19

In doing second order time-independent perturbation theory in non-relativistic quantum mechanics one has to calculate the overlap between states

$$E^{(2)}_n ~=~ \sum_{m \neq n}\frac{|\langle m | H' | n \rangle|^2}{E_n^{(0)}-E_m^{(0)}}$$

(where $E^{(k)}_n$ represents the kth order correction to the nth energy level).

For the Hydrogenic atom the spectrum of the Hamiltonian consists of a discrete set corresponding to "bound states" (the negative energy states) and a continuum of "scattering states" (the positive energy states).

Is there an example where the overlap between a bound state and the scattering states makes a measurable contribution to the energy in the perturbative regime? Should the scattering states be included in the perturbative calculations? Are there experimental results that backs this up? (For example, perhaps the electron-electron interaction in highly excited states of Helium).

As an aside, what "is" the Hilbert space of the Hydrogen atom in the position representation? I've often read the basis of eigenstates of the Hydrogen atom Hamiltonian isn't complete without the scattering states, but I've not seen any convincing argument of this. I've read that the radial bound states are dense in $L^2((0, \infty))$ (e.g. here), so including the scattering states the Hilbert space must strictly contain this.

3 Answers3

14

The scattering states must be included in the perturbative calculations if the result is to be highly accurate. In particular, it is not justified to ignore the continuous spectrum at energies close to the dissociation threshold.

The Hilbert space in the position representation is the space of square integrable functions on $R^3\setminus\{0\}$ with respect to the inner product $$\langle\phi|\psi\rangle:=\int \frac{dx}{|x|}\phi(x)^*\psi(x).$$ The bound states alone are not dense in this space.

For thorough treatments of the hydrogen spectrum see the books
G R. Gilmore, Lie Groups, Lie Algebras & Some of Their Applications, Wiley 1974, Dover, 2002. pp. 427-430
and
A.O. Barut and R. Raczka, Theory of group representations and applications, 2nd. ed., Warzawa 1980. Chapter 12.2.

8

Your main question was "Is there an example where the overlap between a bound state and the scattering states makes a measurable contribution to the energy in the perturbative regime?"

Actually, I disagree with the statement of the other answer that the scattering states must be included in the perturbative calculations only if the result is to be highly accurate. In fact, as for the hydrogen atom, if you take, as a simple example, the perturbative potential $\epsilon/r$, than the continuum in second order accounts for TWO THIRDS of the total contribution! If you take $\epsilon/r^2$, it is even 75%!

For other examples, see the following articles:

  1. Here you can see, that "the continuum contribution to the sum is sometimes quite large, accounting, in extreme cases, for all but a few percent of the total."

  2. For a textbook example, you can have a look at Schiff (p.263-265), where the Stark effect of hydrogen atom is worked out with the continuum contribution.

If you are interested, I can provide you with more material.

wondering
  • 788
5

I agree with the conclusion of user 'wondering', continuum states can indeed give a significant (often quite large) contribution to perturbation energies and wave functions. I would just like to give a few examples.

