The Wronskian and Linear Independence

Suppose you find two solutions to a second-order linear differential equation. How do you know whether you have found enough to build every possible solution? This question sits at the heart of solving Second-Order Linear Equations. Its answer is a single, computable determinant called the Wronskian. The idea matters for a key reason. Linear differential equations of order $n$ have solution spaces of dimension exactly $n$ . But "dimension $n$ " is an abstract fact. It stays abstract until you can certify that $n$ particular solutions actually span that space. The Wronskian converts an infinite-dimensional-looking question into a single algebraic computation. That question is: are these functions independent at every point? Abel's identity reveals something striking. This computation has a strikingly simple closed form. It is governed only by the coefficient of the equation. This article develops the Wronskian rigorously and proves Abel's identity. It then shows exactly how the two combine to certify a fundamental set of solutions. This sets up the machinery used in Homogeneous Equations with Constant Coefficients and Variation of Parameters.

Key takeaways

The Wronskian $W[y_1,y_2](t) = y_1 y_2' - y_2 y_1'$ is a determinant testing linear independence of solutions.
For solutions of a linear ODE, $W$ is either identically zero or never zero on the interval — no in-between.
Abel's identity gives $W(t)$ explicitly: $W(t) = W(t_0)\,e^{-\int_{t_0}^t p(s)\,ds}$ , no need to solve the ODE.
$W(t_0)\neq 0$ at even one point certifies $\{y_1,y_2\}$ as a fundamental set spanning all solutions.
Linear independence of arbitrary functions does not require $W\neq0$ ; the converse needs the ODE structure.
The Wronskian generalizes to order $n$ via an $n\times n$ determinant of a solution matrix.
Key idea: for solutions of a linear ODE, one number at one point decides independence everywhere.

Where this lives today

The Wronskian is the standard certificate of a fundamental set of solutions in Fundamental Sets and General Solutions, letting one confirm in one line that a candidate general solution $c_1 y_1 + c_2 y_2$ really captures every solution. It reappears explicitly as the denominator in the Variation of Parameters formula for particular solutions of nonhomogeneous linear equations, where a nonzero Wronskian is exactly what makes the method's integrals well-defined. In Homogeneous Equations with Constant Coefficients, the Wronskian of the exponential (or exponential-times-polynomial, or exponential-times-trig) solutions built from the characteristic roots is computed once and for all, giving a clean algebraic proof that the standard solution basis is genuinely a basis. Beyond ODE theory, the Wronskian and its generalizations underlie tests for algebraic independence of special functions (Bessel, Legendre, hypergeometric functions), appear in Sturm–Liouville and spectral theory to control node/oscillation counts of eigenfunctions, and show up in the theory of integrable systems and soliton solutions, where "Wronskian solutions" of the KdV and related equations are built directly from determinants of this type. In control theory and signal processing, Wronskian-type conditions test whether a family of exponential or sinusoidal modes are independent, which matters for observability and identifiability of linear systems.

:::quiz[If $y_1, y_2$ solve $y'' + p(t)y' + q(t)y=0$ on an interval $I$ with $p,q$ continuous, and $W[y_1,y_2](2)=0$ , what can you conclude about $W$ at $t=5\in I$ ?]

$W(5)$ could be zero or nonzero, need more information
$W(5) = 0$ necessarily ✓
$W(5) \neq 0$ necessarily
The question doesn't make sense without knowing $y_1,y_2$ explicitly

:::quiz[Two solutions $y_1,y_2$ of a second-order linear homogeneous ODE with continuous coefficients satisfy $W[y_1,y_2](t_0)\neq0$ for some $t_0$ . What does the theorem let you conclude?]

