Notes on Rates of Change: From Averages to Exponential Approximations

Uneven data spacing and the pitfall of inferring concavity from raw increments

• We examine a population data set with f(t) representing population (in thousands) over time t (years).
• Reported differences between entries are not uniform in time:
  • The first two entries differ by 22,000 (i.e., Δf = 22).
  • The next pair differs by 60,000 (Δf = 60).
  • The following change is 18 (interpreted as +18, i.e., 18,000).
• Pitfall: If you just compare successive increases (e.g., 22 vs 60), you might wrongly infer concavity, but this ignores uneven time gaps.
• Key lesson: compute the average rate of change over each time interval to compare growth more fairly.

Average rate of change (ARC) on intervals

• General formula:
  \text{ARC}_{[a,b]} = \frac{f(b) - f(a)}{b - a}.
• Example intervals (using years as given, populations in thousands):
  • From 1992 to 1994 (t = 2 to t = 4):
    \text{ARC}_{[1992,1994]} = \frac{f(1994) - f(1992)}{4 - 2} = \frac{215 - 193}{2} = 11.
  • Interpretation: the population increased by an average of 11 thousand per year over these two years.
  • From 1994 to 2005 (t = 4 to t = 15):
    \text{ARC}_{[1994,2005]} = \frac{f(2005) - f(1994)}{15 - 4} = \frac{275 - 215}{11} = \frac{60}{11} \approx 5.45.$
  • Interpretation: averaged growth 5.45 thousand per year over 11 years (a slower rate than the prior interval).
  • From 2005 to 2012 (t = 15 to t = 22):
    \text{ARC}_{[2005,2012]} = \frac{f(2012) - f(2005)}{22 - 15} = \frac{f(2012) - 275}{7}.
  • Note: the last interval’s slope depends on the unknown f(2012); it shows how ARC depends on interval length.
• Takeaway: Uneven spacing can hide the true pattern; parsing ARC on each interval is essential to understand growth behavior.

Graphical interpretation: tangent vs secant and the role of concavity

• Secant line: the line connecting two data points; its slope is the average rate of change on that interval.
  • Example: slope of the secant between (1992, f(1992)) and (1994, f(1994)) is 11 (thousand per year).
  • Slope of the secant between (1994, f(1994)) and (2005, f(2005)) is about 5.45 (thousand per year).
• Tangent line: line with slope f'(a) at a point a; used to approximate near a.
• Concavity effects (for a concave down function):
  • Tangent line tends to overestimate values of f between nearby points.
  • Secant line tends to underestimate values of f between the two end points.
• Practical note: two natural approximation methods exist for estimating f' near a specific x-value; their behavior depends on the local concavity of f.

Estimating the instantaneous rate f'(15) from data around x = 15 (the year 2005)

• Data near x = 15 (year 2005): 1994 (t=4) to 2005 (t=15):
  • ARC on [1994,2005] ≈ 5.45 per year; this is a natural estimate for f'(15) (the instantaneous rate around 2005).
• Two reasonable estimation approaches discussed:
  1) Use the interval containing t = 15 to approximate the instantaneous rate: take f'(15) ≈ 5.45.
  2) Use a tangent-line approximation at t = 15 by drawing a line through (15, f(15) = 275) with slope f'(15) ≈ 5.45; use it to project nearby values.
• Tangent-line prediction for f(14):
  f(14) \approx f(15) - f'(15) \cdot (14 - 15) = 275 - 5.45 \approx 269.55.
• Another possible approach: use the secant slope from another nearby interval to estimate f'(15) (e.g., slope from 1992 to 2005): m_{secant}^{1992-2005} = \frac{f(2005) - f(1992)}{15 - 2} = \frac{275 - 193}{13} = \frac{82}{13} \approx 6.31.
  • If used, f(14) would be approximately 275 - 6.31 ≈ 268.69.
• Note from teaching perspective: problems often prefer a clearly defined method (ARC on a nearby interval, then a tangent-line extension) rather than open-endedJustification; both are reasonable, but a test may specify a method.
• A quick takeaway value often cited: using the interval [1994,2005] gives a reasonable f'(15) ≈ 5.45 thousand per year, yielding f(14) around 269.5 thousand (about 269–270 thousand).

Secant vs tangent interpretations and how they relate to concavity

• If the actual function is concave down (f'' < 0) near the interval around x = 15:
  • The tangent line at x = 15 will overestimate f for x ≠ 15 in the neighborhood.
  • The secant line across the interval will underestimate f for x between the endpoints.
• If the function were concave up (f'' > 0): invert the roles—tangent would underestimate, secant would overestimate between endpoints.
• In practice, you often don’t know the exact concavity from sparse data, so you rely on the two approximations (tangent and secant) to bracket the true function locally.
• This underscores how local curvature (second derivative) matters for how well linear approximations fit beyond the immediate point.

Local extrema and points of inflection (second derivative concepts)

• What happens when f' changes sign (i.e., f' crosses through zero):
  • If f' changes from positive to negative at c: local maximum at c (the graph rises, then falls).
  • If f' changes from negative to positive at c: local minimum at c (the graph falls, then rises).
  • If f' changes sign with a cusp (f' undefined at c): possible sharp corner; standard smooth calculus assumptions may fail there.
• Sign of the second derivative f'':
  • f''(x) > 0: graph is concave up (cup-shaped).
  • f''(x) < 0: graph is concave down (cap-shaped).
  • Inflection point: a point where concavity changes, i.e., f'' changes sign.
• Extrema vs inflection points:
  • Local extrema are critical in the sense of f' = 0 and sign changes of f'; inflection points are about changes in concavity (f'': sign change).
• Preview: local maxima/minima are discussed in depth in later chapters; the focus here is recognizing how f' and f'' guide where the graph goes up, down, and how it bends.
• Practical modeling takeaway: inflection points and extrema give structural features to search for when fitting or interpreting models to data.

