Separable Equations

Suppose a hot cup of coffee cools in a room at constant temperature. Or a population grows subject to limited resources. Can we write down, in closed form, exactly how the quantity changes over time? Both situations reduce to a first-order differential equation. Its right-hand side factors into two pieces. One piece depends only on the independent variable. The other depends only on the unknown function. This factoring is not a curiosity. It is the single most exploitable structural feature a first-order equation can have. It converts a differential equation into two ordinary integrals. Separable equations matter because they give the first genuinely algorithmic method for solving nonlinear differential equations. This contrasts with the qualitative sketching of First-Order ODEs and Direction Fields. Before separable equations, a student can only say "the slope field looks like this." After separable equations, a student can produce an explicit formula. They can check it by substitution. They can use it to predict the future of the system exactly. This is the bridge from geometric intuition to symbolic computation. The big picture is this: "separability" is the first of a hierarchy of special structures. These are separable, then linear, then exact. Each makes first-order equations solvable by quadrature (that is, by integration). Each structure has its own recognition test and its own solution recipe. Separable equations are the simplest member of this family. They require only algebraic factoring and Integration Techniques already in hand. This article gives the precise definition of a separable equation. It proves the separation-of-variables theorem rigorously. This includes the existence and uniqueness statement needed to justify the method. It works two nontrivial examples with attention to domain and singular solutions. It shows exactly what goes wrong if the separability hypothesis is dropped.

Key takeaways

A first-order ODE is separable if it can be written as $\frac{dy}{dx} = g(x)h(y)$ .
Method: divide by $h(y)$ (where nonzero), then integrate both sides separately in $x$ and $y$ .
Points where $h(y) = 0$ give constant (equilibrium) solutions that the algebra can silently discard.
The resulting implicit relation $H(y) = G(x) + C$ defines $y$ implicitly; solving for $y$ explicitly is not guaranteed.
Local existence and uniqueness of solutions follows from continuity and differentiability of $g,h$ (Picard–Lindelöf).
Key idea: factoring the slope function into an $x$ -part and a $y$ -part turns a differential equation into two ordinary integrals.

Where this lives today

Separable equations are the working engine behind population models (logistic growth, as in Example 1), radioactive decay and pharmacokinetics ( $dy/dt = -ky$ ), Newton's law of cooling, and Torricelli's law for draining tanks. In physics, separable ODEs appear when solving for terminal velocity under quadratic drag, and in circuit theory for simple RC discharge. In probability and statistics, separable differential equations arise in deriving certain characteristic-function and survival-function relationships (e.g., the exponential distribution's memoryless property emerges from a separable hazard-rate equation). In numerical analysis and scientific computing, separable equations serve as the benchmark test cases against which numerical ODE solvers (Euler's method, Runge–Kutta) are validated, since the exact closed-form solution is known. The technique also foreshadows Exact Equations and Linear First-Order Equations, both of which generalize separability: an exact equation replaces the product structure $g(x)h(y)$ with a more general condition on partial derivatives, while a linear equation trades separability for an integrating-factor trick that always works regardless of factorization.

Ask the tutor: How do I find the exact interval on which a separable equation's explicit solution is valid, before I even solve the initial value problem?

\frac{dy}{dx} = f(x,y)

Intuition

Think of $dy/dx = g(x)h(y)$ as saying: the instantaneous steepness of the solution curve is a product of a factor that depends only on where you are horizontally ( $g(x)$ ) and a factor that depends only on where you are vertically ( $h(y)$ ). This is a strong restriction — most functions $f(x,y)$ do not factor this way (think of $f(x,y) = x + y$ , which is a sum, not a product, of separate pieces). When the restriction does hold, the direction field has a striking symmetry: moving along any horizontal line $y = y_0$ , the slopes are all proportional to $g(x)$ with the same constant of proportionality $h(y_0)$ ; moving along any vertical line $x = x_0$ , the slopes are all proportional to $h(y)$ scaled by $g(x_0)$ . The algebraic trick is to divide both sides by $h(y)$ , formally separating the differentials, and then recognize that each side is now a function of one variable alone, so each side can be integrated independently. The subtlety — and the one place careless students lose points — is that dividing by $h(y)$ requires $h(y) \neq 0$ . Wherever $h(y_0) = 0$ , the constant function $y \equiv y_0$ is itself a solution (since both sides of the ODE vanish), but this equilibrium solution can vanish from view once you divide it out. A complete answer must always list these separately.

