I found it counterintuitive that it was predicted in simulation and then confirmed with experiments, rather than the other way around.

With nanopore experiments, it's not unusual to see unusual behavior that could possibly be explained by macromolecule interactions, contamination, or electrical interference. There are often more unknowns than knowns due to the scale of the system, novel materials, measurement precision, unique physics, etc etc... What usually happens is the undesired measurement gets filed away for 'later study'. Unless you have some theoretical framework or model that explains it, you may never go back. So probably the experimental part was previously dismissed as an error or contamination, but now it is interesting because there is an explanation

Years ago, one of the HP people involved with early ink-jet printers told me that intuition about fluid behavior totally fails at that scale, and they had to use computational fluid dynamics to figure out how to get ink droplets to behave.

That is how it's supposed to work. You build a theory based on known phenomena, then you look for any new phenoma it predicts and test them in experiments. Without the new phenomena, your theory is more likely to be just "overfitting" the data. A famous example of overfitting the data - adding more cruft to account for unexplained aspects - is this: https://en.wikipedia.org/wiki/Deferent_and_epicycle

This is how lots of new science is discovered. One example is the Theory of Relativity. Sure it was not a software simulation per se but it was still simulated on paper. It was not confirmed by experiment until several years latter.

If I'm not mistaken I think what OP is referring to is an apparent emergent phenomena that contradicted theory, whereas I believe GR was discovered through theory (ie Einstein's "happiest thought" was in an elevator shaft).

It’s nice way to do engineering as well - Fully modeling every aspect of a collision first is often cheaper than doing multiple crash tests.

> I probably would have written it off as floating point shenanigans.

Pretty sure anyone in actual scientific computing stays far, far away from floating point numbers and sticks to arbitrary precision ones instead.

Double precision floating point numbers are the most common in scientific computing. Sometimes software will mix in single precision for more speed. The molecular dynamics package GROMACS can use mixed single/double precision, for example:

http://manual.gromacs.org/documentation/2016/install-guide/i...

Error analysis can tell you whether the roundoff error would grow exponentially with time steps for the algorithm you use. If you avoid this kind of conditions in scientific computing, then double precision floating point number would work perfectly fine.

And by the way, scientific computing is a diverse discipline, or a federation of disciplines. Say, there may be little overlap between the algorithms, tools, and environments that a physicist uses for scientific computing and those used by a bioinformatician. So I don't think it is universally true in scientific computing that arbitrary precision is preferable to double precision.

I'm quite curious what niche of scientific computing you've dealt with that gave you that impression.