A student makes rectangles from an imaginary 24-unit length of border. They try dimensions 2×10, 4×8 and 6×6. Their areas are 20, 32 and 36 square units. The square seems best—but how do we prove no other permitted rectangle has a larger area, including ones with non-integer side lengths? The answer can come from the arithmetic–geometric mean inequality, often shortened to AM–GM.
This Mathematics in Bukit Timah article connects averages, positive numbers, exact inequalities, equality conditions, fixed-sum products, minimum-perimeter shapes and proof-based optimisation. Every rectangle, fictional classroom display and numerical material budget is an invented Mathematics model, not a surveyed Bukit Timah plot, legal boundary, building specification or actual supplier quotation. The goal is to explain what makes an optimum mathematically guaranteed.
Quick answer: For nonnegative real numbers a and b, (a+b)/2≥√(ab), with equality exactly when a=b. If a rectangle has perimeter 24 units, its two adjacent side lengths x and y satisfy x+y=12. Hence xy≤[(x+y)/2]²=6²=36 square units. Equality requires x=y=6: the square is the unique maximum-area rectangle under these ideal conditions.
Article contents
- What AM–GM actually says
- Why the Bukit Timah application must be hypothetical
- Arithmetic mean is the familiar average
- Worked example 1: compare two means
- A picture of squares can motivate the theorem
- The basic proof starts with a nonnegative square
- Worked example 2: check the equality condition
- A second proof uses an ordinary squared difference
- Worked example 3: test a zero input
- Negative numbers need separate care
- Worked example 4: reject an invalid domain
- The fixed-sum product theorem follows immediately
- Worked example 5: product with sum twelve
- Why checking many pairs is not a proof of maximum
- Worked example 6: a noninteger competitor
- A fixed-perimeter rectangle turns into a fixed sum
- Worked example 7: perimeter twenty-four
- A change of perimeter changes the scale
- Worked example 8: perimeter forty
- Minimum perimeter for a fixed area is the reverse question
- Worked example 9: area thirty-six
- Units must remain consistent
- Worked example 10: convert a mixed-unit rectangle
- Completing the square proves the same maximum
- Worked example 11: verify by a quadratic graph
- Differentiation is another valid route
- Worked example 12: compare the three methods
- Equality conditions identify the actual optimiser
- Worked example 13: exclude square shapes
- Optimisation is always subject to the actual constraints
- Worked example 14: whole-number side lengths
- A fixed product creates a minimum sum
- Worked example 15: product thirty-six
- The expression x plus a reciprocal term fits AM–GM
- Worked example 16: a minimum with no calculus
- A different coefficient changes the equality point
- Worked example 17: compare nearby values
- Arithmetic and geometric means also compare spread
- Worked example 18: equal mean, different geometric mean
- Three-number AM–GM is a generalisation
- Worked example 19: maximum product for three terms
- Equal sharing is optimal for an unweighted product under a fixed sum
- Worked example 20: four positive variables
- A product bound is not a sum bound
- Worked example 21: reject a false universal equality
- Equality in a proof must be feasible
- Worked example 22: a≥9 under fixed sum
- AM–GM can compare a rectangle’s side sum and area
- Worked example 23: expand the squared inequality
- A negative square proves a maximum
- Worked example 24: find the exact product deficit
- Optimisation may involve a fixed three-sided fence
- Worked example 25: maximise with three new sides
- Why equality changes under a weighted physical constraint
- Worked example 26: compare an incorrect square
- The third-side result has a calculus check
- Worked example 27: a complete squared form
- A ratio constraint can defeat the simple equal-input conclusion
- Worked example 28: minimum aspect ratio three
- Positive-variable conditions also exclude boundary collapse
- Worked example 29: no positive minimum area under fixed perimeter
- AM–GM may give a bound in the wrong direction for a question
- Worked example 30: two opposite objectives
- Weighted arithmetic–geometric mean is an extension
- Worked example 31: a one-third weighted mean
- Three-number AM–GM also gives a minimum sum
- Worked example 32: product two hundred sixteen
- A finite list of shapes cannot prove a general optimum
- A mathematical optimum is not necessarily the best real design
- Checking dimensions is part of a proof’s practical meaning
- Worked example 33: compare centimetres and square centimetres
- An optimum might be a bound rather than an attained value
- Worked example 34: exclude equality explicitly
- AM–GM is a bridge between algebra and calculus
- Choosing the right method saves working
- Common error 1: forget nonnegative assumptions
- Common error 2: quote a bound but omit equality
- Common error 3: optimise the wrong quantities
- Common error 4: use geometric mean for unrelated units
- Common error 5: confuse maximum with minimum
- Common error 6: assume a physical optimum from an ideal rectangle
- Primary preparation: compare products with a fixed sum
- Lower Secondary: squares, products and factorisation
- Additional Mathematics and enrichment pathways
- A twenty-minute family diagnostic
- Seven days of practice and retrieval
- Practice questions with checked answers
- Frequently asked questions
- Official context and connected Mathematics in Bukit Timah
What AM–GM actually says
The arithmetic mean of a and b is (a+b)/2. The geometric mean is √(ab) when a,b are nonnegative.
