This repository was archived by the owner on Jul 24, 2024. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 289
Expand file tree
/
Copy pathring_exp.lean
More file actions
1560 lines (1340 loc) · 56 KB
/
Copy pathring_exp.lean
File metadata and controls
1560 lines (1340 loc) · 56 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/-
Copyright (c) 2019 Tim Baanen. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Authors: Tim Baanen
-/
import tactic.norm_num
import control.traversable.basic
/-!
# `ring_exp` tactic
A tactic for solving equations in commutative (semi)rings,
where the exponents can also contain variables.
More precisely, expressions of the following form are supported:
- constants (non-negative integers)
- variables
- coefficients (any rational number, embedded into the (semi)ring)
- addition of expressions
- multiplication of expressions
- exponentiation of expressions (the exponent must have type `ℕ`)
- subtraction and negation of expressions (if the base is a full ring)
The motivating example is proving `2 * 2^n * b = b * 2^(n+1)`,
something that the `ring` tactic cannot do, but `ring_exp` can.
## Implementation notes
The basic approach to prove equalities is to normalise both sides and check for equality.
The normalisation is guided by building a value in the type `ex` at the meta level,
together with a proof (at the base level) that the original value is equal to
the normalised version.
The normalised version and normalisation proofs are also stored in the `ex` type.
The outline of the file:
- Define an inductive family of types `ex`, parametrised over `ex_type`,
which can represent expressions with `+`, `*`, `^` and rational numerals.
The parametrisation over `ex_type` ensures that associativity and distributivity are applied,
by restricting which kinds of subexpressions appear as arguments to the various operators.
- Represent addition, multiplication and exponentiation in the `ex` type,
thus allowing us to map expressions to `ex` (the `eval` function drives this).
We apply associativity and distributivity of the operators here (helped by `ex_type`)
and commutativity as well (by sorting the subterms; unfortunately not helped by anything).
Any expression not of the above formats is treated as an atom (the same as a variable).
There are some details we glossed over which make the plan more complicated:
- The order on atoms is not initially obvious.
We construct a list containing them in order of initial appearance in the expression,
then use the index into the list as a key to order on.
- In the tactic, a normalized expression `ps : ex` lives in the meta-world,
but the normalization proofs live in the real world.
Thus, we cannot directly say `ps.orig = ps.pretty` anywhere,
but we have to carefully construct the proof when we compute `ps`.
This was a major source of bugs in development!
- For `pow`, the exponent must be a natural number, while the base can be any semiring `α`.
We swap out operations for the base ring `α` with those for the exponent ring `ℕ`
as soon as we deal with exponents.
This is accomplished by the `in_exponent` function and is relatively painless since
we work in a `reader` monad.
- The normalized form of an expression is the one that is useful for the tactic,
but not as nice to read. To remedy this, the user-facing normalization calls `ex.simp`.
## Caveats and future work
Subtraction cancels out identical terms, but division does not.
That is: `a - a = 0 := by ring_exp` solves the goal,
but `a / a := 1 by ring_exp` doesn't.
Note that `0 / 0` is generally defined to be `0`,
so division cancelling out is not true in general.
Multiplication of powers can be simplified a little bit further:
`2 ^ n * 2 ^ n = 4 ^ n := by ring_exp` could be implemented
in a similar way that `2 * a + 2 * a = 4 * a := by ring_exp` already works.
This feature wasn't needed yet, so it's not implemented yet.
## Tags
ring, semiring, exponent, power
-/
-- The base ring `α` will have a universe level `u`.
-- We do not introduce `α` as a variable yet,
-- in order to make it explicit or implicit as required.
universes u
namespace tactic.ring_exp
open nat
/--
The `atom` structure is used to represent atomic expressions:
those which `ring_exp` cannot parse any further.
For instance, `a + (a % b)` has `a` and `(a % b)` as atoms.
The `ring_exp_eq` tactic does not normalize the subexpressions in atoms,
but `ring_exp` does if `ring_exp_eq` was not sufficient.
Atoms in fact represent equivalence classes of expressions,
modulo definitional equality.
The field `index : ℕ` should be a unique number for each class,
while `value : expr` contains a representative of this class.
The function `resolve_atom` determines the appropriate atom
for a given expression.
-/
meta structure atom : Type := (value : expr) (index : ℕ)
namespace atom
/--
The `eq` operation on `atom`s works modulo definitional equality,
ignoring their `value`s.
The invariants on `atom` ensure indices are unique per value.
Thus, `eq` indicates equality as long as the `atom`s come from the same context.
-/
meta def eq (a b : atom) : bool := a.index = b.index
/--
We order `atom`s on the order of appearance in the main expression.
-/
meta def lt (a b : atom) : bool := a.index < b.index
meta instance : has_repr atom := ⟨λ x, "(atom " ++ repr x.2 ++ ")"⟩
end atom
section expression
/-!
### `expression` section
In this section, we define the `ex` type and its basic operations.
First, we introduce the supporting types `coeff`, `ex_type` and `ex_info`.
For understanding the code, it's easier to check out `ex` itself first,
then refer back to the supporting types.
The arithmetic operations on `ex` need additional definitions,
so they are defined in a later section.
-/
/--
Coefficients in the expression are stored in a wrapper structure,
allowing for easier modification of the data structures.
The modifications might be caching of the result of `expr.of_rat`,
or using a different meta representation of numerals.
-/
@[derive decidable_eq, derive inhabited]
structure coeff : Type := (value : ℚ)
/-- The values in `ex_type` are used as parameters to `ex` to control the expression's structure. -/
@[derive decidable_eq, derive inhabited]
inductive ex_type : Type
| base : ex_type
| sum : ex_type
| prod : ex_type
| exp : ex_type
open ex_type
/--
Each `ex` stores information for its normalization proof.
