When the clock does not matter
A great many laws of nature have a quiet symmetry: they do not depend on what time it is. A radioactive nucleus decays at the same rate today as a century from now; a warm cup cools by a rule that mentions the temperature gap but never the date; a population breeds according to its current size, not the year on the calendar. When the right-hand side of a first-order equation contains the unknown but not the independent variable — when dy/dt = f(y) with no t showing up on its own — the equation is called autonomous. The word means "self-governing": the rate of change is dictated entirely by the present state, with no external schedule.
This is not a small or rare special case — it is the typical shape of a fundamental model, and it buys you something remarkable. Because t never appears explicitly, the slope field you met two guides ago looks identical along every vertical line: the arrow at (t, y) has the same slope f(y) as the arrow at (t + 5, y), since shifting time changes nothing. The whole rich two-dimensional picture is just one column of information, copied sideways forever. That redundancy is your invitation to throw away a dimension entirely.
Collapsing the plane onto the phase line
Here is the move that makes autonomous equations so easy to read. Since the slope field is the same in every time-column, you lose nothing by keeping just one column — a single vertical y-axis — and recording, at each height y, only one fact: which way does the solution move? The sign of f(y) answers it. Where f(y) is positive, dy/dt is positive, so y is rising — draw an arrow pointing up. Where f(y) is negative, y is falling — draw an arrow pointing down. Where f(y) is zero, y holds still. That decorated axis is the phase line, and it captures the qualitative behavior of every solution at once.
The special heights where f(y) = 0 are the heart of the picture. At such a y, the rate of change is exactly zero, so a solution that starts there never moves — it sits as a flat, horizontal line forever. These constant solutions are the equilibrium solutions (also called steady states, fixed points, or critical points of the equation), and finding them is purely algebra: set f(y) = 0 and solve for y. No calculus, no integration — the hardest steady states are just the roots of a function. Between consecutive equilibria the sign of f(y) cannot flip without passing through zero, so on each open interval the arrows all point the same way, and any solution starting there marches steadily toward one end of the interval.
There is also a guarantee hiding here, and it is worth naming. Because the equation satisfies the hypotheses of the existence-uniqueness theorem when f is smooth, two solution curves can never cross, and in particular a non-equilibrium solution can never reach an equilibrium in finite time — it can only approach one forever. That is why a population creeping up toward its carrying capacity gets ever closer but never quite arrives, and why the arrows on the phase line tell the whole truth: a solution trapped between two equilibria is pinned there for all time, sliding monotonically from one toward the other.
Stable, unstable, and the sign of the derivative
Not all equilibria are alike, and the difference is the most useful idea in this guide. Picture a marble. Resting at the bottom of a bowl, it is at equilibrium, and if you nudge it, it rolls back — that is a stable equilibrium. Balanced on the peak of a dome, it is also at equilibrium, but the faintest nudge sends it racing away — that is an unstable equilibrium. The stability of an equilibrium asks exactly this: if a solution starts a little off the steady state, does it return or run away?
The phase line answers it by inspection. At a stable equilibrium the arrows on both sides point inward, toward it: below it y rises, above it y falls, so any nearby solution is squeezed back. At an unstable equilibrium the arrows point outward, away from it, and the smallest deviation grows. (A third kind, semi-stable, has arrows pointing the same way on both sides — attracting from one side, repelling from the other.) You can decide every case just from the pattern of up-and-down arrows you already drew.
There is a clean analytic test that says the same thing with one derivative, and it leans on an idea from Volume I. Near an equilibrium y* you can replace f(y) by its linear approximation f(y) ≈ f'(y*)·(y − y*), since f(y*) = 0. Let u = y − y* be the small deviation; then du/dt ≈ f'(y*)·u, an exponential whose fate is decided by the sign of one number. If f'(y*) < 0 the deviation decays — the equilibrium is stable. If f'(y*) > 0 it grows — unstable. If f'(y*) = 0 the linear test is silent and you must look closer (this is where semi-stable cases hide). So the whole question of stability comes down to the sign of the derivative of f at the steady state.
