3 * Implementation of symbolic matrices */
6 * GiNaC Copyright (C) 1999-2001 Johannes Gutenberg University Mainz, Germany
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
37 #ifndef NO_NAMESPACE_GINAC
39 #endif // ndef NO_NAMESPACE_GINAC
41 GINAC_IMPLEMENT_REGISTERED_CLASS(matrix, basic)
44 // default constructor, destructor, copy constructor, assignment operator
50 /** Default ctor. Initializes to 1 x 1-dimensional zero-matrix. */
51 matrix::matrix() : inherited(TINFO_matrix), row(1), col(1)
53 debugmsg("matrix default constructor",LOGLEVEL_CONSTRUCT);
59 void matrix::copy(const matrix & other)
61 inherited::copy(other);
64 m = other.m; // STL's vector copying invoked here
67 void matrix::destroy(bool call_parent)
69 if (call_parent) inherited::destroy(call_parent);
78 /** Very common ctor. Initializes to r x c-dimensional zero-matrix.
80 * @param r number of rows
81 * @param c number of cols */
82 matrix::matrix(unsigned r, unsigned c)
83 : inherited(TINFO_matrix), row(r), col(c)
85 debugmsg("matrix constructor from unsigned,unsigned",LOGLEVEL_CONSTRUCT);
86 m.resize(r*c, _ex0());
91 /** Ctor from representation, for internal use only. */
92 matrix::matrix(unsigned r, unsigned c, const exvector & m2)
93 : inherited(TINFO_matrix), row(r), col(c), m(m2)
95 debugmsg("matrix constructor from unsigned,unsigned,exvector",LOGLEVEL_CONSTRUCT);
102 /** Construct object from archive_node. */
103 matrix::matrix(const archive_node &n, const lst &sym_lst) : inherited(n, sym_lst)
105 debugmsg("matrix constructor from archive_node", LOGLEVEL_CONSTRUCT);
106 if (!(n.find_unsigned("row", row)) || !(n.find_unsigned("col", col)))
107 throw (std::runtime_error("unknown matrix dimensions in archive"));
108 m.reserve(row * col);
109 for (unsigned int i=0; true; i++) {
111 if (n.find_ex("m", e, sym_lst, i))
118 /** Unarchive the object. */
119 ex matrix::unarchive(const archive_node &n, const lst &sym_lst)
121 return (new matrix(n, sym_lst))->setflag(status_flags::dynallocated);
124 /** Archive the object. */
125 void matrix::archive(archive_node &n) const
127 inherited::archive(n);
128 n.add_unsigned("row", row);
129 n.add_unsigned("col", col);
130 exvector::const_iterator i = m.begin(), iend = m.end();
138 // functions overriding virtual functions from bases classes
143 basic * matrix::duplicate() const
145 debugmsg("matrix duplicate",LOGLEVEL_DUPLICATE);
146 return new matrix(*this);
149 void matrix::print(std::ostream & os, unsigned upper_precedence) const
151 debugmsg("matrix print",LOGLEVEL_PRINT);
153 for (unsigned r=0; r<row-1; ++r) {
155 for (unsigned c=0; c<col-1; ++c)
156 os << m[r*col+c] << ",";
157 os << m[col*(r+1)-1] << "]], ";
160 for (unsigned c=0; c<col-1; ++c)
161 os << m[(row-1)*col+c] << ",";
162 os << m[row*col-1] << "]] ]]";
165 void matrix::printraw(std::ostream & os) const
167 debugmsg("matrix printraw",LOGLEVEL_PRINT);
168 os << "matrix(" << row << "," << col <<",";
169 for (unsigned r=0; r<row-1; ++r) {
171 for (unsigned c=0; c<col-1; ++c)
172 os << m[r*col+c] << ",";
173 os << m[col*(r-1)-1] << "),";
176 for (unsigned c=0; c<col-1; ++c)
177 os << m[(row-1)*col+c] << ",";
178 os << m[row*col-1] << "))";
181 /** nops is defined to be rows x columns. */
182 unsigned matrix::nops() const
187 /** returns matrix entry at position (i/col, i%col). */
188 ex matrix::op(int i) const
193 /** returns matrix entry at position (i/col, i%col). */
194 ex & matrix::let_op(int i)
197 GINAC_ASSERT(i<nops());
202 /** expands the elements of a matrix entry by entry. */
203 ex matrix::expand(unsigned options) const
205 exvector tmp(row*col);
206 for (unsigned i=0; i<row*col; ++i)
207 tmp[i] = m[i].expand(options);
209 return matrix(row, col, tmp);
212 /** Search ocurrences. A matrix 'has' an expression if it is the expression
213 * itself or one of the elements 'has' it. */
