Quantum mechanics, Part 2

Building on what we’ve learned in the first part of quantum mechanics, we can now explore more complex concepts in quantum theory. In this second part, we will cover the theory that allows us to do more advanced calculations in quantum mechanics

Introduction to matrix mechanics¶

In studying complex multi-state quantum systems, numerical methods are often the only way to solve a variety of problems, as solving the Schrödinger equation by hand becomes impossible. One important feature of these numerical methods is utilizing the Heisenberg picture of quantum mechanics, also known as matrix mechanics.

In matrix mechanics, we describe the system not through its total wavefunction, but by its operators. The quantum state $|\Psi\rangle$ stays constant; the operators (energy, momentum, etc.) are what evolve through time. In particular, the operators can be expressed in specific bases (bases are plural of “basis”). For discrete operators, as we have seen for the spin matrices, we can choose a discrete basis. In our case, the $\hat S_z$ operator can be expressed using the basis of the two spin states, as given by:

Meanwhile, for continuous (differential) operators, we can choose a functional basis. For instance, we can use the Legendre, Laguerre, and Hermite polynomials, all of which form a complete basis set, to find the matrix representation of the operator. Within this section, we will give a brief introduction on determining the matrix representation of various operators, based on the book Basic Theory of Lasers and Masers by Jacques Vanier.

The Hamiltonian in general bases¶

In the time-independent regime, the state of a quantum system is represented by its state-vector $|\Psi\rangle$, which can be expanded into a set of eigenstates $|\psi_n\rangle$ by $|\Psi\rangle = \displaystyle \sum_n |\psi_n\rangle$. Each eignestate satisfies eigenvalue equation $\hat H |\psi_n \rangle = E_n |\psi_n\rangle$, where $\hat H$ is the Hamiltonian and $E_n$ is the energy eigenvalue of a particular eigenstate. We may extract the energy eigenvalues $E_n$ as follows. First, we multiply both sides by a bra $\langle\psi_m|$:

Now, the eigenstates $|\psi_n\rangle$ can theoretically be expressed in any basis, but we typically choose to use a complete orthonormal basis. Such bases include Fourier series as well as the Legendre, Laguerre, and Hermite polynomials. The specific basis doesn’t matter; it simply matters that a complete and orthonormal basis satisfies $\langle \psi_m | \psi_n \rangle = \delta_{mn}$. Therefore, we have $\langle \psi_m | E_n | \psi_n \rangle = E_n \langle \psi_m |\psi_n\rangle$ (since $E_n$ is a constant and can be factored out of the expression). Given that we have an orthonormal basis, $E_n \langle \psi_m |\psi_n\rangle$ is only nonzero when $m = n$, in which case we have $E_n \langle \psi_n |\psi_n \rangle = E_n$, and thus:

Thus, to find $E_n$, we “only” need to calculate $\langle \psi_n | \hat H |\psi_n\rangle$, which can also be written in terms of the Schrödinger formalism as:

But returning to the matrix representation - we may write the set of eigenvalues $E_n$ with the following matrix equation:

This is a diagonal matrix - so one might wonder, why bother writing this when each side can just be written as a vector multiplied by an identity matrix? The reason is that if we don’t know the eigenstates, we can always choose to expand our quantum state $|\Psi\rangle$ in some other complete and orthonormal basis instead, which we will refer to as $|\phi_n\rangle$ (since we can always choose any basis to write out our state-vector). As both bases are complete and orthogonal, we can then express a state in our new basis $|\phi_n\rangle$ in terms of our eigenstate basis $|\psi_n\rangle$ as a linear sum:

Where all the $a_{nk}$'s are constant coefficients. We can write this in matrix form as:

We can extract out the components of $A_{nk}$ using another orthogonality argument. If we take the components of our new basis $|\phi_n\rangle$, then multiply by a bra $\langle \psi_j|$, then we have:

Where $\langle \psi_j |\psi_k\rangle = \delta_{jk}$ since our bases are othogonal. The only case were $\langle \psi_j | \phi_n \rangle = A_{nk} \delta_{jk}$ does not vanish is when $j = k$ and therefore we have:

Thus we have $A_{nk} = |\phi_n\rangle \langle \psi_k|$. We may also derive an expression for the inverse, i.e. $A^{-1}$. Recall that the inverse of $A$ must satisfy $A A^{-1} = I$ where $I$ is the identity matrix. We recall that the identity matrix is given by $I = \displaystyle \sum_n |\phi_n \rangle \langle \phi_n |$. Then, we have:

Now, if we take the inner product of both sides with the ket $|\psi_k\rangle$, then:

Where we were able to swap the order of the inner product since the inner product is commutative, and we were able to pull the $|\psi_k\rangle$ into the sum since the sum does not sum over the index $k$ (so $|\psi_k\rangle$ can essentially be treated as a constant in the sum). Now if we take the inner product with the bra $\langle \phi_n|$ we have:

Where again, since $|\psi_k\rangle$ is not summed over, we were able to pull it out of the sum, and we used orthogonality to collapse the sum into a single term.

