Chapter 19 — Key Takeaways

The Big Idea

A differential equation expresses how a quantity changes — its rate — as a function of the quantity itself and the independent variable. Solving one means recovering the function from its rate equation, which is antidifferentiation with feedback: the right-hand side depends on the unknown function, not just on the independent variable. Because so many laws of nature are written as rates, differential equations are the native language of dynamic systems (§19.1).

One habit above all (§19.2): every candidate solution can be checked by substitution. Differentiate it, plug into the equation, and confirm both sides agree. ODEs let you verify an answer even when you could never have guessed it.

The Solution Methods

Separable equations (§19.3). If $\dfrac{dy}{dx} = g(x)\,h(y)$, gather $y$ on one side and $x$ on the other and integrate each:

$$\int \frac{dy}{h(y)} = \int g(x)\,dx.$$

The master case is $y' = ky \Rightarrow y = y_0 e^{kt}$ — exponential growth ($k>0$) or decay ($k<0$). Nonlinear separable equations can blow up in finite time (e.g. $y' = xy^2$).

First-order linear equations and the integrating factor (§19.4). For $\dfrac{dy}{dx} + P(x)y = Q(x)$, multiply by

$$\mu(x) = e^{\int P(x)\,dx}$$

so the left side collapses to $\dfrac{d}{dx}(\mu y)$, then integrate: $\mu y = \int \mu Q\,dx + C$. The integrating factor reverses the product rule, just as $u$-substitution reverses the chain rule. Every solution splits into a particular (steady) part plus a homogeneous (transient) part that decays.

Slope (direction) fields (§19.5). Every first-order equation $y' = F(x,y)$ assigns a slope to each point of the plane. Drawing short segments of those slopes shows all solutions at once, with no integral computed. Solution curves flow tangent to the segments; you can read off equilibria and long-term stability even when no closed form exists.

Euler's method (§19.6). Step along the local slope:

$$y_{n+1} = y_n + h\,F(x_n, y_n), \qquad x_{n+1} = x_n + h.$$

This is the linearization (tangent-line) idea of Chapter 11 iterated. Local error per step is $O(h^2)$; global error over a fixed interval is $O(h)$ — first-order accurate. For too large a step it goes unstable (e.g. $y' = -2y$ explodes when $h > 1$). Production code uses higher-order, adaptive methods — Heun ($O(h^2)$), RK4 ($O(h^4)$), and adaptive RK45 in scipy's solve_ivp.

The Standard Models

Exponential vs. logistic — the key contrast (§19.3, §19.7). Exponential growth has constant per-capita rate and runs away forever. Logistic growth makes the per-capita rate fall toward a ceiling, the carrying capacity $K$: it tracks the exponential while $P \ll K$, grows fastest at the inflection point $P = K/2$ (maximum rate $rK/4$), then levels off into the S-curve (sigmoid). The half-capacity inflection is independent of $r$ and $P_0$.

The SIR Model (Anchor Example, §19.9)

A three-compartment epidemic model, conserving $S + I + R = N$:

$$\frac{dS}{dt} = -\beta S I, \qquad \frac{dI}{dt} = \beta S I - \gamma I, \qquad \frac{dR}{dt} = \gamma I.$$

• $\beta$ — transmission rate (mass-action: contacts $\propto SI$); $\gamma$ — recovery rate, with $1/\gamma$ the mean infectious period.
• Basic reproduction number $\boxed{R_0 = \dfrac{\beta N}{\gamma}}$, from linearizing $I' \approx (\beta N - \gamma)I$ at the start ($S \approx N$). Epidemic grows if $R_0 > 1$, dies out if $R_0 < 1$.
• The infectious curve peaks when $S$ falls to $S^* = \gamma/\beta = N/R_0$ (where $dI/dt = 0$).
• The outbreak ends with susceptibles to spare — the final attack rate is below $100\%$, smaller residue for larger $R_0$. Do not confuse the threshold for growth ($R_0 > 1$) with the final attack rate.
• Herd-immunity threshold: transmission collapses once an immune fraction $p > 1 - 1/R_0$. (Measles, $R_0\approx 15 \Rightarrow p \approx 93\%$.)
• The system is nonlinear ($\beta SI$ is a product of unknowns) and has no closed form — hence numerical integration. This anchor climaxes in the Chapter 39 capstone, where you fit $\beta$, $\gamma$ to data and simulate interventions.

Common Errors to Avoid

• Mixing "rate out." It is $(\text{outflow rate}) \times \dfrac{y}{V}$ — the concentration leaves, not the raw amount. If inflow ≠ outflow, $V$ itself varies with time and must go inside the concentration (§19.4).
• Forgetting the constant of integration / the initial condition. A separable or linear equation gives a family of solutions; pin down the member with $y(x_0) = y_0$ (§19.3–19.4).
• Wrong integrating factor. It is $e^{\int P\,dx}$ from the coefficient of $y$, after putting the equation in standard form $y' + Py = Q$ (divide through if the leading coefficient is not $1$) (§19.4).
• Trusting Euler with a large step. First-order accuracy plus possible instability; halving $h$ only halves the error (§19.6).
• Misreading $R_0$. $R_0 > 1$ predicts growth, not universal infection; the epidemic burns out with susceptibles remaining (§19.9).
• Global vs. local existence. A nonlinear IVP's solution is only guaranteed near the initial point; it may blow up in finite time (§19.3).

Connections

• The Fundamental Theorem of Calculus (Chapter 14) is the engine underneath every solution: solving an ODE almost always reduces to computing an integral. Separable equations integrate two sides; the integrating factor integrates $\mu Q$.
• Integration techniques (Chapters 15–16) are the workhorses: $u$-substitution for separable forms, partial fractions for the logistic equation (§19.7).
• Linearization (Chapter 11) is exactly Euler's method (§19.6) and the early-outbreak SIR approximation (§19.9).
• Case Study 1 (epidemiology) develops the SIR model into the 2020 COVID policy story; Case Study 2 (pharmacokinetics) shows the mixing/linear equation governing drug dosing — the salt tank in a new disguise.
• Looking ahead: Chapter 39 builds its capstone on the SIR model you set up here; the gallery models (oscillator, Lotka–Volterra, RC circuit, compound interest) recur across physics, biology, and economics.

The Recurring Themes, Made Concrete Here

• Calculus is the mathematics of change — this chapter does nothing but turn rate laws into the quantities they govern.
• Hand computation builds understanding; machine computation builds power — you separated, integrated, and verified by hand, while Euler and solve_ivp cracked the nonlinear SIR system no formula can touch.
• The Fundamental Theorem of Calculus sits beneath every step: solving a differential equation is antidifferentiation with feedback.

Closing Part III

This chapter completes Part III: Integration. Across Chapters 13–19 you built the definite integral, proved the Fundamental Theorem, mastered the techniques of integration, tamed improper integrals, applied integration to geometry and physics, and now solved the equations that govern dynamic systems. Part IV (Chapters 20–24) turns to sequences and series — adding infinitely many numbers, Taylor series, and the road to Euler's identity $e^{i\pi} + 1 = 0$. You have just learned to speak nature's native language; next, learn to add infinitely many numbers.