Dirac Bra-Ket Notation

Dirac bra-ket notation is the sentence structure of quantum mechanics — the compact algebraic language, set down by P. A. M. Dirac in the 1930s and named in his 1939 paper, in which every state, measurement, and operator is a manipulation of vectors on a complex Hilbert space $\mathcal{H}$ . A physical state is a ket $|\psi\rangle\in\mathcal{H}$ ; paired with it is a bra $\langle\psi|$ , an element of the dual space $\mathcal{H}^{*}$ that eats a ket and returns a complex number. The pun is deliberate and load-bearing: a bra and a ket clamped together make a bra-c-ket $\langle\phi|\psi\rangle$ , the inner product — the amplitude from which all probabilities flow. The genius of the notation is that it separates the kind of an object from its components. A ket is an abstract arrow in $\mathcal{H}$ ; only after you pick an orthonormal basis $\{|i\rangle\}$ does it grow coordinates $\psi_i=\langle i|\psi\rangle$ — mirroring the physics, in which a state exists before any measuring device is chosen and concrete numbers appear only on projection. The resolution of identity $\sum_i|i\rangle\langle i|=I$ is the hinge between the abstract and the concrete — slip it into any expression and an abstract statement unfolds into a basis sum, reproducing every formula of matrix mechanics on demand. Three composite objects recur. The inner product $\langle\phi|\psi\rangle$ gives amplitudes and, squared, Born probabilities; the outer product $|\psi\rangle\langle\phi|$ is an operator, and outer products of basis kets are the matrix units from which every operator is assembled; expectation values collapse to the tidy sandwich $\langle\psi|A|\psi\rangle$ . Throughout quantum information one works in finite dimension, $\mathcal{H}=\mathbb{C}^d$ , where every bra is realized by an inner product and the subtleties of unbounded operators evaporate — leaving pure linear algebra in a uniquely convenient syntax, the one underlying the qubit, superposition, measurement, and density-operator formalisms to come.

Key takeaways

A quantum state is a ket $|\psi\rangle$ , a vector in a complex Hilbert space $\mathcal{H}$ ; its conjugate-dual is a bra $\langle\psi|$ , a linear functional that maps kets to complex numbers.
Clamped together, $\langle\phi|\psi\rangle$ is the inner product — a complex amplitude obeying conjugate symmetry $\langle\phi|\psi\rangle=\langle\psi|\phi\rangle^{*}$ , linearity in the ket, and positivity $\langle\psi|\psi\rangle\ge 0$ .
Reversed, $|\psi\rangle\langle\phi|$ is an operator; the special case $|\psi\rangle\langle\psi|$ is a rank-one projector with $P^2=P=P^\dagger$ .
The resolution of identity $\sum_i|i\rangle\langle i|=I$ for any orthonormal basis is the workhorse trick: insert it anywhere to turn abstract objects into component sums.
Fix a basis and kets become column vectors, bras their conjugate-transpose rows, operators matrices — Dirac notation is the basis-free algebra behind that bookkeeping, and it encodes the Born rule $p(i)=|\langle i|\psi\rangle|^2$ in one line.
Key idea: a ket $|\psi\rangle$ is a vector and a bra $\langle\phi|$ is the functional that sends it to the number $\langle\phi|\psi\rangle$ ; inner products give amplitudes, outer products give operators, and $\sum_i|i\rangle\langle i|=I$ lets you slide freely between abstract vectors and their components.

The core insight

A ket $|\psi\rangle$ is a vector and a bra $\langle\phi|$ is the linear functional that maps a ket to the number $\langle\phi|\psi\rangle$ ; inner products give amplitudes, outer products give operators, and the completeness relation $\sum_i|i\rangle\langle i|=I$ lets you switch freely between abstract vectors and their components.