The logistic equation: the whole story in one model
One model puts everything above on stage at once, and it is the canonical example for good reason. The logistic equation dP/dt = r·P·(1 − P/K) models a population that grows but cannot grow forever. Read the right-hand side as a story: when P is tiny, the factor (1 − P/K) is near 1, so dP/dt ≈ r·P — pure exponential growth, the population racing upward. As P climbs toward K, that factor shrinks toward zero, throttling the growth; and if P ever overshoots K, the factor goes negative and the population falls back. The constant K is the carrying capacity, the largest population the environment will sustain, and r is the intrinsic growth rate.
Now run the machinery on it without solving a single integral. Set the right-hand side to zero: r·P·(1 − P/K) = 0 gives two equilibria, P = 0 and P = K. Their stability follows from the sign of f'(P), where f(P) = r·P·(1 − P/K) so f'(P) = r·(1 − 2P/K). At P = 0, f'(0) = r > 0 — unstable: an empty habitat is a knife-edge, and the slightest seed population grows away from extinction. At P = K, f'(K) = −r < 0 — stable: the carrying capacity attracts. Every solution that starts with any P between 0 and K therefore rises monotonically and levels off at K; every solution that starts above K sinks back down to K. The phase line has told you the destiny of the population without a formula in sight.
phase line for dP/dt = r P (1 - P/K) ( r, K > 0 )
P
--|-- P = K <== STABLE ( f'(K) = -r < 0 )
^ arrows point IN from both sides
| | 0 < P < K : f > 0, P rises -> toward K
| v
--|-- P = K from above: f < 0, P falls -> toward K
^
| 0 < P < K : f > 0, P rises -> away from 0
--o-- P = 0 <== UNSTABLE ( f'(0) = +r > 0 )
arrows point OUT (here, upward)Euler's method: your first numerical step
When you do want actual numbers and the integral refuses to cooperate, there is a route that never refuses. Recall the very first geometric reading of a first-order equation: dy/dt = f(t, y) hands you the slope of the solution at every point. Euler's method takes that literally. Stand at your known starting point (t0, y0). The equation tells you the slope right there. Follow that straight line for a tiny step of width h, and you land at a new, approximate point. Then ask the equation for the slope again at the new point, follow it for another step h, and repeat. You are walking the solution forward in short straight segments, steering at every step by the slope field itself.
- Start at the initial condition (t0, y0) given by your problem, and pick a step size h — the smaller, the more accurate, but the more steps you must take.
- Evaluate the slope at where you stand: m = f(t_n, y_n), straight from the equation's right-hand side.
- Take one straight step: y_{n+1} = y_n + h·m, and advance the time t_{n+1} = t_n + h. This is the whole update — it is just the point-slope form of a line, the same tangent-line approximation from Volume I.
- Repeat from step 2 at the new point, marching forward as far as you need. A table of (t_n, y_n) values is your approximate solution.
Why does this work, and how well? Each step is exactly the linear (first-order Taylor) approximation of the true solution, so the error you make on a single step is on the order of h^2. But to cross a fixed interval you need about 1/h steps, and those errors accumulate, leaving a total error on the order of h: halve the step, roughly halve the error. That makes Euler's method first-order accurate — honest but slow, the Model T of solvers. Real software uses smarter relatives (Runge–Kutta methods take a few cleverly weighted slope samples per step and achieve error on the order of h^4 for the same effort), but every one of them is the same idea you just learned: read the slope, take a step, repeat.
Two ways of knowing, side by side
Step back and notice that this guide has handed you the two ends of the toolkit named in the very first guide of this rung — the qualitative and the numerical — and they are complements, not rivals. The phase line costs almost nothing and tells you the long-term truth: which steady state wins, from which side, whether a small disturbance heals or grows. It never gives a number. Euler's method costs arithmetic and gives you numbers to any precision you can afford, the actual value of y at t = 7.3. It never tells you why. Run both on the logistic equation and they confirm each other: the phase line predicts the climb to K, and the Euler march fills in the S-shaped curve that gets there.
This pairing closes the first-order rung on a deliberate note. You began by classifying equations and reading their slope fields; you learned to win closed forms when the equation is separable, exact, or linear; and now you can handle the equations that yield no formula at all — reading their long-run behavior off an phase line and grinding out their numbers with Euler's method. The next rung lifts everything to second order, where two equilibria can spiral and oscillate rather than merely attract or repel — but the instincts you built here, looking for steady states and asking whether they hold, are exactly the ones you will carry up the ladder.