Algorithmically complex residually finite groups

Bulletin of Mathematical Sciences, Mar 2017

We construct the first examples of algorithmically complex finitely presented residually finite groups and the first examples of finitely presented residually finite groups with arbitrarily large (recursive) Dehn functions, and arbitrarily large depth functions. The groups are solvable of class 3.

A PDF file should load here. If you do not see its contents the file may be temporarily unavailable at the journal website or you do not have a PDF plug-in installed and enabled in your browser.

Alternatively, you can download the file locally and open with any standalone PDF reader:

https://link.springer.com/content/pdf/10.1007%2Fs13373-017-0103-z.pdf

Algorithmically complex residually finite groups

Bull. Math. Sci. Algorithmically complex residually finite groups Olga Kharlampovich 0 1 2 Alexei Myasnikov 0 1 2 Mark Sapir 0 1 2 B Mark Sapir 0 1 2 Alexei Myasnikov 0 1 2 Communicated by Efim Zelmanov. 0 Department of Mathematics, Vanderbilt University , Nashville, TN 37240 , USA 1 Stevens Institute of Technology , Hoboken, NJ 07030 , USA 2 Department of Mathematics and Statistics, Hunter College, City University of New York , New York, NY 10065 , USA We construct the first examples of algorithmically complex finitely presented residually finite groups and the first examples of finitely presented residually finite groups with arbitrarily large (recursive) Dehn functions, and arbitrarily large depth functions. The groups are solvable of class 3. 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310 1.1 The problem and previous approaches for a solution . . . . . . . . . . . . . . . . . . . . . . 310 1.2 The “yes” and “no” parts of the McKinsey algorithm . . . . . . . . . . . . . . . . . . . . . 313 O. Kharlampovich: Partially supported by NSF Grant DMS-0700811, A. Myasnikov: Partially supported by NSF Grants DMS-0700811 and DMS-0914773. M. Sapir: Partially supported by NSF Grant DMS-1500180 and by BSF (USA-Israel) Grant 2010295. Word problem; Depth function; Dehn function; Minsky machine Contents 1.3 The time function of the algorithm Ayes: the Dehn function . . . . . . . . . . . . . . . . . . 315 1.4 The time function of the algorithm Ano: the depth function . . . . . . . . . . . . . . . . . . 315 1.5 Methods of proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316 1.6 Structure of the paper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316 2 Turing machines and Minsky machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317 2.1 Turing machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317 2.2 Universally halting Turing machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319 2.3 Minsky machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322 3 Simulation of Minsky machines by semigroups . . . . . . . . . . . . . . . . . . . . . . . . . . . 324 3.1 The construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324 3.2 Residually finite finitely presented semigroups . . . . . . . . . . . . . . . . . . . . . . . . . 326 3.3 Residually finite semigroups with large depth function . . . . . . . . . . . . . . . . . . . . . 328 4 Simulation of Minsky machines in solvable groups . . . . . . . . . . . . . . . . . . . . . . . . 333 4.1 The construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333 4.2 A residually finite finitely presented group with large depth function . . . . . . . . . . . . . 346 5 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347 5.1 Universal theories of sets of finite groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347 5.2 Distortion of pro-finitely closed subgroups of finitely presented groups . . . . . . . . . . . . 349 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350 1 Introduction 1.1 The problem and previous approaches for a solution It is well known that finitely presented residually finite algebraic systems of finite signature (say, semigroups or groups) are much simpler algorithmically than arbitrary finitely presented algebraic systems. For example, the word problem in every such algebraic system is decidable. In this paper we discuss the question “how complicated finitely presented residually finite algebraic systems can be in the cases of groups and semigroups?”. The most “common” residually finite groups and semigroups, are linear (say, over a field). These groups and semigroups are algorithmically “tame”: the word problem there is decidable in polynomial time and even log-space [ 34 ]. The Dehn function is a well-known indicator of complexity of the word problems: the smaller the Dehn function the easier the word problem. The converse implication does not hold, however. One can construct groups with decidable word problem and very large Dehn functions [ 39,59 ]. However, no examples of finitely presented residually finite groups or even semigroups with superexponential Dehn function are known. Thus one of the main open problems in this area is how large could the Dehn function of a residually finite finitely presented group or semigroup be. The question for groups was known since early 90s. It was open for so long because all known methods to construct algorithmically hard groups produced either non-residually finite groups or groups where the question about their residual finiteness is very difficult. Not much is known even for linear groups (note that Gersten asked [ 19,20 ] if there exists a uniform upper bound for Dehn functions of linear groups). Let us briefly discuss the previous attempts to solve the problem and the reasons why these methods did not work. 1.1.1 Method 1. Known groups with large Dehn functions One could hope that some of the known finitely presented groups with very large Dehn function may turn out to be residually finite, which would shed some light on how to produce residually finite groups with even larger Dehn functions. Unfortunately, this is not the case, all these groups are non-residually finite. For example, the Dehn function of the one relator group G(1,2) = a, b | b−1a−1bab−1ab = a2 introduced by Baumslag in [2] is bigger than any iterated exponent (see Gersten [ 18 ]). Platonov [50] proved that it is equivalent to the function exp(log 2n)(1), where exp(m)(x ) is the function defined by exp(0)(n) = n and exp(k+1)(n) = exp(exp(k)(n))). However, G(1,2) is not residually finite (and in fact has very few finite quotients) [2]. Furthermore, the word problem in G(1,2) is decidable in polynomial time [ 41 ]. 1.1.2 Method 2. Using subgroups with very large distortion Consider a finitely presented group G and a “badly” distorted finitely generated subgroup H . Let T = G, t | t −1ht = h (h ∈ H ) be the HNN extension of G where the free letter t centralizes H . It was noticed by Bridson and Häfliger [11, Theorem 6.20.III] that the Dehn function of T is at least as large as the distortion function of H in G. The following result puts some limitations on this method of constructing complicated residually finite groups. Lemma 1.1 If the group T is residually finite then H is closed in the pro-finite topology of G. Proof In the notation above, suppose T is residually finite and u ∈ G\H . Then w = [u, t ] = 1 in T (by the standard properties of HNN extensions). Hence there exists a homomorphism φ from T onto a finite group Tw such that φ (w) = 1. But this implies φ (u) ∈/ φ (H ) (since every element of φ (H ) commutes with φ (t )). Hence there exists a normal subgroup of finite index N < G such that u does not belong to N H . In other words H must be closed in the pro-finite topology of G. Consider the following typical examples of residually finite groups G with highly distorted subgroups H . The first one is Wise’s version [64] of Rips’ construction [53] which for every finitely presented group Q gives a finitely presented residually finite small cancellation group G with a short exact sequence 1 → N → G → Q → 1 where N finitely generated. It is easy to see that the distortion function of N in G is at least as large as the Dehn function in Q, so choosing Q properly one can get a finitely presented residually finite group G with a highly distorted subgroup N . Now, the subgroup N is normal in the HNN extension T . So it is closed in the pro-finite topology of T only if Q = G/N is residually finite. By Lemma 1.1, T can be residually finite only if Q is residually finite. In other words to construct a complicated finitely presented residually finite group T one has to have the initial group Q complicated, finitely presented and residually finite as well. The second example is the standard Mikhailova construction. In this case highly distorted subgroups of the direct product of two free groups F2 × F2 can be obtained as equalizers of two homomorphisms φ1 : E1 → M and φ2 : E2 → M where E1, E2 are finitely generated subgroups of F2 and M is finitely presented (see Sect. 5.2 below). But by Lemma 5.3 below the equalizer is closed in the pro-finite topology only if M is residually finite. Thus as in the previous example, in order to construct an algorithmically complex residually finite finitely presented group using Mikhailova’s construction and the HNN extensions as above, one needs to have already a finitely presented residually finite algorithmically complex group M . The third example is Cohen’s [12] construction of highly distorted subgroups employing the modular machines. One can also prove that in that construction the subgroup will be pro-finitely closed only if the modular machine is very easy. One can also try to use the hydra groups [ 10,14 ] to construct HNN extensions as above with Dehn functions bigger than any prescribed Ackermann function. The question of whether these groups are residually finite was open when the first version of this paper was written, and is now answered in negative in [49]. 1.1.3 Method 3. Boone–Novikov constructions One of the standard ways to produce algorithmically complicated groups is by simulating Turing machines using free constructions (HNN extensions and amalgamated products) which goes back to the seminal papers by Boone and Novikov (see, for example, [54]). There are currently many versions of that construction (for a recent survey see [57]). But in fact, it can be shown that for each known version of the proof of Boone–Novikov theorem using free constructions, even for easy Turing machines the corresponding group is non-residually finite. This is, for instance, the main idea of the example in [ 31 ]. Here is an even easier example. Let G = G(M ) be a group constructed by any of these constructions. Then for every input word u of the Turing machine M there exists a word w = w(u) obtained by inserting some copies of u in w(∅) (here ∅ denotes the empty word), so that u is accepted by M if and only if w(u) = 1 in G. Now consider M that accepts a word an if and only if n = 0 (that machine is actually one of the basic building blocks in [59]). Then w(an) = 1 in G if and only if n = 0. Suppose that there exists a homomorphism φ onto a finite group H that separates w(∅) = w(a0) from 1. Then φ (a) has finite order, say, s, in H . Therefore φ (w(∅)) = φ (w(a0)) = φ (w(as )) = φ (1) = 1, a contradiction. Hence G(M ) is not residually finite. 1.1.4 Method 4. Residually finite groups obtained by free constructions In general, the question about residual finiteness of free constructions is very difficult. Currently there are only two large classes of groups where the question was settled: these are ascending HNN extensions of free groups [6] and certain groups acting “nicely” on CAT(0)-cubical complexes including small cancelation groups (see the recent work of Wise, for example, [ 25 ] and references therein). All these groups have easy word problem and uniformly bounded Dehn functions. The reason for the lack of more examples is that groups obtained by free constructions from “nice” groups contain a lot of extra elements and it is not at all clear how to separate these elements from 1 by homomorphisms onto finite groups. In the two cases when it could be done, it was possible to reformulate the problem in the language of algebraic geometry and geometric topology, respectively. 1.1.5 Method 5. Other groups with complicated word problem There are several other constructions of groups which simulate various algorithmic problems including undecidable ones, but each of these also always either produce non-residually finite groups or groups with simple word problem. For example, the group in [ 37 ] is based on the R. Thompson group V which is infinite and simple (hence not residually finite). 1.1.6 The main results of the paper. In this paper, we construct finitely presented residually finite semigroups and groups with arbitrarily complex word problem, and also easy word problem but arbitrarily large (of course recursive) Dehn function, and arbitrarily large (recursive) depth function. We also give applications of these results to the questions about decidability of the universal theory of finite solvable groups and to distortion of pro-finitely closed subgroups of residually finite groups. 1.1.7 What next? We expect the approach used in this paper to be useful in solving other problems that are still open. For example, the residually finite version of the Higman embedding theorem [ 24 ] would be very desirable. It is known [ 15,21,38 ] that a finitely generated recursively presented residually finite group may have undecidable word problem, and hence cannot be embedded into a finitely presented residually finite group. But it is not known whether undecidability of the word problem is the only obstacle for such an embedding. Thus it would be very interesting to find out whether every finitely generated residually finite group with decidable word problem embeds into a finitely presented residually finite group. Note that usually a version of Boone–Novikov theorem precedes a version of Higman theorem, hence we can consider this paper as a step toward the residually finite version of Higman’s theorem. 