Click here to load reader
Upload
selamitsp
View
4
Download
0
Embed Size (px)
DESCRIPTION
EulerMethod Excel Implementation Elementary Differential Equations
Citation preview
7/21/2019 EulerMethod Excel Implementation Elementary Differential Equations
http://slidepdf.com/reader/full/eulermethod-excel-implementation-elementary-differential-equations 1/7
Math 240 Home, Textbook Contents, Online Homework Home
Geometric Interpretation - Numerical Methods
Initial Discussion
So far we have learned some algebraic techniques for solving first order differential equations of
various special forms. We will learn more techniques in the future, but we will still be restricted to
solving just selected equations of special form. Most first order equations can't be solved explicitly,even in integral form. That being the case, it is often most useful to have a numerical technique to
approximate the solution. We will consider the two simplest numerical methods for approximating
solutions, Euler's method and the improved Euler's method. These techniques build on the
geometric interpretation of differential equations that we developed in the last lab. We will consider
the example
Geometrically, this solution is the curve
that follows the arrows of the slope field
and passes through the point (0,1). We
can build an approximation to this curveas follows. Suppose we want to
approximate the value of at
Start at the point (0,1). At this point the
slope is So our arrow
points down with slope We follow
this arrow as pictured at the left to get
Since the
exact value of , this
is a reasonably good approximation.
You might observe that this is exactly
the same thing as using the tangent line
approximation from Calculus. And just
like the tangent line approximation, it works
reasonably well as long as your target (in this
case 0.1) is close enough to your starting (in
this case 0). It breaks down if you try to get too
far away from your initial value. For example, if
you try to approximate using this technique,
you get , but the exact value
is . As you can see from the
picture above, the slope field changes at each
point causing the yellow solution curve to bend up
instead of just continuing in the same downwarddirection of the initial arrow. To deal with the
problem of changing slopes, we need to make many small steps following each arrow for a short
distance and then computing the new slope (the new arrow), repeating the process until we reach
our target value. We call the
horizontal length of each short step
the step size in this process. It will
be helpful to introduce some
notation as we start repeating our
approximation process for multiple
steps. For the initial value problem
so the starting point is , we
mentary Differential Equations https://www.math.ksu.edu/math240/book/chap1/numerical.php
7 2.1.2016 14:38
7/21/2019 EulerMethod Excel Implementation Elementary Differential Equations
http://slidepdf.com/reader/full/eulermethod-excel-implementation-elementary-differential-equations 2/7
call our step size and then define
For example, to approximate using a step size of 0.1 we compute
Implementing Euler's Method in Excel
This sort of repetitive calculation is exactly what computers were invented to do. While you can
carry out the assignment below by hand, you will probably want to program this algorithm into a
computer. A spreadsheet is well suited to this process, though if you want to cook up a program in
your favorite language to carry out the computations instead, be my guest.
Open Excel (or your favorite spreadsheet if you are working on your own computer). We'll start by
labeling the columns to keep track of what goes where. Enter x into cell A1, y' into cell B1, y into
cell C1, and h into cell H1. Next put the initial values and step-size into the appropriate starting
spots with 0 in cell A2, 1 in cell C2, and 0. 1 in cell H2. Then put in the formulas with =A2+$H$2 in
cell A3, =2*A2- C2 in cell B3, and =C2+$H$2*B3 in cell C3 (we'll discuss the meaning of the $ signs in
these formulas below). You now have the x column set to add the step size h as you move down
each row, the y' column to compute the slope based on the starting x and y values, and the y
column to add h times y' to the previous y. Now click and drag to select cells A3..C3 and then click
on the small square at the lower right of the box around those cells and drag it down for a dozen
rows or so. This will copy the formulas down and automates the repetitive calculation of each step.
Note that if you click now on cell A4 it shows the formula =A3+$H$2. Comparing this to the original
formula you copied from cell A3, =A2+$H$2, you see the spreadsheet recognizes that as you pull
down the formula, you want to update the reference from cell A2 to cell A3 as you pull down.
However, the reference to cell H2 hasn't changed (which is good because you still wanted toreference the same step-size). This is what the $ signs did - they told the spreadsheet the reference
was absolute, not relative, and shouldn't update when you pulled the formula down.
Naturally, you can use this technique to handle other equations. The only thing you need to edit is
the formula for y' in column B. Do be sure to copy the edited formula down the whole column.
