Transcendence theory in Lean 4

6 The four exponentials theorem in transcendence degree one

The four exponentials conjecture holds when the four logarithms generate a field of transcendence degree one: Theorem 6.31, proved independently by Waldschmidt and by Brownawell. The proof here is Waldschmidt’s of 1973 [ Wal73 ] with the tools of his 1971 paper [ Wal71 ] , so it needs neither a lower bound for linear forms in logarithms nor Philippon’s zero estimate. The printed 1973 proof meets one hypothesis of its Lemme 2 with Gel’fond’s bound for linear forms in two logarithms; the zero count of Theorem 6.12, which needs no separation between the frequencies, makes that hypothesis unnecessary.

The chapter has three parts: Gel’fond’s transcendence criterion in Waldschmidt’s form, the zero count, and the construction of 1973. The construction proves by contradiction that its hypotheses cannot hold together, so several statements of the last part have hypotheses that are never satisfied.

Notation for the construction. \(N\) is a large integer, and

\[ S=\Bigl\lfloor \frac{N^{2}}{\sqrt{\log N}}\Bigr\rfloor ,\qquad S'=\lfloor S/2\rfloor ,\qquad t_1=\Bigl\lfloor \frac{N}{\sqrt{\log N}}\Bigr\rfloor ,\qquad t_2=\bigl\lfloor N\sqrt{\log N}\bigr\rfloor , \]
\[ R_1=14t_1,\qquad R_2=14t_2. \]

For complex coefficients \(c(i,j,h)\), \(i\lt S\) and \(j,h\lt 2N\), the auxiliary function is

\[ F(z)=\sum _{i\lt S}\ \sum _{j,h\lt 2N}c(i,j,h)\, z^{i}e^{(jx_1+hx_2)z}. \]

Presentation data for \(x_1,x_2,y_1,y_2\in \mathbb {C}\) are: \(\omega ,\omega _1\in \mathbb {C}\) with \(\omega \) transcendental; \(Q\in \mathbb {Z}[X][Y]\), monic in \(Y\) of degree \(d\ge 1\), with \(Q(\omega ,\omega _1)=0\) and minimal there (no non-zero \(A\in \mathbb {Z}[X][Y]\) with \(\deg _YA\lt d\) vanishes at \((\omega ,\omega _1)\)); and \(D,E_i,G_j,H_{ij}\in \mathbb {Z}[X][Y]\) with \(D(\omega ,\omega _1)\neq 0\) and

\[ x_iD=E_i,\qquad y_jD=G_j,\qquad e^{x_iy_j}D=H_{ij}\qquad \text{at }(\omega ,\omega _1). \]

For \(M\in \mathbb {N}\) and integers \(q(i,j,h,\mu ,\nu )\), \(\mu \lt M\), \(\nu \lt d\), put \(c_q(i,j,h)=\sum _{\mu \lt M,\, \nu \lt d}q(i,j,h,\mu ,\nu )\, \omega ^{\mu }\omega _1^{\nu }\), and let \(F_q\) be the auxiliary function with these coefficients (with \(T\) in place of \(2N\) where a statement says so). The growth functions with constant \(k\gt 0\) are

\[ \sigma _1(x)=k\cdot \begin{cases} x-3+9\sqrt{\log 3},& x\le 3,\\ x^{2}\sqrt{\log x},& x\gt 3,\end{cases}\qquad \sigma _2(x)=k\cdot \begin{cases} x-3+9/\sqrt{\log 3},& x\le 3,\\ x^{2}/\sqrt{\log x},& x\gt 3.\end{cases} \]

Given functions \(\sigma _1,\sigma _2\), a polynomial \(P\in \mathbb {Z}[X]\) is small at \(\omega \) for \((N,C)\) if \(P\neq 0\), all its coefficients have absolute value at most \(e^{\sigma _1(N)}\), \(\deg P\le \sigma _2(N)\), and \(|P(\omega )|\lt e^{-C\sigma _1(N)\sigma _2(N)}\).

6.1 Gel’fond’s transcendence criterion

Lemma 6.1 Height of a divisor
✓
#

Let \(P\in \mathbb {Z}[X]\) be non-zero and let \(Q\in \mathbb {Z}[X]\) divide \(P\). If every coefficient of \(P\) has absolute value at most \(H\), then every coefficient of \(Q\) has absolute value at most \(e^{\deg P}H\).

Source: [ Wal71 , § 3, Lemme 1 ] (Gel’fond; a generalisation of a lemma of Popken and Koksma).

Proof ▶

Gel’fond’s inequality \(H(P_1)H(P_2)\le e^{\deg (P_1P_2)}H(P_1P_2)\), where \(H\) is the largest absolute value of a coefficient, follows by comparing heights with Mahler measure. Apply it to \(P=QR\), where \(H(R)\ge 1\).

