mechanical engineering question
chapter 21 of advanced engineering mathematics 10th edition
Question: Discuss error at two of the methods taught in this chapter. Why does the error behave differently and what are the computational costs of reducing error?
You can use error at Euler's method and Improved Euler's method to answer the question. thank you
Experience with Euler’s1 method raises some interesting
questions about numerical approxima-
tions for the solutions of differential equations.
1. What determines the amount of numerical error in an
approximation?
2. Why is that halving the step size tends to decrease the
numerical error in Euler’s method by one-half and the numerical
error in the modified Euler method by one-quarter?
3. Are some differential equations more difficult to approximate
numerically than others? If so, can this be predicted without doing
numerical experiments?
None of these questions has a simple answer, but we will at least
be able to offer a partial answer for each.
Any approximation of a function necessarily allows a possibility
of deviation from the correct value of the function. Error is the
term used to denote the amount by which an approximation
fails to equal the exact solution (exact solution minus
approximation).2 Error occurs in an approximation for several
reasons.
Truncation error in a numerical method is error that is caused
by using simple approximations to represent exact mathematical
formulas. The only way to completely avoid truncation
error is to use exact calculations. However, truncation error can
be reduced by applying the same approximation to a larger number of
smaller intervals or by switching to a better approx-
imation. Analysis of truncation error is the single most important
source of information about the theoretical characteristics that
distinguish better methods from poorer ones. With a com-
bination of theoretical analysis and numerical experiments, it is
possible to estimate truncation error accurately.
Round-off error in a numerical method is error that is caused by
using a discrete number of significant digits to represent real
numbers on a computer. Since computers can retain a large
number of digits in a computation, round-off error is problematic
only when the approximation requires that the computer subtract two
numbers that are nearly identical. This is exactly
what happens if we apply an approximation to intervals that are too
small. Thus, the effort to decrease truncation error can have the
unintended consequence of introducing significant
round-off error.
Practitioners of numerical approximation are most concerned with
truncation error, but they also try to restrict their efforts at
decreasing truncation error to improvements that do
not introduce significant round-off error. In our study of
approximation error, we consider only truncation error. We seek
information about error on both a local and global scale.
Local
truncation error is the amount of truncation error that occurs in
one step of a numerical approximation. Global truncation error is
the amount of truncation error that occurs in the
use of a numerical approximation to solve a problem.
Taylor’s Theorem and Approximations
The principal tool in the determination of truncation error is
Taylor’s theorem, a key result in calculus that provides a formula
for the amount of error in using a truncated Taylor series
to represent the corresponding function. Taylor’s theorem in its
most general form applies to approximations of any degree; here we
present only the information specifically needed to
analyze the error in Euler’s method.
Theorem 1 Let y be a function of one variable having a
continuous second derivative on some
interval I = [0, tf ] and let f be a function of two variables
having continuous first partial
derivatives on the rectangular region R = I × [y0 − r, y0 + s],
with r, s > 0.
1. Given T and t in I, there exists some point τ ∈ I such that
.................. (1)
2. Given points (T, Y ) and (T, y) in R, there exists some point
(T, η) ∈ R such that
f(T, y) = f(T, Y ) + fy(T, η)(y − Y
),................................. (2)
where fy is the partial derivative of f with respect to its second
argument.
Theorem 1 serves to quantify the idea that the difference in
function values for a smooth function should vanish as the
evaluation points become closer.
One can be a little more restrictive when specifying the range of
possible values for τ and η; however, nothing is gained by doing
so. We cannot use Theorem 1 to compute the error in
an approximation. The theorem provides formulas for the error, but
the catch is that there is no way to determine τ and η without
knowing the exact solution. It may seem that this catch
makes the formulas useless, but this is not the case. We do know
that τ and η are confined to a given closed interval, and therefore
we can compute worst-case values for the quantities y"(τ )
and fy(T, η).
