3 * Implementation of symbolic matrices */
6 * GiNaC Copyright (C) 1999-2003 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
38 #include "operators.h"
46 GINAC_IMPLEMENT_REGISTERED_CLASS(matrix, basic)
49 // default ctor, dtor, copy ctor, assignment operator and helpers:
52 /** Default ctor. Initializes to 1 x 1-dimensional zero-matrix. */
53 matrix::matrix() : inherited(TINFO_matrix), row(1), col(1)
58 void matrix::copy(const matrix & other)
60 inherited::copy(other);
63 m = other.m; // STL's vector copying invoked here
66 DEFAULT_DESTROY(matrix)
74 /** Very common ctor. Initializes to r x c-dimensional zero-matrix.
76 * @param r number of rows
77 * @param c number of cols */
78 matrix::matrix(unsigned r, unsigned c)
79 : inherited(TINFO_matrix), row(r), col(c)
86 /** Ctor from representation, for internal use only. */
87 matrix::matrix(unsigned r, unsigned c, const exvector & m2)
88 : inherited(TINFO_matrix), row(r), col(c), m(m2) {}
90 /** Construct matrix from (flat) list of elements. If the list has fewer
91 * elements than the matrix, the remaining matrix elements are set to zero.
92 * If the list has more elements than the matrix, the excessive elements are
94 matrix::matrix(unsigned r, unsigned c, const lst & l)
95 : inherited(TINFO_matrix), row(r), col(c)
100 for (lst::const_iterator it = l.begin(); it != l.end(); ++it, ++i) {
104 break; // matrix smaller than list: throw away excessive elements
113 matrix::matrix(const archive_node &n, lst &sym_lst) : inherited(n, sym_lst)
115 if (!(n.find_unsigned("row", row)) || !(n.find_unsigned("col", col)))
116 throw (std::runtime_error("unknown matrix dimensions in archive"));
117 m.reserve(row * col);
118 for (unsigned int i=0; true; i++) {
120 if (n.find_ex("m", e, sym_lst, i))
127 void matrix::archive(archive_node &n) const
129 inherited::archive(n);
130 n.add_unsigned("row", row);
131 n.add_unsigned("col", col);
132 exvector::const_iterator i = m.begin(), iend = m.end();
139 DEFAULT_UNARCHIVE(matrix)
142 // functions overriding virtual functions from base classes
147 void matrix::print(const print_context & c, unsigned level) const
149 if (is_a<print_tree>(c)) {
151 inherited::print(c, level);
155 if (is_a<print_python_repr>(c))
156 c.s << class_name() << '(';
158 if (is_a<print_latex>(c))
159 c.s << "\\left(\\begin{array}{" << std::string(col,'c') << "}";
163 for (unsigned ro=0; ro<row; ++ro) {
164 if (!is_a<print_latex>(c))
166 for (unsigned co=0; co<col; ++co) {
167 m[ro*col+co].print(c);
169 if (is_a<print_latex>(c))
174 if (!is_a<print_latex>(c))
179 if (is_a<print_latex>(c))
186 if (is_a<print_latex>(c))
187 c.s << "\\end{array}\\right)";
191 if (is_a<print_python_repr>(c))
197 /** nops is defined to be rows x columns. */
198 size_t matrix::nops() const
200 return static_cast<size_t>(row) * static_cast<size_t>(col);
203 /** returns matrix entry at position (i/col, i%col). */
204 ex matrix::op(size_t i) const
206 GINAC_ASSERT(i<nops());
211 /** returns writable matrix entry at position (i/col, i%col). */
212 ex & matrix::let_op(size_t i)
214 GINAC_ASSERT(i<nops());
216 ensure_if_modifiable();
220 /** Evaluate matrix entry by entry. */
221 ex matrix::eval(int level) const
223 // check if we have to do anything at all
224 if ((level==1)&&(flags & status_flags::evaluated))
228 if (level == -max_recursion_level)
229 throw (std::runtime_error("matrix::eval(): recursion limit exceeded"));
231 // eval() entry by entry
232 exvector m2(row*col);
234 for (unsigned r=0; r<row; ++r)
235 for (unsigned c=0; c<col; ++c)
236 m2[r*col+c] = m[r*col+c].eval(level);
238 return (new matrix(row, col, m2))->setflag(status_flags::dynallocated |
239 status_flags::evaluated);
242 ex matrix::subs(const lst & ls, const lst & lr, unsigned options) const
244 exvector m2(row * col);
245 for (unsigned r=0; r<row; ++r)
246 for (unsigned c=0; c<col; ++c)
247 m2[r*col+c] = m[r*col+c].subs(ls, lr, options);
249 return matrix(row, col, m2).basic::subs(ls, lr, options);
254 int matrix::compare_same_type(const basic & other) const
256 GINAC_ASSERT(is_exactly_a<matrix>(other));
257 const matrix &o = static_cast<const matrix &>(other);
259 // compare number of rows
261 return row < o.rows() ? -1 : 1;
263 // compare number of columns
265 return col < o.cols() ? -1 : 1;
267 // equal number of rows and columns, compare individual elements
269 for (unsigned r=0; r<row; ++r) {
270 for (unsigned c=0; c<col; ++c) {
271 cmpval = ((*this)(r,c)).compare(o(r,c));
272 if (cmpval!=0) return cmpval;
275 // all elements are equal => matrices are equal;
279 bool matrix::match_same_type(const basic & other) const
281 GINAC_ASSERT(is_exactly_a<matrix>(other));
282 const matrix & o = static_cast<const matrix &>(other);
284 // The number of rows and columns must be the same. This is necessary to
285 // prevent a 2x3 matrix from matching a 3x2 one.
