Is (0/0)log(0/0) identified in information theory?

1.8k Views Asked by At

I encounter this question when studying the Theil index for income inequality, which is invented by Theil who borrows a lot from information theory.

My question is just as that in the title. Can we calculate (0/0)log(0/0)? I know in information theory we have 0log(0)=0 and 0log(0/0)=0. But what about (0/0)log(0/0)?

Thanks!

1

There are 1 best solutions below

1
On

I know in information theory we have $0 \log(0)=0 $

Actually, what we have is that in several formulas (esp. the entropy) appears the term $ p_i \log(p_i)$ - that expression is not defined when $p_i=0$, because $0 \log(0)$ is not defined, but it's readily seen that the formulas remain valid if, we take the convention $x \log(x)|_{x=0}=0$ (this is also reasonable by a limit argument). Or, writen more concisely (but also more confusingly) $0 \log(0)=0 $. But it must be understood that this "rule" only applies for that kind of expression. Elsewhere, we fall into absurd consequences.

and $0 \log(0/0)=0$

... and this might be one. I'm not sure where you've read this, and it's not clear what this means. If you mean,as before $x \log(x/x)|_{x=0}=0$ ... well, that's obviously true in the limit, because we get $0 \log(1) = 0 \times 0 =0$.

what about $(0/0)\log(0/0)$ ?

Again, the question as stated makes little sense. Obviously, $(x/x) \log(x/x) $ tends to $1 \log(1)=1 \times 0 =0$ as $x\to 0$.

But to take as such a general "rule" that $(0/0)\log(0/0)=0$ would be as wrong/objectionable and prone to errors as assuming $0/0=1$.