9 Baker’s theorem, by Schneider–Lang in several variables
Baker’s theorem in its qualitative, inhomogeneous form (Theorem 9.24): if \(\ell _1,\dots ,\ell _n\) are logarithms of algebraic numbers, linearly independent over \(\mathbb {Q}\), then \(1,\ell _1,\dots ,\ell _n\) are linearly independent over \(\overline{\mathbb {Q}}\). The route is the one Bertrand and Masser found [ BM80 , Mas81 ] , as Waldschmidt writes it in Chapter 4 of [ Wal00 ] : Baker’s theorem for a basis of a number field (the book’s Theorem 4.5), from the criterion of Schneider–Lang for \(\mathbb {C}^{d_0}\times (\mathbb {C}^{\times })^{d_1}\) with \(d_0\le 1\) (its Corollary 4.2). The criterion is proved directly, as in the book’s §4.6: an auxiliary function from Siegel’s lemma, Liouville’s inequality at the points of a grid, a Schwarz lemma for Cartesian products (the book’s Proposition 4.7) at the first derivative that does not vanish, and a contradiction. No zero estimate is needed. The Schwarz lemma comes from one-variable Hermite division, one coordinate at a time.
Step 5 of the book’s §4.6 needed a repair: vanishing to a given order in each coordinate does not survive the change of variables that the proof makes, while vanishing to a given total order does; only the constants change. No new mathematics is claimed.
Notation. \(\iota \) is a finite set, \(n=|\iota |\) unless said otherwise, and \(\mathbb {C}^{\iota }\) carries the sup norm, so that its closed balls are polydiscs. For \(w,z\in \mathbb {C}^{\iota }\), \(\langle w,z\rangle =\sum _\nu w_\nu z_\nu \). For a function \(F\) on \(\mathbb {C}^{\iota }\) and a list \(L=(L(0),\dots ,L(k-1))\) of \(k\) coordinate directions,
where \(D^{k}F\) is the \(k\)-th Fréchet derivative and \(e_\nu \) are the coordinate vectors; every derivative of \(F\) of total order \(k\) vanishes at \(z\) when \(D^{k}F(z)=0\). Given \(x_1,\dots ,x_{d_1}\in \mathbb {C}^{\iota }\), \(k_0\in \iota \) and integers \(T_0,T_1\ge 0\), the exponential monomials are \(z_{k_0}^{\tau }e^{\langle t_1x_1+\dots +t_{d_1}x_{d_1},\, z\rangle }\) for \(0\le \tau \le T_0\) and \(t\in \{ 0,\dots ,T_1\} ^{d_1}\), and for coefficients \(p=(p_{\tau ,t})\)
The grid points of a family \((y_j)_{j\in \iota }\) in \(\mathbb {C}^{\iota }\) are the points \(\sum _js_jy_j\) with integers \(0\le s_j\lt S_1\).
9.1 The Cartesian Schwarz lemma
Let \(\mathbb {K}\) be a nontrivially normed field, \(F\) a normed space over \(\mathbb {K}\), and \(g\colon \mathbb {K}^{\iota }\to F\) a function of class \(C^{k}\). For \(i\in \iota \) and a function \(h\) on \(\mathbb {K}^{\iota }\), let \(\partial _ih(z)\) be the derivative at \(w=z_i\) of \(w\mapsto h(z[i:=w])\), where \(z[i:=w]\) is \(z\) with its coordinate \(z_i\) replaced by \(w\). Then for every list \(L\) of \(k\) coordinate directions and every \(z\in \mathbb {K}^{\iota }\)
the partial derivative in the direction \(L(k-1)\) being taken first.
Source: Standard.
Let \(F\) be an entire function on \(\mathbb {C}^{\iota }\) with \(|F|\le M\) on the closed polydisc of radius \(\rho \gt 0\) about \(p\). For a list \(L\) of \(k\) coordinate directions, let \(\sigma _\nu \) be the number of times \(\nu \) occurs in \(L\). Then
Source: Classical; in the form of [ Wal00 , p. XVII ] .
Through Lemma 9.1, by Cauchy’s inequality in one variable at a time.
Let \(K\) be a field of characteristic \(0\), \(E\subset K\) a finite set and \(S_0\in \mathbb {N}\), and write \(q^{(k)}\) for the \(k\)-th formal derivative of \(q\in K[X]\). There are polynomials \(b_{\zeta ,k}\in K[X]\) (\(\zeta \in E\), \(0\le k\lt S_0\)) of degree less than \(|E|S_0\) such that, for \(\zeta '\in E\) and \(0\le k'\lt S_0\), \(b_{\zeta ,k}^{(k')}(\zeta ')\) is \(1\) if \((\zeta ',k')=(\zeta ,k)\) and \(0\) otherwise, and every \(q\in K[X]\) of degree less than \(|E|S_0\) satisfies
Source: Standard (Hermite interpolation).
