The Four Postulates of Quantum Mechanics

The four postulates are not philosophy. They are the operational manual. Every theorem in the rest of this track — why gates must be unitary, why a two-qubit system has a four-dimensional state space, why measurement is irreversible, why entanglement exists, why no-cloning holds — derives from these four rules. Internalise them once, and you stop reasoning about quantum mechanics from scratch; you start reasoning from the postulates.

The postulates come in two styles of writing. Physics textbooks state them with slightly more philosophical baggage about "isolated systems" and "observables"; Nielsen and Chuang state them in the sharper, more computational form — the version you will see here. The computational form is cleaner because it strips the philosophy and keeps only the parts that a computer scientist needs to build algorithms and proofs.

One specific guarantee about this chapter: by the end, you will be able to derive, from the postulates alone, why a two-qubit state has exactly 2^2 = 4 complex amplitudes, why the Bell state (|00\rangle + |11\rangle)/\sqrt{2} cannot be written as a product of single-qubit states, and why applying a unitary to a normalised state always gives another normalised state. If any of those facts feel mysterious right now — good. By the end of this chapter they will be obvious.

The anchor — what the postulates do together

Before the four statements, the story:

1. You prepare a state. (Postulate 1 says what a state is.)
2. You evolve it using gates. (Postulate 2 says how.)
3. You measure to get a classical output. (Postulate 3 says the rule.)
4. You do all of this on multi-qubit systems. (Postulate 4 says how to combine them.)

That loop is quantum computing. Every algorithm in this track, from the trivial (apply an X gate, measure) to the profound (Shor's factoring), fits this shape. The four postulates are the formal rules that make the shape meaningful.

With the shape in view, here are the four rules.

Postulate 1 — the state space

Three words deserve attention.

"Hilbert space" means a complex inner-product space that is complete (every Cauchy sequence converges). For finite-dimensional systems — every quantum-computing system made of n qubits — completeness is automatic, and "Hilbert space" is synonymous with "finite-dimensional complex inner-product space." For infinite-dimensional systems (a particle on a line, for example, with continuous position), completeness is a genuine technical condition; you do not need to worry about it in this track.

For a single qubit, \mathcal{H} = \mathbb{C}^2. You met this in chapter 10. For n qubits, \mathcal{H} = \mathbb{C}^{2^n} — a space of dimension 2^n. (The factor of 2^n comes from postulate 4 below.)

"Unit vector" restates the normalisation requirement you met in chapter 10: \langle\psi|\psi\rangle = 1. The postulate is asserting that the physical state of the system is not an arbitrary vector, but a vector of length 1. Kets that differ only in length represent the same physical state after normalisation.

"Completely described by" is the important phrase. It says that |\psi\rangle contains all the information there is to know about the system. If two systems are both in state |\psi\rangle, they are indistinguishable by any experiment. There are no hidden variables within this framework that carry extra information — the state vector is the whole description.

Refinement — physical states are rays, not vectors

Postulate 1 says the state is a unit vector, but chapter 10's going-deeper section showed that two vectors differing only by a global phase e^{i\gamma} give identical statistics in every measurement. So the physical state is really an equivalence class of unit vectors under global phase — a ray in \mathcal{H}. This is the form the postulate takes in most physics textbooks; the computational form keeps unit vectors as the working representation and remembers the global-phase equivalence as an implicit rule.

Why the ray refinement matters in practice: you can freely multiply an entire quantum circuit's state by a global phase without changing any observable outcome. Gate optimisers and compilers use this constantly — they absorb and factor global phases to simplify circuits. The refinement is not pedantry; it is a real tool.

Postulate 2 — time evolution

A unitary operator is a linear operator U satisfying

Equivalently: U is invertible, and its inverse is its adjoint. The adjoint U^\dagger is obtained by taking the transpose and then complex-conjugating every entry (chapter 7 treats adjoints carefully; for now the formula above is enough).

Why unitary and not something more general?

Short answer: preserving normalisation under linear evolution forces unitarity.

Suppose evolution is some linear operator U: |\psi(t_1)\rangle = U|\psi(t_0)\rangle. Postulate 1 says the state at every time must be normalised. So you need

Compute the left side:

Why the U^\dagger appears on the bra: if |\psi(t_1)\rangle = U|\psi(t_0)\rangle, then taking the Hermitian conjugate gives \langle\psi(t_1)| = \langle\psi(t_0)|U^\dagger. The bra of a transformed ket is the original bra multiplied from the right by the adjoint.

For this to equal 1 for every normalised |\psi(t_0)\rangle, the operator U^\dagger U must be the identity. That is the definition of unitarity.