The AM–GM inequality states that the arithmetic mean is at least the geometric mean for these nonnegative inputs.
It does not say the two means are always equal. Equality occurs when a=b, including the zero-zero boundary case if nonnegative numbers are allowed.
Why the Bukit Timah application must be hypothetical
It is easy to imagine garden boundaries and floor plans in Bukit Timah, but the rectangular designs below are ideal mathematical drawings.
Real gardens may have sloping boundaries, planning restrictions, access needs, curves, drainage, material waste and measurement uncertainty. None of those constraints is included unless explicitly stated in a problem.
The results are exact for the fictional optimisation model, not a recommendation about an actual property or construction project.
Arithmetic mean is the familiar average
For positive numbers four and nine, the arithmetic mean is (4+9)/2=6.5.
It lies between the two numbers and represents their equal-weight numerical average. It is a sum-based measure of centre.
The geometric mean instead comes from a product: √(4×9)=√36=6. The two means need not match.
Worked example 1: compare two means
Take a=4 and b=9.
Arithmetic mean is 13/2=6.5. Geometric mean is √36=6.
Thus 6.5≥6, with strict inequality because four and nine are different. The inequality holds as expected.
A picture of squares can motivate the theorem
Take a square with side length (a+b)/2. Its area is [(a+b)/2]².
A rectangle with sides a and b has area ab. The inequality says the square of the average side length is at least the rectangle’s area for the same combined side sum.
This geometric comparison is one reason AM–GM appears naturally in fixed-perimeter design questions.
The basic proof starts with a nonnegative square
For any real a,b≥0, the expression (√a−√b)²≥0.
Expand: a−2√(ab)+b≥0. Rearranging gives a+b≥2√(ab).
Divide by positive two to obtain (a+b)/2≥√(ab). Equality occurs when √a=√b, equivalently a=b.
Worked example 2: check the equality condition
Take a=b=7.
Arithmetic mean is (7+7)/2=7. Geometric mean is √49=7.
They are equal. If we change b to nine, the means become eight and √63, and the geometric mean is smaller. Equality is tied to equal inputs, not just approximately similar numbers.
A second proof uses an ordinary squared difference
For a,b≥0, (a−b)²≥0.
Expand to a²−2ab+b²≥0. Add 4ab to both sides to obtain (a+b)²≥4ab.
Because a+b and √ab are nonnegative, taking square roots gives a+b≥2√ab, hence AM≥GM.
Worked example 3: test a zero input
Take a=0 and b=9.
Arithmetic mean is 4.5. Geometric mean is √0=0. Therefore AM≥GM holds.
Equality does not occur because a≠b. The theorem remains valid for nonnegative inputs, although many optimisation applications restrict side lengths to strictly positive numbers.
Negative numbers need separate care
The real geometric mean √ab may be undefined if ab<0, and the familiar AM–GM inequality is not a theorem for arbitrary signed real pairs.
For a=−1 and b=−4, ab=4 gives geometric mean two, while arithmetic mean is −2.5. The claimed AM≥GM would be false.
The nonnegative-domain condition matters. A memorised inequality applied outside its domain can produce a nonsense optimisation conclusion.
Worked example 4: reject an invalid domain
A pupil tries AM–GM on a=−2 and b=−8.
Their arithmetic mean is −5; √(ab)=√16=4. The inequality −5≥4 is false.
That does not disprove the correctly stated theorem. It shows why both inputs must be nonnegative in the real two-variable AM–GM version.
The fixed-sum product theorem follows immediately
Suppose a,b≥0 with a+b=S fixed, where S≥0.
AM–GM gives √ab≤S/2. Squaring gives ab≤S²/4.
Equality is possible when a=b=S/2. Thus the product of two nonnegative quantities with a fixed sum is largest when they are equal.
Worked example 5: product with sum twelve
Let positive a+b=12.
