3
$\begingroup$

I wondered whether my understanding of entropy is correct, that a 256-bit counter that starts at 0 and counts to 2^256 - 1 by a +1 increment has a 256-bit entropy. I am asking this because I felt it awkward that for RNGs, developers often only specify an entropy.

So is it correct that a non-repeating 256-bit counter has 256-bits of entropy, because the numbers are equally distributed in their appearance?

$\endgroup$
3
  • 6
    $\begingroup$ Entropy is first and foremost a property of a process, and not of a variable or value. E.g. throwing a 16-sided perfect dice once 'produces' 4 bits of entropy. As such you might want to clarify what you mean, as the 'entropy of a counter' is ill-defined. See e.g. the first half of this answer for a description of entropy which might help you refine your question. $\endgroup$
    – Morrolan
    Commented Dec 13, 2022 at 16:14
  • $\begingroup$ I mean I have a process in which I start counting up from 0 to 2^256 - 1, and the process does not repeat itself. I am not quite sure what more information would be needed to answer the question. Let's say this was my first grader implementation of an RNG. $\endgroup$
    – kaiya
    Commented Dec 13, 2022 at 16:21
  • $\begingroup$ Or, turn the question around, what other properties of a RNG decide about whether the RNG is good? $\endgroup$
    – kaiya
    Commented Dec 13, 2022 at 16:26

2 Answers 2

11
$\begingroup$

So is it correct that a non-repeating 256-bit counter has 256-bits of entropy, because the numbers are equally distributed in their appearance?

Not necessarily. Entropy can be viewed as a measurement of "how much information does the attacker not know about the system". To take a simple example, a fair coin flipped in secret generates one bit of entropy; that same coin flipped in the open (where everyone can see it) generates no entropy (even though it is the same coin); whether it comes up heads or tails, the attacker knows which, and can take that into account.

In your example, if we assume that the attacker knows about when the counter is started, and knows the approximate count rate, he knows the approximate value of the counter, and hence the amount of entropy in that counter is much less than 256 bits.

$\endgroup$
4
  • 1
    $\begingroup$ Thanks! So the entropy would be 256-bit only if the observer wasn't able to recognize the pattern? But how are we able to distinguish non-predictable patterns from predictable ones? (I guess now we are talking PRNGs) $\endgroup$
    – kaiya
    Commented Dec 13, 2022 at 16:32
  • $\begingroup$ Anyhow, thanks for pointing out the relation with predictability, which were not obvious when only looking at the formulas of (Shannon) entropy, as are often taught. $\endgroup$
    – kaiya
    Commented Dec 13, 2022 at 16:54
  • 1
    $\begingroup$ @kaiya: actually, it is implicit with the standard formula for Shannon entropy; however we need to remember that the probability distribution used in that formula is from the attacker's perspective $\endgroup$
    – poncho
    Commented Dec 13, 2022 at 17:40
  • 1
    $\begingroup$ @kaiya: Anything you wrongly treat as unpredictable represents a hole in your algorithm. So, the standard approach in crypto is to assume the observer has access to the algorithm source code, but not the key. $\endgroup$
    – Brian
    Commented Dec 14, 2022 at 14:52
2
$\begingroup$

Actually not at all. It's only a few bits equivalent to the semantic content of the descriptive para-phrase integers 0 through 2^256 - 1 i.e. the set $( n \in \mathbb{Z}, 0 \le n \le 2^{256}-1 )$. That's only 14 symbols thus not many bits. It's a rough (upper bound) approximation to the Kolmogorov complexity of your counter.

Given that repetitively incrementing anything by one is extremely auto correlated, that process has an effective Kolmogorov complexity of zero and can be ignored. The only method for injecting entropy into your system would be to initialize the counter to a secret value. That would create the 256 bits of entropy.

And yes there are problems and nuances with the way the $\log_2$ Shannon formula is commonly used here. It only applies to to independent data samples that are not explicitly predictable from prior values, e.g. a fixed delta like incrementing by one. And also without an understanding of the generator at the point the sequence is read.

$\endgroup$
2
  • $\begingroup$ Thanks, I haven't yet looked into Kolmogorov complexity. Do I understand it correctly that, when using a random initialization value, only the process of creating the initial ransom has an entropy of 2^256, or did you mean the entire process in this case? $\endgroup$
    – kaiya
    Commented Dec 13, 2022 at 18:30
  • $\begingroup$ @kaiya Err, sorry. I'm not quite sure I understand your comment... $\endgroup$
    – Paul Uszak
    Commented Dec 14, 2022 at 13:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.