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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
/* */
/* This file is part of the program and library */
/* SCIP --- Solving Constraint Integer Programs */
/* */
/* Copyright 2002-2022 Zuse Institute Berlin */
/* */
/* Licensed under the Apache License, Version 2.0 (the "License"); */
/* you may not use this file except in compliance with the License. */
/* You may obtain a copy of the License at */
/* */
/* http://www.apache.org/licenses/LICENSE-2.0 */
/* */
/* Unless required by applicable law or agreed to in writing, software */
/* distributed under the License is distributed on an "AS IS" BASIS, */
/* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. */
/* See the License for the specific language governing permissions and */
/* limitations under the License. */
/* */
/* You should have received a copy of the Apache-2.0 license */
/* along with SCIP; see the file LICENSE. If not visit scipopt.org. */
/* */
/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
/**@file scip_expr.h
* @ingroup PUBLICCOREAPI
* @brief public functions to work with algebraic expressions
* @author Ksenia Bestuzheva
* @author Benjamin Mueller
* @author Felipe Serrano
* @author Stefan Vigerske
*/
extern "C" NDEBUG
/* If NDEBUG is defined, the function calls are overwritten by defines to reduce the number of function calls and
* speed up the algorithms.
*/
/** @} */
/**@addtogroup PublicExprMethods
* @{
*/
/**@name Expressions */
/**@{ */
/** creates and captures an expression with given expression data and children */
SCIP_RETCODE ;
/** creates and captures an expression with given expression data and up to two children */
SCIP_RETCODE ;
/** creates and captures an expression representing a quadratic function */
SCIP_RETCODE ;
/** creates and captures an expression representing a monomial
*
* @note In deviation from the actual definition of monomials, we also allow for negative and rational exponents.
* So this function actually creates an expression for a signomial that has exactly one term.
*/
SCIP_RETCODE ;
/** appends child to the children list of expr
*
* @attention Only use if you really know what you are doing. The expression handler of the expression needs to be able to handle an increase in the number of children.
*/
SCIP_RETCODE ;
/** overwrites/replaces a child of an expressions
*
* The old child is released and the newchild is captured, unless they are the same (=same pointer).
*/
SCIP_RETCODE ;
/** remove all children of expr
*
* @attention Only use if you really know what you are doing. The expression handler of the expression needs to be able to handle the removal of all children.
*/
SCIP_RETCODE ;
/** duplicates the given expression and its children */
SCIP_RETCODE ;
/** duplicates the given expression, but reuses its children */
SCIP_RETCODE ;
/** copies an expression including children to use in a (possibly different) SCIP instance */
SCIP_RETCODE ;
/** creates an expression from a string
*
* We specify the grammar that defines the syntax of an expression.
* Loosely speaking, a `Base` will be any "block", a `Factor` is a `Base` to a power,
* a `Term` is a product of `Factors` and an `Expression` is a sum of `Terms`.
*
* The actual definition:
* <pre>
* Expression -> ["+" | "-"] Term { ("+" | "-" | "number *") ] Term }
* Term -> Factor { ("*" | "/" ) Factor }
* Factor -> Base [ "^" "number" | "^(" "number" ")" ]
* Base -> "number" | "<varname>" | "(" Expression ")" | Op "(" OpExpression ")
* </pre>
* where `[a|b]` means `a` or `b` or none, `(a|b)` means `a` or `b`, `{a}` means 0 or more `a`.
*
* Note that `Op` and `OpExpression` are undefined.
* `Op` corresponds to the name of an expression handler and `OpExpression` to whatever string the expression handler accepts (through its parse method).
