New results on the hardness of diffie-hellman bits

We generalize and extend results obtained by Boneh and Venkatesan in 1996 and by González Vasco and Shparlinski in 2000 on the hardness of computing bits of the Diffie-Hellman key, given the public values. Specifically, while these results could only exclude (essentially) error-free predictions, we here exclude any non-negligible advantage, though for larger fractions of the bits. We can also demonstrate a trade-off between the tolerated error rate and the number of unpredictable bits. Moreover, by changing computational model, we show that even a very small proportion of the most significant bits of the Diffie-Hellman secret key cannot be retrieved from the public information by means of a Las Vegas type algorithm, unless the corresponding scheme is weak itself.