Subversion Repositories gelsvn

Rev

Rev 2 | Rev 45 | Go to most recent revision | Show entire file | Ignore whitespace | Details | Blame | Last modification | View Log | RSS feed

Rev 2 Rev 12
Line 8... Line 8...
8 
using namespace std;
 8 
using namespace std;
9 
 
 9 
 
10 
 
 10 
 
11 
namespace
 11 
namespace
12 
{
 12 
{
13 
	// During experiments 925 iterations were observed for a hard problem
 13 
  // During experiments 925 iterations were observed for a hard problem
14 
	// Ten times that should be safe.
 14 
  // Ten times that should be safe.
15 
	const int KMAX = 10000;
 15 
  const int KMAX = 10000;
16 
 
 16 
 
17 
	// The threshold below is the smallest that seems to give reliable
 17 
  // The threshold below is the smallest that seems to give reliable
18 
	// solutions.
 18 
  // solutions.
19 
	const float EV_THRESH = 0.0000001;
 19 
  const float EV_THRESH = 0.0000001;
20 
}
 20 
}
21 
 
 21 
 
22 
namespace CGLA
 22 
namespace CGLA
23 
{
 23 
{
24 
 
 24 
 
25 
	template <class MT>
 25 
  template <class MT>
26 
	int power_eigensolution(const MT& Ap, MT& Q, MT& L, int max_sol)
 26 
  int power_eigensolution(const MT& Ap, MT& Q, MT& L, int max_sol)
27 
	{
 27 
  {
- 
 
 28 
    L = MT(0);
- 
 
 29 
 
28 
		typedef typename MT::VectorType VT;
 30 
    typedef typename MT::VectorType VT;
29 
		MT A = Ap;
 31 
    MT A = Ap;
30 
		int n = s_min(MT::get_v_dim(), max_sol);
 32 
    int n = s_min(MT::get_v_dim(), max_sol);
31 
 
 33 
 
32 
		for(int i=0;i<n;++i)
 34 
    for(int i=0;i<n;++i)
33 
			{
 35 
      {
34 
				// Seed the eigenvector estimate
 36 
	// Seed the eigenvector estimate
35 
				VT q(1);
 37 
	VT q(1);
36 
				float l=0,l_old;
 38 
	float l=0,l_old;
37 
 
 39 
 
38 
				// As long as we haven't reached the max iterations and the
 40 
	// As long as we haven't reached the max iterations and the
39 
				// eigenvalue has not converged, do
 41 
	// eigenvalue has not converged, do
40 
				int k=0;
 42 
	int k=0;
41 
				for(; k<2 || k<KMAX && (l-l_old > EV_THRESH * l) ; ++k)
 43 
	for(; k<2 || k<KMAX && (l-l_old > EV_THRESH * l) ; ++k)
42 
					{
 44 
	  {
43 
						// Multiply the eigenvector estimate onto A
 45 
	    // Multiply the eigenvector estimate onto A
44 
						const VT z = A * q;
 46 
	    const VT z = A * q;
45 
 
 47 
 
46 
						// Check that we did not get the null vector.
 48 
	    // Check that we did not get the null vector.
47 
						float z_len = z.length();
 49 
	    float z_len = z.length();
48 
						if(z_len < EV_THRESH)
 50 
	    if(z_len < EV_THRESH)
49 
							return i;
 51 
	      return i;
50 
 
 52 
 
51 
						// Normalize to get the new eigenvector
 53 
	    // Normalize to get the new eigenvector
52 
						q = z/z_len;
 54 
	    q = z/z_len;
53 
 
 55 
 
54 
						// Record the old eigenvalue estimate and get a new estimate.
 56 
	    // Record the old eigenvalue estimate and get a new estimate.
55 
						l_old = l;
 57 
	    l_old = l;
56 
						l = dot(q, A * q);
 58 
	    l = dot(q, A * q);
57 
					}
 59 
	  }
58 
				// If we hit the max iterations, we also don't trust the eigensolution
 60 
	// If we hit the max iterations, we also don't trust the eigensolution
59 
				if(k==KMAX)
 61 
	if(k==KMAX)
60 
					return i;
 62 
	  return i;
61
63
62 
				// Update the solution by adding the eigenvector to Q and
 64 
	// Update the solution by adding the eigenvector to Q and
63 
				// the eigenvalue to the diagonal of L.
 65 
	// the eigenvalue to the diagonal of L.
64 
				Q[i] = q;
 66 
	Q[i] = q;
65 
				L[i][i] = l;
 67 
	L[i][i] = l;
66 
 
 68 
 
67 
				// Update A by subtracting the subspace represented by the
69 
	// Update A by subtracting the subspace represented by the
68 
				// eigensolution just found. This is called the method of
70 
	// eigensolution just found. This is called the method of
69 
				// deflation.
 71 
	// deflation.
70 
				MT B;
 72 
	MT B;
71 
				outer_product(q,q,B);
 73 
	outer_product(q,q,B);
72 
				A = A - l * B;
 74 
	A = A - l * B;
73 
			}
 75 
      }
74 
		return n;
 76 
    return n;
75 
	}
 77 
  }
76 
 
 78 
 
77 
	/* There is no reason to put this template in a header file, since
79 
  /* There is no reason to put this template in a header file, since
78 
		 we will only use it on matrices defined in CGLA. Instead, we
80 
     we will only use it on matrices defined in CGLA. Instead, we
79 
		 explicitly instantiate the function for the square matrices
 81 
     explicitly instantiate the function for the square matrices
80 
		 of CGLA */
 82 
     of CGLA */
81 
	template int power_eigensolution<Mat2x2f>(const Mat2x2f&,
 83 
  template int power_eigensolution<Mat2x2f>(const Mat2x2f&,
82 
																						Mat2x2f&,Mat2x2f&,int);
 84 
					    Mat2x2f&,Mat2x2f&,int);
83
85
84 
	template int power_eigensolution<Mat3x3f>(const Mat3x3f&,
 86 
  template int power_eigensolution<Mat3x3f>(const Mat3x3f&,
85 
																						Mat3x3f&,Mat3x3f&,int);
 87 
					    Mat3x3f&,Mat3x3f&,int);
86 
	template int power_eigensolution<Mat4x4f>(const Mat4x4f&,
 88 
  template int power_eigensolution<Mat4x4f>(const Mat4x4f&,
87 
																						Mat4x4f&,Mat4x4f&,int);
 89 
					    Mat4x4f&,Mat4x4f&,int);
88 
}
 90 
}
89 
 
 91