For those interested in going a little deeper than the press release to trace some of the twists and turns of this story, this FAQ gives links to the primary references and addresses a few key questions.
Summary with links to the press releases and the scientific papers
What is the key data plot and how is it interpreted?
Why were the reported 1999 eclipse times off by one hour?
So what is causing the period changes in KIC 9832227?
If this binary star is not the next to merge, which one is?
Given the advantage of hindsight, did we jump the gun in our first report? (Question added September 11)
Summary with links to the press releases and the scientific papers A January 2017 press release announced the prediction by Larry Molnar and his team of the impending merger of a binary star system known as KIC 9832227. (Supplemental images for that release.) The prediction was based on observations of changes in the orbital period as measured from 1999 to 2016. A September 2018 press release updates the story with the discovery by Quentin Socia and his team of a one hour offset in the 1999 eclipse timing, which invalidates the basis for the merger prediction.
What is the key data plot and how is it interpreted? The fundamental quantity that is measured is the time of mid-eclipse. Each estimate of this time is based on brightness measurements from many (often hundreds) of images. In the plot below, this time is given relative to what would be expected for a fixed orbital period. The horizontal axis spans nearly 20 years of observations while the vertical axis spans a range of 100 minutes. The merging star model is shown as a purple line. The slope of this line gradually changes from positive to negative, reflecting a decrease in the orbital period. This is turn implies a decreasing orbital separation.
This updated plot shows two new things. On the lower right side, the symbols marked in red are Calvin College data taken after the merger prediction was made. They appear to support the merger prediction as they lie close to the extrapolation of the purple line. On the upper left side, the two black squares are the new estimates presented by Socia and his team. The one in the upper left corner is a corrected value for the red triangle (the 1999 data from NSVS) found below it, offset by 60 minutes. (The next section describes how the offset came about.) The second black square, below and to the right of the first, is from the newly analyzed Vulcan data set. The plot can be viewed at full resolution by clicking on it.
With the new data points on the left, the slope there is not as different from the slope on the right side as originally thought. Therefore the basis for the merging model extrapolation (the purple line) is fundamentally altered.
Why were the reported 1999 eclipse times off by one hour? The 1999 data point is computed using observations from the Northern Sky Variability Survey (NSVS). This data archive is available online and is described in a paper published by Wozniak et al. in April of 2004. A key part of the description is the units used to record the time of observation, called Modified Julian Date. In Table 6 of that paper, the "modification" is defined to be the subtraction of the number 2400000. This was used by Molnar et al. to compute the original estimate for 1999, the red triangle on the plot. On the other hand, Socia et al. used the offset 2400000.5 to compute the black square, taking the definition of "modification" from Table 6 of a preprint version of the paper of Wozniak et al. that appeared in January of 2004. The definitions differ by exactly 0.5 days or 12 hours. When placed on the plot the difference is only 1 hour as the extra 11 hours is one full orbit of the binary star (which brings the system back to the point it was at the beginning of the 11 hours).
So two versions of the paper yield two possible answers. How can one tell which is the correct answer and which is the typographical error? Normally, one would expect the published version to be the definitive word. But to establish an unbiased answer to the question, Molnar computed the elevation of the star at the time of the first measurement at the location of the NSVS telescope (which happened to be Los Alamos, New Mexico) once for each definition. As it turned out, the definition in the published version would have required imaging the star while it was below the horizon! Therefore, the preprint version (and the black square) must be the correct answer.
So what is causing the period changes in KIC 9832227? Even in the revised plot, there are real changes in the orbital period with time. Most clearly, the slope on the left side of the plot is less steep than on the right. But in addition, the slope for a few years in the middle of the plot is nearly zero. Together this means the period briefly increased before finally decreasing. If this is not the expected signature of merging, then what is the cause? Part of the argument of the 2017 paper was that other known mechanisms for period change were ruled out (leaving merging as the last remaining option). It remains true that other known mechanisms do not apply. The range of variation is too great for variable starspots too work. Spectroscopy along with the eclipse information rule out a third body model.
Therefore, the period change mechanism must be one that has not been imagined yet. There are precedents for similar variations in some other star systems as well, although they are likewise not understood. Given how thoroughly this star system has been monitored, further observations are desirable in the hopes of finding some pattern that will give a clue to the unknown mechanism.
If this binary star is not the next to merge, which one is? Another part of the argument of the 2017 paper was that distinctive period variations should be visible for a century preceding merger (based on the example of the one documented merger, that of V1309 Sco in 2008). Since in addition red novae are observed in our galaxy about once per decade, the next binary star to merge should be recognizable if its period changes are monitored over the course of a decade. Massive surveys of galactic stars are now underway that may already or will soon contain the evidence for the next merging star. In conclusion, the program laid out in the 2017 paper for how to search for the next merging star remains as interesting as ever. The nontrivial task of sifting the data in these surveys holds the hope of an exciting reward.
Given the advantage of hindsight, did we jump the gun in our first report? In particular, given that we know real changes in period (of mysterious origin) are not uncommon, did we jump the gun in proposing the merger hypothesis?
The answer to this has several parts. We begin by noting we were well aware of random variations. The hope of understanding this astronomical puzzle was in fact my motivation to begin studying contact binary timing in the first place. Only with the publication in 2011 of Tylenda’s paper about the contact binary V1309 Sco that merged and became a red nova in 2008 did we consider that understanding binary mergers was also a topic to be pursued. It was a missed opportunity for the astronomy community that the rapid changes in period of this system were not noticed before the outburst, and so there was never a chance to measure the fundamental properties of the binary (total mass, mass ratio, orbital inclination). As a result, this system cannot be used as a strong test of models of stellar merger that are now being developed.
We do not in general consider it a mistake to look for an important new signal against a background of another sort. Discovery science is almost always done at the margin of detectability. By this, I mean if one waits until the equipment or the circumstances permit a discovery with overwhelming signal to noise, then one has waited until after someone else has already reported the discovery.
At the same time, when one is working in the margins there is an obligation to be cautious before drawing conclusions. And we were in fact cautious in this case. Four years elapsed between our first noticing the declining orbital period and when we published our paper. During this time we worked hard to eliminate all known mechanisms that affect the orbital timing in some systems (as described in the paper). In particular, we ruled out starspots and a third body as causes. Furthermore, we waited until the rate of change was more extreme than had been found in other systems. And during these years we looked to see if the system continued to follow the very specific prediction made by the merging star hypothesis. If the period variations are truly random, there is no reason to expect further variations to follow a specific prediction.
In making the choice when to publish, though, we balanced minimizing the chance of random behavior (but not eliminating it) with the urgency of allowing follow up to be done in a timely way. Certainty could be had by waiting until the star exploded (if indeed it did), but the chance to observe fully the preexplosion state is then lost. This applied both for professional observatories and for amateurs as the star was bright enough for amateurs to monitor ongoing period changes as well. We preferred the possibility of being wrong to that of missing the next V1309 Sco-type outburst.
The final story entailed a string of unlikely coincidences. It is unusual for a published paper to misstate the definition of time used. It is more unusual that using this random offset should produce a data set consistent with a merging hypothesis (as the equation for this has only one shape and few degrees of freedom). It is yet more unusual that the extrapolation of such a baseless hypothesis would accurately prediction further timing variations in the subsequent observations. Yet each of these things happened in this case.
In summary, we do not think we jumped the gun on this idea. We were cautious and yet took a necessary risk of being incorrect. The next time around we will be cautious again. One additional measure we can add is to look for internal consistency on definitions of time!
Posted September 7, 2018 by L. Molnar