On a Class of Self-similar Polycyclic Groups A. C. Dantas, E. de Melo, R. N. de Oliveira, S. N. Sidki Abstract . A group G G is self-similar if it admits a triple ( G , H , f ) (G,H,f) where H H is a subgroup of G G and f : H → G f:H\to G a simple homomorphism, that is, the only subgroup K K of H H , normal in G G and f f -invariant ( K f ≤ K K^{f}\leq K ) is trivial. The group G G then has two chains of subgroups: G 0 = G , H 0 = H , G k = ( H k − 1 ) f , H k = H ∩ G k ​ for ​ ( k ≥ 1 ) . G_{0}=G,\ H_{0}=H,\ G_{k}=(H_{k-1})^{f},\ H_{k}=H\cap G_{k}\ \text{for }(k\geq 1). We define a family of self-similar polycyclic groups, denoted S ​ S ​ P SSP , where each subgroup G k G_{k} is self-similar with respect to the triple ( G k , H k , f ) (G_{k},H_{k},f) for all k k . By definition, a group G G belongs to this S ​ S ​ P SSP family provided f : H → G f:H\rightarrow G is a monomorphism, H k H_{k} and G k + 1 G_{k+1} are normal subgroups of index p p in G k G_{k} ( p p a prime or infinite) and G k = H k ​ G k + 1 G_{k}=H_{k}G_{k+1} . When G G is a finite p p -group in the class S ​ S ​ P SSP , we show that the above conditions follow simply from [ G : H ] = p [G:H]=p and f f is a simple monomorphism. We show that if the Hirsch length of G G is n n , then G G has a polycyclic generating set { a 1 , … , a n } \{a_{1},\ldots,a_{n}\} which is self-similar under the action of f : a 1 → a 2 → … → a n f:a_{1}\rightarrow a_{2}\rightarrow\ldots\rightarrow a_{n} , and then G G is either a finite p p -group or is torsion-free. Surprisingly, the arithmetic of n n modulo 3 3 has a strong impact on the structure of G G . This fact allows us to prove that G G is nilpotent metabelian whose center is free p p -abelian ( p p prime or infinite) of rank at least n / 3 n/3 . We classify those groups G G where H H has nilpotency class at most 2 2 . Furthermore, when p = 2 p=2 , we prove that G G is a finite 2 2 -group of nilpotency class at most 2 2 , and classify all such groups. 2010 Mathematics Subject Classification 20D05; 20D45. With an appendix by Bettina Eick 1. Introduction It has been customary to define self-similarity for a group G as the existence of both a subgroup H of G having finite index m in G , and of a homomorphism f:H\to G . Then the group G is self-similar provided the only normal subgroup K of G which is f -invariant subgroup (that is, K^{f}\leq K ) is trivial. In this case, the triple (G,H,f) is called simple. Such a simple triple leads to a faithful representation of G as a state-closed group of automorphisms of a regular one-rooted tree of degree m , having as basis the set of right cosets of H in G . The restriction on the finiteness of m is unnecessary. The Kaloujnine-Krasner wreath embedding theorem G\to H\wr(G/H) extends recursively, using f , to a representation of G into the automorphism group of the one rooted regular tree of valency m . The simplicity of self-similarity ensures that the representation is faithful (see [ 7 ] ). Strictly speaking, this is the notion of transitive self-similar group, where the action of the group on the first level of the tree is transitive. A generalization to intransitive actions was introduced recently in [ 1 ] . Self-similarity for infinite index m was pursued sporadically in [ 8 ] from 1984, [ 9 ] from 2004, and [ 10 ] from 2005. The following result, which holds for a general m is the starting point for the developments in this paper. Proposition 1.1 ( [ 2 ] , Proposition 23) . Let G be a group, H a subgroup of G , and (G,H,f) a simple triple. Denote H^{f} by G(1) , H\cap H^{f} by H(1) , the center of G by Z(G) and f|_{H(1)} by f_{1} . Suppose that G decomposes as G=Z(G)HH^{f} . Then, (G(1),H(1),f_{1}) is a simple triple. The result fails to be inductive, as we cannot guarantee a corresponding factorization of G(1) . Our definition of the class SSP of polycyclic groups, imposes such a factorization. Specifically, given a self-similar group G with a simple triple (G,H,f) and the chains of subgroups: G_{0}=G,\ H_{0}=H,\ G_{k}=(H_{k-1})^{f},\ H_{k}=H\cap G_{k}\ \text{for }(k\geq 1). we define G as an SSP group if: (1) f:H\to G a monomorphism; (2) (G_{k},H_{k},f) is a simple triple for any k ; (3) H_{k} and G_{k+1} are normal subgroups of G_{k} , and G_{k}=H_{k}G_{k+1} ; (4) G_{k}/H_{k} is cyclic of order p , a prime number or infinite. The group G will be said to be an SSP -lifying, or simply lifting, of H . We have directly from the definition that a finite SSP group is a finite p -group admitting an injective virtual endomorphism of degree p . The converse also holds. Theorem 1.2 . Let G be a finite p -group with a simple triple (G,H,f) where f is injective and [G:H]=p . Then G is an SSP group. By the Auslander-Swan theorem, polycyclic groups of finite Hirsch length have faithful representations in \mathrm{GL}(n,\mathbb{Z}) for some n (see [ 4 , Theorem 3.3.1] ). It