In this thesis, we study several variants of the Mean King problem from the perspective of quantum information and optimization. The Mean King problem is a quantum guessing task, where one party prepares a quantum state, another party performs a measurement chosen from a given set of bases, and the first party tries to guess the measurement outcome. We consider four variants of this problem, including both non-interactive and interactive settings. We analyze their optimal guessing performance in both the average-case and the worst-case settings. We formulate the success probabilities of these Mean King games as optimization problems over quantum states and measurement strategies. This approach gives a unified way to study the different variants. For the non-interactive games, we derive the winning operators and show that some of the optimization problems can be reduced to eigenvalue computations. For the interactive games, we formulate them as semidefinite programs. These programs include extra constraints that come from returned information and partial trace conditions. These formulations support both theoretical analysis and numerical computation. We use MATLAB and CVX to compute optimal success probabilities for several choices of measurement bases. These bases include mutually unbiased bases and rotated bases. The results show how the choice of measurement bases affects the guessing probability. The results also show the difference between average-case and worst-case performance across different Mean King variants. In particular, in one variant, the numerical results for two and three mutually unbiased bases are close to the known success probabilities of the 2 → 1 and 3 → 1 quantum random access code tasks. These numerical similarities suggest that Mean King problems and random access codes may share a deeper connection. This observation gives a possible direction for future work.
In this thesis, we study several variants of the Mean King problem from the perspective of quantum information and optimization. The Mean King problem is a quantum guessing task, where one party prepares a quantum state, another party performs a measurement chosen from a given set of bases, and the first party tries to guess the measurement outcome. We consider four variants of this problem, including both non-interactive and interactive settings. We analyze their optimal guessing performance in both the average-case and the worst-case settings. We formulate the success probabilities of these Mean King games as optimization problems over quantum states and measurement strategies. This approach gives a unified way to study the different variants. For the non-interactive games, we derive the winning operators and show that some of the optimization problems can be reduced to eigenvalue computations. For the interactive games, we formulate them as semidefinite programs. These programs include extra constraints that come from returned information and partial trace conditions. These formulations support both theoretical analysis and numerical computation. We use MATLAB and CVX to compute optimal success probabilities for several choices of measurement bases. These bases include mutually unbiased bases and rotated bases. The results show how the choice of measurement bases affects the guessing probability. The results also show the difference between average-case and worst-case performance across different Mean King variants. In particular, in one variant, the numerical results for two and three mutually unbiased bases are close to the known success probabilities of the 2 → 1 and 3 → 1 quantum random access code tasks. These numerical similarities suggest that Mean King problems and random access codes may share a deeper connection. This observation gives a possible direction for future work.
The mean king problem is a quantum mechanical retrodiction problem, in which Alice has to name the outcome of an ideal measurement made in one of several different orthonormal bases. Alice is allowed to prepare the state of the system and to do a final measurement, possibly including an entangled copy. However, Alice gains knowledge about which basis was measured only after she no longer has access to the quantum system or its copy. We give a necessary and sufficient condition on the bases, for Alice to have a strategy to solve this problem, without assuming that the bases are mutually unbiased. The condition requires the existence of an overall joint probability distribution for random variables, whose marginal pair distributions are fixed as the transition probability matrices of the given bases. In particular, in the qubit case the problem is decided by Bell's original three variable inequality. In the standard setting of mutually unbiased bases, when they do exist, Alice can always succeed. However, for randomly chosen bases her success probability rapidly goes to zero with increasing dimension.