214 bool matrix::has(const ex & other) const
216 GINAC_ASSERT(other.bp!=0);
218 // tautology: it is the expression itself
219 if (is_equal(*other.bp)) return true;
221 // search all the elements
222 for (exvector::const_iterator r=m.begin(); r!=m.end(); ++r)
223 if ((*r).has(other)) return true;
228 /** evaluate matrix entry by entry. */
229 ex matrix::eval(int level) const
231 debugmsg("matrix eval",LOGLEVEL_MEMBER_FUNCTION);
233 // check if we have to do anything at all
234 if ((level==1)&&(flags & status_flags::evaluated))
238 if (level == -max_recursion_level)
239 throw (std::runtime_error("matrix::eval(): recursion limit exceeded"));
241 // eval() entry by entry
242 exvector m2(row*col);
244 for (unsigned r=0; r<row; ++r)
245 for (unsigned c=0; c<col; ++c)
246 m2[r*col+c] = m[r*col+c].eval(level);
248 return (new matrix(row, col, m2))->setflag(status_flags::dynallocated |
249 status_flags::evaluated );
252 /** evaluate matrix numerically entry by entry. */
253 ex matrix::evalf(int level) const
255 debugmsg("matrix evalf",LOGLEVEL_MEMBER_FUNCTION);
257 // check if we have to do anything at all
262 if (level == -max_recursion_level) {
263 throw (std::runtime_error("matrix::evalf(): recursion limit exceeded"));
266 // evalf() entry by entry
267 exvector m2(row*col);
269 for (unsigned r=0; r<row; ++r)
270 for (unsigned c=0; c<col; ++c)
271 m2[r*col+c] = m[r*col+c].evalf(level);
273 return matrix(row, col, m2);
278 int matrix::compare_same_type(const basic & other) const
280 GINAC_ASSERT(is_exactly_of_type(other, matrix));
281 const matrix & o = static_cast<matrix &>(const_cast<basic &>(other));
283 // compare number of rows
285 return row < o.rows() ? -1 : 1;
287 // compare number of columns
289 return col < o.cols() ? -1 : 1;
291 // equal number of rows and columns, compare individual elements
293 for (unsigned r=0; r<row; ++r) {
294 for (unsigned c=0; c<col; ++c) {
295 cmpval = ((*this)(r,c)).compare(o(r,c));
296 if (cmpval!=0) return cmpval;
299 // all elements are equal => matrices are equal;
304 // non-virtual functions in this class
311 * @exception logic_error (incompatible matrices) */
312 matrix matrix::add(const matrix & other) const
314 if (col != other.col || row != other.row)
315 throw (std::logic_error("matrix::add(): incompatible matrices"));
317 exvector sum(this->m);
318 exvector::iterator i;
319 exvector::const_iterator ci;
320 for (i=sum.begin(), ci=other.m.begin(); i!=sum.end(); ++i, ++ci)
323 return matrix(row,col,sum);
327 /** Difference of matrices.
329 * @exception logic_error (incompatible matrices) */
330 matrix matrix::sub(const matrix & other) const
332 if (col != other.col || row != other.row)
333 throw (std::logic_error("matrix::sub(): incompatible matrices"));
335 exvector dif(this->m);
336 exvector::iterator i;
337 exvector::const_iterator ci;
338 for (i=dif.begin(), ci=other.m.begin(); i!=dif.end(); ++i, ++ci)
341 return matrix(row,col,dif);
345 /** Product of matrices.
347 * @exception logic_error (incompatible matrices) */
348 matrix matrix::mul(const matrix & other) const
350 if (this->cols() != other.rows())
351 throw (std::logic_error("matrix::mul(): incompatible matrices"));
353 exvector prod(this->rows()*other.cols());
355 for (unsigned r1=0; r1<this->rows(); ++r1) {
356 for (unsigned c=0; c<this->cols(); ++c) {
357 if (m[r1*col+c].is_zero())
359 for (unsigned r2=0; r2<other.cols(); ++r2)
360 prod[r1*other.col+r2] += (m[r1*col+c] * other.m[c*other.col+r2]).expand();
363 return matrix(row, other.col, prod);
367 /** operator() to access elements.
369 * @param ro row of element
370 * @param co column of element
371 * @exception range_error (index out of range) */
372 const ex & matrix::operator() (unsigned ro, unsigned co) const
374 if (ro>=row || co>=col)
375 throw (std::range_error("matrix::operator(): index out of range"));
381 /** Set individual elements manually.
