Chapter II · Functions of real variables

Polynomials and their graphs

§23 30 min

The shapes of xmx^m for even and odd mm, why a chord of y=x2y = x^2 lies above the curve, and why the highest power of a polynomial wins for large xx.

Why this matters

Polynomials are the simplest functions that are not straight lines, and the first family in Hardy’s catalogue of functions (§23). Three ideas from this section return again and again: the graph of xmx^m depends on whether mm is even or odd; a curve can be shown to bend one way by comparing it with its chords; and in a polynomial the highest power eventually swamps everything else. The last is the first appearance of a theme of the whole book, the relative rate of growth of functions.

The idea

A polynomial in xx is a function

y=a0xm+a1xm−1+⋯+am−1x+am,y = a_0 x^m + a_1 x^{m-1} + \dots + a_{m-1}x + a_m,

with constant coefficients a0,…,ama_0, \dots, a_m. Every polynomial is built from the powers x,x2,x3,…x, x^2, x^3, \dots by two operations on graphs: multiplying every ordinate by a constant, and adding the ordinates of two graphs above each xx. So the powers come first.

Explore The family y=xmy = x^m

Try this. Move mm from 11 to 1010. Compare the even values with the odd ones. Then look at the stretch between −1-1 and 11, and outside it, as mm grows.

y=xmy = x^mHardy: Examples XI, 1
xy
  • xᵐ
  • trace point (drag)
At x=0.72x = 0.72: y=0.5184y = 0.5184
What to notice

Every even power passes through (−1,1)(-1, 1), (0,0)(0, 0), (1,1)(1, 1) and is symmetric about the axis of yy; every odd power passes through (−1,−1)(-1, -1), (0,0)(0, 0), (1,1)(1, 1) and is symmetric about the origin. As mm grows the curve hugs the axis of xx more and more closely between −1-1 and 11 and climbs more and more steeply outside, approaching the shape of a box.

The symmetry comes straight from algebra: for even mm, (−x)m=xm(-x)^m = x^m, so the points (x,y)(x, y) and (−x,y)(-x, y) are on the graph together; for odd mm, (−x)m=−xm(-x)^m = -x^m, so (x,y)(x, y) and (−x,−y)(-x, -y) are. The flattening and steepening come from 0<xm+1<xm0 < x^{m+1} < x^m when 0<x<10 < x < 1, and xm+1>xmx^{m+1} > x^m when x>1x > 1.

Hardy raises a question here that he cannot yet answer: plotting points, however many, does not prove that the graph of x2x^2 is an unbroken curve. That needs a definition of continuity, which comes in Chapter V. One property, though, can be proved now with nothing but algebra: the parabola bends upwards everywhere.

Explore Every chord of y=x2y = x^2 lies above the curve

Try this. Drag P0P_0 and P1P_1 along the parabola and move λ\lambda. Compare the length of the vermilion gap with the formula in the read-out.

y=x2y = x^2
xy
  • the curve y = x²
  • the chord P₀P₁
  • the gap
  • P₀, P₁ (drag)
At x=λx0+μx1=0.6x = \lambda x_0 + \mu x_1 = 0.6 the chord is at height 2.522.52 and the curve at 0.360.36. The gap is 2.162.16, and λμ(x1−x0)2=0.4⋅0.6⋅9=2.16\lambda\mu(x_1 - x_0)^2 = 0.4\cdot0.6\cdot9 = 2.16.
What to notice

The gap is never negative, and it equals λμ(x1−x0)2\lambda\mu(x_1 - x_0)^2 exactly. It is largest at the middle of the chord (λ=μ=12\lambda = \mu = \tfrac12) and zero only at the ends. Wherever the chord is, the curve is below it.

Precise statements

Definition Polynomial Hardy: §23

A polynomial in xx is a function of the form a0xm+a1xm−1+⋯+ama_0x^m + a_1x^{m-1} + \dots + a_m, where mm is a non-negative integer and a0,…,ama_0, \dots, a_m are constants. If a0≠0a_0 \neq 0, mm is its degree and a0xma_0x^m its leading term.

Theorem The parabola is convex Hardy: §23

Let x0≠x1x_0 \neq x_1 and let λ,μ>0\lambda, \mu > 0 with λ+μ=1\lambda + \mu = 1. Then

(λx02+μx12)−(λx0+μx1)2=λμ(x1−x0)2>0.(\lambda x_0^2 + \mu x_1^2) - (\lambda x_0 + \mu x_1)^2 = \lambda\mu(x_1 - x_0)^2 > 0.

That is, every point of the chord joining two points of y=x2y = x^2, other than its ends, lies above the curve.

