Introduction to ODEs: Classification and First Models
What a differential equation is, how to classify it (ODE or PDE, order, linear or nonlinear, single equation or system), and the first two models of the course: population growth and decay, and free fall. Sections 1 to 6 follow page 2 of the class notes item by item; Robinson formalises and extends each step, and material beyond the notes is labelled as a complement.
Physical laws are rarely statements about quantities themselves; they are statements about how quantities change. Newton's second law relates force to the rate of change of momentum, a mass balance relates the net inflow to the rate of change of the stored mass, and a population model relates the number of births per year to the current population. Writing any of these laws down produces an equation in which the unknown is a function, and that function appears together with some of its derivatives. Such an equation is a differential equation.
A differential equation is an equation involving an unknown function (of one or more independent variables) and its derivatives.
The independent variables are the arguments of the unknown function (e.g. $t$ in $x(t)$, $x$ in $y(x)$, both $x$ and $y$ in $G(x,y)$). The unknown function itself is the dependent variable. A differential equation relates a dependent variable, its derivatives and the independent variables. [Robinson, Sec. 3.1, p. 11]
Notation. The class notes write every derivative with a prime, also in time: $x'(t) = dx/dt$. Robinson keeps the prime for derivatives with respect to $x$ and uses a dot for derivatives with respect to time:
$$\dot z = \frac{dz}{dt}, \qquad \ddot z = \frac{d^2 z}{dt^2}, \qquad y' = \frac{dy}{dx}, \qquad y'' = \frac{d^2 y}{dx^2}.$$Both notations appear on this page; they mean the same thing.
1.1 Ordinary versus partial differential equations
The most important distinction depends on how many independent variables there are. With one independent variable only ordinary derivatives can occur; with two or more, the derivatives are partial.
- An ODE is a differential equation in which the unknown function depends on only one independent variable. Example: $x'(t) = x(t)$, with dependent variable $x$ and independent variable $t$. The function $x(t) = e^t$ is a solution, since $\frac{d}{dt}e^t = e^t$.
- A PDE is a differential equation in which the unknown function depends on more than one variable. Example: the heat equation for $u(x,t)$, $$\frac{\partial u}{\partial t} = \frac{\partial^2 u}{\partial x^2}.$$ (The notes write the derivatives with $d$; with two independent variables they are partial derivatives.)
Robinson gives the same definitions with a wider set of examples [Robinson, Sec. 3.1, pp. 11–12]:
| Equation | Name / meaning | Type | Source |
|---|---|---|---|
| $\dfrac{dy}{dx} = f(x)$ | Prescribed slope of the graph of $y$ | ODE, independent variable $x$ | p. 11 |
| $m\,\ddot{\mathbf x} = \mathbf f(t)$ | Newton's second law for a particle at $\mathbf x(t) = (x,y,z)$ | ODE (system of 3), variable $t$ | p. 12 |
| $-\dfrac{\hbar^2}{2m}\dfrac{d^2\psi}{dx^2} + V(x)\psi = E\psi$ | Schrödinger equation, complex $\psi(x)$ | ODE, variable $x$ | p. 12 |
| $\dfrac{\partial h}{\partial t} = k\,\dfrac{\partial^2 h}{\partial x^2}$ | Heat equation for the temperature $h(x,t)$ of a rod | PDE, variables $x, t$ | p. 12 |
| $\rho\Big(\dfrac{\partial \mathbf u}{\partial t} + (\mathbf u\cdot\nabla)\mathbf u\Big) - \mu\Delta\mathbf u + \nabla p = \mathbf f,\ \ \nabla\cdot\mathbf u = 0$ | Navier–Stokes equations for a fluid velocity $\mathbf u(x_1,x_2,x_3,t)$ | PDE system, 4 variables | Eqs. (3.1)–(3.3), p. 12 |
The Navier–Stokes equations (3.3) are the model behind every computational fluid dynamics code used to design wings, intakes and nozzles. They are PDEs; this course deals with ODEs only [Robinson, p. 12]. ODEs still appear constantly in aerospace work: flight-path and attitude dynamics ($t$ is the only independent variable), structural vibration modes, and the self-similar reductions of the flow equations, such as the boundary-layer equation of Section 2.
The order of an ODE is the order of the highest derivative in the ODE.
Robinson makes this precise [Robinson, Sec. 3.2, Eq. (3.4), p. 13]: an $n$th-order ODE for $y(t)$ is an equation of the form
$$F\!\left(\frac{d^n y}{dt^n},\ \frac{d^{n-1} y}{dt^{n-1}},\ \dots,\ \frac{dy}{dt},\ y,\ t\right) = 0, \tag{3.4}$$in which $d^n y/dt^n$ really occurs in $F$. The last condition is not pedantry. If $F(\ddot y,\dot y,y,t) = y - t$, then "$F = 0$" is the algebraic relation $y = t$, which contains no derivative at all and so is not a differential equation [p. 13]. The examples of the class notes: $x' = x$ is first order, the heat equation is second order (in $x$), and Newton's law $m\,y'' = F$ (Section 6) is second order.
| Equation | Physical meaning | Order |
|---|---|---|
| $\dfrac{dy}{dx} = f(x)$ | Slope of a graph | 1 |
| $\tfrac12 m\dot x^2 + V(x) = E$ | Energy conservation: kinetic plus potential energy is constant | 1 |
| $m\dfrac{d^2x}{dt^2} = F$ | Newton's second law | 2 |
| $\psi''' + \tfrac12\psi\psi'' = 0$ | Fluid boundary layers (known in fluid mechanics as the Blasius equation for flow along a flat plate) | 3 |
2.1 Complement (Robinson): autonomous equations
If the independent variable $t$ does not occur explicitly in the equation, as in $\dfrac{dy}{dt} = f(y)$, the equation is autonomous. Otherwise it is non-autonomous.
