Quantum simulation is a fundamental application of quantum computing that promises to overcome the exponential scaling limitations of classical algorithms in modeling many-body systems.
Among various quantum algorithms, quantum Krylov subspace diagona...
Quantum simulation is a fundamental application of quantum computing that promises to overcome the exponential scaling limitations of classical algorithms in modeling many-body systems.
Among various quantum algorithms, quantum Krylov subspace diagonalization (QKSD) has emerged as a promising method for the early fault-tolerant quantum computing (EFTQC) era.
QKSD projects a Hamiltonian onto a subspace spanned by real-time evolved states, efficiently estimating extremal eigenvalues with modest circuit depth.
However, its practical reliability has not been well understood, which can be clarified by analyzing the effect of statistical sampling noise and the ill-conditioning.
This dissertation establishes a theoretical framework for sampling error and cost analysis in QKSD, which had not been previously formalized.
Using random matrix modeling and non-asymptotic perturbation bounds for the generalized eigenvalue problem, it quantifies how finite-shot measurement noise propagates to eigenvalue estimation and derives the first explicit expression for QKSD’s computational cost in terms of system size, sampling accuracy, and spectral gap.
The basis-thresholding method is introduced to mitigate instability caused by small singular values in the overlap matrix, ensuring numerical robustness and predictable convergence behavior.
To further improve efficiency, two error-reduction strategies are introduced: the Hamiltonian shifting technique, which eliminates redundant operator terms, and the iterative coefficient splitting (ICS) method, which optimizes quantum uncertainty by balancing operator weights.
These strategies drastically reduce the total sampling cost while maintaining precision, as confirmed by numerical experiments on molecular Hamiltonians and Hubbard-type models.
Beyond error mitigation, this work generalizes the Krylov framework to filtered-state preparation, interpreting the Krylov subspace as a spectral filter that preferentially amplifies the ground-state component.
This insight leads to the development of filtered quantum phase estimation (FQPE), a unifying framework that integrates Krylov filtering with phase estimation to achieve improved overlap amplification and reduced resource requirement.
Along with the Krylov filter, classically inspired filters such as Gaussian and Chebyshev filters are systematically analyzed to quantify the trade-offs among circuit depth, success probability, and accuracy.
Through rigorous analysis and practical design strategies, this dissertation enhances the reliability and resource efficiency of QKSD, bridging the gap between near-term heuristic methods and asymptotically optimal quantum algorithms.
The results contribute to the foundation for predictable eigenvalue estimation and quantum advantage realization in the early fault-tolerant era.