Chapter II · Functions of real variables

Curves in a plane

§32 30 min

Describing curves by an equation f(x, y) = 0 or by parametric equations x = f(t), y = F(t); families of curves through given points; and what elimination of the parameter gains and loses.

Why this matters

So far a curve has mostly been the graph of y=f(x)y = f(x). But many curves are not graphs of one-valued functions (a circle fails the vertical line test), and many relations between xx and yy cannot be solved for yy at all, such as y5−y−x=0y^5 - y - x = 0. §32 therefore adopts the equation f(x,y)=0f(x, y) = 0 as the standard way to describe a plane curve, and adds a second way: let both coordinates be functions of a third variable tt. Both descriptions are used throughout analysis, for instance for implicit functions in Chapter V and for lengths of curves in Chapter VI.

The idea

Two ways of describing the same curve:

  • By one equation, f(x,y)=0f(x, y) = 0: the curve is the set of points satisfying it. This includes y=f(x)y = f(x) as the special case y−f(x)=0y - f(x) = 0, and also circles, conics and algebraic curves that cannot be solved for yy. Solving can even hide part of the curve: Hardy notes that solving the circle x2+y2+2Gx+2Fy+C=0x^2 + y^2 + 2Gx + 2Fy + C = 0 for yy gives y=−F±F2−x2−2Gx−Cy = -F \pm \sqrt{F^2 - x^2 - 2Gx - C}, two formulas, while the original single equation describes the whole circle more simply.
  • By parametric equations, x=f(t)x = f(t), y=F(t)y = F(t): as the parameter tt varies, the point (x,y)(x, y) moves and traces the curve. The circle is x=acos⁡tx = a\cos t, y=asin⁡ty = a\sin t; as tt goes from 00 to 2π2\pi the point goes round once, and beyond that it goes round again.

Equations can also be combined. If f=0f = 0 and ϕ=0\phi = 0 are two curves, any point on both satisfies f+λϕ=0f + \lambda\phi = 0 for every number λ\lambda. So a single parameter λ\lambda gives a whole family of curves through the common points.

Explore A family of curves pivoting on fixed points

Try this. With the ellipse and hyperbola, move λ\lambda slowly from −3-3 to 33. Look for the values of λ\lambda where the curve breaks into straight lines. Then try two circles and stop at λ=−1\lambda = -1.

(14x2+y2−1)+0.5 (x2−y2−1)=0(\tfrac14x^2 + y^2 - 1) + 0.5\,(x^2 - y^2 - 1) = 0
xy
  • f + λφ = 0
  • f = 0
  • φ = 0
  • common points of f = 0 and φ = 0
At each marked point f=0f = 0 and ϕ=0\phi = 0, so f+λϕ=0+0.5⋅0=0f + \lambda\phi = 0 + 0.5\cdot 0 = 0: the curve passes through all 4 of them, whatever λ\lambda is.
What to notice

Every member passes through the four marked points, the intersections of the two thin curves f=0f = 0 and ϕ=0\phi = 0. Between them the members change type: ellipses, hyperbolas and, at λ=−1\lambda = -1, −14-\tfrac14 and 11, pairs of straight lines. For two circles, λ=−1\lambda = -1 cancels the squares and leaves the common chord x=1x = 1.

Precise statements

Definition The curve f(x,y)=0f(x, y) = 0 Hardy: §32

The curve (or locus) f(x,y)=0f(x, y) = 0 is the set of points (x,y)(x, y) whose coordinates satisfy the equation. The equation defines yy as a (possibly many-valued) function of xx, and also xx as a function of yy.

Definition Parametric equations Hardy: §32

If x=f(t)x = f(t) and y=F(t)y = F(t), where ff and FF are functions of a variable tt, the set of points (f(t),F(t))(f(t), F(t)), for all tt for which both are defined, is the locus defined by the parametric equations. Eliminating tt between the two equations gives an equation ϕ(x,y)=0\phi(x, y) = 0 satisfied by every point of the locus.

