Chronology |
Current Month |
Current Thread |
Current Date |

[Year List] [Month List (current year)] | [Date Index] [Thread Index] | [Thread Prev] [Thread Next] | [Date Prev] [Date Next] |

*From*: John Denker <jsd@av8n.com>*Date*: Fri, 01 Sep 2006 00:30:15 -0400

Jeffrey Schnick wrote:

I set up a spreadsheet ... for doing

uncertainty propagation in the case of a function of up to six variables ... Monte Carlo.

Yup.

(Excel 2003 SP2--I haven't tested it with old versions of Excel)

I tried it with gnumeric under Linux. Works fine. I had to widen one of the

columns, and slide some of the small graphs so they didn't occlude each other.

Total hassle: about 4 seconds.

I suspect that in some versions of excel, users would need to make sure some

optional math extensions had been installed.

The spreadsheet is the file gaussEPv5.xls.zip at

http://www.anselm.edu/internet/physics/cbphysics/math.html

The spreadsheet may be of use to folks who want to go beyond sig figs but don't want to bog

students down with algebraic uncertainty propagation.

I don't want to go "beyond" sig figs; I want to go in the other direction

entirely. Sig figs rules destroy your data. Example: try propagating

the uncertainty in the following elementary calculation, using sig figs:

x = (((2 + 0.4) + 0.4) + 0.4) + 0.4

where each of the addends has 10% uncertainty, normally distributed. Hint:

The right answer is x = 3.6, with less than 6% uncertainty. Alas the usual

"sig figs "rules" produce the ludicrous answer x = 2. Indeed the "rules" set

each of the parenthesized sub-expressions to 2.

Help prevent cruelty to data: Don't use sig figs.

It provides more information than algebraic methods in that one can see what the

calculated function's distribution looks like, in particular, whether or not it is reasonably

close to being gaussian.

There are many situations where Monte Carlo is /infinitely/ more informative

than algebraic techniques, because it is feasible in situations where mortals

cannot even get started with algebraic techniques.

Important examples include cases where the input uncertainties are non-Gaussian

and/or non-independent. See

http://www.av8n.com/physics/uncertainty.htm#sec-mg-mass

for an example of this.

=========

Minor pedagogical suggestion concerning the About page: All the customers really

need to do to get started is to change the semantics by changing the formula in

cell a2. That is, don't tell them to start by renaming the function (f) and

renaming the variables (x1, x2, etc.). That's just window dressing. Window

dressing is nice, but it can come later.

Another, more-important point: Students generally need to walk before they run.

Monte Carlo is the gold standard, but it is too sophisticated to be a starting point.

So start them out with the /Crank Three Times/ method. Even the dimmest of intro

students can calculate the reciprocal of 1±.5 by evaluating 1/(1-.5), 1/1, and

1/(1+.5). This does not require any sophistication or any software. Also, it

doesn't pollute their brains with "sig figs" nonsense, and it lays a good foundation

for future progress. Monte Carlo can be (in due course) introduced as the natural

extension and elaboration of Crank Three Times. See

http://www.av8n.com/physics/uncertainty.htm#sec-crank3

**Follow-Ups**:**Re: [Phys-l] Monte Carlo***From:*Bernard Cleyet <bernardcleyet@redshift.com>

- Next by Date:
**Re: [Phys-l] gasoline** - Next by thread:
**Re: [Phys-l] Monte Carlo** - Index(es):