286 return row == o.rows() && col == o.cols();
289 /** Automatic symbolic evaluation of an indexed matrix. */
290 ex matrix::eval_indexed(const basic & i) const
292 GINAC_ASSERT(is_a<indexed>(i));
293 GINAC_ASSERT(is_a<matrix>(i.op(0)));
295 bool all_indices_unsigned = static_cast<const indexed &>(i).all_index_values_are(info_flags::nonnegint);
300 // One index, must be one-dimensional vector
301 if (row != 1 && col != 1)
302 throw (std::runtime_error("matrix::eval_indexed(): vector must have exactly 1 index"));
304 const idx & i1 = ex_to<idx>(i.op(1));
309 if (!i1.get_dim().is_equal(row))
310 throw (std::runtime_error("matrix::eval_indexed(): dimension of index must match number of vector elements"));
312 // Index numeric -> return vector element
313 if (all_indices_unsigned) {
314 unsigned n1 = ex_to<numeric>(i1.get_value()).to_int();
316 throw (std::runtime_error("matrix::eval_indexed(): value of index exceeds number of vector elements"));
317 return (*this)(n1, 0);
323 if (!i1.get_dim().is_equal(col))
324 throw (std::runtime_error("matrix::eval_indexed(): dimension of index must match number of vector elements"));
326 // Index numeric -> return vector element
327 if (all_indices_unsigned) {
328 unsigned n1 = ex_to<numeric>(i1.get_value()).to_int();
330 throw (std::runtime_error("matrix::eval_indexed(): value of index exceeds number of vector elements"));
331 return (*this)(0, n1);
335 } else if (i.nops() == 3) {
338 const idx & i1 = ex_to<idx>(i.op(1));
339 const idx & i2 = ex_to<idx>(i.op(2));
341 if (!i1.get_dim().is_equal(row))
342 throw (std::runtime_error("matrix::eval_indexed(): dimension of first index must match number of rows"));
343 if (!i2.get_dim().is_equal(col))
344 throw (std::runtime_error("matrix::eval_indexed(): dimension of second index must match number of columns"));
346 // Pair of dummy indices -> compute trace
347 if (is_dummy_pair(i1, i2))
350 // Both indices numeric -> return matrix element
351 if (all_indices_unsigned) {
352 unsigned n1 = ex_to<numeric>(i1.get_value()).to_int(), n2 = ex_to<numeric>(i2.get_value()).to_int();
354 throw (std::runtime_error("matrix::eval_indexed(): value of first index exceeds number of rows"));
356 throw (std::runtime_error("matrix::eval_indexed(): value of second index exceeds number of columns"));
357 return (*this)(n1, n2);
361 throw (std::runtime_error("matrix::eval_indexed(): matrix must have exactly 2 indices"));
366 /** Sum of two indexed matrices. */
367 ex matrix::add_indexed(const ex & self, const ex & other) const
369 GINAC_ASSERT(is_a<indexed>(self));
370 GINAC_ASSERT(is_a<matrix>(self.op(0)));
371 GINAC_ASSERT(is_a<indexed>(other));
372 GINAC_ASSERT(self.nops() == 2 || self.nops() == 3);
374 // Only add two matrices
375 if (is_a<matrix>(other.op(0))) {
376 GINAC_ASSERT(other.nops() == 2 || other.nops() == 3);
378 const matrix &self_matrix = ex_to<matrix>(self.op(0));
379 const matrix &other_matrix = ex_to<matrix>(other.op(0));
381 if (self.nops() == 2 && other.nops() == 2) { // vector + vector
383 if (self_matrix.row == other_matrix.row)
384 return indexed(self_matrix.add(other_matrix), self.op(1));
385 else if (self_matrix.row == other_matrix.col)
386 return indexed(self_matrix.add(other_matrix.transpose()), self.op(1));
388 } else if (self.nops() == 3 && other.nops() == 3) { // matrix + matrix
390 if (self.op(1).is_equal(other.op(1)) && self.op(2).is_equal(other.op(2)))
391 return indexed(self_matrix.add(other_matrix), self.op(1), self.op(2));
392 else if (self.op(1).is_equal(other.op(2)) && self.op(2).is_equal(other.op(1)))
393 return indexed(self_matrix.add(other_matrix.transpose()), self.op(1), self.op(2));
398 // Don't know what to do, return unevaluated sum
402 /** Product of an indexed matrix with a number. */
403 ex matrix::scalar_mul_indexed(const ex & self, const numeric & other) const
405 GINAC_ASSERT(is_a<indexed>(self));
406 GINAC_ASSERT(is_a<matrix>(self.op(0)));
407 GINAC_ASSERT(self.nops() == 2 || self.nops() == 3);
409 const matrix &self_matrix = ex_to<matrix>(self.op(0));
411 if (self.nops() == 2)
412 return indexed(self_matrix.mul(other), self.op(1));
413 else // self.nops() == 3
414 return indexed(self_matrix.mul(other), self.op(1), self.op(2));
417 /** Contraction of an indexed matrix with something else. */
418 bool matrix::contract_with(exvector::iterator self, exvector::iterator other, exvector & v) const
420 GINAC_ASSERT(is_a<indexed>(*self));
421 GINAC_ASSERT(is_a<indexed>(*other));
422 GINAC_ASSERT(self->nops() == 2 || self->nops() == 3);
423 GINAC_ASSERT(is_a<matrix>(self->op(0)));
425 // Only contract with other matrices
426 if (!is_a<matrix>(other->op(0)))
429 GINAC_ASSERT(other->nops() == 2 || other->nops() == 3);
431 const matrix &self_matrix = ex_to<matrix>(self->op(0));
432 const matrix &other_matrix = ex_to<matrix>(other->op(0));
434 if (self->nops() == 2) {
436 if (other->nops() == 2) { // vector * vector (scalar product)
438 if (self_matrix.col == 1) {
439 if (other_matrix.col == 1) {
440 // Column vector * column vector, transpose first vector
441 *self = self_matrix.transpose().mul(other_matrix)(0, 0);