383 * @exception range_error (index out of range) */
384 matrix & matrix::set(unsigned ro, unsigned co, ex value)
386 if (ro>=row || co>=col)
387 throw (std::range_error("matrix::set(): index out of range"));
389 ensure_if_modifiable();
390 m[ro*col+co] = value;
395 /** Transposed of an m x n matrix, producing a new n x m matrix object that
396 * represents the transposed. */
397 matrix matrix::transpose(void) const
399 exvector trans(this->cols()*this->rows());
401 for (unsigned r=0; r<this->cols(); ++r)
402 for (unsigned c=0; c<this->rows(); ++c)
403 trans[r*this->rows()+c] = m[c*this->cols()+r];
405 return matrix(this->cols(),this->rows(),trans);
409 /** Determinant of square matrix. This routine doesn't actually calculate the
410 * determinant, it only implements some heuristics about which algorithm to
411 * run. If all the elements of the matrix are elements of an integral domain
412 * the determinant is also in that integral domain and the result is expanded
413 * only. If one or more elements are from a quotient field the determinant is
414 * usually also in that quotient field and the result is normalized before it
415 * is returned. This implies that the determinant of the symbolic 2x2 matrix
416 * [[a/(a-b),1],[b/(a-b),1]] is returned as unity. (In this respect, it
417 * behaves like MapleV and unlike Mathematica.)
419 * @param algo allows to chose an algorithm
420 * @return the determinant as a new expression
421 * @exception logic_error (matrix not square)
422 * @see determinant_algo */
423 ex matrix::determinant(unsigned algo) const
426 throw (std::logic_error("matrix::determinant(): matrix not square"));
427 GINAC_ASSERT(row*col==m.capacity());
429 // Gather some statistical information about this matrix:
430 bool numeric_flag = true;
431 bool normal_flag = false;
432 unsigned sparse_count = 0; // counts non-zero elements
433 for (exvector::const_iterator r=m.begin(); r!=m.end(); ++r) {
434 lst srl; // symbol replacement list
435 ex rtest = (*r).to_rational(srl);
436 if (!rtest.is_zero())
438 if (!rtest.info(info_flags::numeric))
439 numeric_flag = false;
440 if (!rtest.info(info_flags::crational_polynomial) &&
441 rtest.info(info_flags::rational_function))
445 // Here is the heuristics in case this routine has to decide:
446 if (algo == determinant_algo::automatic) {
447 // Minor expansion is generally a good guess:
448 algo = determinant_algo::laplace;
449 // Does anybody know when a matrix is really sparse?
450 // Maybe <~row/2.236 nonzero elements average in a row?
451 if (row>3 && 5*sparse_count<=row*col)
452 algo = determinant_algo::bareiss;
453 // Purely numeric matrix can be handled by Gauss elimination.
454 // This overrides any prior decisions.
456 algo = determinant_algo::gauss;
459 // Trap the trivial case here, since some algorithms don't like it
461 // for consistency with non-trivial determinants...
463 return m[0].normal();
465 return m[0].expand();
468 // Compute the determinant
470 case determinant_algo::gauss: {
473 int sign = tmp.gauss_elimination(true);
474 for (unsigned d=0; d<row; ++d)
475 det *= tmp.m[d*col+d];
477 return (sign*det).normal();
479 return (sign*det).normal().expand();
481 case determinant_algo::bareiss: {
484 sign = tmp.fraction_free_elimination(true);
486 return (sign*tmp.m[row*col-1]).normal();
488 return (sign*tmp.m[row*col-1]).expand();
490 case determinant_algo::divfree: {
493 sign = tmp.division_free_elimination(true);
496 ex det = tmp.m[row*col-1];
497 // factor out accumulated bogus slag
498 for (unsigned d=0; d<row-2; ++d)
499 for (unsigned j=0; j<row-d-2; ++j)
500 det = (det/tmp.m[d*col+d]).normal();
503 case determinant_algo::laplace:
505 // This is the minor expansion scheme. We always develop such
506 // that the smallest minors (i.e, the trivial 1x1 ones) are on the
507 // rightmost column. For this to be efficient it turns out that
508 // the emptiest columns (i.e. the ones with most zeros) should be
509 // the ones on the right hand side. Therefore we presort the
510 // columns of the matrix:
511 typedef std::pair<unsigned,unsigned> uintpair;
512 std::vector<uintpair> c_zeros; // number of zeros in column
513 for (unsigned c=0; c<col; ++c) {
515 for (unsigned r=0; r<row; ++r)
516 if (m[r*col+c].is_zero())
518 c_zeros.push_back(uintpair(acc,c));
520 sort(c_zeros.begin(),c_zeros.end());
521 std::vector<unsigned> pre_sort;
522 for (std::vector<uintpair>::iterator i=c_zeros.begin(); i!=c_zeros.end(); ++i)
523 pre_sort.push_back(i->second);
524 int sign = permutation_sign(pre_sort);
525 exvector result(row*col); // represents sorted matrix
527 for (std::vector<unsigned>::iterator i=pre_sort.begin();
530 for (unsigned r=0; r<row; ++r)
531 result[r*col+c] = m[r*col+(*i)];
535 return (sign*matrix(row,col,result).determinant_minor()).normal();
537 return sign*matrix(row,col,result).determinant_minor();
543 /** Trace of a matrix. The result is normalized if it is in some quotient
544 * field and expanded only otherwise. This implies that the trace of the
545 * symbolic 2x2 matrix [[a/(a-b),x],[y,b/(b-a)]] is recognized to be unity.
