Theory Jordan_Normal_Form
section ‹Jordan Normal Form›
text ‹This theory defines Jordan normal forms (JNFs) in a sparse representation, i.e.,
as block-diagonal matrices. We also provide a closed formula for powers of JNFs,
which allows to estimate the growth rates of JNFs.›
theory Jordan_Normal_Form
imports
Matrix
Char_Poly
Polynomial_Interpolation.Missing_Unsorted
begin
definition jordan_block :: "nat ⇒ 'a :: {zero,one} ⇒ 'a mat" where
"jordan_block n a = mat n n (λ (i,j). if i = j then a else if Suc i = j then 1 else 0)"
lemma jordan_block_index[simp]: "i < n ⟹ j < n ⟹
jordan_block n a $$ (i,j) = (if i = j then a else if Suc i = j then 1 else 0)"
"dim_row (jordan_block n k) = n"
"dim_col (jordan_block n k) = n"
unfolding jordan_block_def by auto
lemma jordan_block_carrier[simp]: "jordan_block n k ∈ carrier_mat n n"
unfolding carrier_mat_def by auto
lemma jordan_block_char_poly: "char_poly (jordan_block n a) = [: -a, 1:]^n"
unfolding char_poly_defs by (subst det_upper_triangular[of _ n], auto simp: prod_list_diag_prod)
lemma jordan_block_pow_carrier[simp]:
"jordan_block n a ^⇩m r ∈ carrier_mat n n" by auto
lemma jordan_block_pow_dim[simp]:
"dim_row (jordan_block n a ^⇩m r) = n" "dim_col (jordan_block n a ^⇩m r) = n" by auto
lemma jordan_block_pow: "(jordan_block n (a :: 'a :: comm_ring_1)) ^⇩m r =
mat n n (λ (i,j). if i ≤ j then of_nat (r choose (j - i)) * a ^ (r + i - j) else 0)"
proof (induct r)
case 0
{
fix i j :: nat
assume "i ≠ j" "i ≤ j"
hence "j - i > 0" by auto
hence "0 choose (j - i) = 0" by simp
} note [simp] = this
show ?case
by (simp, rule eq_matI, auto)
next
case (Suc r)
let ?jb = "jordan_block n a"
let ?rij = "λ r i j. of_nat (r choose (j - i)) * a ^ (r + i - j)"
let ?v = "λ i j. if i ≤ j then of_nat (r choose (j - i)) * a ^ (r + i - j) else 0"
have "?jb ^⇩m Suc r = mat n n (λ (i,j). if i ≤ j then ?rij r i j else 0) * ?jb" by (simp add: Suc)
also have "… = mat n n (λ (i,j). if i ≤ j then ?rij (Suc r) i j else 0)"
proof -
{
fix j
assume j: "j < n"
hence col: "col (jordan_block n a) j = vec n (λi. if i = j then a else if Suc i = j then 1 else 0)"
unfolding jordan_block_def col_mat[OF j] by simp
fix f
have "vec n f ∙ col (jordan_block n a) j = (f j * a + (if j = 0 then 0 else f (j - 1)))"
proof -
define p where "p = (λ i. vec n f $ i * col (jordan_block n a) j $ i)"
have "vec n f ∙ col (jordan_block n a) j = (∑i = 0 ..< n. p i)"
unfolding scalar_prod_def p_def by simp
also have "… = p j + sum p ({0 ..< n} - {j})" using j
by (subst sum.remove[of _ j], auto)
also have "p j = f j * a" unfolding p_def col using j by auto
also have "sum p ({0 ..< n} - {j}) = (if j = 0 then 0 else f (j - 1))"
proof (cases j)
case 0
have "sum p ({0 ..< n} - {j}) = 0"
by (rule sum.neutral, auto simp: p_def col 0)
thus ?thesis using 0 by simp
next
case (Suc jj)
with j have jj: "jj ∈ {0 ..< n} - {j}" by auto
have "sum p ({0 ..< n} - {j}) = p jj + sum p ({0 ..< n} - {j} - {jj})"
by (subst sum.remove[OF _ jj], auto)
also have "p jj = f (j - 1)" unfolding p_def col using jj
by (auto simp: Suc)
also have "sum p ({0 ..< n} - {j} - {jj}) = 0"
by (rule sum.neutral, auto simp: p_def col, auto simp: Suc)
finally show ?thesis unfolding Suc by simp
qed
finally show ?thesis .