Only that $y_1,y_2$ are independent at $t_0$ , nothing about other points
$y_1, y_2$ are linearly independent on $I$ and every solution is $c_1y_1+c_2y_2$ ✓
$y_1$ and $y_2$ must both be exponential functions
$W$ must be constant for all $t$

Ask the tutor: If two solutions of a linear ODE have Wronskian zero at $t=0$ but I only know $p(t)$ blows up at $t=0$ , what can I still conclude on each side of $t=0$ ?

y_1, y_2

Intuition

Think of $(y_1(t), y_1'(t))$ and $(y_2(t), y_2'(t))$ as two vectors in $\mathbb{R}^2$ , one for each solution, at a fixed time $t$ . The Wronskian is exactly the signed area of the parallelogram these two vectors span. If the vectors are parallel — meaning $y_2$ 's "state" is a scalar multiple of $y_1$ 's state — the parallelogram collapses and $W=0$ . If they point in genuinely different directions, the area is nonzero. For solutions of a linear ODE, this state vector $(y(t), y'(t))$ is precisely the point that a two-dimensional first-order system tracks through time, so the Wronskian is measuring whether two solution trajectories ever coincide in direction. The surprising fact — and the reason this article exists — is that for solutions of a linear equation, this area cannot shrink to zero at one instant and be positive at another. Abel's identity says the area evolves by a pure exponential factor, $W(t) = W(t_0) e^{-\int p}$ , an exponential that is never zero. So the parallelogram either starts and stays flat (identically zero Wronskian, dependent solutions) or starts and stays non-degenerate (nonzero Wronskian everywhere, independent solutions). This is the qualitative content behind the theorem below.

Figure

Two solution curves and their Wronskian as parallelogram area

p, q

p, q

Worked examples

Example 1 (using Abel's identity without solving the ODE). Consider

t^2 y'' - t(t+2) y' + (t+2) y = 0, \qquad t > 0,

which one can check has $y_1(t) = t$ as a solution: $y_1'' = 0$ , $y_1' = 1$ , and substituting gives $-t(t+2) + (t+2)t = 0$ . ✓. We are told a second solution $y_2$ exists with $y_2(1) = 2, y_2'(1) = 5$ (consistent initial data for some solution, guaranteed to exist by the existence theorem). We want $W[y_1,y_2](t)$ for all $t>0$ , without finding $y_2$ explicitly. Put the equation in standard form by dividing by $t^2$ :

y'' - \frac{t+2}{t}\, y' + \frac{t+2}{t^2}\, y = 0, \quad\text{so } p(t) = -\frac{t+2}{t} = -1 - \frac{2}{t}.

By Abel's identity with basepoint $t_0 = 1$ ,

W(t) = W(1)\exp\left(-\int_1^t p(s)\,ds\right) = W(1)\exp\left(\int_1^t \left(1 + \frac{2}{s}\right) ds\right).

Compute the integral: $\int_1^t (1 + 2/s)\,ds = (t - 1) + 2\ln t$ . So

W(t) = W(1)\, e^{t-1}\, t^2.

Now compute $W(1)$ directly from the given data: $y_1(1) = 1$ , $y_1'(1) = 1$ , $y_2(1) = 2$ , $y_2'(1) = 5$ , so

W(1) = y_1(1) y_2'(1) - y_2(1) y_1'(1) = (1)(5) - (2)(1) = 3.

Therefore

W(t) = 3 t^2 e^{t - 1}, \qquad t > 0.

Since $3t^2 e^{t-1} > 0$ for every $t > 0$ , the dichotomy theorem tells us $W(t) \neq 0$ for all $t$ in the interval — automatically, without ever solving for $y_2$ — so $\{y_1, y_2\}$ is guaranteed to be a fundamental set on $(0,\infty)$ , and every solution of the ODE on $(0,\infty)$ is $c_1 t + c_2 y_2(t)$ .

Example 2 (checking independence directly, then cross-checking with Abel). Consider the equation

y'' - 3y' + 2y = 0,

with candidate solutions $y_1(t) = e^{t}$ and $y_2(t) = e^{2t}$ . Direct verification: $y_1'' - 3y_1' + 2y_1 = e^t - 3e^t + 2e^t = 0$ ✓, and $y_2'' - 3y_2' + 2y_2 = 4e^{2t} - 6e^{2t} + 2e^{2t} = 0$ ✓. Compute the Wronskian directly:

W(t) = y_1 y_2' - y_2 y_1' = e^t(2e^{2t}) - e^{2t}(e^t) = 2e^{3t} - e^{3t} = e^{3t}.

