Tracy&James | Thursday, 18 July 2019
All models are wrong, but some are useful. I’m sure many of you will have come across that saying before, particularly if you work within a science, engineering or technological industry. I’m fortunate enough to work in one of the country’s top scientific establishments which has an operating budget of getting on for £1 billion a year and only one customer to worry about. With such a budget I’m sure you can imagine how much we spend on our modelling efforts (especially these days when people get a bit twitchy about us wanting to actually test our product ☺ ). As such, you could reasonably expect our models to be perfect, right? Well no, because as already stated – all models are wrong! (They are certainly useful though).
In the SET world, a ‘model’ can be anything from a simple Excel spreadsheet through to a commercial finite element code with meshing software that outputs via a stunning graphical user interface, costing hundreds of thousands of pounds. (Actually there’s a layer above that of non-commercial, lab-developed codes that have taken decades of development and fine tuning by scientists and engineers, that it’s hard to put a value on). In common, they all rely on the implementation of known physics principles and laws, and all need validation – even the most basic ones.
As an example, during a board discussion about casting from the front of a boat, I decided I would write an excel model for determining the path of a thrown ball in the wind. This was a simple model – in the vertical axis there is a constant force arising from gravity plus (or minus) an additional force from drag, which arises from the vertical velocity of the ball – obviously this force switches direction when the maximum height is reached and the fall begins (assuming an upward throw). In the horizontal axis there is just the one force acting – again drag. The physics ‘engine’ of the model was the board’s old favourite F=ma and remembering ‘a’ is the 2nd derivative of position (the first being velocity), then it’s fairly easy to form a couple of differential equations that solve for the x and y positions at any given time.
Thus this model is so easy that it’s got to be correct, right? Well no – and I’ll come on to the reasons why. To validate this model it would be fairly easy to throw some balls, measure the launch angle and velocity, take a wind speed measurement and apply the drag coefficient that you’ve previously measured in a wind tunnel (even for this very simple model we’re going to have to start splashing the cash to book some testing facilities). So where would the errors (and there will be errors) between the model and the real throws come from? Actually, there’s quite a long list e.g. noise (you can’t have the ball move without producing pressure changes in the air around it), heat (friction with the air), what if the ball spins? What if the air into which the ball is thrown isn’t perfectly laminar (it’s almost certain to be turbulent if there’s a wind)? What if the ball deforms (and recovers) as you throw it? What if Tracy throws it on a Tuesday morning etc. etc.? Hopefully you can see that we’re already discussing a hugely complex problem and one that would take a code, containing all the extra physics to solve adequately.
Or you could ‘fudge’ it. Here I would have the experimental path observation, along with the velocity, angle and wind speed measurements. Then I’d just throw figures for the drag coefficient into my model until I came up with a perfect match, and I almost certainly would come up with a calculated trajectory that was impressively close to the real one. I could then declare that the coefficient of drag is ‘X’, and my model is brilliant. But what about noise, heat, spin, turbulence? Well these are no longer of any consequence because my model matches reality. We can forget about them and just use ‘X’ as this is clearly correct (and from this point on you cannot question the validity of ‘X’) – perhaps you can see that this is just the worst kind of science!
I’m hoping that some of you will see the parallels with some of the published, simple, fly casting models within this discussion and various board threads. I’ve seen answers on the board along the lines of ‘well, the effect of viscoelastic losses (for example) is negligible because the model fits!’. As a slight aside, in my lab I have a dynamic mechanical analyser (DMA), this comes with a presentation case of materials that demonstrate the machines capabilities. If you want to test a material that exhibits very high viscoelastic loss there is a sample supplied to do so – it’s made of PVC, yes the stuff that fly-lines are made of.
Also, and I know I go on about this, the lack of validation is astonishing. So much so that I believe that, if fly casting was somehow more consequential, some of the papers published would have been ripped to shreds by now.
Apologies to those who want to read about fishing/casting on the FP, here’s my weekly report – no trips, no fish, I did do some casting with an MPR in my garden though. I’ll do better for next week.
All the best, James.
