Derivatives and Sensitivity In Practice

Work through each problem on paper before expanding the solution details.

Part 1: The Underlying Mechanics Drill

Problem 1: Average Rate of Change for a Linear Function

A server hosting a machine learning inference endpoint incurs compute costs according to the linear function:

f(x) = 3x + 5

where $x$ represents execution runtime in hours, and $f(x)$ represents total cost in dollars.

Calculate the average rate of change $\frac{\Delta y}{\Delta x}$ between $x_1 = 1$ hour and $x_2 = 4$ hours.

Reveal Solution

Step 1: Compute function values at interval endpoints:

At $x_1 = 1$ : $f(1) = 3(1) + 5 = 8$
At $x_2 = 4$ : $f(4) = 3(4) + 5 = 12 + 5 = 17$

Step 2: Calculate differences:

$\Delta x = x_2 - x_1 = 4 - 1 = 3$
$\Delta y = f(4) - f(1) = 17 - 8 = 9$

Step 3: Evaluate the rate fraction:

\frac{\Delta y}{\Delta x} = \frac{9}{3} = \mathbf{3.0 \text{ dollars per hour}}

Takeaway: For any linear function $f(x) = mx + c$ , the average rate of change across any interval is always equal to the constant coefficient $m = 3.0$ .

Problem 2: Power Rule Derivative Computation

Given the cubic monomial function:

f(x) = 4x^3

Compute the exact algebraic derivative $\frac{d}{dx}[4x^3]$ using the power rule.

Reveal Solution

Apply the constant multiplier rule and the power rule:

\frac{d}{dx}[c \cdot x^n] = c \cdot n x^{n-1}

For $c = 4$ and $n = 3$ :

\frac{d}{dx}[4x^3] = 4 \cdot \frac{d}{dx}[x^3] = 4 \cdot (3 x^{3-1}) = 4 \cdot (3x^2) = \mathbf{12x^2}

Takeaway: Multiplying the coefficient by the exponent ( $4 \times 3 = 12$ ) and decreasing the exponent by $1$ ( $3 - 1 = 2$ ) gives the exact derivative $12x^2$ .

Problem 3: Polynomial Derivative Evaluation at an Operating Point

Consider the quadratic function:

f(x) = 2x^2 - 6x + 9

Part A: Compute the general derivative function $f'(x)$ .
Part B: Evaluate the instantaneous rate of change at $x = 3$ .

Reveal Solution

Part A: Compute $f'(x)$ using term-by-term differentiation:

$\frac{d}{dx}[2x^2] = 2(2x) = 4x$
$\frac{d}{dx}[-6x] = -6$
$\frac{d}{dx}[9] = 0$

f'(x) = \mathbf{4x - 6}

Part B: Evaluate at $x = 3$ :

f'(3) = 4(3) - 6 = 12 - 6 = \mathbf{+6.0}

Interpretation: At $x = 3$ , the function is increasing at an instantaneous rate of $+6.0$ units of output for every $1$ unit increase in $x$ . The tangent slope is positive and uphill.

Problem 4: Stationary Point Calculation (Vertex of a Quadratic Curve)

Given the quadratic error curve:

f(x) = x^2 - 8x + 12

Part A: Compute the derivative $f'(x)$ .
Part B: Find the stationary point by solving $f'(x) = 0$ .
Part C: Evaluate the function value at this stationary coordinate.

Reveal Solution

Part A: Compute the derivative:

f'(x) = \frac{d}{dx}[x^2] - \frac{d}{dx}[8x] + \frac{d}{dx}[12] = \mathbf{2x - 8}

Part B: Set derivative to zero and solve for $x$ :

2x - 8 = 0 \implies 2x = 8 \implies \mathbf{x = 4}

Part C: Evaluate $f(4)$ :

f(4) = (4)^2 - 8(4) + 12 = 16 - 32 + 12 = \mathbf{-4.0}

Geometric Conclusion: At $x = 4$ , the curve reaches its global minimum vertex $(4, -4)$ . The tangent line at this point is perfectly flat with slope $f'(4) = 0.0$ .

Problem 5: Directional Sign and Sensitivity Interpretation

An optimization algorithm is evaluating an objective function $f(x)$ . At the current operating point $x_0 = 1.5$ , the calculated derivative is:

f'(1.5) = +5.0

To decrease the value of $f(x)$ , should the algorithm increase or decrease $x$ ? Explain why using the relationship between differentials.

Reveal Solution

Answer: The algorithm must decrease $x$ (nudge $x$ in the negative direction).

Mathematical Rationale: The first-order differential relationship states:

\Delta f \approx f'(x) \cdot \Delta x

Substituting $f'(1.5) = +5.0$ :

\Delta f \approx (+5.0) \cdot \Delta x

If we increase $x$ ( $\Delta x > 0$ ), then $\Delta f > 0$ (the function increases).
If we decrease $x$ ( $\Delta x < 0$ , e.g., $\Delta x = -0.1$ ), then: $\Delta f \approx (+5.0) \cdot (-0.1) = \mathbf{-0.5} \quad (\text{the function decreases})$

Optimization Takeaway: Gradient descent updates variables in the direction of the negative derivative ( $-\frac{df}{dx} = -5.0$ ). When the derivative is positive, step left (decrease $x$ ).

Part 2: Applied Scenario: The VC Valuation Sensitivity Analysis

In Course 1, our venture capital firm models the enterprise post-money valuation of high-growth AI startups.

