Supplementary Figure 1
    
    Supplementary Table 1:  Statistical results for various inversions
  
Comparison of statistical results for a variety of models and radial
parameterizations, where VR is the variance reduction, N=2850 is the
number of data and M is the number of model parameters. In model S6,
we determine lateral variations in S velocity up to degree~6,
and assume that lateral variations in P velocity and density
are proportional to the S velocity variations by constant scaling
factors of 0.55 and 0.2, respectively. In model SP6, we invert for
independent lateral variations in S and P velocity, but
assume that the density model is proportional to the S model.
These two inversions use only the mode data. The third inversion, model
SPRD6, allows for independent lateral variations in S velocity,
P velocity and density, and additional variations in topography
on the free surface, the 660 and the CMB. This inversion
includes constraints imposed by the free-air gravity anomaly. K13
indicates that models are expanded radially in Chebyshev polynomials up
to order 13 (over-parameterized inversion), and K7 models use Chebyshev
polynomials up to order 7. 
        
        
	  | Model | VR (%) | c2/N | c2/(N-M) | M | 
	
	  | S6 (K13) | 89.4 | 2.7 | 3.1 | 378 | 
	
	  | SP6 (K13) | 90.3 | 2.5 | 3.3 | 758 | 
	
	  | SPRD6 (K13) | 91.5 | 2.1 | 3.7 | 1215 | 
	
	  | S6 (K7) | 88.1 | 3.0 | 3.2 | 216 | 
	
	  | SP6 (K7) | 90.1 | 2.5 | 2.9 | 432 | 
	
	  | SPRD6 (K7) | 91.7 | 2.1 | 2.8 | 729 | 
      
Changes in c2/(N-M) indicate whether the increase in the number
of model parameters is warranted by the data. However, this test is not
useful when the models are over-parameterized, as in K13 inversions.
We prefer to over-parameterize our models and damp higher-order
polynomials more strongly rather than parameterizing the models with
lesser degrees of freedom. When the models are expanded radially up to
order 7, which we believe gives a reasonable number of model parameters
with respect to the number of data, c2/(N-M) improves with the
addition of P, and again when density and boundary topography
are added. 
      
      
      
	
      
      
      
        
      Miaki Ishii, Department of Earth and Planetary Sciences, Harvard University, copyright ©1999, all rights reserved
      
      
Last modified: September 8, 1999