The math optional, made finite. Daily Practice

Partial derivatives; equality of mixed partials (Schwarz)

At a Glance

Why This Chapter Matters

Two of the most mark-dense questions in recent Paper 2 Real Analysis sit here: a 15-marker on piecewise mixed partials (2014) and a 20-marker on Euler’s theorem for homogeneous functions (2020). Both are conceptually sharp — one tests whether you can compute fxyfyxf_{xy} \ne f_{yx} via limits and diagnose the discontinuity, the other tests whether you know to apply Euler’s theorem to tanu\tan u rather than uu itself. These patterns are well-established exam favourites and reward specific preparation.

Minimum Theory

Schwarz (Clairaut) theorem. If fxyf_{xy} and fyxf_{yx} both exist and are continuous at a point (a,b)(a,b), then fxy(a,b)=fyx(a,b)f_{xy}(a,b) = f_{yx}(a,b). The converse fails: if the mixed partials are unequal at (a,b)(a,b), at least one must be discontinuous there.

Mixed partials at a singular point (piecewise function). When ff is given by a formula for (x,y)(0,0)(x,y) \ne (0,0) and a separate value at (0,0)(0,0), the standard recipe is: (1) compute fx(0,y)f_x(0,y) and fy(x,0)f_y(x,0) by differentiating the off-origin formula and setting the remaining variable to 00; (2) compute fxy(0,0)f_{xy}(0,0) as limk0[fx(0,k)fx(0,0)]/k\lim_{k \to 0}[f_x(0,k)-f_x(0,0)]/k, and fyx(0,0)f_{yx}(0,0) similarly. Never use the product/quotient rule directly at (0,0)(0,0) for a piecewise function.

Euler’s theorem for homogeneous functions. If f(x,y)f(x,y) is homogeneous of degree nn (i.e., f(tx,ty)=tnf(x,y)f(tx,ty) = t^n f(x,y)), then xfx+yfy=nfx f_x + y f_y = n f. The second-order extension: differentiating the first-order relation with respect to xx and yy separately and combining (multiply /x\partial/\partial x relation by xx, /y\partial/\partial y relation by yy, add) gives x2fxx+2xyfxy+y2fyy=n(n1)fx^2 f_{xx} + 2xy f_{xy} + y^2 f_{yy} = n(n-1)f.

Mixed partials, Schwarz theorem, and Euler's theorem

Question Archetypes

ArchetypeRecognition
mixed-partialsPiecewise ff with formula for (x,y)(0,0)(x,y)\ne(0,0), value 00 at origin; compute fxyf_{xy} and fyxf_{yx} at (0,0)(0,0) and discuss continuity
euler-theoremu=arctan(g(x,y))u = \arctan(g(x,y)) or u=sin1()u = \sin^{-1}(\cdot) where tanu\tan u or sinu\sin u is homogeneous; prove a second-order PDE identity

euler-theorem (1 question(s); 2020)

Recognition Cues — The function uu involves an inverse trig (arctan, arcsin) of a rational function. The question asks you to prove an identity of the form x2uxx+2xyuxy+y2uyy=(something in u)x^2 u_{xx} + 2xy u_{xy} + y^2 u_{yy} = (\text{something in } u). The key is that uu itself is not homogeneous, but tanu\tan u (or sinu\sin u) is — verify this by checking the scaling degrees.

Solution Template

  1. Let z=tanuz = \tan u (or sinu\sin u, etc.). Write out zz explicitly as a ratio of polynomials.
  2. Verify homogeneity of zz. Replace (x,y)(x,y) with (tx,ty)(tx, ty) and read off the degree nn (typically nn = numerator degree minus denominator degree).
  3. Apply first-order Euler: xzx+yzy=nzx z_x + y z_y = nz. Convert to uu using zx=sec2uuxz_x = \sec^2 u \cdot u_x (or zx=cosuuxz_x = \cos u \cdot u_x for z=sinuz = \sin u).
  4. Simplify to get xux+yuy=g(u)x u_x + y u_y = g(u) for a simple trig expression gg.
  5. Differentiate the first-order relation. Apply /x\partial/\partial x to both sides, multiply by xx; apply /y\partial/\partial y, multiply by yy; add.
  6. Use xux+yuy=g(u)x u_x + y u_y = g(u) to simplify. This yields x2uxx+2xyuxy+y2uyy=g(u)(g(u)1)x^2 u_{xx} + 2xy u_{xy} + y^2 u_{yy} = g(u)(g'(u)-1).
  7. Substitute and simplify with trig identities to reach the target form.

Worked Example

2020 Paper 2, 2020-P2-Q3c (20 marks)

If u=tan1x3+y3xyu = \tan^{-1}\dfrac{x^3+y^3}{x-y}, xyx \ne y, show that x2uxx+2xyuxy+y2uyy=(14sin2u)sin2ux^2 u_{xx} + 2xy\,u_{xy} + y^2 u_{yy} = (1-4\sin^2 u)\sin 2u.

