# How is the error in a wavefunction related to the error in energy?

While studying the variational method in McQuarrie's Quantum Chemistry, I came across the following problem: to relate the difference between an approximation $$\phi$$ and the exact ground-state wave function $$\psi_0$$ with the difference between the approximate energy $$E=\int\phi^* \hat H \phi \, \text{d} \tau$$ and the ground-state energy $$E_0=\int\psi_0^* \hat H \psi_0 \, \text{d} \tau$$. For instance, if I know that $$\phi$$ and $$\psi_0$$ differ by at most, say, $$O(\epsilon)$$ over all space, can I say that $$E-E_0 \approx O(\epsilon)$$?

It is intuitive to think that the better approximation $$\phi$$ is, the better the energy approximation gets, but I cannot prove this. In Chemistry SE, I was told to look at this paper, but it does not quite solve the problem.

Without loss of generality, suppose $$\phi$$ and $$\psi_0$$ normalized. By writing $$\phi$$ as a linear combination of $$\psi_n$$, we get $$\phi=\sum_{n=0}^{\infty} c_n\psi_n,\tag{1}$$ such that:

$$\sum_{n=0}^{\infty} c_n^* c_n=1\tag{2}.$$

Therefore: $$\hat H \phi=\sum_{n=0}^{\infty} c_n\hat H\psi_n = \sum_{n=0}^{\infty} c_nE_n\psi_n,\tag{3}$$

hence:

$$E=\int\phi^* \hat H \phi \, \text{d} \tau=\sum_{n=0}^{\infty} c_n^* c_nE_n\tag{4}.$$

From here, I can easily show that $$E \geq E_0$$ (since $$|c_0|^2 =1- \sum_{n=1}^{\infty}|c_n|^2$$), but how can I estimate the difference between the approximate energy and the ground-state energy?

• Wavefunctions are not dimensionless. Their dimension is such that $\int \psi^{*} \psi \, \mathrm{d}\tau=1$. For instance, if the particle is constrained to move in one spatial dimension, then the wavefunction has dimension $L^{-1/2}$. Mar 19, 2021 at 16:43
• Sure. I think that the author uses the fact that $O(\epsilon)=O(k \epsilon)$ if $k$ is a constant, no matter what its unit is (see Big O notation). Thus, I think saying the wavefunctions differ by $O(\epsilon)$ and the energies differ by $O(\epsilon)$ is not, by itself, a contradiction. Mar 19, 2021 at 16:56
• Donald A. McQuarrie in the chapter "Approximate Methods" of his book "Quantum Chemistry" Mar 19, 2021 at 18:28
• Unless you have a PDF of the book, I can't see it. Mar 19, 2021 at 18:29
• sorry for deleting so many of my comments. The warning "comments are not intended for discussion: please move the conversation to chat" came up, and I wanted to avoid this: meta.stackexchange.com/q/353643/391772 Mar 21, 2021 at 3:23

You can, in fact, do better as the answer to this Physics.SE question of yours indicates, assuming you're after an asymptotic expression. I haven't read the book you mention, but if you're familiar with the bra-ket notation, I recommend the discussion of variational methods in Sakurai's Modern Quantum Mechanics. The following will be based on Sakurai's approach, but adapted to address the questions I think you're asking.

Let $$\{|\psi_n\rangle\}$$ be an orthonormal basis, where $$|\psi_0\rangle$$ is the true ground state. We consider a trial wave function (ket) $$|\phi\rangle=|\psi_0\rangle + |\delta \phi\rangle$$, and express it in the energy eigenstate basis, $$|\phi\rangle=\sum_{n=0}^{N} c_n|\psi_n\rangle \tag{1}$$ where $$\sum_{n=0}^{N} |c_n|^2=1$$, such that $$\delta \phi= \sum_{n=1}^N c_n |\psi_n\rangle$$. Here $$N$$ is the dimension of the Hilbert space. You may take $$N\rightarrow \infty$$ as in your question, but a finite dimension allows a clearer discussion of the difference between the wave functions.