*/
SCIP_RETCODE ;
/** captures an expression (increments usage count) */
void ;
/** releases an expression (decrements usage count and possibly frees expression) */
SCIP_RETCODE ;
/** returns whether an expression is a variable expression */
SCIP_Bool ;
/** returns whether an expression is a value expression */
SCIP_Bool ;
/** returns whether an expression is a sum expression */
SCIP_Bool ;
/** returns whether an expression is a product expression */
SCIP_Bool ;
/** returns whether an expression is a power expression */
SCIP_Bool ;
/** print an expression as info-message */
SCIP_RETCODE ;
/** initializes printing of expressions in dot format to a give FILE* pointer */
SCIP_RETCODE ;
/** initializes printing of expressions in dot format to a file with given filename */
SCIP_RETCODE ;
/** main part of printing an expression in dot format */
SCIP_RETCODE ;
/** finishes printing of expressions in dot format */
SCIP_RETCODE ;
/** shows a single expression by use of dot and gv
*
* This function is meant for debugging purposes.
* It's signature is kept as simple as possible to make it
* easily callable from gdb, for example.
*
* It prints the expression into a temporary file in dot format, then calls dot to create a postscript file, then calls ghostview (gv) to show the file.
* SCIP will hold until ghostscript is closed.
*/
SCIP_RETCODE ;
/** prints structure of an expression a la Maple's dismantle */
SCIP_RETCODE ;
/** evaluate an expression in a point
*
* Iterates over expressions to also evaluate children, if necessary.
* Value can be received via SCIPexprGetEvalValue().
* If an evaluation error (division by zero, ...) occurs, this value will
* be set to SCIP_INVALID.
*
* If a nonzero \p soltag is passed, then only (sub)expressions are
* reevaluated that have a different solution tag. If a soltag of 0
* is passed, then subexpressions are always reevaluated.
* The tag is stored together with the value and can be received via
* SCIPexprGetEvalTag().
*/
SCIP_RETCODE ;
/** returns a previously unused solution tag for expression evaluation */
SCIP_Longint ;
/**@} */
/** @name Differentiation
* @anchor SCIP_EXPR_DIFF
*
* @par Gradients (Automatic differentiation Backward mode)
*
* Given a function, say, \f$f(s(x,y),t(x,y))\f$ there is a common mnemonic technique to compute its partial derivatives, using a tree diagram.
* Suppose we want to compute the partial derivative of \f$f\f$ w.r.t. \f$x\f$.
* Write the function as a tree:
*
* f
* |-----|
* s t
* |--| |--|
* x y x y
*
* The weight of an edge between two nodes represents the partial derivative of the parent w.r.t. the children, e.g.,
*
* f
* |
* s
*
* is \f$ \partial_sf \f$.
* The weight of a path is the product of the weight of the edges in the path.
* The partial derivative of \f$f\f$ w.r.t. \f$x\f$ is then the sum of the weights of all paths connecting \f$f\f$ with \f$x\f$:
* \f[ \frac{\partial f}{\partial x} = \partial_s f \cdot \partial_x s + \partial_t f \cdot \partial_x t. \f]
*
* We follow this method in order to compute the gradient of an expression (root) at a given point (point).
* Note that an expression is a DAG representation of a function, but there is a 1-1 correspondence between paths
* in the DAG and path in a tree diagram of a function.
* Initially, we set `root->derivative` to 1.0.
* Then, traversing the tree in Depth First (see \ref SCIPexpriterInit), for every expr that *has* children,
* we store in its i-th child, `child[i]->derivative`, the derivative of expr w.r.t. child evaluated at point multiplied with `expr->derivative`.
*
* For example:
* 1. `f->derivative` = 1.0
* 2. `s->derivative` = \f$\partial_s f \,\cdot\f$ `f->derivative` = \f$\partial_s f\f$
* 3. `x->derivative` = \f$\partial_x s \,\cdot\f$ `s->derivative` = \f$\partial_x s \cdot \partial_s f\f$
*
* However, when the child is a variable expressions, we actually need to initialize `child->derivative` to 0.0
* and afterwards add, instead of overwrite the computed value.