Lemma 6.2 Gel’fond’s lemma
✓
#

Let \(\alpha \in \mathbb {C}\) be transcendental and let \(P\in \mathbb {Z}[X]\) be primitive, with every coefficient of absolute value at most \(H\). Let \(n,\lambda \in \mathbb {R}\) with \(\deg P\le n\le \log H\) and \(\lambda \gt 6\). If \(|P(\alpha )|\lt H^{-\lambda n}\), then there are a primitive irreducible divisor \(Q\) of \(P\) and an integer \(s\ge 1\) with

\[ |Q(\alpha )|\lt H^{-(\lambda -6)n/s},\qquad \deg Q\le n/s, \]

and every coefficient of \(Q\) of absolute value at most \(H^{1/s}e^{2n/s}\).

Source: [ Wal71 , § 3, Lemme 2 ] , after Gel’fond [ Gel52 ] .

Proof ▶

Write \(P=Q^{s}R\) as in Lemma 2.15. The cofactor \(R\) is not too small at \(\alpha \), so \(|Q(\alpha )|^{s}\) is small; the height and degree of \(Q\) follow through Mahler measure.

Lemma 6.3 Small values force divisibility
✓
#

Let \(P,Q\in \mathbb {Z}[X]\) with \(Q\) irreducible, let \(\alpha \in \mathbb {C}\), and let \(H,h\ge 1\) bound the absolute values of the coefficients of \(P\) and of \(Q\) respectively. Put \(d=\deg P\) and \(\delta =\deg Q\). If

\[ \bigl((1+|\alpha |)(d+\delta )\bigr)^{d+\delta }H^{\delta }h^{d}\bigl(|P(\alpha )|+|Q(\alpha )|\bigr)\lt 1, \]

then \(Q\) divides \(P\).

Source: [ Wal71 , § 3, (3.13) ] , citing [ Lang66 , Ch. V, § 2 ] .

Proof ▶

Otherwise \(P\) and \(Q\) are coprime, and their resultant is a non-zero integer of the form \(AP+BQ\) with \(A,B\in \mathbb {Z}[X]\) bounded by Hadamard’s inequality; evaluating at \(\alpha \) contradicts the hypothesis.

Lemma 6.4 Divisibility at one scale
✓
#

Let \(\alpha \in \mathbb {C}\) and \(\varepsilon \gt 0\). There is \(U\) such that for all real \(u\ge U\) and \(1\le v\le u\) the following holds. Let \(P,Q\in \mathbb {Z}[X]\) with \(Q\) irreducible, the coefficients of \(P\) of absolute value at most \(e^{u}\) and \(\deg P\le v\), the coefficients of \(Q\) of absolute value at most \(e^{3u}\) and \(\deg Q\le (1+\varepsilon /2)v\), and \(|P(\alpha )|\lt e^{-(4+\varepsilon )uv}\), \(|Q(\alpha )|\lt e^{-(4+\varepsilon )uv}\). Then \(Q\) divides \(P\).

Source: A step of the proof of [ Wal71 , § 3, Lemme fondamental ] .

Proof ▶

Lemma 6.3, with every size written in terms of \((u,v)\).

Lemma 6.5 The criterion for continuous growth functions
✓
#

Let \(\alpha \in \mathbb {C}\) and \(\varepsilon \gt 0\). Let \(\sigma _1,\sigma _2\colon \mathbb {R}\to \mathbb {R}\) be continuous, strictly increasing and tending to \(+\infty \), and let \(a_1,a_2\ge 1\) be such that for every \(x\ge 1\)

\[ \sigma _2(x)\le \sigma _1(x),\qquad \sigma _i(x+1)\le a_i\, \sigma _i(x)\quad (i=1,2). \]

Put \(C=\max \{ 10+\varepsilon ,(4+\varepsilon )a_1a_2\} \). Suppose that for every integer \(N\gt N_0\) there is a non-zero \(P_N\in \mathbb {Z}[X]\) with every coefficient of absolute value at most \(e^{\sigma _1(N)}\), \(\deg P_N\le \sigma _2(N)\) and \(|P_N(\alpha )|\lt e^{-C\sigma _1(N)\sigma _2(N)}\). Then \(\alpha \) is algebraic.

Source: [ Wal71 , § 3, Lemme fondamental ] and its proof, where continuity of the \(\sigma _i\) is assumed without loss of generality.

Proof ▶

Waldschmidt’s proof. If \(\alpha \) is transcendental, Lemma 6.2 gives small irreducible factors of the \(P_N\). At a suitable scale Lemma 6.4 makes such a factor divide a later \(P_N\), and Lemma 6.1 then bounds its height, against the choice of the scale.

Theorem 6.6 Waldschmidt’s transcendence criterion
✓
#

Let \(\alpha \in \mathbb {C}\) and \(\varepsilon \gt 0\). Let \(\sigma _1,\sigma _2\colon \mathbb {R}\to \mathbb {R}\) be strictly increasing and tending to \(+\infty \), and let \(a_1,a_2\ge 1\) be such that for every \(x\gt 0\)

\[ \sigma _2(x)\le \sigma _1(x),\qquad \sigma _i(x+1)\le a_i\, \sigma _i(x)\quad (i=1,2). \]

Put \(C=\max \{ 10+\varepsilon ,(4+\varepsilon )a_1a_2\} \). Suppose that for every integer \(N\gt N_0\) there is a non-zero \(P_N\in \mathbb {Z}[X]\) with every coefficient of absolute value at most \(e^{\sigma _1(N)}\), \(\deg P_N\le \sigma _2(N)\) and \(|P_N(\alpha )|\lt e^{-C\sigma _1(N)\sigma _2(N)}\). Then \(\alpha \) is algebraic.

