Bourbaki

§ 5. Magnus algebras

In this paragraph, X denotes a set, F(X) the free group constructed on X (Algebra, Chapter I, § 7, no. 5) and A(X) the free associative algebra constructed on X with its total graduation (An(X))n≥0(A^n(X))_{n \geq 0} (cf. Algebra, Chapter III, § 3, no. 1, Example 3). X is identified with its images in F(X) and A(X).

1. Magnus algebras

Let A^(X)\hat{A}(X) be the product module ∏n≥0An(X)\prod_{n \geq 0} A^n(X). We define on A^(X)\hat{A}(X) a multiplication by the rule

(a⋅b)n=∑i=0nai⋅bn−i (a \cdot b)_n = \sum_{i=0}^n a_i \cdot b_{n-i}

where a=(an)a = (a_n) and b=(bn)b = (b_n) are in A^(X)\hat{A}(X). We know (Commutative Algebra, Chapter III, § 2, no. 12, Example 1) that A^(X)\hat{A}(X) is an associative algebra and that A(X)A(X) is identified with the subalgebra of A^(X)\hat{A}(X) consisting of the sequence all of whose terms are zero except for a finite number.

A^(X)\hat{A}(X) is given the product topology of the discrete topologies on the factors

A^n(X); this topology makes A^(X)\hat{A}(X) into a complete Hausdorff topological algebra, when the ring K has the discrete topology, and A(X)A(X) is dense in A^(X)\hat{A}(X)

Let a=(an)∈A^(X)a = (a_n) \in \hat{A}(X); the family (an)n≥0(a_n)_{n \geq 0} is summable and a=∑n≥0ana = \sum_{n \geq 0} a_n.

For every integer m≥0m \geq 0, let A^m(X)\hat{A}_m(X) denote the ideal consisting of the series a=∑n≥mana = \sum_{n \geq m} a_n such that an∈An(X)a_n \in A^n(X) for all n≥mn \geq m. This sequence of ideals is a fundamental system of neighbourhoods of 0 in A^(X)\hat{A}(X) and an integral filtration on A^(X)\hat{A}(X). The order function associated with the above filtration is denoted by ω\omega; then ω(0)=+∞\omega(0) = +\infty and ω(a)=m\omega(a) = m if a=∑n≥mana = \sum_{n \geq m} a_n with an∈An(X)a_n \in A^n(X) for all n≥mn \geq m and am≠0a_m \neq 0 (§ 4, nos. 1 and 2).

A^(X)\hat{A}(X) is called the Magnus algebra of the set X with coefficients in K. If there is any ambiguity over K we write A^K(X)\hat{A}_K(X).

Proposition 1

Let B be a unital associative algebra with a real filtration (Bα)α∈R(B_\alpha)_{\alpha \in \mathbf{R}} such that B is Hausdorff and complete (§ 4, nos. 1 and 2). Let f be a mapping of X into B such that there exists λ>0\lambda > 0 for which f(X)⊂Bλf(X) \subset B_\lambda. Then f can be extended in one and only one way to a continuous unital homomorphism f^\hat{f} of A^(X)\hat{A}(X) into B.

Let f′f' be the unique unital algebra homomorphism of A(X)A(X) into B extending ff (Algebra, Chapter III, § 2, no. 7, Proposition 7). We show that f′f' is continuous: f′(An(X))⊂Bnλf'(A^n(X)) \subset B_{n\lambda} whence f′(A^n(X)∩A(X))⊂Bnλf'(\hat{A}_n(X) \cap A(X)) \subset B_{n\lambda}. Therefore f′f' can be extended in one and only one way by continuity to a homomorphism f^:A^(X)→B\hat{f} : \hat{A}(X) \to B.

