Algorithms
Ansatz
A proposed, parameterized quantum circuit structure used as a starting template in variational quantum algorithms like VQE.
Definition
An ansatz (German for 'approach' or 'starting point') is a chosen quantum circuit structure with adjustable parameters, used as the trainable template in variational quantum algorithms. The classical optimizer in algorithms like VQE adjusts these parameters to find the best-performing version of the circuit.
In depth
An ansatz is the family of quantum circuits that a variational algorithm searches over. The word is German for approach or starting point. A parameterized ansatz is a circuit template with adjustable angles, and training means finding the angle values that make the circuit prepare a good state. The choice of ansatz decides what the algorithm can represent: a state outside its reach can never be found, however well the angles are tuned, and an ansatz that is too flexible can be impossible to train.
Ansatzes fall into two broad types. Problem-inspired ansatzes build in knowledge of the task: unitary coupled cluster for chemistry follows the structure of molecular wave functions, and the QAOA ansatz alternates cost and mixer layers. Hardware-efficient ansatzes, introduced by Kandala and colleagues at IBM in 2017, use layers of the device's native gates and are shallow and easy to run, but they may not match the problem and can suffer barren plateaus. Adaptive methods, such as ADAPT-VQE of Grimsley and coauthors in 2019, grow the circuit one operator at a time, choosing at each step the operator with the largest energy gradient.
Designers balance expressibility, trainability, and cost. More layers and parameters make the ansatz more expressive, but deeper circuits are noisier and have flatter landscapes. Techniques to improve results include preserving symmetries of the problem, such as particle number, to restrict the search to valid states, initializing from a classical solution such as Hartree-Fock, and pruning parameters. The ansatz is where the physics of the problem meets the limits of the hardware.
Technical Definition
Different ansatz choices (such as hardware-efficient ansätze or chemically-inspired ansätze like UCCSD) trade off between expressibility (ability to represent the desired target state), circuit depth, and trainability — including susceptibility to barren plateaus.
Visual Explanation: An Analogy
Think of an ansatz like a recipe template with some ingredient amounts left as variables to be tuned — the basic structure (mix flour, add liquid, bake) is fixed, but the exact quantities are adjusted through experimentation to get the best result.
Key facts
| Item | Detail |
|---|---|
| Meaning | A parameterized circuit family searched by a variational algorithm |
| Problem-inspired | Unitary coupled cluster for chemistry, QAOA layers for optimization |
| Hardware-efficient | Native-gate layers, introduced by Kandala et al. in 2017 |
| Adaptive | ADAPT-VQE grows the circuit step by step (2019) |
| Trade-offs | Expressibility versus trainability and circuit depth |
Where the idea comes from
The ansatz concept was part of VQE from 2014. Hardware-efficient ansatzes in 2017 and adaptive methods in 2019 expanded the options, and barren plateau research from 2018 shaped how ansatzes are chosen.
Real-World Use Cases
- The starting structure for every VQE and QAOA implementation
- Choice of ansatz significantly affects whether an algorithm converges to a good solution
How it differs from related ideas
- A hardware-efficient ansatz is easy to run but may be unstructured, while a chemistry-inspired ansatz is accurate but deep.
- The ansatz is the model, and the optimizer is the training procedure.
- A good ansatz includes the symmetries of the problem.
Frequently asked questions
What is an ansatz?
A parameterized quantum circuit template whose angles are tuned by a classical optimizer in a variational algorithm.
How do I choose an ansatz?
Use problem-specific structure and symmetries where possible, and keep the circuit shallow enough for the hardware and trainable.
What is ADAPT-VQE?
A method that builds the ansatz one operator at a time, picking operators that most reduce the energy.
Related Terms
Keep exploring
- Algorithm: Variational Quantum Eigensolver (VQE)
- Algorithm: Quantum Approximate Optimization Algorithm (QAOA)
- Hardware: IBM Heron (r1, r2, and r3)