qed
} note scalar_to_sum = this
{
fix i j
assume i: "i < n" and ij: "i > j"
hence j: "j < n" by auto
have "vec n (?v i) ∙ col (jordan_block n a) j = 0"
unfolding scalar_to_sum[OF j] using ij i j by auto
} note easy_case = this
{
fix i j
assume j: "j < n" and ij: "i ≤ j"
hence i: "i < n" and id: "⋀ p q. (if i ≤ j then p else q) = p" by auto
have "vec n (?v i) ∙ col (jordan_block n a) j =
(of_nat (r choose (j - i)) * (a ^ (Suc (r + i - j)))) +
(if j = 0 then 0
else if i ≤ j - 1 then of_nat (r choose (j - 1 - i)) * a ^ (r + i - (j - 1)) else 0)"
unfolding scalar_to_sum[OF j]
using ij by simp
also have "… = of_nat (Suc r choose (j - i)) * a ^ (Suc (r + i) - j)"
proof (cases j)
case (Suc jj)
{
assume "i ≤ Suc jj" and "¬ i ≤ jj"
hence "i = Suc jj" by auto
hence "a * a ^ (r + i - Suc jj) = a ^ (r + i - jj)" by simp
}
moreover
{
assume ijj: "i ≤ jj"
have "of_nat (r choose (Suc jj - i)) * (a * a ^ (r + i - Suc jj))
+ of_nat (r choose (jj - i)) * a ^ (r + i - jj) =
of_nat (Suc r choose (Suc jj - i)) * a ^ (r + i - jj)"
proof (cases "r + i < jj")
case True
hence gt: "jj - i > r" "Suc jj - i > r" "Suc jj - i > Suc r" by auto
show ?thesis
unfolding binomial_eq_0[OF gt(1)] binomial_eq_0[OF gt(2)] binomial_eq_0[OF gt(3)]
by simp
next
case False
hence ge: "r + i ≥ jj" by simp
show ?thesis
proof (cases "jj = r + i")
case True
have gt: "r < Suc r" by simp
show ?thesis unfolding True by (simp add: binomial_eq_0[OF gt])
next
case False
with ge have lt: "jj < r + i" by auto
hence "r + i - jj = Suc (r + i - Suc jj)" by simp
hence prod: "a * a ^ (r + i - Suc jj) = a ^ (r + i - jj)" by simp
from ijj have id: "Suc jj - i = Suc (jj - i)" by simp
have binom: "Suc r choose (Suc jj - i) =
r choose (Suc jj - i) + (r choose (jj - i))"
unfolding id
by (subst binomial_Suc_Suc, simp)
show ?thesis unfolding prod binom
by (simp add: field_simps)
qed
qed
}
ultimately show ?thesis using ij unfolding Suc by auto
qed auto
finally have "vec n (?v i) ∙ col (jordan_block n a) j
= of_nat (Suc r choose (j - i)) * a ^ (Suc (r + i) - j)" .
} note main_case = this
show ?thesis
by (rule eq_matI, insert easy_case main_case, auto)
qed
finally show ?case by simp
qed
definition jordan_matrix :: "(nat × 'a :: {zero,one})list ⇒ 'a mat" where
"jordan_matrix n_as = diag_block_mat (map (λ (n,a). jordan_block n a) n_as)"
lemma jordan_matrix_dim[simp]:
"dim_row (jordan_matrix n_as) = sum_list (map fst n_as)"
"dim_col (jordan_matrix n_as) = sum_list (map fst n_as)"
unfolding jordan_matrix_def
by (subst dim_diag_block_mat, auto, (induct n_as, auto simp: Let_def)+)
lemma jordan_matrix_carrier[simp]:
"jordan_matrix n_as ∈ carrier_mat (sum_list (map fst n_as)) (sum_list (map fst n_as))"
unfolding carrier_mat_def by auto
lemma jordan_matrix_upper_triangular: "i < sum_list (map fst n_as)
⟹ j < i ⟹ jordan_matrix n_as $$ (i,j) = 0"
unfolding jordan_matrix_def
by (rule diag_block_upper_triangular, auto simp: jordan_matrix_def[symmetric])
lemma jordan_matrix_pow: "(jordan_matrix n_as) ^⇩m r =
diag_block_mat (map (λ (n,a). (jordan_block n a) ^⇩m r) n_as)"
unfolding jordan_matrix_def
by (subst diag_block_pow_mat, force, rule arg_cong[of _ _ diag_block_mat], auto)
lemma jordan_matrix_char_poly:
"char_poly (jordan_matrix n_as) = (∏(n, a)←n_as. [:- a, 1:] ^ n)"
proof -
let ?n = "sum_list (map fst n_as)"
have "diag_mat
([:0, 1:] ⋅⇩m 1⇩m (sum_list (map fst n_as)) + map_mat (λa. [:- a:]) (jordan_matrix n_as)) =
concat (map (λ(n, a). replicate n [:- a, 1:]) n_as)" unfolding jordan_matrix_def
proof (induct n_as)
case (Cons na n_as)
obtain n a where na: "na = (n,a)" by force
let ?n2 = "sum_list (map fst n_as)"
note fbo = four_block_one_mat
note mz = zero_carrier_mat
note mo = one_carrier_mat
have mA: "⋀ A. A ∈ carrier_mat (dim_row A) (dim_col A)" unfolding carrier_mat_def by auto
let ?Bs = "map (λ(x, y). jordan_block x y) n_as"
let ?B = "diag_block_mat ?Bs"
from jordan_matrix_dim[of n_as, unfolded jordan_matrix_def]
have dimB: "dim_row ?B = ?n2" "dim_col ?B = ?n2" by auto
hence B: "?B ∈ carrier_mat ?n2 ?n2" unfolding carrier_mat_def by simp
show ?case unfolding na fbo
apply (simp add: Let_def fbo[symmetric] del: fbo)
apply (subst smult_four_block_mat[OF mo mz mz mo])
apply (subst map_four_block_mat[OF jordan_block_carrier mz mz mA])
apply (subst add_four_block_mat[of _ n n _ ?n2 _ ?n2], auto simp: dimB B)
apply (subst diag_four_block_mat[of _ n _ ?n2], auto simp: dimB B)
apply (subst Cons, auto simp: jordan_block_def diag_mat_def,
intro nth_equalityI, auto)
done
qed (force simp: diag_mat_def)
also have "prod_list ... = (∏(n, a)←n_as. [:- a, 1:] ^ n)"
by (induct n_as, auto)
finally
show ?thesis unfolding char_poly_defs
by (subst det_upper_triangular[of _ ?n], auto simp: jordan_matrix_upper_triangular)
qed
definition jordan_nf :: "'a :: semiring_1 mat ⇒ (nat × 'a)list ⇒ bool" where
"jordan_nf A n_as ≡ (0 ∉ fst ` set n_as ∧ similar_mat A (jordan_matrix n_as))"
lemma jordan_nf_powE: assumes A: "A ∈ carrier_mat n n" and jnf: "jordan_nf A n_as"
obtains P Q where "P ∈ carrier_mat n n" "Q ∈ carrier_mat n n" and
"char_poly A = (∏(na, a)←n_as. [:- a, 1:] ^ na)"
"⋀ k. A ^⇩m k = P * (jordan_matrix n_as)^⇩m k * Q"
proof -
from A have dim: "dim_row A = n" by auto
assume obt: "⋀P Q. P ∈ carrier_mat n n ⟹ Q ∈ carrier_mat n n ⟹
char_poly A = (∏(na, a)←n_as. [:- a, 1:] ^ na) ⟹
(⋀k. A ^⇩m k = P * jordan_matrix n_as ^⇩m k * Q) ⟹ thesis"
from jnf[unfolded jordan_nf_def] obtain P Q where
simw: "similar_mat_wit A (jordan_matrix n_as) P Q"
and sim: "similar_mat A (jordan_matrix n_as)" unfolding similar_mat_def by blast
show thesis
proof (rule obt)
show "⋀ k. A ^⇩m k = P * jordan_matrix n_as ^⇩m k * Q"
by (rule similar_mat_wit_pow_id[OF simw])
show "char_poly A = (∏(na, a)←n_as. [:- a, 1:] ^ na)"
unfolding char_poly_similar[OF sim] jordan_matrix_char_poly ..