We preserve the hypotheses and notation of Proposition 1 and let u∈A^(X)u \in \hat{A}(X). The element f^(u)\hat{f}(u) is denoted by u((f(x))x∈X)u((f(x))_{x \in X}) and called the result of substituting the f(x) for the x in u. In particular, u((x)x∈X)=uu((x)_{x \in X}) = u. Now let u=(uy)y∈Yu = (u_y)_{y \in Y} be a family of elements of A^1(X)\hat{A}_1(X) and let v∈A^(Y)v \in \hat{A}(Y). The above allows us to define the element v((uy)y∈Y)∈A^(X)v((u_y)_{y \in Y}) \in \hat{A}(X). It is denoted by v∘uv \circ u. As uy((f(x)))∈Bλu_y((f(x))) \in B_\lambda, the elements uy((f(x)))u_y((f(x))) can be substituted for the y in v. The mappings v↦(v∘u)((f(x)))v \mapsto (v \circ u)((f(x))) and v↦v((uy((f(x))))))v \mapsto v((u_y((f(x)))))) are then two continuous homomorphisms of unital algebras of A^(X)\hat{A}(X) into B taking the same value uy((f(x)))u_y((f(x))) at the element y∈Yy \in Y. Therefore (Proposition 1)

(v∘u)((f(x)))=v((uy((f(x)))))) (v \circ u)((f(x))) = v((u_y((f(x))))))

for all v∈A^(Y)v \in \hat{A}(Y).

2. Magnus group

For all a=(an)n≥0a = (a_n)_{n \geq 0} in A^(X)\hat{A}(X), the element a0a_0 of K will be called the constant term of a and denoted by ε(a)\varepsilon(a). Formula (1) shows that ε\varepsilon is an algebra homomorphism of A^(X)\hat{A}(X) into K.

Lemma 1

For an element a of A^(X)\hat{A}(X) to be invertible, it is necessary and sufficient that its constant term be invertible in K.

If aa is invertible in A^(X)\hat{\mathbf{A}}(\mathbf{X}), ε(a)\varepsilon(a) is invertible in KK. Conversely, if ε(a)\varepsilon(a) is invertible in KK, there exists u∈A^1(X)u \in \hat{\mathbf{A}}_1(\mathbf{X}) such that a=ε(a)(1−u)a = \varepsilon(a)(1-u); we write b=(∑n≥0un)ε(a)−1b = \left( \sum_{n \geq 0} u^n \right) \varepsilon(a)^{-1}. Then ab=ba=1ab = ba = 1 and aa is invertible.

The set of elements of A^(X)\hat{\mathbf{A}}(\mathbf{X}) of constant term 1 is therefore a subgroup of the multiplicative monoid A^(X)\hat{\mathbf{A}}(\mathbf{X}), called the Magnus group constructed on X\mathbf{X} (relative to KK). In this chapter it will be denoted by Γ(X)\Gamma(\mathbf{X}) or simply Γ\Gamma. For every integer n≥1n \geq 1, we denote by Γn\Gamma_n the set of a∈Γa \in \Gamma such that ω(a−1)≥n\omega(a-1) \geq n. By Proposition 2 of § 4, no. 5, the sequence (Γn)n≥1(\Gamma_n)_{n \geq 1} is an integral central filtration on Γ\Gamma.

3. Magnus group and free group

Theorem 1

Let rr be a mapping of X\mathbf{X} into A^(X)\hat{\mathbf{A}}(\mathbf{X}) such that ω(r(x))≥2\omega(r(x)) \geq 2 for all x∈Xx \in \mathbf{X}. The unique homomorphism gg of the free group F(X)F(\mathbf{X}) into the Magnus group Γ(X)\Gamma(\mathbf{X}) such that g(x)=1+x+r(x)g(x) = 1 + x + r(x) for all x∈Xx \in \mathbf{X} is injective.

We first prove three lemmas.

Lemma 2

Let nn be a non-zero rational integer. In the ring of formal power series K[[t]]K[[t]] we write (1+t)n=∑j≥0cj,ntj(1+t)^n = \sum_{j \geq 0} c_{j,n} t^j. There exists an integer j≥1j \geq 1 such that cj,n≠0c_{j,n} \neq 0.

