Skip to content

Singular Solutions and Systems of First-Order ODEs

CSIR NET ODE: Clairaut and Lagrange equations, p- and c-discriminants, envelopes, linear systems x' = Ax, matrix exponential, Liouville formula, stability.

  • 23 min read
  • Updated
On this page · 12 sections
  1. Prerequisites
  2. The Hook: A Parabola Drawn with Straight Lines
  3. Part 1: Equations Not Solved for \( y' \)
  4. Singular Solutions and Discriminants
  5. Part 2: Systems of First-Order ODEs
  6. Memory Tricks
  7. Don’t Confuse
  8. Common Mistakes
  9. Practice
  10. Quick Revision Table
  11. Key Takeaways
  12. What’s Next

Prerequisites

The Hook: A Parabola Drawn with Straight Lines

Part 1: Equations Not Solved for \( y' \)

Write \( p = \frac{dy}{dx} \). A first-order equation of higher degree has the form \( F(x, y, p) = 0 \), for example \( p^2 - (x + y)p + xy = 0 \).

CaseMethodExample
Solvable for \( p \)factor into \( p = f_i(x, y) \); solve each; the general solution is the product of the families\( (p - x)(p - y) = 0 \): \( \left(y - \frac{x^2}{2} - c\right)(y - ce^x) = 0 \)
Solvable for \( y \): \( y = g(x, p) \)differentiate in \( x \): \( p = g_x + g_p \frac{dp}{dx} \), an equation in \( x, p \)Clairaut, Lagrange (below)
Solvable for \( x \): \( x = g(y, p) \)differentiate in \( y \): \( \frac{1}{p} = g_y + g_p \frac{dp}{dy} \)\( x = y + p^2 \)

Clairaut’s equation

$$\boxed{y = px + f(p)}$$

Differentiate in \( x \): \( p = p + x\frac{dp}{dx} + f'(p)\frac{dp}{dx} \), so

$$\frac{dp}{dx}\big(x + f'(p)\big) = 0.$$
  • \( \frac{dp}{dx} = 0 \): \( p = c \), giving the general solution \( y = cx + f(c) \), a family of straight lines. (Rule: replace \( p \) by \( c \).)
  • \( x + f'(p) = 0 \): eliminate \( p \) between this and the equation. This gives the singular solution, the envelope of the lines.
Clairaut equationGeneral solutionSingular solution
\( y = px + p^2 \)\( y = cx + c^2 \)\( p = -\frac{x}{2} \): \( x^2 + 4y = 0 \)
\( y = px + \frac{1}{p} \)\( y = cx + \frac{1}{c} \)\( p^2 = \frac{1}{x} \): \( y^2 = 4x \)
\( y = px + a\sqrt{1 + p^2} \)\( y = cx + a\sqrt{1 + c^2} \)\( x^2 + y^2 = a^2 \)
\( y = px - p^3 \)\( y = cx - c^3 \)\( 3p^2 = x \): \( 27y^2 = 4x^3 \)

Check the first row: with \( y = -\frac{x^2}{4} \), \( p = -\frac{x}{2} \) and \( px + p^2 = -\frac{x^2}{2} + \frac{x^2}{4} = -\frac{x^2}{4} = y \).

Some equations become Clairaut after a substitution. With \( u = x^2 \), \( v = y^2 \) we get \( \frac{dv}{du} = \frac{yp}{x} \), so an equation \( y^2 = x^2 P + f(P) \) with \( P = \frac{yp}{x} \) becomes the Clairaut equation \( v = uP + f(P) \) in \( (u, v) \).

Lagrange (d’Alembert) equation

\( y = x\,g(p) + f(p) \) with \( g(p) \ne p \). Differentiating gives \( p - g(p) = \big(x g'(p) + f'(p)\big)\frac{dp}{dx} \), which is linear in \( x \) as a function of \( p \):

$$\frac{dx}{dp} - \frac{g'(p)}{p - g(p)}\,x = \frac{f'(p)}{p - g(p)}.$$

Solve for \( x(p) \); together with \( y = x g(p) + f(p) \) this is the general solution in parametric form. Roots of \( p = g(p) \), lost in the division, can give extra (often singular) solutions.

Singular Solutions and Discriminants

Let the general solution be the one-parameter family \( \phi(x, y, c) = 0 \).

