No exact finite-dimensional CCR

In plain terms

Here is the problem. Position and momentum obey the most famous relation in quantum mechanics: measuring them in opposite orders differs by a fixed universal amount. Everything characteristic of the theory -- uncertainty above all -- flows from it. Now suppose the world has only finitely many distinguishable states, as CSD says. Can the relation survive?

No, and the proof fits in a sentence. Add up the diagonal of both sides. The left side is a difference of two products in opposite orders, and such a difference always sums to zero. The right side is a nonzero constant spread along the diagonal, which sums to something nonzero. Zero cannot equal nonzero.

So any finite theory owes an account of where the relation fails. Not whether -- where.

In CSD

The corpus pays this debt exactly rather than apologising for it. On the truncated oscillator the relation holds perfectly on every energy level except the very top one, where it fails by a single rank-one defect -- of precisely the magnitude the trace argument demands. The defect is not noise to be minimised; it is the exact compensation the counting requires, pushed into the one place a finite ladder must end.

For any state a laboratory can prepare -- negligible population at the ceiling -- the departure is invisible. And the result is stated CSD-free on purpose: the obligation binds every finite-dimensional reconstruction, so a rival programme claiming finiteness inherits the same debt whether it has noticed or not.

Mathematically

No finite matrices satisfy [Q,P] = c times identity with c nonzero: the trace of a commutator vanishes, the trace of c times identity is c times the dimension. Corollary: the constant cannot be i h-bar in finite dimension.

The companion construction gives the sharp positive: a conjugate pair on the N-level oscillator whose commutator is exactly i everywhere except the top Fock level, where a rank-one defect of magnitude N sits -- exactly what makes the trace vanish.