Lecture 8, 9: Tarski problems and limit groups

Transcription

1 Lecture 8, 9: Tarski problems and limit groups Olga Kharlampovich October 21, 28 1 / 51

2 Fully residually free groups A group G is residually free if for any non-trivial g G there exists φ Hom(G, F ), where F is a free group, such that g φ 1. A group G is fully residually free or discriminated by F if for any finite subset M G with 1 M there exists an F -homomorphism φ : G F such that 1 φ(m). Fg such groups are called limit groups. Examples: Fg free groups, free abelian groups, surface groups, extensions of centralizers of fully residually free groups, subgroups of fully residually free groups. 2 / 51

3 Fully residually free groups A group G is residually free if for any non-trivial g G there exists φ Hom(G, F ), where F is a free group, such that g φ 1. A group G is fully residually free or discriminated by F if for any finite subset M G with 1 M there exists an F -homomorphism φ : G F such that 1 φ(m). Fg such groups are called limit groups. Examples: Fg free groups, free abelian groups, surface groups, extensions of centralizers of fully residually free groups, subgroups of fully residually free groups. 2 / 51

4 Fully residually free groups Easy exercises: 1) Residually free groups are torsion free; 2) Fg free groups and free abelian groups are fully residually free; 3) Fg subgroups of fully residually free groups are fully residually free. 4) Fully residually free groups are commutative transitive; 5) F F is NOT fully residually free. 3 / 51

5 Description G is an extension of centralizer of H if G = H, t [C H (u), t] = 1 Lemma If H is a limit group, then G is a limit group. Proof φ : G H is identical on H and t φ = u n, for a large n. Elements in G have canonical form g 1 t m 1 g 2 t m 2... t m k g k+1, where g i C H (u), i = 1,..., k. They are mapped into g 1 u nm 1 g 2 u nm 2... u nm k g k+1. Theorem[Kharlampovich, Miasnikov, 96] That s it. Every f.g. fully residually free group is a subgroup of a group obtained from a free group as a finite series of extensions of centralizers. Moreover, there is an algorithm to find this embedding. 4 / 51

6 Examples Examples of sentences in the theory of F : ( Vaught s identity) x y z(x 2 y 2 z 2 = 1 ([x, y] = 1&[x, z] = 1&[y, z] = 1)) (Torsion free) x(x n = 1 x = 1) (Commutation transitivity) x y z((x 1&y 1&z 1&[x, y] = 1 &[x, z] = 1) [y, z] = 1) CT doesn t hold in F 2 F 2. (CSA) x y([x, x y ] = 1 [x, y] = 1) x, y z(xy = yx (x = z 2 y = z 2 xy = z 2 )) not true in a free abelian group of rank 2. 5 / 51

7 Examples x y z([x, y] = 1 (x = z 2 y = z 2 xy = z 2 )) This implies that if a group G is equivalent to F, then it does not have non-cyclic abelian subgroups. F has Magnus property, namely, for any n, m the following sentence is true: x y( z 1,..., z m+n (x = Π n i=1 z 1 i y ±1 z i y = Π m+n i=n+1 z 1 i x ±1 z i ) z(x = z 1 y ±1 z)) 6 / 51

8 Elementary theory The elementary theory Th(G) of a group G is the set of all first order sentences in the language of group theory which are true in G. Recall, that the group theory language consists of multiplication, inversion 1, and the identity symbol 1. Every group theory sentence is equivalent to one of the type: Φ = X 1 Y 1... X k Y k r s ( u pi (X 1, Y 1,..., X k, Y k ) = 1 p=1 i=1 t v pj (X 1, Y 1,..., X k, Y k ) 1). j=1 where u pi, v pj are group words (products of variables and their inverses). 7 / 51

9 Elementary theory Informally: Th(G) = all the information about G that can be expressed in the first-order logic of group theory. Hence: Th(G) = Th(H) G and H are indistinguishable in the first-order logic (elementary equivalent) Th(G) is decidable the first-order information about G is available (in principle) to us. 8 / 51

10 Tarski s Problems In 1945 Alfred Tarski posed the following problems. Do the elementary theories of free non-abelian groups F n and F m coincide? Is the elementary theory of a free non-abelian group F n decidable? 9 / 51

11 Tarski s Problems In 1945 Alfred Tarski posed the following problems. Do the elementary theories of free non-abelian groups F n and F m coincide? Is the elementary theory of a free non-abelian group F n decidable? 9 / 51

