Home / Learn / Exam prep / GATE Physics / Lesson 4.9
All lessons

Module 4: Quantum mechanics · Lesson 4.9 · about 50 min

The variational method and WKB

Two approximation methods for problems with no small parameter: the variational principle and its upper bound, Gaussian and exponential trial functions, helium with a screened charge, linear trial combinations, then the WKB wavefunction, Bohr-Sommerfeld quantisation with the right half and quarter phases, and tunnelling through smooth barriers.

Key points, meanings and formulas on one page: choose "Save as PDF" in the print box.

Module 4 · about 50 minutes. By the end you will be able to bound a ground-state energy from above with a trial function and get the best value of its parameter, treat helium with a screened charge, set up a linear trial combination, write a WKB wavefunction, quantise a well with Bohr-Sommerfeld and the correct phase, and estimate tunnelling through a smooth barrier. You need expectation values from lesson 4.1, the oscillator of lesson 4.4 and hydrogen from lesson 4.6. Perturbation theory (lesson 4.8) needs a small term; these two methods do not.

Any guess for the ground state gives an energy that is too high, so the best guess in a family is the one with the lowest energy: that is the variational method. When the potential changes slowly compared with the wavelength, the particle behaves almost classically and its phase is just ∫p dx/ħ: that is WKB.

1. The variational principle

Expand the trial state. Write a normalised trial state as |ψ⟩=∑ncn|n⟩ in the (unknown) eigenstates of H, with ∑n|cn|2=1.

Average the energy. ⟨ψ|H|ψ⟩=∑n|cn|2En≥E0∑n|cn|2=E0, because every En≥E0.

Use it. Pick a family ψ(x;α), compute E(α)=⟨H⟩, and set dE/dα=0. The minimum is the best upper bound in the family. If ψ is not normalised, use ⟨ψ|H|ψ⟩/⟨ψ|ψ⟩.

Excited states. If the trial function is orthogonal to the ground state (for example odd, when the ground state is even), the same argument gives ⟨H⟩≥E1.

E0≤⟨ψ|H|ψ⟩⟨ψ|ψ⟩for every trial ψ;an error ϵ in ψ gives an error of order ϵ2 in E

The last point is why the method is so good: the energy is stationary at the true eigenstate, so a rough wavefunction still gives a sharp energy. The wavefunction itself, and averages such as ⟨x2⟩, are far less reliable.

`$\psi(x;\alpha)$`
the trial function, a guess with an adjustable parameter; it must satisfy the boundary conditions (vanish at hard walls, be normalisable)
`$E(\alpha) = \langle H\rangle$`
the trial energy; always at or above the true ground-state energy
`$dE/d\alpha = 0$`
the condition for the best member of the family

2. Trial functions that come up again and again

Oscillator with a Gaussian. ψ=e−bx2 gives ⟨T⟩=ℏ2b/2m and ⟨V⟩=mω2/8b. Minimising, b=mω/2ℏ and E=12ℏω: exact, because the true ground state is in the family.

Oscillator with an exponential. ψ=e−α|x| has ⟨T⟩=ℏ2α2/2m and ⟨x2⟩=1/2α2, so E(α)=ℏ2α22m+mω24α2. A sum Aα2+B/α2 has minimum 2AB, giving E=ℏω/2=0.707ℏω, 41% too high. The kink at the origin costs kinetic energy.

Worked example (infinite well). For a well 0<x<a, try ψ=x(a−x), which vanishes at both walls. Then ⟨ψ|ψ⟩=a5/30 and ⟨ψ|T|ψ⟩=ℏ22m∫0a(a−2x)2dx=ℏ2a36m, so E=5ℏ2ma2. The exact value is π2ℏ2/2ma2=4.935ℏ2/ma2: only 1.3% high, from a parabola with no free parameter at all.

Hydrogen with a Gaussian. ψ=e−br2 gives E(b)=3ℏ2b2m−e24πϵ08bπ, with minimum −83π×13.6 eV =−11.5 eV against the exact −13.6 eV. The Gaussian has the wrong shape at the nucleus (no cusp) and decays too fast far away.

The trial energy as a function of the parameter for four cases. The dashed line is the exact ground-state energy; every curve stays above it, and its minimum is the variational estimate. Drag the slider to see how far off a poor parameter puts you.

