Algorithmic aspects of Galois theory in recent times

Size: px

Start display at page:

Download "Algorithmic aspects of Galois theory in recent times"

Zoe Joseph
6 years ago
Views:

1 Algorithmic aspects of Galois theory in recent times Michael F. Singer Department of Mathematics North Carolina State University Raleigh, NC Bicentenaire de la naissance d Évariste Galois l Institut Henri Poincaré, 28 Octobre 2011

2 Calculating Galois groups... in theory

3 Calculating Galois groups... in theory Calculating Galois groups... in practice

4 Calculating Galois groups... in theory Calculating Galois groups... in practice Calculating Differential Galois groups

5 Calculating Galois groups in theory

6 Calculating Galois groups in theory Can we calculate Galois groups?

7 Calculating Galois groups in theory Can we calculate Galois groups? Yes we can, but...

8 An Algorithm (Kronecker, Gründzuge..., Crelle, 92, 1882)

9 An Algorithm (Kronecker, Gründzuge..., Crelle, 92, 1882) f (x) Z[x], deg(f ) = n gcd(f, f ) = 1

10 An Algorithm (Kronecker, Gründzuge..., Crelle, 92, 1882) f (x) Z[x], deg(f ) = n gcd(f, f ) = 1 1) Let R(Y, X 1,..., X n) = Q σ S n (Y (α σ(1) X α σ(n) X n)) = Q σ S n (Y (α 1 X σ 1 (1) α nx σ 1 (n))) Q[Y, X 1,..., X n]

11 An Algorithm (Kronecker, Gründzuge..., Crelle, 92, 1882) f (x) Z[x], deg(f ) = n gcd(f, f ) = 1 1) Let R(Y, X 1,..., X n) = = Q σ S n (Y (α σ(1) X α σ(n) X n)) σ S n (Y (α 1 X σ 1 (1) α nx σ 1 (n))) Q[Y, X 1,..., X n] 2) Factor R(Y, X 1,..., X n) = R 1 (Y, X 1,..., X n) R t(y, X 1,..., X n)

12 An Algorithm (Kronecker, Gründzuge..., Crelle, 92, 1882) f (x) Z[x], deg(f ) = n gcd(f, f ) = 1 1) Let R(Y, X 1,..., X n) = = Q σ S n (Y (α σ(1) X α σ(n) X n)) σ S n (Y (α 1 X σ 1 (1) α nx σ 1 (n))) Q[Y, X 1,..., X n] 2) Factor R(Y, X 1,..., X n) = R 1 (Y, X 1,..., X n) R t(y, X 1,..., X n) Galois group of f = {σ S n R 1 (Y, X σ(1),..., X σ(n) ) = R 1 (Y, X 1,..., X n)}

13 An Algorithm (Kronecker, Gründzuge..., Crelle, 92, 1882) f (x) Z[x], deg(f ) = n gcd(f, f ) = 1 1) Let R(Y, X 1,..., X n) = = Q σ S n (Y (α σ(1) X α σ(n) X n)) σ S n (Y (α 1 X σ 1 (1) α nx σ 1 (n))) Q[Y, X 1,..., X n] 2) Factor R(Y, X 1,..., X n) = R 1 (Y, X 1,..., X n) R t(y, X 1,..., X n) Galois group of f = {σ S n R 1 (Y, X σ(1),..., X σ(n) ) = R 1 (Y, X 1,..., X n)} High Complexity!

14 Polynomial Time Algorithms Question: Is there an algorithm to compute the Galois group of f (x) Z[x] whose running time is given as a polynomial in the size of f?

15 Polynomial Time Algorithms Question: Is there an algorithm to compute the Galois group of f (x) Z[x] whose running time is given as a polynomial in the size of f? For m Z, size(m) = number of digits log(m) For f (x) = a nx n a 0 Z[x], size(f (x)) = n max i {size(a i )}.