12 Tarski s Problems In 1945 Alfred Tarski posed the following problems. Do the elementary theories of free non-abelian groups F n and F m coincide? Is the elementary theory of a free non-abelian group F n decidable? 9 / 51

13 Solutions Theorem [Kharlampovich and Myasnikov ( ), independently Sela ( )] Th(F n ) = Th(F m ), m, n > 1. Theorem [Kharlampovich and Myasnikov] The elementary theory Th(F ) of a free group F even with constants from F in the language is decidable. 10 / 51

14 Solutions Theorem [Kharlampovich and Myasnikov ( ), independently Sela ( )] Th(F n ) = Th(F m ), m, n > 1. Theorem [Kharlampovich and Myasnikov] The elementary theory Th(F ) of a free group F even with constants from F in the language is decidable. 10 / 51

15 Tarski s Type Problems Long history of Tarski s type problems in algebra. Crucial results on fields, groups, boolean algebras, etc. 11 / 51

16 Tarski s Type Problems Long history of Tarski s type problems in algebra. Crucial results on fields, groups, boolean algebras, etc. 11 / 51

17 Tarski s Type Problems: Complex numbers Complex numbers C Th(C) = Th(F ) iff F is an algebraically closed field. Th(C) is decidable. This led to development of the theory of algebraically closed fields. Elimination of quantifiers: every formula is logically equivalent (in the theory ACF) to a boolean combination of quantifier-free formulas (something about systems of equations). 12 / 51

18 Tarski s Type Problems: Complex numbers Complex numbers C Th(C) = Th(F ) iff F is an algebraically closed field. Th(C) is decidable. This led to development of the theory of algebraically closed fields. Elimination of quantifiers: every formula is logically equivalent (in the theory ACF) to a boolean combination of quantifier-free formulas (something about systems of equations). 12 / 51

19 Tarski s Type Problems: Reals Reals R Th(R) = Th(F ) iff F is a real closed field. Th(R) is decidable. A real closed field = an ordered field where every odd degree polynomial has a root and every element or its negative is a square. Theory of real closed fields (Artin, Schreier), 17th Hilbert Problem (Artin). Elimination of quantifiers (to equations): every formula is logically equivalent (in the theory RCF) to a boolean combination of quantifier-free formulas. 13 / 51

20 Tarski s Type Problems: Reals Reals R Th(R) = Th(F ) iff F is a real closed field. Th(R) is decidable. A real closed field = an ordered field where every odd degree polynomial has a root and every element or its negative is a square. Theory of real closed fields (Artin, Schreier), 17th Hilbert Problem (Artin). Elimination of quantifiers (to equations): every formula is logically equivalent (in the theory RCF) to a boolean combination of quantifier-free formulas. 13 / 51

21 Tarski s Type Problems: Reals Reals R Th(R) = Th(F ) iff F is a real closed field. Th(R) is decidable. A real closed field = an ordered field where every odd degree polynomial has a root and every element or its negative is a square. Theory of real closed fields (Artin, Schreier), 17th Hilbert Problem (Artin). Elimination of quantifiers (to equations): every formula is logically equivalent (in the theory RCF) to a boolean combination of quantifier-free formulas. 13 / 51

22 Tarski s Type Problems: Reals Reals R Th(R) = Th(F ) iff F is a real closed field. Th(R) is decidable. A real closed field = an ordered field where every odd degree polynomial has a root and every element or its negative is a square. Theory of real closed fields (Artin, Schreier), 17th Hilbert Problem (Artin). Elimination of quantifiers (to equations): every formula is logically equivalent (in the theory RCF) to a boolean combination of quantifier-free formulas. 13 / 51

23 Tarski s Type Problems: p-adics Ax-Kochen, Ershov p-adics Q p Th(Q p ) = Th(F ) iff F is p-adically closed field. Th(Q p ) is decidable. Existence of roots of odd degree polynomials in R Hensel s lemma in Q p. Elimination of quantifiers (to equations). 14 / 51

24 Tarski s Type Problems: p-adics Ax-Kochen, Ershov p-adics Q p Th(Q p ) = Th(F ) iff F is p-adically closed field. Th(Q p ) is decidable. Existence of roots of odd degree polynomials in R Hensel s lemma in Q p. Elimination of quantifiers (to equations). 14 / 51

25 Tarski s Type Problems: Groups Tarski s problems are solved for abelian groups (Tarski, Szmielew). For non-abelian groups results are sporadic. Novosibirsk Theorem [Malcev, Ershov, Romanovskii, Noskov] Let G be a finitely generated solvable group. Then Th(G) is decidable iff G is virtually abelian (finite extension of an abelian group). Elementary theories of free semigroups of different ranks are different and undecidable Interpretation of arithmetic. 15 / 51

