Singapore School Mathematics Operating Manual · Chapter 29
Complex Mathematics questions rarely become easy because the learner suddenly sees the whole answer at once.
More often, the question becomes manageable because it is decomposed.
A large target is split into smaller mathematical jobs. Each job is solved under clear assumptions. The result of one job becomes the input to another. At the end, the partial results are recomposed into one complete answer.
This architecture appears everywhere in Singapore school Mathematics. A Primary problem may require finding one unit before reconstructing the whole. A Secondary algebra problem may require forming an equation before solving it. A geometry problem may require proving similarity before calculating a missing length. A trigonometric problem may require finding an angle before finding area. A statistics question may require cleaning data before calculating a mean. A JC problem may require differentiation before optimisation, or parameter identification before integration.
The central operating rule is simple:
do not solve the whole problem as one undifferentiated block when the mathematics naturally contains smaller jobs.
This chapter develops four linked skills: decomposition, subproblem control, interface checking and recomposition.
Decomposition · Subproblems · Interfaces · Recomposition · Practice · Worked answers
1. Decomposition means replacing one large target with smaller solvable targets
Suppose a question asks for the cost of tiling a rectangular floor.
The final target is cost.
But cost may depend on the number of tiles.
The number of tiles may depend on floor area and tile area.
The floor area may depend on missing dimensions.
The missing dimensions may depend on an earlier perimeter condition.
The large problem can therefore be decomposed into:
dimensions → floor area → tile count → total cost.
Each arrow marks a mathematical dependency.
2. Decomposition is not the same as doing random intermediate calculations
A weak solution may calculate every visible quantity “just in case”.
A strong decomposition asks which intermediate values are necessary for the final target.
The difference is purpose.
Subproblems should exist because they unlock the target, not because the numbers are available.
3. The final target determines the decomposition
Consider the same rectangle.
If the target is area, only the dimensions may be needed.
If the target is diagonal length, the area may be irrelevant.
If the target is perimeter, neither diagonal nor area is necessary.
The same givens can produce different dependency trees depending on what is asked.
4. Decomposition can be forward or backward
Forward decomposition starts from the givens and asks what useful quantities can be produced.
Backward decomposition starts from the target and asks what must be known immediately before the target can be found.
The strongest problem solving often uses both.
Backward planning identifies required subgoals. Forward execution then derives them from the givens.
This links directly to Backward Reasoning and Target Decomposition.
5. A good subproblem has a clear input, operation and output
For example:
Input: two points A and B.
Operation: midpoint formula.
Output: midpoint M.
That output can then become the input for another subproblem, such as finding a perpendicular bisector.
Clear subproblem boundaries make long working easier to audit.
6. Subproblems should be as independent as possible
If two quantities can be found independently, calculate and verify them separately before combining them.
Suppose total revenue = adult revenue + child revenue.
Adult revenue and child revenue can be solved as separate branches.
If one branch contains an error, the other remains intact.
This is mathematically similar to modular design in engineering: isolate failures where possible.
7. A subproblem can be algebraic
Suppose a rectangle has perimeter 50 cm and length is 5 cm more than width.
Subproblem 1: find dimensions.
Let width = w and length = w + 5.
2w + 2(w + 5) = 50.
4w + 10 = 50, so w = 10 and length = 15.
If the final target is area, subproblem 2 uses these dimensions: A = 150 cm².
8. A subproblem can be geometric
Suppose a larger triangle contains a smaller triangle with one pair of parallel sides.
Before calculating a missing length, the first subproblem may be proving the triangles similar.
Once similarity is established, the second subproblem uses corresponding side ratios.
The proof is not decoration. It is the interface certificate that makes the ratio step legal.
9. A subproblem can be representational
Suppose a word problem is difficult in prose.
Subproblem 1 may be translating it into a table, equation, diagram or graph.
Only then does the computational subproblem begin.
Representation itself can be a mathematical task.
10. A subproblem can be a domain check
Before solving log(x−3)=2, establish x>3.
