Superposition Principle

The superposition principle is the single postulate that most sharply separates quantum physics from everything before it. It says nothing more — and nothing less — than this: the states of an isolated quantum system are the unit vectors of a complex Hilbert space $\mathcal{H}$ , and because a Hilbert space is a linear vector space, any normalized complex combination of valid states is again a valid state. From two distinguishable configurations $|\psi_0\rangle$ and $|\psi_1\rangle$ the theory manufactures a whole continuum of new states $\alpha|\psi_0\rangle+\beta|\psi_1\rangle$ with no classical counterpart. A bit is either 0 or 1; a qubit can be any point on a sphere of superpositions between them. What makes this more than bookkeeping is that the coefficients are complex amplitudes, not probabilities. A classical hedge about two alternatives is a probability pair $(p_0,p_1)$ with $p_0+p_1=1$ . A superposition stores $(\alpha,\beta)$ with $|\alpha|^2+|\beta|^2=1$ and carries one extra physically real degree of freedom the probabilities do not: the relative phase between $\alpha$ and $\beta$ . When two paths lead to the same outcome, their amplitudes add as complex numbers before the modulus is squared, and the surviving cross term is interference — constructive or destructive. That term is the engine of every quantum algorithm and the fingerprint that separates a genuine superposition from a statistical mixture. Two normalization facts keep the principle well-defined. An overall global phase $e^{i\gamma}$ is unobservable, so physical states are rays, not vectors; and the Born rule ties $|\alpha_k|^2$ to measurement statistics, which is why the amplitudes must be normalized. Superposition, the inner product, and measurement form one interlocking structure.

Key takeaways

If $|\psi_0\rangle$ and $|\psi_1\rangle$ are valid states, so is any normalized complex combination $\alpha|\psi_0\rangle+\beta|\psi_1\rangle$ — the principle is just the linearity of the state space.
The coefficients are amplitudes $\alpha,\beta\in\mathbb{C}$ : their relative magnitude sets the outcome probabilities and their relative phase is a real, measurable degree of freedom.
Superposition is not classical probability: amplitudes add before squaring, so distinct alternatives can interfere and even cancel — something no probability distribution can do.
The Born rule $\Pr(k)=|\langle k|\psi\rangle|^2$ links amplitudes to statistics and forces the normalization $\sum_k|\alpha_k|^2=1$ ; a global phase $e^{i\gamma}$ changes nothing.
One qubit fills the whole Bloch sphere: $|\psi\rangle=\cos\tfrac\theta2|0\rangle+e^{i\phi}\sin\tfrac\theta2|1\rangle$ — the tilt $\theta$ sets the magnitudes, the azimuth $\phi$ the relative phase.
Key idea: quantum states form a complex vector space, so a normalized linear combination of states is again a state — and because the combination happens at the level of complex amplitudes added before squaring, distinct alternatives can interfere.

Where this lives today

Superposition is now an engineered, monitored quantity. In a superconducting processor a transmon qubit is driven onto the Bloch-sphere equator — a genuine $\tfrac1{\sqrt2}(|0\rangle+|1\rangle)$ superposition — and a Ramsey experiment watches the coherence $\alpha\beta^*$ decay, directly measuring the coherence time $T_2$ before the environment turns the pure state into a mixture. Google's 2019 Sycamore demonstration and its successors rest on holding a $53$ -qubit register in a superposition over $2^{53}\approx9\times10^{15}$ computational branches and letting interference sculpt the output distribution. At the other extreme, experiments push superposition toward the macroscopic: SQUID loops carry clockwise and counter-clockwise supercurrents at once, and matter-wave interferometers have interfered single molecules of tens of thousands of atomic mass units, probing whether the principle ever fails at large scale. Every error-correcting code, interferometric sensor, and key-distribution protocol is at bottom a scheme for creating, protecting, and reading out the relative phases of a superposition.