3. Helium, the classic test

Helium has two electrons and Z=2. Ignoring the electron-electron repulsion gives −2Z2×13.6 eV =−108.8 eV; adding it as a first-order term with the unscreened 1s2 state gives +54Z×13.6=34.0 eV, so −74.8 eV. Measured: −79.0 eV.

Trial state. Both electrons in hydrogen-like 1s orbitals with an effective charge ζ: ψ=ζ3πa03e−ζ(r1+r2)/a0. Each electron partly shields the other, so expect ζ<2.

Average energy. In units of the Rydberg (13.6 eV): kinetic 2ζ2, nuclear attraction −4Zζ, repulsion 54ζ. So E(ζ)=2ζ2−4Zζ+54ζ.

Minimise. 4ζ−4Z+54=0 gives ζ=Z−516=2716=1.69.

Result. E=−2ζ2=−2(27/16)2×13.6 eV =−77.5 eV, within 2% of the measured value and, as promised, above it.

Linear trial functions (Rayleigh-Ritz). Take ψ=∑iciϕi with fixed ϕi and the ci as parameters. Minimising gives the secular equation det⁡(Hij−ESij)=0, with Sij=⟨ϕi|ϕj⟩. Its lowest root bounds E0, and each higher root bounds the next level. With an orthonormal basis this is just diagonalising H in a truncated basis, which is how most numerical solvers work. To check a trial answer against a numerical one, try the Schrödinger equation solver.

“A lower trial energy means the method has failed.” It cannot happen. If your ⟨H⟩ comes out below the exact E0, there is an algebra error, or the trial function breaks a boundary condition (for example it does not vanish at a hard wall), or it was not normalised.

“The best trial energy means the best wavefunction for everything.” The energy is accurate to second order in the error of ψ; other averages only to first order. The Gaussian trial for hydrogen gets the energy within 15%, yet its |ψ(0)|2 (which sets the hyperfine splitting) is four times too small.

4. The WKB approximation

In a region where E>V, write p(x)=2m(E−V(x)). If V were constant, ψ=e±ipx/ℏ. When V changes slowly, keep the local momentum:

ψ(x)≈Cp(x)exp⁡(±iℏ∫xp(x′)dx′),ψ(x)≈C′ℏκ(x)exp⁡(±∫xκ(x′)dx′) where E<V

with κ=2m(V−E)/ℏ. The amplitude 1/p is classical: |ψ|2∝1/v, the particle is found more often where it moves slowly. The approximation needs the de Broglie wavelength to change little over one wavelength, |dλ/dx|≪1, which fails at turning points where p=0. There, the connection formulas (from the exact Airy-function solution of a linear potential) join the two sides and fix a phase of π/4 at each smooth turning point.

Quantisation. Matching the waves from the two turning points x1,x2 of a well:

∫x1x2p(x)dx=(n+12)πℏ(two smooth points),(n+34)πℏ(one hard wall),(n+1)πℏ(two hard walls)

with n=0,1,2,…. Equivalently ∮pdx=(n+12)h for a smooth well: the Bohr-Sommerfeld rule with the half-quantum that the old quantum theory missed.

Worked example (oscillator). ∫pdx between turning points is half the area of the phase-space ellipse, πE/ω. Setting it to (n+12)πℏ gives En=(n+12)ℏω, exact. The infinite square well with the (n+1) rule is exact too.

Worked example (V = F|x|). Turning points at ±E/F. ∫−E/FE/F2m(E−F|x|)dx=432mE3/2F. Setting this to (n+12)πℏ:

En=[3πℏF(n+12)42m]2/3,En∝(n+12)2/3

Worked example (the quantum bouncer). A particle on a hard floor in gravity, V=mgx for x>0. One wall and one smooth point: 232mE3/2mg=(n+34)πℏ, so En=[9π28mg2ℏ2(n+34)2]1/3. For n=0 this is 1.842(mg2ℏ2)1/3, against the exact 1.856 from the first Airy zero: 0.8% off, even for the ground state. Ultracold neutrons above a mirror show exactly these levels, about 1.4 peV for the lowest.

