Why do we still need real world experiments?

[Helicobacter]

We have windtunels to measure air flow, Erlenmeier flasks for chemical reactions, and so on.

Why not just load the geometry of the aircraft into a computer simulation and get the results from there. It is much more cost efficient and you can manipulate things faster and easier.

For chemical experiments, you can just load some subatomic particles into an object oriented environment under the four physical laws, input the exact situation and the computer will render the results from the interaction. You can rewind, slow down, and use other convenient features on the results.

Why do we not perform these experiments exclusively on the computer?

 

[Topher925]

Because 99.99999% of the worlds problems can not be solved (or at least we don't know how to solve for it). The best we can do is make an approximation and hope for the best. 99 times out of 100, simulations are incorrect or do not meet the required accuracy. In some cases we don't even understand the physics, let alone the math, of what's going on.

 

[gmax137]

Why do we not perform these experiments exclusively on the computer?

Why didn't you write a program to answer your own question?

 

[minger]

The only people who believe numerical solutions are experimentalists and managers. Truth is, people running FEA and CFD are fully aware of its shortcomings. Did you know that 99% of CFD solutions "approximate" turbulence?

It would be nice to be able to run these kinds of numerical solutions, but the problem is that complex runs can take weeks/months. The 1% of DNS (direct numerical solution) CFD runs that actually attempt to calculate the turbulence can take months on enormous clustered supercomputers.

Also, as we develop methods and best-practices to model complex situations, we always always need a way to verify and validate our solutions.

 

[mgb_phys]

Why do we not perform these experiments exclusively on the computer?

Even for simple cases where Newtonian mechanics apply (like fluid flow) we have to approximate the behaviour by splitting the problem up into little boxes and assuming that these boxes are small enough to be a constant value.
As you need more accuracy, you need more and more boxes - especially in regimes where the flow is turbulent, and so more and more computers/

With more complex behaviours, like the quantum mechanics of chemical reactions, it gets even harder.

 

[Helicobacter]

I have extracted three main drawbacks from your rejoinders.

Complexity of modeling
Insufficient accuracy
These concerns can be remedied with faster computers with more memory. In a little bit over two decades we will have hardware 1 million times better than today. What about then?

Modeling unknown phenomena
Somebody told me that quantum mechanical effects are minor in substances that have a real world size (the result of a die roll is practically determined by the standard model). And can't quantum mechanical effects be modeled with random variates? And turbulence with a state-of-the-art virtual engine?

 

[DaveC426913]

For chemical experiments, you can just load some subatomic particles into an object oriented environment under the four physical laws, input the exact situation and the computer will render the results from the interaction. You can rewind, slow down, and use other convenient features on the results.

Why do we not perform these experiments exclusively on the computer?

Because we don't know how everything works enough to make a simulation that will yield unexpected results (i.e. we learn something we didn't know before).

If we write the program that shows us how stuff works, stuff will only ever work the way we expect it to. We would never learn anything new.

 

[Animastryfe]

Because we don't know how everything works enough to make a simulation that will yield unexpected results (i.e. we learn something we didn't know before).

If we write the program that shows us how stuff works, stuff will only ever work the way we expect it to. We would never learn anything new.

Isn't this too strong? There are plenty of papers on computational results. If we never learn what anything new from simulations, then why simulate, for example, fission and fusion reactions? What do, for example, computational biologists do? I suspect you mean that we have to have a good model or quantitative theory first in order for simulations to be useful, and simulations are there to give us predictions based on those theories that would otherwise be too difficult to do with only pencil and paper.

 

[sardonicus]

Babies and children learn from real world experience, they learn simple physics the hard way, falling, dropping, biting, etc. Can they learn it another way? It would be a different experience to be sure. Extrapolate?

 

[Gannet]

Computer programs are only based on what is known.

True tests are done to discover the unknown.

Or you can let your customers discover what you did not know.

 

[pallidin]

Simulation is a valuable design strategy. Used often in today's higher technologies, thousands, sometimes millions of dollars are saved.
From biological research to NASA, computer simulation can save multiple hours of "real-world" and very expensive tests.
Not an end unto itself to be sure, but a welcome addition to development.

 

[alxm]

Because we don't know how everything works enough to make a simulation that will yield unexpected results (i.e. we learn something we didn't know before).

Oh this isn't true at all. The Navier-Stokes equation, from which more-or-less all of fluid dynamics derives, were found in 1822. Haven't we learned something new about fluid dynamics since? Of course - in fact, we still use wind tunnels all the time. Similarly, the Schrödinger and Dirac equations are now 80 years old, and there's little doubt all of chemistry (and more or less all 'everyday world') things can be described by them. That doesn't mean chemistry is 'done' by any means, or that there aren't surprises left in store.

It's true that if you're working from strictly empirical models, you will never learn something you don't already know, but this is not the case for things founded on solid theory.

As for the original question, it's simply a matter of a vast underestimation of the complexity of the calculations involved. Say I want to calculate the energy of a chemical compound within 5 kcal/mol or so. With powerful computers, that's reasonable for about 200 (light!) atoms or so. For within 0.2 kcal/mol? Maybe 20 atoms. Within experimental error? Maybe 5. Quantum-chemical calculations are currently neither faster, easier, or cheaper than investigating in the laboratory. They are used for investigating things that can't be done experimentally.

(And the most accurate method, full-CI, scales factorially with the number of electrons. That's exponentially faster than any exponentially-fast growth in computing power)