Since our example equation is linear, we can find the exact answer to our initial value
problem is . This allows us to check how close Euler's method comes to
approximating the true solution. In cell D1 of the spreadsheet enter y exact and in cell E1 enter
rel error (which stands for relative error). Then in cell D2 enter =3*exp( - A2) +2*A2- 2 and in cell
E2 enter =( D2- C2) / D2. It will be convenient to right-click on cell E2 and Format Cells as Percentage
since relative error is a percent. Now click and drag to select cells D2 and E2 and drag those
mentary Differential Equations https://www.math.ksu.edu/math240/book/chap1/numerical.php
7 2.1.2016 14:38
7/21/2019 EulerMethod Excel Implementation Elementary Differential Equations
http://slidepdf.com/reader/full/eulermethod-excel-implementation-elementary-differential-equations 3/7
formulas down. This will show you how accurate the approximation produced by Euler's method is.
As you can see, Euler's method isn't particularly accurate. We can improve the approximation by
taking smaller steps. If you change cell H2 to 0. 05 to cut the step size in half, you find that the
error in the value of has been cut from 5.87% to 2.85%. Of course you had to take 10 steps
of size 0.05 to get to 0.5 while it only took 5 steps of size 0.1. So you do twice as much work but
you cut the error roughly in half. This is actually the standard pattern.
The Improved Euler Method (Heun's Method)
Just shrinking the step size isn't always practical as a way to improve accuracy in Euler's method.
Since you use the results from each step in the next round of calculations, roundoff error will build
up as you proceed. If you make the step size too small, you can build up so much roundoff error
that it swamps the "discretization error" built into the approximation. You also slow down your
calculations with small step sizes since it takes so many steps to get where you are going. Some
extra work getting a more accurate approximation than the tangent line approximation will pay off.
This leads to the Improved Euler Method, which is also sometimes called Heun's Method. So how
can we improve our approximation for
, ? Well what we
are looking for is not the tangent line to the
curve, we are looking for the "secant" line
between the points and
. Then the formula for the secant
line is where is the slope of the secant line. So how
can we approximate the slope of the secant
line? Euler's method uses the slope of the
mentary Differential Equations https://www.math.ksu.edu/math240/book/chap1/numerical.php
7 2.1.2016 14:38
7/21/2019 EulerMethod Excel Implementation Elementary Differential Equations
http://slidepdf.com/reader/full/eulermethod-excel-implementation-elementary-differential-equations 4/7
tangent line at the left endpoint
as the approximation to
the slope of the secant line. Geometrically,
there is no reason why the left endpoint
should give a better approximation than the slope at the right endpoint, . In fact,
the average of the slopes of the tangent lines at each endpoints
is better still. Unfortunately, we only know the left endpoint ; the right endpoint
is what we are trying to find (which is why we used the left endpoint in Euler's method). But once
we have carried out Euler's method, we have an approximation to the right endpoint, . Now
we can use this approximation to compute the approximate slope of the tangent line at the right
endpoint and then the approximate slope of the secant line
All those "approximates" in the last sentence may look scary, but the process is actually fairly
straightforward. For the initial value problem
we set and and compute and
as in Euler's method. We then improve our approximation for from to
We repeat this procedure as many times as needed to reach our desired point just as with the
original Euler's method. For example, to approximate using a step size of 0.1 we compute
the following (calculations here have only been reported to 4 decimal places):
Implementing The Improved Euler Method In Excel
Just as for Euler's method, the repeated calculations of the improved Euler method are best handled
with a computer. As before, you can use whatever tool you find most appropriate, but the simplest
mentary Differential Equations https://www.math.ksu.edu/math240/book/chap1/numerical.php
7 2.1.2016 14:38
7/21/2019 EulerMethod Excel Implementation Elementary Differential Equations
http://slidepdf.com/reader/full/eulermethod-excel-implementation-elementary-differential-equations 5/7
is probably the spreadsheet.
Open up your favorite spreadsheet. We set things up similarly to the Euler's method, but with a
couple of extra columns now for the extra computations. Enter x in cell A1, l ef t y' in cell B1, y
t i l de in cell C1, r i ght y' in cell D1, y in cell E1, and h in cell H1. Next enter the initial values 0 in
cell A2, 1 in cell E2, and 0. 1 in cell H2. Then enter the formulas =A2+$H$2 in cell A3, =2*A2- E2 in
cell B3, =E2+$H$2* B3 in cell C3, =2*A3- C3 in cell D3, and =E2+$H$2*( B3+D3) / 2 in cell E3. Now click
and drag to select A3..E3 and then copy the formula down the sheet to repeat the calculations as
far as you want.
