src/HOL/Library/Inner_Product.thy
author huffman
Thu Jun 11 09:03:24 2009 -0700 (2009-06-11)
changeset 31563 ded2364d14d4
parent 31492 5400beeddb55
child 31590 776d6a4c1327
permissions -rw-r--r--
cleaned up some proofs
huffman@29993
     1
(* Title:      Inner_Product.thy
huffman@29993
     2
   Author:     Brian Huffman
huffman@29993
     3
*)
huffman@29993
     4
huffman@29993
     5
header {* Inner Product Spaces and the Gradient Derivative *}
huffman@29993
     6
huffman@29993
     7
theory Inner_Product
haftmann@30663
     8
imports Complex_Main FrechetDeriv
huffman@29993
     9
begin
huffman@29993
    10
huffman@29993
    11
subsection {* Real inner product spaces *}
huffman@29993
    12
huffman@31492
    13
text {*
huffman@31492
    14
  Temporarily relax type constraints for @{term "open"},
huffman@31492
    15
  @{term dist}, and @{term norm}.
huffman@31492
    16
*}
huffman@31492
    17
huffman@31492
    18
setup {* Sign.add_const_constraint
huffman@31492
    19
  (@{const_name "open"}, SOME @{typ "'a::open set \<Rightarrow> bool"}) *}
huffman@31446
    20
huffman@31446
    21
setup {* Sign.add_const_constraint
huffman@31446
    22
  (@{const_name dist}, SOME @{typ "'a::dist \<Rightarrow> 'a \<Rightarrow> real"}) *}
huffman@31446
    23
huffman@31446
    24
setup {* Sign.add_const_constraint
huffman@31446
    25
  (@{const_name norm}, SOME @{typ "'a::norm \<Rightarrow> real"}) *}
huffman@31446
    26
huffman@31492
    27
class real_inner = real_vector + sgn_div_norm + dist_norm + open_dist +
huffman@29993
    28
  fixes inner :: "'a \<Rightarrow> 'a \<Rightarrow> real"
huffman@29993
    29
  assumes inner_commute: "inner x y = inner y x"
huffman@29993
    30
  and inner_left_distrib: "inner (x + y) z = inner x z + inner y z"
huffman@29993
    31
  and inner_scaleR_left: "inner (scaleR r x) y = r * (inner x y)"
huffman@29993
    32
  and inner_ge_zero [simp]: "0 \<le> inner x x"
huffman@29993
    33
  and inner_eq_zero_iff [simp]: "inner x x = 0 \<longleftrightarrow> x = 0"
huffman@29993
    34
  and norm_eq_sqrt_inner: "norm x = sqrt (inner x x)"
huffman@29993
    35
begin
huffman@29993
    36
huffman@29993
    37
lemma inner_zero_left [simp]: "inner 0 x = 0"
huffman@30067
    38
  using inner_left_distrib [of 0 0 x] by simp
huffman@29993
    39
huffman@29993
    40
lemma inner_minus_left [simp]: "inner (- x) y = - inner x y"
huffman@30067
    41
  using inner_left_distrib [of x "- x" y] by simp
huffman@29993
    42
huffman@29993
    43
lemma inner_diff_left: "inner (x - y) z = inner x z - inner y z"
huffman@29993
    44
  by (simp add: diff_minus inner_left_distrib)
huffman@29993
    45
huffman@29993
    46
text {* Transfer distributivity rules to right argument. *}
huffman@29993
    47
huffman@29993
    48
lemma inner_right_distrib: "inner x (y + z) = inner x y + inner x z"
huffman@29993
    49
  using inner_left_distrib [of y z x] by (simp only: inner_commute)