This condition is not a final answer, but it constrains the later algebra and filters candidates.
Domain checking is therefore a legitimate subproblem.
11. A subproblem can be a classification step
In optimisation, first identify stationary points.
Then classify them.
Then compare with endpoints if the domain is closed.
The final global optimum is a recomposed conclusion from several candidate classes.
12. Primary Mathematics already uses decomposition deeply
A unitary-method problem may ask for the cost of 7 items when 4 cost $28.
Subproblem 1: cost of 1 item = $7.
Subproblem 2: cost of 7 items = $49.
The “find one unit” move is a canonical decomposition pattern.
13. Ratio problems often decompose through one-part value
If red:blue = 3:5 and total is 64, then total ratio parts = 8.
Subproblem 1: one part = 64/8 = 8.
Subproblem 2: red = 24, blue = 40.
The decomposition creates a stable reusable interface: the value of one ratio part.
14. Percentage problems decompose through the correct base
If a price rises from $80 to $100:
Subproblem 1: change = 20.
Subproblem 2: identify original base = 80.
Subproblem 3: percentage increase = 20/80 ×100% =25%.
Separating these jobs prevents the common mistake of using the final value as the base.
15. Coordinate geometry naturally decomposes
To find a perpendicular bisector:
Subproblem 1: midpoint.
Subproblem 2: gradient of original segment.
Subproblem 3: perpendicular gradient.
Subproblem 4: line equation through midpoint.
Each step has a distinct mathematical object.
16. Trigonometry often uses chained decomposition
A problem may ask for area of a non-right triangle while giving incomplete information.
Subproblem 1: calculate a missing side using sine or cosine rule.
Subproblem 2: calculate the included angle.
Subproblem 3: use area = 1/2 ab sin C.
The danger is that an early error propagates through every later stage.
This makes interface checking essential.
17. An interface is the handoff from one subproblem to another
If subproblem A outputs “radius = 5 cm”, subproblem B may use that radius to compute area.
The value 5 cm is the interface.
Interfaces should carry both magnitude and meaning.
Writing only “5” can invite unit or role confusion.
18. Interfaces should carry units
If a subproblem outputs time = 30 minutes but the next formula expects hours, the interface is not ready.
Convert 30 minutes to 0.5 hours before handoff.
Unit conversion belongs at the boundary between subproblems when required.
This connects to Mathematical Type Checking.
19. Interfaces should carry domain restrictions
Suppose u = x−2 and later u² = 9.
If the original problem requires x>2, then u>0.
The interface is not merely u=±3 from the squared equation. The active restriction must travel forward.
20. Interfaces should carry exactness status
If a subproblem gives θ≈37.2°, the next stage should not silently treat 37.2° as exact.
Keep sufficient precision or exact form until the final rounding stage where possible.
Approximation status is part of the interface contract.
21. Interfaces should carry proof status
If a geometry ratio depends on two triangles being similar, the similarity must be established before the ratio is used.
A guessed visual similarity is not a valid interface.
The handoff should occur only after the prerequisite has been proved or given.
22. Interfaces can expose where an error entered
Suppose a final answer is wrong.
Rather than re-reading every line, inspect the handoff values between subproblems.
If midpoint is correct but perpendicular gradient is wrong, the fault lies in one module rather than the whole solution.
Decomposition therefore improves diagnosis as well as solving.
23. A useful interface can be checked independently
After finding a side length from trigonometry, ask whether it satisfies triangle inequalities.
After finding a probability branch, ask whether it lies in [0,1].
After finding a mean, ask whether it lies within the data range.
After finding an equation, substitute a known point.
Independent checks at interfaces stop errors from propagating.
24. Error propagation makes early interfaces high-value checkpoints
If a quantity is used in five later calculations, an error in it infects all five.
A quantity used only in the last line has less downstream influence.
Therefore the most important interface checks are often early, central handoffs.
25. Linked question parts are formal subproblem interfaces
Part (a) may establish a result explicitly intended for part (b).
The examination has already decomposed the problem for the learner.