443 // Column vector * row vector, swap factors
444 *self = other_matrix.mul(self_matrix)(0, 0);
447 if (other_matrix.col == 1) {
448 // Row vector * column vector, perfect
449 *self = self_matrix.mul(other_matrix)(0, 0);
451 // Row vector * row vector, transpose second vector
452 *self = self_matrix.mul(other_matrix.transpose())(0, 0);
458 } else { // vector * matrix
460 // B_i * A_ij = (B*A)_j (B is row vector)
461 if (is_dummy_pair(self->op(1), other->op(1))) {
462 if (self_matrix.row == 1)
463 *self = indexed(self_matrix.mul(other_matrix), other->op(2));
465 *self = indexed(self_matrix.transpose().mul(other_matrix), other->op(2));
470 // B_j * A_ij = (A*B)_i (B is column vector)
471 if (is_dummy_pair(self->op(1), other->op(2))) {
472 if (self_matrix.col == 1)
473 *self = indexed(other_matrix.mul(self_matrix), other->op(1));
475 *self = indexed(other_matrix.mul(self_matrix.transpose()), other->op(1));
481 } else if (other->nops() == 3) { // matrix * matrix
483 // A_ij * B_jk = (A*B)_ik
484 if (is_dummy_pair(self->op(2), other->op(1))) {
485 *self = indexed(self_matrix.mul(other_matrix), self->op(1), other->op(2));
490 // A_ij * B_kj = (A*Btrans)_ik
491 if (is_dummy_pair(self->op(2), other->op(2))) {
492 *self = indexed(self_matrix.mul(other_matrix.transpose()), self->op(1), other->op(1));
497 // A_ji * B_jk = (Atrans*B)_ik
498 if (is_dummy_pair(self->op(1), other->op(1))) {
499 *self = indexed(self_matrix.transpose().mul(other_matrix), self->op(2), other->op(2));
504 // A_ji * B_kj = (B*A)_ki
505 if (is_dummy_pair(self->op(1), other->op(2))) {
506 *self = indexed(other_matrix.mul(self_matrix), other->op(1), self->op(2));
517 // non-virtual functions in this class
524 * @exception logic_error (incompatible matrices) */
525 matrix matrix::add(const matrix & other) const
527 if (col != other.col || row != other.row)
528 throw std::logic_error("matrix::add(): incompatible matrices");
530 exvector sum(this->m);
531 exvector::iterator i = sum.begin(), end = sum.end();
532 exvector::const_iterator ci = other.m.begin();
536 return matrix(row,col,sum);
540 /** Difference of matrices.
542 * @exception logic_error (incompatible matrices) */
543 matrix matrix::sub(const matrix & other) const
545 if (col != other.col || row != other.row)
546 throw std::logic_error("matrix::sub(): incompatible matrices");
548 exvector dif(this->m);
549 exvector::iterator i = dif.begin(), end = dif.end();
550 exvector::const_iterator ci = other.m.begin();
554 return matrix(row,col,dif);
558 /** Product of matrices.
560 * @exception logic_error (incompatible matrices) */
561 matrix matrix::mul(const matrix & other) const
563 if (this->cols() != other.rows())
564 throw std::logic_error("matrix::mul(): incompatible matrices");
566 exvector prod(this->rows()*other.cols());
568 for (unsigned r1=0; r1<this->rows(); ++r1) {
569 for (unsigned c=0; c<this->cols(); ++c) {
570 if (m[r1*col+c].is_zero())
572 for (unsigned r2=0; r2<other.cols(); ++r2)
573 prod[r1*other.col+r2] += (m[r1*col+c] * other.m[c*other.col+r2]).expand();
576 return matrix(row, other.col, prod);
580 /** Product of matrix and scalar. */
581 matrix matrix::mul(const numeric & other) const
583 exvector prod(row * col);
585 for (unsigned r=0; r<row; ++r)
586 for (unsigned c=0; c<col; ++c)
587 prod[r*col+c] = m[r*col+c] * other;
589 return matrix(row, col, prod);
593 /** Product of matrix and scalar expression. */
594 matrix matrix::mul_scalar(const ex & other) const
596 if (other.return_type() != return_types::commutative)
597 throw std::runtime_error("matrix::mul_scalar(): non-commutative scalar");
599 exvector prod(row * col);
601 for (unsigned r=0; r<row; ++r)
602 for (unsigned c=0; c<col; ++c)
603 prod[r*col+c] = m[r*col+c] * other;
605 return matrix(row, col, prod);
609 /** Power of a matrix. Currently handles integer exponents only. */
610 matrix matrix::pow(const ex & expn) const
613 throw (std::logic_error("matrix::pow(): matrix not square"));
615 if (is_exactly_a<numeric>(expn)) {
616 // Integer cases are computed by successive multiplication, using the
617 // obvious shortcut of storing temporaries, like A^4 == (A*A)*(A*A).
618 if (expn.info(info_flags::integer)) {
619 numeric b = ex_to<numeric>(expn);
621 if (expn.info(info_flags::negative)) {
628 for (unsigned r=0; r<row; ++r)
632 // This loop computes the representation of b in base 2 from right
633 // to left and multiplies the factors whenever needed. Note
634 // that this is not entirely optimal but close to optimal and
635 // "better" algorithms are much harder to implement. (See Knuth,
636 // TAoCP2, section "Evaluation of Powers" for a good discussion.)
642 b /= _num2; // still integer.
648 throw (std::runtime_error("matrix::pow(): don't know how to handle exponent"));
652 /** operator() to access elements for reading.
654 * @param ro row of element
655 * @param co column of element
656 * @exception range_error (index out of range) */
657 const ex & matrix::operator() (unsigned ro, unsigned co) const
659 if (ro>=row || co>=col)
660 throw (std::range_error("matrix::operator(): index out of range"));
666 /** operator() to access elements for writing.
