| Secondary electron emission as a result of electron-matter interaction is of major concern for many research and technical applications including scanning electron microscopy,spacecraft industry,high energy accelerators and radiation biology.However,even a century after the discovery of secondary electron emission it is still hard to find reliable experimental data due to large deviations in the measured data.In addition,the conventional theoretical approaches failed to describe material dependent yield and no formula exists for the calculation of absolute yield.The available yield formulas were for reduced yield value.Therefore,it exerts a tremendous necessity for a new approach with high accuracy and efficiency to predict the secondary electron yield over a sufficiently wide primary energy range and for elemental solids.Furthermore,the Monte Carlo simulation methods,which proved to be much more accurate than the analytical approaches in various disciplines,need precise knowledge regarding the electron-matter interaction.Most of these techniques rely on the availability of data of optical constants or the electron stopping powers for the target media.Dielectric function data is generally not available for many materials.Also,the experimental data of electron stopping powers is available for few materials,which further limits the use of Monte Carlo methods.Machine learning methods have attracted great attention in recent years because of its ability to explore the patterns present in the data.Specifically,the applications of machine learning methods in conjunction with the first principles calculations in material informatics for the discovery and design of new materials as well as for the prediction of material properties suggest the potential of machine learning methods to be used for solid state material informatics.Instead,the use of machine learning methods with experimental data for material’s property exploration will enhance the accuracy of these techniques for a specific problem.This thesis concerns with the applications of Monte Carlo techniques and machine learning methods for the study electron-interaction with matter,specifically the secondary electron emission yields and the electron stopping powers.It introduces the use of machine learning methods with the experimental data for exploring the nature of secondary electron yield of materials as well as the electron stopping powers.Monte Carlo methods have been used to study the electron emission from the biologically most important material,the liquid water.This dissertation is organized as follows:In Chapter 1,we discussed the background,applications and importance of the study.It summarizes the theoretical models and experimental measurements of secondary electron yield and electron stopping power.It also covers the background and research progress for the simulation of electron-water interaction.In Chapter 2,we presented the Monte Carlo simulation methods for the work carried out in this dissertation,i.e.,Classical Monte Carlo Simulation Code(CMC),Geant4-DNA(G4DNA)Monte Carlo Code for simulation of election interaction with water.In Chapter 3,we report a machine learning approach to predict more reliable secondary electron yield from noisy experimental databases for a large number of materials.A unique aspect of this work is that the machine learning approach has been used to remove the uncertainty present in the measured data and to generate a quantitively sufficient yield database for the development of a mathematical expression of secondary electron yield.The machine learning model developed in this study is an accurate model(based on the Leave-one-material-out cross validation(LOOCV)test results)that can predict yields for unseen element over a wide energy range.In Chapter 4,we used the secondary electron yield database generated by machine learning model to develop a mathematical expression for the calculation of absolute secondary electron yield,which solves a long-standing problem in the applied physics.The formula is free of unknown constants and is independent of the experimentally determined maximum yield and is used for the calculation of absolute yield as a function of incident energy and atomic number.It shows the material dependent variation of secondary electron yield with the primary energy.The yield calculated by the universal formula is within the range of the published experimental data.Comparing with previous formulas,our formula has promising use for application.We have extended this expression for the calculation of yield for compound materials.Comparison with experimental data shows that it can be indeed used for the compound materials.The formula can be useful for a large scientific community dealing with applications of electronprobe interactions with matter as well as for those suffering from the detrimental effects related to the emission of secondary electrons.In Chapter 5,we discussed the use of various machine learning techniques for the accurate predictions based on a very small experimental database of electron stopping power.Generally,machine learning methods require large database for precise predictions.However,for electron stopping power,experimental data is available for only few materials.We applied various algorithms individually as well as their assemblies,known as ensembles,to enhance the prediction accuracy.Based on the model’s performance evaluation tests,we concluded that the ensemble machine learning methods are more accurate than the individual algorithms,especially for the cases where the training database is much smaller.Using this method,we were able to predict the electron stopping power for the elements that were present in the training database as well as for elements beyond the training database over a wide energy range.In Chapter 6,we presented a study of the secondary electron emission from liquid water.Liquid water has a great research importance in radiation biology because it is believed to be tissue equivalent.G4DNA is an open-source Monte Carlo code which is being widely used for the simulation of radiation interaction with water and few other biological materials.In this study we extended our CMC code for the simulation of electron interaction with liquid water and then compared the CMC code with G4DNA code for the prediction of secondary electron yield from liquid water.We have found that,compared to CMC and experimental data,G4DNA underestimates the secondary electron yield.This is due to the reason that in G4DNA the electron inelastic scattering modeling at the low energies is inadequate and generally a cut is employed in the simulation of electron trajectories below 10 eV.But,these low energy electrons are of particular importance for the simulation of radiation damage at cellular level.Therefore,we suggest to include a full optical data down to meV energy-loss range to improve G4DNA.A summary of this thesis is provided in the Chapter 7. |