qed (insert simw[unfolded similar_mat_wit_def Let_def dim], auto)
qed
lemma choose_poly_bound: assumes "i ≤ d"
shows "r choose i ≤ max 1 (r^d)"
proof (cases "i ≤ r")
case False
hence "r choose i = 0" by simp
thus ?thesis by arith
next
case True
show ?thesis
proof (cases r)
case (Suc rr)
from binomial_le_pow[OF True] have "r choose i ≤ r ^ i" by simp
also have "… ≤ r^d" using power_increasing[OF ‹i ≤ d›, of r] Suc by auto
finally show ?thesis by simp
qed (insert True, simp)
qed
context
fixes b :: "'a :: archimedean_field"
assumes b: "0 < b" "b < 1"
begin
lemma poly_exp_constant_bound: "∃ p. ∀ x. c * b ^ x * of_nat x ^ deg ≤ p"
proof (cases "c ≤ 0")
case True
show ?thesis
by (rule exI[of _ 0], intro allI,
rule mult_nonpos_nonneg[OF mult_nonpos_nonneg[OF True]], insert b, auto)
next
case False
hence c: "c ≥ 0" by simp
from poly_exp_bound[OF b, of deg] obtain p where "⋀ x. b ^ x * of_nat x ^ deg ≤ p" by auto
from mult_left_mono[OF this c]
show ?thesis by (intro exI[of _ "c * p"], auto simp: ac_simps)
qed
lemma poly_exp_max_constant_bound: "∃ p. ∀ x. c * b ^ x * max 1 (of_nat x ^ deg) ≤ p"
proof -
from poly_exp_constant_bound[of c deg] obtain p where
p: "⋀ x. c * b ^ x * of_nat x ^ deg ≤ p" by auto
show ?thesis
proof (rule exI[of _ "max p c"], intro allI)
fix x
let ?exp = "of_nat x ^ deg :: 'a"
show "c * b ^ x * max 1 ?exp ≤ max p c"
proof (cases "x = 0")
case False
hence "?exp ≠ of_nat 0" by simp
hence "?exp ≥ 1" by (metis less_one not_less of_nat_1 of_nat_less_iff of_nat_power)
hence "max 1 ?exp = ?exp" by simp
thus ?thesis using p[of x] by simp
qed (cases deg, auto)
qed
qed
end
context
fixes a :: "'a :: real_normed_field"
begin
lemma jordan_block_bound:
assumes i: "i < n" and j: "j < n"
shows "norm ((jordan_block n a ^⇩m k) $$ (i,j))
≤ norm a ^ (k + i - j) * max 1 (of_nat k ^ (n - 1))"
(is "?lhs ≤ ?rhs")
proof -
have id: "(jordan_block n a ^⇩m k) $$ (i,j) = (if i ≤ j then of_nat (k choose (j - i)) * a ^ (k + i - j) else 0)"
unfolding jordan_block_pow using i j by auto
from i j have diff: "j - i ≤ n - 1" by auto
show ?thesis
proof (cases "i ≤ j")
case False
thus ?thesis unfolding id by simp
next
case True
hence "?lhs = norm (of_nat (k choose (j - i)) * a ^ (k + i - j))" unfolding id by simp
also have "… ≤ norm (of_nat (k choose (j - i)) :: 'a) * norm (a ^ (k + i - j))"
by (rule norm_mult_ineq)
also have "… ≤ (max 1 (of_nat k ^ (n - 1))) * norm a ^ (k + i - j)"
proof (rule mult_mono[OF _ norm_power_ineq _ norm_ge_zero])
have "k choose (j - i) ≤ max 1 (k ^ (n - 1))"
by (rule choose_poly_bound[OF diff])
hence "norm (of_nat (k choose (j - i)) :: 'a) ≤ of_nat (max 1 (k ^ (n - 1)))"
unfolding norm_of_nat of_nat_le_iff .
also have "… = max 1 (of_nat k ^ (n - 1))" by (metis max_def of_nat_1 of_nat_le_iff of_nat_power)
finally show "norm (of_nat (k choose (j - i)) :: 'a) ≤ max 1 (real_of_nat k ^ (n - 1))" .
qed simp
also have "… = ?rhs" by simp
finally show ?thesis .
