15.1 Population Models ended with a harvest: a constant effort removes and the stock settles at . Now let the harvester be alive. Its numbers grow only by eating the prey , so the “effort” is a variable that depends on how much prey there is. The state is a point in the plane, and the tools are those of 7.3 Phase Portraits & the Trace–Determinant Plane and 7.7 Nonlinear Systems & Linearization: nullclines, fixed points, a Jacobian at each and a trace-determinant classification. Three models test them: one with a conservation law, one with a limit cycle, and one that decides which of two species survives.
Transcript
- A predator is a harvester whose effort is itself a variable. Scale the Lotka-Volterra model so that x prime is x times one minus y, and y prime is alpha y times x minus 1. The nullclines are x equals 1 and y equals 1, and the fixed points are 0 0, a saddle, and 1 1, where the trace is 0, the determinant is alpha, and the eigenvalues are plus or minus i root alpha. That is a centre of the linearisation, with period 2 pi over root alpha, which cannot say whether orbits close or spiral.
- Dividing the two equations separates the variables, and integrating gives H equals alpha times x minus ln x, plus y minus ln y. This is constant along every solution, so the orbits are closed level curves around 1 1. With alpha equal to 1 and start 0.5, 0.8, H is 2.2163. Prey and predators both swing between 0.4779 and 1.8091, with period 6.5117, longer than the small orbit period of 2 pi, 6.2832. Each time average is exactly 1.
- But this is structurally fragile. Add a little prey crowding, epsilon, to the prey equation. The fixed point moves, the trace becomes minus epsilon, and the determinant is alpha times one minus epsilon. For epsilon equal to 0.1 the eigenvalues are minus 0.05 plus or minus 0.9474 i: a stable spiral. From 0.5, 0.8 the run reaches 0.9957, 0.8996 at t equal to 100. The conservation law was an accident of the exact form.
- The same fragility lives in the numerical method. Forward Euler multiplies the distance from a centre by the square root of 1 plus omega squared h squared at every step, so the orbit spirals outward. Up to time 40, Euler with step 0.01 ends with H up by 4.65 percent, and with step 0.001 by 0.39 percent, first order in h. Fourth order Runge-Kutta with step 0.02 ends at minus 9.4 times 10 to the minus 11. Euler turns a conservative system into an unstable one.
- A predator spends time handling each prey, so it eats a x over 1 plus a T h x per unit time: a Holling type 2 response, saturating at 1 over T h. Put that into logistic prey and you get Rosenzweig-MacArthur. With e equal to 1 and m equal to one half, the predator nullcline is the vertical line x star equals 1, and the coexistence point is x star equals 1, y star equals 2 times 1 minus 1 over K. The determinant is positive, so the trace decides everything: tau equals one half minus 3 over 2 K.
- The trace is negative for K below 3, zero at K equals 3, and positive beyond. At K equal to 3 the eigenvalues are plus or minus 0.4082 i, a Hopf bifurcation with onset period 15.39. At 2.7 trajectories spiral into the point. At 3.3 the fixed point repels, but the flow is trapped, so by the Poincare-Bendixson theorem a limit cycle must exist. It is isolated and attracting, so its amplitude belongs to the model, not to the start, and grows like the square root of K minus 3: a supercritical Hopf bifurcation.
- Now enrich the system by raising K. The prey maximum rises towards K, but the minimum collapses. At K equal to 5, prey run from 0.0357 to 4.4778, with period 21.59. At K equal to 10 the minimum is 2.5 times 10 to the minus 5, a drop by a factor of about 1400, and the predator minimum falls from 0.159 to 0.0036. Such troughs invite chance extinction. This is Rosenzweig's paradox of enrichment: more food, a less stable community.
- Now two species that share a resource, each hurting the other through coefficients alpha 12 and alpha 21. At the corner where species 1 sits at capacity, the eigenvalue for species 2 is r 2 times 1 minus alpha 21 K 1 over K 2, and positive means species 2 can invade. With 0.5 and 0.7 both can invade: coexistence at 0.7692, 0.4615, a stable node. With 1.5 and 1.3 neither can: both corners are stable and the interior point is a saddle, so whoever starts ahead wins. With 0.5 and 1.5, or 1.5 and 0.5, one species excludes the other. The interior point is stable exactly when alpha 12 times alpha 21 is below 1.
- Next, epidemics are another coupled nonlinear system, with the same mass action product S times I. Dividing dI by dS gives an exact conserved quantity, I plus S minus gamma over beta times ln S, like H here. But it does not oscillate: susceptibles are only used up, so the orbit runs to a final state on an axis. What replaces the Hopf bifurcation is a threshold: I first grows only if R 0, beta S 0 over gamma, exceeds 1, a close relative of the invasion tests above.