547 * @return the sum of diagonal elements
548 * @exception logic_error (matrix not square) */
549 ex matrix::trace(void) const
552 throw (std::logic_error("matrix::trace(): matrix not square"));
555 for (unsigned r=0; r<col; ++r)
558 if (tr.info(info_flags::rational_function) &&
559 !tr.info(info_flags::crational_polynomial))
566 /** Characteristic Polynomial. Following mathematica notation the
567 * characteristic polynomial of a matrix M is defined as the determiant of
568 * (M - lambda * 1) where 1 stands for the unit matrix of the same dimension
569 * as M. Note that some CASs define it with a sign inside the determinant
570 * which gives rise to an overall sign if the dimension is odd. This method
571 * returns the characteristic polynomial collected in powers of lambda as a
574 * @return characteristic polynomial as new expression
575 * @exception logic_error (matrix not square)
576 * @see matrix::determinant() */
577 ex matrix::charpoly(const symbol & lambda) const
580 throw (std::logic_error("matrix::charpoly(): matrix not square"));
582 bool numeric_flag = true;
583 for (exvector::const_iterator r=m.begin(); r!=m.end(); ++r) {
584 if (!(*r).info(info_flags::numeric)) {
585 numeric_flag = false;
589 // The pure numeric case is traditionally rather common. Hence, it is
590 // trapped and we use Leverrier's algorithm which goes as row^3 for
591 // every coefficient. The expensive part is the matrix multiplication.
595 ex poly = power(lambda,row)-c*power(lambda,row-1);
596 for (unsigned i=1; i<row; ++i) {
597 for (unsigned j=0; j<row; ++j)
600 c = B.trace()/ex(i+1);
601 poly -= c*power(lambda,row-i-1);
610 for (unsigned r=0; r<col; ++r)
611 M.m[r*col+r] -= lambda;
613 return M.determinant().collect(lambda);
617 /** Inverse of this matrix.
619 * @return the inverted matrix
620 * @exception logic_error (matrix not square)
621 * @exception runtime_error (singular matrix) */
622 matrix matrix::inverse(void) const
625 throw (std::logic_error("matrix::inverse(): matrix not square"));
627 // NOTE: the Gauss-Jordan elimination used here can in principle be
628 // replaced by two clever calls to gauss_elimination() and some to
629 // transpose(). Wouldn't be more efficient (maybe less?), just more
632 // set tmp to the unit matrix
633 for (unsigned i=0; i<col; ++i)
634 tmp.m[i*col+i] = _ex1();
636 // create a copy of this matrix
638 for (unsigned r1=0; r1<row; ++r1) {
639 int indx = cpy.pivot(r1, r1);
641 throw (std::runtime_error("matrix::inverse(): singular matrix"));
643 if (indx != 0) { // swap rows r and indx of matrix tmp
644 for (unsigned i=0; i<col; ++i)
645 tmp.m[r1*col+i].swap(tmp.m[indx*col+i]);
647 ex a1 = cpy.m[r1*col+r1];
648 for (unsigned c=0; c<col; ++c) {
649 cpy.m[r1*col+c] /= a1;
650 tmp.m[r1*col+c] /= a1;
652 for (unsigned r2=0; r2<row; ++r2) {
654 if (!cpy.m[r2*col+r1].is_zero()) {
655 ex a2 = cpy.m[r2*col+r1];
656 // yes, there is something to do in this column
657 for (unsigned c=0; c<col; ++c) {
658 cpy.m[r2*col+c] -= a2 * cpy.m[r1*col+c];
659 if (!cpy.m[r2*col+c].info(info_flags::numeric))
660 cpy.m[r2*col+c] = cpy.m[r2*col+c].normal();
661 tmp.m[r2*col+c] -= a2 * tmp.m[r1*col+c];
662 if (!tmp.m[r2*col+c].info(info_flags::numeric))
663 tmp.m[r2*col+c] = tmp.m[r2*col+c].normal();
674 /** Solve a linear system consisting of a m x n matrix and a m x p right hand
675 * side by applying an elimination scheme to the augmented matrix.