Theorem Curves through the intersections of two curves Hardy: Examples XVIII, 3

For every number λ\lambda, the curve f(x,y)+λϕ(x,y)=0f(x, y) + \lambda\phi(x, y) = 0 passes through every common point of the curves f(x,y)=0f(x, y) = 0 and ϕ(x,y)=0\phi(x, y) = 0.

Theorem A rational parametrisation of the ellipse Hardy: Examples XVIII, 4

Let a,b>0a, b > 0. As tt runs through all real numbers, the point given by

xa=2t1+t2,yb=1−t21+t2\frac{x}{a} = \frac{2t}{1 + t^2}, \qquad \frac{y}{b} = \frac{1 - t^2}{1 + t^2}

describes the ellipse x2/a2+y2/b2=1x^2/a^2 + y^2/b^2 = 1 with exactly one point missing, namely (0,−b)(0, -b). Each other point of the ellipse arises from exactly one tt.

Elimination tells you that every point of the parametric locus lies on the eliminated curve; it does not tell you that every point of that curve is reached. This example shows the difference.

Explore The point that is never reached

Try this. Move tt and watch the point; then widen the range TT step by step up to 6060. Where does the traced arc end? Compare the circle x=acos⁡tx = a\cos t, y=asin⁡ty = a\sin t with range beyond 2π2\pi.

xa=2t1+t2,  yb=1−t21+t2(a=2, b=1.2)\frac{x}{a} = \frac{2t}{1+t^2},\ \ \frac{y}{b} = \frac{1-t^2}{1+t^2}\quad (a = 2,\ b = 1.2)
xy
  • points for −T ≤ t ≤ T
  • the point at t
  • the point (0, −b): no t gives it
At t=1t = 1 the point is (2, 0)(2,\ 0); eliminating tt gives x2a2+y2b2=1\frac{x^2}{a^2} + \frac{y^2}{b^2} = 1, and this point satisfies it (error 00). As t→±∞t \to \pm\infty the point creeps towards (0,−b)(0, -b) from both sides, but no real tt reaches it: the equations give the ellipse minus one point.
What to notice

t=0t = 0 gives the top (0,b)(0, b), t=±1t = \pm1 give (±a,0)(\pm a, 0), and as ∣t∣|t| grows the point slides down both sides towards (0,−b)(0, -b), getting ever closer but never arriving: the eliminated equation is the full ellipse, but the parametric locus lacks one point. The trigonometric circle, by contrast, covers every point, and covers it again and again.

Proofs

Proof Curves through the intersections of two curves
Strategy

Substitute a common point into f+λϕf + \lambda\phi.

Take a common point

Let (x0,y0)(x_0, y_0) lie on both curves, so f(x0,y0)=0f(x_0, y_0) = 0 and ϕ(x0,y0)=0\phi(x_0, y_0) = 0.

Substitute

Then f(x0,y0)+λϕ(x0,y0)=0+λ⋅0=0f(x_0, y_0) + \lambda\phi(x_0, y_0) = 0 + \lambda\cdot0 = 0 whatever λ\lambda is, so (x0,y0)(x_0, y_0) lies on f+λϕ=0f + \lambda\phi = 0.

What this gives in practice

Choosing λ\lambda cleverly simplifies the equation while keeping the common points. With two circles, λ=−1\lambda = -1 removes x2+y2x^2 + y^2 and leaves a straight line through both common points; with two conics, a suitable λ\lambda gives a pair of lines. Solving that simpler equation with one of the originals finds the intersections (Examples XVIII, 1).

Proof A rational parametrisation of the ellipse
Strategy

First check that every point given by the formulas is on the ellipse. Then solve for tt in terms of (x,y)(x, y): the formula for tt works at every point of the ellipse except where its denominator vanishes, which is (0,−b)(0, -b).