The `orig` expression is the expression that was passed to `eval`.
The `pretty` expression is the normalised form that the `ex` represents.
(I didn't call this something like `norm`, because there are already
too many things called `norm` in mathematics!)
The field `proof` contains an optional proof term of type `%%orig = %%pretty`.
The value `none` for the proof indicates that everything reduces to reflexivity.
(Which saves space in quite a lot of cases.)
-/
meta structure ex_info : Type :=
(orig : expr) (pretty : expr) (proof : option expr)
/--
The `ex` type is an abstract representation of an expression with `+`, `*` and `^`.
Those operators are mapped to the `sum`, `prod` and `exp` constructors respectively.
The `zero` constructor is the base case for `ex sum`, e.g. `1 + 2` is represented
by (something along the lines of) `sum 1 (sum 2 zero)`.
The `coeff` constructor is the base case for `ex prod`, and is used for numerals.
The code maintains the invariant that the coefficient is never `0`.
The `var` constructor is the base case for `ex exp`, and is used for atoms.
The `sum_b` constructor allows for addition in the base of an exponentiation;
it serves a similar purpose as the parentheses in `(a + b)^c`.
The code maintains the invariant that the argument to `sum_b` is not `zero`
or `sum _ zero`.
All of the constructors contain an `ex_info` field,
used to carry around (arguments to) proof terms.
While the `ex_type` parameter enforces some simplification invariants,
the following ones must be manually maintained at the risk of insufficient power:
- the argument to `coeff` must be nonzero (to ensure `0 = 0 * 1`)
- the argument to `sum_b` must be of the form `sum a (sum b bs)` (to ensure `(a + 0)^n = a^n`)
- normalisation proofs of subexpressions must be `refl ps.pretty`
- if we replace `sum` with `cons` and `zero` with `nil`, the resulting list is sorted
according to the `lt` relation defined further down; similarly for `prod` and `coeff`
(to ensure `a + b = b + a`).
The first two invariants could be encoded in a subtype of `ex`,
but aren't (yet) to spare some implementation burden.
The other invariants cannot be encoded because we need the `tactic` monad to check them.
(For example, the correct equality check of `expr` is `is_def_eq : expr → expr → tactic unit`.)
-/
meta inductive ex : ex_type → Type
| zero (info : ex_info) : ex sum
| sum (info : ex_info) : ex prod → ex sum → ex sum
| coeff (info : ex_info) : coeff → ex prod
| prod (info : ex_info) : ex exp → ex prod → ex prod
| var (info : ex_info) : atom → ex base
| sum_b (info : ex_info) : ex sum → ex base
| exp (info : ex_info) : ex base → ex prod → ex exp
/--
Return the proof information associated to the `ex`.
-/
meta def ex.info : Π {et : ex_type} (ps : ex et), ex_info
| sum (ex.zero i) := i
| sum (ex.sum i _ _) := i
| prod (ex.coeff i _) := i
| prod (ex.prod i _ _) := i
| base (ex.var i _) := i
| base (ex.sum_b i _) := i
| exp (ex.exp i _ _) := i
/--
Return the original, non-normalized version of this `ex`.
Note that arguments to another `ex` are always "pre-normalized":
their `orig` and `pretty` are equal, and their `proof` is reflexivity.
-/
meta def ex.orig {et : ex_type} (ps : ex et) : expr := ps.info.orig
/--
Return the normalized version of this `ex`.
-/
meta def ex.pretty {et : ex_type} (ps : ex et) : expr := ps.info.pretty
/--
Return the normalisation proof of the given expression.
If the proof is `refl`, we give `none` instead,
which helps to control the size of proof terms.
To get an actual term, use `ex.proof_term`,
or use `mk_proof` with the correct set of arguments.
-/
meta def ex.proof {et : ex_type} (ps : ex et) : option expr := ps.info.proof
/--
Update the `orig` and `proof` fields of the `ex_info`.
Intended for use in `ex.set_info`.
-/
meta def ex_info.set (i : ex_info) (o : option expr) (pf : option expr) : ex_info :=
{orig := o.get_or_else i.pretty, proof := pf, .. i}
/--
Update the `ex_info` of the given expression.
We use this to combine intermediate normalisation proofs.
Since `pretty` only depends on the subexpressions,
which do not change, we do not set `pretty`.
-/
meta def ex.set_info : Π {et : ex_type} (ps : ex et), option expr → option expr → ex et
| sum (ex.zero i) o pf := ex.zero (i.set o pf)
| sum (ex.sum i p ps) o pf := ex.sum (i.set o pf) p ps
| prod (ex.coeff i x) o pf := ex.coeff (i.set o pf) x
| prod (ex.prod i p ps) o pf := ex.prod (i.set o pf) p ps
| base (ex.var i x) o pf := ex.var (i.set o pf) x
| base (ex.sum_b i ps) o pf := ex.sum_b (i.set o pf) ps
| exp (ex.exp i p ps) o pf := ex.exp (i.set o pf) p ps
instance coeff_has_repr : has_repr coeff := ⟨λ x, repr x.1⟩
/-- Convert an `ex` to a `string`. -/
meta def ex.repr : Π {et : ex_type}, ex et → string
| sum (ex.zero _) := "0"
| sum (ex.sum _ p ps) := ex.repr p ++ " + " ++ ex.repr ps
| prod (ex.coeff _ x) := repr x
| prod (ex.prod _ p ps) := ex.repr p ++ " * " ++ ex.repr ps
| base (ex.var _ x) := repr x
| base (ex.sum_b _ ps) := "(" ++ ex.repr ps ++ ")"
| exp (ex.exp _ p ps) := ex.repr p ++ " ^ " ++ ex.repr ps
meta instance {et : ex_type} : has_repr (ex et) := ⟨ex.repr⟩
/--
Equality test for expressions.
