unusual subscriptied assignment mismatch

I narrowed the occurence of the error down to some change in the variable b that happens in the 9th iteration of the loop. For some reason the zeros at the end are no longer kept and the assignment fails.

I figured out, that this change happens exactly when b becomes greater than 512, so I guess it has something to do with matlabs handling of doubles? Further, using the uncommented variable makes the code work just fine.

Please help me solve this problem.

 var1 = 0.1;
 vec = 1:1000;

 % length_var = 100.0000;
 length_var = 1/(var1 * 0.1);

 for ii=1:20
    a = round(ii*(0.25*length_var));
    b = a - 1 + 3*length_var;
    disp(b-a)
    c(ii,:) = vec(a:b);
 end

Answer 1

[Jan]

You get a warning in each iteration:

 Warning: Integer operands are required for colon operator when used as index 

var1 = 0.1;
vec = 1:1000;
length_var = 1/(var1 * 0.1);
for ii=1:20
  a = round(ii*(0.25*length_var));
  b = a - 1 + 3*length_var;
  fprintf('%.16g\n', b - a);       % 298.9999999999999 in the first 8 iterations
  c(ii,:) = vec(a:b);
end

In the 9th iteration b-a becomes 299. Then vec(a:b) has 300 elements, but c(ii,:) has 299 only.

Do not use floating point numbers as indices. You have applied a round for the index a already. Do this for b also. Or even better: Let length_var have an integer value:

fprintf('%.16g', 1/(var1 * 0.1))
99.99999999999999

This is an expected effect when working with floating point numbers stored with limited precision. To get more explanations search in this forum or in the net for "IEEE754".

See also: FAQ: Why is 0.3 - 0.2 ~= 0.1

Answer 2

[Matt J]

   var1 = 0.1;
   vec = 1:1000;

   length_var = round( 1/(var1 * 0.1) );

   c=nan(20,3*length_var);

   for ii=1:20
      a = round(ii*(0.25*length_var));
      b = a - 1 + 3*length_var;
      disp(b-a)
      c(ii,:) = vec(a:b);
   end