Every point is on the ellipse

x2a2+y2b2=4t2+(1−t2)2(1+t2)2=1+2t2+t4(1+t2)2=1.\dfrac{x^2}{a^2} + \dfrac{y^2}{b^2} = \dfrac{4t^2 + (1 - t^2)^2}{(1 + t^2)^2} = \dfrac{1 + 2t^2 + t^4}{(1 + t^2)^2} = 1.

(0,−b)(0, -b) is missed

y=−by = -b would need 1−t2=−(1+t2)1 - t^2 = -(1 + t^2), that is 1=−11 = -1. Impossible, so no tt gives (0,−b)(0, -b).

Recover tt from the point

From the formulas, 1+yb=21+t21 + \dfrac{y}{b} = \dfrac{2}{1 + t^2}, and dividing xa=2t1+t2\dfrac{x}{a} = \dfrac{2t}{1 + t^2} by it gives t=x/a1+y/b=bxa(b+y)t = \dfrac{x/a}{1 + y/b} = \dfrac{bx}{a(b + y)}. So tt is determined by the point: different tt give different points.

Every other point is reached

Let (x,y)(x, y) be on the ellipse with y≠−by \neq -b, and put t=bx/{a(b+y)}t = bx/\{a(b + y)\}. Write X=x/aX = x/a, Y=y/bY = y/b, so X2+Y2=1X^2 + Y^2 = 1 and t=X/(1+Y)t = X/(1 + Y). Then 1+t2=(1+Y)2+X2(1+Y)2=2(1+Y)(1+Y)2=21+Y1 + t^2 = \dfrac{(1 + Y)^2 + X^2}{(1 + Y)^2} = \dfrac{2(1 + Y)}{(1 + Y)^2} = \dfrac{2}{1 + Y}, using X2=1−Y2X^2 = 1 - Y^2. Hence 2t1+t2=X\dfrac{2t}{1 + t^2} = X and 1−t21+t2=21+t2−1=Y\dfrac{1 - t^2}{1 + t^2} = \dfrac{2}{1+t^2} - 1 = Y: this tt gives exactly the point (x,y)(x, y).

Geometrically, after scaling the ellipse to the unit circle X2+Y2=1X^2 + Y^2 = 1, the line joining the missing point (0,−1)(0, -1) to (X,Y)(X, Y) has slope (1+Y)/X=1/t(1 + Y)/X = 1/t. Every line through (0,−1)(0, -1) meets the circle in exactly one more point, except the horizontal tangent there, which would need “t=∞t = \infty”: that is why exactly one point is missed. This “rational parametrisation” is why the substitution t=tan⁡12θt = \tan\tfrac12\theta turns trigonometric integrals into rational ones in Chapter VI.

Worked examples and exercises

Exercise ●○○ Hardy: Examples XVIII, 1

Find the common points of the circles x2+y2=4x^2 + y^2 = 4 and (x−2)2+y2=4(x - 2)^2 + y^2 = 4.

Hint
Subtract one equation from the other (that is, take λ=−1\lambda = -1).

Solution

Subtracting: x2−(x−2)2=0x^2 - (x - 2)^2 = 0, i.e. 4x−4=04x - 4 = 0, so x=1x = 1. Then y2=3y^2 = 3, y=±3y = \pm\sqrt3. The common points are (1,3)(1, \sqrt3) and (1,−3)(1, -\sqrt3): finitely many isolated points, as Hardy says two such equations generally give.

Exercise ●○○ Hardy: Examples XVIII, 2

Trace the curves (x+y)2=1(x + y)^2 = 1, xy=1xy = 1, x2−y2=1x^2 - y^2 = 1.

Hint
The first factorises: (x+y−1)(x+y+1)=0(x + y - 1)(x + y + 1) = 0.