Since equivalence of `atom`s is not the same as equality,
we cannot make a true `(=)` operator for `ex` either.
-/
meta def ex.eq : Π {et : ex_type}, ex et → ex et → bool
| sum (ex.zero _) (ex.zero _) := tt
| sum (ex.zero _) (ex.sum _ _ _) := ff
| sum (ex.sum _ _ _) (ex.zero _) := ff
| sum (ex.sum _ p ps) (ex.sum _ q qs) := p.eq q && ps.eq qs
| prod (ex.coeff _ x) (ex.coeff _ y) := x = y
| prod (ex.coeff _ _) (ex.prod _ _ _) := ff
| prod (ex.prod _ _ _) (ex.coeff _ _) := ff
| prod (ex.prod _ p ps) (ex.prod _ q qs) := p.eq q && ps.eq qs
| base (ex.var _ x) (ex.var _ y) := x.eq y
| base (ex.var _ _) (ex.sum_b _ _) := ff
| base (ex.sum_b _ _) (ex.var _ _) := ff
| base (ex.sum_b _ ps) (ex.sum_b _ qs) := ps.eq qs
| exp (ex.exp _ p ps) (ex.exp _ q qs) := p.eq q && ps.eq qs
/--
The ordering on expressions.
As for `ex.eq`, this is a linear order only in one context.
-/
meta def ex.lt : Π {et : ex_type}, ex et → ex et → bool
| sum _ (ex.zero _) := ff
| sum (ex.zero _) _ := tt
| sum (ex.sum _ p ps) (ex.sum _ q qs) := p.lt q || (p.eq q && ps.lt qs)
| prod (ex.coeff _ x) (ex.coeff _ y) := x.1 < y.1
| prod (ex.coeff _ _) _ := tt
| prod _ (ex.coeff _ _) := ff
| prod (ex.prod _ p ps) (ex.prod _ q qs) := p.lt q || (p.eq q && ps.lt qs)
| base (ex.var _ x) (ex.var _ y) := x.lt y
| base (ex.var _ _) (ex.sum_b _ _) := tt
| base (ex.sum_b _ _) (ex.var _ _) := ff
| base (ex.sum_b _ ps) (ex.sum_b _ qs) := ps.lt qs
| exp (ex.exp _ p ps) (ex.exp _ q qs) := p.lt q || (p.eq q && ps.lt qs)
end expression
section operations
/-!
### `operations` section
This section defines the operations (on `ex`) that use tactics.
They live in the `ring_exp_m` monad,
which adds a cache and a list of encountered atoms to the `tactic` monad.
Throughout this section, we will be constructing proof terms.
The lemmas used in the construction are all defined over a commutative semiring α.
-/
variables {α : Type u} [comm_semiring α]
open tactic
open ex_type
/--
Stores the information needed in the `eval` function and its dependencies,
so they can (re)construct expressions.
The `eval_info` structure stores this information for one type,
and the `context` combines the two types, one for bases and one for exponents.
-/
meta structure eval_info :=
(α : expr) (univ : level)
-- Cache the instances for optimization and consistency
(csr_instance : expr) (ha_instance : expr) (hm_instance : expr) (hp_instance : expr)
-- Optional instances (only required for (-) and (/) respectively)
(ring_instance : option expr) (dr_instance : option expr)
-- Cache common constants.
(zero : expr) (one : expr)
/--
The `context` contains the full set of information needed for the `eval` function.
This structure has two copies of `eval_info`:
one is for the base (typically some semiring `α`) and another for the exponent (always `ℕ`).
When evaluating an exponent, we put `info_e` in `info_b`.
-/
meta structure context :=
(info_b : eval_info) (info_e : eval_info) (transp : transparency)
/--
The `ring_exp_m` monad is used instead of `tactic` to store the context.
-/
@[derive [monad, alternative]]
meta def ring_exp_m (α : Type) : Type := reader_t context (state_t (list atom) tactic) α
/--
Access the instance cache.
-/
meta def get_context : ring_exp_m context := reader_t.read
/--
Lift an operation in the `tactic` monad to the `ring_exp_m` monad.
This operation will not access the cache.
-/
meta def lift {α} (m : tactic α) : ring_exp_m α := reader_t.lift (state_t.lift m)
/--
Change the context of the given computation,
so that expressions are evaluated in the exponent ring,
instead of the base ring.
-/
meta def in_exponent {α} (mx : ring_exp_m α) : ring_exp_m α := do
ctx ← get_context,
reader_t.lift $ mx.run ⟨ctx.info_e, ctx.info_e, ctx.transp⟩
/--
Specialized version of `mk_app` where the first two arguments are `{α}` `[some_class α]`.
Should be faster because it can use the cached instances.
-/
meta def mk_app_class (f : name) (inst : expr) (args : list expr) : ring_exp_m expr := do
ctx ← get_context,
pure $ (@expr.const tt f [ctx.info_b.univ] ctx.info_b.α inst).mk_app args
/--
Specialized version of `mk_app` where the first two arguments are `{α}` `[comm_semiring α]`.
Should be faster because it can use the cached instances.
-/
meta def mk_app_csr (f : name) (args : list expr) : ring_exp_m expr := do
ctx ← get_context,
mk_app_class f (ctx.info_b.csr_instance) args
/--
Specialized version of `mk_app ``has_add.add`.
Should be faster because it can use the cached instances.
-/
meta def mk_add (args : list expr) : ring_exp_m expr := do
ctx ← get_context,
mk_app_class ``has_add.add ctx.info_b.ha_instance args
/--
Specialized version of `mk_app ``has_mul.mul`.
Should be faster because it can use the cached instances.