26 Tarski s Type Problems: Groups Tarski s problems are solved for abelian groups (Tarski, Szmielew). For non-abelian groups results are sporadic. Novosibirsk Theorem [Malcev, Ershov, Romanovskii, Noskov] Let G be a finitely generated solvable group. Then Th(G) is decidable iff G is virtually abelian (finite extension of an abelian group). Elementary theories of free semigroups of different ranks are different and undecidable Interpretation of arithmetic. 15 / 51

27 Tarski s Type Problems: Groups Tarski s problems are solved for abelian groups (Tarski, Szmielew). For non-abelian groups results are sporadic. Novosibirsk Theorem [Malcev, Ershov, Romanovskii, Noskov] Let G be a finitely generated solvable group. Then Th(G) is decidable iff G is virtually abelian (finite extension of an abelian group). Elementary theories of free semigroups of different ranks are different and undecidable Interpretation of arithmetic. 15 / 51

28 Tarski s Type Problems: Groups Tarski s problems are solved for abelian groups (Tarski, Szmielew). For non-abelian groups results are sporadic. Novosibirsk Theorem [Malcev, Ershov, Romanovskii, Noskov] Let G be a finitely generated solvable group. Then Th(G) is decidable iff G is virtually abelian (finite extension of an abelian group). Elementary theories of free semigroups of different ranks are different and undecidable Interpretation of arithmetic. 15 / 51

29 Tarski s Type Problems: elementary classification Remeslennikov, Myasnikov, Oger: elementary classification of nilpotent groups (not finished yet). Typical results: [Myasnikov, Oger] Finitely generated non-abelian nilpotent groups G and H are elementarily equivalent iff G Z H Z Many subgroups are definable by first order formulas, many algebraic invariants are definable. 16 / 51

30 Tarski s Type Problems: elementary classification Remeslennikov, Myasnikov, Oger: elementary classification of nilpotent groups (not finished yet). Typical results: [Myasnikov, Oger] Finitely generated non-abelian nilpotent groups G and H are elementarily equivalent iff G Z H Z Many subgroups are definable by first order formulas, many algebraic invariants are definable. 16 / 51

31 Tarski s Type Problems: elementary classification Remeslennikov, Myasnikov, Oger: elementary classification of nilpotent groups (not finished yet). Typical results: [Myasnikov, Oger] Finitely generated non-abelian nilpotent groups G and H are elementarily equivalent iff G Z H Z Many subgroups are definable by first order formulas, many algebraic invariants are definable. 16 / 51

32 Tarski Problems: in free groups Nothing like that in free groups: no visible logical invariants. Ranks of free non-abelian groups are not definable. Only maximal cyclic subgroups are definable. Elimination of quantifiers (as we know now): to boolean combinations of -formulas! New methods appeared. It seems these methods allow one to deal with a wide class of groups which are somewhat like free groups: hyperbolic, relatively hyperbolic, acting nicely on Λ-hyperbolic spaces, etc. 17 / 51

33 Tarski Problems: in free groups Nothing like that in free groups: no visible logical invariants. Ranks of free non-abelian groups are not definable. Only maximal cyclic subgroups are definable. Elimination of quantifiers (as we know now): to boolean combinations of -formulas! New methods appeared. It seems these methods allow one to deal with a wide class of groups which are somewhat like free groups: hyperbolic, relatively hyperbolic, acting nicely on Λ-hyperbolic spaces, etc. 17 / 51

34 Tarski Problems: in free groups Nothing like that in free groups: no visible logical invariants. Ranks of free non-abelian groups are not definable. Only maximal cyclic subgroups are definable. Elimination of quantifiers (as we know now): to boolean combinations of -formulas! New methods appeared. It seems these methods allow one to deal with a wide class of groups which are somewhat like free groups: hyperbolic, relatively hyperbolic, acting nicely on Λ-hyperbolic spaces, etc. 17 / 51

35 Tarski Problems: in free groups Nothing like that in free groups: no visible logical invariants. Ranks of free non-abelian groups are not definable. Only maximal cyclic subgroups are definable. Elimination of quantifiers (as we know now): to boolean combinations of -formulas! New methods appeared. It seems these methods allow one to deal with a wide class of groups which are somewhat like free groups: hyperbolic, relatively hyperbolic, acting nicely on Λ-hyperbolic spaces, etc. 17 / 51

