Chapter VII · Additional theorems in the differential and integral calculus

Differentials

§160 30 min

How the symbols dxdx, dydy, dzdz get an exact meaning, why the formula dz=zx dx+zy dydz = z_x\,dx + z_y\,dy holds whether or not xx and yy are independent, and how that is used.

Why this matters

So far dy/dxdy/dx has been a single symbol: the derivative, not a quotient. Yet scientists and geometers constantly write things like dA=π(b da+a db)dA = \pi(b\,da + a\,db) or x dx+y dy=0x\,dx + y\,dy = 0 and manipulate the pieces separately. §160 shows how to give the isolated symbols dxdx, dydy, dzdz a precise meaning, so that the approximate equation δz≈fx δx+fy δy\delta z \approx f_x\,\delta x + f_y\,\delta y of the last lesson becomes an exact equation between differentials. Its great advantage, which Hardy singles out, is that the formula does not care whether xx and yy are independent variables or functions of something else.

The idea

Take the area z=xyz = xy of a rectangle and increase its sides by δx\delta x and δy\delta y. The new area exceeds the old by two strips, y δxy\,\delta x and x δyx\,\delta y, plus a small corner δx δy\delta x\,\delta y. The strips are linear in the increments; the corner is of second order. The differential dzdz is, by definition, the linear part: dz=y dx+x dydz = y\,dx + x\,dy.

Explore Strips and a corner

Try this. Press “Halve both increments” several times and watch the two strips and the corner, and the numbers in the read-out.

z=xyz = xy: δz=y δx+x δy+δx δy\delta z = y\,\delta x + x\,\delta y + \delta x\,\delta y
xy
  • original area xy
  • the differential dz = y dx + x dy (two strips)
  • the error δz − dz = δx δy (corner)
δz=2.2\delta z = 2.2, dz=y dx+x dy=1.96dz = y\,dx + x\,dy = 1.96, error δz−dz=δx δy=0.24\delta z - dz = \delta x\,\delta y = 0.24.
Error compared with the larger step: δx δymax⁡(δx,δy)=min⁡(δx,δy)=0.4\dfrac{\delta x\,\delta y}{\max(\delta x, \delta y)} = \min(\delta x, \delta y) = 0.4, which tends to 0 with the step. Relative change: dzz=dxx+dyy=0.49\dfrac{dz}{z} = \dfrac{dx}{x} + \dfrac{dy}{y} = 0.49.
What to notice

Each halving halves the strips but quarters the corner. The differential dzdz (the strips) captures the change to first order; the error δx δy\delta x\,\delta y becomes negligible compared with the step. The relative form dz/z=dx/x+dy/ydz/z = dx/x + dy/y says: percentage errors in the sides add up to the percentage error in the area.

Precise statements

For one variable, y=f(x)y = f(x), we know that δy={f′(x)+ε} δx\delta y = \{f'(x) + \varepsilon\}\,\delta x with ε→0\varepsilon \to 0, so δy≈f′(x) δx\delta y \approx f'(x)\,\delta x.

Definition Differentials of a function of one variable Hardy: §160 (3)–(6)

Let y=f(x)y = f(x) be differentiable. For the independent variable, the differential dxdx is an arbitrary increment, dx=δxdx = \delta x. The differential of yy is defined by

dy=f′(x) δx.dy = f'(x)\,\delta x.

Taking y=xy = x gives dx=δxdx = \delta x, consistently, so dy=f′(x) dxdy = f'(x)\,dx, and dividing, dy/dx=f′(x)dy/dx = f'(x): the derivative is the quotient of the differentials.

So dy/dxdy/dx now has two meanings, the derivative and the quotient of two differentials, and they agree. Note that dydy is not δy\delta y: it is the change along the tangent line, while δy\delta y is the change along the curve.

Definition The total differential Hardy: §160 (7)–(9)

Let z=f(x,y)z = f(x, y), with xx and yy independent and fxf_x, fyf_y continuous. Its differential is

dz=fx δx+fy δy.dz = f_x\,\delta x + f_y\,\delta y.

Since dx=δxdx = \delta x and dy=δydy = \delta y (apply the definition to z=xz = x and z=yz = y), this reads

dz=∂z∂x dx+∂z∂y dy,dz = \frac{\partial z}{\partial x}\,dx + \frac{\partial z}{\partial y}\,dy,

an exact equation corresponding to the approximate equation δz≈fx δx+fy δy\delta z \approx f_x\,\delta x + f_y\,\delta y of §159.