This is never zero, so $\{y_1, y_2\}$ is a fundamental set on $\mathbb{R}$ , and the general solution is $y = c_1 e^t + c_2 e^{2t}$ . Cross-check with Abel's identity: here $p(t) = -3$ (constant), so with $t_0 = 0$ ,

W(t) = W(0) e^{-\int_0^t (-3)\,ds} = W(0) e^{3t}.

Compute $W(0) = y_1(0) y_2'(0) - y_2(0) y_1'(0) = (1)(2) - (1)(1) = 1$ . So Abel's identity predicts $W(t) = e^{3t}$ , exactly matching the direct computation. This numerically confirms both the derivative-based derivation of $W' = -pW$ and the correctness of the direct computation — a useful consistency check whenever both routes are available.

y_1, y_2

Scope, edge cases & common traps

Situation	What is true	Common mistake
$y_1, y_2$ solve the same linear homogeneous ODE with continuous $p,q$	$W\equiv 0$ or $W$ never $0$ on $I$	Assuming this for unrelated functions (see counterexample)
Nonhomogeneous or nonlinear equations	Wronskian dichotomy does not apply	Using $W\neq0$ as an independence test for solutions of a nonlinear ODE
Coefficients $p,q$ have a discontinuity in $I$	Theorem applies separately on each subinterval of continuity	Applying Abel's identity across a singular point of $p$
Arbitrary (non-solution) smooth functions	$W\equiv0$ does not imply dependence in general	Concluding dependence from $W\equiv 0$ without ODE structure
Need only one point	$W(t_0)\neq0$ at a single $t_0$ certifies independence on all of $I$	Trying to check $W\neq0$ at every $t$ by hand

The mathematician behind it

Origins · 1 minJózef Hoene-Wroński and how this was discovered

Józef Hoene-Wroński

1776–1853 · Polish mathematician and philosopher, active largely in France

Wroński was a strikingly unconventional figure: trained first as a soldier and later drawn to mathematics and metaphysics, he spent much of his career in Paris pursuing an ambitious, often mystical, program to found all of mathematics on what he called the "Universal Law of Creation." Amid a body of work that mixed genuine mathematical insight with grandiose philosophical claims, he introduced the determinant that now bears his name in the early 1810s, in the context of studying series expansions and the algebraic dependence of families of functions. The Wronskian's now-standard role — certifying linear independence of solutions to linear differential equations via Abel's identity — was clarified and put on firm analytic footing over the following decades by mathematicians developing the theory of linear ODEs, most notably through the work of Niels Henrik Abel, whose identity gives the determinant its explicit exponential form.

History

- 1802: Niels Henrik Abel born in Norway; his later short life (he died in 1829) still produced the identity now bearing his name.

- 1812: Józef Hoene-Wroński introduces the determinant later named the Wronskian, in work on function series.

- 1827: Abel's identity for the Wronskian of solutions to a linear ODE appears in Abel's work on differential equations.

- 1880s–1900s: The Wronskian becomes a standard tool in the emerging rigorous theory of linear ODEs and Sturm–Liouville theory.

- 20th century: The Wronskian criterion becomes a textbook staple for certifying fundamental sets of solutions in undergraduate ODE courses.

Does the Wronskian being nonzero at just one point really guarantee independence everywhere?

Yes — but only for solutions of a common linear homogeneous ODE with continuous coefficients on the interval. The proof works in two stages. First, a purely algebraic argument at the single point $t_0$ : if $W(t_0)\neq 0$ , the only constants solving the linear system $c_1y_1(t_0)+c_2y_2(t_0)=0$ , $c_1y_1'(t_0)+c_2y_2'(t_0)=0$ are $c_1=c_2=0$ , which already forces independence as functions (since dependence would need these equations to hold with a nonzero pair). Second, Abel's identity shows $W$ can never revisit zero once it's nonzero at one point, since it's a nonzero constant times a never-vanishing exponential. Both ingredients are essential; dropping the ODE hypothesis breaks the argument, as the counterexample with $t^2$ and $t|t|$ shows.

Why does Abel's identity not require actually solving the second-order equation?