The relative rate of change and average relative change

• Relative rate of change (instantaneous):
  • For a function y = f(x), the relative rate of change is
    \frac{1}{y} \frac{d y}{d x} = \frac{f'(x)}{f(x)}.
  • This is the derivative of ln f(x):
    \frac{d}{dx} \ln f(x) = \frac{f'(x)}{f(x)}.
• Relative change over a finite interval (average relative rate of change):
  • A common form (depending on which base value you use) is
    \text{Average relative change over } [a,b] \approx \frac{f(b) - f(a)}{f(a) (b - a)}.
  • An alternative (using the end value in the base) gives a similar idea but with f(b) as the base.
• Example intuition: if a quantity grows by 50% over 5 years, the naive average relative rate is
  \frac{0.50}{5} = 0.10 = 10\% \text{per year},
  but actual growth with compounding would be larger if the rate stayed constant at 10% yearly for five years; the instantaneous rate varies with the model.
• In exponential models y = e^{k x}, the relative rate of change is the constant k:
  \frac{1}{y} \frac{dy}{dx} = k.
• General approach to estimate an exponential rate from data: if you know f(a) and f(b) with a < b, you can compute an average relative change or fit a k such that
  f(b) \approx f(a) e^{k (b-a)} \quad\Rightarrow\quad k \approx \frac{1}{b-a} \ln \left(\frac{f(b)}{f(a)}\right).
• Relative-rate perspective as a bridge between linear and exponential models: it provides a natural way to talk about growth without committing to a specific global model.

Exponential approximation (tangent exponential) and comparison to tangent line

• Given a point a with known value f(a) and derivative f'(a), define the instantaneous relative rate at a:
  k = \frac{f'(a)}{f(a)}.
• Exponential (tangent exponential) model that matches value and relative rate at x = a:
  f(x) \approx f(a) \, e^{k (x - a)}.
• Linear (tangent line) model that matches slope at x = a:
  y \approx f(a) + f'(a) (x - a).
• Example: let a = 4, f(a) = 100, f'(a) = 5.
  • Then the tangent line is y_t(x) = 100 + 5 (x - 4).
  • The exponential tangent is with k = \frac{f'(a)}{f(a)} = \frac{5}{100} = 0.05, so
    f(x) \approx 100 \, e^{0.05 (x - 4)}.
• Near x = 4, the tangent line and the tangent exponential are both good approximations and agree closely with the original function at that point (they’re both tangent there).
• When extrapolating further from the point, the two models may diverge depending on whether the true growth is more linear near the point or truly exponential; the choice of model depends on the context and data.
• Real-world note: exponential growth assumptions have limits; the logistic model is a common alternative for growth that saturates due to limited resources, discussed in later chapters; it’s important to keep model choice aligned with data and context.

Connections and practical takeaways

• For any new concept, relate it to four perspectives: function, data, graph, and formula to build intuition (as the course emphasizes).
• Sparse or uneven data require careful interpretation; do not over-interpret any single slope or pattern without considering interval lengths and spacing.
• When predicting or extrapolating, be explicit about which model you’re using (tangent line vs exponential) and acknowledge potential errors due to curvature (second derivative) not being known far from the data.
• The material ties together crucial ideas: first derivatives (growth/decline), second derivatives (concavity and inflection), and relative rate concepts that connect linear and exponential descriptions of change.

Summary of key formulas (recap)

• Average rate of change over [a,b]:
  \text{ARC}_{[a,b]} = \frac{f(b) - f(a)}{b - a}.
• Tangent line at a:
  y = f(a) + f'(a) (x - a).
• Secant slope between two points (a, f(a)) and (b, f(b)):
  m_{sec} = \frac{f(b) - f(a)}{b - a}.
• Concavity and second derivative: concave up if f''(x) > 0; concave down if f''(x) < 0; inflection where f'' changes sign.
• Local extrema criteria: if f' changes sign from + to - at c, local max; if from - to +, local min.
• Relative rate of change (instantaneous):
  \frac{1}{y} \frac{dy}{dx} = \frac{f'(x)}{f(x)} = \frac{d}{dx} \ln f(x).
• Average relative change over [a,b]:
  \text{Average relative change} \approx \frac{f(b) - f(a)}{f(a) (b - a)}.
• Exponential tangent model (matching point and relative rate):
  f(x) \approx f(a) \, e^{\left( \frac{f'(a)}{f(a)} \right) (x - a)}.
• Example values used in the dataset (illustrative):
  • ARC on [1992,1994]: = \frac{215 - 193}{2} = 11. (thousand per year)
  • ARC on [1994,2005]: = \frac{275 - 215}{11} = \frac{60}{11} \approx 5.45. (thousand per year)
  • ARC on [2005,2012]: = \frac{f(2012) - 275}{7}.$$ (depends on f(2012))
• Example extrapolation with f'(15) ≈ 5.45k/year:
  • f(14) ≈ 275 - 5.45 = 269.55 (thousand)
• Example with a secant-based slope around 1992-2005: 82/13 ≈ 6.31 (thousand per year) which would give f(14) ≈ 275 - 6.31 ≈ 268.69 (thousand)

Conceptual note on modelling realism

• Real-world data often require cautious interpretation; exponential growth is not guaranteed to continue indefinitely (historical examples like stock market patterns show inflection and leveling off).
• The logistic model is a common next-step in discussion of growth that saturates, though it is not the focus of this session; keep it in mind as a useful real-world model.
• The choice between linear tangent approximations and exponential approximations depends on context, data, and how far from the point of tangency you’re predicting; both have domains where they are appropriate.