-/
meta def mk_mul (args : list expr) : ring_exp_m expr := do
ctx ← get_context,
mk_app_class ``has_mul.mul ctx.info_b.hm_instance args
/--
Specialized version of `mk_app ``has_pow.pow`.
Should be faster because it can use the cached instances.
-/
meta def mk_pow (args : list expr) : ring_exp_m expr := do
ctx ← get_context,
pure $ (@expr.const tt ``has_pow.pow
[ctx.info_b.univ, ctx.info_e.univ]
ctx.info_b.α ctx.info_e.α
ctx.info_b.hp_instance).mk_app args
/-- Construct a normalization proof term or return the cached one. -/
meta def ex_info.proof_term (ps : ex_info) : ring_exp_m expr :=
match ps.proof with
| none := lift $ tactic.mk_eq_refl ps.pretty
| (some p) := pure p
end
/-- Construct a normalization proof term or return the cached one. -/
meta def ex.proof_term {et : ex_type} (ps : ex et) : ring_exp_m expr := ps.info.proof_term
/--
If all `ex_info` have trivial proofs, return a trivial proof.
Otherwise, construct all proof terms.
Useful in applications where trivial proofs combine to another trivial proof,
most importantly to pass to `mk_proof_or_refl`.
-/
meta def none_or_proof_term : list ex_info → ring_exp_m (option (list expr))
| [] := pure none
| (x :: xs) := do
xs_pfs ← none_or_proof_term xs,
match (x.proof, xs_pfs) with
| (none, none) := pure none
| (some x_pf, none) := do
xs_pfs ← traverse ex_info.proof_term xs,
pure (some (x_pf :: xs_pfs))
| (_, some xs_pfs) := do
x_pf ← x.proof_term,
pure (some (x_pf :: xs_pfs))
end
/--
Use the proof terms as arguments to the given lemma.
If the lemma could reduce to reflexivity, consider using `mk_proof_or_refl.`
-/
meta def mk_proof (lem : name) (args : list expr) (hs : list ex_info) : ring_exp_m expr := do
hs' ← traverse ex_info.proof_term hs,
mk_app_csr lem (args ++ hs')
/--
Use the proof terms as arguments to the given lemma.
Often, we construct a proof term using congruence where reflexivity suffices.
To solve this, the following function tries to get away with reflexivity.
-/
meta def mk_proof_or_refl (term : expr) (lem : name) (args : list expr) (hs : list ex_info) :
ring_exp_m expr := do
hs_full ← none_or_proof_term hs,
match hs_full with
| none := lift $ mk_eq_refl term
| (some hs') := mk_app_csr lem (args ++ hs')
end
/-- A shortcut for adding the original terms of two expressions. -/
meta def add_orig {et et'} (ps : ex et) (qs : ex et') : ring_exp_m expr :=
mk_add [ps.orig, qs.orig]
/-- A shortcut for multiplying the original terms of two expressions. -/
meta def mul_orig {et et'} (ps : ex et) (qs : ex et') : ring_exp_m expr :=
mk_mul [ps.orig, qs.orig]
/-- A shortcut for exponentiating the original terms of two expressions. -/
meta def pow_orig {et et'} (ps : ex et) (qs : ex et') : ring_exp_m expr :=
mk_pow [ps.orig, qs.orig]
/-- Congruence lemma for constructing `ex.sum`. -/
lemma sum_congr {p p' ps ps' : α} : p = p' → ps = ps' → p + ps = p' + ps' := by cc
/-- Congruence lemma for constructing `ex.prod`. -/
lemma prod_congr {p p' ps ps' : α} : p = p' → ps = ps' → p * ps = p' * ps' := by cc
/-- Congruence lemma for constructing `ex.exp`. -/
lemma exp_congr {p p' : α} {ps ps' : ℕ} : p = p' → ps = ps' → p ^ ps = p' ^ ps' := by cc
/-- Constructs `ex.zero` with the correct arguments. -/
meta def ex_zero : ring_exp_m (ex sum) := do
ctx ← get_context,
pure $ ex.zero ⟨ctx.info_b.zero, ctx.info_b.zero, none⟩
/-- Constructs `ex.sum` with the correct arguments. -/
meta def ex_sum (p : ex prod) (ps : ex sum) : ring_exp_m (ex sum) := do
pps_o ← add_orig p ps,
pps_p ← mk_add [p.pretty, ps.pretty],
pps_pf ← mk_proof_or_refl pps_p ``sum_congr
[p.orig, p.pretty, ps.orig, ps.pretty]
[p.info, ps.info],
pure (ex.sum ⟨pps_o, pps_p, pps_pf⟩ (p.set_info none none) (ps.set_info none none))
/--
Constructs `ex.coeff` with the correct arguments.
There are more efficient constructors for specific numerals:
if `x = 0`, you should use `ex_zero`; if `x = 1`, use `ex_one`.
-/
meta def ex_coeff (x : rat) : ring_exp_m (ex prod) := do
ctx ← get_context,
x_p ← lift $ expr.of_rat ctx.info_b.α x,
pure (ex.coeff ⟨x_p, x_p, none⟩ ⟨x⟩)
/--
Constructs `ex.coeff 1` with the correct arguments.
This is a special case for optimization purposes.