The physical picture

Picture a quantum state as an arrow in a complex vector space rather than a marble in one of two boxes. Classically the state of a bit is one of two points, 0 or 1. Quantum mechanically the arrow may point anywhere on a continuous surface — the Bloch sphere for a single qubit — and every direction is a legitimate superposition of 0 and 1. The tilt away from the north pole fixes how much amplitude sits on $|0\rangle$ versus $|1\rangle$ ; the rotation about the vertical axis fixes the relative phase. The first figure makes this concrete: a specific tilt reads off specific probabilities.

Figure

A qubit state as a point on the Bloch sphere, with amplitude cos(θ/2) on the 0 axis and e^(iφ)sin(θ/2) on the 1 axis. At polar angle θ=60° the amplitudes are cos30°=0.866 and sin30°=0.500, so P(0)=0.75 and P(1)=0.25, while the azimuth φ carries the physically real relative phase

The reason the amplitudes must be complex is that two ways of reaching the same outcome have to be able to cancel. Two paths carrying $+\tfrac1{\sqrt2}$ and $-\tfrac1{\sqrt2}$ add to $1$ as probabilities but to $0$ as amplitudes, so the outcome never happens — destructive interference, impossible for ordinary probabilities and precisely the resource algorithms exploit. The slogan: probabilities add for classical alternatives, amplitudes add for quantum alternatives — add the amplitudes first, square the modulus after. The second figure shows the payoff: a coherent superposition and a 50/50 mixture are indistinguishable in one basis yet completely distinguishable in another.

Figure

The experimental fingerprint of coherence. A superposition and a 50/50 mixture are identical in the Z basis, each giving P=0.5, but differ sharply in the X basis, where the superposition gives P(+)=1.0 and the mixture P(+)=0.5. The gap is the interference term Re(α*β)=+0.5

The core equations

The starting point is the definition itself: superposition is the closure of the state space under normalized complex combination. Two states go in, one comes out, with complex weights.

|\psi\rangle=\alpha\,|\psi_0\rangle+\beta\,|\psi_1\rangle,\qquad \alpha,\beta\in\mathbb{C} \qquad (1)

where: $|\psi_0\rangle,|\psi_1\rangle$ — two valid, typically orthonormal, states; $\alpha,\beta$ — complex probability amplitudes; $|\psi\rangle$ — the resulting superposed state. Reading it: the only freedom quantum mechanics adds over " $|\psi_0\rangle$ or $|\psi_1\rangle$ " is the pair of complex weights — and that freedom is the whole subject. Nothing restricts a superposition to two terms. Over a full orthonormal basis the state is a weighted sum whose weights are extracted by the inner product, and the normalization constraint is what makes those weights probabilities.

|\psi\rangle=\sum_{k}\alpha_k\,|k\rangle,\qquad \alpha_k=\langle k|\psi\rangle,\qquad \sum_k|\alpha_k|^2=1 \qquad (2)

where: $\{|k\rangle\}$ — an orthonormal basis of $\mathcal{H}$ ; $\alpha_k$ — the amplitude on $|k\rangle$ , read off by projecting; $\sum_k|\alpha_k|^2$ — total probability, pinned to 1. Reading it: expanding a state in a basis and normalizing are one act, because the norm is the sum of the squared amplitudes. The link between the abstract amplitudes and anything measurable is the Born rule — the bridge that makes the normalization in (2) mandatory rather than conventional.

