Linear Algebra, Matrices and Eigenvalues
CSIR NET Physical Sciences: special matrices, eigenvalues and eigenvectors, Cayley-Hamilton theorem, diagonalisation, matrix exponentials, Pauli matrices.
- 19 min
On this page · 16 sections
Prerequisites
- Dimensional Analysis and Vector Calculus: the scalar triple product is a \( 3\times3 \) determinant, and rotations act on vectors.
- Determinants, rank and solving \( A\mathbf x = \mathbf b \), from Rank, Determinants and Linear Equations.
- The proof-level treatment of the same results is in Eigenvalues and the Cayley–Hamilton Theorem (Mathematical Sciences). This page is the physicist’s toolkit.
The Hook: Two Masses, Three Springs
Matrix Vocabulary
For an \( n\times n \) complex matrix \( A \), write \( A^T \) for the transpose and \( A^\dagger = (A^*)^T \) for the Hermitian conjugate (adjoint).
| Class | Definition | Eigenvalues | Other facts |
|---|---|---|---|
| Real symmetric | \( A^T = A \), real | real | orthogonal eigenvectors; diagonalised by an orthogonal matrix |
| Hermitian | \( A^\dagger = A \) | real | orthogonal eigenvectors; diagonalised by a unitary matrix |
| Anti-Hermitian | \( A^\dagger = -A \) | purely imaginary (or 0) | \( iA \) is Hermitian |
| Real antisymmetric | \( A^T = -A \), real | purely imaginary pairs \( \pm i\lambda \), and 0 | odd \( n \): \( \det A = 0 \) |
| Unitary | \( A^\dagger A = I \) | \( \lvert\lambda\rvert = 1 \) | preserves inner products; \( \lvert\det A\rvert = 1 \) |
| Real orthogonal | \( A^TA = I \) | \( \lvert\lambda\rvert = 1 \) | \( \det A = \pm1 \); rotations have \( \det A = +1 \) |
| Normal | \( AA^\dagger = A^\dagger A \) | any | exactly the matrices unitarily diagonalisable |
| Idempotent (projector) | \( A^2 = A \) | 0 or 1 | \( \operatorname{tr}A = \operatorname{rank}A \) |
| Nilpotent | \( A^k = 0 \) for some \( k \) | all 0 | \( \operatorname{tr}A = 0 \), \( \det A = 0 \) |
| Involutory | \( A^2 = I \) | \( \pm1 \) | Pauli matrices are examples |
Hermitian, anti-Hermitian and unitary matrices are all normal. Any square matrix splits uniquely as Hermitian plus anti-Hermitian: \( A = \tfrac{1}{2}(A + A^\dagger) + \tfrac{1}{2}(A - A^\dagger) \).
Eigenvalues and Eigenvectors
\( A\mathbf v = \lambda\mathbf v \) with \( \mathbf v \neq \mathbf 0 \) has solutions only when
$$\boxed{\det(A - \lambda I) = 0}$$the characteristic equation, a polynomial of degree \( n \) in \( \lambda \). Two consequences save most of the work:
$$\boxed{\operatorname{tr}A = \sum_i\lambda_i, \qquad \det A = \prod_i\lambda_i}$$For a \( 2\times2 \) matrix the characteristic equation is simply
$$\lambda^2 - (\operatorname{tr}A)\,\lambda + \det A = 0.$$Shortcuts that avoid the determinant
- Triangular or diagonal matrices: the eigenvalues are the diagonal entries.
- Functions of a matrix: if \( A\mathbf v = \lambda\mathbf v \), then \( p(A)\mathbf v = p(\lambda)\mathbf v \) for any polynomial \( p \); also \( A^{-1} \) has eigenvalues \( 1/\lambda \) and \( e^A \) has \( e^{\lambda} \), with the same eigenvectors.
- Shift: \( A + cI \) has eigenvalues \( \lambda + c \).
- Rank-one matrices: \( \mathbf u\mathbf v^T \) has one eigenvalue \( \mathbf v^T\mathbf u \) and \( n - 1 \) zeros. The \( n\times n \) all-ones matrix \( J \) therefore has eigenvalues \( n, 0, \ldots, 0 \), and \( aI + bJ \) has \( a + nb \) (once) and \( a \) (\( n - 1 \) times).
- Traces of powers: \( \operatorname{tr}A^k = \sum_i\lambda_i^k \). For a \( 2\times2 \) matrix, \( \operatorname{tr}A \) and \( \operatorname{tr}A^2 \) fix both eigenvalues.
- \( AB \) and \( BA \) have the same eigenvalues (for square matrices).
