A question on the preciseness of the double used in MTL. Can anyone with kindness help me?
I am using MTL(Matrix Template Library) for Matrix computing in Dev-C++ 4.9.8.0 and updated the gcc version from 3.2 to 4.4.1 by installing tdm-mingw-1.908.0-4.4.1-2 and the boost version is 1.37.0. The system is Windows xp The sample code is : #include <boost/numeric/mtl/mtl.hpp> #include "boost/random/normal_distribution.hpp" #include <boost/random.hpp> #include <boost/random/uniform_real.hpp> int main(int argc, char* argv[]) { using namespace mtl; using namespace mtl::matrix; const unsigned n= 5; dense2D<double> A(n, n), B(n, n); morton_dense<double, doppled_64_row_mask> C(n, n), D(n, n); dense2D<double> F(n, n), E(n, n); hessian_setup(A, 3.0); hessian_setup(B, 1.0); hessian_setup(C, 2.0); hessian_setup(D, 11.0); B(1,1)=9.9999; .......................... std::cout << "The matrices are: B=\n" << B <<"\n"; std::cout << "The result is " <<1.0/3.0 <<"\n"; ……………………. But the result is shown that B(1,1) is 10, and 1.0/3.0 is 0.33. How can I get the highest precision of the type double and dense2D<double> for scientific computation? The same question occurred in random number generation in boost random library. Can anyone with kindness help me?
From: boost-users-bounces@lists.boost.org [mailto:boost-users-bounces@lists.boost.org] On Behalf Of fmingu Sent: Monday, March 29, 2010 1:12 AM To: boost-users Subject: [Boost-users] A question on the preciseness of the double used in MTL. Can anyone with kindness help me? I am using MTL(Matrix Template Library) for Matrix computing in Dev-C++ 4.9.8.0 and updated the gcc version from 3.2 to 4.4.1 by installing tdm-mingw-1.908.0-4.4.1-2 and the boost version is 1.37.0. The system is Windows xp The sample code is : #include <boost/numeric/mtl/mtl.hpp> #include "boost/random/normal_distribution.hpp" #include <boost/random.hpp> #include <boost/random/uniform_real.hpp> int main(int argc, char* argv[]) { using namespace mtl; using namespace mtl::matrix; const unsigned n= 5; dense2D<double> A(n, n), B(n, n); morton_dense<double, doppled_64_row_mask> C(n, n), D(n, n); dense2D<double> F(n, n), E(n, n); hessian_setup(A, 3.0); hessian_setup(B, 1.0); hessian_setup(C, 2.0); hessian_setup(D, 11.0); B(1,1)=9.9999; .......................... std::cout << "The matrices are: B=\n" << B <<"\n"; std::cout << "The result is " <<1.0/3.0 <<"\n"; . But the result is shown that B(1,1) is 10, and 1.0/3.0 is 0.33.
How can I get the highest precision of the type double and dense2D<double> for scientific computation?
The highest significant precision for the floating point you are using can be got using your_stream.precision(2 + std::numeric_limits<double>::digits * 3010/10000); For the usual 64 bit doubles this is equivalent to cout << setprecision(17) ... This should every digit that might possibly be significant. Using C++0X you should be able to use std::numeric_limits<double>::maxdigits10() This is about 2 digits more than the std::numeric_limits<double>::digits10() function which is the maximum number of digits guaranteed accurate decimal digits (for 64 bit double it is 15 decimal digits). HTH Paul --- Paul A. Bristow Prizet Farmhouse Kendal, UK LA8 8AB +44 1539 561830, mobile +44 7714330204 pbristow@hetp.u-net.com
participants (2)
-
fmingu -
Paul A. Bristow