Wednesday, May 14, 2014

A Day in the Life of a Scientist

So my friend Phil posted to Facebook how he tried all kinds of technical fixes to get the speed sensor on his bike to work, and in the end all he had to do was move the base unit a little closer to the sensor. The story hit home for reasons that will become apparent.

When I was a kid, wanting to be a scientist, I read stories about Einstein and Newton and how they wielded elegant, airtight logic to move their incredible hypotheses straight toward undeniable fact. Now that I'm a scientist (sort of ), my experience has been closer to Edison's comment about it being 99% perspiration. No doubt Edison wasn't as smart as Einstein and Newton, but I think his take is a lot closer to typical than what you read in the biographies.

So here is a day in the life of a real scientist, showing you how the sausage is really made. The details have been altered but the story is all too true.

1. Arrive at desk at 8 am. Where did I leave off yesterday? Oh yeah, I was trying to fix up a simulation so that it predicts Effect Y when Cause X is active. It had better, because our experiment said X causes Y.

2. Was I really unable to show X causes Y? I have results here from right before I went home last night, but didn't record the inputs because I was in a hurry not to miss dinner and was sure I'd remember them. So I don't really know what these results mean. Reconstruct the inputs and re-run.

3. During this run, I started to think about causes. What about the interpolation order? Is it high enough? Too high? What about the modulus? I can't even remember where I got this value. I'm going to look it up on Wikipedia. If the Wikipedia value is different, then my value is probably bad. Or maybe Wikipedia is wrong. So I'll check another source. Here's a student project from the Technical University of München. It uses a value within 5% of mine. They are pretty smart over in München and I can imagine them speaking in a foreign accent, so it's probably a good value.

4. Lunchtime, then I have to go on a phone call with the people funding this work. They want to know all about the schedule and budget, so it takes an extra half hour to regain my train of thought once the call is over.    

5. The run is done and it shows Result B, which is different from anything I've seen so far. What if I didn't put on the boundary conditions correctly? To be sure, I'll remove all the boundary conditions and re-apply carefully.

6. Ran again and I'm still not getting the expected results. Maybe I had the boundary conditions right before, and messed them up with this latest re-do. Double check and run again. I think maybe the displacement is going nonlinear. I might need to change to a hyperelastic constitutive theory.  It is a good thing I'm so smart and know about these things. Others would have been defeated by this problem.

7. I am 99% sure there is a bug in this code. There must be. I've checked every conceivable input and they all look good. All this money for software and it's full of bugs. Spend an hour drafting a bug report. In the process of writing the bug report I discover that I was looking at the wrong column of results. Trash the bug report. I am really stumped for new ideas, so I spend several minutes drawing the Greek letter sigma in both its capital and lower-case forms.

8. Maybe the temperature was recorded incorrectly during the experiment. Did they even say where they placed the thermocouple? Did they make sure everything was soaked out before doing the test? Experimenters never document things well enough. If the temperature was this way, then the result would be that way. I think.

9. Run again with a different temperature and the trend is opposite what I expected. Something is interfering with my concentration - a vague feeling of pressure. Then I remember - back during Step 3, I never did get to the bathroom. It can't be put off any longer.

10. It's getting late and my kid has a ball game tonight. I am not sure what the original problem was. Oh yeah, X should cause Y but it's causing Z, or A or B depending on the inputs. Go for a walk outside to clear my head.

11. While sitting on a bench outside, I realize that in converting minutes to hours, I might have multiplied by 60 instead of dividing by 60. I don't think I did that. I have a Ph.D., and nobody would get a Ph.D. who doesn't know when to divide instead of multiplying. But I'd better check it anyway.

12. Back in the office and sure enough, I've entered 60 minutes as 3600 hours instead of 1 hour. Got to change that, but I need to find my original set of inputs. By looking at the timestamps on my files, I deduce that the original inputs are in the file Inputs(13).xlsx,

13. I really need to go, so I'm in a hurry and run the simulation with the corrected inputs, but forget to turn on output so it takes 18 minutes to get nothing. But I'm so close. I can run it once more and if I really need to hurry, I'll skip filling in my timesheet until tomorrow. Run again and get result Y. Victory!! Leave so quickly that I forget to take my container of lunch leftovers out of the refrigerator. I'll discover them in a few weeks when they have grown a green beard.


I am not sure where this all fits into the scientific method, Kuhn's paradigm shifts or Popper's falsifiability criterion.  You can find numerous fallacies and biases here. But this process actually works, at least for me. The best I can guess is that during all the fooling around with inputs and going down dead ends, even though the results are no good, I'm building intuition that I can use later to speed up my reasoning by a large factor, so that after going through ten or so of these cycles, I can answer questions without even having to run a simulation. Without intuition you may as well pack it in.  

No comments:

Post a Comment