Why must a covariance matrix be positive semidefinite?
MathematicsLevel: IntermediateMultivariate probability
Why must a covariance matrix be positive semidefinite?
Hints
- Consider a linear combination.
- Variance cannot be negative.
Answer
For every vector a, aᵀΣa is Var(aᵀX), which cannot be negative. A negative eigenvalue contradicts a valid covariance model.
Source and licence: SciPy — multivariate probability · BSD-3-Clause · answer: Quantoros editorial