An early-stage SaaS startup's valuation $V(r)$ (measured in millions of dollars, $\char36 M$ ) is modeled as a quadratic function of its monthly recurring revenue (MRR) traction rate $r$ (measured in units of $\char36 100\text{k}/\text{month}$ ):

V(r) = 0.5r^2 + 2r + 10

where:

$r$ is the traction rate ( $r = 4$ corresponds to $\char36 400\text{k}/\text{month}$ in MRR).
$V(r)$ is the post-money valuation in millions of dollars (at $r = 0$ , base intellectual property valuation is $V(0) = \char36 10\text{M}$ ).

Problem 6: Step-by-Step Valuation Sensitivity & Marginal Nudge Audit

Part A: Compute the general derivative function $\frac{dV}{dr}$ using the power, linear multiplier, and constant rules.
Part B: Evaluate the instantaneous valuation sensitivity $\frac{dV}{dr}$ at current traction rate $r = 4$ . State the physical units of this derivative.
Part C (Marginal Contract Nudge): The startup is negotiating an enterprise contract that will increase its traction rate by $\Delta r = +0.10$ $Δ r = + 0.10$ (an additional $\char36 10\text{k}/\text{month}$ $$10 k / month$ in MRR).
1. Use the linear tangent approximation $\Delta V_{\text{linear}} \approx \frac{dV}{dr} \cdot \Delta r$ to estimate the valuation increase.
2. Compute the exact valuation increase $\Delta V_{\text{exact}} = V(4.1) - V(4.0)$ .
3. Compare the linear estimate against the exact value and explain why the tangent approximation is accurate for small nudges.
Part D (Scale Economies & Sensitivity Growth): Calculate the sensitivity derivative at $r = 1$ versus $r = 6$ . In 2–3 sentences, explain why sensitivity grows as traction expands, and why venture investors evaluate derivative rates when pricing investment rounds.

Reveal Solution

Part A: Analytical Valuation Derivative

Differentiate $V(r) = 0.5r^2 + 2r + 10$ term-by-term:

$\frac{d}{dr}[0.5r^2] = 0.5 \cdot (2r) = 1.0r = r$
$\frac{d}{dr}[2r] = 2$
$\frac{d}{dr}[10] = 0$

\frac{dV}{dr} = \mathbf{r + 2}

Part B: Instantaneous Sensitivity at $r = 4$

Substitute $r = 4$ into the derivative:

\left. \frac{dV}{dr} \right|_{r=4} = 4 + 2 = \mathbf{6.0}

Physical Units & Meaning: The valuation sensitivity is $\char36 6.0\text{M}$ of valuation per unit of traction rate (or $\char36 60{,}000$ in enterprise valuation for every $\char36 1{,}000/\text{month}$ in MRR).

Part C: Marginal Contract Nudge Comparison

Linear Tangent Approximation: $\Delta V_{\text{linear}} \approx \left(\frac{dV}{dr}\right) \cdot \Delta r = 6.0 \cdot (+0.10) = \mathbf{\char36 0.60\text{M}} \quad (\text{or } \char36 600{,}000)$
Exact Polynomial Valuation Calculation:
- At $r = 4.0$ : $V(4.0) = 0.5(4.0)^2 + 2(4.0) + 10 = 0.5(16) + 8 + 10 = 8 + 8 + 10 = \mathbf{\char36 26.0\text{M}}$
- At $r = 4.1$ : $V(4.1) = 0.5(4.1)^2 + 2(4.1) + 10 = 0.5(16.81) + 8.2 + 10 = 8.405 + 8.2 + 10 = \mathbf{\char36 26.605\text{M}}$
- Exact difference: $\Delta V_{\text{exact}} = 26.605 - 26.000 = \mathbf{\char36 0.605\text{M}} \quad (\text{or } \char36 605{,}000)$
Comparison and Residual Analysis: $\text{Discrepancy} = \Delta V_{\text{exact}} - \Delta V_{\text{linear}} = 0.605 - 0.600 = \mathbf{\char36 0.005\text{M}} \quad (\text{or } \char36 5{,}000)$ The linear tangent approximation captures $99.17\%$ of the exact valuation change. The microscopic $\char36 5{,}000$ difference corresponds precisely to the second-order quadratic term: $a(\Delta r)^2 = 0.5(\Delta r)^2 = 0.5(0.10)^2 = 0.5(0.01) = \mathbf{\char36 0.005\text{M}} \quad (\text{or } \frac{1}{2} V''(r)(\Delta r)^2 = \frac{1}{2}(1.0)(0.01) = \char36 0.005\text{M})$ For infinitesimal nudges ( $dr$ ), this second-order term approaches zero, making the derivative an exact representation of marginal change.

Part D: Scale Economies & Investor Pricing Leverage

At $r = 1$ ( $\char36 100\text{k}/\text{month}$ MRR): $\left.\frac{dV}{dr}\right|_{r=1} = 1 + 2 = \mathbf{\char36 3.0\text{M} \text{ valuation per traction unit}}$
At $r = 6$ ( $\char36 600\text{k}/\text{month}$ MRR): $\left.\frac{dV}{dr}\right|_{r=6} = 6 + 2 = \mathbf{\char36 8.0\text{M} \text{ valuation per traction unit}}$

Investor Synthesis: Because the valuation curve is quadratic ( $0.5r^2$ ), the derivative $\frac{dV}{dr} = r + 2$ scales linearly with traction. Each incremental dollar of recurring revenue creates more enterprise value at high volume than at early launch due to compounding market dominance and operational leverage. Venture investors compute this instantaneous rate of change to price marginal growth and structure follow-on investment rounds.