  • A singular solution is a solution of \( F(x, y, p) = 0 \) that is not obtained from \( \phi(x, y, c) = 0 \) for any value of \( c \). Geometrically it is the envelope of the family of integral curves.
  • p-discriminant: eliminate \( p \) from \( F = 0 \) and \( \frac{\partial F}{\partial p} = 0 \).
  • c-discriminant: eliminate \( c \) from \( \phi = 0 \) and \( \frac{\partial \phi}{\partial c} = 0 \).

The two discriminants contain more than the envelope. The classical factorisation (envelope \( E \), tac-locus \( T \), nodal locus \( N \), cusp locus \( C \)) is:

$$\boxed{\text{p-discriminant} \supseteq E \, T^2 \, C, \qquad \text{c-discriminant} \supseteq E \, N^2 \, C^3}$$
LocusIn p-discIn c-discSolves the ODE?
Envelope \( E \)onceonceyes: this is the singular solution
Tac-locus \( T \) (two non-consecutive curves touch)squaredabsentgenerally no
Nodal locus \( N \) (curves cross themselves)absentsquaredgenerally no
Cusp locus \( C \)oncecubedgenerally no

Procedure. Compute both discriminants. A factor that appears in both, to the first power, and that satisfies the differential equation, is the singular solution. Always substitute to confirm; a discriminant factor is only a candidate.

graph TD
    A["F(x, y, p) = 0"] --> B["General solution phi(x, y, c) = 0"]
    A --> C["p-discriminant: eliminate p from F = 0, F_p = 0"]
    B --> D["c-discriminant: eliminate c from phi = 0, phi_c = 0"]
    C --> E{"Factor common to both, power 1?"}
    D --> E
    E -- yes --> F{"Satisfies the ODE?"}
    F -- yes --> G["Singular solution (envelope)"]
    F -- no --> H["Extraneous locus"]
    E -- "only in p-disc, squared" --> I["Tac-locus"]
    E -- "only in c-disc, squared" --> J["Nodal locus"]

Simpler cases:

  • \( p^2 = y \): general solution \( y = \frac{(x - c)^2}{4} \). Both discriminants give \( y = 0 \), and \( y = 0 \) solves the equation: singular solution, the envelope of the parabolas.
  • \( y' = 3y^{2/3} \) from the previous page: general solution \( y = (x - c)^3 \); \( y = 0 \) is not of this form for any \( c \) and is the envelope. This is why the IVP \( y(0) = 0 \) is not unique.

Part 2: Systems of First-Order ODEs

A system \( \mathbf{x}' = \mathbf{f}(t, \mathbf{x}) \), \( \mathbf{x}(t_0) = \mathbf{x}_0 \in \mathbb{R}^n \) is handled by the same theorems as the scalar case, with \( \lvert \cdot \rvert \) replaced by any norm on \( \mathbb{R}^n \). If \( \mathbf{f} \) is continuous and Lipschitz in \( \mathbf{x} \), the IVP has a unique local solution; Picard iterates and Gronwall work unchanged.

Every \( n \)th-order equation is a system. For \( y^{(n)} = g(t, y, y', \dots, y^{(n-1)}) \), put \( x_1 = y \), \( x_2 = y' \), \dots, \( x_n = y^{(n-1)} \):

$$x_1' = x_2, \ \dots, \ x_{n-1}' = x_n, \ x_n' = g(t, x_1, \dots, x_n).$$

For the linear equation \( y'' + p(t)y' + q(t)y = 0 \) this gives the companion matrix \( \begin{pmatrix} 0 & 1 \\ -q & -p \end{pmatrix} \). This is why an \( n \)th-order IVP needs \( n \) initial conditions.

