2.096/6.336/16.910  Introduction to Numerical Simulation

1) Dec. 12 (Wed) , the last day of classes, is a firm deadline of any submission. We DO NOT accept any kind of submissions after that.

2) Graded Pset 5 is ready for pickup. The pickup box is located at the front door of the 7th floor in building 36, which is closer to the elevator.

3) Graded Pset 6 will be ready before Monday.  

As explained in class, we have added a link to the HKN class evaluation on the navigation menu of our class website.

If you were not in class yesterday Tuesday Dec 4 when a HKN representative explained the process, I would like here to bring to your attention and underline the importance of the HKN class evaluation process. I will not get your feedback until after the grades are assigned and the term is over. I will NOT be able to see your names on your survey. However I am very much looking forward to read your feedback to make this a better experience for your fellow students in the following years.

Therefore I would like to strongly encourage you to evaluate the class from home using the link I have added on the menu bar on the class website (you will need your MIT certificate to log in, the link will shortly become active and will stay active until December 16, 2007).

Thanks

Luca Daniel

In Problem Set 2, you experimentally measured the performance characteristics of your implementation of a matrix inversion routine, and compared with the MATLAB inv command. Many of the homework submissions concluded that one or both of these ran in O(N) time on an NxN matrix.

If you reached this conclusion, you should take a minute to revisit your measurements. In general, people took a perfectly valid approach to their experiments, but they measured with matrices that were too small to really capture the scaling behavior.

I've added an addendum to the solution which describes this in greater detail.

Please follow the link to the problem set 2 solutions and download the newly added PDF file. There's also some matlab code demonstrating an approach to MATLAB performance measurement.

Thanks,
-steve

P.S. Problem Set 2 homeworks will be handed back in class tomorrow (10/18).

Many of you have asked about Problem set 3, Problem 4, Part a, which asks that you show that the  norm of the residual for the third iteration (r^3) over the norm of r^0 is less than or equal to 3*(sqrt(5)-1)/(sqrt(5)+1).
If you believe that you can find a tighter bound, you are absolutely right.
In fact, if the tightest bound you could find for this ratio was greater than 1, this would mean that the norm of your residual might grow with each iteration, which is obviously a bad thing (and happens to be untrue for this algorithm).

So we are modifying the problem (as indeed a few of you have already suggested) such that the coefficient is 2, rather than 3, resulting in a tighter bound.

If you have already done the problem and demonstrated that relationship is constrained to the bound as specified, you may leave it as is -- we will not deduct any points for this.

If you have any questions, feel free to contact us:
Lei Zhang <zhangl@mit.edu>
Steve Leibman <sleibman@mit.edu>

Regards,
-steve

Attention all students in 2.096/6.336/16.910 Intro to Numerical Simulation:

Regularly scheduled office hours begin this week, and will be held in room 36-713.

Monday: 4:30 - 5:30 pm with Lei Zhang
Tuesday: Just after class (~11am) - 1:15pm with Steve Leibman
Thursday: 4:30 - 5:30 pm with Lei Zhang

While we expect these hours to remain constant throughout the semester, it is possible that the schedule will be modified to accommodate demand, so be sure to check the announcements on the front page of the Stellar site prior to attending.