Coherent state

A coherent state is a translated base state of the harmonic oscillator. Coherent states are interesting because they behave classically in some ways.

αˉ=Tα0=eiαp^/0. \begin{align*} \ket{\bar \alpha} = T_{\alpha} \ket{0} = e^{-i \alpha \hat p /\hbar} \ket 0. \end{align*}

Consider the expectation values.

αˉxαˉ=ααˉpαˉ=0TαpTα0=0pTαTα0=0p0=0αˉHαˉ=0Tα(12mp2+12mω2x2)Tα0=012mp2+mω22(TαxTαTαxTα)0=012mp2+mω22(x+α)20=0H0+mω2α0x0+mω22x0200=12ω+12mω2α2. \begin{align*} \braket{\bar\alpha | x | \bar\alpha} &= \alpha \\ \braket{\bar\alpha | p | \bar\alpha} &= \braket{0 | T_\alpha\adj p T_\alpha |0} = \braket{0 | p T_\alpha\adj T_\alpha |0} = \braket{0|p|0} = 0 \\ \braket{\bar\alpha | H | \bar\alpha} &= \braket{0 | T_\alpha\adj \left( \frac{1}{2m} p^2 + \frac{1}{2} m\omega^2 x^2 \right) T_\alpha |0} \\ &= \braket{0| \frac{1}{2m} p^2 + \frac{m\omega^2}{2} \left( T_\alpha\adj x T_\alpha T_\alpha\adj x T_\alpha \right) | 0}\\ &= \braket{0 | \frac{1}{2m} p^2 + \frac{m\omega^2}{2} (x + \alpha)^2 |0} \\ &= \braket{0|H|0} + m \omega^2 \alpha \braket{0 | x | 0} + \frac{m\omega^2}{2} x_0^2 \braket{0|0} \\ &= \frac{1}{2} \hbar\omega + \frac{1}{2} m\omega^2 \alpha^2. \end{align*}

This looks like the classical result, with the addition of the zero-point energy term! We also get an interesting result from the Heisenberg time evolution.

αˉxHαˉ=αˉ+xScosωt+1mωpSsinωtαˉ=αcosωtαˉpHαˉ=αˉpScosωtmωxSsinωtαˉ=mωαsinωt. \begin{align*} \braket{\bar\alpha | x_H | \bar\alpha} &= \braket{\bar\alpha + x_S \cos \omega t + \frac{1}{m\omega} p_S \sin\omega t|\bar\alpha} \\ &= \alpha \cos\omega t \\ \braket{\bar\alpha | p_H | \bar\alpha} &= \braket{\bar\alpha | p_S \cos\omega t - m\omega x_S \sin\omega t | \bar\alpha} \\ &= -m\omega \alpha \sin\omega t. \end{align*}

We see the expectation values oscillate and pH=ddtxH \braket{p_H} = \frac{d }{d t} \braket{x_H}. Now consider the uncertainties

αˉx2αˉ=0(x+α)20=0x20+α200=2mω+α2αˉp2αˉ=0p20=mω2ΔxΔp=2. \begin{align*} \braket{\bar \alpha | x^2 |\bar\alpha} &= \braket{0 | (x + \alpha)^2 | 0} \\ &= \braket{0 | x^2 | 0} + \alpha^2 \braket{0|0} \\ &= \frac{\hbar}{2m\omega} + \alpha^2 \\ \braket{\bar \alpha | p^2 | \bar \alpha} &= \braket{0 | p^2 | 0} = \frac{m\hbar\omega}{2} \\ \Delta x \Delta p &= \frac{\hbar}{2}. \end{align*}

We can repeat the exercise for xHx_H and pHp_H and see that the uncertainty bound is saturated for all time. This is interesting, it means that the wavepacket remains coherent as it oscillates over time. Again, this looks classical.

We can write a coherent state in terms of the energy eigenstates it’s composed of. First write TαT_\alpha in terms of the raising and lowering operators. We apply the Baker-Campbell-Hausdorff formula and use the fact that [a,a][a\adj,a] is a constant.

Tα=eiαp/=eα(aa)/2d where d=mω=eαa/2deαa/2deα2/4d2. \begin{align*} T_\alpha &= e^{-i \alpha p/\hbar} = e^{\alpha (a\adj - a) / \sqrt 2 d} \text{ where } d = \sqrt{\frac{\hbar}{m\omega}} \\ &= e^{\alpha a\adj/\sqrt 2 d} e^{-\alpha a / \sqrt 2 d} e^{-\alpha^2 / 4d^2}. \end{align*}

Then expand a coherent state in terms of this operator. Looking at the last exponential, we notice that a0=0a \ket 0 = 0, so all but the first term of the Taylor expansion will be annihilated.

αˉ=eα2/4d2eαa/2deαa/2d0=eα2/4d2n1n!(α2d)n(a)n0=eα2/4d2n1n!(α2d)nn=ncnn. \begin{align*} \ket{\bar \alpha} &= e^{-\alpha^2 / 4d^2} e^{\alpha a\adj / \sqrt 2d} e^{-\alpha a / \sqrt 2d} \ket 0 \\ &= e^{-\alpha^2/4d^2} \sum_n \frac{1}{n!} \left( \frac{\alpha}{\sqrt 2d} \right) ^n \left( a\adj \right)^n \ket 0 \\ &= e^{-\alpha^2 / 4d^2} \sum_n \frac{1}{\sqrt{n!}} \left( \frac{\alpha}{\sqrt 2d} \right) ^n \ket n \\ &= \sum_n c_n \ket n. \end{align*}

The terms cn2|c_n|^2 form a Poisson distribution with mean λ=α2/2d2\lambda = \alpha^2 / 2d^2. Immediately we see N=λ \braket{N} = \lambda and ΔN=λ\Delta N = \sqrt \lambda. Since H=(N+12)ωH = (N + \frac 12) \hbar\omega, we know ΔH=ωλ\Delta H = \hbar \omega \sqrt \lambda.