Theorem Invariance of the differential Hardy: §160

The formula dz=zx dx+zy dydz = z_x\,dx + z_y\,dy holds also when xx and yy are not independent but are functions x=ϕ(t)x = \phi(t), y=ψ(t)y = \psi(t) of one variable tt (with continuous derivatives), the differentials then being dx=ϕ′(t) dtdx = \phi'(t)\,dt, dy=ψ′(t) dtdy = \psi'(t)\,dt and dz=dzdt dtdz = \frac{dz}{dt}\,dt.

Theorem Reading off partial derivatives Hardy: §160

If xx and yy are independent and dz=λ dx+μ dydz = \lambda\,dx + \mu\,dy for all dxdx, dydy, then λ=∂z/∂x\lambda = \partial z/\partial x and μ=∂z/∂y\mu = \partial z/\partial y.

The same definitions and results hold for any number of variables: du=ux dx+uy dy+uz dzdu = u_x\,dx + u_y\,dy + u_z\,dz, and so on.

Proofs

Proof Invariance of the differential
Strategy

Compute dzdz in the only way available when zz is a function of the single variable tt, namely dz=dzdt dtdz = \frac{dz}{dt}\,dt, and expand dz/dtdz/dt with the chain rule of §157.

zz as a function of tt

Now z=f{ϕ(t),ψ(t)}z = f\{\phi(t), \psi(t)\} is a function of tt alone, so by the one-variable definition dz=dzdt dtdz = \dfrac{dz}{dt}\,dt.

Apply the chain rule

Since fxf_x and fyf_y are continuous, dzdt=fx ϕ′(t)+fy ψ′(t)\dfrac{dz}{dt} = f_x\,\phi'(t) + f_y\,\psi'(t).

Multiply by dtdt

dz=fx ϕ′(t) dt+fy ψ′(t) dt=fx dx+fy dydz = f_x\,\phi'(t)\,dt + f_y\,\psi'(t)\,dt = f_x\,dx + f_y\,dy, because dx=ϕ′(t) dtdx = \phi'(t)\,dt and dy=ψ′(t) dtdy = \psi'(t)\,dt by the one-variable definition. This is the same formula as for independent xx and yy.

Proof Reading off partial derivatives
Strategy

Since dxdx and dydy are arbitrary for independent variables, we may choose one of them to be zero.

Compare with the definition

We have fx dx+fy dy=λ dx+μ dyf_x\,dx + f_y\,dy = \lambda\,dx + \mu\,dy for all dxdx, dydy.

Choose dy=0dy = 0, then dx=0dx = 0

With dy=0dy = 0 and dx=1dx = 1: λ=fx\lambda = f_x. With dx=0dx = 0 and dy=1dy = 1: μ=fy\mu = f_y. (Hardy derives it from §159 instead; the content is the same uniqueness.)

The invariance is what makes the notation so efficient. To differentiate f(x,y)=0f(x, y) = 0 implicitly, just write fx dx+fy dy=0f_x\,dx + f_y\,dy = 0 (valid although yy now depends on xx) and divide: dy/dx=−fx/fydy/dx = -f_x/f_y whenever fy≠0f_y \ne 0. The tangent to the curve f(x,y)=0f(x, y) = 0 at (x0,y0)(x_0, y_0) is therefore (x−x0)fx+(y−y0)fy=0(x - x_0)f_x + (y - y_0)f_y = 0 (Examples LXII, 9–10). And at a maximum of f(x,y)f(x, y) on a curve g(x,y)=0g(x, y) = 0 we have df=0df = 0 for every allowed displacement, which forces fx/gx=fy/gyf_x/g_x = f_y/g_y (Examples LXII, 13), the germ of the method of Lagrange multipliers.

Worked examples and exercises

Exercise Area of an ellipse ●○○ Hardy: Examples LXII, 1

The area of an ellipse with semi-axes aa and bb is A=πabA = \pi ab. Prove that dAA=daa+dbb\dfrac{dA}{A} = \dfrac{da}{a} + \dfrac{db}{b}, and estimate the change of area when aa increases by 1%1\% and bb decreases by 2%2\%.

Hint
dA=π(b da+a db)dA = \pi(b\,da + a\,db). Divide by AA.

