Quantum Superposition
Explained
A qubit in superposition is a weighted combination of the states |0⟩ and |1⟩, with a complex amplitude on each. Measuring it returns exactly one of the two, with a probability equal to that amplitude squared.
What quantum superposition is
A classical bit is either 0 or 1. A qubit has two corresponding basis states, written |0⟩ and |1⟩, and its general state is a linear combination of them: |ψ⟩ = α|0⟩ + β|1⟩. The coefficients α and β are called amplitudes. They are complex numbers, and they are normalized so that |α|² + |β|² = 1. That single line is the whole definition of superposition; everything else on this page follows from it.
The amplitudes are not directly observable. When the qubit is measured in the computational basis, the result is |0⟩ with probability |α|² and |1⟩ with probability |β|², and after the measurement the qubit is in whichever basis state was observed. This rule, the Born rule, is why the squared magnitudes have to sum to one: the two outcomes are the only outcomes, and their probabilities must add up. The state |+⟩ = (|0⟩ + |1⟩)/√2, for instance, gives each outcome with probability 1/2, and so does |−⟩ = (|0⟩ − |1⟩)/√2. Those two states differ only by the sign, or phase, of the second amplitude, and a computational-basis measurement cannot tell them apart. A measurement in a different basis can, which is the first sign that a superposition carries more information than a coin flip does.
For a single qubit the normalized amplitudes can be pictured as a point on the surface of a sphere, the Bloch sphere: |0⟩ at the north pole, |1⟩ at the south pole, and every superposition somewhere in between, with the relative phase between α and β as the longitude. Quantum gates rotate that point. Measurement projects it onto one of the poles. The name comes from the superposition principle of classical wave physics, where two waves passing through the same point add up amplitude by amplitude; the quantum version keeps the addition and adds two things, complex amplitudes and the Born rule for turning them into probabilities.
Why it is not just a coin you have not looked at yet
A hidden coin has a definite face and a probability that only reflects your ignorance. A qubit in |+⟩ has no definite value before measurement, and the difference is testable: apply a Hadamard gate to |+⟩ and the qubit returns to |0⟩ every time, because the two paths that lead to |1⟩ arrive with opposite signs and cancel. A classical probability distribution over {0, 1} cannot cancel anything, since probabilities are never negative. Interference of amplitudes is the physical content of superposition, and it is the resource every quantum algorithm is built on.
How a superposition is created, steered and measured
The five moves that every gate-based quantum algorithm is made of
Start from a basis state
Every circuit begins with the register reset to |0⟩ on each qubit: amplitude 1 on |0…0⟩ and 0 everywhere else. There is no superposition yet, and a measurement here would return all zeros with certainty.
Create the superposition
A Hadamard gate maps |0⟩ to (|0⟩ + |1⟩)/√2. Applied to every qubit of an n-qubit register it produces the uniform superposition over all 2ⁿ bit strings, each with amplitude 1/√2ⁿ. This is the opening move of Grover's algorithm, QAOA and most quantum machine-learning circuits.
Set amplitudes and phases
Rotation gates shape the state. A rotation about the Y axis moves probability between |0⟩ and |1⟩; a rotation about Z leaves the probabilities alone and changes the relative phase. Data encoding in quantum machine learning is exactly this: classical features become rotation angles.
Let the amplitudes interfere
Further gates add amplitudes from different computational paths. Paths that reach the same basis state with the same phase reinforce; paths with opposite phase cancel. An algorithm is a choreography that makes the amplitude of the answer you want grow and the amplitudes of wrong answers shrink.
Measure, then repeat
Measurement returns one basis state, sampled with probability given by the squared amplitude, and the superposition is gone. To estimate the full distribution a circuit is run many times; each run is a shot, and the counts over shots are the algorithm's classical output.
Two Hadamard gates in a row are the smallest complete example. The first turns |0⟩ into an equal superposition; the second sends the |0⟩ component to (|0⟩ + |1⟩)/√2 and the |1⟩ component to (|0⟩ − |1⟩)/√2. The |1⟩ terms cancel, the |0⟩ terms add, and the qubit is back in |0⟩ with certainty. Nothing was measured in between, so nothing collapsed: the amplitudes simply interfered.
Superposition vs probability vs entanglement
Four kinds of "not just a plain 0 or 1", and what separates them
Most confusion about superposition comes from conflating it with one of its neighbours. A probabilistic bit is the closest classical analogue and the one it is most often mistaken for; entanglement is what people usually mean when they say quantum computers do something no classical machine can. The cards below put the four side by side.
