Visual walkthrough — Initial value problems (IVP) vs boundary value problems (BVP)

We build everything around the same tiny equation, $y^{\prime \prime }+y=0$, because it is the smallest equation that shows off all three BVP outcomes. By the end you will see why the location of the side-conditions is the whole story.

---

Step 1 — A differential equation is a whole family of curves

WHAT. Before we split hairs about IVP vs BVP, look at what "$y^{\prime \prime }+y=0$" even means. Read it out loud: "the second derivative of $y$ is the negative of $y$." A function whose bend ($y^{\prime \prime }$) always points back toward zero — that is exactly what a wave does. The full family of solutions is

$y(x)=\underset{\text{knob 1}}{\underbrace{c_1}}\underset{\text{building block }{y}_1}{\underbrace{\cos x}}+\underset{\text{knob 2}}{\underbrace{c_2}}\underset{\text{building block }{y}_2}{\underbrace{\sin x}}.$

• $\cos x$ and $\sin x$ are two independent building-block waves — you cannot make one by scaling the other.
• $c_1,c_2$ are two free knobs (numbers we get to choose). Turning them reshapes and re-sizes the wave.

WHY two knobs. The equation is second order (highest derivative is $y^{\prime \prime }$). Each derivative you must "undo" leaves behind one arbitrary constant, so two derivatives → two constants. We will need exactly two facts to lock both knobs.

PICTURE. Each choice of $(c_1,c_2)$ is one curve. The figure shows a bundle of them — a family.

---

Step 2 — IVP: pin one point, and its slope, and the future is forced

WHAT. An IVP hands you everything at one location $x_0$. For our second-order equation that means a value and a slope at the same spot:

$\underset{\text{height here}}{\underbrace{y(x_0)}}=A,\underset{\text{slope here}}{\underbrace{y^{\prime }(x_0)}}=B.$

Take $x_0=0$, $A=1$, $B=0$: start at height $1$, perfectly flat.

• $y(0)=c_1\cos 0+c_2\sin 0=c_1=1$. (Here $\cos 0=1$, $\sin 0=0$ do the selecting.)
• $y^{\prime }(x)=-c_1\sin x+c_2\cos x$, so $y^{\prime }(0)=c_2=0$.

Both knobs fixed → one curve, $y=\cos x$.

WHY it always works. Knowing height and slope at one point is like knowing a car's position and velocity right now — physics has no choice about the next instant. The tool that guarantees this is the Wronskian, and Picard–Lindelöf is the theorem that says "yes, exactly one."

PICTURE. A single dot with a little slope-stick attached; only one family member threads through both.

---

Step 3 — Why "position + slope at one point" can never be ambiguous

WHAT. Feed the two IVP facts into the family and you get a $2\times 2$ system in the knobs. The matrix that decides everything is built from the building blocks and their slopes, all at the one point $x_0$:

$W(x_0)=\underset{\cos x_0}{\underbrace{y_1(x_0)}}\underset{\cos x_0}{\underbrace{y_2^{\prime }(x_0)}}-\underset{\sin x_0}{\underbrace{y_2(x_0)}}\underset{-\sin x_0}{\underbrace{y_1^{\prime }(x_0)}}={\cos }^2{x}_0+{\sin }^2{x}_0=1.$

This number $W$ is the Wronskian. Notice it came out $=1$, never zero, at every point.

WHY this matters. A $2\times 2$ system has a unique answer exactly when its determinant is nonzero. For independent building blocks the Wronskian is never zero — so an IVP is always uniquely solvable. Hold this thought: the BVP will use a different matrix that can hit zero.

PICTURE. Two arrows (the "value info" and the "slope info") pointing in genuinely different directions — they span the whole plane, so any target is reachable.

---

Step 4 — BVP: split the two facts across two different points

WHAT. A BVP gives you information at two distinct places $a\ne b$:

$y(a)=\alpha ,y(b)=\beta .$

No slope is given — instead we demand the curve hit two targets. Plugging into the family produces a system in the knobs whose matrix is now built from the same building blocks evaluated at two different points:

$$\underset{\text{call this }M}{\underbrace{\left(\begin{array}{cc}{y}_1(a)& y_2(a)\\ y_1(b)& y_2(b)\end{array}\right)}}\left(\begin{array}{c}{c}_1\\ c_2\end{array}\right)=\left(\begin{array}{c}\alpha \\ \beta \end{array}\right).$$

• Row 1 says "at $x=a$ the mixed wave must equal $\alpha$."
• Row 2 says "at $x=b$ it must equal $\beta$."

