On probabilistic identities and coset identities in pro- p p groups Steffen Kionke, Nowras Otmen, Tommaso Toti, Matteo Vannacci, Thomas Weigel Abstract . It is shown that a probabilistic identity on a σ \sigma -compact K K -analytic group G G , K K a non-archimedean local field, is a coset identity. As an application, one concludes that compact groups which are linear over a local field and various pro- p p groups obtained from free constructions satisfy a probabilistic Tits alternative. By means of Lie-theoretic methods, we also study torsion probabilistic identities in virtually free pro- p p and compact p p -adic analytic groups. 2010 Mathematics Subject Classification Primary 20P05; Secondary 20E18, 11M41 1. Introduction 1.1. State of the art The main purpose of this paper is to study the interplay between three important concepts in Group Theory, identities, coset identities and probabilistic identities, in certain classes of pro- p groups. Let \Gamma be a group and let w\in F_{r} be a non-trivial word in the free abstract group of rank r . One says that w is an identity on \Gamma if for all g_{1},\ldots,g_{r}\in\Gamma one has w(g_{1},\ldots,g_{r})=1 . The study of identities in groups has a long tradition, e.g., already W. Burnside raised the question whether a finitely generated group of finite exponent n is necessarily a finite group. Several interesting classes of groups are defined by imposing an identity: abelian, nilpotent and solvable groups, and groups of finite exponent. Moreover, the restricted Burnside problem, which was one of the central problems in Algebra of the twentieth century, was centred around torsion identities. A natural generalisation of identities is the concept of coset identities . One says that w is a coset identity on \Gamma if there exists a finite index subgroup H of \Gamma and elements g_{1},\ldots,g_{r}\in\Gamma such that w(g_{1}h_{1},\ldots,g_{r}h_{r})=1 for all h_{1},\dots,h_{r}\in H . Even though it is clear that any identity is a coset identity, all known examples of groups satisfying a coset identity also satisfy an identity. In recent years probabilistic methods in Group Theory have been used quite successfully. Some difficult group-theoretical results - like the Magnus problem - have been reproved with a much shorter probabilistic proof (see [ DPSS03 ] ). One natural way to introduce probability on a profinite group G is to endow the group G with a Haar probability measure \mu making (G,\mathcal{A},\mu) a probability space, where \mathcal{A} denotes the Borel \sigma -algebra. If \Gamma is residually finite, one may talk about the probability P(\Gamma,w) that k random elements in the profinite completion \widehat{\Gamma} satisfy w . One says that w is a probabilistic identity for \Gamma if P(\Gamma,w) is positive. If \Gamma does not admit any probabilistic identity, one says that it is randomly free . Again, it is clear that a coset identity yields a probabilistic identity. However, to the the authors’ knowledge, all known examples of profinite groups with probabilistic identities are indeed profinite groups with coset identities. The interplay between identities, coset identities and probabilistic identities has been investigated quite intensively. In 2018 Larsen and Shalev proved that groups with infinitely many non-isomorphic non-abelian upper composition factors are randomly free [ LS18 , Thm. 1.5] ; thus, one may restrict the attention to groups with restricted composition factors, e.g. pro- p groups. Their result also implies that finitely generated linear groups satisfy a probabilistic Tits alternative : they are either virtually solvable or randomly free. Moreover, by [ LS16 , Theorem 1.4] the concepts of coset identity and probabilistic identity coincide for groups in this class. Other works investigating the relation between identities and coset identities include [ WZ92 ] , [ dCM07 ] , [ BG07 ] [ Nik23 ] and [ LP00 ] . Already in 2000 Aner Shalev made the following very challenging conjectures (loc. cit.): Conjecture ( [ Sha00 , Conjectures 2 and 3, pg. 7] ) . (a) Let G be a finitely generated pro- p group satisfying some probabilistic identity w . Then G satisfies some identity. (b) Let G be a finitely generated pro- p group satisfying some coset identity. Then G satisfies some identity. Conjecture (b) was answered affirmatively for p -adic analytic pro- p groups in [ BG07 , Prop. 1.9] . In this article we study two main classes of finitely generated pro- p groups, p -adic analytic pro- p groups and certain “free constructions” of pro- p groups (e.g. free pro- p products), and we show that groups in these classes satisfy a probabilistic Tits alternative. In particular, this answers Conjecture (a) above affirmatively for these groups. 1.2. Main results We say that a group word w is an open coset identity in a topological group G , if w is satisfied on the coset of some open subgroup. In profinite groups open coset identities are exactly the coset identities as defined above. Our first main result shows that for \sigma -compact analytic groups over a non-archimedean local field K the concepts of satisfying a probabilistic identity and satisfying a open coset identity coincide. Theorem A . Let G be a \sigma -compact K -analytic group. If w is a probabilistic identity for G , then w is an open coset identity. One of the most studied classes of finitely generated pro- p groups is that of K -analytic pro- p groups, for K a non-archimedean local field; for instance, closed pro- p subgroups of \mathrm{GL}_{n}(\mathbb{Z}_{p}) and \mathrm{GL}_{n}(\mathbb{F}_{p}[\![t]\!]) are such examples. Here we show that a probabilistic Tits alternative holds for this class. By [ BG07 , Thm. 1.3] the following corollary immediately follows. Corollary B . Let G be a compact subgroup of \mathrm{GL}_{n}(K) for a local field K . Then G is either virtually solvable or randomly free. In particular, a p -adic analytic pro- p group is either solvable or it is randomly free. The proof of A is of analytic flavour and relies on the Weierstraß Preparation Theorem. Using B we can also give a new proof of [ LS16 , Thm. 1.4] , showing that a linear group satisfying a probabilistic identity must also satisfy a coset identity; see Corollary 3.5 . Note that our argument also applies to probabilistic identities with parameters. We now concentrate on randomly free groups. One of the first general methods to obtain randomly free groups appears in [ Abé05 ] , where it is proved that weakly branch groups are randomly free by analysing their actions on trees. In [ Sze05 ] , it is shown that also the Nottingham group \mathcal{N}_{p} is randomly free by a different method. Using these results, it follows that free pro- p groups are also randomly free for every odd prime and for p=2 and rank greater than 2. Interestingly, we do not know of