16 Polynomial Time Algorithms Question: Is there an algorithm to compute the Galois group of f (x) Z[x] whose running time is given as a polynomial in the size of f? For m Z, size(m) = number of digits log(m) For f (x) = a nx n a 0 Z[x], size(f (x)) = n max i {size(a i )}. Running Time = number of +,,,

17 Polynomial Time Algorithms Question: Is there an algorithm to compute the Galois group of f (x) Z[x] whose running time is given as a polynomial in the size of f? For m Z, size(m) = number of digits log(m) For f (x) = a nx n a 0 Z[x], size(f (x)) = n max i {size(a i )}. Running Time = number of +,,, Revised Question:Is there an algorithm to compute generators of the Galois group of f (x) Z[x] whose running time is given as a polynomial in the size of f?

18 Polynomial Time Algorithms Question: Is there an algorithm to compute the Galois group of f (x) Z[x] whose running time is given as a polynomial in the size of f? For m Z, size(m) = number of digits log(m) For f (x) = a nx n a 0 Z[x], size(f (x)) = n max i {size(a i )}. Running Time = number of +,,, Revised Question:Is there an algorithm to compute generators of the Galois group of f (x) Z[x] whose running time is given as a polynomial in the size of f? We do not know!

19 Galois (Discours préliminaire): Si maintenant vous me donnez une équation que vous aurez choisie à votre gré et que vous desiriez connaître si elle est ou non soluble par radicaux, je n aurai rien à y faire que de vous indiquer le moyen de répondre à votre question, sans vouloir charger ni moi ni personne de le faire. En un mot les calculs sont impraticables.

20 Galois (Discours préliminaire): Si maintenant vous me donnez une équation que vous aurez choisie à votre gré et que vous desiriez connaître si elle est ou non soluble par radicaux, je n aurai rien à y faire que de vous indiquer le moyen de répondre à votre question, sans vouloir charger ni moi ni personne de le faire. En un mot les calculs sont impraticables. Question: Is there an algorithm to decide if f (x) Z[x] is solvable by radicals whose running time is given as a polynomial in the size of f?

21 Galois (Discours préliminaire): Si maintenant vous me donnez une équation que vous aurez choisie à votre gré et que vous desiriez connaître si elle est ou non soluble par radicaux, je n aurai rien à y faire que de vous indiquer le moyen de répondre à votre question, sans vouloir charger ni moi ni personne de le faire. En un mot les calculs sont impraticables. Question: Is there an algorithm to decide if f (x) Z[x] is solvable by radicals whose running time is given as a polynomial in the size of f? Landau and Miller (Solvability by radicals in polynomial time, J. Comp. Sys. Sci.,1985): If now you give us a polynomial which you have chosen at your pleasure, and if you want to know if it is or is not solvable by radicals, we have presented techniques to answer that question in polynomial time.

22 Ingredients of the Landau-Miller Algorithm

23 Ingredients of the Landau-Miller Algorithm Lenstra-Lenstra-Lovász (1982): Polynomial time algorithm to factor f (x) Q[x].

24 Ingredients of the Landau-Miller Algorithm Lenstra-Lenstra-Lovász (1982): Polynomial time algorithm to factor f (x) Q[x]. Landau (1985): Construct splitting field and Galois group G in time polynomial in G and size(f )

25 Ingredients of the Landau-Miller Algorithm Lenstra-Lenstra-Lovász (1982): Polynomial time algorithm to factor f (x) Q[x]. Landau (1985): Construct splitting field and Galois group G in time polynomial in G and size(f ) Adjoin root of f Q(α 1 ), factor f over Q(α 1 ) f = f 1 f 2 f t Adjoin root of f 1 Q(α 1, α 2 ), factor f over Q(α 1, α 2 ) Stop when f factors completely over K = Q(α 1,..., α l ) K = Q(β), β = r 1 α r nα n, g(x) = min. poly β over Q Galois group = {σ S n g(r 1 α σ(1) r nα σ(n) ) = 0}

