30
$\begingroup$

This Washington Post news article states that with the advent of computer simulation of nuclear tests, live tests are no longer needed.

Generally speaking there are 3 aspects of an explosion resulting from a nuclear weapon test that are tested:

  1. effectiveness
  2. yield
  3. explosive capability.

Russia carried out its last nuclear test in 1990, UK in 1991,the US in 1992, France and China in 1996, India & Pakistan in 1998.

List Of nuclear weapon tests

  • Therefore, does this mean that all aspects of a nuclear explosion can be simulated using super computers, or is there any aspect of a nuclear explosion that cannot be simulated by super computers?
$\endgroup$
  • 9
    $\begingroup$ I have to say that as a matter of practice the people with the actual experience in writing these simulations and comparing the results to reality all (or essentially all) work for the various nuclear armed governments and their work is classified, so the question becomes do you expect a source on Physics to be able to give you a definitive yes/no answer, or just to offer generalities. (In the US these jobs are more or less continually available to people with the right skills who can pass the security tests, but they tend to be one-way doors career-wise.) $\endgroup$ – dmckee Nov 23 '15 at 17:05
  • 1
    $\begingroup$ @dmckee Yes,I realize that individuals with the actual experience may not be able to comment due to various non disclosure rules that they are a signatory to.Ergo,I am satisfied with any non classified answer to this question. $\endgroup$ – DSarkar Nov 23 '15 at 17:30
  • 26
    $\begingroup$ You cannot blow up all that tax payers money just with simulators $\endgroup$ – jean Nov 23 '15 at 18:07
  • 6
    $\begingroup$ RE, "super computers": Functionally, there's nothing that a supercomputer can do that an normal computer could not also do, supercomputers are just faster. And indeed, today's laptop computers would have been considered supercomputers by the standards of the 60's and 70's. $\endgroup$ – RBarryYoung Nov 23 '15 at 21:40
  • 2
    $\begingroup$ @jean: You can spend a lot money on super computer simulations. Just take the human brain project as an example. $\endgroup$ – Bergi Nov 24 '15 at 4:33
36
$\begingroup$

All of it can be simulated to a certain level of precision - given enough computing power AND correct experimental values for all the parameters. The tricky bit without tests is to get experimental values for eg. the thermal conductivity of Plutonium at TPa of pressure.

Experimental tests can also only validate something to a certain level of precision - depending on the instrumentation and repeatability of the device.

edit: Yes, there are quantum effects that can't be predicted - which specific atom will fission first - but that isn't really important from an engineering perspective.

$\endgroup$
  • 15
    $\begingroup$ This is the basic situation with anything that involves complex systems and has dependence on physical parameters that are hard to access in the laboratory. $\endgroup$ – dmckee Nov 23 '15 at 17:08
  • 2
    $\begingroup$ @dmckee yes, but it's also true that an experiment is only a measurement of that single system at one time - it doesn't mean it is a good measurement of the next one $\endgroup$ – Martin Beckett Nov 23 '15 at 18:08
  • 3
    $\begingroup$ Even from a theoretical perspective, the quantum effects don't matter in this case; a supercomputer can't predict which atom will fission first, but no experiment can do so either (except for itself). $\endgroup$ – PyRulez Nov 23 '15 at 22:44
  • $\begingroup$ I can't imagine a program caring which atom fissions first, just so long as the appropriate fractional population does fission (i.e., the atoms are stored as a number density). $\endgroup$ – Kyle Kanos Nov 23 '15 at 23:43
  • 1
    $\begingroup$ @PyRulez Well, there's other quantum effects than just which atom will fission first. It might very well be that there's effects that manifest in complex scenarios like that, just like with every other thing, really. Nobody was capable of modelling how the first orbital reentry would go, and no kind of supercomputer would help that - we had to fix our models with the experimental data. Of course, after the corrections, the simulations became very accurate again, but still - the first time you do something "new" (e.g. more energetic), there will be corrections to your models :D $\endgroup$ – Luaan Nov 24 '15 at 12:18
22
$\begingroup$

To some extent this answer is echoing things that @Martin said... but from my own point of view.

