Context : We report the detection of five high-velocity stars in the core of the globular cluster NGC 2808 .
The stars lie on the red giant branch and show total velocities between 40 and 45 \mathrm { km s ^ { -1 } } .
For a core velocity dispersion \sigma _ { c } = 13.4 \mathrm { km s ^ { -1 } } , this corresponds to up to 3.4 \sigma _ { c } .
These velocities are close to the estimated escape velocity ( \sim 50 \mathrm { km s ^ { -1 } } ) and suggest an ejection from the core .
Two of these stars have been confirmed in our recent integral field spectroscopy data and we will discuss them in more detail here .
These two red giants are located at a projected distance of \sim 0.3 pc from the center .
According to their positions on the color magnitude diagram , both stars are cluster members .

Aims : We investigate several possible origins for the high velocities of the stars and conceivable ejection mechanisms .
Since the velocities are close to the escape velocity , it is not obvious whether the stars are bound or unbound to the cluster .
We therefore consider both cases in our analysis .

Methods : We perform numerical simulations of three-body dynamical encounters between binaries and single stars and compare the resulting velocity distributions of escapers with the velocities of our stars .
If the stars are bound , the encounters must have taken place when the stars were still on the main sequence .
We compare the predictions for a single dynamical encounter with a compact object with those of a sequence of two-body encounters due to relaxation .
If the stars are unbound , the encounter must have taken place recently , when the stars were already in the giant phase .

Results : After including binary fractions and black-hole retention fractions , projection effects , and detection probabilities from Monte-Carlo simulations , we estimate the expected numbers of detections for all the different scenarios .
Based on these numbers , we conclude that the most likely scenario is that the stars are bound and were accelerated by a single encounter between a binary of main-sequence stars and a \sim 10 M _ { \odot } black hole .
Finally , we discuss the origin of previously discovered fast stars in globular clusters , and conclude that the case of NGC 2808 is most likely a representative case for most other detections of fast stars in globular clusters .
We show that with the present analysis we are able to explain high-velocity stars in the clusters M3 and 47 Tucanae with simple dynamical encounters .

Conclusions :