Their product cannot exceed 12²/4=144/4=36.
The maximum is attained at a=b=6. Pairs such as 4 and 8 give product 32, and 2 and 10 give 20, consistent with the upper bound.
Why checking many pairs is not a proof of maximum
A pupil may tabulate integer pairs (1,11),(2,10),…,(6,6) and see the product rise to 36.
That list covers integer sides, not every positive real pair such as 5.5 and 6.5.
The AM–GM inequality proves the bound for all permissible nonnegative real values at once. Mathematical optimisation needs that general argument.
Worked example 6: a noninteger competitor
Take a=5.5 and b=6.5, still with sum twelve.
Their product is 5.5×6.5=35.75, below 36.
AM–GM establishes the same conclusion without testing any specific decimals. The square remains optimal over the full continuous domain.
A fixed-perimeter rectangle turns into a fixed sum
For a rectangle with side lengths x,y>0, perimeter P=2x+2y.
If P is fixed, then x+y=P/2. The area is A=xy.
By AM–GM, A≤(P/4)²=P²/16. Equality occurs when x=y=P/4, forming a square.
Worked example 7: perimeter twenty-four
Let perimeter P=24. Then x+y=12.
Area satisfies xy≤36, and equality occurs at x=y=6.
Therefore the maximum area is 36 square units. This is a theorem over every rectangle with positive real side lengths and that exact perimeter, not simply the largest of three tested examples.
A change of perimeter changes the scale
If perimeter doubles, the optimal square’s side length doubles.
Area, being a product of two lengths, increases by a factor of four. Thus an optimal square with perimeter 48 has area 144, four times the 36 from perimeter 24.
The relation demonstrates quadratic scaling under similar shapes.
Worked example 8: perimeter forty
If an imaginary rectangular display has total boundary length 40 units, then x+y=20.
AM–GM gives area A=xy≤10²=100 square units.
Equality requires x=y=10, giving a square with perimeter 4×10=40. The result can be checked directly.
Minimum perimeter for a fixed area is the reverse question
Suppose rectangle area A=xy is fixed and positive.
AM–GM gives (x+y)/2≥√(xy)=√A. Therefore x+y≥2√A.
Multiply by two: P=2(x+y)≥4√A. Equality occurs when x=y=√A, again a square.
Worked example 9: area thirty-six
For a rectangle whose area is 36 square units, the perimeter satisfies P≥4√36=24 units.
A 6×6 square reaches the bound. A 4×9 rectangle has the same area but perimeter 2(4+9)=26, which is larger.
Fixed area minimises perimeter at equal sides; fixed perimeter maximises area at equal sides. They are complementary optimisation statements.
Units must remain consistent
AM–GM compares numerical lengths expressed in the same unit. If one dimension is in centimetres and another is in metres, convert before applying a fixed-sum perimeter condition.
Area from x×y has squared length units. Perimeter from 2(x+y) has length units.
A result of 36 from a length model might mean 36 cm², 36 m² or 36 square grid units depending on what the problem states.
Worked example 10: convert a mixed-unit rectangle
An imaginary rectangle has side lengths 100 cm and 2 m.
Convert 100 cm=1 m. Perimeter is 2(1+2)=6 m, and area is 1×2=2 m².
If one multiplied the bare numbers 100×2 and called the result 200 m², the unit inconsistency would invalidate the calculation before any optimisation theorem is considered.
Completing the square proves the same maximum
For fixed x+y=12, write y=12−x. Then A=x(12−x)=−x²+12x.
Complete the square: A=36−(x−6)².
Because (x−6)²≥0, A≤36, with equality at x=6. This agrees exactly with AM–GM and gives an independent algebraic proof.
Worked example 11: verify by a quadratic graph
The area function A(x)=36−(x−6)² is a downward-opening parabola.
Its vertex is at (6,36). The permitted domain 0<x<12 keeps both side lengths positive.
The vertex is inside that domain, so maximum area is 36. This matches the inequality proof with no calculator needed.
Differentiation is another valid route
For A(x)=12x−x², differentiate to get A'(x)=12−2x.
Set the derivative to zero: x=6. The second derivative is −2, so this stationary point is a maximum.
This method agrees with AM–GM, but the inequality proof is often shorter for a fixed sum and product. Choosing a method is part of mathematical judgement.
Worked example 12: compare the three methods
Problem: maximise xy subject to x+y=12 and x,y>0.
AM–GM yields xy≤36. Completing the square yields A=36−(x−6)². Differentiation gives A’=12−2x, zero at x=6 with concave-down graph.
