Assume I only have only three algorithms available at my disposable: PBKDF2, SHA512, and HMAC-SHA256. (This means I do not have algorithms like HMAC-Expand available.)

Assume I also have a cryptographically secure random 256 bit key. I would like to expand this key to 512 bits, so that I can split this resulting key in half and now have 2 child keys.

The recommended option is to use PBKDF2, but this advice is typically given when stretching a low-entropy, user inputted password. Given that my key is already high-entropy, is there any issue with just using SHA512(key) to expand my key?

The reason I'd rather use SHA512 is that it does not require a salt like PBKDF2 does.

This answer states that it's acceptable to do something like:

child_key_1 = HMAC-256(master_key, "c1")
child_key_2 = HMAC-256(master_key, "c2")

Have I interpreted that answer correctly, and is there a clear winner amongst all my options?


1 Answer 1


What you are looking for is a "Key Derivation Function" (KDF). Notice you called out PBKDF - Password Based KDF. One very common KDF (it's even in NIST standards) is to use the shared secret - your "master_key" - to hmac a counter. Your proposal is therefore in line with the standard methods of KDF.

  • $\begingroup$ Actually, what he is looking for is a pseudorandom generator (PRG) since he assumes that the given key is already cryptographically secure. $\endgroup$ Mar 11, 2017 at 20:37
  • $\begingroup$ Is it your assumption that KDFs are only for low entropy input? $\endgroup$ Mar 11, 2017 at 22:05
  • $\begingroup$ Not only; using a KDF in this setting is fine. But KDFs usually have stronger properties and what the OP describes exactly matches the definition of a PRF, so I wanted to point this out. See en.wikipedia.org/wiki/Pseudorandom_generator. $\endgroup$ Mar 11, 2017 at 22:13

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.