Initial value problems (IVP) vs boundary value problems (BVP): Question bank

Before the traps, let's nail down the machinery every question below leans on.

Definition The words and symbols this page uses

Side conditions = the extra data (values or slopes) you attach to a differential equation to pick one solution out of the whole family. When they all sit at one point we call them initial conditions; when they are split across two ends we call them boundary conditions. "Side conditions" is just the umbrella word covering both.
Independent variable = the input the solution depends on. We write it as $x$ throughout, but $t$ (time) means exactly the same thing — $x$ and $t$ are interchangeable placeholders. Space problems tend to use $x$ , time problems $t$ ; the maths is identical.
$y_{1} (x), y_{2} (x)$ = two independent basis solutions of the homogeneous equation. Every solution of a 2nd-order linear ODE is a mix of them.
$c_{1}, c_{2}$ = the two arbitrary constants (the "mixing amounts") that a 2nd-order general solution carries. Side conditions exist precisely to fix these two numbers.
$a, b$ = the two boundary points (the left and right ends of the interval), with $a \neq = b$ .
$α, β$ = the two boundary-value targets: $α$ is the demanded value $y (a) = α$ at the left end, $β$ is the demanded value $y (b) = β$ at the right end. They are the numbers the curve must hit.

Formula Building the linear system step by step

Step 1 — start from the general solution. A 2nd-order linear ODE has $y (x) = c_{1} y_{1} (x) + c_{2} y_{2} (x) + y_{p} (x),$ where $y_{p}$ is a particular solution (take $y_{p} = 0$ for homogeneous problems). The unknowns we must fix are $c_{1}, c_{2}$ .

Step 2 — impose the left condition $y (a) = α$ . Substitute $x = a$ : $c_{1} y_{1} (a) + c_{2} y_{2} (a) + y_{p} (a) = α ⟹ y_{1} (a) c_{1} + y_{2} (a) c_{2} = α - y_{p} (a) .$ This is one linear equation in $c_{1}, c_{2}$ — notice $c_{1}, c_{2}$ are the unknowns and the $y_{i} (a)$ are just numbers.

Step 3 — impose the right condition $y (b) = β$ . Substitute $x = b$ : $y_{1} (b) c_{1} + y_{2} (b) c_{2} = β - y_{p} (b) .$

Step 4 — stack the two equations as rows of a matrix. Reading off coefficients: $M (y_{1} (a) y_{1} (b) y_{2} (a) y_{2} (b)) c (c_{1} c_{2}) = b (α - y_{p} (a) β - y_{p} (b)) .$ Each row of $M$ came from one boundary point; each column multiplies one constant. That is why a BVP's fate lives in a $2 \times 2$ matrix.

Step 5 — read off the verdict with linear algebra: $det M \neq = 0 \Rightarrow$ unique $c$ ; $det M = 0 \Rightarrow$ no solution or infinitely many, depending on whether $b$ is compatible.

The IVP cousin. For an IVP both conditions are at one point $x_{0}$ : impose $y (x_{0}) = A$ and $y^{'} (x_{0}) = B$ . The two rows now use $y$ and $y^{'}$ at the same point, so the coefficient matrix has determinant $W = y_{1} (x_{0}) y_{2}^{'} (x_{0}) - y_{2} (x_{0}) y_{1}^{'} (x_{0}),$ the Wronskian — never zero for independent $y_{1}, y_{2}$ , so the IVP system is always uniquely solvable.

Now hold three anchor pictures in your head — literally sketch them once before you start:

IVP picture — one dot on the axis with an arrow leaving it: all data at one spot, march forward. (Draw a single point, one outgoing arrow.)
BVP picture — two dots at the ends of an interval and a curve stretched to touch both: data at the edges, fill the middle. (Draw two points, a curve bridging them.)
The matrix $M$ — the $2 \times 2$ coefficient grid from Step 4 whose determinant decides a BVP's fate; its IVP cousin is the Wronskian $W$ , evaluated at the single initial point. (Draw the grid, one row per boundary point.)

True or false — justify

Every "answer" here is the justification. Say true/false first, then check the reason.

A 2nd-order ODE with exactly 2 side conditions is always uniquely solvable.

False. Counting conditions is not enough — if the two conditions sit at different points the matrix

M

can be singular (

det M = 0

), giving no solution or infinitely many. Uniqueness is guaranteed only for IVPs, where the Wronskian is nonzero.

y (a) = α

and

y^{'} (a) = β

together form a boundary value problem.