All three show x=y=6 and Amax=36. Their agreement cross-checks the model.
Equality conditions identify the actual optimiser
An inequality such as A≤36 is only an upper bound until we establish whether equality can occur inside the feasible domain.
In the rectangle problem, x=y=6 is permitted and achieves the bound, so 36 is a true maximum.
If an extra restriction excluded x=y, the same bound might be unattainable and the optimisation answer would change.
Worked example 13: exclude square shapes
Suppose x+y=12 but a separate rule requires x≤4, with x>0.
AM–GM still says xy≤36, but equality would need x=6, which is now forbidden.
Area A=x(12−x) increases on 0<x≤4, so its maximum in the restricted domain is at x=4, giving A=4×8=32.
Optimisation is always subject to the actual constraints
A mathematical design may specify a fixed boundary length, a minimum side, a maximum width or an integer number of tiles.
The AM–GM square is optimal only when the equality case is allowed. Additional constraints may shift the optimum to a boundary.
A student should list the permitted domain before declaring an unconstrained equality case to be the final answer.
Worked example 14: whole-number side lengths
If x and y must be positive integers with x+y=13, equality x=y=6.5 is not permitted.
The product is maximised by the closest integers, six and seven, giving 42.
AM–GM gives a real upper bound 13²/4=42.25, but the integer feasibility condition prevents attaining it. Comparing nearest integers finishes the discrete problem.
A fixed product creates a minimum sum
Suppose a,b>0 and ab=K>0.
AM–GM gives a+b≥2√K, with equality at a=b=√K.
This is useful for minimum-cost or minimum-length questions that involve a quantity plus its reciprocal. It is the same theorem viewed from another fixed constraint.
Worked example 15: product thirty-six
Let a,b>0 satisfy ab=36.
Then a+b≥2√36=12. The minimum sum is 12 at a=b=6.
Pairs 3 and 12 give sum fifteen; 4 and 9 give thirteen. The symmetric pair is shortest in combined total.
The expression x plus a reciprocal term fits AM–GM
For x>0, consider f(x)=x+9/x.
Both terms are positive, and their product is x×(9/x)=9, which is constant.
AM–GM gives f(x)≥2√9=6, with equality when x=9/x, or x²=9, so x=3.
Worked example 16: a minimum with no calculus
Minimise x+9/x for x>0.
Apply AM–GM to a=x and b=9/x. The minimum is 2√9=6.
Equality requires x=9/x, so x²=9. With x>0, x=3. Evaluate 3+9/3=6 to verify the bound is achieved.
A different coefficient changes the equality point
Minimise 2x+18/x for x>0.
The product of the two positive terms is (2x)(18/x)=36, constant. Their sum is at least 2√36=12.
Equality occurs when 2x=18/x, so 2x²=18, x²=9 and x=3. The minimum is attained at a positive valid input.
Worked example 17: compare nearby values
For f(x)=2x+18/x, at x=3 the value is 6+6=12.
At x=2, f=4+9=13. At x=6, f=12+3=15.
These examples support the AM–GM result, but the proof covers every positive real x and identifies the exact equality condition.
Arithmetic and geometric means also compare spread
Take fixed arithmetic mean m=(a+b)/2>0 for positive a,b.
As the two inputs become more unequal while keeping their sum fixed, their product decreases below m². The geometric mean correspondingly falls below m.
This is a mathematical relationship about two numbers, not a judgement about educational consistency or a statistic drawn from real pupils.
Worked example 18: equal mean, different geometric mean
Pairs (6,6), (5,7) and (2,10) all have arithmetic mean six.
Their geometric means are 6, √35≈5.916 and √20≈4.472.
The equal pair achieves the maximum geometric mean for the fixed arithmetic mean. The product falls as the two values become more unequal.
Three-number AM–GM is a generalisation
For nonnegative a,b,c, (a+b+c)/3≥(abc)^(1/3).
Equality occurs when all three are equal. This is a general mathematical inequality extending the two-variable case.
The theorem can help optimise products when the sum of three positive quantities is fixed. Its proof requires a suitable general inequality argument, not merely multiplying two-variable statements without care.
Worked example 19: maximum product for three terms
Let a,b,c>0 with a+b+c=18.
Three-variable AM–GM gives (abc)^(1/3)≤18/3=6. Cube both sides to get abc≤216.
The maximum product 216 occurs when a=b=c=6. The equality triple satisfies the sum constraint.
