Tangent Planes

The flat plane fitted at one point.

A tangent line, one dimension down

The curve y = x² has slope 2x, which is 4 at the point (2, 4). The tangent line there is y = 4 + 4(x − 2), that is y = 4x − 4. Near x = 2 the line and the curve are hard to tell apart: at x = 2.1 the curve is at 4.41 and the line at 4.4, a gap of 0.01.

So near the point, the line can stand in for the curve. A surface has the same thing one dimension up: a flat plane that rests on it at one point.

xy

The curve y = x² and its tangent line y = 4x − 4 at the marked point (2, 4). The line meets the curve there with the same slope, 4, and stays close to it on either side.

A flat plane resting on a surface

Take the dome f(x, y) = 16(1 − x² − y²) and the point above (0.54, 0.42), at height 16(1 − 0.2916 − 0.1764) = 8.512. Close to that point the dome curves only a little, and a flat plane can be laid against it so that it touches there and tilts the same way the dome does. That plane is the tangent plane at the point.

At (0.54, 0.42) the partial derivatives are ∂f/∂x = −32x = −17.28 and ∂f/∂y = −32y = −13.44: the dome falls in both the x and the y directions, steeply, and the plane falls with it.

The dome z = 16(1 − x² − y²) and a gold patch of its tangent plane at the marked point above (0.54, 0.42). The patch is built from the two partial derivatives there: its height changes by −17.28 per unit in x and by −13.44 per unit in y.

Two slopes fix the plane

A plane through the point (x₀, y₀, z₀) that is not vertical can be written z − z₀ = a(x − x₀) + b(y − y₀). Moving in x alone, it climbs a for each unit; moving in y alone, it climbs b. To rest on the surface, its slope in x must equal ∂f/∂x there and its slope in y must equal ∂f/∂y.

So the tangent plane at (x₀, y₀) is z − z₀ = (∂f/∂x)(x − x₀) + (∂f/∂y)(y − y₀), with both partial derivatives evaluated at (x₀, y₀) and z₀ = f(x₀, y₀). One slope alone is not enough: it fixes a line in the plane, and the plane could still swing about that line.

The bowl at (3, 4)

For f(x, y) = x² + y² at (3, 4): z₀ = 9 + 16 = 25, ∂f/∂x = 2x = 6 and ∂f/∂y = 2y = 8. The tangent plane is z = 25 + 6(x − 3) + 8(y − 4), which multiplies out to z = 6x + 8y − 25. At (3, 4) it gives 18 + 32 − 25 = 25, so the plane passes through the point on the surface.

Cut both with the vertical plane y = 4. The surface becomes the curve z = x² + 16 and the tangent plane becomes the line z = 25 + 6(x − 3), that is z = 6x + 7: the tangent line to that curve at x = 3. Cut with x = 3 instead, and the curve z = 9 + y² meets the line z = 8y − 7 at y = 4. The tangent plane holds the tangent line of every slice through the point.

xz

The bowl z = x² + y² and its tangent plane at (3, 4), both cut along y = 4. The plain curve is the slice z = x² + 16; the gold line is the plane’s slice, z = 6x + 7, touching the curve at (3, 25) with slope 6, the value of ∂f/∂x there.

How close it stays

Step from (3, 4) to (3.1, 4.2). The surface is at 3.1² + 4.2² = 27.25 and the plane at 25 + 0.6 + 1.6 = 27.2, a gap of 0.05. Step a tenth as far, to (3.01, 4.02): the surface is at 25.2205 and the plane at 25.22, a gap of 0.0005, a hundredth of the first.

For this bowl the gap is exactly h² + k² for a step of h in x and k in y, so it shrinks with the square of the step while the step itself shrinks only in proportion. That is what makes the plane the best flat approximation: near the point, f(x, y) ≈ z₀ + (∂f/∂x)(x − x₀) + (∂f/∂y)(y − y₀), and the estimate 27.2 for f(3.1, 4.2) is out by 0.05.

A plane built at one point

The partial derivatives change from point to point, so each point has its own tangent plane, and one plane cannot fit a curved surface everywhere.

Nor need the surface stay on one side of it. For the saddle z = x² − y² at the origin, both partial derivatives are 0 and the tangent plane is z = 0. The saddle lies above that plane along the x-axis, where z = x², and below it along the y-axis, where z = −y², crossing it along the lines y = x and y = −x.

The usual mistakes

Pairing a slope with the wrong variable. ∂f/∂x multiplies (x − x₀) and ∂f/∂y multiplies (y − y₀): for the bowl at (3, 4), z = 25 + 8(x − 3) + 6(y − 4) is a different plane, and it misses the surface’s slope in both directions.

Leaving the partial derivatives as functions. In z = 25 + 2x(x − 3) + 2y(y − 4) the terms are not linear, so it is not a plane; evaluate 2x and 2y at (3, 4) first.

Dropping z₀. z = 6(x − 3) + 8(y − 4) has the right tilt but passes through (3, 4, 0), 25 units below the surface.

A tree measured with a tape and a clinometer

In the application below, a tree’s height is H = d tan θ, a function of the distance d and the angle θ. Its tangent plane at the readings estimates the height from a second pair of readings, and adding the sizes of the two partial-derivative terms gives the largest error each instrument can cause.