Let \(0\lt r\) and \(5r\le R\), let \(Z\) be a finite multiset of complex numbers of modulus at most \(r\), put \(p=|Z|\) and \(P(x)=\prod _{\zeta \in Z}(x-\zeta )\), and let \(u\) be an entire function with \(|u|\le M\) on the closed disc of radius \(R\) about \(0\). Then \(u=\rho +Pq\) for a polynomial \(\rho \) of degree less than \(p\) and an entire function \(q\), such that \(\rho ^{(k)}(\zeta )=u^{(k)}(\zeta )\) whenever \(k\) is less than the multiplicity of \(\zeta \) in \(Z\), and for \(|x|\le R\)
Source: [ Wal00 , Lemma 4.8 c)–d), p. 123 ] in one variable, with the bounds of Step 2.4 of its proof (pp. 125–126).
Induction on the nodes, dividing by one factor \(x-\zeta \) at a time, as in Step 2.4 of the book; \(\rho \) and \(q\) are the book’s \(f_0\) and \(f_n\).
Let \(g\) be an entire function on \(\mathbb {C}^{\iota }\) with \(|g|\le K\) on the closed polydisc of radius \(R\) about \(0\), let \(i\in \iota \) and \(S_0\in \mathbb {N}\), and let \(E\) be a finite set of complex numbers of modulus at most \(r\), where \(0\lt r\) and \(5r\le R\); put \(p=|E|S_0\). There is an entire function \(h\) on \(\mathbb {C}^{\iota }\) such that:
\(|h|\le 3^{p}K\) on the closed polydisc of radius \(R\) about \(0\);
\(|g-h|\le (2r)^{p}(3/R)^{p}K\) on the closed polydisc of radius \(r\) about \(0\);
for every point \(\xi \) and every list \(v\) of \(m\) coordinate directions different from \(i\): if \(D_{(v,i^{k})}\, g(\xi [i:=\zeta ])=0\) for every \(\zeta \in E\) and every \(k\lt S_0\), where \((v,i^{k})\) is the list \(v\) followed by \(k\) times \(i\), then \(D_vh(\xi )=0\).
Source: A step of the proof of [ Wal00 , Proposition 4.7, pp. 122–130 ] : the case \(m=n\) of its Lemma 4.8 (pp. 123–126) in one coordinate, with the bounds of Step 2.4.
\(h\) is the Hermite interpolant of \(g\) in the variable \(z_i\) at the nodes of \(E\), each with multiplicity \(S_0\): Lemma 9.4 on every slice, written in the fixed basis of Lemma 9.3, so that the remainder depends on the other variables only through the jets of \(g\) at the nodes. Only one-variable complex analysis is used.
Let \(n\ge 1\) and let \(f\) be an entire function on \(\mathbb {C}^{n}\). Let \(E_1,\dots ,E_n\) be sets of \(S_1\) complex numbers each, all of modulus at most \(r\gt 0\), and let \(R\ge 5r\). Suppose that \(|f|\le M\) on the closed polydisc of radius \(R\) about \(0\), and that every derivative of \(f\) of total order less than \(nS_0\) vanishes at every point of \(E_1\times \dots \times E_n\). Then on the closed polydisc of radius \(r\) about \(0\)
Source: [ Wal00 , Proposition 4.7, pp. 122–130 ] , in the form used in its §4.6, with \(2\cdot 3^{n}\) in place of \(18^{n}\).
Lemma 9.5 in each coordinate in turn, telescoping the differences. The book assumes vanishing with every exponent below \(S_0\); vanishing by total order, assumed here, is what the change of variables of §4.6 preserves.
9.2 Siegel’s lemma and the Taylor bounds
Let \(J\) and \(\Lambda \) be finite sets, \(\mu =|J|\) and \(\nu =|\Lambda |\), and let \(v_{j\lambda }\in \mathbb {R}\) with \(\sum _\lambda |v_{j\lambda }|\le C\) for every \(j\), where \(C\gt 0\). Let \(X\ge 0\) and \(\ell \ge 1\) be integers with \(\ell ^{\mu }\lt (X+1)^{\nu }\). Then there is \(\xi \in \mathbb {Z}^{\Lambda }\), \(\xi \neq 0\), with \(|\xi _\lambda |\le X\) for every \(\lambda \) and
Source: [ Wal00 , Lemma 4.11, pp. 132–133 ] , with any real \(C\gt 0\) in place of an integer.