-/
meta def ex_one : ring_exp_m (ex prod) := do
ctx ← get_context,
pure $ ex.coeff ⟨ctx.info_b.one, ctx.info_b.one, none⟩ ⟨1⟩
/-- Constructs `ex.prod` with the correct arguments. -/
meta def ex_prod (p : ex exp) (ps : ex prod) : ring_exp_m (ex prod) := do
pps_o ← mul_orig p ps,
pps_p ← mk_mul [p.pretty, ps.pretty],
pps_pf ← mk_proof_or_refl pps_p ``prod_congr
[p.orig, p.pretty, ps.orig, ps.pretty]
[p.info, ps.info],
pure (ex.prod ⟨pps_o, pps_p, pps_pf⟩ (p.set_info none none) (ps.set_info none none))
/-- Constructs `ex.var` with the correct arguments. -/
meta def ex_var (p : atom) : ring_exp_m (ex base) := pure (ex.var ⟨p.1, p.1, none⟩ p)
/-- Constructs `ex.sum_b` with the correct arguments. -/
meta def ex_sum_b (ps : ex sum) : ring_exp_m (ex base) :=
pure (ex.sum_b ps.info (ps.set_info none none))
/-- Constructs `ex.exp` with the correct arguments. -/
meta def ex_exp (p : ex base) (ps : ex prod) : ring_exp_m (ex exp) := do
ctx ← get_context,
pps_o ← pow_orig p ps,
pps_p ← mk_pow [p.pretty, ps.pretty],
pps_pf ← mk_proof_or_refl pps_p ``exp_congr
[p.orig, p.pretty, ps.orig, ps.pretty]
[p.info, ps.info],
pure (ex.exp ⟨pps_o, pps_p, pps_pf⟩ (p.set_info none none) (ps.set_info none none))
lemma base_to_exp_pf {p p' : α} : p = p' → p = p' ^ 1 := by simp
/-- Conversion from `ex base` to `ex exp`. -/
meta def base_to_exp (p : ex base) : ring_exp_m (ex exp) := do
o ← in_exponent $ ex_one,
ps ← ex_exp p o,
pf ← mk_proof ``base_to_exp_pf [p.orig, p.pretty] [p.info],
pure $ ps.set_info p.orig pf
lemma exp_to_prod_pf {p p' : α} : p = p' → p = p' * 1 := by simp
/-- Conversion from `ex exp` to `ex prod`. -/
meta def exp_to_prod (p : ex exp) : ring_exp_m (ex prod) := do
o ← ex_one,
ps ← ex_prod p o,
pf ← mk_proof ``exp_to_prod_pf [p.orig, p.pretty] [p.info],
pure $ ps.set_info p.orig pf
lemma prod_to_sum_pf {p p' : α} : p = p' → p = p' + 0 := by simp
/-- Conversion from `ex prod` to `ex sum`. -/
meta def prod_to_sum (p : ex prod) : ring_exp_m (ex sum) := do
z ← ex_zero,
ps ← ex_sum p z,
pf ← mk_proof ``prod_to_sum_pf [p.orig, p.pretty] [p.info],
pure $ ps.set_info p.orig pf
lemma atom_to_sum_pf (p : α) : p = p ^ 1 * 1 + 0 := by simp
/--
A more efficient conversion from `atom` to `ex sum`.
The result should be the same as `ex_var p >>= base_to_exp >>= exp_to_prod >>= prod_to_sum`,
except we need to calculate less intermediate steps.
-/
meta def atom_to_sum (p : atom) : ring_exp_m (ex sum) := do
p' ← ex_var p,
o ← in_exponent $ ex_one,
p' ← ex_exp p' o,
o ← ex_one,
p' ← ex_prod p' o,
z ← ex_zero,
p' ← ex_sum p' z,
pf ← mk_proof ``atom_to_sum_pf [p.1] [],
pure $ p'.set_info p.1 pf
/--
Compute the sum of two coefficients.
Note that the result might not be a valid expression:
if `p = -q`, then the result should be `ex.zero : ex sum` instead.
The caller must detect when this happens!
The returned value is of the form `ex.coeff _ (p + q)`,
with the proof of `expr.of_rat p + expr.of_rat q = expr.of_rat (p + q)`.
-/
meta def add_coeff (p_p q_p : expr) (p q : coeff) : ring_exp_m (ex prod) := do
ctx ← get_context,
pq_o ← mk_add [p_p, q_p],
(pq_p, pq_pf) ← lift $ norm_num.eval_field pq_o,
pure $ ex.coeff ⟨pq_o, pq_p, pq_pf⟩ ⟨p.1 + q.1⟩
lemma mul_coeff_pf_one_mul (q : α) : 1 * q = q := one_mul q
lemma mul_coeff_pf_mul_one (p : α) : p * 1 = p := mul_one p
/--
Compute the product of two coefficients.
The returned value is of the form `ex.coeff _ (p * q)`,
with the proof of `expr.of_rat p * expr.of_rat q = expr.of_rat (p * q)`.
-/
meta def mul_coeff (p_p q_p : expr) (p q : coeff) : ring_exp_m (ex prod) :=
match p.1, q.1 with -- Special case to speed up multiplication with 1.
| ⟨1, 1, _, _⟩, _ := do
ctx ← get_context,
pq_o ← mk_mul [p_p, q_p],
pf ← mk_app_csr ``mul_coeff_pf_one_mul [q_p],
pure $ ex.coeff ⟨pq_o, q_p, pf⟩ ⟨q.1⟩
| _, ⟨1, 1, _, _⟩ := do
ctx ← get_context,
pq_o ← mk_mul [p_p, q_p],
pf ← mk_app_csr ``mul_coeff_pf_mul_one [p_p],
pure $ ex.coeff ⟨pq_o, p_p, pf⟩ ⟨p.1⟩
| _, _ := do
ctx ← get_context,
pq' ← mk_mul [p_p, q_p],
(pq_p, pq_pf) ← lift $ norm_num.eval_field pq',
pure $ ex.coeff ⟨pq_p, pq_p, pq_pf⟩ ⟨p.1 * q.1⟩
end
section rewrite
/-! ### `rewrite` section
In this section we deal with rewriting terms to fit in the basic grammar of `eval`.
For example, `nat.succ n` is rewritten to `n + 1` before it is evaluated further.