Linearity + normalisation-preservation + invertibility ⟹ unitarity. Two of these three are baked in (linearity by postulate, normalisation by postulate 1). Invertibility is the genuinely physical assumption — the claim that time evolution of a closed quantum system is reversible.

Reversibility and the closed-system caveat

Unitary operators are invertible, so postulate 2 says closed-system quantum evolution is reversible — every U has a U^\dagger that undoes it. Run a quantum circuit forward, then apply the conjugate-transpose in reverse, and you are back where you started.

The word "closed" is load-bearing. A closed system is one that is not interacting with anything outside itself. In reality, qubits interact with their environment constantly — stray photons, thermal vibrations, measurement apparatus — and the combined system's evolution is still unitary, but the evolution of just the qubit looks non-unitary when you ignore the environment. That non-unitary residue is called decoherence, and the description that captures it requires density matrices and channels (Part 13). For this chapter: the postulate applies as stated to any system you can isolate, and what gets violated in practice is "closed," not "unitary."

Connection to Schrödinger — one paragraph

For continuous time evolution, the unitary U(t_1, t_0) takes a specific form:

where H is a Hermitian operator called the Hamiltonian of the system, and \hbar is Planck's reduced constant. Differentiating gives the Schrödinger equation,

You will derive this carefully in chapter 12. For this chapter it is enough to know that continuous evolution in time is always the exponential of -iHt/\hbar for some Hermitian H, and that this is one way to produce the unitary U postulate 2 requires. Discrete quantum circuits — the kind you draw as boxes on wires — simply fix a U per gate and skip the continuous-time story.

Postulate 3 — measurement

This is the most technical postulate. Unpack it slowly.

"Collection \{M_m\}" — measurement is not described by a single object. It is described by a set of operators, one for each possible classical outcome. If a measurement has outcomes \{0, 1\}, there are two operators M_0, M_1. If it has outcomes \{0, 1, 2\} — say, a qutrit measured in a three-state basis — there are three.

The completeness equation \sum_m M_m^\dagger M_m = I is the quantum analogue of "the probabilities of all outcomes sum to 1." To see why, compute:

Why the sum can be pulled inside the bra-ket: inner products are linear in each slot, and a finite sum of operators inside a bra-ket equals the sum of bra-ket sandwiches of each operator. The completeness equation then forces the total to I, which gives \langle\psi|\psi\rangle = 1 for any normalised state.

So probabilities sum to 1 for any normalised state iff \sum_m M_m^\dagger M_m = I. The completeness equation is not an independent assumption — it is the condition required for the probability rule to be consistent.

The state after measurement M_m|\psi\rangle / \sqrt{p(m)} is the collapse rule. Apply the operator M_m to the state, then renormalise. The division by \sqrt{p(m)} is what ensures the post-measurement state is normalised:

The projective measurement special case

The general postulate allows measurement operators that are not projectors. The most common measurements in quantum computing are a restricted special case called projective measurements, where the M_m are orthogonal projectors:

When the measurement operators are projectors, the general postulate simplifies:

• Probability: p(m) = \langle\psi|P_m|\psi\rangle (the extra M_m^\dagger M_m = P_m^\dagger P_m = P_m collapses to a single projector).
• Post-measurement state: P_m|\psi\rangle / \sqrt{p(m)}.

This is the Born rule in full generality. You met the computational-basis version in chapter 10 (p(k) = |\alpha_k|^2 = \langle\psi|k\rangle\langle k|\psi\rangle = \langle\psi|P_k|\psi\rangle where P_k = |k\rangle\langle k|). Chapter 13 treats projective measurement as its own topic.

Why two versions of the postulate?

The general \{M_m\} formulation covers more measurement scenarios than the projective special case. For example, it captures POVMs (positive-operator-valued measures), used in optimal state-discrimination protocols, and measurements that destroy the system (where the post-measurement state does not live in the same space as the pre-measurement state). Projective measurements cover everything you see in standard quantum circuits; POVMs arise when you want to extract more nuanced information, or when the measurement apparatus is itself only partially trustworthy.

For every exercise in this track through chapter 60 or so, "measurement" means "projective measurement in some orthonormal basis." The general form is in your back pocket for when it is needed.

Postulate 4 — composite systems

This postulate does a quiet but enormous amount of work. Two consequences.

