Work

A framework for comparing nuclear warhead authentication protocols

Year: 2019
Type: dissertation
Abstract: Even with the end of the Cold War, nuclear arms control continues to be a cornerstone of strategic stability and international non-proliferation efforts.New treaties are necessary to build upon, or at least maintain, the status-quo, and will rely upon verification technologies and protocols to ensure all sides are dismantling their warheads as promised.The nuclear weapon states refuse to participate in any process which might reveal the design of their warheads to an adversary or would-be proliferator.This makes warhead authentication, the critical verification step where the object to be dismantled is confirmed to be an authentic warhead, especially challenging.Despite several decades of research, there is no agreed means of describing or assessing warhead authentication protocols.This has hindered protocol development, and made it more difficult for the policy and technical communities to communicate what is important and feasible.This thesis presents a framework for describing warhead authentication protocols and quantifying their performance.The framework draws on methods used to assess digital authentication protocols, as well as information theoretic analysis of privacy.A model is developed for describing authentication protocols; showing how authentication questions, physical properties, and measurable data relate to one another.This allows the objectives and assumptions of a protocol to be made explicit, helping to ensure that protocols are compared fairly.It was found that the protocols in the literature have made use of very different assumptions, and that has influenced their choices of measurement technology and concepts of operation.Having established how to describe protocols, the thesis investigates how best to quantify the completeness (type I error rate), soundness (type II error rate), and information privacy of a protocol.While the absolute soundness cannot be calculated without knowledge of all possible hoaxes, a conditional soundness can be estimated using a minimax approach.A new measure of information privacy is presented, based on a change in the KL divergence between an inspector's beliefs and the actual warhead design, when the inspector starts from an incorrect prior.Médard always made time for my questions and was an amazing source of ideas for how to apply information theory to the project.Professor Areg Danagoulian helped me remain focused on the practical matters underlying my project, which I greatly appreciated.All of my committee have been excellent role models as researchers and mentors, which I hope to emulate in the future.I would like to thank all the staff in the department who have helped me in various ways.The list is very long, but I would especially like to say thank you to Heather for looking after us and keeping us on the straight and narrow in those first few years.Thank you as well to everyone who has made the past few years so enjoyable and memorable.This includes everyone in the LNSP group, Ian, my friends in NSE and the MIT rugby club, and especially the Simmons students and House team.Being a GRT was an experience I will always cherish.Last, but by no means least, I would like to thank my parents, Jamie, Lisa, and Alba for their love and support.Your encouragement and help has been of immeasurable value, and gives me confidence as we move on to the next challenges together.lognormal-distributed property with low variance, and a uniformlydistributed nuisance parameter . . . . . . . . . . . . . . . . . . . . .73 17 Calculating the completeness of a Poisson observable of an underlying lognormal-distributed property with high variance, and a uniformlydistributed nuisance parameter . . . . . . . . . . . . . . . . . . . . .74 18 Poisson observable of an invariant property and a a uniformly-distributed nuisance parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . .75 19 Example of linearized neutronics model for assessing soundness . . . .78 20 1D example of soundness calculation, for a Poisson observable of a lognormal property, as defined in Equation 3.22 . . . . . . . . . . . .84 21 1D example of soundness calculation, for a Poisson observable of a Gaussian property, as defined in Equation 3.8 . . . . . . . . . . . . .84 22 Illustration of the authentic object used in the 2D soundness example 86 23 Distribution of risk function results for successful hoaxes in the 2D soundness example, for 𝒞 = 0.8, 𝒮 = 0.7.The most extreme results are taken as the risk function for that completeness, soundness pair. . . .89 24 Completeness -soundness -risk map for the 2D soundness example .90 25 Completeness -soundness -risk map for the 2D soundness example using a one-sided hypothesis test to more easily detect [A] heavy hoaxes, and [B] light hoaxes . . . . . . . . . . . . . . . . . . . . . . . . . . . .90 26 Shannon entropy is a measure of the uniformity, or peakedness, of a distribution, regardless of the ordering of the entries.This makes it an accurate measure of uncertainty but less appropriate as a measure of estimator error . . . . . . . . . . . . .
Language: English
Open Access status: green
OpenAlex ID: W3035239802
Cites: 52
Cited by:
Related to: 10