8
$\begingroup$

Following from this question, I tried to look at the cited article in order to simulate and solve that same problem... without success. Mainly, I still fail to understand how the authors managed to simulate the Hamiltonian evolution through the circuit shown at the bottom of Fig.4. Even exponentiating classically the matrix I do not get values of the gates shown in the Quirk circuit that @Blue linked along his question.

I tried to look at the paper in which the Group Leader Optimization algorithm is explained, but I still have troubles understanding how do they assign the rotation angles to the different gates.

$\endgroup$
  • 1
    $\begingroup$ Good question! Some of those rotation angle values are definitely wrong. @Nelimee tried them out in his program. Have a look at optim_hamil.py. A practical way to get the correct rotation angle values/coefficients is to use some sort of multivariate optimization algorithm. Nelimee used the scipy.optimize module for that purpose. However, I would like to personally understand the Group Leader Optimization Algorithm properly too. The paper: arxiv.org/abs/1004.2242 is too vague! $\endgroup$ – Sanchayan Dutta Jul 20 '18 at 12:19
  • $\begingroup$ The point is that from what I understood they are using genetic algorithms to find the optimal configuration to represent a given quantum subroutine, which already strikes me as odd honestly. $\endgroup$ – FSic Jul 20 '18 at 12:26
  • 1
    $\begingroup$ You can recover the matrices exponentials with this script. And you understood correctly, they use a genetic algorithm (which is a kind of optimisation algorithm) to find a good (and not the optimal) gate decomposition that implements a unitary matrix that is close to the one we want. $\endgroup$ – Nelimee Jul 20 '18 at 12:30
  • $\begingroup$ Thanks for the script! However, with this you obtain directly the values of the exponentiated matrix, which can be directly implemented in Quirk, I was rather trying to understan how they manage to simulate such matrix with the gates they provided in the figure, because I also tried to compute the circuit to see the final matrix, but I end up having something totally different. I do not exclude to have made some miscalculations though. $\endgroup$ – FSic Jul 20 '18 at 12:52
5
$\begingroup$

I don't know why/how the authors of that paper do what they do. However, here's how I'd go about it for this special case (and it is a very special case):

You can write the Hamiltonian as a Pauli decomposition $$ A=15\mathbb{I}\otimes\mathbb{I}+9Z\otimes X+5X\otimes Z-3Y\otimes Y. $$ Update: It should be $+3Y\otimes Y$. But I don't want to redraw all my diagrams etc., so I'll leave the negative sign.

Now, it is interesting to note that every one of these terms commutes. So, that means that $$ e^{iA\theta}=e^{15i\theta}e^{9i\theta Z\otimes X}e^{5i\theta X\otimes Z}e^{-3i\theta Y\otimes Y}. $$ You could work out how to simulate each of these steps individually, but let me make one further observation first: these commuting terms are the stabilizers of the 2-qubit cluster state. That may or may not mean anything to you, but it tells me that a smart thing to do is apply a controlled-phase gate. $$ CP\cdot A\cdot CP=15\mathbb{I}\otimes\mathbb{I}+9\mathbb{I}\otimes X+5X\otimes \mathbb{I}-3X\otimes X. $$ (You may want to check the sign of the last term. I didn't compute it carefully.) So, if we start and end our sequence with controlled-phase gates, then 2 of the terms are easy to get right: we rotate the first qubit about the $x$ axis by an angle $5\theta$, and the second qubit about the $x$ axis by an angle $9\theta$.

The only thing we are left to get right is the $X\otimes X$ rotation. If you think about the structure of $e^{-3i\theta X\otimes X}$, this is like an $x$ rotation on the basis states $\{|00\rangle,|11\rangle\}$, and another one on $\{|01\rangle,|10\rangle\}$. A controlled-not converts these bases into the single-qubit bases controlled off the target qubit. But since both implement the same controlled-rotation but controlled off opposite values, we can just remove the control. Thus, the overall circuit is: enter image description here This simplifies slightly by combining the two controlled-gates at the end: enter image description here Note that I have not included the global phase term here because that's not the sensible way to do it. When you make controlled-($e^{iA\theta}$), you apply the "global phase" as a $z$ rotation on the control qubit.

$\endgroup$
  • $\begingroup$ Hi, do you know any textbook or any paper which discusses the general algorithm for decomposing the Hamiltonian into Pauli gates? I only found this website, but it doesn't explain the logic behind the decomposition nor the general algorithm for larger matrices (if it exists). $\endgroup$ – Sanchayan Dutta Oct 19 '18 at 6:30
  • 2
    $\begingroup$ @Blue The simplest method is based on a Trotter-Suzuki decomposition. Those key words should hopefully provide a fruitful search. However, there are newer, much more efficient, algorithms. I'll need to dig out a reference... $\endgroup$ – DaftWullie Oct 19 '18 at 6:46
  • 2
    $\begingroup$ @Blue Here's a talk that gives a fair introduction+overview, and contains links to some of the relevant papers: bristol.ac.uk/media-library/sites/maths/events/2015/… $\endgroup$ – DaftWullie Oct 19 '18 at 6:50
  • $\begingroup$ I was a bit confused by the second Pauli decomposition and asked a question regarding that here. If you get time, do please consider answering it! $\endgroup$ – Sanchayan Dutta Oct 19 '18 at 11:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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