Variational method


The variational method is a method of finding approximate solution to quantum problems, typically finding eigenvalues and eigenfunctions of a time-independent Hamiltonian.

Method

Consider some time-independent Hamiltonian HH of a function ϕ\phi which can vary freely, called a trial function. We define the Functional:

E[ϕ]=ϕHϕϕϕE[\phi]=\frac{\braket{ \phi | H|\phi }}{\braket{ \phi | \phi } }

This is the mean value of EE in the state ϕ\ket{\phi}. As usual, if ϕ\ket{\phi} is an eigenstate, then E(ϕ)E(\phi) is an energy eigenvalue. The heart of the method is this:

To prove this, let's consider a small variation δϕ\delta \phi of ϕ\phi1:

E[ϕ+δϕ]ϕ+δϕϕ+δϕ=ϕ+δϕHϕ+δϕE[\phi+\delta \phi]\braket{ \phi+\delta \phi | \phi+\delta \phi } =\braket{ \phi+\delta \phi | H|\phi+\delta \phi }

We now look for the eigenvalues of EE for the varied function:

δEϕϕdτ=EδϕϕdτEϕδϕdτ+δϕHϕdτ+ϕHδϕdτ=δϕ(HE)ϕdτ+ϕ(HE)δϕdτ\begin{align} \delta E\int \phi^{*}\phi d\tau&=-E\int \delta\phi^{*}\phi d\tau-E\int \phi^{*}\delta\phi d\tau+\int \delta \phi^{*}H\phi d\tau +\int \phi^{*}H\delta \phi d\tau \\ &=\int \delta \phi^{*}(H-E)\phi d\tau+\int \phi^{*}(H-E)\delta \phi d\tau \end{align}

But we want EE to be stationary, in which case δE=0\delta E=0 always. Thus, the left hand side must be zero. This is our condition to find ϕ\phi; the rest is "just" a matter of solving the equation.

For example, consider the Schrödinger equation Hψn=EnψnH\psi_{n}=E_{n}\psi_{n}, indexed by a quantum number n0n\geq0. The ground state is n=0n=0 with E0E_{0}. We now consider a generic function ϕ\phi. This can be expressed in Fourier series in the {ψn}n\{ \psi_{n} \}_{n} basis as

ϕ=nanψn\phi=\sum_{n}a_{n}\psi_{n}

The Expected value of E(ϕ)E(\phi) for this function is

E[ϕ]=nan2Ennan2=E0+nan2(EnE0)nan2E0E[\phi]=\frac{\sum_{n}\lvert a_{n} \rvert ^{2}E_{n}}{\sum_{n}\lvert a_{n} \rvert ^{2}}=E_{0}+ \frac{\sum_{n}\lvert a_{n} \rvert ^{2}(E_{n}-E_{0})}{\sum_{n}\lvert a_{n} \rvert ^{2}}\geq E_{0}

We hit on a lower bound. In fact, while this doesn't give us a solution, it does gives a bound to restrict what E[ϕ]E[\phi] can be:

E[ϕ]=ϕHϕϕϕE0\boxed{E[\phi]=\frac{\braket{ \phi | H|\phi }}{\braket{ \phi | \phi } }\geq E_{0}}

The variational method consists in minimizing this functional. Thus, the approximations given by the variational method are always overestimates and the actual ground state energy is guaranteed to be lower (or technically equal is the best case scenario).

Footnotes

  1. Since ϕ\phi is a function, δϕ\delta \phi is a directional derivative.