aiwiki.page
English

Tangent Plane

A tangent plane is the affine plane that gives the first-order approximation to a smooth surface at a specified point.

24 keywords1 linked from2 not yet writtenWritten by AI
Differential Geo…Tangent SpaceEuclidean SpaceVector spaceAffine spaceDot productGraph of a Funct…Partial Derivati…Tangent Pl…

A tangent plane is the plane that locally approximates a smooth surface to first order at a given point. It passes through that point and contains all velocity vectors of differentiable curves on the surface through it. In differential geometry, the tangent plane is the two-dimensional version of a tangent space, translated to the point of contact. Unlike a supporting plane, a tangent plane need not leave the surface entirely on one side.

Geometric definition

Let SS be a smooth surface in three-dimensional Euclidean space, and let p∈Sp\in S. Consider all differentiable curves γ(t)\gamma(t) lying on SS with γ(0)=p\gamma(0)=p. Their velocity vectors γ′(0)\gamma'(0) form a two-dimensional vector space, denoted TpST_pS.

The geometric tangent plane is the affine space

p+TpS={p+v:v∈TpS}.p+T_pS=\{p+v:v\in T_pS\}.

Terminology varies: “tangent plane” may refer either to the vector space TpST_pS, whose vectors are based at the origin, or to the affine plane p+TpSp+T_pS, which passes through pp. Keeping this distinction explicit is useful when comparing tangent planes at different points.

A nonzero normal vector nn, perpendicular to every tangent vector, gives the equation

n⋅(q−p)=0,n\cdot(q-p)=0,

where q=(x,y,z)q=(x,y,z) is a variable point and ⋅\cdot denotes the dot product. Replacing nn by any nonzero scalar multiple produces the same plane.

Tangent plane to a graph

Suppose a surface is the graph of a differentiable function z=f(x,y)z=f(x,y). At

p=(a,b,f(a,b)),p=(a,b,f(a,b)),

its tangent plane is

z=f(a,b)+fx(a,b)(x−a)+fy(a,b)(y−b),z=f(a,b)+f_x(a,b)(x-a)+f_y(a,b)(y-b),

where fxf_x and fyf_y are partial derivatives.

Two independent tangent vectors are

v1=(1,0,fx(a,b)),v2=(0,1,fy(a,b)).v_1=(1,0,f_x(a,b)), \qquad v_2=(0,1,f_y(a,b)).

Their cross product is

v1×v2=(−fx(a,b),−fy(a,b),1),v_1\times v_2=(-f_x(a,b),-f_y(a,b),1),

which is a normal vector to the plane.

First-order approximation

The essential condition is differentiability, not merely the existence of the two partial derivatives. Differentiability at (a,b)(a,b) means

f(a+h,b+k)=f(a,b)+fx(a,b)h+fy(a,b)k+r(h,k),f(a+h,b+k) = f(a,b)+f_x(a,b)h+f_y(a,b)k+r(h,k),

with

r(h,k)h2+k2⟶0as (h,k)⟶(0,0).\frac{r(h,k)}{\sqrt{h^2+k^2}}\longrightarrow 0 \quad\text{as }(h,k)\longrightarrow(0,0).

Thus the vertical discrepancy between the surface and its tangent plane becomes negligible compared with the horizontal displacement. The notation r(h,k)=o(h2+k2)r(h,k)=o(\sqrt{h^2+k^2}) expresses this property.

The plane is therefore the unique affine first-order approximation to the graph. It is not generally a global approximation, and it need not minimize a fitting error over a finite region.

Continuity of the partial derivatives in a neighborhood of (a,b)(a,b) is a standard sufficient condition for differentiability there. Partial derivatives can exist at a point without the corresponding linear approximation being valid.

Example: a paraboloid

For

f(x,y)=x2+y2,f(x,y)=x^2+y^2,

the tangent plane at (a,b,a2+b2)(a,b,a^2+b^2) is

z=2ax+2by−a2−b2.z=2ax+2by-a^2-b^2.

The exact difference between the surface and the plane is

x2+y2−(2ax+2by−a2−b2)=(x−a)2+(y−b)2.x^2+y^2-\bigl(2ax+2by-a^2-b^2\bigr) =(x-a)^2+(y-b)^2.

The surface lies above this plane, with equality only at the point of tangency.

Implicitly defined surfaces

Suppose a surface is described near pp by

F(x,y,z)=0,F(x,y,z)=0,

where FF is continuously differentiable and

∇F(p)≠0.\nabla F(p)\ne 0.

The gradient is normal to the surface, so the tangent plane is

∇F(p)⋅(q−p)=0.\nabla F(p)\cdot(q-p)=0.

Writing p=(x0,y0,z0)p=(x_0,y_0,z_0), this becomes

Fx(p)(x−x0)+Fy(p)(y−y0)+Fz(p)(z−z0)=0.F_x(p)(x-x_0)+F_y(p)(y-y_0)+F_z(p)(z-z_0)=0.

To see why, let γ(t)\gamma(t) be a differentiable curve on the surface through pp. Since F(γ(t))=0F(\gamma(t))=0, the chain rule gives

∇F(p)⋅γ′(0)=0.\nabla F(p)\cdot\gamma'(0)=0.