26 Ingredients of the Landau-Miller Algorithm Lenstra-Lenstra-Lovász (1982): Polynomial time algorithm to factor f (x) Q[x]. Landau (1985): Construct splitting field and Galois group G in time polynomial in G and size(f )

27 Ingredients of the Landau-Miller Algorithm Lenstra-Lenstra-Lovász (1982): Polynomial time algorithm to factor f (x) Q[x]. Landau (1985): Construct splitting field and Galois group G in time polynomial in G and size(f ) Sims (1970): Given a group G can determine if it solvable in time polynomial in G. Can determine if Galois group is solvable in time polynomial in G and size(f )

28 Ingredients of the Landau-Miller Algorithm Lenstra-Lenstra-Lovász (1982): Polynomial time algorithm to factor f (x) Q[x]. Landau (1985): Construct splitting field and Galois group G in time polynomial in G and size(f ) Sims (1970): Given a group G can determine if it solvable in time polynomial in G. Can determine if Galois group is solvable in time polynomial in G and size(f ) P alfy (1982): G S n solvable, transitive and primitive implies G < n 3.25

29 Ingredients of the Landau-Miller Algorithm Lenstra-Lenstra-Lovász (1982): Polynomial time algorithm to factor f (x) Q[x]. Landau (1985): Construct splitting field and Galois group G in time polynomial in G and size(f ) Sims (1970): Given a group G can determine if it solvable in time polynomial in G. Can determine if Galois group is solvable in time polynomial in G and size(f ) P alfy (1982): G S n solvable, transitive and primitive implies G < n 3.25 Landau-Miller (1985): Showed how to reduce to the case of equations with transitive, primitive Galois groups.

30 Calculating Galois groups in practice Things that work

31 Mod p techniques f (x) Z[x], monic, deg f = n (f ) = Q (α i α j ) 2 Z

32 Mod p techniques f (x) Z[x], monic, deg f = n (f ) = Q (α i α j ) 2 Z Fact: For p (f ), Gal(f (mod p)) Gal(f )

33 Mod p techniques f (x) Z[x], monic, deg f = n (f ) = Q (α i α j ) 2 Z Fact: For p (f ), Gal(f (mod p)) Gal(f ) R(Y, X 1,..., X n) = Y σ S n (Y (α σ(1) X α σ(n) X n)) = R 1 R t over Q

34 Mod p techniques f (x) Z[x], monic, deg f = n (f ) = Q (α i α j ) 2 Z Fact: For p (f ), Gal(f (mod p)) Gal(f ) R(Y, X 1,..., X n) = Y σ S n (Y (α σ(1) X α σ(n) X n)) = R 1 R t over Q = (R 1,1 R 1,m1 ) (R t,1 R t,mt ) mod p

35 Mod p techniques f (x) Z[x], monic, deg f = n (f ) = Q (α i α j ) 2 Z Fact: For p (f ), Gal(f (mod p)) Gal(f )

36 Mod p techniques f (x) Z[x], monic, deg f = n (f ) = Q (α i α j ) 2 Z Fact: For p (f ), Gal(f (mod p)) Gal(f ) Theorem of Frobenius Let n = n n t, n 1 n 2... n t Density of {p p (f ), f = f 1 f t (mod p), deg(f i ) = n i } 1 G {σ G σ = τ 1 τ t, τ i a cycle of length n i }

37 Mod p techniques f (x) Z[x], monic, deg f = n (f ) = Q (α i α j ) 2 Z Fact: For p (f ), Gal(f (mod p)) Gal(f ) Theorem of Frobenius Let n = n n t, n 1 n 2... n t Density of {p p (f ), f = f 1 f t (mod p), deg(f i ) = n i } 1 G {σ G σ = τ 1 τ t, τ i a cycle of length n i } Advantages: Easy to factor mod p (Berlekamp, 1967) Gives a good probabilistic test for S n, A n; good evidence for other groups.