In the class notes autonomy appears on p. 3, where it is used for the stability of first-order equations (Topic 2). Its physical meaning is that the law does not depend on the clock. If $y(t)$ solves $\dot y = f(y)$, then so does the time-shifted function $z(t) = y(t - s)$ for any fixed $s$. To see this, apply the chain rule with the inner function $u(t) = t - s$, $du/dt = 1$:
$$\frac{dz}{dt} = \frac{dy}{du}\Big|_{u=t-s}\cdot\frac{du}{dt} = f\big(y(t-s)\big)\cdot 1 = f\big(z(t)\big).$$So an experiment started today and the same experiment started tomorrow evolve identically. This is what makes the geometric phase-line method of Topic 2 work.
Any second-order equation $\ddot x = f(\dot x, x, t)$ is equivalent to a first-order system for the two unknowns $x_1 = x$, $x_2 = \dot x$:
$$\dot x_1 = x_2, \qquad \dot x_2 = f(x_2, x_1, t).$$The first equation is the definition of $x_2$; the second is the original equation rewritten. The same trick ($x_k = y^{(k-1)}$, $k = 1,\dots,n$) turns any $n$th-order equation into $n$ first-order ones. This is developed in Topic 4 [Notes, p. 9; Robinson, Sec. 25.1, p. 251 for $n = 2$; Exercise 25.1, p. 252 for general $n$].
3. Linear and nonlinear equations
Linearity is the classification with the largest practical consequences. Robinson's summary is blunt: linear problems are relatively "easy" (an explicit solution can usually be found) and nonlinear problems are "hard" (explicit solutions exist only in very particular cases) [p. 13]. The class notes give two things: the form of a linear equation (in the margin of p. 2) and a test to decide linearity, based on superposition.
3.1 The form of a linear ODE
An ODE of order $n$ is linear if it has the form
$$a_0(x)\,y^{(n)}(x) + a_1(x)\,y^{(n-1)}(x) + \dots + a_{n-1}(x)\,y'(x) + a_n(x)\,y(x) = b(x), \tag{L}$$where the coefficients $a_i$ and the right-hand side $b$ are functions of the independent variable only. It is homogeneous if $b \equiv 0$ and inhomogeneous otherwise.
The lecture note Definition of linear ODE (Sep 9) writes the same form with the indices in the opposite order, $a_n(x)\,y^{(n)} + \dots + a_1(x)\,y' + a_0(x)\,y = b(x)$, and Robinson writes it with $t$ as the independent variable and $f(t)$ on the right, summarising it as "only multiples of $y$ and its derivatives occur" [Robinson, Eq. (3.5), p. 14]. The coefficients may be as nonlinear as they like in the independent variable; what matters is how the unknown appears. Collecting the left-hand side of (L) into an operator,
$$L[y] = a_0\,y^{(n)} + a_1\,y^{(n-1)} + \dots + a_n\,y = \sum_{j=0}^{n} a_j\,y^{(n-j)},$$a linear ODE is simply $L[y] = b$.
3.2 The superposition test of the class notes
Consider $u_1$ and $u_2$, two solutions to an ODE. Then the ODE is linear if, for all $\alpha, \beta \in \mathbb R$, $\alpha u_1 + \beta u_2$ is also a solution to the ODE. If not, the ODE is non-linear.
"Every combination $\alpha u_1 + \beta u_2$ of solutions is a solution" is exactly the superposition principle of homogeneous linear equations. An inhomogeneous linear equation such as $x' - x = 1$ has the form (L) but fails the test as stated: with $\alpha = \beta = 0$ the zero function would have to be a solution, and $0 - 0 \neq 1$. For $b \not\equiv 0$ the test is applied to the homogeneous part $L[y] = 0$. The two examples of the notes below are tested exactly as written there.
3.3 Demonstration 1: a homogeneous linear equation passes the test
If $L[u_1] = 0$ and $L[u_2] = 0$, then $L[\alpha u_1 + \beta u_2] = 0$ for all $\alpha, \beta \in \mathbb R$.
Set $v = \alpha u_1 + \beta u_2$.
- Differentiation is linear: for every $k$, $\ v^{(k)} = \dfrac{d^k}{dx^k}(\alpha u_1 + \beta u_2) = \alpha u_1^{(k)} + \beta u_2^{(k)}$.