Where this lives today

Dirac notation is not a historical relic; it is the working interface of the quantum-information industry. Every quantum-algorithm paper states its inputs, gates, and measurements in kets and bras, and the software that runs real hardware mirrors it directly: a Qiskit or Cirq circuit prepares a register in $|0\cdots 0\rangle$ , applies unitaries, and measures in the computational basis — precisely the bra-ket pipeline $\langle x|U_n\cdots U_1|0\cdots 0\rangle$ . Entanglement is diagnosed by whether a ket factors — the Bell state $|\Phi^{+}\rangle=\tfrac{1}{\sqrt2}(|00\rangle+|11\rangle)$ is famous exactly because it does not — and teleportation and superdense coding are one-line manipulations of such kets. Error correction lives here too: a stabilizer code is fixed by the Pauli operators (outer-product sums) that leave the codespace projector $\Pi=\prod\tfrac12(I+g_k)$ invariant, and syndrome extraction is the expectation of those projectors. Open-system modelling leans on the same outer-product structure through the density operator $\rho=\sum_k p_k|\psi_k\rangle\langle\psi_k|$ . From a whiteboard proof to a cryostat running superconducting qubits, the notation Dirac fixed in 1939 remains the medium in which the field thinks.

The physical picture

Picture a ket $|\psi\rangle$ as a column of complex numbers and its bra $\langle\psi|$ as the same column laid on its side and complex-conjugated — a row. The whole grammar of the notation is the geometry of sliding rows against columns: push a row against a column and everything contracts to a single number, the inner product; push a column against a row the other way and it fans out into a full square array, the outer product operator. The brackets make each object's orientation visible, so you always know whether you hold a vector, a number, or a matrix.

Figure

A ket is a column of amplitudes and its bra is the conjugate-transposed row — note that i becomes minus i under conjugation. Row against column contracts to one number, the inner product, here equal to 1; column against row fans out into a 2 by 2 matrix, the projector onto 0 with diagonal entries 1 and 0

The number a bracket returns is a physical amplitude: read $\langle\phi|\psi\rangle$ as "how much of $|\phi\rangle$ lives inside $|\psi\rangle$ ", zero if orthogonal, one if they coincide and are normalized. Quantum mechanics declares the squared magnitude of that overlap to be a probability, so the inner product is literally the amplitude that a system prepared as $|\psi\rangle$ is found in $|\phi\rangle$ . The completeness relation is the spoken instruction "drop in a complete set of intermediate states" — inserting it splits any global question into a sum over basis outcomes, as the second figure shows in numbers.

Figure

Inserting a complete set of basis states between a bra and a ket splits an amplitude into a sum over outcomes, and the squared amplitudes are Born probabilities — for an equal superposition of 0 and 1 carrying a relative phase i, both computational outcomes have probability 0.5 while the observable ⟨Y⟩ equals plus 1

The core equations

Everything below is the algebra of two moves — clamping a bra onto a ket to make a number, and clamping a ket onto a bra to make an operator — plus the one relation, completeness, that lets you insert a basis wherever you please. We start with how a ket acquires components.

|\psi\rangle=\sum_i \psi_i\,|i\rangle,\qquad \psi_i=\langle i|\psi\rangle \qquad (1)

where: $|\psi\rangle$ — a ket, a vector in $\mathcal{H}$ ; $|i\rangle$ — an orthonormal basis ket labelled by $i$ ; $\psi_i$ — the complex component (amplitude) of $|\psi\rangle$ along $|i\rangle$ . Reading: to read off a coordinate, project with the matching bra — the amplitude is the overlap $\langle i|\psi\rangle$ . Clamping two kets' components together through their bra gives the inner product; the order matters, because swapping the two arguments conjugates the result.

