Topology Convergence Minimum

A convergence claim becomes useful only after it identifies the tests that define the limit. Norm convergence controls every unit-vector input uniformly; strong operator convergence controls each fixed input; weak operator convergence controls each fixed matrix element. These statements are different in infinite dimension, and choosing the wrong one can turn a valid approximation into an unjustified prediction about spectra, energies, or dynamics.

Helpful background. Sequences, Series, and Convergence supplies elementary limiting arguments. Functional Analysis Essentials explains why boundedness and completeness matter.

Vector limits and operator limits

For vectors, $\psi_n\to\psi$ in norm means $\|\psi_n-\psi\|\to0$. Weak convergence means $\langle\phi,\psi_n\rangle\to\langle\phi,\psi\rangle$ for every fixed $\phi$. An orthonormal sequence $e_n$ converges weakly to zero because the coefficients of any fixed Hilbert vector tend to zero, but $\|e_n\|=1$ rules out norm convergence to zero.

For bounded operators on one Hilbert space, the three basic tests are

$$\begin{aligned} \text{norm:}&\quad\|B_n-B\|\to0,\\ \text{strong:}&\quad\|(B_n-B)\psi\|\to0,\\ \text{weak:}&\quad\langle\phi,(B_n-B)\psi\rangle\to0. \end{aligned}$$

The strong test must hold for every fixed $\psi$, and the weak test for every fixed pair $\phi,\psi$. Norm implies strong, and strong implies weak. The converses fail. These definitions apply to sequences here; general operator-algebra topology may require nets rather than sequences when metrizability or separability is unavailable.

Finite-rank truncation gives strong but not norm convergence

Let $P_n$ project onto $\operatorname{span}\{e_1,\ldots,e_n\}$ in $\ell^2(\mathbb N)$. For each fixed $\psi=\sum c_je_j$,

$$\|(I-P_n)\psi\|^2=\sum_{j>n}|c_j|^2\longrightarrow0.$$

Thus $P_n\to I$ strongly. Nevertheless, $\|(I-P_n)e_{n+1}\|=1$ and $\|I-P_n\|=1$ for every $n$. The input exposing the error changes with $n$, which strong convergence does not forbid. This is a direct warning against interpreting basis-cutoff convergence for a fixed state as a uniform guarantee for all states.

For a fixed compact operator $K$, strong convergence of these projections does imply $\|(I-P_n)K\|\to0$. Compactness makes the image of the unit ball approximable by a finite set, allowing pointwise control to become uniform there. This additional hypothesis is why compact spectral problems behave better under truncation than general continuous-spectrum problems.

Weak convergence does not preserve operator products

Let $S$ be the unilateral shift, $Se_j=e_{j+1}$. Then $S^n\to0$ weakly: matrix elements vanish first on finite sequences and then on all vectors by density and the bound $\|S^n\|=1$. Its adjoints also converge weakly to zero. Yet

$$(S^n)^*S^n=I$$

for every $n$. Multiplying the weak limits would incorrectly give zero. Strong limits preserve products when both factors converge strongly, because uniform boundedness controls one factor in the usual difference estimate. Strong convergence alone, however, need not be preserved by taking adjoints; the strong-star topology asks separately for convergence of operators and adjoints.

The same example cautions against equating weak vector convergence with convergence of all probabilities: $e_n\rightharpoonup0$ does not imply $\langle e_n,Ie_n\rangle\to0$. A matrix element with both vectors varying is not one of the fixed tests in weak convergence.

Unbounded Hamiltonians are compared through resolvents

An operator-norm difference $\|H_n-H\|$ may be meaningless when domains differ. For self-adjoint operators on a common Hilbert space, strong resolvent convergence means

$$(H_n-zI)^{-1}\psi\longrightarrow(H-zI)^{-1}\psi$$

for every $\psi$, for a nonreal $z$; the self-adjoint resolvent identities then give the corresponding convergence at other nonreal points. Norm resolvent convergence strengthens this to operator-norm convergence. Resolvents are bounded, with norm at most $1/|\operatorname{Im}z|$, so these tests avoid subtracting operators on incompatible domains.

Strong resolvent convergence controls bounded continuous spectral functions on each fixed vector, including the associated unitary evolutions. It does not supply convergence of unbounded energy moments without extra bounds. Spectral projections converge strongly at intervals whose endpoints carry no limiting spectral mass; an endpoint eigenvalue can obstruct the naive statement. Precise results and their proofs belong to the spectral and resolvent treatments, not to a blanket rule that all spectral data converge.

Exercise: a matrix-element check with an operator norm surprise

Let $Q_n=|e_n\rangle\langle e_n|$. Determine its strong and norm limits. Then explain why the same sequence, regarded as density operators, does not converge to zero in trace norm.

Solution

For fixed $\psi$, $\|Q_n\psi\|=|\langle e_n,\psi\rangle|\to0$, so the strong limit is zero. But $\|Q_n\|=1$ and $\|Q_n\|_1=\operatorname{Tr}Q_n=1$. There is no convergence to zero in either of these norms. A sequence can therefore lose its apparent mass under fixed-vector tests without tending to a normalized limiting state. Operational comparison of density operators uses the appropriate state topology, such as trace norm, not strong operator convergence alone.

References

• Hall, Brian C., Quantum Theory for Mathematicians, Graduate Texts in Mathematics 267, Springer, 2013, doi:10.1007/978-1-4614-7116-5 — Hilbert-space and spectral convergence distinctions.
• Teschl, Gerald, Mathematical Methods in Quantum Mechanics: With Applications to Schrödinger Operators, 2nd ed., Graduate Studies in Mathematics 157, American Mathematical Society, 2014, author’s edition — weak and strong convergence, compactness, and strong and norm resolvent convergence.