Strange (lack of) round-off error using norm

Hello,

I have found this behaviour:

>> format hex
>> p=[sqrt(2) sqrt(2)]*1e3;
>> sqrt(sum(p.^2))
ans =
   409f400000000001
>> norm(p)
ans =
   409f400000000000

I understand the round-off error using sqrt. What I do not understand is how does norm do to avoid it.

Thanks,

Mariano

[David Goodmanson]

Hi Mariano

I ran this on my PC and got

ans = 409f400000000001
ans = 409f400000000001

Not sure what to say .....

[Walter Roberson]

R2018b on my Mac, I do see the different values.

Answer 1

[Mariano]

I should have said version and computer:

R2018b on Windows 64bits.