Classical bit
Holds 0 or 1, nothing else. Reading it never changes it, copying it is free, and two bits are always described by their two separate values
Probabilistic bit
Is 0 with probability p and 1 with probability 1 − p. The uncertainty is in your knowledge, not in the bit: it already has a value, and probabilities can only add, never cancel
Qubit in superposition
Is α|0⟩ + β|1⟩ with complex amplitudes. Has no definite value until measured, and amplitudes with opposite phase cancel. |+⟩ and |−⟩ give the same measurement statistics and are different states
Entangled qubits
A superposition of a multi-qubit register that cannot be written as one state per qubit. In (|00⟩ + |11⟩)/√2 each qubit alone looks random, yet the two outcomes always agree
Superposition and entanglement are related but not the same thing. Every entangled state is a superposition of a multi-qubit register, but not every multi-qubit superposition is entangled: |+⟩ ⊗ |+⟩, two qubits each in an equal superposition, is a superposition of all four two-bit strings and is not entangled at all, because it factors into one state per qubit. Entanglement appears when a two-qubit gate such as CNOT acts on a superposition and ties the qubits' amplitudes together so that no such factoring exists. Superposition is the raw material; entanglement is one of the structures gates can build from it.
What superposition does and does not give you
What it enables
- An n-qubit register in superposition carries 2ⁿ amplitudes at once, so a single gate acts on all of them in one step; describing that state on a classical machine needs 2ⁿ complex numbers
- Amplitudes interfere: an algorithm can arrange for wrong answers to cancel and the right one to reinforce, which is how Grover's algorithm reaches a marked item in about √N steps instead of N
- Classical data can be encoded into amplitudes and phases, giving quantum kernels and quantum neural networks feature spaces that are exponentially large in the number of qubits
- Estimating a probability encoded as an amplitude takes O(1/ε) circuit runs with amplitude estimation, against O(1/ε²) samples for classical Monte Carlo at the same precision ε
What it does not
- One measurement yields one basis state, not the whole superposition. The 2ⁿ amplitudes cannot be read out; at most n classical bits come back from n qubits, a limit known as the Holevo bound
- An unknown superposition cannot be copied (the no-cloning theorem), so there is no reading it twice or storing a backup before a measurement destroys it
- Superposition is fragile. Interaction with the environment turns it into an ordinary classical mixture, and every gate adds a little error, which is why near-term circuits must stay shallow
- Speedups come from the structure of a specific algorithm, not from superposition alone. For most problems no algorithm is known that turns 2ⁿ amplitudes into fewer than 2ⁿ steps
The sentence to stop repeating
"A quantum computer tries all answers at once." It is true that a function applied to a uniform superposition of inputs produces a superposition of all its outputs in one pass. It is false that this helps by itself: measuring that state returns one random input-output pair, which a classical computer could have produced with one evaluation. Every genuine speedup, from Grover's search to amplitude estimation, comes from the steps after that pass, where interference concentrates amplitude on the answer before anyone measures. Superposition sets the stage; interference does the work.
How superposition is made and lost on hardware
Superconducting circuits, trapped ions, neutral atoms and photons all realize a qubit differently, but the operations on its superposition are the same abstract rotations, and the enemy is the same: decoherence. Superposition only exists while the qubit is isolated from everything that could learn its state. The environment "measures" it continuously, and every such interaction turns a little of the superposition into ordinary classical randomness. The rows below are the vocabulary hardware specifications use for that process.
Note: the ratio between how long a superposition survives and how long one gate takes sets how deep a circuit can be before its interference pattern is washed out. Noisy intermediate-scale (NISQ) algorithms are designed around that ratio: shallow circuits, many shots, and a classical optimizer doing the heavy lifting between runs. Schrödinger's cat, a superposition of a macroscopic object, is the thought experiment that shows why decoherence wins so quickly at large scales: the more particles a system has, the more ways the environment has to find out which branch it is in.
Superposition in the algorithms of our library
Where the concept turns into a circuit you can run
Every algorithm on Superpositions Studio opens by putting a register into superposition and closes by measuring it. What happens in between is the algorithm, and the pages linked below each show one way of making interference do useful work: on a search problem, an optimization, a classification, or an estimate.
Search and optimization
Grover's algorithm starts from the uniform superposition over all N = 2ⁿ candidates and applies amplitude amplification: an oracle flips the phase of the marked item, a reflection about the mean turns that phase into a larger amplitude, and after about √N rounds the marked item dominates the measurement. QAOA and quantum annealing do the same with an optimization objective: the superposition over all bit strings is steered, by problem-dependent phases and a mixing step, toward low-cost strings, and each measurement is a candidate solution.
Machine learning and estimation
A quantum support vector machine and a hybrid quantum neural network encode each data point as rotation angles, so that a sample becomes a superposition whose amplitudes and phases depend on its features. The overlap between two such states is a kernel; trainable rotations on top of the encoding are the quantum layer of the network. Quantum amplitude estimation goes the other way: it prepares a superposition in which the quantity of interest is an amplitude, then reads that amplitude out with fewer circuit runs than sampling would need.
