Intuition The one core ideaIf you walk along a line where a function's value never changes, and you also draw the arrow pointing "straight uphill" the fastest, those two directions always meet at a perfect right angle. This whole topic is just carefully proving that "no change" and "most change" are perpendicular.
Before we can even state the main result , we need to earn every symbol it uses. Below, each piece is built from nothing: plain words → a picture → why the topic needs it . Read top to bottom; each idea rests on the one above it.
Definition A multivariable functionf f ( x , y ) machine that takes two numbers in (a point ( x , y ) one number out .
f : R 2 → R f R 2 R
Picture it as a landscape. Stand on a flat floor whose position is ( x , y ) f ( x , y ) height of the ground at that spot.
Intuition Why we need this firstEverything in this topic — level curves, gradients, tangents — is a statement about this landscape . If you can picture "a spot on the floor" and "the height there," you already have the stage on which everything happens.
The little symbols:
R − 2 0 3.7 π R 2 R n n n n = 2 n = 3
Definition Level curve / level setPick one fixed height and call it c c = 25 level curve is every floor-spot whose height equals exactly c f ( x , y ) = c . = c f ( x , y , z ) = c level surface .
Picture it as a contour line on a hiking map. Every point on one contour is at the same altitude . Walk along it and you never climb or descend.
Intuition Why the topic needs itThe whole claim is "gradient ⟂ level curve." So we must be crystal-clear on what a level curve is : the set of directions along which f not change . Hold that phrase — "no change" — it is half the story.
Common mistake "The level curve is the graph off Why it feels right: both come from the same f The fix: the graph is the 3D hill surface z = f ( x , y ) level curve lives flat on the floor (x y shadow of one horizontal slice of the hill. The gradient lives on that same flat floor.
Before the gradient, we need to know how to measure "how steeply does f which direction do we change?
Definition Partial derivativeSymbol: ∂ x ∂ f f x f x Meaning: freeze y f x f y x
Picture it as walking on a grid line. Standing on the hill, walk only east : how quickly does your altitude rise? That number is f x only north : that's f y
partial and not an ordinary derivative?
An ordinary derivative d x d one input. Here f one direction at a time so we can measure it unambiguously.
Worked example Computing partialsFor f ( x , y ) = x 2 + y 2
f x = 2 x y 2 0 f y = 2 y x 2 ( 3 , 4 ) f x = 6 f y = 8
∇ f ∇ gradient collects all the partials into one vector :
∇ f = ( ∂ x ∂ f , ∂ y ∂ f ) = ( f x , f y ) . ∇ f = ( f x , f y , f z ) arrow that lives on the flat floor (the input space), not on the hill.
Picture it as an uphill arrow drawn on the map. At every floor-spot, ∇ f steepest increase of height, and its length says how steep.
Intuition Why bundle the partials this way?Two separate numbers (f x f y vector , they suddenly encode the single best uphill direction — which is neither pure-east nor pure-north but the diagonal that combines them. That combined arrow is exactly what the topic claims is perpendicular to level curves.
Worked example Gradient of the bowlf = x 2 + y 2 ∇ f = ( 2 x , 2 y ) ( 3 , 4 ) ( 6 , 8 ) straight away from the origin , i.e. outward, the steepest-uphill direction for a bowl. Notice ( 6 , 8 ) = 2 ⋅ ( 3 , 4 )
The claim uses the word "perpendicular." We measure that with the dot product .
A vector like u = ( u 1 , u 2 ) u u
The magic case we need:
Intuition Why this is THE tool for "perpendicular""Perpendicular" is a statement about an angle , and the dot product is the one gadget that turns an angle into simple arithmetic (a 1 b 1 + a 2 b 2 ∇ f 0
Worked example Right-angle check( 6 , 8 ) ⋅ ( − 4 , 3 ) = ( 6 ) ( − 4 ) + ( 8 ) ( 3 ) = − 24 + 24 = 0. ( 6 , 8 ) ( − 4 , 3 )
Recall Edge case: the zero vectorIf ∇ f = ( 0 , 0 ) 0 everything , so "perpendicular" loses meaning. These are exactly the critical points , handled separately in Lagrange Multipliers and optimization. Away from them, the argument is clean.
Definition Parametrized curver ( t ) r ( t ) = ( x ( t ) , y ( t )) moving point whose location depends on time t t r ( t )
Definition Velocity / tangent vectorr ′ ( t ) r ′ ( t ) = ( x ′ ( t ) , y ′ ( t )) velocity arrow — it points in the direction of motion and is tangent to the path.
Picture it as an ant walking a route on the map. At any instant the ant faces some direction; that facing arrow is r ′ ( t )
Intuition Why the topic needs a moving pointTo say "∇ f a direction along the level curve . The trick: send the ant walking inside the level curve. Then its velocity r ′ ( t ) is a tangent-to-the-level-curve direction — precisely the thing we'll dot against ∇ f
The final tool. The ant walks the path r ( t ) height f ( r ( t ))
Intuition Why the chain rule, and why it seals the proofWe know two facts and need to connect them: (a) how the height depends on position (that's ∇ f how position depends on time (that's r ′ Multivariable Chain Rule is exactly the bridge that multiplies these together. Now the punchline: if the ant walks inside a level curve , its height never changes, so the left side is 0 ∇ f ⋅ r ′ = 0 ∇ f ⊥ r ′
This same dot product is also the definition of the Directional Derivative D u f = ∇ f ⋅ u
Function f of x and y as a height landscape
Level curve f = c, no change in height
Partial derivatives f_x and f_y
Gradient grad f, the uphill arrow
Zero dot product means perpendicular
Parametrized curve r of t
Tangent vector r prime of t
Chain rule d dt f of r = grad f dot r prime
Gradient perpendicular to level curves
What does f : R 2 → R A machine taking a point ( x , y )
What is a level curve f ( x , y ) = c All floor-spots at the same fixed height c f
How do you compute f x Freeze y f x
What is ∇ f The vector ( f x , f y )
What does a ⋅ b = 0 The two (nonzero) arrows are perpendicular, because cos 9 0 ∘ = 0
Compute ( 6 , 8 ) ⋅ ( − 4 , 3 ) − 24 + 24 = 0
What is r ′ ( t ) r ( t ) The velocity/tangent arrow, pointing along the direction of motion.
State the multivariable chain rule for d t d f ( r ( t )) ∇ f ( r ( t )) ⋅ r ′ ( t )
Why is ∇ f = ( 0 , 0 ) It is a critical (flat) point; its dot product with everything is 0