Dimension multiplies. If \dim \mathcal{H}_A = d_A and \dim \mathcal{H}_B = d_B, then \dim(\mathcal{H}_A \otimes \mathcal{H}_B) = d_A \cdot d_B. For n qubits — which is n copies of \mathbb{C}^2 — the joint space is \mathbb{C}^{2 \cdot 2 \cdots 2} = \mathbb{C}^{2^n}, a space with 2^n complex amplitudes. This is the source of the often-quoted but rarely-explained claim that "a quantum computer has 2^n amplitudes for n qubits." It is not parallelism; it is dimension counting under postulate 4.

Product states do not exhaust the composite space. The state |\psi_A\rangle \otimes |\psi_B\rangle has the form "a specific A-state combined with a specific B-state." Because the joint space is a tensor product (not a Cartesian product), most of its vectors are not of this form. A vector in \mathcal{H}_A \otimes \mathcal{H}_B that cannot be written as |\psi_A\rangle \otimes |\psi_B\rangle for any choice of single-system states is called entangled. Entanglement is a consequence of postulate 4.

Chapter 8 (tensor products) covered the mechanics. Quick summary for a two-qubit system \mathcal{H}_A = \mathcal{H}_B = \mathbb{C}^2. The basis of the joint space is \{|00\rangle, |01\rangle, |10\rangle, |11\rangle\} — four basis kets built from the two one-qubit kets of each subsystem. Any two-qubit state is a normalised linear combination:

If the state is a product |\psi_A\rangle \otimes |\psi_B\rangle = (\alpha|0\rangle + \beta|1\rangle) \otimes (\gamma|0\rangle + \delta|1\rangle), expanding gives

so the product states are exactly those where c_{00}c_{11} = c_{01}c_{10}. Most states fail this condition — hence most states are entangled.

The entanglement example — the Bell state

Consider the state

Its coefficients are c_{00} = c_{11} = 1/\sqrt{2} and c_{01} = c_{10} = 0. The product test: c_{00}c_{11} = 1/2, c_{01}c_{10} = 0. These are unequal, so |\Phi^+\rangle is not a product state. It is one of the four Bell states, the simplest entangled states in two-qubit space, and a central object in the rest of this track.

Physically, the two qubits of a Bell state are correlated in a way classical probability cannot match: measure either qubit and you immediately know what the other will give, even though individually each qubit looks maximally random. Entanglement is a property, and it is a direct gift of postulate 4.

How the four postulates fit together

The loop at the start of this chapter is now precise.

1. Prepare an n-qubit state |\psi_0\rangle — a unit vector in \mathbb{C}^{2^n} (postulates 1 and 4).
2. Evolve it through a sequence of gates U_1, U_2, \ldots, U_k, each a unitary on \mathbb{C}^{2^n} (postulate 2). The final state before measurement is |\psi_k\rangle = U_k \cdots U_2 U_1 |\psi_0\rangle.
3. Measure using a set of measurement operators \{M_m\} — typically projective measurement onto the computational basis of one or more qubits (postulate 3). Outcome m occurs with probability p(m); the state collapses to M_m|\psi_k\rangle/\sqrt{p(m)}.

That is quantum computing. Every algorithm — Deutsch-Jozsa, Grover, Shor, quantum phase estimation, variational quantum eigensolvers — is a specific choice of initial state, sequence of unitaries, and measurement. The creativity lies in the choice; the rules are fixed.

Two consequences worth calling out now, to be proven fully later:

• No-cloning (chapter 57). No unitary operator can copy an unknown quantum state — the operation U(|\psi\rangle \otimes |0\rangle) = |\psi\rangle \otimes |\psi\rangle fails to be linear on the space of input states, so no such U exists. The impossibility is a direct consequence of postulate 2 (evolution is linear and unitary).
• No-cloning implies no-faster-than-light-signaling. The correlations an entangled state produces cannot be used to send information instantaneously; the no-communication theorem (chapter 60) shows this directly from postulates 3 and 4.

Worked examples

Common confusions

• "The postulates describe what quantum mechanics is." More accurately, the postulates describe what quantum mechanics does — they are operational rules for predicting outcomes of experiments. Metaphysical interpretations (many-worlds, Copenhagen, relational, Bohmian) are philosophical layers on top. A quantum-computing reader does not need to adopt any single interpretation; the postulates work identically in all of them.

• "Unitary evolution and measurement are the same kind of process." They are very different. Unitary evolution is deterministic and reversible; it preserves superposition. Measurement is probabilistic and (generally) irreversible; it destroys superposition. One does not reduce to the other. In the density-matrix framework (Part 13) you will see that measurement can be viewed as unitary evolution of the qubit-plus-environment followed by ignoring the environment — but from the perspective of just the qubit, the two processes remain distinct.