In my experience of (Monte Carlo) simulation, the model you implement captures your knowledge of the physics of the situation; and if your knowledge is "perfect", your calculation, with sufficient compute power at your fingertips, will also be "perfect".

Several problems arise:

  1. Incomplete physics knowledge: if you don't know certain aspects of the behavior of your system (materials properties have a funny way of going non-linear at very high strain rates, for example) then you cannot put the right knowledge into the model. This happened to me years ago: a model assumed linear behavior and made certain predictions; when the subsequent observations didn't agree with the predictions, we went looking for the discrepancy and found a previously unknown non-linearity.
  2. Boundary conditions: for systems that have chaotic behavior (like the weather), the smallest changes in initial conditions will change the evolution of the system. To some extent this can be alleviated by repeating the simulation thousands of times with tiny changes in the input - this would give an idea of the variability that can be expected in the output based on this incomplete knowledge
  3. Incomplete systems knowledge: as soon as you have multiple sub-systems that co-exist in a large scale simulation, interactions can appear that you may not have considered. Certain components may be described in a parametric way ("when the current into the base of this transistor goes high, the collector current will increase by beta times the base current"). But if neutron radiation comes along, it may change the properties of the transistor in ways that were not considered in the model. In a system of any complexity, the potential for, and impact of, such interactions becomes significant.

All the above says that simulations can only go so far: and that confidence in the performance of a system can only be gained by testing (components).

That said - I would not be the one suggesting that we do more nuclear tests. I'd be pretty happy if these things didn't work at all.

$\endgroup$
12
$\begingroup$

I think anyone who says "there's no need to do experiments, we can simulate everything!" either:

  1. Doesn't know what they're talking about
  2. Is trying to sell snake oil
  3. Is a scientific fraud trying to push pseudoscience as actual science

I have never seen a serious, honest scientist claim that simulation is sufficient substitute for empirical evidence, even for very simple experiments.

Washington Post news article

Looks like an excellent example of #1.

There is a reason that a critical component of the scientific method is experimental verification of hypotheses. To claim that simulation is as good as an experiment is ridiculous: What if the simulation is wrong? What if there's bugs in the code? What if the assumptions you based it on are false? What if the approximations inherent in your theory (all scientific theory by necessity involves some approximation of reality) cause catastrophic divergence from real-world result? What if there are hidden variables nobody was aware of?

And most importantly, how on Earth will you know the results are real unless you've went and tried it? Even with a competent simulator there are issues, and academics produce notoriously poor-quality code. Without experimental validation, you might as well be making it up. From the article:

But a former nuclear weapons designer, who spoke on the condition of anonymity because he is still in the government, offered a more cautious view. “To say the calculations are better than underground testing is silly,” he said. “If you want to know if something works, you have to test it. The calculations are good, but the issue is one of risk. How good do you think the calculations are?”

And:

Sen. Jon Kyl (R-Ariz.), who has long opposed the treaty, said: “Computer simulation is a part of the stockpile stewardship program, which scientists say has been helpful. One told me it produced good news and bad news. The good news is that it tells us a lot more about these weapons than we ever knew before. The bad news is that it tells us the weapons have bigger problems that we realized. While computers are helpful, they’re not a substitute for testing. That’s why, even though we’re not testing right now, we should not give up the legal right to test.”

And:

Jeffrey G. Lewis, a nuclear weapons expert and the director of the East Asia Nonproliferation Program at the Monterey Institute of International Studies, said that years of underground nuclear tests helped show weapons designers that the bombs worked under certain conditions, “but they could never fully explain how or why.”

Emphasis mine - although Jeffrey Lewis is apparently in favor of the policy, his remark illustrates the crucial point: The universe operates on complex laws. Not all of these laws are known. It is impossible to guarantee that a simulation based on known laws will agree to reality. In some cases the disagreement may be insignificant, but there is no way to know a priori that it will be for any given case.

EDIT to add yet one more problem with relying on simulations: If you look at the past experiments conducted to validate simulations, you may see that they conform very closely and get the idea that the possible sources of error I mention are insignificant, and that I am making a mountain out of a molehill. However, you would be ignoring certain important nuances.