* The complete example would then be:
*
* 1. `f->derivative` = 1.0, `x->derivative` = 0.0, `y->derivative` = 0.0
* 2. `s->derivative` = \f$\partial_s f \,\cdot\f$ `f->derivative` = \f$\partial_s f\f$
* 3. `x->derivative` += \f$\partial_x s \,\cdot\f$ `s->derivative` = \f$\partial_x s \cdot \partial_s f\f$
* 4. `y->derivative` += \f$\partial_y s \,\cdot\f$ `s->derivative` = \f$\partial_y s \cdot \partial_s f\f$
* 5. `t->derivative` = \f$\partial_t f \,\cdot\f$ `f->derivative` = \f$\partial_t f\f$
* 6. `x->derivative` += \f$\partial_x t \,\cdot\f$ `t->derivative` = \f$\partial_x t \cdot \partial_t f\f$
* 7. `y->derivative` += \f$\partial_y t \,\cdot\f$ `t->derivative` = \f$\partial_y t \cdot \partial_t f\f$
*
* Note that, to compute this, we only need to know, for each expression, its partial derivatives w.r.t a given child at a point.
* This is what the callback `SCIP_DECL_EXPRBWDIFF` should return.
* Indeed, from "derivative of expr w.r.t. child evaluated at point multiplied with expr->derivative",
* note that at the moment of processing a child, we already know `expr->derivative`, so the only
* missing piece of information is "the derivative of expr w.r.t. child evaluated at point".
*
* An equivalent way of interpreting the procedure is that `expr->derivative` stores the derivative of the root w.r.t. expr.
* This way, `x->derivative` and `y->derivative` will contain the partial derivatives of root w.r.t. the variable, that is, the gradient.
* Note, however, that this analogy is only correct for leave expressions, since the derivative value of an intermediate expression gets overwritten.
*
*
* \par Hessian (Automatic differentiation Backward on Forward mode)
*
* Computing the Hessian is more complicated since it is the derivative of the gradient, which is a function with more than one output.
* We compute the Hessian by computing "directions" of the Hessian, that is \f$H\cdot u\f$ for different \f$u\f$.
* This is easy in general, since it is the gradient of the *scalar* function \f$\nabla f u\f$, that is,
* the directional derivative of \f$f\f$ in the direction \f$u\f$: \f$D_u f\f$.
*
* This is easily computed via the so called forward mode.
* Just as `expr->derivative` stores the partial derivative of the root w.r.t. expr,
* `expr->dot` stores the directional derivative of expr in the direction \f$u\f$.
* Then, by the chain rule, `expr->dot` = \f$\sum_{c:\text{children}} \partial_c \text{expr} \,\cdot\f$ `c->dot`.
*
* Starting with `x[i]->dot` = \f$u_i\f$, we can compute `expr->dot` for every expression at the same time we evaluate expr.
* Computing `expr->dot` is the purpose of the callback `SCIP_DECL_EXPRFWDIFF`.
* Obviously, when this callback is called, the "dots" of all children are known
* (just like evaluation, where the value of all children are known).
*
* Once we have this information, we compute the gradient of this function, following the same idea as before.
* We define `expr->bardot` to be the directional derivative in direction \f$u\f$ of the partial derivative of the root w.r.t `expr`,
* that is \f$D_u (\partial_{\text{expr}} f) = D_u\f$ (`expr->derivative`).
*
* This way, `x[i]->bardot` = \f$D_u (\partial_{x_i} f) = e_i^T H_f u\f$.
* Hence `vars->bardot` contain \f$H_f u\f$.
* By the chain rule, product rule, and definition we have
* \f{eqnarray*}{
* \texttt{expr->bardot} & = & D_u (\partial_{\text{expr}} f) \\
* & = & D_u ( \partial_{\text{parent}} f \cdot \partial_{\text{expr}} \text{parent} ) \\
* & = & D_u ( \texttt{parent->derivative} \cdot \partial_{\text{expr}} \text{parent} ) \\
* & = & \partial_{\text{expr}} \text{parent} \cdot D_u (\texttt{parent->derivative}) + \texttt{parent->derivative} \cdot D_u (\partial_{\text{expr}} \text{parent}) \\
* & = & \texttt{parent->bardot} \cdot \partial_{\text{expr}} \text{parent} + \texttt{parent->derivative} \cdot D_u (\partial_{\text{expr}} \text{parent})
* \f}
*
* Note that we have computed `parent->bardot` and `parent->derivative` at this point,
* while \f$\partial_{\text{expr}} \text{parent}\f$ is the return of `SCIP_DECL_EXPRBWDIFF`.