Counting states. Inverting the rule, the number of bound states of a well with energies below E is about 1πℏ∫pdx+12, a quick way to estimate how many levels a finite well holds.

5. Tunnelling through a smooth barrier

Inside a barrier the WKB wave decays as e−∫κdx. For a thick barrier, the transmission is

T≈exp⁡(−2∫x1x2κ(x)dx),κ(x)=2m(V(x)−E)ℏ

taken between the two turning points. For a rectangular barrier this is e−2κL, the leading factor of the exact result.

Worked example (rectangular). An electron 1 eV below the top of a barrier 1 nm wide: κ=2me(1 eV)/ℏ=5.12 nm⁻¹, 2κL=10.25, T≈3.6×10−5. Double the width and T falls to about 10−9: tunnelling is exponentially sensitive to width and to V−E.

Worked example (field emission). A metal surface with work function W in a field ℰ: outside, V−E=W−eℰx, a triangle of height W and width L=W/eℰ. Since ∫0L1−x/Ldx=2L/3, the exponent is 43κ0L=42mW3/23ℏeℰ, with κ0=2mW/ℏ. This is the Fowler-Nordheim law: current grows as e−const/ℰ. Alpha decay is the same idea with a Coulomb barrier, and the Gamow factor explains why half-lives range over more than 20 orders of magnitude for alpha energies between 4 and 9 MeV.

“WKB only works for highly excited states.” It is derived for large n, but it is exact for the oscillator and the square well, and within 1% for the bouncer’s ground state. It does fail where the potential changes sharply on the scale of a wavelength, for example a step.

“Use the (n + ½) rule for every well.” The phase depends on the kind of turning point. Hard walls do not give the π/4 shift, so a half-oscillator (wall at x=0) has the rule (n+34) and keeps only the odd oscillator levels, E=(2n+32)ℏω, which is again exact.

  • Variational: ⟨H⟩≥E0 for any trial state; minimise over parameters; energy error is second order in the wavefunction error.
  • Oscillator: Gaussian gives 12ℏω (exact); e−α|x| gives ℏω/2. Well with x(a−x): 5ℏ2/ma2 (1.3% high). Hydrogen with a Gaussian: −11.5 eV.
  • Helium: ζ=Z−5/16=27/16, E=−2ζ2×13.6 eV =−77.5 eV (measured −79.0).
  • WKB wave ∝p−1/2e±i∫pdx/ℏ; valid when the wavelength changes slowly; breaks at turning points.
  • ∫pdx=(n+12)πℏ (smooth-smooth), (n+34)πℏ (wall-smooth), (n+1)πℏ (wall-wall). For V∝|x|, En∝(n+12)2/3.
  • Tunnelling: T≈e−2∫κdx; a triangular barrier gives 43κ0L.

Practise: GATE questions on this lesson

They start easy and climb to top-rank level. More on the variational method and WKB practice page and the GATE Physics practice page.

Next: lesson 4.10 lets the perturbation depend on time: transitions, Rabi oscillations and Fermi’s golden rule. Back to the GATE Physics course.

Check yourself

1A trial function gives ⟨H⟩ = −12.9 eV for a system whose true ground state is unknown. What can you say for sure?
2What is the WKB quantisation rule for a smooth potential well with two ordinary turning points?
3In the WKB tunnelling formula T ≈ exp(−2∫κ dx), what is κ?

Sources

  • D. J. Griffiths and D. F. Schroeter, Introduction to Quantum Mechanics, 3rd ed. (2018), chapters 8 and 9: the variational principle and the WKB approximation
  • J. J. Sakurai and J. Napolitano, Modern Quantum Mechanics, 3rd ed. (2021), sections 2.5 and 5.4: WKB and variational methods
  • L. D. Landau and E. M. Lifshitz, Quantum Mechanics (Non-relativistic Theory), 3rd ed. (1977), sections 46 to 50: the quasi-classical case
  • GATE Physics (PH) syllabus, quantum mechanics section, published by the organising institute each year

Your ticks stay in this browser. Save them as a file on the course home.

Updated 2026-11-22. Not affiliated with the IITs, IISc or the GATE organising institute. The general aptitude part of GATE is not covered here. Check the official GATE website for the current pattern. All questions are original.