Partial derivatives
A function of two variables can change in every direction at once. How fast does it change when only one input moves, and does the order of two such derivatives matter?
Where you are 8 in Partial derivatives
Why this exists
The derivative of The derivative at a point measures the rate of change of the output when the one input moves. A function of two variables (Functions of several variables and level sets) can have its input moved in any direction of the plane. The simplest choice is to move one input and freeze the other: what is left is a function of one variable, and all of one-variable calculus applies to it.
Amara and Samir run a blacksmith’s forge. Amara’s thermal camera reads T(x, y), the temperature in degrees Fahrenheit of a steel plate at x inches along its long edge and y inches along its short edge. A step along each edge changes the temperature at its own rate: the partial derivatives T_x and T_y (Example 3).
Definition
A point (a, b) is an interior point of D \subseteq \mathbb{R}^2 if some open disk B_r(a, b) of Limits and continuity in several variables lies inside D. A set all of whose points are interior points is open.
Partial derivatives. Let f : D \to \mathbb{R} and let (a, b) be an interior point of D. The partial derivatives of f with respect to x and to y at (a, b) are
each defined when its limit exists. Both are one-variable limits: with g(x) = f(x, b) and G(y) = f(a, y), f_x(a, b) = g'(a) and f_y(a, b) = G'(b). Other notations are \dfrac{\partial f}{\partial x}(a, b), and for z = f(x, y) also \dfrac{\partial z}{\partial x} and z_x. Where f_x exists at every point of a set, it is a function there, the partial derivative function f_x; the same for f_y. For f of n variables (Vectors in n dimensions), \dfrac{\partial f}{\partial x_i} is defined the same way, moving only the ith input.
Second-order partial derivatives. The partial derivatives of f_x and f_y, where they exist, are the four second-order partial derivatives
Subscripts are read left to right, in the order the derivatives are taken: f_{xy} is first x, then y. Leibniz notation is read from the function outward, right to left:
Also \displaystyle f_{xx} = \frac{\partial^2 f}{\partial x^2}. f_{xy} and f_{yx} are the mixed partials; higher orders follow the same pattern, f_{xyy} = (f_{xy})_y. A function is of class C^k on an open set U, for an integer k \ge 1, if all its partial derivatives of order at most k exist and are continuous on U.
Intuition
The plane y = b cuts the graph z = f(x, y) in the trace z = f(x, b), and f_x(a, b) is the slope of that curve at x = a: rise in z per unit step in x. Likewise f_y(a, b) is the slope of the trace in the plane x = a (Figure 1). The two tangent lines through (a, b, f(a, b)) have directions \langle 1, 0, f_x(a, b) \rangle and \langle 0, 1, f_y(a, b) \rangle; the plane through them is the candidate tangent plane of Tangent planes and linear approximation.
f_{xx}(a, b) is the second derivative of the trace z = f(x, b), its concavity. The mixed partial f_{xy}(a, b) measures how the slope in the x direction changes as you step in the y direction. On the saddle z = xy of Functions of several variables and level sets, f_x = y: the x slope grows by 1 per unit step in y, so f_{xy} = 1.
Both mixed partials come from the four corners of a small rectangle (Figure 2), through the double difference
Read by columns, \Delta is a y step at a + h minus the same y step at a; read by rows, a difference of x steps. That symmetry is why f_{xy} = f_{yx} for well-behaved f.
Results
Result 1 (rules). At points where x \mapsto f(x, b) is built by sums, products, quotients and compositions of differentiable functions, f_x(a, b) is computed by the one-variable rules with y = b held constant. The same holds for f_y and in any number of variables.
Result 2 (mean value theorem along a line). Let f_x exist at every point of an open disk B, and let (x_1, y) and (x_2, y) lie in B with x_1 < x_2. Then f(x_2, y) - f(x_1, y) = (x_2 - x_1) f_x(c, y) for some c with x_1 < c < x_2. In particular, if f_x = 0 on B, then on B the value f(x, y) depends on y alone. The same holds with x and y exchanged.
Result 3 (Clairaut’s theorem). Let f be defined on an open disk B about (a, b). Suppose f_x, f_y, f_{xy} and f_{yx} exist at every point of B, and f_{xy} and f_{yx} are continuous at (a, b). Then
So if f is of class C^2 on an open set U, then f_{xy} = f_{yx} on U. If f is of class C^k, a partial derivative of order at most k does not depend on the order in which the derivatives are taken, in any number of variables.
Why it is true
Result 1. f_x(a, b) = g'(a) for g(x) = f(x, b), a function of one variable in which b is a fixed number, and the one-variable rules apply to g.
Result 2. A disk contains the segment between any two of its points \mathbf{p} and \mathbf{q}, by the triangle inequality: for 0 \le t \le 1 and center \mathbf{c},
So g(x) = f(x, y) is differentiable at every point of [x_1, x_2], with g'(x) = f_x(x, y), hence continuous there, and the mean value theorem (The mean value theorem) gives c in (x_1, x_2) with g(x_2) - g(x_1) = g'(c)(x_2 - x_1).
Result 3. Let B have radius r and take \displaystyle 0 < \lvert h \rvert < \frac{r}{2}. Every point of the square with opposite corners (a, b) and (a + h, b + h) is within 2\lvert h \rvert of (a, b), so lies in B. Let \Delta = \Delta(h, h).
By rows: let \varphi(x) = f(x, b + h) - f(x, b) for x between a and a + h. Then \varphi'(x) = f_x(x, b + h) - f_x(x, b) exists, and the mean value theorem gives \xi strictly between a and a + h with
The function y \mapsto f_x(\xi, y) has derivative f_{xy}(\xi, y), so the mean value theorem again gives \eta between b and b + h with \Delta = h^2 f_{xy}(\xi, \eta).