* Hence the only information we need to compute is \f$D_u (\partial_{\text{expr}} \text{parent})\f$.
* This is the purpose of the callback `SCIP_DECL_EXPRBWFWDIFF`.
*
* @{
*/
/** evaluates gradient of an expression for a given point
*
* Initiates an expression walk to also evaluate children, if necessary.
* Value can be received via SCIPgetExprPartialDiffNonlinear().
* If an error (division by zero, ...) occurs, this value will
* be set to SCIP_INVALID.
*/
SCIP_RETCODE ;
/** evaluates Hessian-vector product of an expression for a given point and direction
*
* Evaluates children, if necessary.
* Value can be received via SCIPgetExprPartialDiffGradientDirNonlinear().
* If an error (division by zero, ...) occurs, this value will
* be set to SCIP_INVALID.
*/
SCIP_RETCODE ;
/**@} */ /* end of differentiation methods */
/**@name Expressions
* @{
*/
/** possibly reevaluates and then returns the activity of the expression
*
* Reevaluate activity if currently stored is no longer uptodate (some bound was changed since last evaluation).
*
* The owner of the expression may overwrite the methods used to evaluate the activity,
* including whether the local or global domain of variables is used.
* By default (no owner, or owner doesn't overwrite activity evaluation),
* the local domain of variables is used.
*
* @note If expression is set to be integral, then activities are tightened to integral values.
* Thus, ensure that the integrality information is valid (if set to TRUE; the default (FALSE) is always ok).
*/
SCIP_RETCODE ;
/** compare expressions
* @return -1, 0 or 1 if expr1 <, =, > expr2, respectively
* @note The given expressions are assumed to be simplified.
*/
int ;
/** compute the hash value of an expression */
SCIP_RETCODE ;
/** simplifies an expression
*
* This is largely inspired by Joel Cohen's
* *Computer algebra and symbolic computation: Mathematical methods*,
* in particular Chapter 3.
* The other fountain of inspiration are the simplifying methods of expr.c in SCIP 7.
*
* Note: The things to keep in mind when adding simplification rules are the following.
* I will be using the product expressions (see expr_product.c) as an example.
* There are mainly 3 parts of the simplification process. You need to decide
* at which stage the simplification rule makes sense.
* 1. Simplify each factor (simplifyFactor()): At this stage we got the children of the product expression.
* At this point, each child is simplified when viewed as a stand-alone expression, but not necessarily when viewed as child of a product expression.
* Rules like SP2, SP7, etc are enforced at this point.
* 2. Multiply the factors (mergeProductExprlist()): At this point rules like SP4, SP5 and SP14 are enforced.
* 3. Build the actual simplified product expression (buildSimplifiedProduct()):
* At this point rules like SP10, SP11, etc are enforced.
*
* During steps 1 and 2 do not forget to set the flag `changed` to TRUE when something actually changes.