The point (λx0+μx1,λx02+μx12)(\lambda x_0 + \mu x_1, \lambda x_0^2 + \mu x_1^2) runs along the chord from P1P_1 (at λ=0\lambda = 0) to P0P_0 (at λ=1\lambda = 1), so the identity covers every interior point of the chord.

Theorem The leading term dominates

Let P(x)=a0xm+a1xm−1+⋯+amP(x) = a_0x^m + a_1x^{m-1} + \dots + a_m with a0≠0a_0 \neq 0 and m≥1m \ge 1, and put K=∣a1∣+∣a2∣+⋯+∣am∣K = |a_1| + |a_2| + \dots + |a_m|. If ∣x∣≥1|x| \ge 1, then

∣a1xm−1+⋯+am∣≤K∣a0∣ ∣x∣ ∣a0xm∣.|a_1x^{m-1} + \dots + a_m| \le \frac{K}{|a_0|\,|x|}\,|a_0x^m|.

So when ∣x∣≥1|x| \ge 1 and ∣x∣>NK/∣a0∣|x| > NK/|a_0|, all the other terms together are less than 1/N1/N of the leading term.

Proofs

Proof The parabola is convex
Strategy

Expand both sides and use λ+μ=1\lambda + \mu = 1 to make the expression symmetric. Hardy’s identity is just the algebra of a weighted average.

Rewrite the chord height

Since λ+μ=1\lambda + \mu = 1, the chord height is λx02+μx12=(λ+μ)(λx02+μx12)=λ2x02+λμ(x02+x12)+μ2x12\lambda x_0^2 + \mu x_1^2 = (\lambda + \mu)(\lambda x_0^2 + \mu x_1^2) = \lambda^2x_0^2 + \lambda\mu(x_0^2 + x_1^2) + \mu^2x_1^2.

Expand the curve height

(λx0+μx1)2=λ2x02+2λμx0x1+μ2x12(\lambda x_0 + \mu x_1)^2 = \lambda^2x_0^2 + 2\lambda\mu x_0x_1 + \mu^2x_1^2.

Subtract

The squares cancel and leave λμ(x02+x12−2x0x1)=λμ(x1−x0)2\lambda\mu(x_0^2 + x_1^2 - 2x_0x_1) = \lambda\mu(x_1 - x_0)^2.

Sign

λ>0\lambda > 0, μ>0\mu > 0 and (x1−x0)2>0(x_1 - x_0)^2 > 0 because x0≠x1x_0 \neq x_1. So the chord is strictly above the curve at every interior point.

Proof The leading term dominates
Strategy

When ∣x∣≥1|x| \ge 1, every power ∣x∣m−k|x|^{m-k} with k≥1k \ge 1 is at most ∣x∣m−1|x|^{m-1}. Bound the lower terms by that, and compare with ∣a0∣ ∣x∣m|a_0|\,|x|^m.

Each lower term is at most ∣ak∣ ∣x∣m−1|a_k|\,|x|^{m-1}

For 1≤k≤m1 \le k \le m and ∣x∣≥1|x| \ge 1 we have ∣x∣m−k≤∣x∣m−1|x|^{m-k} \le |x|^{m-1}, so ∣akxm−k∣≤∣ak∣ ∣x∣m−1|a_kx^{m-k}| \le |a_k|\,|x|^{m-1}.

Add them up

By the triangle inequality, ∣a1xm−1+⋯+am∣≤(∣a1∣+⋯+∣am∣)∣x∣m−1=K∣x∣m−1|a_1x^{m-1} + \dots + a_m| \le (|a_1| + \dots + |a_m|)|x|^{m-1} = K|x|^{m-1}.

Compare with the leading term

K∣x∣m−1=K∣a0∣ ∣x∣⋅∣a0∣ ∣x∣mK|x|^{m-1} = \dfrac{K}{|a_0|\,|x|}\cdot|a_0|\,|x|^m, which is the inequality claimed.

The ratio can be made small

If moreover ∣x∣>NK/∣a0∣|x| > NK/|a_0|, then K/(∣a0∣ ∣x∣)<1/NK/(|a_0|\,|x|) < 1/N, so the other terms together are less than 1/N1/N of the leading term.

The bound is crude (Hardy’s own numbers below do far better), but it shows the mechanism: the ratio of the rest to the leading term behaves like a constant divided by xx. The next widget shows it in pictures. On logarithmic axes, where the horizontal coordinate is log⁡10x\log_{10}x and the vertical one log⁡10(size)\log_{10}(\text{size}), each term cxkcx^k is a straight line of slope kk, and the steepest line always wins in the end.

Explore Which term is biggest?