\Pr(k)=|\langle k|\psi\rangle|^2=|\alpha_k|^2 \qquad (3)

where: $\Pr(k)$ — the probability of outcome $k$ when measuring in the basis $\{|k\rangle\}$ ; $|\alpha_k|^2$ — the squared modulus of that amplitude. Reading it: the amplitude is not itself a probability; its squared modulus is, which is why phases can hide inside amplitudes and still influence a different measurement. For a single qubit the two complex amplitudes, stripped of unphysical redundancy, collapse onto two real angles — the Bloch-sphere coordinates of the figure above.

|\psi(\theta,\phi)\rangle=\cos\tfrac{\theta}{2}\,|0\rangle+e^{i\phi}\sin\tfrac{\theta}{2}\,|1\rangle \qquad (4)

where: $\theta\in[0,\pi]$ — polar angle fixing the amplitude magnitudes; $\phi\in[0,2\pi)$ — azimuth carrying the relative phase. Reading it: a qubit stores a continuous point on a sphere yet a single measurement returns one bit — the tension at the heart of quantum information. The measurable difference between a superposition and a mixture lives in a single term, exposed by measuring $|\psi\rangle=\alpha|0\rangle+\beta|1\rangle$ against $|+\rangle=\tfrac1{\sqrt2}(|0\rangle+|1\rangle)$ .

\Pr(+)=\tfrac12\big(1+2\,\mathrm{Re}(\alpha^*\beta)\big)=\tfrac12+\mathrm{Re}(\alpha^*\beta) \qquad (5)

where: $|+\rangle$ — the $X$ -eigenstate we project onto; $\mathrm{Re}(\alpha^*\beta)$ — the interference (coherence) term. Reading it: a classical mixture lacks this term and always gives $\tfrac12$ , so any deviation from $\tfrac12$ is the signature of a genuine superposition. The same distinction, written with operators, is the difference between a rank-one projector and a diagonal average — the language of decoherence and error correction.

\rho_{\text{pure}}=|\psi\rangle\langle\psi|=\begin{pmatrix}|\alpha|^2 & \alpha\beta^*\\ \alpha^*\beta & |\beta|^2\end{pmatrix}\ \neq\ \tfrac12\!\begin{pmatrix}1&0\\0&1\end{pmatrix}=\rho_{\text{mix}} \qquad (6)

where: $\rho_{\text{pure}}$ — the density operator of a superposition, with nonzero off-diagonal coherences $\alpha\beta^*$ ; $\rho_{\text{mix}}$ — that of a 50/50 mixture, purely diagonal. Reading it: coherence is literally an off-diagonal matrix element, and decoherence erases it. Finally, superpositions survive quantum dynamics because that dynamics is linear — the property that turns superposition into a computational resource.

U\big(\alpha|\psi_0\rangle+\beta|\psi_1\rangle\big)=\alpha\,U|\psi_0\rangle+\beta\,U|\psi_1\rangle \qquad (7)

where: $U$ — any unitary (indeed any linear) operator. Reading it: one application of $U$ to a superposed input produces the superposition of all corresponding outputs at once — quantum parallelism — with interference selecting the answer afterward.

Derivation AWhy normalized amplitudes are exactly the Born probabilities

The goal here is to show that the normalization condition $\sum_k|\alpha_k|^2=1$ is not an arbitrary tidiness rule but a direct consequence of the linear structure of the superposition together with the inner product that defines length in Hilbert space. We take a two-term superposition of orthonormal states, expand its squared norm honestly, and watch the cross terms vanish — leaving a sum of squared moduli that must equal one, which is exactly what the Born rule needs to read those moduli as probabilities.

Step 1. Write the bra of the state as the conjugate transpose of the ket: $\langle\psi|=\alpha^*\langle\psi_0|+\beta^*\langle\psi_1|$ . Why: the inner product is antilinear in its first slot, so every amplitude in the bra is complex-conjugated — this conjugation is exactly what will turn amplitudes into real, nonnegative probabilities.

Step 2. Multiply out the four terms of $\langle\psi|\psi\rangle$ , giving $|\alpha|^2\langle\psi_0|\psi_0\rangle+|\beta|^2\langle\psi_1|\psi_1\rangle+\alpha^*\beta\langle\psi_0|\psi_1\rangle+\beta^*\alpha\langle\psi_1|\psi_0\rangle$ . Why: the linear combination distributes over the inner product, so the norm carries not only the diagonal weights but also two cross terms that encode overlap between the components.