Solution

dA=Aa da+Ab db=πb da+πa dbdA = A_a\,da + A_b\,db = \pi b\,da + \pi a\,db, and dividing by A=πabA = \pi ab gives dA/A=da/a+db/bdA/A = da/a + db/b. With da/a=0.01da/a = 0.01 and db/b=−0.02db/b = -0.02, dA/A=−0.01dA/A = -0.01: the area decreases by about 1%1\%. (Exactly: 1.01×0.98=0.98981.01 \times 0.98 = 0.9898, a decrease of 1.02%1.02\%; the difference is the product of the two small changes, as in the widget’s corner.)

Exercise The chain rule with differentials ●○○ Hardy: Examples LXII, 5

Let zz be a function of uu and vv, which are functions of xx and yy. Show that ∂z∂x=∂z∂u∂u∂x+∂z∂v∂v∂x\dfrac{\partial z}{\partial x} = \dfrac{\partial z}{\partial u}\dfrac{\partial u}{\partial x} + \dfrac{\partial z}{\partial v}\dfrac{\partial v}{\partial x}, and similarly for yy.

Hint
Write dz=zu du+zv dvdz = z_u\,du + z_v\,dv (valid although u,vu, v are not independent), then substitute du=ux dx+uy dydu = u_x\,dx + u_y\,dy and dv=vx dx+vy dydv = v_x\,dx + v_y\,dy.

Solution

By invariance, dz=zu du+zv dv=zu(ux dx+uy dy)+zv(vx dx+vy dy)=(zuux+zvvx) dx+(zuuy+zvvy) dydz = z_u\,du + z_v\,dv = z_u(u_x\,dx + u_y\,dy) + z_v(v_x\,dx + v_y\,dy) = (z_u u_x + z_v v_x)\,dx + (z_u u_y + z_v v_y)\,dy. Since xx and yy are independent, the coefficients of dxdx and dydy are zxz_x and zyz_y, by the theorem on reading off partial derivatives.

Exercise Implicit differentiation and tangents ●○○ Hardy: Examples LXII, 9–10

Find dy/dxdy/dx on the circle x2+y2=25x^2 + y^2 = 25 and on the folium x3+y3=3xyx^3 + y^3 = 3xy, and the tangent to the circle at (3,4)(3, 4).

Hint
Take the differential of each side. For the folium, collect the terms in dydy.

Solution

Circle: 2x dx+2y dy=02x\,dx + 2y\,dy = 0, so dy/dx=−x/ydy/dx = -x/y; at (3,4)(3, 4) the slope is −34-\tfrac34 and the tangent is 3(x−3)+4(y−4)=03(x - 3) + 4(y - 4) = 0, that is 3x+4y=253x + 4y = 25.

Folium: 3x2 dx+3y2 dy=3y dx+3x dy3x^2\,dx + 3y^2\,dy = 3y\,dx + 3x\,dy, so (y2−x) dy=(y−x2) dx(y^2 - x)\,dy = (y - x^2)\,dx and dy/dx=(y−x2)/(y2−x)dy/dx = (y - x^2)/(y^2 - x), wherever y2≠xy^2 \ne x.

Exercise Triangles of constant area ●●● Hardy: Examples LXII, 3

The sides a,b,ca, b, c of a triangle vary so that its area Δ\Delta stays constant; regard aa as a function of bb and cc. Prove that ∂a∂b=−cos⁡Bcos⁡A\dfrac{\partial a}{\partial b} = -\dfrac{\cos B}{\cos A}.

Hint
Use Δ=12bcsin⁡A\Delta = \tfrac12 bc\sin A and a2=b2+c2−2bccos⁡Aa^2 = b^2 + c^2 - 2bc\cos A, with cc fixed and dΔ=0d\Delta = 0. At the end use the projection formula c=acos⁡B+bcos⁡Ac = a\cos B + b\cos A.

Solution

Keep cc fixed. From dΔ=0d\Delta = 0: 12c(sin⁡A db+bcos⁡A dA)=0\tfrac12 c(\sin A\,db + b\cos A\,dA) = 0, so dA=−sin⁡Abcos⁡A dbdA = -\dfrac{\sin A}{b\cos A}\,db. From the cosine rule:

2a da=2b db−2ccos⁡A db+2bcsin⁡A dA=2 db(b−ccos⁡A−csin⁡2Acos⁡A)=2 db bcos⁡A−ccos⁡A.2a\,da = 2b\,db - 2c\cos A\,db + 2bc\sin A\,dA = 2\,db\Bigl(b - c\cos A - \frac{c\sin^2 A}{\cos A}\Bigr) = 2\,db\,\frac{b\cos A - c}{\cos A}.

