Linear algebra
Introduced| English | Meaning |
|---|---|
| rank/ræŋk/ | The dimension of the column space of a matrix. |
| nullity/ˈnʌlɪti/ | The dimension of the null space of a matrix: how many free variables the equation Ax = 0 has. |
| eigenspace/ˈaɪdʒənspeɪs/ | The subspace consisting of zero and the eigenvectors for one eigenvalue. |
Two measurements, one constraint
- Suppose two sensors report $x_1+2x_2=3$ and $2x_1+4x_2=6$. The second equation repeats the first; two equations need not determine two unknowns.
- Their matrix $A=\begin{pmatrix}1&2\\2&4\end{pmatrix}$ has no zero entries, yet it is singular. We will connect rank 秩, solution sets and eigenvectors. Work over $\mathbb R$ unless another field is stated.
Columns describe a linear map
- An $m\times n$ real matrix maps $\mathbb R^n$ to $\mathbb R^m$. Its column space is the span of its columns: exactly the outputs $A\mathbf x$ it can produce.
- The columns are independent exactly when $A\mathbf c=\mathbf0$ forces $\mathbf c=\mathbf0$. A basis is an independent spanning list; every vector in that space has unique coordinates relative to that basis.
Reduce the whole augmented matrix
- Swap rows, multiply a row by a nonzero scalar, or add a multiple of another row. Apply each operation to both $A$ and the right-hand side in $[A\mid\mathbf b]$ to preserve the solutions.
- A row with zero coefficients and a nonzero right-hand side means no solution. Once a real system is consistent, a free variable gives infinitely many solutions; without free variables, its solution is unique.
- For the sensors, subtract twice row 1 from row 2:
- Set $x_2=t$: all solutions are $(3-2t,t)$, $t\in\mathbb R$. Changing the second reading to 7 instead gives a final row $[0\;0\mid1]$, so no solution exists.
For $A=\begin{pmatrix}1&2\\2&4\end{pmatrix}$ and $\mathbf b=(3,7)$, how many real solutions does $A\mathbf x=\mathbf b$ have?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Subtract twice the first augmented row from the second: the last row is [0, 0 | 1], meaning 0 = 1. A free variable in the coefficient matrix does not cure inconsistency.
Count pivots, then free variables
- Rank is the dimension of the column space of a matrix. It equals the number of pivots. Nullity 零度 is the dimension of the kernel of a linear map. The kernel consists of the solutions of $A\mathbf x=\mathbf0$.
- For $n$ columns, rank + nullity $=n$, with rank at most $\min(m,n)$. A $3\times5$ matrix of rank 2 has nullity 3; a consistent $A\mathbf x=\mathbf b$ then has solution set $\mathbf x_p+\ker A$, an affine space of dimension 3.
How many real solutions?
Use consistency and coefficient rank to classify real linear systems.
A $4 \times 5$ matrix has rank 3. What is the dimension of its null space?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
Rank–nullity: rank + nullity = number of columns. With 5 columns and rank 3, the null space has dimension 5 − 3 = 2.
A real $3\times4$ matrix has rank 2. Describe the solution set of $A\mathbf x=\mathbf b$ when $\mathbf b\in\operatorname{Col}A$ and when $\mathbf b\notin\operatorname{Col}A$. Explain why rank alone does not settle consistency.
Nullity is $4-2=2$. If $\mathbf b\in\operatorname{Col}A$, a particular solution $\mathbf x_p$ exists and all solutions form $\mathbf x_p+\ker A$, an affine plane in $\mathbb R^4$ with infinitely many points. If $\mathbf b\notin\operatorname{Col}A$, there are no solutions. Rank fixes the kernel dimension but does not say whether the given right-hand side is an attainable output.
Which columns form the basis?
- Row reduction preserves the kernel and the dependence relations between columns, but it can change the column space. For our sensor matrix, the reduced matrix is $R=\begin{pmatrix}1&2\\0&0\end{pmatrix}$.
- The pivot is in column 1: take column 1 of the original $A$ for a basis of $\operatorname{Col}A$. Thus $\operatorname{Col}A=\operatorname{span}\{(1,2)\}$, whereas $\operatorname{Col}R=\operatorname{span}\{(1,0)\}$. Both kernels are $\operatorname{span}\{(-2,1)\}$.