Solution

(x+y)2=1(x + y)^2 = 1 means x+y=1x + y = 1 or x+y=−1x + y = -1: a pair of parallel straight lines, not a single curve. xy=1xy = 1 is the rectangular hyperbola y=1/xy = 1/x, in the first and third quadrants, with the axes as asymptotes. x2−y2=1x^2 - y^2 = 1 is a rectangular hyperbola with vertices (±1,0)(\pm1, 0) and asymptotes y=±xy = \pm x; it is xy=1xy = 1 turned through half a right angle and shrunk by a factor 2\sqrt2. Indeed, in the coordinates u=(x−y)/2u = (x - y)/\sqrt2, v=(x+y)/2v = (x + y)/\sqrt2 (axes turned through 45∘45^\circ) we have x2−y2=2uvx^2 - y^2 = 2uv, so the curve is uv=12uv = \tfrac12, that is (2u)(2v)=1(\sqrt2u)(\sqrt2v) = 1.

Exercise ●●○ Hardy: Examples XVIII, 4

What loci are represented by (α) x=at+bx = at + b, y=ct+dy = ct + d, and (β) x/a=2t/(1+t2)x/a = 2t/(1 + t^2), y/b=(1−t2)/(1+t2)y/b = (1 - t^2)/(1 + t^2), as tt varies through all real values?

Hint
(α): if a≠0a \neq 0, solve the first for tt and substitute. What if a=c=0a = c = 0?

Solution

(α) Eliminating tt: c(x−b)−a(y−d)=0c(x - b) - a(y - d) = 0. If aa and cc are not both 00, this is the straight line through (b,d)(b, d) in the direction (a,c)(a, c), and every point of it is reached: the point (b+as,d+cs)(b + as, d + cs) comes from t=st = s. If a=c=0a = c = 0, the “locus” is the single point (b,d)(b, d).

(β) By the theorem above, the ellipse x2/a2+y2/b2=1x^2/a^2 + y^2/b^2 = 1 without the point (0,−b)(0, -b). (If a=ba = b it is a circle with one point removed.)

Exercise Solving the circle for y ●○○ Hardy: §32

Hardy writes the circle x2+y2+2Gx+2Fy+C=0x^2 + y^2 + 2Gx + 2Fy + C = 0 as y=−F±F2−x2−2Gx−Cy = -F \pm \sqrt{F^2 - x^2 - 2Gx - C}. For which xx is this defined, and what part of the circle does each sign give?

Hint
The circle has centre (−G,−F)(-G, -F) and radius ρ=G2+F2−C\rho = \sqrt{G^2 + F^2 - C}. Rewrite F2−x2−2Gx−CF^2 - x^2 - 2Gx - C in terms of ρ\rho and x+Gx + G.

Solution

F2−x2−2Gx−C=ρ2−(x+G)2F^2 - x^2 - 2Gx - C = \rho^2 - (x + G)^2. So the formula is defined for ∣x+G∣≤ρ|x + G| \le \rho, i.e. for xx between −G−ρ-G - \rho and −G+ρ-G + \rho, the horizontal extent of the circle. The ++ sign gives the upper semicircle (above the centre’s level y=−Fy = -F), the −- sign the lower one; they meet at the two ends x=−G±ρx = -G \pm \rho. Neither is the whole circle: the single equation f(x,y)=0f(x, y) = 0 is the better description, as Hardy says.

Exercise Finding a pair of lines ●●○ Hardy: Examples XVIII, 3

For the ellipse 14x2+y2=1\tfrac14x^2 + y^2 = 1 and the hyperbola x2−y2=1x^2 - y^2 = 1, find every λ\lambda for which 14x2+y2−1+λ(x2−y2−1)=0\tfrac14x^2 + y^2 - 1 + \lambda(x^2 - y^2 - 1) = 0 is a pair of straight lines, and use one of them to find the four common points.

Hint
The member is (14+λ)x2+(1−λ)y2=1+λ(\tfrac14 + \lambda)x^2 + (1 - \lambda)y^2 = 1 + \lambda. When does such an equation factor into lines?

