Proposed by: Implicit in ElGamal (1985) and earlier; first systematically studied and named by Stefan Brands (1993) / surveyed by Dan Boneh (1998)
Category: Group-based
Mathematical form
Let G=⟨g⟩ be cyclic of prime order q.
Assumption. The distributions
(g,ga,gb,gab)and(g,ga,gb,gc),a,b,c←Zq
are computationally indistinguishable: for every PPT D,
Pr[D(g,ga,gb,gab)=1]−Pr[D(g,ga,gb,gc)=1]≤ε(n).
Where it holds / fails
Believed to hold: prime-order subgroups of Fp∗, prime-order elliptic-curve groups without efficient pairings (P-256, Curve25519, Ristretto255).
Provably false: groups with an efficiently computable symmetric pairing e (test e(ga,gb)=e(g,gab)) — this is exactly why [[dlin]] and [[sxdh]] were introduced; also false in Zp∗ itself (Legendre-symbol leak) — always work in the prime-order subgroup.
Best known attacks
Nothing better than computing discrete logs. Quantum-broken via Shor.
Importance
IND-CPA security of ElGamal encryption; IND-CCA security of Cramer–Shoup (1998, first practical CCA scheme without random oracles, >3000 citations).
Naor–Reingold PRF (1997), Ristretto/VRFs, PAKE protocols (SPAKE2, OPAQUE), mix-nets, and the analyses of X25519-based protocols (TLS 1.3, Signal) — typically as gap/oracle variants.
Boneh’s survey “The Decision Diffie–Hellman Problem” (1998, >2000 citations).