a direct proof of this fact. As far as the authors are aware, the assertion that the free pro- 2 group of rank 2 is randomly free does not follow from previous results and requires C below. While it is clear that a group with a randomly free quotient is randomly free, it is not at all clear if a group which has a finite-index randomly free subgroup is itself randomly free. In Section 4 , we prove a particular instance of that; namely, that virtually non-abelian free pro- p groups are randomly free and we deduce that several classes of “free constructions” of pro- p groups are randomly free. It was previously unknown how to construct new randomly free pro- p groups and the statements therein provide different ways with which one can achieve that. We summarize these results in the next theorems. Theorem C . Let G be a virtually free pro- p group. Then G is either virtually procyclic or it is randomly free. This is an application of B together with some Representation Theory. Note that there are examples of pro- p groups for which a probabilistic Tits alternative does not hold; see [ dCM07 , Corollary 10] for an example of a finitely generated pro- p group which satisfies an identity, but is not solvable. Combining C and B we can prove that the following “free constructions” involving pro- p groups satisfy a probabilistic Tits alternative. Theorem D . Let \mathcal{G} one of the following classes of pro- p groups: (i) finitely generated Demushkin groups, (ii) non-trivial free products of pro- p groups, (iii) pro- p analogues of limit groups (in the sense of [ KZ11 ] ), and let G be a group in \mathcal{G} . Then G is either solvable or randomly free. We are able to prove a similar dichotomy for proper amalgamated free pro- p products when p is odd, however, it does not hold in general when p=2 ; see Corollary 4.5 and the subsequent comment. Note that all groups considered in Theorems C and D have positive rank gradient and in 4.9 we ask if all such pro- p groups are randomly free. We remark that a finitely generated pro- p group of positive rank gradient cannot satisfy a coset identity by [ Nik23 , Thm. 1.7] . See the end of Section 4 for a detailed discussion. We want to emphasise that torsion identities hold a special place in the world of identities due to historical reasons (cfr. Burside problem). We remark also that A answers [ LP00 , Conjecture 3] positively for compact subgroups of \mathrm{GL}_{n}(K) where K is a local field. It is therefore worth having other techniques to deal with torsion identities in other groups. In Section 5 we use Lie-theoretic methods to give a different proof of some of the previous results for torsion identities. In Section 5.1 , we study the measure of conjugacy classes in profinite groups and we show that the set of torsion elements has measure zero in virtually free pro- p groups. We believe that the technique developed there can be applied to other classes of groups. Finally, in Section 5.2 , we study torsion identities in compact p -adic analytic groups. In [ FAR25 ] , the authors ask if finitely generated pro- p groups which are not p -adic analytic can have the set of torsion of elements of positive measure. Here, we tackle the ‘dual’ question; namely, using A and Lie-theoretic methods, we obtain necessary and sufficient conditions for a p -adic analytic pro- p group G to have \mathrm{Tor}(G) of positive measure. Corollary E . Let G be a p -adic analytic pro- p group and \mu its normalised Haar measure. Then \mu(\mathrm{Tor}(G))>0 if, and only if, G is solvable and there exists an open uniform pro- p subgroup U of G and t\in\mathrm{Tor}(G) such that C_{U}(t)=1 . This is a direct consequence of the more general 5.7 . Moreover, even though having torsion of positive measure can only occur in the solvable case due to Corollary B, in Corollary 5.8 we give another proof of it. In an ongoing work, we consider the problem of classifying which solvable p -adic analytic pro- p groups can have torsion of positive measure. Notational conventions As it is customary when working with profinite groups, all subgroups will be assumed to be closed and all homomorphisms will be assumed to be continuous, unless stated otherwise. For a group G , we write \mathrm{Tor}(G) to denote its set of torsion elements. 2. Preliminaries In this section, we recall the necessary definitions, notations and basic results that we will need throughout the paper. 2.1. Probabilistic identities A profinite group G , being a compact Hausdorff topological group, admits a unique normalized Haar measure \mu_{G} whose underlying \sigma -algebra is generated by open subsets of G . Let w=w(x_{1},\ldots,x_{r}) be a non-trivial word in the free group of rank r and define, for a profinite group G , the associated word map \begin{matrix}w_{G}\colon&G^{r}&\longrightarrow&G\\ &(g_{1},\dots,g_{r})&\longmapsto&w(g_{1},\dots,g_{r}).\end{matrix} Since w_{G} is continuous, the subset X(G,w)=\{(g_{1},\dots,g_{r})\in G^{r}\mid w(g_{1},\dots,g_{r})=1\} is closed in G^{r} , hence measurable, and we call P(G,w):=\mu_{G^{r}}(X(G,w)) the probability that G satisfies w . Definition 2.1 . Let G be a profinite group and let w be a non-trivial word. We say that w is a probabilistic identity on G if P(G,w)>0 . We say that G is randomly free if it does not satisfy any probabilistic identity. We remark that the definition of randomly free given above is equivalent to saying that, for every n\geq 1 , the probability that an n -tuple of elements of G generates an abstract free subgroup of rank n is 1 ( [ LS18 , Lemma 1.4] ). Further, we say that w=w(x_{1},\ldots,x_{r}) is a coset identity if there exists H\leq_{o}G and elements g_{1},\ldots,g_{r}\in G such that w(g_{1}h_{1},\ldots,g_{r}h_{r})=1 for any h_{1},\ldots,h_{r}\in H . Note that every coset identity is a probabilistic identity, since g_{1}H\times\ldots\times g_{r}H\subseteq X(G,w) implies P(G,w)\geq 1/[G:H]^{r}>0 . Note that w=w(x_{1},\dots,x_{r}) is a coset identity on a residually finite group \Gamma if and only if it is a coset identity on its profinite completion \widehat{\Gamma} . Indeed, if H is a finite index subgroup of \Gamma and g_{1},\dots,g_{r}\in\Gamma such that w(g_{1}h_{1},\dots,g_{r}h_{r})=1 for every h_{1},\dots,h_{r}\in H , then the closure \overline{H} of H in \widehat{\Gamma} is an open subgroup of \widehat{\Gamma} and w(g_{1}\bar{h}_{1},\dots,g_{r}\bar{h}_{r})=1 for every \bar{h}_{1},\dots,\bar{h}_{r}\in\overline{H} . Conversely, let H\leq_{o}\widehat{\Gamma} and let g_{1},\ldots,g_{r}\in G such that w(g_{1}h_{1},\ldots,g_{r}h_{r})=1 for any h_{1},\ldots,h_{r}\in H . Since \Gamma is dense in its profinite completion, g_{i}H\cap\Gamma is non-empty and open in the profinite topology of \Gamma for every i=1,\dots,r . Therefore, there is N\leq\Gamma of finite index and \gamma_{i}\in\Gamma such that \gamma_{i}N\subseteq g_{i}H so that w is a coset identity on \Gamma . We collect in the following lemma some useful properties of the Haar measure for profinite groups. Note that Equation 2.1 was already known for countably based profinite groups, see for example [ LS03 , Lemma 11.1.1] . Lemma 2.2 . Let G be a profinite group and let X be a closed subset of G . If K\unlhd G and \pi\colon G\to G/K is the natural projection, then (2.1) \mu_{G}(X)\leq\mu_{G/K}(\pi(X)). If moreover G is countably based, then (2.2) \mu_{G}(X)=\underset{N\unlhd_{o}G}{\mathrm{inf}}\frac{|XN/N|}{|G/N|}=\underset{i\geq 1}{\mathrm{inf}}\frac{|XN_{i}/N_{i}|}{|G/N_{i}|}, for every \{N_{i}\}_{i\geq 1} descending chain of open normal subgroups of G with trivial intersection. Proof. It is well known, for example by [ Hal50 , Section 63, Theorem C] , that the normalized Haar measure of the quotient group G/K is the pushforward of \mu_{G} by \pi , that is \mu_{G/K}(E)=\mu_{G}(\pi^{-1}(E)) for every Borel subset E\subseteq G/K . If X\subseteq G is closed, it is in particular compact and so \pi(X) is a closed hence measurable subset of G/K . So we have \mu_{G/K}(\pi(X))=\mu_{G}(\pi^{-1}(\pi(X))\geq\mu(X), since X\subseteq\pi^{-1}(\pi(X)) . Equation 2.2 follows from [ LS03 , p. 206] . ∎ If w is a non-trivial word, since \pi(X(G,w))\subseteq X(G/K,w) , Equation 2.1 implies that P(G,w)\leq P(G/K,w) . This observation immediately implies the following corollary, which will be frequently used in the rest of the article. Corollary 2.3 . A profinite group that projects onto a randomly free group is itself randomly free. 2.2. K -analytic groups Let K be a non-archimedean local field, with norm \lvert\cdot\rvert , and let R be its ring of integers. Then R is a compact discrete valuation ring and we fix an uniformiser \pi\in R that generates its maximal ideal \mathfrak{m} . Note that, for x\in K\smallsetminus\{0\} , then |x|=d^{k} , where d\in(0,1) and k is the integer such that x\in\pi^{k}R\smallsetminus\pi^{k+1}R . We follow the exposition in [ Ser65 , Pt. 2] . For r\in\mathbb{R}^{n}_{>0} and x\in K^{n} , by |x|\leq r we mean that |x_{i}|\leq r_{i} for each 1\leq i\leq n . For an open subset U of K^{m} , we say that a function f:U\to K^{n} is analytic if, for each x\in U , there exists r\in\mathbb{R}_{>0}^{m} and power series F_{i}\in R\llbracket X_{1},\ldots,X_{m}\rrbracket , with i=1,\ldots,n such that B(x,r)=x+\{y\in K^{m}:|y|0 , so it suffices to study the fibers of analytic maps, which we carry out in the next lemma. Lemma 3.3 . Let M and N be K -analytic manifolds and let f\colon M\to N be an analytic morphism. Let y\in N . The fibre S=f^{-1}(y) is either locally negligible or has a non-empty interior. Proof. Let x\in S , so f(x)=y . Assume that S does not contain an open neighbourhood of x . We show that this entails local negligibility of S at x . The statement is purely local; hence, by picking centered charts at x and y we can assume that f\colon R^{m}\to R^{n} is given by convergent power series (f_{1},\dots,f_{n}) in m variables and x=0 and y=0 . Step 1: Reduction to n=1 . If f=(f_{1},\dots,f_{n}) is such that S=f^{-1}(0) is not a neighbourhood of x=0 , then one of the sets f_{i}^{-1}(0) is not a neighbourhood of 0 . If f_{i}^{-1}(0) is locally negligible at x , then the smaller set f^{-1}(0) is locally negligible at 0 as well. Step 2: Use 2.5 to deduce local negligibility. Let f(\underline{X})=\sum_{\alpha\in\mathbb{N}_{0}^{m}}c_{\alpha}\underline{X}^{\alpha} be given in multi-index notation, i.e., \underline{X}^{\alpha}=\prod_{i=1}^{m}X_{i}^{\alpha_{i}} . Since f converges on R^{m} , the absolute values |c_{\beta}| tend to zero with |\beta|\to\infty . Consider the graded lexicographic ordering on \mathbb{N}_{0}^{m} . Since f(0)=0 , the constant term is 0, and since f is non-zero, one may find a minimal \alpha\in\mathbb{N}_{0}^{m} with c_{\alpha}\neq 0 . Let r be the total degree of \alpha . The homogeneous summand f_{r} of f of degree r is non-zero, hence there are \lambda_{1},\dots,\lambda_{m-1}\in R with f_{r}(\lambda_{1},\dots,\lambda_{m-1},X_{m})\neq 0 . Substituting X_{i}+\lambda_{i}X_{m} for X_{i} for all i2 , then G surjects onto C_{2}\amalg C_{2}\amalg C_{2} , which is also virtually non-abelian free pro-2. ∎ Before continuing, we recall a fact that will be useful in a few of the upcoming proofs. If G is a pro- p group and H\leq G is a proper subgroup, then there exists N\unlhd_{o}G such that HN is a proper open subgroup of G . Since G/N is nilpotent, the normal closure of HN/N in G/N is a proper normal subgroup of G/N , whose preimage in G gives us a proper open normal subgroup of G that contains H . More generally, this shows that the normal closure H^{G} of H in G is a proper normal subgroup of G . For the remaining proofs in this section, we use profinite Bass-Serre theory; we refer the reader to [ Rib17 ] for the basic definitions and for a deeper exposition on the topic. A finite directed graph is a quintuple (\Gamma,V(\Gamma),E(\Gamma),d_{0},d_{1}) where: \Gamma is a finite set which is a disjoint union of its subsets V(\Gamma) and E(\Gamma) –called the vertex set and edge set of \Gamma , respectively–and functions d_{i}\colon\Gamma\to V(\Gamma) , for i\in\{0,1\} , such that their restriction to V(\Gamma) is the identity– d_{0}(e) and d_{1}(e) are called the initial and terminal vertices of an edge e\in E(\Gamma) , respectively. Let (\mathcal{G},\Gamma) be a graph of pro- p groups over a finite connected graph \Gamma , so to each m\in\Gamma we associate a pro- p group \mathcal{G}(m) . If m is a vertex (resp. edge), we say that \mathcal{G}(m) is a vertex group (resp. edge group). Furthermore, if e is an edge and d_{0}(e) and d_{1}(e) are its initial and terminal vertices, respectively, then we have monomorphisms \partial_{0}\colon\mathcal{G}(e)\to\mathcal{G}(d_{0}(e)) and \partial_{1}\colon\mathcal{G}(e)\to\mathcal{G}(d_{1}(e)) . Let T be a maximal subtree of \Gamma . The fundamental pro- p group of (\mathcal{G},\Gamma) will be denoted by \Pi=\Pi_{1}(\mathcal{G},\Gamma) and is defined as \left(\coprod_{v\in V(\Gamma)}\mathcal{G}(v)\right)\amalg F, where F is the free pro- p group on the set \{t_{e}\mid e\in E(\Gamma)\} , modulo the normal closure N of the set \{t_{e}\mid e\in E(T)\}\cup\{\partial_{0}(x)^{-1}t_{e}\partial_{1}(x)t_{e}^{-1}\mid x\in\mathcal{G}(e),e\in E(\Gamma)\}. We remark that the resulting group is independent of the choice of maximal subtree. If all the vertex and edge groups embed in \Pi , then we say that (\mathcal{G},\Gamma) is injective. That is not always the case, but we can replace (\mathcal{G},\Gamma) by an injective graph of groups (\tilde{\mathcal{G}},\Gamma) such that \Pi_{1}(\mathcal{G},\Gamma)=\Pi_{1}(\tilde{\mathcal{G}},\Gamma) ; notice that the underlying graph does not change (essentially, we replace the vertex and edge groups by