Trainability · Algorithms
Matchgate circuits have no barren plateau
Theorem statement
For the matchgate family whose dynamical Lie algebra is $\mathfrak{so}(2n)$ (with $n\ge 3$), the single-ideal loss variance $$\operatorname{Var}_\theta[\ell]=\frac{P_{\mathfrak g}(\rho)\,P_{\mathfrak g}(O)}{\dim\mathfrak g},\qquad \dim\mathfrak g=n(2n-1),$$ carries an algebra dimension that grows only polynomially in $n$. When the purity product $P_{\mathfrak g}(\rho)\,P_{\mathfrak g}(O)$ stays bounded below by an inverse polynomial in $n$, the variance is likewise inverse-polynomially bounded below and does not decay as $C\,b^{-n}$ with $b>1$, so the family has no barren plateau. The transverse-field Ising chain realizes this concretely: for the highest-weight state $\rho=|0\rangle\!\langle0|^{\otimes n}$ and a two-body Ising observable $O$ the purities are $P_{\mathfrak g}(\rho)=n/2^n$ and $P_{\mathfrak g}(O)=2^n$, giving the exact inverse-linear variance $\operatorname{Var}=1/(2n-1)$.
Sources
- Does provable absence of barren plateaus imply classical simulability? Or, why we need to rethink variational quantum computing
M. Cerezo, Martin Larocca, Diego Garcia-Martin, N. L. Diaz, Paolo Braccia, Enrico Fontana, Manuel S. Rudolph, Pablo Bermejo, Aroosa Ijaz, Supanut Thanasilp, Eric R. Anschuetz, Zoe Holmes, 2023
- Classification of dynamical Lie algebras of 2-local spin systems on linear, circular and fully connected topologies
Roeland Wiersema, Efekan Kökcü, Alexander F. Kemper, Bojko N. Bakalov, 2024
- A Lie Algebraic Theory of Barren Plateaus for Deep Parameterized Quantum Circuits
Michael Ragone, Bojko N. Bakalov, Frederic Sauvage, Alexander F. Kemper, Carlos Ortiz Marrero, Martin Larocca, M. Cerezo, 2023
Lean context
- Lean declaration
QuantumAlg.matchgateSO_polyDLA_family_dichotomy
Copy a short prompt with the import, theorem name, citations, and public source link.