36 Algebraic sets G - a group generated by A, F (X ) - free group on X = {x 1, x 2,... x n }. A system of equations S(X, A) = 1 in variables X and coefficients from G (viewed as a subset of G F (X )). A solution of S(X, A) = 1 in G is a tuple (g 1,..., g n ) G n such that S(g 1,..., g n ) = 1 in G. V G (S), the set of all solutions of S = 1 in G, is called an algebraic set defined by S. 18 / 51

37 Algebraic sets G - a group generated by A, F (X ) - free group on X = {x 1, x 2,... x n }. A system of equations S(X, A) = 1 in variables X and coefficients from G (viewed as a subset of G F (X )). A solution of S(X, A) = 1 in G is a tuple (g 1,..., g n ) G n such that S(g 1,..., g n ) = 1 in G. V G (S), the set of all solutions of S = 1 in G, is called an algebraic set defined by S. 18 / 51

38 Algebraic sets G - a group generated by A, F (X ) - free group on X = {x 1, x 2,... x n }. A system of equations S(X, A) = 1 in variables X and coefficients from G (viewed as a subset of G F (X )). A solution of S(X, A) = 1 in G is a tuple (g 1,..., g n ) G n such that S(g 1,..., g n ) = 1 in G. V G (S), the set of all solutions of S = 1 in G, is called an algebraic set defined by S. 18 / 51

39 Algebraic sets G - a group generated by A, F (X ) - free group on X = {x 1, x 2,... x n }. A system of equations S(X, A) = 1 in variables X and coefficients from G (viewed as a subset of G F (X )). A solution of S(X, A) = 1 in G is a tuple (g 1,..., g n ) G n such that S(g 1,..., g n ) = 1 in G. V G (S), the set of all solutions of S = 1 in G, is called an algebraic set defined by S. 18 / 51

40 Radicals and coordinate groups The maximal subset R(S) G F (X ) with V G (R(S)) = V G (S) is the radical of S = 1 in G. The quotient group G R(S) = G[X ]/R(S) is the coordinate group of S = 1. Solutions of S(X ) = 1 in G G-homomorphisms G R(S) G. 19 / 51

41 Radicals and coordinate groups The maximal subset R(S) G F (X ) with V G (R(S)) = V G (S) is the radical of S = 1 in G. The quotient group G R(S) = G[X ]/R(S) is the coordinate group of S = 1. Solutions of S(X ) = 1 in G G-homomorphisms G R(S) G. 19 / 51

42 Radicals and coordinate groups The maximal subset R(S) G F (X ) with V G (R(S)) = V G (S) is the radical of S = 1 in G. The quotient group G R(S) = G[X ]/R(S) is the coordinate group of S = 1. Solutions of S(X ) = 1 in G G-homomorphisms G R(S) G. 19 / 51

43 Zariski topology Zariski topology Zariski topology on G n formed by algebraic sets in G n as a sub-basis for the closed sets. Zariski topology on F with constants in the language: closed sets = algebraic sets. Coordinate groups of algebraic sets over F are residually free. G is residually free if for any non-trivial g G there exists a homomorphism φ Hom(G, F ) such that g φ / 51

44 Irreducible components Equationally Noetherian groups The following conditions are equivalent: G is equationally Noetherian, i.e., every system S(X ) = 1 over G is equivalent to some finite part of itself. the Zariski topology over G n is Noetherian for every n, i.e., every proper descending chain of closed sets in G n is finite. Every chain of proper epimorphisms of coordinate groups over G is finite. 21 / 51

45 Equationally Noetherian groups [R. Bryant (1977), V.Guba (1986)]: Free groups are equationally Noetherian. Proof Let H 0 H 1... be a sequence of epimorphisms between fg groups. Then the sequence Hom(H 0, F ) Hom(H 1, F )... eventually stabilizes because we embed F in SL 2 (Q) and the sequence of algebraic varieties Hom(H 0, SL 2 (Q)) Hom(H 1, SL 2 (Q))... eventually stabilizes. Theorem Linear groups over a commutative, Noetherian, unitary ring are equationally Noetherian. 22 / 51

46 Equationally Noetherian groups [Dahmani, Groves, 2006] Torsion free relatively hyperbolic groups with abelian parabolics are equationally Noetherian (for torsion free hyperbolic groups proved by Sela). 23 / 51