their canonical images in the fundamental group). Moreover, we say that an injective graph of groups is reduced if the monomorphisms of edge groups to vertex groups are not surjective. When \Gamma is finite, we can also replace a non-reduced graph of groups by a reduced one with the same fundamental group, which is achieved by collapsing the ‘superfluous’ edges. Details for these procedures may be found in [ Rib17 , Sections 6.4, 7.2] . For a vertex v\in V(\Gamma) , we denote by X_{v} the set of edges in \Gamma incident to v and we say that v is a pending vertex if |X_{v}|=1 . If e\in X_{v} , let J(v,e)\subseteq\{0,1\} be the set of indices i such that d_{i}(e)=v . Corollary 4.3 . Let (\mathcal{G},\Gamma) be an injective and reduced graph of pro- p groups over a finite connected graph \Gamma and let T be a maximal subtree of \Gamma . Define \tilde{\mathcal{G}}(v):=\mathcal{G}(v)/\langle\partial_{i}(\mathcal{G}(e))\mid e\in X_{v},\hskip 1.99997pti\in J(v,e)\rangle^{\mathcal{G}(v)}. If \Gamma is not a tree, suppose that there is a vertex v such that \tilde{\mathcal{G}}(v) is non-trivial or that |E(\Gamma)\smallsetminus E(T)|\geq 2 . If \Gamma is a tree, suppose that |V(\Gamma)|\geq 2 and, if p=2 , that \tilde{\mathcal{G}}(v) is non-trivial for at least 3 vertices or that |\tilde{\mathcal{G}}(v)|\geq 4 for one vertex. Then \Pi_{1}(\mathcal{G},\Gamma) is randomly free. Proof. Define a graph of pro- p groups (\tilde{\mathcal{G}},\Gamma) by setting \tilde{\mathcal{G}}(v) as the vertex groups and letting the edge groups be trivial. Fix T a maximal subtree of \Gamma . Thus, (\tilde{\mathcal{G}},\Gamma) is a graph of pro- p groups and \tilde{\Pi}=\Pi_{1}(\tilde{\mathcal{G}},\Gamma)=\left(\coprod_{v\in V(\Gamma)}\tilde{\mathcal{G}}(v)\right)\amalg L, where L is a free pro- p group with free generating set \{t_{e}\mid e\in E(\Gamma)\smallsetminus E(T)\} . With respect to T , we have that \Pi=\Pi_{1}(\mathcal{G},\Gamma) is the group \left(\coprod_{v\in V(\Gamma)}\mathcal{G}(v)\right)\amalg F, where F is the free pro- p group on the set \{t_{e}\mid e\in E(\Gamma)\} , modulo the normal closure N of the set \{t_{e}\mid e\in E(T)\}\cup\{\partial_{0}(x)^{-1}t_{e}\partial_{1}(x)t_{e}^{-1}\mid x\in\mathcal{G}(e),e\in E(\Gamma)\}. By the universal property of the free pro- p product, we have a surjective homomorphism \left(\coprod_{v\in V(\Gamma)}\mathcal{G}(v)\right)\amalg F\to\tilde{\Pi}. By construction of (\tilde{\mathcal{G}},\Gamma) , this map factors through N and we obtain a surjection \Pi\to\tilde{\Pi} . We divide the proof in two cases. Suppose \Gamma is not a tree. If every \tilde{\mathcal{G}}(v) is trivial and |E(\Gamma)\smallsetminus E(T)|\geq 2 , then \tilde{\Pi} is a non-abelian free pro- p group and thus randomly free. If there exists a vertex v such that \tilde{\mathcal{G}}(v) is non-trivial, then \tilde{\Pi} is a non-trivial free pro- p product and thus randomly free by Corollary 4.2 . Suppose \Gamma is a tree. For p odd, since |V(\Gamma)|\geq 2 , there at least two pending vertices in \Gamma , say v_{1} and v_{2} . The graph of pro- p groups (\tilde{\mathcal{G}},\Gamma) is injective and reduced by assumption, so \tilde{\mathcal{G}}(v_{1}) and \tilde{\mathcal{G}}(v_{2}) are non-trivial. Again, \tilde{\Pi} is a non-trivial free pro- p product. The case p=2 may be proved by a similar argument. ∎ The following is not complicated, but we include a proof for completeness. Lemma 4.4 . Let \Gamma be a connected graph such that |V(\Gamma)|=|E(\Gamma)|=n\geq 2 and suppose that \Gamma has no pending vertices. Then \Gamma is a circuit. Proof. We prove it by induction on n . The case n=2 is clear. Suppose now that n>2 . It is a standard fact that \Gamma has exactly one cycle. If we remove an edge e in this cycle, then T=\Gamma\smallsetminus\{e\} is a tree. In that case, T has at least two pending vertices; let v_{1} and v_{2} be two of them. We conclude that d_{1-i}(e)=v_{1} and d_{i}(e)=v_{2} for some i\in\{0,1\} , otherwise it would contradict the fact that \Gamma has no pending vertices. It follows that the valency of both v_{1} and v_{2} are equal to 2. If we collapse the edge e to v_{2} , we obtain a circuit by induction. Upon reinserting e , we conclude that \Gamma is a circuit. ∎ A few comments are due. First, notice that the assumptions on the graph are not strong, though one very important case is left out: if |V(\Gamma)|=1 and |E(\Gamma)|=|E(T)|+1 , then the corresponding fundamental group is a proper pro- p HNN-extension. Together with free products and with free products with amalgamation, HNN-extensions constitute the other basic type of free construction, so it would be desirable to understand under which conditions they produce randomly free groups. Notice that the pro- p HNN-extensions not covered by the Corollary 4.3 are the ones in which the base subgroup is generated by the associated subgroups. Second, if |V(\Gamma)|\geq 2 and |E(\Gamma)|=|E(T)|+1 , then we can consider two cases. If \Gamma has a pending vertex v , then by the assumption that the graph of groups is injective and reduced, we have that \tilde{\mathcal{G}}(v)\neq 1 and the fundamental group of the graph of groups is going to be randomly free. If \Gamma does not have a pending vertex, then \Gamma is a circuit by Lemma 4.4 . So, for the odd prime case, the cases left to understand are HNN-extensions and fundamental groups \Pi_{1}(\mathcal{G},\Gamma) where \Gamma is a circuit and every vertex group is generated by the embeddings of the edge groups incident to it. For p=2 , one has to also consider the case in which \Gamma is a tree with only two pending vertices; the added difficulty comes from the fact that C_{2}\amalg C_{2} is solvable, making it harder to find a suitable randomly free homomorphic image. An example of a graph for which the above corollary does not apply is the following. Let F_{i}=\langle x_{i},y_{i}\rangle be free pro- p groups of rank 2 for i\in\{1,2,3\} and consider the graph of groups given by \langle y_{1}\rangle=\langle x_{2}\rangle \langle x_{1}\rangle=\langle x_{3}\rangle \langle y_{3}\rangle=\langle y_{2}\rangle F_{1} F_{2} F_{3} where the edges in the chosen maximal subtree are dashed; for ease of notation, the equalities on the edges indicate how the edge group is embedded in each vertex group (which consequently indicates how the amalgamation is carried out). The hypotheses of Corollary 4.3 are not satisfied; that said, the fundamental group \Pi of the above graph of groups surjects onto the free pro- p group L=\langle a,b,c\rangle of rank 3 by mapping x_{1} and x_{3} to a , y_{1} and x_{2} to b , y_{2} and y_{3} to c and the stable letter coming from the edge outside of the maximal subtree to 1, so it is randomly free. We refer the reader to [ RZ10 , Section 9.2] for the construction of amalgamated free pro- p products. Given pro- p groups A , B and