Recognising that structure prevents redundant recalculation and clarifies how marks are staged.
26. “Hence” is an interface signal
When a question says “hence”, it often tells the learner that a previous result should become an input to the next step.
Ignoring the handoff can lead to a longer or disallowed route.
27. Reusing a result does not remove the need to understand its meaning
If part (a) gives gradient m=−2, part (b) may ask for a perpendicular line.
The learner must know that perpendicular gradient is 1/2, not merely copy −2 forward.
An interface provides data, not the next operation automatically.
28. Recomposition combines solved subproblems into the final answer
Suppose adult revenue is $420 and child revenue is $280.
Recomposition gives total revenue = $700.
This stage can be mathematically simple, but it must match the final target.
A problem can have every subproblem correct and still fail if the results are recombined incorrectly.
29. Recomposition requires compatible types
You can add adult revenue and child revenue because both are monetary totals.
You cannot add area and perimeter.
You cannot average probabilities and lengths without a defined model.
Recomposition is subject to the same type constraints as every other operation.
30. Recomposition should preserve the final unit
If the final target is cost, the answer should be in currency.
If it is an angle, use degrees or radians as appropriate.
If it is probability, the result should be dimensionless and lie in [0,1].
If it is a count, whole-number interpretation may be required.
31. Recomposition can require choosing among candidates
Suppose two algebraic roots survive an equation but only one satisfies the geometry.
The final recomposition is not “both”.
It is the intersection of algebraic candidates with the original contextual constraints.
32. Recomposition can require adding mutually exclusive branches
In probability, the event “one red and one blue” may be decomposed into RB and BR.
If these branches are mutually exclusive, recomposition adds their probabilities.
Different subproblem structures imply different recomposition operations.
33. Recomposition can require inclusion-exclusion
If two counted categories overlap, simple addition double-counts the intersection.
The correct recomposition is |A∪B|=|A|+|B|−|A∩B|.
The interface between counting branches must record whether branches overlap.
34. Recomposition can require optimisation across candidate classes
For a closed-interval maximum:
Subproblem 1 finds stationary points.
Subproblem 2 evaluates stationary-point values.
Subproblem 3 evaluates endpoints.
Recomposition compares all candidate values and selects the greatest.
No single subproblem owns the global answer.
35. Recomposition can fail through duplicate use of one result
If the same region is included in two area subproblems and both areas are added, the overlap is double-counted.
If the same probability path appears in two branches, the total is inflated.
If the same solution is generated by symmetric cases, list it once as a solution value unless paths themselves are being counted.
36. Recomposition should answer the wording, not merely display intermediate values
A question asks for minimum number of buses.
The arithmetic gives 3.2.
The final recomposition includes the discrete decision: 4 buses.
Returning to context is part of the final assembly.
37. A complete recomposition should not lose assumptions
If a result was derived under x>0, the final statement should not present it as valid for all real x.
If a model assumes constant speed, the final interpretation should remain conditional on that model.
If a geometric theorem required parallel lines, the final conclusion cannot survive if that condition is removed.
38. Decomposition can create a dependency graph rather than a chain
Some problems branch.
To find the equation of a circle, centre and radius may be found independently.
To find total probability, several event branches may be calculated separately.
To optimise a function, stationary-point and endpoint branches may run in parallel.
The architecture is therefore often a directed graph rather than a straight list of steps.
39. Some subproblems can be solved in parallel
If two branches do not depend on each other, their order does not matter mathematically.
This flexibility can be useful under examination time pressure.
A learner can solve the easier branch first and return to the harder one later, provided the recomposition requirements remain clear.
40. Other subproblems have strict dependency order
You cannot compute a tangent equation before knowing the point and gradient.
You cannot calculate a scale factor before identifying corresponding sides.
You cannot classify an algebraic candidate before generating it.
Some interfaces impose a hard solving order.
The next Batch 08 chapter develops that dependency architecture explicitly.
41. Decomposition reduces cognitive load
A six-step problem feels difficult partly because six relationships are competing for working memory.
