To reconstruct an exponential model P(t) = Ab^t from two positive observations, compare their outputs to find the constant growth or decay factor b, then use either observation to determine the starting value A. If observations are at times t₁ and t₂, with t₁ ≠ t₂, then b^(t₂ − t₁) = P(t₂)/P(t₁). The calculation is exact when the observations fit the assumed model.
For example, suppose a hypothetical quantity is 120 at t = 0 and 405 at t = 3. The three-interval multiplier is 405/120 = 27/8, so the multiplier per interval is 3/2. The model is P(t) = 120(3/2)^t. This guide shows how to handle measurements that do not start at zero, decay models, doubling times and the limits of fitting a curve from just two observations.
Exponential models multiply over equal intervals
A model P(t) = Ab^t uses A as its value at t = 0 and b as the multiplier for one unit of time. Because b is positive, the output stays positive when A > 0.
The defining feature is a constant ratio over equal time steps. If b = 3/2, the quantity is multiplied by 1.5 each unit; it rises by 50% from the preceding value. If b = 1/2, it halves at each step.
This differs from a linear model, which adds the same amount per unit of time. The distinction can be tested with consecutive observations: an exponential progression 120, 180, 270 adds 60 then 90, yet multiplies by the same 1.5 each step.
Recognising the model type matters before algebra begins. Two points can always be fitted by a suitable positive exponential when the times differ and outputs are positive, but that fit alone does not prove a real process follows exponential growth.
Why two observations can determine A and b
There are two unknown positive parameters in P(t) = Ab^t. Two observations P(t₁) = p₁ and P(t₂) = p₂ give p₁ = Ab^t₁ and p₂ = Ab^t₂.
Divide the second equation by the first. The common A cancels, leaving p₂/p₁ = b^(t₂ − t₁). Provided t₂ ≠ t₁ and both measurements are positive, this relation determines a unique positive b.
Once b is known, substitute into either observation to recover A = p₁/b^t₁. Check both observations in the resulting formula.
This is a useful general inverse problem: the shape of the change comes from comparing observations, while the overall scale comes from anchoring one observation. A student who guesses the starting value from the first stated data point can be wrong when that point does not occur at t = 0.
The complete worked model with t = 0 and t = 3
Suppose P(0) = 120 and P(3) = 405. Since P(0) = Ab⁰ = A, the first observation immediately gives A = 120.
Use the second observation: 405 = 120b³. Divide to obtain b³ = 405/120 = 27/8. The unique positive cube root is b = 3/2.
Therefore the reconstructed model is P(t) = 120(3/2)^t. It represents a 50% increase per unit interval, provided the exponential assumption remains appropriate.
Check P(3): 120 × (27/8) = 405. Check P(0): 120 × 1 = 120. Both match the observations. Exact fractions make this verification cleaner than prematurely rounding 1.5 or the final model coefficients.
What the multiplier predicts between the observations
For the reconstructed model, P(1) = 120(3/2) = 180 and P(2) = 120(9/4) = 270. The sequence from t = 0 to t = 3 is 120, 180, 270, 405.
Each step multiplies the previous value by 3/2. The increases are 60, 90 and 135, so the increments are not constant. This is why the same two observations should not be connected with a straight-line growth assumption without additional justification.
Between observed times the model supplies interpolated predictions. For example, at t = 1.5, P(t) = 120(3/2)^1.5, which is well defined as a positive real number.
These predictions belong to the mathematical model, not independently measured facts. An educational explanation should distinguish what the two observations actually establish from what the chosen model assumes between and beyond them.
A second example with neither observation at zero
Suppose P(2) = 270 and P(5) = 911.25. The times differ by three units, so divide the measurements: P(5)/P(2) = 911.25/270 = 27/8.
Therefore b³ = 27/8 and b = 3/2. Now recover the initial value from the observation at t = 2: 270 = Ab² = A(9/4).