Step 3. Impose orthonormality, $\langle\psi_i|\psi_j\rangle=\delta_{ij}$ : the diagonal overlaps become 1 and the cross overlaps vanish, leaving $\langle\psi|\psi\rangle=|\alpha|^2+|\beta|^2$ . Why: for orthonormal components the interference between different basis states washes out of the norm, so total probability is a plain Pythagorean sum of the squared amplitudes.

Step 4. Demand a physical (unit) state, $|\alpha|^2+|\beta|^2=1$ , and read $|\alpha|^2,|\beta|^2$ as $\Pr(0),\Pr(1)$ . Why: only a unit vector has $|\alpha_k|^2$ summing to one, so only then can the Born rule call each $|\alpha_k|^2$ a probability — normalization and the probability interpretation are the same requirement. *Quick checks: for $\alpha=\beta=\tfrac1{\sqrt2}$ the squared norm is $\tfrac12+\tfrac12=1$ ; the state $\tfrac15(3|0\rangle+4|1\rangle)$ gives $\tfrac{9}{25}+\tfrac{16}{25}=1$ . Cross terms cancel identically for orthonormal components.*

Go deeper · Derivation B · 1 minInterference: amplitudes add before they are squared

Derivation BInterference: amplitudes add before they are squared

The aim of this derivation is to make the difference between a superposition and a classical mixture quantitative rather than philosophical, by computing one measurement probability two ways. We prepare $|\psi\rangle=\alpha|0\rangle+\beta|1\rangle$ and ask for the probability that an $X$ -basis measurement returns $|+\rangle=\tfrac1{\sqrt2}(|0\rangle+|1\rangle)$ . Doing it with amplitudes produces an extra cross term; redoing it for a statistical mixture makes that term vanish. The presence or absence of that one term is the entire operational content of the word "coherent".

Step 1. Compute the amplitude of the outcome by projecting: $\langle+|\psi\rangle=\tfrac1{\sqrt2}(\langle0|+\langle1|)(\alpha|0\rangle+\beta|1\rangle)=\tfrac1{\sqrt2}(\alpha+\beta)$ . Why: because $|+\rangle$ overlaps both $|0\rangle$ and $|1\rangle$ , the two amplitudes are summed here — before any squaring — which is where interference can occur.

Step 3. Expand the modulus-square: $\tfrac12|\alpha+\beta|^2=\tfrac12(|\alpha|^2+|\beta|^2+\alpha^*\beta+\beta^*\alpha)=\tfrac12(1+2\,\mathrm{Re}(\alpha^*\beta))$ . Why: the diagonal pieces sum to one by normalization, and the two off-diagonal pieces combine into $2\,\mathrm{Re}(\alpha^*\beta)$ — the interference term that depends on the relative phase of $\alpha$ and $\beta$ .

Step 4. Contrast with a 50/50 mixture, which measures $|0\rangle$ or $|1\rangle$ first and then projects: $\Pr_{\text{mix}}(+)=\tfrac12\cdot\tfrac12+\tfrac12\cdot\tfrac12=\tfrac12$ . Why: a mixture has no fixed relative phase to carry a $\mathrm{Re}(\alpha^*\beta)$ term, so its $X$ -basis probability is pinned at $\tfrac12$ — the deviation of a real experiment from $\tfrac12$ measures coherence. *Quick checks: $|+\rangle$ ( $\alpha=\beta=\tfrac1{\sqrt2}$ ) gives $\Pr(+)=\tfrac12+\tfrac12=1$ , consistent with $\langle+|+\rangle=1$ ; $|-\rangle$ ( $\alpha=-\beta=\tfrac1{\sqrt2}$ ) gives $\Pr(+)=\tfrac12-\tfrac12=0$ , consistent with $\langle+|-\rangle=0$ .*

Go deeper · Derivation C · 1 minFrom four real numbers to the Bloch sphere

Derivation CFrom four real numbers to the Bloch sphere

This derivation explains why a single qubit, whose two complex amplitudes seem to hold four real numbers, actually lives on a two-dimensional sphere. Two of those four parameters are physically inert: normalization removes one, and the unobservability of the global phase removes another. What remains is exactly the polar angle and azimuth of equation (4) — and tracking which parameter dies at which step is the cleanest way to see that magnitudes and relative phase are the only physical content of a qubit superposition.