For the ground state of a perturbed hydrogen with Hamiltonian $$ H=-\frac{1}{2}\nabla^2-\frac{1}{r}+W\equiv H^{(0)}+W \ , $$ and $$ \psi^{(0)}_{100}(\vec{r})=2\exp(-r)Y_{00}(\theta,\phi) \ \ , \ \ E^{(0)}_1=-\frac{1}{2} \ , $$ the solution of the first-order Schrödinger equation $$ (H^{(0)}-E^{(0)}_1)|\psi^{(1)}\rangle+(W-E^{(1)})|\psi^{(0)}_{100}\rangle=0 $$ can be written as an expansion over all (bound and scattering) hydrogenic eigenstates: $$ \begin{aligned} |\psi^{(1)}\rangle &= \frac{{\cal{P}}^\perp}{E^{(0)}_1-H^{(0)}}W|\psi^{(0)}_{100}\rangle \\ &= \sum_{l=0}^{\infty}\sum_{m=-l}^{+l} \left[ \sum_{\substack{n=l+1 \\ n\neq1}}^\infty\frac{\langle\psi^{(0)}_{nlm}|W|\psi^{(0)}_{100}\rangle}{E^{(0)}_1-E^{(0)}_n}|\psi^{(0)}_{nlm}\rangle+\int_0^\infty\mathrm{d}\varepsilon\frac{\langle\psi_{lm}^{(0)}(\varepsilon)|W|\psi^{(0)}_{100}\rangle}{E^{(0)}_1-\varepsilon}|\psi_{lm}^{(0)}(\varepsilon)\rangle \right] \ , \end{aligned} $$ where ${\cal{P}}^\perp=I-|\psi^{(0)}_{100}\rangle\langle\psi^{(0)}_{100}|$ is projecting onto the orthogonal complement space of $|\psi^{(0)}_{100}\rangle$, and intermediate normalization was assumed, so $\langle\psi^{(0)}_{100}|\psi^{(1)}\rangle=0$ and ${\cal{P}}^\perp|\psi^{(1)}\rangle=|\psi^{(1)}\rangle$. Zeroth-order states are orthonormal in the following sense: $$ \begin{aligned} \langle\psi^{(0)}_{nlm}|\psi^{(0)}_{n'l'm'}\rangle&=\delta_{nn'}\delta_{ll'}\delta_{mm'} \ , \\ \langle\psi^{(0)}_{lm}(\varepsilon)|\psi^{(0)}_{l'm'}(\varepsilon')\rangle&=\delta(\varepsilon-\varepsilon')\delta_{ll'}\delta_{mm'} \ , \\ \langle\psi^{(0)}_{nlm}|\psi^{(0)}_{l'm'}(\varepsilon')\rangle&=0 \ . \end{aligned} $$ The continuum part of the zeroth-order spectrum affects second- and third-order energies via the relations $$ E^{(2)}=\langle\psi^{(0)}_{100}|W|\psi^{(1)}\rangle \ \ , \ \ E^{(3)}=\langle\psi^{(1)}|W|\psi^{(1)}\rangle-E^{(1)}\langle\psi^{(1)}|\psi^{(1)}\rangle \ . $$ In the following, examples for $W$ are considered where $\psi^{(1)}$, $E^{(2)}$ and $E^{(3)}$ can be given in a closed analytical form, which makes it possible to assess the importance of continuum states in the sum-over-states expansions for each case. The bound state part of $E^{(2)}$ (denoted by $E^{(2)}_\text{b}$) will be calculated numerically, and compared to the analytically known $E^{(2)}=E^{(2)}_\text{b}+E^{(2)}_\text{c}$. We take $\lambda\geq0$ in all cases below.

$\bullet$ Case 1: $W={\cal{F}}z$

This is just the textbook case of the quadratic Stark effect in the ground state of hydrogen (see e.g. Ch. 33 of Schiff). From symmetry considerations, it is obvious that $E^{(1)}=0$ and that $\psi^{(1)}$ must have $p_z$ symmetry. After substituting the Ansatz $\psi^{(1)}(\vec{r})=\chi(r)Y_{10}(\theta,\phi)$, the first-order Schrödinger equation is reduced to an ordinary differential equation which is easily solved. We obtain $$ \psi^{(1)}(\vec{r})=-\frac{2}{\sqrt{3}}{{\cal{F}}}r\left[1+\frac{r}{2}\right]\exp\left(-r\right)Y_{10}(\theta,\phi) \ , $$ from which the second-order energy is readily found: $$ E=-\frac{1}{2}-\frac{9}{4}{\cal{F}}^2+{\cal{O}}({\cal{F}}^4) \ . $$ The third-order energy is zero once again due to parity.