Thus A = 270 × 4/9 = 120. The model is again P(t) = 120(3/2)^t.
The first listed observation was not P(0). If a student mistakenly sets A = 270, the result would fail the t = 5 check. This is an important diagnostic: the value of A is tied to the model’s time origin, not to which data point appears first in the question.
The direct formula for unequal observation times
With positive observations p₁ and p₂ at different times t₁ and t₂, the multiplier is b = (p₂/p₁)^[1/(t₂ − t₁)]. The formula follows directly by taking the appropriate positive root of b^(t₂ − t₁) = p₂/p₁.
Then A = p₁/b^t₁. These formulas work for integer, fractional or other real observation times if the same continuous exponential model is intended.
For school calculations, substituting the observations and simplifying the ratio before taking powers is usually more transparent than copying an unfamiliar formula mechanically.
Also check that the two times are not identical. A ratio divided by a zero time difference does not identify a per-unit multiplier. Two different readings at exactly the same time would instead conflict with the idea of a single deterministic P(t) under the given data.
A decay model uses the same algebra
Suppose a hypothetical amount starts at 800 and becomes 100 after three equal time intervals. Let P(t) = Ab^t with A = 800.
The later observation gives 100 = 800b³, so b³ = 1/8. Therefore b = 1/2 and P(t) = 800(1/2)^t.
The quantity is multiplied by one-half each time unit. The predictions are P(1) = 400, P(2) = 200 and P(3) = 100.
Nothing in the reconstruction technique changes between growth and decay. The difference is the multiplier: b > 1 means growth, 0 < b < 1 means decay, and b = 1 means the model remains constant. The base should not be negative in the usual real-valued exponential model.
Why a negative base is not an appropriate default
In the standard positive exponential model, b > 0. This allows b^t to be defined as a real, continuous expression for arbitrary real t.
A negative base such as (−2)^t is not real for every real exponent. Even at integer times it would alternate sign, which is not the ordinary constant-percentage growth or decay model.
If all observed quantities represent positive population counts, masses or concentrations, choose positive A and b within the mathematical modelling assumptions.
The condition b > 0 matters when solving a ratio equation with an even time difference. For example, b² = 9/4 has algebraic square roots ±3/2, but the exponential base in this model is positive, so select b = 3/2.
The model can include a constant case
If two positive observations at different times have exactly the same output, their ratio is 1. Then b^(t₂ − t₁) = 1, which with b > 0 implies b = 1.
The fitted model is P(t) = A, a constant function. It does not describe growth or decay even though it still fits the algebraic family Ab^t.
For example, P(1) = 50 and P(4) = 50 lead to b = 1 and A = 50. Both observations are reproduced.
The student should not divide by ln b when b = 1 in a doubling-time calculation: ln 1 = 0, and a truly constant positive amount never doubles from its starting value at a finite time. Recognise this special case before using logarithmic time formulas.
Percentage change is based on the previous value
For b = 3/2, the quantity grows by 50% of its current amount per time unit. The multiplier is 1 + 0.50 = 1.50.
For b = 1/2, the amount decays by 50% of its current amount per time unit. The multiplier is 1 − 0.50 = 0.50.
It is incorrect to say that a model with multiplier 1.5 adds 50 units or adds half of its initial value every step. From 120, the first increase is 60, but the next increase is 90.
Percentages in repeated growth or decay depend on the evolving amount. The model P(t) = Ab^t encodes that multiplicative relationship precisely, whereas an additive formula describes a different process.
Two observations do not prove the exponential assumption
The model P(t) = Ab^t can exactly pass through two positive measurements at distinct times. But a linear, quadratic or many other functions can also be constructed to pass through the same two points.
More observations are needed to evaluate whether equal-step ratios remain approximately constant. In real data, measurement noise may mean no single two-parameter exponential exactly matches every observation.