WHY it can fail. Unlike the Wronskian, this matrix $M$ mixes two locations, and there is no guarantee its determinant stays nonzero. The magic number is

$\det M=y_1(a)y_2(b)-y_2(a)y_1(b).$

PICTURE. Two pins standing at $x=a$ and $x=b$; we must lasso a family curve through both.

---

Step 5 — Case A: $\det M\ne 0$ → exactly one curve fits

WHAT. Choose $a=0$, $b=\frac{\pi }{2}$ (a quarter-wave apart), $\alpha =\beta =0$. Compute

$\det M=\cos 0\cdot \sin \frac{\pi }{2}-\sin 0\cdot \cos \frac{\pi }{2}=1\cdot 1-0\cdot 0=1\ne 0.$

Nonzero determinant → the system has one and only one answer: $c_1=0$, $c_2=0$, i.e. the flat line $y=0$. Boring, but unique — that is the point.

WHY. When the two pins land on a rising part of the wave, the family bends enough between them that only one member can touch both. Same reliable behaviour an IVP always enjoys.

PICTURE. Two pins a quarter-wave apart; a single curve passes through, all others miss.

---

Step 6 — Case B: $\det M=0$ with matching target → infinitely many

WHAT. Now put the pins a half-wave apart: $a=0$, $b=\pi$, both targets $0$.

$\det M=\cos 0\cdot \sin \pi -\sin 0\cdot \cos \pi =1\cdot 0-0\cdot (-1)=0.$

Working it through: $y(0)=c_1=0$, and $y(\pi )=-c_1=0$ — which is automatically true and says nothing about $c_2$. So $c_2$ is a free knob:

$y=c_2\sin x\text{for }\text{any }{c}_2.$

WHY. $\sin x$ starts at $0$ and returns to $0$ after exactly half a wave. Every scaled copy of it obeys both boundary demands. This is precisely the eigenvalue situation of Sturm-Liouville Theory: $\det M=0$ is the equation whose special lengths are eigenvalues, and the surviving $\sin x$ is the eigenfunction.

PICTURE. A fan of $\sin$-waves of different heights, all threading both pins.

---

Step 7 — Case C: $\det M=0$ with clashing target → no solution

WHAT. Keep the pins half-a-wave apart but move the second target: $a=0$, $b=\pi$, $\alpha =0$, $\beta =1$.

• $y(0)=c_1=0$.
• $y(\pi )=-c_1=0$, but we demanded it equal $1$. So $0=1$. Impossible.

Same singular $\det M=0$, yet now the family simply cannot reach the target — no curve exists.

WHY. Every family member that is $0$ at $x=0$ is forced back to $0$ at $x=\pi$. Asking it to arrive at $1$ instead is asking a returning wave to break its own rhythm. It can't.

PICTURE. All candidate curves cross zero at $x=\pi$; the amber target dot floats untouched above them.

---

Step 8 — The rule that summarises all three cases

WHAT. Everything above collapses into a single decision:

WHY. The location of the side-conditions is what changes the deciding matrix from the always-safe Wronskian to the sometimes-singular $M$. That single change is the entire IVP-vs-BVP story. This machinery is exactly what Separation of Variables (PDE) feeds into when solving the Heat Equation and Wave Equation, and it is why Green's Functions need $\det M\ne 0$ to exist.

PICTURE. A flowchart of the verdict, in blueprint style.

---

The one-picture summary

One frame, three pins-configurations, one equation $y^{\prime \prime }+y=0$: a quarter-wave gap (unique), a half-wave gap with matching ends (infinite fan), a half-wave gap with a clashing end (empty target).

Recall Feynman retelling — the whole walkthrough in plain words

We started with a wobbling curve family: "bend always points home" gives sines and cosines, with two dials to turn. IVP: stand at one spot, say how high and how steep — the curve has no freedom left, so exactly one wave fits, always. The reason is a number called the Wronskian that is built at that single spot and is stubbornly never zero. BVP: instead of steepness, you plant two flags at two different spots and demand the wave touch both. Now the deciding number is a different one, $\det M$, made from the same wave-blocks but read at two places. If the flags sit a quarter-wave apart, the wave bends enough that only one fits — clean. But plant them a half-wave apart and the sine returns to zero all by itself: if both flags are at zero, every height of sine works (infinitely many); if one flag is lifted off zero, no sine can reach it (none). Same equation, same count of facts — only the placement changed, and that is the entire difference between IVP and BVP.

Connections

• Wronskian and Linear Independence — the never-zero number behind IVP uniqueness (Steps 3, 8).
• Sturm-Liouville Theory — where $\det M=0$ becomes eigenvalues (Step 6).
• Picard-Lindelöf Existence and Uniqueness Theorem — the IVP guarantee (Step 2).
• Separation of Variables (PDE), Heat Equation, Wave Equation, Green's Functions — where these BVP verdicts get used.