1.2 The “yes” and “no” parts of the McKinsey algorithm One of the initial motivations for studying residually finite groups, semigroups and other algebraic structures was McKinsey’s algorithm solving the word problem in finitely presented residually finite algebraic structures. Let G = X | R be a residually finite finitely presented algebraic structure of finite signature (say, groups, semigroups, rings, etc.) Let us recall McKinsey’s algorithm solving the word problem in G (see [ 36,40 ]). Let F (X ) be the free algebraic structure of signature freely generated by X . To solve the word problem in G one runs in parallel two separate algorithms Ayes and Ano, such that starting with a given pair of elements w, w ∈ F (X ) Ayes stops if and only if w = W in G and Ano stops if and only if w = w in G. The algorithm Ayes enumerates one by one all consequences of the defining relations R and waits until w = w appears in the list. The algorithm Ano enumerates all homomorphisms φ1, φ2, . . . , of G into finite algebraic structures of signature and waits until φi (w) = φi (w ). Note that although Ayes and Ano enumerate complimentary sets of elements of F (X ) × F (X ), these algorithms are quite different and their complexity functions can be very different too. To estimate the complexity of algorithms and algorithmic problems, recall the basic definitions of the complexity theory. Let S be a set of objects (words, numbers, etc.) equipped with a size function S → N so that there are only finitely many objects in S of any given size. For every algorithm A computing some partial function φ : S → N, by TA we denote the time function of A which is defined as the maximal running time of A on inputs of size at most n from the domain of φ (thus we simply ignore inputs not from the domain of φ). The time complexity of an algorithmic problem, i.e., the membership problem in a subset X ⊆ S is the minimum of TA for all algorithms A solving the problem, i.e., computing the indicator function of X . It is not that easy to define what “minimum of a set of functions” means. But we can define a pseudo-order on the set of functions as follows. For two functions f, g we write f g ( f is smaller than g) is there exists a constant C such that for any n, f (n) ≤ Cg(C n) + C n + C. (The functions are equivalent if f g and g f.) We do not need to address the question whether every set of functions has a minimum with respect to this order. We can certainly say whether the time complexity is polynomial (exponential, etc.). Let now G be a finitely presented residually finite group. Although it is universally assumed that Ayes and Ano are very slow in general, there were no examples of groups G for which these algorithms were actually very slow. Moreover it was not known if for some universal recursive function f (n) the time complexity of the word problem in any finitely presented residually finite group (or semigroup) does not exceed f (n). In the case of finitely presented linear groups it is well known that the word problem can be solved in deterministic polynomial time [ 34,61 ]. This applies to most finitely presented groups (where “most” means “with overwhelming probability” in one of several probabilistic models): recent results of Agol [1] and Ollivier and Wise [45] together with the older result of Olshanskii [46] imply that most finitely presented groups are linear (even over Z). One of our main results is the following theorem (an immediate corollary of Theorem 4.21 below): Theorem For every recursive set of natural numbers X there exists a finitely presented residually finite solvable group G of class 3 such that the word problem in G is as hard as the membership problem in X . Remark 1.2 Note that if we replace “finitely presented” assumption by “recursively presented”, then residually finite groups are known to be very complicated. As we have mentioned before, recursively presented finitely generated residually finite groups may have undecidable word problem [ 15,21,38 ]. See also the survey [ 16 ] where it is shown how to construct complicated residually finite groups using the method of Golod–Shafarevich. Note also that although our groups are not linear, they are (Abelian of prime exponent)-by-linear since they are solvable of class 3 with the second derived subgroup Abelian of prime exponent (one can assume that the exponent is equal to any prime number, say, 2). 1.3 The time function of the algorithm Ayes: the Dehn function In [ 39 ] Madlener and Otto constructed finitely presented groups with arbitrarily large Dehn functions. For residually finite groups, the situation is different. Nilpotent groups are examples of residually finite groups with arbitrarily high polynomial Dehn function [4]. The Baumslag–Solitar groups x , y | x y = x k , k ≥ 2, are examples of residually finite (even linear) groups with exponential Dehn function. No examples of residually finite groups with bigger Dehn functions were known. This gap is filled by the following theorem (see Theorem 4.21). For every recursive function g(n) there exists a finitely presented residually finite solvable group G of class 3 such that the Dehn function of G is bigger than g(n). In addition we can assume that the time complexity of the word problem is polynomial. 1.4 The time function of the algorithm Ano: the depth function The time function of the algorithm Ano can be estimated in terms of the depth function introduced by Bou-Rabee [9]. Recall that if G = X is a finitely generated residually finite group or semigroup, the depth function ρG (n) is the smallest function such that every two words w =G w of length at most n are separated by a homomorphism to a finite group (semigroup) H with |H | ≤ ρG (n). That function does not depend on the choice of finite generating set X (up to the natural equivalence). It is easy to see that for every finitely generated linear group or semigroup G, ρG is at most polynomial. Since finitely generated metabelian groups are subgroups of direct products of linear groups [62] the depth function of every finitely generated metabelian group is at most polynomial. By the recent result of Agol [1] based on the earlier results of Wise [63], every small cancelation group is a subgroup of a Right Angled Artin group, hence linear and has polynomial depth function. In fact one can have much smaller bounds for many linear groups. For example, for the free group F2, n3 by [9]. A lower bound for the depth function for a free group is equivalent ρF2 (n) 2 to n 3 by a result of Kassabov and Matucci [ 26 ]. There are some finitely presented groups for which the depth function is unknown and very interesting. For example the ascending HNN extensions of free groups are known to be residually finite and even virtually residually nilpotent (proved by Borisov and the third author [6,7]) but the only upper bound one can deduce from the proof is exponential. Although many of these groups have small cancelation presentations and so are covered by the results from [1], there are some groups of this kind for which the depth function is not known. For finitely generated infinitely presented groups (even amenable ones) the situation is much clearer. Using the method of Kassabov and Nikolov [ 27 ] and the result of Nikolov and Segal [44] one can construct a finitely generated residually finite group with arbitrary large recursive depth function. In this paper, we show that a similar result holds for finitely presented solvable of class 3 groups (see Theorem 4.22). For every recursive function f one can construct a residually finite finitely presented solvable of class 3 group with depth function greater than f . In addition, again, one can assume that the time complexity of the word problem in G is polynomial. 1.5 Methods of proof As we have shown above (see Sect. 1.1.3) all versions of the Boone–Novikov construction ([ 12,37,54,59 ]) do not produce complicated residually finite groups. Instead, we simulate Minsky machines. We first simulate Minsky machines in semigroups and then “embed” these semigroups into solvable groups of class 3. The first simulation of Minsky machines in semigroups were studied by Gurevich [ 23 ] (see also [55,58]). Simulations of Minsky machines in groups, including solvable groups could be found in [60] and [ 28 ] (see also [ 32 ]). In this paper we closely follow the construction by the first author from her unpublished thesis [ 29 ]. Of course that construction also often leads to non-residually finite groups. But it turned out that the difficulty can be overcome by modifying the Minsky machine first. In this paper, we use the fact that every Turing machine recognizing a recursive set is equivalent to a sym-universally halting Minsky machine, i.e., Minsky machine whose symmetrization halts on every non-accepted configuration (see Theorem 2.7). 1.6 Structure of the paper The paper is organized as follows. Section 2 contains preliminary results about Turing and Minsky machines that are needed further. In Sect. 3, we simulate sym-universally halting Minsky machines in residually finite finitely presented semigroups and prove the analogs of the above theorems for semigroups. The advantage of using Minsky machines instead of the general Turing machines is that the resulting semigroups are much “smaller”, and all non-zero elements of these semigroups are basically subwords of words corresponding to configurations of the Minsky machines. In the construction, we use the ideas from the papers of the third author [55,58] who proved that these semigroups are minimal in the following sense: the varieties generated by these semigroups are minimal varieties of semigroups containing finitely presented semigroups with undecidable word problem (this leads to a complete description of varieties with decidable word problem where periodic groups are locally finite [55]). In Sect. 4 we use semigroups from Sect. 3 to construct the groups with large Dehn and depth functions. The idea of such a construction came from the paper [ 28 ] of the first author solving a problem formulated by Adyan [ 33 ] by constructing a finitely presented group with undecidable word problem and satisfying a non-trivial identity. We use the simplification of that construction from the unpublished dissertation of the first author [ 29 ]. It turned out that if we start with residually finite semigroups from Sect. 3, we often get residually finite finitely presented groups whose Dehn and depth functions resemble the corresponding functions of the semigroups we start with. Finally, Sect. 5 contains two applications of our main results and methods. In Sect. 5.1, we strengthen the well-known result of Slobodskoi about undecidability of the universal theory of finite groups by showing that the universal theory of any set of finite groups that contains all finite solvable groups of class 3 is undecidable (see Theorem 5.1). This gives the first proper variety of groups where the set of all finite groups has undecidable universal theory. In Sect. 5.2, we construct pro-finitely closed subgroups of the direct product of two free groups with arbitrary large time complexity of the membership problem, distortion function and ther relative depth function (defined there). 2 Turing machines and Minsky machines 2.1 Turing machines In this paper, we shall consider several types of machines. A machine M in general has an alphabet and a set of words in that alphabet called configurations. It also has a finite set of commands. Each command is a partial injective transformation of the set of configurations. A machine is called deterministic if the domains of its commands are disjoint. A machine usually has a distinguished configuration sacc called the accept configuration and a set I = I (M ) of input configurations which recursively enumerate words in some finite alphabet A. A machine halts if none of the commands is applicable to the configuration. We always assume that a machine halts on every accept configuration but it may halt on non-accept configurations too. A computation of M is a finite or infinite sequence of configurations and commands from P: w1 −θ→1 w2 −θ→2 . . . −θ→l wl+1, . . . such that θi (wi ) = wi+1 for every i = 1, . . . , l, . . . . If the computation is finite and wl+1 is the last configuration, then l is called the length of the computation, and we say that wl+1 is obtained from w1 by applying . A configuration is called accepted by M if there exists a computation connecting that configuration with sacc (that computation is called accepting). The time function TM (n) of M is the minimal function such that every accepted configuration of length ≤ n has an accepting computation of length ≤ TM (n). Let us give a definition of a Turing machine (see [59]). A Turing machine M with k tapes consists of hardware (the tape alphabet A = ik=1 Ai , and the state alphabet Q = iK=1 Qi )1 and program P (a list of commands, defined below). A configuration of a Turing machine M is a word α1u1q1v1ω1 α2u2q2v2ω2 · · · αK u K qK vK ωK (we included spaces to make the word more readable) where ui , vi are words in Ai , qi ∈ Qi and αi , ωi are special symbols (not from A ∪ Q). This machine has k tapes. For every configuration c the content of tape number i is the subword αi ui qi vi ωi . A command simultaneously replaces subwords ai qi bi by words a q b i i i where qi , qi ∈ Qi , ai , ai , are either letters from Ai ∪ {αi } or empty, bi , bi are either letters from Ai ∪ {ωi } or empty. A command cannot insert or erase αi or ωi , so if, say, ai = αi , then ai = αi . Note that with every command θ one can consider the inverse command θ −1 which undoes what θ does. For the Turing machine we choose stop states qi0 in each Qi , then a configuration w is accepted if there exists a computation starting with w and ending with a configuration where all state symbols are qi0 and all tapes are empty (which is the accept configuration for the Turing machine). Thus for a k-tape Turing machine sacc is equal 0 0 to α1q1 ω1 · · · αk qk ωk . Also we choose start states qi1 in each Qi . Then an input configuration corresponding to a word u over A1 is a configuration inp(u) of the form α1uq11ω1 α2q21ω2 · · · αK qK1 ωK . Thus the input set I consists of all these words. We say that a word u over A1 is accepted by M if the configuration inp(u) is accepted. The set of all words accepted by M is called the language accepted by M . For every machine M , the machine Sym(M ) is made from M by adding the inverses of all commands of M . Two configurations w, w are called equivalent, written w ≡M w , if there exists a computation of Sym(M ) connecting these configurations. Clearly, ≡M is an equivalence relation. For example, every two accepted input configurations are equivalent, because both are equivalent to sacc. The following general lemma is easy but useful. Lemma 2.1 Suppose that M is deterministic. Then (a) Any reduced computation of Sym(M ) is a concatenation of two (possibly empty) parts 1 2−1, both 1 and 2 uses only commands of M . (b) Two configurations w, w of M are equivalent if and only if there exist two computations of M connecting w, w with the same configuration w of M . 