qed
qed
lemma jordan_block_poly_bound:
assumes i: "i < n" and j: "j < n" and a: "norm a = 1"
shows "norm ((jordan_block n a ^⇩m k) $$ (i,j)) ≤ max 1 (of_nat k ^ (n - 1))"
(is "?lhs ≤ ?rhs")
proof -
from jordan_block_bound[OF i j, of k, unfolded a]
show ?thesis by simp
qed
theorem jordan_block_constant_bound: assumes a: "norm a < 1"
shows "∃ p. ∀ i j k. i < n ⟶ j < n ⟶ norm ((jordan_block n a ^⇩m k) $$ (i,j)) ≤ p"
proof (cases "a = 0")
case True
show ?thesis
proof (rule exI[of _ 1], intro allI impI)
fix i j k
assume *: "i < n" "j < n"
{
assume ij: "i ≤ j"
have "norm ((of_nat (k choose (j - i)) :: 'a) * 0 ^ (k + i - j)) ≤ 1" (is "norm ?lhs ≤ 1")
proof (cases "k + i > j")
case True
hence "?lhs = 0" by simp
also have "norm (…) ≤ 1" by simp
finally show ?thesis .
next
case False
hence id: "?lhs = (of_nat (k choose (j - i)) :: 'a)" and j: "j - i ≥ k" by auto
from j have "k choose (j - i) = 0 ∨ k choose (j - i) = 1" by (simp add: nat_less_le)
thus "norm ?lhs ≤ 1"
proof
assume k: "k choose (j - i) = 0"
show ?thesis unfolding id k by simp
next
assume k: "k choose (j - i) = 1"
show ?thesis unfolding id unfolding k by simp
qed
qed
}
thus "norm ((jordan_block n a ^⇩m k) $$ (i,j)) ≤ 1" unfolding True
unfolding jordan_block_pow using * by auto
qed
next
case False
hence na: "norm a > 0" by auto
define c where "c = inverse (norm a ^ n)"
define deg where "deg = n - 1"
have c: "c > 0" unfolding c_def using na by auto
define b where "b = norm a"
from a na have "0 < b" "b < 1" unfolding b_def by auto
from poly_exp_max_constant_bound[OF this, of c deg]
obtain p where "⋀ k. c * b ^ k * max 1 (of_nat k ^ deg) ≤ p" by auto
show ?thesis
proof (intro exI[of _ p], intro allI impI)
fix i j k
assume ij: "i < n" "j < n"
from jordan_block_bound[OF this]
have "norm ((jordan_block n a ^⇩m k) $$ (i, j))
≤ norm a ^ (k + i - j) * max 1 (real_of_nat k ^ (n - 1))" .
also have "… ≤ c * norm a ^ k * max 1 (real_of_nat k ^ (n - 1))"
proof (rule mult_right_mono)
from ij have "i - j ≤ n" by auto
show "norm a ^ (k + i - j) ≤ c * norm a ^ k"
proof (rule mult_left_le_imp_le)
show "0 < norm a ^ n" using na by auto
let ?lhs = "norm a ^ n * norm a ^ (k + i - j)"
let ?rhs = "norm a ^ n * (c * norm a ^ k)"
from ij have ge: "n + (k + i - j) ≥ k" by arith
have "?lhs = norm a ^ (n + (k + i - j))" by (simp add: power_add)
also have "… ≤ norm a ^ k" using ge a na using less_imp_le power_decreasing by blast
also have "… = ?rhs" unfolding c_def using na by simp
finally show "?lhs ≤ ?rhs" .
qed
qed simp
also have "… = c * b ^ k * max 1 (real_of_nat k ^ deg)" unfolding b_def deg_def ..
also have "… ≤ p" by fact
finally show "norm ((jordan_block n a ^⇩m k) $$ (i, j)) ≤ p" .
qed
qed
definition norm_bound :: "'a mat ⇒ real ⇒ bool" where
"norm_bound A b ≡ ∀ i j. i < dim_row A ⟶ j < dim_col A ⟶ norm (A $$ (i,j)) ≤ b"
lemma norm_boundI[intro]:
assumes "⋀ i j. i < dim_row A ⟹ j < dim_col A ⟹ norm (A $$ (i,j)) ≤ b"
shows "norm_bound A b"
unfolding norm_bound_def using assms by blast
lemma jordan_block_constant_bound2:
"∃p. norm (a :: 'a :: real_normed_field) < 1 ⟶
(∀i j k. i < n ⟶ j < n ⟶ norm ((jordan_block n a ^⇩m k) $$ (i, j)) ≤ p)"
using jordan_block_constant_bound by auto
lemma jordan_matrix_poly_bound2:
fixes n_as :: "(nat × 'a) list"
assumes n_as: "⋀ n a. (n,a) ∈ set n_as ⟹ n > 0 ⟹ norm a ≤ 1"
and N: "⋀ n a. (n,a) ∈ set n_as ⟹ norm a = 1 ⟹ n ≤ N"
shows "∃c1. ∀k. ∀e ∈ elements_mat (jordan_matrix n_as ^⇩m k).
