precision problem

[someone]

CBLAS function dgemv (all operations of type double) returns a vector,
some of whose components are:

-4.7195321e-16
2.6001879e-16
-6.1531184e-16
2.3054204e-16
-2.2817035e-16

Should I assume these to be essentially 0?
MATLAB does similar computation and returns values:

0.00000004361293
0.000000080393098
0.000000036770758
0.000000009778969
0.000000016177475

which look something like x.ye-8. Which of the two should I believe as
these components are used in further computations.

Thanks

[Kai-Uwe Bux]

someone wrote:

CBLAS function dgemv (all operations of type double) returns a vector,
some of whose components are:

-4.7195321e-16
2.6001879e-16
-6.1531184e-16
2.3054204e-16
-2.2817035e-16

Should I assume these to be essentially 0?
MATLAB does similar computation and returns values:

0.00000004361293
0.000000080393098
0.000000036770758
0.000000009778969
0.000000016177475

which look something like x.ye-8. Which of the two should I believe as
these components are used in further computations.

a) This is like: I asked two of my best friends which color my new shirt is.
One said green the other said blue. Whom should I believe?

How are we supposed to know which result is better without even knowing what
computation was done?

b) No please *do not post* your computation: the whole thing is *off topic*
anyway. See the FAQ for which questions legitimately go here.
Best

Kai-Uwe Bux

[Jim Langston]

"someone" <bo***********@gmail.com> wrote in message
news:11**********************@g10g2000cwb.googlegr oups.com...

CBLAS function dgemv (all operations of type double) returns a vector,
some of whose components are:

-4.7195321e-16
2.6001879e-16
-6.1531184e-16
2.3054204e-16
-2.2817035e-16

Should I assume these to be essentially 0?
MATLAB does similar computation and returns values:

0.00000004361293
0.000000080393098
0.000000036770758
0.000000009778969
0.000000016177475

which look something like x.ye-8. Which of the two should I believe as
these components are used in further computations.
It depends on the formulas you used. If these were the same formulas for
both, I would note that in your first one:
-4.7195321e-16 and 0.00000004361293

that CBLAS says it's a negative number, MATLAB says it's a positive number.
That's a pretty big difference. Look at your original formula (I don't care
to see it) and see if it should be a negative or positive value. If it
should be positive, use MATLAB, if it should be negative, use CBLAS. If you
can't decide, neither can I.

[res]

You can do some experiments to figure out how are they implemented
internally.

It's possible that one is using 64-bit double, and the other is using 80-bit
long double.

"someone" <bo***********@gmail.com> wrote in message
news:11**********************@g10g2000cwb.googlegr oups.com...

Which of the two should I believe as these components are used in further
computations.