29. Integration schemes for Cauchy problems#

OLD-Contents

Schemes.

  • Euler methods, CN

  • RK

  • Multi-step: AB, AM, BDF,…

  • Other methods:

    • Newmark-beta, Verlet (Leapfrog),…

Concepts.

  • Convergence:

  • Consistence:

  • Stability:

    • zero-stability

    • A-stability

    • L-stability

  • Stiffness

Theoretical results

  • Dahlquist barriers

  • Lax-Richtmyer

References

Many models of real-life systems are governed by Ordinary Differential Equations (ODEs). While simple linear equations yield elegant analytical solutions, as shown in sections about Linear ODEs with constant coefficients and Linear ODEs in state-space representation, real-world engineering problems demand numerical approximations, as no practical closed-form solution exists.

The choice of a time-integration scheme is far more than a software implementation detail; it represents a fundamental trade-off between accuracy, computational cost, and fidelity: a naive choice of solver can damp out physically vital high-frequency oscillations, generate artificial energy out of nothing, or completely explode due to numerical instability.

This chapter tries to provide some rigorous mathematical foundation needed to analyze, select, (and build? Do we need any new algorithm? Do not re-invent the wheel!) robust time-integration schemes.

Contents.

Foundations of numerical integration of initial value problems

First, the canonical problem1

\[\begin{split} \left\{\begin{aligned} & \dot{\mathbf{y}} = \mathbf{f}(t, \mathbf{y}(t)) \\ & \mathbf{y}(0) = \mathbf{y}_0 \end{aligned}\right. \end{split}\]

is discussed, along with the definition of

  • local truncation error and consistency

  • global truncation error and convergence

Common integration schemes

For first-order equation.

  • One-step methods(I):

    • Explicit: Explicit Euler, Heun (special case of RK)

    • Implicit: Implicit Euler, Crank-Nicolson

  • One-step methods(II): Runge-Kutta (RK) methods

  • Linear multi-step methods:

    • Adams-Bashforth (AB)

    • Adams-Moulton (AM)

    • Backward Differentiation Formulas (BDF)

Stability analysis

Linear stability on test equations, \(\dot{y} = \lambda y\) for the first-order system, \(\ddot{y} = \dots\) for the second-order system. Through one-step transfer function \(G(z)\), \(z = \lambda \, dt\), \(y_{n+1} = G(z) y_{n}\),

  • Stability region of numerical methods

  • 0-ztability

  • A-stability

  • L-stability

Theorems and limits

  • Dahlquist barriers

  • Lax theorem: consistency + stability = convergence

Integration and conservation laws

Some physical systems satisfy some conservation laws. As an example, Hamiltonian systems satisfy energy conservation or - equivalently - volume conservation in phase space.

Simplectic integrators are numerical methods that preserves these conservation laws of the original system. …

Examples

  • Example: torsional dynamics of an elastic beam: finite element method of the continuous model is built; spectral decomposition of the discrete problem is discussed and compared with the eigenvalues of the continuous system; diagonal structural damping is introduced; previous results are used for an aware numerical integration in time of the dynamics of the beam subject to a harmonic torsional moment, using fast mode truncation and their static recovery.

1

A second-order canonical problem \( \mathbf{M} \ddot{\mathbf{y}} + \mathbf{C} \dot{\mathbf{y}} + \mathbf{K} \mathbf{y} = \mathbf{f}(t, \mathbf{y}(t))\), with proper initial conditions, is also discussed. This problem often arises with some interesting properties - symmetry of matrices - usually arises from a Lagrangian approach to classical mechanics systems. Some numerical methods targeting this form of equations exist, like [Leapfrog], [Verlet], or [Newmark] methods.