Re: [Phys-L] Half-Life measurement : uncertainties, correlations, SVD

[John Denker <jsd@av8n.com>]

On 10/19/21 10:21 AM, Paul Nord wrote:

> It shouldn't make much difference, but it's not difficult to imagine
> two or three minutes of lost time that could have been avoided.

That's a checkable hypothesis!
Clip the first three minutes off of the sean_data and re-run
the analysis. World's easiest theoretical investigation.

As Paul predicted, it doesn't make "much" difference.
It somewhat increases the percentage uncertainty on the fast
amplitude and fast decay constant, but not enormously. In
particular, three minutes is not enough to explain the difference
between the fast component activity in the unclipped sean_data
versus the my_data. And of course it does not begin to explain the
difference in slow component activity. So there must be something
else. Perhaps different geometry.

Maaaybe different detector efficiency, but then you would need
to explain why the background levels are similar.

Hmmmmm.


> Students used two different setups.  [...]  We give them a little
> freedom to screw this up.  So they may not have been careful about
> getting the sample close to the detector.

Yes, freedom is good.
That includes the freedom to screw up.
It's especially good if they /learn/ from their mistakes.

There's an old proverb that says:
  Learn from the mistakes of others;
  it's cheaper and quicker than making all the mistakes yourself.

In the real world, one would analyze the data from trial runs in
order to figure out where the pitfalls are. In class, there never
seems to be enough time for that.

Is there any way to recreate (or approximate) reality for the
students?  Maybe something like this: In advance, as homework in
preparation for doing the experiment, hand them a couple of ugly
data sets to analyze. Ask why are these data sets different? Why
are they ugly? What steps can you take so that your data will
come out less ugly?