- Similar matrices \( B = S^{-1}AS \) share eigenvalues, trace, determinant and rank.
Multiplicity and diagonalisability
The algebraic multiplicity of \( \lambda \) is its multiplicity as a root; the geometric multiplicity is the number of independent eigenvectors, \( n - \operatorname{rank}(A - \lambda I) \). Always geometric \( \le \) algebraic. \( A \) is diagonalisable exactly when they are equal for every eigenvalue, and then
$$A = SDS^{-1}, \qquad D = \operatorname{diag}(\lambda_1, \ldots, \lambda_n),$$with the eigenvectors as the columns of \( S \). Distinct eigenvalues guarantee diagonalisability. The standard counter-example is
$$N = \begin{pmatrix}1 & 1\\ 0 & 1\end{pmatrix},$$with \( \lambda = 1 \) twice but only one eigenvector \( (1, 0) \). It is a Jordan block, and \( N^k = \begin{pmatrix}1 & k\\ 0 & 1\end{pmatrix} \).
Commuting matrices
If \( [A, B] = 0 \) and both are diagonalisable, they can be diagonalised by the same similarity transformation: they share a complete set of eigenvectors. In quantum mechanics this is the statement that commuting observables have common eigenstates, and it is how a degeneracy in one operator is resolved by another.
The Cayley–Hamilton Theorem
Every square matrix satisfies its own characteristic equation. If \( p(\lambda) = \det(A - \lambda I) \), then
$$\boxed{p(A) = 0}$$For a \( 2\times2 \) matrix: \( A^2 - (\operatorname{tr}A)A + (\det A)I = 0 \).
Three uses:
- Inverse. If \( A^2 - tA + dI = 0 \) with \( d \ne 0 \), multiply by \( A^{-1} \): \( A^{-1} = (tI - A)/d \).
- High powers. Every power \( A^k \) reduces to a polynomial of degree at most \( n - 1 \) in \( A \).
- Functions. For a \( 2\times2 \) matrix with distinct eigenvalues, \( f(A) = \alpha I + \beta A \), where \( \alpha, \beta \) solve \( f(\lambda_i) = \alpha + \beta\lambda_i \) for both eigenvalues.
The minimal polynomial is the lowest-degree monic polynomial with \( m(A) = 0 \). It divides the characteristic polynomial and has every eigenvalue as a root; \( A \) is diagonalisable exactly when the minimal polynomial has no repeated roots.
Functions of Matrices
For a diagonalisable \( A = SDS^{-1} \), \( f(A) = S\,f(D)\,S^{-1} \), with \( f \) applied to each eigenvalue. The most important case is the exponential, defined by its power series \( e^A = \sum_kA^k/k! \).
- \( \det(e^A) = e^{\operatorname{tr}A} \) (always, diagonalisable or not).
- \( e^Ae^B = e^{A+B} \) only if \( [A, B] = 0 \).
- \( e^{iH} \) is unitary when \( H \) is Hermitian. This is why time evolution \( e^{-iHt/\hbar} \) conserves probability.
- If \( A^2 = I \): \( e^{i\theta A} = \cos\theta\,I + i\sin\theta\,A \).
- If \( A^2 = -I \): \( e^{\theta A} = \cos\theta\,I + \sin\theta\,A \).
Pauli Matrices
$$\sigma_x = \begin{pmatrix}0 & 1\\ 1 & 0\end{pmatrix}, \qquad \sigma_y = \begin{pmatrix}0 & -i\\ i & 0\end{pmatrix}, \qquad \sigma_z = \begin{pmatrix}1 & 0\\ 0 & -1\end{pmatrix}$$- Each is Hermitian, unitary and traceless, with \( \sigma_i^2 = I \), eigenvalues \( \pm1 \) and determinant \( -1 \).
- \( \sigma_i\sigma_j = \delta_{ij}I + i\varepsilon_{ijk}\sigma_k \), so \( [\sigma_i, \sigma_j] = 2i\varepsilon_{ijk}\sigma_k \) and \( \{\sigma_i, \sigma_j\} = 2\delta_{ij}I \).
- For vectors \( \mathbf a, \mathbf b \): \( (\boldsymbol\sigma\cdot\mathbf a)(\boldsymbol\sigma\cdot\mathbf b) = (\mathbf a\cdot\mathbf b)I + i\boldsymbol\sigma\cdot(\mathbf a\times\mathbf b) \).