Source: [ Wal71 , § 3, Lemme fondamental ] , with constant \(a_i\) as in Remark 2 there; a refinement of Gel’fond’s criterion [ Gel52 , Ch. III, § 4, Lemma VII ] .

Proof ▶

Replace \(\sigma _1,\sigma _2\) by their piecewise linear interpolations between the integers and apply Lemma 6.5.

6.2 Zeros of exponential polynomials

Lemma 6.7 Values from low-order derivatives
✓
#

Let \(w_1,\dots ,w_l\in \mathbb {C}\) be pairwise distinct with \(|w_j|\le W\) for some \(W\ge 0\). For each \(j\) let \(P_j\in \mathbb {C}[z]\) be zero or of degree less than \(q_j\), and put \(g(z)=\sum _jP_j(z)e^{w_jz}\) and \(n=\sum _jq_j\). Let \(R\ge 0\). If \(|g^{(s)}(0)|\le D\) for every \(s\lt n\), then

\[ |g(u)|\le n\, (W+1)^{n+1}e^{R(W+1)}D\qquad \text{whenever }|u|\le R. \]

Source: [ Wal71 , § 4, (4.5)–(4.13) ] .

Proof ▶

Expand \(g\) in its Taylor series at \(0\) and bound every derivative there with Lemma 2.10.

Lemma 6.8 The rescaled inequality
✓
#

Let \(f(z)=\sum _{j=1}^{l}\sum _{i\lt q_j}b_{j,i}z^{i}e^{\omega _jz}\) with pairwise distinct \(\omega _j\) and \(b_{j,i}\) not all zero; put \(n=\sum _jq_j\) and \(\Omega =\max _j|\omega _j|\), and suppose \(\Omega \gt 0\). Let \(z_0\in \mathbb {C}\), \(\rho \ge 0\), let \(S\) be a finite set of points of the disc \(|z-z_0|\le \rho \), and put \(\sigma =\sum _{z\in S}\operatorname {ord}_zf\) and \(x=\rho \Omega \). Then for every \(R\gt x+1\)

\[ \sigma \log \frac{R-x}{x+1}\le \log \Bigl(n!\, 2^{n+1}\frac{R}{R-1}\Bigr)+2R. \]

Source: [ Wal71 , § 4, (4.14) ] and the rescaling that follows it.

Proof ▶

Lemmas 2.8 and 6.7 bound \(\max _{s\lt n}|g^{(s)}(0)|\) against the maximum of \(|g|\) on \(|u|=R\) in opposite directions; apply both to \(g(z)=f(z_0+z/\Omega )\), which is not identically zero by Lemma 2.6.

Lemma 6.9 From the rescaled inequality to the count
✓
#

Let \(n\ge 2\) be an integer and let \(x\ge 0\), \(\lambda \gt 0\) and \(\sigma \ge 0\) be real. If for every \(R\gt x+1\)

\[ \sigma \log \frac{R-x}{x+1}\le \log \Bigl(n!\, 2^{n+1}\frac{R}{R-1}\Bigr)+2R, \]

then

\[ \sigma \lt \frac{n}{\lambda }+2\, \frac{1+n^{\lambda }}{\lambda \log n}\, (1+x). \]

Source: After [ Wal71 , § 4, (4.14) ] .

Proof ▶

For most parameters take \(R=n^{\lambda }(1+x)+x\), so that \(\log \frac{R-x}{x+1}=\lambda \log n\). The remaining cases, among them \(2\le n\le 5\), where the bound \(n!\, 2^{n}\le n^{n}\) used in [ Wal71 ] fails, need other choices of \(R\).

Lemma 6.10 Few zeros beat the bound
✓
#

For every integer \(n\ge 1\), real \(x\ge 0\) and \(\lambda \gt 0\), and every integer \(0\le \sigma \le n-1\),

\[ \sigma \lt \frac{n}{\lambda }+2\, \frac{1+n^{\lambda }}{\lambda \log n}\, (1+x), \]

where for \(n=1\) the second term is read as \(0\).

Source: Elementary.

Proof ▶

For \(n=1\), \(\sigma =0\lt 1/\lambda \); for \(\lambda \le 1\), \(n/\lambda \ge n\gt \sigma \). For \(\lambda \gt 1\), study the right-hand side as a function of \(\lambda \log n\).

