Alright, on topic, I've been meaning to post about this lately and making the time for it today.
So not only can individual galaxies act as lenses, but entire clusters can too. This makes images on much larger scales, and can stretch individual galaxies out into arcs many times longer than wide. This takes the entire mass of the cluster - both luminous and dark - and the difference between the mass modeling and observations give the dark matter in the cluster, the stuff we can't directly measure from electromagnetic radiation. The presence of strong lensing features put strong constraints on the total mass distribution of the clusters and in general many images can be formed. Not only can you find long arcs, but these arcs can often be seen more than once. As an example, the cluster in the image below gives us many lines of sight to the blue galaxy, which shows up many times in this image of the lensing cluster. In fact these are all the same object!
The inset shows how these lensed arcs can be inverted to give back a view of what the source may look like. A good lens model will give consistent results between all of the images and as we can see aside from projection effects, the same features are identifiable and present in all of these projections. Now, what's interesting is that not only do we get multiple views of these background objects, the path length along each of these lines of sight is different. This means the light takes a different amount of time to get to us from along each path, so there is a time delay between the images as well. So if something happens at the source, we would see it first in image A, then after some time the change would show up in image B, then in image C and etc. So we could watch a "replay" of an event through a lens like this.
This is particularly significant because in the winter of 2014, multiple images of a supernova were found occurring in the background galaxy seen through the cluster MACS J1149.5+2223. We are able to detect this distant supernova only because of the magnification boost that the lensing effect provides, since it's too distant to see otherwise. The supernova is seen in a cross-like configuration below:
Now, since the host galaxy is multiply imaged, there should be another path the light could take to get to us, which will show up later, after we see this quad-view of the supernova. The mass models of the cluster all predicted it should happen a little more than a year later, and in fact a recurrence of the supernova, now called SN Refsdal after a pioneer of strong lensing studies was seen in December 2015.http://frontierfields.org/2016/01/29/predicted-reappearance-of-supernova-refsdal-confirmed/http://www.spacetelescope.org/news/heic1525/
The supernova should also have been visible in 1998, though no archival images from the time-frame exist.
This is amazing and jaw dropping for a few reasons. First, we are seeing the replay of an event that occurred 10 billion years ago in it's host galaxy, the details of which have been deflected and warped through the curved space-time of this cluster. Second, using gravitational mass models for this cluster we were able to predict the appearance of this object a second time accurately, so it shows that we have a good handle on the gravitational potential of this cluster that the models were all able to give consistent results. Third, we can learn a lot about cosmology, specifically the expansion rate of the universe, and the distribution of matter within the cluster itself. Finally, it's an exquisite example of scientific observation and theory in action. The models made a prediction and observations were able to bear it out.
This event is on the cutting edge of ongoing research, and I wish more media outlets had covered it. This has to be one of the most interesting and awe-inspiring releases from astrophysics I've ever come across.