Abel's identity is derived by finding the differential equation the Wronskian itself satisfies, not by solving for $y_1,y_2$ individually. Differentiating $W=y_1y_2'-y_2y_1'$ and substituting $y_1''=-py_1'-qy_1$ , $y_2''=-py_2'-qy_2$ (using only that $y_1,y_2$ solve the original equation), the $q$ -terms cancel exactly, leaving the strikingly simple first-order linear equation $W'=-pW$ . That equation only involves $p$ , never $q$ , and is solved in closed form using an integrating factor. So the entire computation bypasses $y_1, y_2$ 's explicit formulas, needing only $p(t)$ and one initial value $W(t_0)$ .

Can the Wronskian be zero at every point yet the solutions still be independent, for actual ODE solutions?

No — this is precisely what the dichotomy half of the theorem rules out, provided $y_1,y_2$ genuinely solve a common linear homogeneous ODE with continuous coefficients on the interval. If $W\equiv0$ , Step 6 of the proof shows there exist constants $c_1,c_2$ , not both zero, and a point $t_1$ where the linear combination $z=c_1y_1+c_2y_2$ satisfies $z(t_1)=z'(t_1)=0$ ; by uniqueness of solutions to the IVP, $z\equiv0$ on the whole interval, exhibiting an explicit nontrivial dependence relation. So for actual solutions, $W\equiv0$ really does force dependence — the earlier counterexample only breaks this because its functions are not both solutions of one shared linear ODE.

Does the Wronskian test extend beyond second order?

Yes. For $n$ functions $y_1,\dots,y_n$ that are $n-1$ times differentiable, the Wronskian is the $n\times n$ determinant built from the functions and their first $n-1$ derivatives, as in the definition. If the $y_i$ all solve a common linear homogeneous $n$ th-order ODE $y^{(n)}+p_{n-1}(t)y^{(n-1)}+\cdots+p_0(t)y=0$ with continuous coefficients, the identical Abel's identity argument applies (with $p_{n-1}$ playing the role of $p$ ), giving the same dichotomy: $W\equiv0$ or $W$ never zero, and a nonzero value at one point certifies a fundamental set of $n$ solutions spanning the full $n$ -dimensional solution space.

What if the coefficients

p, q

are not continuous on the whole interval, e.g. have a singularity at one point?

The theorem, exactly as stated, requires continuity of $p,q$ on the interval $I$ where it is applied, because that continuity is what guarantees the existence-uniqueness theorem for the underlying IVP, which the whole proof leans on in Steps 5 and 6. If $p$ or $q$ has a singularity at some point $t^*$ , you apply the theorem separately on each maximal subinterval of continuity not containing $t^*$ (e.g. $(0,\infty)$ and $(-\infty,0)$ if the singularity is at $0$ ). Abel's identity itself, $W(t)=W(t_0)\exp(-\int_{t_0}^tp)$ , remains valid only within such a subinterval; crossing a singularity of $p$ can change the qualitative behavior entirely and needs separate analysis (this is the setting of regular singular points and Frobenius theory).

Is a nonzero Wronskian equivalent to linear independence for general (non-solution) functions?

Only one direction survives in general: if the Wronskian of $y_1, y_2$ (viewed as arbitrary sufficiently differentiable functions, not necessarily tied to any ODE) is nonzero at some point, then they are linearly independent — this direction is a purely algebraic fact about the $2\times2$ linear system and needs no ODE hypothesis at all. But the converse fails for general functions: independent functions can have identically vanishing Wronskian, as the $t^2, t|t|$ example demonstrates. The converse only becomes true once you additionally know the functions solve a common linear homogeneous ODE with continuous coefficients, which is exactly the extra structural hypothesis Abel's identity exploits.

How is the Wronskian used practically in solving nonhomogeneous equations?

In Variation of Parameters, given a fundamental set $y_1, y_2$ of the homogeneous equation, a particular solution to $y''+py'+qy=g(t)$ is sought as $Y=u_1y_1+u_2y_2$ with $u_1', u_2'$ solved from a linear system whose determinant is exactly $W[y_1,y_2](t)$ . Because $y_1,y_2$ form a fundamental set, the theorem guarantees $W(t)\neq0$ for every $t$ in the interval, so the system is always solvable and the resulting formulas $u_1'=-y_2g/W$ , $u_2'=y_1g/W$ never involve dividing by zero. This is the direct, load-bearing application of the dichotomy theorem in constructing explicit solution formulas.