38 Mod p techniques f (x) Z[x], monic, deg f = n (f ) = Q (α i α j ) 2 Z Fact: For p (f ), Gal(f (mod p)) Gal(f ) Theorem of Frobenius Let n = n n t, n 1 n 2... n t Density of {p p (f ), f = f 1 f t (mod p), deg(f i ) = n i } 1 G {σ G σ = τ 1 τ t, τ i a cycle of length n i } Advantages: Easy to factor mod p (Berlekamp, 1967) Gives a good probabilistic test for S n, A n; good evidence for other groups. Disadvantages: Asymptotic result Groups not determined by distribution of cycle patterns - already in deg. 8

39 Invariant Theoretic Techniques

40 Invariant Theoretic Techniques Example: f (x) = x 3 + bx + c Q[x] Gal(f ) S 3

41 Invariant Theoretic Techniques Example: f (x) = x 3 + bx + c Q[x] Gal(f ) S 3 f (x) = (x + α)(x + βx + γ), α, β, γ Q = Gal(f ) = S 2 or {id}

42 Invariant Theoretic Techniques Example: f (x) = x 3 + bx + c Q[x] Gal(f ) S 3 f (x) = (x + α)(x + βx + γ), α, β, γ Q = Gal(f ) = S 2 or {id} f (x) irreducible = Gal(f ) acts transitively on the roots α 1, α 2, α 3 Group Theory = Gal(f ) = S 3 or A 3

43 Invariant Theoretic Techniques Example: f (x) = x 3 + bx + c Q[x] Gal(f ) S 3 f (x) = (x + α)(x + βx + γ), α, β, γ Q = Gal(f ) = S 2 or {id} f (x) irreducible = Gal(f ) acts transitively on the roots α 1, α 2, α 3 Group Theory = Gal(f ) = S 3 or A 3 Let F (z) = z 2 + 4b c 2 = (z + δ)(z δ) δ = (α 1 α 2 )(α 2 α 3 )(α 3 α 1 ) δ is an invariant of A 3 but not of S 3 so Gal(f ) = A 3 F (z) factors over Q.

44 Invariant Theoretic Techniques Example: f (x) = x 3 + bx + c Q[x] Gal(f ) S 3 f (x) = (x + α)(x + βx + γ), α, β, γ Q = Gal(f ) = S 2 or {id} f (x) irreducible = Gal(f ) acts transitively on the roots α 1, α 2, α 3 Group Theory = Gal(f ) = S 3 or A 3 Let F (z) = z 2 + 4b c 2 = (z + δ)(z δ) δ = (α 1 α 2 )(α 2 α 3 )(α 3 α 1 ) δ is an invariant of A 3 but not of S 3 so Gal(f ) = A 3 F (z) factors over Q. Reduce calculation of Galois groups to factorization of associated polynomials

45 Why does this work?

46 Why does this work? I. A finite group is determined by its permutation representations.

47 Why does this work? I. A finite group is determined by its permutation representations. Given H G, ρ : G S N such that G acts transitively but H does not.

48 Why does this work? I. A finite group is determined by its permutation representations. Given H G, ρ : G S N such that G acts transitively but H does not. II. One can find permutation representations of G = Gal(K /k) in K. Let ρ : G S N, then β 1,..., β N K such that σ(α i ) = α ρ(σ)(i), for all σ G

49 Why does this work? I. A finite group is determined by its permutation representations. Given H G, ρ : G S N such that G acts transitively but H does not. II. One can find permutation representations of G = Gal(K /k) in K. Let ρ : G S N, then β 1,..., β N K such that σ(α i ) = α ρ(σ)(i), for all σ G Given Gal(K /k) G, to show Gal(K /k) = G: For each maximal subgroup H G, find a representation as in I. Find β 1,..., β N K as in II. Form F H (z) = Q (z β i ) k[z]. If F H (z) is irreducible for each H, then Gal(K /k) = G.

