Convergence or Asymptotic Behavior Check¶
Convergence diagnostic — instantiates Solvable Baseline Decomposition
Watches the correction terms as orders are added to tell an expansion that is homing in from one that is only asymptotic — and finds the order where truncation is optimal.
Adding more correction terms only helps if the series is actually going somewhere. Convergence or Asymptotic Behavior Check is the diagnostic that watches the sequence of term magnitudes as orders accumulate and asks a structural question: is this expansion converging (terms shrinking without bound, so more orders always help), merely asymptotic (terms shrink for a while, reach a minimum, then grow), or outright diverging (terms growing from the start)? Its defining move is looking at the shape of the term sequence itself — the ratio and trend of successive terms — rather than at the recomposed answer's accuracy against any external truth. For an asymptotic series it identifies the optimal truncation order: the term after which adding more makes things worse, and the smallest error the series can ever deliver. It does not add the terms or decide the physics — it certifies whether the correction ladder is trustworthy to climb and how far.
Example¶
A physicist expands the period of a pendulum at large swing amplitude as a power series in the amplitude — the harmonic (small-angle) result is the baseline, and each correction accounts for more of the pendulum's nonlinearity. The refinement machinery is happy to keep producing terms. This check studies their magnitudes. For moderate amplitudes it sees the terms shrinking steadily, term-to-term ratios comfortably below one: the series is converging, and refinement can safely continue toward the error budget. But as it probes amplitudes near the pendulum's tipping point, it sees the term magnitudes stop shrinking and begin to grow after a certain order — the signature of a series whose radius of convergence has been exceeded. The verdict is not "the answer is wrong" but "past order N, adding terms increases error; truncate at N and the residual there is the best this expansion can do." That single diagnosis prevents the refinement loop from confidently marching a divergent series off a cliff.
How it works¶
- Track term magnitudes. Record the size of each successive correction term and the ratio between consecutive terms as orders are added.
- Classify the behavior. Shrinking-without-bound ⇒ convergent; shrinking-then-growing ⇒ asymptotic; growing ⇒ divergent.
- Locate optimal truncation. For an asymptotic series, find the order of the smallest term — the point of least error and the place to stop.
- Report the floor. State the smallest residual the series can reach and whether that floor is inside the error budget, so downstream steps know the expansion's ceiling on accuracy.
Tuning parameters¶
- Terms examined — how many orders are computed before judging behavior. Too few and a slow onset of divergence is missed; more orders cost computation.
- Ratio threshold — how far below 1 the term ratio must stay to be called convergent. Strict thresholds flag borderline series early; loose ones tolerate marginal shrinkage.
- Truncation rule — whether to stop at the smallest term, one before it, or a conservative margin earlier. Earlier truncation trades a little accuracy for safety against a mis-located minimum.
- Behavior confidence — how many consecutive shrinking (or growing) terms must be seen before committing to a verdict; more evidence resists being fooled by a single anomalous term.
When it helps, and when it misleads¶
Its strength is catching the most seductive failure in the whole method: a series that improves for several orders and feels convergent but is secretly asymptotic, so that a diligent team refining "to higher accuracy" is actually making the answer worse. Many of the most important perturbation series in physics are exactly this kind — the quantum electrodynamics perturbation series is famously asymptotic (divergent) yet superbly accurate when truncated near its optimal order.[1]
Its failure mode is judging too early: a series can shrink for many orders before turning, so a check that looks at only a handful of terms may certify convergence that later fails. It can also be fooled by an anomalous small term that mimics a minimum. And it speaks only to the series' internal behavior — a beautifully convergent expansion can still converge to the wrong answer if the baseline was mischosen, which this check cannot see. The discipline is to compute enough terms to see the trend, truncate conservatively before a suspected minimum, and pair the verdict with an external check against known cases rather than trusting internal convergence alone.
How it implements the components¶
convergence_monitor— its core act: tracking the term sequence's magnitudes and ratios to classify the expansion as convergent, asymptotic, or divergent.residual_error_budget— it reports the smallest residual the series can reach (the asymptotic floor) and whether that floor lies inside the tolerable error, bounding what the expansion can deliver.
It watches the series but does not build it: the refinement_and_recomposition_rule and correction_order_sequence that generate the terms belong to Successive-Order Refinement (its nearest twin, which adds terms to improve the answer where this check only judges whether adding them helps) and First-Order Correction Pass. It also does not map where in parameter space the method fails (validity_range_boundary, Validity Boundary Scan).
Related¶
- Instantiates: Solvable Baseline Decomposition — supplies the verdict on whether the correction series is safe to climb and where to truncate.
- Consumes: Successive-Order Refinement supplies the sequence of correction terms whose magnitudes this check monitors.
- Sibling mechanisms: Zeroth-Order Model Selection · Delta Term Isolation · Dimensionless Small-Parameter Check · First-Order Correction Pass · Successive-Order Refinement · Benchmark Backtest · Validity Boundary Scan · Fallback Trigger Rule · Residual Comparison Test
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Watches the correction terms as orders are added to tell an expansion that is homing in from one that is only asymptotic — and finds the order where truncation is optimal, making its operative form a computation, comparison, model, or analytic representation used to infer, estimate, or choose.
Independent corroboration: The frozen evidence defines Convergence or Asymptotic Behavior Check as 'Watches the correction terms as orders are added to tell an expansion that is homing in from one that is only asymptotic — and finds the order where truncation is optimal', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Mathematics
Origin pattern: Single lineage
Present-day reach: Specialized
Rationale: Asymptotic analysis cohered inspection of successive term magnitudes and ratios to distinguish convergent, divergent, and optimally truncated asymptotic expansions.
Related originating lineages:
- Physics — Perturbative physics, including the famously asymptotic QED expansion, supplied canonical high-stakes applications of optimal truncation.
Review resolution: Successive-term inspection, remainder behavior, and optimal truncation are canonical mathematical asymptotic-analysis methods, with perturbative physics a major application lineage.
Review outcome: Researched adjudication after independent review; high confidence.
Sources consulted:
- NIST Digital Library of Mathematical Functions, Remainder Terms and Stokes Phenomenon
- NIST Digital Library of Mathematical Functions, Sums and Sequences
References¶
[1] An asymptotic series improves as terms are added up to an optimal order, then diverges. Freeman Dyson's 1952 argument that the perturbation series of quantum electrodynamics is divergent (yet extraordinarily accurate when truncated) is the canonical example of why an expansion's internal term behavior, not just its early improvement, must be checked. withdrawn registry ↩