If n>0n > 0, then cn,n=1c_{n,n} = 1 by the binomial formula. Suppose that n<0n < 0 and let m=−nm = -n. If cj,n=0c_{j,n} = 0 for all j≥1j \geq 1, then (1+t)n=1(1+t)^n = 1, whence, taking the inverse, (1+t)m=1(1+t)^m = 1, which contradicts the formula cm,m=1c_{m,m} = 1.

Lemma 3

Let x1,…,xsx_1, \ldots, x_s be elements of X\mathbf{X} such that s≥1s \geq 1 and xi≠xi+1x_i \neq x_{i+1} for 1≤i≤s−11 \leq i \leq s-1; let n1,…,nsn_1, \ldots, n_s be non-zero rational integers. Then the element ∏i=1s(1+xi)ni\prod_{i=1}^s (1+x_i)^{n_i} of A^(X)\hat{\mathbf{A}}(\mathbf{X}) is ≠1\neq 1.

Let mm be a maximal ideal of KK and kk the field K/mK/m; let p:A^K(X)→A^k(X)p : \hat{\mathbf{A}}_K(\mathbf{X}) \to \hat{\mathbf{A}}_k(\mathbf{X}) be the unique continuous homomorphism of unital KK-algebras such that p(x)=xp(x) = x for x∈Xx \in \mathbf{X} (no. 1, Proposition 1). It suffices to prove that p(∏i=1s(1+xi)ni)≠1p\left( \prod_{i=1}^s (1+x_i)^{n_i} \right) \neq 1 and the problem is reduced to the case where KK is a field.

In the notation of Lemma 2:

∏i=1s(1+xi)ni=∑b1≥0⋯∑bs≥0cb1,n1⋯cbs,nsx1b1⋯xsbs. \prod_{i=1}^s (1+x_i)^{n_i} = \sum_{b_1 \geq 0} \cdots \sum_{b_s \geq 0} c_{b_1, n_1} \cdots c_{b_s, n_s} x_1^{b_1} \cdots x_s^{b_s}.

By Lemma 2, there exist integers ai>0a_i > 0 such that cai,ni≠0c_{a_i, n_i} \neq 0 (1≤i≤s1 \leq i \leq s). By Algebra, Chapter I, § 7, no. 4, Proposition 6, no monomial x1b1⋯xsbsx_1^{b_1} \cdots x_s^{b_s} such that bi≥0b_i \geq 0 and (b1,…,bs)≠(a1,…,as)(b_1, \ldots, b_s) \neq (a_1, \ldots, a_s) can be equal to x1a1…xsasx_1^{a_1} \ldots x_s^{a_s}. It follows that the coefficient of x1a1…xsasx_1^{a_1} \ldots x_s^{a_s} in ∏i=1s(1+xi)ni\prod_{i=1}^s (1 + x_i)^{n_i} is ca1,n1…cas,ns≠0c_{a_1, n_1} \ldots c_{a_s, n_s} \neq 0, which implies the result.

Lemma 4

Let σ\sigma be the continuous endomorphism of A^(X)\hat{A}(X) such that σ(x)=x+r(x)\sigma(x) = x + r(x) for x∈Xx \in X (no. 1, Proposition 1). Then σ\sigma is an automorphism and σ(A^m(X))=A^m(X)\sigma(\hat{A}_m(X)) = \hat{A}_m(X) for all m∈Nm \in \mathbf{N}.