\langle\phi|\psi\rangle=\sum_i \phi_i^{*}\,\psi_i,\qquad \langle\phi|\psi\rangle=\langle\psi|\phi\rangle^{*} \qquad (2)

where: $\langle\phi|$ — the bra, the conjugate-dual functional of $|\phi\rangle$ ; $\phi_i^{*}$ — the complex conjugate of the $i$ -th component of $|\phi\rangle$ ; $(\cdot)^{*}$ — complex conjugation. Reading: the bra carries the conjugates and is linear in the ket, anti-linear in the bra — which makes $\langle\psi|\psi\rangle=\sum_i|\psi_i|^2$ a real, non-negative squared length. The single most-used identity in the formalism says that summing the rank-one projectors of any orthonormal basis rebuilds the identity operator.

\sum_i |i\rangle\langle i|=I \qquad (3)

where: $|i\rangle\langle i|$ — the rank-one projector onto basis vector $|i\rangle$ ; $I$ — the identity operator on $\mathcal{H}$ . Reading: since $I$ changes nothing, you may insert $\sum_i|i\rangle\langle i|$ anywhere for free — it changes only the form, expanding an abstract object into a basis sum. An outer product of a normalized ket with its own bra is a projector — geometrically, dropping a shadow onto the line through $|\psi\rangle$ .

P_\psi=|\psi\rangle\langle\psi|,\qquad P_\psi^{2}=P_\psi,\qquad P_\psi^\dagger=P_\psi \qquad (4)

where: $P_\psi$ — the projector onto the line spanned by normalized $|\psi\rangle$ ; $\dagger$ — Hermitian conjugation (conjugate transpose). Reading: projecting twice equals projecting once ( $P^2=P$ ) and a projector is its own dagger — the two properties that define an orthogonal projector. Observables sandwiched between a bra and a ket give expectation values, and any operator is a weighted sum of matrix units $|i\rangle\langle j|$ .

\langle A\rangle_\psi=\langle\psi|A|\psi\rangle,\qquad A=\sum_{i,j}A_{ij}\,|i\rangle\langle j|,\quad A_{ij}=\langle i|A|j\rangle \qquad (5)

where: $A$ — an operator (observable, if Hermitian); $\langle A\rangle_\psi$ — the mean value of $A$ in state $|\psi\rangle$ ; $A_{ij}$ — the matrix element of $A$ in the basis $\{|i\rangle\}$ . Reading: the outer products $|i\rangle\langle j|$ are the matrix units, so this equation is the dictionary between operators and matrices. Finally, the dagger reverses every bra-ket, conjugates every scalar, and reverses product order — the rule that keeps Hermitian conjugation consistent.

\big(|\psi\rangle\langle\phi|\big)^\dagger=|\phi\rangle\langle\psi|,\qquad (A|\psi\rangle)^\dagger=\langle\psi|A^\dagger \qquad (6)

where: $A^\dagger$ — the Hermitian conjugate of $A$ . Reading: a ket daggers into its bra and vice versa, so $(c\,|\psi\rangle\langle\phi|)^\dagger=c^{*}\,|\phi\rangle\langle\psi|$ — reverse the order and conjugate the scalar.

Derivation AFrom completeness to the matrix product rule

The goal is to prove that composing operators in Dirac notation reproduces ordinary matrix multiplication, which is what licenses the entire column–row–matrix picture. We take two operators $A$ and $B$ , ask for the $(i,k)$ entry of their product, and let the single trick of the trade — inserting a resolution of identity in the gap between them — do all the work. The payoff is that the abstract product $AB$ and the concrete row-times-column rule are literally the same statement, so one may compute in whichever picture is convenient. The boxed result is $(AB)_{ik}=\sum_j A_{ij}B_{jk}$ .

Result.

(AB)_{ik}=\sum_j A_{ij}\,B_{jk}

Step 1. Write the target entry as a bra-ket sandwich, $(AB)_{ik}=\langle i|AB|k\rangle$ , using equation (5) applied to the operator $AB$ . Why: the definition of a matrix element is exactly "clamp the operator between the two basis states", so nothing has been assumed yet — this is pure notation.