677 * @param vars n x p matrix, all elements must be symbols
678 * @param rhs m x p matrix
679 * @return n x p solution matrix
680 * @exception logic_error (incompatible matrices)
681 * @exception invalid_argument (1st argument must be matrix of symbols)
682 * @exception runtime_error (inconsistent linear system)
684 matrix matrix::solve(const matrix & vars,
688 const unsigned m = this->rows();
689 const unsigned n = this->cols();
690 const unsigned p = rhs.cols();
693 if ((rhs.rows() != m) || (vars.rows() != n) || (vars.col != p))
694 throw (std::logic_error("matrix::solve(): incompatible matrices"));
695 for (unsigned ro=0; ro<n; ++ro)
696 for (unsigned co=0; co<p; ++co)
697 if (!vars(ro,co).info(info_flags::symbol))
698 throw (std::invalid_argument("matrix::solve(): 1st argument must be matrix of symbols"));
700 // build the augmented matrix of *this with rhs attached to the right
702 for (unsigned r=0; r<m; ++r) {
703 for (unsigned c=0; c<n; ++c)
704 aug.m[r*(n+p)+c] = this->m[r*n+c];
705 for (unsigned c=0; c<p; ++c)
706 aug.m[r*(n+p)+c+n] = rhs.m[r*p+c];
709 // Gather some statistical information about the augmented matrix:
710 bool numeric_flag = true;
711 for (exvector::const_iterator r=aug.m.begin(); r!=aug.m.end(); ++r) {
712 if (!(*r).info(info_flags::numeric))
713 numeric_flag = false;
716 // Here is the heuristics in case this routine has to decide:
717 if (algo == solve_algo::automatic) {
718 // Bareiss (fraction-free) elimination is generally a good guess:
719 algo = solve_algo::bareiss;
720 // For m<3, Bareiss elimination is equivalent to division free
721 // elimination but has more logistic overhead
723 algo = solve_algo::divfree;
724 // This overrides any prior decisions.
726 algo = solve_algo::gauss;
729 // Eliminate the augmented matrix:
731 case solve_algo::gauss:
732 aug.gauss_elimination();
733 case solve_algo::divfree:
734 aug.division_free_elimination();
735 case solve_algo::bareiss:
737 aug.fraction_free_elimination();
740 // assemble the solution matrix:
742 for (unsigned co=0; co<p; ++co) {
743 unsigned last_assigned_sol = n+1;
744 for (int r=m-1; r>=0; --r) {
745 unsigned fnz = 1; // first non-zero in row
746 while ((fnz<=n) && (aug.m[r*(n+p)+(fnz-1)].is_zero()))
749 // row consists only of zeros, corresponding rhs must be 0, too
750 if (!aug.m[r*(n+p)+n+co].is_zero()) {
751 throw (std::runtime_error("matrix::solve(): inconsistent linear system"));
754 // assign solutions for vars between fnz+1 and
755 // last_assigned_sol-1: free parameters
756 for (unsigned c=fnz; c<last_assigned_sol-1; ++c)
757 sol.set(c,co,vars.m[c*p+co]);
758 ex e = aug.m[r*(n+p)+n+co];
759 for (unsigned c=fnz; c<n; ++c)
760 e -= aug.m[r*(n+p)+c]*sol.m[c*p+co];
762 (e/(aug.m[r*(n+p)+(fnz-1)])).normal());
763 last_assigned_sol = fnz;
766 // assign solutions for vars between 1 and
767 // last_assigned_sol-1: free parameters
768 for (unsigned ro=0; ro<last_assigned_sol-1; ++ro)
769 sol.set(ro,co,vars(ro,co));
778 /** Recursive determinant for small matrices having at least one symbolic
779 * entry. The basic algorithm, known as Laplace-expansion, is enhanced by
780 * some bookkeeping to avoid calculation of the same submatrices ("minors")
781 * more than once. According to W.M.Gentleman and S.C.Johnson this algorithm
782 * is better than elimination schemes for matrices of sparse multivariate
783 * polynomials and also for matrices of dense univariate polynomials if the
784 * matrix' dimesion is larger than 7.
786 * @return the determinant as a new expression (in expanded form)
787 * @see matrix::determinant() */
788 ex matrix::determinant_minor(void) const
790 // for small matrices the algorithm does not make any sense:
791 const unsigned n = this->cols();
793 return m[0].expand();
795 return (m[0]*m[3]-m[2]*m[1]).expand();
797 return (m[0]*m[4]*m[8]-m[0]*m[5]*m[7]-
798 m[1]*m[3]*m[8]+m[2]*m[3]*m[7]+
799 m[1]*m[5]*m[6]-m[2]*m[4]*m[6]).expand();
801 // This algorithm can best be understood by looking at a naive
802 // implementation of Laplace-expansion, like this one:
804 // matrix minorM(this->rows()-1,this->cols()-1);
805 // for (unsigned r1=0; r1<this->rows(); ++r1) {
806 // // shortcut if element(r1,0) vanishes
807 // if (m[r1*col].is_zero())
809 // // assemble the minor matrix
810 // for (unsigned r=0; r<minorM.rows(); ++r) {
811 // for (unsigned c=0; c<minorM.cols(); ++c) {
813 // minorM.set(r,c,m[r*col+c+1]);
815 // minorM.set(r,c,m[(r+1)*col+c+1]);
818 // // recurse down and care for sign:
820 // det -= m[r1*col] * minorM.determinant_minor();
822 // det += m[r1*col] * minorM.determinant_minor();
824 // return det.expand();
825 // What happens is that while proceeding down many of the minors are
826 // computed more than once. In particular, there are binomial(n,k)
827 // kxk minors and each one is computed factorial(n-k) times. Therefore
828 // it is reasonable to store the results of the minors. We proceed from
829 // right to left. At each column c we only need to retrieve the minors
830 // calculated in step c-1. We therefore only have to store at most
831 // 2*binomial(n,n/2) minors.