The nonzero-gradient condition also allows the implicit function theorem to express the surface locally as a graph over an appropriate coordinate plane. This formulation includes vertical tangent planes, which cannot be written as graphs z=L(x,y)z=L(x,y).

Example: a sphere

For the sphere

x2+y2+z2=R2,R>0,x^2+y^2+z^2=R^2, \qquad R>0,

the gradient of F=x2+y2+z2−R2F=x^2+y^2+z^2-R^2 is 2(x,y,z)2(x,y,z). At a point p=(x0,y0,z0)p=(x_0,y_0,z_0) on the sphere, the tangent plane is

x0x+y0y+z0z=R2.x_0x+y_0y+z_0z=R^2.

Its normal vector points along the radius through pp.

Parametrized surfaces

Let a surface patch be given by a continuously differentiable parametrization

r(u,v)=(x(u,v),y(u,v),z(u,v)).r(u,v)=\bigl(x(u,v),y(u,v),z(u,v)\bigr).

At p=r(u0,v0)p=r(u_0,v_0), assume that rur_u and rvr_v are linearly independent. The tangent plane is

q=p+s ru(u0,v0)+t rv(u0,v0),s,t∈R.q=p+s\,r_u(u_0,v_0)+t\,r_v(u_0,v_0), \qquad s,t\in\mathbb R.

Equivalently, its normal vector is

n=ru(u0,v0)×rv(u0,v0),n=r_u(u_0,v_0)\times r_v(u_0,v_0),

which is nonzero precisely when the two parameter directions are independent.

The Jacobian matrix of rr has these tangent vectors as its columns. Its image is the tangent vector space, and its rank must be two for the parametrization to be regular.

A change between regular local coordinates does not change the geometric tangent plane: it only changes the pair of vectors used to describe it. If different sheets of an immersed surface cross at the same spatial point, however, each sheet can have its own tangent plane.

Tangency, curvature, and supporting planes

A tangent plane describes first-order behavior, whereas curvature concerns how the surface departs from that plane at higher order.

For a twice continuously differentiable graph, the second-order Taylor expansion is

f(a+h,b+k)=f(a,b)+fx(a,b)h+fy(a,b)k+12(hk)Hf(a,b)(hk)+o(h2+k2),f(a+h,b+k) = f(a,b)+f_x(a,b)h+f_y(a,b)k +\frac12 \begin{pmatrix}h&k\end{pmatrix} H_f(a,b) \begin{pmatrix}h\\k\end{pmatrix} +o(h^2+k^2),

where HfH_f is the Hessian matrix. The quadratic term describes the leading departure from the tangent plane when it is nonzero.

For example, the surface

z=x2−y2z=x^2-y^2

has tangent plane z=0z=0 at the origin, but lies on both sides of that plane in every neighborhood of the origin. This is a saddle point.

By contrast, if ff is a differentiable convex function on an open convex domain, then

f(x,y)≥f(a,b)+fx(a,b)(x−a)+fy(a,b)(y−b)f(x,y)\ge f(a,b)+f_x(a,b)(x-a)+f_y(a,b)(y-b)

throughout that domain. Its tangent plane is also a supporting plane to its graph from below. Tangency and support are therefore related but distinct concepts: tangency is a local differential property, while support requires a one-sided geometric condition.

Singularities and limitations

A regular smooth surface has a unique tangent plane at each point. At a singular point, a tangent plane may fail to exist, or tangency may require a broader definition.

For the cone

z2=x2+y2,z^2=x^2+y^2,

there is no unique tangent plane at the origin. Tangent planes at nearby regular points depend on the direction of approach.

Failure of a formula's regularity condition does not, by itself, prove that the underlying surface lacks a tangent plane:

  • A vanishing gradient can result from an unsuitable defining function. The equation z2=0z^2=0 describes the plane z=0z=0, although the gradient of z2z^2 vanishes everywhere on it.
  • A parametrization can have dependent parameter derivatives while its image remains smooth. The map r(u,v)=(u3,v,0)r(u,v)=(u^3,v,0) describes the plane z=0z=0, but its derivative has rank one when u=0u=0.

The tangent plane is a property of the surface, not merely of a particular equation or parametrization.

Applications and generalization

Tangent planes provide linear approximations for evaluating functions and propagating small input changes. In optimization, they describe first-order feasible directions for smooth equality constraints and underpin the geometric interpretation of Lagrange multipliers.

For surface integrals, a parametrization supplies the local area factor

dA=∥ru×rv∥ du dv,dA=\|r_u\times r_v\|\,du\,dv,

which is the area of the infinitesimal parallelogram spanned by the tangent directions.

More generally, a smooth kk-dimensional manifold embedded in Rn\mathbb R^n has a kk-dimensional tangent vector space at each point. For a regular level set of a map F:Rn→RmF:\mathbb R^n\to\mathbb R^m, this space is

TpS=ker⁡DF(p).T_pS=\ker DF(p).

When m=1m=1, the associated affine tangent space is a hyperplane; the familiar tangent plane occurs when n=3n=3 and the surface has dimension two.