Ladder operators and the algebraic solution

Write the Hamiltonian as Ĥ = ½mω²(x̂² + p̂²/m²ω²) — a sum of two squares, with u = x̂ and v = p̂/mω. Over ordinary numbers that factorises as (u − iv)(u + iv). Try it on operators and the order starts to matter: (u − iv)(u + iv) = u² + v² + i(uv − vu) = u² + v² + i[x̂, p̂]/mω, and since [x̂, p̂] = iħ the leftover is i(iħ)/mω = −ħ/mω. The factorisation misses, and it misses by a constant rather than by an operator. Roll the scale factor √(mω/2ħ) into each bracket to get a and a†, and the miss becomes exactly one half: a†a = Ĥ/ħω − ½, or Ĥ = ħω(a†a + ½). Nothing has been assumed and nothing solved yet — this is a rearrangement of the same operator. But it has already produced the zero-point term and said where that term comes from: not from a wall, not from a boundary condition, but from two observables refusing to commute.

Multiply in the other order: aa† = Ĥ/ħω + ½, so [a, a†] = aa† − a†a = 1. That one line is the entire engine. Define N̂ = a†a, so Ĥ = ħω(N̂ + ½), and then [N̂, a†] = a†[a, a†] = a† and [N̂, a] = −a, which in energies reads [Ĥ, a†] = ħω a†. Now let Ĥ|ψ⟩ = E|ψ⟩ and act on a†|ψ⟩: Ĥ(a†|ψ⟩) = (a†Ĥ + ħω a†)|ψ⟩ = (E + ħω)(a†|ψ⟩). So a†|ψ⟩ is again an eigenstate, with energy exactly one ħω higher, and the same computation with a lowers it by one ħω. No wavefunction has appeared, no differential equation has been solved, and no Hermite polynomial has been mentioned. The evenly spaced ladder is a consequence of an algebraic identity — which is why every strictly quadratic system, a lattice mode, a mode of the electromagnetic field, a trapped ion, inherits the same spectrum.

Raising can continue forever, since nothing bounds a bound state's energy above. Lowering cannot, and the reason is a norm. For a normalised eigenstate of energy E, ‖a|ψ⟩‖² = ⟨ψ|a†a|ψ⟩ = E/ħω − ½, and a squared norm is never negative, so E ≥ ħω/2 for every state whatsoever. Now suppose E = (ν + ½)ħω with ν not a whole number — say ν = 1.7. Then a|ψ⟩ is nonzero, since its squared norm is 1.7, and it sits at ν = 0.7; applying a again gives a state whose squared norm is 0.7 × 1.7 = 1.19, so it too is nonzero, and it sits at ν = −0.3, an energy of 0.2ħω — below the floor the norm has just established. Contradiction. The only escape is that the descent terminates, and it can terminate only on a state annihilated by a, so a|0⟩ = 0 and ν = 0 exactly. Every rung above is reached in whole steps, giving Eₙ = (n + ½)ħω with n = 0, 1, 2, …. Quantisation here comes from positivity of the inner product, not from fitting waves into a box.

a|n⟩ is proportional to |n−1⟩, but with what number in front? Take the norm: ‖a|n⟩‖² = ⟨n|a†a|n⟩ = ⟨n|N̂|n⟩ = n, so a|n⟩ = √n |n−1⟩ once the free phase is fixed real. For the raise, use the commutator instead of a fresh integral: ‖a†|n⟩‖² = ⟨n|aa†|n⟩ = ⟨n|(a†a + 1)|n⟩ = n + 1, so a†|n⟩ = √(n+1)|n+1⟩. Two things follow at once. The lowering factor vanishes at n = 0, which is the bottom rung stated as an equation rather than argued for. And the raising factor beats the lowering factor by exactly one in the square, at every rung — that difference of one is [a, a†] itself. These roots are not decoration. The dipole matrix element ⟨n+1|x̂|n⟩ = √((n+1)ħ/2mω) is what makes vibrational transition strengths climb with n, so dropping the factor leaves the energies intact and the spectroscopy wrong.

The algebra hands you energies but no wavefunctions, and it needs exactly one differential equation to get them. Put p̂ = −iħ d/dx into a|0⟩ = 0: √(mω/2ħ)(x + (ħ/mω) d/dx)ψ₀ = 0, so dψ₀/ψ₀ = −(mω/ħ)x dx and ψ₀ ∝ exp(−mωx²/2ħ). That is first order and separable — a line of work, against the second-order equation and two-step recursion the series method has to grind through. Normalising with the Gaussian integral gives ψ₀ = (mω/πħ)¹⁄⁴exp(−x²/2x₀²), where x₀ = √(ħ/mω) is the oscillator length. Every excited state then follows by differentiating and multiplying rather than by solving anything: ψₙ = (a†)ⁿψ₀/√(n!). One application of a† gives ψ₁ = √2 (x/x₀)ψ₀; two give ψ₂ ∝ (2x²/x₀² − 1)ψ₀, which is the Hermite polynomial H₂ = 4ξ² − 2 up to a factor of two, obtained without writing a recursion relation.

Expectation values become step counts. With x̂ = √(ħ/2mω)(a + a†), squaring gives terms in a², a†², a†a and aa†; the first two connect |n⟩ to |n±2⟩ and integrate to zero, so ⟨x̂²⟩ = (ħ/2mω)(n + (n+1)) = (2n+1)ħ/2mω with no integral performed. The same move gives ⟨p̂²⟩ = (2n+1)mħω/2, and with it the virial split and the uncertainty product. The cost is narrow but real. First, a and a† are not observables — they are not Hermitian, so 'measuring a' means nothing. Second, every step above used [Ĥ, a†] = ħω a†, and that identity holds only for a strictly quadratic Ĥ. Add a cubic term βx̂³ and the commutator picks up pieces in a², a†² and a†a, so a† no longer maps an eigenstate to an eigenstate: levels shift and the spacing closes up as n grows, which is the anharmonicity a real bond shows. What survives is the language — a† and a become the creation and annihilation operators for a quantum of a field, and the ladder index becomes particle number.