Successive-Order Refinement¶
Iterative refinement — instantiates Solvable Baseline Decomposition
Climbs the correction ladder order by order, recomposing baseline plus accumulated terms and stopping when the residual falls inside its error budget — or when adding orders stops paying.
One correction is rarely the whole story. Successive-Order Refinement is the loop that goes beyond the leading term: it adds the second-order correction, then the third, recomposing the baseline plus all accumulated terms at each step and checking how much closer to the target each addition brings the result. Its defining move is a budgeted stopping rule: it does not add orders until they run out, it adds them until the remaining error drops inside a declared budget — or until the marginal gain from one more order no longer justifies the cost and risk. Where the first-order pass delivers one term and halts, this mechanism owns the ladder and the decision of how high to climb. The output is a recomposed answer at a chosen order, carrying an explicit statement of the residual it still holds. Its whole discipline is knowing when to stop refining, so the correction series improves the answer instead of degenerating into an ever-growing pile of terms.
Example¶
A climate group starts from a solved baseline: a zero-dimensional energy-balance model where incoming solar radiation equals outgoing thermal radiation, giving a single equilibrium temperature. That is order zero. Successive-Order Refinement then climbs. First correction: add radiative forcing from raised CO₂, recompute equilibrium. Second: add the water-vapor feedback, which amplifies the first. Third: add the ice-albedo feedback. At each order it recomposes — baseline temperature plus the accumulated feedback corrections — and compares the remaining discrepancy against paleoclimate and observational constraints, which set the error budget. After the major feedbacks are in, the next candidate correction (a small higher-order cloud-feedback adjustment) moves the answer by less than the residual budget requires, and its own uncertainty is larger than its effect. The refinement stops there: the group reports the temperature at that order with its residual band, rather than chasing terms whose uncertainty exceeds their contribution.
How it works¶
- Add the next order. Compute the correction at the next order and recompose it with the baseline and all prior corrections.
- Measure the shrink. Assess how much the new term reduced the residual — the gap between the current recomposed answer and the target's tolerance.
- Check against budget. Compare the remaining residual to the declared error budget; if it fits, stop.
- Watch marginal gains. If successive terms stop shrinking the residual meaningfully (or start growing), halt even if the budget is not met — that is a signal to escalate, not to keep climbing.
Tuning parameters¶
- Error budget — how small the residual must become before refinement stops. A tight budget forces more orders and more cost; a loose one stops early and accepts a coarser answer.
- Maximum order — a hard ceiling on how high to climb regardless of budget, guarding against runaway refinement of a series that will not converge.
- Marginal-gain threshold — how little a new term may contribute before it is judged not worth adding. Set it high to stop early; low to squeeze out every improvement.
- Recomposition bookkeeping — whether each order is kept as a separately labeled term or merged into a running total. Separate terms preserve interpretability and the baseline's visibility; merging is compact but opaque.
When it helps, and when it misleads¶
Its strength is disciplined progression: it turns "keep making the model more realistic" into a bounded climb with an explicit stopping rule, so the answer improves in a way you can audit and defend — and it stops before the corrections outgrow their value.
Its failure mode is correction pileup: assuming that more orders always mean a better answer. Higher-order terms can grow rather than shrink, and adding them makes the result worse — the phenomenon that Runge's classic example shows for high-degree interpolation, where piling on terms produces wilder error, not less.[n1] The related misuse is refining past the point where each term's own uncertainty exceeds its contribution, manufacturing false precision from noise. The discipline is the budgeted, marginal-gain-watching stopping rule: refine only while terms are still shrinking the residual toward a declared target, and treat a series that stops shrinking as a reason to check convergence or escalate — not to add another order.
How it implements the components¶
refinement_and_recomposition_rule— it iteratively recomposes baseline plus accumulated corrections, adding orders while keeping each contribution traceable.residual_error_budget— it defines the acceptable remaining error and uses it as the stopping rule that decides how high to climb.
It consumes rather than establishes the ordering: the correction_order_sequence and the leading term are set by First-Order Correction Pass, its nearest twin — that pass computes one leading correction, this mechanism climbs the higher-order ladder above it. It does not diagnose whether the series converges (convergence_monitor, Convergence or Asymptotic Behavior Check) nor test the recomposed answer against known cases (benchmark_case_set, Benchmark Backtest).
Related¶
- Instantiates: Solvable Baseline Decomposition — supplies the higher-order recomposed answer and the budgeted stopping decision.
- Consumes: First-Order Correction Pass supplies the ordering and the leading correction this mechanism builds higher orders upon.
- Sibling mechanisms: Zeroth-Order Model Selection · Delta Term Isolation · Dimensionless Small-Parameter Check · First-Order Correction Pass · Convergence or Asymptotic Behavior Check · Benchmark Backtest · Validity Boundary Scan · Fallback Trigger Rule · Residual Comparison Test
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Successive-Order Refinement operates as an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution because it climbs the correction ladder order by order, recomposing baseline plus accumulated terms and stopping when the residual falls inside its error budget — or when adding orders stops paying.
Independent corroboration: The frozen evidence defines Successive-Order Refinement as 'Climbs the correction ladder order by order, recomposing baseline plus accumulated terms and stopping when the residual falls inside its error budget — or when adding orders stops paying', so its operative form is Analysis, Modeling & Optimization.
Nearest alternative: Protocol, Workflow & Routine — Successive-Order Refinement includes features of a repeatable ordered procedure or handoff sequence that coordinates action, but its defining operation is an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Mathematics
Origin pattern: Single lineage
Present-day reach: Specialized
Rationale: Accumulating correction terms order by order is the classical perturbation and asymptotic-expansion method.
Related originating lineages:
- Computer Science & Software Engineering — Computer science and software-engineering practice supplies a parallel or contributing lineage for the mechanism's defining operation: climbs the correction ladder order by order, recomposing baseline plus accumulated terms and stopping when the residual falls inside its error budget — or when adding orders stops….
- Engineering & Design — Numerical error budgets determine useful truncation.
- Physics — Perturbation series approximate systems around solvable baselines.
Review resolution: The blind reviewers agree that mathematics is the primary origin and differ only on alternate origin disagreement, encyclopedia synthesis disagreement. I preserve every independently explained alternate from both records rather than imposing a numeric cap. I retain single_lineage because the combined evidence shows one traceable formative lineage. The broader reach of specialized records portability separately from historical provenance; encyclopedia_synthesis=true preserves the affirmative synthesis judgment where either reviewer identified one.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] Runge's phenomenon — increasing the degree of a polynomial interpolant on equally spaced points can make the approximation oscillate and worsen rather than improve — is the cautionary archetype for correction pileup: more orders do not guarantee a better answer, which is why a budgeted stopping rule matters. ↩