Quantum Coherence

Mathematically, coherence lives in the off-diagonal elements of the density matrix. A pure superposition state |psi> = alpha|0> + beta|1> has a density matrix rho = |psi><psi| with off-diagonal terms alphabeta and beta*alpha that encode the phase relationship between the two amplitudes. When these off-diagonal elements vanish, rho becomes diagonal and the system is in a classical mixture: it is in state |0> with probability |alpha|^2 or |1> with probability |beta|^2, with no interference possible. Quantitative measures of coherence include the l1-norm of coherence (sum of absolute values of off-diagonal elements) and the relative entropy of coherence (distance from the diagonal state in information-theoretic terms). Both measures are zero for classical states and maximal for equal superpositions.

The practical limit coherence places on quantum hardware is captured by two timescales: T1, the energy relaxation time (how long before an excited qubit decays to its ground state), and T2, the dephasing time (how long before phase information is randomized). T2 is always at most 2*T1 and is often much shorter. For a quantum algorithm to succeed on a circuit of depth d, the total gate time must be well below T2. Modern superconducting qubits have T1 and T2 in the range of 100 microseconds to milliseconds; trapped ions reach milliseconds to seconds; some nuclear spin systems reach minutes. The ratio of T2 to gate time, often called the number of coherent operations, is the most direct figure of merit for near-term hardware.

Physical decoherence arises from multiple sources depending on the qubit modality. Superconducting qubits lose coherence through photon emission into transmission lines, coupling to two-level systems in dielectric materials, magnetic flux noise from surface spins, and phonon interactions with the substrate. Trapped ions decohere primarily through photon scattering from off-resonant laser light and magnetic field fluctuations that shift energy levels. Nitrogen-vacancy centers in diamond are affected by nearby 13C nuclear spins and surface charges. Understanding and mitigating these specific channels, through better materials, shielding, and dynamical decoupling pulse sequences, has driven roughly a thousandfold improvement in qubit coherence times over the past two decades.

Coherence is the resource that separates quantum computation from classical probabilistic computation. Without phase coherence between superposition branches, there is no interference, and without interference, there is no way to amplify correct answers while canceling wrong ones. Every known quantum speedup, from Shor’s factoring algorithm to Grover’s search to quantum simulation, relies on constructive interference among coherent paths. This is why error correction is necessary for large-scale quantum computation: it maintains artificial coherence by detecting and reversing the small incremental decoherence that occurs with each gate, effectively extending the logical coherence time far beyond the physical qubit’s natural limit.