The moment you start eschewing empirical validation, the character of the research work: It is a lot easier to make predictions when you know nobody will actually test those predictions and catch you if you're wrong. Moreover, you fall into a bizarre trap similar to skipping controls: You have some simulations that you validate, and sometimes they turn out false and you have to scrap your manuscript, but you have other simulations that you don't validate, and they never need to be re-done. What is the point of the validation then, if all it does is slow you down? (To connect with the analogy - "what is the point of doing controls, if all they do is make experiments work less often?")

Of course, the answer is evident: The experiment is worthless without the controls (this holds even if over your scientific career you have performed 1000 controlled experiments, and never once have the controls given you any new information). Likewise, the simulation is worthless without the empirical validation. But if one accepts that empirical validation is dispensable, then a contradiction arises: Why not dispense with validation even more? After all, you only win and you "can't lose". Because of this, and coupled with various human flaws to which all scientists are subject, once empirical validation stops being "always mandatory, no excuses", it won't be long before everybody is coming up with made up theories that have nothing to do with reality and no trace of genuine science remains in the community. This is the big danger that "really" makes it important to do empirical validation of simulations.

This isn't to be interpreted as "we have to do nuclear testing or all science will be ruined forever". I don't personally mind that the testing stops. However, once the decision to stop testing stops, a serious person should admit to themselves that certain scientific knowledge which was to be obtained from those tests is now inaccessible, and can no longer be obtained. Saying "we're stopping the tests... but it's okay, we can do simulations!" is like asking to have your cake and eat it too. It's not happening, and you are fooling yourself if you think it is.

And I'm also not trying to claim that making simulations is a pointless business. It's very useful work which can produce valuable scientific results. But it is important to remember that results from a simulation apply only to the simulation, not reality, because reality is not the simulation, does not necessarily resemble the simulation, and most likely differs from the simulation in unpredictable ways. One cannot take the results of a simulation, and claim they apply to the real world: One can only make the argument for performing a proposed real-world experiment, based on the results of the simulation.

But if the experiments are not an option, then the simulation can no longer be connected to reality. At that point the researcher is stuck investigating only the phenomena of his simulated world, not the real one. Once again, there is nothing wrong with that, and arguably the whole discipline of mathematics is concerned with investigating an imaginary world, yet it has produced many findings useful in the real world. However, with the word "science", one is wont to associate investigation of the real world, rather than an imagined, simulated world.


Also, as you can easily see from reading the article, the decision to stop testing is politically motivated. Saying that you decided to stop testing is expedient in the current political climate (even though it's unscientific). Removing the strict test of empirical validation is expedient because it allows easier manipulation of results. Neither On the other hand, a stalwart devotion to scientific principles is unlikely to be appreciated by much of the electorate, and since nuclear war is unlikely during the current administration, losing the ability to design effective nuclear weapons (and likewise for nuclear power plants) is not seen as a big deal. Therefore, the administration makes a [politically] correct decision to stop testing, and claim that simulations are good enough anyway.

$\endgroup$
  • $\begingroup$ I don't believe the claim is that the simulation is perfect, just that it is good enough to maintain the status quo. If someone wanted to build a nuclear weapon of radically different design to those that have been tested, that would be a different matter - but since that would also represent a serious blow to non-proliferation, it probably wouldn't be a good idea anyway. That's a political argument, certainly, but there's a difference between politics determining the context for a scientific decision and politics overriding a scientific decision as you suggest. $\endgroup$ – Harry Johnston Nov 25 '15 at 20:38
  • $\begingroup$ @HarryJohnston To be clear, the issue is not how close the errors are, but that unless you do an experiment, you have no idea how much the error will be in each particular instance. Unless you do the experiment, as far as you know, the simulation could be absolutely wrong. Yes, in the past, usually simulations have turned out to be correct. But it is epistemologically inadmissible to go from this to arguing that the simulation will be as good as the experiment and then argue against doing the experiment. If there is a strong reason for not doing the experiment... $\endgroup$ – Superbest Nov 26 '15 at 0:42
  • $\begingroup$ ...(maybe it is too expensive, or maybe it is unethical, or maybe your funding agency doesn't want you to do it) it is up to you whether you will take the simulation as a kind of poor man's experiment. But if you take results of the simulation and use them to investigate hypotheses about reality (as opposed to hypotheses about the abstract model which you are simulating) then you are not doing science anymore, you are doing something else. That "something else" might still be productive and worth funding, but to call it science just seems silly to me, for the philosophical reasons I gave. $\endgroup$ – Superbest Nov 26 '15 at 0:45
  • $\begingroup$ As for the politics, I felt compelled to touch upon it because the question is motivated by a US govt decision, and the salient aspects of the matter have to do with politics, not science. As such seeking a scientific explanation for the decision is futile - the events are not moved by science, but by politics. If divorced from the context, the question is a valid one that can be answered with a scientific explanation, but the explanation would be at odds with the question's context, which again is unscientific and political. $\endgroup$ – Superbest Nov 26 '15 at 0:48
  • $\begingroup$ then you are not doing science anymore - I think that's the important point, really. They're not doing science (in the sense of "scientific research") and I don't think they claim to be. They're doing engineering. "Will the bomb still work if we do such and such" is analogous to "will the bridge still stay up if we do such and such". Not a research question, unless the answer turns out to depend on some aspect of science we're unsure about. $\endgroup$ – Harry Johnston Nov 26 '15 at 1:04
5
$\begingroup$