*
* \par Definition of simplified expressions
*
* An expression is simplified if it
* - is a value expression
* - is a var expression
* - is a product expression such that
* - SP1: every child is simplified
* - SP2: no child is a product
* - SP4: no two children are the same expression (those should be multiplied)
* - SP5: the children are sorted [commutative rule]
* - SP7: no child is a value
* - SP8: its coefficient is 1.0 (otherwise should be written as sum)
* - SP10: it has at least two children
* - TODO?: at most one child is an `abs`
* - SP11: no two children are `expr*log(expr)`
* (TODO: we could handle more complicated stuff like \f$xy\log(x) \to - y * \mathrm{entropy}(x)\f$, but I am not sure this should happen at the simplification level;
* similar for \f$(xy) \log(xy)\f$, which currently simplifies to \f$xy \log(xy)\f$)
* - SP12: if it has two children, then neither of them is a sum (expand sums)
* - SP13: no child is a sum with a single term
* - SP14: at most one child is an `exp`
* - is a power expression such that
* - POW1: exponent is not 0
* - POW2: exponent is not 1
* - POW3: its child is not a value
* - POW4: its child is simplified
* - POW5: if exponent is integer, its child is not a product
* - POW6: if exponent is integer, its child is not a sum with a single term (\f$(2x)^2 \to 4x^2\f$)
* - POW7: if exponent is 2, its child is not a sum (expand sums)
* - POW8: its child is not a power unless \f$(x^n)^m\f$ with \f$nm\f$ being integer and \f$n\f$ or \f$m\f$ fractional and \f$n\f$ not being even integer
* - POW9: its child is not a sum with a single term with a positive coefficient: \f$(25x)^{0.5} \to 5 x^{0.5}\f$
* - POW10: its child is not a binary variable: \f$b^e, e > 0 \to b\f$; \f$b^e, e < 0 \to b := 1\f$
* - POW11: its child is not an exponential: \f$\exp(\text{expr})^e \to \exp(e\cdot\text{expr})\f$
* - is a signedpower expression such that
* - SPOW1: exponent is not 0
* - SPOW2: exponent is not 1
* - SPOW3: its child is not a value
* - SPOW4: its child is simplified
* - SPOW5: (TODO) do we want to distribute signpowers over products like we do for powers?
* - SPOW6: exponent is not an odd integer: (signpow odd expr) -> (pow odd expr)
* - SPOW8: if exponent is integer, its child is not a power
* - SPOW9: its child is not a sum with a single term: \f$\mathrm{signpow}(25x,0.5) \to 5\mathrm{signpow}(x,0.5)\f$
* - SPOW10: its child is not a binary variable: \f$\mathrm{signpow}(b,e), e > 0 \to b\f$; \f$\mathrm{signpow}(b,e), e < 0 \to b := 1\f$
* - SPOW11: its child is not an exponential: \f$\mathrm{signpow}(\exp(\text{expr}),e) \to \exp(e\cdot\text{expr})\f$
* - TODO: what happens when child is another signed power?
* - TODO: if child ≥ 0 -> transform to normal power; if child < 0 -> transform to - normal power
*
* TODO: Some of these criteria are too restrictive for signed powers; for example, the exponent does not need to be
* an integer for signedpower to distribute over a product (SPOW5, SPOW6, SPOW8). Others can also be improved.
* - is a sum expression such that
* - SS1: every child is simplified
* - SS2: no child is a sum
* - SS3: no child is a value (values should go in the constant of the sum)
* - SS4: no two children are the same expression (those should be summed up)
* - SS5: the children are sorted [commutative rule]
* - SS6: it has at least one child
* - SS7: if it consists of a single child, then either constant is != 0.0 or coef != 1
* - SS8: no child has coefficient 0
* - SS9: if a child c is a product that has an exponential expression as one of its factors, then the coefficient of c is +/-1.0
* - SS10: if a child c is an exponential, then the coefficient of c is +/-1.0
* - it is a function with simplified arguments, but not all of them can be values
* - TODO? a logarithm doesn't have a product as a child
* - TODO? the exponent of an exponential is always 1
*
* \par Ordering Rules (see SCIPexprCompare())
* \anchor EXPR_ORDER
* These rules define a total order on *simplified* expressions.
* There are two groups of rules, when comparing equal type expressions and different type expressions.
*
* Equal type expressions:
* - OR1: u,v value expressions: u < v ⇔ val(u) < val(v)
* - OR2: u,v var expressions: u < v ⇔ `SCIPvarGetIndex(var(u))` < `SCIPvarGetIndex(var(v))`
* - OR3: u,v are both sum or product expression: < is a lexicographical order on the terms
* - OR4: u,v are both pow: u < v ⇔ base(u) < base(v) or, base(u) = base(v) and expo(u) < expo(v)
* - OR5: u,v are \f$u = f(u_1, ..., u_n), v = f(v_1, ..., v_m)\f$: u < v ⇔ For the first k such that \f$u_k \neq v_k\f$, \f$u_k < v_k\f$, or if such a \f$k\f$ doesn't exist, then \f$n < m\f$.