σ(x)≡xmod  A^2(X)\sigma(x) \equiv x \mod \hat{A}_2(X) for x∈Xx \in X, whence, for n≥1n \geq 1 and x1,…,xnx_1, \ldots, x_n in XX,

σ(x1)…σ(xn)≡x1…xnmod  A^n+1(X); \sigma(x_1) \ldots \sigma(x_n) \equiv x_1 \ldots x_n \mod \hat{A}_{n+1}(X);
it follows by linearity that σ(a)≡a\sigma(a) \equiv a modulo A^n+1(X)\hat{A}_{n+1}(X) for all a∈An(X)a \in A^n(X) and in particular σ(An(X))⊂A^n(X)\sigma(A^n(X)) \subset \hat{A}_n(X). It follows that σ(An(X))⊂A^n(X)\sigma(A^n(X)) \subset \hat{A}_n(X) for m≥nm \geq n, whence σ(A^n(X))⊂A^n(X)\sigma(\hat{A}_n(X)) \subset \hat{A}_n(X). In other words, σ\sigma is compatible with the filtration (A^m(X))(\hat{A}_m(X)) on A(X)A(X) and its restriction to the associated graded ring is the identity. Hence σ\sigma is bijective (Commutative Algebra, Chapter III, § 2, no. 8, Corollary 3 to Theorem 1).

Finally we prove Theorem 1. Let w≠1w \neq 1 be an element of F(X)F(X). By Algebra, Chapter I, § 7, no. 5, Proposition 7, there exist x1,…,xsx_1, \ldots, x_s in XX and non-zero rational integers n1,…,nsn_1, \ldots, n_s such that s≥1s \geq 1, xi≠xi+1x_i \neq x_{i+1} (1≤i≤s−11 \leq i \leq s-1) and

w=x1n1…xsns. w = x_1^{n_1} \ldots x_s^{n_s}.
In the notation of Lemma 4,
g(w)=∏(1+σ(xi))ni=σ(∏(1+xi)ni), g(w) = \prod (1 + \sigma(x_i))^{n_i} = \sigma(\prod (1 + x_i)^{n_i}),
hence g(w)≠1g(w) \neq 1 by Lemmas 3 and 4.

4. Lower central series of a free group

We shall prove the two following theorems:

Theorem 2

Suppose that in the ring KK the relation n.1=0n.1 = 0 implies n=0n = 0 for every integer nn. Let rr be a mapping of XX into A^(X)\hat{A}(X) such that ω(r(x))≥2\omega(r(x)) \geq 2 for x∈Xx \in X and let gg be the homomorphism of F(X)F(X) into the Magnus group Γ(X)\Gamma(X) such that g(x)=1+x+r(x)g(x) = 1 + x + r(x) for x∈Xx \in X. For all n≥1n \geq 1, CnF(X)C^n F(X) is the inverse image under gg of the subgroup 1+A^n(X)1 + \hat{A}_n(X) of Γ(X)\Gamma(X).

Theorem 3

For all x∈Xx \in X, let c(x)c(x) be the canonical image of xx in F(X)/(F(X),F(X))F(X)/(F(X), F(X)). Let gg be the graded Lie Z\mathbf{Z}-algebra associated with the filtration (CnF(X))n≥1(C^n F(X))_{n \geq 1} of F(X)F(X) (§ 4, no. 6). The unique homomorphism of the free Lie Z\mathbf{Z}-algebra LZ(X)L_\mathbf{Z}(X) into gg which extends cc is an isomorphism.

Loosely speaking, the graded Lie Z\mathbf{Z}-algebra associated with the free group F(X)F(X) (with the lower central series) is the free Lie Z\mathbf{Z}-algebra LZ(X)L_\mathbf{Z}(X).

We write F(X)=F,Γ(X)=Γ,A^(X)=A^,A^Z(X)=A^Z,CnF(X)=Cn,Γn=1+A^n(X)F(X) = F, \Gamma(X) = \Gamma, \hat{A}(X) = \hat{A}, \hat{A}_\mathbf{Z}(X) = \hat{A}_\mathbf{Z}, C^n F(X) = C^n, \Gamma_n = 1 + \hat{A}_n(X) and let α:LZ(X)→g\alpha : L_\mathbf{Z}(X) \to g be the homomorphism introduced in the statement of Theorem 3.

(A) Preliminary reductions.

