Direction fields and Euler's method — visual - numerical intuition first

1. What is a direction field?

WHY it works: The ODE is a constraint "your slope here must equal $f(x,y)$ ." The segment just visualises that constraint. A solution is any curve that obeys the constraint everywhere.

2. Euler's method — derive it from scratch

We can't always solve the ODE with formulas. But we can follow the arrows numerically.

Derivation (first principles — local linear approximation):

The derivative is the slope of the tangent line. Near $x_n$ , the true solution is well-approximated by its tangent: $y(x_n+h) \approx y(x_n) + h\cdot y'(x_n).$

This is just the first-order Taylor expansion: $y(x_n+h)=y(x_n)+h\,y'(x_n)+\tfrac{h^2}{2}y''(\xi).$

But the ODE gives us $y'(x_n)=f(x_n,y_n)$ . Substitute and drop the $h^2$ remainder:

3. Worked example: walk it by hand

Worked example Solve

$\dfrac{dy}{dx}=x+y$ , $y(0)=1$ , with $h=0.5$ , find $y(1)$ . Here $f(x,y)=x+y$ , $x_0=0,\ y_0=1$ .

Step 1 ( $n=0$ ): slope $=f(0,1)=0+1=1$ . $y_1=y_0+h f = 1 + 0.5(1)=1.5,\qquad x_1=0.5$ Why this step? We move right by $h=0.5$ along slope $1$ , so rise $=0.5\cdot1=0.5$ .

Step 2 ( $n=1$ ): slope $=f(0.5,1.5)=0.5+1.5=2$ . $y_2 = 1.5 + 0.5(2)=2.5,\qquad x_2=1.0$ Why this step? New slope recomputed at the new point — Euler always uses the slope where it currently stands.

Answer: $y(1)\approx 2.5$ . Check vs exact: the true solution is $y=2e^x-x-1$ , so $y(1)=2e-2\approx 3.44$ . Euler under-shoots because the curve bends upward ( $y''>0$ ) and tangents lie below it. Smaller $h$ closes the gap.

Figure — Direction fields and Euler's method — visual - numerical intuition first

Worked example Forecast-then-Verify

Forecast: For the same problem with $h=0.25$ (4 steps), will $y(1)$ be bigger or smaller, closer or farther from $3.44$ ? Verify: Steps give $y\approx 1,\,1.25,\,1.5625,\,1.953,\,2.441$ → $y(1)\approx 2.44$ ? Recompute carefully: $y_1=1+0.25(1)=1.25$ ; $f=0.25+1.25=1.5\Rightarrow y_2=1.625$ ; $f=0.5+1.625=2.125\Rightarrow y_3=2.156$ ; $f=0.75+2.156=2.906\Rightarrow y_4=2.883$ . So $y(1)\approx 2.88$ — closer to $3.44$ than the coarse $2.50$ . ✔ Smaller $h$ , better.

4. Common mistakes (Steel-man + fix)

5. Active recall

Recall Quick self-test (hide answers)

Q: What does $f(x,y)$ in $y'=f(x,y)$ physically give you? → the slope of the solution at $(x,y)$ .
Q: Euler update formula? → $y_{n+1}=y_n+h f(x_n,y_n)$ .
Q: Order of global error? → $O(h)$ (first order).
Q: Where does Euler evaluate the slope? → at the current/left point.

Recall Feynman: explain to a 12-year-old

Imagine you're walking in fog and a magic compass tells you which way is "downhill" right where you stand. You can't see the whole hill, but you take a small step in that direction, ask the compass again, take another small step, and so on. The path you trace is your best guess of the trail. The ODE is the magic compass; Euler's method is taking tiny steps. Tiny steps = closer to the real trail, but you do more work.

6. Connections

Separable First-Order ODEs — exact solutions to compare Euler against.
Taylor's Theorem — the source of the $h\,y'$ approximation and the $O(h^2)$ remainder.
Runge-Kutta Methods (RK4) — higher-order successors that sample multiple slopes per step.
Existence and Uniqueness (Picard–Lindelöf) — when a single solution curve through a point exists.
Stability and Stiff Equations — when explicit Euler explodes and you need implicit methods.

What does

f(x,y)

y'=f(x,y)

tell you geometrically?

The slope of the solution curve at the point

(x,y)

Define a direction (slope) field.

At each point

(x,y)

a short segment of slope

f(x,y)

; solutions are curves tangent to these segments everywhere.

What is an isocline?

The locus

f(x,y)=m

where all field segments share the same slope

m

State Euler's method update.

y_{n+1}=y_n+h\,f(x_n,y_n)

with

x_{n+1}=x_n+h

Derive Euler from Taylor.

y(x_n+h)=y_n+h y'_n+\tfrac{h^2}{2}y''(\xi)

; replace

y'_n=f(x_n,y_n)

and drop the

O(h^2)

term.

What is the local truncation error of Euler?

\tfrac{h^2}{2}y''(\xi)

, i.e.

O(h^2)

per step.

What is the global error order of Euler?

O(h)

— first order; halving

h

roughly halves total error.

Which point's slope does explicit Euler use?

The current (left) point

(x_n,y_n)

, not the new one.

Why does Euler undershoot a convex (

y''>0

) solution?

Tangent lines lie below a convex curve, so each step lands too low.

Common error: omitting which factor?

Multiplying the slope by the step

h

;

\Delta y=h f

, not

f

Concept Map

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Dekho, ek first-order ODE $\frac{dy}{dx}=f(x,y)$ tumhe seedha $y$ nahi batati — wo har point $(x,y)$ par solution curve ka slope batati hai. Matlab plane ke har jagah ek chhota sa arrow (dash) bana sakte ho jiska tilt $f(x,y)$ ke barabar ho. In saare dashes ka collection hi direction field (slope field) hai. Solution curve wahi hai jo har point par us dash ke "tangent" ho — bas arrows ko follow karte jao, manzil mil jayegi.

Jab formula se solve nahi hota, tab Euler's method kaam aata hai. Idea simple hai: agar tum ek point par khade ho aur slope $f$ pata hai, to thoda sa step $h$ aage chalo us slope ke direction me. Taylor expansion se: $y(x+h)\approx y+h\,y'$ , aur $y'=f(x,y)$ , to formula banta hai $y_{n+1}=y_n+h\,f(x_n,y_n)$ . Yaad rakhne ka tarika: "Stand, Slope, Step" — khade ho, slope padho, step lo. Dhyan rahe slope ko $h$ se multiply karna zaroori hai, kyunki slope "rise per run" hai aur rise $=$ slope $\times h$ .

Error ki baat: har ek step me chhoti si galti $\sim h^2$ hoti hai, aur $L/h$ steps lagte hain, to total error $\sim h$ — isliye Euler ko first-order method kehte hain. $h$ aadha karo to error lagभग aadha. Lekin bahut zyada steps lene par rounding error bhi badhta hai, aur stiff equations me blow up ho sakta hai — isliye aage RK4 jaise better methods aate hain.

Exam aur intuition dono ke liye golden point: pehle field ka picture socho (kaunsi taraf curve jhuk raha hai), phir Euler ke number nikaalo, aur exact solution se compare karke dekho ki Euler overshoot kar raha hai ya undershoot — convex curve ( $y''>0$ ) me Euler hamesha neeche reh jaata hai.