Multivariate Normal Distribution with Covariance Matrix

The multivariate normal distribution describes a d-dimensional random vector using a mean vector and covariance matrix. The compact relation f(x;μ,Σ)=(2π)⁻ᵈᐟ²|Σ|⁻¹ᐟ² exp[−½(x−μ)ᵀΣ⁻¹(x−μ)] shows how the defined quantities interact; x and μ use component units; Σ_ij uses the product of the corresponding units; the density uses the reciprocal product unit. Interpretation depends on the model context: Describe joint measurement errors, correlated features and multivariate statistical models. Also inspect whether a quantity enters linearly, quadratically, inversely or through another function. This lets you predict how the result changes before calculating. The equation is therefore both a calculation rule and a statement about the model's structure.

Assign each given value to a symbol and convert units consistently first. For d=2, μ=(0,0) and Σ=diag(1,4), the density at the mean is f(μ)=1/(2π√4)=1/(4π)≈0.0796 per unit area. Check: |Σ|=4 and the exponential is exp(0)=1 at the centre; larger variances lower the peak. Then verify the result independently: The density requires a symmetric positive-definite covariance matrix, which represents spread and linear dependence Compare the final unit and order of magnitude with the context, and round only after completing the calculation. This check can reveal not only arithmetic mistakes but also a wrong model or a missed factor. When several quantities enter the relation, vary one mentally and ask whether the result follows the expected trend.

The key step is: The density requires a symmetric positive-definite covariance matrix, which represents spread and linear dependence. This leads to f(x;μ,Σ)=(2π)⁻ᵈᐟ²|Σ|⁻¹ᐟ² exp[−½(x−μ)ᵀΣ⁻¹(x−μ)]. The derivation separates definitions, subject-specific assumptions and the final algebraic step; correct algebra cannot replace a missing condition. For d=2, μ=(0,0) and Σ=diag(1,4), the density at the mean is f(μ)=1/(2π√4)=1/(4π)≈0.0796 per unit area. Diagonal entries are variances and off-diagonal entries are covariances Describe joint measurement errors, correlated features and multivariate statistical models. Reconstruct the central idea before memorizing the compact form. That makes it easier to recognize when a similar-looking formula does not apply.

The density requires a symmetric positive-definite covariance matrix, which represents spread and linear dependence. Diagonal entries are variances and off-diagonal entries are covariances. The Mahalanobis distance in the exponent accounts jointly for scales and correlations. If Σ is singular, this is not an ordinary d-dimensional density; the distribution is degenerate on a lower-dimensional subspace. Also check x and μ use component units; Σ_ij uses the product of the corresponding units; the density uses the reciprocal product unit. Violating the validity range can produce a precise but substantively wrong result. Verify that assumptions about geometry, smoothness, independence, symmetry or distribution actually hold in the problem. If a condition is not met, label the result as an approximation or choose a more general method. Extra decimal places cannot repair a model mismatch. This check matters especially when a neighbouring formula uses different assumptions.

The two relations are neighbours in the subject, but answer different questions. Multivariate Normal Distribution with Covariance Matrix: The multivariate normal distribution describes a d-dimensional random vector using a mean vector and covariance matrix. Normal Distribution (Gaussian Bell Curve): The normal distribution is the most important probability distribution, describing many natural phenomena. Choose based on the unknown and on which assumptions the data satisfy. Describe joint measurement errors, correlated features and multivariate statistical models. As a check, use the example For d=2, μ=(0,0) and Σ=diag(1,4), the density at the mean is f(μ)=1/(2π√4)=1/(4π)≈0.0796 per unit area and identify which extra quantity the companion relation would need. This clarifies when one equation is enough and when both belong in a solution.