2ⁿ
Basis states an n-qubit register superposes
|α|²
Probability of reading |0⟩ from α|0⟩ + β|1⟩
H
The gate that turns |0⟩ into (|0⟩ + |1⟩)/√2
1
Basis state returned by a single measurement
n
Classical bits recoverable from n qubits (Holevo bound)
√N
Grover iterations to find one of N items
Each linked page reports what the algorithm did on a concrete problem against a classical baseline, with the circuits, qubit counts and shot budgets used. The comparisons in the Quantum Solutions Library are the place to see how far the interference advantage carries on today's hardware, and where it does not yet.
FAQ
Common questions about quantum superposition, qubits and measurement
What is quantum superposition in simple terms?
A qubit in superposition is not 0 or 1 but a weighted combination of both, with a number called an amplitude on each. You never see the combination directly: measuring the qubit gives 0 or 1, and the probability of each is the square of its amplitude. Until that measurement the qubit has no definite value, and the two amplitudes can cancel or reinforce each other, which is what makes it different from a coin you have not looked at.
Is superposition the same as probability?
No. A probabilistic bit already has a value that you do not know, and probabilities are never negative, so they can only add. Amplitudes are complex numbers that can cancel. The states (|0⟩ + |1⟩)/√2 and (|0⟩ − |1⟩)/√2 give identical measurement statistics yet are different states, and a Hadamard gate sends one to |0⟩ and the other to |1⟩. No probability distribution behaves that way.
What is the difference between superposition and entanglement?
Superposition is a property of any quantum state: it is a combination of basis states with amplitudes. Entanglement is a property of a multi-qubit superposition that cannot be split into one state per qubit. Two qubits each in (|0⟩ + |1⟩)/√2 are in superposition but not entangled; (|00⟩ + |11⟩)/√2 is both. Entanglement is created by two-qubit gates acting on superpositions.
How many states can n qubits be in superposition over?
All 2ⁿ computational basis states at once, each with its own complex amplitude. Ten qubits span 1,024 basis states, twenty span about a million, fifty span about 10¹⁵. Writing such a state down classically needs 2ⁿ numbers, which is why simulating even a few dozen qubits exactly is hard, and also why you cannot simply read all 2ⁿ amplitudes back out.
What happens to a superposition when you measure it?
The measurement returns one basis state, chosen at random with probability equal to the squared amplitude of that state, and the qubit is left in that basis state. The other amplitudes are gone. To learn anything about the distribution you run the same circuit many times, called shots, and look at the counts. Algorithms are designed so the answer you want has a large amplitude before the measurement happens.
Does superposition make quantum computers faster?
Not by itself. Putting 2ⁿ inputs into superposition and evaluating a function on all of them in one pass sounds like parallelism, but a measurement returns only one input-output pair. Speedups such as Grover's √N search or amplitude estimation's O(1/ε) scaling come from interference: gate sequences that make wrong answers cancel and right answers reinforce before the measurement. Superposition is necessary for that, but the algorithm is what makes it pay.
Related pages
Algorithms, applied use cases and benchmarks connected to this page
Algorithms
Grover's Algorithm for Quantum Search
Learn and benchmark Grover's algorithm for quantum search with reproducible circuits, oracle constructions, and implementation notes.
Algorithms
QAOA for QUBO and Ising Optimization
Explore the Quantum Approximate Optimization Algorithm for QUBO and Ising problems with reproducible benchmarks, parameter schedules, and hardware notes.
Algorithms
Quantum Annealing for QUBO and Ising Problems
Map combinatorial optimization problems to QUBO and Ising form and benchmark quantum annealing with reproducible runs, baselines, and hardware notes.
Algorithms
Quantum Support Vector Machine (QSVM)
Build and benchmark quantum support vector machine pipelines with quantum kernels, classical SVM baselines, and reproducible machine learning workflows.
Algorithms
Quantum Neural Networks: QNN & HQNN
Understand quantum neural networks (QNN) and hybrid QNN (HQNN): how each works, when to use which, real benchmarks, and NISQ hardware requirements.
Algorithms
Quantum Amplitude Estimation (QAE)
Estimate probabilities, expectations, and VaR/CVaR with IQAE's quadratic speedup over Monte Carlo sampling, benchmarked with a real proof of concept.
Ready to put superposition to work?
Run Grover, QAOA, QSVM, HQNN and amplitude estimation on your own problem, compare against classical baselines, and export reproducible code and reports.
Powered by Superpositions Studio: transparent, reproducible quantum computing