Scribd Logo
Upload

Welcome to Scribd!

  • 55 views

    Lec15 PDF

    Uploaded by

    Milind Jain
    This document discusses numerical methods for solving initial value problems (IVPs) of ordinary differential equations (ODEs), including: 1. One-step methods like the explicit and implicit Euler methods and Runge-Kutta methods. Convergence requires consistency and stability. 2. Multi-step methods like Adams-Bashforth and Adams-Moulton methods. A consistent multi-step scheme converges if and only if it is stable. 3. The concept of stability is important for convergence - a method must bound errors from small perturbations. Consistency alone is not enough to guarantee convergence.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Lec15 PDF

    Uploaded by

    Milind Jain
    55 views83 pages
    This document discusses numerical methods for solving initial value problems (IVPs) of ordinary differential equations (ODEs), including: 1. One-step methods like the explicit and implicit Euler methods and Runge-Kutta methods. Convergence requires consistency and stability. 2. Multi-step methods like Adams-Bashforth and Adams-Moulton methods. A consistent multi-step scheme converges if and only if it is stable. 3. The concept of stability is important for convergence - a method must bound errors from small perturbations. Consistency alone is not enough to guarantee convergence.

    Original Title

    lec15.pdf

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document discusses numerical methods for solving initial value problems (IVPs) of ordinary differential equations (ODEs), including: 1. One-step methods like the explicit and implicit Euler methods and Runge-Kutta methods. Convergence requires consistency and stability. 2. Multi-step methods like Adams-Bashforth and Adams-Moulton methods. A consistent multi-step scheme converges if and only if it is stable. 3. The concept of stability is important for convergence - a method must bound errors from small perturbations. Consistency alone is not enough to guarantee convergence.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    55 views83 pages

    Lec15 PDF

    Uploaded by

    Milind Jain
    This document discusses numerical methods for solving initial value problems (IVPs) of ordinary differential equations (ODEs), including: 1. One-step methods like the explicit and implicit Euler methods and Runge-Kutta methods. Convergence requires consistency and stability. 2. Multi-step methods like Adams-Bashforth and Adams-Moulton methods. A consistent multi-step scheme converges if and only if it is stable. 3. The concept of stability is important for convergence - a method must bound errors from small perturbations. Consistency alone is not enough to guarantee convergence.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    You are on page 1of 83

    ODE: IVP problems

    Finite difference
    approximation
    One-step methods
    * Euler explicit method
    * Euler implicit method
    * Runge-Kutta methods
    Euler mehtod: first order one- step method for the IVP
    y’(t)=f(t,y), y(a)=c, t Є[a, b]
    Example:
    let us denote the approximate solution by ȳ(t)
    When do we say the numerical scheme is convergent?

    There are three type of solutions:


    1. Exact solution of the ODE
    2. Exact solution of the difference equation ( obtained by hypothetical infinite precision
    computer)
    3. Solution obtained using computer with finite precision
    What is the secret behind this convergence??

    Is that because the truncation error appeared in the Taylor’s goes to zero as
    h goes to zero?

    impact of round-off error,


    The main ingredients involved are
    truncation error, global error, consistency,
    stability.
    Let us define more specifically what is truncation error:
    In any case we need to write Truncation error at t_i=exact ODE at t_i +
    O(h), this gives an idea why we divide by h.
    :( abuse of notations here ! T_k to d_i
    (Here difference equation is solved in infinite precision)
    Impact of round off error in Euler one-step method. ( difference
    equation solved in finite precision )

    Abuse of notations!!!
    Does Euler method converge because of consistency???

    Or whether the gloabl error goes to zero as truncation error goes to zero
    always??? ( its crazy!!)

    The following example will make you think!!


    It is not convergeing !!
    Does the above method converge theoretically? We have seen any way
    numerically not!

    Once you have verified that your method is convergent, is it secure to


    approach a computer to perform the computation? Or are you sure you will
    get a convergent solution in the computer?
    Therefore the conistency is not enough for convergence!!!

    Some additional property is needed for the convergence. We


    identify it as “Stability”.

    In fact there are many notions of stability.

    A convincing one in this context is the following:

    Definition: (stability) A finite difference scheme is stable if the error produced by


    a small perturbation in the difference scheme remains bounded.
    Ref: Kendall E. Atkinson pp 349

    Does this stability analysis ensure a bound on the round-off error produced
    during the computation of the numerical solution?

    In fact we do not consider the whole round-off error analysis, we only consider
    the perturbation in the initial datum.

    There are several articles available dealing with round-off error analysis:

    Sylvie Boldo, Florian Faissole, Alexandre Chapoutot “Round-off Error


    Analysis of Explicit One-Step Numerical Integration Methods, HAL 2017.
    Order of accuracy:
    Zero stability: Abuse of notations!
    Absolute stability specifies upto what level we need to reduce the size of
    h to get a well behaved approximate solution. We have seen even though
    method is convergent and zero stable the method produces unrealistic
    solution for coarse mesh.
    For linear problem the bound on round-off error is obtained
    through the bound in exact solution of difference equation.
    For the ode y’(t)=f(t,y), y(0)=c, The Eulers
    ’ method so far discussed is stable
    with respect to any of the notions.
    So far we have discussed the one-step explicit Euler method.
    Conditionally Stable Unconditionally Stable
    Converging solution for a Diverging solution for an
    STABLE ODE UNSTABLE ODE
    Abuse of notations again :( t_n to x_n and [a,b] to [x_0, X_M]
    Consistency again!

    Note: Theorem 12.2 can be used to find the error bound on the approximated
    solution of a given one step method, provided the conditions of theorem 12.2 are
    prescribed.
    One-step methods of high order accuracy
    One-step method of high order accuracy: One-step implicit method or
    trapezium rule method
    How to show the convergence of this method? Or how to
    find an error bound of the solution? Therorem 12.2 will
    help.
    More on one-step methods:
    (Or midpoint method)
    So far in this slide we were discussing about one-step method. We have
    compared it with its general form. We found implicit and explicit methods
    and high order methods.

    Somehow the consistency+ the concept of stability helps to


    prove the convergence!!

    Runge kutta methods are derived in another way as well!!

    For multi-step method we strict to one definition of stability and


    prove a consitent scheme is stable if and only if it is convergent
    One-step high order method conitnued...
    Taylor method
    Multi-step methods

    * Adams-Bashforth methods
    * Adams-Moulton methods
    * Predictor-corrector methods
    Consistency, Stability and Convergence of Multi-step
    methods.
    This theorem connects two notions of stability!
    For proof refer to the book of Henrici in the previous slide
    For proof refer to the book of Henrici in the previous slide
    Where f_n=f(t_n,y(t_n)).

    You might also like