Rényi entropy of magic

Lorenzo Leone, Salvatore F. E. Oliviero, Alioscia Hamma (or view on SciRate)

Main ideas

Measures of magic

Renyi entropy

Consider \(\langle \psi | P | \psi \rangle ^2\) that’s normalized to be a probability distribution.

Then define Renyi entropies (p log p)

The 2-entropy is bounded by log(d).

The Renyi entropy is between the min-relative entropy and the stabilizer nullity.

The main benefit is that it’s computable for various n.

Magic Power

How much “magic” is produced by unitary evolution?

Let’s consider the magic of a unitary (the magic generated from an average stabilizer state)

The linear magic power is zero iff the unitary commutes with a matrix related to the projector onto \(S_4\).

Most unitaries are concentrated in magic power around log(d/4).

The magic power lower-bounds the t-count (with high probability, \(t(U) = \Omega(n)\).

This gives a way to characterize t-counts as “how much extraclassical computation” you are doing. Three T gates are worth more than Toffoli gates.

Out-of-time-order correlation

We need order N gates to get 8-OTOC (related to quantum chaos). How much magic is needed to drive a quantum system towards chaos?

We can talk about a random Clifford circuit, “doped” by layers of non-clifford one-qubit gates. The magic power increases with the number of doping layers.

The Renyi entropy is directly related to the 2-OTOC functions.

tags: simulating-quantum-circuits