To find $E^{(2)}_\text{b}$, we need to compute the matrix elements of $z=r\sqrt{4\pi/3} \, Y_{10}(\theta,\phi)$: $$ \begin{aligned} \langle\psi^{(0)}_{n10}|z|\psi^{(0)}_{100}\rangle &= \frac{1}{\sqrt{3}}\int_0^\infty\mathrm{d}r \, r^2 \, R_{n1}(r) \, r \, R_{10}(r) \\ &= \frac{1}{\sqrt{3n(n^2-1)}}\frac{192n^2}{(n+1)^5}\frac{(4)_{n-2}}{(n-2)!}\sum_{k=0}^{n-2}\frac{(-(n-2))_{k}(5)_{k}}{(4)_{k}k!}\left(\frac{2}{n+1}\right)^k \\ &=\frac{16}{\sqrt{3}}\left(\frac{n-1}{n+1}\right)^n\sqrt{\frac{n^7}{(n^2-1)^5}} \\ &\equiv f(n)\ , \end{aligned} $$ where $(x)_n$ is the rising factorial. The above radial matrix element is a special case of a more general formula presented here. The sum over bound states is then calculated numerically: $$ \frac{1}{{\cal{F}}^2}E^{(2)}_\text{b}=-\sum_{n=2}^\infty\frac{2n^2f^2(n)}{n^2-1}\approx-1.83163 \ . $$ This implies $E^{(2)}_\text{c}/{\cal{F}}^2\approx-0.41837$, in good agreement with the value given in Ruffa, Am. J. Phys. 41 234 (1973); in other words, roughly $19\%$ of $E^{(2)}$ comes from the continuum. It may be tempting to explain this as a consequence of $H$ not having any exact bound states in the presence of a homogeneous electric field (only resonances); however, this cannot be true, as the same effect will be observed for Hamiltonians with genuine bound states.

$\bullet$ Case 2: $W=\lambda r$

Contrary to the previous case, the potential $-1/r+\lambda r$ (which is just a rescaled Cornell potential) gives rise purely to bound states. As far as I know, no closed form analytical solution can be found, but PT is easy to apply. We find $E^{(1)}=3\lambda/2$ and that $\psi^{(1)}$ must have $s$ symmetry; the first-order Schrödinger equation is again easily solved to yield $$ \psi^{(1)}(\vec{r})=\lambda\left[3-r^2\right]\exp(-r)Y_{00}(\theta,\phi) \ , $$ $$ E=-\frac{1}{2}+\frac{3}{2}\lambda-\frac{3}{2}\lambda^2+\frac{27}{4}\lambda^3+{\cal{O}}(\lambda^4) \ . $$ For e.g. $\lambda=0.01$, this perturbative result compares very favorably with the exact numerical ground state energy $E\approx-0.4851437$.

To compute the sum over bound states in $E^{(2)}$, we use another special case of the aforementioned general matrix element formula: $$ \begin{aligned} \langle\psi^{(0)}_{n00}|r|\psi^{(0)}_{100}\rangle &=\frac{1}{\sqrt{n}}\frac{24n^2}{(n+1)^4}\frac{(2)_{n-1}}{(n-1)!}\sum_{k=0}^{n-1}\frac{(-(n-1))_{k}(4)_{k}}{(2)_{k}k!}\left(\frac{2}{n+1}\right)^k \\ &=\frac{3}{2}\delta_{n1}-8\left(\frac{n-1}{n+1}\right)^n\sqrt{\frac{n^5}{(n^2-1)^4}}(1-\delta_{n1}) \\ &\equiv f(n) \ . \end{aligned} $$ The sum is again computed numerically: $$ \frac{1}{\lambda^2}E^{(2)}_\text{b}=-\sum_{n=2}^\infty\frac{2n^2f^2(n)}{n^2-1}\approx-1.074881 \ , $$ and it shows that roughly $28\%$ of the second-order energy comes from continuum states.

