Home / Learn / Exam prep / GATE Physics / Lesson 1.1
All lessons

Module 1: Mathematical physics · Lesson 1.1 · about 50 min

Linear vector spaces and Dirac notation

Vector spaces beyond arrows: bases and dimension, inner products and Cauchy-Schwarz, Gram-Schmidt, kets and bras, outer products, projectors and completeness, change of basis, Hermitian and unitary operators, function spaces and tensor products, the way GATE and CSIR NET test them.

Key points, meanings and formulas on one page: choose "Save as PDF" in the print box.

Module 1 · about 50 minutes. By the end you will be able to decide whether a set is a vector space, count its dimension, use inner products and Gram-Schmidt in spaces of arrows, complex vectors and functions, and calculate fluently in Dirac notation: bras and kets, outer products, projectors, completeness and change of basis. This is the language of all of quantum mechanics. If 2 × 2 matrices and eigenvalues feel rusty, first skim the IIT JAM lesson on matrices and eigenvalues and vector algebra.

A vector is anything you can add and scale: arrows, columns of complex numbers, wavefunctions, even matrices. An inner product adds lengths and angles, and Dirac notation lets you do all of it with one set of symbols.

1. What makes a vector space

A vector space over the numbers 𝔽 (real ℝ or complex ℂ) is a set where you can add any two members and multiply any member by a number, and the result stays in the set, with the familiar rules: addition is commutative and associative, there is a zero vector, every vector has a negative, and scalars distribute (a(u→+v→)=au→+av→, (a+b)v→=av→+bv→).

That is all. Nothing says a vector must be an arrow. Spaces you will meet in GATE:

  • ℝn and ℂn: columns of numbers (spin states live in ℂ2).
  • Polynomials of degree at most n: adding two keeps the degree at most n.
  • Square-integrable functions on an interval or on the whole line: the home of wavefunctions.
  • m×n matrices; Hermitian matrices (a real vector space, not a complex one).

The subspace test. A subset of a known vector space is itself a vector space exactly when it contains the zero vector and is closed under addition and scaling. That is the quickest way to answer “which of these sets is a vector space?”:

  • Traceless matrices: yes, because the trace is linear.
  • The plane x+y+z=1: no, it misses the origin.
  • Polynomials of degree exactly 3: no, x3+(−x3+x)=x.
  • Matrices with determinant zero: no, diag(1,0)+diag(0,1)=I.

2. Bases, dimension and components

Vectors v→1,…,v→k are linearly independent if c1v→1+⋯+ckv→k=0 forces every ci=0. A basis is an independent set that spans the space; every basis has the same number of vectors, the dimension. Once a basis is chosen, each vector is a list of numbers, its components.

For numerical vectors, independence is a determinant (or rank) question. For functions, look for hidden relations first: cosh⁡x=12ex+12e−x makes {ex,e−x,cosh⁡x} dependent, and {sin2⁡x,cos2⁡x,1} is dependent too. The Wronskian determinant settles harder cases.

Counting dimensions is a favourite NAT question. Count free real (or complex) parameters:

Space Dimension
polynomials of degree at most n n+1
real symmetric n×n matrices n(n+1)/2
real antisymmetric n×n matrices n(n−1)/2
complex Hermitian n×n matrices (over ℝ) n2
traceless Hermitian n×n matrices (over ℝ) n2−1

The last row gives 3 for n=2 (the Pauli matrices) and 8 for n=3 (the Gell-Mann matrices).

3. Inner products: length and angle

An inner product assigns a number ⟨u|v⟩ to each pair of vectors with three properties:

⟨u|v⟩=⟨v|u⟩∗,⟨u|(av+bw)⟩=a⟨u|v⟩+b⟨u|w⟩,⟨v|v⟩>0 for v≠0

Physics puts the complex conjugate on the left slot: ⟨au|v⟩=a∗⟨u|v⟩. The standard examples are

⟨u|v⟩=∑kuk∗vk  (in ℂn),⟨f|g⟩=∫abf∗(x)g(x)w(x)dx  (functions, weight w>0)

The norm is ‖v‖=⟨v|v⟩. Two inequalities hold in every inner-product space:

|⟨u|v⟩|≤‖u‖‖v‖  (Cauchy-Schwarz),‖u+v‖≤‖u‖+‖v‖  (triangle)

Cauchy-Schwarz becomes an equality only when one vector is a multiple of the other. In quantum mechanics it is the engine behind the uncertainty principle.