Step 3. Distribute the sum and read off two matrix elements per term: $\sum_j\langle i|A|j\rangle\langle j|B|k\rangle=\sum_j A_{ij}B_{jk}$ . Why: each sandwich $\langle i|A|j\rangle$ and $\langle j|B|k\rangle$ is by definition a scalar entry, and their product summed over $j$ is precisely the row- $i$ -of- $A$ dotted into column- $k$ -of- $B$ — the schoolbook multiplication rule.

Go deeper · Derivation B · 1 minThe Born rule from projectors and completeness

Derivation BThe Born rule from projectors and completeness

The goal is to show that the measurement probabilities of an orthonormal basis are the squared amplitudes $p(i)=|\langle i|\psi\rangle|^2$ and that they automatically sum to one for any normalized state — the two facts that make Dirac notation a probability calculus, not merely a linear-algebra convenience. We start from the measurement postulate in projector form, use conjugate symmetry to fold the sandwich into a squared magnitude, then use completeness to collapse the total probability back to the norm — short precisely because the notation was built for it. The boxed results are $p(i)=|\langle i|\psi\rangle|^2$ and $\sum_i p(i)=1$ .

Result.

p(i)=|\langle i|\psi\rangle|^2,\qquad \sum_i p(i)=1

Step 1. Write the probability of outcome $i$ as the expectation of its projector, $p(i)=\langle\psi|P_i|\psi\rangle=\langle\psi|i\rangle\langle i|\psi\rangle$ , with $P_i=|i\rangle\langle i|$ . Why: the projective-measurement postulate assigns each outcome the expectation value of the projector onto its subspace — writing that expectation as a bra-ket sandwich is what exposes the amplitude inside.

Step 2. Recognize the amplitude and its conjugate: $\langle\psi|i\rangle\langle i|\psi\rangle=\langle i|\psi\rangle^{*}\langle i|\psi\rangle=|\langle i|\psi\rangle|^2$ . Why: conjugate symmetry (equation 2) says $\langle\psi|i\rangle=\langle i|\psi\rangle^{*}$ , so the sandwich is a number times its own conjugate — a manifestly real, non-negative squared magnitude, exactly what a probability must be.

Step 3. Sum over outcomes and collapse with completeness: $\sum_i p(i)=\langle\psi|\big(\sum_i|i\rangle\langle i|\big)|\psi\rangle=\langle\psi|\psi\rangle=1$ . Why: the bracketed sum is the identity by equation (3), so the whole expression falls back to the norm $\langle\psi|\psi\rangle$ , which is 1 for a physical state — total probability is conserved as a theorem, not an extra assumption.

Go deeper · Derivation C · 1 minExpectation value as a probability-weighted eigenvalue average

Derivation CExpectation value as a probability-weighted eigenvalue average

The goal is to connect the compact quantum formula $\langle A\rangle_\psi=\langle\psi|A|\psi\rangle$ to the ordinary statistical mean of a random variable, showing that a quantum expectation is nothing more exotic than each eigenvalue weighted by its Born probability. We take a Hermitian observable in its spectral form and insert it into the sandwich. This is the bridge between the operator formalism and laboratory statistics — why $\langle A\rangle$ predicts the average of many repeated measurements. The boxed result is $\langle A\rangle_\psi=\sum_a a\,|\langle a|\psi\rangle|^2$ .

Result.

\langle A\rangle_\psi=\sum_a a\,|\langle a|\psi\rangle|^2=\sum_a a\,p(a)

Step 1. Insert the spectral decomposition $A=\sum_a a\,|a\rangle\langle a|$ into the sandwich: $\langle\psi|A|\psi\rangle=\langle\psi|\big(\sum_a a\,|a\rangle\langle a|\big)|\psi\rangle$ . Why: a Hermitian operator has an orthonormal eigenbasis $\{|a\rangle\}$ with real eigenvalues $a$ , so writing $A$ as a weighted sum of eigen-projectors is exact, not an approximation.