50 Differential Galois Groups

51 Differential Galois Groups What are Diffferential Galois Groups and what do they measure?

52 Differential Galois Groups What are Diffferential Galois Groups and what do they measure? Calculating Differential Galois Groups... in theory

53 Differential Galois Groups What are Diffferential Galois Groups and what do they measure? Calculating Differential Galois Groups... in theory Calculating Differential Galois Groups... in practice

54 Picard-Vessiot Theory

55 Picard-Vessiot Theory Consider a linear differential equation L(y) = d n y dz n + a n 1(z) d n 1 y dz n a 0y = 0, a i (z) C(z)

56 Picard-Vessiot Theory Consider a linear differential equation L(y) = d n y dz n + a n 1(z) d n 1 y dz n a 0y = 0, a i (z) C(z) z 0 a nonsingular point solutions y 1,..., y n anal. near z 0, lin. indep. /C.

57 Picard-Vessiot Theory Consider a linear differential equation L(y) = d n y dz n + a n 1(z) d n 1 y dz n a 0y = 0, a i (z) C(z) z 0 a nonsingular point solutions y 1,..., y n anal. near z 0, lin. indep. /C. PV-extension K = C(z)(y 1,..., y n, y 1,..., y n,..., y (n 1) 1,..., y (n 1) n ).

58 Picard-Vessiot Theory Consider a linear differential equation L(y) = d n y dz n + a n 1(z) d n 1 y dz n a 0y = 0, a i (z) C(z) z 0 a nonsingular point solutions y 1,..., y n anal. near z 0, lin. indep. /C. PV-extension K = C(z)(y 1,..., y n, y 1,..., y n,..., y (n 1) 1,..., y (n 1) n ). PV-group DGal(K /k) = {σ : K K σ is a C(z) diff. autom. of K } DGal(K /k) leaves Soln(L) invariant DGal(K /k) GL n(c). DGal(K /k) is Zariski-closed.

59 Picard-Vessiot Theory Consider a linear differential equation L(y) = d n y dz n + a n 1(z) d n 1 y dz n a 0y = 0, a i (z) C(z) z 0 a nonsingular point solutions y 1,..., y n anal. near z 0, lin. indep. /C. PV-extension K = C(z)(y 1,..., y n, y 1,..., y n,..., y (n 1) 1,..., y (n 1) n ). PV-group DGal(K /k) = {σ : K K σ is a C(z) diff. autom. of K } DGal(K /k) leaves Soln(L) invariant DGal(K /k) GL n(c). DGal(K /k) is Zariski-closed. Galois Correspondence: H Zariski closed DGal(K /C(z)) Diff. field, C(z) F K F

60 What do Differential Galois Groups Measure?

61 What do Differential Galois Groups Measure? Algebraic Dependence: K - a PV-extension of C(z) with PV-group G tr. deg. C(z) K = dim C G

62 What do Differential Galois Groups Measure? Algebraic Dependence: K - a PV-extension of C(z) with PV-group G tr. deg. C(z) K = dim C G Example: L(y) = y + 1 z λ2 + (1 )y = 0, z λ 2 DGal = SL n(c) 1 2 / Z tr. deg. C(z) C(z)(J λ, Y λ, J λ, Y λ) = 3

63 Solvability

64 Solvability L(y) = 0 is solvable in terms of liouvillian functions if there exists a tower of fields C(z) = K 0... K n such that K i+1 = K i (t i ) with t i algebraic over K i, or t i K i, i.e., t i = R u i, u i K i, or t i /t i K i, i.e., t i = e R u i, u i K i. with K K n, where K is the PV-extension associated with L(y) = 0.

65 Solvability L(y) = 0 is solvable in terms of liouvillian functions if there exists a tower of fields C(z) = K 0... K n such that K i+1 = K i (t i ) with t i algebraic over K i, or t i K i, i.e., t i = R u i, u i K i, or t i /t i K i, i.e., t i = e R u i, u i K i. with K K n, where K is the PV-extension associated with L(y) = 0. Example: L(y) = d 2 y dx 1 dy 2 2x dx xy K 0 = C(x) K 1 = K 0 ( x) K 2 = K 1 (er x ) {er x, e R x } is a basis for Soln(L = 0)