Linear systems: structure

$$\mathbf{x}' = A(t)\mathbf{x} + \mathbf{g}(t), \qquad A, \mathbf{g} \text{ continuous on an interval } I.$$
  1. Global existence. For any \( t_0 \in I \), \( \mathbf{x}_0 \), there is a unique solution on all of \( I \) (the right side is Lipschitz on every strip).
  2. Solution space. The solutions of the homogeneous system \( \mathbf{x}' = A(t)\mathbf{x} \) form a vector space of dimension exactly \( n \). The map \( \mathbf{x} \mapsto \mathbf{x}(t_0) \) is an isomorphism onto \( \mathbb{R}^n \).
  3. Fundamental matrix. A matrix \( \Phi(t) \) whose columns are \( n \) independent solutions: \( \Phi' = A\Phi \). The general solution is \( \mathbf{x} = \Phi(t)\mathbf{c} \), and the solution with \( \mathbf{x}(t_0) = \mathbf{x}_0 \) is \( \Phi(t)\Phi(t_0)^{-1}\mathbf{x}_0 \).
  4. Liouville (Abel–Jacobi) formula.
$$\boxed{\det \Phi(t) = \det \Phi(t_0)\, \exp\left(\int_{t_0}^{t} \operatorname{tr} A(s)\,ds\right)}$$

So \( \det \Phi \) is either never zero or identically zero on \( I \): solutions independent at one \( t \) are independent at every \( t \).

  1. Variation of parameters. A particular solution of the non-homogeneous system is
$$\mathbf{x}_p(t) = \Phi(t)\int_{t_0}^{t} \Phi(s)^{-1}\mathbf{g}(s)\,ds.$$

The scalar version of 4 and 5 (Abel’s formula for the Wronskian, variation of parameters for \( y'' + py' + qy = g \)) is on Linear ODEs and Variation of Parameters.

Constant coefficients: the matrix exponential

$$e^{tA} = \sum_{k=0}^{\infty} \frac{t^k A^k}{k!}, \qquad \frac{d}{dt}e^{tA} = Ae^{tA}, \qquad e^{0A} = I.$$

So \( \Phi(t) = e^{tA} \) is the fundamental matrix with \( \Phi(0) = I \), and

$$\boxed{\mathbf{x}(t) = e^{(t - t_0)A}\mathbf{x}_0 + \int_{t_0}^{t} e^{(t - s)A}\mathbf{g}(s)\,ds.}$$
PropertyHolds?
\( e^{(t+s)A} = e^{tA}e^{sA} \); \( \left(e^{tA}\right)^{-1} = e^{-tA} \)always
\( \det e^{tA} = e^{t\,\operatorname{tr}A} \)always (Liouville)
\( e^{A + B} = e^A e^B \)if \( AB = BA \); can fail otherwise
\( e^{PDP^{-1}} = Pe^{D}P^{-1} \)always
eigenvalues of \( e^{tA} \) are \( e^{t\lambda_i} \)always

Computing \( e^{tA} \) for \( 2 \times 2 \) matrices.

CaseFormula
Distinct eigenvalues \( \lambda_1 \ne \lambda_2 \)\( e^{tA} = \frac{\lambda_1 e^{\lambda_2 t} - \lambda_2 e^{\lambda_1 t}}{\lambda_1 - \lambda_2}\,I + \frac{e^{\lambda_1 t} - e^{\lambda_2 t}}{\lambda_1 - \lambda_2}\,A \)
Repeated eigenvalue \( \lambda \)\( e^{tA} = e^{\lambda t}\big(I + t(A - \lambda I)\big) \) (Cayley–Hamilton: \( (A - \lambda I)^2 = 0 \))
\( A = \begin{pmatrix} a & -b \\ b & a \end{pmatrix} \)\( e^{tA} = e^{at}\begin{pmatrix} \cos bt & -\sin bt \\ \sin bt & \cos bt \end{pmatrix} \)
Nilpotent \( N \), \( N^k = 0 \)\( e^{tN} = I + tN + \dots + \frac{t^{k-1}N^{k-1}}{(k-1)!} \) (finite sum)

Phase portraits and stability in the plane

For \( \mathbf{x}' = A\mathbf{x} \), \( A \) real \( 2 \times 2 \) with \( \det A \ne 0 \), let \( \tau = \operatorname{tr}A \), \( \Delta = \det A \). The eigenvalues are the roots of \( \lambda^2 - \tau\lambda + \Delta = 0 \).