Step 3. Read the result as a classical mean: $\langle A\rangle_\psi=\sum_a a\,p(a)$ . Why: this is the textbook average of the random variable "measured value of $A$ " over its probability distribution — the quantum sandwich and the statistical mean are the same number, which is what makes $\langle A\rangle$ physically testable.

Limiting cases

The one notation specializes cleanly across every structure of quantum information; each row below is a sentence in Dirac's language that you will read hundreds of times.

Case	Result	Meaning
Orthonormal basis	$\langle i\vert j\rangle=\delta_{ij}$	Basis kets are unit vectors, mutually orthogonal; the Kronecker delta packs normalization and orthogonality into one symbol.
Normalized state	$\langle\psi\vert\psi\rangle=1$	Physical pure states have unit norm, so total measurement probability is one; a non-unit ket is normalized by dividing by $\sqrt{\langle\psi\vert\psi\rangle}$ .
Orthogonal states	$\langle\phi\vert\psi\rangle=0$	Perfectly distinguishable — a single measurement can tell them apart with certainty, because zero overlap means zero amplitude.
Global phase	$\vert\psi'\rangle=e^{i\theta}\vert\psi\rangle\ \Rightarrow\ \vert\langle\phi\vert\psi'\rangle\vert=\vert\langle\phi\vert\psi\rangle\vert$	An overall phase leaves every probability untouched, so $e^{i\theta}\vert\psi\rangle$ and $\vert\psi\rangle$ are the same physical state — a ray, not a vector.
Continuous basis	$\langle x\vert x'\rangle=\delta(x-x')$	In infinite dimension the Kronecker delta becomes a Dirac delta and the sum an integral; the finite- $d$ case of quantum information avoids these subtleties.

Common trapA bra is just a ket "written sideways"

Not quite — the bra also complex-conjugates. For $|\psi\rangle=\alpha|0\rangle+\beta|1\rangle$ the bra is $\langle\psi|=\alpha^{*}\langle 0|+\beta^{*}\langle 1|$ , not $\alpha\langle 0|+\beta\langle 1|$ ; drop the conjugate and $\langle\psi|\psi\rangle$ stops being a real length. The second, equally common, blunder is confusing the two products: $\langle\psi|\phi\rangle$ is a number (bra then ket) while $|\phi\rangle\langle\psi|$ is an operator (ket then bra) — same letters, different objects, and only the order tells them apart. When in doubt, count orientations: row meets column makes a scalar, column meets row makes a matrix.

Worked examples

1 · Reading a qubit completely (fully worked, confidence builder)

Take the single-qubit state $|\psi\rangle=\tfrac{1}{\sqrt2}\big(|0\rangle+i|1\rangle\big)$ . Write its bra, confirm normalization, then find the measurement probabilities in the computational (Z) basis and the $\pm$ (X) basis, and the three Pauli expectation values.

Step 1. The bra conjugates every amplitude: since the $|1\rangle$ coefficient is $i$ , its conjugate $-i$ appears, giving $\langle\psi|=\tfrac{1}{\sqrt2}\big(\langle 0|-i\langle 1|\big)$ . This single sign is the whole content of the pitfall above.

Step 2. Normalization is $\langle\psi|\psi\rangle=\tfrac12\big(\langle 0|-i\langle 1|\big)\big(|0\rangle+i|1\rangle\big)=\tfrac12\big(1+i\langle 0|1\rangle-i\langle 1|0\rangle+1\big)=\tfrac12(1+1)=\mathbf{1}$ , using $\langle 0|1\rangle=0$ — properly normalized.

Step 3. Z-basis probabilities are squared amplitudes: $p(0)=|\langle 0|\psi\rangle|^2=|1/\sqrt2|^2=\mathbf{0.5}$ and $p(1)=|\langle 1|\psi\rangle|^2=|i/\sqrt2|^2=\mathbf{0.5}$ — the phase $i$ is invisible to a Z measurement.