norm e ≤ c1 + of_nat k ^ (N - 1)"
proof -
from jordan_matrix_carrier[of n_as] obtain d where
jm: "jordan_matrix n_as ∈ carrier_mat d d" by blast
define f where "f = (λn (a::'a) i j k. norm ((jordan_block n a ^⇩m k) $$ (i,j)))"
let ?g = "λk c1. c1 + of_nat k ^ (N-1)"
let ?P = "λn (a::'a) i j k c1. f n a i j k ≤ ?g k c1"
define Q where "Q = (λn (a::'a) k c1. ∀i j. i<n ⟶ j<n ⟶ ?P n a i j k c1)"
have "⋀ c c' k n a i j. c ≤ c' ⟹ ?P n a i j k c ⟹ ?P n a i j k c'" by auto
hence Q_mono: "⋀n a c c'. c ≤ c' ⟹ ∀k. Q n a k c ⟹ ∀k. Q n a k c'"
unfolding Q_def by arith
{ fix n a assume na: "(n,a) ∈ set n_as"
obtain c where c: "norm a < 1 ⟶ (∀i j k. i < n ⟶ j < n ⟶ f n a i j k ≤ c)"
apply (rule exE[OF jordan_block_constant_bound2])
unfolding f_def using Jordan_Normal_Form.jordan_block_constant_bound2
by metis
define c1 where "c1 = max 1 c"
then have "c1 ≥ 1" "c1 ≥ c" by auto
have "∃c1. ∀k i j. i < n ⟶ j < n ⟶ ?P n a i j k c1"
proof rule+
fix i j k assume "i < n" "j < n"
then have "0<n" by auto
let ?jbs = "map (λ(n,a). jordan_block n a) n_as"
have sq_jbs: "Ball (set ?jbs) square_mat" by auto
have "jordan_matrix n_as ^⇩m k = diag_block_mat (map (λA. A ^⇩m k) ?jbs)"
unfolding jordan_matrix_def using diag_block_pow_mat[OF sq_jbs] by auto
show "?P n a i j k c1"
proof (cases "norm a = 1")
case True {
have nN:"n-1 ≤ N-1" using N[OF na] True by auto
have "f n a i j k ≤ max 1 (of_nat k ^ (n-1))"
using Jordan_Normal_Form.jordan_block_poly_bound True ‹i<n› ‹j<n›
unfolding f_def by auto
also have "... ≤ max 1 (of_nat k ^ (N-1))"
proof (cases "k=0")
case False then show ?thesis
by (subst max.mono[OF _ power_increasing[OF nN]], auto)
qed (simp add: power_eq_if)
also have "... ≤ max c1 (of_nat k ^ (N-1))" using ‹c1≥1› by auto
also have "... ≤ c1 + (of_nat k ^ (N-1))" using ‹c1≥1› by auto
finally show ?thesis by simp
} next
case False {
then have na1: "norm a < 1" using n_as[OF na] ‹0<n› by auto
hence "f n a i j k ≤ c" using c ‹i<n› ‹j<n› by auto
also have "... ≤ c1" using ‹c≤c1›.
also have "... ≤ c1 + of_nat k ^ (N-1)" by auto
finally show ?thesis by auto
}
qed
qed
}
hence "∀na. ∃c1. na ∈ set n_as ⟶ (∀k. Q (fst na) (snd na) k c1)"
unfolding Q_def by auto
from choice[OF this] obtain c'
where c': "⋀ na k. na ∈ set n_as ⟹ Q (fst na) (snd na) k (c' na)" by blast
define c where "c = max 0 (Max (set (map c' n_as)))"
{ fix n a assume na: "(n,a) ∈ set n_as"
then have Q: "∀ k. Q n a k (c' (n,a))" using c'[OF na] by auto
from na have "c' (n,a) ∈ set (map c' n_as)" by auto
from Max_ge[OF _ this] have "c' (n,a) ≤ c" unfolding c_def by auto
from Q_mono[OF this Q] have "⋀ k. Q n a k c" by blast
}
hence Q: "⋀k n a. (n,a) ∈ set n_as ⟹ Q n a k c" by auto
have c0: "c ≥ 0" unfolding c_def by simp
{ fix k n a e
assume na:"(n,a) ∈ set n_as"
let ?jbk = "jordan_block n a ^⇩m k"
assume "e ∈ elements_mat ?jbk"
from elements_matD[OF this] obtain i j
where "i < n" "j < n" and [simp]: "e = ?jbk $$ (i,j)"
by (simp only:pow_mat_dim_square[OF jordan_block_carrier],auto)
hence "norm e ≤ ?g k c" using Q[OF na] unfolding Q_def f_def by simp
}
hence norm_jordan:
"⋀k. ∀(n,a) ∈ set n_as. ∀e ∈ elements_mat (jordan_block n a ^⇩m k).
norm e ≤ ?g k c" by auto
{ fix k
let ?jmk = "jordan_matrix n_as ^⇩m k"
have "dim_row ?jmk = d" "dim_col ?jmk = d"
using jm by (simp only:pow_mat_dim_square[OF jm])+
let ?As = "(map (λ(n,a). jordan_block n a ^⇩m k) n_as)"
have "⋀e. e ∈ elements_mat ?jmk ⟹ norm e ≤ ?g k c"
proof -
fix e assume e:"e ∈ elements_mat ?jmk"
obtain i j where ij: "i < d" "j < d" and "e = ?jmk $$ (i,j)"
using elements_matD[OF e] by (simp only:pow_mat_dim_square[OF jm],auto)
have "?jmk = diag_block_mat ?As"
using jordan_matrix_pow[of n_as k] by auto
hence "elements_mat ?jmk ⊆ {0} ∪ ⋃ (set (map elements_mat ?As))"
using elements_diag_block_mat[of ?As] by auto
hence e_mem: "e ∈ {0} ∪ ⋃ (set (map elements_mat ?As))"
using e by blast
show "norm e ≤ ?g k c"
proof (cases "e = 0")
case False
then have "e ∈ ⋃ (set (map elements_mat ?As))" using e_mem by auto
then obtain n a
where "e ∈ elements_mat (jordan_block n a ^⇩m k)"
and na: "(n,a) ∈ set n_as" by force
thus ?thesis using norm_jordan na by force
qed (insert c0, auto)
qed
}
thus ?thesis by auto
qed
lemma norm_bound_bridge:
"∀e ∈ elements_mat A. norm e ≤ b ⟹ norm_bound A b"
unfolding norm_bound_def by force
lemma norm_bound_mult: assumes A1: "A1 ∈ carrier_mat nr n"
and A2: "A2 ∈ carrier_mat n nc"
and b1: "norm_bound A1 b1"
and b2: "norm_bound A2 b2"
shows "norm_bound (A1 * A2) (b1 * b2 * of_nat n)"
proof
let ?A = "A1 * A2"
let ?n = "of_nat n"
fix i j
assume i: "i < dim_row ?A" and j: "j < dim_col ?A"
define v1 where "v1 = (λ k. row A1 i $ k)"
define v2 where "v2 = (λ k. col A2 j $ k)"
from assms(1-2) have dim: "dim_row A1 = nr" "dim_col A2 = nc" "dim_col A1 = n" "dim_row A2 = n" by auto
{
fix k
assume k: "k < n"
have n: "norm (v1 k) ≤ b1" "norm (v2 k) ≤ b2"
using i j k dim v1_def v2_def
b1[unfolded norm_bound_def, rule_format, of i k]
b2[unfolded norm_bound_def, rule_format, of k j] by auto
have "norm (v1 k * v2 k) ≤ norm (v1 k) * norm (v2 k)" by (rule norm_mult_ineq)
also have "… ≤ b1 * b2" by (rule mult_mono'[OF n], auto)
finally have "norm (v1 k * v2 k) ≤ b1 * b2" .
