Qualitative Models and Numerical Approximation

Learn what the law says before solving it

An exact formula is useful, but it is not the only way to understand a differential equation. The equation itself specifies a slope at every permissible point in the coordinate plane. Those local slopes reveal where solutions rise, fall, flatten, or move away from a threshold. This qualitative reading is especially valuable when the equation cannot be solved with elementary functions or when measurements make a numerical answer more appropriate than a symbolic one.

Consider the autonomous equation

$$\frac{dy}{dt}=y(4-y).$$

The right-hand side depends only on $y$, not explicitly on time. Its zeros, $y=0$ and $y=4$, are equilibrium solutions because a solution at either height has slope zero. For $0<y<4$, both factors are positive, so solutions rise. For $y>4$, the first factor is positive and the second is negative, so solutions fall. A solution beginning positive therefore moves toward $y=4$ rather than growing forever.

Equilibrium is a behavior, not only a value

An equilibrium is called stable when nearby solutions move toward it and unstable when nearby solutions move away. For the model above, $y=4$ is stable: values just below it rise and values just above it fall. The equilibrium $y=0$ is unstable within the positive domain because small positive values rise away from zero. These words summarize the direction information in the sign chart; they are not labels to apply by visual guesswork.

Qualitative analysis also tells you what to question in an answer. A proposed solution starting at $y(0)=2$ should remain between $0$ and $4$ and increase toward $4$. If a calculator output suddenly becomes negative or rises beyond $4$ without a changed model, something in the method or data entry needs attention. Such predictions are not optional preamble. They are the standard against which a later numerical table can be judged.

The units still matter. If $y$ measures thousands of organisms, the constant $4$ must be measured in the same units, while the equation’s overall coefficient would carry reciprocal-time units if one were present. A model with $y(4-y)$ may be dimensionally simplified for instruction, but an applied model should make units consistent. Qualitative shape does not excuse dimensional inconsistency.

Euler’s method follows one short tangent step

Euler’s method approximates a solution by repeatedly following the currently known tangent slope for a short time step. Starting from $(t_n,y_n)$ with step size $\Delta t$, the update is

$$y_{n+1}=y_n+f(t_n,y_n)\Delta t,$$

when $dy/dt=f(t,y)$. The expression $f(t_n,y_n)\Delta t$ is the predicted change in $y$ over that short interval. It is rate times time, so it has the same units as $y$. This makes Euler’s method a local linear approximation rather than a mysterious recurrence rule.

For $y'=y(4-y)$ with $y(0)=1$ and $\Delta t=0.5$, the initial slope is $1(4-1)=3$. The first Euler estimate is

$$y_1=1+3(0.5)=2.5.$$

At $t=0.5$, the method recomputes the slope from the estimated height: $2.5(4-2.5)=3.75$. The next estimate is $y_2=2.5+3.75(0.5)=4.375$. The overshoot above the stable equilibrium is a warning that a step of half a time unit is fairly coarse for this rapidly changing model. The next slope becomes negative and pulls the estimate back down, creating a numerical oscillation that is not necessarily part of the true solution.

Step size is a modeling choice

Reducing $\Delta t$ usually makes Euler’s polygonal path track a smooth solution more closely because each tangent approximation is used over less time. It also requires more calculations and can accumulate rounding error. The word “usually” matters: numerical accuracy must be assessed by comparison, refinement, or an error analysis, not merely asserted because a smaller number was used.

One practical check is to calculate with step sizes $\Delta t$ and $\Delta t/2$ over the same interval. If the answers agree to the required precision, confidence increases; if they differ substantially, the coarser estimate is not yet reliable. A graph provides another check. Euler points should follow the slope directions predicted by the differential equation. A table with no units or no stated step size cannot communicate what approximation it actually represents.

Retrieval check

For $dy/dt=(y-2)(y-5)$, make a sign chart and classify the equilibria at $y=2$ and $y=5$. Then use one Euler step of size $0.1\,\mathrm{s}$ from $y(0)=3$; show the rate, the predicted change, and the new estimate with units appropriate to your chosen quantity. Finally, say whether the estimate’s direction agrees with the sign chart. The next lessons switch representations again, using parametric and polar descriptions when a single graph $y=f(x)$ is not the most natural language for a curve.