Equal sharing is optimal for an unweighted product under a fixed sum
For n positive quantities with a fixed sum S, general AM–GM gives their product at most (S/n)ⁿ.
Equality requires all n quantities to equal S/n.
This is an exact mathematical statement when all positive real choices are allowed. It does not imply that every real project should divide resources equally; actual objectives and constraints may differ.
Worked example 20: four positive variables
Suppose four positive numbers sum to 20. Their product is at most (20/4)⁴=5⁴=625.
Equality is achieved at 5,5,5,5.
A proposed unequal tuple 2,4,6,8 has product 384, below the bound. The theorem covers every other positive tuple as well.
A product bound is not a sum bound
AM–GM compares a sum-based mean with a product-based mean. It does not say that a+b equals 2√ab.
The equality holds only at equal positive inputs. For unequal values, the arithmetic mean is strictly greater.
Writing equality in place of an inequality discards the essence of the theorem and can produce a false optimiser.
Worked example 21: reject a false universal equality
A pupil claims (a+b)/2=√ab for every positive a,b.
Use a=1,b=9. The arithmetic mean is five; the geometric mean is three.
Since 5≠3, the claim is false. The correct statement is 5≥3, with equality only when a=b.
Equality in a proof must be feasible
Some problems set a+b=12 but also require a≥9. Then equal values six and six are not permitted.
The AM–GM bound ab≤36 remains valid, but the feasible maximum occurs at the closest allowed value to equality, which may be a boundary.
A complete optimisation solution separates universal bound from attainable constrained optimum.
Worked example 22: a≥9 under fixed sum
Let a+b=12 with 9≤a<12 and b=12−a>0.
The area-like product f(a)=a(12−a) is decreasing for a≥9, so its maximum occurs at a=9.
Then b=3 and product is 27. The unconstrained bound 36 is not attainable within this restricted set.
AM–GM can compare a rectangle’s side sum and area
For positive x,y, (x+y)²≥4xy.
This equivalent squared form avoids radicals and can be useful in algebraic proofs.
If x+y=12, then 144≥4xy, so xy≤36. Equality again requires x=y.
Worked example 23: expand the squared inequality
Prove (x+y)²≥4xy for real x,y.
Subtract: (x+y)²−4xy=x²−2xy+y²=(x−y)²≥0.
This identity holds for all real x,y. The AM–GM interpretation as a comparison of nonnegative geometric and arithmetic means still uses nonnegative inputs.
A negative square proves a maximum
For fixed x+y=12, the product is xy=x(12−x)=36−(x−6)².
The term −(x−6)² is nonpositive, so adding it to 36 cannot increase the result.
This proof is another way of writing the same nonnegative-square logic. AM–GM and completing the square share a mathematical core.
Worked example 24: find the exact product deficit
Let x+y=12 and define d=x−6, so y=6−d.
Then xy=(6+d)(6−d)=36−d².
The gap below the maximum is exactly d². If x=5.5, d=−0.5 and the product deficit is 0.25, giving product 35.75.
Optimisation may involve a fixed three-sided fence
Imagine a fictional rectangular display bordering an existing straight boundary, so only three of its sides require new edging.
Let x be the two equal widths perpendicular to the boundary and y the third edged length. A fixed new-edge budget 24 gives 2x+y=24.
The enclosed area is A=xy. This is not the same constraint as the four-side perimeter problem, so the optimal dimensions need not form a square.
Worked example 25: maximise with three new sides
Set u=2x, v=y. Then u+v=24 and A=xy=(uv)/2.
AM–GM gives uv≤(24/2)²=144. Therefore A≤72 square units.
Equality requires u=v=12, so 2x=12 and y=12. Hence x=6, y=12, giving maximum area 72.
Why equality changes under a weighted physical constraint
The three-sided constraint is 2x+y=24. The equal quantities in AM–GM are 2x and y, not directly x and y.
Therefore equality requires 2x=y. The optimal rectangle is twice as long as its width.
A student who imposes x=y by habit will solve a different optimisation problem and obtain a smaller feasible area.
Worked example 26: compare an incorrect square
Under 2x+y=24, suppose x=y. Then 3x=24, so x=y=8 and area is 64.
The actual optimum x=6,y=12 has area 72, which is larger.
Both shapes satisfy the fictional three-sided edging constraint. The equality condition determines why the non-square rectangle is better in this model.
The third-side result has a calculus check
For y=24−2x, A(x)=x(24−2x)=24x−2x².
Differentiate: A'(x)=24−4x, zero at x=6. The second derivative −4 is negative, so the stationary point is a maximum.
