Hacker News Reader

Comment by evancoop | Hacker News Reader

How many fascinating analogs exist for this problem? "...the corrosion on the glass also forms millions of tiny pits. The authors think those pits could serve as tiny reaction chambers..."

In other words, the experimental conditions, as intended, were almost "too perfect." The simulation of reality requires some amount of unspecified noise with respect to CONDITIONS, in this case, corroded glass.

How many experiments, on the terminal or the bench, are run with noise in the underlying test conditions?

View on HN

I'm not aware of any bench experiment that eliminates all noise (I think that would be physically impossible) but we can certainly isolate things from the external environment very well, and the physical results you get from those tend to be very "clean" (IE, if you're measuring a physical parameter which is required to be an integer, and computing a mean on your observed values, the mean will be near the integer and the variance will be tiny compared to the magnitude of the integer.

On the computer, many "experiments" (really simulations) are 100% deterministic and therefore have perfectly predictable noise characteristics. Most simulations are not deterministic for a wide range of engineering reasons (order of summation in a distributed environment, inability to specify random seed) but are nondeterministic in a statistically useful way (IE, you can run a few times and get a good idea of the real result).

> How many experiments, on the terminal or the bench, are run with noise in the underlying test conditions?

All of them, approximately. There are a few journals like http://www.orgsyn.org/ that consist entirely of rigorously vetted methods, but that is definitely an outlier. The significance of the noise will depend on the specifics, of course.

I'm fairly sure there have been a few high profile retractions of 'metal-free catalysis' that were ultimately traced to metal impurities in the reagents. There was also an incident with the DOE where in the process of refurbishing some of the nuclear arsenal they found that they couldn't reproduce one of the necessary ingredients due to some then unknown change. I am blanking on the (code)name of the material they were trying to reproduce though.

You're thinking of FOGBANK. Apparently they needed some of the impurities present in the original manufacturing runs to get the properties they wanted. Restarting FOGBANK production was problematic because a) the manufacturing process is in itself tricky and nasty and then b) once they restarted it they discovered the previously overlooked impurities.

Apparently they've resolved that.

My inorganic chemistry prof was fond of saying organic chemistry experiments "usually fail if the labware is too clean"!B-))

Reply on news.ycombinator.com