j=1} f(\beta_j)+ \sum^{r}_{j=1} f^{(1)}(\beta_j)+...+ \sum^{r}_{j=1}  f^{(s+p+1)}(\beta_j)
> $$
>
> All the sums where the derivatives have an order $<p$ will be zero. If now we use the result $\sum^{r}_{j=1}f^{(t)}(\beta_j)=pk_t$, we end up with
>
> $$
> \sum^{r}_{j=1} F(\beta_j)=p\sum^{p+s}_{t=p}k_t
> $$

### Annotation 14 (page 3)

Beside the passage: “f(t)(0) = 0, (t = O, 1, , p - 2), f(P-1)(0) = COCrp, f(t)(0) = pKt, (t = p, p + 1, , p + s),”

**Annotation** by Luís Batalha (2016-02-23):

> As shown in a comment above, $f(x)$ can be written as
>
> $$
> f(x)=\frac{c^s(wx^{p+s}+...+w_rx^{p-1})}{(p-1)!}
> $$
>
> Since the lowest term in $f(x)$ is of order $p-1$, $f^{(t)}(x)$ for $t<p-1$ will always have terms in $x$ and so $f^{(t)}(0)$ will of course be zero.
>
> $f^{(p-1)}(0)=\frac{(p-1)!c^s w_r}{(p-1)!}$ where $w_r$ is the last term in the binomial expansion of $(cx^r+...+c_r)^p$ and so $w_r=c_r^p$.
>
> Finally for $t\geq p$, $f^{(t)}(0)$ will only have one term that will not vanish, the term of order $t$. The result is simply a product of $p$, $c^s$ and the binomial expansion term $w_t$. All the terms are integers and so we can group in the form $pK_t$, where $K_t$ in an integer.

### Annotation 15 (page 3)

Beside the passage: “where the Kt are integers. If p is chosen greater than each of k, C, Cr (possible since the number of primes is infinite), the desired result follows from (11). Finally, the right side of (12) equals f f { J Cr3 (7-1) e(lT):jdr. j=1 c o (p-1)! This is a finite sum, each term of which may be made as small as we wish by …”

**Annotation** by Erik José Amézquita Morataya (2016-09-04):

> Isn't a $\tau$ missing? I mean, shouldn't the integral be
> $$\int_0^1\frac{\{c^r\beta_j\tau\theta(\tau\beta_j)\}^p}{\beta_j\tau(p-1)!}e^{(1-\tau)\beta_j}d\tau$$?

### Annotation 16 (page 3)

Beside the passage: “{Cr 0io(-rj)} lrm 3 = 0. P-409 (p - 1)!”

**Annotation** by Luís Batalha (2016-02-23):

> We have finally reached a contradiction since the left side of (12) is an integer and the right side is as small as we wish by choosing a very large $p$, which means that $\pi$ cannot be an algebraic number!

**Reply** by Matt Yang (2016-02-23):

> How do you prove the limit is zero?

**Reply** by Luís Batalha (2016-02-23):

> One way of understanding this limit is thinking about the exponential as $\sum_{n=0}^{\infty}\frac{x^n}{n!}=e^x$. This is a convergent series which means that the weight of the terms with large $n$ will have to tend to zero, otherwise it would diverge.
>
> $$
> \lim_{n \rightarrow \infty}  \frac{x^n}{n!} = 0
> $$
>
> This is exactly the same limit we have in the paper if we do $x=c^r\beta_j\theta(\tau\beta_j)$.

## Full text

Extracted from the PDF. [Annotation N] marks where each annotation above sits.

#### Page 1

1939] THE TRANSCENDENCE OF ir 469

for XI- X_X2, with (x, y) in a neighborhood of CO, and y-yo(x) and y'-yoi(x) not both zero, then CO furnishes a proper (or improper) strong relative minimum for Jf= f2f(x, y, y')dx, G(x, y, y, y, y')-f(x, y y, y')- )-(y-y)fV,(x, y, y') - (y' -')fy, (x, y, y'). Examples were given which satisfy these sufficient conditions but which do not yield to classical methods. 3. The finite abstract groups in which every subgroup is invariant (Hamiltonian groups) are well known, as are also the groups in which every subgroup is abelian. Classifying the finite abstract groups according to the number of complete sets of non-invariant conjugate subgroups contained in them, Professor Sigley discussed the class of groups which contain a single complete set of non-invariant conjugate subgroups. Necessary and sufficient conditions that a group belong to this class were obtained. 4. It was shown by Professor Finkel how the derivatives of ax and sin-' x may be derived directly by the A-process, without reverting to the derivatives of the logarithmic and sine functions. 5. Mr. Gillam sketched a development of three-dimensional analytic euclidean geometry (free of coordinates) with "point" as the single undefined element and "distance" as the only primitive relation. Lines and planes were defined as determinantal loci. 6. Mr. Wulftange presented a new method for the numerical solution of quadratic congruences which is believed to possess certain practical advantages. L. M. BLUMENTHAL, Secretary

THE TRANSCENDENCE [Annotation 1] OF c IVAN NIVEN,* University [Annotation 2] of Pennsylvania

