Intuition The big picture (WHY this theorem exists)When you write an ODE like y ′ = f ( x , y ) y' = f(x,y) y ′ = f ( x , y ) y ( x 0 ) = y 0 y(x_0)=y_0 y ( x 0 ) = y 0 assuming a solution exists and that it's the only one. But why should it? Picard–Lindelöf gives the price of admission : if f f f exists and is unique on some small interval around x 0 x_0 x 0
The key idea: a solution to the IVP is the same as a fixed point of an integral operator. "Not too wild" (Lipschitz) makes that operator a contraction (provided the interval is short enough), and contractions have exactly one fixed point. That's the whole soul of the theorem.
Definition Initial Value Problem (IVP)An IVP is the pair
y ′ ( x ) = f ( x , y ) , y ( x 0 ) = y 0 . y'(x) = f(x,y), \qquad y(x_0) = y_0. y ′ ( x ) = f ( x , y ) , y ( x 0 ) = y 0 . y ( x ) y(x) y ( x ) x 0 x_0 x 0
Definition Lipschitz condition iny y y f ( x , y ) f(x,y) f ( x , y ) Lipschitz continuous in y y y on a region R R R L ≥ 0 L \ge 0 L ≥ 0 Lipschitz constant ) such that
∣ f ( x , y 1 ) − f ( x , y 2 ) ∣ ≤ L ∣ y 1 − y 2 ∣ for all ( x , y 1 ) , ( x , y 2 ) ∈ R . |f(x,y_1) - f(x,y_2)| \le L\,|y_1 - y_2| \quad \text{for all } (x,y_1),(x,y_2)\in R. ∣ f ( x , y 1 ) − f ( x , y 2 ) ∣ ≤ L ∣ y 1 − y 2 ∣ for all ( x , y 1 ) , ( x , y 2 ) ∈ R . Meaning: as y y y f f f faster than linearly . The slope in the y y y L L L
Note the two hypotheses play complementary roles: continuity bounds the slope (M M M Lipschitz condition is what makes the iteration converge and pins down a single solution. (Continuity alone still gives existence — the weaker Peano theorem — but it can lose uniqueness.)
Intuition Honest bookkeeping: what each hypothesis really does in the proofA common textbook slogan says "continuity ⟹ existence, Lipschitz ⟹ uniqueness." That's a useful first picture, but in the Picard-iteration proof the Lipschitz estimate is used for both convergence (existence) and uniqueness. The clean split "continuity alone ⟹ existence" belongs to a different proof (Peano, via Arzelà–Ascoli), not to Picard iteration.
h h h M M M ( x 0 , y 0 ) (x_0,y_0) ( x 0 , y 0 ) M M M Δ x \Delta x Δ x M Δ x M\,\Delta x M Δ x b b b f f f
Solution is bounded by lines of slope ± M \pm M ± M ∣ y ( x ) − y 0 ∣ ≤ M ∣ x − x 0 ∣ . |y(x)-y_0| \le M\,|x-x_0|. ∣ y ( x ) − y 0 ∣ ≤ M ∣ x − x 0 ∣. ∣ y − y 0 ∣ ≤ b |y-y_0|\le b ∣ y − y 0 ∣ ≤ b M ∣ x − x 0 ∣ ≤ b M|x-x_0|\le b M ∣ x − x 0 ∣ ≤ b ∣ x − x 0 ∣ ≤ b / M |x-x_0|\le b/M ∣ x − x 0 ∣ ≤ b / M ∣ x − x 0 ∣ ≤ a |x-x_0|\le a ∣ x − x 0 ∣ ≤ a ∣ x − x 0 ∣ ≤ min ( a , b / M ) = h |x-x_0|\le \min(a,\,b/M)=h ∣ x − x 0 ∣ ≤ min ( a , b / M ) = h Why min? The tighter of the two constraints wins.