- For a unit vector \( \hat{\mathbf n} \): \( (\boldsymbol\sigma\cdot\hat{\mathbf n})^2 = I \), so \( e^{i\theta\boldsymbol\sigma\cdot\hat{\mathbf n}} = \cos\theta\,I + i\sin\theta\,\boldsymbol\sigma\cdot\hat{\mathbf n} \). The eigenvalues of \( \boldsymbol\sigma\cdot\mathbf a \) are \( \pm\lvert\mathbf a\rvert \).
- Together with \( I \) they form a basis: any \( 2\times2 \) matrix is \( a_0I + \mathbf a\cdot\boldsymbol\sigma \) with \( a_0 = \tfrac{1}{2}\operatorname{tr}A \). If \( A \) is Hermitian, all four coefficients are real, and the eigenvalues are \( a_0 \pm \lvert\mathbf a\rvert \).
The spin-\( \tfrac{1}{2} \) operators are \( \mathbf S = \tfrac{\hbar}{2}\boldsymbol\sigma \) (see Quantum Mechanics (Core)).
Rotation Matrices
A rotation by angle \( \theta \) about the \( z \)-axis is
$$R_z(\theta) = \begin{pmatrix}\cos\theta & -\sin\theta & 0\\ \sin\theta & \cos\theta & 0\\ 0 & 0 & 1\end{pmatrix}.$$Every proper rotation in three dimensions (real orthogonal, \( \det R = +1 \)) has eigenvalues \( 1, e^{i\theta}, e^{-i\theta} \). The eigenvector with eigenvalue 1 is the rotation axis, and the angle follows from the trace:
$$\boxed{\operatorname{tr}R = 1 + 2\cos\theta}$$A real orthogonal matrix with \( \det = -1 \) is a rotation combined with a reflection; it has an eigenvalue \( -1 \).
Linear Systems in One Paragraph
For \( A\mathbf x = \mathbf b \) with \( A \) square: if \( \det A \ne 0 \) there is a unique solution. If \( \det A = 0 \), the homogeneous system \( A\mathbf x = \mathbf 0 \) has non-trivial solutions (a null space of dimension \( n - \operatorname{rank}A \), by rank–nullity), and the inhomogeneous system has either no solution or infinitely many, depending on whether \( \mathbf b \) lies in the column space (rank of the augmented matrix equals rank of \( A \)). The eigenvalue condition \( \det(A - \lambda I) = 0 \) is exactly the demand for a non-trivial solution of a homogeneous system.
Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| Hermitian: \( A^\dagger = A \) (eigenvalues real) | Unitary: \( A^\dagger = A^{-1} \) (eigenvalues of modulus 1) | |
| Algebraic multiplicity (root count) | Geometric multiplicity (eigenvector count) | |
| \( \det e^A = e^{\operatorname{tr}A} \) always | \( e^{A}e^{B} = e^{A+B} \) only when \( [A,B] = 0 \) | |
| Characteristic polynomial (degree \( n \)) | Minimal polynomial (divides it; no repeated roots iff diagonalisable) | |
| Similar matrices: same eigenvalues | Same eigenvalues does not imply similar (\( I \) and the Jordan block \( N \)) | |
| \( \operatorname{tr}(AB) = \operatorname{tr}(BA) \) | \( \operatorname{tr}(ABC) \ne \operatorname{tr}(BAC) \) in general (only cyclic shifts allowed) |
Common Mistakes
Practice
All questions below are original practice questions written for this page. They are not from any CSIR NET paper.
Level 1: Foundation
- What are the eigenvalues of \( \begin{pmatrix}2 & 5 & 7\\ 0 & 3 & 1\\ 0 & 0 & -1\end{pmatrix} \)?
- A \( 2\times2 \) matrix has trace 5 and one eigenvalue 2. What is its determinant?
- What are the possible eigenvalues of a projection matrix?
Answers
- 2, 3 and −1. Triangular: read the diagonal.
- 6. The other eigenvalue is \( 5 - 2 = 3 \), and \( \det = 2\times3 \).
- 0 and 1. \( P^2 = P \) gives \( \lambda^2 = \lambda \).
Level 2: Part B Style
Practice Q1. The eigenvalues of the \( 4\times4 \) matrix with every entry equal to 1 are:
(a) 1, 1, 1, 1 (b) 4, 0, 0, 0 (c) 4, 1, 1, 1 (d) 2, 2, 0, 0
Answer: (b) 4, 0, 0, 0
\( J = \mathbf u\mathbf u^T \) with \( \mathbf u = (1,1,1,1) \) has rank 1, so three eigenvalues are 0. The trace is 4, so the remaining eigenvalue is 4, with eigenvector \( \mathbf u \). Option (c) has trace 7, but \( \operatorname{tr}J = 4 \); option (d) has the right trace but rank 2; (a) confuses \( J \) with \( I \).
