Actually, it is not necessary Cholesky.
When we try to generate Brownian motions that are correlated, we can use that formula.
Suppose you have 2 independent Brownian motions W1 and W2, we construct that
W3=p W1 + sqrt(1-p^2) W2; then Cov ( W1 , W3 )=p.
Any application ?
Of courses, you can model “CMS10Y – CMS2Y range accrual”, The correlation is p between cms10 and cms2