Lemma 6.11 The degenerate cases
✓
#

Let \(f\), \(n\) and \(\Omega \) be as in Lemma 6.8, without the hypothesis \(\Omega \gt 0\), and suppose that \(n\le 1\) or \(\Omega =0\). Then for every finite set \(S\subset \mathbb {C}\)

\[ \sum _{z\in S}\operatorname {ord}_zf\le n-1. \]

Source: Elementary; the degenerate cases of [ Wal71 , § 4, Lemme 3 ] .

Proof ▶

If \(n=1\), then \(f=be^{\omega z}\) with \(b\neq 0\), which has no zeros. If \(\Omega =0\), the only frequency is \(0\) and \(f\) is a non-zero polynomial of degree less than \(n\).

Theorem 6.12 Zeros of an exponential polynomial in a disc
✓
#

Let \(\omega _1,\dots ,\omega _l\in \mathbb {C}\) be pairwise distinct, let \(q_1,\dots ,q_l\in \mathbb {N}\), and let \(b_{j,i}\in \mathbb {C}\) (\(1\le j\le l\), \(0\le i\lt q_j\)) be not all zero. Put

\[ f(z)=\sum _{j=1}^{l}\sum _{i=0}^{q_j-1}b_{j,i}\, z^{i}e^{\omega _jz},\qquad n=\sum _jq_j,\qquad \Omega =\max _j|\omega _j|. \]

Let \(z_0\in \mathbb {C}\), \(\rho \ge 0\) and \(\lambda \gt 0\). Then for every finite set \(S\) of points of the disc \(|z-z_0|\le \rho \)

\[ \sum _{z\in S}\operatorname {ord}_zf\lt \frac{n}{\lambda }+2\, \frac{1+n^{\lambda }}{\lambda \log n}\, (1+\rho \Omega ), \]

where for \(n=1\) the second term is read as \(0\). No separation between the \(\omega _j\) is assumed.

Source: [ Wal71 , § 4, Lemme 3, (4.2) ] ; compare [ Tij71 ] .

Proof ▶

For \(n\ge 2\) and \(\Omega \gt 0\), Lemmas 6.8 and 6.9; otherwise Lemmas 6.11 and 6.10.

Lemma 6.13 A non-zero derivative on the grid
✓
#

Let \(x_1,x_2\) be linearly independent over \(\mathbb {Q}\), and likewise \(y_1,y_2\). Let \(S,T,R_1,R_2,S'\in \mathbb {N}\) and let \(c(i,j,h)\in \mathbb {C}\) (\(i\lt S\), \(j,h\lt T\)) be not all zero; put

\[ G(z)=\sum _{i\lt S}\sum _{j,h\lt T}c(i,j,h)\, z^{i}e^{(jx_1+hx_2)z},\qquad n=ST^{2}. \]

If for some \(\lambda \gt 0\)

\[ \frac{n}{\lambda }+2\, \frac{1+n^{\lambda }}{\lambda \log n}\Bigl(1+(R_1|y_1|+R_2|y_2|)\, T(|x_1|+|x_2|)\Bigr)\le R_1R_2S', \]

then \(G^{(s)}(ay_1+by_2)\neq 0\) for some \(a\lt R_1\), \(b\lt R_2\) and \(s\lt S'\).

Source: [ Wal73 , Lemme 6 ] , with the zero estimate [ Wal71 , § 4, Lemme 3 ] in place of Gel’fond’s zero lemma.

Proof ▶

Otherwise the \(R_1R_2\) points \(ay_1+by_2\), distinct because \(y_1,y_2\) are independent, would be zeros of order at least \(S'\) of \(G\). The frequencies \(jx_1+hx_2\) are distinct, so \(G\) is not identically zero (Lemma 2.6), and Theorem 6.12 counts fewer zeros.

6.3 The construction of 1973

Lemma 6.14 A rank-one matrix of logarithms
✓
#

Let \(l_{11},l_{12},l_{21},l_{22}\) be non-zero complex numbers with every \(e^{l_{ij}}\) algebraic, \(l_{11}l_{22}=l_{12}l_{21}\) and \(\operatorname {trdeg}_{\mathbb {Q}}\mathbb {Q}[l_{11},l_{12},l_{21},l_{22}]\le 1\), and suppose that neither the rows nor the columns of \((l_{ij})\) are linearly dependent over \(\mathbb {Q}\). Then there are \(x_1,x_2,y_1,y_2\in \mathbb {C}\), with \(x_1,x_2\) linearly independent over \(\mathbb {Q}\) and likewise \(y_1,y_2\), such that every \(e^{x_iy_j}\) is algebraic and \(\operatorname {trdeg}_{\mathbb {Q}}\mathbb {Q}[x_1,x_2,y_1,y_2]\le 1\).

Source: Elementary; as in [ Wal73 , § I ] .

Proof ▶

Take \(x=(l_{11},l_{21})\) and \(y=(1,l_{12}/l_{11})\), so that \(x_iy_j=l_{ij}\).

Lemma 6.15 Presenting the field
✓
#

Let \(x_1,x_2\) be linearly independent over \(\mathbb {Q}\), and likewise \(y_1,y_2\). If every \(e^{x_iy_j}\) is algebraic and \(\operatorname {trdeg}_{\mathbb {Q}}\mathbb {Q}[x_1,x_2,y_1,y_2]\le 1\), then \(x_1,x_2,y_1,y_2\) have presentation data.

