Flight Testing a Centrifuge

On 13 January, Ivanka Barzashka and I gave a briefing at the AAAS on our work regarding Iran’s uranium enrichment capacity.  Joshua Pollack also gave a briefing, which he has described.  Joshua’s analysis is thorough and interesting but I think I would use a different distinction than the “actual” and “nominal” values that he defines.

Pollack shows how the estimates of the capability of Iran’s centrifuge, the IR-1, have declined over time.  That is intriguing but I worry that it makes the calculations that Ivanka and I and others have performed using data reported from International Atomic Energy Agency (IAEA) on-site inspections seem like the next step in a series of similar estimates.  They are not.  There are two very different types of approaches being taken here.  Here I present an analogy that I think might make the differences clear.

Imagine that, instead of centrifuges, we were trying to estimate the performance of a North Korean long-range missile.  One approach is to ask people, perhaps in the intelligence services, who might know and simply cite those values.  That is essentially the origin of most past estimates of Iran’s centrifuge capability.  The problem with these estimates is that they are of unascertainable credibility and when they contradict each other, we have no way to arbitrate among the competing claims.

We might also try to calculate the missile’s performance based on its technical characteristics.  Before the missile ever left the ground, we might have photographs of the missile or individual stages.  From this information we can develop estimates of what the missile might do.

The Koreans use existing rockets as stages for bigger rockets.  If an upper stage looks like a SCUD or a Nodong or some other well-tested rocket, it probably is or at least is very similar.  If that rocket has flown and we know its performance, we are on pretty safe ground using that performance to predict how it would perform as a stage in a new rocket.  We might see a lower stage and note that it is very similar to some Chinese rocket.  Knowing that the North  Koreans crib foreign designs for everything they can, we might safely assume it is a copy of the Chinese model and the performance of the Chinese model could be assumed.  Similarly, if we know that the Iranian centrifuge is a copy of some particular European design, then we can figure that it has the same capacity.  (Of course, we may be mistaken about which European design is being copied and the Iranians may not be able to reproduce the design.)

If there is no analog, that does not mean we are helpless.  Knowledge of the state of North Korean technology allows us to make estimates about what the performance might be.  For example, if we think we know what high temperature materials they have available and what rocket fuels they have experience with, then we can make reasonable estimates of what the specific impulse or efficiency of their rocket engines might be.  Then, with a photograph showing the size of the rocket, we could calculate its range and payload.  With a centrifuge, if we think we know the strength of the materials Iran has available, we can estimate the maximum rotational speeds and, along with the dimensions of the centrifuge, we can calculate what the separative capacity of the centrifuge ought to be if they did the best possible job with the available technology.

Eventually, however,the North Koreans test the missile and the situation changes dramatically.  Radar tracking will very accurately measure key performance parameters.  For example, acceleration and rate of increase in acceleration allow calculation of the specific impulse.  We can also measure stage burn times and range.  Tracking a warhead into reentry will allow calculation of its mass.  Of course, the radar data provide little information about how the North Koreans achieve that performance.  The radar-tracking will reveal whether the rocket is solid or liquid fuel, for example, but doesn’t tell us what type of liquid fuel is used, the materials in the engines, whether a liquid fueled rocket has a single large engine or multiple smaller engines.  (Once we have the performance data, we can, of course, go back and try to make more refined guesses about those characteristics.)

The statements about centrifuge performance are analogous.  Most have been based on citations of people who might know.  Some are calculations based on best-available data about the dimensions and materials used in the centrifuge.  But, once the IAEA began on-site inspections, that is equivalent to a missile flight test. The numbers that come out are of an entirely different nature, they are no longer estimates, but measurements.

There may still be debate about what these measurements mean for potential capability or whether the country may be intentionally underperforming.  For example, we can speculate that the North Koreans add some concrete to their rocket to intentionally make the performance appear lower than it actually is.  And what if the third stage of the rocket explodes?  Then we have the measured performance, but we must also consider what the rocket could do if the problem that led to the explosion were corrected.

