Methods for Reducing Ancilla-Overhead in Block Encodings

Abstract: Block encodings are a fundamental primitive in quantum algorithms, but can often have large ancilla overhead. In this work, we introduce novel techniques for reducing this overhead in two distinct ways. In Part I, we prove the existence of a "space-time tradeoff" by deriving an algorithm that, for any block encoding, approximately uncomputes all but one of its ancilla (freeing up those ancillae for reuse in later parts of a quantum algorithm). In Part II, we evaluate the minimum number of ancillae required to perform coherent multiplication of block encodings. We prove that logarithmic ancillae is optimal for exact multiplication of block encodings. However, in certain block encoding regimes, we show that approximate multiplication of block encodings can be achieved to high-precision with just one ancilla.