Jump to content

Linear algebra (Osnabrück 2024-2025)/Part I/Lecture 16/refcontrol



The determinant

Suppose an n×n-matrix is given. Can we see "at a glance“ whether it is invertible?MDLD/invertible (matrix) Does there exist an expression in the n2 entries of the matrix to decide this? This question has a positive answer in terms of the determinant.


Let K be a field,MDLD/field and let M denote an n×n-matrixMDLD/matrix over K with entries aij. For i∈{1,…,n}, let Mi denote the (n−1)×(n−1)-matrix, which arises from M, when we remove the first column and the i-th row. Then one defines recursively the determinant of M by

det⁡M={a11, for n=1,∑i=1n(−1)i+1ai1det⁡Mi for n≥2.

The determinant is only defined for square matrices. For small n, the determinant can be computed easily.



For a 2×2-matrixMDLD/matrix

M=(abcd),

we have

det⁡(abcd)=ad−cb.
For a 3×3-matrix, we can use the rule of Sarrus to compute the determinant. We repeat the first column as the fourth column and the second column as the fifth column. The products of diagonals from up to down enter with a positive sign, and the products of the other diagonals enter with a negative sign.

For a 3×3-matrixMDLD/matrix M=(a11a12a13a21a22a23a31a32a33), we have

det⁡(a11a12a13a21a22a23a31a32a33)=a11a22a33+a12a23a31+a13a21a32−a13a22a31−a11a23a32−a12a21a33.

This is called the rule of Sarrus.


LemmaLemma 16.4 change

For an upper triangular matrixMDLD/upper triangular matrix

M=(b1∗⋯⋯∗0b2∗⋯∗⋮⋱⋱⋱⋮0⋯0bn−1∗0⋯⋯0bn),
we have
det⁡M=b1b2⋯bn−1bn.
In particular, for the

identity matrixMDLD/identity matrix we get

det⁡En=1.

Proof  

This follows with a simple induction directly from the recursive definition of the determinant.MDLD/determinant (rec)

◻



Multilinear and alternatings mappings

Wir führen zwei Begriffe ein, the we im Moment hauptsächlich zum weiteren Verständnis the Determinante brauchen.


Let K be a field,MDLD/field and let V1,…,Vn and W be vector spacesMDLD/vector spaces over K. A mappingMDLD/mapping

Φ:V1×⋯×Vn⟶W

is called multilinear if, for every i∈{1,…,n} and every (n−1)-tuple (v1,…,vi−1,vi+1,…,vn) with vj∈Vj, the induced mapping

Vi⟶W,vi⟼Φ(v1,…,vi−1,vi,vi+1,…,vn),

is

K-linear.MDLD/linear (map)

For n=2, this property is called bilinear. For example, the multiplikation in a field K, that is, the mapping

K×K⟶K,(x,y)⟼xy,

is bilinear. Also, for a K-vector spaceMDLD/vector space V and its dual spaceMDLD/dual space V∗, the evaluation mapping

V×V∗⟶K,(v,f)⟼f(v),

is bilinear.


LemmaLemma 16.6 change

Let K be a field,MDLD/field and let V1,…,Vn and W be vector spacesMDLD/vector spaces over K. Let

Φ:V1×⋯×Vn⟶W

be a multilinear mapping,MDLD/multilinear mapping and let vi1,…,vimi∈Vi and aij∈K for i=1,…,n and j=1,…,mi. Then

Φ(∑j=1m1a1jv1j,…,∑j=1mnanjvnj)=∑(j1,…,jn)∈{1,…,m1}×⋯×{1,…,mn}a1j1⋯anjnΦ(v1j1,…,vnjn).

Proof

◻



Let K be a field,MDLD/field let V and W denote K-vector spaces,MDLD/vector spaces and let n∈ℕ. A multilinear mappingMDLD/multilinear mapping

Φ:Vn=V×⋯×V⏟n-mal⟶W

is called alternating if the following holds: Whenever in v=(v1,…,vn), two entries are identical, that is, vi=vj for a pair i≠j, then

Φ(v)=0.