Practice Q2. For real \( \theta \), \( e^{i\theta\sigma_x} \) equals:
(a) \( \begin{pmatrix}\cos\theta & i\sin\theta\\ i\sin\theta & \cos\theta\end{pmatrix} \) (b) \( \begin{pmatrix}\cos\theta & \sin\theta\\ -\sin\theta & \cos\theta\end{pmatrix} \) (c) \( \begin{pmatrix}e^{i\theta} & 0\\ 0 & e^{-i\theta}\end{pmatrix} \) (d) \( \begin{pmatrix}\cosh\theta & \sinh\theta\\ \sinh\theta & \cosh\theta\end{pmatrix} \)
Answer: (a)
\( \sigma_x^2 = I \), so \( e^{i\theta\sigma_x} = \cos\theta\,I + i\sin\theta\,\sigma_x \). Option (c) is \( e^{i\theta\sigma_z} \); (b) is \( e^{i\theta\sigma_y} \); (d) is \( e^{\theta\sigma_x} \) without the \( i \).
Practice Q3. A real \( 3\times3 \) orthogonal matrix has determinant \( +1 \) and trace 2. The angle of the rotation it represents is:
(a) \( 30^\circ \) (b) \( 45^\circ \) (c) \( 60^\circ \) (d) \( 90^\circ \)
Answer: (c) 60°
\( \operatorname{tr}R = 1 + 2\cos\theta = 2 \Rightarrow \cos\theta = \tfrac{1}{2} \Rightarrow \theta = 60^\circ \).
Practice Q4. For \( N = \begin{pmatrix}1 & 1\\ 0 & 1\end{pmatrix} \), which statement is correct?
(a) \( N^{10} = \begin{pmatrix}1 & 10\\ 0 & 1\end{pmatrix} \) and \( N \) is not diagonalisable (b) \( N^{10} = \begin{pmatrix}1 & 1\\ 0 & 1\end{pmatrix} \) and \( N \) is diagonalisable (c) \( N^{10} = \begin{pmatrix}1 & 2^{10}\\ 0 & 1\end{pmatrix} \) and \( N \) is not diagonalisable (d) \( N^{10} = I \)
Answer: (a)
Write \( N = I + E \) with \( E = \begin{pmatrix}0 & 1\\ 0 & 0\end{pmatrix} \) and \( E^2 = 0 \). Since \( I \) and \( E \) commute, \( N^{10} = I + 10E \). The only eigenvalue is 1, and \( \operatorname{rank}(N - I) = 1 \) leaves one eigenvector, so \( N \) is not diagonalisable.
Level 3: Part C Style
Practice Q5. Let \( A = \begin{pmatrix}2 & 1 & 1\\ 1 & 2 & 1\\ 1 & 1 & 2\end{pmatrix} \). Consider:
- The eigenvalues of \( A \) are 4, 1, 1.
- \( A^2 - 5A + 4I = 0 \).
- \( A^{-1} = \frac{1}{4}(5I - A) \).
- \( A \) is not diagonalisable because it has a repeated eigenvalue.
Which are correct?
(a) 1 only (b) 1 and 2 only (c) 1, 2 and 3 only (d) all four
Answer: (c) 1, 2 and 3 only
Statement 1: \( A = I + J \) with \( J \) the all-ones \( 3\times3 \) matrix (eigenvalues 3, 0, 0). So \( A \) has eigenvalues 4, 1, 1. True. Check: trace 6, determinant 4.
Statement 2: \( A \) is real symmetric, hence diagonalisable, so its minimal polynomial has simple roots: \( (A - I)(A - 4I) = 0 \), i.e. \( A^2 - 5A + 4I = 0 \). Directly, \( (A - I)(A - 4I) = J(J - 3I) = J^2 - 3J = 3J - 3J = 0 \). True.
Statement 3: multiply statement 2 by \( A^{-1} \): \( A - 5I + 4A^{-1} = 0 \). True. Explicitly \( A^{-1} = \frac{1}{4}\begin{pmatrix}3 & -1 & -1\\ -1 & 3 & -1\\ -1 & -1 & 3\end{pmatrix} \).
Statement 4: false. A real symmetric matrix is always diagonalisable; the eigenvalue 1 has a two-dimensional eigenspace (all vectors with components summing to zero).
Practice Q6. A \( 2\times2 \) Hermitian matrix \( H \) satisfies \( \operatorname{tr}H = 3 \) and \( \operatorname{tr}H^2 = 5 \). Which of the following are correct?
- The eigenvalues of \( H \) are 1 and 2.
- \( \det H = 2 \).