huffman@29993
    50
huffman@29993
    51
lemma inner_scaleR_right: "inner x (scaleR r y) = r * (inner x y)"
huffman@29993
    52
  using inner_scaleR_left [of r y x] by (simp only: inner_commute)
huffman@29993
    53
huffman@29993
    54
lemma inner_zero_right [simp]: "inner x 0 = 0"
huffman@29993
    55
  using inner_zero_left [of x] by (simp only: inner_commute)
huffman@29993
    56
huffman@29993
    57
lemma inner_minus_right [simp]: "inner x (- y) = - inner x y"
huffman@29993
    58
  using inner_minus_left [of y x] by (simp only: inner_commute)
huffman@29993
    59
huffman@29993
    60
lemma inner_diff_right: "inner x (y - z) = inner x y - inner x z"
huffman@29993
    61
  using inner_diff_left [of y z x] by (simp only: inner_commute)
huffman@29993
    62
huffman@29993
    63
lemmas inner_distrib = inner_left_distrib inner_right_distrib
huffman@29993
    64
lemmas inner_diff = inner_diff_left inner_diff_right
huffman@29993
    65
lemmas inner_scaleR = inner_scaleR_left inner_scaleR_right
huffman@29993
    66
huffman@29993
    67
lemma inner_gt_zero_iff [simp]: "0 < inner x x \<longleftrightarrow> x \<noteq> 0"
huffman@29993
    68
  by (simp add: order_less_le)
huffman@29993
    69
huffman@29993
    70
lemma power2_norm_eq_inner: "(norm x)\<twosuperior> = inner x x"
huffman@29993
    71
  by (simp add: norm_eq_sqrt_inner)
huffman@29993
    72
huffman@30046
    73
lemma Cauchy_Schwarz_ineq:
huffman@29993
    74
  "(inner x y)\<twosuperior> \<le> inner x x * inner y y"
huffman@29993
    75
proof (cases)
huffman@29993
    76
  assume "y = 0"
huffman@29993
    77
  thus ?thesis by simp
huffman@29993
    78
next
huffman@29993
    79
  assume y: "y \<noteq> 0"
huffman@29993
    80
  let ?r = "inner x y / inner y y"
huffman@29993
    81
  have "0 \<le> inner (x - scaleR ?r y) (x - scaleR ?r y)"
huffman@29993
    82
    by (rule inner_ge_zero)
huffman@29993
    83
  also have "\<dots> = inner x x - inner y x * ?r"
huffman@29993
    84
    by (simp add: inner_diff inner_scaleR)
huffman@29993
    85
  also have "\<dots> = inner x x - (inner x y)\<twosuperior> / inner y y"
huffman@29993
    86
    by (simp add: power2_eq_square inner_commute)
huffman@29993
    87
  finally have "0 \<le> inner x x - (inner x y)\<twosuperior> / inner y y" .
huffman@29993
    88
  hence "(inner x y)\<twosuperior> / inner y y \<le> inner x x"
huffman@29993
    89
    by (simp add: le_diff_eq)
huffman@29993
    90
  thus "(inner x y)\<twosuperior> \<le> inner x x * inner y y"
huffman@29993
    91
    by (simp add: pos_divide_le_eq y)
huffman@29993
    92
qed
huffman@29993
    93
huffman@30046
    94
lemma Cauchy_Schwarz_ineq2:
huffman@29993
    95
  "\<bar>inner x y\<bar> \<le> norm x * norm y"
huffman@29993
    96
proof (rule power2_le_imp_le)
huffman@29993
    97
  have "(inner x y)\<twosuperior> \<le> inner x x * inner y y"
huffman@30046
    98
    using Cauchy_Schwarz_ineq .
