Singapore School Mathematics Operating Manual · Chapter 38
A decision can be correct and still be fragile.
If a maximum capacity is 100 and the calculated requirement is 99.9, the answer is technically feasible under an exact model. But a small measurement error could make it fail. If the requirement is 60, the same decision has much more room.
This chapter develops margin, slack and robustness. Margin asks how far a result lies from a boundary. Slack measures unused capacity or remaining room inside a constraint. Robustness asks whether the conclusion survives reasonable changes in inputs, assumptions or uncertainty.
1. A boundary divides decision regimes
If x≤100 is required, x=100 is the boundary.
Values below are feasible; values above are not.
The closer x lies to 100, the smaller the decision margin.
2. Slack is unused room inside a constraint
If budget is $500 and spending is $430, slack is $70.
If capacity is 40 and occupancy is 37, slack is 3 places.
Slack converts a yes/no feasibility result into a quantitative measure of safety.
3. Positive slack indicates interior feasibility
For constraint x≤10, define slack as 10−x.
x=7 gives slack 3.
x=10 gives slack 0.
x=12 gives negative slack −2, signalling violation.
4. Zero slack marks an active constraint
At x=10, the constraint x≤10 is exactly binding.
Any increase would violate it.
Active constraints are often where optimal solutions or regime changes occur.
5. Distance from failure is more informative than pass/fail alone
Two designs both satisfy x≤100.
Design A gives x=99.8; Design B gives x=80.
Both pass, but B has a much larger margin.
When input uncertainty exists, that distinction can be decisive.
6. Uncertainty should be compared with the margin
Suppose x=99.8±0.5 under threshold 100.
The uncertainty interval reaches 100.3, so feasibility is not guaranteed.
If x=80±0.5, the entire interval remains safely below 100.
7. A robustness margin is a buffer against perturbation
If a conclusion remains unchanged after all plausible input changes, it is robust.
The size of the admissible perturbation before failure is a practical robustness margin.
8. Margin can be absolute or relative
Absolute margin from 100 for x=90 is 10.
Relative margin can be expressed as 10/100=10% of the threshold.
Relative measures help compare systems operating at different scales.
9. Direction matters
For an upper-bound constraint x≤100, smaller x increases slack.
For a lower-bound requirement x≥75, larger x increases margin.
Margin should be defined relative to the direction of failure.
10. Two-sided constraints have two margins
If 20≤x≤30 and x=24, lower margin is 4 and upper margin is 6.
The nearest failure boundary is 4 units away.
The minimum of the two margins can act as a simple robustness measure.
11. Feasible regions generalise slack
For x≥0, y≥0 and x+y≤10, a point inside the triangle has room before crossing one or more boundaries.
A point on an edge has zero slack for that active constraint.
12. Optimisation often drives slack to zero
If profit increases with production and production is capped by capacity, the maximum may occur at full capacity.
The capacity constraint becomes active.
This is not universally true, but active constraints frequently identify optimum boundaries.
13. A solution at a boundary may be mathematically valid but operationally delicate
If a model uses rounded inputs and the optimum sits exactly at capacity, tiny unmodelled effects can cause violation.
The mathematical answer should therefore be interpreted with its margin.
14. Discrete decisions have step margins
A bus holds 40 passengers.
For 79 passengers, two buses have one spare place.
For 80, slack is zero.
For 81, the decision jumps to three buses.
One passenger changes the regime because the previous margin was exhausted.
15. Rounding can consume hidden margin
If a quantity reported as 9.9 actually lies in an interval extending above 10, the visible central value may suggest slack that does not truly exist.
Use the full uncertainty range near boundaries.
16. Percentage thresholds need a reference base
If a score must exceed 80%, raw-mark slack depends on the total possible marks.
On a 50-mark paper, 80% is 40 marks. On a 100-mark paper, it is 80.
Normalisation and margin must use the same reference.
17. Graphical margin can be seen as vertical or horizontal distance
In y=f(x), a threshold y=T can be drawn as a horizontal line.