1 denotes disjoint union. Proof Part (a) of the lemma follows from the fact that in any reduced computation of Sym(M ) inverses of commands of M cannot be followed by commands of M (since M is deterministic). For Part (b), if w is equivalent to w , then by Part (a) there is a computation of the form 1 2−1 connecting w and w . Then applying 1 to w, and 2 to w produces the same configuration w . Definition 2.2 (a) We say that an algorithmic problem A is as hard as an algorithmic problem B if for any decision algorithm for A which solves the problem in time TA there exists an algorithm for B that solves it in time TA . (b) We say that a language Y polynomially reduces to a language X if there exists a deterministic Turing machine C checking membership in Y which uses an oracle checking membership in X and runs in polynomial time (in terms of the length of the word being checked. (c) We say that languages X and Y are polynomially equivalent if there are polynomial reductions of X to Y and vice versa. (d) We say that a machine M polynomially reduces (resp. is polynomially equivalent) to a machine M if the configuration equivalence problem of M polynomially reduces (resp. is polynomiall equivalent) to the configuration equivalence problem for M . 2.2 Universally halting Turing machines A deterministic machine M is called universally halting if it does not have infinitely long computations (see [8]). We say that a computation of Sym(M ) is reduced if no command is followed by its inverse. We call a deterministic machine M symuniversally halting if Sym(M ) does not have infinitely long reduced computations that start at a non-accepted configuration. It is proved in [13] that for every recursive set X of natural numbers, that is accepted by a deterministic Turing machine M there exists a universally halting deterministic Turing machine M with one tape accepting X . From the construction, it is clear that M polynomially reducible to M . One can also convert a sym-universally halting Turing machine into a 1-tape symuniversally halting Turing machine: Lemma 2.3 Let M be a deterministic sym-universally halting Turing machine recognizing a language X . Then there exists a one-tape deterministic sym-universally halting Turing machine M recognizing X and polynomially equivalent to M . Moreover there exists a map φ from the set C (M ) of configurations of M to the set C (M ) of configurations of M such that (1) For every word u in the alphabet of X , φ (inpM (u)) = inpM (u) (2) For every c ∈ C (M ), |φ (c)| = O(|c|) (3) φ (c) ≡M φ (c ) if and only if c ≡M c (4) If a command θ of M takes c to c , then there exists a computation of M of length at most O(|c|) that takes φ (c) to φ (c ). Proof The proof is basically by inspection of the proof from [48, Theorem 2.1]. Recall the way to convert a k-tape Turing machine M into a 1-tape Turing machine M . The tape alphabet of M consists of all letters occurring in the configurations of M . A configuration of M is a word αc1qc2ω where c = c1c2 is a configuration of M or differs from a configuration of M by at most two letters: the left and right neighbor of some state letter. The map φ takes each configuration c to which a command of M is applicable to the configuration αcqθ ω where qθ is a state letter of M corresponding to the command θ that is applicable to c (note that since M is deterministic, θ is determined by c). If no command of M is applicable to c, we set φ (c) = αcqω where q is a distinuished state letter of M . A command θ of M that substitutes ai qi bi by ai qi bi , i = 1, . . . , k, is simulated as follows: the letter qθ moves from right to left, and every time it meets qi , it checks if it is a part of the subword ai qi bi , and if so, replaces it by a q b . After all these i i i substitutions the letter qθ returns to the right end of the configuration (next to the letter α) and becomes ready to simulate the next command of M or becomes the distinguished state letter q (for more details see [48]). It is easy to check (using Lemma 2.1) that the new machine is polynomially equivalent to the old one and properties (1)–(4) hold. It is also easy to check that if the original machine is sym-universally halting, the new one is also sym-universally halting. Theorem 2.4 For every recursive language X there exists a deterministic symuniversally halting Turing machine M with one tape recognizing X . Moreover if M is any deterministic Turing machine recognizing X then we can additionally assume that M polynomially reduces to M . Proof Let M be a deterministic universally halting Turing machine with k tapes recognizing X . Consider the new Turing machine M constructed as follows. M has one more tape than M , called the history tape. The alphabet A of this tape is in one-to-one correspondence with the set of commands P of M : A = {[θ ], θ ∈ P}. An input configuration of M does not have letters from A and its subword written on the first k tapes is an input configuration of M . With every command θ of M we associate a command θ of M . It does what θ would do on the first k tapes of M and inserts [θ ] on the history tape of M next to the right of qk+1. After the first k tapes of M form the accept configuration sacc(M ), the machine erases letters from the tape alphabet A on the history tape and halts, producing the accept configuration sacc(M ) of M (thus sacc(M ) = s0αk+1qk0+1ωk+1). Let P be the program of M . We shall modify M further to obtain a new (k + 1)tape Turing machine M . It has the same tape alphabets as M and all state letters of M are also state letters of M . The input and accept configurations are also the same. The program P of M contains a copy P˜ of P (the set of the main commands) and some new commands. After each main command θ˜ of P˜ which does the same as the corresponding command θ of M , but changes the state letters to state letters which are not in M , M executes the history written on the history tape backward, without erasing the history tape. It just scans the history tape from left to right, reading the symbols written there one by one and executing on the first k tapes the inverses of the commands written on the history tape. If at the end of the scanning the history tape, the word written on the first k tapes is an input configuration of M , M executes on the first k tapes the history written on the history tape in the natural order, scanning the history tape from right to left. After that M changes the state letters to what θ would do, and is ready to execute the next main command. We do not give precise definition of the program of M because it is obvious on the one hand and long on the other hand. The machine M is deterministic and universally halting. Moreover for every input configuration c of M , M accepts c if and only if c is accepted by M , hence if and only if c corresponds to a word from X . Note that since M is deterministic, and no commands of it are applicable to the accept state, M accepts the same language as M . Let c be a configuration of M that is not accepted by M . By Lemma 2.1(a) every reduced computation of Sym(M ) starting at c is a concatenation of a computation of M followed by a computation of the machine M −1 obtained from M by replacing every command with its inverse. Since M is universally halting, there are only finitely many computations of M starting with c. Claim 1 There are finitely many computations of M −1 starting with any non-accepted configuration c. Equivalently, there are only finitely many computations of M ending with c. Indeed, since c is not accepted, none of the tape letters on the history tape is erased during any computation ending in c. Therefore by the definition of M every computation ending at c and having length ≥|c| must arrive at a configuration c = c1c2 where c1 is an input configuration of M and c is the content of the history tape of c with the state letter moved next to ωk+1. And, moreover, this should happen at most |c| steps before arriving to c. The suffix c2 of c is completely determined by c. The sequence of commands from used to get from c to c is in one-to-one correspondence with the sequence of tape letters of c2 to the right of the state letter. Therefore c is completely determined by c. If the length of the computation is at least 2|c|, then a configuration of the form c1c2 must occur in it before c , where c2 differs from c2 only by the state letter (which, as in c2 is next to ωk+1. This is impossible because between every two arrivals at such configurations, every computation of M must execute one of the main commands, and increase the number of tape letters on the history tape. Thus we proved Claim 2 Every computation ending at c is of length less than 2|c|. Claim 2 implies Claim 1 and the fact that M is sym-universally halting. To prove that M polynomially reduces to M , let c, c be two configurations of M . In order to check whether c and c are equivalent first check whether c is accepted. For this we need to run the program of M for at most 2|c| steps and see whether we first get a configuration c of the form c1c2 where c1 is an input configuration of M and c2 is of the form αk+1vqk+1ωk+1 (qk+1 ∈ Qk+1) and then a configuration of the form c1αqk+1vωk+1. If so, then check the equivalence c1 ≡M sacc(M ) using the oracle that checks equivalence of configurations of M . The answer is “yes” if and only if c is accepted. That process takes linear (in |c|) number of steps and one oracle query. Similarly, we check if c is accepted. If Both c and c are accepted, then c ≡M c . If one of them is accepted and another one is not, then these configurations are not equivalent. Finally suppose that both c and c are not accepted. By Lemma 2.1, c ≡M c if and only if there exist two computations 1 starting with c1 and 2 starting with c2 such that the end configurations of these computations are the same. We can assume that either 1 or 2 has length >2(|c| + |c |) (that can be checked in time linear in |c| + |c |). Without loss of generality assume that the length of 1 is bigger than 2|c|. Then after at most 2|c| steps of 1 we arrive at a configuration of the form c1c2 where c1 is a configuration of c such that there exists a computation of length <|c2| of M starting with an input configuration d of M and ending at c1, and the sequence of commands used in this computation is in one-to-one correspondence with the tape letters in c2. Therefore c is equivalent to a configuration of the form dαk+1qk+1ωk+1 where qk+1 ∈ Qk+1 which is an input configuration of M (and we need linear time in terms of |c| to find this configuration). Hence we can assume that c = dαk+1qk+1ωk+1. If the longest computation of M starting at c has length <2|c |, then the longest configurations we can reach by one of these computations is at most 2|c |, and it would take at most O((|c | + |c|)2) steps of M to reach any of these configurations starting at c. Thus it would take polynomial time to check whether c ≡M c in this case. Thus we can assume that there exists a computation of M that starts at c and has length >2|c |. Then, as before we can assume that c = d αk+1qk+1ωk+1, where d is an input configuration of M . Now if order to check if c ≡M c it is enough to check whether d ≡M d which is one oracle query. It remain to apply Lemma 2.3 and convert M to a 1-tape Turing machine. 2.3 Minsky machines The hardware of a k-glass Minsky machine MMk , k ≥ 2, consists of k glasses containing coins. We assume that these glasses are of infinite height. The machine can add a coin to a glass, and remove a coin from a glass (provided the glass is not empty). The commands of a Minsky machine are numbered started at 0. A configuration of a k-glass Minsky machine is a k + 1-tuple (i ; 1, . . . , K ) where i is the number of the command that is to be executed, j is the number of coins in the glass # j . We can write a number in the unary notation: the number n is written as 1 . . . 1 (n ones). Clearly then we can view a configuration as a word in the alphabet consisting of digits 1 and symbols (,), ; and comma “,”. The accept configuration is s0 = (0; 0, . . . , 0) (the command number is 0, all glasses are empty) and input configurations have the form (1; m, 0, . . . , 0). Let us describe commands of Minsky machines more precisely. Each basic command has one of the following forms: • Put a coin in each of the glasses ##n1, . . . , nl and go to command # j . We shall encode this command as i ; → Add(n1, . . . , nl ); j where i is the number of the command; • Provided the glasses ##n1, . . . , nl are not empty, take a coin from each of these glasses and go to instruction # j . This command is encoded as • Stop. This command is encoded as i ; → 0; Remark 2.5 This defines deterministic Minsky machines. We will also need nondeterministic Minsky machines. Those will have two or more commands with the same number. Remark 2.6 We can also use (composite) commands that are not literally commands described above but can be easily split into a few basic commands, such as “Put coins in glasses ##i1, . . . , il provided glasses ##n1, . . . , nm are empty”. Theorem 2.7 Let X be a recursively enumerable set of natural numbers. Then the following holds: (a) there exists a 2-glass deterministic Minsky machine MM2 which “enumerates” X in the following sense: for every m ∈ N, if m ∈ X , then MM2 takes configuration (1; 2m , 0) to the accept configuration (0; 0, 0), if x ∈/ X , then starting with (1; 2m , 0) it works infinitely long time. (b) If X is recursive, then there exists a 2-glass deterministic Minsky machine MM2 that recognizes X and is sym-universally halting. (c) In (a) and (b) we can also assume for every computation of MM2 starting with a configuration c and each of the glasses, that glass is emptied after at most O(|c|) steps (here |c| denotes the size of configuration c, i.e., the total number of coins in all glasses of the configuration). (d) If M is a deterministic Turing machine recognizing X , then we can assume that MM2 polynomially reduces to M . Proof The proof of Part (a) can be found in [ 35 ]. To prove (b) let us first recall the way to convert a 1-tape Turing machine M into a 2-glass Minsky machine MM2 [ 35 ], and then prove that if M is sym-universally halting, then so is MM2. Suppose that the tape alphabet of M has m letters. For simpliciy consider the case when m = 2. The general case is absolutely similar. So suppose that the set of tape letters is {1, 2}. Let q1, . . . , qs be the set Q of state letters. With every configuration c = αuqi vω of M we associate the following configuratiion of a 3-glass Minsky machine MM3: φ (s) = (i ; nu , nv, 0) where nu is the word u viewed as a natural number written in base 3, and v is the word v read from right to left. Now every command aqi b → a q j b of M is interpreted by MM3 as follows. If a, a are not empty, the machine MM3 needs to check whether nu ≡ a (mod 3), nv ≡ b (mod 3) and if so, then replace the last digit of nu by a and the last digit of nv by b . If, say, a is empty, then the machine should just multiply nu by 3 and add a . For example, the command θ of the form 1qi 2 → q j 1 is interpreted by a sequence P(θ ) of commands of MM3 as follows. The commands of P(θ ) will be numbered i.1 through i.l for some l. In order to check that nu ≡ 1 (mod 3), the machine should remove one coin from the first glass, then repeatedly keep removing 3 coins from the first glass while adding 1 coin to the third glass. If at the end the first glass is empty, then, indeed, nu ≡ 1 (mod 3). If this is the case (otherwise the command is not applicable and the machine halts), we remove the coins from the third glass, and check if nv ≡ 2 (mod 3). If so, we multiply the number of coins in the second glass by 3 (keep adding 3 coins to the second glass while removing a coin from the third glass until the third glass is empty), and then add two coins to the second glass. It is easy to see that if c is a configuration of M , then after applying P(θ ) to φ (c), we get φ (θ (c)). The length of the computation of P(θ ) connecting φ (c) with φ (θ (φ)) is O(|c|). Let MM3 be the resulting 3-glass Minsky machine. In order to convert it into a 2-glass Minsky machine, we associate with every configuration (c = i, m, n, p) of MM3 the following configuration ψ (c) of a 2-glass Minsky machine: (i ; 2m 3n5 p, 0). Now removing (adding) a coin from (to) glass number j of MM3 where j = 1, 2 or 3 is simulated by dividing (multiplying) the number of coins in the first glass by 2,3 or 5 respectively. Notice that the following property of MM3 holds: (*) If is a computation of MM3 and c1, c2, . . . , cm are configurations occurring in that computation, and m > |c1| for some universal constant , then one of the configurations c2, . . . , cm must be of the form φ (c ) where c is a configuration of M . Moreover if φ (c ) and φ (c ) are two consecutive configurations of that form in , then there exists a command θ of M such that θ (c ) = c . Therefore if there exists an infinite computation of MM3 or MM3−1 starting with some configuration c of MM3, then there exists an infinite computation of M (resp. M −1) starting with some configuration c . Moreover if c is not accepted by MM3, then c is not accepted by M . Thus if M is sym-universally halting, then MM3 is sym-universally halting. The proof for MM2 is similar. This gives Part (b). Part (c) of the theorem immediately follows from the construction. Part (d) is proved as follows. Suppose that c, c are two configurations of MM3 (for MM2 the proof is similar). By (*) in at most O(|w|) steps of MM3 either c turns into a configuration of the form φ (d) for some configuration d of M or MM3 halts. In the latter case, we check whether c is equivalent to c in O(|c|) steps. So we can assume that both c and c are equivalent to configurations φ (d) and φ (d ) for some configurations d, d of M , and the lengths of d, d are O(|c| + |c |). Again by (*), c is equivalent to c if and only if d and d are equivalent configurations of M . Thus we need to use the oracle once. 3 Simulation of Minsky machines by semigroups 3.1 The construction Here we will show how to simulate a Minsky machine by a semigroup. The construction is based on the following general idea which applies also in the case of solvable groups considered later. (1) First, with every configuration ψ we associate a word w(ψ ). Then with every command κ of the Minsky machine M we associate a finite set of defining relations Rκ . The semigroup S(M ) is defined by the relations from the union R of all Rκ (which is finite since we have only a finite number of commands) and usually some auxiliary relations Q which are in a sense “independent” of R but make the semigroup “smaller”. We need Q, for example, to make sure S(M ) satisfies a particular identity. We say that the semigroup S(M ) simulates M if the following holds for arbitrary configurations ψ1, ψ2 of M : ψ1 ≡M ψ2 if and only if w(ψ1) = w(ψ2) in S(M ). Usually, in order to prove the property (1) one has to prove the following two properties. Property 3.1 If a configuration c can be obtained from a configuration c by a command κ of M then the word w(c ) can be obtained from the word w(c) by applying defining relations of S(M ) from the set Rκ . Property 3.2 If a word w(c ) can be obtained from a word w(c) by applying the defining relations of S(M ) then c ≡M c . It is easy to see that Properties 3.1 and 3.2 imply property (1). There is an easy way to interpret Minsky machines in a semigroup S(M ). Let MMk be a Minsky machine with k glasses and commands ##1, 2, . . . , N , 0 (here the command number 0 is the stop command, it is the command with domain {0; 0, . . . , 0)}). Then S(MMk ) is generated by the elements q0, . . . , qN and {ai , Ai | i = 1, . . . , k}. The set of defining relations of S(MMk ) consists of all relations ai a j = a j ai , ai A j = A j ai , Ai A j = A j Ai , i = j, (2) which we shall call commutativity relations, all relations of the form x y = 0 where x y is a two-letter word which is not a subword of a word of the form qi a11 . . . akk A1 . . . Ak modulo the commutativity relations (2) (for example qi q j = Ai ai = ai q j = Ai q j = 0), which we shall call 0-relations, and relations associated with commands of M according to the following table, Command of MMk i, n1 > 0, . . . , nm > 0 → Sub(n1, . . . , nm ); j qi an1 . . . anm = q j These will be called the Minsky relations. (3) The words in S(MMk ) corresponding to configurations of M are the following: w(i ; 1, . . . , k ) = qi a11 . . . akk A1 . . . Ak . The proof that Properties 3.1 and 3.2 hold in S(MMk ) follows easily from Lemma 2.1, see [ 32,55 ]. 3.2 Residually finite finitely presented semigroups The following obvious lemma shows that the auxiliary 0-relations make the semigroup S(MMk ) really small: it just does not have too many elements which are not related to configurations of MMk . Lemma 3.3 Every word W in the generators of S(MMk ) that is not equal to 0 in S(MMk ) is, modulo the commutativity relations, a subword of some word of the form w(i ; 1, . . . , k ). Lemma 3.4 Suppose that a word W is not 0 in S(MMk ). By Lemma 3.3 W is a subword of a word of the form w(i ; 1, . . . , k ) (up to the commutativity relations). Suppose that W does not contain either qi or one of the A j . Then there are at most O(|W |) different (up to the commutativity relations) words that are equal to W in S(MMk ). All these words are subwords of words of the form w(i ; 1, . . . , k ) such that the configurations (i ; 1, . . . , k ) and (i , 1, . . . , k ) of M are equivalent. Proof If W does not contain qi , then the only relations that apply to W are the commutativity relations, so the only words that are equal to W in S(M ) are the words obtained from W by the use of commutativity relations. Suppose that W contains qi but does not contain one of the A j . Without loss of generality, we can assume that W contains every letter from w(i ; 1, . . . , k ) except some of the A j ’s. Every application of a Minsky relation to W corresponds to a command of the Minsky machine, applied to the configuration c = (i ; 1, . . . , k ). Let c = c1 → c2 → . . . be any computation of Sym(MMk ) starting with c. Then the sequence of commands of MMk applied in that computation has the form 1 2−1 where 1, 2 are computations of MMk (by Lemma 2.1). Each computation i corresponds to a sequence Y of applications of Minsky relations and commutativity relations (by Property 3.1). If that sequence of relations can be applied to W , then this computation never checks whether glass # j is empty. By Property (c) of Theorem 2.7, the lengths of 1 and 2 must be at most O(|W |). This implies the statement of the lemma. Recall that s0 is the accept configuration of MMk , i.e., s0 = (0; 0, . . . , 0). Lemma 3.5 Suppose that the Minsky machine MMk is sym-universally halting. Then (a) Every element z of S(MMk ) which is not equal to 0 or w(s0) has finitely many divisors, i.e., elements y such that z = pyq for some p, q ∈ S(MMk ) ∪ {1}. (b) For every configuration c of MMk the word w(c) is equal to w(s0) in S(MMk ) if and only if ψ is accepted by MMk . Proof (a) If z is represented by a word w that contains one of the qi and all letters A j , then it must be equal to one a word of the form w(i ; 1, . . . , k ) modulo commutativity relations (by Lemma 3.3). In that case applying relations of S(MMk ) to w amounts to applying commands of Sym(MMk ) to the configuration c = (i ; 1, . . . , k ) (by Properties 3.1 and 3.2). Thus every divisor of z is represented by a subword of one of the words w(c ) such that c ≡MMk c . Also note that c cannot be an accepted configuration of MMk because otherwise z would be equal to w(s0) in S(MMk ). Since MMk is sym-universally halting, the number of configurations that are equivalent to c is finite. Hence the number of divisors of z is finite too. If a word w representing z does not contain a q-letter or one of the Ai , then we can apply Lemma 3.4. (b) This follows from Properties 3.1 and 3.2. Remark 3.6 Note that for every element z of any semigroup S the set of all nondivisors of z in S is an ideal (denoted by N (z)). By definition of a divisor, N (z) does not contain z. The Rees quotient semigroup S/N (z) consists of all divisors of z and 0 with a natural multiplication. In particular, if z has only finitely many divisors then S/N (z) is finite. Also note that for every ideal I of S, if z ∈/ I , then z is separated from every other element of S by the natural homomorphism from S to S/I . Lemma 3.7 If MMk is sym-universally halting, then S(MMk ) is residually finite. Proof Suppose that MMk is sym-universally halting. Let z1 = z2 be two different elements of S(MMk ). We need to show that there exists a homomorphism from S(MMk ) to a finite semigroup separating z1 and z2. First suppose that either z1 or z2 is not in {0, w(s0)}. Then by Remark 3.6 z1 and z2 are separated by one of the natural homomorphisms from S(MMk ) to S(MMk )/N (z1) or S(MMk )/N (z1) which are finite semigroups by Lemma 3.5 (a). Thus we can assume that z1 = 0, z2 = w(s0). Consider the (finite) set U of all subwords of the word A1 A2 . . . Ak , including the empty word ∅. We identify words in U which are equal modulo the commutativity relations. For each u ∈ U let us introduce a symbol κu . Now consider the finite set L = {0, αi , Ai , κu | i = 1, . . . , k, u ∈ U } with the following operation: κu αi = κu if u does not contain the letter Ai otherwise κu αi = 0, κu Ai = κu Ai , if u does not contain Ai and κu Ai = 0 otherwise, αi Ai = Ai , αi2 = αi , αai A j = A j αi , αi α j = α j αi , Ai A j = A j Ai for every i = j between 1 and k, all other products are equal to 0. Then it is easy to see that L is a finite semigroup and the map qm → κ, ai → αi , Ai → Ai extends to a homomophism from S(MMk ) to L separating z1 and z2. Recall that the Dehn function of a finite semigroup presentation X | R is the minimal function f (n) such that for any words u, v which are equal in S and such that |u| + |v| ≤ n, there exists a derivation of length at most f (n) of this equality from the defining relations. For a finitely presented semigroup, a Dehn function does not depend on the choice of finite presentation (up to equivalence), and the equivalence class of that function is called the Dehn function of the semigroup. Remark 3.8 The time complexity of the word problem is bounded from above in terms of the Dehn function of a finitely presented semigroup S: given the Dehn function f (n) of a semigroup presentation P, in order to check whether w = w (mod P) with |w| + |w | ≤ n, we just need to check all sequences of length ≤ f (n) of applications of defining relations w → w1 → . . .. moreover the word problem is decidable if and only if the Dehn function is recursive [ 39 ]. Theorem 3.9 For every recursive set of natural numbers X and every recursive func tion g(n) there exists a finitely presented residually finite semigroup S such that the word problem in S is as hard as the membership problem in X and polynomially reduces to it; the Dehn function S is bigger than g(n). Proof By Theorems 2.4 and 2.7 there exists a sym-universally halting 2-glass Minsky machine that recognizes X and whose configuration equivalence problem polynomially reduces to the membership problem in X . By Lemma 3.5, the problem of recognizing equality to w(s0) in S(MM2) is at least as hard as the membership problem in X . By Lemma 3.7, S(MM2) is residually finite. Remark 3.10 The proof of Theorem 3.9 could be simplified a little if instead of the semigroup S(MMk ) we consider the semigroup S˜(MMk ) obtained from S(MMk ) by adding one relation q0 = 0. That is S˜(MMk ) is the Rees quotient of S(M Mk ) by the ideal generated by q0. Indeed, if MMk is sym-universally halting, then in S˜(MMk ) every non-zero element has only finitely many divisors, and so it is residually finite by [ 22 ]. The word problem in S˜(MMk ) and the word problem in S(MMk ) are polynomially equivalent. That follows from the fact that no command of MMk apply to a configuration of the form (0; m, n). The semigroup S˜(MMk ) is used in the next subsection. 