huffman@29993
    99
  thus "\<bar>inner x y\<bar>\<twosuperior> \<le> (norm x * norm y)\<twosuperior>"
huffman@29993
   100
    by (simp add: power_mult_distrib power2_norm_eq_inner)
huffman@29993
   101
  show "0 \<le> norm x * norm y"
huffman@29993
   102
    unfolding norm_eq_sqrt_inner
huffman@29993
   103
    by (intro mult_nonneg_nonneg real_sqrt_ge_zero inner_ge_zero)
huffman@29993
   104
qed
huffman@29993
   105
huffman@29993
   106
subclass real_normed_vector
huffman@29993
   107
proof
huffman@29993
   108
  fix a :: real and x y :: 'a
huffman@29993
   109
  show "0 \<le> norm x"
huffman@29993
   110
    unfolding norm_eq_sqrt_inner by simp
huffman@29993
   111
  show "norm x = 0 \<longleftrightarrow> x = 0"
huffman@29993
   112
    unfolding norm_eq_sqrt_inner by simp
huffman@29993
   113
  show "norm (x + y) \<le> norm x + norm y"
huffman@29993
   114
    proof (rule power2_le_imp_le)
huffman@29993
   115
      have "inner x y \<le> norm x * norm y"
huffman@30046
   116
        by (rule order_trans [OF abs_ge_self Cauchy_Schwarz_ineq2])
huffman@29993
   117
      thus "(norm (x + y))\<twosuperior> \<le> (norm x + norm y)\<twosuperior>"
huffman@29993
   118
        unfolding power2_sum power2_norm_eq_inner
huffman@29993
   119
        by (simp add: inner_distrib inner_commute)
huffman@29993
   120
      show "0 \<le> norm x + norm y"
huffman@29993
   121
        unfolding norm_eq_sqrt_inner
huffman@29993
   122
        by (simp add: add_nonneg_nonneg)
huffman@29993
   123
    qed
huffman@29993
   124
  have "sqrt (a\<twosuperior> * inner x x) = \<bar>a\<bar> * sqrt (inner x x)"
huffman@29993
   125
    by (simp add: real_sqrt_mult_distrib)
huffman@29993
   126
  then show "norm (a *\<^sub>R x) = \<bar>a\<bar> * norm x"
huffman@29993
   127
    unfolding norm_eq_sqrt_inner
huffman@29993
   128
    by (simp add: inner_scaleR power2_eq_square mult_assoc)
huffman@29993
   129
qed
huffman@29993
   130
huffman@29993
   131
end
huffman@29993
   132
huffman@31492
   133
text {*
huffman@31492
   134
  Re-enable constraints for @{term "open"},
huffman@31492
   135
  @{term dist}, and @{term norm}.
huffman@31492
   136
*}
huffman@31492
   137
huffman@31492
   138
setup {* Sign.add_const_constraint
huffman@31492
   139
  (@{const_name "open"}, SOME @{typ "'a::topological_space set \<Rightarrow> bool"}) *}
huffman@31446
   140
huffman@31446
   141
setup {* Sign.add_const_constraint
huffman@31446
   142
  (@{const_name dist}, SOME @{typ "'a::metric_space \<Rightarrow> 'a \<Rightarrow> real"}) *}
huffman@31446
   143
huffman@31446
   144
setup {* Sign.add_const_constraint
huffman@31446
   145
  (@{const_name norm}, SOME @{typ "'a::real_normed_vector \<Rightarrow> real"}) *}
huffman@31446
   146
wenzelm@30729
   147
interpretation inner:
huffman@29993
   148
  bounded_bilinear "inner::'a::real_inner \<Rightarrow> 'a \<Rightarrow> real"
huffman@29993
   149
proof
huffman@29993
   150
  fix x y z :: 'a and r :: real
huffman@29993
   151
  show "inner (x + y) z = inner x z + inner y z"
huffman@29993
   152
    by (rule inner_left_distrib)
huffman@29993
   153
  show "inner x (y + z) = inner x y + inner x z"
huffman@29993