-/
/-- Given a proof that the expressions `ps_o` and `ps'.orig` are equal,
show that `ps_o` and `ps'.pretty` are equal.
Useful to deal with aliases in `eval`. For instance, `nat.succ p` can be handled
as an alias of `p + 1` as follows:
```
| ps_o@`(nat.succ %%p_o) := do
ps' ← eval `(%%p_o + 1),
pf ← lift $ mk_app ``nat.succ_eq_add_one [p_o],
rewrite ps_o ps' pf
```
-/
meta def rewrite (ps_o : expr) (ps' : ex sum) (pf : expr) : ring_exp_m (ex sum) :=
do
ps'_pf ← ps'.info.proof_term,
pf ← lift $ mk_eq_trans pf ps'_pf,
pure $ ps'.set_info ps_o pf
end rewrite
/--
Represents the way in which two products are equal except coefficient.
This type is used in the function `add_overlap`.
In order to deal with equations of the form `a * 2 + a = 3 * a`,
the `add` function will add up overlapping products,
turning `a * 2 + a` into `a * 3`.
We need to distinguish `a * 2 + a` from `a * 2 + b` in order to do this,
and the `overlap` type carries the information on how it overlaps.
The case `none` corresponds to non-overlapping products, e.g. `a * 2 + b`;
the case `nonzero` to overlapping products adding to non-zero, e.g. `a * 2 + a`
(the `ex prod` field will then look like `a * 3` with a proof that `a * 2 + a = a * 3`);
the case `zero` to overlapping products adding to zero, e.g. `a * 2 + a * -2`.
We distinguish those two cases because in the second, the whole product reduces to `0`.
A potential extension to the tactic would also do this for the base of exponents,
e.g. to show `2^n * 2^n = 4^n`.
-/
meta inductive overlap : Type
| none : overlap
| nonzero : ex prod → overlap
| zero : ex sum → overlap
lemma add_overlap_pf {ps qs pq} (p : α) : ps + qs = pq → p * ps + p * qs = p * pq := λ pq_pf, calc
p * ps + p * qs = p * (ps + qs) : symm (mul_add _ _ _)
... = p * pq : by rw pq_pf
lemma add_overlap_pf_zero {ps qs} (p : α) : ps + qs = 0 → p * ps + p * qs = 0 := λ pq_pf, calc
p * ps + p * qs = p * (ps + qs) : symm (mul_add _ _ _)
... = p * 0 : by rw pq_pf
... = 0 : mul_zero _
/--
Given arguments `ps`, `qs` of the form `ps' * x` and `ps' * y` respectively
return `ps + qs = ps' * (x + y)` (with `x` and `y` arbitrary coefficients).
For other arguments, return `overlap.none`.
-/
meta def add_overlap : ex prod → ex prod → ring_exp_m overlap
| (ex.coeff x_i x) (ex.coeff y_i y) := do
xy@(ex.coeff _ xy_c) ← add_coeff x_i.pretty y_i.pretty x y
| lift $ fail "internal error: add_coeff should return ex.coeff",
if xy_c.1 = 0
then do
z ← ex_zero,
pure $ overlap.zero (z.set_info xy.orig xy.proof)
else pure $ overlap.nonzero xy
| (ex.prod _ _ _) (ex.coeff _ _) := pure overlap.none
| (ex.coeff _ _) (ex.prod _ _ _) := pure overlap.none
| pps@(ex.prod _ p ps) qqs@(ex.prod _ q qs) := if p.eq q
then do
pq_ol ← add_overlap ps qs,
pqs_o ← add_orig pps qqs,
match pq_ol with
| overlap.none := pure overlap.none
| (overlap.nonzero pq) := do
pqs ← ex_prod p pq,
pf ← mk_proof ``add_overlap_pf
[ps.pretty, qs.pretty, pq.pretty, p.pretty]
[pq.info],
pure $ overlap.nonzero (pqs.set_info pqs_o pf)
| (overlap.zero pq) := do
z ← ex_zero,
pf ← mk_proof ``add_overlap_pf_zero
[ps.pretty, qs.pretty, p.pretty]
[pq.info],
pure $ overlap.zero (z.set_info pqs_o pf)
end
else pure overlap.none
section addition
lemma add_pf_z_sum {ps qs qs' : α} : ps = 0 → qs = qs' → ps + qs = qs' := λ ps_pf qs_pf, calc
ps + qs = 0 + qs' : by rw [ps_pf, qs_pf]
... = qs' : zero_add _
lemma add_pf_sum_z {ps ps' qs : α} : ps = ps' → qs = 0 → ps + qs = ps' := λ ps_pf qs_pf, calc
ps + qs = ps' + 0 : by rw [ps_pf, qs_pf]
... = ps' : add_zero _
lemma add_pf_sum_overlap {pps p ps qqs q qs pq pqs : α} :
pps = p + ps → qqs = q + qs → p + q = pq → ps + qs = pqs → pps + qqs = pq + pqs := by cc
lemma add_pf_sum_overlap_zero {pps p ps qqs q qs pqs : α} :
pps = p + ps → qqs = q + qs → p + q = 0 → ps + qs = pqs → pps + qqs = pqs :=
λ pps_pf qqs_pf pq_pf pqs_pf, calc
pps + qqs = (p + ps) + (q + qs) : by rw [pps_pf, qqs_pf]
... = (p + q) + (ps + qs) : by cc
... = 0 + pqs : by rw [pq_pf, pqs_pf]
... = pqs : zero_add _
lemma add_pf_sum_lt {pps p ps qqs pqs : α} :
pps = p + ps → ps + qqs = pqs → pps + qqs = p + pqs := by cc
lemma add_pf_sum_gt {pps qqs q qs pqs : α} :
qqs = q + qs → pps + qs = pqs → pps + qqs = q + pqs := by cc
/--
Add two expressions.