3.3 Residually finite semigroups with large depth function Recall the definition of the depth function ρ: for every finitely generated residually finite semigroup S and every number n, ρS(n) is defined as the smallest number such that for every two different elements z, z in S of word length ≤ n there exists a homomorphism φ from S onto a finite semigroup B of cardinality at most ρS(n) such that φ (z) = φ (z ). The following lemma from [ 22 ] follows from Remark 3.6. Lemma 3.11 Suppose that every non-zero element of a semigroup S with 0 has finitely many divisors. Then S is residually finite. Theorem 3.12 For every recursive set of natural numbers X and every recursive function g(n) there exists a finitely presented residually finite semigroup S such that the depth function of S is bigger than g(n). In addition, the word problem in S and the membership problem in X polynomially reduce to each other. Proof Let MM2 be a sym-universally halting 2-glass Minsky machine with N + 1 commands numbered 0, . . . , N . We need the following property of MM2: Thus we can assume that both w1 and w2 start with a q-letters. For every where αi = αi (w1) ∈ {0, 1} we denote li by li (w), and αi by αi (w). Then for every word w obtained from w by applying the relations of S˜(MM4) we have α j (w ) = α j (w), j = 1, 2, 3, 4, u(w) is obtained from u(w1) by applying relations of S(MM2). Claim There are only finitely many words that are equal to u(w1) in S˜(MM2). Indeed, if one of the numbers α1 or α2 is 0, the Claim is true by Theorem 2.7 (c). If both α1 and α2 are equal to 1, since w1 is not equal to 0 in S˜(MM4), the configuration (i ; l1, l2) is not accepted by MM2 (here we use the relation q0 = 0), and the Claim is true because MM2 is sym-universally halting. Let Y be the set of all words that are equal to w1 in S˜(MM4). Then the Claim implies that the set of numbers l3(w) − l4(w), w ∈ Y , is finite. Let D(w1) be the maximum of all these numbers. The number D(w2) is defined similarly. Let D be the maximum of D(w1), D(w2). Let us add the relations a3D = a32D, a4D = a42D to S˜(MM4). Let S¯ be the resulting semigroup, and ψ : S˜(MM4) → S¯ be the corresponding homomorphism. Then it is easy to see that ψ (w1) = ψ (w2). Notice that in S¯, every non-zero element has finite number of divisors. Indeed, it is true for S˜(MM2) (see Remark 3.10) and the number of different elements of S¯ of the form v(w) is finite. Hence we can again use Lemma 3.11. The function ρ(n) for the semigroup S˜(MM4) is at least as large as the following function (n) associated with the machine MM4: (n) is the smallest number such that for every non-accepted input configuration of M of length ≤ n, the machine MM4 halts after at most (n) steps (we call this function the co-time function of MM4). Indeed let c be an input configuration of length at most n such that MM4 halts after exactly (n) steps starting at c. Suppose that the word w(c) in S˜(MM4) corresponding to the configuration c can be separated from 0 in a homomorphic image E of S˜(MM4) with at most (n) − 1 elements. Then the images of a3, a4 in that semigroup satisfy z D = z2D for some D < T (n). Note that • the halting computation has > D steps, • the letter a3 does not occur in w(c), • every old command of MM4 adds one coin in glass 3, Therefore there exists a word W which is equal to w(c) in S˜(MM4) and which has the form Modulo relations corresponding to the commands (4), this word is equal to q j a1l1 a2l2 A1 A2a3D A3 A4. q j a1l1 a2l2 A1 A2a32D A3a4D A4. The image of the latter word in E is equal to q j a1l1 a2l2 A1 A2a3D A3a4D A4 q j a1l1 a2l2 A1 A2 A3 A4. which, again modulo the relations corresponding to the commands (4), is equal to Here j > 1 by our assumption that command number 1 cannot be used in the middle of a computation consisting of old commands. Hence the latter word is equal to 0 by the relations corresponding to the commands (5) and the relation q0 = 0, a contradiction. Note that the co-time function of a Turing machine recognizing a recursive set can be larger than any given recursive function. Indeed, we can assume that the Turing machine has a history tape as in Sect. 2.2, so if the machine does not accept the input, the last configuration in the halting computation has the history of computation written on the tape. Then after the Turing machine halts without accepting, we can make it compute some large recursive function, taking as a variable the word written on the history tape. It remains to note that the co-time function of a Minsky machine simulating that Turing machine cannot be smaller. In the next section, we shall use the following properties of the semigroups studied in this section. Let Sˇ be the semigroup given by all non-Minsky defining relations of S(MMk ) and let Sˇ¯ be the semigroup given by all non-Minsky defining relations of the semigroup S¯ from the proof of Theorem 3.12. Lemma 3.15 (a) The growth function of Sˇ is polynomial of degree k. (b) The semigroup Sˇ satisfies the following property: where α, α , β j , β j ∈ {0, 1} and w = w in Sˇ, then qiα = qiα (i.e., either α = α = 0 or i = i and α = α ), m j = m j , β j = β j , j = 1, . . . , k. (c) The semigroup Sˇ¯ satisfies the following two properties (Q1) If where i, i = 0, α, α , β j , β j ∈ {0, 1}, and w = w in Sˇ¯, then qiα = qiα , β j = β j , j = 1, . . . , k. (Q2) For every word w the equality w Ai = 0 in Sˇ¯ implies wai Ai = 0 in Sˇ¯. Proof (a) Indeed every non-zero element of length ≤ n of Sˇ is represented (modulo the commutativity relations) by a word of the form qi a1m1 . . . akmk A11 . . . Akk where m j ≤ n, j ∈ {0, 1}. To prove Properties (P) and (Q1) notice that the exponents of qi , a j , A j do not change when we apply the defining relations of these semigroups to w. To prove (Q2) notice that word the w = qiαa1m1 . . . akmk A1β1 . . . Akβk is equal to 0 in Sˇ¯ if and only if i = 0 and then apply (Q1). 4 Simulation of Minsky machines in solvable groups Recall that a variety of algebraic structures is a class of all algebraic structures of a given signature satisfying a given set of identities (also called laws). Equivalently, by a theorem of Birkhoff [ 35 ] a variety is a class of algebraic structures closed under taking cartesian products, homomorphic images and substructures. Every variety contains free objects (called relatively free algebraic structures). One can define algebraic structures that are finitely presented in a variety as factor-structures by congruence relations generated by finite number of equalities. Every finitely presented algebraic structure which belongs to a variety V is finitely presented inside V but the converse is very rarely true. See [ 32 ] for a survey of algorithmic problems for varieties of different algebraic structures (mostly semigroups, groups, associative and Lie algebras). In this section we concentrate on varieties of groups (see [ 43 ]). The most well known varieties are the variety of Abelian groups A given by the identity [x , y] = 1, the variety of nilpotent groups of class c, Nc given by the identity [. . . [x1, x2], . . . , xc+1] = 1, etc. The class of Abelian groups of finite exponent d, Ad , is also a variety, given by two identities [x , y] = 1, x d = 1. If U and V are two varieties of groups then the class of groups consisting of extensions of groups from U by groups from V is again a variety (the product of U and V) denoted by U V. The product of varieties is associative [ 43 ]. For example the variety of all solvable groups of class c is the product of c copies of the variety A. If V is a variety of groups, then ZV is the variety consisting of all central extensions of groups from V. For example N2 = ZA and, more generally, Nc+1 = ZNc for every c ≥ 1. 4.1 The construction Let MMk be a Minsky machine with k glasses and N + 1 commands (numbered 0, . . . , N ). We are going to construct a group G(MMk ) simulating MMk . The group G(MMk ) will be very close to the semigroup S(MMk ) constructed above. The main idea will be to replace the product in S(MMk ) by another, derived, operation and make sure that with respect to the new operation the semigroup S(MMk ) “embeds” into our group, in such a way that the “image” of S(MMk ) is “almost the whole” group. The group will be generated by the x -letters which will be related to the letters qi from S(MMk ), and also a-letters a1, . . . , ak , A-letters A1, . . . , Ak and some other a- and A-letters that help us impose the necessary commutativity relations that, in particular, make the group solvable, and contain “very few” extra elements. The group we are going to construct will be a semidirect product of an elementary Abelian normal subgroup generated (as a normal subgroup) by the x -letters by a semidirect product of an Abelian subgroup generated (as a normal subgroup) by A-letters and an Abelian subgroup generated by a-letters. Thus we should have a way to ensure that in a subgroup generated by two sets of letters Z ∪ Y , the normal subgroup generated by Z is Abelian. This is done with the help of the following lemma due to Baumslag [3] and Remeslennikov [51]. In that lemma we denote ua = a−1ua and ua+b = ua ub (note that although ua+b is not necessarily equal to ub+a , the equality will hold if the normal subgroup generated by u is Abelian, which is going to be the case every time we apply this lemma). Lemma 4.1 ([3,51]). Suppose that a group H is generated by three sets X, F = {ai | i = 1, . . . , m}, F = {ai | i = 1, . . . , m} such that (1) The subgroup generated by F ∪ F is Abelian; (2) For every a ∈ F and every x ∈ X we have x f (a) = x a for some monic polynomial f of a which has at least two terms (everywhere below f (t ) = t − 1 and hence x f (a) = a−1x ax −1 = [a, x −1], so we will not mention f again); (3) [x1a1α1 ...amαm , x2] = 1, for every x1, x2 ∈ X , and every α1, . . . , αm ∈ {0, 1, −1}. Then the normal subgroup generated by X in the group H = X ∪ F ∪ F is Abelian, and H is metabelian. If the elements ai and ai and the set X satisfy the conditions of Lemma 4.1 we will call ai , BR-conjoints to ai with respect to X (and the polynomial f ), i = 1, . . . , m. Consider the free commutative monoid generated by letters A0, . . . , Ak . Let U be the set of all divisors of the element A0 A1 . . . Ak in that monoid, and U be the set of all symbols q j u, u ∈ U , j = 0, . . . , N . Also fix a prime p (say, p = 2). The generating set of our group G = G(MMk ) will consist of three subsets: L0 = {x (u) | u ∈ U, i = 0, . . . , N }; L1 = { Ai | i = 0, . . . , k, }; L2 = {ai , ai , a˜i , a˜i | i = 1, . . . , k}. We introduce notation for some subgroups of the group G. Denote Hi = Li , i = 0, 1, 2. Denote also M0 = {a˜i , a˜i , A0 | i = 1, . . . , k}, Mi = {ai , ai , Ai }, i = 1, . . . , k. The group G(MMk ) has the following finite set of defining relations: (G1) Relations saying that H0 and H1 are Abelian groups of exponent p, and H2 is an Abelian group. (G2) Any y ∈ Mi , z ∈ M j , i = j ∈ {0, . . . , k}, commute. (G3) For every i = 1, . . . , k, (ai )−1 is a BR-conjoint to ai−1 with respect to { Ai }. (G4) The elements of the set {(a˜i )−1 | i = 1, . . . , k} are BR-conjoints to elements of the set {a˜i−1 | i = 1, . . . , k} with respect to { A0}. (G5) a) If u ∈ U does not contain Ai for some i = 0, . . . , k, then [x (u), Ai ] = x (u Ai ). b) For every i = 1, . . . , k, if u does not contain Ai , then x (u)ai −1 = x (u)ai (where x a−1 = xa x −1), c) For every i = 0, . . . , k, if u contains Ai , z ∈ Mi , then [x (u), z] = 1. (G6) x (q j )ai = x (q j )a˜i , x (q j )ai = x (q j )a˜i , j = 0, . . . , N , i = 1, . . . , k. (G7) [x (u)z , x (v)] = 1, where z = a1α1 . . . akαk , αi ∈ {−1, 0, 1}, u, v ∈ U Remark 4.2 Relations (G7) together with (G1) and (G5) b) imply that for every subset I ⊆ {1, . . . , k} the letters {ai | i ∈ I } are BR-conjoints of {ai | i ∈ I } with respect to the set of all x (u)’s where u does not contain letters Ai , i ∈ I . (G8) Relations constructed from the program of the machine MMk . For every f ∈ G denote f ∗ ai = f −1 f ai f −ai−1 f (ai )−1 , i = 1, . . . , k, also let f ∗ Ai = [ f, Ai ], i = 0, . . . , k. We denote (. . . (t1 ∗ t2) ∗ . . .) ∗ tm by t1 ∗ . . . ∗ tm , and t1 ∗ t2 ∗ . . . ∗ t2 by t1 ∗ t2(n). i, n1 > 0, . . . , nm > 0 → Sub(n1, . . . , nm ); j x(qi A0) ∗ an1 ∗ . . . ∗ anm = x(q j A0) x (qi A0) ∗ a1(m1) ∗ . . . ak(mk ) ∗ A(1α1) ∗ . . . ∗ A(kαk ) = x (q j A0) ∗ a1(n1) ∗ . . . ∗ ak(nk) ∗ A(1β1) ∗ . . . A(kβk ) where αi , βi ∈ {0, 1} is true in G(MMk ) if and only if the equality qi a1m1 . . . akmk A1α1 . . . Akαk = q j a1n1 . . . akmk A1β1 . . . Akβk is true in the semigroup S(MMk ) (in particular, αi = βi for every i ). (c) The equality x (qi ) ∗ a1(m1) ∗ . . . ak(mk ) ∗ A(1α1) ∗ . . . ∗ A(kαk ) = x (q j ) ∗ a1(n1) ∗ . . . ∗ ak(nk ) ∗ A(1β1) ∗ . . . A(kβk ) where αi , βi ∈ {0, 1} is true in G(MMk ) if and only if the equality qi a1m1 . . . akmk A1α1 . . . Akαk = q j a1n1 . . . akmk A1β1 . . . Akβk is true in the semigroup Sˆ (i.e., these words coincide, see Lemma 3.15 (b)). Proof The proof of Part (a) is divided into several lemmas. Lemma 4.4 The subgroup H1 ∪ H2 of G is metabelian and a semidirect product of the Abelian normal subgroup H1H2 of exponent p, and H2. Proof Indeed by relations (G2), k i=1 Mi , i = 0, . . . , k = ai , ai , Ai × a˜i , a˜i , A0, i = 1, . . . , k . Using relations (G1), (G3), (G4), we can apply Lemma 4.1 to each of the factors in that direct product and conclude that each of them is metabelian and a semidirect product of the Abelian normal subgroup of exponent p generated by the intersection of { Ai | i = 0, . . . , k} with that factor, and the Abelian group generated by the a-letters from that factor. Lemma 4.5 The normal subgroup T of G generated as a normal subgroup by all the elements x (u), u ∈ U, is Abelian of exponent p. Proof Relations (G5) a) of the group G imply that every element x (u), u ∈ U, is a product of elements x (q j )z , z ∈ H1, i = 0, . . . , N . Therefore, it is enough to show that x (qk )x (qt )z = x (qt )z x (qk ) for any z ∈ H1, H2 and any k, t . To simplify these equalities notice that z = z0z1 . . . zk where zi ∈ Mi by (G2). Therefore equalities (7) are equivalent to x (qk )z0 x (qt )z1...zk = x (qt )z1...zk x (qk )z0 . (7) (8) We can represent element x (q j )zi , i ≥ 1, as a product of elements of the form x (q j )aip(ai )q and x (q j Ai )a˜ip(a˜i )q . Indeed we have the following sequence of equalities deduced using (G2), (G5), (G6): =(G=5) =a=),=c), =(G=6) x (q j )air1+r2 (ai )s1+s2 Ait2 ...airk (ai )sk Aitk (x (q j Ai )t1 )a˜ir1 (a˜i )s1 (9) Repeating this argument k times, one proves that x (q j )z1z2...zk can be represented as a product of elements of the form x (u)y where u ∈ U , y ∈ H2. A similar proof (using also (G4)) gives that x (q j )z0 is a product of elements of that form. It remains to note that elements of the form x (u)y , u ∈ U, y ∈ H2 commute by Remark 4.2 and Lemma 4.1. Remark 4.6 Note that equalities (9) and similar equalities when x (q j ) is replaced by r x (u), u ∈ U , imply the following: if y is a product of elements of the form ai l (ai )sl Ai and l rl = l sl = 0, then [x (u), y] is equal to 1 if u contains Ai or is equal to a product of conjugates of elements x (u Ai ) by elements from a˜i × a˜i otherwise. Similarly, suppose that y is a product of elements from M0, each factor containing A0, and the total exponent of every a˜i (resp. a˜i ) is 0. Then [x (u), y] = 1 provided u contains A0 and is a product of conjugates of x (u A0) by elements from ai , ai provided u does not contain A0. By construction, the group G is a semidirect product of T and the metabelian group H1H2 H2. By Lemma 4.4, G is solvable of class 3 