We study possible advantages of randomized and quantum computing over deterministic computing for scalar initial-value problems for ordinary differential equations of order k. For systems of equations of the first order this question has been settled modulo some details in \cite{Kacewicz05}. A speed-up over deterministic computing shown in \cite{Kacewicz05} is related to the increased regularity of the solution with respect to that of the right-hand side function. For a scalar equation of order k (which can be transformed into a special system of the first order), the regularity of the solution is increased by k orders of magnitude. This leads to improved complexity bounds depending on k for linear information in the deterministic setting, see \cite{Szczesny05}. This may suggest that in the randomized and quantum settings a speed-up can also be achieved depending on k. We show in this paper that a speed-up dependent on k is not possible in the randomized and quantum settings. We establish lower complexity bounds, showing that the randomized and quantum complexities remain at the some level as for systems of the first order, no matter how large k is. Thus, the algorithms from \cite{Kacewicz05} remain (almost) optimal, even if we restrict ourselves to a subclass of systems arising from scalar equations of order k.
We study possible advantages of randomized and quantum computing over deterministic computing for scalar initial-value problems for ordinary differential equations of order k. For systems of equations of the first order this question has been settled modulo some details in \cite{Kacewicz05}. A speed-up over deterministic computing shown in \cite{Kacewicz05} is related to the increased regularity of the solution with respect to that of the right-hand side function. For a scalar equation of order k (which can be transformed into a special system of the first order), the regularity of the solution is increased by k orders of magnitude. This leads to improved complexity bounds depending on k for linear information in the deterministic setting, see \cite{Szczesny05}. This may suggest that in the randomized and quantum settings a speed-up can also be achieved depending on k. We show in this paper that a speed-up dependent on k is not possible in the randomized and quantum settings. We establish lower complexity bounds, showing that the randomized and quantum complexities remain at the some level as for systems of the first order, no matter how large k is. Thus, the algorithms from \cite{Kacewicz05} remain (almost) optimal, even if we restrict ourselves to a subclass of systems arising from scalar equations of order k.
The complexity of initial-value problems is well studied for systems of equations of first order. In this paper, we study the ε-complexity for initial-value problems for scalar equations of higher order. We consider two models of computation, the randomized model and the quantum model. We construct almost optimal algorithms adjusted to scalar equations of higher order, without passing to systems of first order equations. The analysis of these algorithms allows us to establish upper complexity bounds. We also show (almost) matching lower complexity bounds. The ε-complexity in the randomized and quantum setting depends on the regularity of the right-hand side function, but is independent of the order of equation. Comparing the obtained bounds with results known in the deterministic case, we see that randomized algorithms give us a speed-up by 1/2, and quantum algorithms by 1 in the exponent. Hence, the speed-up does not depend on the order of equation, and is the same as for the systems of equations of first order. We also include results of some numerical experiments which confirm theoretical results.
We study the complexity of a two-point boundary value problem. We concentrate on the linear problem of order k with separated boundary conditions. Right-hand side functions are assumed to be r times differentiable with all derivatives bounded by a constant. We consider three models of computation: deterministic with standard and linear information, randomized and quantum. In each setting, we construct an algorithm for solving the problem, which allows us to establish upper complexity bounds. In the deterministic setting, we show that the use of linear information gives us a speed-up of at least one order of magnitude compared with the standard information. For randomized algorithms, we show that the speed-up over standard deterministic algorithms is by 1/2 in the exponent. For quantum algorithms, we can achieve a speed-up by one order of magnitude. We also provide lower complexity bounds. They match upper bounds in the deterministic setting with the standard information, and almost match upper bounds in the randomized and quantum settings. In the deterministic setting with the linear information, a gap still remains between the upper and lower complexity bounds.
This thesis describes quantum algorithms for Hamiltonian simulation, ordinary differential equations (ODEs), and partial differential equations (PDEs). Product formulas are used to simulate Hamiltonians which can be expressed as a sum of terms which can each be simulated individually. By simulating each of these terms in sequence, the net effect approximately simulates the total Hamiltonian. We find that the error of product formulas can be improved by randomizing over the order in which the Hamiltonian terms are simulated. We prove that this approach is asymptotically better than ordinary product formulas and present numerical comparisons for small numbers of qubits. The ODE algorithm applies to the initial value problem for time-independent first order linear ODEs. We approximate the propagator of the ODE by a truncated Taylor series, and we encode the initial value problem in a large linear system. We solve this linear system with a quantum linear system algorithm (QLSA) whose output we perform a post-selective measurement on. The resulting state encodes the solution to the initial value problem. We prove that our algorithm is asymptotically optimal with respect to several system parameters. The PDE algorithms apply the finite difference method (FDM) to Poisson's equation, the wave equation, and the Klein-Gordon equation. We use high order FDM approximations of the Laplacian operator to develop linear systems for Poisson's equation in cubic volumes under periodic, Neumann, and Dirichlet boundary conditions. Using QLSAs, we output states encoding solutions to Poisson's equation. We prove that our algorithm is exponentially faster with respect to the spatial dimension than analogous classical algorithms. We also consider how high order Laplacian approximations can be used for simulating the wave and Klein-Gordon equations. We consider under what conditions it suffices to use Hamiltonian simulation for time evolution, and we propose an algorithm for these cases that uses QLSAs for state preparation and post-processing.