- Substitute into $L$ and split the sum: $$L[v] = \sum_{j=0}^{n} a_j\big(\alpha u_1^{(n-j)} + \beta u_2^{(n-j)}\big) = \alpha\sum_{j=0}^{n} a_j u_1^{(n-j)} + \beta\sum_{j=0}^{n} a_j u_2^{(n-j)} = \alpha L[u_1] + \beta L[u_2].$$
- With $L[u_1] = L[u_2] = 0$: $\ L[v] = \alpha\cdot 0 + \beta\cdot 0 = 0$. $\square$
Let $u_1$ and $u_2$ be solutions, $u_1' = u_1$ and $u_2' = u_2$, and call $u = \alpha u_1 + \beta u_2$. Then
$$u' = \alpha u_1' + \beta u_2' = \alpha u_1 + \beta u_2 = u,$$Reading note: the notes write the middle step as $\alpha u_1 + \alpha u_2$; the second coefficient is $\beta$.
so $u$ is again a solution for every $\alpha$, $\beta$: the ODE is linear. In the form (L): $n = 1$, $a_0 = 1$, $a_1 = -1$, $b = 0$.
The same property gives the structure of the solutions of an inhomogeneous equation: if $L[y_p] = b$ and $L[y_h] = 0$, then $L[y_h + y_p] = L[y_h] + L[y_p] = 0 + b = b$. General solution = homogeneous solution + one particular solution (Topic 3; Robinson Sec. 14.1).
3.4 Demonstration 2: $u'' = (u')^2 + 1$ fails the test
Let $u_1$ and $u_2$ be solutions, $u_1'' = (u_1')^2 + 1$ and $u_2'' = (u_2')^2 + 1$, and let $v = \alpha u_1 + \beta u_2$ [Notes, p. 2]. By linearity of differentiation,
$$v' = \alpha u_1' + \beta u_2', \qquad v'' = \alpha u_1'' + \beta u_2''.$$For $v$ to be a solution we would need $v'' = (v')^2 + 1$. Compute both sides.
- Left side, replacing $u_1''$ and $u_2''$ by their equations: $$v'' = \alpha\big((u_1')^2 + 1\big) + \beta\big((u_2')^2 + 1\big) = \alpha\,(u_1')^2 + \beta\,(u_2')^2 + \alpha + \beta.$$
- Right side, expanding the square $(a + b)^2 = a^2 + 2ab + b^2$: $$(v')^2 + 1 = \big(\alpha u_1' + \beta u_2'\big)^2 + 1 = \alpha^2(u_1')^2 + 2\alpha\beta\,u_1'u_2' + \beta^2(u_2')^2 + 1.$$
- Compare. The left side has $\alpha$ where the right side has $\alpha^2$, $\beta$ where it has $\beta^2$, no cross term $2\alpha\beta\,u_1'u_2'$, and the constant $\alpha + \beta$ instead of $1$. The equality cannot hold for all $\alpha, \beta$: the ODE is non-linear.
For an inhomogeneous linear equation, combinations with $\alpha + \beta = 1$ are still solutions ($L[\alpha u_1 + \beta u_2] = (\alpha + \beta)\,b = b$). For $u'' = (u')^2 + 1$ even these fail. Put $\beta = 1 - \alpha$ and subtract the two sides of steps 1 and 2; since $\beta(1 - \beta) = (1 - \alpha)\alpha$ and $2\alpha\beta = 2\alpha(1 - \alpha)$,
$$v'' - \big((v')^2 + 1\big) = \alpha(1 - \alpha)(u_1')^2 + \beta(1 - \beta)(u_2')^2 - 2\alpha\beta\,u_1'u_2' = \alpha(1 - \alpha)\big(u_1' - u_2'\big)^2,$$which is not zero when $0 < \alpha < 1$ and $u_1' \neq u_2'$. Such solutions exist: $w = u'$ satisfies $w' = w^2 + 1$, solved by $w = \tan(t + c)$ because $\frac{d}{dt}\tan s = \sec^2 s = \tan^2 s + 1$, so $u = -\ln\cos(t + c) + d$. Take $u_1 = -\ln\cos t$, $u_2 = -\ln\cos(t - \pi/4)$ on $-\pi/4 < t < \pi/2$ and $\alpha = \beta = \tfrac12$. At $t = 0$: $u_1' = 0$, $u_2' = -1$, $u_1'' = 1$, $u_2'' = 2$, so $v'' = \tfrac32$ while $(v')^2 + 1 = \tfrac14 + 1 = \tfrac54$. The gap $\tfrac14 = \tfrac12\cdot\tfrac12\cdot\big(0 - (-1)\big)^2$ is the one the formula predicts.
3.5 A practical test, and the lecture examples
Read each term of the equation and ask what it does to the unknown $y$. The equation is linear exactly when every term is one of:
- a function of the independent variable alone (it goes into $b$);
- a function of the independent variable multiplied by $y$ or by one derivative $y^{(k)}$, to the first power.
Anything else makes the equation nonlinear: powers such as $y^2$ or $(y')^2$, products such as $y\,y'$, and nonlinear functions of the unknown such as $\sin y$, $e^{y}$ or $\sqrt{y'}$.