ConditionEigenvaluesOrigin is a
\( \Delta < 0 \)real, opposite signssaddle (unstable)
\( \Delta > 0 \), \( \tau^2 - 4\Delta > 0 \)real, same signnode: stable if \( \tau < 0 \), unstable if \( \tau > 0 \)
\( \Delta > 0 \), \( \tau^2 = 4\Delta \)real, equalstar or improper (degenerate) node
\( \Delta > 0 \), \( \tau^2 - 4\Delta < 0 \), \( \tau \ne 0 \)complex, \( \operatorname{Re} \ne 0 \)spiral (focus): stable if \( \tau < 0 \)
\( \Delta > 0 \), \( \tau = 0 \)purely imaginarycentre (stable, not asymptotically)
graph TD
    A["2x2 system x' = Ax, det A not 0"] --> B{"det A negative?"}
    B -- yes --> S["Saddle"]
    B -- no --> C{"tr A = 0?"}
    C -- yes --> Ce["Centre"]
    C -- no --> D{"sign of tr^2 - 4 det"}
    D -- positive --> N["Node"]
    D -- zero --> DN["Degenerate or star node"]
    D -- negative --> Sp["Spiral"]
    N --> T{"tr A negative?"}
    Sp --> T
    DN --> T
    T -- yes --> St["Asymptotically stable"]
    T -- no --> U["Unstable"]

Stability of \( \mathbf{x}' = A\mathbf{x} \) in any dimension.

  • Every solution tends to 0 as \( t \to \infty \) (asymptotic stability) iff every eigenvalue has negative real part. For \( 2 \times 2 \): iff \( \tau < 0 \) and \( \Delta > 0 \).
  • Every solution is bounded on \( [0, \infty) \) (stability) iff every eigenvalue has \( \operatorname{Re}\lambda \le 0 \), and those with \( \operatorname{Re}\lambda = 0 \) are semisimple (algebraic multiplicity = geometric multiplicity, i.e. no Jordan block of size \( > 1 \)).
  • \( A = \begin{pmatrix} 0 & 1 \\ 0 & 0 \end{pmatrix} \) has both eigenvalues 0, yet \( \mathbf{x}(t) = (c_1 + c_2 t, \ c_2) \) is unbounded when \( c_2 \ne 0 \). The Jordan block is the reason.

Memory Tricks

  • Clairaut: “replace \( p \) by \( c \)” for the lines; “differentiate in \( p \)” for the envelope. \( x + f'(p) = 0 \) is the derivative of \( px + f(p) \) with respect to \( p \).
  • “E T² C and E N² C³”. The envelope appears once in both discriminants; the tac-locus is squared in \( p \) only; the nodal locus is squared in \( c \) only; the cusp is in both (powers 1 and 3).
  • Liouville = trace. \( \det e^{tA} = e^{t\,\operatorname{tr}A} \): you never need the eigenvalues to get the determinant.
  • Trace–determinant: “D negative, saddle; T zero, centre; T negative, sink”.
  • Zero real part needs a diagonal block. Imaginary or zero eigenvalues give bounded solutions only if they come without Jordan blocks.

Don’t Confuse

ThisvsThat
Singular solution (not in the family for any \( c \))Particular solution (one value of \( c \))
p-discriminant factorSingular solution (must appear in both discriminants and satisfy the ODE)
Tac-locus (squared, p-disc only)Nodal locus (squared, c-disc only)
\( e^{A+B} = e^Ae^B \) when \( AB = BA \)general \( A, B \) (fails)
Stable (bounded): \( \operatorname{Re}\lambda \le 0 \), semisimple on the axisAsymptotically stable: all \( \operatorname{Re}\lambda < 0 \)
Centre (\( \tau = 0 \), \( \Delta > 0 \))Spiral (\( \tau \ne 0 \), \( \tau^2 < 4\Delta \))

Common Mistakes

Practice

All questions below are original practice questions written for this page. They are not from any CSIR NET paper. In Level 3, as in Part C, one or more options may be correct.

Level 1: Foundation

  1. Write the general and singular solutions of \( y = px + \frac{1}{p} \).
  2. Find \( \det e^{tA} \) for \( A = \begin{pmatrix} 1 & 5 \\ 7 & -4 \end{pmatrix} \).
  3. Classify the origin for \( \mathbf{x}' = \begin{pmatrix} 0 & 1 \\ -1 & 0 \end{pmatrix}\mathbf{x} \).
Answers
  1. General: \( y = cx + \frac{1}{c} \). Singular: \( x - \frac{1}{p^2} = 0 \), so \( p = \frac{1}{\sqrt{x}} \) (taking \( x > 0 \)) and \( y = \sqrt{x} + \sqrt{x} = 2\sqrt{x} \), i.e. \( y^2 = 4x \).
  2. \( \operatorname{tr}A = -3 \), so \( \det e^{tA} = e^{-3t} \).
  3. \( \tau = 0 \), \( \Delta = 1 \): a centre. Solutions are \( (c_1\cos t + c_2\sin t, \ -c_1\sin t + c_2\cos t) \), circles about the origin.