Example 1
Suppose we want to approximate the function ln(0.5 + t) near the
point T = 0.5. We have
y(t) = ln(0.5 + t), y(T) = 0, y'(t) = 1 /(0.5 + t) , y'(T) = 1,
y"(t) = −1/(0.5 + t)^2
.
Let’s assume that our goal4
is to find an upper bound for the largest possible error in
approximating
y(t) by (t − 0.5) with 0 ≤ t ≤ 1. The approximation formula of
Equation (1) yields
and the approximation error E is defined by
E = (t − 0.5) − y(t).
We therefore have
with 0 ≤ τ ≤ 1. Now let
Given the range of possible τ values, the worst case is |y"(τ )| =
M; thus,
|E| ≤ 2(t − 0.5)2 , t ∈ [0, 1].
Local Truncation Error for Euler’s Method
Consider an initial value problem
y'= f(t, y(t)), y(0) = y0,
where f has continuous first partial derivatives on some region R
defined by 0 ≤ t ≤ tf and
y0 − r ≤ y ≤ y0 + s. Euler’s method is a scheme for obtaining an
approximate value yn+1 for
y(tn+1) using only the approximation yn for y(tn) and the
function f that calculates the slope
of the solution curve through any point. Specifically, the method
is defined by the formula
yn+1 = yn + hf(tn, yn), where h = tn+1 − tn.
We define the global truncation error at step n in any numerical
approximation of (3) by
En = y(tn) - yn: .............. (5)
Our aim is to find a worst-case estimate of the global truncation
error in the numerical scheme for Euler's method, assuming that the
correct solution and numerical approximation stay within
R. This is a difficult task because we have so little to work with.
We'll start by trying to determine the relationship between the
error at time tn+1 and the error at time tn. Figure 1
shows the relationships among the relevant quantities involved in
one step of the approximation. The approximation and exact solution
at each of the two time steps are related by the error definitions.
The approximations at the two time steps are related by Euler's
method. We still need to have a relationship between the exact
solution values at the two time steps, and this is
where Theorem 1 is needed.
The comparison of y(tn+1) and y(tn) must begin with Equation
(1). Combining Equation (1), with T = tn, and Equation (4)
yields
y(tn+1) = y(tn) + hf(tn; y(tn)) + (h^2/2)y"(t ): ..............
(6)
This result is a step in the right direction, but it is not yet
satisfactory. A useful comparison
of y(tn+1) with y(tn) can have terms consisting entirely of known
quantities and error terms,
but not quantities, such as f(tn; y(tn)), that need to be evaluated
at points that are not known
exactly. These quantities have to be approximated by known
quantities. This is where Equation
(2) comes into the picture. Substituting
f(tn, y(tn)) = f(tn, yn) + fy(tn, η)[y(tn) − yn] = f(tn, yn) −
fy(tn, η)En .into Equation (6) gives us
y(tn+1) = y(tn) + hf(tn, yn) − hfy(tn, η)En +(h^2/2)y"(τ )
................... (7)
Equation (7) meets our needs because every term is either a
quantity of interest, a quantity
that can be evaluated, or an error term. Subtracting this equation
from the Euler formula (4)
yields
yn+1 − y(tn+1) = yn − y(tn) + hfy(tn, η)En −(h^2/2)y"(τ ),
or
En+1 = [1 + hfy(tn, η)]En −(h^2/2)y"(τ ) ................ (8)
This result indicates the relationship between the errors at
successive steps. The local truncation error is defined to be the
error in step n + 1 when there is no error in step n; hence,
the local truncation error for Euler’s method is (−h^2y"(τ ))/2.
The local truncation error has
two factors of h, and we say that it is O(h^2) The quantity [1 +
hfy(tn, η)]En represents the error at step n + 1 caused by the
error at step n. This propagated error is larger than En if
fy > 0 and smaller than En if fy < 0. Having fy < 0 is
generally a good thing because it causes truncation errors to
diminish as they propagate...............
for full pdf comment your email id .....thanks
Get Answers For Free
Most questions answered within 1 hours.