Through Iron & Ice: Searching for Sterile Neutrinos at the IceCube Neutrino Observatory

Abstract

Despite the rapid progression in our understanding of neutrinos over the last half century, much is left unknown about their properties. This leaves neutrinos as the most promising portal for Beyond Standard Model (BSM) physics, and neutrinos have already provided fruitful surprises. A number of neutrino experiments in the last three decades have observed anomalous oscillation signals consistent with a mass-squared splitting of $Δm^2 \sim 1\ \textrm{eV}^2$, motivating the existence and search for sterile neutrinos. On the other hand, other experiments have failed to see such a signal. In this thesis, we present two analyses. The first is an update to the sterile neutrino global fits with the inclusion of recent experimental data. We find that the 3+1 model provides a better fit to the global data set compared to the null, with an improvement of $Δχ^2 = 51$ with the addition of only 3 degrees of freedom, corresponding to $6.6σ$. While a substantial improvement, we also find a irreconcilable tension between the data sets of $5.1σ$, calculated using the parameter goodness-of-fit test. This motivates the exploration of expanded models: a 3+2 model, and a 3+1+Decay model. In the 3+2 model, we find negligible improvement to the fit, and an even worse tension of $5.5σ$. In the more exotic 3+1+Decay model, we find the tension reduced to $3.6σ$. While a substantial improvement compared to the 3+1 model with the introduction of only one additional parameter, the tension is still too large to assuage concerns. The second analysis is the results of an expanded IceCube sterile neutrino search. A previous sterile neutrino search found no evidence for sterile neutrinos, finding a p-value of 8%. Of the three sterile mixing angles, $θ_{14}, θ_{24}$, and $θ_{34}$, only $θ_{24}$ was fitted for, as $θ_{14}$ was negligible and $θ_{34} = 0$ was considered a conservative assumption. We present results of an analysis where we include $θ_{34}$ to the fitted model. Both a frequentist and Bayesian analysis were conducted, with fits done in terms of the mass-squared splitting $Δm_{41}^2$ and the mixing matrix parameters $|U_{\mu4}|^2$ and $|U_{\tau4}|^2$. The frequentist analysis finds a best fit at $Δm_{41}^2 = 5.0\ \textrm{eV}^2$, $|U_{\mu4}|^2 = 0.04$, and $|U_{\tau4}|^2 = 0.006$, with a p-value of 5.2% assuming Wilks' Theorem with 3 degrees of freedom. Pseudoexperiments are indicating a smaller p-value 2.7%. The Bayesian analysis finds a similar best fit point at $Δm_{41}^2 = 5.0\ \textrm{eV}^2$, $|U_{\mu4}|^2 = 0.02$, and $|U_{\tau4}|^2 = 0.006$, with a Bayes factor indicating a ``Very Strong'' preference for this sterile hypothesis over the null hypothesis.