The energy in the trial wave function is (since $$|\phi\rangle$$ is normalized) \begin{align}\tag{2} E &= \langle \phi | \hat{H} | \phi\rangle = \left( \langle \psi_0| + \langle \delta\phi| \right) \hat{H} \left( |\psi_0\rangle + | \delta \phi\rangle\right)\\\tag{3} &= \langle \psi_0| \hat{H} |\psi_0\rangle + \langle \psi_0| \hat{H} | \delta \phi\rangle + \langle \delta\phi| \hat{H} |\psi_0\rangle + \langle \delta\phi| \hat{H} | \delta \phi\rangle\\ &= E_0 + \langle \delta\phi| \hat{H} | \delta \phi\rangle, \tag{4} \end{align} where we used that $$|\delta \phi\rangle$$ is orthogonal to $$|\psi_0\rangle$$. Thus the difference between the approximate and exact ground state energies can be written $$\delta E = E -E_0 = \langle \delta\phi| \hat{H} | \delta \phi\rangle = \sum_{n=1}^N E_n |c_n|^2. \tag{5}$$

Now, if we think about $$|\phi\rangle$$ as a (long?) vector, where each element corresponds to $$c_n$$ ($$n=0,1,2,\dots$$), the difference between $$|\psi\rangle$$ and the exact ground state can be written $$|\delta\phi\rangle = |\phi\rangle - |\psi_0\rangle = (0, c_1, c_2, c_3, \dots, c_N). \tag{6}$$ You now have to decide how you want to define the meaning of "$$\phi$$ and $$\psi_0$$ differ by at most $$O(\epsilon)$$". (There are many distance metrics for Hilbert spaces.) One simple way is to directly say that the largest element ($$c_{n_0}$$, say) of the $$|\delta \phi\rangle$$ vector is $$O(\epsilon)$$. Alternatively, we can require $$\sqrt{\langle \delta \phi | \delta \phi \rangle} = \sqrt{\sum_{n=1}^N |c_n|^2} = \sqrt{1-|c_0|^2} = \mathcal{O} (\epsilon), \tag{7}$$ which is achieved if $$|c_n|\lesssim \mathcal{O}(\epsilon)\quad \forall n>0$$. In both cases we find $$\delta E = \sum_{n=1}^N E_n |c_n|^2 \lesssim \sum_{n=1}^N E_n \mathcal{O}\left(\epsilon^2\right). \tag{8}$$ If the spectrum of $$\hat{H}$$ is bounded (i.e. each $$E_n$$ is finite), $$E_n$$ can be treated as a constant that is unimportant for asymptotic properties, yielding $$\delta E \sim \sum_{n=1}^N \mathcal{O}\left(\epsilon^2\right) = \mathcal{O}\left(\epsilon^2\right). \tag{9}$$ Thus, an error $$\mathcal{O}(\epsilon)$$ in the trial wave function leads to an energy error that scales as $$\mathcal{O}(\epsilon^2)$$. Note, however, that this expression is not itself useful to estimate the magnitude of the energy difference, which depends non-universally on $$E_n$$. A trivial way to see this is to simply multiply $$\hat{H}$$ by some large number, which will affect the energy difference in Eq. (5), but not the difference between states in Eq. (6). Neither will it change the error in Eq. (9). If you need the magnitude of $$\delta E$$, you'll want to know $$E_0$$ independently, e.g. from experiment or some other theoretical technique (but if you have another theoretical technique, using a variational one might be unnecessary). The real utility of the variational method comes from the fact that you can just guess trial wave functions until you're tired, pick the one with lowest energy $$E$$, and be sure that it's the best approximation of $$|\psi_0\rangle$$ that you've tried so far.

To avoid confusion, perhaps it's worth mentioning that the overall units of Eqs. (8,9) are the same. In (9) I've absorbed the units of energy into the $$\mathcal{O}$$, whereas $$\mathcal{O}$$ is dimensionless in (8). This is common practice, since $$\mathcal{O}$$ is used as a shorthand.

• I gave +1 earlier, but I'm just commenting to say sorry for submitting that edit before finishing it (i.e. also changing the reference numbers at the end!). By the way, I remember your first answer on this site was an absolute masterpiece about the Heisenberg model. I wonder if you could offer any help with these Heisenberg and Ising model questions which have unfortunately gone unanswered for far too long: mattermodeling.stackexchange.com/q/1548/5, mattermodeling.stackexchange.com/q/1678/5, mattermodeling.stackexchange.com/q/1733/5. I'm trying to clear up the queue! Mar 19, 2021 at 20:14
• @NikeDattani I can give an answer to the first one. The second one might be answerable by diving into the referenced paper, but I have no real inroad to the question about the Wolff algorithm. Mar 23, 2021 at 17:52
• Thanks! Removing those two questions from the unanswered queue would be a huge help, and would earn you two necromancer badges. About the Wolff algorithm, someone from Physics.SE is now discussing it with the OP here, and the discussion is quite interesting (apparently the machine learning model was trained on 2-body correlations but works decently on 4-body correlations too): chat.stackexchange.com/rooms/119831/… Mar 23, 2021 at 18:01