Let γ\gamma denote the homomorphism of FF into Γ\Gamma defined by γ(x)=1+x\gamma(x) = 1 + x for x∈Xx \in X. By Lemma 4 there exists an automorphism σ\sigma of the algebra A^\hat{A} compatible with the filtration on A^\hat{A} and such that σ(1+x)=g(x)\sigma(1 + x) = g(x) for all x∈Xx \in X; then σ(Γn)=Γn\sigma(\Gamma_n) = \Gamma_n for all nn. As the homomorphisms gg and σ∘γ\sigma \circ \gamma of FF into Γ\Gamma coincide on X,g=σ∘γX, g = \sigma \circ \gamma and hence g−1(Γn)=γ−1(Γn)g^{-1}(\Gamma_n) = \gamma^{-1}(\Gamma_n). Under the hypotheses of Theorem 2, Z\mathbf{Z} can be identified with a subring of KK; the Magnus algebra A^Z\hat{A}_\mathbf{Z} is therefore identified with a subring of A^\hat{A} and the filtration on A^Z\hat{A}_\mathbf{Z} is induced by that on A^\hat{A}. As γ\gamma maps FF into A^Z\hat{A}_\mathbf{Z}, we see that it suffices to prove Theorems 2 and 3 under the supplementary hypotheses K=Z,r=0K = \mathbf{Z}, r = 0 and hence g=γg = \gamma, hypotheses which we shall henceforth make.

(B) Surjectivity of α\alpha.

As XX generates the group F=C1F = C^1, the set c(X)c(X) generates the Z\mathbf{Z}-module g1=C1/C2g^1 = C^1/C^2. But g1g^1 generates the Lie Z\mathbf{Z}-algebra gg (§ 4, no. 6, Proposition 5) and hence c(X)c(X) generates gg, which proves that α\alpha is surjective.

(C) We identify the graded algebra gr(A^)\mathrm{gr}(\hat{A}) with A(X)A(X) under the canonical isomorphisms An(X)→A^n/A^n+1A^n(X) \to \hat{A}_n/\hat{A}_{n+1}. For every integer n≥1n \geq 1, we write Fn=γ−1(Γn)F^n = \gamma^{-1}(\Gamma_n); we know (§ 4, no. 5) that (Fn)n≥1(F^n)_{n \geq 1} is an integral central filtration on FF. Let g′g' denote the associated graded Lie Z\mathbf{Z}-algebra (§ 4, no. 4). Let ff be the Lie algebra homomorphism of g′g' into A(X)A(X) associated with γ\gamma (§ 4, no. 5, Proposition 3). Now Cn⊂FnC^n \subset F^n for every integer n≥1n \geq 1 (§ 4, no. 6, Proposition 4) and hence there is a canonical homomorphism ε\varepsilon of g=⨁n≥1Cn/Cn+1g = \bigoplus_{n \geq 1} C^n/C^{n+1} into g′=⨁n≥1Fn/Fn+1g' = \bigoplus_{n \geq 1} F^n/F^{n+1}

LZ(X)→αg→εg′→fA(X). L_\mathbf{Z}(X) \xrightarrow{\alpha} g \xrightarrow{\varepsilon} g' \xrightarrow{f} A(X).

We write β=f∘ε\beta = f \circ \varepsilon; we give β\beta explicitly as follows: if uu is the class modulo Cn+1C^{n+1} of an element ww of CnC^n, then γ(w)−1\gamma(w) - 1 is of order ≥n\geq n in A^\hat{A} and β(u)\beta(u) is the homogeneous component of γ(w)−1\gamma(w) - 1 of degree nn. In particular,

β(c(x))=xfor all x∈X. \beta(c(x)) = x \quad \text{for all } x \in X.

(D) Proof of Theorems 2 and 3.