Step 1. Write both amplitudes in polar form, $\alpha=r_0e^{i\gamma_0}$ and $\beta=r_1e^{i\gamma_1}$ , exposing four real parameters $r_0,r_1,\gamma_0,\gamma_1$ . Why: separating magnitude from phase lets us retire the redundant parameters one at a time instead of all at once.

Step 2. Impose normalization $r_0^2+r_1^2=1$ and solve it with $r_0=\cos\tfrac\theta2,\ r_1=\sin\tfrac\theta2$ for a single angle $\theta\in[0,\pi]$ . Why: the unit-norm constraint ties the two magnitudes together, so one real parameter is spent and the half-angle parametrization automatically respects $r_0,r_1\ge0$ .

Step 3. Factor out the global phase $e^{i\gamma_0}$ , which multiplies the entire state and is therefore unobservable, leaving only the relative phase $\phi=\gamma_1-\gamma_0$ . Why: because $e^{i\gamma_0}|\psi\rangle$ and $|\psi\rangle$ give identical probabilities and expectation values, the overall phase carries no physics and a second real parameter is retired.

Step 4. Collect the survivors: $|\psi\rangle=\cos\tfrac\theta2|0\rangle+e^{i\phi}\sin\tfrac\theta2|1\rangle$ , a point $(\theta,\phi)$ on the Bloch sphere. Why: exactly two real degrees of freedom remain — magnitude tilt and relative phase — which is why the qubit's state space is a sphere and not a four-dimensional blob. *Quick checks: $\theta=\tfrac\pi2,\phi=0$ gives $|+\rangle$ ; $\theta=\tfrac\pi2,\phi=\pi$ gives $|-\rangle$ ; $\theta=0$ gives $|0\rangle$ ; and the normalization $\cos^2\tfrac\theta2+\sin^2\tfrac\theta2=1$ holds for every $\theta$ .*

Limiting cases

The superposition principle has sharp edges. The table collects the regimes where its content is tested — including those that quietly reproduce classical physics.

Case	Result	Meaning
No relative phase ( $\alpha,\beta$ real, equal, same sign)	$\tfrac1{\sqrt2}(\vert 0\rangle+\vert 1\rangle)=\vert +\rangle$	equal-weight superposition with zero relative phase; a definite $X$ -eigenstate, maximally distinct from a mixture in the $X$ basis
Relative phase $\pi$	$\tfrac1{\sqrt2}(\vert 0\rangle-\vert 1\rangle)=\vert -\rangle$	same magnitudes as $\vert +\rangle$ but flipped phase; orthogonal to $\vert +\rangle$ , proving relative phase is fully physical
Global phase only	$e^{i\gamma}\vert \psi\rangle\sim\vert \psi\rangle$	multiplying the whole state by one phase changes no statistics; physical states are rays, not vectors
One amplitude zero	$\alpha\vert 0\rangle+0\cdot\vert 1\rangle=\vert 0\rangle$	a "superposition" with a vanishing amplitude is just a basis state — the principle smoothly contains classical configurations
Decohered ( $\alpha\beta^*\to0$ )	$\rho_{\text{pure}}\to\rho_{\text{mix}}$	when coupling to an environment erases the off-diagonal coherence, the superposition degrades into a classical mixture

Common trapA superposition is not "the system secretly in one state we just haven't looked at"

