Smoothing parameter optimization using gradient descent. The map shows a sampled representation of the underlying λ parameter space in terms of the median value of the reduced chi square results. Initial values, tracks, and convergence locations of the (λ₁, λ₂) parameters during the optimization are represented by black circles, black lines, and white crosses, respectively. The red cross marks the global minimum in the sampled parameter space. Initial locations that start off too far from the global best solution (λ₁ = 3.5, λ₂ = 0.6) might converge to local minima with less accurate fit results.