$A=\begin{pmatrix}1&2\\2&4\end{pmatrix}$ row-reduces to $R=\begin{pmatrix}1&2\\0&0\end{pmatrix}$. Which is a basis of the original column space $\operatorname{Col}A$?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Use the pivot position to select column 1 of A, not R. Both original columns lie on the line through (1,2), and the second is twice the first. The two-column list is dependent.
A student uses the nonzero column $(1,0)$ of the reduced sensor matrix as a basis for the original column space. Correct the claim, and find the common kernel of $A=\begin{pmatrix}1&2\\2&4\end{pmatrix}$ and its row-reduced form.
The pivot position is column 1, so use the original column $(1,2)$. The original column space is its span, not the horizontal line spanned by $(1,0)$. Both homogeneous systems reduce to $x_1+2x_2=0$, giving kernel $\{t(-2,1):t\in\mathbb R\}$. Row operations preserve the kernel but need not preserve the column space.
Determinants need a square matrix
- For an $n\times n$ matrix, invertible $\iff\det A\ne0\iff\ker A=\{\mathbf0\}\iff\operatorname{rank}A=n$. Rectangular matrices still have rank and a kernel, but no determinant or two-sided inverse.
- A triangular matrix has determinant equal to the product of its diagonal entries. Swapping two rows reverses the sign; adding a multiple of a different row leaves it unchanged; multiplying one row by $c$ multiplies the determinant by $c$.
Find $\det \begin{pmatrix} 2 & 1 \\ 4 & 5 \end{pmatrix}$.
Difficulty 2/5 · 140 coins + a small bonus on first completion today
For a 2 × 2 matrix, det = ad − bc = 2 · 5 − 1 · 4 = 10 − 4 = 6.
If $A$ is square and $\det A = 0$, then
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Zero determinant means a nontrivial kernel.
$\det A=6$. Swap two rows, then multiply one row of the resulting matrix by 3. What is the new determinant?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
The swap contributes −1 and the row scaling contributes 3: (−1) × 3 × 6 = −18. This scales one row, not the whole matrix.
Eigenvectors and their space
- For a square matrix over a chosen field $\mathbb F$ ($\mathbb R$ or $\mathbb C$), an eigenvector is a nonzero $\mathbf v$ satisfying $A\mathbf v=\lambda\mathbf v$. Eigenvalues in that field are the roots of $\det(\lambda I-A)$.
- An eigenspace 特征空间 is the subspace consisting of zero and the eigenvectors for one eigenvalue. It is $E_\lambda=\ker(A-\lambda I)$. Zero belongs to every eigenspace, but it is never an eigenvector.
Diagonalisation changes coordinates
- An $n\times n$ matrix is diagonalisable over $\mathbb F$ exactly when it has a basis of $n$ eigenvectors in $\mathbb F^n$. Put them in the columns of $P$: $AP=PD$, so $P^{-1}AP=D$, with the eigenvalues in the matching order.
- Having $n$ distinct eigenvalues in that field is sufficient, not necessary. Row reduction is different: $\operatorname{diag}(2,3)$ row-reduces to $I$, changing its eigenvalues. Row operations do not generally preserve eigenvalues; similarity does.
A repeated root needs enough eigenvectors
- Algebraic multiplicity counts a root in the characteristic polynomial; geometric multiplicity is its eigenspace dimension. For each eigenvalue, $1\le\text{geometric}\le\text{algebraic}$.
- Diagonalisation over $\mathbb F$ requires the characteristic polynomial to split into linear factors over $\mathbb F$ and equality of those multiplicities for every eigenvalue. Equivalently, the eigenspace dimensions must add to $n$.
- For $J=\begin{pmatrix}1&1\\0&1\end{pmatrix}$, eigenvalue 1 has algebraic multiplicity 2. Solving $(J-I)\mathbf v=\mathbf0$ forces $v_2=0$: its eigenspace has dimension 1, too few for a basis of $\mathbb R^2$.
- For $I_2$, the same eigenvalue has eigenspace $\mathbb R^2$, dimension 2. The identity is already diagonal; a repeated eigenvalue alone does not prevent diagonalisation.