Intuition Turn the ODE into an integral equationDifferentiation is fragile; integration is robust. Integrate y ′ = f ( x , y ) y'=f(x,y) y ′ = f ( x , y ) x 0 x_0 x 0 x x x y ( x 0 ) = y 0 y(x_0)=y_0 y ( x 0 ) = y 0 y ( x ) = y 0 + ∫ x 0 x f ( t , y ( t ) ) d t . y(x) = y_0 + \int_{x_0}^{x} f(t,\,y(t))\,dt. y ( x ) = y 0 + ∫ x 0 x f ( t , y ( t )) d t . ⟺ \iff ⟺ unchanged by the operator T [ y ] ( x ) = y 0 + ∫ x 0 x f ( t , y ( t ) ) d t T[y](x)=y_0+\int_{x_0}^x f(t,y(t))\,dt T [ y ] ( x ) = y 0 + ∫ x 0 x f ( t , y ( t )) d t T [ y ] = y T[y]=y T [ y ] = y
Build a sequence of Picard iterates :
y 0 ( x ) = y 0 , y n + 1 ( x ) = y 0 + ∫ x 0 x f ( t , y n ( t ) ) d t . y_0(x)=y_0,\qquad y_{n+1}(x) = y_0 + \int_{x_0}^{x} f\big(t,\,y_n(t)\big)\,dt. y 0 ( x ) = y 0 , y n + 1 ( x ) = y 0 + ∫ x 0 x f ( t , y n ( t ) ) d t .
Step A — T T T On ∣ x − x 0 ∣ ≤ h |x-x_0|\le h ∣ x − x 0 ∣ ≤ h ∣ f ∣ ≤ M |f|\le M ∣ f ∣ ≤ M ∣ y n + 1 ( x ) − y 0 ∣ ≤ ∫ x 0 x ∣ f ( t , y n ) ∣ d t ≤ M h ≤ M ⋅ b M = b , |y_{n+1}(x)-y_0| \le \int_{x_0}^x |f(t,y_n)|\,dt \le M\,h \le M\cdot\frac{b}{M}=b, ∣ y n + 1 ( x ) − y 0 ∣ ≤ ∫ x 0 x ∣ f ( t , y n ) ∣ d t ≤ M h ≤ M ⋅ M b = b , R R R f f f (Here is where M M M h ≤ b / M h\le b/M h ≤ b / M
Step B — Why this converges (Lipschitz at work). Then
∣ y n + 1 ( x ) − y n ( x ) ∣ ≤ ∫ x 0 x ∣ f ( t , y n ) − f ( t , y n − 1 ) ∣ d t ≤ L ∫ x 0 x ∣ y n − y n − 1 ∣ d t . |y_{n+1}(x)-y_n(x)| \le \int_{x_0}^{x} |f(t,y_n)-f(t,y_{n-1})|\,dt \le L\int_{x_0}^x |y_n-y_{n-1}|\,dt. ∣ y n + 1 ( x ) − y n ( x ) ∣ ≤ ∫ x 0 x ∣ f ( t , y n ) − f ( t , y n − 1 ) ∣ d t ≤ L ∫ x 0 x ∣ y n − y n − 1 ∣ d t . ∣ y n + 1 ( x ) − y n ( x ) ∣ ≤ M L n ∣ x − x 0 ∣ n + 1 ( n + 1 ) ! , |y_{n+1}(x)-y_n(x)| \le M\,\frac{L^{n}\,|x-x_0|^{n+1}}{(n+1)!}, ∣ y n + 1 ( x ) − y n ( x ) ∣ ≤ M ( n + 1 )! L n ∣ x − x 0 ∣ n + 1 , M L ( e L h − 1 ) < ∞ \dfrac{M}{L}(e^{L h}-1)<\infty L M ( e L h − 1 ) < ∞ uniformly (Weierstrass M-test) to a limit y y y existence , and note it already uses Lipschitz — Lipschitz is not "uniqueness only."