C , where C is a subgroup of both A and B , we say that the amalgamated free pro- p product A\amalg_{C}B with amalgamated subgroup C is proper if there is an injection A\ast_{C}B\to A\amalg_{C}B, where \ast is the abstract free product with amalgamation. In other words, we have that A and B embed in A\amalg_{C}B . This consideration is necessary because, in constrast with the abstract case, we do not always have these embeddings in the profinite setting (see, for instance, [ RZ10 , Example 9.2.9] ). Note that A\amalg_{C}B can be expressed as the fundamental pro- p group of \hbox to71.37pt{\vbox to18.9pt{\pgfpicture\makeatletter\hbox{\hskip 35.53577pt\lower-2.2pt\hbox to0.0pt{\lxSVG@begingroup@{_scopebegin} \lxSVG@begingroup@{stroke} \lxSVG@begingroup@{fill} \lxSVG@setlinewidth{\the\pgflinewidth}\lxSVG@begingroup@{stroke-width} \lx@inpgf@ignorespaces\nullfont\hbox to0.0pt{\lxSVG@begingroup@{_scopebegin} {}{{}}{}{{{}}{\lx@inpgf@ignorespaces}{}{\lx@inpgf@ignorespaces}{}{}{}{}{}}{}{{}}{}{{{}}{\lx@inpgf@ignorespaces}{}{\lx@inpgf@ignorespaces}{}{}{}{}{}}\lxSVG@begingroup@{_scopebegin} \color[rgb]{0,0,0}{}\lxSVG@fillstroke\lxSVG@drawpath@unclipped{M -39.37 0 M -36.6 0 C -36.6 1.53 -37.84 2.77 -39.37 2.77 C -40.9 2.77 -42.14 1.53 -42.14 0 C -42.14 -1.53 -40.9 -2.77 -39.37 -2.77 C -37.84 -2.77 -36.6 -1.53 -36.6 0 Z M -39.37 0 M 39.37 0 M 42.14 0 C 42.14 1.53 40.9 2.77 39.37 2.77 C 37.84 2.77 36.6 1.53 36.6 0 C 36.6 -1.53 37.84 -2.77 39.37 -2.77 C 40.9 -2.77 42.14 -1.53 42.14 0 Z M 39.37 0}{} \lx@inpgf@ignorespaces \lxSVG@closescope {}{{}}{} {}{}{}{}{{{}{}}}{{\lx@inpgf@ignorespaces}}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M -39.37 0 L 39.37 0}{fill:none} \lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} \lxSVG@begingroup@{fill} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lxSVG@begingroup@{_scopebegin} \lxSVG@begingroup@{fill} \lxSVG@fill\lxSVG@drawpath@unclipped{M -10.05 3.04 h 20.1 v 18.68 h -20.1 Z}{stroke:none} \lx@inpgf@ignorespaces \lxSVG@closescope }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{-3.93124pt}{5.533pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{60}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}{{\lx@inpgf@ignorespaces}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{-32.20276pt}{6.533pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{60}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}{{\lx@inpgf@ignorespaces}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{24.40936pt}{6.533pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{60}\lxSVG@closescope }}} \lxSVG@closescope }}} \lxSVG@closescope {\lx@inpgf@ignorespaces}{\lx@inpgf@ignorespaces}{\lx@inpgf@ignorespaces}\hss}\lxSVG@discardpath\lxSVG@closescope \hss}}\lxSVG@closescope\endpgfpicture}}. Corollary 4.5 . Let G=A\amalg_{C}B be a proper free pro- p product with amalgamation of pro- p groups such that C is a proper subgroup of A and B . In case that p=2 and |A:C|=|B:C|=2 , we also assume that C is solvable. Then G is either solvable or randomly free. Proof. If p is odd, we apply Corollary 4.3 in order to conclude that G is randomly free. Now, let p=2 . We divide the proof in several steps. First, suppose that A and B are finite. Denote by a=|A| , b=|B| and c=|C| . By [ Pap14 , Lemma 5.2.1] , the abstract amalgamated free product H=A\ast_{C}B has a normal free subgroup L of index a power of 2 and of rank n=\cfrac{ab}{c^{2}}-\cfrac{a}{c}-\cfrac{b}{c}+1. This implies that H is residually- 2 . Since G is the pro-2 completion of H and L is open in G in the pro-2 topology, we have that G is virtually free pro-2. If |A:C| and |B:C| are both equal to 2, then n=1 and G is solvable as an extension of solvable groups. Otherwise, it follows that n>1 and that G is randomly free by C . Second, suppose that either A or B is infinite. Since G is a proper amalgamated free pro-2 product, we can find an indexing set \Lambda and open normal subgroups \{U_{1\lambda}\}_{\lambda\in\Lambda} and \{U_{2\lambda}\}_{\lambda\in\Lambda} of A and B , respectively, such that \bigcap_{\lambda}U_{i\lambda}=1 for i\in\{1,2\} and U_{1\lambda}\cap C=U_{2\lambda}\cap C for each \lambda\in\Lambda . If |A:C|>2 , by choosing \lambda such that |A/U_{1\lambda}:CU_{1\lambda}/U_{1\lambda}|>2 , we get a surjection A\amalg_{C}B\to A/U_{1\lambda}\amalg_{CU_{1\lambda}/U_{1\lambda}}B/U_{2\lambda}, where the image is randomly free by the first part. Now, suppose that |A:C|=2 and |B:C|>2 . Similarly to the previous case, by choosing \lambda such that |B/U_{2\lambda}:CU_{2\lambda}/U_{2\lambda}|>2 , we obtain that A\amalg_{C}B projects onto a randomly free group. Finally, suppose that |A:C|=|B:C|=2 and that C solvable. Since C is normal in both A and B , we have G/C\simeq C_{2}\amalg C_{2} so that G is solvable. ∎ The above ‘dichotomy’ result cannot be strenghtened for p=2 . Indeed, given words w\in F_{m},v\in F_{n} , we denote by w\circ v\in F_{mn} the word (w\circ v)(x_{1},\ldots,x_{nm})=w(v(x_{1},\ldots,x_{n}),\ldots,v(x_{n(m-1)+1},\ldots,x_{nm})). Note that w\circ v is the trivial word if and only if w or v are trivial. Now, let H be the finitely generated pro-2 group constructed in [ dCM07 , Corollary 10] and consider the group G=(C_{2}\times H)\amalg_{H}(C_{2}\times H) . This is a proper amalgamated free pro-2 product by [ RZ10 , Theorem 9.2.4] , which is not solvable and satisfies a non-trivial identity of the form w\circ v where v is a non-trivial identity of the infinite dihedral group and w is a non-trivial identity of H . We recall the following definition of pro- p analogues of limit groups which first appeared in [ KZ11 ] . Let p be a prime number and let \mathcal{G}_{0} be the class of free pro- p groups of finite rank. Define \mathcal{G}_{n} recursively as follows: a group G_{n} is in \mathcal{G}_{n} if G_{n}=G_{n-1}\amalg_{C}A , where G_{n-1}\in\mathcal{G}_{n-1} , A is free abelian pro- p of finite rank and C is a self-centralizing procyclic subgroup of G_{n-1} and also a proper direct summand of A . A pro- p group H is in the class \mathcal{L} if it is a finitely generated pro- p subgroup of a group in \mathcal{G}_{n} for some n\geq 0 . The weight of H is the smallest such n . Our aim will be to prove that groups in \mathcal{L} are either solvable or randomly free. In light of [ SZ14 , Theorem B] , the class \mathcal{L} is suitable for a combinatorial approach. We begin with a lemma. Lemma 4.6 . Let H\in\mathcal{L} be such that d(H)>1 and let x\in H . Then |H:\langle x\rangle^{H}|\geq 3 . Proof. Since H is not procyclic, then \langle x\rangle^{H} is necessarily a proper normal subgroup of H , as remarked earlier, so we can restrict ourselves to the case p=2 . If d(H)=2 , then it is either the free pro-2 group of rank 2 or \mathbb{Z}_{2}^{2} ( [ KZ11 , Theorem 7.3] ) and for both the conclusion holds. Suppose d(H)>2 , let N=\langle x\rangle^{H} and let \Phi(H) be the Frattini subgroup of H . If x\in\Phi(H) , the statement follows since d(H)>2 . Suppose now that x\notin\Phi(H) and let H^{\prime}=[H,H] . Notice that d(H/H^{\prime})>2 , so \langle x\rangle^{H}H^{\prime}/H^{\prime}=\langle x\rangle H^{\prime}/H^{\prime} is a procyclic subgroup of a group which is at best 3-generated. Thus, we have |H:\langle x\rangle^{H}|\geq 4 . ∎ Corollary 4.7 . The groups in the class \mathcal{L} are either abelian or randomly free. Proof. This clearly holds for groups of weight 0. Let G\in\mathcal{L} be a group of weight n\geq 1 . By [ SZ14 , Theorem B] , we have that G is the fundamental pro- p group G=\Pi_{1}(\mathcal{G},\Gamma) of a finite graph of pro- p groups with procyclic or trivial edge groups and finitely generated vertex groups. Since \Gamma is finite, we may assume (\mathcal{G},\Gamma) to be a reduced injective graph of groups. Notice that passing to an injective graph of groups could change the vertex and edge groups; however, since they are homomorphic images of the initial ones, it remains true that the edge groups are procyclic or trivial and that the vertex groups are finitely generated subgroups of G and thus contained in the class \mathcal{L} . We divide the proof in two cases, which essentially involve dealing with the graphs of groups not covered by Corollary 4.3 . Before continuing, we recall two facts about the groups in \mathcal{L} which we will use often: they are commutative-transitive , that is, if [g,h]=[h,k]=1 , then [g,k]=1 ( [ KZ11 , Theorem 5.1] ), and their 2-generated subgroups are either non-abelian free pro- p of rank 2 or abelian ( [ KZ11 , Theorem 7.3] ). For the rest of the proof, we denote the free pro- p group of rank 2 by F_{2} . Case 1: p=2 and \Gamma is a tree with only 2 pending vertices. In other words, (\mathcal{G},\Gamma) is of the form \mathcal{G}(e_{1}) \mathcal{G}(e_{n-1}) \mathcal{G}(v_{1}) \mathcal{G}(v_{2}) \cdots \mathcal{G}(v_{n-1}) \mathcal{G}(v_{n}) where the edge groups are either trivial or procyclic and the vertex groups are finitely generated. If n=2 , the edge group is procyclic and so G is a proper free pro-2 product with amalgamation by [ Rib71 , Theorem 3.2] . Thus, using Lemma 4.6 , Corollary 4.5 and the commutative-transitive property, we conclude that G is either abelian or randomly free. If n=3 , we may assume \mathcal{G}(v_{1}) and \mathcal{G}(v_{3}) are isomorphic to \mathbb{Z}_{2} , the edge groups incident to them embed as 2\mathbb{Z}_{2} and \mathcal{G}(v_{2})=\langle\partial_{1}(\mathcal{G}(e_{1}),\partial_{0}(e_{2})\rangle . The reason we can make these assumptions is the following: if either pending vertex group is not procyclic, then by Lemma 4.6 we have that one vertex v is such that (in the notation of Corollary 4.3 ) |\tilde{\mathcal{G}}(v)|\geq 4 , so we may apply the aforementioned corollary; if \mathcal{G}(v_{2}) is not generated by its incident edge groups, then once again we are under the hypotheses of Corollary 4.3 . In this case, it follows that G is 2-generated, so it is either abelian or randomly free. Suppose now n>3 . Similarly to the case n=3 , we can make the same assumptions on \mathcal{G}(v_{1}) and \mathcal{G}(v_{n}) and their incident edge groups and that the other vertex groups are generated by the embeddings of its incident edge groups. Since the graph of groups is reduced, the \mathcal{G}(v_{i}) are strictly 2-generated for 2\leq i\leq n-1 (if they were procyclic, we would have either that \partial_{1}(\mathcal{G}(e_{i-1}))\leq\partial_{0}(\mathcal{G}(e_{i})) or \partial_{0}(\mathcal{G}(e_{i}))\leq\partial_{1}(\mathcal{G}(e_{i-1})) , both contradicting our assumption on the vertex groups). This means that they are isomorphic either to F_{2} or to \mathbb{Z}_{2}^{2} . If they are all isomorphic to \mathbb{Z}_{2}^{2} , then G is abelian by the commutative-transitive property. Otherwise, let 2\leq i_{0}\leq n-1 be minimal such that \mathcal{G}(v_{i_{0}}) is F_{2} . Without loss of generality, we can have two situations: i_{0}=2 , in which case we get \langle x^{2}\rangle \langle y\rangle \langle z\rangle \langle x\rangle \langle x^{2},y\rangle \langle y,z\rangle \cdots where \mathcal{G}(v_{i_{0}})=\langle x^{2},y\rangle and we get a surjection to C_{2}\amalg\mathbb{Z}_{2} ; or 20 . In fact, if d(G)=d , then we can take k=d by recalling that in this case the Frattini subgroup \Phi(G) is open in G and elements g_{1},\ldots,g_{d}\in G topologically generate G if, and only if, their reductions modulo \Phi(G) generate G/\Phi(G) . It follows that if G=\overline{\langle g_{1},\ldots,g_{d}\rangle} , then g_{1}\Phi(G)\times\ldots\times g_{d}\Phi(G)\subseteq X_{d} . Suppose also that G is randomly free. This implies that \mu(Y_{k})=1 for every positive integer k and, moreover, that \mu(Y_{d}\cap X_{d})>0 . So, for all the examples of finitely generated randomly free pro- p groups considered above, we can not only find dense free abstract subgroups, but we can do so with positive probability. Recall that the rank gradient of a finitely generated pro- p group G is defined as \mathrm{RG}(G)=\underset{H\leq_{o}G}{\mathrm{inf}}\frac{d(H)-1}{|G:H|}, where d(H) denotes the minimal number of generator of the pro- p group H . Moreover, if M\leq_{o}G , then using the Schreier inequality it can be shown that \mathrm{RG}(M)=|G:M|\mathrm{RG}(G) . In particular, pro- p groups that have an open subgroup of positive rank gradient are of positive rank gradient. Note that many groups that we have shown to be randomly free have positive rank gradient, see for example [ GGK20 , Section 1] . On the other hand, by [ Nik23 , Theorem 1.7] , a finitely generated pro- p group of positive rank gradient contains a free non-abelian dense subgroup, so it cannot satisfy any coset identity. We do not know if this result can be generalised to probabilistic identities, so we leave it as a problem. Problem 4.9 . Let G be a finitely generated pro- p group of positive rank gradient. Is G randomly free? 5. Torsion probabilistic identities 5.1. Conjugacy classes and virtually free pro- p groups In this section, we study how to measure conjugacy classes in countably based profinite groups. As an application, using a Lie-theoretic argument that we think is of its own interest, we will give an alternative proof of C for torsion words w=x^{k} , k\in\mathbb{N} . Given x\in G , we denote by x^{G}:=\{x^{g}\mid g\in G\} its conjugacy class in G . It is clear that x^{G} is a closed subset of G . Proposition 5.1 . Let G be a countably base profinite group and let x\in G . Given \{N_{i}\}_{i\geq 1} a descending chain of open normal subgroups of G with trivial intersection, define \displaystyle c_{1}(x)=\underset{i\geq 1}{\mathrm{sup}}|C_{G/N_{i}}(xN_{i})|; \displaystyle c_{2}(x)=\underset{N\unlhd_{o}G}{\mathrm{sup}}|C_{G/N}(xN)|; \displaystyle c_{3}(x)=\underset{K\unlhd