A Riccati equation is nonlinear but has a useful hidden linear structure. Its quadratic term can be removed by taking a logarithmic derivative of a solution of a second-order linear equation. This does not make singularities disappear: a zero of the linear solution becomes a pole of the Riccati solution. Tracking those zeros is part of solving the initial-value problem.
A different substitution works when one Riccati solution is already known. It produces a first-order linear equation, but it divides by the difference from that known solution. We will recover the excluded solution explicitly and explain why uniqueness makes the division legitimate for every other solution on a common interval.
The logarithmic-derivative transformation
Let be an open interval, let be nowhere zero, and let be real. Consider
The associated linear equation is
Theorem 1.2.1
(Riccati solutions and a linear equation)
If is a nontrivial real solution of (1), then on each connected component of ,
solves the Riccati equation. Conversely, every real solution on an interval arises this way from a nonvanishing solution of (1) on . Multiplying by a nonzero constant leaves unchanged.
For initial data , take the unique linear solution with
The Riccati solution's maximal interval within is the connected component containing on which this is nonzero. At any finite endpoint of that component,
In particular it diverges in magnitude and cannot be continued as a finite solution across . An endpoint at the boundary of requires separate information about the coefficients and is not asserted to be a pole.
Proof
(Both directions and the zero set)
On an interval where , differentiating gives
Substituting (1) simplifies the first two terms and yields . Conversely, given a solution on , choose and set
Then is nonzero and . Differentiating once more and using the Riccati equation gives (1), while .
The linear equation has continuous coefficients on and its initial-value problem has a unique solution on all of . The chosen normalization makes nontrivial. Any zero is simple: if both and vanished, uniqueness would force . Thus , and continuity of gives the displayed pole asymptotic.
The quotient solves the initial-value problem on the nonzero component. Uniqueness for the Riccati equation identifies any other solution with it on their common interval. At a zero endpoint the quotient diverges, so no finite solution can cross; there is no earlier obstruction inside the component. This proves maximality within .
Proof X-ray
Moving the blow-up into the zeros of a linear solution
The proof transfers every difficulty to the linear equation (1), where none can arise: a linear equation with continuous coefficients has a solution on all of . The hypothesis that is and nowhere zero is used twice, to write the coefficient in (1) and to divide by in the quotient.
The non-obvious move is the simplicity of the zeros. It comes from uniqueness for the linear initial-value problem, not for the Riccati equation: would give , contradicting . Simplicity is what makes the pole of order one with leading coefficient .
The argument closes with Riccati uniqueness: any solution through equals the quotient where both exist, and the quotient is unbounded at a zero of , so no solution can be finite beyond it. If vanished somewhere, neither (1) nor the quotient would be defined there, which is the situation of the counterexample below.
When a particular Riccati solution is known
Theorem 1.2.2
(Reduction to a first-order linear equation)
Suppose solves the Riccati equation on an interval . For every other solution on , the difference is nowhere zero, and
is nowhere zero and satisfies the linear equation
Conversely, if solves this equation and is nonzero on an interval, then is a Riccati solution there. The solution itself is not represented by any finite and must be retained separately.
Proof
(Subtract before taking the reciprocal)
For two solutions on a common interval, uniqueness implies that either they agree everywhere or never meet. Thus the reciprocal is defined for a different solution. Writing and subtracting the two equations gives
Consequently . Reversing these algebraic steps proves the converse wherever . If vanishes, the formula is undefined; one must examine the interval rather than pass through that point by algebraic cancellation.
Worked examples
Worked example 1.2.3
(A tangent solution and its maximal interval)
For with , take , , . Equation (1) is , with , , so . Therefore
The linear solution continues through its zeros, but the Riccati solution tends to opposite infinities at the two endpoints. Other components of define other local quotient solutions; they are not continuations through either pole of this initial-value solution.
Worked example 1.2.4
(A known equilibrium turns a nonlinear equation linear)
For , choose . Then , so
For , , and
The denominator vanishes at the finite right endpoint and the solution tends to there. The equilibrium is absent from the reciprocal formula and must be added. The other equilibrium is represented: corresponds to the constant solution .
Worked example 1.2.5
(Which initial values blow up)
For with , take , , . Equation (1) is with , , so
A zero of needs , which is possible only when , because takes exactly the values in . For the maximal interval is ; gives , as in the previous example. For it is , so the pole lies in the past. For , and the solution is global. Edge case : never vanishes and . Here the linearization keeps both equilibria, unlike the reciprocal substitution with .
Counterexample mode
A zero quadratic coefficient is not a Riccati pole
One might expect the solution to misbehave where the hypothesis fails. Consider with . Separation on the initial interval gives
The coefficient vanishes at , but the solution is regular there. Formula (1) contains and the quotient contains , so the preceding theorem cannot be applied across zero as stated. This does not imply a singularity of the original equation; it means the chosen transformation has lost its hypothesis.
The theorem also does not say that every solution must have a finite pole. For example, has the global constant solutions , and is another global solution. Whether a zero occurs in the associated linear solution depends on its initial data.
Common mistake
(Continuing a solution through its pole)
The quotient is defined on every component of , and for the formula is defined on as well. That piece is a different solution with different data, not the continuation of the solution with . A solution of an initial-value problem lives on an interval containing , and it cannot be continued past a point where it is unbounded.
Common mistake
(Losing the known solution in the reciprocal substitution)
After the general solution never equals , since that would need . For with , the formula misses the equilibrium ; the solution set is the formula together with .
Exercises
Exercise 1.2.6
(Solve with using the associated linear equation. Give the maximal interval and locate its pole.)
Foundation
Solution(2)
- 1
Set up the linear problem
Here , , so (1) is with , . Hence .
- 2
Form the quotient and read off the zero
. The zero of is , so the maximal interval containing is , and as .
Exercise 1.2.7
(For with , use the particular solution and verify that the resulting solution is on all of .)
Foundation
Solution(3)
- 1
Find the initial value of
, and gives .
- 2
Check that never vanishes
Key step
for all , so on and is defined everywhere.
- 3
Simplify
, a global solution.
Exercise 1.2.8
(For , , derive the associated linear equation and find the maximal solution interval.)
Core
Solution(3)
- 1
Identify the coefficients
, , , so and . Equation (1) is .
- 2
Apply the normalization
, , so .
- 3
Form the quotient and find the component
. The component of containing is , the maximal interval.
Exercise 1.2.9
(Suppose two solutions of the same linear equation (1) are nonzero on a connected interval and give the same logarithmic-derivative Riccati solution. Show that and differ by a nonzero constant factor.)
CoreProof
Solution(2)
- 1
Cancel the common factor
and give , that is, .
- 2
Differentiate the ratio
∎, so by connectedness is a constant, which is nonzero because is nonzero.