* `0 + qs = 0`
* `ps + 0 = 0`
* `ps * x + ps * y = ps * (x + y)` (for `x`, `y` coefficients; uses `add_overlap`)
* `(p + ps) + (q + qs) = p + (ps + (q + qs))` (if `p.lt q`)
* `(p + ps) + (q + qs) = q + ((p + ps) + qs)` (if not `p.lt q`)
-/
meta def add : ex sum → ex sum → ring_exp_m (ex sum)
| ps@(ex.zero ps_i) qs := do
pf ← mk_proof ``add_pf_z_sum [ps.orig, qs.orig, qs.pretty] [ps.info, qs.info],
pqs_o ← add_orig ps qs,
pure $ qs.set_info pqs_o pf
| ps qs@(ex.zero qs_i) := do
pf ← mk_proof ``add_pf_sum_z [ps.orig, ps.pretty, qs.orig] [ps.info, qs.info],
pqs_o ← add_orig ps qs,
pure $ ps.set_info pqs_o pf
| pps@(ex.sum pps_i p ps) qqs@(ex.sum qqs_i q qs) := do
ol ← add_overlap p q,
ppqqs_o ← add_orig pps qqs,
match ol with
| (overlap.nonzero pq) := do
pqs ← add ps qs,
pqqs ← ex_sum pq pqs,
qqs_pf ← qqs.proof_term,
pf ← mk_proof ``add_pf_sum_overlap
[pps.orig, p.pretty, ps.pretty, qqs.orig, q.pretty, qs.pretty, pq.pretty, pqs.pretty]
[pps.info, qqs.info, pq.info, pqs.info],
pure $ pqqs.set_info ppqqs_o pf
| (overlap.zero pq) := do
pqs ← add ps qs,
pf ← mk_proof ``add_pf_sum_overlap_zero
[pps.orig, p.pretty, ps.pretty, qqs.orig, q.pretty, qs.pretty, pqs.pretty]
[pps.info, qqs.info, pq.info, pqs.info],
pure $ pqs.set_info ppqqs_o pf
| overlap.none := if p.lt q
then do
pqs ← add ps qqs,
ppqs ← ex_sum p pqs,
pf ← mk_proof ``add_pf_sum_lt
[pps.orig, p.pretty, ps.pretty, qqs.orig, pqs.pretty]
[pps.info, pqs.info],
pure $ ppqs.set_info ppqqs_o pf
else do
pqs ← add pps qs,
pqqs ← ex_sum q pqs,
pf ← mk_proof ``add_pf_sum_gt
[pps.orig, qqs.orig, q.pretty, qs.pretty, pqs.pretty]
[qqs.info, pqs.info],
pure $ pqqs.set_info ppqqs_o pf
end
end addition
section multiplication
lemma mul_pf_c_c {ps ps' qs qs' pq : α} :
ps = ps' → qs = qs' → ps' * qs' = pq → ps * qs = pq := by cc
lemma mul_pf_c_prod {ps qqs q qs pqs : α} :
qqs = q * qs → ps * qs = pqs → ps * qqs = q * pqs := by cc
lemma mul_pf_prod_c {pps p ps qs pqs : α} :
pps = p * ps → ps * qs = pqs → pps * qs = p * pqs := by cc
lemma mul_pp_pf_overlap {pps p_b ps qqs qs psqs : α} {p_e q_e : ℕ} :
pps = p_b ^ p_e * ps → qqs = p_b ^ q_e * qs →
p_b ^ (p_e + q_e) * (ps * qs) = psqs → pps * qqs = psqs
:= λ ps_pf qs_pf psqs_pf, by simp [symm psqs_pf, pow_add, ps_pf, qs_pf]; ac_refl
lemma mul_pp_pf_prod_lt {pps p ps qqs pqs : α} :
pps = p * ps → ps * qqs = pqs → pps * qqs = p * pqs := by cc
lemma mul_pp_pf_prod_gt {pps qqs q qs pqs : α} :
qqs = q * qs → pps * qs = pqs → pps * qqs = q * pqs := by cc
/--
Multiply two expressions.