Dirichlet’s box principle.
Let \(E\) be a complex normed space and \(G\colon E\to \mathbb {C}\) analytic at every point, with \(|G|\le M\) on the closed ball of radius \(R\) about \(0\), and let \(0\lt r\lt R\). For every \(z\in E\) with \(\| z\| \le r\) and every \(T\in \mathbb {N}\)
Source: The tail bound in the proof of [ Wal00 , Lemma 4.13, pp. 134–135 ] , with \(1/(1-r/R)\) in place of \(1+\sqrt{T}\).
Cauchy’s inequality on the line through \(z\) bounds each term, and the terms are summed as a geometric series.
Let \(\Lambda \) be a finite set, let \(\varphi _\lambda \) (\(\lambda \in \Lambda \)) be entire functions on \(\mathbb {C}^{\iota }\) with \(|\varphi _\lambda |\le B_\lambda \) on the closed polydisc of radius \(r\gt 0\) about \(0\), and let \(T\in \mathbb {N}\). There are \(u_{\tau \lambda }\in \mathbb {C}\), for \(\lambda \in \Lambda \) and \(\tau \in \{ 0,\dots ,T-1\} ^{\iota }\), with \(|u_{\tau \lambda }|\le B_\lambda \), such that for all \(c\in \mathbb {C}^{\Lambda }\) and \(z\in \mathbb {C}^{\iota }\) the function \(F=\sum _\lambda c_\lambda \varphi _\lambda \) satisfies
Source: A step of the proof of [ Wal00 , Proposition 4.10, p. 135 ] , with Cauchy’s inequalities (p. XVII).
Expand the Taylor polynomial in monomials; Lemma 9.2 bounds the coefficients.
Let \(\Lambda \) be a finite set, \(L=|\Lambda |\), and let \(\varphi _\lambda \) (\(\lambda \in \Lambda \)) be entire functions on \(\mathbb {C}^{\iota }\) with \(|\varphi _\lambda |\le B_\lambda \) on the closed polydisc of radius \(R\) about \(0\) and \(\sum _\lambda B_\lambda \le C\), where \(C\gt 0\). Let \(0\lt r\lt R\), and let \(T\), \(X\) and \(\ell \ge 1\) be integers with \(\ell ^{2T^{n}}\lt (X+1)^{L}\). Then there are integers \(p_\lambda \), not all zero, with \(|p_\lambda |\le X\), such that on the closed polydisc of radius \(r\) about \(0\)
Source: [ Wal00 , § 4.5: Lemmas 4.11–4.13 and the proof of Proposition 4.10, pp. 132–136 ] , with \(T\), \(X\) and \(\ell \) left free.
Let \(n\ge 1\) and \(L\) be integers and \(N,U,V,r,R\) real numbers with \(r\gt 0\), \(W=N+U+V\ge 12n^{2}\), \(er\le R\le re^{W/6}\) and
Then there are integers \(T\), \(X\) and \(\ell \ge 1\) with \(X\le e^{N}\) and \(\ell ^{2T^{n}}\lt (X+1)^{L}\) such that
Source: The numerical part of the proof of [ Wal00 , Proposition 4.10, pp. 135–136 ] .
The book’s choices: \(\frac43W\le T\log (R/r)\lt \frac43W+\log (R/r)\), \(X=\lfloor e^{N}\rfloor \), and \(\ell =\lceil \frac83T^{n}e^{W}\rceil \) cells for the box principle.
Let \(n=|\iota |\ge 1\), let \(\Lambda \) be a finite set, and let \(\varphi _\lambda \) (\(\lambda \in \Lambda \)) be entire functions on \(\mathbb {C}^{\iota }\). Let \(N,U,V,r\gt 0\) and \(R\) be real numbers with \(W=N+U+V\ge 12n^{2}\) and \(er\le R\le re^{W/6}\). Suppose that \(|\varphi _\lambda |\le B_\lambda \) on the closed polydisc of radius \(R\) about \(0\), with \(\sum _\lambda B_\lambda \le e^{U}\), and that
Then there are integers \(p_\lambda \), not all zero, with \(|p_\lambda |\le e^{N}\), such that \(\bigl|\sum _\lambda p_\lambda \varphi _\lambda \bigr|\le e^{-V}\) on the closed polydisc of radius \(r\) about \(0\).
