Mike Jonas
I have put a bit more work into Chaos Theory (aka the Butterfly Effect), and whether it can apply in the real world. It seems that just about every time someone tries to demonstrate Chaos Theory, they do it with a model. Trouble is, their findings then always turn out to have come from the model itself. As with the turtles that prop up our planet, the climate modelling world sometimes feels like it is propped up by circular logic all the way down.
The strictest version of Chaos Theory cannot apply to the real world, because it refers to variations in initial conditions, and the real world only has one trajectory. In other words, there is only one version of Earth’s climate at any one point in time, so it can’t have a variation in initial conditions. Unfortunately, the theory has morphed into something which many in climate science think applies to Earth’s climate, and when they run climate models they get confirmation. Well, they think it looks like confirmation.
First, let’s be clear about what we are looking at. Chaos Theory says that Earth’s weather and climate are very sensitive to local conditions: a tiny variation grows exponentially into a large difference. We know that weather and climate show natural variations on many physical scales and many time scales – there is turbulence in both the atmosphere and the ocean. We know that turbulence can be very difficult to predict, and we know that differential equations such as Navier-Stokes can be useful for understanding it. We don’t know that a tiny variation will grow exponentially. My suspicion is that the exponential growth described by Chaos Theory is an artefact of the models, not a natural phenomenon, and that it is being confused with natural variation. Further, that if we ditched Chaos Theory we might end up with a better understanding of our world.
So I did an extensive search of the literature, using Grok of course because it is amongst other things a powerful search engine. Amongst all the attempts to demonstrate Chaos Theory using models, there were some that tried to do it using atmospheric analogues: they look for pairs of very similar situations in historical records to see whether they then diverge rapidly as per Chaos Theory. The concept is that the difference between the two situations is a proxy for Chaos Theory’s variation in initial conditions. Unsurprisingly, they found divergence (otherwise weather forecasting would be easy), but equally unsurprisingly the situations were never close enough to be a proper test – Chaos Theory is based on infinitesimal differences.
And then Grok served up a bombshell. Some of its AI mates have got into climate modelling, and when some of their model runs showed the rapid divergence expected from Chaos Theory their handlers did a very interesting test: they repeated the exact same model runs on a more accurate computer. The divergence was eliminated or severely reduced. The divergence, in their words, “is caused by numerical noise, which is an artifact of the computation rather than a real atmospheric process“. If chaos can be created by using a less accurate computer, imagine how much chaos can be created by the crude iterations of a GCM. (GCMs are General Circulation Models, which have been the mainstream climate models for many years).
I had already written most of a paper covering the other tests, and obviously I needed to add this in. I couldn’t claim that all chaos in models was numerical noise, even though it could be (and probably is?). Although I suspected that GCMs’ iterative errors would be orders of magnitude larger, I couldn’t address that in any detail without a lot of work. So I put it mainly into the “Discussion” section, on the basis that it did not have to be argued and referenced as assiduously there as in the “Analysis” section, and that one day I might get around to doing that detailed work. The paper is Chaos Theory remains void for the real climate and is now published, and here is the abstract:
Weather and climate model results, and studies related to Chaos Theory, are examined for evidence that Chaos Theory applies in the real world rather than only to models. No such evidence is found, and in some cases rapid growth in model divergence as expected from Chaos Theory has been found to be an artefact of the models. Results from models coded with mathematical formulae that exhibit rapid growth in divergence cannot be used to assess the applicability of Chaos Theory to the real world, because that would necessarily involve circular logic: the model results are only exhibiting an emergent property of the formulae as used in the models. Tests with atmospheric analogues have been unable to assess the applicability of Chaos Theory to the real world, because they could not emulate the infinitesimal perturbations at the core of Chaos Theory. Some AI models have shown that their rapid growth in divergence, as expected from Chaos Theory, is caused only or primarily by numerical noise, and is therefore an artefact of the computation rather than a real atmospheric process. General Circulation Models use processes with inaccuracies similar to numerical noise, which could have a similar effect. Other AI models do not exhibit rapid growth in divergence. The model findings apply only to the models, so they cannot be used to assess Chaos Theory’s relevance to the real world. In the real world, Chaos Theory remains void (scientifically inapplicable).
Even if every model run in history is shown to have been the source of its own chaos, that still would not prove that Chaos Theory does not apply in the real world. Even if Chaos Theory does apply in the real world, it seems impossible to demonstrate it for the reason given above, hence the “remains void” in the title. What I suspect will happen (I would like to be wrong) is that Chaos Theory will be changed to something that actually can be demonstrated in the real climate, and then it will be represented as having always been correct.
I would really like to see a change in some of the thinking in climate science, so I added a paragraph arguing that at least some of the weather and climate variability that is regarded as chaos may ultimately be resolvable by better observations and better understanding, exactly as happens in other complex fields. In other fields of science, the normal attitude to unknowns is that the relevant factors are not yet understood, not that they are deterministic chaos. It seems reasonable to take a similar approach in the field of weather and climate.
Then I would like to see GCMs not being used for climate modelling. What a massive waste of time and money they have been.