T T T Why it feels right: the estimate sup ∣ T [ u ] − T [ v ] ∣ ≤ L h sup ∣ u − v ∣ \sup|T[u]-T[v]|\le L\,h\,\sup|u-v| sup ∣ T [ u ] − T [ v ] ∣ ≤ L h sup ∣ u − v ∣ The fix: T T T K = L h < 1 K=L\,h<1 K = L h < 1 shrink the interval to h < 1 / L h<1/L h < 1/ L . Without h < 1 / L h<1/L h < 1/ L
Shrink h h h additionally require h ≤ α / L h\le \alpha/L h ≤ α / L α < 1 \alpha<1 α < 1 K = L h < 1 K=L h<1 K = L h < 1 T T T [ x 0 − h , x 0 + h ] [x_0-h,x_0+h] [ x 0 − h , x 0 + h ] h = min ( a , b / M ) h=\min(a,b/M) h = min ( a , b / M ) Weighted norm / factorial bound (no shrinking): use the bound above (or the equivalent Bielecki weighted sup-norm ∥ u ∥ = sup e − 2 L ∣ x − x 0 ∣ ∣ u ( x ) ∣ \|u\|=\sup e^{-2L|x-x_0|}|u(x)| ∥ u ∥ = sup e − 2 L ∣ x − x 0 ∣ ∣ u ( x ) ∣ T T T whole interval ∣ x − x 0 ∣ ≤ h |x-x_0|\le h ∣ x − x 0 ∣ ≤ h h h h L h ≥ 1 Lh\ge1 L h ≥ 1
Uniqueness. If y y y y ~ \tilde y y ~ ϕ ( x ) = ∣ y − y ~ ∣ \phi(x)=|y-\tilde y| ϕ ( x ) = ∣ y − y ~ ∣ ϕ ( x ) ≤ L ∫ x 0 x ϕ ( t ) d t , \phi(x)\le L\int_{x_0}^x \phi(t)\,dt, ϕ ( x ) ≤ L ∫ x 0 x ϕ ( t ) d t , Grönwall's inequality forces ϕ ≡ 0 \phi\equiv 0 ϕ ≡ 0 L h < 1 Lh<1 L h < 1
Worked example 1. Computeh h h y ′ = x 2 + y 2 y'=x^2+y^2 y ′ = x 2 + y 2 y ( 0 ) = 0 y(0)=0 y ( 0 ) = 0 a = 1 , b = 1 a=1,\ b=1 a = 1 , b = 1 R = { ∣ x ∣ ≤ 1 , ∣ y ∣ ≤ 1 } R=\{|x|\le1,|y|\le1\} R = { ∣ x ∣ ≤ 1 , ∣ y ∣ ≤ 1 }
Bound M M M on R R R ∣ f ∣ = ∣ x 2 + y 2 ∣ ≤ 1 + 1 = 2 = M |f|=|x^2+y^2|\le 1+1=2=M ∣ f ∣ = ∣ x 2 + y 2 ∣ ≤ 1 + 1 = 2 = M Why? Both squares are at most 1 1 1 Lipschitz L L L ∂ f / ∂ y = 2 y \partial f/\partial y = 2y ∂ f / ∂ y = 2 y ∣ 2 y ∣ ≤ 2 |2y|\le 2 ∣2 y ∣ ≤ 2 R R R L = 2 L=2 L = 2 Why this gives Lipschitz? Bounded y y y L = sup ∣ ∂ y f ∣ L=\sup|\partial_y f| L = sup ∣ ∂ y f ∣ Interval: h = min ( a , b / M ) = min ( 1 , 1 2 ) = 1 2 h=\min(a,b/M)=\min(1,\tfrac12)=\tfrac12 h = min ( a , b / M ) = min ( 1 , 2 1 ) = 2 1 Contraction check: L h = 2 ⋅ 1 2 = 1 Lh=2\cdot\tfrac12=1 L h = 2 ⋅ 2 1 = 1 not < 1 <1 < 1 ∣ x ∣ ≤ 1 2 |x|\le \tfrac12 ∣ x ∣ ≤ 2 1
Worked example 2. Where can uniqueness FAIL:y ′ = y 2 / 3 y'=y^{2/3} y ′ = y 2/3 y ( 0 ) = 0 y(0)=0 y ( 0 ) = 0
f = y 2 / 3 f=y^{2/3} f = y 2/3 continuous ⟹ Peano gives existence. ✔But ∂ f / ∂ y = 2 3 y − 1 / 3 → ∞ \partial f/\partial y = \tfrac23 y^{-1/3}\to\infty ∂ f / ∂ y = 3 2 y − 1/3 → ∞ y → 0 y\to 0 y → 0 Not Lipschitz near y = 0 y=0 y = 0
Indeed two solutions exist: y ≡ 0 y\equiv 0 y ≡ 0 y = ( x 3 ) 3 y=\left(\tfrac{x}{3}\right)^3 y = ( 3 x ) 3 Why this matters: drop Lipschitz, lose uniqueness — exactly what the theorem warns.