The three equations analysed in class (corrected note dated Sep 9, presented in the lecture of Sep 10):
| ODE | Order | Verdict | Justification against the form (L) |
|---|---|---|---|
| $u'(t) = u(t)$ | 1 | Linear, homogeneous | Rewrite as $1\cdot u' + (-1)\cdot u = 0$: $a_0 = 1$, $a_1 = -1$, $b = 0$. Superposition: Section 3.3. |
| $u''(t) = \big(u'(t)\big)^2 + 1$ | 2 | Nonlinear | The derivative $u'$ is squared. No choice of $a_i(t)$, $b(t)$ reproduces $(u')^2$. Superposition fails: Section 3.4. |
| $y'(x) = 4\big(y(x)\big)^2 + x^2$ | 1 | Nonlinear | The unknown $y$ itself is squared. The term $x^2$ is harmless: it is a function of $x$ only and would sit in $b(x)$. |
The lecture of Sep 8 classified these examples and the solution was flagged as needing correction; the corrected solution was given on Sep 10 together with the note Definition of linear ODE. The table above follows the corrected note. The third example is a useful trap: a term like $x^2$ looks "nonlinear", but nonlinearity in the independent variable is irrelevant; only nonlinearity in the unknown counts.
3.6 Complement (Robinson): what "solving" can mean
Robinson lists the kinds of answer one can expect, in decreasing order of explicitness [Robinson, Sec. 3.4, pp. 14–16]:
| Type | Example | Comment |
|---|---|---|
| Fully explicit | $y(t) = 3\cos 5t + 8\sin t$ (3.6) | Combination of elementary functions; rare. |
| Explicit with an integral | $y(t) = 1 + \displaystyle\int_0^t e^{-s^2}\,ds$ (3.7) | Still $y$ as a function of $t$, but the integral has no elementary antiderivative. |
| Implicit | $\ln y + 4\ln x - y - 2x + 4 = 0$ (3.8) | Relates $x$ and $y$ without derivatives; cannot be solved for $y$. |
| Qualitative | Phase portrait of (3.9), Fig. 1.1 | Describes the behaviour without any formula. |
| Numerical | MATLAB ode45 for $\dot x = t - x^2$, $x(0) = 0$ | A table of approximate values at discrete times. |
A system of ODEs has two or more dependent variables. Example:
$$\begin{cases} x'(t) = 2x(t) - y(t),\\[2pt] y'(t) = -x(t) + 3y(t). \end{cases}$$The two unknowns $x(t)$, $y(t)$ share the independent variable $t$. The system is first order, linear and homogeneous (each right-hand side is a combination of $x$ and $y$ with constant coefficients and nothing else), and autonomous. In matrix form,
$$\begin{pmatrix} x \\ y \end{pmatrix}' = \begin{pmatrix} 2 & -1 \\ -1 & 3 \end{pmatrix}\begin{pmatrix} x \\ y \end{pmatrix}.$$It passes the superposition test of Section 3.2: if $(x_1, y_1)$ and $(x_2, y_2)$ are solutions, then $(x, y) = \alpha(x_1, y_1) + \beta(x_2, y_2)$ satisfies $x' = \alpha x_1' + \beta x_2' = \alpha(2x_1 - y_1) + \beta(2x_2 - y_2) = 2x - y$, and in the same way $y' = -x + 3y$.
Preview of Topic 4: one second-order equation for $x$
Solve the first equation for $y$: $y = 2x - x'$. Differentiating, $y' = 2x' - x''$. Substitute both into the second equation:
$$2x' - x'' = -x + 3\,(2x - x') = 5x - 3x' \quad\Longrightarrow\quad x'' - 5x' + 5x = 0.$$This is the elimination method of the class notes, p. 9, which turns $x' = ax + by$, $y' = cx + dy$ into $x'' - (a + d)x' + (ad - bc)x = 0$. Check: $a = 2$, $b = -1$, $c = -1$, $d = 3$ give $a + d = 5$ and $ad - bc = 6 - 1 = 5$ ✓. Second-order equations of this kind are solved in Topic 3 and systems in Topic 4.
4.1 Complement (Robinson): a nonlinear system and its phase portrait
Robinson's first example (which he interprets as two competing species only much later, p. 326) has unknowns $x(t)$ and $y(t)$ [Robinson, Eq. (3.9), p. 15]:
$$\dot x = x\,(4 - 2x - y), \qquad \dot y = y\,(9 - 3x - 3y). \tag{3.9}$$Each species grows when rare (the constant terms $4x$ and $9y$), suffers from crowding by its own kind ($-2x^2$, $-3y^2$) and from competition with the other ($-xy$, $-3xy$). The system is first order, autonomous and nonlinear (it contains $x^2$, $y^2$ and $xy$), so the superposition test fails.
Where the solutions settle
A point where both right-hand sides vanish is a constant (equilibrium) solution. From $x(4 - 2x - y) = 0$ we need $x = 0$ or $y = 4 - 2x$; from $y(9 - 3x - 3y) = 0$ we need $y = 0$ or $y = 3 - x$. Combining the four options:
- $x = 0,\ y = 0$: the point $(0,0)$;
- $x = 0,\ y = 3 - x$: the point $(0,3)$;
- $y = 4 - 2x,\ y = 0$: the point $(2,0)$;
- $y = 4 - 2x,\ y = 3 - x$: subtracting, $0 = 1 - x$, so $x = 1$ and $y = 2$: the point $(1,2)$.
