Toolbox

T1: What is Determinant

Reminder: This post contains 4022 words · 12 min read · by Xianbin

What is Determinant?

In Wiki, it says that

In mathematics the determinant is a scalar-valued function of the entries of a square matrix.

In Geometry, Determinant is a Volume

Since it is a volume, the output must be a value!

Let us put it more simple. Determinant is function ff: Rn××RnR\mathbb{R}^n\times \cdots \times \mathbb{R}^n \to \mathbb{R}.

Let a1,a2\textbf{a}_1,\textbf{a}_2 be vectors of size two.

f(a1,a2)=Det(a1,a2)f(\textbf{a}_1,\textbf{a}_2) = \textup{Det}(\textbf{a}_1,\textbf{a}_2)

If we increase a1\textbf{a}_1 to be ca1c\textbf{a}_1 where c>1c > 1, then the volume increases cc times.

property(1)

1):f(ca1,a2)=cf(a1,a2)1): f(c\textbf{a}_1,\textbf{a}_2) = cf(\textbf{a}_1,\textbf{a}_2)

That is simple, right?

By similar tricks, we have that

2):f(a1+v,a2)=f(a1,a2)+f(v,a2)2): f(\textbf{a}_1+\textbf{v},\textbf{a}_2) = f(\textbf{a}_1,\textbf{a}_2) +f(\textbf{v},\textbf{a}_2)

Now, let us consider the exchange of two vectors in ff.

f(a1,,ak,,ai,,an)f(\textbf{a}_1,\ldots, \textbf{a}_k, \ldots,\textbf{a}_i,\ldots,\textbf{a}_n)

We know that the absolute value cannot change. Notice that due to the definition of determinant, the notion sign will change.

Then,

3):f(a1,,ak,,ai,,an)=f(a1,,ai,,ak,,an)3): f(\textbf{a}_1,\ldots, \textbf{a}_k, \ldots,\textbf{a}_i,\ldots,\textbf{a}_n) = - f(\textbf{a}_1,\ldots, \textbf{a}_i, \ldots,\textbf{a}_k,\ldots,\textbf{a}_n)

Finally, we know that if the vectors all unit vectors, we have

(4)f(e1,,ei,,ek,,en)=1(4) f(\textbf{e}_1,\ldots, \textbf{e}_i, \ldots,\textbf{e}_k,\ldots,\textbf{e}_n)=1 where eiej\textbf{e}_i \perp \textbf{e}_j for any iji\not = j. That is because, the volume can only be one.

Now, let us see Det(a1,a2)\textup{Det}(\textbf{a}_1,\textbf{a}_2). By property (1),(2),(3),(4).

Det(a1,a2)=Det((v1v2),(v3v4))=Det((v10)+(0v2),(v3v4))=Det((v10),(v3v4))+Det((0v2),(v3v4))=v1Det((10),(v3v4))+v2Det((01),(v3v4))=v1v3Det((10),(10))+v1v4Det((10),(01))+v2v3Det((01),(10))+v2v4Det((01),(01))=v1v4v2v3\textup{Det}(\textbf{a}_1,\textbf{a}_2) = \textup{Det}\left( {v_1 \choose v_2}, {v_3 \choose v_4} \right)\\= \textup{Det}\left( {v_1 \choose 0} + {0 \choose v_2}, {v_3 \choose v_4} \right)\\= \textup{Det}\left( {v_1 \choose 0} , {v_3 \choose v_4} \right) + \textup{Det}\left( {0 \choose v_2}, {v_3 \choose v_4} \right)\\= v_1\textup{Det}\left( {1 \choose 0} , {v_3 \choose v_4} \right) + v_2\textup{Det}\left( {0 \choose 1}, {v_3 \choose v_4} \right)\\ = v_1v_3 \textup{Det}\left( {1 \choose 0}, {1 \choose 0}\right) +v_1v_4\textup{Det}\left( {1 \choose 0}, {0 \choose 1}\right) \\+v_2v_3\textup{Det}\left( {0 \choose 1}, {1 \choose 0}\right) + v_2v_4\textup{Det}\left( {0 \choose 1}, {0 \choose 1}\right) \\= v_1 v_4 - v_2v_3

Leibniz Formula for Determinants

Let us calculate Det(a1,,an)\textbf{Det}(\textbf{a}_1,\ldots,\textbf{a}_n).