The mathematical reconstruction therefore assumes the exponential form and determines the parameters under that assumption. It should not be presented as proof that a real-world phenomenon always grows exponentially.
This distinction is valuable for students working with applications. A model is an explicit hypothesis about structure that can be tested against further observations and revised when the evidence does not support it.
How to find doubling time
For a growth model P(t) = Ab^t with b > 1, doubling time T satisfies P(T) = 2P(0). Therefore Ab^T = 2A, so b^T = 2.
Taking logarithms gives T = ln 2/ln b. For b = 3/2, the doubling time is ln 2/ln(3/2), about 1.710 time units.
This is not necessarily a whole number of steps. The continuous formula permits fractional time, whereas a strictly discrete problem may ask for the first whole interval by which the amount has at least doubled.
Read the time-unit convention carefully. If t measures days, T is in days. If t measures years, the same numerical time represents years.
Half-life is a related but different quantity
For a decay model with 0 < b < 1, half-life T satisfies P(T) = P(0)/2, so b^T = 1/2.
Hence T = ln(1/2)/ln b. Both logarithms are negative, so the ratio is positive.
For b = 1/2, T = 1 interval exactly. For b = 4/5, the half-life is ln(1/2)/ln(4/5), approximately 3.106 time units.
A student should not reuse the growth doubling-time formula without considering what the target ratio is. The general relation for changing from P(0) to qP(0) is T = ln q/ln b when b > 0 and b ≠ 1, provided the target is meaningful for the intended model.
A threshold question is an inverse problem
Suppose the reconstructed growth model is P(t) = 120(3/2)^t, and the question asks when it first reaches 600 in continuous time.
Set 120(3/2)^t = 600, so (3/2)^t = 5. Taking logarithms gives t = ln 5/ln(3/2), approximately 3.969.
If the question instead asks for the first whole-number time t at which P(t) is at least 600, test the adjacent integers. At t = 3 the amount is 405, and at t = 4 it is 607.5, so the first whole time is t = 4.
The continuous crossing time and the first whole interval are related but distinct answers. Do not round a threshold time to the nearest integer automatically; the wording and inequality determine the correct whole-number result.
Model outputs can be checked without logarithms
When a follow-up asks for P(4) under P(t) = 120(3/2)^t, simply substitute t = 4. We get P(4) = 120 × 81/16 = 607.5.
Logarithms are not needed because the exponent is known. They become useful when the exponent is unknown, as in solving P(t) = 600 for t.
This method-selection habit matters under timed conditions. A student who takes logarithms for every exponential calculation introduces extra steps and opportunities for rounding error.
First identify what is unknown: if the output is sought for a given t, evaluate the model; if t is sought for a given output, invert the exponential using logarithms after confirming the sign conditions.
The logarithm requires a positive ratio
To solve b^t = q with a positive exponential base, the right side q must be positive. In a positive-value model, q represents the target amount divided by A.
For example, P(t) = 120(3/2)^t can never equal −10 or zero for any finite real t. Its output is positive for every real t.
Attempting t = ln(−10/120)/ln(3/2) would take the logarithm of a negative number and has no real meaning in this setting.
Before using logarithms, verify that the target amount and ratio are positive and that b ≠ 1 where a logarithmic inversion is needed. Domain awareness is part of solving an exponential equation correctly.
How changing the time origin affects A
A is the value at the chosen origin t = 0, not necessarily the first physically recorded measurement. If time is relabelled, the numerical parameter A changes even when the underlying model describes the same process.
For P(t) = 120(3/2)^t, suppose we introduce s = t − 2 so that s = 0 corresponds to the old t = 2. Then t = s + 2 and P = 120(3/2)^(s+2) = 270(3/2)^s.
The new initial value in s-coordinates is 270, while the multiplier remains 3/2. Both formulas describe the same quantities at corresponding times.
This clarifies a common misconception: the exponent’s zero point is part of the model’s definition. The ratio b is tied to the chosen time unit, while A is tied to the chosen time origin.