By the projection formula bcos⁡A−c=−acos⁡Bb\cos A - c = -a\cos B, so a da=−acos⁡B db/cos⁡Aa\,da = -a\cos B\,db/\cos A, that is ∂a/∂b=−cos⁡B/cos⁡A\partial a/\partial b = -\cos B/\cos A. By symmetry ∂a/∂c=−cos⁡C/cos⁡A\partial a/\partial c = -\cos C/\cos A.

Exercise A constrained maximum ●●○ Hardy: Examples LXII, 12–13

Find the largest value of f=xyf = xy on the line g=x+y−1=0g = x + y - 1 = 0, first by substitution, then by the condition fx/gx=fy/gyf_x/g_x = f_y/g_y.

Hint
At the maximum, df=y dx+x dy=0df = y\,dx + x\,dy = 0 for every displacement along the line, where dx+dy=0dx + dy = 0.

Solution

Substitution: f=x(1−x)f = x(1 - x), maximal at x=12x = \tfrac12, value 14\tfrac14.

Differentials: on the line dy=−dxdy = -dx, so df=(y−x) dxdf = (y - x)\,dx, which vanishes for all dxdx only if x=yx = y. That is fx/gx=y/1=x/1=fy/gyf_x/g_x = y/1 = x/1 = f_y/g_y. With x+y=1x + y = 1 we again get x=y=12x = y = \tfrac12. The method needs no solving for yy, which is its point when the constraint is complicated.

Check yourself

Check yourself Enter a number

A rectangle has sides x=3x = 3, y=2y = 2. They increase by δx=0.1\delta x = 0.1 and δy=0.05\delta y = 0.05. What is δz−dz\delta z - dz for the area z=xyz = xy?

Why

dz=y dx+x dy=0.2+0.15=0.35dz = y\,dx + x\,dy = 0.2 + 0.15 = 0.35, while δz=3.1×2.05−6=0.355\delta z = 3.1 \times 2.05 - 6 = 0.355. The difference is δx δy=0.005\delta x\,\delta y = 0.005, the corner of the picture.

Check yourself True or false?

If z=f(x,y)z = f(x, y) and xx, yy are both functions of tt, the formula dz=fx dx+fy dydz = f_x\,dx + f_y\,dy is still correct (with dx=x′(t) dtdx = x'(t)\,dt, dy=y′(t) dtdy = y'(t)\,dt).

Why

True; this is the invariance Hardy stresses. Dividing by dtdt gives exactly the chain rule of §157. What changes is only that dxdx and dydy are no longer arbitrary: they are tied to dtdt.

Check yourself Choose one

For y=x2y = x^2 at x=1x = 1 with δx=0.1\delta x = 0.1, which is correct?

Why

dy=f′(1) δx=2×0.1=0.2dy = f'(1)\,\delta x = 2 \times 0.1 = 0.2 is the change along the tangent; δy=1.12−1=0.21\delta y = 1.1^2 - 1 = 0.21 is the change along the curve. In Hardy’s treatment a differential is an ordinary finite number, not an “infinitely small quantity”; what is small is the difference δy−dy=(δx)2\delta y - dy = (\delta x)^2 compared with δx\delta x.

Check yourself Enter a number

On the circle x2+y2=25x^2 + y^2 = 25, what is dy/dxdy/dx at the point (3,4)(3, 4)?

Why

2x dx+2y dy=02x\,dx + 2y\,dy = 0 gives dy/dx=−x/y=−3/4dy/dx = -x/y = -3/4.

Hardy’s language

the formula which expresses dz in terms of dx and dy is the same whether the variables x and y are independent or not.

— G. H. Hardy, §160

Summary

Summary
  • For independent variables, dx=δxdx = \delta x is an arbitrary increment; for y=f(x)y = f(x), dy=f′(x) dxdy = f'(x)\,dx, so dy/dxdy/dx is a genuine quotient equal to the derivative.
  • For z=f(x,y)z = f(x, y): dz=zx dx+zy dydz = z_x\,dx + z_y\,dy, the linear part of δz\delta z; the error δz−dz\delta z - dz is small compared with the step.
  • The formula holds whether or not xx and yy are independent; that is what makes differentials convenient for implicit differentiation, changes of variable and constrained maxima.
  • If dz=λ dx+μ dydz = \lambda\,dx + \mu\,dy for independent x,yx, y, then λ=zx\lambda = z_x and μ=zy\mu = z_y.