was shown in [ 5 ] that the affine group \mathrm{Affine}(n,\mathbb{Z})=\mathbb{Z}^{n}\rtimes\mathrm{GL}(n,\mathbb{Z}) has a faithful representation on an m -ary tree (where m is finite) which is both finite-state and self-similar. Being of finite-state is inherited by subgroups, but this is not true for self-similarity in general; though, for \mathrm{Affine}(n,\mathbb{Z}) , we observe from its action on the tree that its restriction to any subgroup containing the module \mathbb{Z}^{n} induces faithful self-similarity of the subgroup. It is known from [ 2 ] that finitely generated torsion free nilpotent groups of class at most 2 afford faithful transitive self-similar representations on m -trees, where m is finite. Furthermore, Berlatto and dos Santos have presented in [ 3 ] a detailed proof that the N_{3,4} group of Bludov-Gusev, which is 4-generated torsion-free nilpotent of class 3, is not self-similar, be it transitive or not, for any finite m . There are very few general results where a self-similar group of a certain variety embeds constructively in a larger self-similar group of the same variety. One such general result is Proposition 1.3 ( [ 6 ] , Section 3) . Suppose that (G,H,f) is a simple triple where the index [G:H]=m is finite or infinite. Consider \widehat{H}=\underset{k-1}{G\times\cdots\times G}\times H, and define the homomorphisms F:\widehat{H}\to G^{k} , by F:(g_{1},\ldots,g_{k-1},h)\mapsto(h^{f},g_{1},\ldots,g_{k-1}). Then, the triple (G^{k},\widehat{H},F) is simple. In the above proposition, if f is a monomorphism, then F is also a monomorphism. Consequently, this method provides a way to construct new SSP groups. This paper is organized as follows: Section 2 provides a proof of Theorem 1.2 and an exposition of preliminary properties of SSP -groups of Hirsch length n . Notable among these properties are the following: Theorem 1.4 . Let G be an SSP -group of Hirsch length n . (i) There exists a polycyclic generating subset S=\{a_{1},\dots,a_{n}\} of G such that its elements have the same order p (finite or infinite), and S is self-similar under f:a_{i}\to a_{i+1} for 1\leq i\leq n-1 ; (ii) G is a nilpotent group. Considering the generator set S from the preceding theorem, let c be the maximum index s such that [a_{1},a_{2}]=\dots=[a_{1},a_{s}]=e . Then c will be called the cut-off point for the group G , and G is then said to be of G(n,c) -type. Given a non-abelian group G of G(n,c) -type, its subgroups H and H^{f} have the same cut-off point c , and both are SSP -groups of type G(n-1,c) . In Section 3, we construct groups of G(n,c) -type for 2\leq n\leq 6 . In particular, we prove that G(4,2) -type groups do not exist. Furthermore, we show that for n\leq 5 the groups have nilpotency class at most 2, and there exist groups of G(6,4) -type of nilpotency class 3. We comment that Hand-calculations of groups of G(n,c) - type were done up to n=12 , and no groups with a nilpotency class greater than 3 were found. This leads to the problem about the existence of an upper bound for the nilpotency class of such groups in general and whether, indeed, the upper bound is 3. In Section 4, we analyze the structure of groups of G(n,c) -type under the additional condition of being metabelian, and then progress to prove that all SSP -groups are indeed metabelian, having a triple factorization determined by the modulo 3 arithmetic of n . Theorem 1.5 . Let G be a G(n,c) -group, n\geq 3 , and write n=3q(n)+r(n) , where q(n) , r(n) are non-negative integers and r(n)\leq 2 . Then, (i) G is metabelian; (ii) Let n\geq 3 . Then c=c(n,j)=2q(n)+r(n)+j , for some 0\leq j\leq q(n) ; (iii) Let G be a non-abelian. Then any sequence of SSP -liftings of G terminates after a finite number of steps. (iv) [a_{1},a_{c(n,j)+1}] is a non-trivial element of \langle a_{q(n)+1-j},\dots,a_{q(n)+1+r(n)+2j}\rangle ; (v) Z(G)=\langle a_{q(n)+1-j},\dots,a_{c(n,j)}\rangle has rank at least n/3 ; (vi) G has the triple decomposition G=\langle a_{1},\dots,a_{q(n)-j}\rangle\cdot Z(G)\cdot\langle a_{c(n,j)+1},\dots,a_{n}\rangle, where the three factors are free abelian p -groups. In Section 5 we provide faithful matrix presentations for class 2 nilpotent groups of G(n,c) -type. In Section 6 we prove Theorem 1.6 . Let G be a group of G(n+1,c) -type, and suppose its subgroup H has nilpotency class 2. Then, G is nilpotent of class at most 3. As a consequence, we obtain Corollary 1.7 . Let G be an SSP-group where p=2 . Then G is a finite 2-group of nilpotency class at most 2. 