*
* Different type expressions:
* - OR6: u value, v other: u < v always
* - OR7: u sum, v var or func: u < v ⇔ u < 0+v;
* In other words, if \f$u = \sum_{i=1}^n \alpha_i u_i\f$, then u < v ⇔ \f$u_n\f$ < v or if \f$u_n\f$ = v and \f$\alpha_n\f$ < 1.
* - OR8: u product, v pow, sum, var or func: u < v ⇔ u < 1*v;
* In other words, if \f$u = \prod_{i=1}^n u_i\f$, then u < v ⇔ \f$u_n\f$ < v.
* Note: since this applies only to simplified expressions, the form of the product is correct.
* Simplified products do *not* have constant coefficients.
* - OR9: u pow, v sum, var or func: u < v ⇔ u < v^1
* - OR10: u var, v func: u < v always
* - OR11: u func, v other type of func: u < v ⇔ name(type(u)) < name(type(v))
* - OR12: none of the rules apply: u < v ⇔ ! v < u
*
* Examples:
* - x < x^2 ?: x is var and x^2 power, so none applies (OR12).
* Hence, we try to answer x^2 < x ?: x^2 < x ⇔ x < x or if x = x and 2 < 1 ⇔ 2 < 1 ⇔ False. So x < x^2 is True.
* - x < x^-1 --OR12→ ~(x^-1 < x) --OR9→ ~(x^-1 < x^1) --OR4→ ~(x < x or -1 < 1) → ~True → False
* - x*y < x --OR8→ x*y < 1*x --OR3→ y < x --OR2→ False
* - x*y < y --OR8→ x*y < 1*y --OR3→ y < x --OR2→ False
*
* \par Algorithm
*
* The recursive version of the algorithm is
*
* EXPR simplify(expr)
* for c in 1..expr->nchildren
* expr->children[c] = simplify(expr->children[c])
* end
* return expr->exprhdlr->simplify(expr)
* end
*
* Important: Whatever is returned by a simplify callback **has** to be simplified.
* Also, all children of the given expression **are** already simplified.
*/
SCIP_RETCODE ;
/** replaces common sub-expressions in a given expression graph by using a hash key for each expression
*
* The algorithm consists of two steps:
*
* 1. traverse through all given expressions and compute for each of them a (not necessarily unique) hash
*
* 2. initialize an empty hash table and traverse through all expression; check for each of them if we can find a
* structural equivalent expression in the hash table; if yes we replace the expression by the expression inside the
* hash table, otherwise we add it to the hash table
*
* @note the hash keys of the expressions are used for the hashing inside the hash table; to compute if two expressions
* (with the same hash) are structurally the same we use the function SCIPexprCompare().
*/
SCIP_RETCODE ;
/** computes the curvature of a given expression and all its subexpressions
*
* @note this function also evaluates all subexpressions w.r.t. current variable bounds
* @note this function relies on information from the curvature callback of expression handlers only,
* consider using function @ref SCIPhasExprCurvature() of the convex-nlhdlr instead, as that uses more information to deduce convexity
*/
SCIP_RETCODE ;
/** computes integrality information of a given expression and all its subexpressions
*
* The integrality information can be accessed via SCIPexprIsIntegral().
*/
SCIP_RETCODE ;
/** returns the total number of variable expressions in an expression
*
* The function counts variable expressions in common sub-expressions only once, but
* counts variables appearing in several variable expressions multiple times.
*/
SCIP_RETCODE ;
/** returns all variable expressions contained in a given expression
*
* The array to store all variable expressions needs to be at least of size
* the number of unique variable expressions in the expression which is given by SCIPgetExprNVars().
*
* If every variable is represented by only one variable expression (common subexpression have been removed)
* then SCIPgetExprNVars() can be bounded by SCIPgetNTotalVars().