The figure, which Robinson calls a phase diagram or phase portrait, shows that every solution with both species present tends to $(1,2)$ as $t \to +\infty$: the species coexist [p. 15; species reading, p. 326]. The vector notation $\dot{\mathbf x} = \mathbf f(\mathbf x, t)$ and the full theory of systems are the subject of Topic 4.
5. Example 1: population growth and decay
5.1 The model
$x(t)$ is the population at time $t$, with $x(t) > 0$ for all $t$. The rate of change of $x$ is proportional to $x$ itself, with proportionality constant $k \in \mathbb R$. This gives the first-order linear ODE
$$x'(t) = k\,x(t), \qquad x(0) = x_0 > 0.$$Where the law comes from. Robinson derives it from a balance [Robinson, Sec. 8.3, Eq. (8.5), p. 62]: in a short time $\Delta t$ a population $x$ gains about $\beta x\,\Delta t$ births and loses about $\delta x\,\Delta t$ deaths, with constant per-capita rates $\beta$, $\delta$. Hence $\Delta x \approx (\beta - \delta)\,x\,\Delta t$; dividing by $\Delta t$ and letting $\Delta t \to 0$ gives $x' = kx$ with $k = \beta - \delta$ (Malthus, 1798).
| Symbol | Meaning | SI unit |
|---|---|---|
| $x(t)$ | Population (number of individuals) | dimensionless count |
| $x_0$ | Initial population, $x(0)$ | dimensionless count |
| $k$ | Net growth rate per individual | s−1 (in practice yr−1) |
| $t$ | Time | s |
Linearity: in the form (L) with $t$ as variable, $1\cdot x' + (-k)\cdot x = 0$, so $a_0 = 1$, $a_1 = -k$, $b = 0$.
5.2 Solution by integration from $0$ to $t$
Because $x(\tau) > 0$ we may divide by it: $\ \dfrac{x'(\tau)}{x(\tau)} = k$. The class notes integrate both sides from $0$ to $t$ (the dummy variable is $\tau$, so that $t$ can be the upper limit) and evaluate the left side in two equivalent ways [Notes, p. 2].
- Route 1: substitution. Put $u = x(\tau)$, so $du = x'(\tau)\,d\tau$. The limits change with it: $\tau = 0$ gives $u = x(0) = x_0$ and $\tau = t$ gives $u = x(t)$. An antiderivative of $1/u$ for $u > 0$ is $\ln u$, so by Barrow's rule $$\int_0^t \frac{x'(\tau)}{x(\tau)}\,d\tau = \int_{x_0}^{x(t)} \frac{du}{u} = \Big[\ln u\Big]_{x_0}^{x(t)} = \ln x(t) - \ln x_0 = \ln\frac{x(t)}{x_0}.$$ The notes abbreviate this step as $\dfrac{dx}{dt} = kx \Rightarrow \displaystyle\int_0^t \frac{dx}{x} = \int_0^t k\,dt$.
- Route 2: recognise a derivative. By the chain rule with outer function $\ln$ and inner function $x(\tau) > 0$, $$\frac{d}{d\tau}\ln\big(x(\tau)\big) = \frac{1}{x(\tau)}\cdot x'(\tau),$$ so the integrand is the derivative of $\ln x(\tau)$, and Barrow's rule gives directly $$\int_0^t \big(\ln x(\tau)\big)'\,d\tau = \ln x(t) - \ln x(0) = \ln\frac{x(t)}{x_0}.$$
- Right side. An antiderivative of the constant $k$ is $k\tau$: $$\int_0^t k\,d\tau = \Big[k\tau\Big]_0^t = kt - k\cdot 0 = kt.$$
- Equate and solve. $\ln\dfrac{x(t)}{x_0} = kt$. Exponentiate both sides: $\dfrac{x(t)}{x_0} = e^{kt}$.
- Check the ODE: $x'(t) = x_0\,k\,e^{kt} = k\,x(t)$ ✓ (chain rule on $e^{kt}$).
- Check the initial value: $x(0) = x_0e^0 = x_0$ ✓.
- Dimensions: $[kt] = \text{s}^{-1}\cdot\text{s} = 1$, so the exponential is dimensionless and $x$ has the units of $x_0$ ✓.
5.3 Growth or decay: the sign of $k$
| Case | Behaviour | As $t \to \infty$ | Time scale |
|---|---|---|---|
| $k > 0$ | Population grows | $x(t) \to \infty$ | Doubling time $d$: $2 = e^{kd}$, so $d = \ln 2/k$ |
| $k < 0$ | Population decays | $x(t) \to 0$ | Half-life: $\tfrac12 = e^{kh}$, so $h = \ln 2/|k|$ |
| $k = 0$ | No change | $x(t) \equiv x_0$ | — |
The notes record the first two rows, "if $k > 0$ population increases, if $k < 0$ population decreases" [Notes, p. 2]; the limits, the time scales and the case $k = 0$ are an extension. Neither time scale depends on $x_0$: a population of a thousand and one of a million double in the same time. The decaying case is also the law of radioactive decay (Problem 1.3, plutonium-239).