⟨u|v⟩
the overlap of two vectors: how much of v points along u (a complex number in general)
‖v‖
the length of v; for a wavefunction, ‖ψ‖² = 1 means total probability 1
⟨u|v⟩ = 0
u and v are orthogonal: completely independent directions
|⟨u|v⟩|²
for normalised states, the probability that a system in state v is found in state u
Physical meaning
The inner product is what turns a list of amplitudes into measurable probabilities. Conjugating the left slot is what makes ⟨v|v⟩ a real, positive length for complex vectors.

Worked example. For |u⟩=(1,i)T and |v⟩=(i,1)T: ⟨u|v⟩=1⋅i+(−i)⋅1=0. Orthogonal. Without the conjugate you would get 2i, and ⟨u|u⟩ would come out as 1+i2=0 for a non-zero vector, which is nonsense. Conjugation is not optional.

Not every formula is an inner product. u→TGv→ on ℝ2 is one exactly when the symmetric matrix G is positive definite: diag(1,−1) fails (it gives ⟨v|v⟩=−1 for (0,1)), while (2111) (trace 3, determinant 1, both eigenvalues positive) works.

4. Gram-Schmidt: building an orthonormal basis

Any basis can be straightened into an orthonormal one. Take the vectors one at a time and remove from each its components along the ones already made:

u→1=v→1,u→2=v→2−⟨u1|v2⟩⟨u1|u1⟩u→1,u→3=v→3−⟨u1|v3⟩⟨u1|u1⟩u→1−⟨u2|v3⟩⟨u2|u2⟩u→2

and finally divide each u→k by its norm. Move the sliders: the part of v→2 along v→1 is the projection, and what is left, u→2, is at right angles.

Splitting v₂ into its projection on v₁ and the orthogonal remainder u₂, the one step Gram-Schmidt repeats. At 0° or 180° the remainder vanishes: the vectors are dependent.

By Pythagoras, ‖u2‖2=‖v2‖2−|⟨u1|v2⟩|2/‖u1‖2, a quick check on any calculation.

Worked example (functions). On [−1,1] with ⟨f|g⟩=∫−11fgdx, orthogonalise 1,x,x2. First u1=1. Next ⟨1|x⟩=0 (odd), so u2=x. Then ⟨x|x2⟩=0 and ⟨1|x2⟩/⟨1|1⟩=(2/3)/2=1/3, so u3=x2−13. Up to normalisation these are the Legendre polynomials P0,P1,P2: Gram-Schmidt on powers of x, with different weights and intervals, produces every classical orthogonal polynomial (Hermite with weight e−x2 on the whole line, Laguerre with e−x on [0,∞)).

A trap with subspaces. To project onto a plane spanned by two vectors that are not orthogonal, you cannot add the two separate projections. Orthogonalise first, or (in ℝ3) subtract the projection on the normal. For the plane spanned by (1,1,0) and (0,1,1), the normal is (1,−1,1), and (1,0,0) has squared projection 1−13=23 onto the plane, not 12+0.

5. Dirac notation

Dirac’s notation separates a vector from the numbers that describe it, so the same symbols work in every basis and every space.

  • A ket |ψ⟩ is a vector (a column).
  • A bra ⟨ϕ| is the linear map that sends |ψ⟩ to the number ⟨ϕ|ψ⟩ (a row: the conjugate transpose of |ϕ⟩).
  • The bra of c|ψ⟩ is c∗⟨ψ|: the ket-to-bra map is antilinear.

Put a ket in front of a bra and you get an operator, the outer product:

(|a⟩⟨b|)|ψ⟩=|a⟩⟨b|ψ⟩,tr(|a⟩⟨b|)=⟨b|a⟩,(|a⟩⟨b|)†=|b⟩⟨a|

For |a⟩=(1,0)T and |b⟩=(i,1)T, |a⟩⟨b|=(10)(−i, 1)=(−i100).