668 * @param ro row of element
669 * @param co column of element
670 * @exception range_error (index out of range) */
671 ex & matrix::operator() (unsigned ro, unsigned co)
673 if (ro>=row || co>=col)
674 throw (std::range_error("matrix::operator(): index out of range"));
676 ensure_if_modifiable();
681 /** Transposed of an m x n matrix, producing a new n x m matrix object that
682 * represents the transposed. */
683 matrix matrix::transpose(void) const
685 exvector trans(this->cols()*this->rows());
687 for (unsigned r=0; r<this->cols(); ++r)
688 for (unsigned c=0; c<this->rows(); ++c)
689 trans[r*this->rows()+c] = m[c*this->cols()+r];
691 return matrix(this->cols(),this->rows(),trans);
694 /** Determinant of square matrix. This routine doesn't actually calculate the
695 * determinant, it only implements some heuristics about which algorithm to
696 * run. If all the elements of the matrix are elements of an integral domain
697 * the determinant is also in that integral domain and the result is expanded
698 * only. If one or more elements are from a quotient field the determinant is
699 * usually also in that quotient field and the result is normalized before it
700 * is returned. This implies that the determinant of the symbolic 2x2 matrix
701 * [[a/(a-b),1],[b/(a-b),1]] is returned as unity. (In this respect, it
702 * behaves like MapleV and unlike Mathematica.)
704 * @param algo allows to chose an algorithm
705 * @return the determinant as a new expression
706 * @exception logic_error (matrix not square)
707 * @see determinant_algo */
708 ex matrix::determinant(unsigned algo) const
711 throw (std::logic_error("matrix::determinant(): matrix not square"));
712 GINAC_ASSERT(row*col==m.capacity());
714 // Gather some statistical information about this matrix:
715 bool numeric_flag = true;
716 bool normal_flag = false;
717 unsigned sparse_count = 0; // counts non-zero elements
718 exvector::const_iterator r = m.begin(), rend = m.end();
720 lst srl; // symbol replacement list
721 ex rtest = r->to_rational(srl);
722 if (!rtest.is_zero())
724 if (!rtest.info(info_flags::numeric))
725 numeric_flag = false;
726 if (!rtest.info(info_flags::crational_polynomial) &&
727 rtest.info(info_flags::rational_function))
732 // Here is the heuristics in case this routine has to decide:
733 if (algo == determinant_algo::automatic) {
734 // Minor expansion is generally a good guess:
735 algo = determinant_algo::laplace;
736 // Does anybody know when a matrix is really sparse?
737 // Maybe <~row/2.236 nonzero elements average in a row?
738 if (row>3 && 5*sparse_count<=row*col)
739 algo = determinant_algo::bareiss;
740 // Purely numeric matrix can be handled by Gauss elimination.
741 // This overrides any prior decisions.
743 algo = determinant_algo::gauss;
746 // Trap the trivial case here, since some algorithms don't like it
748 // for consistency with non-trivial determinants...
750 return m[0].normal();
752 return m[0].expand();
755 // Compute the determinant
757 case determinant_algo::gauss: {
760 int sign = tmp.gauss_elimination(true);
761 for (unsigned d=0; d<row; ++d)
762 det *= tmp.m[d*col+d];
764 return (sign*det).normal();
766 return (sign*det).normal().expand();
768 case determinant_algo::bareiss: {
771 sign = tmp.fraction_free_elimination(true);
773 return (sign*tmp.m[row*col-1]).normal();
775 return (sign*tmp.m[row*col-1]).expand();
777 case determinant_algo::divfree: {
780 sign = tmp.division_free_elimination(true);
783 ex det = tmp.m[row*col-1];
784 // factor out accumulated bogus slag
785 for (unsigned d=0; d<row-2; ++d)
786 for (unsigned j=0; j<row-d-2; ++j)
787 det = (det/tmp.m[d*col+d]).normal();
790 case determinant_algo::laplace:
792 // This is the minor expansion scheme. We always develop such
793 // that the smallest minors (i.e, the trivial 1x1 ones) are on the
794 // rightmost column. For this to be efficient, empirical tests
795 // have shown that the emptiest columns (i.e. the ones with most
796 // zeros) should be the ones on the right hand side -- although
797 // this might seem counter-intuitive (and in contradiction to some
798 // literature like the FORM manual). Please go ahead and test it
799 // if you don't believe me! Therefore we presort the columns of
801 typedef std::pair<unsigned,unsigned> uintpair;
802 std::vector<uintpair> c_zeros; // number of zeros in column
803 for (unsigned c=0; c<col; ++c) {
805 for (unsigned r=0; r<row; ++r)
806 if (m[r*col+c].is_zero())
808 c_zeros.push_back(uintpair(acc,c));
810 std::sort(c_zeros.begin(),c_zeros.end());
811 std::vector<unsigned> pre_sort;
812 for (std::vector<uintpair>::const_iterator i=c_zeros.begin(); i!=c_zeros.end(); ++i)
813 pre_sort.push_back(i->second);
814 std::vector<unsigned> pre_sort_test(pre_sort); // permutation_sign() modifies the vector so we make a copy here
815 int sign = permutation_sign(pre_sort_test.begin(), pre_sort_test.end());
816 exvector result(row*col); // represents sorted matrix
818 for (std::vector<unsigned>::const_iterator i=pre_sort.begin();
821 for (unsigned r=0; r<row; ++r)
822 result[r*col+c] = m[r*col+(*i)];
826 return (sign*matrix(row,col,result).determinant_minor()).normal();
828 return sign*matrix(row,col,result).determinant_minor();
834 /** Trace of a matrix. The result is normalized if it is in some quotient
835 * field and expanded only otherwise. This implies that the trace of the
836 * symbolic 2x2 matrix [[a/(a-b),x],[y,b/(b-a)]] is recognized to be unity.