   154
    by (rule inner_right_distrib)
huffman@29993
   155
  show "inner (scaleR r x) y = scaleR r (inner x y)"
huffman@29993
   156
    unfolding real_scaleR_def by (rule inner_scaleR_left)
huffman@29993
   157
  show "inner x (scaleR r y) = scaleR r (inner x y)"
huffman@29993
   158
    unfolding real_scaleR_def by (rule inner_scaleR_right)
huffman@29993
   159
  show "\<exists>K. \<forall>x y::'a. norm (inner x y) \<le> norm x * norm y * K"
huffman@29993
   160
  proof
huffman@29993
   161
    show "\<forall>x y::'a. norm (inner x y) \<le> norm x * norm y * 1"
huffman@30046
   162
      by (simp add: Cauchy_Schwarz_ineq2)
huffman@29993
   163
  qed
huffman@29993
   164
qed
huffman@29993
   165
wenzelm@30729
   166
interpretation inner_left:
huffman@29993
   167
  bounded_linear "\<lambda>x::'a::real_inner. inner x y"
huffman@29993
   168
  by (rule inner.bounded_linear_left)
huffman@29993
   169
wenzelm@30729
   170
interpretation inner_right:
huffman@29993
   171
  bounded_linear "\<lambda>y::'a::real_inner. inner x y"
huffman@29993
   172
  by (rule inner.bounded_linear_right)
huffman@29993
   173
huffman@29993
   174
huffman@29993
   175
subsection {* Class instances *}
huffman@29993
   176
huffman@29993
   177
instantiation real :: real_inner
huffman@29993
   178
begin
huffman@29993
   179
huffman@29993
   180
definition inner_real_def [simp]: "inner = op *"
huffman@29993
   181
huffman@29993
   182
instance proof
huffman@29993
   183
  fix x y z r :: real
huffman@29993
   184
  show "inner x y = inner y x"
huffman@29993
   185
    unfolding inner_real_def by (rule mult_commute)
huffman@29993
   186
  show "inner (x + y) z = inner x z + inner y z"
huffman@29993
   187
    unfolding inner_real_def by (rule left_distrib)
huffman@29993
   188
  show "inner (scaleR r x) y = r * inner x y"
huffman@29993
   189
    unfolding inner_real_def real_scaleR_def by (rule mult_assoc)
huffman@29993
   190
  show "0 \<le> inner x x"
huffman@29993
   191
    unfolding inner_real_def by simp
huffman@29993
   192
  show "inner x x = 0 \<longleftrightarrow> x = 0"
huffman@29993
   193
    unfolding inner_real_def by simp
huffman@29993
   194
  show "norm x = sqrt (inner x x)"
huffman@29993
   195
    unfolding inner_real_def by simp
huffman@29993
   196
qed
huffman@29993
   197
huffman@29993
   198
end
huffman@29993
   199
huffman@29993
   200
instantiation complex :: real_inner
huffman@29993
   201
begin
huffman@29993
   202
huffman@29993
   203
definition inner_complex_def:
huffman@29993
   204
  "inner x y = Re x * Re y + Im x * Im y"
huffman@29993
   205
huffman@29993
   206
instance proof
huffman@29993
   207
  fix x y z :: complex and r :: real
huffman@29993
   208
  show "inner x y = inner y x"
huffman@29993
   209
    unfolding inner_complex_def by (simp add: mult_commute)
huffman@29993
   210
  show "inner (x + y) z = inner x z + inner y z"
huffman@29993
   211
    unfolding inner_complex_def by (simp add: left_distrib)
huffman@29993
   212
  show "inner (scaleR r x) y = r * inner x y"
huffman@29993
   213
    unfolding inner_complex_def by (simp add: right_distrib)
huffman@29993
   214
  show "0 \<le> inner x x"