Source: The opening of [ Wal73 , § III ] .

Proof ▶

The number \(\omega =x_1y_1\) is non-zero with \(e^{\omega }\) algebraic, so it is transcendental by Theorem 3.2. Every \(x_i\), \(y_j\) and \(e^{x_iy_j}\) is then algebraic over \(\mathbb {Q}(\omega )\), and Lemma 2.17 presents them.

Lemma 6.16 The size estimate
✓

For all \(c,r\in \mathbb {N}\) there is \(\kappa \gt 0\) such that for all integers \(N\ge 3\) and \(S,T,a,b,m\in \mathbb {N}\) with

\[ S\le \frac{N^{2}}{\sqrt{\log N}},\quad m\le S,\quad T\le rN,\quad a\le r\frac{N}{\sqrt{\log N}},\quad b\le rN\sqrt{\log N}, \]

one has

\[ c^{\, c(1+m+S+T(a+b))}(1+m+S+T+a+b)^{c(1+m+S)}\le e^{\kappa N^{2}\sqrt{\log N}}. \]

Source: The parameter estimates in the proofs of [ Wal73 , Lemmes 4 and 7 ] .

Proof ▶
Lemma 6.17 Derivatives at a grid point, presented
✓
#

Let \(x_1,x_2,y_1,y_2\in \mathbb {C}\) with every \(e^{x_iy_j}\) algebraic. Let \(\omega ,\omega _1\in \mathbb {C}\), let \(\varphi \) be evaluation at \((\omega ,\omega _1)\), and let \(D,E_i,G_j,H_{ij}\in \mathbb {Z}[X][Y]\) with \(\varphi (D)\neq 0\), \(x_i\varphi (D)=\varphi (E_i)\), \(y_j\varphi (D)=\varphi (G_j)\) and \(e^{x_iy_j}\varphi (D)=\varphi (H_{ij})\). There is \(c\in \mathbb {N}\) such that for all \(S,T,a,b,m\in \mathbb {N}\) there are \(\Lambda \in \mathbb {C}\), \(\Lambda \neq 0\), with \(|\Lambda |\le c^{\, c(1+m+S+T(a+b))}\), and \(P_{ijh}\in \mathbb {Z}[X][Y]\) (\(i\lt S\), \(j,h\lt T\)) of degree at most \(c(1+m+S)\) in \(X\) and in \(Y\) and of length at most

\[ c^{\, c(1+m+S+T(a+b))}\, (1+m+S+T+a+b)^{c(1+m+S)}, \]

such that for all \(f_{ijh}\in \mathbb {C}\)

\[ \Lambda \cdot \frac{d^{m}}{dz^{m}}\Bigl(\sum _{i,j,h}f_{ijh}\, z^{i}e^{(jx_1+hx_2)z}\Bigr)\Big|_{z=ay_1+by_2}=\sum _{i,j,h}f_{ijh}\, \varphi (P_{ijh}). \]

Source: A step of the proofs of [ Wal73 , Lemmes 4 and 7 ] .

Proof ▶

Expand the derivative by Leibniz’s rule. The powers of the algebraic numbers \(e^{x_iy_j}\) reduce through integer relations (Lemma 2.3), and lengths are tracked with Lemmas 2.11 and 2.12.

Lemma 6.18 Reduced presentations
✓

In the setting of Lemma 6.17, let moreover \(Q\in \mathbb {Z}[X][Y]\) be monic in \(Y\) with \(\varphi (Q)=0\). There is \(c\in \mathbb {N}\) such that for all \(S,T,M,a,b,m\in \mathbb {N}\) there are \(\Lambda \neq 0\) with \(|\Lambda |\le c^{\, c(1+m+S+T(a+b))}\) and \(R_{ijh\mu \nu }\in \mathbb {Z}[X][Y]\) (\(i\lt S\), \(j,h\lt T\), \(\mu \lt M\), \(\nu \lt \deg _YQ\)) of \(Y\)-degree less than \(\deg _YQ\) and \(X\)-degree at most \(M+c(1+m+S)\), and of length at most

\[ c^{\, c(1+m+S+T(a+b))}\, (1+m+S+T+a+b)^{c(1+m+S)}, \]

such that for every integer array \(q\)

\[ \Lambda \, F_q^{(m)}(ay_1+by_2)=\varphi \Bigl(\sum _{i,j,h,\mu ,\nu }q(i,j,h,\mu ,\nu )\, R_{ijh\mu \nu }\Bigr), \]

where \(F_q\) is formed with \(T\) in place of \(2N\).

Source: A step of the proofs of [ Wal73 , Lemmes 4 and 7 ] .

Proof ▶

Put the coefficients \(c_q\) into Lemma 6.17 and reduce modulo \(Q\) with Lemma 2.13.