The notes assume $x > 0$, natural for a population. For $\dot x = \lambda x$ with $x(t_0) = x_0$ of either sign, Robinson argues that $x \equiv 0$ is the only solution through $x_0 = 0$ (uniqueness by Picard's theorem, Topic 2, Section 2), so no other solution can reach $0$; $x(t)$ keeps the sign of $x_0$, the moduli in $\ln|x|$ can be dropped, and the same computation gives $x(t) = x_0\,e^{\lambda(t - t_0)}$ [Robinson, Sec. 8.2, pp. 61–62].
Population of Great Britain and Ireland: 1801: 16 345 646; 1851: 27 533 755; 1901: 41 609 091 (Eq. 8.6).
Fitting $k$ from 1801 and 1851, $x(1851) = x(1801)e^{50k}$:
$$k = \frac{\ln x(1851) - \ln x(1801)}{50} = \frac{\ln(27\,533\,755/16\,345\,646)}{50} = \frac{0.52145}{50} = 0.01043\ \text{yr}^{-1}.$$- Prediction for 1901: $x(1801)e^{100k} = 46.4$ million (true: 41.6 million) — reasonable.
- Prediction for 2001: $x(1801)e^{200k} = 131.6$ million (true: about 63 million, UK plus Republic of Ireland) — more than twice too large.
- Doubling time: $\ln 2/0.01043 = 66.5$ years. Robinson rounds $k$ to $0.010$ first and quotes 69 years; both describe the same fit.
5.4 The logistic correction
In reality a population does not grow exponentially forever, but reaches a flat maximum: we use a logistic correction.
The logistic equation $x' = kx\,(1 - x/M) = kx - \tfrac{k}{M}x^2$ keeps the exponential rate for small populations and stops the growth at a maximum value $M$ [Robinson, Sec. 8.5, Eq. (8.12), p. 66]. The class notes study it on p. 3, with the methods of the next block: its stationary points and their stability (Topic 2, Section 4.4) and its explicit solution by separation of variables (Topic 2, Section 6.3).
6. Example 2: Newton's second law and free fall
6.1 The model
Let $y(t)$ be the height of the object, measured on a vertical axis pointing up. The only force is the weight, $F = -mg$, and the acceleration is $y''(t)$, so
$$-m\,g = m\,y''(t) \quad\Longrightarrow\quad \frac{d^2y}{dt^2} = -g,$$a second-order linear ODE. The mass cancels.
Hypotheses. Constant gravity $g \approx 9.81\ \text{m/s}^2$, no air resistance, motion along a vertical line: one degree of freedom, the height $y(t)$ [m]. In the form (L): $a_0 = 1$, $a_1 = a_2 = 0$, $b = -g$; the equation is inhomogeneous with constant right-hand side.
6.2 Integrating twice
The notes write the answer as $y = -\iint g\,dt\,dt$. With the antiderivatives and the limits shown:
- Write $y'' = (y')'$. An antiderivative of the constant $-g$ is $-g\tau$, so integrating from $0$ to $t$ and using Barrow's rule, $$y'(t) - y'(0) = \int_0^t (-g)\,d\tau = \Big[-g\tau\Big]_0^t = -gt.$$ With $v_0 = y'(0)$: $\ y'(t) = v_0 - gt$.
- An antiderivative of $v_0 - g\tau$ is $v_0\tau - \tfrac12 g\tau^2$ (check: $\frac{d}{d\tau}\big(v_0\tau - \tfrac12 g\tau^2\big) = v_0 - g\tau$), so $$y(t) - y(0) = \int_0^t (v_0 - g\tau)\,d\tau = \Big[v_0\tau - \tfrac12 g\tau^2\Big]_0^t = v_0t - \tfrac12 g t^2.$$
Two integrations produce two constants, one per order of the equation. The result of the notes, written without initial conditions, corresponds to $y_0 = v_0 = 0$ (an object released from rest at the origin):
$$y(t) = -\iint g\,dt\,dt = -\tfrac12 g\,t^2.$$- Check: $y' = -gt$, $y'' = -g$ ✓; $y(0) = 0$, $y'(0) = 0$ ✓.
- Dimensions: $[g t^2] = \text{m s}^{-2}\cdot\text{s}^2 = \text{m}$ ✓.
- Mass independence: $m$ cancelled — without drag every body falls the same way.
6.3 Complement (Robinson): from momentum balance to $F = ma$
Newton's second law states that the change $\Delta p$ in the momentum of a body equals the applied force $F$ times the time $\Delta t$ over which it acts, $\Delta p = F\,\Delta t$ [Robinson, Sec. 5.3, p. 30]. Dividing by $\Delta t$ and letting $\Delta t \to 0$ (Robinson notes this step is somewhat imprecise) gives
$$\frac{dp}{dt} = F(t).$$With $p = mv$, the product rule gives
$$\frac{dp}{dt} = \frac{d}{dt}(mv) = \frac{dm}{dt}\,v + m\,\frac{dv}{dt}.$$For constant mass $dm/dt = 0$, so $m\,\dfrac{dv}{dt} = F(t)$: the familiar $F = ma$. (For variable-mass bodies such as rockets, $F = d(mv)/dt$ does not apply: the momentum carried away by the exhaust must be included in a balance over the whole system, which is outside this course.)