Figure

Direction field for

dy/dx = xy

: slopes are the product of an

x

-factor and a

y

-factor

Central theorem

g: I \to \mathbb{R}

g

dy/dx = g(x)h(y)

Worked examples

Example 1 (logistic-type equation with asymmetric initial data). Solve

\frac{dy}{dx} = y(3-y), \qquad y(0) = 5.

Here $g(x) \equiv 1$ and $h(y) = y(3-y)$ , so the equation is separable. Equilibrium solutions occur where $h(y)=0$ , i.e. $y \equiv 0$ and $y \equiv 3$ . Since the initial value $y_0 = 5$ lies in $J = (3,\infty)$ where $h(y) = y(3-y) < 0$ and is continuous and nonvanishing, the theorem applies on $J = (3,\infty)$ . Separate variables:

\frac{dy}{y(3-y)} = dx.

Partial fractions: $\dfrac{1}{y(3-y)} = \dfrac{1}{3}\left(\dfrac{1}{y} + \dfrac{1}{3-y}\right)$ . Integrating,

\frac{1}{3}\Big(\ln|y| - \ln|3-y|\Big) = x + C_1 \quad\Longrightarrow\quad \ln\left|\frac{y}{3-y}\right| = 3x + C_2.

For $y > 3$ , $\dfrac{y}{3-y} < 0$ , so $\left|\dfrac{y}{3-y}\right| = \dfrac{y}{y-3}$ . Exponentiate:

\frac{y}{y-3} = Ae^{3x}, \qquad A = e^{C_2} > 0.

Apply $y(0) = 5$ : $\dfrac{5}{2} = A$ , so $A = 5/2$ . Solve for $y$ :

y = A e^{3x}(y-3) \implies y\left(1 - Ae^{3x}\right) = -3Ae^{3x} \implies y = \frac{3Ae^{3x}}{Ae^{3x}-1}.

With $A = 5/2$ :

y(x) = \frac{\frac{15}{2}e^{3x}}{\frac{5}{2}e^{3x}-1} = \frac{15e^{3x}}{5e^{3x}-2}.

Check: at $x=0$ , $y(0) = 15/(5-2) = 15/3 = 5$ . ✓ Differentiating and substituting back into $y(3-y)$ confirms the ODE (routine quotient-rule verification). Note the solution blows up as $5e^{3x} \to 2^+$ , i.e. as $x \to \left(\tfrac{1}{3}\ln\tfrac{2}{5}\right)^+ \approx -0.305^+$ ; the maximal interval of existence to the right is $\left(\tfrac13\ln\tfrac25,\infty\right)$ , and $y(x) \to 3^+$ as $x \to \infty$ , consistent with $y\equiv 3$ being a stable equilibrium approached from above.

Example 2 (trigonometric separable equation, non-polynomial $g$ and $h$ ). Solve

\frac{dy}{dx} = \frac{x \sec y}{1+x^2}, \qquad y(1) = \frac{\pi}{4}.

Here $g(x) = \dfrac{x}{1+x^2}$ (continuous on $I=\mathbb{R}$ ) and $h(y) = \sec y$ , continuous and nonvanishing on $J = (-\pi/2,\pi/2)$ , which contains $y_0 = \pi/4$ . The theorem applies with these $I, J$ . Separate:

\cos y\, dy = \frac{x}{1+x^2}\,dx.

Integrate both sides:

\sin y = \frac{1}{2}\ln(1+x^2) + C.