Lemma 6.19 The linear system
✓
#

Let \(x_1,x_2,y_1,y_2\in \mathbb {C}\) have presentation data, and suppose every \(e^{x_iy_j}\) is algebraic. There is \(\kappa _1\gt 0\) such that for every large \(N\) there are an integer \(M\) with \(0\lt M\le \kappa _1S\) and an \(R\times U\) integer matrix \(B\), where \(U=S(2N)^{2}Md\) and \(2R\le U\), with entries of absolute value at most \(e^{\kappa _1N^{2}\sqrt{\log N}}\), such that for every integer vector \(q\) with \(Bq=0\)

\[ F_q^{(m)}(ay_1+by_2)=0\qquad \text{for all }a\lt t_1,\ b\lt t_2,\ m\lt S. \]

Source: The equation count of [ Wal73 , Lemme 4 ] .

Proof ▶

The rows of \(B\) are the coefficients of the reduced polynomials of Lemma 6.18, bounded by Lemma 6.16; \(M\) is taken large enough for twice as many unknowns as equations.

Lemma 6.20 Solving the system
✓
#

Let \(\omega ,\omega _1\in \mathbb {C}\) and \(Q\in \mathbb {Z}[X][Y]\) with \(d=\deg _YQ\ge 1\), such that no non-zero \(A\in \mathbb {Z}[X][Y]\) with \(\deg _YA\lt d\) vanishes at \((\omega ,\omega _1)\), and let \(\kappa _1\gt 0\). There is \(\kappa \ge \kappa _1\) such that for every large \(N\), every \(M\) with \(0\lt M\le \kappa _1S\), and every \(R\times U\) integer matrix \(B\) with \(U=S(2N)^{2}Md\), \(2R\le U\) and entries of absolute value at most \(e^{\kappa _1N^{2}\sqrt{\log N}}\), there is an integer vector \(q\) with \(Bq=0\) such that every \(|q(i,j,h,\mu ,\nu )|\le e^{\kappa N^{2}\sqrt{\log N}}\), some \(c_q(i,j,h)\neq 0\), and every \(|c_q(i,j,h)|\le e^{\kappa N^{2}\sqrt{\log N}}\).

Source: Siegel’s lemma [ Lang66 , Ch. I, § 2 ] , as used in [ Wal73 , Lemme 4 ] .

Proof ▶

Lemma 2.5 gives \(q\neq 0\). A non-zero block \(q(i,j,h,\cdot ,\cdot )\) is a polynomial of \(Y\)-degree less than \(d\), so by minimality \(c_q(i,j,h)\neq 0\).

Lemma 6.21 The auxiliary function
✓
#

Let \(x_1,x_2,y_1,y_2\in \mathbb {C}\) have presentation data, and suppose every \(e^{x_iy_j}\) is algebraic. There is \(\kappa \gt 0\) such that for every large \(N\) there are \(M\le \kappa S\) and integers \(q(i,j,h,\mu ,\nu )\) (\(i\lt S\), \(j,h\lt 2N\), \(\mu \lt M\), \(\nu \lt d\)) of absolute value at most \(e^{\kappa N^{2}\sqrt{\log N}}\), such that the coefficients \(c_q(i,j,h)\) are not all zero and of absolute value at most \(e^{\kappa N^{2}\sqrt{\log N}}\), and

\[ F_q^{(m)}(ay_1+by_2)=0\qquad \text{for all }a\lt t_1,\ b\lt t_2,\ m\lt S. \]

Source: [ Wal73 , Lemme 4 ] , with Siegel’s lemma.

Proof ▶

Lemma 6.19, then Lemma 6.20.

Lemma 6.22 The numbers of the extrapolation
✓
#

Let \(X\ge 0\), \(Y\ge 1\) and \(\kappa \ge 0\) be real, and let \(N,S,t_1,t_2,s\) be natural numbers with \(s\le S\) and \(N\ge 32(13+\kappa +\log Y+2XY)\). Write \(w=\sqrt{\log N}\) and suppose \(Sw\le N^{2}\le 2Sw\), \(N\le 2t_1w\) and \(Nw\le 2t_2\). Then

\[ 2\, s!\, (N-1)^{-t_1t_2S}\cdot S(2N)^{2}e^{\kappa N^{2}w}\, (YN^{2}w)^{S}\, e^{2NX\cdot YN^{2}w}\le e^{-N^{4}w/32}. \]

Source: The parameter estimate in the proof of [ Wal73 , Lemme 5 ] .

Proof ▶
Lemma 6.23 Extrapolation
✓
#

Let \(x_1,x_2\in \mathbb {C}\), let \(y_1,y_2\) be linearly independent over \(\mathbb {Q}\), and let \(\kappa \gt 0\). There are \(\kappa '\gt 0\) and \(N_0\) such that for every \(N\gt N_0\) and all coefficients with \(|c(i,j,h)|\le e^{\kappa N^{2}\sqrt{\log N}}\): if \(F^{(m)}(ay_1+by_2)=0\) for all \(a\lt t_1\), \(b\lt t_2\) and \(m\lt S\), then

\[ |F^{(s)}(ay_1+by_2)|\le \exp \bigl(-N^{4}\sqrt{\log N}/\kappa '\bigr)\qquad \text{for all }s\lt S',\ a\lt R_1,\ b\lt R_2. \]

Source: [ Wal73 , Lemme 5 ] .