and, moreover, belongs to A2pA. Remark 4.7 The proof of Lemma 4.5 shows that T is generated (as an Abelian group) by elements of the form x (u)y where u ∈ U and y ∈ H2. Lemma 4.8 The quotient of G(MMk ) over the center satisfies the identity [[x1, y1], [x2, y2], . . . , [xk+2, yk+2]] = 1. This means that G belongs to the variety ZNk+1A. Proof Let P be the derived subgroup of G(MMk ). By Lemma 4.5, every element of P is a product of an element of T and an element of H1H2 . It also follows from Lemma 4.5 that [ P, P] ⊆ T , hence by Remark 4.7, it is generated by elements of the form x (u)y , u ∈ U, y ∈ H2, the word u contains at least one Ai , i = 0, . . . , k. Since T is Abelian, the subgroup [ P, P, . . . , P] is generated by the commutators k+2 x (u)y , h1h,21,1 , . . . , h1h,2k,k for some h1,i ∈ H1, y, h2,i ∈ H2. An easy induction shows that every such commutator is a conjugate of where y ∈ H2. Let h ∈ H1, u ∈ U, y ∈ H2. Suppose that h = Ai11 · . . . · Aits where ti = 0. Consider t s [x (u), h y ]. Then Remark 4.6 implies that [x (u), h y ] is a product of elements of the form x (u )y where u ∈ U contains letters Ai1 , . . . , Ais and it may not be equal to 1 only if one of the letters Ai j does not occur in u. Therefore the commutator (10) is either equal to 1 or is a product of elements of the form x (u )y where the word u ∈ U contains all letters A0, A1, . . . , Ak , y ∈ H2. But every such x (u ) is in the center of G(MMk ) by (G5) c). Hence [ P, . . . , P] is contained in the center of G(MMk ). (10) We now prove Parts (b) and (c) of Theorem 4.3. For every configuration c = (i ; m1, . . . , mk ) of MMk let wG (c) = x (qi A0) ∗ a1(m1) ∗ . . . ak(mk ) ∗ A(1α1) ∗ . . . ∗ A(kαk ). To prove (b), as we mentioned before Property 3.1, we need to prove Properties 3.1 and 3.2. Property 3.1 for G(MMk ) is proved in the same way as for the semigroup S(MMk ) (see [ 32,55 ]),since the only property of S(MMk ) used there was that the word w = qi a1l1 . . . aklk A1α1 . . . Akαk is equal in S(MMi ) to any word obtained from w by permuting ai with a j , Ai with A j and ai with A j (i = j ). The same is true for words of the form wG (c) in G(MMk ) by the definition of the operation ∗, relations (G1), (G2) and Lemma 4.5. In order to prove Property 3.2 we will define a new group G that is an image of G under some homomorphism that is injective on the elements from Parts (b) and (c). Let Sˇ be the semigroup with the same generating set as S(MMk ) subject all the relations of S(MMk ) except the Minsky relations (3). That semigroup does not depend on MMk . Thus non-zero elements in Sˇ have the form qiα1 a1l1 . . . aklk A1α1 . . . Akαk where l j ∈ N, α j ∈ {0, 1}. Let W be the set of all non-zero elements of Sˇ containing a q-letter. Let ψ be the natural homomorphism from Sˇ onto S(MMk ). Let W0 = ψ (W ). We will need the set of vectors {1, 2, 3}k ⊂ Zk with coordinates 1, 2, 3. Its elements will be denoted by i . The j -th coordinate of the vector i will be denoted by i j , the standard unit vectors (0, . . . , 1, . . . , 0) are denoted by e j . Consider the set of symbols {z(i , u) | i ∈ {1, 2, 3}k , u ∈ W ∪ W0} and the multiplicative Abelian group T1 of exponent p freely generated by this set. For each letter from L1 ∪ L2, we define an automorphism of T1. The group G will be the semidirect product of T1 and the group generated by these automorphisms. For simplicity we will denote automorphisms corresponding to letters from L1 ∪ L2 by the same letters (the automorphisms are just conjugations by these letters). Let us start with automorphisms a j , a j . We have to define z(i , w)a j and z(i , w)a j for every i and every w ∈ W ∪ W0. This definition does not depend on whether w belongs to W or W0. First suppose that w does not contain A j . Then ⎧ ⎪⎪⎨ z(i , w)z(i + e j , w)z(i + 2e j , w)z(i , wa j ) z(i , w)a j = ⎪ z(i , w)z(i − e j , w)−1 ⎪⎩ z(i − 2e j , w) z(i , w)a j = z(i , w)−1z(i , w)a j . if i j = 1; if i j = 2; if i j = 3. If w contains letter A j , then let z(i , w)a j = z(i , w)a j = z(i , w). It is easy to prove that a j is an automorphism by constructing the automorphism a −j1 (view (11) as a triangular system of linear equations and solve it by backward substitution). For example: ⎧ ⎪⎪⎪ z(i − e j , w)−1z(i , w)−1z(i , wa j )−1, ⎪⎨ z(i , w)a −j1 = ⎪ z(i , w)z(i + e j , w), if i j = 2 ⎪⎪⎪⎩ z(i + 2e j , w), if i j = 1. if i j = 3 ⎧ ⎪⎪⎨ z(i , w)z(i + e j , w)z(i + 2e j , w)z(i , va j A j ), z(i , w)a˜ j = ⎪ z(i , w)z(i − e j , w)−1, ⎩⎪ z(i − 2e j , w), z(i , w)a˜ j = z(i , w)−1z(i , w)a˜ j . The automorphisms corresponding to A j , j = 1, . . . , k, are defined as follows: if w ∈ W ∪ W0 does not contain A j and if w contains A j . Finally the automorphism corresponding to A0 is defined as follows: A j A0 = z i , w z i , w A j = z i , w z i , ψ (w) if i j = 1; if i j = 2; if i j = 3. if w ∈ W and if w ∈ W0. The following lemma is obtained by a straightforward application of the definition of the automorphisms above and the definition of the operation ∗. This lemma implies that G satisfies (G8) if we replace x (u) by z(1, u) where 1 is the vector (1, 1, . . . , 1) (since the corresponding relations hold in S(MMk )). If w ∈ W ∪ W0, d ∈ {a j , A j | j = 1, . . . , k} then by wd we mean the product of w and d in S(MMk ) provided w ∈ W0, or in Sˆ provided w ∈ W . The proof of the following lemma is by inspection of various cases and mostly left to the reader. Lemma 4.9 The following relations hold in G. For every d ∈ {a j , A j | j = 1, . . . , k}, w ∈ W ∪ W0 z(1, w) ∗ d = z(1, wd) where ∗ is defined in (G8). Here we set z(1, 0) = 1 (where 0 is the zero element in S(MMk ) or Sˆ, 1 in the right hand side is the identity element in G(MMk )). Proof For example, if d = A1 then z(1, w) ∗ d = [z(1, w), A1]. It is equal to 1 if w contains A1, and it is equal to z(1, w)−1z(1, w)A1 = z(1, w)−1(z(1, w)z(1, w A j )) = z(1, w A j ) if w does not contain A1. Thus in both cases z(1, w) ∗ d = z(1, wd). By definition, G is the semidirect product of T1 and the subgroup of Aut(T1) generated by the automorphisms corresponding to the elements from L1 ∪ L2. From the definition of the automorphisms and Lemma 4.9, it follows that G is generated by the elements z(1, u), u ∈ U , and the automorphisms corresponding to elements of L1 ∪ L2. It is easy to check that all the relations (G1)-(G8) hold in G, therefore Lemma 4.10 The map that sends every a- or A-letter to itself, every x (u) with u ∈ U containing A0, u = v A0, to z(1, ψ (v)) and every x (u) with u not containing A0 to z(1, u) extends to a homomorphism γ from G to G. Lemma 4.11 The homomorphism γ is surjective and the preimage of T1 is T . Proof We only need to define pre-images of elements z(i , w) ∈ G¯ , w ∈ W ∪ W0. By the definition of γ , we have γ (x (u)) = z(1, u) for every u ∈ U which does not contain A0, and γ (x (u A0)) = z(1, ψ (u )) so x (u), u ∈ U , are preimages of all z(1, u). The preimage x (i , w ) of z(i , w) for every i and w is defined by induction on the length of w and the sum of i j (the base of induction, where i = 1 is obvious). x (i + e j , w) = x (i , w)−(a j )−1 , x (i + 2e j , w) = x (i , w)a −j1 , x (i , wa j ) = x (i , w) ∗ a j . We also set x (i , w A j ) = x (i , w) ∗ A j for any i . This covers the case when w contains A j and i j = 2 or 3. It is easy to see that for every i and w ∈ W ∪ W0, we have γ (x (i , w)) = z(i , w). This proves the lemma. In G(MMk ), consider the set P0 of elements where αi ∈ {0, 1} and the set P of elements z(1, ψ (qi )) ∗ a1(m1) ∗ . . . ak(mk ) ∗ A(1α1) ∗ . . . ∗ A(kαk ) z(1, qi ) ∗ a1(m1) ∗ . . . ak(mk ) ∗ A(1α1) ∗ . . . ∗ A(kαk ) By construction P ∩ P0 = ∅, elements (12) are different if and only if elements qi a1m1 . . . akmk A1α1 . . . Akαk from S(MMk ) are different, and elements (13) are different if and only if the corresponding elements qi a1m1 . . . akmk A1α1 . . . Aαk of Sˇ are different. By Lemma 4.9 the k element x (qi A0) ∗ a1(m1) ∗ . . . ak(mk ) ∗ A(1α1) ∗ . . . ∗ Ak(αk ) is equal to the element (12) and the element x (qi ) ∗ a1(m1) ∗ . . . ak(mk ) ∗ A(1α1) ∗ . . . ∗ A(kαk ) is equal to the element (13). This completes the proof of Property 3.2 and Theorem 4.3 (b), (c). We shall need a few more properties of the group G(MMk ). Lemma 4.12 Let elements x (i , w), w ∈ W ∪ W0, from G(MMk ) be defined as in the proof of Lemma 4.11. Let y ∈ L1 ∪ L2, w ∈ W ∪ W0. Then x (i , u)y is a product of one or several elements of the form x (i , w ) such that every letter a j occurs in w at least as many times as in w (in particular if for some R > 0, w belongs to the ideal VR defined in Lemma 3.7, then w ∈ VR . Proof Indeed it is easy to check that for every i ∈ {1, 2, 3}{1,...,k}, x (i , w) satisfies the same equalities as elements z(i , w) from the definition of automorphism of G with z replaced by x everywhere. Then the statement of the lemma for y ∈ ∪ j≥1 M j , follows from the way x (i , u) are constructed. For y ∈ M0, one needs to use (G2), (G5) c), and (G6). Lemma 4.13 The normal subgroup T of G = G(MMk ) generated by the elements x (u), u ∈ U is the direct product of cyclic subgroups generated by the elements x i , w , i ∈ {1, 2, 3}{1,...,k}, w ∈ W ∪ W0. Proof By Lemma 4.12 elements x (i , w) span T . We defined elements x (i , w), w ∈ W ∪ W0 in such a way that they are pre-images of the corresponding elements z(1, w) in G under γ . Thus the elements x (i , w), i ∈ {1, 2, 3}{1,...,k}, w ∈ W ∪ W0 are linearly independent since their images under γ are linearly independent in T1 (here we temporarily view T , T1 as vector spaces over the field with p elements). The proof of Lemma 4.12 actually gives the following two facts (the proof of the first of them also employs Lemma 3.15 (a)). Lemma 4.14 If v is a word in a- and A-letters (i.e. over L1 ∪ L2), u ∈ U , then x (i , u)v is a product in G of elements x ( j , w) as in Lemma 4.12 where the length of each w does not exceed |v| + O(1) hence the total number of different x ( j , w) occurring in this product is polynomial in |v| of degree at most k. Lemma 4.15 Let v, w ∈ W . Suppose v = w is a Minsky relation of the semigroup S(MMk ). Then the corresponding relation from (G8) has the form x (1, v)(x (1, w))−1 = 1. The conjugate of x (1, v)x (1, w)−1 by an element g ∈ H1, H2 is a product m (x (i , vm )x (i , wm )−1), where for each m, vm = vum , wm = wum for some word um whose length does not exceed the length of g. Remark 4.16 Instead of semigroup S(MMk ) we could use the semigroup S˜(MMk ) and construct a group G˜ (MMk ) which is a quotient of G(MMk ) by the subgroup spanned by x (i , w) where w ∈ W0 ∪ W contains q0. It is easy to see that the construction of G(MMk ), Lemmas 4.4, 4.5, 4.8–4.15 (with notation modified in the appropriate way by “killing” q0) and Theorem 4.18 hold for the group G˜ (MMk ) as well. Moreover let S¯ be any homomorphic image of S obtained by adding defining relations to S(MMk ). Let Sˇ¯ be the semigroup given by the same presentation as S¯ excluding the Minsky relations. Suppose that Properties (Q1), (Q2) from Lemma 3.15 hold in Sˇ¯. Then we can replace S(MMk ) by S¯, and Sˇ by Sˇ¯ in the construction of G(MMk ). The resulting group will satisfy all the lemmas and the theorem mentioned in the previous paragraph (with S(MMk ) replaced by S¯, and Sˇ replaced by Sˇ¯). To verify this statement, one just needs to routinely check the proofs of these lemmas and the theorem. 4.1.1 A finitely presented solvable group with undecidable word problem By Theorem 2.7, there exists a 2-glass Minsky machine which computes a nonrecursive partial function. The corresponding group G(MMk ) has undecidable word 2 problem and belongs to the variety A pA ∩ ZN 3A by Theorem 4.3. Hence we obtain the following: Theorem 4.17 See ([28]). There exists a finitely presented group with undecidable 2 word problem that belongs to the variety A pA ∩ ZN 3A. The proof of this theorem presented in this paper is simpler than the original proof in [ 28 ]. 4.1.2 Residually finite finitely presented groups Theorem 4.18 If a Minsky machine MMk is sym-universally halting then the group G(MMk ) is residually finite. The word problem in G(MMk ) and the configuration equivalence problem for MMk are polynomially reducible to each other. Proof Let MMk be a sym-universally halting Minsky machine. Let g = 1 ∈ G(MMk ). We use the notation from the definition of G(MMk ). There exists a natural homomorphism ζ from G(MMk ) to the metabelian group H1H2 H2 with kernel T . Since every finitely generated metabelian group is residually finite, we can assume that ζ (g) = 1. Hence g ∈ T . By Lemma 4.13, g is a product of elements of the form x (i , w), i ∈ {1, 2, 3}{1,...,k}, u ∈ W ∪ W0. (14) Hence g = g0g1 where g0 (resp. g1) is a product of elements (14) with w ∈ W0 (resp. w ∈ W ). Suppose that g1 is not 1. Let T be the subgroup of G(MMk ) generated by elements (14) with w ∈ W0. Then T is a normal subgroup of G(MMk ) by Lemma 4.12. Let G (MMk ) = G(MMk )/ T . This group is a semidirect product of T / T and the metabelian group H1H2 H1. Let D be the sum of lengths of words w ∈ W that appear in the factors of g1. Let YD be the set of all words in Sˇ where at least one a-letter appears at least D times, and 0. Then YD is an ideal in Sˇ, and the image of the set of elements (14) with w ∈ YD in G (MMk ) form a normal subgroup N of G (MMk ) of finite index (because T is an Abelian group of finite exponent p). That normal subgroup does not contain g by Theorem 4.3 (c). Then G (MMk )/N is a semidirect product of a finite group and the metabelian group H1H2 H2. Hence G (MMk )/N is residually finite and g can be separated from 1 by a homomorphism from G(MMk ) onto a finite group. Finally suppose that g1 = 1. Let w1, . . . , wl be the elements from W0 that appear in the representation of g as a product of elements (14). Let E be the set of words that is equal to one of the w j in S(MMk ). Since MMk is sym-universally halting, E is finite. Let D be the maximal length of a word in E . Let, as above, YD be the ideal in Sˇ consisting of 0 and all elements where one of the a-letters appears at least D times. Let Z D be the set of non-zero elements of S(MMk ) that are images of words from YD under the natural homomorphism Sˇ → S(MMk ). Then Z D does not contain w1, . . . , wl . Consider the subgroup F of T spanned by all elements (14) with w ∈ Z D ∪ YD. From Lemma 4.12, it follows that F is a normal subgroup of G(MMk ) of finite index in T . Since Z D does not contain w1, . . . , wl , the subgroup F does not contain g. The factor-group G(MMk )/F is a semidirect product of a finite group and the metabelian group H1H2 H2, and