The distance between the curve and threshold at a chosen x gives visual decision margin.
Where the curve approaches the threshold, robustness shrinks.
18. Root problems have sign margins
If f(x) is far from zero, small perturbations may not change the sign.
If f(x) is near zero, sign classification is fragile.
This is useful in numerical root finding and inequality checks.
19. Classification margins appear in geometry
An angle of 40° is comfortably acute.
An angle of 89.9° is acute but close to the 90° boundary.
Uncertainty matters far more in the second case.
20. Classification margins appear in statistics too
If a rule labels values above a cutoff as “high”, a result barely above the cutoff has low classification margin.
Labels near arbitrary thresholds should not be interpreted as fundamentally different states unless the context justifies it.
21. A robust optimum is not always the numerically largest optimum
Two solutions may have nearly equal objective values.
One may sit deep inside the feasible region while the other relies on several tight constraints.
If robustness matters, the slightly lower objective can be preferable in real applications.
School Mathematics can introduce the idea without making a real-world recommendation.
22. Margin supports prioritised checking
If a final value is far from every boundary, a quick check may be enough.
If it is close to a threshold, preserve precision, inspect assumptions and propagate uncertainty carefully.
23. Slack can identify which constraint matters
Suppose three constraints have slack 20, 8 and 0.
The zero-slack constraint is active and deserves attention first.
The 20-unit slack constraint is currently less restrictive.
24. A change in one constraint can shift the active owner
If capacity expands, another constraint may become the new bottleneck.
Decision structure depends on which boundary is closest or active.
25. Margin and sensitivity should be considered together
A large margin can still be fragile if the output is extremely sensitive to uncertain inputs.
A small margin can be acceptable if inputs are known exactly and the model is exact.
Robustness depends on both distance to failure and response to perturbation.
26. Error budget should fit inside decision margin
If the margin is 0.5 but total uncertainty can be ±1.0, the decision is not guaranteed.
If uncertainty is ±0.05, the same margin is much safer.
The Error Budgets chapter quantifies this relationship.
27. Redundancy is most valuable near low margins
Close calls deserve independent verification because a small error can flip the conclusion.
Far from boundaries, the same checking effort may have lower value.
28. Margin can be designed, not only measured
In an abstract optimisation exercise, one may choose a solution satisfying constraints with extra room.
This deliberately creates robustness.
29. A margin audit
Ask:
What boundary defines failure? How far is the result from it? Is the margin absolute or relative? Which constraint has the least slack? Does uncertainty fit inside the margin? Is the output sensitive? Would a small perturbation change the decision? Is a zero-slack result being treated with appropriate caution?
30. Independent practice
1. For x≤50 and x=42, find slack.
2. For x≥75 and x=81, find margin above threshold.
3. For 20≤x≤30 and x=27, find both margins.
4. A result is 99.7±0.5 under upper threshold 100. Is feasibility guaranteed?
5. A bus holds 40 passengers. Find slack with 2 buses carrying 78 passengers.
6. Why is x=100 under constraint x≤100 less robust than x=80?
7. If total uncertainty is ±0.2 and margin is 3, what qualitative conclusion follows?
8. Why should a low-margin answer receive stronger checking?
31. Worked answers
1. 50−42=8.
2. 81−75=6.
3. Lower margin=7; upper margin=3.
4. No. The interval extends to 100.2.
5. Capacity=80, so slack=2 places.
6. It has zero slack; any increase violates the constraint.
7. The uncertainty is much smaller than the margin, so the threshold classification is robust under that uncertainty.
8. A small error or perturbation can change the decision.
32. Continue through Batch 10
Use Redundancy and Fault Tolerance to strengthen close decisions, Residuals and Model Mismatch when observed performance sits outside the expected margin, and Evidence Sufficiency and Verification Depth to decide when further checking no longer changes the conclusion.
Return to the BTT Mathematics Hub for Batch 10.