Proof ▶

Lemma 2.9 for the grid \(\{ ay_1+by_2:a\lt t_1,\ b\lt t_2\} \), with the numbers of Lemma 6.22.

Lemma 6.24 Norms down to \(\mathbb {Q}(\omega )\)
✓
#

Let \(\omega ,\omega _1\in \mathbb {C}\) and let \(Q\in \mathbb {Z}[X][Y]\) be monic in \(Y\) of degree \(d\), with \(Q(\omega ,\omega _1)=0\) and minimal there. There is \(c\in \mathbb {N}\) such that for every \(A\in \mathbb {Z}[X][Y]\) of \(Y\)-degree less than \(d\), length at most \(b\) and \(X\)-degree at most \(e\), with \(A(\omega ,\omega _1)\neq 0\), there is a non-zero \(P\in \mathbb {Z}[X]\) of length at most \((cb)^{d}\) and degree at most \(d(e+c)\) with

\[ |P(\omega )|\le |A(\omega ,\omega _1)|\, \bigl(cb\max (1,|\omega |)^{e}\bigr)^{d}. \]

Source: The core of the proof of [ Wal73 , Lemme 7 ] .

Proof ▶

\(P\) is the determinant of multiplication by \(A\) on \(\mathbb {Z}[X][Y]/(Q)\) in the basis \(1,Y,\dots ,Y^{d-1}\): the norm from \(\mathbb {Q}(\omega ,\omega _1)\) to \(\mathbb {Q}(\omega )\), taken without building the field extension.

Lemma 6.25 From a small value to a small polynomial
✓
#

Let \(x_1,x_2,y_1,y_2\in \mathbb {C}\) have presentation data, suppose every \(e^{x_iy_j}\) is algebraic, and let \(\kappa ,\kappa '\gt 0\). There is \(k\gt 0\) such that for every \(C\) there is \(N_0\) with the following property. Let \(N\gt N_0\), \(M\le \kappa S\), and let \(q(i,j,h,\mu ,\nu )\) be integers of absolute value at most \(e^{\kappa N^{2}\sqrt{\log N}}\). If \(a\lt R_1\), \(b\lt R_2\), \(s\lt S'\) and

\[ 0\lt |F_q^{(s)}(ay_1+by_2)|\le \exp \bigl(-N^{4}\sqrt{\log N}/\kappa '\bigr), \]

then some \(P\in \mathbb {Z}[X]\) is small at \(\omega \) for \((N,C)\) with respect to the growth functions of constant \(k\).

Source: [ Wal73 , Lemme 7 ] .

Proof ▶

Lemma 6.18 writes the value, times a power of \(D(\omega ,\omega _1)\), as \(A(\omega ,\omega _1)\); take for \(P\) the norm of Lemma 6.24. The sizes come from Lemma 6.16.

Lemma 6.26 The core of the construction
✓
#

Let \(x_1,x_2\) be linearly independent over \(\mathbb {Q}\), and likewise \(y_1,y_2\). Suppose every \(e^{x_iy_j}\) is algebraic and \(\operatorname {trdeg}_{\mathbb {Q}}\mathbb {Q}[x_1,x_2,y_1,y_2]\le 1\). Then there are a transcendental \(\omega \in \mathbb {C}\) and \(k\gt 0\) such that for every \(C\) and every large \(N\) there are coefficients \(c(i,j,h)\), not all zero, with the following property: whenever \(a\lt R_1\), \(b\lt R_2\), \(s\lt S'\) and \(F^{(s)}(ay_1+by_2)\neq 0\), some \(P\in \mathbb {Z}[X]\) is small at \(\omega \) for \((N,C)\) with respect to the growth functions of constant \(k\).

Source: [ Wal73 , § III, (4) and Lemmes 4, 5, 7 ] .

Proof ▶

Lemma 6.15 gives presentation data and Lemma 6.21 the auxiliary function. Lemma 6.23 makes its derivatives small on the larger grid, and Lemma 6.25 turns a non-zero value into \(P\).

Lemma 6.27 The growth functions
✓
#

For every \(k\gt 0\) the growth functions \(\sigma _1,\sigma _2\) are strictly increasing and tend to \(+\infty \), and for every \(x\gt 0\)

\[ \sigma _2(x)\le \sigma _1(x),\qquad \sigma _i(x+1)\le 3\, \sigma _i(x)\quad (i=1,2). \]

Source: Elementary; the functions of the end of [ Wal73 , § III ] .

