sum(w) and ones(1,size(w,2))*w' results totally different numbers

hi, I have a vector w and it is a matrix with one row and 10 columns.

>> sum(w)

ans =

   -0.1563

Then, I calculate the following, which is the product of a Ones vector and w, in which in matrix algebra, this is equivalent to summation of w's elements. >> ones(1,size(w,2))*w'

ans =

     1

Both results are different. Would you please let me know why?

w =

   1.0e+15 *

Columns 1 through 6

   -0.0497    2.4484   -3.2273    0.1944    0.1592    0.4407

Columns 7 through 10

   -0.4389    0.1548    0.1592    0.1592

Answer 1

[James Tursa]

The answer is on the order of eps of the numbers you are dealing with. You have massive cancellation going on in the operation, and the order of the operation will make a difference (apparently sum and mtimes are doing the calculation in a slightly different order). The answer will have a lot of garbage bits. E.g., what do you get when you do this:

eps(max(abs(w)))

And then compare that result with your answer and you will see that they are about the same size.

[Jan]

Exactly. Try this (insert all digits, if you have them):

w = [-0.0497e15, 2.4484e15, -3.2273e15, 0.1944e15, 0.1592e15, ...
      0.4407e15, -0.4389e15, 0.1548e15, 0.1592e15, 0.1592e15];
format long g
sum(w)
sum(sort(w))
[dummy, index] = sort(abs(w));
sum(w(index))

Do you see a difference already? For a sum with error correction see FEX: XSum

Answer 2

[bym]

try

ones(size(w))*w.'

[Cedric Wannaz]

It's not that; it is related to the way MATLAB computes the product between a vector whose elements are ~1e15 and an other whose elements are 1's.

[Abalfazl Zareei]

The same answer:

>> ones(size(w))*w.'

ans =

     1

Answer 3

[Ahmed A. Selman]

The line

ones(1,size(w,2))*w'

means creating a ones matrix with dimensions (1,size(w,2)) multiplied with w', and

sum(w)

means summation of the vector w.

I really don't know if it is right to compare these two, entirely different things.

So use

sum(ones(1,size(w,2))*w')

and compare :)

[Jan]

Multiplying an [1xN] with an [Nx1] vector means the dot-product. This is mathematically the same as summing the multiplied elements. Using a vector of 1's as one of the vectors results in a sum. Therefore the shown procedures are not different in theory.

[Ahmed A. Selman]

Indeed it is the same. I thought w was N-by-N, my bad. Thank you for the correction.