2. Preliminary Results on SSP -groups In this section, we prove Theorem 1.2 and list some results on SSP -groups that will be used throughout the text without explicit reference. The following lemma is a straightforward consequence of Proposition 1.1 ; we include a proof for convenience. Lemma 2.1 . Let G be a finite p -group with a simple triple (G,H,f) where [G:H]=p . Then f:H\cap H^{f}\rightarrow H^{f} is also a simple virtual endomorphism of degree p . Proof. Since [G:H]=p we have that H\vartriangleleft G and G=HH^{f} . Moreover, since |\frac{H^{f}}{H\cap H^{f}}|=|\frac{HH^{f}}{H}| we have that H\cap H^{f} has index p in H^{f} . Let K be a subgroup of H\cap H^{f} , normal in H^{f} , with K^{f}\leq K . Note that K^{H} is a normal subgroup of G . Let k^{h}\in K^{H} where k\in K and h\in H . Applying f on k^{h} we obtain that (k^{h})^{f}=(k^{f})^{h^{f}}\in K^{H^{f}}=K . Then (K^{H})^{f}\leq K . Since f is a simple endomorphism we conclude that K=1 , as desired. ∎ Proof of Theorem 1.2 : Proof. Since f is injective and [G:H]=p we have that H and H^{f} are both maximal subgroups of G and then normal in G with G=HH^{f} . Now, as |H^{f}|<|G| we use induction on the order of G and apply Lemma 2.1 to deduce that H^{f} , and consequently G , are SSP groups. ∎ Theorem 2.2 . Let G be an SSP -group of Hirsch length n . Then, the following results hold. (1) The trivial group is the only f -invariant subgroup of G . Proof. This is because f is injective and G_{i+1}=(H_{i})^{f} is not contained in H for any 1\leq i\leq n-1 . ∎ (2) There exists a generating subset S=\{a_{1},\ldots,a_{n}\} of G such that a_{i+1}=(a_{i})^{f} for 1\leq i\leq n-1 . The set S is called the canonical basis for G . Furthermore, S_{i}=\{a_{i+1},...,a_{n}\} is a generating set for G_{i} and \{a_{i+1},...,a_{n-1}\} is a generating set for H_{i} . Strictly speaking, it is the set of cyclic subgroups \{\langle a_{1}\rangle,\ldots,\langle a_{n}\rangle\} that is canonical. Proof. Note that since f is injective (G,H^{f},f^{-1}) is a simple triple. Using induction on n we have that G_{1}=H^{f} has a basis \{a_{2},\ldots,a_{n}\} with the desired properties. Now, we define a_{1}=(a_{2})^{f^{-1}} . ∎ (3) For n\geq 3 , [a_{i},a_{i+1}] = 1 for all i . If [a_{1},a_{2}]=\cdots=[a_{1},a_{s}]=1 , then \{a_{1},\ldots,a_{s}\} is a commutative set. Proof. G_{n-2}=\langle a_{n-1},a_{n}\rangle is abelian and then [a_{n-1},a_{n}]=1 . Now, since [a_{i},a_{i+1}]^{f^{n-(i+1)}}=[a_{n-1},a_{n}] the result follows. ∎ (4) Given a word W(a_{i_{1}},\ldots,a_{i_{s}}) in G then W(a_{1},\ldots,a_{l})^{f}=W(a_{2},\ldots,a_{l+1}) , which is defined provided l\leq n-1 . Whenever necessary, we denote [a_{1},a_{i}]^{f^{t}} by [a_{1},a_{i}]_{t} . Proof. This is because f is injective. ∎ (5) Groups of SSP-type are nilpotent. Proof. Let G be of G(n,c) -type. Proceed by induction on n . The assertion is clearly true for n=2 . Since H , H^{f} are normal subgroups of G , and of G(n-1,c) type, by induction, they are both nilpotent. The result follows since G=HH^{f} . ∎ Define c as the maximum index s such that [a_{1},a_{2}]=\cdots=[a_{1},a_{s}]=1 . Then, 2\leq c\leq n . We call c the cut-off point of the group G , and say G is of G(n,c) -type. We note that G abelian is equivalent to c=n . The cut-off point c is an important parameter in an algorithmic construction of non-abelian SSP -groups. Specifically, for a non-abelian SSP -group G of type G(n,c) , its subgroup H inherits the same cut-off point c . Consequently, H is an SSP -group of type G(n-1,c) . Theorem 2.3 . For n\geq 3 , the polycyclic collection formulas of G are consequences of the closure under the application of powers of f to the set of relations i) a_{1}^{p}=e (where p a prime number or infinite) ii) [a_{1},a_{j}]=e\quad(2\leq j\leq c), iii) [a_{1},a_{j}]=a_{2}^{i_{2}}\dots a_{j-1}^{i_{j-1}}\quad(c+1\leq j\leq n) called the essential relations of G with respect to the set of generators S=\{a_{1},\ldots,a_{n}\} . Proof. For a general polycyclic group P with subnormal series \{P_{i}\ (0\leq i\leq n-1)\} and polycyclic basis B=\{b_{1},b_{2},\dots,b_{n}\} , where b_{i} generates P_{i} modulo P_{i+1} ( 0\leq i\leq n-1 ), its presentation with respect to B is expressed by polycyclic collection formulas having one of three forms: (1) b_{i}^{m_{i}}=b_{i+1}^{u_{i+1}}\dots b_{n}^{u_{n}} (if m_{i} is finite), (2) b_{i+1}b_{i}=b_{i}^{v_{i}}b_{i+1}^{v_{i+1}}\dots b_{n}^{v_{n}} , (3) b_{i+1}^{-1}b_{i}=b_{i}^{w_{i}}b_{i+1}^{w_{i+1}}\dots b_{n}^{w_{n}} , where u_{j} , v_{j} , w_{j} are integers. In the case of SSP -groups, since p=o(a_{i}) is either a prime number or infinite, the first formula is superfluous. The second formula follows from [a_{1},a_{n}]\in\langle a_{2},\dots,a_{n-1}\rangle and by induction on n . While the third formula follows from [(a_{1})^{-1},a_{n}]\in[a_{n},a_{1}]\langle a_{2},\dots,a_{n-1}\rangle=\langle a_{2},\dots,a_{n-1}\rangle and by induction on n . ∎ 3. First Examples (1) Let G be an SSP -group, n\geq 3 , and the cut-off point c=n-1 . Then, the essential relations are i) a_{1}^{p}=e, ii) [a_{1},a_{i}]=e\quad(2\leq i\leq n-1), iii) [a_{1},a_{n}]=a_{2}^{k_{2}}\dots a_{n-1}^{k_{j-1}}\neq e . for some integers k_{2},\dotsc,k_{n-1} . Clearly, G splits as \langle a_{1},\dotsc,a_{n-1}\rangle\rtimes\langle