For an alternating mapping, there is only one vector space occurring several times in the product on the left.


LemmaLemma 16.8 change

Let K be a field,MDLD/field let V and W denote K-vector spaces,MDLD/vector spaces and let n∈ℕ. Suppose that

Φ:Vn=V×⋯×V⏟n-mal⟶W

is an alternating mapping.MDLD/alternating mapping Then

Φ(v1,…,vr−1,vr,vr+1,…,vs−1,vs,vs+1,…,vn)=−Φ(v1,…,vr−1,vs,vr+1,…,vs−1,vr,vs+1,…,vn).
This means that if we swap two vectors, them the sign is changing.

Proof  

Due to the definition of alternating and Lemma 16.6 , we have

0=Φ(v1,…,vr−1,vr+vs,vr+1,…,vs−1,vr+vs,vs+1,…,vn)=Φ(v1,…,vr−1,vr,vr+1,…,vs−1,vs,vs+1,…,vn)+Φ(v1,…,vr−1,vs,vr+1,…,vs−1,vr,vs+1,…,vn).
◻



The determinant is an alternating mapping

We want to show that the recursively defined determinant is a multilinear and alternating mapping. To make sense of this, we identify

Matn(K)≅(Kn)n,

that is, we identify a matrix with the n-tuple of its rows. Thus, in the following, we consider a matrix as a column tuple

(v1⋮vn),

where the entries vi are row vectors of length n.


TheoremTheorem 16.9 change

Let K be a field, and n∈ℕ+. Then the determinantMDLD/determinant

Matn(K)=(Kn)n⟶K,M⟼det⁡M,

is multilinear.MDLD/multilinear This means that for every k∈{1,…,n}, and for every choice of n−1 vectors v1,…,vk−1,vk+1,…,vn∈Kn, and for any u,w∈Kn, the identity

det⁡(v1⋮vk−1u+wvk+1⋮vn)=det⁡(v1⋮vk−1uvk+1⋮vn)+det⁡(v1⋮vk−1wvk+1⋮vn)

holds, and for s∈K, the identity

det⁡(v1⋮vk−1suvk+1⋮vn)=sdet⁡(v1⋮vk−1uvk+1⋮vn)
holds.

Proof  

Let

M:=(v1⋮vk−1uvk+1⋮vn),M′:=(v1⋮vk−1wvk+1⋮vn) and M~:=(v1⋮vk−1u+wvk+1⋮vn),

where we denote the entries and the matrices arising from deleting a row in an analogous way. In particular, u=(ak1,…,akn) and w=(ak′1′,…,ak′n′). We prove the statement by induction over n,  For i≠k, we have a~i1=ai1=a'i1 and

det⁡M~i=det⁡Mi+det⁡M'i

due to the induction hypothesis. For i=k, we have Mk=Mk′=M~k and a~k1=ak1+a'k1. Altogether, we get

det⁡M~=∑i=1n(−1)i+1a~i1det⁡M~i=∑i=1,i≠kn(−1)i+1ai1(det⁡Mi+det⁡M'i)+(−1)k+1(ak1+a'k1)(det⁡M~k)=∑i=1,i≠kn(−1)i+1ai1det⁡Mi+∑i=1,i≠kn(−1)i+1ai1det⁡M'i+(−1)k+1ak1det⁡Mk+(−1)k+1a'k1det⁡Mk=∑i=1n(−1)i+1ai1det⁡Mi+∑i=1,i≠k,n(−1)i+1ai1det⁡M'i+(−1)k+1a'k1det⁡Mk=∑i=1n(−1)i+1ai1det⁡Mi+∑i=1n(−1)i+1a'i1det⁡M'i=det⁡M+det⁡M′.

The compatibility with the scalar multiplication is proved in a similar way, see Exercise 16.22 .

