A composite pulse replaces a single rotation by a short sequence of rotations about different axes, chosen so that systematic errors cancel. An amplitude miscalibration or a detuning then barely affects the final operation, and the sequence works without knowing how large the error is. Composite pulses were developed in NMR in the late 1970s and 1980s [1, 2]. Today they are routine in trapped ions, neutral atoms, spins and superconducting qubits as a simple, analytic form of robust control.

The idea

Write \(\theta_\phi\) for a rotation by angle \(\theta\) about an axis at angle \(\phi\) in the \(xy\)-plane. Two systematic errors dominate:

  • Pulse-length (amplitude) errors — every rotation angle is scaled, \(\theta \to \theta(1+\epsilon)\);
  • Off-resonance errors — a detuning tilts the rotation axis out of the \(xy\)-plane.

A composite pulse chooses angles and phases so that the leading error terms of the individual rotations cancel in the product. The first example — inverting a spin with \(90_x\,180_y\,90_x\) instead of \(180_x\) — is far less sensitive to amplitude errors than a single pulse [1].

Classic sequences

BB1 corrects amplitude errors [3]. A target rotation \(\theta_0\) is complemented by

$$\theta_0 \;\cdot\; \pi_{\phi}\; 2\pi_{3\phi}\; \pi_{\phi}, \qquad \phi = \arccos\!\Big(-\frac{\theta}{4\pi}\Big),$$

which leaves an error of order \(\epsilon^3\) for any initial state. The correction block can be placed before, after, or in the middle of the main pulse.

CORPSE corrects off-resonance errors and SCROFULOUS amplitude errors. Together with BB1, these families were introduced to quantum computing as fully compensating rotations, which work for any input state [4].

Arbitrary order. Systematic constructions cancel errors to any order \(\epsilon^n\), independently of the initial state and without knowing \(\epsilon\) [5]. Short sequences tailored to a single gate — for example robust NOT gates — trade generality for duration [6].

Point-to-point versus universal

Some composite pulses only need to work for one initial state, such as inversion from \(|0\rangle\) to \(|1\rangle\). These point-to-point pulses can be shorter and more robust. Universal (fully compensating) rotations must work for every input state, which quantum gates require. Mixing the two up is a common source of disappointing results.

Beyond hard pulses

  • Smooth composite pulses. Sequences of shaped instead of rectangular pulses keep robustness while reducing bandwidth [7]. A single shaped pulse with a time-dependent phase can be viewed as the continuous limit of a composite sequence [8].
  • Robust dynamical decoupling. Composite-pulse ideas make decoupling sequences robust to their own pulse errors. In universally robust sequences the number of pulses grows only linearly with the order of compensation [9]; see Dynamical decoupling.
  • Numerical design. Optimizing a pulse over a range of error parameters (Robust & ensemble optimal control) can be seen as an unrestricted search for “composite” solutions. It often finds shorter or more robust pulses than analytic families, at the cost of interpretability.

Strengths and limitations

Strengths

  • Analytic, simple to implement and calibrate, independent of the size of the error.
  • Robust for any initial state (universal sequences), compatible with standard hardware.

Limitations

  • Longer than a single pulse, so more exposed to decoherence and time-dependent noise; see Filter functions & noise-tailored control.
  • Each family targets specific error types. Simultaneous robustness against several errors needs longer sequences or numerical design.