a_{n}\rangle . Furthermore, G has nilpotency class 2 and Z(G)=\langle a_{2},\dotsc,a_{n-1}\rangle. (2) There are no groups of G(4,2) -type. Thus, for a non-abelian SSP -group G of rank 4 , H=\langle a_{1},a_{2},a_{3}\rangle is abelian and the cut-off point c of G is c=3. Proof. Let G be an SSP -group of G(4,2) -type. We have [a_{1},a_{3}]=(a_{2})^{x}\neq e and so [a_{2},a_{4}]=(a_{3})^{x}\neq e . Furthermore, we have [a_{1},a_{4}]=(a_{2})^{u}(a_{3})^{v} . The Hall-Witt identity [a_{1},a_{3},a_{4}]\cdot[a_{4},a_{1},a_{3}]\cdot[a_{3},a_{4},a_{1}]=e holds. The second and third terms are trivial, thus [a_{1},a_{3},a_{4}]=[(a_{2})^{x},a_{4}]=[a_{2},a_{4}]^{x}=(a_{3})^{x^{2}}=e , that is a contradiction. ∎ (3) The possible cut-off points for n=5 are c=4 and 5 , and we know that the corresponding groups are realizable. Proof. Let G be an SSP -group of G(5,3) -type. Then, [a_{1},a_{3}]=e, [a_{1},a_{4}]=a_{2}^{u}a_{3}^{v}\neq e [a_{2},a_{5}]=a_{3}^{u}a_{4}^{v}\neq e [a_{1},a_{5}]=a_{2}^{x}a_{3}^{y}a_{4}^{z}. Then, G^{\prime}\leq\langle a_{2},a_{3},a_{4}\rangle, is an abelian group. The Hall-Witt relation [a_{1},a_{4},a_{5}]\cdot[a_{5},a_{1},a_{4}]\cdot[a_{4},a_{5},a_{1}]=e holds. As the second and third terms are trivial, we have e=[a_{1},a_{4},a_{5}]=[a_{2}^{u}a_{3}^{v},a_{5}]==[a_{2}^{u},a_{5}]=(a_{3}^{u}a_{4}^{v})^{u}=a_{3}^{u^{2}}a_{4}^{uv} u=0 . Similarly, the Hall-Witt relation [a_{1},a_{2},a_{5}]\cdot[a_{5},a_{1},a_{2}]\cdot[a_{2},a_{5},a_{1}]=e, implies v=0 . Thus, [a_{1},a_{4}]=e ; that is a contradiction. ∎ (4) Let n=6 . Suppose G is of G(6,c) -type. Then G_{1} is of G(5,4) or G(5,5) -type, and therefore for G , the possible cut-off points are c=4,5,6 . We know that G is realizable for c=5,6 . (a) Suppose G is an SSP-group of G(6,4) -type. Then, [a_{1},a_{i}]=e\quad\text{for }1\leq i\leq 4; there exist integers u_{2},u_{3},u_{4} such that [a_{1},a_{5}]=a_{2}^{u_{2}}a_{3}^{u_{3}}a_{4}^{u_{4}}\neq e [a_{2},a_{6}]=a_{3}^{u_{2}}a_{4}^{u_{3}}a_{5}^{u_{4}}\neq e [a_{1},a_{6}]=a_{2}^{x_{2}}a_{3}^{x_{3}}a_{4}^{x_{4}}a_{5}^{x_{5}}. We note that H_{1}=H\cap H^{f}=\langle a_{2},a_{3},a_{4},a_{5}\rangle is an abelian normal subgroup of G , and G/H_{1} abelian; that is, G is metabelian. We note further that \langle a_{3},a_{4}\rangle is central in G . Now, as we have done in the analysis of groups of G(5,3) -type, we employ the two Hall-Witt relations [a_{1},a_{5},a_{6}][a_{6},a_{1},a_{5}][a_{5},a_{6},a_{1}]=e, [a_{1},a_{2},a_{6}][a_{6},a_{2},a_{1}][a_{1},a_{6},a_{2}]=e, to conclude that u_{2}=0=u_{4}. Therefore, [a_{1},a_{5}]=a_{3}^{u_{3}},\quad u_{3}\neq 0, [a_{2},a_{6}]=a_{4}^{u_{3}}, and [a_{1},a_{6}]=a_{2}^{x_{2}}a_{3}^{x_{3}}a_{4}^{x_{4}}a_{5}^{x_{5}} for some integers x_{i} , (2\leq i\leq 5) . We compute, [a_{1},a_{6},a_{1}]=[(a_{5})^{x_{5}},a_{1}]=[a_{5},a_{1}]^{x_{5}}=a_{3}^{-u_{3}x_{5}}, [a_{1},a_{6},a_{6}]=[(a_{2})^{x_{2}},a_{6}]=([a_{1},a_{5}]_{1})^{x_{2}}=a_{4}^{u_{3}x_{2}}. Therefore, G has nilpotency class at most 3 , where \gamma_{2}(G)=\langle a_{3}^{u_{3}},a_{4}^{u_{3}},a_{2}^{x_{2}},a_{5}^{x_{5}}\rangle, \gamma_{3}(G)=\langle a_{3}^{-u_{3}x_{5}},a_{4}^{u_{3}x_{2}}\rangle. In particular, \gamma_{3}(G) is trivial if and only if x_{2}=x_{5}=0 . (b) Suppose o(a_{i})=2 . Then as \{a_{2},a_{3},a_{4},a_{5}\} is commutative, [a_{1},a_{6}]=a_{2}^{x_{2}}a_{3}^{x_{3}}a_{4}^{x_{4}}a_{5}^{x_{5}} is either trivial or an involution, and therefore \langle a_{1},a_{6}\rangle is either abelian or dihedral of order 8 . Thus, \gamma_{3}(G) is trivial. (c) The group G decomposes as G=(\langle a_{1},a_{2},a_{3},a_{4}\rangle\rtimes\langle a_{5}\rangle)\rtimes\langle a_{6}\rangle. where [a_{i},a_{5}]=e\quad(2\leq i\leq 4), a_{1}^{a_{5}}=a_{1}a_{3}^{u_{3}},\quad u_{3}\neq 0, [a_{i},a_{6}]=e\quad(3\leq i\leq 5), a_{1}^{a_{6}}=a_{1}a_{2}^{x_{2}}a_{3}^{x_{3}}a_{4}^{x_{4}}a_{5}^{x_{5}}, a_{2}^{a_{6}}=a_{2}a_{4}^{u_{3}}. (d) We construct G starting anew with the presentation H=\langle a_{1},a_{2},a_{3},a_{4},a_{5}\mid a_{i}^{p}=e,[a_{i},a_{5}]=e\quad(2\leq i\leq 4),[a_{1},a_{5}]=a_{3}^{u_{3}},\quad u_{3}\neq 0\rangle, and by defining the action of a_{6} on H as an extension of [a_{i},a_{6}]=e\quad(3\leq i\leq 5), a_{1}^{a_{6}}=a_{1}a_{2}^{x_{2}}a_{3}^{x_{3}}a_{4}^{x_{4}}a_{5}^{x_{5}}, a_{2}^{a_{6}}=a_{2}a_{4}^{u_{3}}. To prove that a_{6} is an automorphism of H we check (i) \{a_{1},a_{2},a_{3},a_{4},a_{5}\}^{a_{6}} generates H , (ii) [a_{i},a_{5}]^{a_{6}}=e\quad(2\leq i\leq 4) , [a_{1},a_{5}]^{a_{6}}=(a_{3}^{u_{3}})^{a_{6}}. Part (i) is straightforward. Part (ii) reduces to checking [a_{2},a_{5}]^{a_{6}}=e,\quad[a_{1},a_{5}]^{a_{6}}=a_{3}^{u_{3}} which are also clear. To prove o(a_{6})=p , we note that the centralizer of a_{2} in \langle a_{6}\rangle is trivial, since (a_{2})^{a_{6}^{k}}=a_{2}a_{4}^{ku_{3}} for all integers k , and from u_{3}\neq 0 . Finally, we show that the function f defined on \{a_{1},\dotsc,a_{6}\} by a_{i}\mapsto a_{i+1} (1\leq i\leq 5) extends to the intended similarity on G . (e) Let p be odd or infinite. Then, G has nilpotency class 3 if and only if the pair (x_{2},x_{5}) is different from (0,0) . 