As before, you can apply this method for other equations just by editing the formulas for the slopes
in columns B and D. And as before, you can add columns for the exact values and the relative error.
Enter y exact in cell F1 and rel error in cell G1. Enter the formulas =3*exp( - A2)+2*A2- 2 in cell
F2 and =( F2- E2) / F2 in cell G2. Click and drag to select F2..G2 and then copy those formulas down
the sheet to show the exact values and the relative error. As before it looks nicer if you set the
format of the cells in column G to Percentage.
Note that the improved Euler's method is twice as much work as the original Euler's method, since
you do an extra improvement step each row. On the other hand, the error for is now just
0.2%, which is much closer than the 2.85% error we got by doing twice as much work with a
smaller step size with the original Euler's method. We can improve the accuracy of our
approximation further by using a smaller step size of . The result of this calculation is off by
only 0.05% of the true value. While halving the step size (hence doubling the work) halved the
error for Euler's method, it reduces the error in the improved Euler's method by about a factor of 4.
This is another advantage of the improved Euler's method.
mentary Differential Equations https://www.math.ksu.edu/math240/book/chap1/numerical.php
7 2.1.2016 14:38
7/21/2019 EulerMethod Excel Implementation Elementary Differential Equations
http://slidepdf.com/reader/full/eulermethod-excel-implementation-elementary-differential-equations 6/7
Concluding Discussion
We could repeat the process of improving the estimate by using both and from the improved
Euler's method to get a New And Improved Euler's method. And of course we could then improve
the estimate another time and yet another time. This leads to a class of methods called
Runge-Kutta methods. To find the best approximations of the slope by taking weighted averages of
different simpler approximations at different points requires that we replace the simple tangent line
approximation by a Taylor series approximation and do some more algebra. But since that's all been
done, you can just look up the formulas and enter them into a spreadsheet in the same manner as
we've done here.
Of course, applying these methods requires you to pick the correct step size. Looking back at the
geometry you should realize there is no reason you have to always use the same step size. Indeed,
you will be better off if you vary the step size to be small where the solution is curving quickly and a
larger step size (hence fewer steps so quicker execution and less roundoff) where the solution is
moving in more of a straight line. In developing code to track the motion of space probes (which
have long straight paths between planets and complicated curves as the swing around planets),
Fehlberg developed a further refinement using two different Runge-Kutta methods where you
compare their answers to get an approximation for the error. This approximation can then be used
to adjust the step size automatically. These are called Runge-Kutta-Fehlberg methods and are
perhaps the best of the common techniques for finding careful approximations to initial value
problems.
While the RKF techniques are recommended for careful approximations, the methods we've learned
here, besides introducing the basic ideas, are also useful when you need rough approximations
quickly. Being so simple, Euler's method and the improved Euler's method will run quicker than
more accurate techniques. The physics model of many video games is implemented using thesemethods. To update the screen 60 times a second, you need to compute the position of many
different objects as quickly as possible. And you just need things to look and feel right to the
players; you don't need precise answers (for example, any error less than one pixel is irrelevant). In
this situation, using Euler's method or the improved Euler's method is the best choice.
Exercises
Approximate for each of the initial value problems using Euler's method, first with a step size
of and then with a step size of .
1.
2.
mentary Differential Equations https://www.math.ksu.edu/math240/book/chap1/numerical.php
7 2.1.2016 14:38
7/21/2019 EulerMethod Excel Implementation Elementary Differential Equations
http://slidepdf.com/reader/full/eulermethod-excel-implementation-elementary-differential-equations 7/7
3.
4.
5.
6.
Approximate for each of the initial value problems using the improved Euler's method, first
with a step size of and then with a step size of .
7.
8.
9.
10.
11.
12.
Find the exact solution to each of these equations and use that to find an exact value for for
each of the initial value problems and compare the errors in the different methods used above.
13.
14.
15.
16.
17.
18.
If you have any problems with this page, please contact [email protected].
©2010, 2014 Andrew G. Bennett
mentary Differential Equations https://www.math.ksu.edu/math240/book/chap1/numerical.php