False. Both conditions live at the same point

a

, so this is an IVP regardless of one being a derivative. "Boundary" refers to distinct locations, not to using

y

y^{'}

"Initial" conditions must be specified at

t = 0

False. "Initial" means all conditions at one common point

x_{0}

— that point can be anything (

x_{0} = 3

t = 7

, etc.). The defining feature is one location, not the number zero.

An IVP for a well-behaved ODE can fail to have a solution.

False (essentially). Under mild continuity/Lipschitz conditions, Picard–Lindelöf guarantees a unique local solution — the nonzero Wronskian at the start point pins the constants

c_{1}, c_{2}

down uniquely.

A BVP always has a solution because you gave it enough data.

False. Data at the edges can be self-contradictory (e.g.

y^{''} + y = 0

y (0) = 0

y (π) = 1

forces

0 = 1

). Enough count of conditions does not guarantee compatibility of the system

M c = b

det M = 0

for a BVP, then it definitely has infinitely many solutions.

False.

det M = 0

means either no solution or infinitely many — which one depends on the right-hand side

b

(the target values

α, β

). Singular matrix ⇒ non-unique, but non-unique includes the empty case.

det M \neq = 0

, the BVP has exactly one solution.

True. A nonsingular

2 \times 2

system

M c = b

has a unique

(c_{1}, c_{2})

, hence one curve satisfying both boundary conditions.

The Wronskian of two independent solutions can be zero at the initial point.

False. For genuinely linearly independent solutions the Wronskian is nonzero everywhere on the interval (Abel's theorem: it's either always zero or never zero). That non-vanishing is exactly why the IVP system for

c_{1}, c_{2}

is always solvable.

For an IVP, the relevant matrix is

M

evaluated at two points.

False. For an IVP the data is at one point, so the discriminating object is the Wronskian

y_{1} y_{2}^{'} - y_{2} y_{1}^{'}

at that single point

x_{0}

— a matrix whose rows use

y

and

y^{'}

, not two locations.

Spot the error

Each line states a flawed claim; the reveal names the mistake and repairs it. As you read, keep the matrix $M$ picture (one row per point) in view — most errors here are a row secretly collapsing.

y^{''} + y = 0

y (0) = 1

y (π) = 1

is an IVP with a unique answer."

Error: it's a BVP (two distinct points

0

and

π

). With basis

y_{1} = cos x, y_{2} = sin x

det M = cos 0 sin π - sin 0 cos π = 0

, so it is singular — not a comfortably-unique IVP at all.

"Since

y^{''} + y = 0

y (0) = 0

y (π) = 0

forces

c_{1} = 0

, the only solution is

y \equiv 0

Error: with

y = c_{1} cos x + c_{2} sin x

, the second condition

y (π) = - c_{1} = 0

repeats the information

c_{1} = 0

and says nothing about

c_{2}

. So

y = c_{2} sin x

works for every

c_{2}

— infinitely many solutions, not just the trivial one.

"An eigenvalue problem is different from a BVP."

Error: an eigenvalue problem is a BVP where

det M (λ) = 0

is enforced to get nontrivial solutions. The

λ

making the matrix singular are the eigenvalues — this is exactly Sturm-Liouville Theory.

"To solve a BVP I march forward from the left boundary like an IVP."

Error: marching only uses left-edge data, but you don't know

y^{'} (a)

for a BVP. You must solve globally so the curve also lands on the right-edge target — a two-point matching, not a forward roll.

det M = 0

came from bad arithmetic; recompute and it'll be nonzero."

Error: for special interval lengths (like

L = π

with

y^{''} + y = 0

)

det M = 0

is structurally exact, not a slip. It marks a resonance/eigenvalue length — no recomputation removes it.

"For

y^{''} = 0

[0, L]

with both ends fixed, uniqueness might fail."

Error: with basis

y_{1} = 1, y_{2} = x

det M = y_{1} (0) y_{2} (L) - y_{2} (0) y_{1} (L) = 1 \cdot L - 0 \cdot 1 = L \neq = 0

for

L > 0

, so the linear profile is always unique — steady heat conduction has one answer. (If you order the rows the other way you get

- L

; only the sign flips, the nonzero conclusion is identical.)

Why questions

Why can a BVP have no solution while a same-order IVP always does?

An IVP evolves forward from a fully-known state (position + velocity), so nature determines the rest — the Wronskian is nonzero. A BVP demands the curve hit a distant target, which may be geometrically impossible, so the edge matrix

M

can be singular and the system

M c = b

incompatible.