Source: [ Wal00 , Proposition 4.10 with Lemmas 4.11–4.13, pp. 131–136 ] .
9.3 Exponential monomials
Let \(K\) be a number field with an embedding \(\varphi \colon K\to \mathbb {C}\), let \(w\in K^{\iota }\), \(\tau \in \mathbb {N}\), \(k_0\in \iota \), and \(g(z)=z_{k_0}^{\tau }\exp \bigl(\sum _\nu \varphi (w_\nu )z_\nu \bigr)\). Let \(q\in \mathbb {C}^{\iota }\) and \(v\in K\), with \(\varphi (v)=q_{k_0}\) if \(\tau \gt 0\). Suppose that \(A\ge 1\), \(\overline{\left|w_\nu \right|}\le A\) for every \(\nu \), \(\overline{\left|v\right|}\le B\), and that \(\delta w_\nu \) and \(\delta v\) are algebraic integers for an integer \(\delta \). Then for every list \(L\) of \(k\) coordinate directions
for some \(\gamma \in K\) with \(\overline{\left|\gamma \right|}\le A^{k}(B+k)^{\tau }\) and \(\delta ^{k+\tau }\gamma \) an algebraic integer.
Source: [ Wal00 , Lemma 4.9, p. 130 ] , in the form used in its §4.6.
Let \(x_1,\dots ,x_{d_1}\in \mathbb {C}^{\iota }\) be linearly independent over \(\mathbb {Q}\), let \(k_0\in \iota \) and \(T_0,T_1\in \mathbb {N}\), and let the coefficients \(p_{\tau ,t}\in \mathbb {C}\) be not all zero. Then \(D^{k}F_p(0)\neq 0\) for some \(k\).
Source: [ Wal00 , Exercises 2.4–2.5, p. 60 ] , in the form used in its §4.6.
The frequencies \(\sum _it_ix_i\) are pairwise distinct. On a suitable line, \(F_p\) is a one-variable exponential polynomial with distinct frequencies, which is not identically zero by Lemma 2.6.
Let \(K\) be a number field with an embedding \(\varphi \colon K\to \mathbb {C}\), let \(k_0\in \iota \), and let \(x_1,\dots ,x_{d_1}\) and \(y_j\) (\(j\in \iota \)) be vectors in \(\mathbb {C}^{\iota }\). Suppose that \(x_{i\nu }=\varphi (\xi _{i\nu })\) and \(e^{\langle x_i,y_j\rangle }=\varphi (a_{ij})\) with \(\xi _{i\nu },a_{ij}\in K\), and that \(\eta _j\in K\) satisfy \(\varphi (\eta _j)=y_{jk_0}\) if \(T_0\gt 0\). Suppose that an integer \(\delta \) makes every \(\delta \xi _{i\nu }\), \(\delta \eta _j\) and \(\delta a_{ij}\) an algebraic integer, and that all their houses are at most \(H\ge 1\). Let \(T_1\ge 1\) and \(X\in \mathbb {R}\), and let the coefficients \(p_{\tau ,t}\) be integers with \(|p_{\tau ,t}|\le X\). Then at every grid point \(\sum _js_jy_j\) and for every list \(L\) of \(k\) coordinate directions, \(D_LF_p\bigl(\sum _js_jy_j\bigr)=\varphi (\gamma )\) for some \(\gamma \in K\) with
and \(\delta ^{k+T_0+d_1nT_1S_1}\gamma \) an algebraic integer.
Source: A step of [ Wal00 , § 4.6, step 2, p. 137 ] , with Lemma 4.9 (pp. 130–131).
Lemma 9.13 for each monomial, at the grid point, where \(e^{\langle \sum _it_ix_i,\, \sum _js_jy_j\rangle }=\varphi \bigl(\prod _{i,j}a_{ij}^{t_is_j}\bigr)\).
Let \(k_0\in \iota \) and \(d_0\le 1\), and let \(x_1,\dots ,x_{d_1}\) and \(y_j\) (\(j\in \iota \)) be vectors in \(\mathbb {C}^{\iota }\) such that every coordinate \(x_{i\nu }\) and every \(e^{\langle x_i,y_j\rangle }\) is algebraic and, if \(d_0=1\), every \(y_{jk_0}\) is algebraic. There is \(C\ge 1\) with the following property. Let \(T\ge 1\) and \(S_1\) be integers and \(N\ge 0\), and let \(F_p\), with \(T_0=d_0T\) and \(T_1=T\), have integer coefficients with \(|p_{\tau ,t}|\le e^{N}\). If \(\Delta =D_LF_p\bigl(\sum _js_jy_j\bigr)\neq 0\) for a list \(L\) of \(k\) coordinate directions and a grid point \(\sum _js_jy_j\), then
Source: [ Wal00 , (4.14), § 4.6, step 2, p. 137 ] , with a bracket of a slightly different shape.