Step 4. For the X basis use $|\pm\rangle=\tfrac{1}{\sqrt2}(|0\rangle\pm|1\rangle)$ . Then $\langle +|\psi\rangle=\tfrac12(1+i)$ and $\langle -|\psi\rangle=\tfrac12(1-i)$ , so $p(+)=\tfrac14(1+1)=\mathbf{0.5}$ and likewise $p(-)=\mathbf{0.5}$ — unbiased in X as well as Z.

Step 5. The expectation values follow: $\langle Z\rangle=p(0)-p(1)=0$ and $\langle X\rangle=p(+)-p(-)=0$ . For Y, note $Y|\psi\rangle=|\psi\rangle$ (direct check: $Y$ sends $\tfrac{1}{\sqrt2}[1,\,i]^{\mathsf T}$ to itself), so $\langle Y\rangle=\langle\psi|\psi\rangle=\mathbf{+1}$ . The Bloch vector is $(\langle X\rangle,\langle Y\rangle,\langle Z\rangle)=(0,+1,0)$ — exactly the $+1$ eigenket of $Y$ , pointing along $+y$ .

2 · Operators from outer products, and an amplitude through a gate

Show that the Pauli- $X$ operator is the outer-product sum $|0\rangle\langle 1|+|1\rangle\langle 0|$ and confirm its action. Then for $|\psi\rangle=\tfrac{\sqrt3}{2}|0\rangle+\tfrac12|1\rangle$ compute $\langle Z\rangle$ , and finally the probability of measuring $0$ after a Hadamard gate.

Step 1. Apply $X=|0\rangle\langle 1|+|1\rangle\langle 0|$ to $|0\rangle$ : the first term gives $|0\rangle\langle 1|0\rangle=0$ and the second gives $|1\rangle\langle 0|0\rangle=|1\rangle$ , so $X|0\rangle=\mathbf{|1\rangle}$ and, symmetrically, $X|1\rangle=|0\rangle$ . In matrix units this is $\left(\begin{smallmatrix}0&1\\1&0\end{smallmatrix}\right)$ — Pauli- $X$ , the bit flip, read straight off the outer products.

Step 2. For $|\psi\rangle=\tfrac{\sqrt3}{2}|0\rangle+\tfrac12|1\rangle$ the Z-probabilities are $p(0)=|\tfrac{\sqrt3}{2}|^2=\tfrac34$ and $p(1)=|\tfrac12|^2=\tfrac14$ , so $\langle Z\rangle=\tfrac34-\tfrac14=\mathbf{0.5}$ . This matches a Bloch polar angle $\theta=60^\circ$ , since $\cos\theta=0.5$ while the $|0\rangle$ amplitude is $\cos(\theta/2)=\tfrac{\sqrt3}{2}$ .

Step 3. The Hadamard is $H=\tfrac{1}{\sqrt2}\big(|0\rangle\langle 0|+|0\rangle\langle 1|+|1\rangle\langle 0|-|1\rangle\langle 1|\big)$ . Keeping only the two terms of $H$ that begin with $|0\rangle$ , the amplitude to find $0$ afterward is $\langle 0|H|\psi\rangle=\tfrac{1}{\sqrt2}\big(\langle 0|\psi\rangle+\langle 1|\psi\rangle\big)=\tfrac{1}{\sqrt2}\big(\tfrac{\sqrt3}{2}+\tfrac12\big)$ .

Step 4. Square it: $p(0)=\tfrac12\big(\tfrac{\sqrt3+1}{2}\big)^2=\tfrac{4+2\sqrt3}{8}=\tfrac{2+\sqrt3}{4}\approx\mathbf{0.933}$ , with $p(1)=\tfrac{2-\sqrt3}{4}\approx0.067$ summing to $1$ . The Hadamard rotates the state close to $|+\rangle$ , so a subsequent Z measurement returns $0$ about 93% of the time — a vivid look at how gates reshape the amplitudes that bra-ket sandwiches read out.

