Imagine you're tuning a radio. You know the station is somewhere near 91.5 FM, give or take 0.1 — that's your energy estimate δ. Your antenna already picks up a faint signal from the right station — that's your guiding state overlap γ. The question is: how many times do you need to sweep the dial to lock onto the station perfectly? This paper proves you need exactly O(log(1/ε)/γδ) sweeps, and no strategy — no matter how clever — can do it in fewer. The committed claim: given a Hamiltonian H with a unique ground state |ψ₀⟩, an energy estimate within δ, a spectral gap of at least 3δ, and a guiding state with overlap γ, you can prepare an ε-approximation of the ground state using O(log(1/ε)/γδ) applications of the time-evolution operator e^{iH} and the guiding-state preparation unitary A. This matches the known lower bound, closing the complexity question for this primitive. Two distinct algorithms achieve the same bound. The first interleaves amplitude amplification with error reduction, extending the HMdW03 framework — a technique that alternates between boosting the probability of the target state and sharpening the approximation quality. The second uses the composition of quantum transducers, a more algebraic approach that chains quantum subroutines as composable black boxes. Having two independent constructions reaching the same bound is a strong structural signal: it suggests the bound is natural, not an artifact of one particular proof technique. This paper is the state-preparation companion to two recent ground-state-energy estimation papers by overlapping author sets [JW26, SdW26]. Together, the trio resolves both halves of the ground-state problem — estimating the energy and preparing the state — at optimal query complexity. The energy estimation papers established the lower bounds and the estimation algorithms; this paper closes the loop on preparation. The ladder here is the theoretical complexity landscape, not an experimental benchmark. Prior approaches to ground-state preparation either achieved suboptimal scaling in one of the three parameters (ε, γ, δ) or required additional assumptions like a priori knowledge of the spectral gap beyond what's needed here. The 3δ gap condition is mild — it's saying the ground state is reasonably isolated, which is the regime where the problem is well-posed. The log(1/ε) dependence on precision is exponentially better than polynomial dependence, which is what you'd get from naive phase estimation approaches without the guiding state. Integrity is strong in the specific sense that matters for theory papers: this is a mathematical proof, not a simulation or experiment. The claim is a theorem with a proof. The validation is whether the proof is correct, and the dual-algorithm construction provides internal cross-validation. There's no benchmark gaming possible — either the bound holds or it doesn't. The limitation is that this is purely query-complexity analysis; it says nothing about gate complexity, coherence requirements, or practical implementation on real hardware. The milestone question is where theory meets reality. This result tells you the fundamental cost of ground-state preparation in the oracle model. The next barrier is translating this into practical gate-level circuits for specific Hamiltonians of interest — chemistry, materials science, optimization — where the constants hidden in the O(·) notation and the cost of implementing e^{iH} dominate actual runtime. The gap between query-optimal and practically useful remains substantial, but you can't close it without first knowing the floor, and now we know the floor.