* `x * y = (x * y)` (for `x`, `y` coefficients)
* `x * (q * qs) = q * (qs * x)` (for `x` coefficient)
* `(p * ps) * y = p * (ps * y)` (for `y` coefficient)
* `(p_b^p_e * ps) * (p_b^q_e * qs) = p_b^(p_e + q_e) * (ps * qs)`
(if `p_e` and `q_e` are identical except coefficient)
* `(p * ps) * (q * qs) = p * (ps * (q * qs))` (if `p.lt q`)
* `(p * ps) * (q * qs) = q * ((p * ps) * qs)` (if not `p.lt q`)
-/
meta def mul_pp : ex prod → ex prod → ring_exp_m (ex prod)
| ps@(ex.coeff _ x) qs@(ex.coeff _ y) := do
pq ← mul_coeff ps.pretty qs.pretty x y,
pq_o ← mul_orig ps qs,
pf ← mk_proof_or_refl pq.pretty ``mul_pf_c_c
[ps.orig, ps.pretty, qs.orig, qs.pretty, pq.pretty]
[ps.info, qs.info, pq.info],
pure $ pq.set_info pq_o pf
| ps@(ex.coeff _ x) qqs@(ex.prod _ q qs) := do
pqs ← mul_pp ps qs,
pqqs ← ex_prod q pqs,
pqqs_o ← mul_orig ps qqs,
pf ← mk_proof ``mul_pf_c_prod
[ps.orig, qqs.orig, q.pretty, qs.pretty, pqs.pretty]
[qqs.info, pqs.info],
pure $ pqqs.set_info pqqs_o pf
| pps@(ex.prod _ p ps) qs@(ex.coeff _ y) := do
pqs ← mul_pp ps qs,
ppqs ← ex_prod p pqs,
ppqs_o ← mul_orig pps qs,
pf ← mk_proof ``mul_pf_prod_c
[pps.orig, p.pretty, ps.pretty, qs.orig, pqs.pretty]
[pps.info, pqs.info],
pure $ ppqs.set_info ppqs_o pf
| pps@(ex.prod _ p@(ex.exp _ p_b p_e) ps) qqs@(ex.prod _ q@(ex.exp _ q_b q_e) qs) := do
ppqqs_o ← mul_orig pps qqs,
pq_ol ← in_exponent $ add_overlap p_e q_e,
match pq_ol, p_b.eq q_b with
| (overlap.nonzero pq_e), tt := do
psqs ← mul_pp ps qs,
pq ← ex_exp p_b pq_e,
ppsqqs ← ex_prod pq psqs,
pf ← mk_proof ``mul_pp_pf_overlap
[pps.orig, p_b.pretty, ps.pretty, qqs.orig, qs.pretty, ppsqqs.pretty, p_e.pretty, q_e.pretty]
[pps.info, qqs.info, ppsqqs.info],
pure $ ppsqqs.set_info ppqqs_o pf
| _, _ := if p.lt q then do
pqs ← mul_pp ps qqs,
ppqs ← ex_prod p pqs,
pf ← mk_proof ``mul_pp_pf_prod_lt
[pps.orig, p.pretty, ps.pretty, qqs.orig, pqs.pretty]
[pps.info, pqs.info],
pure $ ppqs.set_info ppqqs_o pf
else do
pqs ← mul_pp pps qs,
pqqs ← ex_prod q pqs,
pf ← mk_proof ``mul_pp_pf_prod_gt
[pps.orig, qqs.orig, q.pretty, qs.pretty, pqs.pretty]
[qqs.info, pqs.info],
pure $ pqqs.set_info ppqqs_o pf
end
lemma mul_p_pf_zero {ps qs : α} : ps = 0 → ps * qs = 0 :=
λ ps_pf, by rw [ps_pf, zero_mul]
lemma mul_p_pf_sum {pps p ps qs ppsqs : α} : pps = p + ps →
p * qs + ps * qs = ppsqs → pps * qs = ppsqs := λ pps_pf ppsqs_pf, calc
pps * qs = (p + ps) * qs : by rw [pps_pf]
... = p * qs + ps * qs : add_mul _ _ _
... = ppsqs : ppsqs_pf
/--
Multiply two expressions.
* `0 * qs = 0`
* `(p + ps) * qs = (p * qs) + (ps * qs)`
-/
meta def mul_p : ex sum → ex prod → ring_exp_m (ex sum)
| ps@(ex.zero ps_i) qs := do
z ← ex_zero,
z_o ← mul_orig ps qs,
pf ← mk_proof ``mul_p_pf_zero [ps.orig, qs.orig] [ps.info],
pure $ z.set_info z_o pf
| pps@(ex.sum pps_i p ps) qs := do
pqs ← mul_pp p qs >>= prod_to_sum,
psqs ← mul_p ps qs,
ppsqs ← add pqs psqs,
pps_pf ← pps.proof_term,
ppsqs_o ← mul_orig pps qs,
ppsqs_pf ← ppsqs.proof_term,
pf ← mk_proof ``mul_p_pf_sum
[pps.orig, p.pretty, ps.pretty, qs.orig, ppsqs.pretty]
[pps.info, ppsqs.info],
pure $ ppsqs.set_info ppsqs_o pf
lemma mul_pf_zero {ps qs : α} : qs = 0 → ps * qs = 0 :=
λ qs_pf, by rw [qs_pf, mul_zero]
lemma mul_pf_sum {ps qqs q qs psqqs : α} : qqs = q + qs → ps * q + ps * qs = psqqs →
ps * qqs = psqqs := λ qs_pf psqqs_pf, calc
ps * qqs = ps * (q + qs) : by rw [qs_pf]
... = ps * q + ps * qs : mul_add _ _ _
... = psqqs : psqqs_pf
/--
Multiply two expressions.
* `ps * 0 = 0`
* `ps * (q + qs) = (ps * q) + (ps * qs)`
-/
meta def mul : ex sum → ex sum → ring_exp_m (ex sum)
| ps qs@(ex.zero qs_i) := do
z ← ex_zero,
z_o ← mul_orig ps qs,
pf ← mk_proof ``mul_pf_zero [ps.orig, qs.orig] [qs.info],
pure $ z.set_info z_o pf
| ps qqs@(ex.sum qqs_i q qs) := do
psq ← mul_p ps q,
psqs ← mul ps qs,
psqqs ← add psq psqs,
psqqs_o ← mul_orig ps qqs,
pf ← mk_proof ``mul_pf_sum
[ps.orig, qqs.orig, q.orig, qs.orig, psqqs.pretty]
[qqs.info, psqqs.info],
pure $ psqqs.set_info psqqs_o pf
end multiplication
section exponentiation
lemma pow_e_pf_exp {pps p : α} {ps qs psqs : ℕ} :
pps = p ^ ps → ps * qs = psqs → pps ^ qs = p ^ psqs :=
λ pps_pf psqs_pf, calc
pps ^ qs = (p ^ ps) ^ qs : by rw [pps_pf]
... = p ^ (ps * qs) : symm (pow_mul _ _ _)
... = p ^ psqs : by rw [psqs_pf]
/--
Compute the exponentiation of two coefficients.
The returned value is of the form `ex.coeff _ (p ^ q)`,
with the proof of `expr.of_rat p ^ expr.of_rat q = expr.of_rat (p ^ q)`.