Physics of the portrait

Bra-ket notation is simultaneously a piece of mathematics and a way of thinking — it grew out of one man's insistence that the symbols of quantum mechanics should carry the physics, so that manipulating them correctly guarantees physically correct answers.

Origins · 1 minPaul Dirac and how this was discovered

Paul Dirac

1902–1984 · architect of relativistic quantum mechanics and the symbolic method

Paul Adrien Maurice Dirac, trained as an electrical engineer in Bristol before turning to physics at Cambridge, was the great formalist of the quantum revolution — the man who, more than anyone, gave quantum mechanics its permanent notation. By the 1930 first edition of The Principles of Quantum Mechanics he had built an abstract "symbolic method" in which states and observables were manipulated directly, without reference to any particular representation. The bra and ket vectors themselves were christened in a short, decisive 1939 paper, A new notation for quantum mechanics, where he split the symmetric bracket $\langle\phi\mid\psi\rangle$ into a "bra" $\langle\phi|$ and a "ket" $|\psi\rangle$ — a pun on "bracket" that became one of the most durable jokes in physics. His conviction was that good notation is not cosmetic: by making the type of each object visible, it makes whole classes of error impossible and turns derivations into near-mechanical symbol manipulation. Famously reticent, he shared the 1933 Nobel Prize with Schrödinger, and his notation, refined through the book's four editions to 1958, is now the first thing every physics student learns to write.

History

1925

— Dirac recasts quantum mechanics with non-commuting q-numbers

1930

— the first edition of The Principles of Quantum Mechanics introduces the symbolic method

1939

— the paper A new notation for quantum mechanics defines the bra and the ket

1958

— the fourth edition makes bra-ket notation the field's standard language

Going further

Texts. Nielsen & Chuang, Quantum Computation and Quantum Information (10th anniversary ed.), §2.1–2.2 · Preskill, Quantum Computation lecture notes (Ph219), Chapter 2 · Watrous, The Theory of Quantum Information, Chapter 1 · Wilde, Quantum Information Theory (2nd ed.), Chapter 3 · Kitaev, Shen & Vyalyi, Classical and Quantum Computation, §8 · Mermin, Quantum Computer Science, Appendix A · Dirac, A new notation for quantum mechanics, Math. Proc. Camb. Phil. Soc. 35, 416 (1939). All equations are standard results re-derived here from scratch. The key identities are: the component formula $\psi_i=\langle i|\psi\rangle$ ; the matrix-product identity $(AB)_{ik}=\sum_j A_{ij}B_{jk}$ ; the Born rule $p(i)=|\langle i|\psi\rangle|^2$ with $\sum_i p(i)=1$ ; the expectation identity $\langle A\rangle=\sum_a a\,|\langle a|\psi\rangle|^2$ ; the projector properties $P_\psi^2=P_\psi=P_\psi^\dagger$ ; and the worked-example results $\langle Y\rangle=+1$ for the $Y$ eigenket and $p(0)=(2+\sqrt3)/4\approx0.933$ after the Hadamard gate. The 1939 Dirac paper is the historical origin of the bra and ket terms.

Prerequisites. Complex Vector Spaces and Linear Algebra · Inner Product, Norm, and Normalization. Leads to. The Qubit and Its State Space · The Born Rule · Density Operators · Observables and Hermitian Operators.

Ask the tutor: Want to see any state's bra, probabilities, or Bloch vector worked out step by step, or watch a resolution of identity crack open a circuit amplitude? Ask and we will compute it together.

What is the actual difference between a bra and a ket?

A ket $|\psi\rangle$ is a vector — an element of $\mathcal{H}$ , concretely a column of complex amplitudes. A bra $\langle\phi|$ is not a vector but a function: it takes any ket and returns a complex number, the inner product $\langle\phi|\psi\rangle$ . In finite dimension every such function can itself be written with a vector (its components are the conjugates of $|\phi\rangle$ 's), which is why a bra "looks like" a conjugate-transposed row. But the roles differ: kets are the states, bras are the questions you ask of them — keeping column versus row, vector versus functional, straight is what makes the whole algebra typecheck.