Apply $y(1) = \pi/4$ : $\sin(\pi/4) = \dfrac{\sqrt2}{2}$ , and $\dfrac12 \ln(1+1) = \dfrac12\ln 2$ , so

C = \frac{\sqrt2}{2} - \frac12 \ln 2.

Thus the implicit solution is

\sin y = \frac12 \ln(1+x^2) + \frac{\sqrt2}{2} - \frac12\ln 2.

Since $y_0 = \pi/4 \in (-\pi/2,\pi/2)$ where $\sin$ is invertible, we may solve explicitly there:

y(x) = \arcsin\!\left(\frac12 \ln(1+x^2) + \frac{\sqrt2}{2} - \frac12\ln 2\right),

valid on the set of $x$ where the argument of $\arcsin$ lies in $[-1,1]$ . As $x \to \pm\infty$ , $\ln(1+x^2)\to\infty$ , so the argument eventually exceeds $1$ ; the solution's maximal interval of existence is therefore a bounded interval around $x=1$ , determined by solving

\frac12\ln(1+x^2) = 1 - \frac{\sqrt2}{2} + \frac12\ln2

for the endpoint values of $x$ — a concrete illustration that solutions of nonlinear separable equations can have finite escape even when $g$ is defined for all $x$ .

h(y) \neq 0

Scope, edge cases & common traps

Situation	What to do
$h(y_0) = 0$ at the initial point	Check if $y \equiv y_0$ solves the ODE directly; if so it is a valid equilibrium solution, list it explicitly.
$h(y) = 0$ for some $y$ but not at $y_0$	Solve normally on the interval $J$ around $y_0$ where $h\neq 0$ ; the zero of $h$ marks a boundary the solution curve cannot cross (by uniqueness, distinct solutions cannot intersect).
Implicit relation $H(y) = G(x)+C$ cannot be solved explicitly	State the implicit solution as final answer; this is a fully legitimate closed form.
Antiderivatives $G$ or $H$ have no elementary closed form	Leave the answer as a definite integral (still "solved by quadrature").
Domain of $g$ or $h$ restricted (e.g. $\ln y$ , $\sqrt{y}$ )	The maximal interval of existence may be strictly smaller than where the original expression is defined; always re-check via $H^{-1}(G(x)) \in J$ .

The mathematician behind it

Origins · 1 minGottfried Wilhelm Leibniz and how this was discovered

Gottfried Wilhelm Leibniz

1646–1716 · Co-founder of calculus, developer of differential notation

Leibniz introduced the differential notation $dy$ , $dx$ that makes the separation-of-variables manipulation ( $dy/h(y) = g(x)\,dx$ ) visually and computationally natural. His correspondence in the 1690s with Johann Bernoulli contains explicit treatments of what we now call separable equations, arising from geometric problems such as finding curves of given tangent-line properties.

History

- 1691 — Leibniz and Jacob Bernoulli solve early separable-type equations arising from geometric tangent problems.

- 1694 — Leibniz formalizes the differential notation and technique of "separating" variables in letters to Johann Bernoulli.

- 1728–1740s — Leonhard Euler systematizes separable and related first-order equations as part of his broader calculus of differential equations.

- 1820s–1830s — Cauchy establishes rigorous existence theorems for ODEs, providing the modern justification (later sharpened by Lipschitz and Picard–Lindelöf) for why the separation manipulation is valid.

- 1876–1898 — Rudolf Lipschitz and Émile Picard formalize the Lipschitz condition and Picard iteration, giving the precise existence-uniqueness framework used to justify separable-equation solutions rigorously today.

Do I always need to check that h(y) ≠ 0 before dividing?

