Computer experiments are the only way we can make predictions for complex systems. We specify our assumptions, check them by "predicting" the past, and then use our model to predict the future.
Actually, computer use has nothing to do with it. The logic is exactly the same as any quantitative prediction, from the predicted return of Halley's Comet onwards, except that now we can look at systems far too complex to model by hand.
No, but simulations and experiments are two different processes. Experiments can be simulated and you can experiment with simulations. What's relative is the physical process and the degree of randomization. In other words are you trying to prove or disprove?
Yes your absolutely right meeting the stated criteria. But if you actually do any type of programming. You will have noticed that modbus protocol from one type of duplex isn't the same as another. Even though the computational equivalents may be the same. I can't count the times I've ran identical programs on identical cpus with different results. That's basic programming 101. Your job if you accept the challenge is to do a rung by rung,and subroutine assessment. "That's where the rubber meets the road"
Confession: I adore bit logic its true and accurate and easier to correct. Word logic is easier when it comes to solving complexities. But a nightmare to correct. Anyway a code is a code that's a code. You can create your own.
"Yes I know you don't understand"
So what's your point?
It is more interesting that the skeptic apparently believes that laboratory experiments, or field experiments, conducted under the same conditions would always produce identical results (or else why would he be so troubled that there would be some slight variations between simulations?). This is a ridiculous proposition, there is always some slight variability between real-world experiments, and scientists are fairly capable of using statistics to sort out if experimental uncertainty is responsible for a difference between two replicate measurements. So even if two model runs performed with identical initial conditions produced slightly different results (especially if there were some random variables or Monte Carlo-type procedure included in the calculations), it is not as though that variability would be so different from physical experiments.
The real test of a model is whether the variability in the output is similar to variability observed in the real world for those parameters that the model simulates. For example, if you are talking about a global climate model, the model should reproduce the same order of magnitude of short-term global mean temperature variations that are observed in our one available realization of global mean temperature. (Would the skeptic, for instance, argue that if we could wind back time to 1900 and re-run the planet's climate with the same forcings for 1900-2000 we would get the identical plot of global mean temperature? I suggest that if he thinks that would happen he is an idiot.) But, getting back to models and reality, in the case of climate models, they are progressing to the stage where they are starting to do not only capture the trends, but also show the short-term fluctuations. This, of course, does not please climate skeptics at all, since they don't accept the proposition that models actually have enough of the physics correct that they accurately simulate internal climate variability.
Anyway, computer experiments are valid. Even if the results disagree with independent measurements of variables that are simulated, the experiment is a "success" in the sense it can lead to improvement of the model. Or, to use the tropical tropospheric hot spot as an example, disagreement between the model output and empirical data can point out flaws in field measurements.
1. I cannot see any problem with using numerical results in science.
2. Results may be exactly reproducible or they may not. That depends on the model. My first model used a pseudo-random binary sequence to produce the "random" inputs. If your model is deterministic and the same seed is used for the pseudo-random generators then you probably will get the same results each time.
On the other hand, if you use true random numbers then the model results probably won't be reproduced exactly each time. It depends, somewhat, on what you are modelling. If you are looking fr means and variances then you might get close but if you are looking for accurate scenarios then you might not.
The problem with the climate science models is typified by what I inferred from gcnp58 when he or she said: "Or, to use the tropical tropospheric hot spot as an example, disagreement between the model output and empirical data can point out flaws in field measurements."
When you start to think that reality needs changing to fit your models then you are starting to drift away from science. For the models to be useful they have to reflect reality; reality does not have to work like any model.
The current situation is:
● We have a climate.
● We have climate models.
● The IPCC will not use the models for making predictions - only "projections".
● The various projections in the IPCC reports have always been of the style: Temperature is OK now but soon it is going to skyrocket. So far, they have been wrong in FAR, SAR, TAR, AR4 and draft AR5.
● We have politicians intending to spend literally trillions based on the predictions that they think are coming from the models.
Here is my full quote: "Now apply that to a climate model. You can have the same model using the same assumptions and same inputs and it NEVER produces the same output."
Your "questions" are a classic example of misdirection. Of course a deterministic computer model will output the same values for fixed inputs. And you can conduct controlled experiments on such a model by varying one input and keeping others the same.
And if you feel climate models can do experiments, what type of experiment to do: natural, controlled or field??
_______________________________________...
Edit: Are climate models deterministic?
_______________________________________...
Edit2: I'll also throw in the fact that some solutions to the Navier–Stokes equations like turbulence remain unsolved and are one the greatest physics problems of our time and one of the million dollar offers (for a solution) from the The Clay Mathematics Institute.
If computer "experiments" weren't accurate or reproducible, then our current planes wouldn't fly. Planes are just one example. Nearly all products manufactured today are rendered on a computer to ensure they will work as expected.
People who use computer models to make predictions, which they themselves admit that many uncertainties exist (clouds, aerosols etc.), then claim that their results are "certain" and anyone who disagrees with them are "deniers" are anti-science. Pure and simple.
A denier that blocks me from answering his "questions," asks "Why is using a climate model referred to as an experiment?" He thing goes on to make the bizarre claims that computer results are not reproducible and that "You can have the same model using the same assumptions and same inputs and it NEVER produces the same output."
So my question is, should numerical experiments be used in science? It seems to me that a lot of science would need to be thrown out if we get rid of them. Thirty years ago, when I was in plasma physics they were certainly running lots of them. A very famous result in statistical mechanics is the Fermi-Pasta-Ulam problem, which was run "experimentally" (to quote Fermi et al) on the Maniac computer 60 years ago. In the 1940s Ulam first started using Monte Carlo methods.
As a secondary question, are the results from running a numerical experiment reproducible? That is, if I run the same code on the same machine with the same input variables and all random number generators set to the same starting point, will I get the same results?
Actually, to test this, I just ran a "computer experiment." I am working on some code that models the sublimation of ice, and I ran it 5 times and got exactly the same results. Of course I think there's a Catch-22 here: I can only falsify his hypothesis on computer experiments not being reproducible by running computer experiments, and he doesn't think computer experiments are real experiments.
