Access Type

Open Access Dissertation

Date of Award

January 2020

Degree Type


Degree Name



Physics and Astronomy

First Advisor

Gil Paz


The radiative decay $\bar{B}\to X_s\gamma$ and semileptonic heavy meson decay $D\to \pi l \nu$ are important flavor physics probes of new physics. However, these decays are plagued with nonperturbative uncertainties that are needed to be controlled to obtain a theoretically clean description. In this dissertation, we provide effective field theory and machine learning approaches to controlling these uncertainties.\par

In $\bar B\to X_s\gamma$, the largest uncertainty on the total rate and the CP asymmetry arises from resolved photon contributions. These appear first at order $1/m_b$ and are related to operators other than $Q_{7\gamma}$ in the effective weak Hamiltonian. One of the three leading contributions, $Q^q_1-Q_{7\gamma}$, is described by a non-local function whose moments are related to Heavy Quark Effective Theory (HQET) parameters.\par

The extraction of higher-order moments requires the knowledge of higher-dimensional HQET and non-relativistic quantum chromo- dynamics (NRQCD) operators. We present a general method that allows for an easy construction of HQET or NRQCD operators that contain any number of covariant derivatives. As an application of our method, we list these terms in the $1/M^4$ NRQCD Lagrangian, where $M$ is the mass of the spin-half field. Then we use the recent progress in our knowledge of these parameters to reevaluate the resolved photon contribution to $\bar B\to X_s\gamma$ total decay rate and Charge and Parity asymmetry.\par

The decay rate of semileptonic $D\to \pi l \nu$ is proportional to the hadronic form factors, which parameterize how the quark $c \to d$ transition is realized in $D \to \pi$ meson decays. Currently, these form factors cannot be determined analytically in the whole range of available momentum transfer $q^2$. Thus, the form factors are parameterized by phenomenological models. We developed a machine learning approach with artificial neural networks trained from experimental pseudo-data to predict the shape of these form factors with a prescribed uncertainty. This provides the first model-independent parameterization of $D\to \pi l \nu$ vector form factor shape in the literature.