Level 2: Part B Style

Practice Q1. The singular solution of \( y = px - p^2 \) is:

(a) \( x^2 = 4y \) (b) \( x^2 = -4y \) (c) \( y = 0 \) (d) \( y^2 = 4x \)

Answer: (a)

Clairaut with \( f(p) = -p^2 \): \( x - 2p = 0 \), \( p = \frac{x}{2} \), \( y = \frac{x^2}{2} - \frac{x^2}{4} = \frac{x^2}{4} \). Check: \( y' = \frac{x}{2} = p \) and \( px - p^2 = \frac{x^2}{4} \).

Practice Q2. For \( A = \begin{pmatrix} 3 & 1 \\ 0 & 3 \end{pmatrix} \), \( e^{tA} \) equals:

(a) \( \begin{pmatrix} e^{3t} & e^{t} \\ 0 & e^{3t} \end{pmatrix} \) (b) \( e^{3t}\begin{pmatrix} 1 & t \\ 0 & 1 \end{pmatrix} \) (c) \( \begin{pmatrix} e^{3t} & 0 \\ 0 & e^{3t} \end{pmatrix} \) (d) \( e^{3t}\begin{pmatrix} 1 & 1 \\ 0 & 1 \end{pmatrix} \)

Answer: (b)

\( A = 3I + N \) with \( N = \begin{pmatrix} 0 & 1 \\ 0 & 0 \end{pmatrix} \), \( N^2 = 0 \), and \( 3I \) commutes with \( N \). So \( e^{tA} = e^{3t}(I + tN) \). Option (a) exponentiates entrywise, which is wrong for non-diagonal matrices.

Practice Q3. Let \( \Phi(t) \) be a fundamental matrix of \( \mathbf{x}' = \begin{pmatrix} t & 1 \\ \sin t & 2 - t \end{pmatrix}\mathbf{x} \) with \( \det \Phi(0) = 5 \). Then \( \det \Phi(1) \) is:

(a) 5 (b) \( 5e \) (c) \( 5e^2 \) (d) \( 5e^{1 + \cos 1} \)

Answer: (c)

\( \operatorname{tr}A(t) = t + 2 - t = 2 \). Liouville: \( \det \Phi(1) = 5\exp\left(\int_0^1 2\,dt\right) = 5e^2 \). The off-diagonal entries play no part.

Practice Q4. For \( \mathbf{x}' = \begin{pmatrix} -1 & 2 \\ -3 & -1 \end{pmatrix}\mathbf{x} \), the origin is:

(a) a saddle (b) a centre (c) a stable spiral (d) an unstable node

Answer: (c)

\( \tau = -2 \), \( \Delta = 1 + 6 = 7 \), \( \tau^2 - 4\Delta = 4 - 28 < 0 \). Complex eigenvalues \( -1 \pm i\sqrt{6} \) with negative real part: a stable spiral.

Level 3: Part C Style

Practice Q5. Consider \( p^2(2 - 3y)^2 = 4(1 - y) \), \( p = \frac{dy}{dx} \), whose general solution is \( (x - c)^2 = y^2(1 - y) \). Which statements are true?

(a) \( y = 1 \) is a singular solution. (b) \( y = \frac{2}{3} \) is a singular solution. (c) \( y = 0 \) is a nodal locus of the family and does not satisfy the equation. (d) The p-discriminant is \( (2 - 3y)^2(1 - y) = 0 \).