Let \(k_0\in \iota \), \(d_0\le 1\) and \(d=d_0+d_1\), and let \(x_1,\dots ,x_{d_1}\) and \(y_j\) (\(j\in \iota \)) be vectors in \(\mathbb {C}^{\iota }\) with \(\sum _{i,\nu }|x_{i\nu }|\le c_x\) and \(\sum _j\| y_j\| \le c_y\). Let \(S_1,T\ge 1\) be integers and \(U,N\gt 0\) and \(E\) real numbers; put \(r=(c_y+2)S_1\) and \(R=Er\), and assume
Then there are integers \(p_{\tau ,t}\), not all zero, with \(|p_{\tau ,t}|\le e^{N}\), such that, with \(T_0=d_0T\) and \(T_1=T\):
\(\bigl|D_LF_p\bigl(\sum _js_jy_j\bigr)\bigr|\le k!\, e^{-U}\) at every grid point and for every list \(L\) of \(k\) coordinate directions;
\(|F_p(w)|\le (T+1)^{d}e^{N}\rho ^{T}e^{c_xT\rho }\) whenever \(\rho \ge 1\) and \(\| w\| \le \rho \).
Source: [ Wal00 , § 4.6, step 3 and (4.16), pp. 138–139 ] .
9.4 The criterion of Schneider–Lang
Let \(n=|\iota |\ge 1\) and let \((y_j)_{j\in \iota }\) be a basis of \(\mathbb {C}^{\iota }\). There is \(c\ge 1\) with the following property. Let \(F\) be an entire function on \(\mathbb {C}^{\iota }\), let \(S_0\), \(S_1\ge 1\) and \(M\ge nS_0\) be integers, and let \(\rho \ge 1\) and \(B\) be real numbers. Suppose that every derivative of \(F\) of total order less than \(M\) vanishes at every grid point, and that \(|F(w)|\le B\) whenever \(\| w\| \le cS_1\rho \). Then at every grid point and for every list \(L\) of \(M\) coordinate directions
Source: [ Wal00 , § 4.6, step 5, p. 140 ] , with Proposition 4.7 (p. 122) and Cauchy’s inequalities (p. XVII).
Theorem 9.6 for \(F\bigl(\sum _jz_jy_j\bigr)\) on the grid \(\{ 0,\dots ,S_1-1\} ^{n}\), then Lemma 9.2 on the unit polydisc about the grid point. This is where the book’s vanishing hypothesis, with every exponent below \(S_0\), is not preserved by the change of variables \(z\mapsto \sum _jz_jy_j\); vanishing by total order is.
Let \(n\ge 1\) be an integer and \(C\ge 1\). There is an integer \(s_0\) such that for every integer \(S_1\ge s_0\) there is \(e_0\) with the following property. Let \(T\ge 1\) and \(E\ge e_0\) be integers with \(\log T\le n\log S_1+n\log E\), and put
Then for every integer \(k\lt nET\)
Source: [ Wal00 , § 4.6, step 4 and (4.17), p. 139 ] .
Let \(n\ge 1\) and \(d\) be integers, \(C\ge 1\), \(c_x\in \mathbb {R}\) and \(c_F\ge 1\). There is an integer \(s_0\) such that for every integer \(S_1\ge s_0\) there is \(e_0\) with the following property. Let \(T\ge 1\) and \(E\ge e_0\) be integers with \(\log T\le n\log S_1+n\log E\), and let \(U\) and \(N\) be as in Lemma 9.19. Then for all integers \(u\ge ET\) and \(M\le 2nu\), with \(\rho '=c_FS_1u/T\),
Source: [ Wal00 , § 4.6, step 6 and (4.19), pp. 140–141 ] .
The book gives these asymptotics without proof; here they hold uniformly in \(u\ge ET\), the book’s \(S_0'\).