838 * @return the sum of diagonal elements
839 * @exception logic_error (matrix not square) */
840 ex matrix::trace(void) const
843 throw (std::logic_error("matrix::trace(): matrix not square"));
846 for (unsigned r=0; r<col; ++r)
849 if (tr.info(info_flags::rational_function) &&
850 !tr.info(info_flags::crational_polynomial))
857 /** Characteristic Polynomial. Following mathematica notation the
858 * characteristic polynomial of a matrix M is defined as the determiant of
859 * (M - lambda * 1) where 1 stands for the unit matrix of the same dimension
860 * as M. Note that some CASs define it with a sign inside the determinant
861 * which gives rise to an overall sign if the dimension is odd. This method
862 * returns the characteristic polynomial collected in powers of lambda as a
865 * @return characteristic polynomial as new expression
866 * @exception logic_error (matrix not square)
867 * @see matrix::determinant() */
868 ex matrix::charpoly(const symbol & lambda) const
871 throw (std::logic_error("matrix::charpoly(): matrix not square"));
873 bool numeric_flag = true;
874 exvector::const_iterator r = m.begin(), rend = m.end();
875 while (r!=rend && numeric_flag==true) {
876 if (!r->info(info_flags::numeric))
877 numeric_flag = false;
881 // The pure numeric case is traditionally rather common. Hence, it is
882 // trapped and we use Leverrier's algorithm which goes as row^3 for
883 // every coefficient. The expensive part is the matrix multiplication.
888 ex poly = power(lambda,row)-c*power(lambda,row-1);
889 for (unsigned i=1; i<row; ++i) {
890 for (unsigned j=0; j<row; ++j)
893 c = B.trace() / ex(i+1);
894 poly -= c*power(lambda,row-i-1);
904 for (unsigned r=0; r<col; ++r)
905 M.m[r*col+r] -= lambda;
907 return M.determinant().collect(lambda);
912 /** Inverse of this matrix.
914 * @return the inverted matrix
915 * @exception logic_error (matrix not square)
916 * @exception runtime_error (singular matrix) */
917 matrix matrix::inverse(void) const
920 throw (std::logic_error("matrix::inverse(): matrix not square"));
922 // This routine actually doesn't do anything fancy at all. We compute the
923 // inverse of the matrix A by solving the system A * A^{-1} == Id.
925 // First populate the identity matrix supposed to become the right hand side.
926 matrix identity(row,col);
927 for (unsigned i=0; i<row; ++i)
928 identity(i,i) = _ex1;
930 // Populate a dummy matrix of variables, just because of compatibility with
931 // matrix::solve() which wants this (for compatibility with under-determined
932 // systems of equations).
933 matrix vars(row,col);
934 for (unsigned r=0; r<row; ++r)
935 for (unsigned c=0; c<col; ++c)
936 vars(r,c) = symbol();
940 sol = this->solve(vars,identity);
941 } catch (const std::runtime_error & e) {
942 if (e.what()==std::string("matrix::solve(): inconsistent linear system"))
943 throw (std::runtime_error("matrix::inverse(): singular matrix"));
951 /** Solve a linear system consisting of a m x n matrix and a m x p right hand
952 * side by applying an elimination scheme to the augmented matrix.
954 * @param vars n x p matrix, all elements must be symbols
955 * @param rhs m x p matrix
956 * @return n x p solution matrix
957 * @exception logic_error (incompatible matrices)
958 * @exception invalid_argument (1st argument must be matrix of symbols)
959 * @exception runtime_error (inconsistent linear system)
961 matrix matrix::solve(const matrix & vars,
965 const unsigned m = this->rows();
966 const unsigned n = this->cols();
967 const unsigned p = rhs.cols();
970 if ((rhs.rows() != m) || (vars.rows() != n) || (vars.col != p))
971 throw (std::logic_error("matrix::solve(): incompatible matrices"));
972 for (unsigned ro=0; ro<n; ++ro)
973 for (unsigned co=0; co<p; ++co)
974 if (!vars(ro,co).info(info_flags::symbol))
975 throw (std::invalid_argument("matrix::solve(): 1st argument must be matrix of symbols"));
977 // build the augmented matrix of *this with rhs attached to the right
979 for (unsigned r=0; r<m; ++r) {
980 for (unsigned c=0; c<n; ++c)
981 aug.m[r*(n+p)+c] = this->m[r*n+c];
982 for (unsigned c=0; c<p; ++c)
983 aug.m[r*(n+p)+c+n] = rhs.m[r*p+c];
986 // Gather some statistical information about the augmented matrix:
987 bool numeric_flag = true;
988 exvector::const_iterator r = aug.m.begin(), rend = aug.m.end();
989 while (r!=rend && numeric_flag==true) {
990 if (!r->info(info_flags::numeric))
991 numeric_flag = false;
995 // Here is the heuristics in case this routine has to decide:
996 if (algo == solve_algo::automatic) {
997 // Bareiss (fraction-free) elimination is generally a good guess:
998 algo = solve_algo::bareiss;
999 // For m<3, Bareiss elimination is equivalent to division free
1000 // elimination but has more logistic overhead
1002 algo = solve_algo::divfree;
1003 // This overrides any prior decisions.