By isolating one subproblem, the learner temporarily reduces the active state.
The completed result is then compressed into a labelled interface value.
This is one reason clear working can improve reasoning rather than merely presentation.
42. Labels are memory compression
Writing “midpoint M=(5,9)” is more useful than leaving the coordinate pair unexplained.
The label stores semantic meaning with the number.
Later, the learner can retrieve the mathematical role without reconstructing how the value was obtained.
43. Diagrams can externalise decomposition
A geometry diagram separates regions, labels known values and shows dependency relationships spatially.
A bar model can separate ratio parts.
A probability tree separates branches.
A table separates variables and cases.
Good representations make decomposition visible.
44. Algebraic substitution is a decomposition device
For x⁴−5x²+4=0, let u=x².
The quartic becomes u²−5u+4=0.
Subproblem 1 solves for u.
Subproblem 2 reconstructs x from each u.
The substitution converts one difficult object into a simpler intermediate object.
45. Completing the square is decomposition of structure
x²−6x+13 becomes (x−3)²+4.
The transformation separates the variable square from the constant floor.
This exposes minimum value and axis of symmetry.
Representation change can therefore decompose hidden structure even when no new variable is introduced.
46. Proofs can be decomposed into obligations
To prove two triangles congruent by SAS:
subproblem 1 proves one side equality;
subproblem 2 proves the included angle equality;
subproblem 3 proves the second side equality;
recomposition invokes SAS.
The theorem acts as the final interface rule.
47. Existence-and-uniqueness claims decompose naturally
To prove exactly one solution exists:
subproblem 1 proves at least one solution exists;
subproblem 2 proves no two distinct solutions can exist.
Recomposition yields uniqueness.
This is a logical decomposition rather than a numerical one.
48. Contradiction proofs also have architecture
Subproblem 1 assumes the negation of the target.
Subproblem 2 combines that assumption with accepted facts.
Subproblem 3 derives an impossibility.
Recomposition rejects the assumption and establishes the target.
49. Statistics problems often decompose into data preparation and analysis
Before computing a mean, data may need unit standardisation or frequency interpretation.
Before fitting a model, variables may need to be identified and plotted.
Before comparing distributions, relevant summary measures must be selected.
Mixing cleaning, calculation and interpretation into one step increases error risk.
50. Modelling problems decompose into world → mathematics → world
Subproblem 1: identify variables and assumptions.
Subproblem 2: construct the mathematical model.
Subproblem 3: solve the model.
Subproblem 4: interpret the result in context.
Subproblem 5: evaluate whether the model remains valid.
The entire modelling cycle is a decomposition architecture.
51. Examination questions often reward visible decomposition
Method marks commonly correspond to meaningful intermediate stages.
A completely opaque calculator answer can hide valid reasoning from the marker.
Showing key subproblems and interfaces makes the mathematical route legible.
This is not about producing excessive prose. It is about exposing the structure that earns credit.
52. Over-decomposition can become inefficient
Not every arithmetic line needs to become a named subproblem.
Breaking a simple task into ten artificial stages increases friction.
Good decomposition follows genuine mathematical boundaries: different quantities, different methods, different domains, different candidate classes or different proof obligations.
53. Under-decomposition creates hidden coupling
If a learner tries to perform unit conversion, substitution, equation solving and interpretation in one calculator entry, a mistake becomes hard to locate.
Separating these stages makes each easier to verify.
54. The best granularity depends on learner fluency
A beginner may need more explicit subproblems.
An expert can compress familiar routines into one mental unit.
But even experts reopen compressed modules when a problem becomes unstable or unfamiliar.
Decomposition is therefore adaptive.
55. Error diagnosis should follow the dependency graph backward
If the final answer is wrong, inspect the immediate inputs to the final step.
If one input is wrong, inspect what produced it.
Continue backward until the first incorrect interface is found.
This is more efficient than restarting from the first line every time.
56. Repair should target the broken module
If the error is in ratio-part calculation, there is no need to reteach percentage interpretation.