So far in this derivation we have set αR\alpha \in \mathbb R and used the translation operator. We can generate a coherent state to allow αC\alpha \in \mathbb C using the displacement operator to define

αˉ=D(α)0=eαaαa0. \begin{align*} \coh \alpha = D(\alpha) \ket 0 = e^{\alpha a\adj - \alpha\conj a} \ket 0. \end{align*}

Another way to define coherent states is as eigenstates of the lowering operator

aαˉ=aeαaαa0=[a,eαaαa]0,since a0=0=[a,αaαa]eαaαa0,since [a,a] constant=eαaαa0aαˉ=ααˉ. \begin{align*} a\coh\alpha &= a e^{\alpha a\adj - \alpha\conj a} \ket 0 \\ &= [a,e^{\alpha a\adj - \alpha\conj a}] \ket 0, && \text{since } a \ket 0 = 0 \\ &= [a, \alpha a\adj - \alpha\conj a] e^{\alpha a\adj - \alpha\conj a} \ket 0, && \text{since } [a,a\adj] \text{ constant} \\ &= e^{\alpha a\adj - \alpha\conj a} \ket 0 \\ a\coh \alpha &= \alpha \coh \alpha. \end{align*}

We see complex α\alpha is allowed since aa is not hermitian. We can interpret the real and imaginary parts physically by considering the position and momentum of a generalized coherent state.

αˉxαˉ=d2αˉa+aαˉ=d2(α+α)αˉαˉ=2d[α],αˉpαˉ=i2dαˉaaαˉ=2d[α]. \begin{align*} \braket{\bar\alpha | x | \bar\alpha} &= \frac{d}{\sqrt 2} \braket{\bar \alpha | a\adj + a | \bar\alpha} \\ &= \frac{d}{\sqrt 2} (\alpha\conj + \alpha) \braket{\bar \alpha | \bar\alpha} \\ &= \sqrt 2 d\, \Re[\alpha], \\ \braket{\bar\alpha | p | \bar\alpha} &= \frac{i\hbar}{\sqrt 2d} \braket{\bar\alpha|a\adj-a|\bar\alpha} = \sqrt 2\frac{\hbar}{d} \Im[\alpha]. \end{align*}

The real component of α\alpha corresponds to a translation in xx, and the complex component to a translation in pp. In position space, we see (where d=h/mωd = \sqrt{h/m\omega})

αaαa=(x02d+ip0d2)12d(xipmω)(x02dip0d2)12d(x+ipmω)=id(p0dxx0pdmω)=i(x0pp0x). \begin{align*} \alpha a\adj - \alpha\conj a &= \left( \frac{x_0}{\sqrt 2d} + i \frac{p_0d}{\sqrt 2\hbar} \right) \frac{1}{\sqrt 2d} \left( x-i \frac{p}{m\omega} \right) - \left( \frac{x_0}{\sqrt 2d} - i \frac{p_0d}{\sqrt 2\hbar} \right) \frac{1}{\sqrt 2d} \left( x + i \frac{p}{m\omega} \right) \\ &= \frac{i}{d\hbar} \left( p_0 d x - \frac{x_0 p}{dm\omega} \right) \\ &= \frac{i}{\hbar} (x_0 p - p_0 x). \end{align*}

Now consider the Heisenberg time evolution.

DH(α,t)=eiHt/DS(α)eiHt/=eαaH(t)αaH(t)=exp(αeiωtaSαeiωtaS)=DS(αeiωt). \begin{align*} D_H(\alpha,-t) &= e^{-i Ht/\hbar} D_S(\alpha) e^{iHt/\hbar} \\ &= e^{\alpha a_H\adj(-t) - \alpha\conj a_H(-t)} \\ &= \exp \left( \alpha e^{-i\omega t} a_S\adj - \alpha\conj e^{i\omega_t} a_S \right) \\ &= D_S(\alpha e^{-i\omega t}). \end{align*}

Then apply this to the Schrödinger time evolution of the coherent state

αˉ,t=eiHt/DS(α)0=eiHt/DS(α)eiHt/eiHt/0=DH(α,t)eiωt/20=eiωt/2DS(αeiωt)0=eiωt/2αeiωt. \begin{align*} \ket{\bar\alpha,t} &= e^{-iHt/\hbar} D_S(\alpha) \ket 0 \\ &= e^{-iHt/\hbar} D_S(\alpha) e^{iHt/\hbar} e^{-iHt/\hbar} \ket 0 \\ &= D_H(\alpha,-t) e^{-i\omega t/2} \ket 0\\ &= e^{-i\omega t/2} D_S(\alpha e^{-i\omega t}) \ket 0 \\ &= e^{-i\omega t/2} \ket{\overline{\alpha e^{-i \omega t}}}. \end{align*}

We can also show that coherent states form an (overcomplete) basis. Each coherent state is a linear combination of every other other coherent state – no two states are orthogonal. This makes sense intuitively, every gaussian contains some component of every other gaussian. We can thus write a number state in terms of coherent states.

The overcompleteness is interesting. Whereas there are countably infinite number states, there are uncountable coherent states. This is a possible consequence of coherent states being eigenstates of a non-hermitian operator.