833 // Unique flipper counter for partitioning into minors
834 std::vector<unsigned> Pkey;
836 // key for minor determinant (a subpartition of Pkey)
837 std::vector<unsigned> Mkey;
839 // we store our subminors in maps, keys being the rows they arise from
840 typedef std::map<std::vector<unsigned>,class ex> Rmap;
841 typedef std::map<std::vector<unsigned>,class ex>::value_type Rmap_value;
845 // initialize A with last column:
846 for (unsigned r=0; r<n; ++r) {
847 Pkey.erase(Pkey.begin(),Pkey.end());
849 A.insert(Rmap_value(Pkey,m[n*(r+1)-1]));
851 // proceed from right to left through matrix
852 for (int c=n-2; c>=0; --c) {
853 Pkey.erase(Pkey.begin(),Pkey.end()); // don't change capacity
854 Mkey.erase(Mkey.begin(),Mkey.end());
855 for (unsigned i=0; i<n-c; ++i)
857 unsigned fc = 0; // controls logic for our strange flipper counter
860 for (unsigned r=0; r<n-c; ++r) {
861 // maybe there is nothing to do?
862 if (m[Pkey[r]*n+c].is_zero())
864 // create the sorted key for all possible minors
865 Mkey.erase(Mkey.begin(),Mkey.end());
866 for (unsigned i=0; i<n-c; ++i)
868 Mkey.push_back(Pkey[i]);
869 // Fetch the minors and compute the new determinant
871 det -= m[Pkey[r]*n+c]*A[Mkey];
873 det += m[Pkey[r]*n+c]*A[Mkey];
875 // prevent build-up of deep nesting of expressions saves time:
877 // store the new determinant at its place in B:
879 B.insert(Rmap_value(Pkey,det));
880 // increment our strange flipper counter
881 for (fc=n-c; fc>0; --fc) {
887 for (unsigned j=fc; j<n-c; ++j)
888 Pkey[j] = Pkey[j-1]+1;
890 // next column, so change the role of A and B:
899 /** Perform the steps of an ordinary Gaussian elimination to bring the m x n
900 * matrix into an upper echelon form. The algorithm is ok for matrices
901 * with numeric coefficients but quite unsuited for symbolic matrices.
903 * @param det may be set to true to save a lot of space if one is only
904 * interested in the diagonal elements (i.e. for calculating determinants).
905 * The others are set to zero in this case.
906 * @return sign is 1 if an even number of rows was swapped, -1 if an odd
907 * number of rows was swapped and 0 if the matrix is singular. */
908 int matrix::gauss_elimination(const bool det)
910 ensure_if_modifiable();
911 const unsigned m = this->rows();
912 const unsigned n = this->cols();
913 GINAC_ASSERT(!det || n==m);
917 for (unsigned r1=0; (r1<n-1)&&(r0<m-1); ++r1) {
918 int indx = pivot(r0, r1, true);
922 return 0; // leaves *this in a messy state
927 for (unsigned r2=r0+1; r2<m; ++r2) {
928 if (!this->m[r2*n+r1].is_zero()) {
929 // yes, there is something to do in this row
930 ex piv = this->m[r2*n+r1] / this->m[r0*n+r1];
931 for (unsigned c=r1+1; c<n; ++c) {
932 this->m[r2*n+c] -= piv * this->m[r0*n+c];
933 if (!this->m[r2*n+c].info(info_flags::numeric))
934 this->m[r2*n+c] = this->m[r2*n+c].normal();
937 // fill up left hand side with zeros
938 for (unsigned c=0; c<=r1; ++c)
939 this->m[r2*n+c] = _ex0();
942 // save space by deleting no longer needed elements
943 for (unsigned c=r0+1; c<n; ++c)
944 this->m[r0*n+c] = _ex0();
954 /** Perform the steps of division free elimination to bring the m x n matrix
955 * into an upper echelon form.
957 * @param det may be set to true to save a lot of space if one is only
958 * interested in the diagonal elements (i.e. for calculating determinants).
959 * The others are set to zero in this case.
