9

In Trotterization, the typical Hamiltonian considered is:

$$ H = \sum_{p, q} h_{pq} a^{\dagger}_p a_q + \sum_{p, q, r, s} a^{\dagger}_p a^{\dagger}_q a_r a_s $$

Which is then converted into a sequence of gates by the Jordan Wigner transformation.

However, how do we choose the ordering of the $ pq$ and $pqrs$ terms? For example, if our Hamiltonian was:

$$ H = (a^{\dagger}_1 a_2 + a^{\dagger}_2 a_1) + (a^{\dagger}_2 a_3 + a^{\dagger}_3 a_2) + (a^{\dagger}_1 a^{\dagger}_2 a_2 a_3 + a^{\dagger}_3 a^{\dagger}_2 a_2 a_1)$$

There are at least $ 3! = 6$ ways to order the Hamiltonian, a number which quickly explodes as the size of the Hamiltonian grows.

Note that the terms don't have clean commutation rules - sometimes the terms will commute or anticommute. Is there an approach to ordering the Hamiltonian that maximizes simulation accuracy?

C. Kang
  • 1,716
  • 8
  • 23
  • @gIS you are correct; unfortunately, Trotterization exponentiates the matrices, which incurs errors because $ e^{A + B} \neq e^A e^B $ when $A, B$ don't commute – C. Kang Sep 01 '20 at 23:59
  • @Aman somewhat off topic, but I personally like Microsoft's docs – C. Kang Sep 01 '20 at 23:59
  • Your Hamiltonian needs to be Hermitian, which your example is currently not. – DaftWullie Sep 02 '20 at 06:51
  • @DaftWullie good point, let me revise the question to have it. I should note that I'm using molecular Hamiltonians, which are Hermitian in nature – C. Kang Sep 02 '20 at 14:46
  • Am I right in thinking that when you use Jordan Wigner, your particular example translates to a triangle with $XX+YY$ couplings between each pair of qubits? In which case, symmetry would suggest that ordering shouldn't make a difference. – DaftWullie Sep 02 '20 at 14:58
  • @DaftWullie ah, the Hamiltonian was chosen arbitrarily - in this case, consider an ansatz of just an electron in 1. If the 1/2 term was applied first, then the 2/3 term, both would be applied. However, if the 2/3 term was applied first, there would be no impact (as there is no electron yet in the 2 spot). – C. Kang Sep 02 '20 at 16:18

1 Answers1

1

The question of how to order the terms when Trotterizing is an active area of research. It turns out to be beneficial to put commuting terms next to each other. It can also be beneficial to randomise the ordering of terms in each Trotter step (as this suppresses errors), or to even not include certain low value terms. A paper that gives many references to the work that has been done in this area is the following: Compilation by stochastic Hamiltonian sparsification.

John
  • 472
  • 2
  • 5