Why is the inner product linear in the ket but conjugate-linear in the bra?

This is the physics convention (the mathematicians' is often the mirror image), and it exists so that norms come out real and positive. If we demanded linearity in both slots, then $\langle\psi|\psi\rangle$ for $|\psi\rangle=i|0\rangle$ would give $i^2\langle 0|0\rangle=-1$ , a negative "length" — nonsense. Conjugating the bra side gives $\langle\psi|\psi\rangle=|i|^2\langle 0|0\rangle=+1$ . The asymmetry is exactly what guarantees $\langle\psi|\psi\rangle=\sum_i|\psi_i|^2\ge 0$ , so that $\sqrt{\langle\psi|\psi\rangle}$ is a genuine norm and Born probabilities are non-negative.

Why can I insert the identity anywhere — isn't multiplying by I trivial?

It is trivial in value and profound in use. Multiplying by $I$ never changes what an expression equals, but writing $I$ as $\sum_i|i\rangle\langle i|$ changes what it looks like: it introduces a sum over a basis you get to choose. Drop it between two operators and you get the matrix-product rule; drop it inside a norm and you get $\sum_i|\psi_i|^2=1$ ; drop it between the stages of a circuit and you get a sum over all intermediate computational paths — the discrete analogue of a Feynman path sum. It works because completeness is an exact identity: you pay nothing to insert it and gain a decomposition every time.

Does every ket have a corresponding bra?

In the finite-dimensional spaces of quantum information, yes, and uniquely — this is the Riesz representation theorem: every linear functional on $\mathcal{H}$ is an inner product with some fixed vector, so bras and kets are in one-to-one correspondence and you pass between them by daggering. In infinite dimension the correspondence frays: a position eigenket $|x\rangle$ is not normalizable and lives in a "rigged Hilbert space", with $\langle x|x'\rangle=\delta(x-x')$ a distribution rather than a number. Quantum computing sidesteps all of this by working in $\mathbb{C}^d$ , where the notation is pure, well-defined linear algebra.

If a global phase changes nothing, why keep track of it at all?

Because relative phases are everything, and a global phase is only globally irrelevant. The state $e^{i\theta}|\psi\rangle$ gives the same probabilities as $|\psi\rangle$ for every measurement, so the two are physically identical — states are really rays, and the qubit space is the Bloch sphere, not $\mathbb{C}^2$ . But the moment $|\psi\rangle$ is one branch of a superposition, its phase is measured relative to the other branches and becomes observable, as interference. So discard an overall phase freely, but never discard a phase sitting between two terms of a superposition — that is where quantum algorithms hide their answers.

How do bras and kets work for multiple qubits?

Combine subsystems with the tensor product: a two-qubit ket is $|\psi\rangle\otimes|\phi\rangle$ , abbreviated $|\psi\rangle|\phi\rangle$ or $|\psi\phi\rangle$ , with computational basis $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$ . Bras tensor the same way, and inner products factor: $(\langle\psi|\langle\phi|)(|\psi'\rangle|\phi'\rangle)=\langle\psi|\psi'\rangle\,\langle\phi|\phi'\rangle$ . A state is a product state exactly when it factors like this; when it cannot — as with $\tfrac{1}{\sqrt2}(|00\rangle+|11\rangle)$ — it is entangled, and that failure to factor is the algebraic signature the Schmidt decomposition and partial trace go on to quantify.

Dirac Bra-Ket Notation

Where this lives today

The physical picture

The core equations

Limiting cases

Worked examples

1 · Reading a qubit completely (fully worked, confidence builder)

2 · Operators from outer products, and an amplitude through a gate

Physics of the portrait

Going further

Before this

Where this leads