You can simulate everything to certain precision, but there is at least one aspect that remains unsimulated: reality.

What I mean are the following points:

  • When you want to calculate anything a little bit more complicated, at some point, you will use approximations. There are always good reasons for making certain assumptions, neglecting terms and effects, etc. - but sometimes somebody misses something. This is also true for for atomic bombs. I doubt that this is a big issue once the chain reaction is on its way, but it might influence whether the bomb explodes at all.

  • When you do a simulation, you are assuming that the device you build looks exactly as you planned. In reality, this will never happen. Obviously, you can try to simulate this, if you know what might get wrong and simulate these wrongly built devices, but if there are serious systematic errors, your simulations won't catch them.

Therefore, I doubt that in order to truly know what is going to happen, you can solely rely on simulations. However, you can certainly reduce the need for experimental tests a lot.

Have a look at space exploration. They are doing extensive simulations, because building spacecraft is incredibly expensive, yet sometimes things go horribly wrong...

$\endgroup$
2
$\begingroup$

This may not be what you're looking for, but don't forget the non-immediate impact of nuclear weapons. Simulations can only go so far to simulate fallout, persistent radiation, flow of radioactive particles through the nearby area, environmental impact, sociological impact, etc.

They're also going to have trouble trying to simulate every possible target environment. If you're trying to level a military base, you can make a pretty good guess at what most of the structures are made of and how they're built, but there may be one or two structures that survive due to lucky placement, superior engineering, etc. And you may be wrong about the target's design.

Finally, you really can't simulate things like the local weather to absolute precision, which could throw off simulations of fallout and so forth (even a brief, 1 mph wind difference could cause a drastic difference if a major population center is right on the edge of danger). These probably aren't a very big effect on immediate damage though; like throwing a peanut at an oncoming supernova, I rather doubt a little wind or slight temperature difference is going to substantially alter the blast wave from a nuke.

To be fair, some of these impacts can't really be tested live, since that would involve nuking major cities, rainforests and purple unicorns to test the impact. Meaning the only valid testing is through simulation based on models from smaller tests of various kinds. Still, fallout and environmental effects can be tested in remote areas of relatively insignificant environments, and real-world testing will probably show things that weren't predicted by the simulations.

$\endgroup$
2
$\begingroup$

I love the quote "Nobody believes an analysis except the person who did it. Everybody believes a test except the person who did it." Both simulation and test have serious shortcomings. You can't do many tests because they are expensive. Simulations do not incorporate all the physics. It is easy to decide incorrectly that some effect is not important and ignore it or to have some aspect of the manufacturing process be important when you didn't think about it. Ideally (from an engineering point of view, not a budgetary one) you do lots of both. When the simulations cannot reproduce the tests, you find out what is wrong with the simulations. Then you update the simulation and decide what test to do next. When the budget is constrained, as it always is, you have to decide what mix of simulation and test to buy. As simulations get better, it is rational to buy more of them and do less tests. Then you worry that the simulation missed some important point. It isn't easy.

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.