• "Postulate 3 means measurement outcomes are random." More precisely, outcome probabilities are determined by the state; which outcome you observe on a single run is probabilistic. On many repetitions the frequencies converge to the p(m) values given by the postulate. Randomness lives at the single-run level; statistics live at the ensemble level.

• "The tensor product is just multiplying dimensions." Dimensions do multiply — \dim(\mathcal{H}_A \otimes \mathcal{H}_B) = d_A d_B — but the tensor product is more than a product. It is a bilinear construction: (\lambda|\psi_A\rangle) \otimes |\psi_B\rangle = \lambda(|\psi_A\rangle \otimes |\psi_B\rangle) and similarly on the other side, and (|\psi_A\rangle + |\phi_A\rangle) \otimes |\psi_B\rangle = |\psi_A\rangle \otimes |\psi_B\rangle + |\phi_A\rangle \otimes |\psi_B\rangle. The resulting space contains all linear combinations of products, not just products themselves — which is why entanglement exists. Chapter 8 drilled this.

• "Postulate 1 just says states are vectors." It says they are unit vectors in a Hilbert space, completely describing the system. The three qualifiers are all load-bearing. Non-unit vectors have to be renormalised before yielding probabilities; non-Hilbert-space objects (mere vector spaces without an inner product) cannot support the Born rule; and "completely describing" rules out hidden-variable theories within the standard framework.

• "The postulates apply only to microscopic systems." The postulates apply to any quantum system — microscopic, mesoscopic, and in principle macroscopic. The reason you do not see superpositions of rocks is not a limit of the postulates; it is decoherence, which is the effective non-unitary evolution a subsystem undergoes when it interacts strongly with its environment. The postulates always hold for the full system; it is the effective description of an open subsystem that looks non-quantum.

Going deeper

The four postulates above are all you need for most of this track. The rest of this section is optional depth: how density matrices extend postulate 1 to mixed states, how POVMs extend postulate 3 beyond projective measurement, how the Schrödinger equation is a continuous-time unpacking of postulate 2, why decoherence is how the real world drifts away from closed-system evolution, and how Bose statistics specialise postulate 4 for identical particles. These topics recur throughout Parts 9, 13, and 16 — the section below is a map, not a destination.

The density matrix formulation

Postulate 1 as stated requires a unit vector |\psi\rangle. This is enough to describe pure states — states about which you have maximal quantum information. It is not enough to describe two common scenarios:

1. A statistical mixture — "the system is in |\psi_1\rangle with probability p_1, else |\psi_2\rangle with probability p_2, ...". Classical ignorance of which pure state you have.
2. A subsystem of a larger entangled state — even if the whole system is in a pure state, individual subsystems are not, in general, describable by a pure-state vector.

The generalisation is the density matrix (also called density operator) \rho: a positive semidefinite operator with \operatorname{Tr}(\rho) = 1. A pure state |\psi\rangle corresponds to \rho = |\psi\rangle\langle\psi|; a statistical mixture corresponds to \rho = \sum_i p_i |\psi_i\rangle\langle\psi_i|.

In the density-matrix formulation, the postulates generalise:

• Postulate 1 (density-matrix form): state is a density matrix \rho with \rho = \rho^\dagger, \rho \geq 0, \operatorname{Tr}(\rho) = 1.
• Postulate 2: \rho \to U\rho U^\dagger under unitary evolution.
• Postulate 3: p(m) = \operatorname{Tr}(M_m^\dagger M_m \rho); post-measurement state is M_m \rho M_m^\dagger / p(m).
• Postulate 4: \rho_{AB} = \rho_A \otimes \rho_B for a product state; more generally \rho_{AB} lives in the tensor-product space and need not factorise.

Density matrices are the fuller language. Part 13 develops them; the pure-state postulates of this chapter are the restriction to situations where the density matrix happens to be |\psi\rangle\langle\psi|.

POVMs — the fullest generalisation of postulate 3

The measurement operators \{M_m\} of postulate 3 are already more general than projective measurements, but the fullest generalisation is a positive operator-valued measure (POVM): a set of positive operators \{E_m\} with \sum_m E_m = I. The probability of outcome m is p(m) = \langle\psi|E_m|\psi\rangle, with no post-measurement state specified (POVMs describe what information is extracted, not what remains).

POVMs are strictly more general than projective measurements and are the right tool for optimal state discrimination, entanglement detection, and quantum tomography. Chapter 57 treats them in detail.