4. Metabelian SSP -groups The Hall-Witt identity for a metabelian group G has the simplified form [x,y,z][z,x,y][y,z,x]=e for all x,y,z in G . From the Hall-Witt identity, we will be using in our analysis of metabelian groups of G(n+1,c) -type the relations W(1,i,n+1):\quad[a_{1},a_{i},a_{n+1}][a_{n+1},a_{1},a_{i}][a_{i},a_{n+1},a_{1}]=e, where 1\leq i\leq n+1 . Given the symmetry i\mapsto n+2-i on the interval [2,n] , we refer to the pair W(1,i,n+1) , W(1,n+2-i,n+1) as symmetric . The relations W(1,i,n+1) will be important in identifying c as a numerical function c(n,j) in terms of the decomposition n=3q(n)+r(n),\quad\text{where }q(n)\geq 0,\ r(n)=0,1,2. Define the numerical function c(n,0)=2q(n)+r(n) c(n,j)=c(n,0)+j,\quad\text{where }0\leq j\leq q(n), Note that c(n,q(n))=n . Given a metabelian G(n,c) -type group G . It will be shown that c=c(n,j) for some 0\leq j\leq q(n) . Computations in Section 3 confirm the coincidence of c with the corresponding values of c(n,j) : c(3,0)=2=c(2,0);\quad c(4,0)=3=c(3,1);\quad c(5,0)=4=c(4,1)=4, c(6,0)=4,\quad c(6,1)=5,\quad c(6,2)=6. Theorem 4.1 . Let G be a metabelian group of type G(n,c) , n\geq 3 . Then, (1) c=c(n,j) for some 0\leq j\leq q(n) , (2) [a_{1},a_{c+1}]\in\langle a_{q(n)+1-j},\dotsc,a_{q(n)+1+r(n)+2j}\rangle , a non-trivial element of G . (3) Let G be non-abelian. Then sequences of liftings of G have a bounded number of steps. Proof. The assertion is true for 3\leq n\leq 6 , and for G of type G(n,c) , where c=n-1 (that is, j=q(n)-1 ). (see the examples). Assume assertions (1) and (2) to be true for n . We may assume 0\leq j\leq q(n)-1 . Suppose G lifts to \widehat{G} an SSP -group of type G(n+1,c) . Then, we have in \widehat{G} , (i) [a_{1},a_{c+1}]=a_{q(n)+1-j}^{u}\ a_{q(n)+2-j}^{u^{\prime}}\dotsm\ a_{q(n)+1+r(n)+2j-1}^{v^{\prime}}\ a_{q(n)+1+r(n)+2j}^{v}\neq e , (ii) [a_{1},a_{n+1}]=a_{2}^{z_{2}}\dotsm a_{n}^{z_{n}} . Consider W(1,c+1,n+1):\quad[a_{1},a_{c+1},a_{n+1}][a_{n+1},a_{1},a_{c+1}][a_{c+1},a_{n+1},a_{1}]=e. Evaluate the first of the three terms: \displaystyle[a_{1},a_{c+1},a_{n+1}] \displaystyle=[a_{q(n)+1-j}^{u}\ a_{q(n)+2-j}^{u^{\prime}}\dotsm\ a_{q(n)+1+r(n)+2j-1}^{v^{\prime}}\ a_{q(n)+1+r(n)+2j}^{v},\ a_{n+1}] \displaystyle=[a_{q(n)+1-j}^{u},\ a_{n+1}] (since a_{n+1} commutes with the elements a_{q(n)+2-j},\ \dotsc,\ a_{q(n)+1+r(n)+2j} ) \displaystyle=[a_{1}^{u},\ a_{n+1-q(n)+j}]_{q(n)-j} \displaystyle=[(a_{1})^{u},\ a_{2q(n)+r(n)+j+1}]_{q(n)-j} \displaystyle=[a_{1},\ a_{c+1}]^{u}_{q(n)-j}. The next term: [a_{n+1},\ a_{1},\ a_{c+1}]=[a_{2}^{z_{2}}\dotsm a_{n}^{z_{n}},\ a_{c+1}]=[a_{1}^{z_{2}}\dotsm a_{n-1}^{z_{n}},\ a_{c}]_{1}=e. Now, since [a_{c+1},\ a_{n+1}]=[a_{1},\ a_{n-c+1}]_{c}=[a_{1},\ a_{n-(2q(n)+r(n)+j)+1}]_{c} =[a_{1},\ a_{q(n)+1-j}]_{c}, and q(n)+1-j\leq c , we have [a_{c+1},\ a_{n+1}]=e , and the third term is [a_{c+1},\ a_{n+1},\ a_{1}]=e. Therefore, the above Hall-Witt relation reduces to [a_{1},\ a_{c+1},\ a_{n+1}]=[a_{1},\ a_{c+1}]^{u}_{q(n)-j}=e, [a_{1},\ a_{c+1}]^{u}=e, and thus u=0 , follows, since [a_{1},\ a_{c+1}] is non-trivial. Likewise, expanding the relation W(1,n+1-c,n+1) symmetric to W(1,c+1,n+1) , yields v=0. Hence, in \widehat{G} , the expression in (i) reduces to (ii)^{*}\quad[a_{1},a_{c+1}]=a_{q(n)+2-j}^{u^{\prime}}\dotsm a_{q(n)+1+r(n)+2j-1}^{v^{\prime}}\quad\text{non-trivial}. Lemma 4.2 . There exists 0\leq j^{\prime}\leq q(n+1)-1 such that q(n)+2-j=q(n+1)+1-j^{\prime}, and, q(n)+1+r(n)+2j-1=q(n+1)+1+r(n+1)+2j^{\prime}. In other words, j^{\prime}=q(n+1)-q(n)+(j-1) implies that q(n)+1+r(n)+2j-1=q(n+1)+1+r(n+1)+2j^{\prime}. Proof. Using that 3q(n+1)+r(n+1)=n+1 , we have that \displaystyle q(n+1)+1+r(n+1)+2j^{\prime} \displaystyle=q(n+1)+1+r(n+1)+2(q(n+1)-q(n)+(j-1)) \displaystyle=q(n+1)+1+r(n+1)+2q(n+1)-2q(n)+2(j-1) \displaystyle=3q(n+1)+r(n+1)-2q(n)+2(j-1)+1 \displaystyle=q(n)+1+r(n)+2j-1. ∎ If j^{\prime}<0 , then j=0 and q(n+1)=q(n) . The last is equivalent to n=3q(n) or 3q(n)+1 . Thus, from (i), [a_{1},a_{c+1}] is non-trivial and it belongs to the subgroups \langle a_{q(n)+1}\rangle , or \langle a_{q(n)+1},a_{q(n)+2}\rangle . However, by (ii)^{*} , both possibilities collapse in \widehat{G} , which is a contradiction. Since c^{\prime}=c=2q(n)+r(n)+j and j^{\prime}=q(n+1)-q(n)+(j-1) we reach c^{\prime}=2q(n+1)+r(n+1)+j^{\prime} . In view of the above lemma, we have in \widehat{G} , [a_{1},a_{c+1}]=[a_{1},a_{c^{\prime}+1}]=a_{q(n+1)+1-j^{\prime}}^{x}\dotsm a_{q(n+1)+1+r(n+1)+2j^{\prime}}^{y}\quad\text{non-trivial}. This proves item (2) . As for item (3) , this simply follows from the fact that given c , the number of solutions (n,j) for c(n,j)=c is finite. ∎ 4.1. The center Z(G) of G Proposition 4.3 . Let