Why does

det M = 0

signal eigenvalue behaviour?

A singular

M

means the homogeneous BVP has a nonzero solution — a nontrivial curve satisfying both zero boundary conditions. Those special parameters (interval length, or

λ

) are precisely eigenvalues, feeding Sturm-Liouville Theory and separation of variables.

Why is the Wronskian the "IVP version" of

det M

Both are the determinant of the coefficient matrix that fixes

c_{1}, c_{2}

M c = b

. For an IVP the two rows come from

y

and

y^{'}

at one point (that's the Wronskian); for a BVP the two rows come from

y

at two points (that's

M

). Same job, different data layout.

Why does spatial separation of the conditions create the trouble, not their number?

When both conditions are at one point, the rows of the coefficient matrix are "independent by construction" (value vs slope). When split across points, the two rows can accidentally align (e.g.

sin

vanishing at both

0

and

π

), collapsing the determinant.

Why do the Heat Equation and Wave Equation naturally produce BVPs in space but IVPs in time?

Space has physical edges (rod ends, string ends) where you fix values — boundaries. Time has a starting instant where you know the state and let it evolve — an initial point. So PDEs are typically initial-boundary value problems.

Why does checking "how many conditions" mislead students about solvability?

Because for IVPs the count is sufficient, students overgeneralise. For BVPs the count only guarantees you can write the system

M c = b

; whether it's solvable and unique depends entirely on

det M

and the target vector

b

Edge cases

What happens to a BVP as the two boundary points merge,

b \to a

The two rows of

M

become identical (same point), so

det M \to 0

— the problem degenerates. In the limit you'd need

y

and

y^{'}

a

instead, i.e. it collapses toward an IVP structure.

A "BVP" where you accidentally place both conditions at the same point — what is it really?

It is really an IVP (or an over/under-determined single-point problem). With genuinely distinct data at one point it behaves like an IVP; with the same condition twice it's underdetermined.

y^{''} = 0

[0, L]

with

y (0) = T_{0}

y (L) = T_{1}

: does anything special happen as

L \to 0

det M = L \to 0

, so uniqueness is lost in the limit — you'd be fixing two temperatures at essentially the same spot, which is contradictory unless

T_{0} = T_{1}

. Nonzero

L

keeps it well-posed.

For

y^{''} + 4 y = 0

y (0) = 0

y (L) = 0

: which

L

give nontrivial solutions and what happens between them?

Writing

y = c_{1} cos 2 x + c_{2} sin 2 x

y (0) = c_{1} = 0

leaves

y = c_{2} sin 2 x

; nontrivial solutions exist exactly when

sin 2 L = 0

, i.e.

2 L = nπ

, so

L = \frac{nπ}{2}

. For any other

L

det M \neq = 0

and the only solution is the trivial

y \equiv 0

A BVP with a nonzero forcing term

f (x)

but a singular

M

— is it hopeless?

Not automatically. It's solvable iff the target vector

b

is compatible with the singular system (a solvability/orthogonality condition, related to Green's Functions); otherwise no solution exists. Singular

M

just removes the guarantee.

The trivial solution

y \equiv 0

: for which problem types is it always available?

For any homogeneous linear BVP with zero boundary data (like

y (0) = y (L) = 0

y \equiv 0

always satisfies it. The interesting question is whether a nonzero solution also exists — that's when

det M = 0

Zero-order "equation" (no derivatives), just

y = g (x)

with a boundary condition — IVP or BVP?

Neither in the usual sense: there are no arbitrary constants

c_{1}, c_{2}

to fix, so side conditions are redundant or contradictory. The IVP/BVP distinction only bites when the general solution carries arbitrary constants.

Recall One-sentence summary of every trap above

The verdict (IVP vs BVP) is set by where the side conditions live; solvability is then set by the determinant of the coefficient matrix in $M c = b$ — the always-nonzero Wronskian for IVPs, the possibly-singular $M$ for BVPs — and singularity means non-uniqueness, which splits into "no solution" or "infinitely many" depending on the target $b$ .

Connections

Parent: IVP vs BVP
Wronskian and Linear Independence — the IVP determinant.
Sturm-Liouville Theory — where $det M = 0$ becomes eigenvalues.
Picard-Lindelöf Existence and Uniqueness Theorem — why IVPs are reliably unique.
Green's Functions — solvability of forced singular BVPs.