$\bullet$ Case 3: $W=\lambda/r^2$

The potential $-1/r+\lambda/r^2$ is just a rescaled Kratzer potential, for which the Schrödinger equation can be solved analytically (see e.g. Landau & Lifshitz III); the exact ground state energy reads $$ \begin{aligned} E =-\frac{2}{\left(1+\sqrt{1+8\lambda}\right)^2} =-\frac{1}{2}+2\lambda-10\lambda^2+56\lambda^3+{\cal{O}}(\lambda^4) \ . \end{aligned} $$ The PT series is convergent for $\lambda<1/8$. Of course, these corrections could also be found from the solution of the first-order Schrödinger equation: $$ \psi^{(1)}(\vec{r})=4\lambda\left[r+\ln(2r)+\gamma-3\right]\exp(-r)Y_{00}(\theta,\phi) \ , $$ where $\gamma$ is the Euler-Mascheroni constant.

The bound part of $E^{(2)}$ is computed with the help of another special case of the aforementioned formula: $$ \begin{aligned} \langle\psi^{(0)}_{n00}|r^{-2}|\psi^{(0)}_{100}\rangle &=\frac{1}{n^{3/2}}\frac{4}{n+1}\frac{(2)_{n-1}}{(n-1)!}\sum_{k=0}^{n-1}\frac{(-(n-1))_{k}}{(2)_{k}}\left(\frac{2}{n+1}\right)^k \\ &=-\frac{2}{n^{3/2}}\left[\left(\frac{n-1}{n+1}\right)^n-1\right]\\ &\equiv f(n) \ , \end{aligned} $$ leading to the numerical sum $$ \frac{1}{\lambda^2}E^{(2)}_\text{b}=-\sum_{n=2}^\infty\frac{2n^2f^2(n)}{n^2-1}\approx-1.56001 \ . $$ This shows that roughly $84\%$ of the second-order energy is produced by the continuum.

$\bullet$ Case 4: $W=\lambda \delta(\vec{r})$

This is a confusing, ill-posed eigenvalue problem. Delta function interactions often appear as approximations to various processes in atomic physics (fine and hyperfine structure, vacuum polarization in bound state QED, finite nuclear size effects, etc.), to be used in first-order PT with the obvious result $E^{(1)}=\lambda/\pi$. However, the higher-order energy corrections all turn out to be divergent. The exact eigenvalues of this Hamiltonian are similarly divergent when attacked with Green's function techniques, but end up finite and "trivial" when calculated variationally, or in some regularization scheme: they simply reduce to the eigenvalues of the unperturbed, hydrogenic Hamiltonian! I wrote a very long (probably too long) answer about this anomaly and why one should not worry too much about it.

The divergence of $E^{(2)}$ and $E^{(3)}$ is immediately evident if one tries to compute them with $$ \psi^{(1)}(\vec{r})=\frac{\lambda}{\pi}\left[2r+2\ln(2r)-\frac{1}{r}+2\gamma-5\right]\exp(-r)Y_{00}(\theta,\phi) \ ; $$ see the aforementioned answer for a derivation of $\psi^{(1)}$.

What makes this problem very deceptive is that all infinities of the PT come from the integral over continuum states, while the sum over bound states is finite: $$ |\psi^{(0)}_{nlm}(0)|^2=\delta_{l0}\frac{1}{\pi n^3} \ \ , \ \ |\psi^{(0)}_{lm}(\varepsilon;0)|^2=\delta_{l0}\frac{1}{\pi}\frac{1}{\displaystyle1-\exp\left(-\frac{2\pi}{\sqrt{2\varepsilon}}\right)} \ , $$ $$ \begin{aligned} \left(\frac{\pi}{\lambda}\right)^2E^{(2)}_\text{b} &=-\sum_{n=2}^\infty\frac{2}{n(n^2-1)} \\ &= -\sum_{n=2}^\infty\left[\frac{1}{n(n-1)}-\frac{1}{n(n+1)}\right] \\ &=-\frac{1}{2} \ , \end{aligned} $$ $$ \begin{aligned} \left(\frac{\pi}{\lambda}\right)^2E^{(2)}_\text{c}&=-\lim_{{\cal{K}}\rightarrow\infty}\int_0^{\cal{K}}\mathrm{d}\varepsilon\frac{2}{2\varepsilon+1} \frac{1}{\displaystyle1-\exp\left(-\frac{2\pi}{\sqrt{2\varepsilon}}\right)} \\ &=-\lim_{{\cal{K}}\rightarrow\infty}\left[\frac{\sqrt{2{\cal{K}}}}{\pi}+\frac{1}{2}\ln(2{\cal{K}}+1)+\frac{1}{2}+{\cal{O}}({\cal{K}}^{-1/2})\right] \ . \end{aligned} $$ If one only considered the bound state part of the hydrogenic spectrum, then it would be easy to arrive at the wrong conclusion that $E^{(2)}$ is finite.