Then y=24−12=12 and area 72. This confirms the AM–GM result independently.
Worked example 27: a complete squared form
Rewrite A(x)=24x−2x² as A(x)=72−2(x−6)².
The squared term is nonnegative, so area is at most 72. Equality occurs at x=6, giving y=12.
The inequality, completed-square and derivative proofs all identify the same optimum under the stated domain 0<x<12.
A ratio constraint can defeat the simple equal-input conclusion
Suppose a rectangle must have its length at least three times its width.
For fixed perimeter 24, the AM–GM square solution x=y=6 is not feasible under y≥3x. The optimum must be found within the restricted subset.
One can still use the AM–GM bound as a broad upper bound, but an additional inequality determines the actual optimum.
Worked example 28: minimum aspect ratio three
Let x be width and y=12−x length, with y≥3x and x>0.
Then 12−x≥3x, so x≤3. The area A=x(12−x) increases on 0<x≤3, so its maximum is at x=3.
Then y=9 and A=27, below the unrestricted square’s area 36.
Positive-variable conditions also exclude boundary collapse
For x+y=12 with x,y>0, values x→0 or y→0 make the product approach zero.
A rectangle with a zero side is degenerate and usually excluded from an actual positive-dimension problem.
The maximum at x=y=6 is safely in the permitted interior, but minimum questions over an open positive domain may have only an infimum rather than an attained minimum.
Worked example 29: no positive minimum area under fixed perimeter
Take x+y=12 with x,y>0 and area A=xy.
Choose x very small and y=12−x. Then A approaches zero as x approaches zero from above, but never equals zero for positive x,y.
Thus there is no attained minimum area in this open domain; the infimum is zero. A student should not write “minimum zero” as though a positive-side rectangle reaches it.
AM–GM may give a bound in the wrong direction for a question
For a+b=S, AM–GM supplies an upper bound on ab.
If the task asks to minimise ab over positive a,b with that fixed sum, equality at a=b instead gives the maximum, not the minimum.
Always read the objective: minimise sum given product, or maximise product given sum. These are the natural AM–GM pairings.
Worked example 30: two opposite objectives
Case A: a+b=12, a,b>0. Product has maximum 36 and infimum zero.
Case B: ab=36, a,b>0. Sum has minimum 12 and is unbounded above, because one variable can grow while the other shrinks to preserve the product.
The same inequality describes different extremal directions under the different fixed constraints.
Weighted arithmetic–geometric mean is an extension
For positive a,b and 0≤w≤1, a weighted version states wa+(1−w)b≥aʷb¹⁻ʷ.
Equality occurs at a=b when 0<w<1. The weights need not be equal, but they sum to one.
This can help with asymmetric optimisation problems. It is advanced enrichment and requires attention to positive bases and the precise weights.
Worked example 31: a one-third weighted mean
Take a=8,b=1 and w=1/3.
The weighted arithmetic mean is (1/3)8+(2/3)1=10/3.
The weighted geometric mean is 8¹/³×1²/³=2. Thus 10/3≥2, confirming the inequality for this pair.
Three-number AM–GM also gives a minimum sum
If abc=K>0 and a,b,c>0, the three-number AM–GM theorem gives a+b+c≥3K¹/³.
Equality holds when all three equal K¹/³. This is the natural reverse of the fixed-sum maximum product problem.
The cube-root power is needed because a product of three equal terms is the cube of their common value.
Worked example 32: product two hundred sixteen
Suppose a,b,c>0 and abc=216.
Then sum a+b+c≥3∛216=3×6=18.
Equality occurs at a=b=c=6, which has product 216 and sum 18. A tuple such as 3,6,12 shares product 216 but has sum 21.
A finite list of shapes cannot prove a general optimum
Comparing rectangles 2×10, 4×8 and 6×6 suggests that the equal-sided choice maximises area under a fixed full perimeter.
But infinitely many real-valued side pairs satisfy x+y=12. The proof uses a universal inequality or exact quadratic identity.
This distinction connects to the Bukit Timah induction and proof guide and counterexamples lesson, where evidence and logical proof are treated separately.
A mathematical optimum is not necessarily the best real design
A square minimises perimeter for a fixed rectangular area, but an actual site could require access corridors, ventilation, clearance, material thickness, visibility or a prescribed aspect ratio.
Those factors would change the feasible set and possibly the objective function.
An exact optimisation result should be labelled with the conditions under which it is valid. Our numbers are teaching models, not building or planning advice for Bukit Timah.