Try this. With Hardy’s x10+30x5+700x4x^{10} + 30x^5 + 700x^4, set x=4x = 4 and read the table. Then choose the three functions x12x^{12}, 106x610^6x^6, 1015x10^{15}x and find the values of xx where the order of size changes.

x10+30x5+700x4x^{10} + 30x^5 + 700x^4
log₁₀ xlog₁₀ (size)
  • x¹⁰ (slope 10)
  • 30x⁵ (slope 5)
  • 700x⁴ (slope 4)
  • the whole polynomial
termvalue at x = 4share of the total
x10x^{10}1.049 × 10⁶83.3%
30x530x^5307202.4%
700x4700x^417920014.2%
The largest term here is x10x^{10}. On these axes c xkc\,x^k is the straight line log⁡c+klog⁡x\log c + k\log x, so the term with the highest power is steepest and is eventually the largest, whatever the coefficients.
What to notice

At x=4x = 4 the term x10x^{10} is already about 83% of the total, as Hardy says. For the second set, 1015x10^{15}x is the largest at x=1x = 1 and x=10x = 10, but the lines cross: x12x^{12} overtakes it near x=23x = 23 and 106x610^6x^6 near x=63x = 63. The coefficients only decide where the crossings are; the powers decide who wins.

Worked examples and exercises

Exercise ●○○ Hardy: Examples XI, 1

Check Hardy’s claims that, at x=4x = 4, x10>1 000 000x^{10} > 1\,000\,000 while 30x5<35 00030x^5 < 35\,000 and 700x4<180 000700x^4 < 180\,000. What fraction of x10+30x5+700x4x^{10} + 30x^5 + 700x^4 is the first term at x=4x = 4, and at x=10x = 10?

Hint
45=10244^5 = 1024, so 410=102424^{10} = 1024^2.

Solution

410=1 048 576>1064^{10} = 1\,048\,576 > 10^6; 30⋅45=30 720<35 00030 \cdot 4^5 = 30\,720 < 35\,000; 700⋅44=700⋅256=179 200<180 000700 \cdot 4^4 = 700 \cdot 256 = 179\,200 < 180\,000. The total is 1 258 4961\,258\,496, of which the first term is about 83.3%83.3\%. At x=10x = 10: 101010^{10}, 3 000 0003\,000\,000 and 7 000 0007\,000\,000, so the first term is 1010/(1.001×1010)≈99.9%10^{10}/(1.001 \times 10^{10}) \approx 99.9\% of the total.

Exercise ●●○ Hardy: Examples XI, 2

Compare the sizes of x12x^{12}, 1 000 000 x61\,000\,000\,x^6 and 1 000 000 000 000 000 x1\,000\,000\,000\,000\,000\,x when x=1x = 1, 1010, 100100. For which positive xx is each the largest?

Hint
Find where each pair is equal: for instance x12=1015xx^{12} = 10^{15}x when x11=1015x^{11} = 10^{15}.

Solution

At x=1x = 1: 11, 10610^6, 101510^{15}. At x=10x = 10: 101210^{12}, 101210^{12}, 101610^{16}. At x=100x = 100: 102410^{24}, 101810^{18}, 101710^{17}.

The pairs are equal at: x12=106x6x^{12} = 10^6x^6 when x=10x = 10; x12=1015xx^{12} = 10^{15}x when x=1015/11≈23.1x = 10^{15/11} \approx 23.1; 106x6=1015x10^6x^6 = 10^{15}x when x=109/5≈63.1x = 10^{9/5} \approx 63.1. So 1015x10^{15}x is the largest for 0<x<1015/110 < x < 10^{15/11}, and x12x^{12} is the largest for x>1015/11x > 10^{15/11}. The middle function 106x610^6x^6 is never the largest: by the time it overtakes 1015x10^{15}x (at x≈63x \approx 63), x12x^{12} has already overtaken both.

Exercise ●○○ Hardy: Examples XI, 3

Show that the graph of y=ax2+2bx+cy = ax^2 + 2bx + c (with a≠0a \neq 0) is the parabola y=ax2y = ax^2 moved so that its vertex is at (−b/a,(ac−b2)/a)\bigl(-b/a, (ac - b^2)/a\bigr).

Hint
Complete the square: ax2+2bx=a(x+b/a)2−b2/aax^2 + 2bx = a(x + b/a)^2 - b^2/a.

Solution

ax2+2bx+c=a(x+ba)2+ac−b2aax^2 + 2bx + c = a\bigl(x + \tfrac ba\bigr)^2 + \dfrac{ac - b^2}{a}. With new coordinates x′=x+b/ax' = x + b/a and y′=y−(ac−b2)/ay' = y - (ac - b^2)/a, measured from axes through (−b/a,(ac−b2)/a)\bigl(-b/a, (ac - b^2)/a\bigr) parallel to the old ones, the equation becomes y′=ax′2y' = ax'^2. Moving the axes does not change the shape of the curve, so the graph is a copy of y=ax2y = ax^2 with its vertex at that point; it opens upwards if a>0a > 0 and downwards if a<0a < 0.

