280642 research outputs found
Sort by
RHINE: R-process Heating Implementation in hydrodynamic simulations with NEural networks
No full text# Overview RHINE (R-process Heating Implementation in hydrodynamic simulations with NEural networks) is a machine-learning based Fortran code for modeling r-process heating in astrophysical hydrodynamic simulations. The code uses trained neural networks to provide fast and accurate estimates of r-process related rates of change for a set of characteristic quantities. RHINE is designed for integration into hydrodynamic simulations of astrophysical environments with r-process viable conditions such as neutron-star mergers. While maintaining a high accuracy comparable to detailed nuclear reaction networks, it avoids their large computational demands. # Features - Provides nuclear rates of change for the electron fraction, mass fractions of neutrons, protons, alphas, and remaining ('heavy') nuclei, average mass number of heavy nuclei, average mass excess per baryon, and energy-loss rates due to neutrino emission from beta-decays.- Trained on large datasets from full nuclear network calculations.- Lightweight and fast: neural network inference is orders of magnitude faster than full nuclear reaction networks. # Modules RHINE contains two modules `RHINE_neural()` and `RHINE_model()`. 1. Module `RHINE_neural()` provides - subroutine `load_model()` to load specific neural network, - function `forward()` to evaluate a neural network, i.e. obtain the output for given input quantities, - functions `scale_data()` and `rescale_data()` for scaling and rescaling functions for data standardization, - functions `expo()`, `sigmoid()`, and `elu()` for the activation functions. 2. Module `RHINE_model()` provides - subroutine `load_model_RHINE()` to load all neural networks, - subroutine `run_RHINE()` to provide all source terms used to update the hydro quantities within a time step dt, - subroutine `get_derivative()` to predict all variables and time derivatives given all input quantities, - subroutine `get_QSE()` to predict the abundances in QSE regime given the density, temperature, and electron fraction, - subroutine `get_ma()` to predict the mass excess and fraction of neutrino losses, - subroutine `normalization()` to ensure physical consistency of the source terms such as mass and charge conservation. # Usage 1. Load the model ```fortranuse RHINE_model call load_model_RHINE( )``` 2. Predict the source terms ```fortrancall run_RHINE( )``` 3. See also more detailed demonstration in `demo.f90`
Experiences from the CBM collaboration: CAD to ROOT conversion for Detector Geometries
No full textFully automated conversion from CAD geometries directly into their ROOT geometry equivalents is a topic of wide interest in particle physics experiment communities for some time. Tessellation of the surface of an intricate geometry is a powerful approach towards this goal, by potentially providing a shared geometrical representation with very good convergence even for the case of complex geometries. However, using tessellated geometries also requires significant computational effort for particle tracking inside and through tessellated objects.In this paper, we first discuss the experiment and the methodology involved in tessellation and conversion. We report on the application and first experience of using two different software approaches. The two tools, VecGeom and TGeoArbN, were used for simulation of the same tessellated subdetector component. Our observations in this simulation with respect to obtained results and simulation speed are reported along with our general observation about the handling of these tools
Criticality analysis of nuclear binding energy neural networks
No full textMachine learning methods, in particular deep learning methods such as artificial neural networks (ANNs) with many layers, have become widespread and useful tools in nuclear physics. However, these ANNs are typically treated as ‘black boxes’, with their architecture (width, depth, and weight/bias initialization) and the training algorithm and parameters chosen empirically by optimizing learning based on limited exploration. We test a non-empirical approach to understanding and optimizing nuclear physics ANNs by adapting a criticality analysis based on renormalization group flows in terms of the hyperparameters for weight/bias initialization, training rates, and the ratio of depth to width. This treatment utilizes the statistical properties of neural network initialization to find a generating functional for network outputs at any layer, allowing for a path integral formulation of the ANN outputs as a Euclidean statistical field theory. We use a prototypical example to test the applicability of this approach: a simple ANN for nuclear binding energies. We find that with training using a stochastic gradient descent optimizer, the predicted criticality behavior is realized, and optimal performance is found with critical tuning. However, the use of an adaptive learning algorithm leads to somewhat superior results without concern for tuning and thus obscures the analysis. Nevertheless, the criticality analysis offers a way to look within the black box of ANNs, which is a first step towards potential improvements in network performance beyond using adaptive optimizers
Ground-state energy of the quantum-mechanical harmonic oscillator: a method and apparatus for direct high-precision measurements in Penning traps
No full textExperimental study of production of nuclei heavier than in a copper target bombarded with 200 MeV/u ions
No full text