} note bound = this
have "?A $$ (i,j) = row A1 i ∙ col A2 j" using dim i j by simp
also have "… = (∑ k = 0 ..< n. v1 k * v2 k)" unfolding scalar_prod_def
using dim i j v1_def v2_def by simp
also have "norm (…) ≤ (∑ k = 0 ..< n. b1 * b2)"
by (rule sum_norm_le, insert bound, simp)
also have "… = b1 * b2 * ?n" by simp
finally show "norm (?A $$ (i,j)) ≤ b1 * b2 * ?n" .
qed
lemma norm_bound_max: "norm_bound A (Max {norm (A $$ (i,j)) | i j. i < dim_row A ∧ j < dim_col A})"
(is "norm_bound A (Max ?norms)")
proof
fix i j
have fin: "finite ?norms" by (simp add: finite_image_set2)
assume "i < dim_row A" and "j < dim_col A"
hence "norm (A $$ (i,j)) ∈ ?norms" by auto
from Max_ge[OF fin this] show "norm (A $$ (i,j)) ≤ Max ?norms" .
qed
lemma jordan_matrix_poly_bound: fixes n_as :: "(nat × 'a)list"
assumes n_as: "⋀ n a. (n,a) ∈ set n_as ⟹ n > 0 ⟹ norm a ≤ 1"
and N: "⋀ n a. (n,a) ∈ set n_as ⟹ norm a = 1 ⟹ n ≤ N"
shows "∃ c1. ∀ k. norm_bound (jordan_matrix n_as ^⇩m k) (c1 + of_nat k ^ (N - 1))"
using jordan_matrix_poly_bound2 norm_bound_bridge N n_as
by metis
lemma jordan_nf_matrix_poly_bound: fixes n_as :: "(nat × 'a)list"
assumes A: "A ∈ carrier_mat n n"
and n_as: "⋀ n a. (n,a) ∈ set n_as ⟹ n > 0 ⟹ norm a ≤ 1"
and N: "⋀ n a. (n,a) ∈ set n_as ⟹ norm a = 1 ⟹ n ≤ N"
and jnf: "jordan_nf A n_as"
shows "∃ c1 c2. ∀ k. norm_bound (A ^⇩m k) (c1 + c2 * of_nat k ^ (N - 1))"
proof -
let ?cp2 = "∏(n, a)←n_as. [:- a, 1:] ^ n"
let ?J = "jordan_matrix n_as"
from jnf[unfolded jordan_nf_def]
have sim: "similar_mat A ?J" by auto
then obtain P Q where sim_wit: "similar_mat_wit A ?J P Q" unfolding similar_mat_def by auto
from similar_mat_wit_pow_id[OF this] have pow: "⋀ k. A ^⇩m k = P * ?J ^⇩m k * Q" .
from sim_wit[unfolded similar_mat_wit_def Let_def] A
have J: "?J ∈ carrier_mat n n" and P: "P ∈ carrier_mat n n" and Q: "Q ∈ carrier_mat n n"
unfolding carrier_mat_def by force+
have "∃c1. ∀ k. norm_bound (?J ^⇩m k) (c1 + of_nat k ^ (N - 1))"
by (rule jordan_matrix_poly_bound[OF n_as N])
then obtain c1 where
bound_pow: "⋀ k. norm_bound ((?J ^⇩m k)) (c1 + of_nat k ^ (N - 1))" by blast
obtain bP where bP: "norm_bound P bP" using norm_bound_max[of P] by auto
obtain bQ where bQ: "norm_bound Q bQ" using norm_bound_max[of Q] by auto
let ?n = "of_nat n :: real"
let ?c2 = "bP * ?n * bQ * ?n"
let ?c1 = "?c2 * c1"
{
fix k
have Jk: "?J ^⇩m k ∈ carrier_mat n n" using J by simp
from norm_bound_mult[OF mult_carrier_mat[OF P Jk] Q
norm_bound_mult[OF P Jk bP bound_pow] bQ, folded pow]
have "norm_bound (A ^⇩m k) (?c1 + ?c2 * of_nat k ^ (N - 1))" (is "norm_bound _ ?exp")
by (simp add: field_simps)
} note main = this
show ?thesis
by (intro exI allI, rule main)
qed
end
context
fixes f_ty :: "'a :: field itself"
begin
lemma char_matrix_jordan_block: "char_matrix (jordan_block n a) b = (jordan_block n (a - b))"
unfolding char_matrix_def jordan_block_def by auto
lemma diag_jordan_block_pow: "diag_mat (jordan_block n (a :: 'a) ^⇩m k) = replicate n (a ^ k)"
unfolding diag_mat_def jordan_block_pow
by (intro nth_equalityI, auto)
lemma jordan_block_zero_pow: "(jordan_block n (0 :: 'a)) ^⇩m k =
(mat n n (λ (i,j). if j ≥ i ∧ j - i = k then 1 else 0))"
proof -
{
fix i j
assume *: "j - i ≠ k"
have "of_nat (k choose (j - i)) * 0 ^ (k + i - j) = (0 :: 'a)"
proof (cases "k + i - j > 0")
case True thus ?thesis by (cases "k + i - j", auto)
next
case False
with * have "j - i > k" by auto
thus ?thesis by (simp add: binomial_eq_0)
qed
}
thus ?thesis unfolding jordan_block_pow by (intro eq_matI, auto)
qed
end
lemma jordan_matrix_concat_diag_block_mat: "jordan_matrix (concat jbs) = diag_block_mat (map jordan_matrix jbs)"
unfolding jordan_matrix_def[abs_def]
by (induct jbs, auto simp: diag_block_mat_append Let_def)