Among the proofs of the transcendence of e, which are in general variations [Annotation 3] and simplifications of the original proof of Hermite, perhaps the simplest is that of A. Hurwitz.t His solution of the problem contains an ingenious device which we now employ to obtain a relatively simple proof of the transcendence of 7r. We assume that ir is an algebraic number, and show that this leads to a contradiction. Since the product of two algebraic numbers is an algebraic number, the quantity i7r is a [Annotation 4] root of an algebraic equation with integral coefficients [Annotation 5] (1) 01(x) = O, whose roots are aci=ir, a2, a3, * * a.n Using Euler's relation e$r+1 =0, we have

(2) (eal + 1)(ea2 + 1) * (ean + 1) = 0. We now construct an algebraic equation with integral coefficients whose roots are the exponents in the expansion of (2). First consider the exponents (3) ai + a2, ai + as, a2 + C . . * an-1 + an * Harrison Research Fellow. t A. Hurwitz, Beweis der Transendenz der Zahl e, Mathematische Annalen, vol. 43, 1893, pp. 220-221 (also in his Mathematische Werke, vol. 2, pp. 134-135).

#### Page 2

470 THE TRANSCENDENCE OF 7w [October, By equation (1), the elementary symmetric functions of a,, a2, , *? are rational numbers. Hence the elementary symmetric functions of the quantities (3) are rational numbers. It follows that the quantities (3) are roots of (4) [Annotation 6] 02(x) = 0, an algebraic equation with integral coefficients. Similarly, the sums of the a's taken three at a time are the nC3 roots of (5) 03(X) = 0 Proceeding thus, we obtain (6) 04(X) = 0, 05(X) = 0, , . . v 0,(X) = 0, algebraic equations with integral coefficients, whose roots are the sums of the a's taken 4, 5, , n at a time respectively. The product equation (7) 61(X)02(X) . ..(x) = 0 has roots which are precisely the exponents in the expansion of (2).

The deletion of zero roots (if any) from equation (7) gives (8) 0(X) = cxr [Annotation 7] + ClXr-l + + Cr 0, whose roots /i, 02 O * r are the non-vanishing exponents in the expansion of (2), and whose coefficients are integers. Hence (2) may be written in the form (9) e#1 + e#2 + ... + er + k= 0 where k is a positive integer.

We define ( 10) ~~~~~~~~csxP1I{6( x) } P~ (10) f(x) - (sp- i)! 1

where s = rp -1, and p is a prime to be specified. Also we define (11) F(x) = f(x) + f(')(x) + f(2)(x) + . . . + f(s+p+l)(x) noting, with thanks to Hurwitz, that the derivative of e-xF(x) is e-xf(x). Hence we may write

e-xF(x) - eF (0) = f - e-f(t)d [Annotation 8] .

The substitution t = -x produces

F(x) - exF(O)= [Annotation 9] - X e(1-7)x f(rx)dT .

Let x range over the values 31,/2, .2 , * r and add the resulting equations. Using (9). we obtain

#### Page 3

1939] THE TRANSCENDENCE OF 1r 471

rr r 1 (12) 1 F + kF(O) = -E: j e (l-T)#if( rfj)dr. j=l j=j [Annotation 10]

This result gives us the contradiction we desire. For we shall choose the prime p to make the left side a non-zero integer, and the right side as small as we please. By (10), we have

Zf(t)@D =0, for 0 < t < p. j=l [Annotation 11] Also by (10) the polynomial obtained by multiplying f(x) by (p - 1)! has integral coefficients. Since the product of p consecutive positive integers is divisible by p!, the pth and higher derivatives of (p -l1)!f(x) are polynomials in x with integral coefficients divisible by p!. Hence the pth and higher derivatives of f(x) are polynomials with integral [Annotation 12] coefficients each of which is divisible by p. That each of these coefficients is also divisible by c8 is obvious from the definition (10). Thus we have shown that, for t_p, the quantity f(t)(Bij) is a polynomial in /3s of degree at most s, each of whose coefficients is divisible by pco. By (8), a symmetric function of ,i, 132, Oir with integral coefficients and of degree at most s is an integer provided each coefficient is divisible by c8 (by the fundamental theorem on symmetric functions). Hence

f(t)() = pk, (t= p,p + 1, , p + s), j=l

where the kt are integers. It follows that r p+s F F(o j) [Annotation 13] = p E kt .

In ordrtcmpltehet=p In order to complete the proof that the left side of (12) is a non-zero integer, we now show that kF(O) is an integer prime to p. From (10) it is clear that

f(t)(0) = 0, (t = O, 1, , p - 2), f(P-1)(0) [Annotation 14] = COCrp, f(t)(0) = pKt, (t = p, p + 1, , p + s),

where the Kt are integers. If p is chosen greater than each of k, C, Cr (possible since the number of primes is infinite), the desired result follows from (11). Finally, the right side of (12) equals f f { J Cr3 (7-1) e(lT):jdr. j=1 c o (p-1)! This is a finite sum, each term of which may be made as small as we wish [Annotation 15] by choosing p very large, because

{Cr 0io(-rj)} lrm [Annotation 16] 3 = 0. P-409 (p - 1)!

---

Source: [Fermat's Library](https://fermatslibrary.com/) — https://fermatslibrary.com/s/the-transcendence-of-pi. More annotated papers: https://fermatslibrary.com/llms.txt