If the error is in unit conversion, the algebra method may already be sound.
Precise decomposition supports precise remediation.
57. Mixed practice tests whether decomposition can be generated independently
A learner may follow a decomposed worked example successfully while still being unable to create the decomposition in a new problem.
Transfer requires changed numbers, changed wording, changed diagrams and delayed practice.
The learner must learn to identify the hidden subproblem architecture, not merely imitate one worksheet format.
58. A practical decomposition audit
Before solving a complex problem, ask:
What is the final target?
What quantities must be known immediately before it?
Which of those can be solved independently?
What conditions or proofs must be established before each handoff?
What units, domains and approximation status must travel across interfaces?
How will the partial results be recombined?
What intermediate results deserve independent checking because many later steps depend on them?
59. Independent practice
1. A rectangle has perimeter 46 cm and length 3 cm more than width. Decompose the problem and find its area.
2. Adult:child tickets are in ratio 3:5 and 96 tickets are sold. Adult tickets cost $12 and child tickets $7. Decompose and find total revenue.
3. A line segment joins A(2,4) and B(8,10). Decompose the steps needed to find its perpendicular bisector.
4. A right triangle has hypotenuse 13 cm and one leg 5 cm. The final target is area. Identify the necessary subproblem and solve.
5. Solve x⁴−5x²+4=0 by decomposition.
6. A bag contains 3 red and 2 blue counters. Two are drawn without replacement. Decompose the event “one red and one blue” into branches and find its probability.
7. A measured time is 30 minutes, but a speed formula requires hours. What interface operation is required?
8. A geometry subproblem gives θ≈37.24°. Why should the interface not simply become θ=37° before the next stage?
9. A final optimisation compares one stationary point and two endpoints. What is the recomposition operation?
10. Explain how decomposition helps locate an error in a six-step solution.
11. Give one example of two subproblems that can be solved in parallel.
12. Give one example of a hard dependency where one subproblem must be completed before the next can begin.
60. Worked answers
1. Let width=w and length=w+3. From 2w+2(w+3)=46, 4w+6=46, so w=10 and length=13. Recompose into area: 10×13=130 cm².
2. Total parts=8, so one part=96/8=12. Adults=36 and children=60. Adult revenue=36×12=$432. Child revenue=60×7=$420. Total revenue=$852.
3. Find midpoint M=((2+8)/2,(4+10)/2)=(5,7). Find gradient AB=(10−4)/(8−2)=1. Perpendicular gradient=−1. Use point-gradient form through M: y−7=−(x−5), so y=−x+12.
4. First find missing leg: b²=13²−5²=169−25=144, so b=12 cm. Then area=1/2×5×12=30 cm².
5. Let u=x². Then u²−5u+4=0, so (u−1)(u−4)=0 and u=1 or4. Reconstruct x: x=±1 or±2.
6. Branches are RB and BR. P(RB)=(3/5)(2/4)=3/10. P(BR)=(2/5)(3/4)=3/10. Recompose by addition because the branches are mutually exclusive: total=3/5.
7. Convert 30 minutes to 0.5 hours before passing the value into the speed formula.
8. Premature rounding may create avoidable downstream error. Preserve more precision until the final required rounding stage.
9. Evaluate the objective at all three candidates and compare them to choose the global maximum or minimum required.
10. Check the handoff values between subproblems. Once the first incorrect interface is found, the fault is localised without restarting every correct earlier stage.
11. In total revenue, adult revenue and child revenue can be calculated independently after the ticket counts are known.
12. A tangent equation cannot be formed until its point and gradient are known.
61. Continue through Batch 08
Next, use Dependency Order, Solving Sequence and Mathematical Workflows to distinguish hard prerequisites from flexible parallel branches. Then use Normalisation, Rescaling and Reference Values when variables should be compared on a common scale, and Sampling, Resolution and Aliasing when sparse observations may hide underlying behaviour.
Return to the BTT Mathematics Hub for Batch 08 after the batch is completed.