Reducing parametric uncertainties through information geometry methods
M. Imbrišak, A. E. Lovell, M. R. Mumpower · 2025
Read Reducing parametric uncertainties through information geometry methods by M. Imbrišak, A. E. Lovell, M. R. Mumpower (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-19474.txt.
Introduction
Imbrišak, Lovell, and Mumpower wrote in twenty twenty-five on cutting error in fission models. They use information geometry, which brings shape math to data problems. Their target is a code that simulates how fission fragments decay. Blur in the input mass data blurs the code's fitted values. They gauge how tighter mass measures would sharpen those values. A toy model with simple noise assumptions came first.
Real test data on californium fission came next. Fitted errors fell by up to about fifteen percent. Two width values of the model gained the most precision. The work shows which measures buy the most certainty. That guidance can rank future fission tests by value. Better values improve forecasts for reactors and nuclear data libraries.
Catalog record
| CREATOR | M. Imbrišak, A. E. Lovell, M. R. Mumpower |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-19474
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Reducing parametric uncertainties through information geometry methods M. Imbrišak, A. E. Lovell, M. R. Mumpower Abstract Information geometry is a study of applying differential geometry methods to challenging statistical problems, such as uncertainty quantification. In this work, we use information geometry to study how measurement uncertainties in pre-neutron emission mass distributions affect the parameter estimation in the Hauser-Feshbach fission fragment decay code, CGMF . We quantify the impact of reduced uncertainties on the pre-neutron mass yield of specific masses to these parameters, for spontaneous fission of 252 Cf, first using a toy model assuming Poissonian uncertainties, then an experimental measurement taken from Göök et al., 2014 in EXFOR. We achieved a reduction of up to ∼ 15 % \sim 15\% in CGMF parameter errors, predominantly in w 0 ( 1 ) w_{0}^{(1)} and w 1 ( 0 ) w_{1}^{(0)} . I Introduction The fission of a heavy nucleus is a process in which the parent nucleus undergoes a complex process of splitting into two or more lighter fission fragments. Immediately following the scission of the compound nucleus, prompt neutrons and \gamma rays are emitted on a very short timescale, reaching either a ground state or long-lived isomeric state. On longer timescales, these nuclei can further \beta decay, emitting delayed neutrons and \gamma rays. The emissions of these prompt and delayed neutrons and \gamma rays depend on the initial conditions of the fission fragments, e.g. their mass, charge, excitation energy, spin, and parity. Because of the short timescales of the scission process, these initial conditions–before prompt emission–are impossible to determine directly from experiment and must be inferred through subsequent prompt and delayed measurements or from theory. In nuclear fission reactions, the parent nucleus undergoes a complex process of splitting into two or more lighter nuclei. This process results in a wide distribution of fission fragments characterized by variations in mass, charge, and kinetic energy. The pre-scission phase, which occurs before the actual splitting, has been the focus of extensive theoretical investigations. Different models can be used to describe this phase. For example, in the macroscopic-microscopic approach Randrup and Möller 2011 ; Möller and Ichikawa 2015 ; Mumpower et al. 2020 , the nucleus is treated as a quantum charged liquid drop while the purely microscopic approach Schunck et al. 2014 ; Bulgac et al. 2016 starts from phenomenological descriptions of nucleon-nucleon forces. Recently, Regnier et al. 2016 ; Sierk 2017 ; Usang et al. 2017 ; Regnier et al. 2019 , researchers have made predictions regarding primary fission fragment yields. These predictions take into account the pre-neutron fission fragment mass distribution. The CGMF code Talou et al. 2021 models the de-excitation of fission fragments on an event-by-event basis using the Hauser-Feshbach statistical model of nuclear reactions Hauser and Feshbach 1952 . It samples initial conditions of the fission fragments in mass, charge, total kinetic energy, spin, and parity, that have been parametrized and fit to available experimental data. Prompt neutrons and \gamma rays are emitted probabilistically, and their energies and momenta are recorded for each event, along with the initial conditions of the fission fragments. These event histories can be used to analyze various mean values, distributions, and correlations