- Euler’s laws of motion
- Discretization of Euler equations
- Stability of central difference scheme
Both these conditions apply to PDEs for continuous systems (with the possibility of shocks) that have been discretized. However, the discrete element method is discretizes a set of coupled ODEs that apply to each rigid particle in a system. Does it even make sense to talk about the CFL condition for discrete elements?
I suggest we should just use the term “stability condition” for discrete element methods. Even though a system of discrete particles can exhibit wave-like behavior, it is not straightforward to connect the behavior of the system to an equivalent PDE that can be analyzed using the CFL/von Neumann techniques.
First, let us explore the ODEs governing a single rigid body and then examine the numerical stability of the ODE during integration.
Euler’s laws of motion
Discrete element methods model systems of objects as rigid bodies that interact via contact. The motion of a single rigid body with respect to an inertial frame can be described using Euler’s equations of motion:
where is the linear momentum, is the angular momentum, is the mass, is the mass moment of inertia, is the linear velocity, is the angular velocity, is the force, is the torque, and is the time. Note that in this case all quantities are computed with respect to the center of mass of the body.
For numerical purposes, it is common to express the linear and angular velocities in terms of derivatives of spatial position () and orientation () as
Discretization of Euler equations
Note that both Euler equations have the form
where . If we know the initial conditions and , we can integrate the two equations simultaneously using a Velocity-Verlet scheme:
Note that the angular momentum equation is typically expressed in body coordinates to avoid a time-varying moment of inertia and solved accordingly.
A more common approach is to just use a standard central difference approach, i.e.,
Stability of central difference scheme
It is not very easy to explore the stability of the general second-order ODE in the presence of arbitrary external forces and torques. Instead, we can draw upon the large literature on the subject from contact problems in finite element analysis (see Nonlinear Finite Elements for Continua and Structures, Chapter 6, for an excellent exposition) and explore the more relevant problem of contact between two rigid bodies.
For simplicity let us follow the approach taken by Y. T. Feng (On the central difference algorithm in discrete element modelling of impact, IJNME 64:1959-1980, 2005). We assume, as is popular in discrete element methods, that the contact between two rigid bodies can be modeled using a one-dimensional spring-damper system where the spring has a stiffness and the damper has a damping coefficient .
One-dimensional unforced system
If there are no external forcings, the two-body system can be described by the equation
This ODE has solutions of the form
Therefore we have
If the solutions are real and not oscillatory (overdamped). On the other hand, if the solutions are imaginary and therefore oscillatory (underdamped). The critical damping value is given by
where is the natural frequency of the system. Then the original ODE can be written in terms of and as
Stability using the Hurwitz matrix approach
Using a central difference scheme for second derivatives and an upwind scheme for first derivatives, we have
This is a linear difference equation that has solutions of the form (see Recurrence_relation for a proof). Plugging in these solutions, we get
We can now apply the z-transform to this equation because (see Wikipedia article on recurrence relations) to get
We can easily generate the Hurwitz matrix () of this polynomial. Stability requires the leading principal minors of this matrix be positive. In this case, we have
and the stability requirements
For systems with positive damping, the only non-trivial condition above is the second one which leads to a critical timestep of
Variations of this expression are often used to determine the stable time step in DEM calculations.
In the next part of this series we will explore a slightly more realistic model of DEM calculations and see what can be said about the stability of that model.
If you have questions/comments/corrections, please contact banerjee at parresianz dot com dot zen (without the dot zen).
📅 16.04.2017 📁 DEM