Step 1 — Set z=tanuz = \tan u. Then z=x3+y3xyz = \dfrac{x^3+y^3}{x-y}.

Step 2 — Verify homogeneity. Replacing (x,y)(tx,ty)(x,y) \to (tx,ty): numerator t3(x3+y3)\to t^3(x^3+y^3), denominator t(xy)\to t(x-y). So z(tx,ty)=t2z(x,y)z(tx,ty) = t^2 z(x,y). Degree n=2n = 2.

Step 3 — First-order Euler. By Euler’s theorem, xzx+yzy=2zx z_x + y z_y = 2z. Since z=tanuz = \tan u, we have zx=sec2uuxz_x = \sec^2 u \cdot u_x and zy=sec2uuyz_y = \sec^2 u \cdot u_y. Substituting:

sec2u(xux+yuy)=2tanu.\sec^2 u \,(x u_x + y u_y) = 2\tan u.

xux+yuy=2tanusec2u=2sinucosu=sin2u.(1)x u_x + y u_y = \frac{2\tan u}{\sec^2 u} = 2\sin u \cos u = \sin 2u. \qquad(1)

Denote g(u)=sin2ug(u) = \sin 2u.

Step 4 — Differentiate (1) with respect to xx, multiply by xx.

ux+xuxx+yuyx=g(u)ux.u_x + x u_{xx} + y u_{yx} = g'(u) u_x.

xux+x2uxx+xyuyx=g(u)xux.(2)x u_x + x^2 u_{xx} + xy\,u_{yx} = g'(u)\, x u_x. \qquad(2)

Step 5 — Differentiate (1) with respect to yy, multiply by yy.

xuxy+uy+yuyy=g(u)uy.x u_{xy} + u_y + y u_{yy} = g'(u) u_y.

xyuxy+yuy+y2uyy=g(u)yuy.(3)xy\,u_{xy} + y u_y + y^2 u_{yy} = g'(u)\, y u_y. \qquad(3)

Step 6 — Add (2) and (3). Using uxy=uyxu_{xy} = u_{yx}:

(xux+yuy)g(u)+(x2uxx+2xyuxy+y2uyy)=g(u)(xux+yuy)g(u).\underbrace{(x u_x + y u_y)}_{g(u)} + (x^2 u_{xx} + 2xy\,u_{xy} + y^2 u_{yy}) = g'(u)\underbrace{(x u_x + y u_y)}_{g(u)}.

x2uxx+2xyuxy+y2uyy=g(u)(g(u)1).x^2 u_{xx} + 2xy\,u_{xy} + y^2 u_{yy} = g(u)(g'(u)-1).

Step 7 — Evaluate. g=sin2ug = \sin 2u, g=2cos2ug' = 2\cos 2u. So g(u)1=2cos2u1=2(12sin2u)1=14sin2ug'(u)-1 = 2\cos 2u - 1 = 2(1-2\sin^2 u)-1 = 1-4\sin^2 u. Therefore:

x2uxx+2xyuxy+y2uyy=sin2u(14sin2u).x^2 u_{xx} + 2xy\,u_{xy} + y^2 u_{yy} = \sin 2u \cdot (1-4\sin^2 u).

x2uxx+2xyuxy+y2uyy=(14sin2u)sin2u.\boxed{x^2 u_{xx} + 2xy\,u_{xy} + y^2 u_{yy} = (1-4\sin^2 u)\sin 2u.}

Common Traps


mixed-partials (1 question(s); 2014)

Recognition Cues — A function defined piecewise: one formula for (x,y)(0,0)(x,y) \ne (0,0), typically f=xyh(x,y)/(x2+y2)f = xy \cdot h(x,y)/(x^2+y^2) for some polynomial hh, and f(0,0)=0f(0,0) = 0. The question asks for fxy(0,0)f_{xy}(0,0) and fyx(0,0)f_{yx}(0,0) and discusses their continuity. The factor xyxy in the numerator is the key shortcut: ff and its first partials vanish on the coordinate axes.

Solution Template

  1. Use the xyxy factor. Note f(h,0)=0f(h,0) = 0 for all hh and f(0,k)=0f(0,k) = 0 for all kk (from the explicit formula).
  2. Compute fx(0,0)f_x(0,0) and fy(0,0)f_y(0,0) by limit definition. Both are 00 (from step 1).
  3. Compute fx(0,y)f_x(0,y) for y0y \ne 0. Differentiate the off-origin formula with respect to xx, then set x=0x = 0.
  4. Compute fxy(0,0)f_{xy}(0,0) by limit definition: fxy(0,0)=limk0[fx(0,k)fx(0,0)]/kf_{xy}(0,0) = \lim_{k\to 0}[f_x(0,k)-f_x(0,0)]/k.
  5. Compute fy(x,0)f_y(x,0) for x0x \ne 0. Differentiate with respect to yy, set y=0y=0.
  6. Compute fyx(0,0)f_{yx}(0,0) similarly.
  7. Compare fxyf_{xy} and fyxf_{yx}. If they differ, cite Schwarz: at least one mixed partial is discontinuous at (0,0)(0,0). Show the discontinuity using different approach paths.