47 Irreducible components If the Zariski topology is Noetherian then every algebraic set can be uniquely presented as a finite union of its irreducible components: V = V 1... V k. Recall, that a closed subset V is irreducible if it is not a union of two proper closed (in the induced topology) subsets. The following is an immediate corollary of the decomposition of algebraic sets into their irreducible components. 24 / 51

48 Irreducible components If the Zariski topology is Noetherian then every algebraic set can be uniquely presented as a finite union of its irreducible components: V = V 1... V k. Recall, that a closed subset V is irreducible if it is not a union of two proper closed (in the induced topology) subsets. The following is an immediate corollary of the decomposition of algebraic sets into their irreducible components. 24 / 51

49 Irreducible components If the Zariski topology is Noetherian then every algebraic set can be uniquely presented as a finite union of its irreducible components: V = V 1... V k. Recall, that a closed subset V is irreducible if it is not a union of two proper closed (in the induced topology) subsets. The following is an immediate corollary of the decomposition of algebraic sets into their irreducible components. 24 / 51

50 Irreducible components Embedding theorem Let G be equationally Noetherian. Then for every system of equations S(X ) = 1 over G there are finitely many irreducible systems S 1 (X ) = 1,..., S m (X ) = 1 (that determine the irreducible components of the algebraic set V (S)) such that G R(S) G R(S1 )... G R(Sm) 25 / 51

51 Irreducible components Theorem V F (S) is irreducible F R(S) is discriminated by F. Recall, that a group G is discriminated by a subgroup H if for any finite subset M G with 1 M there exists an H-homomorphism φ : G H such that 1 φ(m). 26 / 51

52 Irreducible components Proof We will prove that V (S) is irreducible if and only if F R(S) is discriminated in F (by F -homomorphisms). Suppose V (S) is not irreducible and V (S) = n i=1 V (S i) is its decomposition into irreducible components. Then R(S) = n i=1 R(S i), and hence there exist s i R(S i ) \ {R(S), R(S j ), j i}. The set {s i, i = 1,... n} cannot be discriminated in F (by F -homomorphisms). Suppose now s 1,..., s n are elements such that for any retract f : F R(S) F there exists i such that f (s i ) = 1; then V (S) = m i=1 V (S s i). 27 / 51

53 Irreducible components Proof We will prove that V (S) is irreducible if and only if F R(S) is discriminated in F (by F -homomorphisms). Suppose V (S) is not irreducible and V (S) = n i=1 V (S i) is its decomposition into irreducible components. Then R(S) = n i=1 R(S i), and hence there exist s i R(S i ) \ {R(S), R(S j ), j i}. The set {s i, i = 1,... n} cannot be discriminated in F (by F -homomorphisms). Suppose now s 1,..., s n are elements such that for any retract f : F R(S) F there exists i such that f (s i ) = 1; then V (S) = m i=1 V (S s i). 27 / 51

54 Irreducible components Proof We will prove that V (S) is irreducible if and only if F R(S) is discriminated in F (by F -homomorphisms). Suppose V (S) is not irreducible and V (S) = n i=1 V (S i) is its decomposition into irreducible components. Then R(S) = n i=1 R(S i), and hence there exist s i R(S i ) \ {R(S), R(S j ), j i}. The set {s i, i = 1,... n} cannot be discriminated in F (by F -homomorphisms). Suppose now s 1,..., s n are elements such that for any retract f : F R(S) F there exists i such that f (s i ) = 1; then V (S) = m i=1 V (S s i). 27 / 51

55 Diophantine problem in free groups Theorem [Makanin, 1982] There is an algorithm to verify whether a given system of equations has a solution in a free group (free semgroup) or not. He showed that if there is a solution of an equation S(X, A) = 1 in F then there is a short solution of length f ( S ) where f is some fixed computable function. Extremely hard theorem! Now it is viewed as a major achievement in group theory, as well as in computer science. 28 / 51

56 Diophantine problem in free groups Theorem [Makanin, 1982] There is an algorithm to verify whether a given system of equations has a solution in a free group (free semgroup) or not. He showed that if there is a solution of an equation S(X, A) = 1 in F then there is a short solution of length f ( S ) where f is some fixed computable function. Extremely hard theorem! Now it is viewed as a major achievement in group theory, as well as in computer science. 28 / 51

57 Diophantine problem in free groups Theorem [Makanin, 1982] There is an algorithm to verify whether a given system of equations has a solution in a free group (free semgroup) or not. He showed that if there is a solution of an equation S(X, A) = 1 in F then there is a short solution of length f ( S ) where f is some fixed computable function. Extremely hard theorem! Now it is viewed as a major achievement in group theory, as well as in computer science. 28 / 51

