The client-side prediction
If a game employs server-authoritative physics, it usually also needs to employ client-side prediction of player-controlled objects in order to hide the effects of lag. However, this is a more difficult problem than it appears at the outset.
Let's say the server periodically broadcasts the state of all objects. If a player is predicting their own object (and sending inputs to the server), what happens when the player receives a server update? Usually, especially when moving, they snap backward to where they were a moment ago in a motion deemed "rubber-banding". This is because there is always a lag between the client and server.
Client sends input to server and begins moving (resulting in picture 2).
Server receives message and starts moving player (resulting in picture 3).
Server sends current position to client (snapping client back to the old position as shown by the dotted circle). The way Source engine and other games handle this involves a fair bit of bookkeeping...