Worked example 3. A clean linear case:y ′ = y y'=y y ′ = y y ( 0 ) = 1 y(0)=1 y ( 0 ) = 1
f = y f=y f = y ∣ ∂ y f ∣ = 1 |\partial_y f|=1 ∣ ∂ y f ∣ = 1 L = 1 L=1 L = 1 R 2 \mathbb{R}^2 R 2 Picard iterates: y 0 = 1 y_0=1 y 0 = 1 y 1 = 1 + ∫ 0 x 1 d t = 1 + x y_1=1+\int_0^x 1\,dt=1+x y 1 = 1 + ∫ 0 x 1 d t = 1 + x y 2 = 1 + ∫ 0 x ( 1 + t ) d t = 1 + x + x 2 2 y_2=1+\int_0^x(1+t)dt=1+x+\tfrac{x^2}{2} y 2 = 1 + ∫ 0 x ( 1 + t ) d t = 1 + x + 2 x 2
Why beautiful: the iterates are exactly the partial sums of e x e^x e x , converging to the unique solution y = e x y=e^x y = e x L h < 1 Lh<1 L h < 1 h < 1 h<1 h < 1 x x x
Common mistake "Continuity off f f Why it feels right: in beginning calculus, continuous functions behave nicely, so you expect "one and only one." The fix: continuity gives only existence (Peano). Example 2 (y 2 / 3 y^{2/3} y 2/3 Lipschitz condition — the extra control on how fast f f f y y y
Common mistake "The solution is guaranteed on the whole rectangle∣ x − x 0 ∣ ≤ a |x-x_0|\le a ∣ x − x 0 ∣ ≤ a Why it feels right: you defined f f f R R R The fix: the solution might escape the box before x x x a a a h = min ( a , b / M ) h=\min(a,b/M) h = min ( a , b / M ) ∣ y − y 0 ∣ |y-y_0| ∣ y − y 0 ∣ ≤ b \le b ≤ b local .
Common mistake "Lipschitz inx x x Why it feels right: x x x The fix: the operator iterates over y y y ∣ f ( t , y 1 ) − f ( t , y 2 ) ∣ |f(t,y_1)-f(t,y_2)| ∣ f ( t , y 1 ) − f ( t , y 2 ) ∣ in y y y , not x x x x x x
L h < 1 Lh<1 L h < 1 Why it feels right: the slick Banach proof needs K = L h < 1 K=Lh<1 K = L h < 1 The fix: L h < 1 Lh<1 L h < 1 one proof route; you can always satisfy it by shrinking h h h stated theorem only needs continuity + Lipschitz; h = min ( a , b / M ) h=\min(a,b/M) h = min ( a , b / M )
Recall Feynman: explain to a 12-year-oldImagine you're playing a video-game car that must start at one spot, and at every point a rule (an arrow) tells you which way to drive next. Existence says: a road actually starts from your spot — you can drive somewhere. Uniqueness says: there's only ONE road, you can't suddenly split into two cars. The catch: the arrows must not change too suddenly as you move up/down (that's the "Lipschitz" rule). If the arrows go crazy-steep (like the y 2 / 3 y^{2/3} y 2/3
Mnemonic Remember the roles"Continuous gets you Started, Lipschitz gets you Settled (converged + unique)."