huffman@29993
   215
    unfolding inner_complex_def by (simp add: add_nonneg_nonneg)
huffman@29993
   216
  show "inner x x = 0 \<longleftrightarrow> x = 0"
huffman@29993
   217
    unfolding inner_complex_def
huffman@29993
   218
    by (simp add: add_nonneg_eq_0_iff complex_Re_Im_cancel_iff)
huffman@29993
   219
  show "norm x = sqrt (inner x x)"
huffman@29993
   220
    unfolding inner_complex_def complex_norm_def
huffman@29993
   221
    by (simp add: power2_eq_square)
huffman@29993
   222
qed
huffman@29993
   223
huffman@29993
   224
end
huffman@29993
   225
huffman@29993
   226
huffman@29993
   227
subsection {* Gradient derivative *}
huffman@29993
   228
huffman@29993
   229
definition
huffman@29993
   230
  gderiv ::
huffman@29993
   231
    "['a::real_inner \<Rightarrow> real, 'a, 'a] \<Rightarrow> bool"
huffman@29993
   232
          ("(GDERIV (_)/ (_)/ :> (_))" [1000, 1000, 60] 60)
huffman@29993
   233
where
huffman@29993
   234
  "GDERIV f x :> D \<longleftrightarrow> FDERIV f x :> (\<lambda>h. inner h D)"
huffman@29993
   235
huffman@29993
   236
lemma deriv_fderiv: "DERIV f x :> D \<longleftrightarrow> FDERIV f x :> (\<lambda>h. h * D)"
huffman@29993
   237
  by (simp only: deriv_def field_fderiv_def)
huffman@29993
   238
huffman@29993
   239
lemma gderiv_deriv [simp]: "GDERIV f x :> D \<longleftrightarrow> DERIV f x :> D"
huffman@29993
   240
  by (simp only: gderiv_def deriv_fderiv inner_real_def)
huffman@29993
   241
huffman@29993
   242
lemma GDERIV_DERIV_compose:
huffman@29993
   243
    "\<lbrakk>GDERIV f x :> df; DERIV g (f x) :> dg\<rbrakk>
huffman@29993
   244
     \<Longrightarrow> GDERIV (\<lambda>x. g (f x)) x :> scaleR dg df"
huffman@29993
   245
  unfolding gderiv_def deriv_fderiv
huffman@29993
   246
  apply (drule (1) FDERIV_compose)
huffman@29993
   247
  apply (simp add: inner_scaleR_right mult_ac)
huffman@29993
   248
  done
huffman@29993
   249
huffman@29993
   250
lemma FDERIV_subst: "\<lbrakk>FDERIV f x :> df; df = d\<rbrakk> \<Longrightarrow> FDERIV f x :> d"
huffman@29993
   251
  by simp
huffman@29993
   252
huffman@29993
   253
lemma GDERIV_subst: "\<lbrakk>GDERIV f x :> df; df = d\<rbrakk> \<Longrightarrow> GDERIV f x :> d"
huffman@29993
   254
  by simp
huffman@29993
   255
huffman@29993
   256
lemma GDERIV_const: "GDERIV (\<lambda>x. k) x :> 0"
huffman@29993
   257
  unfolding gderiv_def inner_right.zero by (rule FDERIV_const)
huffman@29993
   258
huffman@29993
   259
lemma GDERIV_add:
huffman@29993
   260
    "\<lbrakk>GDERIV f x :> df; GDERIV g x :> dg\<rbrakk>
huffman@29993
   261
     \<Longrightarrow> GDERIV (\<lambda>x. f x + g x) x :> df + dg"
huffman@29993
   262
  unfolding gderiv_def inner_right.add by (rule FDERIV_add)
huffman@29993
   263
huffman@29993
   264
lemma GDERIV_minus:
huffman@29993
   265
    "GDERIV f x :> df \<Longrightarrow> GDERIV (\<lambda>x. - f x) x :> - df"
huffman@29993
   266
  unfolding gderiv_def inner_right.minus by (rule FDERIV_minus)
huffman@29993
   267
huffman@29993
   268
lemma GDERIV_diff:
huffman@29993
   269
    "\<lbrakk>GDERIV f x :> df; GDERIV g x :> dg\<rbrakk>
huffman@29993
   270
     \<Longrightarrow> GDERIV (\<lambda>x. f x - g x) x :> df - dg"