Worked example: A Tree Measured With a Tape and a Clinometer: an Estimate From the Tangent Plane, and the Error From Each Instrument

Question A forester stands on level ground d meters from the foot of a tree and sights the top from ground level at an angle of elevation θ, so that the tree is H(d, θ) = dtanθ meters tall. She measures d = 30 meters and θ = 45°. (a) Find the height, and the tangent plane approximation to H near these readings, with θ in radians. A second forester stands 31 meters from the tree and reads 44°. Use the tangent plane to estimate the height from these readings, and compare the estimate with the exact value. (b) The first forester's tape is accurate to within 0.2 meters and her clinometer to within 1°. Estimate the largest error in her height, and find how accurate the clinometer must be, with the same tape, for that error to be at most 0.5 meters.

  1. 1.(a) At the readings, H(30, 45°) = 30tan 45° = 30 × 1 = 30 meters.

    45 degd = 30 mH(a) H = 30 × tan 45 deg = 30 m
    45 degd = 30 mH(a) H = 30 × tan 45 deg = 30 m
    (a) From 30 meters away the top is at 45°, so the tree is 30tan 45° = 30 meters tall.
  2. 2.Holding θ constant, ∂ H∂ d = tanθ = 1. Holding d constant, ∂ H∂ θ = dsec2θ = 30 × 2 = 60 meters per radian. So the tangent plane is H ≈ 30 + 1 × (d − 30) + 60(θ − π4).

    45 degd = 30 mH(a) H = 30 × tan 45 deg = 30 mrates: 1 m per m of d, 60 m per radian
    45 degd = 30 mH(a) H = 30 × tan 45 deg = 30 mrates: 1 m per m of d, 60 m per radian
    At these readings ∂ H∂ d = 1 meter per meter and ∂ H∂ θ = 60 meters per radian, which is about 1.05 meters per degree.
  3. 3.A change of −1° is −π180 ≈ −0.01745 radians, so at d = 31 and θ = 44° the plane gives H ≈ 30 + 1 − 60 × 0.01745 = 31 − 1.047 = 29.95 meters. Exactly, 31tan 44° ≈ 29.94 meters, so the estimate is out by about 0.02 meters.

    45 degd = 30 mH(a) H = 30 × tan 45 deg = 30 mrates: 1 m per m of d, 60 m per radian31 m, 44 deg: 30 + 1 − 60 × 0.01745 = 29.95 m
    45 degd = 30 mH(a) H = 30 × tan 45 deg = 30 mrates: 1 m per m of d, 60 m per radian31 m, 44 deg: 30 + 1 − 60 × 0.01745 = 29.95 m
    The second forester, 31 meters away, reads 44°; the tangent plane gives 29.95 meters, against 29.94 meters exactly.
  4. 4.(b) The largest error comes when both errors push the height the same way: |Δ H| ≈ 1 × 0.2 + 60 × π180 = 0.2 + 1.047 ≈ 1.25 meters. The clinometer causes more than five times as much of it as the tape.

    45 degd = 30 mH±1.05 m(a) H = 30 × tan 45 deg = 30 mrates: 1 m per m of d, 60 m per radian31 m, 44 deg: 30 + 1 − 60 × 0.01745 = 29.95 m(b) 1 × 0.2 + 60 × 0.01745 = 1.25 m
    45 degd = 30 mH±1.05 m(a) H = 30 × tan 45 deg = 30 mrates: 1 m per m of d, 60 m per radian31 m, 44 deg: 30 + 1 − 60 × 0.01745 = 29.95 m(b) 1 × 0.2 + 60 × 0.01745 = 1.25 m
    (b) One degree either way moves the sight line about 1.05 meters up or down the trunk; the tape adds at most 0.2 meters.
  5. 5.With the tape unchanged and the clinometer accurate to within δ radians, the largest error is about 0.2 + 60δ, and 0.2 + 60δ ≤ 0.5 needs δ ≤ 0.005 radians, which is 0.005 × 180π ≈ 0.29°. Check: 0.2 + 60 × 0.005 = 0.2 + 0.3 = 0.5 meters.

    45 degd = 30 mH±1.05 m(a) H = 30 × tan 45 deg = 30 mrates: 1 m per m of d, 60 m per radian31 m, 44 deg: 30 + 1 − 60 × 0.01745 = 29.95 m(b) 1 × 0.2 + 60 × 0.01745 = 1.25 m0.2 + 60 × 0.005 = 0.5: 0.005 rad = 0.29 deg
    45 degd = 30 mH±1.05 m(a) H = 30 × tan 45 deg = 30 mrates: 1 m per m of d, 60 m per radian31 m, 44 deg: 30 + 1 − 60 × 0.01745 = 29.95 m(b) 1 × 0.2 + 60 × 0.01745 = 1.25 m0.2 + 60 × 0.005 = 0.5: 0.005 rad = 0.29 deg
    For a largest error of 0.5 meters, the angle must be read to within 0.005 radians, about 0.29°.

Answer: (a) 30 meters; H ≈ 30 + (d − 30) + 60(θ − π4); the estimate for 31 meters and 44° is 29.95 meters, against 29.94 meters exactly; (b) about 1.25 meters, most of it from the clinometer, which must be accurate to within about 0.29°

Common mistakes

  • Using the angle error in degrees, 60 × 1 = 60 meters. The partial derivative dsec2θ is in meters per radian, and 1° is only 0.01745 radians.
  • Letting the two errors cancel, as in 1.047 − 0.2. The two errors are independent and either may have either sign, so the largest possible error adds their sizes.

More partial derivatives problems, worked step by step →

Practice Tangent Planes in the app