we can complete the proof as above. To prove that the configuration equivalence problem in MMk polynomially reduces to the word problem in G(MMk ) we notice that the length of a word w(c) corresponding to an input configuration c of MMk in the semigroup S(MMk ) is |c| + O(1). If w(c) is the word of length n in S(MMk ), then the corresponding element in G(MMk ) according to relations (G8), can be represented as a similar word wg(c) with respect to the ∗ operation. Every time when we evaluate the ∗ operation, we rewrite the word as a product of x (q j A0)u , u = u1u2 . . . uk , ui = aiki Ai or ui = ai i , j = 0, . . . , k where u k is a subword of w(c). Using commutativity relations from Lemma 4.5, we can collect all x (q j A0)u with the same u. Elements x (q j A0)u have order p. Therefore, we have a product of conjugates (x (q j A0)u )r , where 1 ≤ r ≤ p − 1, and k1 + k2 + k3 ≤ n. The number of such elements is at most O(n3), each of them can be written as a group word of length at most 2n + 1. Therefore for w(c) of length n, the corresponding element in G(MMk ) can be represented as a word of length at most O(n4). Now two configurations c and c of MMk are equivalent if and only if w(c) = w(c ) in S(MMk ) and if and only if wg(c) = wg(c ) in G(MMk ) by Properties 3.1 and 3.2. Thus checking whether c ≡MMk c can be done in polynomial time in terms of |c|+|c | with using the oracle responsible for the word problem in G(MMk ) only once. To get a polynomial reduction in the other direction we consider any element g of represented by a word w of length ≤n in G(MMk ). We need to check if g = 1. First check if g is in the normal subgroup T . By construction G(MMk ) is the semidirect product of T and a finitely generated metabelian group (generated by a- and A-letters). Since metabelian groups embed into finite direct products of linear groups over fields [62], the membership g ∈ T can be checked in polynomial time. If the answer is “no”, then g = 1. Now suppose that g is in T , then we represent w in G(MMk ) as a product of a fewer than n conjugates x (i , u)v where u ∈ U , and v is a word in a-letters and A-letters whose length is bounded by n. By Lemmas 4.14 and 4.11 then w is a product in G(MMk ) of elements of the form (12) and (13) whose number is bounded by a polynomial in n and whose lengths (in terms of the operation ∗) are bounded by n. Since different words of that form are linearly independent (by Lemma 4.11) in order to check whether g = 1, we need to verify equalities of words of the form (12) and (13) which by Theorem 4.3 is equivalent to verifying equalities of corresponding words in S(MMk ), which, in turn, reduces to verifying equivalence of the corresponding configurations of MMk by Properties 3.1 and 3.2 which hold for S(MMk ). Thus the word problem in G(MMk ) polynomially reduces to the configuration equivalence problem for MMk . Remark 4.19 Note that if instead of the semigroup S(MMk ) we could start with any semigroup S¯ satisfying Properties (Q1), (Q2) of Lemma 3.15. By Remark 4.16, the resulting group G¯ satisfies all the properties mentioned in that remark. The proof of Theorem 4.18 shows that if in S¯ every non-zero element has finitely many divisors, then the group G¯ is residually finite. Moreover if an element w ∈ W0 has finite number of divisors in S¯, then there exists a homomorphism φ from G¯ to a finite group with φ (x (1, w)) = 1. Lemma 4.20 Let d(n) be the Dehn function of G(MMk ) and t (n) be the time function of MMk . Then t (n) Proof Let Gˇ be the group given by the presentation of G(MMk ) except the defining relations (G8). Let g be an element of Gˇ which is equal to 1 in G(MMk ). Let w be a word in generators of Gˇ that represents g in Gˇ , |w| = n. Then w is equal in Gˇ to a product of conjugates of relators (G8). This representation can be obtained as follows. Consider a minimal area van Kampen diagram over the presentation of G(MMk ) with boundary label w. We can read off of this diagram a representation of w as a product of conjugates of all relations (G1)–(G8). Now remove from that product all conjugates of relators (G1)–(G7). The remaining product is . Let t (n) be the time function of MMk . Let c be an accepted configuration of MMk such that |c| = n and the length of the computation connecting c with the accept configuration s0 of MMk is t (n). Let w(c) and wg(c) be the corresponding elements in S(MMk ) and Gˇ respectively. Let Tˇ be the normal subgroup of Gˇ generated (as a normal subgroup) by the x -letters. Since the configuration c is accepted by MMk , we have that wg(c)−1wg(s0) = 1 modulo the relations (G8). Therefore wg(c)−1wg(s0) is a product of conjugates of the relations (G8). Since the length of wg(c) does not exceed |c|4 for some uniform constant (see the proof of Theorem 4.18), the number of factors in the product does not exceed d( n4). By Lemma 4.15, wg(c) = x (1, uc), wg(c ) = x (1, uc ) for some words uc, uc in the generators of S(MMk ), and we can rewrite the product and obtain a product of elements of the form x (i , u)x (i , v)−1 where v is obtained from u by applying a Minsky relator of S(MMk ) once. This product is also equal to x (1, uc)x (1, uc )−1. Since elements x (i , u), u ∈ W ∪ W0 form a basis of T viewed as a vector space of the field with p elements (Lemma 4.13), this implies that there exists an accepting computation for the configuration c of length at most d( n4) and inequality (15) follows. Theorem 4.21 For every recursive set of natural numbers X and every recursive function g(n) there exists a finitely presented residually finite solvable of class 3 group G such that the word problem in G is as hard as the membership problem in X and polynomially reduces to it; the Dehn function G is bigger than g(n). Proof By Theorems 2.4 and 2.7 there exists a sym-universally halting 2-glass Minsky machine MM2 whose configuration equivalence problem polynomially reduces to the membership problem for X . By Lemma 3.5, the time complexity of the problem of recognizing equality to 0 in the semigroup S(MM2) is as large as f (n) and the word problem in S(MM2) polynomially reduces to the membership problem for X . The first statement now follows from Theorems 4.18 and 3.9. For the Dehn function part of the theorem we use Lemma 4.20. Thus we just need to modify the Minsky machine MM4 so that the new machine MMm , m > 4, has the same complexity of the configuration equivalence problem but time function larger than g(n4). This can be done in the following straightforward way. After the machine MM4 is supposed to stop, we make MMm compute a recursive function that is greater than g(n4), then stops. We leave it as an exercise for the reader to determine the exact value of m and the program of MMm . 4.2 A residually finite finitely presented group with large depth function Theorem 4.22 For every recursive function f and a recursive set X of natural num bers, one can construct two residually finite finitely presented solvable of class 3 groups G1, G2. Both groups have depth functions greater than f . The group G1 has word problem as hard as the membership problem for X . The group G2 has the word problem decidable in polynomial time. Proof Consider the Minsky machine MM4 constructed in the proof of Theorem 3.12, the semigroup S˜(MMk ) and the corresponding group G˜ (MM4) (it is obtained from G(MM4) by imposing the relation xu = 1 for every u ∈ U containing q0). Let us prove that it is residually finite. Take an element g ∈ G(MM4) such that g = 1. As in the proof of Theorem 4.18 we can assume g ∈ T . By Lemma 4.13, g is a product of elements of the form 14. Hence g = g0g1 where g0 (resp. g1) is a product of elements (14) with u ∈ W0 (resp. W ). Suppose that g1 is not 1. Let T be the subgroup of G˜ (MM4) generated by elements (14) with w ∈ W0. Then T is a normal subgroup of G˜ (MM4) by Lemma 4.12 (and Remark 4.16). Let G˜ (MM4) = G˜ (MM4)/ T . This group is a semidirect product of T / T and the metabelian group H1H2 H1. Let D be the sum of lengths of words u ∈ W that appear in the factors of g1. Let YD be the set of all words in Sˇ where at least one a-letter appears at least D times, and 0. Then YD is an ideal in Sˇ, and the image of the set of elements (14) with w ∈ YD in G˜ (MM4) form a normal subgroup R of G (MN ) of finite index (because T is an Abelian group of finite exponent p). That normal subgroup does not contain g by Theorem 4.3 (c) (and Remark 4.16. Then G˜ (MM4)/R is a semidirect product of a finite group and the metabelian group H1H2 H2. Hence G˜ (MM4)/R is residually finite and g can be separated from 1 by a homomorphism from G˜ (MM4) onto a finite group. Thus we may assume that g1 = 1. Let w1, . . . , wl be the elements from W0 that appear in the representation of g as a product of elements (14). Let E be the set of elements of Sˆ (i.e., words modulo the commutativity relations) which are equal to one of the w j in S(MM4). Note that every word in E starts with a q-letter by definition of elements (14). For each w ∈ E let D(w) be as in the proof of Theorem 3.12. Let D be the maximum of these numbers D(w) and, as in the proof of Theorem 3.12, let S¯ be the semigroup S˜(MM4) with additional defining relations a3D = a32D, a4D = a42D. As we mentioned in the proof of Theorem 3.12, every non-zero element of S¯ has finitely many divisors. It is easy to see that S¯ satisfies Properties (Q1), (Q2) of Lemma 3.15. Therefore by Remark 4.16, we can built a group G¯ starting with S¯ instead of S˜(MM4) and all the statements mentioned in Remark 4.16 remain true for G¯ . The natural homomorphism δ : S˜(MM4) → S¯ extends to a homomorphism δ¯ : G˜ (MMk ) → G¯ . By Theorem 4.3 and Lemma 4.13 the images under δ¯ of all elements (14) with w ∈ {w1, . . . , wl } form a finite linearly independent set. Hence δ¯(g) = 1 in G¯ . Since every non-zero element of S¯ has finitely many divisors, the group G¯ is residually finite by Remark 4.19. Hence there exists a homomorphism of G˜ (MM4) onto a finite groups separating g from 1. Thus G˜ (MM4) is indeed residually finite. The fact that ρG (n) ≥ f (n) is proved the same way as in the proof of Theorem 3.12 (one only needs to replace the product by operation * everywhere in that proof). 5 Applications In this section we present two applications of our results and methods. One application concerns the universal theory of finite solvable groups of a given class, the second application concerns with the membership problem in pro-finitely closed subgroups of residually finite groups. 5.1 Universal theories of sets of finite groups In this section we will prove the following result. For the class of all finite groups in was proved by Slobodskoi [60] (the idea of Slobodskoi’s proof came from Gurevich’s paper [ 23 ] where the same result was proved for semigroups, see also [58]). 2 Theorem 5.1 The universal theories of the class of finite groups from A pA ∩ ZN5A and the class of all periodic groups are recursively inseparable. In particular, the universal theory of any set of finite groups containing all finite solvable of class 3 groups is undecidable. Proof Consider a partial function f : N → N such that f is one-to-one and computable on its domain, and the domain is recursively enumerable but not recursive. Let MM2 be a Minsky machine computing the function f . We can assume that for every number not in the domain of f the machine MM2 works indefinitely long and never stops. As in the proof of Theorem 3.12, we can assume that the start command number 1 cannot be used in the middle of a computation of MM2. Consider the 4-glass Minsky machine MM4 described in the proof of Theorem 3.12. Let S (MM4) be the semigroup given by the same defining relations as S˜(MM4) except the relation q0 = 0 is substituted by the relation qi A3 A4 = 0 for every i . Let G (MM4) be the group corresponding to S (MM4) in the same way G(MMk ) corresponds to S(MMk ). Since Properties (Q1), (Q2) of Lemma 3.15 obviously hold 2 for S (MM4) we can use Remark 4.16. Then G (MM4) belongs to A pA ∩ ZN5A and simulates MM4 as described in Theorem 4.3. Let R be the (finite) set of defining relations of G (MM4). Let X be the set of numbers such that MM4 accepts the configuration ( , 0, 0, 0). Let X be the set of numbers such that MM4 works infinitely long starting with the configuration ( , 0, 0, 0). Then X and X are recursively inseparable by the choice of MM2 and MM4. For any configuration ( , 0, 0, 0) of MM4 consider the corresponding element w( ) = q1 ∗ a1( ) ∗ A1 ∗ A2 ∗ a3 ∗ A3 ∗ A4. Suppose ∈ X . Then the Minsky machine MM2 halts starting at ( , 0, 0, 0). Since the function f is one-to-one, there are only finite number of computations of Sym(MM2) starting at the configuration c = (1; , 0, 0, 0). Then as in the proof of Theorem 4.22, we can find a a homomorphism φ from G˜ (MM4) to a finite group such that φ (x (1, w)) = 1. Hence the universal formula & R → x (1, w) = 1 does not hold in the finite group 2 H from A pA ∩ ZN5A. Now suppose that ∈ X . Consider any periodic homomorphic image H of G (MM4). Let t¯ be the image of t ∈ G (Mn) in H . Then there exists a number D such that for every element x ∈ T¯ , x ∗ a¯ 3(D) = x ∗ a¯ 3(2D). (16) Since MM4 works infinitely long starting at the configuration ( , 0, 1, 0), by Theorem 4.3 the following equality is true for some i, k1, k2: w( ) = x¯(1, qi A0) ∗ a¯ 1(k1) ∗ a¯ 2(k2) ∗ a¯ 3(D) ∗ A¯1 ∗ A¯2 ∗ A¯3 ∗ A¯4. Then by (16) and Theorem 4.3 = x¯1,qi A0 ∗ a¯ 1(k1) ∗ a¯ 2(k2) ∗ A¯1 ∗ A¯2 ∗ A¯3 ∗ A¯4 = 1. w¯ ( ) = x¯1,qi A0 ∗ a¯ 1(k1) ∗ a¯ 2(k2) ∗ a¯ 3(2D) ∗ a¯ 4(D) ∗ A¯1 ∗ A¯2 ∗ A¯3 ∗ A¯4 = x¯1,qi A0 ∗ a¯ 1(k1) ∗ a¯ 2(k2) ∗ a¯ 3(D) ∗ a¯ 4(D) ∗ A¯1 ∗ A¯2 ∗ A¯3 ∗ A¯4 since qi A3 A4 = 0 in S (Mn). Hence the universal formula & R → w( ) = 1 holds in H . Thus the set of universal formulas &R → w( ) = 1 that do not hold in some finite 2 group from A pA ∩ ZN5A and the set of such formulas which hold in every periodic group are recursively inseparable. Remark 5.2 Note that the universal theory of finite metabelian groups is decidable because every finitely generated metabelian group is residually finite (the connection is explained in [ 32 ]). On the other hand, the universal theory of all finite nilpotent groups is undecidable [ 30 ]. The description of all (finitely based) varieties of groups where the universal theory of finite groups is decidable is currently out of reach. In fact our Theorem 5.1 gives the first example of a proper variety of groups where the universal theory of finite groups is undecidable. From Zelmanov’s solution of the restricted Burnside problem [65,66], it immediately follows that the universal theory of finite groups in every finitely based periodic variety of groups is decidable. That result and simulations of Minsky machines in semigroups (as in Sect. 3) were used by the third author [56] to obtain a complete description of all finitely based varieties of semigroups where finite semigroups have decidable universal theory. For more information on that problem, see [ 32 ]. 