Lawsuit alleges scientific misconduct at U.S. nuclear weapons lab

Peter Williams claims tweaks to a program for modeling a bomb’s explosive trigger make its predictions unreliable. CLAUDIA WILLIAMS

By Adrian ChoJun. 24, 2020 ,

An unusual lawsuit alleges scientific misconduct at Lawrence Livermore National Laboratory in California, one of the United States’s three nuclear weapons labs. Peter Williams, a 50-year-old physicist, worked at Livermore from January 2016 until May 2017, when he says he was fired in retaliation for complaining that his superiors were mishandling a computer program that simulates the detonation of high explosives, undermining their ability to predict how a particular nuclear weapon would perform if used. Williams, who now works at a private research lab, has sued Livermore and seven individuals for reinstatement and $600,000 in damages.

Researchers familiar with the labs say Williams’s allegations should be taken seriously. “If there’s been a cover-up, that’s something that ought to be looked into,” says Raymond Jeanloz, a geophysicist at the University of California, Berkeley, who has been involved with the weapons labs. But he also says the labs implement internal reviews and other measures to ensure the integrity of their work and head off the kind of problem Williams alleges. “This is exactly the kind of thing the people at the lab worry about,” Jeanloz says. Livermore declined to comment on the suit, but in a statement said: “Rigorous debate is a part of the scientific process—the Laboratory does not retaliate against individuals for holding differing opinions.”

The suit, which Williams filed on 22 May, seems quixotic. He is representing himself; to make his case, he needs documents that only the lab can provide; and his complaint centers on a differential equation. Williams spent only a short time at Livermore before he was fired. (In a 12-month performance review Williams included in his suit, his superiors state he wasn’t keeping up with assignments.) Before joining the lab he did two postdocs, taught at City College of San Francisco and Sonoma State University, and worked for 8 years at Agilent Technologies. But Williams is a talented scientist, says Craig Wheeler, an astrophysicist at the University of Texas, Austin, who was his graduate adviser and has subsequently published with him. “He’s a deep, independent thinker,” Wheeler says. “He’s definitely not a crackpot.”

At Livermore, Williams was given the unclassified task of modeling the behavior of a high explosive. Called PBX 9502, the polymer-bonded explosive is intended to be used in refurbishing a 40-year-old thermonuclear warhead called the W80, which is about the size of a garbage can and fits on a cruise missile. The explosive, when detonated, would compress the warhead’s plutonium pit to set off a nuclear fission explosion, which in turn would trigger an even more powerful fusion explosion. Modeling a high explosive is difficult, researchers say, because its behavior depends on its density, which can vary from batch to batch and within a single piece. But such modeling is crucial to be sure that stored weapons will work as intended because the United States gave up nuclear testing in 1992 under the still unratified Comprehensive Nuclear-Test–Ban Treaty.

PBX 9502 is especially tricky because it’s an insensitive explosive, meaning it won’t go up if smacked or set on fire. But that also means that as the detonation front moves through the material, the chemical reactions behind it progress more slowly than in older explosives, Williams says. So instead of assuming the reactions are instantaneous, researchers must model their evolution, too.

To make sure they can do so correctly, researchers detonate samples of the stuff in different geometries, observe the explosions with ultra–high-speed cameras and other tools, and compare the results with simulations. For example, in one test, the sample resembles a mallet, with a thin cylindrical handle that leads to a fatter cylindrical head. Because the detonation front cannot instantly turn the corner where skinny cylinder meets fat, the explosion doesn’t consume the entire sample, but leaves a ring of material whose thickness reveals the speed of the detonation front and chemical reactions.

Livermore used such test explosions to refine a modeling program called ARES-CHEETAH. But Williams says he became aware that one of his supervisors was essentially changing parameters in the program after the fact to make sure the simulation fit the data from each experimental setup. That’s “pseudoscience,” Williams charges, as it guarantees the model will look accurate, even if it isn’t. He says he repeatedly asked the researcher to explain the rationale for the tweaking. “It became clear that he didn’t have a rationale, he had a motive,” Williams says. “And the motive was to keep money flowing to help develop and improve CHEETAH.”

Such “curve fitting” renders the program incapable of making meaningful predictions, he says. If CHEETAH can’t properly model the bench tests of PBX 9502, larger, far more expensive integrated test explosions of the weapon’s nonnuclear components could fail, he says. And modelers may not be able to correctly calculate the yield of the refurbished weapon, he says.

That would be a big problem, says Robert Rosner, a theoretical astrophysicist at the University of Chicago and former director of Argonne National Laboratory who is on Livermore’s advisory board. The uncertainty in the yields of various weapons directly affects calculations of how big the U.S. nuclear arsenal should be and could determine the course of future arms control negotiations, Rosner says. “The extent to which we can decrease [the stockpile] is completely related to the ability to be certain about the yield.”

But Rosner says he’s neither surprised nor alarmed to hear researchers had to adjust the high-explosives models to fit different experiments. What Williams sees as unacceptable tweaking is an inevitable part of the empirical modeling of any complex system, such as the climate or a star, Rosner says. “The idea that the model has to be tuned, I find that unsurprising,” he says. “I would be stunned if it didn’t.”

Both Rosner and Jeanloz say they suspect Williams may not know the whole story behind ARES-CHEETAH, as the lab dismissed him before he obtained a clearance to work on classified material. The weapons program also takes explicit measures to ensure that researchers adhere to scientific standards, Jeanloz says, such as running its own classified peer-reviewed journals and encouraging researchers to keep a hand in unclassified research. The rivalry between Livermore and Los Alamos National Laboratory in New Mexico provides a particularly valuable check, scientists say.

Yet Jeanloz and Rosner agree that models like CHEETAH need careful scrutiny. That’s because some policymakers may view uncertainties in modeling as a reason to resume nuclear testing. In fact, Jeanloz says, some arms control experts argue that to discourage calls for new tests, the United States should make no changes to its existing nuclear arms—even ones that make them safer, such as switching to insensitive explosives.

Williams does not yet have a court date. He’s suing, he says, not out of anger, but out of a sense of scientific duty: “I couldn’t look myself in the mirror if I didn’t do it.”
Posted in:
Science and Policy
Scientific Community

doi:10.1126/science.abd4839