1) January 10, 2005: A few of the very first book CDs were prepared using an improper file format that causes naming (all capital letters) and other associated problems. The publisher worked to remedy this in all warehoused copies, and hopefully none were sold.
Users can also download the proper CD volume here (updated version from 2007).
Some distributions of the MATLAB code for examples 2.1 and 2.4 had bugs. Correct versions are here:
2) January 11, 2005: In Chapter 10, the printed MATLAB code associated with the roughening matrix for Exercise 10.4 was ommitted from the textbook. It can be found on the CD as: Homework/chap10/makerough.m .
The code itself is:
%
% Construct a two-dimensional, second-order roughening matrix
% for a PSCALE by PSCALE square nodal model (with
% model wrap-around)
%
for i=1:PSCALE
for j=1:PSCALE
L((j-1)*PSCALE+i,(j-1)*PSCALE+i)=-4;
%
% Up
%
if (i > 1)
L((j-1)*PSCALE+i,(j-1)*PSCALE+i-1)=1;
else
L((j-1)*PSCALE+i,(j-1)*PSCALE+PSCALE)=1;
end
%
% Down
%
if (i < PSCALE)
L((j-1)*PSCALE+i,(j-1)*PSCALE+i+1)=1;
else
L((j-1)*PSCALE+i,(j-1)*PSCALE+1)=1;
end
%
% Left
%
if (j > 1)
L((j-1)*PSCALE+i,(j-2)*PSCALE+i)=1;
else
L((j-1)*PSCALE+i,(PSCALE-1)*PSCALE+i)=1;
end
%
% Right
%
if (j < PSCALE)
L((j-1)*PSCALE+i,j*PSCALE+i)=1;
else
L((j-1)*PSCALE+i,i)=1;
end
end
end
3) February 7, 2005: In Chapter 3, page 42, the equation after (3.5) should have been numbered, and the "dx" at the end of the integral has been truncated to "d".
4) February 18, 2005: In Appendix B, page 270, equation B.82. tobs = u0-mean/s n^(1/2) should have s/ n^(1/2) in its denominator (so it is like equations B.83 and B.72).
5) April 11, 2005: On page 33, (2.73) should have the sum from 1 to m.
6) August 29, 2005: On page 13, there is an extraneous "d" on the second line of (1.33).
7) September 14, 2005: On page 271, in Exercise B.4, the last inequality should be x >= 2 (not x >= 0).
8) September 19, 2005: On page 34, the right hand sides of (2.76), (2.77), and (2.78) should have leading minus signs to be consistent with our definition of the residual vector (2.74). Note, however, that this factor has no effect on the resulting forumulation of the IRLS problem because the minus sign divides out when the gradient is set equal to zero in (2.80).
9) September 20, 2005: On pages 254-255, in (B.8), the proper normalization constant for the PDF is 2^(-1/2} not 2^(-3/2). In the accompanying plot, Figure B.4, the maximum of the plotted PDF at x=0 is thus 2^(-1/2), and the lambda in its caption should be a sigma to be consistent with (B.8).
10) September 21, 2005: On page 42, in (3.6), there is an "x" at the beginning of the second line of the equation that should be at the end of the first line to make "dx".
11) September 23, 2005: On page 51, the normalization of Ie just above (3.59) should be by 10^37, not 10^31.
12) September 26, 2005: On page 27, the denominator of (2.53) should be: sqrt(C_{i,i}), where the covariance matrix C=sigma^2*inv(G'*G), similarly, the denominator of (2.54) should be: sqrt(C_{i,i}), where C=s^2*inv(G'*G)
13) February 20, 2008: On page 33 the upper limit of the summation in (2.73) should be "m", not "n".
14) March 31, 2009: On pages 69, 71, and 72, the reshape statements in (4.97), (4.98), and (4.100) should be transposed. As a result, the results displayed in Figures 4.3 and 4.4 on page 70 should be transposed. In general, if a model vector is constructed from successive column elements this transpose is necessary for the reshape command to work as desired. Conversely, if a model vector is constructed from successive row elements, then the transpose is not appropriate.
In the case of (4.100) and (4.102) this makes no practical difference because the resolution matrices are symmetric.