Solution

Ax2+By2=KAx^2 + By^2 = K is a pair of lines when K=0K = 0 and AA, BB have opposite signs (λ=−1\lambda = -1: −34x2+2y2=0-\tfrac34x^2 + 2y^2 = 0, so y=±3/8 xy = \pm\sqrt{3/8}\,x), or when one of AA, BB vanishes and the other has the same sign as KK (λ=−14\lambda = -\tfrac14: 54y2=34\tfrac54y^2 = \tfrac34, the lines y=±3/5y = \pm\sqrt{3/5}; λ=1\lambda = 1: 54x2=2\tfrac54x^2 = 2, the lines x=±8/5x = \pm\sqrt{8/5}). With λ=1\lambda = 1: x2=85x^2 = \tfrac85, and then y2=x2−1=35y^2 = x^2 - 1 = \tfrac35. The four common points are (±8/5,±3/5)(\pm\sqrt{8/5}, \pm\sqrt{3/5}), matching the marked points in the widget.

Check yourself

Check yourself Choose one

The locus (x+y)2=1(x + y)^2 = 1 is

Why

(x+y)2=1(x + y)^2 = 1 means x+y=±1x + y = \pm1: two lines of slope −1-1. A square equal to 11 allows two signs; forgetting one gives the last option.

Check yourself Enter a number

In the parametrisation x/a=2t/(1+t2)x/a = 2t/(1 + t^2), y/b=(1−t2)/(1+t2)y/b = (1 - t^2)/(1 + t^2), which value of tt gives the point (a,0)(a, 0)?

Why

y=0y = 0 needs t2=1t^2 = 1, and x=ax = a needs 2t/(1+t2)=12t/(1 + t^2) = 1, so t=1t = 1 (while t=−1t = -1 gives (−a,0)(-a, 0)). Check with t=bx/{a(b+y)}=ab/(ab)=1t = bx/\{a(b + y)\} = ab/(ab) = 1.

Check yourself True or false?

If eliminating tt from x=f(t)x = f(t), y=F(t)y = F(t) gives ϕ(x,y)=0\phi(x, y) = 0, then every point of the curve ϕ(x,y)=0\phi(x, y) = 0 is (f(t),F(t))(f(t), F(t)) for some tt.

Why

False. Elimination only shows that the parametric points lie on ϕ=0\phi = 0. The rational parametrisation of the ellipse misses (0,−b)(0, -b); and x=t2x = t^2, y=t2y = t^2 gives only the half-line y=x≥0y = x \ge 0 of the line y=xy = x.

Check yourself Choose one

Why does f+λϕ=0f + \lambda\phi = 0 pass through every common point of f=0f = 0 and ϕ=0\phi = 0?

Why

It is pure substitution, valid for any functions ff, ϕ\phi and every λ\lambda; λ\lambda is not chosen to make it work. What may fail is the converse: f+λϕ=0f + \lambda\phi = 0 has other points too.

Hardy’s language

Summary

Summary
  • The standard description of a plane curve is one equation f(x,y)=0f(x, y) = 0; it includes y=f(x)y = f(x) and covers curves that are not graphs or cannot be solved for yy.
  • Parametric equations x=f(t)x = f(t), y=F(t)y = F(t) trace a curve as tt varies; eliminating tt gives an equation the points satisfy, but the parametric locus may be only part of that curve.
  • x/a=2t/(1+t2)x/a = 2t/(1 + t^2), y/b=(1−t2)/(1+t2)y/b = (1 - t^2)/(1 + t^2) gives the ellipse minus (0,−b)(0, -b), and t=bx/{a(b+y)}t = bx/\{a(b + y)\} recovers tt.
  • f+λϕ=0f + \lambda\phi = 0 passes through all common points of f=0f = 0 and ϕ=0\phi = 0; choosing λ\lambda well simplifies finding them.