The Schrödinger equation — postulate 2 in continuous time

The unitary U(t_1, t_0) of postulate 2 can be written as e^{-iH(t_1-t_0)/\hbar} for a Hermitian operator H. Differentiating in t_1 gives the Schrödinger equation:

H is the Hamiltonian — the total-energy operator of the system. The Schrödinger equation is the continuous-time version of postulate 2; discrete quantum circuits are the integrated form, where each gate is a specific time-integrated unitary.

For a static Hamiltonian, U(t) = e^{-iHt/\hbar}; for a time-varying Hamiltonian, U is given by the time-ordered exponential, which you will meet when studying quantum control. The Hamiltonian formulation is important because it is how physical hardware implements gates — you engineer H (a magnetic field, a laser pulse, a microwave tone) for the right duration to realise the desired unitary.

Decoherence and open quantum systems

The postulates assume the system is closed — isolated from its environment. Real quantum hardware is open: qubits interact with stray photons, thermal vibrations, control electronics, and the measurement apparatus. The combined (system + environment) is still closed and evolves unitarily, but the state of just the system drifts away from the closed-system trajectory.

The quantitative description is the quantum channel formalism — a completely positive, trace-preserving map \mathcal{E}(\rho) = \sum_k K_k \rho K_k^\dagger with \sum_k K_k^\dagger K_k = I (Kraus representation). This generalises postulate 2 from unitary evolution (\rho \to U\rho U^\dagger) to arbitrary physically realisable evolution. Decoherence is the channel that describes how coherent superpositions decay into classical mixtures when the environment is tracked out.

Building a useful quantum computer requires managing decoherence. Error-correction codes (Part 11) use entanglement to protect logical information from decoherence; fault-tolerant thresholds quantify how much noise is tolerable. All of this sits on top of the four postulates, extending them to the noisy-subsystem regime.

Bose-Einstein statistics and symmetric tensor products

Postulate 4 says the state space of a composite system is the full tensor product \mathcal{H}_A \otimes \mathcal{H}_B. For identical particles, the full tensor product is too large: quantum mechanics further requires the state to be either totally symmetric (for bosons) or totally antisymmetric (for fermions) under particle exchange.

Satyendra Nath Bose derived the symmetric-tensor-product statistics in 1924 by noticing that if you count photon configurations by pretending photons are distinguishable, you get the wrong blackbody law; if you count only configurations that respect the symmetric-product subspace, you get Planck's law exactly. Einstein extended this to massive particles. The integer-spin particles — photons, gluons, the Higgs boson, helium-4 nuclei — are now called bosons in Bose's honour.

For quantum computing, the symmetric-tensor-product story mostly runs silently in the background. The qubits you manipulate are physical systems — individual atoms, individual superconducting circuits — and their identical-particle statistics is built into how the physical implementation works. Some platforms (photonic quantum computing, boson sampling) expose the symmetric-tensor-product structure directly; others (gate-model superconducting qubits) hide it at the hardware level. Either way, postulate 4 as stated here is the unrestricted version; the physical symmetry constraints live inside the specific Hilbert space you choose.

Where this leads next

• Unitary Evolution — chapter 12. A full chapter on postulate 2: proof of unitarity from norm preservation, the matrix characterisation, inverse and adjoint, composition of unitaries, and gate-decomposition previews.
• Projective Measurement — chapter 13. Postulate 3 specialised to the projective case: the Born rule proved from the postulates, measurement in any orthonormal basis, expectation values, and the connection to observables.
• The Bloch Sphere — chapter 14. The geometric picture of single-qubit states and operations, coming directly from the single-qubit version of postulates 1 and 2.
• Density Matrices Introduction — chapter 39. The extension of postulate 1 to mixed states and reduced states — the fuller picture previewed above.

References

1. Nielsen and Chuang, Quantum Computation and Quantum Information, §2.2 — Cambridge University Press. The canonical CS-flavoured statement of the four postulates; the phrasing above follows theirs closely.
2. John Preskill, Lecture Notes on Quantum Computation, Chapter 2-3 — theory.caltech.edu/~preskill/ph229. Clear treatment of postulates with emphasis on the open-systems generalisation.
3. John Watrous, The Theory of Quantum Information — cs.uwaterloo.ca/~watrous/TQI. Mathematically precise development of states, channels, measurements, and tensor products.
4. Wikipedia, Mathematical formulation of quantum mechanics — the physics-flavoured statement of the postulates, with historical notes and links to observables and spectra.
5. Wikipedia, Hilbert space — the mathematical background for postulate 1.
6. Wikipedia, Satyendra Nath Bose — context for the Bose-counting and identical-particle aside.