G be a metabelian group of G(n,c) -type. Define \Sigma_{1}(n,j)=\langle a_{1},\dots,a_{c}\rangle and \Sigma_{n}(n,j)=\Sigma_{1}(n,j)^{f^{q(n)-j}}=\langle a_{q(n)+1-j},\dots,a_{n}\rangle . Then, (1) \Sigma_{1}(n,j) and \Sigma_{n}(n,j) are free abelian p -groups of rank c and are maximal abelian in G ; (2) Z(G)=\Sigma_{1}(n,j)\cap\Sigma_{n}(n,j)=\langle a_{q(n)+1-j},\dots,a_{c}\rangle has rank at least n/3 ; (3) G=\Sigma_{1}(n,j)\cdot\Sigma_{n}(n,j)=\langle a_{1},\dots,a_{q(n)-j}\rangle\,Z(G)\,\langle a_{c+1},\dots,a_{n}\rangle ; (4) If G is non-abelian, then [a_{1},a_{c+1}]_{k} is a linearly independent subset of Z(G) for all 0\leq k\leq q(n)-j-1 . Proof. (1) The first part is clear. We now prove that \Sigma_{1}(n,j) and \Sigma_{n}(n,j) are maximal abelian in G . The proof proceeds by induction on n . The assertion is true for n\leq 2 . Let n\geq 3 and G be non-abelian. Let x=a_{c(n,j)+1}^{m_{1}}\dots a_{n}^{m_{n}}. Suppose \langle a_{1},\dots,a_{c(n,j)},x\rangle is abelian. Note that \langle a_{c(n,j)+1},\ldots,a_{n}\rangle is an abelian subgroup, and therefore x commutes with a_{c(n,j)+1} . On the other hand, \langle a_{2},\dots,a_{c(n,j)+1}\rangle is a maximal abelian subgroup of \langle a_{2},\dots,a_{n}\rangle . By induction, \langle a_{2},\dots,a_{c(n,j)+1},x\rangle is not abelian, which leads to a contradiction. (2) We know that \Sigma_{1}(n,j) is abelian and that c=c(n,j)=2q(n)+r(n)+j . Then, Z(n,j)=\Sigma_{1}(n,j)\cap\Sigma_{n}(n,j)=\langle a_{q(n)-j+1},\dots,a_{c}\rangle is a central subgroup of G . (3) Since G decomposes as G=\langle a_{1},\dots,a_{q(n)-j}\rangle\,Z(n,j)\,\langle a_{c+1},\dots,a_{n}\rangle, a product of three independent abelian groups, given g\in G , we decompose g uniquely as g=g_{1}g_{2}g_{3} with g_{i} in the i -th component above. Thus, g\in Z(G) if and only if g^{\prime}=g_{1}g_{3}\in Z(G) , equivalently, g_{1} commutes with \langle a_{c+1},\dots,a_{n}\rangle , and g_{3} with \langle a_{1},\dots,a_{q(n)-j}\rangle . Now, g_{1}=a_{1}^{i_{1}}\dotsm a_{q(n)-j}^{i_{q(n)-j}} commutes with a_{c+1} if and only if a_{1}^{i_{1}} does (and so, i_{1}=0 ). Similarly, we show g_{1} , g_{3} are trivial. We note that \operatorname{rank}(Z(G))=c-(q(n)-j)=2q(n)+r(n)+j-(q(n)-j)=q(n)+r(n)+2j\geq n/3 . (4) Let k^{*}=q(n)-j-1 . Since [a_{1},a_{c+1}]_{k}\in\langle a_{q(n)+1-j+k},\dots,a_{q(n)+1+r(n)+2j+k}\rangle, and q(n)+1+r(n)+2j+k^{*}=q(n)+1+r(n)+2j+(q(n)-j-1)=c(n,j), we have [a_{1},a_{c+1}]_{k}\in\langle a_{q(n)+1-j+k},\dots,a_{c(n,j)}\rangle, and therefore, k^{*} is maximum k such that [a_{1},a_{c+1}]_{k}\in Z(G). We continue the proof by writing w_{c+1}=[a_{1},a_{c+1}]=(a_{l})^{u_{l}}\dots(a_{c})^{u_{c}}, where u_{i}\in\mathbb{Z}_{p} and u_{l}\neq 0 . The leading term of w_{c+1} is (a_{l})^{u_{l}} . Define \text{Base}\{w_{c+1}\}=\{a_{l},\dots,a_{c}\}. Then, w_{c+1+k}=[a_{1},a_{c+1}]_{k}=(a_{l+k})^{u_{l}}\dots(a_{c+k})^{u_{c}}, for 0\leq k\leq q(n)-j-1 . The leading term of w_{c+1+k} is (a_{l+k})^{u_{l}} , and \text{Base}(w_{c+1+k})=\{a_{l+k},\dots,a_{c+k}\}. Thus, \bigcup\{\text{Base}(w_{c+1+k})\mid 0\leq k\leq q(n)-j-1\} is contained in the commutative set \{a_{q+1-j},\dots,a_{n-1}\} . Since the leading terms of w_{c+1+k} for 0\leq k\leq q(n)-j-1 form the set \{(a_{l+k})^{u_{l}}\mid 0\leq k\leq q(n)-j-1\}, and u_{l}\neq 0 , the independence of \{[a_{1},a_{c+1}]\mid 0\leq k\leq q(n)-j-1\} follows. ∎ Example. Let G be the Heisenberg group G=\langle a_{1},a_{2},a_{3}\mid(a_{i})^{m}=1,[a_{1},a_{3}]=a_{2},[a_{1},a_{2}]=[a_{2},a_{3}]=e\rangle, where m is finite or infinite. Then G is a SSP of G(3,2) -type where f:a_{1}\to a_{2}\to a_{3} . More generally, using Proposition 1.3 , we can construct the group P_{k} ( k\geq 2 ) with 3k -generators, having a triple decomposition into 3 abelian subgroups: P_{k}=\langle b_{1},\ldots,b_{k}\rangle\cdot\langle b_{k+1},\ldots,b_{2k}\rangle\cdot\langle b_{2k+1},\ldots,b_{3k}\rangle, where the second factor is central, and where the Heisenberg relations hold: [b_{1},b_{2k+1}]=b_{k+1},\ldots,[b_{k},b_{3k}]=b_{2k} . Thus P_{k}=\langle b_{1},b_{2k+1}\rangle\times\cdots\times\langle b_{k},b_{3k}\rangle is isomorphic to G^{k} . If we consider n=3k , then P_{k} is SSP of G(n,2k) -type with canonical base S=\{b_{1},b_{2},\ldots,b_{k},b_{k+1},\ldots,b_{2k},b_{2k+1},\ldots,b_{3k}\} . Moreover, the cut-off point of P_{k} is c(n,j)=2q(n)=2k ( j=0 ) and an interesting observation is that by Theorem 4.1 P_{k} does not lift to a SSP group of type G(n+1,c) since c(n+1,j)=2q(n+1)+r(n+1)+j=2q(n)+1+j^{\prime} . Theorem 4.4 . Groups of SSP-type are metabelian. Proof. We proceed by induction on the number of canonical polycyclic generators n of an SSP -group. From Section 2, the assertion is true for 1\leq n\leq 6 . Suppose G is a metabelian group of G(n,c) -type and \widehat{G} is a lifting of G . Then, G=\langle a_{1},\dots,a_{c}\rangle\langle a_{c+1},\dots,a_{n}\rangle , a product