And for the interval: "h h h m in, because the m ountain (M M M b ox both box you in" → h = min ( a , b / M ) h=\min(a,b/M) h = min ( a , b / M ) L h < 1 Lh<1 L h < 1
Picard–Lindelöf needs which TWO hypotheses on f f f f f f continuous (so bounded by
M M M ) AND
f f f Lipschitz in
y y y on the rectangle.
In the Picard-iteration proof, what does continuity give and what does Lipschitz give? Continuity bounds
∣ f ∣ ≤ M |f|\le M ∣ f ∣ ≤ M (lets iteration start, controls the box); Lipschitz gives BOTH uniform convergence (existence) AND uniqueness.
Is "continuity ⟹ existence, Lipschitz ⟹ uniqueness" exactly true for Picard iteration? No — in Picard's proof Lipschitz is used for existence too. The clean continuity-only existence is a different proof (Peano).
State the Lipschitz condition in y y y ∣ f ( x , y 1 ) − f ( x , y 2 ) ∣ ≤ L ∣ y 1 − y 2 ∣ |f(x,y_1)-f(x,y_2)|\le L|y_1-y_2| ∣ f ( x , y 1 ) − f ( x , y 2 ) ∣ ≤ L ∣ y 1 − y 2 ∣ for all
( x , y i ) (x,y_i) ( x , y i ) in the region.
On what interval is the solution guaranteed? ∣ x − x 0 ∣ ≤ h |x-x_0|\le h ∣ x − x 0 ∣ ≤ h with
h = min ( a , b / M ) h=\min(a,\,b/M) h = min ( a , b / M ) .
Why is the interval restricted to b / M b/M b / M Solution slope is bounded by
M M M , so
∣ y − y 0 ∣ ≤ M ∣ x − x 0 ∣ |y-y_0|\le M|x-x_0| ∣ y − y 0 ∣ ≤ M ∣ x − x 0 ∣ ; staying inside height
b b b forces
∣ x − x 0 ∣ ≤ b / M |x-x_0|\le b/M ∣ x − x 0 ∣ ≤ b / M .
When is the operator T T T Only when
K = L h < 1 K=L\,h<1 K = L h < 1 ; otherwise shrink
h h h below
1 / L 1/L 1/ L , or use a weighted (Bielecki) norm to get contraction for any
h h h .
The IVP is equivalent to which fixed-point problem? y = T [ y ] y=T[y] y = T [ y ] where
T [ y ] ( x ) = y 0 + ∫ x 0 x f ( t , y ( t ) ) d t T[y](x)=y_0+\int_{x_0}^x f(t,y(t))\,dt T [ y ] ( x ) = y 0 + ∫ x 0 x f ( t , y ( t )) d t .
Define the Picard iterates. y 0 ( x ) = y 0 y_0(x)=y_0 y 0 ( x ) = y 0 ,
y n + 1 ( x ) = y 0 + ∫ x 0 x f ( t , y n ( t ) ) d t y_{n+1}(x)=y_0+\int_{x_0}^x f(t,y_n(t))\,dt y n + 1 ( x ) = y 0 + ∫ x 0 x f ( t , y n ( t )) d t .
Which inequality cleanly proves uniqueness without needing L h < 1 Lh<1 L h < 1 Grönwall:
ϕ ( x ) ≤ L ∫ x 0 x ϕ d t \phi(x)\le L\int_{x_0}^x\phi\,dt ϕ ( x ) ≤ L ∫ x 0 x ϕ d t with
ϕ = ∣ y − y ~ ∣ ≥ 0 \phi=|y-\tilde y|\ge0 ϕ = ∣ y − y ~ ∣ ≥ 0 forces
ϕ ≡ 0 \phi\equiv0 ϕ ≡ 0 .