Having computed $A$ and $A^{-1}$, we will now show that our transformation of basis actually leads to a very nice expression for the matrix representation of the Hamiltonian. The Hamiltonian in our new basis, which we write as $\hat{\mathscr{H}}$, is given by $\hat{\mathscr{H}} = A \hat H A^{-1}$ (this is a standard result of linear algebra). If we then substitute our expressions for $A$ and $A^{-1}$, we have:

Where the last step comes from $\langle \psi_n | \hat H |\psi_n\rangle = E_n$ (the index does not matter at this point because we use only one index in the expression). Thus we find that:

Let us demonstrate the same with, in the functional picture. Consider a system described by wavefunction $\psi(\mathbf{r})$. By the postulates of quantum mechanics, the wavefunction may be expanded in any complete and orthonormal set of functions $\phi_n(\mathbf{r})$, such that:

Where we assume all $\phi_n$ are normalized. From here, we may prepare a matrix $A_{mn}$ with elements given by:

Now, $A_{mn}$ is not necessarily a diagonal matrix, so it does not necessarily give the energy eigenvalues. However, if we diagonalize it, we are left with the matrix $E_{mn}$, which is diagonal (that is, for all $m \neq n$, $E_{mn} = 0$, and whose diagonals $E_{nn} = E_n$ are equal to the energy eigenvalues.

The matrix operator approach therefore condenses the difficult problem of solving the Schrödinger equation into a more straightforward problem of finding the correct matrix $A$ that diagonalizes the Hamiltonian in a particular basis; from there, we can simply read off the energy eigenvalues from the diagonal.

Example: quantum harmonic oscillator¶

We will use the matrix representation approach to solve for the quantum harmonic oscillator, which will act as a toy model. We want to obtain a matrix representation for the Hamiltonian of the quantum harmonic oscillator and find the energy eigenvalues.

The well-known Hamiltonian for the quantum harmonic oscillator is given by:

We will now need to pick a basis to be able to obtain its matrix representation. In theory, when we don’t know the precise eigenstates of the matrix, any set of basis functions would do (as long as they are complete and orthogonal) - but luckily for us, we already know the eigenstates of the Hamiltonian. So for demonstrative purposes, it is easiest to choose the basis of the eigenstates of the Hamiltonian, which we can write as $|\psi_1\rangle, |\psi_2\rangle, |\psi_3\rangle, \dots, |\psi_n\rangle$. Then, we have $E_n = \langle \psi_n | \hat H |\psi_n\rangle$. But recall that in the example of the quantum harmonic oscillator, $\hat H |\psi_n\rangle = \hbar \omega \left(n + \dfrac{1}{2}\right) |\psi_n\rangle$. Thus, $\langle \psi_n | \hat H | \psi_n\rangle$ is given by:

Thus our resulting energy matrix becomes^[3]:

And thus our energy eigenvalues can be found (as expected) by just reading off the diagonals the diagonals, from which we can surmise that:

Let us now show that even though this matrix representation method for the quantum harmonic oscillator is a “toy model”, it can still predict real-world results. The transition energy $\Delta E$ from our result is the difference between energy levels $n_1, n_2$, and thus:

From which we may derive the spectral lines for various different transitions, given that $\Delta E = hc/\lambda$, with:

The angular frequency of the vibrations, $\omega$, can be found through $\omega = \sqrt{k/\mu}$ where $\mu$ is the reduced mass of the molecule and $k$ is its force constant. Let us compute the vibrational transitions of carbon dioxide, which, although not strictly speaking a diatomic molecule, can be somewhat treated as such. The reduced mass of the triatomic carbon dioxide molecule is given by^[5]:

Where $m_{\mathrm{C}}$ is the mass of the carbon atom and similarly $m_{\mathrm{O}}$ is the mass of the oxygen atom. The force constant of $\mathrm{CO2}$ is approximately $\text{1680 N/m}$^[4] and thus $\omega \approx \text{251.47 THz}$. Substituting these values, we find that the spectral lines are given by: