2004ApJ...610..261G

We have performed Chandra observations during the past three years of five of the M31 supersoft X-ray sources discovered with ROSAT. Surprisingly, only one of these sources has been detected, despite a predicted detection of about 20-80 counts for these sources. This has motivated a thorough check of the ROSAT M31 Survey I data, including a relaxation of the hardness ratio requirement used to select supersoft sources. This increases the number of supersoft sources identified in Survey I by seven. We then carried out a comparison with the ROSAT M31 Survey II data set, which had hitherto not been explicitly investigated for supersoft X-ray sources. We find that most of the ROSAT Survey I sources are not detected, and only two new supersoft sources are identified. The low detection rate in the ROSAT Survey II and our Chandra observations imply that the variability timescale of supersoft sources is a few months. If the majority of these sources are close binary supersoft sources with shell hydrogen burning, it further implies that half of these sources predominantly experience large mass transfer rates.