Give an IVP that has existence but NOT uniqueness, and why. y ′ = y 2 / 3 , y ( 0 ) = 0 y'=y^{2/3},y(0)=0 y ′ = y 2/3 , y ( 0 ) = 0 ;
f f f continuous but not Lipschitz at
y = 0 y=0 y = 0 ; solutions
y ≡ 0 y\equiv0 y ≡ 0 and
y = ( x / 3 ) 3 y=(x/3)^3 y = ( x /3 ) 3 .
How do you usually verify the Lipschitz condition in practice? Show
∂ f / ∂ y \partial f/\partial y ∂ f / ∂ y is bounded on a convex region; then
L = sup ∣ ∂ y f ∣ L=\sup|\partial_y f| L = sup ∣ ∂ y f ∣ (via MVT).
For y ′ = x 2 + y 2 , y ( 0 ) = 0 y'=x^2+y^2,y(0)=0 y ′ = x 2 + y 2 , y ( 0 ) = 0 ∣ x ∣ , ∣ y ∣ ≤ 1 |x|,|y|\le1 ∣ x ∣ , ∣ y ∣ ≤ 1 M M M L L L h h h L h < 1 Lh<1 L h < 1 M = 2 M=2 M = 2 ,
L = 2 L=2 L = 2 ,
h = 1 / 2 h=1/2 h = 1/2 ;
L h = 1 Lh=1 L h = 1 , NOT
< 1 <1 < 1 — needs factorial/Bielecki argument.
Lipschitz condition — drives convergence AND uniqueness in Picard's proofBanach fixed-point theorem (contraction mapping) — needs K = L h < 1 K=Lh<1 K = L h < 1 Bielecki weighted norm — makes T T T h h h Grönwall's inequality — clean uniqueness without L h < 1 Lh<1 L h < 1 Peano existence theorem — existence from continuity alone (different proof, no uniqueness)Picard iteration (method of successive approximations) Linear ODEs first order — global Lipschitz ⟹ global unique solutionsMaximal interval of existence — patching local solutions / continuation
alone gives existence via
Intuition Hinglish mein samjhoDekho, jab bhi tum ek differential equation likhte ho jaise y ′ = f ( x , y ) y' = f(x,y) y ′ = f ( x , y ) y ( x 0 ) = y 0 y(x_0)=y_0 y ( x 0 ) = y 0 f f f ∣ x − x 0 ∣ ≤ h |x-x_0|\le h ∣ x − x 0 ∣ ≤ h h = min ( a , b / M ) h=\min(a,b/M) h = min ( a , b / M )
Do conditions hain. Pehli: f f f continuous hona chahiye — isse ∣ f ∣ ≤ M |f|\le M ∣ f ∣ ≤ M f f f Lipschitz in y y y hona chahiye, matlab ∣ f ( x , y 1 ) − f ( x , y 2 ) ∣ ≤ L ∣ y 1 − y 2 ∣ |f(x,y_1)-f(x,y_2)|\le L|y_1-y_2| ∣ f ( x , y 1 ) − f ( x , y 2 ) ∣ ≤ L ∣ y 1 − y 2 ∣ dono ke liye chahiye: iterates ka uniform convergence (existence) AUR uniqueness. Sirf-continuity-se-existence wala clean version alag proof hai (Peano).
Asli jadoo: ODE ko integral equation y ( x ) = y 0 + ∫ x 0 x f ( t , y ( t ) ) d t y(x)=y_0+\int_{x_0}^x f(t,y(t))\,dt y ( x ) = y 0 + ∫ x 0 x f ( t , y ( t )) d t T T T contraction tabhi banta hai jab K = L ⋅ h < 1 K=L\cdot h<1 K = L ⋅ h < 1 L h ≥ 1 Lh\ge1 L h ≥ 1 L h = 1 Lh=1 L h = 1