of neutrons, \gamma rays, and fission fragments. Currently, CGMF models binary fission while accounting for multi-chance fission and the emission of pre-fission neutrons. In parallel with the development of CGMF and other similar codes Verbeke et al. 2015 ; Verbeke et al. 2018 ; Litaize et al. 2015 ; Okumura et al. 2018 ; Lovell et al. 2021 , it is necessary to conduct modern fission experiments (see e.g. Geppert-Kleinrath et al. 2019 ; Göök et al. 2014 ; Hambsch and Oberstedt 1997 ; Hensle et al. 2020 ) to better understand the fission process as well as constrain phenomenological inputs to theoretical models. The impact of the experimental uncertainties on these inputs should be quantified, but to date, there have been very few studies of this nature, e.g. Jaffke et al. 2018 ; Randrup et al. 2019 . However, uncertainty quantification as a whole has been increasing across the nuclear theory community, e.g. Schindler and Phillips 2009 ; Dobaczewski et al. 2014 ; Coello Pérez and Papenbrock 2015 ; Schunk2015 ; McDonnell et al. 2015 ; Furnstahl et al. 2015a ; Furnstahl et al. 2015b ; Wesolowski et al. 2016 ; Melendez et al. 2017 ; Lovell and Nunes 2018 ; Catacora-Rios et al. 2019 ; Phillips et al. 2021 ; Semposki et al. 2022 ; Kejzlar et al. 2023 ; Sharma et al. 2024 . Information geometry is an interdisciplinary field that introduces methods of differential geometry to statistical problems like uncertainty quantification Amari 1982 ; Amari 2016 . While its initial applications focused on machine learning, neural networks Ma et al. 1997 ; Amari 1998 , and to thermodynamics problems Ruppeiner 1979 , its methods have started being successfully applied to various complex problems in biology Transtrum et al. 2015 , chemistry Transtrum 2016 and physics Nikšić and Vretenar 2016 ; Tisanić et al. 2020 . In the context of uncertainty quantification in nuclear physics, information geometry has been used in the framework of nuclear energy density functionals to reduce model parameters and analyze self-consistent models Nikšić and Vretenar 2016 ; Nikšić et al. 2017 ; Imbrišak and Nomura 2023a ; Imbrišak and Nomura 2023b , offering a bridge between microscopic details and statistical insights. Motivated by this successful application in energy density functionals, we investigate its applications to uncertainty quantification of CGMF model parameters. In Sec. II , we present our method for updating measurement errors based on theoretical uncertainties constrained using information geometry. Next, in Sec. III , we give a brief overview of the method that CGMF uses to sample fission fragment mass distributions. Then, we describe the use of these CGMF -sampled distributions to test our method on well-behaved Poissonian errors in Sec. IV.1 and demonstrate its application to real data in Sec. IV.2 . Finally, we conclude in Sec. V . II Model-based measurement error updates We use the following conventions to make the procedure more transparent: (1) the measurements are indexed by letters i,j,\cdots , (2) the model parameters are indexed by Greek letters, (3) the eigenvalues of the parameter covariance matrix are indexed by letters a,b,\cdots , (4) the space of residuals is labeled \mathcal{D} , the space of the model parameters \mathcal{M} and the space of covariance matrix eigenvalues by \Lambda , and (5) the Einstein summation convention is used whenever possible. In the standard minimization procedure, we model a set of N measurements \{y^{i}\}_{i\in\{1,\cdots,N\}}=\{y^{1},\cdots,y^{N}\} with uncertainties \{\sigma^{i}\}_{i\in\{1,\cdots,N\}}=\{\sigma^{1},\cdots,\sigma^{N}\} by a model f^{1}(\theta),\cdots,f^{N}(\theta) described by parameters \{\theta^{\mu}\}_{\mu\in\{1,\cdots,M\}}=\{\theta^{1},\cdots,\theta^{M}\} . We assume that the standardized differences between measurements and model evaluations, i.e. residuals, follow the normal distribution and are identically distributed \displaystyle r^{i}(\theta) \displaystyle=\frac{y^{i}-f^{i}(\theta)}{\color[rgb]{0,0,0}{\sigma^{i}}}\sim\mathcal{N}(0,1)\color[rgb]{0,0,0}{.} (1) The best-fitting parameters \theta are found by minimizing the \chi^{2} value \displaystyle\chi^{2}(\theta) \displaystyle=\frac{1}{2}\sum\limits_{i=1}^{N}\left(r^{i}(\theta)\right)^{2}, (2) which is a result of the standard likelihood maximization technique. This technique also yields a theoretical lower bound to the parameter errors. By the Cramér-Rao bound ( Cramér 1999 , see, e.g.,) , the inverse of the parameter covariance matrix is bounded by the Fisher information matrix Nielsen 2013 \displaystyle g_{\mu\nu} \displaystyle=\color[rgb]{0,0,0}{\sum\limits_{i=1}^{N}\partial_{\mu}\frac{f^{i}}{\sigma^{i}}\partial_{\nu}\frac{f^{i}}{\sigma^{i}},} (3) where we