The Wronskian

What the Wronskian is

Given two differentiable functions y₁(x), y₂(x), build the 2×2 matrix

M(x) = [[y₁(x), y₂(x)], [y₁'(x), y₂'(x)]]

and take its determinant. The result is a scalar function of x:

W(y₁, y₂)(x) = y₁(x) y₂'(x) − y₂(x) y₁'(x).

That single number at each x carries a remarkable amount of information. If y₁ and y₂ are solutions of a linear homogeneous ODE with continuous coefficients on an interval I, then W either vanishes everywhere on I or vanishes nowhere on I — and the second alternative is exactly the condition that {y₁, y₂} forms a fundamental solution set, a basis for the two-dimensional space of solutions.

The construction generalises: for n functions you take the n×n determinant whose i-th row is the (i−1)-th derivative of all n functions. For n-th order linear ODEs with continuous coefficients the same dichotomy holds.

Why a determinant detects independence

Suppose y₁ and y₂ are linearly dependent — say c₁ y₁ + c₂ y₂ = 0 for some non-zero constants. Differentiating gives c₁ y₁' + c₂ y₂' = 0. Stacked, this is the linear system

M(x) · (c₁, c₂)ᵀ = 0.

A non-trivial (c₁, c₂) exists at every x iff det M(x) = W(x) = 0. So linear dependence forces W ≡ 0. The contrapositive: if W(x₀) ≠ 0 at even one point, the functions are linearly independent.

For arbitrary smooth functions the converse — W ≡ 0 ⇒ dependence — can fail (see common pitfalls below). What rescues the test is the extra constraint imposed by being solutions of a linear ODE: the Picard-Lindelöf uniqueness theorem forces the Wronskian to be either identically zero or never zero, removing the pathological case.

Comparison with related independence tests

Test	Object	Detects	Bidirectional?	When to use
Wronskian W(y₁, y₂)	two functions	linear independence	Yes for ODE solutions; no in general	solutions of a linear ODE
Determinant det[v₁ | v₂]	two vectors in ℝⁿ	linear independence	Yes (vectors are always "rigid")	finite-dimensional linear algebra
Gram determinant det⟨vᵢ, vⱼ⟩	vectors in inner-product space	linear independence	Yes	function spaces, L²
Cauchy-Schwarz equality	two vectors	parallelism (proportionality)	Yes	signal processing, statistics
Rank of derivative matrix	functions and higher derivatives	linear independence	Yes if non-degenerate	generalising to n functions
Casoratian (discrete analogue)	sequences satisfying a recurrence	linear independence	Yes for recurrence solutions	linear difference equations

Worked example: e^x and e^{2x}

Take y₁(x) = e^x, y₂(x) = e^{2x}. Both solve y'' − 3y' + 2y = 0 (characteristic roots r = 1, 2). Build the matrix:

M(x) = [[e^x, e^{2x}], [e^x, 2e^{2x}]]

and compute the determinant:

W(x) = e^x · 2e^{2x} − e^{2x} · e^x = 2e^{3x} − e^{3x} = e^{3x}.

For every real x, e^{3x} > 0. So W is never zero — the pair is a fundamental solution set, and the general solution is y = c₁ e^x + c₂ e^{2x}.

Cross-check with Abel's identity. The ODE in standard form is y'' + p(x) y' + q(x) y = 0 with p = −3. Abel says W' = −p · W = 3W, so W(x) = W(0) e^{3x}. Evaluating at x = 0: W(0) = 1 · 2 − 1 · 1 = 1. So W(x) = e^{3x} — matching directly.

A second worked example: sin x and cos x

Now take y₁ = sin x, y₂ = cos x, both solutions of y'' + y = 0. The matrix is

[[sin x, cos x], [cos x, −sin x]]

and the determinant is W(x) = sin x · (−sin x) − cos x · cos x = −sin²x − cos²x = −1. A constant — non-zero everywhere — so sin and cos are linearly independent, and the general solution to y'' + y = 0 is c₁ sin x + c₂ cos x. Abel: p = 0 in y'' + y = 0, so W' = 0, hence W is constant. Both routes agree.

Abel's identity: how W evolves

For a second-order linear ODE in standard form y'' + p(x) y' + q(x) y = 0, the Wronskian of any two solutions satisfies the first-order linear ODE

W'(x) = −p(x) · W(x),

which integrates to W(x) = W(x₀) · exp(−∫_{x₀}^{x} p(t) dt). The exponential factor is always strictly positive, so the sign of W is constant: zero at one point ⇔ zero everywhere; non-zero at one point ⇔ non-zero everywhere. This is exactly why testing the Wronskian at a single convenient x₀ is enough.

The proof is a direct calculation. Differentiate W = y₁ y₂' − y₂ y₁':

W' = y₁' y₂' + y₁ y₂'' − y₂' y₁' − y₂ y₁'' = y₁ y₂'' − y₂ y₁''.

Substitute y₁'' = −p y₁' − q y₁ and y₂'' = −p y₂' − q y₂:

W' = y₁(−p y₂' − q y₂) − y₂(−p y₁' − q y₁) = −p (y₁ y₂' − y₂ y₁') = −p W.

Done — and notice how the q terms cancel exactly. Only the first-derivative coefficient p(x) appears in Abel's formula.