This is the most common and most damaging misreading. A coherent superposition $\tfrac1{\sqrt2}(|0\rangle+|1\rangle)$ is a single definite pure state — it is exactly $|+\rangle$ , with certainty $\Pr(+)=1$ in the $X$ basis. It is not a system that is "really" 0 or 1 with us merely ignorant of which; that ignorance is a statistical mixture, which gives $\Pr(+)=\tfrac12$ . The two are experimentally different, and the difference is the interference term $\mathrm{Re}(\alpha^*\beta)$ , the off-diagonal coherence of the density matrix. Two further traps ride alongside: the relative phase between $\alpha$ and $\beta$ is physical and measurable, while the global phase $e^{i\gamma}$ is not — so $|\psi\rangle$ and $e^{i\gamma}|\psi\rangle$ are the same state. Never say "the qubit is 0 and 1 at the same time" as if both were latent facts; say "the qubit is in a definite superposition whose statistics depend on the basis you measure".

Worked examples

1 · Normalizing a state and reading its probabilities (confidence builder)

You are handed the unnormalized vector $3|0\rangle+4|1\rangle$ . Turn it into a physical state and give the computational-basis probabilities.

Step 1. Squared norm (orthonormality kills cross terms): $\langle\chi|\chi\rangle=|3|^2+|4|^2=9+16=\mathbf{25}$ .

Step 2. Divide by $\sqrt{25}=5$ : $|\psi\rangle=\tfrac15(3|0\rangle+4|1\rangle)$ . Scaling by a positive real changes no relative amplitude or phase — same physical state, now unit length.

Step 3. Born probabilities: $\Pr(0)=(3/5)^2=\tfrac{9}{25}=\mathbf{0.36}$ , $\Pr(1)=(4/5)^2=\tfrac{16}{25}=\mathbf{0.64}$ , summing to $1$ . The state is $\tfrac15(3|0\rangle+4|1\rangle)$ with $\Pr(0)=0.36$ , $\Pr(1)=0.64$ .

2 · A complex amplitude, its relative phase, and the interference it produces

Step 1. Squared norm: $\langle\chi|\chi\rangle=|2|^2+|1+i|^2=4+(1+i)(1-i)=\mathbf{6}$ , so $|\psi\rangle=\tfrac1{\sqrt6}\big(2|0\rangle+(1+i)|1\rangle\big)$ with $\alpha=2/\sqrt6$ , $\beta=(1+i)/\sqrt6$ .

Step 2. Probabilities: $\Pr(0)=|\alpha|^2=\tfrac{4}{6}=\tfrac23$ , $\Pr(1)=|\beta|^2=\tfrac{2}{6}=\tfrac13$ , summing to $1$ . The relative phase of the $|1\rangle$ amplitude is $\arg(1+i)=\pi/4$ — a physical feature untouched by normalization.

Step 3. Interference term: $\alpha^*\beta=\dfrac{2}{\sqrt6}\cdot\dfrac{1+i}{\sqrt6}=\dfrac{2(1+i)}{6}=\dfrac{1+i}{3}$ , so $\mathrm{Re}(\alpha^*\beta)=\tfrac13$ . Then by equation (5), $\Pr(+)=\tfrac12+\tfrac13=\mathbf{\tfrac56\approx0.833}$ .

Step 4. Contrast with a mixture. Any mixture with these populations gives $\Pr_{\text{mix}}(+)=\tfrac23\cdot\tfrac12+\tfrac13\cdot\tfrac12=\tfrac12$ — always $\tfrac12$ in the $X$ basis, regardless of populations, because it has no phase. The coherent state gives $\Pr(+)=\tfrac56$ ; any mixture gives $\tfrac12$ ; the difference $\tfrac13=\mathrm{Re}(\alpha^*\beta)$ is the coherence.