66 Solvability L(y) = 0 is solvable in terms of liouvillian functions if there exists a tower of fields C(z) = K 0... K n such that K i+1 = K i (t i ) with t i algebraic over K i, or t i K i, i.e., t i = R u i, u i K i, or t i /t i K i, i.e., t i = e R u i, u i K i. with K K n, where K is the PV-extension associated with L(y) = 0. Example: L(y) = d 2 y dx 1 dy 2 2x dx xy K 0 = C(x) K 1 = K 0 ( x) K 2 = K 1 (er x ) {er x, e R x } is a basis for Soln(L = 0) Thm: L(y) = 0 solvable in terms of liouvillian functions DGal contains a solvable subgroup of finite index.

67 Calculating Differential Galois Groups... in theory L(y) = y (n) + a n 1 y (n 1) a 0 y a i Q(x)

68 Calculating Differential Galois Groups... in theory L(y) = y (n) + a n 1 y (n 1) a 0 y a i Q(x) One can decide if L(y) = 0 has algebraic solutions n = 2: Schwarz, Klein :: Baldassari-Dwork, van Hoeij-Weil,... n 2: Jordan, Boulanger, Painlevé :: Risch, S.

69 Calculating Differential Galois Groups... in theory L(y) = y (n) + a n 1 y (n 1) a 0 y a i Q(x) One can decide if L(y) = 0 has algebraic solutions n = 2: Schwarz, Klein :: Baldassari-Dwork, van Hoeij-Weil,... n 2: Jordan, Boulanger, Painlevé :: Risch, S. One can decide if L(y) = 0 is solvable in terms of liouvillian functions. n = 2: Pepin :: Kovacic. n 2: Marotte :: S., Ulmer,... (n = 3 van Hoeij, Weil, Ulmer,..)

70 Calculating Differential Galois Groups... in theory L(y) = y (n) + a n 1 y (n 1) a 0 y a i Q(x) One can decide if L(y) = 0 has algebraic solutions n = 2: Schwarz, Klein :: Baldassari-Dwork, van Hoeij-Weil,... n 2: Jordan, Boulanger, Painlevé :: Risch, S. One can decide if L(y) = 0 is solvable in terms of liouvillian functions. n = 2: Pepin :: Kovacic. n 2: Marotte :: S., Ulmer,... (n = 3 van Hoeij, Weil, Ulmer,..) Can characterize when L(y) = 0 is solvable in terms of linear DE of lower order and decide for n = 3. Can decide if L(y) = 0 solvable in terms of Airy, Bessel, Cylinder, Kummer, Laguerre, Whittaker... (van Hoeij et al)

71 Calculating Differential Galois Groups... in theory L(y) = y (n) + a n 1 y (n 1) a 0 y a i Q(x) One can decide if L(y) = 0 has algebraic solutions n = 2: Schwarz, Klein :: Baldassari-Dwork, van Hoeij-Weil,... n 2: Jordan, Boulanger, Painlevé :: Risch, S. One can decide if L(y) = 0 is solvable in terms of liouvillian functions. n = 2: Pepin :: Kovacic. n 2: Marotte :: S., Ulmer,... (n = 3 van Hoeij, Weil, Ulmer,..) Can characterize when L(y) = 0 is solvable in terms of linear DE of lower order and decide for n = 3. Can decide if L(y) = 0 solvable in terms of Airy, Bessel, Cylinder, Kummer, Laguerre, Whittaker... (van Hoeij et al) One can compute the Galois group. (Hrushovsky)

72 Calculating Differential Galois Groups... in practice

73 Calculating Differential Galois Groups... in practice Ideas based on Tannakian philosophy:

74 Calculating Differential Galois Groups... in practice Ideas based on Tannakian philosophy: A linear algebraic group G is determined by its linear representations.

75 Calculating Differential Galois Groups... in practice Ideas based on Tannakian philosophy: A linear algebraic group G is determined by its linear representations. Can recover G from its category of fin. dim. G-modules Can construct all fin. dim. G-modules from a single faithful G-module via sums, submodules, quotients and duals.

