zlatps.f man page

 ZLATPS solves one of the triangular systems

    A * x = s*b,  A**T * x = s*b,  or  A**H * x = s*b,

 with scaling to prevent overflow, where A is an upper or lower
 triangular matrix stored in packed form.  Here A**T denotes the
 transpose of A, A**H denotes the conjugate transpose of A, x and b
 are n-element vectors, and s is a scaling factor, usually less than
 or equal to 1, chosen so that the components of x will be less than
 the overflow threshold.  If the unscaled problem will not cause
 overflow, the Level 2 BLAS routine ZTPSV is called. If the matrix A
 is singular (A(j,j) = 0 for some j), then s is set to 0 and a
 non-trivial solution to A*x = 0 is returned.

  A rough bound on x is computed; if that is less than overflow, ZTPSV
  is called, otherwise, specific code is used which checks for possible
  overflow or divide-by-zero at every operation.

  A columnwise scheme is used for solving A*x = b.  The basic algorithm
  if A is lower triangular is

       x[1:n] := b[1:n]
       for j = 1, ..., n
            x(j) := x(j) / A(j,j)
            x[j+1:n] := x[j+1:n] - x(j) * A[j+1:n,j]
       end

  Define bounds on the components of x after j iterations of the loop:
     M(j) = bound on x[1:j]
     G(j) = bound on x[j+1:n]
  Initially, let M(0) = 0 and G(0) = max{x(i), i=1,...,n}.

  Then for iteration j+1 we have
     M(j+1) <= G(j) / | A(j+1,j+1) |
     G(j+1) <= G(j) + M(j+1) * | A[j+2:n,j+1] |
            <= G(j) ( 1 + CNORM(j+1) / | A(j+1,j+1) | )

  where CNORM(j+1) is greater than or equal to the infinity-norm of
  column j+1 of A, not counting the diagonal.  Hence

     G(j) <= G(0) product ( 1 + CNORM(i) / | A(i,i) | )
                  1<=i<=j
  and

     |x(j)| <= ( G(0) / |A(j,j)| ) product ( 1 + CNORM(i) / |A(i,i)| )
                                   1<=i< j

  Since |x(j)| <= M(j), we use the Level 2 BLAS routine ZTPSV if the
  reciprocal of the largest M(j), j=1,..,n, is larger than
  max(underflow, 1/overflow).

  The bound on x(j) is also used to determine when a step in the
  columnwise method can be performed without fear of overflow.  If
  the computed bound is greater than a large constant, x is scaled to
  prevent overflow, but if the bound overflows, x is set to 0, x(j) to
  1, and scale to 0, and a non-trivial solution to A*x = 0 is found.

  Similarly, a row-wise scheme is used to solve A**T *x = b  or
  A**H *x = b.  The basic algorithm for A upper triangular is

       for j = 1, ..., n
            x(j) := ( b(j) - A[1:j-1,j]' * x[1:j-1] ) / A(j,j)
       end

  We simultaneously compute two bounds
       G(j) = bound on ( b(i) - A[1:i-1,i]' * x[1:i-1] ), 1<=i<=j
       M(j) = bound on x(i), 1<=i<=j

  The initial values are G(0) = 0, M(0) = max{b(i), i=1,..,n}, and we
  add the constraint G(j) >= G(j-1) and M(j) >= M(j-1) for j >= 1.
  Then the bound on x(j) is

       M(j) <= M(j-1) * ( 1 + CNORM(j) ) / | A(j,j) |

            <= M(0) * product ( ( 1 + CNORM(i) ) / |A(i,i)| )
                      1<=i<=j

  and we can safely call ZTPSV if 1/M(n) and 1/G(n) are both greater
  than max(underflow, 1/overflow).