# How to calculate entropy of a random number?

I am having a set of random numbers and I want to calculate entropy of them. I searched many entropy calculation formula, but I didn't get it. Can you elaborate a little bit?

• Many formulae? Is there more than one? Commented Sep 4, 2017 at 10:38
• Can you elaborate on what exactly you do not understand please?
– dusk
Commented Sep 4, 2017 at 11:27
• If your numbers were auto correlated, that would be interesting. Are they? Commented Sep 4, 2017 at 13:38
• @Paul Uszak: you are right, there is not more that one applicable formula; there's none, for the reason in the second sentence in that answer.
– fgrieu
Commented Sep 4, 2017 at 14:54

In a cryptographic sense this is not really possible. The entropy of the numbers is determined by the way they have been chosen. From only a list of numbers, say $(1, 2, 3, 4)$, we cannot just determine the entropy.

But if we instead say that we choose four numbers uniformly from 1 to 10, we can calculate the entropy. Recall the definition of (Shannon) entropy:

$$H(X) = -\sum_{i=1}^n {\mathrm{P}(x_i) \log_b \mathrm{P}(x_i)}$$

Here $X$ is the state (e.g. $(1, 2, 3, 4)$). The $x_i$ values are the possible states, e.g. $\{(1, 1, 1, 1), (1, 1, 1, 2), \ldots, (10, 10, 10, 10)\}$. $P(x_i)$ is the probability that the random number $X = x_i$. When the probability distribution is uniform, then the probabilities are all equal to $\frac{1}{n}$.

Notice that this is a calculation not over the state itself, but over the probability distribution of all the possible states. In other words: The entropy is determined not on what the numbers are, but how they are chosen.

It is possible to estimate (not calculate) the entropy of a series of data, but this is more relevant in the field of data processing. This is not relevant in cryptography.

• And what is x exactly if the source isn't a CSPRNG? Commented Sep 4, 2017 at 11:35
• I expanded the answer to elaborate $x$ and the other variables. Not sure if it is really clear. What do you think?
– dusk
Commented Sep 4, 2017 at 11:45
• Usually in cryptography we are concerned with min-entropy, $-\max_i \log_2 P(x_i)$, not Shannon entropy, since it better reflects a smart adversary's chance at breaking a system, rather than a random uninformed adversary's chance at breaking it. I also like to emphasize the interpretation that a probability distribution formalizes either a physical process or an adversary's state of knowledge. Commented Sep 4, 2017 at 14:54