Many fundamental problems in quantum information theory can be formulated as optimization tasks over quantum states and measurements. While the convex structure of quantum theory often enables powerful semidefinite programming formulations, a major challenge persists: the exponential growth of Hilbert space dimension with system size. This dissertation discusses techniques to address these and related obstacles across several core problems in quantum information. We first study optimization over separable quantum states. Such problems are typically NP-hard and standard semidefinite programming hierarchies suffer from the curse of dimensionality. We introduce a hybrid algorithm that leverages a quantum co-processor to reduce the effective dimension of the problem. While several existing approaches provide only a numerical bound on the optimal value, our method is guaranteed to return an explicit feasible solution: a valid separable state, even in instances where it is not optimal. We apply this framework to approximate the minimum separable energy of many-body Hamiltonians, define an entanglement measure for their ground spaces, and demonstrate numerical scalability to systems of up to 28 qubits. We next consider the general Bayes framework, as introduced by Helstrom, for state discrimination when, instead of a classical description of the candidate states, one has access to their source code: the quantum circuit that prepares them. We show that the typical semidefinite program for the discrimination problem can be reformulated in terms of the Gram matrix of these states, reducing the variable dimensions from dL to NL, where d is the Hilbert space dimension, N is the number of candidate states, and L is the number of possible guesses. Importantly, we further introduce a quantum pre-processing procedure which efficiently constructs the reduced problem from the source code, enabling our method to operate directly on quantum data. We consider two applications. First, we characterize the optimal identifications for quantum changepoint problems under several reward structures whose associated discrimination problems were previously computationally inaccessible. Second, we consider a quantum error classification problem and show how our reduction makes it tractable for systems of hundreds of qubits. We also analyze structural measurement strategies. As a counterpart to the well-known pretty good measurement for minimum-error state discrimination, we introduce the pretty bad measurement for the exclusion variant, in which the goal is to identify a state that was not sent, rather than to identify the one that was, and prove that both are guaranteed to perform at least as well as blind guessing in their respective objectives. Finally, we consider evaluating Boolean functions under severe memory constraints and prove that a simple local "greedy" measurement strategy is globally optimal precisely for affine Boolean functions, while more generally achieving at least the square of the optimal global success probability. Together, these results demonstrate that high-dimensional quantum optimization problems often admit hidden structure enabling dimension reduction, hybrid computation, and provable approximation guarantees.
The mean king problem is a conditional retrodiction problem. In this problem Alice prepares a two prime-dimensional particles state and avails one of the particles to the king who measures its state in one of mutually unbiased bases of his choice. The king tells Alice his choice of basis after she completes a control measurement on his particle. Conditioned on this knowledge, she now infers the state observed by the king by utilizing the outcome of her control measurement. In the extended mean king problem, studied in this paper, the king does not tell Alice his measurement basis, but instead both the king and Alice repeat their measurements. Proper ordering of these allows Alice to deduce both the basis used by the king and the outcome of his first measurement, with the king reticent throughout, i.e., this protocol effects a (almost) complete retrodiction of the king's first measurement.
In the King’s Problem, a physicist is asked to prepare a d-state quantum system in any state of her choosing and give it to a king who measures one of (d+1) sets of mutually unbiased observables on it. The physicist is then allowed to make a control measurement on the system, following which the king reveals which set of observables he measured and challenges the physicist to predict correctly all the eigenvalues he found. This paper obtains an upper bound on the physicist’s probability of success at this task if she is allowed to make measurements only on the system itself (the “conventional” solution) and not on the system as well as any ancillary systems it may have been coupled to in the preparation phase, as in the perfect solutions proposed recently. An optimal conventional solution, with a success probability of 0.7, is constructed in d = 4; this is to be contrasted with the success probability of 0.902 for the optimal conventional solution in d = 2. The gap between the best conventional solution and the perfect solution grows quite rapidly with increasing d.
Random access codes are a type of communication task that is widely used in quantum information science. The optimal average success probability that can be achieved through classical strategies is known for any random access code. However, only a few cases are solved exactly for quantum random access codes. In this paper, we provide bounds for the fully general setting of n independent variables, each selected from a d-dimensional classical alphabet and encoded in a D -dimensional quantum system subject to an arbitrary quantum measurement. The bound recovers the exactly known special cases, and we demonstrate numerically that even though the bound is not tight overall, it can still yield a good approximation.
Random access codes are a type of communication task that is widely used in quantum information science. The optimal average success probability that can be achieved through classical strategies is known for any random access code. However, only a few cases are solved exactly for quantum random access codes. In this paper, we provide bounds for the fully general setting of n independent variables, each selected from a d-dimensional classical alphabet and encoded in a D-dimensional quantum system subject to an arbitrary quantum measurement. The bound recovers the exactly known special cases, and we demonstrate numerically that even though the bound is not tight overall, it can still yield a good approximation.