Checking dimensions is part of a proof’s practical meaning
The inequality [(x+y)/2]²≥xy compares quantities with dimensions of length squared when x,y are lengths.
Its two sides therefore both represent area units. Comparing a length directly with an area without conversion is meaningless.
Dimensional consistency does not prove an optimisation theorem, but it helps detect errors when applying one to an invented measurement story.
Worked example 33: compare centimetres and square centimetres
Suppose x=4 cm and y=9 cm. Their arithmetic mean is 6.5 cm; geometric mean is 6 cm.
Their product area is 36 cm². The square of the arithmetic mean is 42.25 cm².
The valid squared inequality is 42.25 cm²≥36 cm². Comparing 6.5 cm directly with 36 cm² would mix different dimensions.
An optimum might be a bound rather than an attained value
When a constraint uses strict inequalities or excludes equality, a theorem may establish a limit that cannot be reached.
For instance, positive x+y=12 with x≠y has product strictly less than 36, though it can approach 36 as the sides become arbitrarily close.
The supremum is 36, but no permitted unequal pair attains it. A careful answer distinguishes “upper bound” from “maximum achieved.”
Worked example 34: exclude equality explicitly
Suppose x,y>0, x+y=12 and x≠y.
AM–GM gives xy≤36, but equality needs x=y, which is forbidden. Therefore xy<36 for all feasible pairs.
Choose x=6+ε, y=6−ε for small nonzero ε. Product is 36−ε², approaching 36 from below without reaching it.
AM–GM is a bridge between algebra and calculus
Calculus optimisation searches for stationary points and checks the domain. AM–GM produces certain bounds and equality conditions without differentiation.
Both methods rely on identifying the right variables and constraints. Neither can rescue a model that incorrectly translates a word problem.
The earlier Bukit Timah integration and exact area guide and differentiation lesson develop the calculus side.
Choosing the right method saves working
For x+9/x with x>0, AM–GM identifies the minimum six immediately.
For a problem with several additional constraints or a nonstandard objective, completing the square, graph analysis, inequalities or calculus may be easier.
The tutor should teach method selection, not force AM–GM into every optimisation problem simply because the lesson’s title mentions it.
Common error 1: forget nonnegative assumptions
AM–GM’s usual real form requires nonnegative inputs, and many reciprocal optimisation examples require strict positivity.
Using it on arbitrary negative terms can make the inequality false or the geometric mean undefined.
Write a>0,b>0 or a,b≥0 before applying the intended version.
Common error 2: quote a bound but omit equality
A pupil proves xy≤36 when x+y=12, then concludes “the maximum is 36” without showing whether equality is possible.
Check x=y=6 satisfies the sum and gives product 36. Only then has the maximum been established.
In an additional restricted problem, equality could be disallowed and the bound may be unattainable.
Common error 3: optimise the wrong quantities
For a three-sided fence 2x+y=24, the directly fixed sum is 2x+y, not x+y.
AM–GM applied to 2x and y gives equality 2x=y, leading to x=6,y=12.
Assuming x=y would give area 64 and miss the true maximum 72 under that model.
Common error 4: use geometric mean for unrelated units
An arithmetic mean of one side in metres and another in centimetres is numerically meaningless until units are aligned.
Convert both to the same length unit before calculating a fixed sum or product bound.
Only then interpret the squared result in a consistent area unit.
Common error 5: confuse maximum with minimum
For a+b fixed, the equal pair maximises ab. For ab fixed, the equal pair minimises a+b.
The two statements sound similar but optimise different quantities under different constraints.
Draw or write the objective and fixed condition at the top of the working to keep the direction correct.
Common error 6: assume a physical optimum from an ideal rectangle
A square can be mathematically optimal for fixed perimeter and unrestricted positive sides, but local design constraints may prohibit it.
The remedy is to state the feasibility assumptions and avoid inventing details about real sites.
The answer should say “under the given ideal model” where that qualifier is materially important.
Primary preparation: compare products with a fixed sum
Younger pupils can list whole-number pairs summing to twelve and calculate their products.
They can notice that 6×6 gives more than 4×8 or 2×10. That visual pattern prepares the concept of a maximum.
Formal proof over every real-valued pair can wait until algebra, square roots and inequalities are secure.
Lower Secondary: squares, products and factorisation
Students can derive (a−b)²≥0 and expand it into (a+b)²≥4ab.
They should understand why a square is nonnegative and how dividing by positive constants preserves inequality direction.
These prerequisites make AM–GM more meaningful than memorising an acronym with no proof.