lemma jordan_nf_diag_block_mat: assumes Ms: "⋀ A jbs. (A,jbs) ∈ set Ms ⟹ jordan_nf A jbs"
shows "jordan_nf (diag_block_mat (map fst Ms)) (concat (map snd Ms))"
proof -
let ?Ms = "map (λ (A, jbs). (A, jordan_matrix jbs)) Ms"
have id: "map fst ?Ms = map fst Ms" by auto
have id2: "map snd ?Ms = map jordan_matrix (map snd Ms)" by auto
{
fix A B
assume "(A,B) ∈ set ?Ms"
then obtain jbs where mem: "(A,jbs) ∈ set Ms" and B: "B = jordan_matrix jbs" by auto
from Ms[OF mem] have "similar_mat A B" unfolding B jordan_nf_def by auto
}
from similar_diag_mat_block_mat[of ?Ms, OF this, unfolded id id2] Ms
show ?thesis
unfolding jordan_nf_def jordan_matrix_concat_diag_block_mat by force
qed
lemma jordan_nf_char_poly: assumes "jordan_nf A n_as"
shows "char_poly A = (∏ (n,a) ← n_as. [:- a, 1:] ^ n)"
unfolding jordan_matrix_char_poly[symmetric]
by (rule char_poly_similar, insert assms[unfolded jordan_nf_def], auto)
lemma jordan_nf_block_size_order_bound: assumes jnf: "jordan_nf A n_as"
and mem: "(n,a) ∈ set n_as"
shows "n ≤ order a (char_poly A)"
proof -
from jnf[unfolded jordan_nf_def]
have "similar_mat A (jordan_matrix n_as)" by auto
from similar_matD[OF this] obtain m where "A ∈ carrier_mat m m" by auto
from degree_monic_char_poly[OF this] have A: "char_poly A ≠ 0" by auto
from mem obtain as bs where nas: "n_as = as @ (n,a) # bs"
by (meson split_list)
from jordan_nf_char_poly[OF jnf]
have cA: "char_poly A = (∏(n, a)←n_as. [:- a, 1:] ^ n)" .
also have "… = [: -a, 1:] ^ n * (∏(n, a)← as @ bs. [:- a, 1:] ^ n)" unfolding nas by auto
also have "[: -a,1 :] ^ n dvd …" unfolding dvd_def by blast
finally have "[: -a,1 :] ^ n dvd char_poly A" by auto
from order_max[OF this A] show ?thesis .
qed
lemma similar_mat_jordan_block_smult: fixes A :: "'a :: field mat"
assumes "similar_mat A (jordan_block n a)"
and k: "k ≠ 0"
shows "similar_mat (k ⋅⇩m A) (jordan_block n (k * a))"
proof -
let ?J = "jordan_block n a"
let ?Jk = "jordan_block n (k * a)"
let ?kJ = "k ⋅⇩m jordan_block n a"
from k have inv: "k ^ i ≠ 0" for i by auto
let ?A = "mat_diag n (λ i. k^i)"
let ?B = "mat_diag n (λ i. inverse (k^i))"
have "similar_mat_wit ?Jk ?kJ ?A ?B"
proof (rule similar_mat_witI)
show "jordan_block n (k * a) = ?A * ?kJ * ?B"
by (subst mat_diag_mult_left[of _ _ n], force, subst mat_diag_mult_right[of _ n],
insert k inv, auto simp: jordan_block_def field_simps intro!: eq_matI)
qed (auto simp: inv field_simps k)
hence kJ: "similar_mat ?Jk ?kJ"
unfolding similar_mat_def by auto
have "similar_mat A ?J" by fact
hence "similar_mat (k ⋅⇩m A) (k ⋅⇩m ?J)" by (rule similar_mat_smult)
with kJ show ?thesis
using similar_mat_sym similar_mat_trans by blast
qed
lemma jordan_matrix_Cons: "jordan_matrix (Cons (n,a) n_as) = four_block_mat
(jordan_block n a) (0⇩m n (sum_list (map fst n_as)))
(0⇩m (sum_list (map fst n_as)) n) (jordan_matrix n_as)"
unfolding jordan_matrix_def by (simp, simp add: jordan_matrix_def[symmetric])
lemma similar_mat_jordan_matrix_smult: fixes n_as :: "(nat × 'a :: field) list"
assumes k: "k ≠ 0"
shows "similar_mat (k ⋅⇩m jordan_matrix n_as) (jordan_matrix (map (λ (n,a). (n, k * a)) n_as))"
proof (induct n_as)
case Nil
show ?case by (auto simp: jordan_matrix_def intro!: similar_mat_refl)
next
case (Cons na n_as)
obtain n a where na: "na = (n,a)" by force
let ?l = "map (λ (n,a). (n, k * a))"
let ?n = "sum_list (map fst n_as)"
have "k ⋅⇩m jordan_matrix (Cons na n_as) = k ⋅⇩m four_block_mat
(jordan_block n a) (0⇩m n ?n)
(0⇩m ?n n) (jordan_matrix n_as)" (is "?M = _ ⋅⇩m four_block_mat ?A ?B ?C ?D")
by (simp add: na jordan_matrix_Cons)
also have "… = four_block_mat (k ⋅⇩m ?A) ?B ?C (k ⋅⇩m ?D)"
by (subst smult_four_block_mat, auto)
finally have jm: "?M = four_block_mat (k ⋅⇩m ?A) ?B ?C (k ⋅⇩m ?D)" .