1005 algo = solve_algo::gauss;
1008 // Eliminate the augmented matrix:
1010 case solve_algo::gauss:
1011 aug.gauss_elimination();
1013 case solve_algo::divfree:
1014 aug.division_free_elimination();
1016 case solve_algo::bareiss:
1018 aug.fraction_free_elimination();
1021 // assemble the solution matrix:
1023 for (unsigned co=0; co<p; ++co) {
1024 unsigned last_assigned_sol = n+1;
1025 for (int r=m-1; r>=0; --r) {
1026 unsigned fnz = 1; // first non-zero in row
1027 while ((fnz<=n) && (aug.m[r*(n+p)+(fnz-1)].is_zero()))
1030 // row consists only of zeros, corresponding rhs must be 0, too
1031 if (!aug.m[r*(n+p)+n+co].is_zero()) {
1032 throw (std::runtime_error("matrix::solve(): inconsistent linear system"));
1035 // assign solutions for vars between fnz+1 and
1036 // last_assigned_sol-1: free parameters
1037 for (unsigned c=fnz; c<last_assigned_sol-1; ++c)
1038 sol(c,co) = vars.m[c*p+co];
1039 ex e = aug.m[r*(n+p)+n+co];
1040 for (unsigned c=fnz; c<n; ++c)
1041 e -= aug.m[r*(n+p)+c]*sol.m[c*p+co];
1042 sol(fnz-1,co) = (e/(aug.m[r*(n+p)+(fnz-1)])).normal();
1043 last_assigned_sol = fnz;
1046 // assign solutions for vars between 1 and
1047 // last_assigned_sol-1: free parameters
1048 for (unsigned ro=0; ro<last_assigned_sol-1; ++ro)
1049 sol(ro,co) = vars(ro,co);
1058 /** Recursive determinant for small matrices having at least one symbolic
1059 * entry. The basic algorithm, known as Laplace-expansion, is enhanced by
1060 * some bookkeeping to avoid calculation of the same submatrices ("minors")
1061 * more than once. According to W.M.Gentleman and S.C.Johnson this algorithm
1062 * is better than elimination schemes for matrices of sparse multivariate
1063 * polynomials and also for matrices of dense univariate polynomials if the
1064 * matrix' dimesion is larger than 7.
1066 * @return the determinant as a new expression (in expanded form)
1067 * @see matrix::determinant() */
1068 ex matrix::determinant_minor(void) const
1070 // for small matrices the algorithm does not make any sense:
1071 const unsigned n = this->cols();
1073 return m[0].expand();
1075 return (m[0]*m[3]-m[2]*m[1]).expand();
1077 return (m[0]*m[4]*m[8]-m[0]*m[5]*m[7]-
1078 m[1]*m[3]*m[8]+m[2]*m[3]*m[7]+
1079 m[1]*m[5]*m[6]-m[2]*m[4]*m[6]).expand();
1081 // This algorithm can best be understood by looking at a naive
1082 // implementation of Laplace-expansion, like this one:
1084 // matrix minorM(this->rows()-1,this->cols()-1);
1085 // for (unsigned r1=0; r1<this->rows(); ++r1) {
1086 // // shortcut if element(r1,0) vanishes
1087 // if (m[r1*col].is_zero())
1089 // // assemble the minor matrix
1090 // for (unsigned r=0; r<minorM.rows(); ++r) {
1091 // for (unsigned c=0; c<minorM.cols(); ++c) {
1093 // minorM(r,c) = m[r*col+c+1];
1095 // minorM(r,c) = m[(r+1)*col+c+1];
1098 // // recurse down and care for sign:
1100 // det -= m[r1*col] * minorM.determinant_minor();
1102 // det += m[r1*col] * minorM.determinant_minor();
1104 // return det.expand();
1105 // What happens is that while proceeding down many of the minors are
1106 // computed more than once. In particular, there are binomial(n,k)
1107 // kxk minors and each one is computed factorial(n-k) times. Therefore
1108 // it is reasonable to store the results of the minors. We proceed from
1109 // right to left. At each column c we only need to retrieve the minors
1110 // calculated in step c-1. We therefore only have to store at most
1111 // 2*binomial(n,n/2) minors.
1113 // Unique flipper counter for partitioning into minors
1114 std::vector<unsigned> Pkey;
1116 // key for minor determinant (a subpartition of Pkey)
1117 std::vector<unsigned> Mkey;
1119 // we store our subminors in maps, keys being the rows they arise from
1120 typedef std::map<std::vector<unsigned>,class ex> Rmap;
1121 typedef std::map<std::vector<unsigned>,class ex>::value_type Rmap_value;
1125 // initialize A with last column:
1126 for (unsigned r=0; r<n; ++r) {
1127 Pkey.erase(Pkey.begin(),Pkey.end());
1129 A.insert(Rmap_value(Pkey,m[n*(r+1)-1]));
1131 // proceed from right to left through matrix
1132 for (int c=n-2; c>=0; --c) {
1133 Pkey.erase(Pkey.begin(),Pkey.end()); // don't change capacity
1134 Mkey.erase(Mkey.begin(),Mkey.end());
1135 for (unsigned i=0; i<n-c; ++i)
1137 unsigned fc = 0; // controls logic for our strange flipper counter
1140 for (unsigned r=0; r<n-c; ++r) {
1141 // maybe there is nothing to do?
1142 if (m[Pkey[r]*n+c].is_zero())
1144 // create the sorted key for all possible minors
1145 Mkey.erase(Mkey.begin(),Mkey.end());
1146 for (unsigned i=0; i<n-c; ++i)
1148 Mkey.push_back(Pkey[i]);
1149 // Fetch the minors and compute the new determinant
1151 det -= m[Pkey[r]*n+c]*A[Mkey];
1153 det += m[Pkey[r]*n+c]*A[Mkey];
1155 // prevent build-up of deep nesting of expressions saves time:
1157 // store the new determinant at its place in B:
1159 B.insert(Rmap_value(Pkey,det));
1160 // increment our strange flipper counter
1161 for (fc=n-c; fc>0; --fc) {
1163 if (Pkey[fc-1]<fc+c)
1167 for (unsigned j=fc; j<n-c; ++j)
1168 Pkey[j] = Pkey[j-1]+1;
1170 // next column, so change the role of A and B:
1179 /** Perform the steps of an ordinary Gaussian elimination to bring the m x n
1180 * matrix into an upper echelon form. The algorithm is ok for matrices
1181 * with numeric coefficients but quite unsuited for symbolic matrices.
1183 * @param det may be set to true to save a lot of space if one is only
1184 * interested in the diagonal elements (i.e. for calculating determinants).
1185 * The others are set to zero in this case.
