Figure 3:

Asymptotic subspace recovery (4) of the ith component as a function of noise variances ${\sigma }_1^2$ and ${\sigma }_2^2$ occurring in proportions p₁ = 70% and p₂ = 30%, where the sample-to-dimension ratio is c = 10 and the subspace amplitude is θ_i = 1. Contours are overlaid in black and the region where A(β_i) ≤ 0 is shown as zero (the prediction of Conjecture 1). Along the dashed cyan line, the average noise variance is ${\stackrel{‒}{\sigma }}^2\approx 1.74$ and the best performance occurs when ${\sigma }_1^2={\sigma }_2^2={\stackrel{‒}{\sigma }}^2$. Along the dashed green curve, the average inverse noise variance is $\mathcal{I}\approx 0.575$ and the best performance again occurs when ${\sigma }_1^2={\sigma }_2^2$.