Yes. Dividing by $h(y)$ is only valid where $h(y) \neq 0$ , and the resulting family $H(y) = G(x) + C$ is built entirely from that restricted domain. Any $y$ -value where $h(y) = 0$ gives a constant function $y \equiv y^*$ that trivially satisfies the ODE (both sides equal zero), but this solution generally cannot be recovered from the formula $H(y)=G(x)+C$ for any finite choice of $C$ , since $H$ itself was defined using $1/h$ . Omitting this check is the most common source of "lost solutions" in textbook exercises, and can also hide uniqueness failures, as the counterexample with $y^{2/3}$ demonstrates.

Is the "general solution" H(y) = G(x) + C really the complete solution set?

Not quite — it is the complete set of solutions taking values strictly inside $J$ , where $h$ is continuous and nonzero, together with any equilibrium solutions from zeros of $h$ found separately. If $h$ has multiple zeros, the real line for $y$ splits into several open intervals $J_1, J_2, \dots$ on each of which the theorem applies separately, potentially producing different-looking implicit families on each piece, glued together only at the equilibrium levels. A fully complete answer lists all equilibrium solutions plus the family on each maximal $y$ -interval separately.

What if I can't solve H(y) = G(x) + C for y explicitly?

This is common and entirely acceptable — the implicit relation is a complete, rigorous solution of the differential equation. Many separable equations (e.g. $dy/dx = e^{-x^2}/\cos y$ ) lead to transcendental implicit relations with no elementary explicit inverse. In applied contexts, one can still extract qualitative information (monotonicity, asymptotic behavior, equilibrium stability) directly from the implicit equation or by implicit differentiation, without ever solving for $y$ explicitly.

How is separability different from being "solvable by an integrating factor" (linear equations)?

Separable equations require the very specific product structure $f(x,y) = g(x)h(y)$ ; this is a strong and often absent condition — for instance $dy/dx = x+y$ fails to have this form for any choice of $g,h$ . Linear equations, $dy/dx + p(x)y = q(x)$ , are solved by an integrating factor $\mu(x) = e^{\int p(x)dx}$ that works regardless of whether the equation is separable; indeed linear equations are separable only in the special homogeneous case $q\equiv 0$ . The two methods overlap only on that special case and are otherwise complementary tools for different equation shapes.

Can separable equations have solutions that only exist on a finite interval, even though g and h are defined everywhere?

Yes, exactly as illustrated in Example 2. Even when $g$ and $h$ are smooth on the whole real line, the explicit solution $\varphi(x) = H^{-1}(G(x))$ is only valid where $G(x)$ stays inside the range of $H$ . If $H$ has a bounded range (e.g. $H(y) = \sin y$ , range $[-1,1]$ ) while $G(x)$ is unbounded (e.g. grows like $\ln(1+x^2)$ ), the solution necessarily reaches the boundary of $H$ 's range at some finite $x$ , producing finite-time blow-up or loss of the solution's real-valuedness — a genuinely nonlinear phenomenon absent from linear equations with smooth coefficients on all of $\mathbb{R}$ .

Does the order in which I integrate matter, i.e. can I integrate the y-side first or the x-side first?

No — since the two integrals $\int dy/h(y)$ and $\int g(x)\,dx$ are with respect to entirely independent variables, they may be computed in either order, or even simultaneously; the final implicit equation $H(y) = G(x) + C$ is the same regardless. What does matter is being consistent about the single overall constant of integration: writing two separate constants (one per side) and then combining them into one is fine, but forgetting the constant altogether on one side is the error to avoid.

What role does the theorem's use of Picard–Lindelöf actually play, since the separation algorithm seems purely algebraic?

The algebraic manipulation of separating and integrating produces a candidate relation between $x$ and $y$ , but by itself does not certify that this candidate is the only solution, or even a genuine differentiable solution, of the original IVP. Picard–Lindelöf (via the local Lipschitz condition on $h$ ) is what guarantees, independently of the algebra, that a unique solution exists near the initial point — so that the candidate produced by separation, once verified by direct substitution, must in fact be that unique solution. Without this existence-uniqueness backing, as the counterexample shows, the algebra can produce one valid solution while missing others.