G}{\mathrm{sup}}|C_{G/K}(xK)|, where c_{3}(x)=\infty if C_{G/K}(xK) is infinite for some K\unlhd G . Then c_{1}(x)=c_{2}(x)=c_{3}(x):=c(x) and (5.1) \mu(x^{G})=\frac{1}{c(x)}. Proof. Since for any N\unlhd_{o}G one has x^{G}N/N=(xN)^{G/N} , it follows that |x^{G}N/N|=|(xN)^{G/N}|=|G/N:C_{G/N}(xN)|=\frac{|G/N|}{|C_{G/N}(xN)|}. By Lemma 2.2 , one obtains that \mu(x^{G})=\underset{i\geq 1}{\mathrm{inf}}\frac{1}{|C_{G/N_{i}}(xN_{i})|}=\underset{N\unlhd_{o}G}{\mathrm{inf}}\frac{1}{|C_{G/N}(xN)|}. Therefore c_{1}(x)=c_{2}(x) and \mu(x^{G})=1/c_{i}(x) for i=1,2 . Since c_{2}(x)\leq c_{3}(x) , it remains to verify that c_{3}(x)\leq c_{2}(x) . Let K\unlhd G and consider the quotient group G/K\simeq\underset{N\unlhd_{o}G}{\varprojlim}\frac{G}{KN} . Since C_{G/K}(xK) is a closed subgroup of G/K , one has that (5.2) C_{G/K}(xK)\simeq\underset{N\unlhd_{o}G}{\varprojlim}\frac{C_{G/K}(xK)KN}{KN}. Moreover, C_{G/K}(xK)KN/KN\leq C_{G/KN}(xKN) for every N\unlhd_{o}G . Therefore Equation 5.2 implies the following chain of inequalities |C_{G/K}(xK)|=\underset{N\unlhd_{o}G}{\mathrm{sup}}\left|\frac{C_{G/K}(xK)KN}{KN}\right|\leq\underset{N\unlhd_{o}G}{\mathrm{sup}}|C_{G/KN}(xKN)|\leq c_{2}(x). Since the previous inequality holds for every K\unlhd G , it follows that c_{3}(x)\leq c_{2}(x) . ∎ An immediate consequence of Proposition 5.1 is the following sufficient condition to have conjugacy classes of measure zero in a countably based profinite group. Corollary 5.2 . Let G be a countably based profinite group and let x\in G . Suppose that there exists K\unlhd G such that |C_{G/K}(xK)|=\infty . Then \mu(x^{G})=0 . Next we introduce the Lie rings in which we are interested. Given an abstract or profinite group G , let L(G)=\bigoplus_{i\geq 1}\frac{\gamma_{i}(G)}{\gamma_{i+1}(G)} be the associated Lie ring with respect to its lower central series \gamma_{i}(G) . Recall that for a given Lie ring L a Lie ring endomorphism \alpha\colon L\to L is called regular if x^{\alpha}\neq x for every x\in L , x\neq 0 . Moreover, by a result of V.A. Kreknin, a Lie ring that admits a finite order automorphism that is regular needs to be solvable ( [ Khu93 , Theorem 4.3.1] ). Lemma 5.3 . Let F be a finitely generated non-abelian free abstract group and let \alpha\in\mathrm{Aut}(F) be of finite order. Then there is n\in\mathbb{N} such that \alpha has infinitely many fixed points on F/\gamma_{n+1}(F) . Proof. Suppose that F is the free group on X , where |X|\geq 2 . L(F) is not solvable since by [ Ser65 , Theorem 4.6.1] it is the free Lie ring on X . Then, being of finite order, \alpha cannot be regular on L(F) . That is, there is a non-trivial element x\gamma_{n+1}(F)\in\gamma_{n}(F)/\gamma_{n+1}(F) such that x^{\alpha}\gamma_{n+1}(F)=x\gamma_{n+1}(F) . Since F/\gamma_{n+1}(F) is torsion-free and every element of \{x^{i}\gamma_{n+1}(F)\}_{i\geq 1} is fixed by \alpha , this yields the claim. ∎ We show that the same statement holds in the pro- p setting. Lemma 5.4 . Let F be a finitely generated non-abelian free pro- p group and let \alpha\in\mathrm{Aut}(F) be of finite order. Then there is n\in\mathbb{N} such that \alpha has infinitely many fixed points on F/\gamma_{n+1}(F) . Proof. Let E be an abstract non-abelian free group whose pro- p completion is F . If we denote by E_{i} the i -th term of the lower central series of E and by F_{i} the analogous for F , we have that F_{i}\cap E=E_{i} by [ Lub82 , Lemma 2.6] . This implies that the Lie ring homomorphism \phi\colon L(E)\to L(F) defined componentwise as \displaystyle\phi\colon \displaystyle E_{i}/E_{i+1}\to F_{i}/F_{i+1} \displaystyle xE_{i+1}\to xF_{i+1} is injective. Indeed, if x,y\in E_{i} such that xF_{i+1}=yF_{i+1} , then xy^{-1}\in F_{i+1}\cap E=E_{i+1} , that is xE_{i+1}=yE_{i+1} . It follows that L(F) is not solvable since it contains a sub-Lie ring that is not solvable. Since F/F_{i} is torsion-free for every i , the argument used in the proof of Lemma 5.3 can be transferred verbatim. ∎ It is now possible to measure conjugacy classes of torsion elements in a particular class of profinite groups. Proposition 5.5 . Let G be a countably based profinite group and let F\unlhd G be a finitely generated non-abelian free pro- p group. Then \mu(t^{G})=0 for every torsion element t\in G . Proof. Let c_{t}\in\mathrm{Aut}(G) be the conjugation by t . Since F\unlhd G , we have that {c_{t}}_{|_{F}}\in\mathrm{Aut}(F) and |c_{t}|=|t|<\infty . By Lemma 5.4 , c_{t} has infinitely many fixed points on F/\gamma_{n+1}(F) for some n . Recall that \gamma_{n+1}(F) is a characteristic subgroup of F that is normal in G , therefore it is a normal subgroup of G . Then C_{G/\gamma_{n+1}(F)}(t\gamma_{n+1}(F)) is infinite since it contains C_{F/\gamma_{n+1}(F)}(t\gamma_{n+1}(F)) . Therefore \mu(t^{G})=0 by Corollary 5.2 . ∎ In particular, if G is a finitely generated virtually free pro- p group that is not virtually procyclic, then \mu(t^{G})=0 for every torsion element t\in G . Moreover, there are just finitely many conjugacy classes of torsion elements by [ HZ08 , Lemma 8] . Therefore one obtains an alternative proof of C for torsion words w=x^{k} , k\in\mathbb{N} . 5.2. Torsion in compact p -adic analytic groups We now focus on torsion identities in compact p -adic analytic groups. We start by introducing some notions from [ Ser65 ] that will be used throughout this section. Given a K -analytic group G , we denote by \mathcal{L}(G) its associated K -Lie algebra, defined as the tangent space at the identity 1_{G} . If f\colon G_{1}\to G_{2} is an analytic map of analytic groups that preserves the identity, we denote by df\colon\mathcal{L}(G_{1})\to\mathcal{L}(G_{2}) its differential. If f is also a group homomorphism, we have that df is a Lie algebra homomorphism. When K=\mathbb{Q}_{p} , it is possible to describe \mathcal{L}(G) in terms of open uniform pro- p subgroups of G as in [ DdSMS99 , Ch. 9] . We recall that a pro- p group H is uniform if it is finitely generated, powerful and torsion-free. Moreover, a uniform pro- p group H can be endowed with a \mathbb{Z}_{p} -Lie algebra structure and we use the notation L_{H} when we refer to H with this structure. If G is a p -adic analytic group, then \mathcal{L}(G)\simeq\mathbb{Q}_{p}\otimes L_{H} for every open uniform pro- p subgroup H . Moreover, if f\colon G_{1}\to G_{2} is an homomorphism of p -adic analytic groups, then df=1\otimes f_{0} where f_{0} is the restriction of f to a suitable open uniform pro- p subgroup of G_{1} . Indeed, since \mathcal{L}(G_{i}) does not depend on the choice of open uniform subgroup, one may take an open uniform subgroup H_{2} of G_{2} and choose an open uniform subgroup H_{1} in f^{-1}(H_{2}) , in such a way that f_{0}=f|_{H_{1}} allows us to properly define df . Given an automorphism \alpha\in\mathrm{Aut}(G) , we say that \alpha is fixed-point-free on X\subseteq G if \alpha