Predators as a dynamic harvest
Keep the prey growth and add a loss , where is the functional response (prey eaten per predator per unit time, per available prey). The predator converts eaten prey into births at rate and dies at rate :
Constant and give the Lotka-Volterra model (1925 and 1926, from memory), solved by one clever integral. A logistic with a saturating gives Rosenzweig-MacArthur (1963, from memory), solved by geometry and a bifurcation.
Lotka-Volterra and its conserved quantity
Scale time and the two populations so that the equations are
Prey grow when predators are fewer than , predators grow when prey exceed , and is the ratio of the two time scales. The nullclines are , for and , for , so the fixed points are and . The Jacobian is
The origin is a saddle (eigenvalues and ). At the trace is , the determinant and the eigenvalues : a centre of the linearisation, with period . A centre is the borderline case where the linearisation is silent, since a tiny nonlinear term can tip it into a spiral. Here nothing does, because of a conservation law.
Derivation. Divide the equations:
The variables separate. Integrating, , so
is constant along every solution. (Check: .) Each of and has a single minimum, at , and grows without bound as the variable goes to or . So the level sets of are closed curves around : a smooth bowl, and the solutions are the contour lines. The centre is a true nonlinear centre. This is the energy argument of 7.8 Conservative Systems & Lyapunov Functions in a new setting: plays the role of energy, with a surface that is not quadratic.
Numbers. With and start , . The orbit swings between and in both variables with period , longer than the small-orbit period : unlike the harmonic oscillator, the period depends on the energy. A separate scipy run (tolerance ) agrees with the sim to four decimals.
Volterra’s averages. Since and returns to its starting value over a period, : the time average of is . Likewise gives an average prey level of . The averages equal the fixed point whatever the amplitude, and for the orbit above both are (within in the sim). In the unscaled model , the averages are and ; a per-capita kill rate on both species replaces , , so the average prey goes up and the average predator goes down (Volterra’s principle, from memory).
Structural fragility and the numerics that expose it
Lotka-Volterra is structurally fragile: a small change in the equations changes the phase portrait. Add a little crowding, . The fixed point moves to and the Jacobian is
The trace is no longer . For , the eigenvalues are , a stable spiral decaying at rate . From the run reaches at (RK4 and scipy agree). The conserved quantity was an accident of the exact form, and a neutral orbit, which remembers its start forever, is a poor model of a population cycle anyway.
The same fragility lives in the method. Forward Euler multiplies the distance from a centre by at every step. With , start and , Euler with ends at and with at (first order in ), while RK4 with gives . Euler turns a conservative system into an unstable one, see 4.7 Energy & Symplectic Integrators.
Fixed points, Jacobians and the classification
Every planar model in this lesson is handled by one routine, which is the planar case of 7.7 Nonlinear Systems & Linearization.
- Find the nullclines of and , and intersect them to get the fixed points.
- At each, form and compute , .
- Classify. : saddle. and : node, stable if . and : spiral, stable if . with : linear centre.
- Trust the result except on the borderline (, or repeated roots), where the nonlinear terms decide. (Hartman and Grobman, from memory: near a hyperbolic fixed point the nonlinear flow is topologically a copy of the linear one.)
A saturating predator: Holling type II
A predator spends time handling each prey. With attack rate and handling time , it eats per unit time (Holling’s disc equation, 1959, from memory)
For small this is linear (the Lotka-Volterra term), and for large it saturates at . Scale so that the half-saturation is and take logistic prey. The Rosenzweig-MacArthur model is
We use and . The predator nullcline is the vertical line . The prey nullcline is the hump : it starts at , peaks at and vanishes at . At the predator growth rate is , positive iff , so predators invade only then, and the nullclines meet at
Jacobian at the coexistence point. Using :
The determinant is , so the point is never a saddle, and the sign of the trace decides everything. Simplifying with ,
It is negative for , zero at , positive beyond. Geometrically, the point is stable when the predator nullcline cuts the hump to the right of its top, and unstable when it cuts the rising side. At the eigenvalues are : a Hopf bifurcation (3.9 Bifurcations, 7.9 Limit Cycles & Hopf Bifurcations) with onset period .
| eigenvalues | fixed point | |||
|---|---|---|---|---|
| stable spiral | ||||
| borderline | ||||
| unstable spiral | ||||
| unstable spiral |
The sim’s numerical Jacobian and numpy agree on every digit shown.
The limit cycle and the paradox of enrichment
Above the fixed point repels, but the flow is trapped in a bounded region (the prey self-limit, the predator starves without prey). The planar Poincaré-Bendixson theorem (from memory: a bounded trajectory that does not approach a fixed point approaches a periodic orbit) forces a limit cycle. Unlike the Lotka-Volterra orbits it is isolated and attracting, so amplitude and period belong to the model, not to the start.
