Take the rule xn+1=0.5xn+2. Its equilibrium is where output equals input: x=0.5x+2 gives x∗=4, and indeed 0.5(4)+2=4. Now nudge it by starting at 5. You get 4.5, then 4.25, then 4.125. The distance from 4 goes 1, 0.5, 0.25, 0.125, halving each round, so the system comes back: the equilibrium is stable.
Why halving? For a rule f(x)=mx+c, a point at distance d from x∗ lands at distance md, because f(x∗+d)=x∗+md. If ∣m∣<1, distances shrink; if ∣m∣>1, they grow. For a curved rule, the slope at the equilibrium, f′(x∗), plays the role of m for small nudges. A negative slope can still be stable: xn+1=−0.5xn+6 also rests at 4, and from 5 it goes 3.5, then 4.25, overshooting to alternate sides with each swing half the last.
Being an equilibrium and being stable are separate facts. The rule xn+1=2xn−4 also rests at 4, since 2(4)−4=4. But start at 5 and you get 6, 8, 12: the gap doubles each round. Both rules sit still at 4, and only the first comes back after a push.
Think of it like this. A marble in a bowl versus a marble on an upside-down bowl. Both can sit still at the centre, but a tap sends the first rolling back and the second rolling off. The slope of the rule plays the part of the bowl's shape. The comparison holds for small taps; a hard enough push can knock even the first marble out of its bowl.
Core idea. At an equilibrium, a stable system pulls small disturbances back toward the point; an unstable one pushes them away.