Worked Example

2014 Paper 2, 2014-P2-Q3b (15 marks)

Obtain 2f(0,0)xy\dfrac{\partial^2 f(0,0)}{\partial x\,\partial y} and 2f(0,0)yx\dfrac{\partial^2 f(0,0)}{\partial y\,\partial x} for f(x,y)={xy(3x22y2)x2+y2,(x,y)(0,0)0,(x,y)=(0,0).f(x,y) = \begin{cases}\dfrac{xy(3x^2-2y^2)}{x^2+y^2}, & (x,y)\ne(0,0) \\ 0, & (x,y)=(0,0)\end{cases}. Also discuss the continuity of fxyf_{xy} and fyxf_{yx} at (0,0)(0,0).

Step 1 — First partials at origin. Since f(h,0)=0f(h,0) = 0 for all hh and f(0,k)=0f(0,k) = 0 for all kk (the formula gives xy()/(x2+y2)xy(\ldots)/(x^2+y^2) which vanishes on the axes):

fx(0,0)=limh0f(h,0)0h=0,fy(0,0)=0.f_x(0,0) = \lim_{h\to 0}\frac{f(h,0)-0}{h} = 0, \qquad f_y(0,0) = 0.

Step 2 — fx(0,y)f_x(0,y) for y0y \ne 0. Differentiate the off-origin formula with respect to xx and evaluate at x=0x=0:

fx(x,y)=y(9x22y2)x2+y22x2y(3x22y2)(x2+y2)2.f_x(x,y) = \frac{y(9x^2-2y^2)}{x^2+y^2} - \frac{2x^2 y(3x^2-2y^2)}{(x^2+y^2)^2}.

At x=0x = 0, y0y \ne 0: fx(0,y)=y(2y2)y2=2yf_x(0,y) = \dfrac{y \cdot (-2y^2)}{y^2} = -2y.

Step 3 — Compute fxy(0,0)f_{xy}(0,0).

fxy(0,0)=limk0fx(0,k)fx(0,0)k=limk02k0k=2.f_{xy}(0,0) = \lim_{k\to 0}\frac{f_x(0,k)-f_x(0,0)}{k} = \lim_{k\to 0}\frac{-2k - 0}{k} = -2.

Step 4 — fy(x,0)f_y(x,0) for x0x \ne 0. Similarly differentiate with respect to yy and set y=0y = 0:

fy(x,0)=3x.f_y(x,0) = 3x.

Step 5 — Compute fyx(0,0)f_{yx}(0,0).

fyx(0,0)=limh0fy(h,0)fy(0,0)h=limh03h0h=3.f_{yx}(0,0) = \lim_{h\to 0}\frac{f_y(h,0)-f_y(0,0)}{h} = \lim_{h\to 0}\frac{3h-0}{h} = 3.

fxy(0,0)=2,fyx(0,0)=3.\boxed{f_{xy}(0,0) = -2, \qquad f_{yx}(0,0) = 3.}

Step 6 — Continuity discussion. Since fxy(0,0)fyx(0,0)f_{xy}(0,0) \ne f_{yx}(0,0), Schwarz’s theorem requires at least one to be discontinuous at (0,0)(0,0). In fact, both are.

Away from (0,0)(0,0), fCf \in C^\infty so fxy=fyxf_{xy} = f_{yx}; call this common function g(x,y)g(x,y).

The limits of gg along these two paths are 33 and 2-2 respectively, hence different. Therefore lim(x,y)(0,0)g(x,y)\lim_{(x,y)\to(0,0)} g(x,y) does not exist, and both fxyf_{xy} and fyxf_{yx} are discontinuous at (0,0)(0,0).

Both fxy and fyx are discontinuous at (0,0).\boxed{\text{Both } f_{xy} \text{ and } f_{yx} \text{ are discontinuous at }(0,0).}

Common Traps

Marks-Aware Writing

A 15-mark answer for the mixed-partials question requires all of: (1) compute first partials at origin by limit definition (no points for using the formula), (2) compute fx(0,y)f_x(0,y) by differentiating off-origin and setting x=0x=0, (3) compute fxy(0,0)f_{xy}(0,0) as a limit, (4) same for fyx(0,0)f_{yx}(0,0) via fy(x,0)f_y(x,0), (5) state the values, (6) discuss continuity with path-dependent limits. Omitting any of steps 1–5 costs 2–3 marks each.

A 20-mark Euler answer requires: (1) identify the homogeneous function z=tanuz = \tan u and verify degree, (2) state and apply first-order Euler to get xux+yuy=sin2uxu_x + yu_y = \sin 2u, (3) differentiate this relation twice and add, (4) substitute g(u)(g(u)1)g(u)(g'(u)-1), (5) trig simplification of 2cos2u1=14sin2u2\cos 2u - 1 = 1-4\sin^2 u. All five algebraic steps must be visible for full marks.

Practice Set

We've mapped all 13 years of this exam. Get new chapters, tools, and solutions as we release them — free.