1186 * @return sign is 1 if an even number of rows was swapped, -1 if an odd
1187 * number of rows was swapped and 0 if the matrix is singular. */
1188 int matrix::gauss_elimination(const bool det)
1190 ensure_if_modifiable();
1191 const unsigned m = this->rows();
1192 const unsigned n = this->cols();
1193 GINAC_ASSERT(!det || n==m);
1197 for (unsigned r1=0; (r1<n-1)&&(r0<m-1); ++r1) {
1198 int indx = pivot(r0, r1, true);
1202 return 0; // leaves *this in a messy state
1207 for (unsigned r2=r0+1; r2<m; ++r2) {
1208 if (!this->m[r2*n+r1].is_zero()) {
1209 // yes, there is something to do in this row
1210 ex piv = this->m[r2*n+r1] / this->m[r0*n+r1];
1211 for (unsigned c=r1+1; c<n; ++c) {
1212 this->m[r2*n+c] -= piv * this->m[r0*n+c];
1213 if (!this->m[r2*n+c].info(info_flags::numeric))
1214 this->m[r2*n+c] = this->m[r2*n+c].normal();
1217 // fill up left hand side with zeros
1218 for (unsigned c=0; c<=r1; ++c)
1219 this->m[r2*n+c] = _ex0;
1222 // save space by deleting no longer needed elements
1223 for (unsigned c=r0+1; c<n; ++c)
1224 this->m[r0*n+c] = _ex0;
1234 /** Perform the steps of division free elimination to bring the m x n matrix
1235 * into an upper echelon form.
1237 * @param det may be set to true to save a lot of space if one is only
1238 * interested in the diagonal elements (i.e. for calculating determinants).
1239 * The others are set to zero in this case.
1240 * @return sign is 1 if an even number of rows was swapped, -1 if an odd
1241 * number of rows was swapped and 0 if the matrix is singular. */
1242 int matrix::division_free_elimination(const bool det)
1244 ensure_if_modifiable();
1245 const unsigned m = this->rows();
1246 const unsigned n = this->cols();
1247 GINAC_ASSERT(!det || n==m);
1251 for (unsigned r1=0; (r1<n-1)&&(r0<m-1); ++r1) {
1252 int indx = pivot(r0, r1, true);
1256 return 0; // leaves *this in a messy state
1261 for (unsigned r2=r0+1; r2<m; ++r2) {
1262 for (unsigned c=r1+1; c<n; ++c)
1263 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();
1264 // fill up left hand side with zeros
1265 for (unsigned c=0; c<=r1; ++c)
1266 this->m[r2*n+c] = _ex0;
1269 // save space by deleting no longer needed elements
1270 for (unsigned c=r0+1; c<n; ++c)
1271 this->m[r0*n+c] = _ex0;
1281 /** Perform the steps of Bareiss' one-step fraction free elimination to bring
1282 * the matrix into an upper echelon form. Fraction free elimination means
1283 * that divide is used straightforwardly, without computing GCDs first. This
1284 * is possible, since we know the divisor at each step.
1286 * @param det may be set to true to save a lot of space if one is only
1287 * interested in the last element (i.e. for calculating determinants). The
1288 * others are set to zero in this case.
1289 * @return sign is 1 if an even number of rows was swapped, -1 if an odd
1290 * number of rows was swapped and 0 if the matrix is singular. */
1291 int matrix::fraction_free_elimination(const bool det)
1294 // (single-step fraction free elimination scheme, already known to Jordan)
1296 // Usual division-free elimination sets m[0](r,c) = m(r,c) and then sets
1297 // m[k+1](r,c) = m[k](k,k) * m[k](r,c) - m[k](r,k) * m[k](k,c).
1299 // Bareiss (fraction-free) elimination in addition divides that element
1300 // by m[k-1](k-1,k-1) for k>1, where it can be shown by means of the
1301 // Sylvester identity that this really divides m[k+1](r,c).
1303 // We also allow rational functions where the original prove still holds.
1304 // However, we must care for numerator and denominator separately and
1305 // "manually" work in the integral domains because of subtle cancellations
1306 // (see below). This blows up the bookkeeping a bit and the formula has
1307 // to be modified to expand like this (N{x} stands for numerator of x,
1308 // D{x} for denominator of x):
1309 // 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)}
1310 // -N{m[k](r,k)}*N{m[k](k,c)}*D{m[k](k,k)}*D{m[k](r,c)}
1311 // 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)}
1312 // where for k>1 we now divide N{m[k+1](r,c)} by
1313 // N{m[k-1](k-1,k-1)}
1314 // and D{m[k+1](r,c)} by
1315 // D{m[k-1](k-1,k-1)}.
1317 ensure_if_modifiable();
1318 const unsigned m = this->rows();
1319 const unsigned n = this->cols();
1320 GINAC_ASSERT(!det || n==m);
1329 // We populate temporary matrices to subsequently operate on. There is
1330 // one holding numerators and another holding denominators of entries.
1331 // This is a must since the evaluator (or even earlier mul's constructor)
1332 // might cancel some trivial element which causes divide() to fail. The
1333 // elements are normalized first (yes, even though this algorithm doesn't
1334 // need GCDs) since the elements of *this might be unnormalized, which
1335 // makes things more complicated than they need to be.
