Jack Hale

Containers for Portable, Productive, and Performant Scientific Computing

Containers are an emerging technology that holds promise for improving productivity and code portability in scientific computing. The authors examine Linux container technology for the distribution of a nontrivial scientific computing software stack and its execution on a spectrum of platforms from laptop computers through high-performance computing systems. For Python code run on large parallel computers, the runtime is reduced inside a container due to faster library imports. The software distribution approach and data that the authors present will help developers and users decide on whether container technology is appropriate for them. The article also provides guidance for vendors of HPC systems that rely on proprietary libraries for performance on what they can do to make containers work seamlessly and without performance penalty.

Calculating the Malliavin derivative of some stochastic mechanics problems

The Malliavin calculus is an extension of the classical calculus of variations from deterministic functions to stochastic processes. In this paper we aim to show in a practical and didactic way how to calculate the Malliavin derivative, the derivative of the expectation of a quantity of interest of a model with respect to its underlying stochastic parameters, for four problems found in mechanics. The non-intrusive approach uses the Malliavin Weight Sampling (MWS) method in conjunction with a standard Monte Carlo method. The models are expressed as ODEs or PDEs and discretised using the finite difference or finite element methods. Specifically, we consider stochastic extensions of; a 1D Kelvin-Voigt viscoelastic model discretised with finite differences, a 1D linear elastic bar, a hyperelastic bar undergoing buckling, and incompressible Navier-Stokes flow around a cylinder, all discretised with finite elements. A further contribution of this paper is an extension of the MWS method to the more difficult case of non-Gaussian random variables and the calculation of second-order derivatives. We provide open-source code for the numerical examples in this paper.

The Fluid Mechanics of Membrane Filtration

Reverse osmosis and nanofiltration membranes remove colloids, macromolecules, salts, bacteria and even some viruses from water. In crossflow filtration, contaminated water is driven parallel to the membrane, and clean permeate passes through. A large pressure gradient exists across the membrane, with permeate flow rates two to three orders of magnitude smaller than that of the crossflow. Membrane filtration is hindered by two mechanisms, concentration polarization and caking. During filtration, the concentration of rejected particles increases near the membrane surface, forming a concentration polarization layer. Both diffusive and convective transport drive particles back into the bulk flow. However, the increase of the apparent viscosity in the concentration polarization layer hinders diffusion of particles back into the bulk and results in a small reduction in permeate flux. Depending on the number and type of particles present in the contaminated water, the concentration polarization will either reach a quasi-steady state or particles will begin to deposit onto the membrane. In the later case, a cake layer eventually forms on the membrane, significantly reducing the permeate flux. Contradictive theories suggest that the cake layer is either a porous solid or a very viscous (yield stress) fluid. New and refined models that shed light on these theories are presented.Copyright

A volume-averaged nodal projection method for the Reissner-Mindlin plate model

Abstract We introduce a novel meshfree Galerkin method for the solution of Reissner–Mindlin plate problems that is written in terms of the primitive variables only (i.e., rotations and transverse displacement) and is devoid of shear-locking. The proposed approach uses linear maximum-entropy basis functions for field variables approximation and is built variationally on a two-field potential energy functional wherein the shear strain, written in terms of the primitive variables, is computed via a volume-averaged nodal projection operator that is constructed from the Kirchhoff constraint of the three-field mixed weak form. The meshfree approximation is constructed over a set of scattered nodes that are obtained from an integration mesh of three-node triangles on which the meshfree stiffness matrix and nodal force vector are numerically integrated. The stability of the method is rendered by adding bubble-like enrichment to the rotation degrees of freedom. Some benchmark problems are presented to demonstrate the accuracy and performance of the proposed method for a wide range of plate thicknesses.