960 * @return sign is 1 if an even number of rows was swapped, -1 if an odd
961 * number of rows was swapped and 0 if the matrix is singular. */
962 int matrix::division_free_elimination(const bool det)
964 ensure_if_modifiable();
965 const unsigned m = this->rows();
966 const unsigned n = this->cols();
967 GINAC_ASSERT(!det || n==m);
971 for (unsigned r1=0; (r1<n-1)&&(r0<m-1); ++r1) {
972 int indx = pivot(r0, r1, true);
976 return 0; // leaves *this in a messy state
981 for (unsigned r2=r0+1; r2<m; ++r2) {
982 for (unsigned c=r1+1; c<n; ++c)
983 this->m[r2*n+c] = (this->m[r0*n+r1]*this->m[r2*n+c] - this->m[r2*n+r1]*this->m[r0*n+c]).expand();
984 // fill up left hand side with zeros
985 for (unsigned c=0; c<=r1; ++c)
986 this->m[r2*n+c] = _ex0();
989 // save space by deleting no longer needed elements
990 for (unsigned c=r0+1; c<n; ++c)
991 this->m[r0*n+c] = _ex0();
1001 /** Perform the steps of Bareiss' one-step fraction free elimination to bring
1002 * the matrix into an upper echelon form. Fraction free elimination means
1003 * that divide is used straightforwardly, without computing GCDs first. This
1004 * is possible, since we know the divisor at each step.
1006 * @param det may be set to true to save a lot of space if one is only
1007 * interested in the last element (i.e. for calculating determinants). The
1008 * others are set to zero in this case.
1009 * @return sign is 1 if an even number of rows was swapped, -1 if an odd
1010 * number of rows was swapped and 0 if the matrix is singular. */
1011 int matrix::fraction_free_elimination(const bool det)
1014 // (single-step fraction free elimination scheme, already known to Jordan)
1016 // Usual division-free elimination sets m[0](r,c) = m(r,c) and then sets
1017 // m[k+1](r,c) = m[k](k,k) * m[k](r,c) - m[k](r,k) * m[k](k,c).
1019 // Bareiss (fraction-free) elimination in addition divides that element
1020 // by m[k-1](k-1,k-1) for k>1, where it can be shown by means of the
1021 // Sylvester determinant that this really divides m[k+1](r,c).
1023 // We also allow rational functions where the original prove still holds.
1024 // However, we must care for numerator and denominator separately and
1025 // "manually" work in the integral domains because of subtle cancellations
1026 // (see below). This blows up the bookkeeping a bit and the formula has
1027 // to be modified to expand like this (N{x} stands for numerator of x,
1028 // D{x} for denominator of x):
1029 // N{m[k+1](r,c)} = N{m[k](k,k)}*N{m[k](r,c)}*D{m[k](r,k)}*D{m[k](k,c)}
1030 // -N{m[k](r,k)}*N{m[k](k,c)}*D{m[k](k,k)}*D{m[k](r,c)}
1031 // D{m[k+1](r,c)} = D{m[k](k,k)}*D{m[k](r,c)}*D{m[k](r,k)}*D{m[k](k,c)}
1032 // where for k>1 we now divide N{m[k+1](r,c)} by
1033 // N{m[k-1](k-1,k-1)}
1034 // and D{m[k+1](r,c)} by
1035 // D{m[k-1](k-1,k-1)}.
1037 ensure_if_modifiable();
1038 const unsigned m = this->rows();
1039 const unsigned n = this->cols();
1040 GINAC_ASSERT(!det || n==m);
1049 // We populate temporary matrices to subsequently operate on. There is
1050 // one holding numerators and another holding denominators of entries.
1051 // This is a must since the evaluator (or even earlier mul's constructor)
1052 // might cancel some trivial element which causes divide() to fail. The
1053 // elements are normalized first (yes, even though this algorithm doesn't
1054 // need GCDs) since the elements of *this might be unnormalized, which
1055 // makes things more complicated than they need to be.
