问题
i am trying to create gauss-legendre code according to the following algorithm:
for n points
That is,it is created a 2n equation system (if we demand to be accurate for polynominals of order 2n-1 ,
ti are roots of the legendre polynominals of order n.The legendre poynominals are given :
and wi :
My code is :
#include <iostream>
#include <cstdio>
#include <cstdlib>
#include <iomanip>
#include <cmath>
using namespace std;
const double pi=3.14;
//my function with limits (-1,1)
double f(double x){
double y;
y=(pi/4.0)*(log((pi*(x+1.0))/4.0 +1.0));
return y;
}
double legendre (int n){
    double *L,*w,*t;
    double x,sum1,sum2,result;
    L=new double [n];
    w=new double [n];
    t=new double [n];
        while(n<10){
         L[0]=1;
         L[1]=x;
        //legendre coef
        for (int i=1;i<=10;i++){
        L[i+1]=((2.0*i+1.0)*x*L[i] - i*L[i-1])/(i+1.0);
        }
        //weights w
        w=0;
        for (int i=1;i<=10;i++){
        w[i]+=(2.0*(1.0-x*x))/(i*i*(L[i-1]*L[i-1]));
        }
        //sums  w*t
        for (int i=1;i<=10;i++){
            sum1=0.0; //for k=1,3,5,2n-1
            for (int k=1;k<=2*n-1;k+=2){
                sum1+=w[i]*(pow(t[i],k));
            }
                sum1=0;
                sum2=0.0;//for k=0,2,4,2n-2
                for(int k=0;k<=2*n-2;k+=2){
                    sum2+=w[i]*(pow(t[i],k));
                }
                sum2=2.0/n;
        }
    }
    result=w*f(*t);
    return result;
}
int main()
{
    double eps=1e-8;//accuracy
    double exact=0.8565899396;//exact solution for the integral
    double error=1.0;
    double result;
    int n=1;//initial point
    while (fabs(result-exact)>eps) {
        result=legendre(n);
        cout <<"\nFor n = "<<n<<",error = "<<fabs(error-exact)<<",value = "<<result;
    n++;
    }
    return 0;
}
My problems are:
1) The compiler gives me :error: invalid operands of types ‘double*’ and ‘double’ to binary ‘operator*’ --> at result=w*f(*t);
2) I am not sure if i have done the whole thing right.I mean ,if i combined all the things together and if i implemented right the algorithm.
回答1:
I do not know the algorithm but your code is wrong.
First :
        while(n<10)
        {
         L[0]=1;
         L[1]=x;
        //legendre coef
        for (int i=1;i<=10;i++){
        L[i+1]=((2.0*i+1.0)*x*L[i] - i*L[i-1])/(i+1.0);
        }
You must change the value of n (increment, decrement, etc.) otherwise this is an endless loop.
Second :
//weights w
    w=0;
    for (int i=1;i<=10;i++){
    w[i]+=(2.0*(1.0-x*x))/(i*i*(L[i-1]*L[i-1]));
    }
w is a pointer. If you want to reset it, use memset(w,0,sizeof(double)*n); Do not make it equal to 0.
Last:
result=w*f(*t);
Since you are using the w and t pointers as arrays, you have to provide some sort of index like result=w[ind] * f(t[ind]);. Here you are simply multiplying the value of pointer w, not the value that is pointed by w (the value of w is 0 form your code by the way) with the first value of the double array pointed by t.
Also I could not get any relation between your code and the formulas in the question. So my humble advise is do not use C or C++ for this. If you must, then do not use pointers, because it seems you are not familiar with them. You can easily have std::vector instead of pointers.
回答2:
w is a pointer, and you are trying to multiply it with something... you must use index
w[index] * f(*t)
also *t is the first element of t array. Is that what you mean?
回答3:
Regarding your algorithm, the x (the abscissa values) are supposed to be the zeros of the Legendre polynomial. I didn't see you define them anywhere. They're a bit of a pain to define. I was doing something similar and found this (it's a Matlab file, not a C++ file) that defines N xi and wi values. The algorithm works fine: http://www.mathworks.com/matlabcentral/fileexchange/4540-legendre-gauss-quadrature-weights-and-nodes
回答4:
Both alglib and gsl have implementations of gaussian quadrature. Both are GPL licensed, though.
来源:https://stackoverflow.com/questions/5329848/gauss-legendre-in-c