3
$\begingroup$

We are interested in simulating the 1d Ising model Hamiltonian using a Quantum Circuit (QC). A similar question was posted before with no answers. Here we will assume, for simplicity, 3 lattice sites and $J=-1$.

Generically, the Hamiltonian is given as $$ H = -J \sum_{ij} \sigma_i^z \otimes \sigma_j^z.\tag{1} $$ For our case of interest this Hamiltonian becomes: $$ H = \sigma^z \otimes \sigma^z \otimes \mathbb{1} + \mathbb{1} \otimes \sigma^z \otimes \sigma^z.\tag{2} $$ Obviously I have not included any periodic boundary conditions. There are only three lattice sites so there are only two interaction terms. In what follows I will replace $\sigma^z$ with $Z$ implying the corresponding quantum gate.

The evolution operator corresponding to this Hamiltonian is given as $$\tag{3} U(t) = e^{-i (Z \otimes Z\otimes \mathbb{1} + \mathbb{1}\otimes Z \otimes Z)t}. $$ Should these operators not commute we would have to use the Trotter-Suzuki formula. However, they do commute and as a result there is no need to use it.

In each of the two summands there exists a unit operator which can be completely ignored from the circuit. Now, for the operator $Z \otimes Z$ the curcuit would read $$ \mathrm{CNOT} R_z(2t) \mathrm{CNOT}: $$ enter image description here

My question is whether as the generalization to the 3 lattice sites Hamiltonian is as simple forward as this: enter image description here

Of course the $R_Z$ gate runs for $2t$ according the unitary $U(t)$. Finally, does this generalize as simply to the $n$ lattice site Ising model?

$\endgroup$

1 Answer 1

2
$\begingroup$

If you know the circuit that corresponds to the unitary $U_{zz}$ for $Z_1Z_2$, then you also can also be sure that the same circuit will correspond to $U_{zz}$ for $Z_2 Z_3$, except that the circuit will apply to qubits 2 and 3 instead of to qubits 1 and 2. This generalizes to any two qubits in your $n$-site Ising model, for example $U_{zz}$ for $Z_4Z_{13}$ will have the same circuit as $U_{zz}$ for $Z_1Z_2$ except that the gates will apply to qubits 4 and 13 instead of 1 and 2.

The last remaining aspect of your question is perhaps themost interesting one. If you can factorize a unitary $U$ into $U=U_AU_B$, and you know the circuit that corresponds to $U_A$ and $U_B$, then yes you implement each circuit one after the other, and since in your case $U_A$ and $U_B$ commute, it doesn't even matter the order in which you implement them, meaning that you could also switched the order so that your sub-circuit for $U_{zz}$ on qubits 1 and 2 would be switched with your sub-circuit for $U_{zz}$ on qubits 2 and 3. This also generalizes to $n$-sites in your model, or $n$ different unitary operators and their corresponding $n$ sub-circuits!

If you're ever not sure about something like this, you can always double-check yourself in Octave-Online, for example try this code:

Z = [ 1 0 ; 0 -1];                % Define  Sigma Z
Z1 = kron(Z,eye(4));              % Define Z1 = Z \otimes I \otimes I
Z2 = kron(eye(2),kron(Z,eye(2))); % Define Z2 = I \otimes Z \otimes I
Z3 = kron(eye(4),Z);              % Define Z3 = I \otimes I \otimes Z
UA = expm(-1i*Z1*Z2);             % Define UA = exp(-i*Z1*Z2*I)
UB = expm(-1i*Z2*Z3)              % Define UB = exp(-i*I*Z2*Z3)
isequal(UA*UB,UB*UA)              % Check if UA and UB commute

The answer you'll get is 1, whcih means that $U_A$ and $U_B$ do multiplicatively commute, and therefore it's okay to switch $U_A$ and $U_B$.

You can also verify that this circuit will correctly implement $U_AU_B$ by defining the gates and circuits as follows:

CNOT = [1 0 0 0 ; 0 1 0 0 ; 0 0 0 1; 0 0 1 0 ];
CNOT_12 = kron(CNOT,eye(2));                    % CNOT on q1q2
CNOT_23=kron(eye(2),CNOT);                      % CNOT on q2q3
RZ_piBy2=[exp(-1i*pi/4) 0 ; 0 exp(1i*pi/4)]*2;  % Rz(pi/2) for 2 units of time
RZ_qubit2=kron(eye(2),kron(RZ_piBy2,eye(2)));   % Rz on q2
RZ_qubit3=kron(eye(4),RZ_piBy2);                % Rz on q3
CNOT_12*RZ_qubit2*CNOT_12*CNOT_23*RZ_qubit3*CNOT_23

You can also see some related questions on the topic:

$\endgroup$
2
  • $\begingroup$ But is this the optimal circuit? In the # of CNOTS for example or $R$s. $\endgroup$
    – Marion
    Commented Dec 31, 2021 at 13:53
  • $\begingroup$ That's a diffetent question. I'd recommend to post it separately since the question I answered asked nothing about optimality! $\endgroup$ Commented Dec 31, 2021 at 17:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.