of two abelian subgroups. Now, given the form of c=2q(n)+r(n)+j , it follows that \langle a_{c+1},...,a_{n+1}\rangle is abelian and therefore \widehat{G} is a product of two abelian subgroups. It follows from Ito’s theorem that \widehat{G} is metabelian. ∎ Observations. Let G be a group of G(n,c) -type. Then, (1) Since \operatorname{rank}(Z(G))\geq n/3 , \operatorname{rank}(Z(G))=1 if and only if n=3 . In particular, if p is a prime number, and G=C_{p}\wr C_{p} , then |Z_{i+1}(G)/Z_{i}(G)|=p for i=1,\dots,p-2 and |G/Z_{p-1}(G)|=p^{2} . Thus, G is an SSP -group if and only if p=2 , thus G is a dihedral group of order 8 . Also G/Z_{i}(G) is not an SSP -group, for m odd and i=0,\dots,p-3 . (2) Suppose G is a G(n,c) -group which lifts to \widehat{G} a G(n+1,c) -group. Since G=\langle a_{1},\dots,a_{q(n)-j}\rangle\,Z(G)\,\langle a_{c+1},\dots,a_{n}\rangle and Z(G)=\langle a_{q(n)-j+1},\dots,a_{c(n,j)}\rangle, in calculating the Hall-Witt relations W(1,l,n+1) in the lifting \widehat{G} , we may restrict a_{l} to Z(G)\setminus Z(\widehat{G}) (that is, l=q(n)-j+1 ), or a_{l} to G\setminus Z(G) (that is, l in \{1,\dots,q(n)-j\}\cup\{c+1,\dots,n\} ). If l=q(n)-j+1 , then [a_{n+1},a_{1},a_{q(n)-j+1}][a_{q(n)-j+1},a_{n+1},a_{1}]=e, and as [a_{n+1},a_{1}]\in\langle a_{2},\dots,a_{n}\rangle , [a_{q(n)-j+1},a_{n+1},a_{1}]=e. If 1\leq l\leq q(n)-j , then [a_{1},a_{l}]=e and so [a_{n+1},a_{1},a_{l}][a_{l},a_{n+1},a_{1}]=e. If c+1\leq l\leq n , then [a_{l},a_{n}]=e , and so [a_{1},a_{l},a_{n+1}][a_{n+1},a_{1},a_{l}]=e. Altogether: the W(1,l,n+1) relations restrict to one of three kinds: [a_{q(n)-j+1},a_{n+1},a_{1}]=e; \quad[a_{n+1},a_{1},a_{l}]\cdot[a_{l},a_{n+1},a_{1}]=e,\quad\text{for }1\leq l\leq q(n)-j; \quad[a_{n+1},a_{1},a_{l}]\cdot[a_{1},a_{l},a_{n+1}]=e,\quad\text{for }c+1\leq l\leq n. 5. SSP -groups of nilpotency class 2. Let R(n,j) denote the set of statements in the group G of G(n,c) -type: i) a_{1}^{p}=e . ii) [a_{1},a_{i}]=e , for any 2\leq i\leq c(n,j) . iii) [a_{1},a_{c(n,j)+1+k}]\in\langle a_{q(n)+1-j},\ldots,a_{q(n)+1+r(n)+2j+k}\rangle for any 0\leq k\leq q(n)-j-1 and, [a_{1},a_{c(n,j)+1}] is nontrivial. Theorem 5.1 . Let G be a non-abelian group of G(n,c) -type. Then G is of nilpotency class 2 if and only if (1) c=c(n,j) for some 0\leq j\leq q(n)-1 ; (2) G has the presentation G=\langle a_{1},...,a_{n}\ |\ R(n,j)^{\langle f\rangle}\rangle. Proof. It is straightforward to see that a group with such a presentation is nilpotent of class 2. In the other direction, suppose G has nilpotency class 2. The assertion is true for n=3 . Suppose it is true for n . As G is metabelian, we have c=c(n,j)=2q(n)+r(n)+j,\ \text{and}\ Z(G)=\langle a_{q(n)+1-j},\ldots,a_{c(n,j)}\rangle. Since \widehat{G} is of G(n+1,c) -type, its essential relations are those R(n,j) acquired from some group G of G(n,c) -type, \displaystyle(ii)\ \ \displaystyle[a_{1},a_{i}]=e\quad(2\leq i\leq c(n,j)), \displaystyle(iii)\ \ \displaystyle[a_{1},a_{c(n,j)+1+k}]=a_{q(n)+1-j}^{x_{k}}\ldots a_{q(n)+1+r(n)+2j+k}^{y_{k}},\quad(0\leq k\leq q(n)-j-1), together with \displaystyle(iv)\ \ \displaystyle[a_{1},a_{n+1}]=a_{2}^{z_{2}}\cdots a_{q(n)+1-j}^{z_{q(n)+1-j}}\cdots a_{n-q(n)+j}^{z_{n-q(n)+j}}\cdots a_{n-1}^{z_{n-1}} for some integers x_{k} , y_{k} , z_{l} . Relations (iii) and (iv) reduce in \widehat{G} , in three stages to: (1) x_{k}=y_{k}=0\quad(0\leq k\leq q(n)-j-1) ; (2) z_{2}=\ldots=z_{q(n)+1-j}=0 ; (3) z_{n-q(n)+j}=\ldots=z_{n-1}=0 . These reductions are obtained as follows: e=[a_{1},a_{c+1+k},a_{n+1}]=[a_{q(n)+1-j}^{x_{k}},a_{n+1}]=([a_{1},a_{c+1}]^{x_{k}})_{(q(n)-j)} produces x_{k}=0 ; e=[a_{n+1-c-k},a_{n+1},a_{1}]=[a_{c+1}^{y_{k}},a_{1}]=[a_{1},a_{c+1}]^{-y_{k}} produces y_{k}=0 ; e=[a_{1},a_{n+1},a_{c+1}]\ \text{produces}\ z_{2}=0, \vdots e=[a_{1},a_{n+1},a_{n+1}]\ \text{produces}\ z_{q(n)+1-j}=0; e=[a_{1},a_{n+1},a_{q(n)}]\ \text{produces}\ z_{n-1}=0, \vdots e=[a_{1},a_{n+1},a_{1}]\ \text{produces}\ z_{n-q(n)+j}=0. ∎ 5.1. Matrix representations of class 2 nilpotent groups of G(n,c) -type We produce in this subsection faithful linear representations over \mathbb{Z}_{p} of nilpotent groups of G(n,c) -type. This goes to show that the presentations given for groups in Theorem 5.1 are consistent, and therefore all have Hirsch length n . As we have seen, if G is a classe 2 nilpotent group of G(n,c) -type, G has a factorization by abelian subgroups G=V\rtimes K , where V is a normal subgroup of G , and V itself factors as V=W\oplus Z . The abelian subgroups are : W = \langle a_{1},...,a_{q(n)-j}\rangle ; Z = \langle a_{q(n)+1-j},...,a_{c(n,j)}\rangle , the center of G ; K=\langle a_{c(n,j)+1},...,a_{n}\rangle . Following the characterization of G , we consider V as an \mathbb{Z}_{p}[K] module written additively and construct matrix [a_{s}] representing elements of K . Let u_{k} be the additive form of the right-hand side of the equation [a_{1},a_{c(n,j)+1+k}]=(a_{q(n)+1-j})^{x_{k}}....