Changing the time unit changes the multiplier
Suppose t measures years in P(t) = 120(3/2)^t. If we introduce m = 12t to count months, then t = m/12.
The same model becomes P(m) = 120[(3/2)^(1/12)]^m. The multiplier per month is (3/2)^(1/12), not 3/2.
This is why comparing b-values from different time units without conversion is misleading. A 50% annual multiplier is not a 50% monthly multiplier.
The model remains mathematically identical after conversion if the exponent is adjusted consistently. For assessment questions, state units beside each time variable and interpret the multiplier per unit of that variable.
Use an observation-ratio check
Once a model has been reconstructed, compare ratios over the actual time spacing. For P(t) = 120(3/2)^t, the ratio P(5)/P(2) must equal (3/2)^3 = 27/8.
Indeed P(2) = 270 and P(5) = 911.25. Their ratio is 3.375, exactly 27/8.
A student may calculate a correct initial value A but an incorrect b if they divide observations without accounting for the number of time intervals between them. A three-unit change produces b³, not b.
The ratio check exposes this problem quickly. The exponent in the observation ratio is the difference in the time coordinates, even if neither observation occurs at the origin.
What to do if the measurements cannot fit a positive exponential
If the problem states A > 0 and b > 0 but one measured output is zero or negative, no model of the form Ab^t can fit that observation at a finite real time.
Likewise, two different positive outputs recorded at the exact same time cannot both be reproduced by one deterministic function P(t). Additional interpretation or measurement-error treatment would be needed.
These are mathematical consistency checks, not reasons to force negative bases or undefined logarithms into the solution.
A strong learner states the assumptions and recognises when the given evidence violates them. Correct modelling includes knowing when there is no admissible parameter pair under the stated model.
Practice questions and answers
Question one: P(0) = 120 and P(3) = 405. Reconstruct P(t) = Ab^t. Answer: P(t) = 120(3/2)^t.
Question two: P(2) = 270 and P(5) = 911.25. Answer: the same model, P(t) = 120(3/2)^t.
Question three: P(0) = 800 and P(3) = 100. Answer: P(t) = 800(1/2)^t.
Question four: in the first model, find P(4). Answer: 607.5.
Question five: find the first non-negative whole t with 120(3/2)^t ≥ 600. Answer: t = 4.
Question six: find the doubling time for b = 3/2. Answer: ln 2/ln(3/2), approximately 1.710 time units.
For each, identify the observed times, divide to obtain the multiplier over their actual separation, then recover the origin value.
How tutoring can distinguish the first wrong decision
A child who sets A equal to P(2) may not understand the time origin. Another who calculates b = P(5)/P(2) may have missed the three-interval exponent. Another may reconstruct correctly but fail when asked for doubling time because the logarithmic inversion is insecure.
These are different prerequisite gaps. Teaching should target the exact conceptual step, then use a changed example to check transfer rather than repeating the opening numbers.
At eduKateSG, the maximum three-student tutorial format near Sixth Avenue MRT, generally with 1.5-hour weekly lessons subject to suitable placement, supports close inspection of each student’s method choices.
The goal is a model that the student can explain and test, not merely a fitted equation copied from a worked example.
Syllabus guidance and connected reading
Exponential functions, logarithms and models support a useful progression from algebra to applications. Specific modelling extensions and assessment types depend on the student’s school topic order and examination pathway.
Consult the appropriate SEAB lists for 2026 O-Level Additional Mathematics 4049, 2027 SEC G3 K341 or 2027 SEC G2 K232, as appropriate.
Read the linked guides How to Find the First Whole Day an Exponential Threshold Is Crossed, How to Spot a Hidden Quadratic in an Exponential Equation and How to Track Domain Restrictions Through Composite Functions. Explore the Additional Mathematics archive or contact eduKate Singapore about small-group tutorials.
Properly taught kids shine a bright light into the future.