does not fix any x\in X\smallsetminus{1} . We say that \alpha is uniformly fixed-point-free if it is fixed-point-free on every open uniform pro- p subgroup. Note that for \alpha\in\mathrm{Aut}(G) the differential d\alpha\colon\mathcal{L}(G)\to\mathcal{L}(G) is a \mathbb{Q}_{p} -Lie algebra automorphism. Lemma 5.6 . Let G be a p -adic analytic group and \alpha\in\mathrm{Aut}(G) . The following are equivalent: (i) If \alpha is fixed-point-free on an open uniform pro- p subgroup H ; (ii) \alpha is uniformly fixed-point-free; (iii) d\alpha is fixed-point-free. Proof. Suppose that \alpha fixes a non-trivial element of a uniform open pro- p subgroup K of G . Since K is torsion-free and H\cap K is of finite index in K , there is a natural number n such that 1\neq x^{n}\in H\cap K . This is a contradiction since \alpha is fixed-point-free on H . Therefore \alpha is uniformly fixed-point-free. Now, let d\alpha=1\otimes\alpha_{0} where \alpha_{0} is the restriction of \alpha to an open uniform pro- p subgroup H_{0} . For h\in H_{0} and \lambda\in\mathbb{Q}_{p} , we have d\alpha(\lambda\otimes h)=\lambda\otimes\alpha(h) . If d\alpha is fixed-point-free , then \alpha(h)\neq h for every 1\neq h\in H_{0} . Therefore \alpha is uniformly fixed-point-free by (i) . On the other hand, if \alpha is uniformly fixed-point-free, then \lambda\otimes\alpha(h)\neq\lambda\otimes h for every \lambda\in\mathbb{Q}_{p} and h\in H_{0} . Therefore, d\alpha is fixed-point-free. ∎ Given an element g of a group G , let c_{g}\in\rm{Aut}(G) be the conjugation by g . We are ready to give a characterisation of torsion probabilistic identities in compact p -adic analytic groups. Theorem 5.7 . The following are equivalent for a compact p -adic analytic group G : (i) \mu(\rm{Tor}(G))>0 ; (ii) there is t\in\rm{Tor}(G) such that c_{t} is a uniformly fixed-point-free automorphism of G ; (iii) there is t\in\rm{Tor}(G) such that dc_{t} is a fixed-point-free automorphism of \mathcal{L}(G) . Proof. By Lemma 5.6 , it is clear that (ii) is equivalent to (iii) . Suppose P(G,x^{m})>0 for some m\geq 1 . Then there exists an open uniform pro- p subgroup H\unlhd_{o}G and t\in{\rm{Tor}}(G) such that every element in Ht has order m by A . If u\in C_{H}(t) , then 1=(ut)^{m}=uu^{t}\cdots u^{t^{m-1}}=u^{m} implies u=1 , since H is torsion-free. So c_{t} is a uniformly fixed-point-free automorphism of G . Now, suppose that there is a torsion element t , |t|=m , such that the conjugation c_{t} is uniformly fixed-point-free. Consider the analytic map \displaystyle f\colon G \displaystyle\to G \displaystyle g \displaystyle\mapsto g^{t}g^{-1} and note that every element of the form tf(g)=t^{g^{-1}} has order m . Since the differential df=dc_{t}-\mathrm{id} has trivial kernel by our assumption of c_{t} , we have that df\colon\mathcal{L}(G)\to\mathcal{L}(G) is an isomorphism. By [ Ser65 , Pt. 2, Ch. III.9, Theorem 2] , f is a local isomorphism at 1_{G} , therefore its image contains an open subgroup H . In particular, the word x^{m} is satisfied on the coset tH . ∎ As an application, we give a different proof of B in the compact p -adic analytic case for torsion words. Corollary 5.8 . Let G be a compact p -adic analytic group that is not virtually solvable. Then \mu(\mathrm{Tor}(G))=0 . Proof. Suppose that the set of torsion elements of G has positive Haar measure. By 5.7 and [ Khu93 , Theorem 4.3.1] , \mathcal{L}(G) is solvable since it admits a finite order fixed-point-free automorphism. In particular, if H is an open uniform pro- p subgroup of G , then L_{H} is solvable as a Lie subalgebra of \mathcal{L}(G) . By [ DdSMS99 , Corollary 7.16] , H is solvable. This is a contradiction since G is not virtually solvable. ∎ In particular, for a p -adic analytic pro- p group G , we can have \mu(\mathrm{Tor}(G))>0 only if G is solvable. In the virtually nilpotent case, we can give explicit examples of groups satisfying (and not satisfying) probabilistic torsion identities. We make essential use of A , that is, that a probabilistic torsion identity is also a coset identity. Example 5.9 . (1) Let U=\mathbb{Z}_{3}^{2} and consider t\in\mathrm{GL}_{2}(\mathbb{Z}_{3}) to be the matrix \begin{pmatrix}0&-1\\ 1&-1\end{pmatrix}. Notice that t has order 3 and its minimal polynomial is x^{2}+x+1 , whose roots are the primitive third roots of unity. Let G=U\rtimes\langle t\rangle . By standard calculations, one can check that the every element in the coset Ut has order 3. (2) Let U=\mathbb{Z}_{5}^{5} and consider t\in\mathrm{GL}_{5}(\mathbb{Z}_{5}) to be the matrix \begin{pmatrix}0&0&0&-1&0\\ 1&0&0&-1&0\\ 0&1&0&-1&0\\ 0&0&1&-1&0\\ 0&0&0&0&1\end{pmatrix}. Let G=U\rtimes\langle t\rangle . If G satisfies any probabilistic torsion identity, it must be x^{5} . Indeed, a non-trivial torsion element in G must be of the form ut^{j} , with u\in U\smallsetminus\{1\} and 1\leq j\leq 4 . Since x=(ut^{j})^{5}=uu^{t}u^{t^{2}}u^{t^{3}}u^{t^{4}} is in U and U is torsion-free, we must have x=1 . For any H\leq_{o}G , we can find in H a normal subgroup of the form 5^{n_{1}}\mathbb{Z}_{5}\times\cdots\times 5^{n_{2}}\mathbb{Z}_{5} , with each n_{i} a non-negative integer, so Ht^{j} will have an element of infinite order for any 1\leq j\leq 4 . It follows that x^{5} is not a coset identity for G and thus not a probabilistic one. (3) Let p\geq 3 and let \zeta be a primitive p -th root of unity. We consider the ring R=\mathbb{Z}_{p}[\zeta] . Let U=\left\{\begin{pmatrix}1&x&z\\ 0&1&y\\ 0&0&1\end{pmatrix}\mid x,y,z\in R\right\} be the Heisenberg group over R and we define t=\begin{pmatrix}1&0&0\\ 0&\zeta&0\\ 0&0&\zeta^{2}\end{pmatrix} . Then G=\langle t,U\rangle splits as a semidirect product G=U\rtimes\langle t\rangle . Since conjugation with t provides a fixed-point-free automorphism of U , x^{p} is a coset identity for G . We remark that all compact p -adic analytic groups G which satisfy a torsion probabilistic identity of the form x^{p} are virtually nilpotent. Indeed, by 5.7 , there exists an open uniform pro- p subgroup U of G and t\in\mathrm{Tor}(G) such that C_{U}(t)=1 and consequently dc_{t} is a fixed-point-free automorphism of \mathcal{L}(G) of order p . Then \mathcal{L}(G) is nilpotent by [ Zha89 , Theorem 3] and thus so is U . As a final application of A , we show that solvable just-infinite p -adic analytic groups are either torsion-free or they satisfy a torsion probabilistic identity. Corollary 5.10 . Let G be a just-infinite p -adic analytic pro- p group with torsion. Then either G satisfies a torsion probabilistic identity or it is randomly free. Proof. If G is not randomly free, then it is solvable by A . In that case, it is virtually abelian and we can choose a maximal abelian open uniform subgroup U of G . Being maximal, we have that C_{U}(t)=1 for any t\in{\rm{Tor}}(G) and thus G satisfies a torsion probabilistic identity by 5.7 . ∎