The Department of Energy has reduced power levels at its last remaining plutonium-production reactor by more than 10 percent, the third cutback in 14 months, after a stormy confrontation with scientists who warned that the reactor was still operating beyond the demonstrated capacity of its emergency cooling system.

DOE officials confirmed that power levels at the L reactor, one of three government reactors at the Savannah River Plant in South Carolina, were reduced Feb. 12. Savannah River managers called it a "routine" action reflecting a change in the way reactor temperatures are measured.

But other sources said the reduction was ordered only at the insistence of scientists working for the E.I. du Pont de Nemours & Co., DOE's contractor at Savannah River, who had discovered a fundamental error in the calculations that have been used for decades to justify reactor operating levels.

According to several sources, the issue erupted in an acrimonous meeting Feb. 8 at which a top Du Pont scientist threw his security badge on the table in front of Savannah River manager Robert L. Morgan and threatened to "go public" unless DOE reduced power levels.

"Morgan blinked," one source said.

All three of the weapons-production reactors there were cut to 50 percent power last March after a National Academy of Sciences panel warned that a severe accident could overpower emergency cooling systems. The action reduced L reactor's operating level from 2,200 megawatts thermal to 1,100 megawatts.

Du Pont's new calculations, however, showed that the reactor needed to be cut back even more -- to 960 megawatts -- to ensure a "conservative" safety margin while scientists complete a formal assessment of the reactor's ability to withstand accidents.

DOE's initial reluctance to take that step infuriated Du Pont scientists and has raised new questions about the department's commitment to safety. According to several sources, DOE managers at Savannah River resisted the latest cutback because they did not want to admit a mistake and risk kindling public pressure to close the South Carolina reactors.

The L reactor became the last plutonium producer in the nation's aging bomb complex last week, when the department announced that it would not restart the N reactor in Washington state.

"The scientists redid the calculations and they said, 'Whoops, we made a mistake,' " said one source. "DOE wanted to hide the error and the fact that they didn't know what they were doing."

A DOE official acknowledged that the department did not want to "raise the specter of continuing problems at Savannah River. Morgan's view is that it was not a significant change and why bother?"

The incident demonstrates an increasingly knotty problem for the DOE, which is caught in a vise between its weapons-production mission and safety concerns that have mounted since the Chernobyl nuclear accident two years ago in the Soviet Union.

The N reactor became an immediate focus of attention because its nuclear reaction is controlled by graphite, like the Chernobyl reactor. Its retirement leaves the government with only the three Savannah River reactors, which are of a different design.

The Savannah River reactors -- designated by the letters L, P and K -- operate at lower temperatures and pressures than commercial power reactors, which is considered a significant safety advantage. But they lack the concrete containment structures that commercial reactors must have to prevent the release of radioactivity in the event of a major accident, and they are showing signs of corrosion caused by age.

Of the three, the L reactor has generated the most concern because it uses uranium fuel to produce plutonium, a process that results in greater internal heat. The K and P reactors use lithium fuel to produce tritium, a component of nuclear weapons that must be replenished frequently because it decays rapidly.

Seeking to defend its nuclear program against accusations of slipshod science and the subversion of safety to production goals, DOE has since opened many of its highly guarded plants to outside review.

It was one such review, by the academy, that led to last year's 50 percent power reduction at Savannah River. The academy said that reactor operators could not "convincingly demonstrate" that backup cooling systems would cope with a severe accident.

The problem is that the Savannah River reactors, built in the 1950s and initially designed to operate at about 350 megawatts thermal power, have been upgraded repeatedly to increase output. Additional safety equipment was added, but the department has never done the intricate calculations needed to prove that the equipment could protect against a disastrous accident at power levels that have ranged up to nearly 3,000 megawatts.

The power reduction last March was intended to keep decay heat in the reactor below the boiling point of water, a standard that both DOE and the academy panel believed was conservative enough to assure safety until additional research was done. At the time, DOE officials said that the reactors would remain at the lower levels "until the contractor has enough data to show we've got adequate margins of safety at higher power."

Instead, Du Pont scientists discovered that even a 50 percent cutback in power was not sufficient to keep decay heat below the boiling point. The reason is that reducing power in low-pressure reactors, such as those at Savannah River, does not proportionately reduce heat within the reactor.

One source likened the phenomenon to what happens inside an electric coffee pot as the water begins to boil. "When it is boiling actively, heat transfer is very good and the electric coil will actually begin to cool down," he said.

The Du Pont scientists discovered, however, that the phenomenon was not taken into account in the mathematical formula used to set power levels at Savannah River. "They found that they had been running the fuel elements hotter than they thought they had," said one source, who said the formula had been used at Savannah River for decades.

The new calculations showed that the L reactor, which was running at 1,100 megawatts of thermal power, needed to be reduced to 960 megawatts to assure the safety margin agreed upon by DOE and the academy.

According to sources, the Du Pont scientists told DOE managers late last year that they could not defend the higher power level and urged an immediate reduction. After weeks of intensifying conflict, the issue came to a head in the Feb. 8 meeting, when Joseph Spencer, head of the Du Pont laboratory, threw his security badge on the table and, in the words of one source, "dared Morgan to pick it up."

"It was a question of scientific integrity," one source said. "The technical people were pretty stung by the academy's report," which questioned the scientific basis for reactor safety margins that DOE has claimed since the 1950s.

Spencer could not be reached for comment, and calls to Morgan's office were referred to the DOE press office at Savannah River.

Roger Rollins, reactor branch chief at Savannah River, denied that the meeting was acrimonious and explained the cutback as a procedural change.

"It had nothing to do with methodology," he said. "The primary factor was the way we incorporate our cooling water temperature into the limits calculations."

Rollins acknowledged that DOE officials were reluctant to announce the cutback, which was officially confirmed last week after repeated press inquiries. "We are sensitive to the way the rest of the world is sensitive to changes in power levels," he said. "Our concern was: How are we going to be able to explain this?"

Officials at the National Academy of Sciences say they have not been notified of the latest power cutback or the reasons for it, despite what they thought was an understanding with DOE that they would be kept informed.

"This is the first I've heard of it," said Steve Blush, who directed the panel that recommended last year's cutbacks. "This is directly in line with the kind of thing we asked them to do, although it took them awhile to do it."

DOE spokesman C. Anson Franklin said the department had no such understanding with the academy. "We had a policy of notifying them of any finding that would have a bearing on their review," he said. "Their review is over."