3 · Destructive interference makes an outcome vanish

Two paths deliver amplitudes $\tfrac1{\sqrt2}$ and $-\tfrac1{\sqrt2}$ to one detector. Adding amplitudes first: $A=\tfrac1{\sqrt2}-\tfrac1{\sqrt2}=0$ , so $\Pr(\text{click})=|A|^2=\mathbf{0}$ — the detector never fires, though each path alone gives $\tfrac12$ and the classical sum of probabilities is $\tfrac12+\tfrac12=1$ . The amplitudes cancel to $\Pr=0$ , versus the classical $\Pr=1$ : destructive interference in one line.

Physics of the portrait

Paul Dirac insisted superposition was the true first principle of quantum mechanics, and opened his textbook with it.

Origins · 1 minPaul Dirac and how this was discovered

Paul Dirac

1902–1984 · the superposition principle and the transformation theory of quantum mechanics

Paul Adrien Maurice Dirac, trained as an electrical engineer in Bristol before turning to physics at Cambridge, gave quantum mechanics much of its permanent language. His 1927 transformation theory unified Heisenberg's matrix mechanics and Schrödinger's wave mechanics into a single picture of states as vectors and observables as operators — the framework in which "superposition" is simply vector addition. The Principles of Quantum Mechanics (1930) opened not with an equation but with the superposition principle as its primary postulate: physical states form a linear space, and the possibility of preparing a system "intermediate" between two classically exclusive configurations is, for Dirac, the irreducible novelty of the theory. He introduced the bra-ket notation that makes $\alpha|\psi_0\rangle+\beta|\psi_1\rangle$ and $\langle k|\psi\rangle$ everyday tools, predicted antimatter from the relativistic wave equation that bears his name, and shared the 1933 Nobel Prize with Schrödinger. His conviction that superposition — the linearity of the state space — is the conceptual heart of quantum mechanics is exactly this article's point of view.

History

1926

— Schrödinger's wave mechanics lets wavefunctions be added linearly

1926

— Born reads the squared amplitude as a probability, fixing what the coefficients mean

1930

— Dirac's Principles of Quantum Mechanics opens with the superposition principle as postulate one

1932

— von Neumann formalizes states as rays in a complex Hilbert space

1935

— Schrödinger's cat dramatizes superposition at macroscopic scale

Going further

Texts. Nielsen & Chuang, Quantum Computation and Quantum Information, §2.1–2.2 (the state postulate and the Bloch sphere) · Preskill, Quantum Information Lecture Notes, Ch. 2 · Watrous, The Theory of Quantum Information, Ch. 1 · Wilde, Quantum Information Theory, Ch. 3 · Kitaev, Shen & Vyalyi, Classical and Quantum Computation, Ch. 7 · Mermin, Quantum Computer Science, Ch. 1 · Dirac, The Principles of Quantum Mechanics, Ch. 1 (the original statement of the principle). Every equation is standard and independently checkable. Key numerical facts: $|1+i|^2=(1+i)(1-i)=2$ , so $\|2|0\rangle+(1+i)|1\rangle\|^2=6$ with $\Pr(0)=\tfrac23,\Pr(1)=\tfrac13$ ; $\mathrm{Re}(\alpha^*\beta)=\tfrac13$ , hence $\Pr(+)=\tfrac56$ for that state; $9+16=25$ , giving $\Pr=0.36,0.64$ ; and $\langle+|-\rangle=0$ . The $2^{53}$ and $T_2$ figures are quoted as orders of magnitude from public superconducting-qubit demonstrations.

Prerequisites. The Qubit and Its State Space · Dirac Bra-Ket Notation · Inner Product, Norm, and Normalization. Leads to. Global and Relative Phase · The Born Rule · The Measurement Postulate · Pure versus Mixed States.

Ask the tutor: Ask me why $\tfrac1{\sqrt2}(|0\rangle+|1\rangle)$ and a 50/50 mixture give identical computational-basis statistics yet split apart the instant you measure in the $X$ basis.

Is a superposition just the system being "really" in one state that we haven't looked at yet?