Orthonormal bases, completeness and projectors. If ⟨m|n⟩=δmn and the set is complete, every vector expands as |ψ⟩=∑ncn|n⟩ with cn=⟨n|ψ⟩. Since that holds for every |ψ⟩:

∑n|n⟩⟨n|=1ˆ

This resolution of the identity is the most useful line in the subject: insert it anywhere. For example ⟨ϕ|ψ⟩=∑n⟨ϕ|n⟩⟨n|ψ⟩=∑nϕn∗ψn, and an operator’s matrix is Amn=⟨m|Aˆ|n⟩, so that Aˆ=∑mnAmn|m⟩⟨n|.

Each Pˆn=|n⟩⟨n| is a projector: Pˆ2=Pˆ, Pˆ†=Pˆ, eigenvalues 0 and 1. Sums of projectors onto orthogonal states are projectors; differences are not ((Pˆ1−Pˆ2)2=Pˆ1+Pˆ2). The operator 1ˆ−2|ψ⟩⟨ψ| is a reflection: Hermitian, unitary, squares to 1ˆ, and has determinant −1.

Change of basis. The same operator has different matrices in different bases. With |±⟩=(|0⟩±|1⟩)/2, σz swaps |+⟩ and |−⟩, so in the {|+⟩,|−⟩} basis its matrix is (0110). The eigenvalues (±1), trace and determinant do not change; only the eigenbasis makes the matrix diagonal. If Umn=⟨m|n′⟩ links the bases, A′=U†AU.

Worked example (expectation value). Aˆ=|1⟩⟨1|+2|2⟩⟨2|+3|3⟩⟨3| and |ψ⟩=13(|1⟩+2i|2⟩+2|3⟩).

  1. Norm: (1+4+4)/9=1, already normalised.
  2. Aˆ|ψ⟩=13(|1⟩+4i|2⟩+6|3⟩).
  3. ⟨ψ|Aˆ|ψ⟩=19(1⋅1+(−2i)(4i)+2⋅6)=19(1+8+12)=219≈2.33: the eigenvalues weighted by |cn|2.

6. Operators: adjoint, Hermitian, unitary

The adjoint Aˆ† is defined by ⟨ϕ|Aˆψ⟩=⟨Aˆ†ϕ|ψ⟩; in an orthonormal basis its matrix is the conjugate transpose. The rules (AˆBˆ)†=Bˆ†Aˆ† and (cAˆ)†=c∗Aˆ† settle most “is it Hermitian?” questions:

  • AˆBˆ+BˆAˆ, AˆBˆAˆ and i[Aˆ,Bˆ] are Hermitian when Aˆ,Bˆ are.
  • [Aˆ,Bˆ] itself is anti-Hermitian, which is why [xˆ,pˆ]=iℏ carries an i.

Hermitian operators (Hˆ†=Hˆ) have real eigenvalues and orthogonal eigenvectors for different eigenvalues, and in finite dimensions their eigenvectors form an orthonormal basis (the spectral theorem): Hˆ=∑nλn|n⟩⟨n|. Unitary operators (Uˆ†Uˆ=1ˆ) preserve every inner product, map orthonormal bases to orthonormal bases, and have eigenvalues of modulus 1. Having |det⁡Uˆ|=1 is not enough to be unitary: diag(2,12) is a counter-example. The next lesson, Hermitian and unitary matrices, diagonalisation, works these out in detail.

Commuting operators share eigenvectors. If [Aˆ,Bˆ]=0 and Aˆ|a⟩=a|a⟩, then Bˆ|a⟩ is also an eigenvector of Aˆ with eigenvalue a. Without degeneracy that forces Bˆ|a⟩∝|a⟩. With degeneracy, Bˆ can be anything inside each eigenspace: the matrices commuting with diag(1,1,2) form a space of dimension 22+12=5.

“A bra is just the transpose of the ket.” It is the conjugate transpose, and the map from kets to bras is antilinear: the bra of c|ψ⟩ is c∗⟨ψ|. Forget the conjugate and inner products stop being positive ((1,i) would have “length” zero), probabilities can come out complex, and Aˆ†Aˆ would not be Hermitian. When a calculation in ℂn gives a complex norm, look for a missing conjugate first.