The Lie algebra homomorphism β∘α:LZ(X)→A(X)\beta \circ \alpha : L_\mathbf{Z}(X) \to A(X) restricted to XX is the identity by (3) and hence is the canonical injection (§ 3, no. 1). Therefore α\alpha is injective and hence bijective by (B); this proves Theorem 3. As β∘α=f∘ε∘α\beta \circ \alpha = f \circ \varepsilon \circ \alpha is injective and α\alpha is bijective, ε\varepsilon is injective. For all n≥1n \geq 1,

εn:Cn/Cn+1→Fn/Fn+1 \varepsilon_n : C^n/C^{n+1} \to F^n/F^{n+1}

is injective and hence

Cn∩Fn+1=Cn+1. C^n \cap F^{n+1} = C^{n+1}.

C^1 = F = F^1; if C^n = F^n, then C^n \cap F^{n+1} = F^{n+1} whence C^{n+1} = F^{n+1} which proves Theorem 2 by induction on n \geqslant 1.

Corollary

⋂n⩾1CnF(X)={e}. \bigcap_{n \geqslant 1} C^n F(X) = \{e\}.

Applying Theorem 2 with K = \mathbf{Z} and r = 0,

⋂n⩾1CnF(X)=⋂n⩾1g−1(1+A^n(X))=g−1(⋂n⩾1(1+A^n(X)))=g−1(1)={e}. \bigcap_{n \geqslant 1} C^n F(X) = \bigcap_{n \geqslant 1} g^{-1}(1 + \hat{A}_n(X)) = g^{-1}\left(\bigcap_{n \geqslant 1} (1 + \hat{A}_n(X))\right) = g^{-1}(1) = \{e\}.

Remark

Let H be a Hall set relative to X (§ 2, no. 10). Let M be the magma defined by the law of composition (x, y) \mapsto (x, y) = x^{-1}y^{-1}xy on F(X) and let ϕ\phi be the homomorphism of M(X) into M whose restriction to X is the identity. The elements of ϕ(H)\phi(H) are called the basic commutators of F(X) associated with the Hall set H. For every integer n \geqslant 1, let H_n be the subset of H consisting of the elements of length n; we know (§ 2, no. 11, Theorem 1) that the canonical mapping of H_n into L_\mathbf{Z}(X) is a basis of the Abelian group L_\mathbf{Z}^n(X). Moreover, ϕ(Hn)⊂Cn\phi(H_n) \subset C^n; for all m \in H_n, let ϕn(m)\phi_n(m) denote the class mod. C^{n+1} of ϕ(m)∈Cn\phi(m) \in C^n. Theorem 3 then shows that ϕn\phi_n is a bijection of Hn onto a basis of the Abelian group* C^n/C^{n+1}. It follows immediately that, for all w \in F(X) and all i \geqslant 1, there exists a unique element αi\alpha_i of Z(Hi)\mathbf{Z}^{(H_i)} such that, for n \geqslant 1,

w=∏i=1n∏m∈Hiϕ(m)αi(m)mod  .Cn+1, w = \prod_{i=1}^n \prod_{m \in H_i} \phi(m)^{\alpha_i(m)} \mod. C^{n+1},

where the product is calculated according to the total ordering given on H.

Example

Suppose that X is a set with two elements x, y and let H_1 = \{x, y\}, H_2 = \{xy\}. Every element w of F(X) can therefore be written

w≡xayb(x,y)cmod  .C3with a,b,c in Z. w \equiv x^a y^b (x, y)^c \mod. C^3 \quad \text{with } a, b, c \text{ in } \mathbf{Z}.
For w = (xy)^n, a = b = n and c = n(1 - n)/2 (cf. Exercise 9), whence
(xy)n≡xnyn(x,y)n(1−n)/2mod  .C3. (xy)^n \equiv x^n y^n (x, y)^{n(1-n)/2} \mod. C^3.

5. p-Filtration of Free Groups

In this no., p denotes a prime number and we assume that K = \mathbf{F}_p. Let γ\gamma be the homomorphism of F(X) into Γ(X)\Gamma(X) defined by γ(x)=1+x\gamma(x) = 1 + x for x in X; we write Fn(p)(X)=γ−1(1+A^n(X))F_n^{(p)}(X) = \gamma^{-1}(1 + \hat{A}_n(X)). The sequence (Fn(p)(X))n⩾1(F_n^{(p)}(X))_{n \geqslant 1} is an integral central filtration on F(X), which is separated since γ\gamma is injective (no. 3, Theorem 1). It is called the p-filtration on F(X).

