Hacker News Reader

Comment by userbinator | Hacker News Reader

For e.g. a hash function, you'd want either a minimum or something like a 1st percentile.

As far as I know, all cryptographic hash functions are sensitive to single bit-flips by design.

It is definitely a design goal, but I would be very surprised if sha2 did not have 2 inputs that only differ by a single bit, and have the same output.

Finding those inputs is essentially impossible, but for a true 'random oracle ' they are likely to exist.

Note: correct me if I’m wrong, but to be relevant to this theorem and definition of sensitivity, the two strings would have to be for the same circuit and therefore be the same length, which is much harder than finding hash collisions in general.

And I thought they had ways to construct hash functions so that all the inputs of the same length have a different output?

> And I thought they had ways to construct hash functions so that all the inputs of the same length have a different output?

That's trivially impossible for fixed-size hashes, by the pigeon hole principle.

Not if the output is longer than the input, such as for passwords

I doubt any study of hash functions restricts itself to studying a very tiny subset of all possible (infinite) inputs. The security of hash functions is always studied under the assumption of arbitrary inputs.

Sorry, I meant all inputs whose length equals the output length.

The point being, for at least that case you can guarantee a sensitivity of one bit.

Yes, but sensitivity is the wrong measure to prove that. It’s the difference between a big-O and small-o complexity.

Reply on news.ycombinator.com