why is 1+ nan*1i = NaN + NaNi ?

Dear all,

I am trying to figure out how to store some information using complex numbers and retrieving that information later on.

For an operation like

1+nan*1i,

I would like to be able to retrieve "1" by doing

real(1+nan*1i)

and retrieve "nan" by doing

imag(1+nan*1i).

The first case does not work because matlab stores 1+ nan*1i as NaN + NaNi. Is there any workaround to this?

Thanks

Answer 1

[the cyclist]

I don't envision any workaround that would specifically make NaNs work. Speaking mathematically very loosely, the NaN in your original expression could be a "complex NaN", such that the NaN-ness leaks over to the real part.

I can imagine, without deep thinking, a couple workarounds, that are annoying in their own way, but might work in your particular situation.

• Use another matrix dimension to keep real and imaginary parts separate
• Define your own complex number object, that has the properties you outline above.

Answer 2

[Walter Roberson]

A = complex(1,nan)
real(A), imag(A)
B = A

However if you were to do

2*A

then you would get complex(NaN, NaN). You can use complex() to force unusual structures such as complex(5,0) but any arithmetic on it is going to demote it to complex(NaN, NaN)

[Patrick Mboma]

Hi Walter,

Thanks for the reply. I did not know about the complex function. Thanks also for making me aware of it.

Your solution will surely work if I have a scalar. But if I start stacking items in a vector, I am not sure what the results will be. For now I think I will use the safe option of just creating another dimension to keep the numbers separate.