76 Calculating Differential Galois Groups... in practice Ideas based on Tannakian philosophy: A linear algebraic group G is determined by its linear representations. Can recover G from its category of fin. dim. G-modules Can construct all fin. dim. G-modules from a single faithful G-module via sums, submodules, quotients and duals. A linear differential equation is an avatar for the representation theory of its PV-group.

77 Calculating Differential Galois Groups... in practice Ideas based on Tannakian philosophy: A linear algebraic group G is determined by its linear representations. Can recover G from its category of fin. dim. G-modules Can construct all fin. dim. G-modules from a single faithful G-module via sums, submodules, quotients and duals. A linear differential equation is an avatar for the representation theory of its PV-group. Given L(y) with V = Soln(L(y)), for any fin. dim. G-module W, can construct an L(y) with Soln( L(y)) = W.

78 Calculating Differential Galois Groups... in practice Ideas based on Tannakian philosophy: A linear algebraic group G is determined by its linear representations. Can recover G from its category of fin. dim. G-modules Can construct all fin. dim. G-modules from a single faithful G-module via sums, submodules, quotients and duals. A linear differential equation is an avatar for the representation theory of its PV-group. Given L(y) with V = Soln(L(y)), for any fin. dim. G-module W, can construct an L(y) with Soln( L(y)) = W. (i) W V L = L L, Soln( L) = W

79 Calculating Differential Galois Groups... in practice Ideas based on Tannakian philosophy: A linear algebraic group G is determined by its linear representations. Can recover G from its category of fin. dim. G-modules Can construct all fin. dim. G-modules from a single faithful G-module via sums, submodules, quotients and duals. A linear differential equation is an avatar for the representation theory of its PV-group. Given L(y) with V = Soln(L(y)), for any fin. dim. G-module W, can construct an L(y) with Soln( L(y)) = W. (i) W V L = L L, Soln( L) = W (ii) V = Soln(L(y)), W = Soln( L(y)), V W = (0) V W = Soln(LCLM(L, L)(y))

80 Calculating Differential Galois Groups... in practice Ideas based on Tannakian philosophy: A linear algebraic group G is determined by its linear representations. Can recover G from its category of fin. dim. G-modules Can construct all fin. dim. G-modules from a single faithful G-module via sums, submodules, quotients and duals. A linear differential equation is an avatar for the representation theory of its PV-group. Given L(y) with V = Soln(L(y)), for any fin. dim. G-module W, can construct an L(y) with Soln( L(y)) = W. (i) W V L = L L, Soln( L) = W (ii) V = Soln(L(y)), W = Soln( L(y)), V W = (0) V W = Soln(LCLM(L, L)(y)) (iii) Sym m (V ) = {y 1 y m y i V } = Soln(L s m (y))

81 Thm. Assume L(y) = y + r(x)y. L(y) = 0 is solvable in terms of liouvillian functions

82 Thm. Assume L(y) = y + r(x)y. L(y) = 0 is solvable in terms of liouvillian functions L s 6 factors.

83 Thm. Assume L(y) = y + r(x)y. L(y) = 0 is solvable in terms of liouvillian functions L s 6 factors. Thm. Assume L(y) = y + r(x)y. DGal = Valentiner Group A SL 3 6 of order 1080

84 Thm. Assume L(y) = y + r(x)y. L(y) = 0 is solvable in terms of liouvillian functions L s 6 factors. Thm. Assume L(y) = y + r(x)y. DGal = Valentiner Group A SL 3 6 of order 1080 L s 2 and L s 3 are irreducible L s 4 = L 9 L 6, L 9, L 6 irreducible.

Differential Groups and Differential Relations. Michael F. Singer Department of Mathematics North Carolina State University Raleigh, NC USA

Differential Groups and Differential Relations Michael F. Singer Department of Mathematics North Carolina State University Raleigh, NC 27695-8205 USA http://www.math.ncsu.edu/ singer Theorem: (Hölder,