Qubitization and Linear Combination of Unitaries

Block-encoding and qubitization give near-optimal Hamiltonian simulation scaling, the fault-tolerant route Kronos targets for materials.

Beyond product formulas

Product formulas like Trotter are simple but scale poorly in the number of Hamiltonian terms and required precision. The modern approach writes H as a linear combination of unitaries (LCU) and builds a block encoding: a larger unitary that contains H/alpha in a corner, accessed through ancilla qubits.

H = sum_k c_k U_k          # LCU: U_k unitary (Pauli strings), c_k >= 0
alpha = sum_k c_k          # the '1-norm' of the Hamiltonian

# PREPARE:  |0> -> sum_k sqrt(c_k/alpha) |k>
# SELECT :  |k>|psi> -> |k> U_k |psi>
# Block encoding U_H = PREPARE^dag . SELECT . PREPARE
#   <0|_anc U_H |0>_anc = H / alpha

Qubitization and the query cost

Iterating the block encoding with reflections (the Szegedy-style walk operator) 'qubitizes' the problem: the eigenphases of the walk operator encode arccos of the eigenvalues of H/alpha. Simulating for time t then costs a number of queries that is near-optimal:

query_cost(t, eps)  =  O( alpha*t  +  log(1/eps) )
#   alpha  = Hamiltonian 1-norm (the effective 'block-encoding cost')
#   t      = evolution time,  eps = target error
# This is optimal in t and in eps (additive log term) -- unlike Trotter's
# polynomial-in-1/eps scaling.

Why it matters for first-wall chemistry

• Precision-friendly: extra digits of accuracy cost only logarithmically, important for chemically accurate energies.
• Scaling is governed by the Hamiltonian 1-norm alpha, which drives most fault-tolerant resource estimates.
• Requires clean logical qubits and many T gates, so it is a post-error-correction technique, not NISQ.

Kronos treats qubitization as the reference architecture for future materials simulation: it is what our fault-tolerant resource estimates assume when counting logical qubits and T gates for a tungsten defect cluster. It does not run on any device we can access today, and we say so plainly.