[cig-commits] r13096 - seismo/3D/ADJOINT_TOMO/flexwin

[carltape at geodynamics.org]

Author: carltape
Date: 2008-10-15 21:19:02 -0700 (Wed, 15 Oct 2008)
New Revision: 13096

Modified:
   seismo/3D/ADJOINT_TOMO/flexwin/seismo_subs.f90
Log:
Important modification to the definition of dlnA.  The previous definition was based on a first-order approximation to ln(A1/A2), which is (A1-A2)/A1.  There is no need for this approximation to be made, so now we simply use ln(A1/A2).


Modified: seismo/3D/ADJOINT_TOMO/flexwin/seismo_subs.f90
===================================================================
--- seismo/3D/ADJOINT_TOMO/flexwin/seismo_subs.f90	2008-10-15 23:46:37 UTC (rev 13095)
+++ seismo/3D/ADJOINT_TOMO/flexwin/seismo_subs.f90	2008-10-16 04:19:02 UTC (rev 13096)
@@ -721,7 +721,7 @@
   double precision, dimension(*), intent(in) :: d, s
   integer, intent(in) :: npts,i1,i2
   double precision, intent(in) :: dt
-  double precision, intent(out) ::  tshift,cc_max,dlnA
+  double precision, intent(out) :: tshift,cc_max,dlnA
 
   double precision, dimension(NDIM) :: d_win, s_win
   integer :: ishift
@@ -738,8 +738,14 @@
   ! calculate dlnA : definition of Dahlen and Baig (2002), Eq. 3,17,18 : dlnA = Aobs/Asyn - 1
   ! NOTE 1: these records are unshifted
   ! NOTE 2: this measurement will reflect any noise in the data, too
-  dlnA = sqrt( ( sum( d_win(i1:i2)*d_win(i1:i2) )) / (sum( s_win(i1:i2)*s_win(i1:i2) )) ) - 1.0
+  !dlnA = sqrt( ( sum( d_win(i1:i2)*d_win(i1:i2) )) / (sum( s_win(i1:i2)*s_win(i1:i2) )) ) - 1.0
 
+  ! CHT revision 15-oct-2008
+  ! The previously used expression for dlnA is a first-order perturbation of ln(A1/A2) = (A1-A2)/A2
+  ! The new expression is better suited to getting values between -1 and 1,
+  ! with dlnA = 0 indicating perfect fit, as before.
+  dlnA = 0.5 * log( sum(d_win(i1:i2)*d_win(i1:i2)) / sum(s_win(i1:i2)*s_win(i1:i2)) )
+
 end subroutine calc_criteria
 
 !------------------------------------------------------------------------