58 Complexity The original Makanin s algorithm is very inefficient - not even primitive recursive. Plandowski gave a much improved version (for free semigroups): P-space. Gutierrez devised a P-space algorithm for free groups. 29 / 51

59 Complexity The original Makanin s algorithm is very inefficient - not even primitive recursive. Plandowski gave a much improved version (for free semigroups): P-space. Gutierrez devised a P-space algorithm for free groups. 29 / 51

60 Complexity The original Makanin s algorithm is very inefficient - not even primitive recursive. Plandowski gave a much improved version (for free semigroups): P-space. Gutierrez devised a P-space algorithm for free groups. 29 / 51

61 Complexity Theorem [Kharlampovich, Lysenok, Myasnikov, Touikan (2008)] The Diophantine problem for quadratic equations in free groups is NP-complete. Theorem [announced by Lysenok] The Diophantine problem in free semigroups (groups) is NP-complete. If so - many interesting consequences for algorithmic group theory and topology. 30 / 51

62 Complexity Theorem [Kharlampovich, Lysenok, Myasnikov, Touikan (2008)] The Diophantine problem for quadratic equations in free groups is NP-complete. Theorem [announced by Lysenok] The Diophantine problem in free semigroups (groups) is NP-complete. If so - many interesting consequences for algorithmic group theory and topology. 30 / 51

63 Existential and Universal theories of F Theorem [Makanin] The existential Th (F ) and the universal Th (F ) theories of F are decidable. It was known long before that non-abelian free groups have the same existential and universal theories. Main question was: what are finitely generated groups G with Th (G) = Th (F )? 31 / 51

64 Existential and Universal theories of F Theorem [Makanin] The existential Th (F ) and the universal Th (F ) theories of F are decidable. It was known long before that non-abelian free groups have the same existential and universal theories. Main question was: what are finitely generated groups G with Th (G) = Th (F )? 31 / 51

65 Existential and Universal theories of F Theorem [Makanin] The existential Th (F ) and the universal Th (F ) theories of F are decidable. It was known long before that non-abelian free groups have the same existential and universal theories. Main question was: what are finitely generated groups G with Th (G) = Th (F )? 31 / 51

66 Groups universally equivalent to F Unification Theorem 1 Let G be a finitely generated group and F G. Then the following conditions are equivalent: 1) G is discriminated by F (fully residually free), i.e. for any finite subset M G there exists a homomorphism G F injective on M. 2) G is universally equivalent to F ; 3) G is the coordinate group of an irreducible variety over F. 4) G is a Sela s limit group. 5) G is a limit of free groups in Gromov-Hausdorff metric. 6) G embeds into an ultrapower of free groups. 32 / 51

67 Groups universally equivalent to F Unification Theorem 1 Let G be a finitely generated group and F G. Then the following conditions are equivalent: 1) G is discriminated by F (fully residually free), i.e. for any finite subset M G there exists a homomorphism G F injective on M. 2) G is universally equivalent to F ; 3) G is the coordinate group of an irreducible variety over F. 4) G is a Sela s limit group. 5) G is a limit of free groups in Gromov-Hausdorff metric. 6) G embeds into an ultrapower of free groups. 32 / 51

68 Groups universally equivalent to F Unification Theorem 1 Let G be a finitely generated group and F G. Then the following conditions are equivalent: 1) G is discriminated by F (fully residually free), i.e. for any finite subset M G there exists a homomorphism G F injective on M. 2) G is universally equivalent to F ; 3) G is the coordinate group of an irreducible variety over F. 4) G is a Sela s limit group. 5) G is a limit of free groups in Gromov-Hausdorff metric. 6) G embeds into an ultrapower of free groups. 32 / 51

69 Groups universally equivalent to F Unification Theorem 1 Let G be a finitely generated group and F G. Then the following conditions are equivalent: 1) G is discriminated by F (fully residually free), i.e. for any finite subset M G there exists a homomorphism G F injective on M. 2) G is universally equivalent to F ; 3) G is the coordinate group of an irreducible variety over F. 4) G is a Sela s limit group. 5) G is a limit of free groups in Gromov-Hausdorff metric. 6) G embeds into an ultrapower of free groups. 32 / 51