* If, in addition, non-active variables have been removed from the expression, e.g., by simplifying,
* then SCIPgetExprNVars() can be bounded by SCIPgetNVars().
*
* @note function captures variable expressions
*/
SCIP_RETCODE ;
/** @} */
/**@name Expression Handler Callbacks
* @{
*/
/** calls the print callback for an expression
*
* @see SCIP_DECL_EXPRPRINT
*/
;
/** calls the curvature callback for an expression
*
* @see SCIP_DECL_EXPRCURVATURE
*
* Returns unknown curvature if callback not implemented.
*/
;
/** calls the monotonicity callback for an expression
*
* @see SCIP_DECL_EXPRMONOTONICITY
*
* Returns unknown monotonicity if callback not implemented.
*/
;
/** calls the eval callback for an expression with given values for children
*
* Does not iterates over expressions, but requires values for children to be given.
* Value is not stored in expression, but returned in `val`.
* If an evaluation error (division by zero, ...) occurs, this value will
* be set to `SCIP_INVALID`.
*/
SCIP_RETCODE ;
/** calls the eval and fwdiff callback of an expression with given values for children
*
* Does not iterates over expressions, but requires values for children and direction to be given.
*
* Value is not stored in expression, but returned in `val`.
* If an evaluation error (division by zero, ...) occurs, this value will be set to `SCIP_INVALID`.
*
* Direction is not stored in expression, but returned in `dot`.
* If an differentiation error (division by zero, ...) occurs, this value will be set to `SCIP_INVALID`.
*/
SCIP_RETCODE ;
/** calls the interval evaluation callback for an expression
*
* @see SCIP_DECL_EXPRINTEVAL
*
* Returns entire interval if callback not implemented.
*/
;
/** calls the estimate callback for an expression
*
* @see SCIP_DECL_EXPRESTIMATE
*
* Returns without success if callback not implemented.
*/
;
/** calls the initial estimators callback for an expression
*
* @see SCIP_DECL_EXPRINITESTIMATES
*
* Returns no estimators if callback not implemented.
*/
;
/** calls the simplify callback for an expression
*
* @see SCIP_DECL_EXPRSIMPLIFY
*
* Returns unmodified expression if simplify callback not implemented.
*
* Does not simplify descendants (children, etc). Use SCIPsimplifyExpr() for that.
*/
;
/** calls the reverse propagation callback for an expression
*
* @see SCIP_DECL_EXPRREVERSEPROP
*
* Returns unmodified `childrenbounds` if reverseprop callback not implemented.
*/
;
/** @} */
/**@name Expression Iterator */
/**@{ */
/** creates an expression iterator */
SCIP_RETCODE ;
/** frees an expression iterator */
void ;
/** @} */
/**@name Quadratic Expressions */
/**@{ */
/** checks whether an expression is quadratic
*
* An expression is quadratic if it is either a square (of some expression), a product (of two expressions),
* or a sum of terms where at least one is a square or a product.
*
* Use SCIPexprGetQuadraticData() to get data about the representation as quadratic.
*/
SCIP_RETCODE ;
/** frees information on quadratic representation of an expression
*
* Before doing changes to an expression, it can be useful to call this function.
*/
void ;
/** evaluates quadratic term in a solution
*
* \note This requires that every expressiion used in the quadratic data is a variable expression.
*/
SCIP_Real ;
/** prints quadratic expression */
SCIP_RETCODE ;
/** checks the curvature of the quadratic expression
*
* For this, it builds the matrix Q of quadratic coefficients and computes its eigenvalues using LAPACK.
* If Q is
* - semidefinite positive -> curv is set to convex,
* - semidefinite negative -> curv is set to concave,
* - otherwise -> curv is set to unknown.
*
* If `assumevarfixed` is given and some expressions in quadratic terms correspond to variables present in
* this hashmap, then the corresponding rows and columns are ignored in the matrix Q.
*/
SCIP_RETCODE ;
/** @} */
/** @} */
}
/* SCIP_SCIP_EXPR_H_ */