めもめも

このブログに記載の内容は個人の見解であり、必ずしも所属組織の立場、戦略、意見を代表するものではありません。

CIMが基底状態を実現するメカニズム



一般的なアニーリングでは、安定的なスピンの重ね合わせ状態が最初にあって、それを基底状態に向かって状態を変化(収束)させていきます。

CIMは、光パルスを利用することで、パルスのエネルギーを徐々に増やして、「不安定なスピンの状態(そもそも光パルスでスピンをエミュレートできていない状態)」から「安定的なスピンの状態(光パルスでスピンを安定的にエミュレートできる状態)」が構成されるタイミングで、うまいこと、「イジングハミルトニアンの基底状態だけで安定的なスピンが構成される」という絶妙な状況を作り出していると考えられます。


以下、私見。

FPGA を用いた結合では、スピン間の Entanglement は存在しないのはその通りですが、個々のパルスレベルで安定化するタイミングで全体の最適化状態が選べるというのは、それなりに絶妙な仕組みかと。逆の見方をすると、Entanglement を考えずにいい感じに最適化状態が選べるということは、個々のパルスの量子状態をデジタルコンピューターで数値シミュレーションすれば、Entanglement を利用していない、すなわち、デジタルコンピューターでシミュレーションできる手法でありながら、従来のデジタルコンピューターよりもよい性能を出す新しいアルゴリズムが見つかったりしませんかね?


ーーーーー
追記(2018/06/25)

とかとか考えていたら、このような論文が発表されました。

[1806.08422] Emulating the coherent Ising machine with a mean-field algorithm

不安定な光パルスの状態を温度Tのボルツマン平均で表現することで、CIMにおける光パルスの成長を現象論的に再現することに成功したようです。

Calculations on Coherent States and Squeezed States (4)

Quantum Optics

Quantum Optics

This is intended to fill some gaps in calculation details in the book above.

5.1.1 Degenerate Parametric Amplifier

Preparation: basics of the interaction picture

Suppose that the Hamiltonian consists of the time-independent part H_0 plus the time-dependent interaction part V(t).

 H=H_0+V(t)

Define the state vector and the interaction term in the interaction picture in conjunction with the ones in the Schroedinger picture

 \displaystyle{\mid\Psi_I(t)\rangle} := e^{\frac{i}{\hbar}H_0t}{\mid\Psi(t)\rangle}

 \displaystyle V_I(t) := e^{\frac{i}{\hbar}H_0t}V(t)e^{-\frac{i}{\hbar}H_0t}

You can prove that they follow the equation of motion below.

 \displaystyle \frac{d}{dt}{\mid\Psi_I(t)\rangle} = -\frac{i}{\hbar}V_I(t){\mid\Psi_I(t)\rangle}

To calculate the expectation value for an operator O, define the operator in the interaction picture and combine it with the state in the interaction picture.

 \displaystyle O_I(t) := e^{\frac{i}{\hbar}H_0t}Oe^{-\frac{i}{\hbar}H_0t}

 \displaystyle \langle O(t)\rangle = {\langle\Psi_I(t)\mid}O_I(t){\mid\Psi_I(t)\rangle}

Since the operator in the interaction picture is the same as the one in the Hisenberg picture with H_0, it follows the equation of motion below.

 \displaystyle \frac{d}{dt}O_I(t) = \frac{i}{\hbar}[H_0, O_I(t)]

In the special case, when the interaction operator V_I(t) and your operators O_I(t) in the interaction picture are time-independent, you have the following relationships.

 \displaystyle{\mid\Psi_I(t)\rangle} = e^{-\frac{i}{\hbar}V_It}{\mid\Psi(0)\rangle}

 \displaystyle \langle O(t)\rangle = {\langle\Psi(0)\mid}e^{\frac{i}{\hbar}V_It}O_Ie^{-\frac{i}{\hbar}V_It}{\mid\Psi(0)\rangle} --- (4.1)

Especially, the latter means that the operator O in the Heisenberg picture is obtained as

 \displaystyle O_H(t) = e^{\frac{i}{\hbar}V_It}O_Ie^{-\frac{i}{\hbar}V_It} --- (4.2)

Hence it follows the equations of motion below.

