A pure quantum state is a ray in this space, usually represented by a unit state vector \( |\psi\rangle \). Vectors that differ only by an overall phase describe the same physical state. The structure is named for David Hilbert, and John von Neumann set out its role in quantum mechanics in 1932. For a single qubit, the space is \( \mathbb{C}^2 \), spanned by \( |0\rangle \) and \( |1\rangle \).

\[ |\psi\rangle = \sum_i c_i |i\rangle, \qquad \sum_i |c_i|^2 = 1 \]

The complex amplitudes \( c_i \) express a superposition. Measuring in that basis returns outcome \( i \) with probability \( p_i = |c_i|^2 \) (the Born rule).

For a closed system, evolution is a unitary map, \( |\psi(t)\rangle = U(t)|\psi(0)\rangle \). When the Hamiltonian \( H \) does not depend on time, \( U(t) = e^{-iHt/\hbar} \). Unitary maps preserve inner products, so probabilities keep summing to one. Their linearity is also what underlies the No-Cloning Theorem.

Joint systems are described by the tensor product of their individual spaces. For \( n \) qubits, the quantum state space is \( (\mathbb{C}^2)^{\otimes n} \), with dimension \( 2^n \). A register of \( n \) qudits with \( d \) levels each has dimension \( d^n \).

A classical simulation that stores every amplitude needs \( 2^n \) complex numbers. At 16 bytes each, 50 qubits already requires about 18 petabytes. That is the concrete meaning of "exponential": the cost of writing the state down, not a guaranteed speedup.

• Readout is small. By Holevo's 1973 bound, an ensemble of states on an \( n \)-qubit system, measured without entanglement assistance or other side information, has at most \( n \) bits of accessible classical information. A measurement returns one outcome, not the whole state vector.
• Most of the space is out of reach. Poulin and co-authors showed in 2011 that states generated from a fiducial state by polynomial-time evolution under local many-body Hamiltonians with bounded local dimension occupy an exponentially small volume in Hilbert space.
• Low entanglement can be compressed. Tensor networks represent weakly entangled states with far fewer parameters than \( 2^n \), which is why some large systems stay classically tractable.
• Algorithms must be designed. A big space has to be steered so that wrong answers cancel and right ones reinforce. Quantum machine learning proposals that map data into Hilbert space (feature maps) state a hypothesis about usefulness, not a guarantee.

In an array of neutral atoms, each atom used as a two-level system contributes a factor of two, so \( N \) atoms span a \( 2^N \)-dimensional space. In an analog quantum simulation such as QuEra's Aquila (available on Amazon Braket since 2022), the atoms evolve under a Hamiltonian set by the programmed laser pulses and atom positions, and that evolution takes place in this space.

The Rydberg blockade shows that the full \( 2^N \) is not always the relevant number. In the ideal blockade limit, configurations with adjacent excitations are excluded and the dynamics are restricted to the corresponding constrained subspace; for finite blockade they are energetically penalized rather than strictly forbidden. For a chain with an ideal nearest-neighbor blockade, the allowed configurations are counted by Fibonacci numbers, growing as roughly \( 1.618^N \). Knowing which subspace a protocol occupies matters for interpreting results and for judging how hard a classical simulation would be.

A pure qubit state has the form \( \alpha|0\rangle + \beta|1\rangle \) with complex amplitudes and \( |\alpha|^2 + |\beta|^2 = 1 \). After removing normalization and global phase, two real parameters remain, which the Bloch sphere displays.

No. The state carries amplitudes across many basis states, but a measurement returns a single outcome, and Holevo's theorem caps the accessible information from an ensemble on an \( n \)-qubit system at \( n \) bits, absent entanglement assistance or other side information. Useful algorithms rely on interference to make the desired outcome likely.

• Hilbert space is a complete complex inner-product space whose rays, represented by unit vectors, describe pure quantum states; a qubit is modeled by \( \mathbb{C}^2 \).
• Joint systems combine by tensor product, so the dimension for n qubits is \( 2^n \), which is why exact classical simulation becomes costly.
• A large Hilbert space does not by itself give speedups: readout is limited, most states are unreachable by efficient local dynamics from simple initial states, and algorithms must use interference.
• In neutral-atom arrays, effects such as the Rydberg blockade confine dynamics to a smaller subspace than the full \( 2^N \).