Let \(1\le n\lt d\) be integers, \(C\ge 1\), \(c_x,c_y\ge 0\) and \(c_F\ge 1\). There are integers \(S_1,T,E\ge 1\) and real numbers \(U,N\gt 0\) such that, with \(r=(c_y+2)S_1\) and \(R=Er\):
the four assumptions of Lemma 9.17 hold;
the inequality of Lemma 9.19 holds for every integer \(k\lt nET\);
the inequality of Lemma 9.20 holds for all integers \(u\ge ET\) and \(M\le 2nu\), with \(\rho '=c_FS_1u/T\).
Source: [ Wal00 , § 4.6: the conditions (4.15)–(4.19) and the choice of parameters in step 6, pp. 138–141 ] .
Let \(x_1,\dots ,x_{d_1}\in \mathbb {C}^{\iota }\) have algebraic coordinates and be linearly independent over \(\mathbb {Q}\), and let \((y_j)_{j\in \iota }\) be a basis of \(\mathbb {C}^{\iota }\). Either let \(d_0=0\), or let \(d_0=1\) and fix a coordinate \(k\) such that every \(y_{jk}\) is algebraic. If \(|\iota |\lt d_0+d_1\), then the numbers
are not all algebraic.
Source: Schneider (1949), Lang [ Lang66 ] ; this form is [ Wal00 , Corollary 4.2, p. 117 ] for \(d_0\le 1\), proved directly as in its §4.6 (pp. 136–141), on the route of Bertrand–Masser [ BM80 ] and Masser [ Mas81 ] .
Suppose they are all algebraic. Lemma 9.21 chooses the parameters, Lemma 9.17 an auxiliary function \(F_p\), and Lemma 9.14 shows that it is not identically zero. By Lemma 9.16 and condition 2, its derivatives of order less than \(nET\) vanish at the grid points. At the first order at which one does not, Lemma 9.18 bounds it from above and Lemma 9.16 from below, against condition 3.
9.5 Baker’s theorem
Let \(K\subset \mathbb {C}\) be a number field, \((\beta _k)\) a basis of \(K\) over \(\mathbb {Q}\), and \(\ell _k\) logarithms of algebraic numbers. If \(\sum _k\beta _k\ell _k\) is algebraic, then \(\ell _k=0\) for every \(k\).
Source: [ Wal00 , Theorem 4.5 with Lemma 4.6, pp. 119–121 ] ; the route is that of Bertrand–Masser [ BM80 ] and Masser [ Mas81 ] .
The matrix \((\sigma (\beta _k))\) of the embeddings \(\sigma \) of \(K\) is invertible, the trace form being non-degenerate. With \(\lambda _\sigma =\sum _k\sigma (\beta _k)\ell _k\), Theorem 9.22 is applied on the coordinates where \(\lambda _\sigma \neq 0\): with \(d_0=1\) when \(\lambda _\sigma \neq 0\) at the inclusion \(K\subset \mathbb {C}\), and with \(d_0=0\) otherwise.
Let \(\ell _1,\dots ,\ell _n\) be logarithms of algebraic numbers, linearly independent over \(\mathbb {Q}\), and let \(\beta _0,\beta _1,\dots ,\beta _n\) be algebraic numbers, not all zero. Then
Equivalently, \(1,\ell _1,\dots ,\ell _n\) are linearly independent over \(\overline{\mathbb {Q}}\).
Source: Baker [ Bak66 ] ; see [ Bak75 , Theorem 2.1 ] . The route is that of Bertrand–Masser [ BM80 ] , as in [ Wal00 , Chapter 4 ] .
Write \(\beta _i=\sum _kc_{ik}b_k\) in a basis \((b_k)\) of the number field generated by the \(\beta _i\), with \(c_{ik}\in \mathbb {Q}\). Then \(\sum _kb_kL_k=-\beta _0\) is algebraic, where the \(L_k=\sum _ic_{ik}\ell _i\) are logarithms of algebraic numbers (Lemma 2.4), so every \(L_k\) is \(0\) by Theorem 9.23. Independence over \(\mathbb {Q}\) gives \(c_{ik}=0\), hence \(\beta _1=\dots =\beta _n=0\), and then \(\beta _0=0\).
Let \(x\) and \(y\) be logarithms of algebraic numbers, linearly independent over \(\mathbb {Q}\), and let \(a,b\) be algebraic numbers, not both zero. Then \(ax+by\) is transcendental.
Source: Baker [ Bak66 ] ; the proof follows [ Wal00 , Chapter 4 ] , after Bertrand–Masser [ BM80 ] and Masser [ Mas81 ] .
If \(ax+by=\beta \) were algebraic, Theorem 9.24 with \(\beta _0=-\beta \), \(\beta _1=a\) and \(\beta _2=b\) would fail.