 \displaystyle \frac{d}{dt}O_H(t) = \frac{i}{\hbar}\left[V_I, O_H(t) \right] --- (4.3)

Degenerate Parametric Amplifier

 \displaystyle H_0=\hbar\omega a^\dagger a

 \displaystyle V(t) = -\frac{i\hbar}{2}\chi(a^2e^{2i\omega t}-a^{\dagger 2}e^{-2i\omega t})

Using the relationship (1.1), you can prove that:

 \displaystyle V_I(t) = -\frac{i\hbar}{2}\chi(a^2-a^{\dagger 2})

So this is a special case of the time-independent V_I.

Calculate a,\,a^\dagger in the interaction picture.

 \displaystyle \frac{d}{dt}a_I(t) = \frac{i}{\hbar}[H_0,a_I(t)] = e^{\frac{i}{\hbar}H_0t}  \frac{i}{\hbar}[H_0,a]   e^{-\frac{i}{\hbar}H_0t} = -i\omega a_I(t)

 \displaystyle \frac{d}{dt}a^\dagger_I(t) = \frac{i}{\hbar}[H_0,a^\dagger_I(t)] = e^{\frac{i}{\hbar}H_0t}  \frac{i}{\hbar}[H_0,a^\dagger]   e^{\frac{i}{\hbar}H_0t} = i\omega a^\dagger_I(t)

Hence,

 a_I(t) = ae^{-i\omega t},\ a^\dagger_I(t) = a^\dagger e^{i\omega t}

Now we define the rotated coordinates at a fixed time t=t_0:

 Y_1+iY_2 = (X_1+iX_2)e^{i\omega t_0} = 2ae^{i\omega t_0}

Convert it to the interaction picture and set t=t_0.

 Y_{1I}(t_0)+iY_{2I}(t_0) = 2a_I(t_0)e^{i\omega t_0} = 2a = X_1+iX_2

This means that the (time-independent) operators X_1=a+a^\dagger and X_2=-i(a-a^\dagger) represent the rotated coordinates at t=t_0 in the interaction picture.

Now you can use (4.2) and (4.3) to figure out the rotated coordinates in the Heisenberg picture to understand the physical effects on the initial state {\mid\Psi_S(0)\rangle}.

 \displaystyle \frac{d}{dt}X_{1H}(t) = \frac{i}{\hbar}\left[V_I,X_{1H}(t)\right] = e^{\frac{i}{\hbar}V_It} \frac{i}{\hbar}\left[V_I,X_1\right]e^{-\frac{i}{\hbar}V_It}

  \displaystyle= e^{\frac{i}{\hbar}V_It}\frac{1}{2}\chi\left[a^2-a^{\dagger 2},a+a^\dagger\right]e^{-\frac{i}{\hbar}V_It}=e^{\frac{i}{\hbar}V_It} 2(a+a^\dagger) e^{-\frac{i}{\hbar}V_It}=\chi X_{1H}

Likewise,

 \displaystyle \frac{d}{dt}X_{2H}(t) = -\chi X_{2H}

Hence,

 \displaystyle X_{1H}(t) = e^{\chi t} X_1

 \displaystyle X_{2H}(t) = e^{-\chi t} X_2

So you can see the squeezing effect on the rotated coordinates.

Since X_1,\,X_2 are the rotated coordinates in the interaction picture, and they are time-independent, you can apply (4.1) to calculate the expectation value as

 \langle X_1\rangle = {\langle\Psi(0)\mid}e^{\frac{i}{\hbar}V_It}X_1e^{-\frac{i}{\hbar}V_It}{\mid\Psi(0)\rangle}

 \langle X_2\rangle = {\langle\Psi(0)\mid}e^{\frac{i}{\hbar}V_It}X_2e^{-\frac{i}{\hbar}V_It}{\mid\Psi(0)\rangle}

And now you can see that the time evolution \displaystyle e^{-\frac{i}{\hbar}V_It} has the same effect as the squeeze operator S(-\chi t)

 \displaystyle e^{-\frac{i}{\hbar}V_It} = e^{-\frac{\chi t}{2}(a^2-a^{\dagger 2})} = S(-\chi t)

This explains the squeezing effect obtained above.