(a_{q(n)+1+r(n)+2j+k})^{y_{k}}, where 0\leq k\leq q(n)-j-1 . The action of K on V is as follows: \displaystyle[a_{c(n,j)+1}]:a_{1}\mapsto a_{1}+u_{0},a_{i}\mapsto a_{i}\quad(2\leq i\leq c(n,j)) [a_{c(n,j)+2}]:a_{1}\mapsto a_{1}+u_{1},a_{2}\mapsto a_{2}+(u_{0})^{f},a_{i}\mapsto a_{i}\quad(3\leq i\leq c(n,j)) where we recall that u_{0} is non-trivial. In general, \displaystyle[a_{c(n,j)+k}]:a_{1} \displaystyle\mapsto a_{1}+u_{k-1},\ldots,a_{k}\mapsto a_{k}+(u_{0})^{f^{k-1}},a_{i}\mapsto a_{i}\quad(k+1\leq i\leq c(n,j)) Thus, the matrix representation in block form of [a_{c(n,j)+k}] is \displaystyle\begin{bmatrix}I_{(q(n)-j)\times(q(n)-j)}&[B_{k}]\\ 0_{(q(n)+r(n)+2j)\times(q(n)-j)}&I_{(q(n)+r(n)+2j)\times(q(n)+r(n))}\end{bmatrix} where \displaystyle[B_{k}]=\begin{bmatrix}u_{k}&0&0&\ldots&0\\ 0&u_{k-1}&0&\ldots&0\\ \vdots&0&\ddots&\vdots&\vdots\\ \vdots&\vdots&\ldots&u_{0}&0\\ 0&0&\ldots&0&0\end{bmatrix} The groups G constructed above are of affine sort and can be represented matricially as \displaystyle\begin{bmatrix}1&V\\ 0_{1\times(q-j)}&[K]\end{bmatrix} where V is in row form. Finally, in the affine representation of G , the subgroup which corresponds to H is \langle a_{1},...,a_{c(n,j)}\rangle\langle[a_{c(n,j)+1}],...,[a_{n-1}]\rangle , that to H^{f} is \langle a_{2},...,a_{c(n,j)+1}\rangle\langle[a_{c(n,j)+2}],...,[a_{n}]\rangle, and what corresponds to the monomorphism f is the extension of the obvious isomorphisms \displaystyle f_{0}:\langle a_{1},...,a_{c(n,j)}\rangle \displaystyle\to\langle a_{2},...,a_{c(n,j)+1}\rangle,\ f_{1}:\langle[a_{c(n,j)+1}],...,[a_{n-1}]\rangle \displaystyle\to\langle[a_{c(n,j)+2}],...,[a_{n}]\rangle. 6. Lifting SSP groups having nilpotency class 2 We begin this section with a solution of a specific f -equation over \mathbb{Z}_{p} . Let B=\langle b_{1},\ldots,b_{n}\rangle be a multiplicative free abelian p -group ( p prime or infinite) of rank n\geq 2 , which is self-similar by f:b_{i}\rightarrow b_{i+1}\quad(1\leq i\leq n-1). Let w=b_{1}^{x_{1}}\ldots b_{s}^{x_{s}} be a non-trivial element of B such that 1\leq s\leq n-1 . Given 1\leq t\leq n-s , and u,v non-negative in \mathbb{Z}_{p} (if p is a prime number, we view \mathbb{Z}_{p} as \{0,\ldots,p-1\} ), let w^{u+vf^{t}}=\left(b_{1}^{x_{1}u}\ldots b_{s}^{x_{s}u}\right)\left(b_{1+t}^{x_{1}v}\ldots b_{s+t}^{x_{s}v}\right). Proposition 6.1 . Suppose w is nontrivial, and w^{u+vf^{t}}=e . Then, u=v=0 . Proof. w=b_{1}^{x_{1}}\ldots b_{s}^{x_{s}} , w^{f^{t}}=b_{1+t}^{x_{1}}\ldots b_{s+t}^{x_{s}}. (1) The assertion is true as the set of indices \{1,\ldots,s\} , \{1+t,\ldots,s+t\} of the pair w , w^{f^{t}} are disjoint (that is, s<1+t ). (2) Suppose 1+t\leq s . Then, from w^{u+vf^{t}}=\left(b_{1}^{x_{1}}\ldots b_{t}^{x_{t}}\right)^{u}\left(b_{1+t}^{x_{1+t}u+x_{1}v}\ldots b_{s}^{x_{s}u+x_{s-t}v}\right)\left(b_{1+s}^{x_{1+s-t}}\ldots b_{s+t}^{x_{s}}\right)^{v} we conclude (i) \{x_{1},\ldots,x_{t}\}u=\{0\} , (ii) L=\{x_{t+1}u+x_{1}v,\ldots,x_{s}u+x_{s-t}v\}=\{0\} , (iii) \{x_{1+s-t},\ldots,x_{s}\}v=\{0\} . Write s modulo t as s=s^{\prime}t+r^{\prime} with 1\leq s^{\prime} , 0\leq r^{\prime}\leq t-1 , and accordingly, partition L into s^{\prime} t -blocks followed by an r^{\prime} -block. Thus, L=\bigcup_{i=1}^{s^{\prime}}L_{i}\cup M, where \displaystyle L_{1} \displaystyle=\{x_{t+1}u+x_{1}v,\ldots,x_{2t}u+x_{t}v\}, \displaystyle L_{2} \displaystyle=\{x_{2t+1}u+x_{t+1}v,\ldots,x_{3t}u+x_{2t}v\}, \displaystyle L_{i} \displaystyle=\{x_{it+1}u+x_{(i-1)t+1}v,\ldots,x_{(i+1)t}u+x_{it}v\}, \displaystyle M \displaystyle=\{x_{s^{\prime}t+1}u+x_{(s^{\prime}-1)t+1}v,\ldots,x_{s}u+x_{s-t}v\}. Suppose u is not zero. Then, from (i) x_{1}=\ldots=x_{t}=0. Therefore, L_{1}=\{x_{t+1}u+x_{1}v,\ldots,x_{2t}u+x_{t}v\}=\{x_{t+1}u,\ldots,x_{2t}u\}=\{0\}; thus, x_{t+1}=\ldots=x_{2t}=0. By induction, L_{s^{\prime}-1}=\{x_{(s^{\prime}-1)t+1}u+x_{(s^{\prime}-2)t+1}v,\ldots,x_{s^{\prime}t}u+x_{(s^{\prime}-1)t}v\}=\{x_{(s^{\prime}-1)t+1}u,\ldots,x_{s^{\prime}t}u\}=\{0\}. Further, M=\{x_{s^{\prime}t+1}u+x_{(s^{\prime}-1)t+1}v,\ldots,x_{s}u+x_{s-t}v\}=\{x_{s^{\prime}t+1}u,\ldots,x_{s}u\}=\{0\}. Hence, w=e; a contradiction. If u=0 , then we directly obtain v=0 . ∎ Theorem 6.2 . Let G be a group of type G(n,c(n,j)) of nilpotency class 2, having essential relations R(n,j) , as in Theorem 5.1 . Suppose \widehat{G} is a lifting of G . Then \widehat{G}=\langle a_{1},...,a_{n+1}\ |\ \widehat{R}^{\langle F\rangle}\rangle , where F:a_{1}\to a_{2}\to\ldots\to a_{n}\to a_{n+1}, and \widehat{R} is the set of essential relations : (i) (a_{1})^{p}=e; (ii) [a_{1},a_{i}]=e\ \ (2\leq i\leq c(n,j)); (iii) [a_{1},a_{c(n,j)+1+k}]\in\langle a_{q(n)+2-j},\ldots,a_{q(n)+r(n)+2j+k}\rangle\ \ 0\leq k\leq q(n)-j-1\rangle , and is nontrivial if k=0 ; (iv) [a_{1},a_{n+1}]\in\langle a_{q(n)+1-j},\ldots,a_{c(n,j)+1}\rangle. We note that (iii) implies that for 1\leq t