5.2 Distortion of pro-finitely closed subgroups of finitely presented groups Let G be a group generated by a finite set X , H ≤ G be a subgroup generated by a finite set Y . Recall that the distortion function f H,G (n) is defined as the minimal number k such that every element of H represented as a word w of length ≤n in the alphabet X can be represented as a word of length ≤k in the alphabet Y [ 17 ]. Distortion functions with respect to two different sets of generators for the same group are equivalent. By [ 17 ] in a group G with decidable word problem, the distortion function fG,H is recursive if and only if the membership problem in H is decidable. Recall that H is closed in the pro-finite topology of G if H is the intersection of some subgroups of G of finite index. If G is finitely presented and H is closed in the pro-finite topology of G, then there exists a McKinsey-type algorithm A(G, H ) solving the membership problem for H (and thus the fG,H is recursive). For every word w in the alphabet X , the “yes” part Ayes(G, H ) of the algorithm lists all words in Y , rewrites them as words in X , and then applies relations of G to check whether one of these words is equal to w. The “no” part Ano(G, H ) of the algorithm lists all homomorphisms φ of G into finite groups and checks whether φ (w) ∈/ φ (H ). As in Sect. 1.2, one can asks what is the complexity of the “yes” and “no” parts of that algorithm, in particular, and of the membership problem for H in general. The time complexity of Ayes(G, H ) can be estimated in terms of the distortion function fG,H (n) and the time complexity of Ano(G, H ) can be estimated estimated in terms of the relative depth function ρG,H (n) which is defined as the minimal number r such that for every word w of length ≤n in X which does not represent an element of H there exists a homomorphism φ from G to a finite group of order ≤r such that φ (w) ∈/ φ (H ). As for the word problem in residually finite finitely presented groups (discussed above), there were no examples of finitely generated subgroups of finitely presented groups that are closed in the pro-finite topology but have “arbitrary bad” distortion or “arbitrary bad” relative depth function. Mikhailova’s construction [ 42 ] shows that finitely generated subgroups of the residually finite group F2 × F2 (here F2 is a free group of rank 2) could be very distorted. In fact the set of possible distortion functions of subgroups of F2 × F2 coincides, up to a natural equivalence, with the set of Dehn functions of finitely presented groups [47]. Finitely generated subgroups of F2 × F2 are equalizers of pairs of homomorphisms φ : Fk → G, ψ : Fn → G (where Fk , Fn are some subgroups of F2), i.e. the subgroups of the form {(x , y) ∈ Fk × Fn | φ (x ) = ψ (y)} (see, for example, [52] or[5]). The equalizer subgroup is finitely generated if and only if G is finitely presented [5]. It is easy to prove (see Lemma 5.3 below) that if G is residually finite, then the equalizer is closed in the pro-finite topology of F2 × F2. In fact we have the following more general statement: Lemma 5.3 Let P be a class of finite groups closed under direct products and subgroups. Let G be a finitely generated group, let N be a normal subgroup of G, and let φ, ψ be two homomorphisms G → G/N . If G/N is residually P, then the equalizer E (φ, ψ ) = {(g, h ∈ G × G | φ (g) = ψ (h)} is closed in the pro-P topology on G × G. Proof Suppose (u, v) ∈ G × G but (u, v) ∈/ E (φ , ψ ), so φ (u) = ψ (v). Since G/N is residually P there is a homomorphism η : G/N → K onto a finite group K ∈ P such that ηφ (u) = ηψ (v) in K . Therefore the image of the pair (u, v) under (ηφ , ηψ ) is not in the image of the subgroup E (φ , ψ ) in K × K . Hence the subgroup E (φ , ψ ) is closed in the pro-P topology on G × G. Lemma 5.3 and Theorems 4.22 and 4.21 immediately imply Corollary 5.4 For every recursive function f (n) there exists a finitely generated sub group H ≤ F2 × F2 that is closed in the pro-finite topology of F2 × F2 and whose distortion function f F2×F2,H , the relative depth function, and the time complexity of the membership problem are at least f (n). Remark 5.5 Since the groups we construct are solvable of class 3, a similar corollary is true with F2 replaced by the free solvable group of class 3 of finite rank (although the rank is not necessarily 2 because not every free solvable group of class 3 embeds into a 2-generated group that is solvable of class 3). Acknowledgements The authors are grateful to Jean-Camille Birget and Friedrich Otto for pointing to the references [13], to Ben Steinberg for pointing to the reference [ 34 ],to Rostislav Grigorchuk for pointing to the references [ 15,21 ] and to Tim Riley for pointing to the references [ 19,20 ]. We are also grateful to Markus Lohrey and Ralph Strebel for their comments. We are especially grateful to the anonymous referees whose numerous suggestions helped us improve the paper. 1. Agol, I.: The virtual Hacken conjecture (with an appendix by I. Agol, D. Groves and J. Manning), arXiv:1204.2810, (2012) 2. Baumslag, G.: A non-cyclic one-relator group all of whose finite factor groups are cyclic. J. Aust. Math. Soc. 10, 497–498 (1969) 3. Baumslag, G.: Subgroups of finitely presented metabelian groups. J. Aust. Math. Soc. Ser. A 16(1), 98–110 (1973) 4. Baumslag, G., Miller III, C.F., Short, H.: Isoperimetric inequalities and the homology of groups. Invent. Math. 113(3), 531–560 (1993) 5. Baumslag, G., Roseblade, J.E.: Subgroups of direct products of free groups. J. Lond. Math. Soc. 30, 44–52 (1984) 6. Borisov, A., Sapir, M.: Polynomial maps over finite fields and residual finiteness of mapping tori of group endomorphisms. Invent. Math. 160(2), 341–356 (2005) 7. Borisov, A., Sapir, M.: Polynomial maps over p-adics and residual properties of mapping tori of group endomorphisms. Int. Math. Res. Not. IMRN 16, 3002–3015 (2009) 8. Birget, J.-C.: Infinite string rewriting systems and complexity. J. Symb. Comput. 25(6), 759–793 (1998) 9. Bou-Rabee, K.: Quantifying residual finiteness. J. Algebra 323, 729–737 (2010) 10. Brady, N., Dison, W., Riley, T.: Hyperbolic hydra. Groups Geom. Dyn. 7(4), 961–976 (2013) 11. Bridson, M., Haefliger, A.: Metric Spaces of Non-Positive Curvature. Springer, Berlin (1999) 12. Cohen, D.E.: Combinatorial Group Theory: A Topological Approach. London Mathematical Society Student Texts, 14. Cambridge University Press, Cambridge (1989) 13. Davis, M.D.: A note on universal Turing machines. Automata studies, Annals of Mathematics Studies, no. 34, pp. 167–175. Princeton University Press, Princeton (1956) 44. Nikolov, N., Segal, D.: Finite index subgroups in pro-finite groups. C. R. Math. Acad. Sci. Paris 337(5), 303–308 (2003) 45. Ollivier, Y., Wise, D.T.: Cubulating random groups at density less than 1/6. Trans. Am. Math. Soc. 363(9), 4701–4733 (2011) 46. Olshanskii, AYu.: Almost every group is hyperbolic. Int. J. Algebra Comput. 2(1), 1–17 (1992) 47. Olshanskii, A., Sapir, M.: Length and area functions on groups and quasi-isometric Higman embeddings. Int. J. Algebra Comput. 11, 137–170 (2001) 48. Papadimitriou, C.H.: Computational Complexity. Addison-Wesley Publishing Company, Reading (1994) 49. Pueschel, K.: Hydra group doubles are not residually finite, arXiv:1507.02554 50. Platonov, A.N.: An isoperametric function of the Baumslag–Gersten group. Vestnik Moskov. Univ. Ser. I Mat. Mekh. 2004, no. 3, pp. 12–17, translation in Moscow Univ. Math. Bull. 59 (2004), no. 3, p. 1217 (2005) 51. Remeslennikov, V.: Studies on infinite solvable and finitely approximable groups. Mat. Zametki 17(5), 819–824 (1975) 52. Remak, R.: Uber der Zerlegung der endlichen Gruppen in direkte unzerlegbare Faktoren. J. Reine Angew. Math. 139, 293308 (1911) 53. Rips, E.: Subgroups of small cancellation groups. Bull. Lond. Math. Soc. 14, 45–47 (1982) 54. Rotman, J.J.: An Introduction to the Theory of Groups, 4th edn. Graduate Texts in Mathematics, 148. Springer, New York (1995) 55. Sapir, M.: Algorithmic problems in varieties of semigroups. Algebra i Logika 27(4), 440–463 (1988) 56. Sapir, M.: Weak word problem for finite semigroups. Monoids and Semigroups with Applications (Berkeley, CA, 1989), pp. 206–219. World Science Publisher, River Edge (1991) 57. Sapir, M.: Asymptotic invariants, complexity of groups and related problems. Bull. Math. Sci. 1(2), 277–364 (2011) 58. Sapir, M.: Minsky machines and algorithmic problems, accepted in Essays Dedicated to Yuri Gurevich on the Occasion of His 75th Birthday, LNCS. Springer, Berlin (2015) 59. Sapir, M., Birget, J.C., Rips, E.: Isoperimetric and isodiametric functions of groups. Ann. Math. 156(2), 345–466 (2002) 60. Slobodskoi, A.M.: Undecidability of the universal theory of finite groups. Algebra Log. 20(2), 207–230 (1981) 61. Waack, St: On the parallel complexity of linear groups. RAIRO Inform. Theor. Appl. 25, 323–354 (1991) 62. Wehrfritz, B.A.F.: On finitely generated soluble linear groups. Math. Z. 170(2), 155–167 (1980) 63. Wise, D.T.: The structure of groups with a quasiconvex hierarhy. Preprint (2011) 64. Wise, D.T.: A residually finite version of Rips’s construction. Bull. Lond. Math. Soc. 35(1), 23–29 (2003) 65. Zel’manov, E.I.: The solution of the restricted Burnside problem for groups of odd exponent. Izv. Akad. Nauk. SSSR. Ser. Mat., 54(1), 42–59 (1990). Transl. in Math. USSR-Izv. 36(1), 41–60 (1991) 66. Zel’manov, E.I.: The solution of the restricted Burnside problem for 2-groups. Mat. Sb. 182(4), 568–592 (1991) 14. Dison , W. , Riley , T. : Hydra groups . Comment. Math. Helv . 88 ( 3 ), 507 - 540 ( 2013 ) 15. Dyson , V.H.: A family of groups with nice word problems. Collection of articles dedicated to the memory of Hanna Neumann, VIII . J. Aust. Math. Soc . 17 , 414 - 425 ( 1974 ) 16. Ershov , M. : Golod-Shafarevich groups: a survey . Int. J. Algebra Comput . 22 ( 5 ), 1230001 ( 2012 ) 17. Farb , B. : The extrinsic geometry of subgroups and the generalized word problem . Proc. Lond. Math. Soc . 68 ( 3 ), 577 - 593 ( 1994 ) 18. Gersten , S.M.: Dehn functions and l1-norms of finite presentations. Algorithms and Classification in Combinatorial Group Theory, pp. 195 - 225 . Springer, Berlin ( 1992 ) 19. Gersten , S.M. : Isoperimetric and isodiametric functions of finite presentations . Geometric Group Theory , vol. 1 ( Sussex , 1991 ), pp. 79 - 96 , London Math. Soc. Lecture Note Ser., 181 , Cambridge University Press, Cambridge ( 1993 ) 20. Gersten , S.M. , Riley , T.R.: Some duality conjectures for finite graphs and their group theoretic consequences . Proc. Edinb. Math. Soc . 48 ( 2 ), 389 - 421 ( 2005 ) 21. Grigorchuk , R.: Groups with intermediate growth functions and their applications , Doctor's Thesis (Russian) , Moscow Steklov Mathematical Institute ( 1985 ) 22. Golubov , E.A. : Finite separability in semigroups . Dokl. Akad. Nauk SSSR 189 , 20 - 22 ( 1969 ) 23. Gurevich , Y.S.: The problem of equality of words for certain classes of semigroups . Algebra i Log . Sem . 5 ( 5 ), 25 - 35 ( 1966 ) 24. Higman , G.: Subgroups of finitely presented groups . Proc. R. Soc. Ser. A 262 , 455 - 475 ( 1961 ) 25. Hsu , T. , Wise , D. : Cubulating graphs of free groups with cyclic edge groups . Amer. J . Math. 132 ( 5 ), 1153 - 1188 ( 2010 ) 26. Kassabov , M. , Matucci , F. : Bounding the residual finiteness of free groups . Preprint , arXiv 27. Kassabov , M. , Nikolov , N.: Generation of polycyclic groups . J. Group Theory 12 ( 4 ), 567 - 577 ( 2009 ) 28. Kharlampovich , O.G. : Finitely presented solvable group with unsolvable word problem . Sov. Math. Izvest . 45 ( 4 ), 852 - 873 ( 1981 ) 29. Kharlampovich , O.G. : The word problem for groups and Lie algebras , Doctor's Thesis (Russian) , Moscow Steklov Mathematical Institute ( 1990 ) 30. Kharlampovich , O.G. : The universal theory of the class of finite nilpotent groups is undecidable . Mat. Zametki 33 ( 4 ), 499 - 516 ( 1983 ) 31. Kharlampovich , O.G. , Sapir , M.V. : A non-residually finite, relatively finitely presented group in the variety N2A . Combinatorial and Geometric Group Theory (Edinburgh , 1993 ), pp. 184 - 189 , London Math. Soc. Lecture Note Ser., 204 , Cambridge Universiy Press, Cambridge ( 1995 ) 32. Kharlampovich , O. , Sapir , M. : Algorithmic problem in varieties . Int. J. Algebra Comput . 5 ( 4-5 ), 379 - 602 ( 1995 ) 33. Kourovskaja tetrad ' (Unsolved Problems in Group Theory), 5th edn. Novosibirsk , ( 1976 ) 34. Lipton , R.J. , Zalcstein , Y. : Word problems solvable in logspace . J. Assoc. Comput. Mach . 24 , 522 - 526 ( 1977 ) 35. Malcev , A.I. : Algorithms and Recursive Functions . Nauka, Moscow ( 1965 ) 36. Malcev , A.I. : On Homomorphisms onto finite groups (Russian) . Uchen. Zap. Ivanovskogo Gos. Ped. Inst . 18 ( 1958 ), pp. 49 - 60 . English translation in : Amer. Math. Soc. Transl. Ser . 2 , 119 , pp. 67 - 79 ( 1983 ) 37. McKenzie , R. , Thompson , R.J.: An elementary construction of unsolvable word problems in group theory. Word problems: decision problems and the Burnside problem in group theory (Conf ., Univ. California, Irvine, Calif. 1969 ; dedicated to Hanna Neumann), Studies in Logic and the Foundations of Math., 71 , p. 457478 . Amsterdam ( 1973 ) 38. Meskin , S.: A finitely generated residually finite group with an unsolvable word problem . Proc. Am. Math. Soc . 43 ( 1 ), 8 - 10 ( 1974 ) 39. Madlener , K. , Otto , F. : Pseudonatural algorithms for the word problem for finitely presented monoids and groups . J. Symb. Comput . 1 ( 4 ), 383 - 418 ( 1985 ) 40. McKinsey , J.: The decision problem for some classes of sentences without quantifiers . J. Symb. Log . 8 , 61 - 76 ( 1973 ) 41. Miasnikov , A. , Ushakov , A. , Won , D. : The word problem in Baumslag group is polynomial time decidable . J. Algebra 345 , 324 - 342 ( 2011 ) 42. Mikhailova , K.A. : The occurrence problem for direct products of groups . Dokl. Akad. Nauk SSSR 119 , 1103 - 1105 ( 1958 ) 43. Neumann , H.: Varieties of Groups. Springer, Berlin ( 1967 )


This is a preview of a remote PDF: https://link.springer.com/content/pdf/10.1007%2Fs13373-017-0103-z.pdf

Olga Kharlampovich, Alexei Myasnikov, Mark Sapir. Algorithmically complex residually finite groups, Bulletin of Mathematical Sciences, 2017, 309-352, DOI: 10.1007/s13373-017-0103-z