But the flight test is what it is and actual numbers come out.   If there is a big discrepancy between the estimates made before the flight test based on the technical characteristics of the rocket and the measurements from the flight test data itself, then it might be useful to consider why the first calculations were so far off. However, it is not useful to average the two to get a better estimate;  the two numbers are simply of different types.

Similarly, the measurements that come from IAEA on-site inspections give us data from which we can calculate centrifuge performance.  Just because other estimates or guesses predate the measurements does not mean that they somehow represent an established precedent from which we measure a deviation.

One thought on “Flight Testing a Centrifuge

  1. First of all, thanks for the kind words about my presentation, and allow me to repay you the same compliment about yours.

    Second, I think the flight-test analogy is quite sensible. But I would refine it slightly: the operations at Natanz are comparable to a flight test with encrypted telemetry — we have some information about it, some of it quite useful, but not the “ground truth” in actual or potential possession of the operators.

    This tweak is not the result of any disagreements about facts, but rather is the result of a difference in perspective or interpretation. We agree that the IAEA reports are the richest and most credible form of evidence that we currently have on the IR-1′s separative power. Nevertheless, since we lack access to the actual measurements (and since the reports are not peer-reviewed scientific publications, meant to obviate the need for every single reader to review the measurements personally), we do have to evaluate them as a form of evidence. They are not immune from defects, and do not contain every piece of information that we might like to have — more on this point in a moment. Nor can they be subjected to the test of attempted duplication by others.

    So, in deriving estimates of separative power per machine from the IAEA reports, it is still necessary to involve other evidence and sources of insight — including “horse sense” about what assumptions are credible and which ones are not. It is clear from your previous exchange with ISIS that there are persistent differences in that area.

    As a result, what you see as a sharp, qualitative difference between two domains — _inferring_ performance based on assorted scraps of insight vs. _measuring_ performance from IAEA documents — I tend to see as a graduated, quantitative difference within a single domain: it’s a question of more data and less uncertainty.

    Having more data has not eliminated all uncertainty. For example, you have described how varying a single (but important) assumption changes the result by a factor of two. On page 10 of the revised (Dec. 8, 2009) version of “Calculating the Capacity of Fordow,” you wrote:

    “The effective separative capacity of an IR-1 is 0.44 kg-SWU/year. This is the basis for the 0.5 kg-SWU/year that we used in our quick calculation in the Bulletin article and is about a quarter or fifth of the value typically used to estimate Iranian enrichment potential. If we take as a worst case that the total amount of enriched material has a U-235 concentration of 0.49 percent, a case that we consider highly unlikely, then the answer doubles to 0.88 kg-SWU/year. Note that the concentration of the product for the period discussed will become available with the results of the next physical inventory
    verification of the agency.”

    This manner of formulating the result is quite appealing. Here, we have a range of not-wildly-implausible estimates, given as an interval with a best estimate (i.e., most credible) figure — although I’m not quite sure whether you meant 0.44 or 0.5. You’ve also noted what additional evidence will narrow that interval further; in the meantime, you have applied professional analytical judgment to determine what regions of the interval are relatively likely or unlikely.

    (As an aside, the only additional piece of information one might have liked to see here would have been an overall confidence level for the estimate, communicating how likely you consider your range of estimates to contain the true figure. This might be 90%, 95%, or perhaps even 99%. The absence of such figures, I’ve concluded somewhat belatedly, make it somewhat misleading to try to aggregate multiple estimates mathematically. In a controlled setting, everyone would produce estimates based on a common confidence level.)

    Of course, based on what else is not in the reports, there are other key assumptions that could be varied, further widening the interval. Drawing on your debate with ISIS, I would point to the average number of machines actually enriching in a given period as being of particular importance. (One need not suppose intentional cloaking of capacity to recognize this as an issue, I believe.)

    None of which is intended as criticism. Your paper was commendable for the clarity of its assumptions and calculations — you “showed all work” in a way that others haven’t (yet). On the other hand, it isn’t the sole possible credible result, for the simple reason that some assumptions are required, and these assumptions are not incontestable.

Leave a Reply