1056 matrix tmp_n(*this);
1057 matrix tmp_d(m,n); // for denominators, if needed
1058 lst srl; // symbol replacement list
1059 exvector::iterator it = this->m.begin();
1060 exvector::iterator tmp_n_it = tmp_n.m.begin();
1061 exvector::iterator tmp_d_it = tmp_d.m.begin();
1062 for (; it!= this->m.end(); ++it, ++tmp_n_it, ++tmp_d_it) {
1063 (*tmp_n_it) = (*it).normal().to_rational(srl);
1064 (*tmp_d_it) = (*tmp_n_it).denom();
1065 (*tmp_n_it) = (*tmp_n_it).numer();
1069 for (unsigned r1=0; (r1<n-1)&&(r0<m-1); ++r1) {
1070 int indx = tmp_n.pivot(r0, r1, true);
1079 // tmp_n's rows r0 and indx were swapped, do the same in tmp_d:
1080 for (unsigned c=r1; c<n; ++c)
1081 tmp_d.m[n*indx+c].swap(tmp_d.m[n*r0+c]);
1083 for (unsigned r2=r0+1; r2<m; ++r2) {
1084 for (unsigned c=r1+1; c<n; ++c) {
1085 dividend_n = (tmp_n.m[r0*n+r1]*tmp_n.m[r2*n+c]*
1086 tmp_d.m[r2*n+r1]*tmp_d.m[r0*n+c]
1087 -tmp_n.m[r2*n+r1]*tmp_n.m[r0*n+c]*
1088 tmp_d.m[r0*n+r1]*tmp_d.m[r2*n+c]).expand();
1089 dividend_d = (tmp_d.m[r2*n+r1]*tmp_d.m[r0*n+c]*
1090 tmp_d.m[r0*n+r1]*tmp_d.m[r2*n+c]).expand();
1091 bool check = divide(dividend_n, divisor_n,
1092 tmp_n.m[r2*n+c], true);
1093 check &= divide(dividend_d, divisor_d,
1094 tmp_d.m[r2*n+c], true);
1095 GINAC_ASSERT(check);
1097 // fill up left hand side with zeros
1098 for (unsigned c=0; c<=r1; ++c)
1099 tmp_n.m[r2*n+c] = _ex0();
1101 if ((r1<n-1)&&(r0<m-1)) {
1102 // compute next iteration's divisor
1103 divisor_n = tmp_n.m[r0*n+r1].expand();
1104 divisor_d = tmp_d.m[r0*n+r1].expand();
1106 // save space by deleting no longer needed elements
1107 for (unsigned c=0; c<n; ++c) {
1108 tmp_n.m[r0*n+c] = _ex0();
1109 tmp_d.m[r0*n+c] = _ex1();
1116 // repopulate *this matrix:
1117 it = this->m.begin();
1118 tmp_n_it = tmp_n.m.begin();
1119 tmp_d_it = tmp_d.m.begin();
1120 for (; it!= this->m.end(); ++it, ++tmp_n_it, ++tmp_d_it)
1121 (*it) = ((*tmp_n_it)/(*tmp_d_it)).subs(srl);
1127 /** Partial pivoting method for matrix elimination schemes.
1128 * Usual pivoting (symbolic==false) returns the index to the element with the
1129 * largest absolute value in column ro and swaps the current row with the one
1130 * where the element was found. With (symbolic==true) it does the same thing
1131 * with the first non-zero element.
1133 * @param ro is the row from where to begin
1134 * @param co is the column to be inspected
1135 * @param symbolic signal if we want the first non-zero element to be pivoted
1136 * (true) or the one with the largest absolute value (false).
1137 * @return 0 if no interchange occured, -1 if all are zero (usually signaling
1138 * a degeneracy) and positive integer k means that rows ro and k were swapped.
1140 int matrix::pivot(unsigned ro, unsigned co, bool symbolic)
1144 // search first non-zero element in column co beginning at row ro
1145 while ((k<row) && (this->m[k*col+co].expand().is_zero()))
1148 // search largest element in column co beginning at row ro
1149 GINAC_ASSERT(is_ex_of_type(this->m[k*col+co],numeric));
1150 unsigned kmax = k+1;
1151 numeric mmax = abs(ex_to_numeric(m[kmax*col+co]));
1153 GINAC_ASSERT(is_ex_of_type(this->m[kmax*col+co],numeric));
1154 numeric tmp = ex_to_numeric(this->m[kmax*col+co]);
1155 if (abs(tmp) > mmax) {
1161 if (!mmax.is_zero())
1165 // all elements in column co below row ro vanish
1168 // matrix needs no pivoting
1170 // matrix needs pivoting, so swap rows k and ro
1171 ensure_if_modifiable();
1172 for (unsigned c=0; c<col; ++c)
1173 this->m[k*col+c].swap(this->m[ro*col+c]);
1178 /** Convert list of lists to matrix. */
1179 ex lst_to_matrix(const ex &l)
1181 if (!is_ex_of_type(l, lst))
1182 throw(std::invalid_argument("argument to lst_to_matrix() must be a lst"));
1184 // Find number of rows and columns
1185 unsigned rows = l.nops(), cols = 0, i, j;
1186 for (i=0; i<rows; i++)
1187 if (l.op(i).nops() > cols)
1188 cols = l.op(i).nops();
1190 // Allocate and fill matrix
1191 matrix &m = *new matrix(rows, cols);
1192 for (i=0; i<rows; i++)
1193 for (j=0; j<cols; j++)
1194 if (l.op(i).nops() > j)
1195 m.set(i, j, l.op(i).op(j));
1201 #ifndef NO_NAMESPACE_GINAC
1202 } // namespace GiNaC
1203 #endif // ndef NO_NAMESPACE_GINAC