Which matrix is NOT diagonalisable over the reals?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Eigenvalue 2 twice with a one-dimensional eigenspace.
The dimension of an eigenspace always equals the algebraic multiplicity of its eigenvalue.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
False. Geometric multiplicity (eigenspace dimension) is at least 1 and at most the algebraic multiplicity, and can be smaller: the matrix [[1, 1], [0, 1]] has eigenvalue 1 twice but a one-dimensional eigenspace.
Compare $B=\begin{pmatrix}1&1&0\\0&1&0\\0&0&2\end{pmatrix}$ with $D=\operatorname{diag}(1,1,2)$. Both have eigenvalues 1, 1, 2. Prove that only one is diagonalisable over $\mathbb R$.
For $B$, $(B-I)\mathbf v=0$ forces $v_2=v_3=0$, so $\dim E_1=1$. Also $E_2=\operatorname{span}\{(0,0,1)\}$ has dimension 1. Their dimensions sum to 2, below 3, so $B$ is not diagonalisable. For $D$, $E_1=\operatorname{span}\{(1,0,0),(0,1,0)\}$ and $E_2=\operatorname{span}\{(0,0,1)\}$ supply a basis. The shared spectrum, trace 4 and determinant 2 do not decide diagonalisation.
The field can change the answer
- The real rotation matrix $T=\begin{pmatrix}0&-1\\1&0\end{pmatrix}$ has characteristic polynomial $\lambda^2+1$. It has no real eigenvalues and no real eigenvector basis, so it is not diagonalisable over $\mathbb R$.
- Over $\mathbb C$, its distinct eigenvalues $i,-i$ give a complex eigenvector basis, so it is diagonalisable there. A real symmetric matrix always has an orthonormal real eigenvector basis: $Q^TAQ=D$ with $Q^TQ=I$.
Diagonalisable over ℝ?
Decide whether each 2 × 2 real matrix, written row by row, has a basis of real eigenvectors.
Over which fields is $\begin{pmatrix}0&-1\\1&0\end{pmatrix}$ diagonalisable?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
The characteristic polynomial is λ² + 1. It has no real roots, but has two distinct roots i and −i over ℂ, giving a complex eigenvector basis.
Use an eigenbasis to compute powers
- For $S=\begin{pmatrix}1&2\\2&1\end{pmatrix}$, $S(1,1)=3(1,1)$ and $S(1,-1)=-(1,-1)$. The eigenvectors are perpendicular; divide each by $\sqrt2$ to make unit vectors.
- With $Q=\dfrac1{\sqrt2}\begin{pmatrix}1&1\\1&-1\end{pmatrix}$, we have $Q^TSQ=\operatorname{diag}(3,-1)$. For any nonnegative integer $k$, $S^k\mathbf v=\lambda^k\mathbf v$ on an eigenvector, and $S^k=QD^kQ^T$.
Let $S=\begin{pmatrix}1&2\\2&1\end{pmatrix}$ and $\mathbf v=(1,1)$. Find $c$ such that $S^3\mathbf v=c\mathbf v$.
Difficulty 2/5 · 140 coins + a small bonus on first completion today
S v = 3 v. Apply S three times: S³ v = 3³ v = 27 v. There is no need to multiply out the full matrix cube.
Check the hypotheses first
- Count columns in rank–nullity, and check consistency before counting solutions. Read pivot positions from the reduced matrix, but choose original columns for the original column-space basis.
- Use determinant and invertibility equivalences only for square matrices. For diagonalisation, state the field and count independent eigenvectors; distinguish row operations from similarity transformations.
For $A=\begin{pmatrix}1&0&1\\0&1&1\end{pmatrix}$, assess the claim: “Two pivots mean that A is invertible and all its columns are independent.” Find its kernel and solve $A\mathbf x=(2,3)$.
The map $A:\mathbb R^3\to\mathbb R^2$ has rank 2 and nullity 1. It is onto but not one-to-one; it has no determinant or two-sided inverse. Its three columns are dependent because column 3 is column 1 plus column 2. The kernel is $\operatorname{span}\{(-1,-1,1)\}$. Setting $x_3=t$ gives all solutions $(2-t,3-t,t)$, $t\in\mathbb R$.