Exercise05_sol.pdf

4
- 1 - AMS 147 Computational Methods and Applications Exercise #5 1. Consider the total error of the first order numerical differentiation method in the IEEE double precision system E T h ( = = fl fx + h ( = ( = fl fx ( = ( = h Numerical value obtained with finite precision f x ( = Exact value By selecting a proper step size h we can achieve the minimum total error. What is the order of magnitude of the optimal step size? What is the order of magnitude of the minimum total error? Answer: The optimal step size is about 10 -8 : arg min h E T h (= 10 8 The minimum total error is about 10 -8 : min h E T h (= 10 8 2. Let Th ( = be the numerical approximation to I = fx ( =dx a b obtained using the composite trapezoidal rule with step size h. The error is defined as Error h ( = = Th ( = I . Suppose ƒ(x) is infinitely differentiable. What is the order of Error(h)? When the exact value I is unknown, we cannot calculate the error using Error h ( = = Th ( = I . Write out how to use Th ( = and T h 2 to estimate Error h ( = . Answer: Error h (= = Oh 2 ( = Error h (= Th (= T h 2 1 1 2 2

Transcript of Exercise05_sol.pdf

  • - 1 -

    AMS 147 Computational Methods and Applications

    Exercise #5

    1. Consider the total error of the first order numerical differentiation method in the IEEE double

    precision system

    ET h( ) =fl f x + h( )( ) fl f x( )( )

    hNumerical value

    obtained withfinite precision

    f x( )Exactvalue

    By selecting a proper step size h we can achieve the minimum total error.

    What is the order of magnitude of the optimal step size?

    What is the order of magnitude of the minimum total error?

    Answer:

    The optimal step size is about 10-8:

    argminh

    ET h( ) 108

    The minimum total error is about 10-8:

    minh

    ET h( ) 108

    2. Let T h( ) be the numerical approximation to I = f x( )dxa

    b

    obtained using the composite

    trapezoidal rule with step size h. The error is defined as Error h( ) = T h( ) I .

    Suppose (x) is infinitely differentiable. What is the order of Error(h)?

    When the exact value I is unknown, we cannot calculate the error using

    Error h( ) = T h( ) I . Write out how to use T h( ) and Th2

    to estimate Error h( ) .

    Answer:

    Error h( ) = O h2( )

    Error h( )T h( ) T

    h

    2

    112

    2

  • AMS 147 Computational Methods and Applications

    - 2 -

    3. Let T h( ) be the numerical approximation to I = f x( )dxa

    b

    obtained using the composite

    Simpsons rule with step size h. The error is defined as Error h( ) = T h( ) I .

    Suppose (x) is infinitely differentiable. What is the order of Error(h)?

    When the exact value I is unknown, we cannot calculate the error using

    Error h( ) = T h( ) I . Write out how to use T h( ) and Th2

    to estimate Error h( ) .

    Answer:

    Error h( ) = O h4( )

    Error h( )T h( ) T

    h

    2

    112

    4

    4. Consider the Euler method for solving y = F y,t( ) .

    Write out the Euler method.

    Is the Euler method explicit or implicit?

    What is the order of the local truncation error of the Euler method?

    What is the order of the global error of the Euler method?

    Answer:

    yn+1 = yn + hF yn , tn( )

    Euler method is explicit.

    Local truncation error = O(h2).

    Global error = O(h)

    5. Consider the backward Euler method for solving y = F y,t( ) .

    Write out the backward Euler method.

  • AMS 147 Computational Methods and Applications

    - 3 -

    Is the backward Euler method explicit or implicit?

    What is the order of the local truncation error of the backward Euler method?

    What is the order of the global error of the backward Euler method?

    Answer:

    yn+1 = yn + hF yn+1, tn+1( )

    Backward Euler method is implicit.

    Local truncation error = O(h2).

    Global error = O(h)

    6. Let y t( ) be the exact solution of y = F y,t( )

    y 0( ) = y0

    .

    Let yn h( ) be the numerical approximation to y nh( ) obtained using the Euler method with

    step size h. The error is Error h( ) = yn h( ) y nh( ) .

    Since the Euler method is first order accurate, we have

    yn h( ) y nh( ) = c1h + o h( )

    y2nh

    2

    y nh( ) = c1

    h

    2+ o h( )

    When the exact solution y nh( ) is unknown, we cannot calculate the error using

    Error h( ) = yn h( ) y nh( ) . Write out how to use yn h( ) and y2nh2

    to estimate Error h( ) .

    Answer:

    Error h( )yn h( ) y2n

    h

    2

    112

    7. Let y t( ) be the exact solution of y = F y,t( )

    y 0( ) = y0

    .

  • AMS 147 Computational Methods and Applications

    - 4 -

    Let yn h( ) be the numerical approximation to y j h( ) obtained using RK4 (the classical fourth

    order Runge-Kutta method) with step size h. The error is Error h( ) = yn h( ) y nh( ) .

    Since RK4 is fourth order accurate, we have

    yn h( ) y nh( ) = c4h4

    + o h( )

    y2nh

    2y nh( ) = c4

    h

    2

    4

    + o h( )

    When the exact solution y nh( ) is unknown, we cannot calculate the error using

    Error h( ) = yn h( ) y nh( ) . Write out how to use yn h( ) and y2nh2

    to estimate Error h( ) .

    Answer:

    Error h( )yn h( ) y2n

    h

    2

    112

    4