This tranquillity wouldn't last forever. G-1a and G-1b were separated by just half a million kilometers, and over the next seven million years gravitational waves would carry away all the angular momentum that kept them apart. When they finally collided, most of their kinetic energy would be converted into an intense flash of neutrinos, faintly tinged with gamma rays, before they merged to form a black hole. At a distance, the neutrinos would be relatively harmless and the "tinge" would carry a far greater sting; even a hundred light years would he uncomfortably close, for organic life. Whether or not the fleshers were still around when it happened, Karpal liked to think that someone would take on the daunting engineering challenge of protecting the Earth's biosphere, by placing a sufficiently large and opaque shield in the path of the gamma ray burst. Now there was a good use for Jupiter. It wouldn't he an easy task, though; Lac G-1 was too far above the ecliptic to be masked by merely nudging either planet into a convenient point on its current orbit.

Lac G-1's fate seemed unavoidable, and the signal reaching TERAGO certainly confirmed the orbit's gradual decay. One small puzzle remained, though: from the first observations, G-la and G-1b had intermittently spiraled together slightly faster than they should have. The discrepancies had never exceeded one part in a thousand—the waves quickening by an extra nanosecond over a couple of days, every now and then—but when most binary pulsars had orbital decay curves perfect down to the limits of measurement, even nanosecond glitches couldn't be written off as experimental error or meaningless noise.

Karpal had imagined that this mystery would be among the first to yield to his solitude and dedication, but a plausible explanation had eluded him, year after year. Any sufficiently massive third body, occasionally perturbing the orbit, should have added its own unmistakable signature to the gravitational radiation. Small gas clouds drifting into the system, giving the neutron stars something they could pump into energy-wasting jets, should have caused Lac G-1 to blaze with X-rays. His models had grown wilder and more daring, but all of them had come unstuck from a lack of corroborative evidence, or from sheer implausibility. Energy and momentum couldn't just be disappearing into the vacuum, but by now he was almost ready to give up trying to balance the books from a hundred light years away.

Almost. With a martyr's sigh, Karpal touched the highlighted name on the screen, and a plot of the waves from Lacerta for the preceding month appeared.

It was clear at a glance that something was wrong with TERAGO. The hundreds of waves on the screen should have been identical, their peaks at exactly the same height, the signal returning like clockwork to the same maximum strength at the same point on the orbit. Instead, there was a smooth increase in the height of the peaks over the second half of the month—which meant that TERAGO's calibration must have started drifting. Karpal groaned, and flipped to another periodic source, a binary pulsar in Aquila. There were alternating weak and strong peaks here, since the orbit was highly elliptical, but each set of peaks remained perfectly level. He checked the data for five other sources. There was no sign of calibration drift for any of them.

Baffled, Karpal returned to the Lac G-1 data. He examined the summary above the plot, and sputtered with disbelief. In his absence, the summary claimed, the period of the waves had fallen by almost three minutes. That was ludicrous. Over 28 days, Lac G-1 should have shaved 14.498 microseconds off its hour-long orbit, give or take a few unexplained nanoseconds. There had to he an error in the analysis software; it must have become corrupted, radiation-damaged, a few random bits scrambled by cosmic rays somehow avoiding detection and repair.

He flipped to a plot showing the period of the waves, rather than the waves themselves. It began as it should have, virtually flat at 3627 seconds, then about 12 days into the data set it began to creep down from the horizontal, slowly at first, but at an ever-increasing rate. The last point on the curve was at 3456 seconds. The only way the neutron stars could have moved into a smaller, faster orbit was by losing some of the energy that kept them apart—and to be three minutes faster, instead of 14 microseconds, they would have needed to lose about as much energy in a month as they had in the past million years.

"Bollocks."

Karpal checked for news from other observatories, but there'd been no activity detected in Lacerta: no X-rays, no UV, no neutrinos, nothing. Lac G-1 had supposedly just shed the energy equivalent of the moon annihilating its antimatter double; even a hundred light years away, that could hardly have passed unnoticed. The missing energy certainly hadn't gone into gravitational radiation; the apparent power increase there was just 17 percent.

And the period had fallen about 5 percent. Karpal did some calculations in his head, then had the analysis software confirm them in detail. The increasing strength of the gravitational waves was exactly what their decreasing period required. Closer, faster orbits produced stronger gravitational radiation, and this impossible data agreed with the formula, every step of the way. Karpal could not imagine a software error or calibration failure that could mangle the data for one source only while magically preserving the correct physical relationship between the power and frequency of the waves.

The signal had to he genuine.

Which meant the energy loss was real.

What was happening out there? Or had happened, a century ago? Karpal looked down a column of figures showing the separation between the neutron stars, as deduced from their orbital period. They'd been moving together steadily at about 48 millimeters a day since observations began. In the preceding twenty-four hours, though, the distance between them had plummeted by over 7,000 kilometers.

Karpal suffered a moment of pure vertiginous panic, but then quickly laughed it off. Such a spectacularly alarming rate of descent couldn't be sustained for long. Apart from gravitational radiation, there were only two ways to steal energy from a massive cosmic flywheel like this: frictional loss to gas or dust, giving rise to truly astronomical temperatures—ruled out by the absence of UV and X-rays—or the gravitational transfer of energy to another system: some kind of invisible interloper, like a small black hole passing by. But anything capable of absorbing more than a fraction of G-1's angular momentum would have shown up on TERAGO by now, and anything less substantial would soon he swept away, like a pebble skipping off a grindstone, or blown apart like an exploding centrifuge.

Karpal had the software analyze the latest data from TERAGO's six nearest detectors, instead of waiting an hour for the full set to arrive. There was still no evidence of any kind of interloper—just the classical signature of a two-body system—but the energy loss showed no sign of halting, or even leveling off.

It was still growing stronger.

How? Karpal suddenly recalled an old idea which he'd briefly considered as an explanation for the minor anomalies. Individual neutrons were always color neutral: they contained one red, one green and one blue quark, tightly bound. But if both cores had "melted" into pools of unconfined quarks able to move about at random, their color would not necessarily average out to neutrality everywhere. Kozuch Theory allowed the perfect symmetry between red, green, and blue quarks to be broken; this was normally an extremely fleeting occurrence, but it was possible that interactions between the neutron stars could stabilize it. Quarks of a certain color could become "locally heavier" in one core, causing them to sink slightly until the attraction of the other quarks buoyed them up; in the other core, quarks of the same color would be lighter, and would rise. Tidal and rotational forces would also come into play.