7. Infinite dimensions and tensor products

Function spaces. On [−π,π] the functions en=einx/2π are orthonormal and complete: expanding in them is the Fourier series (see the JAM lesson on Fourier series). Components are inner products, cn=⟨en|f⟩, and Parseval’s theorem ∑|cn|2=‖f‖2 is Pythagoras with infinitely many axes. For f=x, c1=12π∫−ππxe−ixdx=−i2π, so |c1|2=2π.

The space L2(ℝ) of square-integrable functions is a Hilbert space: an inner-product space in which every Cauchy sequence converges. Two warnings for GATE:

  • The plane wave eikx and the delta function δ(x−x0) are not in L2: their norms are infinite. They are generalised eigenvectors of pˆ and xˆ, normalised to delta functions, ⟨x|x′⟩=δ(x−x′), with completeness ∫|x⟩⟨x|dx=1ˆ.
  • Unbounded operators like xˆ have no normalisable eigenvectors at all: their spectrum is continuous.

Tensor products. Two systems with spaces of dimensions m and n combine into a space of dimension mn (not m+n), with basis |i⟩⊗|j⟩=|ij⟩. A state ∑cij|ij⟩ is a product state |a⟩⊗|b⟩ exactly when the matrix cij has rank 1; for two qubits, when det⁡c=0. So 12(|00⟩+|01⟩+|10⟩−|11⟩) (determinant −12) is entangled. You can build and inspect such states in fibub’s quantum circuit simulator, and check any matrix calculation with the matrix calculator.

  • A vector space needs only addition and scaling; subspace test: contains 0, closed under both. Arrows, ℂn, polynomials, functions and matrices are all vector spaces.
  • Dimension = number of basis vectors. Symmetric n(n+1)/2, antisymmetric n(n−1)/2, Hermitian n2 (real), traceless Hermitian n2−1.
  • Inner product: conjugate on the left, ⟨u|v⟩=∑uk∗vk or ∫f∗gwdx. Cauchy-Schwarz is an equality only for parallel vectors.
  • Gram-Schmidt subtracts projections one at a time; on 1,x,x2 it gives the Legendre polynomials. Never add projections onto non-orthogonal vectors.
  • Bra of c|ψ⟩ is c∗⟨ψ|; tr|a⟩⟨b|=⟨b|a⟩; ∑n|n⟩⟨n|=1ˆ; projectors have eigenvalues 0 and 1.
  • Matrices change with the basis (A′=U†AU); eigenvalues, trace and determinant do not. Commuting operators share eigenvectors.
  • Plane waves and delta functions lie outside L2; tensor-product dimensions multiply.

Practise: exam questions on this lesson

They climb from warm-up to top-rank level. Check each answer, then read the full solution. More on the linear vector spaces practice page, and every topic on the GATE Physics practice page.

Next: lesson 1.2, Hermitian and unitary matrices and diagonalisation. Back to the GATE Physics course home.

Check yourself

1In Dirac notation, what is the bra that goes with the ket 3i|ψ⟩?
2{|n⟩} is an orthonormal basis. What is Σₙ |n⟩⟨n|?
3What is the dimension of the real vector space of 2 × 2 Hermitian matrices?

Sources

  • P. A. M. Dirac, The Principles of Quantum Mechanics, 4th ed. (1958), chapters 1 and 2
  • R. Shankar, Principles of Quantum Mechanics, 2nd ed. (1994), chapter 1
  • G. B. Arfken, H. J. Weber and F. E. Harris, Mathematical Methods for Physicists, 7th ed. (2013), chapters 5 and 6
  • S. Axler, Linear Algebra Done Right, 3rd ed. (2015), chapters 2, 6 and 7
  • GATE Physics (PH) syllabus, Mathematical Physics section, published by the organising institute each year

Your ticks stay in this browser. Save them as a file on the course home.

Updated 2026-11-22. Not affiliated with the IITs, IISc or the GATE organising institute. The general aptitude part of GATE is not covered here. Check the official GATE website for the current pattern. All questions are original.