70 Groups universally equivalent to F Unification Theorem 1 Let G be a finitely generated group and F G. Then the following conditions are equivalent: 1) G is discriminated by F (fully residually free), i.e. for any finite subset M G there exists a homomorphism G F injective on M. 2) G is universally equivalent to F ; 3) G is the coordinate group of an irreducible variety over F. 4) G is a Sela s limit group. 5) G is a limit of free groups in Gromov-Hausdorff metric. 6) G embeds into an ultrapower of free groups. 32 / 51

71 Groups universally equivalent to F Unification Theorem 1 Let G be a finitely generated group and F G. Then the following conditions are equivalent: 1) G is discriminated by F (fully residually free), i.e. for any finite subset M G there exists a homomorphism G F injective on M. 2) G is universally equivalent to F ; 3) G is the coordinate group of an irreducible variety over F. 4) G is a Sela s limit group. 5) G is a limit of free groups in Gromov-Hausdorff metric. 6) G embeds into an ultrapower of free groups. 32 / 51

72 Groups universally equivalent to F Unification Theorem 2 (with coefficients) Let G be a finitely generated group containing a free non-abelian group F as a subgroup. Then the following conditions are equivalent: 1) G is F -discriminated by F ; 2) G is universally equivalent to F (in the language with constants); 3) G is the coordinate group of an irreducible variety over F. 4) G is Sela s limit group. 5) G is a limit of free groups in Gromov-Hausdorff metric. 6) G F -embeds into an ultrapower of F. 33 / 51

73 Groups universally equivalent to F Equivalence 1) 3) has been already proved. We will prove the equivalence 1) 2). Let L A be the language of group theory with generators A of F as constants. Let G be a f.g. group which is F -discriminated by F. Consider a formula X (U(X, A) = 1 W (X, A) 1). If this formula is true in F, then it is also true in G, because F G. If it is true in G, then for some X G m holds U( X, A) = 1 and W ( X, A) 1. Since G is F -discriminated by F, there is an F -homomorphism φ: G F such that φ(w ( X, A)) 1, i.e. W ( X φ, A) 1. Of course U( X φ, A) = 1. Therefore the above formula is true in F. Since in F -group a conjunction of equations [inequalities] is equivalent to one equation [resp., inequality], the same existential sentences in the language L A are true in G and in F. 34 / 51

74 Groups universally equivalent to F Suppose now that G is F -universally equivalent to F. Let G = X, A S(X, A) = 1, be a presentation of G and w 1 (X, A),..., w k (X, A) nontrivial elements in G. Let Y be the set of the same cardinality as X. Consider a system of equations S(Y, A) = 1 in variables Y in F. Since the group F is equationally Noetherian, this system is equivalent over F to a finite subsystem S 1 (Y, A) = 1. The formula Ψ = Y (S 1 (Y, A) = 1 ( w 1 (Y, A) = 1 w k (Y, A) = 1)). is false in G, therefore it is false in F. This means that there exists a set of elements B in F such that S 1 (B, A) = 1 and, therefore, S(B, A) = 1 such that w 1 (B, A) 1 w k (B, A) 1. The map X B that is identical on F can be extended to the F -homomorphism from G to F. 35 / 51

75 limits of free groups Limit groups by Sela H fg, a sequence {φ i } in Hom(H, F ) is stable if, for all h H, h φ i is eventually always 1 or eventually never 1. Ker φ i is: {h H h φ i = 1 for almost all i} G is a limit group if there is a fg H and a stable sequence {φ i } such that: G = H/Ker φ i. 36 / 51

76 We will prove now the equivalence 1) 4). Suppose that H = g 1,..., g k is f.g. and discriminated by F. There exists a sequence of homomorphisms φ n : G F, so that φ n maps the elements in a ball of radius n in the Cayley graph of H to distinct elements in F. This is a stable sequence of homomorphisms φ m. In general, G is a quotient of H, but since the homomorphisms were chosen so that φ n maps a ball of radius n monomorphically into F, G is isomorphic to H and, therefore, H is a limit group. 37 / 51

77 To prove the converse we may assume that a limit group G is non-abelian because the statement is, obviously, true for abelian groups. By definition, there exists a f.g. group H, and a sequence of homomorphisms φ i : H F, so that G = H/Ker φ i. WLOG we can assume that H is f.p. Add a relation one-at-a-time to H to obtain: H H 1 H 2... G. We may assume Hom(H; F ) = Hom(G; F ) and hence that the φ i s are defined on G. Each non-trivial element of G is mapped to 1 by only finitely many φ i, so G is F discriminated. 38 / 51

78 Groups universally equivalent to F We proved the equivalence 1) 2) 3) 4). The other statements we will prove later. 39 / 51