Exercise ●●○ Hardy: Examples XI, 4

Sketch y=x3−3x+1y = x^3 - 3x + 1, y=x2(x−1)y = x^2(x - 1) and y=x(x−1)2y = x(x - 1)^2. How many times does each meet the axis of xx, and how does each behave there?

Hint
Tabulate the first at x=−2,−1,0,1,2x = -2, -1, 0, 1, 2 and look for changes of sign.
Hint
For the other two, factors that appear squared do not change sign.

Solution

x3−3x+1x^3 - 3x + 1 takes the values −1,3,1,−1,3-1, 3, 1, -1, 3 at x=−2,−1,0,1,2x = -2, -1, 0, 1, 2. It changes sign in (−2,−1)(-2, -1), (0,1)(0, 1) and (1,2)(1, 2), so the curve crosses the axis at least three times; a cubic cannot vanish more than three times (Chapter I, or any algebra text), so exactly three. It rises to the left of −1-1, falls between −1-1 and 11, and rises after 11 (you will prove this with derivatives in Chapter VI). For large ∣x∣|x| it behaves like x3x^3.

x2(x−1)x^2(x - 1) is ≤0\le 0 for x<1x < 1 (apart from 00 at x=0x = 0) and positive for x>1x > 1. At x=0x = 0 the curve touches the axis from below without crossing, because of the square x2x^2; at x=1x = 1 it crosses.

x(x−1)2x(x - 1)^2 is negative for x<0x < 0 and ≥0\ge 0 for x>0x > 0. It crosses at x=0x = 0 and touches the axis from above at x=1x = 1. It is the first curve turned through two right angles and moved: x(x−1)2=−[(1−x)2((1−x)−1)]x(x-1)^2 = -\bigl[(1-x)^2((1 - x) - 1)\bigr], the reflection of x2(x−1)x^2(x-1) in the point (12,0)(\tfrac12, 0).

Check yourself

Check yourself True or false?

If we plot enough points of y=x2y = x^2 and they all lie near a smooth curve, we have proved that the graph is a continuous curve.

Why

False, and this is Hardy’s point in §23. Finitely many points say nothing about the infinitely many others between them; the graph of the function that is x2x^2 at rational points and 00 elsewhere agrees with x2x^2 at every point you could plot with a ruler at rational coordinates. Continuity needs the definitions of Chapter V.

Check yourself Enter a number

On y=x2y = x^2 take x0=−1x_0 = -1, x1=3x_1 = 3. How far above the curve is the midpoint of the chord?

Why

With λ=μ=12\lambda = \mu = \tfrac12 the gap is 14(3−(−1))2=4\tfrac14(3 - (-1))^2 = 4. Directly: the chord’s midpoint is at height 12(1+9)=5\tfrac12(1 + 9) = 5 above x=1x = 1, where the curve has height 11.

Check yourself Choose one

For odd mm, the graph of y=xmy = x^m is symmetric

Why

(−x)m=−xm(-x)^m = -x^m for odd mm, so with (x,y)(x, y) the point (−x,−y)(-x, -y) is on the graph: that is symmetry about the origin. Symmetry about the axis of yy is the even case. Symmetry about y=xy = x would need the function to be its own inverse, which is true only for m=1m = 1.

Check yourself Choose one

At x=10x = 10, which of x12x^{12}, 106x610^6x^6, 1015x10^{15}x is the largest?

Why

The values are 101210^{12}, 101210^{12} and 101610^{16}. The highest power wins only eventually, here for x>1015/11≈23.1x > 10^{15/11} \approx 23.1. The first two are equal at x=10x = 10, which may explain the last option.

Hardy’s language

It is the first term which is of really preponderant importance when x is fairly large.

— G. H. Hardy, §23

Summary

Summary
  • A polynomial is a0xm+⋯+ama_0x^m + \dots + a_m; every polynomial graph is built from the powers by scaling and adding ordinates.
  • xmx^m is symmetric about the axis of yy for even mm and about the origin for odd mm; larger mm hugs the axis between −1-1 and 11 and climbs faster outside.
  • Every chord of y=x2y = x^2 lies above the curve: the gap is exactly λμ(x1−x0)2\lambda\mu(x_1 - x_0)^2.
  • For ∣x∣≥1|x| \ge 1 the lower terms of a polynomial are at most K/(∣a0∣ ∣x∣)K/(|a_0|\,|x|) times the leading term, so the highest power dominates for large xx.
  • Plotting points suggests but never proves the shape of a curve; continuity is defined in Chapter V.