Det((v11vn1)(v12vn2),,(vn1vn1))=Det(v11e1+,vn1en,Δ1)=v11Det(e1,Δ1)++vn1Det(en,Δ1)=j1=1nvj1,1Det(ej1,Δ1)\textup{Det}\left(\left( \begin{matrix} v_{11}\\ \vdots\\ v_{n1} \end{matrix}\right) \left( \begin{matrix} v_{12}\\ \vdots\\ v_{n2} \end{matrix}\right), \cdots, \left( \begin{matrix} v_{n1}\\ \vdots\\ v_{n1} \end{matrix}\right) \right) = \\ \textup{Det}\left( v_{11}e_1+\ldots,v_{n1}e_n, \Delta_1 \right)=\\ v_{11}\textup{Det}(e_1,\Delta_1)+\ldots+v_{n1}\textup{Det}(e_n,\Delta_1)\\= \sum_{j_1=1}^n v_{j_1,1}\textup{Det}(e_{j_1},\Delta_1)

where Δ1=((vn2vn2),,(vn1vn1))\Delta_1 = \left(\left( \begin{matrix} v_{n2}\\ \vdots\\ v_{n2} \end{matrix}\right), \cdots, \left( \begin{matrix} v_{n1}\\ \vdots\\ v_{n1} \end{matrix}\right)\right)

Now, we obtain a recursion equality.

Det(a1,Δ1)=j1=1nvj1,1Det(ej1,Δ1)\textup{Det}(\textbf{a}_1,\Delta_1) = \sum_{j_1=1}^n v_{j_1,1}\textup{Det}(e_{j_1},\Delta_1)

We can see that the relationship is this:

First, we swap a2\textbf{a}_2 to ej1e_{j1}, then, we have

Det(a1,a2,Δ2)=j1=1nvj1,1Det(a2,ej1,Δ2)(1)=j1=1nj2=1nvj1,1vj2,2Det(ej2,ej1,Δ2)(1)\textup{Det}(\textbf{a}_1,\textbf{a}_2,\Delta_2) = \\ \sum_{j_1=1}^n v_{j_1,1} \textup{Det}(\textbf{a}_2,e_{j_1},\Delta_2) \cdot (-1) \\ =\sum_{j_1=1}^n \sum_{j_2=1}^n v_{j_1,1} v_{j_2,2} \textup{Det}(e_{j2},e_{j_1},\Delta_2) \cdot (-1)

Then, we swap it back.

Det(a1,Δ1)=Det(a1,a2,Δ2)=j1=1nj2=1nvj1,1vj2,2Det(ej1,ej2,Δ2)(1)(1)=j1=1nj2=1nvj1,1vj2,2Det(ej1,ej2,Δ2)\textup{Det}(\textbf{a}_1,\Delta_1) =\\ \textup{Det}(\textbf{a}_1,\textbf{a}_2,\Delta_2) = \sum_{j_1=1}^n \sum_{j_2=1}^n v_{j_1,1} v_{j_2,2} \textup{Det}(e_{j_1}, e_{j_2},\Delta_2) \cdot (-1) \cdot(-1) \\= \sum_{j_1=1}^n \sum_{j_2=1}^n v_{j_1,1} v_{j_2,2} \textup{Det}(e_{j_1}, e_{j_2},\Delta_2)

We construct Δn1Δ1\Delta_{n-1}\subset\cdots\subset\Delta_{1} such that Δi+1ai+1=Δi\Delta_{i+1} \cup \textbf{a}_{i+1} = \Delta_{i}, where Δn1=ejn\Delta_{n-1} = e_{jn} and i[n2]i\in [n-2].

Then, we have

Det(a1,a2,,an)=j1=1nj2=1njn=1nvj1,1vjn,nDet(ej1,,Δn1)=δ=(j1,,jn)πnsgn(j1,,jn)i=1nvjδ(i),i\textup{Det}(\textbf{a}_1, \textbf{a}_2,\ldots, \textbf{a}_n) = \sum_{j_1=1}^n \sum_{j_2=1}^n \ldots \sum_{j_n=1}^nv_{j_1,1}\ldots v_{j_n,n}\textup{Det}(e_{j_1},\ldots,\Delta_{n-1})\\= \sum_{\delta = (j_1,\ldots,j_n)\in \pi_n}\text{sgn}(j_1,\ldots,j_n) \prod_{i=1}^n v_{j_{\delta(i)}, i}

where πn\pi_n is a permutation of {1,,n}\{1,\ldots,n\}.

Reference

[1] The Bright Side of Mathematics. Linear Algebra 45,46.