1336 matrix tmp_n(*this);
1337 matrix tmp_d(m,n); // for denominators, if needed
1338 lst srl; // symbol replacement list
1339 exvector::const_iterator cit = this->m.begin(), citend = this->m.end();
1340 exvector::iterator tmp_n_it = tmp_n.m.begin(), tmp_d_it = tmp_d.m.begin();
1341 while (cit != citend) {
1342 ex nd = cit->normal().to_rational(srl).numer_denom();
1344 *tmp_n_it++ = nd.op(0);
1345 *tmp_d_it++ = nd.op(1);
1349 for (unsigned r1=0; (r1<n-1)&&(r0<m-1); ++r1) {
1350 int indx = tmp_n.pivot(r0, r1, true);
1359 // tmp_n's rows r0 and indx were swapped, do the same in tmp_d:
1360 for (unsigned c=r1; c<n; ++c)
1361 tmp_d.m[n*indx+c].swap(tmp_d.m[n*r0+c]);
1363 for (unsigned r2=r0+1; r2<m; ++r2) {
1364 for (unsigned c=r1+1; c<n; ++c) {
1365 dividend_n = (tmp_n.m[r0*n+r1]*tmp_n.m[r2*n+c]*
1366 tmp_d.m[r2*n+r1]*tmp_d.m[r0*n+c]
1367 -tmp_n.m[r2*n+r1]*tmp_n.m[r0*n+c]*
1368 tmp_d.m[r0*n+r1]*tmp_d.m[r2*n+c]).expand();
1369 dividend_d = (tmp_d.m[r2*n+r1]*tmp_d.m[r0*n+c]*
1370 tmp_d.m[r0*n+r1]*tmp_d.m[r2*n+c]).expand();
1371 bool check = divide(dividend_n, divisor_n,
1372 tmp_n.m[r2*n+c], true);
1373 check &= divide(dividend_d, divisor_d,
1374 tmp_d.m[r2*n+c], true);
1375 GINAC_ASSERT(check);
1377 // fill up left hand side with zeros
1378 for (unsigned c=0; c<=r1; ++c)
1379 tmp_n.m[r2*n+c] = _ex0;
1381 if ((r1<n-1)&&(r0<m-1)) {
1382 // compute next iteration's divisor
1383 divisor_n = tmp_n.m[r0*n+r1].expand();
1384 divisor_d = tmp_d.m[r0*n+r1].expand();
1386 // save space by deleting no longer needed elements
1387 for (unsigned c=0; c<n; ++c) {
1388 tmp_n.m[r0*n+c] = _ex0;
1389 tmp_d.m[r0*n+c] = _ex1;
1396 // repopulate *this matrix:
1397 exvector::iterator it = this->m.begin(), itend = this->m.end();
1398 tmp_n_it = tmp_n.m.begin();
1399 tmp_d_it = tmp_d.m.begin();
1401 *it++ = ((*tmp_n_it++)/(*tmp_d_it++)).subs(srl);
1407 /** Partial pivoting method for matrix elimination schemes.
1408 * Usual pivoting (symbolic==false) returns the index to the element with the
1409 * largest absolute value in column ro and swaps the current row with the one
1410 * where the element was found. With (symbolic==true) it does the same thing
1411 * with the first non-zero element.
1413 * @param ro is the row from where to begin
1414 * @param co is the column to be inspected
1415 * @param symbolic signal if we want the first non-zero element to be pivoted
1416 * (true) or the one with the largest absolute value (false).
1417 * @return 0 if no interchange occured, -1 if all are zero (usually signaling
1418 * a degeneracy) and positive integer k means that rows ro and k were swapped.
1420 int matrix::pivot(unsigned ro, unsigned co, bool symbolic)
1424 // search first non-zero element in column co beginning at row ro
1425 while ((k<row) && (this->m[k*col+co].expand().is_zero()))
1428 // search largest element in column co beginning at row ro
1429 GINAC_ASSERT(is_exactly_a<numeric>(this->m[k*col+co]));
1430 unsigned kmax = k+1;
1431 numeric mmax = abs(ex_to<numeric>(m[kmax*col+co]));
1433 GINAC_ASSERT(is_exactly_a<numeric>(this->m[kmax*col+co]));
1434 numeric tmp = ex_to<numeric>(this->m[kmax*col+co]);
1435 if (abs(tmp) > mmax) {
1441 if (!mmax.is_zero())
1445 // all elements in column co below row ro vanish
1448 // matrix needs no pivoting
1450 // matrix needs pivoting, so swap rows k and ro
1451 ensure_if_modifiable();
1452 for (unsigned c=0; c<col; ++c)
1453 this->m[k*col+c].swap(this->m[ro*col+c]);
1458 ex lst_to_matrix(const lst & l)
1460 lst::const_iterator itr, itc;
1462 // Find number of rows and columns
1463 size_t rows = l.nops(), cols = 0;
1464 for (itr = l.begin(); itr != l.end(); ++itr) {
1465 if (!is_a<lst>(*itr))
1466 throw (std::invalid_argument("lst_to_matrix: argument must be a list of lists"));
1467 if (itr->nops() > cols)
1471 // Allocate and fill matrix
1472 matrix &M = *new matrix(rows, cols);
1473 M.setflag(status_flags::dynallocated);
1476 for (itr = l.begin(), i = 0; itr != l.end(); ++itr, ++i) {
1478 for (itc = ex_to<lst>(*itr).begin(), j = 0; itc != ex_to<lst>(*itr).end(); ++itc, ++j)
1485 ex diag_matrix(const lst & l)
1487 lst::const_iterator it;
1488 size_t dim = l.nops();
1490 // Allocate and fill matrix
1491 matrix &M = *new matrix(dim, dim);
1492 M.setflag(status_flags::dynallocated);
1495 for (it = l.begin(), i = 0; it != l.end(); ++it, ++i)
1501 ex unit_matrix(unsigned r, unsigned c)
1503 matrix &Id = *new matrix(r, c);
1504 Id.setflag(status_flags::dynallocated);
1505 for (unsigned i=0; i<r && i<c; i++)
1511 ex symbolic_matrix(unsigned r, unsigned c, const std::string & base_name, const std::string & tex_base_name)
1513 matrix &M = *new matrix(r, c);
1514 M.setflag(status_flags::dynallocated | status_flags::evaluated);
1516 bool long_format = (r > 10 || c > 10);
1517 bool single_row = (r == 1 || c == 1);
1519 for (unsigned i=0; i<r; i++) {
1520 for (unsigned j=0; j<c; j++) {
1521 std::ostringstream s1, s2;
1523 s2 << tex_base_name << "_{";
1534 s1 << '_' << i << '_' << j;
1535 s2 << i << ';' << j << "}";
1538 s2 << i << j << '}';
1541 M(i, j) = symbol(s1.str(), s2.str());
1548 } // namespace GiNaC