Most continuous mathematical formulations arising in science and engineering can only be solved numerically and therefore approximately. We shall always assume that we're dealing with a numerical approximation to the solution. There are two major motivations for studying quantum algorithms and complexity for continuous problems. 1. Are quantum computers more powerful than classical computers for important scientific problems? How much more powerful? 2. Many important scientific and engineering problems have continuous formulations. To answer the first question we must know the classical computational complexity of the problem. Knowing the classical complexity of a continuous problem we obtain the quantum computation speedup if we know the quantum complexity. If we know an upper bound on the quantum complexity through the cost of a particular quantum algorithm then we can obtain a lower bound on the quantum speedup. Regarding the second motivation, in this article we'll report on high-dimensional integration, path integration, Feynman path integration, the smallest eigenvalue of a differential equation, approximation, partial differential equations, ordinary differential equations and gradient estimation. We'll also briefly report on the simulation of quantum systems on a quantum computer.
Quantum computers have the potential to solve some complex problems much faster than classical equivalents. Significant research efforts are aimed at determining which algorithms give which advantages for different use-cases. An application where quantum computers can possibly bring great advantages is solving (partial) differential equations, which has a broad set of applications, for instance in wave-propagation models. Few differential equations admit an analytical solution. For most heuristic methods are required to approximate a solution. Well-known heuristic techniques include the finite element and finite difference method, where the considered space is partitioned and systems of linear equations resulting from this partitioning need to be solved. Quantum computing puts forward new methods to solve these systems of linear equations and hence these differential equations. First, the HHL algorithm gives an efficient way to solve a linear system of equations. The HHL algorithm comes with drawbacks, but in this specific use-case some of these objections might be circumvented. Quantum computers furthermore offer the variational approach: an optimization-based path to solving differential equations. With this approach, the devices might even ‘learn’ the noise patterns emerging in present day quantum computers and compensate for it. In this work we revisit quantum methods for solving partial differential equations and consider how well they work in solving differential equations in practice. We also discuss possible caveats and bottlenecks of the approaches.
Quantum computers were first proposed to perform simulation -- in particular, quantum simulation. The premise is that such computers can simulate quantum dynamics exponentially faster than a classical computer. Some of the early proposals of practical applications of quantum computers come from quantum chemistry. Algorithms for more general systems, described by ordinary and partial differential equations, emerged later on. This thesis grounds on an extensive amount of prior research and proposes a set of improvements as well as new algorithms or algorithmic tools, focusing on applications from quantum chemistry and differential equations.First, we propose a technique relating to finding ground-states for electronic systems via the variational quantum eigensolver. The presented method enables solving the problem on early quantum hardware by a partitioning of the computational problem into smaller problems, using only classical resources to perform the partitioning. Next, the thesis considers simulation problems from chemistry in a more general sense. Following the idea that dynamical simulation of many chemical systems is efficient, whereas preparing ground-states is a `hard' problem, our computational framework leverages simulation routines to prepare molecular states as an input to a computation, relying only on `hard' input states for small problem instances. The preparation relies on dynamical scattering of initial fragments supported by a trap potential, allowing to estimate the success probability for certain bonds. A weak measurement scheme enables preparation of molecular states. Furthermore, this dissertation investigates quantum algorithms for differential equations. More specifically, the the Carleman linearization approach is considered to solve nonlinear differential equations. This entails bounding the Carleman truncation error, re-scaling of the solution to avoid an exponentially vanishing success probability, and discussing higher-order discretization. Moreover, this thesis introduced a penalty-based method to enforce constraints on the solution of differential equations, with inspiration from complex absorbing potentials and the quantum Zeno effect. We bound the approximation errors and embed the penalty method into an efficient quantum algorithm, with computational overhead only logarithmic in the penalty's strength. Efficiency is enabled by an interaction-picture approach with fast-forwardable constraint projections.