Proof ▶
Lemma 6.28 The parameters beat the zero count
✓
#

Let \(X,Y_1,Y_2\ge 0\). There is \(N_0\) such that for every integer \(N\gt N_0\), with \(T=2N\), \(\lambda =1/20\) and \(n=ST^{2}\),

\[ \frac{n}{\lambda }+2\, \frac{1+n^{\lambda }}{\lambda \log n}\bigl(1+(R_1Y_1+R_2Y_2)\, TX\bigr)\le R_1R_2S'. \]

Source: The parameters of [ Wal73 , (4) and Lemme 6 ] against the zero estimate [ Wal71 , § 4, Lemme 3 ] .

Proof ▶

\(n/\lambda \) is about \(80N^{4}/\sqrt{\log N}\) and \(R_1R_2S'\) about \(98N^{4}/\sqrt{\log N}\), while the second term is \(O(N^{2.2+\varepsilon })\).

Lemma 6.29 The construction
✓
#

Let \(l_{11},l_{12},l_{21},l_{22}\) satisfy the hypotheses of Lemma 6.14. Then there are a transcendental \(\omega \), functions \(\sigma _1,\sigma _2\) and constants \(a_1,a_2\ge 1\) satisfying the hypotheses of Theorem 6.6, and pairs \(x_1,x_2\) and \(y_1,y_2\), each linearly independent over \(\mathbb {Q}\), with the following property. For every \(C\) and every large \(N\) there are natural numbers \(S,T,R_1,R_2,S'\), coefficients \(c(i,j,h)\in \mathbb {C}\) (\(i\lt S\), \(j,h\lt T\)), not all zero, and \(\lambda \gt 0\) such that

  • the inequality of Lemma 6.13 holds, and

  • whenever \(a\lt R_1\), \(b\lt R_2\), \(s\lt S'\) and \(G^{(s)}(ay_1+by_2)\neq 0\), where

    \[ G(z)=\sum _{i\lt S}\sum _{j,h\lt T}c(i,j,h)\, z^{i}e^{(jx_1+hx_2)z}, \]

    some \(P\in \mathbb {Z}[X]\) is small at \(\omega \) for \((N,C)\) with respect to \(\sigma _1,\sigma _2\).

Source: [ Wal73 , § III, (4) and Lemmes 4, 5, 7 ] .

Proof ▶

Lemma 6.14 gives \(x\) and \(y\), and Lemma 6.26 gives \(\omega \) and the coefficients, with the parameters of the notation above, the growth functions of Lemma 6.27 (\(a_1=a_2=3\)) and the count of Lemma 6.28.

Lemma 6.30 A counterexample gives small polynomials
✓
#

Let \(l_{11},l_{12},l_{21},l_{22}\) be non-zero complex numbers with every \(e^{l_{ij}}\) algebraic, \(l_{11}l_{22}=l_{12}l_{21}\) and \(\operatorname {trdeg}_{\mathbb {Q}}\mathbb {Q}[l_{11},l_{12},l_{21},l_{22}]\le 1\), and suppose that neither the rows nor the columns of \((l_{ij})\) are linearly dependent over \(\mathbb {Q}\). Then there are a transcendental \(\omega \in \mathbb {C}\), functions \(\sigma _1,\sigma _2\) and constants \(a_1,a_2\ge 1\) satisfying the hypotheses of Theorem 6.6, such that for every \(C\in \mathbb {R}\) there is \(N_0\) and for every \(N\gt N_0\) a polynomial \(P_N\in \mathbb {Z}[X]\) that is small at \(\omega \) for \((N,C)\) with respect to \(\sigma _1,\sigma _2\).

Source: [ Wal73 , § III, Lemmes 4–7 ] , with the zero estimate of [ Wal71 , § 4, Lemme 3 ] ; see also [ Bro74 ] .

Proof ▶

Lemma 6.29, with a non-zero derivative supplied by Lemma 6.13.

6.4 The theorem

Theorem 6.31 Four exponentials in transcendence degree one
✓
#

Let \(l_{11},l_{12},l_{21},l_{22}\) be non-zero logarithms of algebraic numbers with

\[ l_{11}l_{22}=l_{12}l_{21},\qquad \operatorname {trdeg}_{\mathbb {Q}}\mathbb {Q}(l_{11},l_{12},l_{21},l_{22})\le 1. \]

Then the two rows, or the two columns, of \(\begin{pmatrix} l_{11} & l_{12} \\ l_{21} & l_{22} \end{pmatrix}\) are linearly dependent over \(\mathbb {Q}\). Equivalently, if \(x_1,x_2\) and \(y_1,y_2\) are pairs of complex numbers linearly independent over \(\mathbb {Q}\) and the four products \(x_iy_j\) are logarithms of algebraic numbers, then \(\operatorname {trdeg}_{\mathbb {Q}}\mathbb {Q}(x_1,x_2,y_1,y_2)\ge 2\).

Source: Waldschmidt [ Wal73 , Cor. 4 ] and Brownawell [ Bro74 , Cor. 7 ] ; stated in this form as [ RW95 , Theorem 1 ] .

Proof ▶

Otherwise Lemma 6.30 gives polynomials too small at a transcendental \(\omega \), and Theorem 6.6 makes \(\omega \) algebraic.