Additional Mathematics and enrichment pathways
The SEAB 2027 G3 Mathematics and Additional Mathematics directory provides official school context.
AM–GM can reinforce optimisation thinking, but general n-variable and weighted AM–GM proofs are enrichment, not a claim about compulsory content for every SEC candidate.
Match the depth to each student’s current algebraic foundations and course scope.
A twenty-minute family diagnostic
Minutes 1–5: Calculate arithmetic and geometric means for four and nine. Explain why 6.5>6.
Minutes 6–10: Prove AM–GM from (√a−√b)²≥0, stating the domain and equality condition.
Minutes 11–15: For fixed perimeter 24, derive x+y=12 and maximum area 36 at x=y=6.
Minutes 16–20: Change the model to three-sided edging 2x+y=24 and find maximum area 72 at x=6,y=12. Ask why the optimal rectangle is no longer square.
Seven days of practice and retrieval
Day one compares arithmetic and geometric means. Day two proves the two-variable inequality. Day three solves fixed-sum product problems.
Day four finds minimum sums with fixed products. Day five applies the rectangle perimeter formula. Day six introduces extra constraints and integer cases.
Day seven uses an unfamiliar fictional design with one additional restriction; the student must identify the feasible set and whether AM–GM equality is attainable.
Practice questions with checked answers
A. For a=4,b=9, find AM and GM. Answer: AM=6.5, GM=6.
B. If a+b=12 with a,b≥0, maximum ab? Answer: 36.
C. A rectangle has perimeter 40. Maximum area? Answer: 100 square units.
D. Rectangle area is 36. Minimum perimeter? Answer: 24 units.
E. For x>0, minimum of x+9/x? Answer: 6 at x=3.
F. For x>0, minimum of 2x+18/x? Answer: 12 at x=3.
G. For a+b+c=18 with positive inputs, maximum abc? Answer: 216.
H. Three-sided edging 2x+y=24, maximum xy? Answer: 72 at x=6,y=12.
Frequently asked questions
What is the arithmetic–geometric mean inequality?
For nonnegative a,b, (a+b)/2≥√ab, with equality when the inputs are equal.
Why does a square maximise area for a fixed rectangular perimeter?
Equal adjacent sides attain AM–GM equality and give the greatest product xy for the fixed sum x+y.
Does AM–GM apply to negative numbers?
The usual real arithmetic–geometric mean inequality requires nonnegative inputs. Applying it to arbitrary negative values can fail.
Is the equality condition optional?
No. An optimal bound is only attained when the equality conditions are satisfied and feasible under the problem’s restrictions.
Why can a three-sided fence have a non-square optimum?
The fixed sum is 2x+y rather than x+y, so AM–GM equality requires 2x=y.
Can an inequality prove an exact optimum without calculus?
Yes. AM–GM and completing-the-square arguments can establish a global bound and equality case directly.
What if the side lengths must be integers?
The continuous equality choice may not be permitted. Check nearby feasible integer values and the relevant constraints.
Is a fixed-sum product minimum also at equal values?
No. Equal inputs maximise the product for a fixed positive sum; positive products can approach zero near a degenerate boundary.
Are these actual Bukit Timah property dimensions?
No. Every perimeter, area and budget is a fictional educational quantity.
How can parents recognise understanding?
The learner can prove the inequality, state its domain, identify the fixed constraint, test equality feasibility and explain why different constraints change the optimiser.
Official context and connected Mathematics in Bukit Timah
Consult the SEAB 2027 G3 Mathematics and Additional Mathematics syllabus directory for current school-level requirements. The AM–GM generalisations here are extensions, not assertions that every proof is examinable at all levels.
Continue with Mathematics in Bukit Timah: Garden Borders, Perimeter and Area, Surface Area and Storage-Box Decisions, Differentiation and Rates of Change and Mathematical Proof and Counterexamples for complementary approaches.
The immutable Clementi Secondary 1 Mathematics tutorial reference anchors first-principles explanations, exact working, misconception diagnosis and independent checks.
The lasting habit is to state the domain and constraints, use an inequality with valid inputs, establish the equality condition and distinguish a theoretical bound from an attainable optimum.
More Mathematics in Bukit Timah from this new four-part batch: Coordinate circle equations, loci and tangency · Second differences, quadratic sequences and nth terms · Complex numbers, Argand diagrams and modulus. Together they build coordinate geometry, algebraic pattern reasoning, proof-based optimisation and higher-Mathematics number systems. Each lesson includes fictional examples, checked working and level-appropriate extensions.