Continuum contributions to PT energies in the general case

The previous examples were special since the first-order wave function could be found analytically, and this took care of all zeroth-order state contributions. This is of course not possible in the general case, since usually we have only numerical solutions already for the zeroth-order problem. The cleanest way to obtain accurate PT energies is to reformulate the calculation as a variational problem: computing the exact second-order energy turns out to be equivalent to finding the minimum of the so-called Hylleraas functional: $$ J[\phi]=\langle\phi|H^{(0)}-E^{(0)}|\phi\rangle+\langle\phi|W-E^{(1)}|\psi^{(0)}\rangle+\langle\psi^{(0)}|W-E^{(1)}|\phi\rangle \ . $$ This optimization problem is not tied to sum-over-states expressions, and can be used to find highly accurate numerical PT energies. See Ch. 4.2 of Simple Theorems, Proofs, and Derivations in Quantum Chemistry by Mayer or Sec. II.a.25.$\beta$ of Quantum Mechanics of One- and Two-Electron Atoms by Bethe & Salpeter for further details.

One more example: the ground state energy of Helium

The prototypical application of the Hylleraas functional approach was the computation of the second-order energy of Helium and similar two-electron ions. The clamped nucleus Hamiltonian of such a system ($Z$-dependence scaled out with $\vec{r}{}'=Z\vec{r}$) reads $$ \begin{aligned} H &= -\frac{1}{2}\nabla_1^2-\frac{1}{2}\nabla_2^2-\frac{Z}{r_1}-\frac{Z}{r_2}+\frac{1}{|\vec{r}_1-\vec{r}_2|} \\ &= Z^2\left[-\frac{1}{2}{\nabla'}_1^2-\frac{1}{2}{\nabla'}_2^2-\frac{1}{r'_1}-\frac{1}{r'_2}+\frac{1}{Z}\frac{1}{|\vec{r}{}'_1-\vec{r}{}'_2|}\right] \\ &\equiv Z^2\left[H^{(0)}+W\right] \ , \end{aligned} $$ the full electron-electron interaction being treated as a perturbation with "coupling constant" $1/Z$. This suggests that the nuclear charge-dependence of the energy is $$ E(Z)=Z^2\sum_{k=0}^{\infty}\epsilon_k Z^{-k} \ , $$ and that $E^{(2)}$ is independent of $Z$. One can easily find $\epsilon_0$ and $\epsilon_1$, but the second- and higher-order energies can only be obtained numerically, and the above variational approach is required to get their accurate values: $$ \epsilon_0=-1 \ \ , \ \ \epsilon_1=\frac{5}{8} \ \ , \ \ \epsilon_2\approx-0.157666429 \ \ , \ \ \epsilon_3\approx0.008699032 \ \ , \ \ ... \ . $$ Using these coefficients, one can already approach the exact non-relativistic clamped-nucleus value $E\approx-2.903724$ for Helium.

I calculated the bound state part of $\epsilon_2$ similarly to the previous cases; without going into details, I found $\epsilon_{2,\text{b}}\approx-0.0631$, which is in good agreement with the old result of Scherr, J. Chem. Phys 33 317 (1960) and Knight, Scherr, Phys. Rev. 128 6 2675 (1962). This shows that only $40\%$ of the second-order energy is attributed to purely bound states, the other $60\%$ comes from states in which at least one of the electrons is in a scattering state.