◻



TheoremTheorem 16.10 change

Let K be a fieldMDLD/field and n∈ℕ+. Then the determinantMDLD/determinant

Matn(K)=(Kn)n⟶K,M⟼det⁡M,

is

alternating.MDLD/alternating

Proof  

We proof the statement by induction over n,  So suppose that n≥2 and set M=(v1⋮vn)=(aij)ij. Let vr and vs with r<s be the relevant row. By definition, we have det⁡M=∑i=1n(−1)i+1ai1det⁡Mi. Due to the induction hypothesis, we have det⁡Mi=0 for i≠r,s, because two rows coincide in these cases. Therefore,

det⁡M=(−1)r+1ar1det⁡Mr+(−1)s+1as1det⁡Ms,

where ar1=as1. The matrices Mr and Ms consist in the same rows, however, the row z=vr=vs is in Mr the (s−1)-th row and in Ms the r-th row. All other rows occur in both matrices in the same order. By swapping altogether s−r−1 times adjacent rows, we can transform Mr into Ms. Due to the induction hypothesis and Lemma 16.8 , their determinants are related by the factor (−1)s−r−1, thus det⁡Ms=(−1)s−r−1det⁡Mr. Using this, we obtain

det⁡M=(−1)r+1ar1det⁡Mr+(−1)s+1as1det⁡Ms=ar1((−1)r+1det⁡Mr+(−1)s+1(−1)s−r−1det⁡Mr)=ar1((−1)r+1+(−1)2s−r)det⁡Mr=ar1((−1)r+1+(−1)r)det⁡Mr=0.
◻


The property of the determinant to be alternating simplifies its computation. In particular, it is clear how the determinant behaves under elementary row operations. If a row is multiplied with a number s, the determinant has to be multiplied with s as well. If two rows are swapped, then the sign of the determinant changes. If a row (or a multiple of a row) is added to another row, then the determinant does not change.


TheoremTheorem 16.11 change

Let K be a field,MDLD/field and let M denote an n×n-matrixMDLD/matrix

over K. Then the following statements are equivalent.
  1. We have det⁡M≠0.
  2. The rows of M are linearly independent.MDLD/linearly independent
  3. M is invertible.MDLD/invertible (matrix)
  4. We have rk⁡M=n.

Proof  

The relation between rank, invertibility and linear independence was proven in Corollary 12.16 . Suppose now that the rows are linearly dependent.MDLD/linearly dependent After exchanging rows, we may assume that vn=∑i=1n−1sivi. Then, due to Theorem 16.9 and Theorem 16.10 , we get

det⁡M=det⁡(v1⋮vn−1∑i=1n−1sivi)=∑i=1n−1sidet⁡(v1⋮vn−1vi)=0.

Now suppose that the rows are linearly independent. Then, by exchanging of rows, scaling and addition of a row to another row, we can transform the matrix successively into the identity matrix. During these manipulations, the determinant is multiplied with some factor ≠0. Since the determinant of the identity matrix is 1, due to Lemma 16.4 , the determinant of the initial matrix is ≠0.

◻



In case K=ℝ, the determinantMDLD/determinant is in tight relation to volumes of geometric objects. If we consider in ℝn vectors v1,…,vn, then they span a parallelotope. This is defined by

P:={s1v1+⋯+snvn∣si∈[0,1]}.

It consists of all linear combinationsMDLD/linear combinations of these vectors, where all the scalars belong to the unit interval. If the vectors are linearly independent, then this is a "voluminous“ body, otherwise it is an object of smaller dimension. Now the relation

vol⁡P=|det⁡(v1,…,vn)|

holds, saying that the volume of the parallelotope is the modulus of the determinant of the matrix, consisting of the spanning vectors as columns (or rows).

For a proof of the relation between determinant and volume just mentioned, see Fact *****.


<< | Linear algebra (Osnabrück 2024-2025)/Part I | >>
PDF-version of this lecture
Exercise sheet for this lecture (PDF)