Proposition 2

Suppose that X is finite. For every integer n \geqslant 1, the group F(X)/F_n^{(p)}(X) is a finite p-group of nilpotency class \leqslant n.

Arguing by induction on nn, it suffices to prove that Fn(p)(X)/Fn+1(p)(X)F_n^{(p)}(X)/F_{n+1}^{(p)}(X) is a finite commutative pp-group for all n⩾1n \geqslant 1. For all w∈Fn(p)(X)w \in F_n^{(p)}(X), the element γ(w)−1\gamma(w) - 1 of A^(X)\hat{A}(X) is of order ⩾n\geqslant n; we denote by δn(w)\delta_n(w) the homogeneous component of γ(w)−1\gamma(w) - 1 of degree nn. The mapping δn:Fn(p)(X)→An(X)\delta_n : F_n^{(p)}(X) \to A^n(X) is a homomorphism with kernel Fn+1(p)(X)F_{n+1}^{(p)}(X) (§ 4, no. 5, Proposition 3) and hence Fn(p)(X)/Fn+1(p)(X)F_n^{(p)}(X)/F_{n+1}^{(p)}(X) is isomorphic to a subgroup of An(X)A^n(X). Since XX is finite, An(X)A^n(X) is a finite-dimensional vector space over Fp\mathbf{F}_p and hence a finite commutative pp-group and so is Fn(p)(X)/Fn+1(p)(X)F_n^{(p)}(X)/F_{n+1}^{(p)}(X).

Proposition 3

For all w≠1w \neq 1 in F(X)F(X), there exist a finite pp-group GG and a homomorphism ff of F(X)F(X) into GG such that f(w)≠1f(w) \neq 1.

There exist elements x1,…,xrx_1, \ldots, x_r of XX and integers n1,…,nrn_1, \ldots, n_r such that w=x1n1…xrnrw = x_1^{n_1} \ldots x_r^{n_r}. Let Y={x1,…,xr}Y = \{x_1, \ldots, x_r\}. The canonical injection of YY into XX extends to a homomorphism α:F(Y)→F(X)\alpha : F(Y) \to F(X); on the other hand, let β\beta be the homomorphism of F(X)F(X) into F(Y)F(Y) whose restriction to YY is the identity and which maps X−YX - Y to {1}\{1\}. Then β(α(y))=y\beta(\alpha(y)) = y for y∈Yy \in Y and hence β∘α\beta \circ \alpha is the identity automorphism of F(Y)F(Y). There obviously exists w′w' in F(Y)F(Y) such that w=α(w′)w = \alpha(w'); then β(w)=w′≠1\beta(w) = w' \neq 1; now ⋂n⩾1Fn(p)(Y)={1}\bigcap_{n \geqslant 1} F_n^{(p)}(Y) = \{1\} and there therefore exists an integer n⩾1n \geqslant 1 such that β(w)∉Fn(p)(Y)\beta(w) \notin F_n^{(p)}(Y). By Proposition 2, the group G=F(Y)/Fn(p)(Y)G = F(Y)/F_n^{(p)}(Y) is a finite pp-group. If ff is the composition of β\beta and the canonical homomorphism of F(Y)F(Y) onto GG, then f(w)≠1f(w) \neq 1.

Corollary

The intersection of the normal subgroups of finite index in F(X)F(X) is {1}\{1\}.

Exercises

In the following exercises we assume the hypotheses and notation of § 5. FF denotes the free group F(X)F(X) and gg the unique homomorphism of FF into the Magnus group Γ(X)\Gamma(X) such that g(x)=1+xg(x) = 1 + x for all x∈Xx \in X (cf. Theorem 1).

See the exercises for § 5.

Transcribed from the printed volume. Not translated.