6.4 Complement (Robinson Exercise 5.6): release from a height $h$
Release the body from rest at height $h$: $y_0 = h$, $v_0 = 0$ in the solution of 6.2, so $y(t) = h - \tfrac12 g t^2$ and $y'(t) = -gt$. The book leaves this case as Exercise 5.6 (an apple falling from height $h$).
- The impact time $t^*$ solves $y(t^*) = 0$: $\ h = \tfrac12 g t^{*2}$, so $t^* = \sqrt{2h/g}$ (the positive root, since $t^* > 0$).
- The impact velocity is $y'(t^*) = -g\sqrt{2h/g} = -\sqrt{2gh}$.
- Dimensions: $[\sqrt{h/g}] = \sqrt{\text{m}/(\text{m s}^{-2})} = \text{s}$ ✓; $[\sqrt{gh}] = \sqrt{\text{m}^2\text{s}^{-2}} = \text{m/s}$ ✓.
- Energy: $\tfrac12 m\,y'(t^*)^2 = \tfrac12 m (2gh) = mgh$: the potential energy lost equals the kinetic energy gained ✓.
A car of mass $m$ moving at $v_0$ brakes with a constant force $k$. Newton: $m\dot v = -k$. Integrating from $0$ to $t$, $v(t) - v_0 = \int_0^t (-k/m)\,d\tilde t = -kt/m$, so $v(t) = v_0 - kt/m$. It stops when $v(t_c) = 0$, i.e. $t_c = m v_0/k$. Integrating $\dot x = v_0 - kt/m$ from $0$ to $t_c$:
$$x(t_c) - x(0) = \Big[v_0\tilde t - \frac{k\tilde t^{\,2}}{2m}\Big]_0^{t_c} = v_0t_c - \frac{kt_c^2}{2m} = \frac{mv_0^2}{k} - \frac{mv_0^2}{2k} = \frac{mv_0^2}{2k}.$$With Robinson's data $m = 1000\ \text{kg}$, $k = 6500\ \text{N}$, $v_0 = 30\ \text{mph} = 13.41\ \text{m/s}$: $t_c = 1000 \times 13.41/6500 = 2.06\ \text{s}$ and the stopping distance is $1000 \times 13.41^2/(2\times 6500) = 13.8\ \text{m}$ (about 45 ft, as in his Fig. 5.7). Dimensions: $[mv_0^2/k] = \text{kg m}^2\text{s}^{-2}/(\text{kg m s}^{-2}) = \text{m}$ ✓.
6.5 Complement (Robinson): free fall with quadratic air drag, terminal velocity
At the speeds of a skydiver or a descending capsule, air resistance is not negligible. Robinson models it with a drag force proportional to the square of the speed and opposite to the motion [Sec. 7.5.2, Eq. (7.5), p. 52]:
$$m\frac{dv}{dt} = mg - kv|v|, \tag{7.5}$$where now $v$ is the downward velocity and $k > 0$ (units kg/m) collects air density, drag coefficient and frontal area. The term $v|v|$ (rather than $v^2$) makes the drag oppose the motion whatever its direction. For a body released from rest, $v \geq 0$ and the equation becomes
$$\frac{dv}{dt} = g - \frac{k}{m}v^2.$$This equation is first order, autonomous and nonlinear ($v^2$). Its right-hand side vanishes at $v^* = \sqrt{mg/k}$, the terminal velocity: if the body ever reaches $v^*$, weight and drag balance and it stays there. Robinson's skydiver ($m = 100\ \text{kg}$, $k \approx 1/3\ \text{kg/m}$, $g \approx 9.8\ \text{m/s}^2$) has $v^* = \sqrt{100 \times 9.8 \times 3} \approx 54.2\ \text{m/s}$ [p. 53].
Explicit solution (Robinson Exercise 8.6, worked here in full)
Write $w_T = \sqrt{mg/k}$, so that $k/m = g/w_T^2$ and $\dot v = \dfrac{g}{w_T^2}\,(w_T^2 - v^2)$. For $0 \le v < w_T$ we may divide (separation of variables, justified in Topic 2):
$$\frac{dv}{w_T^2 - v^2} = \frac{g}{w_T^2}\,dt.$$- Partial fractions: since $w_T^2 - v^2 = (w_T - v)(w_T + v)$, $$\frac{1}{w_T^2 - v^2} = \frac{1}{2w_T}\left(\frac{1}{w_T - v} + \frac{1}{w_T + v}\right).$$ Check: $\dfrac{(w_T + v) + (w_T - v)}{2w_T(w_T - v)(w_T + v)} = \dfrac{2w_T}{2w_T(w_T^2 - v^2)}$ ✓.