huffman@29993
   271
  unfolding gderiv_def inner_right.diff by (rule FDERIV_diff)
huffman@29993
   272
huffman@29993
   273
lemma GDERIV_scaleR:
huffman@29993
   274
    "\<lbrakk>DERIV f x :> df; GDERIV g x :> dg\<rbrakk>
huffman@29993
   275
     \<Longrightarrow> GDERIV (\<lambda>x. scaleR (f x) (g x)) x
huffman@29993
   276
      :> (scaleR (f x) dg + scaleR df (g x))"
huffman@29993
   277
  unfolding gderiv_def deriv_fderiv inner_right.add inner_right.scaleR
huffman@29993
   278
  apply (rule FDERIV_subst)
huffman@29993
   279
  apply (erule (1) scaleR.FDERIV)
huffman@29993
   280
  apply (simp add: mult_ac)
huffman@29993
   281
  done
huffman@29993
   282
huffman@29993
   283
lemma GDERIV_mult:
huffman@29993
   284
    "\<lbrakk>GDERIV f x :> df; GDERIV g x :> dg\<rbrakk>
huffman@29993
   285
     \<Longrightarrow> GDERIV (\<lambda>x. f x * g x) x :> scaleR (f x) dg + scaleR (g x) df"
huffman@29993
   286
  unfolding gderiv_def
huffman@29993
   287
  apply (rule FDERIV_subst)
huffman@29993
   288
  apply (erule (1) FDERIV_mult)
huffman@29993
   289
  apply (simp add: inner_distrib inner_scaleR mult_ac)
huffman@29993
   290
  done
huffman@29993
   291
huffman@29993
   292
lemma GDERIV_inverse:
huffman@29993
   293
    "\<lbrakk>GDERIV f x :> df; f x \<noteq> 0\<rbrakk>
huffman@29993
   294
     \<Longrightarrow> GDERIV (\<lambda>x. inverse (f x)) x :> - (inverse (f x))\<twosuperior> *\<^sub>R df"
huffman@29993
   295
  apply (erule GDERIV_DERIV_compose)
huffman@29993
   296
  apply (erule DERIV_inverse [folded numeral_2_eq_2])
huffman@29993
   297
  done
huffman@29993
   298
huffman@29993
   299
lemma GDERIV_norm:
huffman@29993
   300
  assumes "x \<noteq> 0" shows "GDERIV (\<lambda>x. norm x) x :> sgn x"
huffman@29993
   301
proof -
huffman@29993
   302
  have 1: "FDERIV (\<lambda>x. inner x x) x :> (\<lambda>h. inner x h + inner h x)"
huffman@29993
   303
    by (intro inner.FDERIV FDERIV_ident)
huffman@29993
   304
  have 2: "(\<lambda>h. inner x h + inner h x) = (\<lambda>h. inner h (scaleR 2 x))"
huffman@29993
   305
    by (simp add: expand_fun_eq inner_scaleR inner_commute)
huffman@29993
   306
  have "0 < inner x x" using `x \<noteq> 0` by simp
huffman@29993
   307
  then have 3: "DERIV sqrt (inner x x) :> (inverse (sqrt (inner x x)) / 2)"
huffman@29993
   308
    by (rule DERIV_real_sqrt)
huffman@29993
   309
  have 4: "(inverse (sqrt (inner x x)) / 2) *\<^sub>R 2 *\<^sub>R x = sgn x"
huffman@29993
   310
    by (simp add: sgn_div_norm norm_eq_sqrt_inner)
huffman@29993
   311
  show ?thesis
huffman@29993
   312
    unfolding norm_eq_sqrt_inner
huffman@29993
   313
    apply (rule GDERIV_subst [OF _ 4])
huffman@29993
   314
    apply (rule GDERIV_DERIV_compose [where g=sqrt and df="scaleR 2 x"])
huffman@29993
   315
    apply (subst gderiv_def)
huffman@29993
   316
    apply (rule FDERIV_subst [OF _ 2])
huffman@29993
   317
    apply (rule 1)
huffman@29993
   318
    apply (rule 3)
huffman@29993
   319
    done
huffman@29993
   320
qed
huffman@29993
   321
huffman@29993
   322
lemmas FDERIV_norm = GDERIV_norm [unfolded gderiv_def]
huffman@29993
   323
huffman@29993
   324
end