Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

Newton–Raphson method - Finite difference method

  1. May 11, 2009 #1

    I am trying to solve a nonlinear differential equation with the use of the finite difference method and the Newton-Raphson method. Is there any one that knows where I can find some literature about the subject?

    I am familiar with the use of the finite difference method, when solving linear differential equations. It is the Newton-Raphson method when using the finite difference method that is new for me.

    Thanks in advance
  2. jcsd
  3. May 12, 2009 #2
    Hello Excom,

    finite difference methods (simple one-step methods such as Euler, Trapezoid, Midpoint, or more complex multi-step methods like the Adams' families, or non-linear methods such as Runge-Kutta, etc etc) can ALL be used to solve both linear and non-linear ordinary differential equations (obviously depending on the kind of differential system there are methods that will perform better than others..) but they're all used for solving a general IVP of the form:


    \left. \begin{array}{l}
    \frac {dy} {dx} = f(x,y) \\
    y( x_{0} ) = y_{0}
    \end{array} \right\} \mbox{ze IVP :p}


    (which may be a scalar equation or a system of equations), regardless of whether f is linear or not.

    Newton-Raphson is for solving non-linear algebraic equations, not differential equations. You will have to use Newton-Raphson (or any other technique for solving non-linear equations) within your finite difference method if the said method is implicit, that is, to solve for the current time-step of the solution as a function of the values at previous time-steps. For example, Adams-Moulton methods are implicit so you will have to solve a non-linear algebraic equation (or system of equations) at each time-step. . . but Forward Euler or Trapezoid or even Runge-Kutta or Adams-Bashforth are all explicit difference methods, and there's no need to solve non-linear equations within the method, so no need for Newton-Raphson :)

    If you're still interested in Newton-Raphson, there are loads of resources on the net, just search on google:) eg one link I found:
    http://www.math.ubc.ca/~clarkson/newtonmethod.pdf" [Broken]

    Hope I could be of help, good luck with your non-linear differential equation! xD
    Last edited by a moderator: May 4, 2017
  4. May 15, 2009 #3
    Tanks for your help
  5. Aug 16, 2009 #4
    please give me
  6. Mar 19, 2010 #5
    In my case, you can approximate the denominator term,
    with a forward, backward, or central difference.

    So, just to elaborate, if you have,
    f(x) = x^2
    f(x+h) = (x+h)^2
    f(x-h) = (x-h)^2
    for some small h (gridspace)

    Then, using Central Difference your Newton-Raphson equation becomes,
    x[i+1] = x - f(x)/f'(x)
    = x - f(x) / ( (f(x+h)-f(x-h) )/(2*h) )
    = x - x^2/( ( (x+h)^2 - (x-h)^2 ) / (2*h) )

    For other example, e.g. f(x) = x^4 + x^3 + x + 5, I'm getting faster convergence via the finite difference version of f'(x) than using the analytical version of it.

    I have not encounter any reference for this, but I don't see anything wrong with this.

    All the best! :)
  7. Jun 29, 2011 #6
    Take a look at this I dont know if it can help you :
    Last edited by a moderator: Apr 26, 2017
Share this great discussion with others via Reddit, Google+, Twitter, or Facebook