যখন একটি মাল্টিভিয়ারেট এলোমেলো পরিবর্তনশীল ( এক্স 1 , এক্স 2 , … , এক্স এন )(X1,X2,…,Xn) এর একটি অবিস্মরণীয় কোভারিয়েন্স ম্যাট্রিক্স সি = ( γ আমি জে ) = ( কোভ ( এক্স আই , এক্স জে ) ) থাকেC=(γij)=(Cov(Xi,Xj)) তখন সমস্ত বাস্তব লিনিয়ার সংমিশ্রণের সেট থাকে এক্স i ভিত্তি E = ( এক্স 1 , এক্স 2 , … ,) এর সাথে Xiএকটি এন-n মাত্রিক আসল ভেক্টর স্পেস তৈরি করেXn)E=(X1,X2,…,Xn) and a non-degenerate inner product given by
⟨ এক্স আমি , এক্স ঞ ⟩ =γij .
⟨Xi,Xj⟩=γij .
Its dual basis with respect to this inner product, E∗=(X∗1,X∗2,…,X∗n)E∗=(X∗1,X∗2,…,X∗n), is uniquely defined by the relationships
⟨X∗i,Xj⟩=δij ,
⟨X∗i,Xj⟩=δij ,
the Kronecker delta (equal to 11 when i=ji=j and 00 otherwise).
The dual basis is of interest here because the partial correlation of XiXi and XjXj is obtained as the correlation between the part of XiXi that is left after projecting it into the space spanned by all the other vectors (let's simply call it its "residual", Xi∘Xi∘) and the comparable part of XjXj, its residual Xj∘Xj∘. Yet X∗iX∗i is a vector that is orthogonal to all vectors besides XiXi and has positive inner product with XiXi whence Xi∘Xi∘ must be some non-negative multiple of X∗iX∗i, and likewise for XjXj. Let us therefore write
Xi∘=λiX∗i, Xj∘=λjX∗j
Xi∘=λiX∗i, Xj∘=λjX∗j
for positive real numbers λiλi and λjλj.
The partial correlation is the normalized dot product of the residuals, which is unchanged by rescaling:
ρij∘=⟨Xi∘,Xj∘⟩√⟨Xi∘,Xi∘⟩⟨Xj∘,Xj∘⟩=λiλj⟨X∗i,X∗j⟩√λ2i⟨X∗i,X∗i⟩λ2j⟨X∗j,X∗j⟩=⟨X∗i,X∗j⟩√⟨X∗i,X∗i⟩⟨X∗j,X∗j⟩ .
ρij∘=⟨Xi∘,Xj∘⟩⟨Xi∘,Xi∘⟩⟨Xj∘,Xj∘⟩−−−−−−−−−−−−−−−−√=λiλj⟨X∗i,X∗j⟩λ2i⟨X∗i,X∗i⟩λ2j⟨X∗j,X∗j⟩−−−−−−−−−−−−−−−−−−√=⟨X∗i,X∗j⟩⟨X∗i,X∗i⟩⟨X∗j,X∗j⟩−−−−−−−−−−−−−−√ .
(In either case the partial correlation will be zero whenever the residuals are orthogonal, whether or not they are nonzero.)
We need to find the inner products of dual basis elements. To this end, expand the dual basis elements in terms of the original basis EE:
X∗i=n∑j=1βijXj .
X∗i=∑j=1nβijXj .
Then by definition
δik=⟨X∗i,Xk⟩=n∑j=1βij⟨Xj,Xk⟩=n∑j=1βijγjk .
δik=⟨X∗i,Xk⟩=∑j=1nβij⟨Xj,Xk⟩=∑j=1nβijγjk .
In matrix notation with I=(δij)I=(δij) the identity matrix and B=(βij)B=(βij) the change-of-basis matrix, this states
I=BC .
I=BC .
That is, B=C−1B=C−1, which is exactly what the Wikipedia article is asserting. The previous formula for the partial correlation gives
ρij⋅=βij√βiiβjj=C−1ij√C−1iiC−1jj .
ρij⋅=βijβiiβjj−−−−−√=C−1ijC−1iiC−1jj−−−−−−√ .