Long runs (RK4 in the sim, scipy DOP853 at tolerance for comparison) give the cycle:
| prey min – max | predator min – max | period | |
|---|---|---|---|
Near the bifurcation, the half-width of the prey swing divided by is , , for the first three rows (computed from the table): the cycle grows like , the mark of a supercritical Hopf bifurcation. The period starts at and lengthens.
Now look at enrichment. Raising lifts the prey maximum towards but the minimum collapses: from to it falls from to , a factor of about , and the predator minimum from to . Real troughs like these are a handful of individuals, and chance extinction follows. This is Rosenzweig’s paradox of enrichment (1971, from memory): more food makes the community less stable.
Competing species
Now two species that do not eat each other but share a resource, each hurting the other’s growth. The Lotka-Volterra competition model is
Here counts how many individuals of species one individual of species replaces. The nullclines are or , and or . Take . There are three fixed points on the axes, , , , and a fourth where the two lines cross, inside the quadrant or outside it.
Corners. At the origin : an unstable node, because tiny populations of both grow. At the matrix is triangular, so the eigenvalues are its diagonal entries:
The first is species at its capacity. The second is the growth rate of a rare species among species : positive means species can invade. So species invades at iff , and species invades at iff .
Four cases. With , and four choices of the pair :
| invades | outcome | fixed-point types | |
|---|---|---|---|
| both | coexistence | corners saddles, interior stable node | |
| neither | bistable | both corners stable nodes, interior saddle | |
| species only | species wins | stable node, saddle, no interior point | |
| species only | species wins | stable node, saddle, no interior point |
Worked example (coexistence). For the lines and cross where and . At the crossing the brackets in the equations vanish, so only the “outer” derivatives survive:
The trace is and the determinant , so : a stable node, eigenvalues and (numpy agrees). In general , so the interior point is stable iff and a saddle otherwise. For the determinant is and the eigenvalues and .
Bistability. The stable manifold of the saddle separates the basins of the two corners. In the sim the start ends at and at : whoever is ahead wins. Only a start exactly on that manifold reaches the interior point.
Competitive exclusion. In rows three and four the winner’s line lies beyond the other’s, there is no interior point, and every trajectory ends at the winner’s corner. The slogan (Gause, 1934, experiments on Paramecium, from memory) is that two species cannot share one limiting resource. The content is the invasion test: coexistence needs each species to limit itself more than its competitor.
The exercise: classify fixed points numerically
In the exercise you write a central-difference Jacobian, a trace-determinant classifier, and a bisection for the Hopf value of , using the facts above. The tests are the table entries.
Concept checks
Problems
For , with positive constants, find the interior fixed point, the eigenvalues of the Jacobian there, and a conserved quantity. For , , , give the fixed point and the period of small orbits.
Show worked solution
Setting and gives . The Jacobian there is , with trace and determinant , so . Separating and integrating gives . Check: . With the numbers: fixed point , , period .
Keep but take in the Rosenzweig-MacArthur model. Find , the Hopf value of , and for the coexistence point, the trace, the determinant and the eigenvalues.
Show worked solution
, and vanishes at . For : , , the off-diagonal entries are and , so . The eigenvalues are : an unstable spiral, surrounded by a limit cycle.
Take , and , . Show that the interior fixed point exists, find it, and classify it with the trace and the determinant. What does this say about coexistence when competition is almost as strong between species as within a species?
Show worked solution
The lines meet at , , . With , and , so : a stable node with eigenvalues and . The slow rate means a decay time of : as the determinant vanishes and coexistence becomes weak and slow.
What comes next
Coupled equations gave us a fragile conservation law, a robust Hopf bifurcation and a competition settled by a determinant. 15.3 Epidemic Models is another coupled nonlinear system. The SIR model has the same mass-action product , and dividing and integrating gives an exact conserved quantity, , like here. But it does not oscillate: susceptibles are only used up, so the orbit runs to a final state on an axis. What replaces the Hopf bifurcation is a threshold: first grows iff , a close relative of the invasion tests above.
Key Takeaways
- A predator is a harvester whose effort is itself a variable: nullclines, fixed points and Jacobians describe the plane.
- Lotka-Volterra conserves : closed orbits, period growing with amplitude ( against ), time averages equal to the fixed point.
- That conservation is structurally fragile: prey crowding gives trace , and forward Euler spirals outward ( in at , RK4 ).
- Classify a planar fixed point by (saddle) and by the signs of and ; on the borderline the nonlinear terms decide.
- Rosenzweig-MacArthur has trace , a Hopf bifurcation at and a supercritical limit cycle.
- Enrichment destabilises: from to the prey minimum falls from to .
- Competition has four outcomes set by the invasion conditions; the interior point is stable iff .
- SIR shares the coupling and a conserved quantity but has a threshold and no cycles.