No, and this is the single most important thing to get right. "Really 0 or 1, we just don't know which" describes a classical mixture — a state of ignorance — which predicts 50/50 statistics in every basis. A superposition $|+\rangle=\tfrac1{\sqrt2}(|0\rangle+|1\rangle)$ is a definite pure state predicting a certain outcome ( $\Pr(+)=1$ ) in the $X$ basis. Those predictions differ, and experiments confirm the superposition's, not the mixture's. The superposition is not hidden ignorance of a pre-existing value; it is a genuinely different physical state whose statistics depend on the basis you choose — a distinction captured exactly by the off-diagonal coherences of the density matrix.

Why do the coefficients have to be complex — wouldn't real probabilities do?

Because interference requires it. If states combined with real, nonnegative weights summing to one, you would just be doing classical probability, and probabilities can only pile up, never cancel. Complex amplitudes — added before the modulus is squared — let two contributions with opposite phase cancel to zero, producing destructive interference and every quantum speedup. The extra ingredient a complex amplitude carries over a probability is the phase, and the relative phase between two amplitudes is a genuine, measurable degree of freedom. Real amplitudes are a special case (phases $0$ or $\pi$ ); the full complex freedom is what fills the Bloch sphere.

What is the difference between global phase and relative phase?

Global phase is an overall factor $e^{i\gamma}$ multiplying the entire state; relative phase is the phase between two amplitudes. The global phase is invisible: $e^{i\gamma}|\psi\rangle$ and $|\psi\rangle$ give identical probabilities and expectation values in every basis, which is why physical states are rays, not vectors. The relative phase is the opposite — fully physical. The states $\tfrac1{\sqrt2}(|0\rangle+|1\rangle)$ and $\tfrac1{\sqrt2}(|0\rangle-|1\rangle)$ differ only by a relative phase of $\pi$ , yet they are orthogonal and perfectly distinguishable in the $X$ basis. Tuning a qubit's azimuth $\phi$ on the Bloch sphere sets exactly this relative phase.

Is it correct to say a qubit is "in both states at the same time"?

It is a slogan, not a statement, and it misleads. A qubit in $\tfrac1{\sqrt2}(|0\rangle+|1\rangle)$ is not holding two latent classical values; it is one definite quantum state that simply is not an eigenstate of the computational-basis measurement. "Both at once" invites the false picture of a coin that is secretly heads-and-tails — precisely the mixture the superposition is not. Safer: the qubit is in a definite superposition, and measuring in the computational basis yields 0 or 1 with the Born probabilities while another basis can give a certain answer. The slogan only becomes useful when reinterpreted as quantum parallelism — one unitary acting on all branches — where interference, not simultaneity, does the work.

What actually goes wrong if I forget to normalize a superposition?

The Born rule stops giving probabilities. If $\langle\psi|\psi\rangle\neq1$ , the numbers $|\alpha_k|^2$ no longer sum to one and cannot be read as a distribution. The fix is mechanical: divide by the norm $\sqrt{\langle\psi|\psi\rangle}$ , a positive real number that leaves every ratio of amplitudes — hence every relative magnitude and phase — untouched. That is why normalization changes nothing physical: it rescales an overall factor, like a global phase, carrying no observable content. In practice you compute probabilities as $|\alpha_k|^2/\langle\psi|\psi\rangle$ — normalizing first and squaring after.

Does measuring a superposition destroy it?

A projective measurement in a basis $\{|k\rangle\}$ leaves the system in the outcome eigenstate $|k\rangle$ , so the original superposition of those basis states is gone — the "collapse" of the measurement postulate. But it is destroyed only relative to the basis you measured: measuring $|+\rangle$ in the $X$ basis returns $+$ with certainty and leaves it undisturbed, because $|+\rangle$ is already an eigenstate there. And "destroyed" refers to that single run; the amplitudes are never directly observed, only their squared moduli sampled, which is why reconstructing an unknown state requires many copies (tomography). Superposition is fragile to the right measurement but invisible to a measurement in its own basis.