Let ff be a β-smooth and α-strongly convex function. If we run GD for TT steps (with step size η=1βη = \frac{1}{β}) we have:

||𝐱(T)𝐱||22eTαβ||𝐱(0)𝐱||22||\mathbf{x}^{(T)} − \mathbf{x}^*||_2^2 ≤ e^{−T\frac{α}{β}} ||\mathbf{x}^{(0)}−\mathbf{x}^*||_2^2

κ=βα\kappa = \frac{\beta}{\alpha} is called the condition number of ff


compare: Gradient descent convergence for β-smooth functions, Gradient descent convergence for α-strongly convex functions