have [simp]: "fst (case x of (n :: nat, a) ⇒ (n, k * a)) = fst x" for x by (cases x, auto)
have jmk: "jordan_matrix (?l (Cons na n_as)) = four_block_mat
(jordan_block n (k * a)) ?B
?C (jordan_matrix (?l n_as))" (is "?kM = four_block_mat ?kA _ _ ?kD")
by (simp add: na jordan_matrix_Cons o_def)
show ?case unfolding jmk jm
by (rule similar_mat_four_block_0_0[OF similar_mat_jordan_block_smult[OF _ k] Cons],
auto intro!: similar_mat_refl)
qed
lemma jordan_nf_smult: fixes k :: "'a :: field"
assumes jn: "jordan_nf A n_as"
and k: "k ≠ 0"
shows "jordan_nf (k ⋅⇩m A) (map (λ (n,a). (n, k * a)) n_as)"
proof -
let ?l = "map (λ (n,a). (n, k * a))"
from jn[unfolded jordan_nf_def] have sim: "similar_mat A (jordan_matrix n_as)" by auto
from similar_mat_smult[OF this, of k] similar_mat_jordan_matrix_smult[OF k, of n_as]
have "similar_mat (k ⋅⇩m A) (jordan_matrix (map (λ(n, a). (n, k * a)) n_as))"
using similar_mat_trans by blast
with jn show ?thesis unfolding jordan_nf_def by force
qed
lemma jordan_nf_order: assumes "jordan_nf A n_as"
shows "order a (char_poly A) = sum_list (map fst (filter (λ na. snd na = a) n_as))"
proof -
let ?p = "λ n_as. (∏(n, a)←n_as. [:- a, 1:] ^ n)"
let ?s = "λ n_as. sum_list (map fst (filter (λ na. snd na = a) n_as))"
from jordan_nf_char_poly[OF assms]
have "order a (char_poly A) = order a (?p n_as)" by simp
also have "… = ?s n_as"
proof (induct n_as)
case (Cons nb n_as)
obtain n b where nb: "nb = (n,b)" by force
have "order a (?p (nb # n_as)) = order a ([: -b, 1:] ^ n * ?p n_as)" unfolding nb by simp
also have "… = order a ([: -b, 1:] ^ n) + order a (?p n_as)"
by (rule order_mult, auto simp: prod_list_zero_iff)
also have "… = (if a = b then n else 0) + ?s n_as" unfolding Cons order_linear_power by simp
also have "… = ?s (nb # n_as)" unfolding nb by auto
finally show ?case .
qed simp
finally show ?thesis .
qed
subsection ‹Application for Complexity›
lemma factored_char_poly_norm_bound: assumes A: "A ∈ carrier_mat n n"
and linear_factors: "char_poly A = (∏ (a :: 'a :: real_normed_field) ← as. [:- a, 1:])"
and jnf_exists: "∃ n_as. jordan_nf A n_as"
and le_1: "⋀ a. a ∈ set as ⟹ norm a ≤ 1"
and le_N: "⋀ a. a ∈ set as ⟹ norm a = 1 ⟹ length (filter ((=) a) as) ≤ N"
shows "∃ c1 c2. ∀ k. norm_bound (A ^⇩m k) (c1 + c2 * of_nat k ^ (N - 1))"
proof -
from jnf_exists obtain n_as
where jnf: "jordan_nf A n_as" by auto
let ?cp1 = "(∏ a ← as. [:- a, 1:])"
let ?cp2 = "∏(n, a)←n_as. [:- a, 1:] ^ n"
let ?J = "jordan_matrix n_as"
from jnf[unfolded jordan_nf_def]
have sim: "similar_mat A ?J" by auto
from char_poly_similar[OF sim, unfolded linear_factors jordan_matrix_char_poly]
have cp: "?cp1 = ?cp2" .
show ?thesis
proof (rule jordan_nf_matrix_poly_bound[OF A _ _ jnf])
fix n a
assume na: "(n,a) ∈ set n_as"
then obtain na1 na2 where n_as: "n_as = na1 @ (n,a) # na2"
unfolding in_set_conv_decomp by auto
then obtain p where "?cp2 = [: -a, 1 :]^n * p" unfolding n_as by auto
from cp[unfolded this] have dvd: "[: -a, 1 :] ^ n dvd ?cp1" by auto
let ?as = "filter ((=) a) as"
let ?pn = "λ as. ∏a←as. [:- a, 1:]"
let ?p = "λ as. ∏a←as. [: a, 1:]"
have "?pn as = ?p (map uminus as)" by (induct as, auto)
from poly_linear_exp_linear_factors[OF dvd[unfolded this]]
have "n ≤ length (filter ((=) (- a)) (map uminus as))" .
also have "… = length (filter ((=) a) as)"
by (induct as, auto)
finally have filt: "n ≤ length (filter ((=) a) as)" .
{
assume "0 < n"
with filt obtain b bs where "?as = b # bs" by (cases ?as, auto)
from arg_cong[OF this, of set]
have "a ∈ set as" by auto
from le_1[rule_format, OF this]
show "norm a ≤ 1" .
note ‹a ∈ set as›
} note mem = this
{
assume "norm a = 1"
from le_N[OF mem this] filt show "n ≤ N" by (cases n, auto)
}
qed
qed
end