(security) Xavier Boyen - On the Impossibility of Efficiently Combining Collision Resistant Hash Functions

About a year ago, when the first collision attack against MD5 was made public, I asked my boss (who's a real scientist, unlike me!) why we couldn't simply use two different hashes and club them together, because the chances of finding a hash collision between two different strings, for two different hash functions, ought to be close to impossible.

Looks like I wasn't on crack after all... :-)

http://ai.stanford.edu/~xb/crypto06b/index.html

Let H1,H2 be two hash functions. We wish to construct a new hash function H that is collision resistant if at least one of H1 or H2 is collision resistant. Concatenating the output of H1 and H2 clearly works, but at the cost of doubling the hash output size. We ask whether a better construction exists, namely, can we hedge our bets without doubling the size of the output? We take a step towards answering this question in the negative --- we show that any secure construction that evaluates each hash function once cannot output fewer bits than simply concatenating the given functions.