- Antiderivatives. For $\int \frac{dv}{w_T - v}$ substitute $u = w_T - v$, $du = -dv$: $\int \frac{-du}{u} = -\ln|u| = -\ln(w_T - v)$ (here $w_T - v > 0$). Directly, $\int\frac{dv}{w_T + v} = \ln(w_T + v)$. Hence $$\int\frac{dv}{w_T^2 - v^2} = \frac{1}{2w_T}\ln\frac{w_T + v}{w_T - v}.$$
- Barrow's rule between $v(0) = 0$ and $v(t)$ on the left, $0$ and $t$ on the right: $$\frac{1}{2w_T}\left[\ln\frac{w_T + v}{w_T - v}\right]_{0}^{v(t)} = \frac{1}{2w_T}\left(\ln\frac{w_T + v(t)}{w_T - v(t)} - \ln 1\right) = \frac{g}{w_T^2}\,\big[\tilde t\,\big]_0^t = \frac{g t}{w_T^2}.$$
- Multiply by $2w_T$ and exponentiate: $\dfrac{w_T + v}{w_T - v} = e^{2gt/w_T}$. Solve for $v$: $w_T + v = (w_T - v)e^{2gt/w_T}$, so $v\,(1 + e^{2gt/w_T}) = w_T\,(e^{2gt/w_T} - 1)$ and, dividing numerator and denominator by $e^{gt/w_T}$, $$v(t) = w_T\,\frac{e^{gt/w_T} - e^{-gt/w_T}}{e^{gt/w_T} + e^{-gt/w_T}} = w_T\tanh\!\left(\frac{g t}{w_T}\right).$$
- Long times: $\tanh s \to 1$ as $s \to \infty$, so $v(t) \to w_T$: the body approaches terminal velocity but never exceeds it.
- Short times: $\tanh s = s - s^3/3 + \dots$, so $v(t) \approx w_T \cdot gt/w_T = gt$: the drag-free result of 5.2 is recovered while $v \ll w_T$.
- Vanishing drag: $k \to 0$ gives $w_T \to \infty$ and again $v \to gt$ for every fixed $t$ (this is the limit Robinson asks for in Ex. 8.6).
- Time scale: $w_T/g$ is the time to reach about $76\%$ of $w_T$ ($\tanh 1 \approx 0.76$); for the skydiver $54.2/9.8 \approx 5.5\ \text{s}$.
- Dimensions: $[k v^2] = \text{kg m}^{-1}\cdot\text{m}^2\text{s}^{-2} = \text{N}$ ✓; $[\sqrt{mg/k}] = \sqrt{\text{kg m s}^{-2}/(\text{kg m}^{-1})} = \text{m/s}$ ✓; $[gt/w_T] = 1$ ✓.
In aerodynamic notation the drag is $D = \tfrac12\rho\,C_D A\,v^2$, so $k = \tfrac12\rho C_D A$ and $w_T = \sqrt{2mg/(\rho C_D A)}$. Terminal velocity grows as the air thins ($\rho$ decreases with altitude), which is why a re-entering capsule or a high-altitude jumper falls much faster in the upper atmosphere. That refinement makes $k$ depend on height and is beyond this topic.
7. Summary and lecture map
| Idea | Key statement | Class notes | Robinson | Lecture |
|---|---|---|---|---|
| ODE vs PDE | One independent variable vs several; $x' = x$, heat equation | p. 2 | Sec. 3.1, p. 11 | Sep 8 |
| Order, autonomy | Highest derivative; no explicit $t$ | p. 2 (autonomy p. 3) | Sec. 3.2, p. 13 | Sep 8 |
| Linearity | Form (L); superposition test: $u' = u$ linear, $u'' = (u')^2 + 1$ not | p. 2 | Sec. 3.3, p. 13–14 | Sep 8–10 (corrected) |
| Systems | Two or more dependent variables; $x' = 2x - y$, $y' = -x + 3y$ | p. 2 | Eq. (3.9), p. 15; Ch. 25 | Sep 8 |
| Population growth / decay | $x' = kx \Rightarrow x = x_0e^{kt}$; grows if $k > 0$, decays if $k < 0$ | p. 2 | Sec. 8.2–8.3, p. 61–64 | Sep 8 |
| Logistic correction | $x' = kx(1 - x/M)$, $x \to M$; analysed in Topic 2 | p. 2 (analysis p. 3) | Sec. 8.5, p. 66–68 | Sep 8–10 |
| Free fall | $-mg = my'' \Rightarrow y = -\frac12gt^2$ (from rest at $0$) | p. 2 | Sec. 5.3, Ex. 5.6 | Sep 8 |
| Quadratic drag | $v = w_T\tanh(gt/w_T)$, $w_T = \sqrt{mg/k}$ | — | Sec. 7.5.2, Ex. 8.6 | — |
| IVP, Picard's theorem | Existence and uniqueness of the solution of an IVP | p. 3 | Ch. 6 | Sep 10 (Topic 2) |
Topic 2 follows pages 3 and 4 of the class notes: initial value problems and Picard's theorem, autonomous equations and the stability of their stationary points, and three methods that give explicit solutions (separable equations, the integrating factor and exact equations).