79 Unification theorems for fully residually free groups This result shows that the class of fully residually free groups is quite special - it appeared (and was independently studied) in several different areas of group theory. It turned out that similar results hold for many other groups! (torsion free relatively hyperbolic with abelian parabolics, solvable etc.) F just has to be equationally Noetherian. 40 / 51

80 Unification theorems for fully residually free groups This result shows that the class of fully residually free groups is quite special - it appeared (and was independently studied) in several different areas of group theory. It turned out that similar results hold for many other groups! (torsion free relatively hyperbolic with abelian parabolics, solvable etc.) F just has to be equationally Noetherian. 40 / 51

81 Groups universally equivalent to F Theorem[Kharlampovich, Miasnikov, 96] Every f.g. fully residually free group is a subgroup of a group obtained from a free group as a finite series of extensions of centralizers. 41 / 51

82 Corollaries of our theorem One can study the coordinate groups of systems of equations (fully residually free groups) via their splittings into graphs of groups. Corollary For every freely indecomposable non-abelian finitely generated fully residually free group one can effectively find a non-trivial splitting (as an amalgamated product or HNN extension) over a cyclic subgroup. 42 / 51

83 Corollaries of our theorem Corollary Every finitely generated fully residually free group is finitely presented. There is an algorithm to find a finite presentation. Corollary Every finitely generated residually free group G can be effectively presented as a subgroup of a direct product of finitely many fully residually free groups; hence, G is embeddable into F Z[t]... F Z[t]. 43 / 51

84 limits of free groups [Ch. Champetier and V. Guirardel (2004)] A marked group (G, S) is a group G with a prescribed family of generators S = (s 1,..., s n ). Two marked groups (G, (s 1,..., s n )) and (G, (s 1,..., s n)) are isomorphic as marked groups if the bijection s i s i extends to an isomorphism. For example, ( a, (1, a)) and ( a, (a, 1)) are not isomorphic as marked groups. Denote by G n the set of groups marked by n elements up to isomorphism of marked groups. One can define a metric on G n by setting the distance between two marked groups (G, S) and (G, S ) to be e N if they have exactly the same relations of length at most N (under the bijection S S ) (Grigorchuk, Gromov s metric) Finally, a limit group is a limit (with respect to the metric above) of marked free groups in G n. 44 / 51

85 limits of free groups Example: A free abelian group of rank 2 is a limit of a sequence of cyclic groups with marking ( a, (a, a n )), n. In the definition of a limit group, F can be replaced by any equationally Noetherian group or algebra. A direct limit of a direct system of n-generated finite partial subalgebras of A such that all products of generators eventually appear in these partial subalgebras, is called a limit algebra over A. It is interesting to study limits of free semigroups. 45 / 51

86 The equivalence 2) 6) is a particular case of general results in model theory. 46 / 51

87 5) 6). It is shown by Champetier and Guirardel that a group is a limit group if and only if it is a finitely generated subgroup of an ultraproduct of free groups (for a non-principal ultrafilter), and any such ultraproduct of free groups contains all the limit groups. This implies the equivalence 5) 6). 47 / 51

88 Krull dimension Proposition 3 Let G be a finitely generated fully residually free group. Then G possesses the following properties. HW12 Prove statements 1,2,3,4 1 Each Abelian subgroup of G is contained in a unique maximal finitely generated Abelian subgroup, in particular, each Abelian subgroup of G is finitely generated; 2 G is finitely presented, and has only finitely many conjugacy classes of its maximal non-cyclic Abelian subgroups. 3 G has solvable word problem; 4 Every 2-generated subgroup of G is either free or abelian; 5 G is linear; 6 If rank (G)=3 then either G is free of rank 3, free abelian of rank 3, or a free rank one extension of centralizer of a free group of rank 2 (that is G = x, y, t [u(x, y), t] = 1, where the word u is not a proper power). 48 / 51

89 Krull dimension Theorem [announced by Louder] There exists a function g(n) such that the length of every proper descending chain of closed sets in F n is bounded by g(n). 49 / 51

90 Description of solutions Theorem [Razborov] Given a finite system of equations S(X ) = 1 in F (A) one can effectively construct a finite Solution Diagram that describes all solutions of S(X ) = 1 in F. 50 / 51

91 Solution diagrams F R(S) σ 1 F R(Ωv1 ) F R(Ωv2 ) F R(Ωvn ) σ 2 F R(Ωv21 ) F R(Ωv2m ) F (A) F (T ) F (A) 51 / 51