- \( \operatorname{tr}H^3 = 9 \).
- \( \det e^H = e^2 \).
(a) 1 and 2 only (b) 1, 2 and 3 only (c) 2 and 4 only (d) all four
Answer: (b) 1, 2 and 3 only
Let the eigenvalues be \( \lambda_1, \lambda_2 \). Then \( \lambda_1 + \lambda_2 = 3 \) and \( \lambda_1^2 + \lambda_2^2 = 5 \), so \( 2\lambda_1\lambda_2 = 9 - 5 = 4 \) and \( \lambda_1\lambda_2 = 2 \). The roots of \( \lambda^2 - 3\lambda + 2 = 0 \) are 1 and 2. Statements 1 and 2 true.
Statement 3: \( 1^3 + 2^3 = 9 \). True.
Statement 4: \( \det e^H = e^{\operatorname{tr}H} = e^3 \), not \( e^2 \). False. The distractor confuses \( \det H \) with \( \operatorname{tr}H \).
Practice Q7. Two equal masses \( m \) are joined to each other and to two fixed walls by three springs of constant \( k \), all in a line. The system starts with the left mass displaced by \( a \), the right mass at rest at equilibrium, both velocities zero. Which of the following are correct?
- The normal-mode angular frequencies are \( \sqrt{k/m} \) and \( \sqrt{3k/m} \).
- The left mass moves as \( x_1(t) = \tfrac{a}{2}\left[\cos\omega_1t + \cos\omega_2t\right] \).
- The motion is periodic for every \( k \) and \( m \).
- The total energy is \( ka^2 \).
(a) 1 and 2 only (b) 1, 2 and 4 only (c) 1, 3 and 4 only (d) all four
Answer: (b) 1, 2 and 4 only
Statement 1: \( \ddot{\mathbf x} = -\frac{k}{m}\begin{pmatrix}2 & -1\\ -1 & 2\end{pmatrix}\mathbf x \); eigenvalues \( k/m \) and \( 3k/m \). True.
Statement 2: the initial displacement \( (a, 0) = \tfrac{a}{2}(1, 1) + \tfrac{a}{2}(1, -1) \). Each mode evolves as \( \cos\omega_it \) (zero initial velocity), so \( x_1 = \tfrac{a}{2}(\cos\omega_1t + \cos\omega_2t) \) and \( x_2 = \tfrac{a}{2}(\cos\omega_1t - \cos\omega_2t) \). True.
Statement 3: the frequency ratio is \( \sqrt{3} \), irrational, so \( x_1(t) \) never exactly repeats. False. (It is quasi-periodic.)
Statement 4: initially only the left and middle springs are stretched or compressed by \( a \): \( E = \tfrac{1}{2}ka^2 + \tfrac{1}{2}ka^2 = ka^2 \). True.
Quick Revision Table
| Item | Result |
|---|---|
| Characteristic equation | \( \det(A - \lambda I) = 0 \); \( 2\times2 \): \( \lambda^2 - (\operatorname{tr})\lambda + \det = 0 \) |
| Trace and determinant | sum and product of eigenvalues |
| Hermitian / unitary / anti-Hermitian | real / \( \lvert\lambda\rvert = 1 \) / imaginary |
| Projector / nilpotent / involutory | 0, 1 / all 0 / \( \pm1 \) |
| \( aI + bJ \) (\( n\times n \)) | \( a + nb \) once, \( a \) (\( n - 1 \) times) |
| Cayley–Hamilton | \( p(A) = 0 \); \( A^{-1} = (tI - A)/d \) for \( 2\times2 \) |
| Diagonalisable | geometric = algebraic multiplicity for every \( \lambda \) |
| Exponential | \( \det e^A = e^{\operatorname{tr}A} \); \( A^2 = I \Rightarrow e^{i\theta A} = \cos\theta + i\sin\theta\,A \) |
| Pauli | \( \sigma_i\sigma_j = \delta_{ij} + i\varepsilon_{ijk}\sigma_k \); eigenvalues of \( \boldsymbol\sigma\cdot\mathbf a \): \( \pm\lvert\mathbf a\rvert \) |
| Rotation | eigenvalues \( 1, e^{\pm i\theta} \); \( \operatorname{tr}R = 1 + 2\cos\theta \) |
Key Takeaways
What’s Next
- Linear ODEs and Special Functions: systems of linear ODEs are solved by exactly these eigenvalue methods.
- Dimensional Analysis and Vector Calculus, the previous topic.
- Linear Algebra in Mathematical Sciences, for canonical forms and inner product spaces at proof level.
- Mathematical Methods: Quick Revision.