# MA213: Numerical Analysis

Unit 4: Differentiation and Integrations of Functions   Before we talk about differential equations, we should expect some calculus.  In this unit, we will address one of the most fundamental challenges of floating point arithmetic: finding the slope of a tangent line.  You will see that numerical differentiation is actually harder than numerical integration!  That said, numerical integration is especially interesting for all of the new ideas that we can explore.  We will also create an algorithm (and write a program) that adapts itself to different integration problems.

Unit4 Learning Outcomes
Upon successful completion of this unit, the student will be able to:
- Approximate the slope of a function at a point, and to approximate the definite integral of an integrable function.  - In addition, you will understand a fundamental difficulty of scientific computation: discrete calculus.

4.1 Numerical Differentiation   4.1.1 Finite Difference Formulas   - Reading: University of Arkansas: Mark Arnold’s “Numerical Differentiation is Easy” Link: University of Arkansas: Mark Arnold’s “Numerical Differentiation is Easy” (PDF)

Instructions: Click on the link above, then select the appropriate link to download a pdf of the reading. Carefully read the pdf.  We will play this game again and again: If we approximate a function f by an interpolating polynomial P, then D(f) is approximately D(P).  Here D is the derivative operator, and the formulas are called finite difference formulas.  Every different interpolating polynomial gives a different finite difference formula for the slope of f at a point.  The formulas here are given by uniformly spaced nodes, but you can make up your own formulas for any set of (pairwise distinct) nodes.

This resource should take approximately 1 hours to complete.

• Lecture: YouTube: University of South Florida: Autar Kaw’s “Differentiation of Continuous Functions” Link: YouTube: University of South Florida: Autar Kaw’s “Differentiation of Continuous Functions” (YouTube)

Instructions: Click on the link above, then watch the video lectures in the chapter listed above.  In this case there are 6 lectures that have been split into 9 videos.

This resource should take approximately 1 hour to complete.

4.1.2 In the Face of Rounding Errors   - Reading: University of Arkansas: Mark Arnold’s “Numerical Differentiation is Hard” Link: University of Arkansas: Mark Arnold’s “Numerical Differentiation is Hard" (PDF)

Instructions: Click on the link above, then select the appropriate link to download a pdf of the reading. Please carefully read the pdf.  As easy as it was for us to find finite difference formulas, they depend on h being small.  In finite precision arithmetic there is a limit to how small h can be.  Are you clear about how truncation error and rounding error conspire to compromise your derivative estimate?  Explain why higher order formulas ameliorate this problem.

This resource should take approximately 1 hour to complete.

Instructions: Click on the link above, then select the appropriate link to download a pdf of the reading. Please carefully read the pdf.  Again, if P approximates f, then I(P) is approximately I(f), right?  Well this time the answer is… yes!  Integrating a Lagrange interpolating polynomial for f, can give a very good approximation for the integral of f, even in the face of rounding errors.  What degree polynomial do we need if f oscillates m times over [a,b]?

This resource should take approximately 1 hour to complete.

4.2.2 Composite Newton-Cotes Rules   - Reading: University of Arkansas: Mark Arnold’s “Numerical Integration is Easy” Link: University of Arkansas: Mark Arnold’s “Numerical Integration is Easy” (PDF)

Instructions: Click on the link above, then select the appropriate link to download a pdf of the reading. Carefully read the pdf.  The composite rule rules.  The property that the integral from a to b can be split into the sum of the integrals from a to c and c to b is fundamentally what make quadrature a nice problem.  The challenge now is to compute I(f) quickly.  In many situations, we have values of f already computed on a uniform grid (x_{i+1}-x_i = h, for all i).  Composite Simpson’s rule is a very popular choice among methods which require uniform spacing of nodes; can you argue why?

This resource should take approximately 1 hour to complete.

Instructions: Click on the link above, then watch the video lectures in the chapter listed above.  In this case there are 8 lectures that have been split into 9 videos.

This resource should take approximately 1 hour to complete.

• Lecture: YouTube: University of South Florida: Autar Kaw’s “Simpson’s 1/3 Rule” Link: YouTube: University of South Florida: Autar Kaw’s “Simpson’s 1/3 Rule” (YouTube)

Instructions: Click on the link above, then watch the video lectures in the chapter listed above.  In this case there are 4 lectures that have been split into 6 videos.

This resource should take approximately 1 hour to complete.

Instructions: Click on the link above, then select the appropriate link to download a pdf of the reading. Carefully read the pdf.  If we do have the freedom to evaluate f at arbitrary points in [a,b], then maybe a uniform grid is not optimum.  There may be regions where f varies little, while on other regions f oscillates a lot.  How does the adaptive quadrature idea deal with such a function? Why do we require the error to be halved when we branch to a lower level?  Do you think our (wish) is more or less likely to be true as we branch to lower levels?  Could we create an adaptive method without an error estimate?

This resource should take approximately 2 hours to complete.

4.3 Of Things Not Covered   - Reading: University of Arkansas: Mark Arnold’s “Monte-Carlo Integration” Link: University of Arkansas: Mark Arnold’s “Monte-Carlo Integration” (PDF)

Instructions: Click on the link above, then select the appropriate link to download a pdf of the reading. Carefully read the pdf’s.  There are two families of integration techniques that we have omitted.  One, Gaussian quadrature, requires that we can evaluate f at arbitrary points in [a,b], but achieves double the accuracy of the uniformly spaced Newton-Cotes methods.  The other type of method is philosophically opposite to adaptive or Gaussian quadrature:  evaluate f at random points (not uniform, not optimal, not smart: random).  As you might guess, this type of method cannot compete with those we have discussed… unless we are integrating a function of many variables over a high dimensional region.  These Monte-Carlo methods are methods of last resort, and as such are very important indeed!

This resource should take approximately 1 hour to complete.

Instructions: Click on the link above, then watch the video lectures in the chapter listed above.  In this case there are 7 lectures that have been split into 9 videos.

This resource should take approximately 1.25 hours to complete.

``````Link: University of Arkansas: Mark Arnold’s “[<span
Program</span>](http://www.uark.edu/~arnold/4363/Octave/prog)” (PDF)

Instructions: Click on the link above, then select the appropriate
quadrature program that meets the specifications required on the
assignment.

This resource should take approximately 4 hours to complete.

displayed on the webpage above.
``````

Unit 4 Assessment   - Assessment: The Saylor Foundation's "Unit 4 Assessment" Link: The Saylor Foundation's "Unit 4 Assessment" (PDF) and “Unit 4 Guide to Responding” (PDF).

`````` Instructions: Carefully answer the two questions in this
assessment.  You may consider them essay questions, but back up or
explain with formulas and/or theorems as needed.  When you are
finished, you should check yourself by looking at the response
guide.  You may not have all of the points given in the response
guide, or you may have noted something that was not included in the
guide, and that may be just fine, but this is a time to reflect on
differences between your response and the guide response and
possibly review some material.
``````