Answer: (a), (c) and (d)
  • (d) True. \( Ap^2 = B \) has a repeated root in \( p \) iff \( AB = 0 \), with \( A = (2 - 3y)^2 \), \( B = 4(1 - y) \).
  • (a) True. \( y = 1 \) appears once in each discriminant (c-discriminant: \( y^2(1 - y) = 0 \)), and \( y = 1 \), \( p = 0 \) satisfies the equation.
  • (b) False. \( y = \frac{2}{3} \) is squared in the p-discriminant only: a tac-locus. With \( y = \frac{2}{3} \), \( p = 0 \), the equation reads \( 0 = \frac{4}{3} \).
  • (c) True. \( y = 0 \) is squared in the c-discriminant only, and each curve \( x - c = \pm y\sqrt{1 - y} \) crosses itself at \( y = 0 \). Substituting \( y = 0 \), \( p = 0 \) gives \( 0 = 4 \).

Practice Q6. Let \( A = \begin{pmatrix} 0 & 1 \\ -2 & -3 \end{pmatrix} \) and consider \( \mathbf{x}' = A\mathbf{x} \). Which statements are true?

(a) Every solution tends to \( \mathbf{0} \) as \( t \to \infty \). (b) The origin is a saddle. (c) \( \det e^{tA} = e^{-3t} \). (d) There is a non-zero solution of the form \( e^{-t}\mathbf{v} \).

Answer: (a), (c) and (d)

Characteristic polynomial \( \lambda^2 + 3\lambda + 2 = (\lambda + 1)(\lambda + 2) \): eigenvalues \( -1, -2 \).

  • (a) True. Both eigenvalues are negative (a stable node).
  • (b) False. \( \Delta = 2 > 0 \); saddles need \( \Delta < 0 \).
  • (c) True. \( \operatorname{tr}A = -3 \).
  • (d) True. \( \mathbf{v} = (1, -1) \): \( A\mathbf{v} = (-1, 1) = -\mathbf{v} \).

Practice Q7. Consider \( \mathbf{x}' = A\mathbf{x} \) in \( \mathbb{R}^2 \). Which statements are true?

(a) If \( A = \begin{pmatrix} 0 & 1 \\ 0 & 0 \end{pmatrix} \), every solution is bounded on \( [0, \infty) \). (b) If \( \operatorname{tr}A = 0 \) and \( \det A > 0 \), every solution is periodic. (c) If \( \det A < 0 \), there is a non-zero solution tending to \( \mathbf{0} \) as \( t \to \infty \). (d) If every solution is bounded on \( [0, \infty) \), then every solution tends to \( \mathbf{0} \).

Answer: (b) and (c)
  • (a) False. \( x_1 = c_1 + c_2 t \) is unbounded for \( c_2 \ne 0 \): double eigenvalue 0 with a Jordan block.
  • (b) True. Eigenvalues \( \pm i\sqrt{\Delta} \): every solution is a combination of \( \cos(\sqrt{\Delta}\,t) \) and \( \sin(\sqrt{\Delta}\,t) \), periodic with period \( \frac{2\pi}{\sqrt{\Delta}} \).
  • (c) True. A saddle has one negative eigenvalue \( \lambda \); \( e^{\lambda t}\mathbf{v} \) along its eigenvector tends to 0.
  • (d) False. A centre: all solutions bounded, none (except 0) tends to 0. \( A = 0 \) is another counter-example.

Quick Revision Table

IdeaFact
Clairaut \( y = px + f(p) \)general \( y = cx + f(c) \); singular from \( x + f'(p) = 0 \)
Lagrange \( y = xg(p) + f(p) \)linear in \( x(p) \); check roots of \( p = g(p) \)
Discriminantsp-disc \( \supseteq ET^2C \); c-disc \( \supseteq EN^2C^3 \)
Singular solutionin both discriminants and satisfies the ODE
Linear systemunique solution on the whole interval; solution space dimension \( n \)
Liouville\( \det\Phi(t) = \det\Phi(t_0)\exp\int\operatorname{tr}A \)
Constant \( A \)\( \mathbf{x} = e^{(t - t_0)A}\mathbf{x}_0 + \int e^{(t - s)A}\mathbf{g}(s)\,ds \)
Repeated \( \lambda \) (\( 2 \times 2 \))\( e^{tA} = e^{\lambda t}(I + t(A - \lambda I)) \)
Phase plane\( \Delta < 0 \) saddle; \( \tau = 0 < \Delta \) centre; \( \tau^2 < 4\Delta \) spiral; else node
Stabilityall \( \operatorname{Re}\lambda < 0 \) \( \iff \) all solutions \( \to 0 \)

Key Takeaways

What’s Next