Reduction of order using the Wronskian

Suppose you know one solution y₁(x) of y'' + p(x) y' + q(x) y = 0 and want a second linearly independent one. Set W = y₁ y₂' − y₂ y₁'; Abel gives W(x) = exp(−∫ p dt) up to a constant. Then y₂'/y₁ − y₂ y₁'/y₁² is the derivative of y₂/y₁, so

(y₂ / y₁)' = W / y₁²,

and integrating produces y₂ = y₁ · ∫ exp(−∫ p) / y₁² dx. This is the standard reduction-of-order formula — Abel's identity in disguise. For y'' − 2y' + y = 0 with known y₁ = e^x, the formula gives y₂ = e^x · ∫ e^{2x}/e^{2x} dx = x e^x. The expected "exponential times x" for the repeated-root case appears automatically.

Variation of parameters

For a non-homogeneous equation a(x) y'' + b(x) y' + c(x) y = f(x) with homogeneous solutions y₁, y₂, write a particular solution as y_p = u₁(x) y₁ + u₂(x) y₂. Imposing u₁' y₁ + u₂' y₂ = 0 (the standard side condition that keeps the algebra tractable) and substituting into the ODE yields the linear system

u₁' y₁ + u₂' y₂ = 0; u₁' y₁' + u₂' y₂' = f / a.

Solving by Cramer's rule gives u₁' = −y₂ f / (a W) and u₂' = y₁ f / (a W). The Wronskian sits in the denominator. Its non-vanishing on the interval is what guarantees that integrating gives well-defined u₁ and u₂; the entire variation-of-parameters technique rests on Abel's "zero everywhere or nowhere" dichotomy.

Sturm-Liouville theory and oscillation

For the Sturm-Liouville eigenvalue problem (p y')' + (λ w − q) y = 0, the Wronskian of an eigenfunction at consecutive zeros relates eigenvalues. Sturm's separation theorem — between any two consecutive zeros of one eigenfunction y_n there is exactly one zero of any other eigenfunction y_m with m > n — is proved by tracking the sign of the Wronskian and using its non-vanishing at non-coincident zeros. Likewise the Sturm comparison theorem (eigenvalue monotonicity in q) is a Wronskian sign argument applied to Lagrange's identity.

Where the Wronskian shows up

• ODE solution-space bases. Certifying that c₁ y₁ + c₂ y₂ is the general solution of a second-order homogeneous linear ODE reduces to W ≠ 0 at one point.
• Variation of parameters. The formula for particular solutions of non-homogeneous linear ODEs has W in the denominator.
• Quantum scattering. Transmission and reflection coefficients in 1D scattering off a potential V(x) are expressed as ratios of Wronskians of Jost solutions of the time-independent Schrödinger equation.
• Special-function identities. Wronskians of Bessel, Legendre, Hermite, and Airy functions are constants (or simple closed forms), making them workhorses of asymptotic analysis and connection-formula computations.
• Phase-amplitude methods. The Prüfer transformation and WKB approximation both manipulate the Wronskian (or its dependence on parameters) to recover oscillation counts and bound-state energies.
• Control theory and observability. The Wronskian of basis solutions of a linear system shows up in the observability Gramian; W ≠ 0 over a window ⇔ the state is reconstructible from measurements.

Common pitfalls

• Concluding dependence from W ≡ 0 without the ODE hypothesis. The standard counterexample is y₁ = x² and y₂ = x|x| (or x³ and |x|³). Both are C¹ on ℝ; W(x) = 0 for every x. Yet no constants c₁, c₂ (not both zero) satisfy c₁ x² + c₂ x|x| = 0 for every x — they would have to satisfy c₁ + c₂ = 0 on (0, ∞) and c₁ − c₂ = 0 on (−∞, 0), forcing both to zero. Linear independence holds even though the Wronskian vanishes. The fix: the bidirectional theorem requires y₁, y₂ to be solutions of a single linear ODE with continuous coefficients on the same interval. For arbitrary smooth functions, W ≡ 0 only ever implies dependence under additional regularity hypotheses.
• Forgetting to put the ODE in standard form before applying Abel. Abel's identity W' = −p W uses the coefficient of y' after dividing through by the coefficient of y''. If your equation is x y'' + y' + y = 0, divide first to get y'' + (1/x) y' + (1/x) y = 0; here p = 1/x and W(x) = W(x₀) / (x/x₀). Forgetting to divide by the leading coefficient gives a wrong sign or wrong magnitude.
• Singular points break the theorem. Abel requires p(x) continuous on the interval. At regular singular points (where p has a pole) the Wronskian can cross zero or blow up. For Bessel's equation at x = 0, the irregular behaviour of one solution (Y_0) at x = 0 is exactly why W(J₀, Y₀) = 2/(πx) — non-zero on (0, ∞) but singular at the endpoint.
• Confusing W with the matrix. The Wronskian is the determinant — a scalar function. The matrix [[y₁, y₂], [y₁', y₂']] is sometimes loosely called the "Wronskian matrix" but the test is about its determinant being non-zero.
• Mistaking sign of W for something physical. The sign of W depends on the order of the functions — swapping y₁ and y₂ flips its sign. Only the non-vanishing matters for linear independence; the sign carries no further geometric content.
• Generalising to nonlinear ODEs. The Wronskian test is specifically about linear ODEs. For nonlinear equations the solution space is not a vector space, so "linear independence" is the wrong question; the right concepts are uniqueness, stability, and continuous dependence.