By columns: with \psi(y) = f(a + h, y) - f(a, y) the same two steps give \Delta = h^2 f_{yx}(\xi', \eta') for a point (\xi', \eta') of the square. Hence
Given \varepsilon > 0, continuity at (a, b) gives \delta such that f_{xy} and f_{yx} are within \varepsilon of their values at (a, b) on B_\delta(a, b). Choosing 2\lvert h \rvert < \delta gives \lvert f_{xy}(a, b) - f_{yx}(a, b) \rvert < 2\varepsilon for every \varepsilon, so the two are equal.
A C^2 function satisfies the hypotheses on a disk about each point of U. For C^k, any reordering is a chain of swaps of neighbors, and a swap in places j and j + 1 is the C^2 case, with the other variables held fixed, applied to the partial of order j - 1 taken first.
Worked examples
The other variable held fixed
Let \displaystyle f(x, y) = x^3 y^2 - 4x\sqrt{y} + \cos(\pi x y) for y > 0. Find f_x(1, 4) and f_y(1, 4), then the four second-order partials at (1, 4).
SolutionHolding y fixed, x^3 y^2 is the constant y^2 times x^3, \displaystyle 4x\sqrt{y} is a constant times x, and the chain rule gives \cos(\pi x y) the inner derivative \pi y (Result 1):
Holding x fixed, \displaystyle \frac{d}{dy}\sqrt{y} = \frac{1}{2\sqrt{y}} and the inner derivative is \pi x:
Since \sin 4\pi = 0, f_x(1, 4) = 48 - 8 = 40 and f_y(1, 4) = 8 - 1 = 7. Differentiating again, with the product rule on y \sin(\pi x y) and x \sin(\pi x y),
and f_{yx}, computed from f_y, is the same expression. With \cos 4\pi = 1: f_{xx}(1, 4) = 96 - 16\pi^2, \displaystyle f_{yy}(1, 4) = \frac{17}{8} - \pi^2 and f_{xy}(1, 4) = f_{yx}(1, 4) = 23 - 4\pi^2, as Result 3 predicts for a function of class C^2 on the half plane y > 0.
A partial derivative from the definition
Let \displaystyle f(x, y) = \sqrt[3]{x^3 + y^3}, the real cube root. Find f_x(0, 0) and f_y(0, 0), and decide whether f_x(1, -1) exists.
SolutionWhere x^3 + y^3 \ne 0, Result 1 gives f_x = \dfrac{x^2}{(x^3 + y^3)^{2/3}}. At the origin this is \displaystyle \frac00: the cube root is not differentiable at 0, so Result 1 does not apply, and we use the definition. Since \displaystyle f(h, 0) = \sqrt[3]{h^3} = h,
and as f(x, y) = f(y, x), also f_y(0, 0) = 1.
At (1, -1), f(1, -1) = 0 and (1 + h)^3 - 1 = h(3 + 3h + h^2), so for h \ne 0
The numerator tends to \displaystyle \sqrt[3]{3} and the denominator to 0 through positive values, so the quotient tends to \infty: f_x(1, -1) does not exist, and the trace z = f(x, -1) has a vertical tangent at x = 1.
Rates from a table at the forge
Amara’s camera gives these temperatures T(x, y), in degrees Fahrenheit, on Samir’s plate.
| T(x, y) | x = 8 | x = 10 | x = 12 |
|---|---|---|---|
| y = 4 | 812 | 838 | 858 |
| y = 6 | 826 | 850 | 868 |
| y = 8 | 834 | 856 | 872 |
Estimate T_x(10, 6), T_y(10, 6) and T_{xy}(10, 6), and then T(10.4, 6).
SolutionAlong the row y = 6 only x changes. The forward and backward quotients \displaystyle \frac{868 - 850}{2} = 9 and \displaystyle \frac{850 - 826}{2} = 12 estimate T_x(10, 6) from each side; their average is the centered quotient
degrees per inch. Down the column x = 10, \displaystyle T_y(10, 6) \approx \frac{856 - 838}{4} = 4.5 degrees per inch.
For the mixed partial, estimate T_x on the rows y = 8 and y = 4, \displaystyle \frac{872 - 834}{4} = 9.5 and \displaystyle \frac{858 - 812}{4} = 11.5, and take their centered quotient in y:
degrees per inch per inch. This is the double difference of the four corners over 4 \cdot 4, \displaystyle \frac{872 - 858 - 834 + 812}{16} = \frac{-8}{16}: the plate warms toward larger x less steeply on the row y = 8 than on y = 4.
The tangent line of the trace z = T(x, 6), as in Linear approximation and differentials, gives T(10.4, 6) \approx 850 + 0.4 \cdot 10.5 = 854.2 degrees.
Implicit partial derivatives
The equation x z^3 + y^2 z - xy = 3 holds at (1, 2, 1). Near (1, 2) it defines z as a function of (x, y) with z(1, 2) = 1 and with partial derivatives, by the implicit function theorem (proved in Real Analysis), because the z-derivative of the left side, 3xz^2 + y^2, is 7 \ne 0 there. Find z_x(1, 2) and z_y(1, 2).
SolutionDifferentiate both sides with respect to x, holding y fixed and treating z as z(x, y), as in Implicit differentiation. The product rule on x z^3 and the chain rule on z^3 give
At (1, 2, 1), \displaystyle z_x = \frac{2 - 1}{3 + 4} = \frac17. With respect to y, holding x fixed,
so \displaystyle z_y(1, 2) = \frac{1 - 4}{7} = -\frac37. Both denominators are the z-derivative the theorem needed to be nonzero.
Proof: the wave equation
Let c > 0 and let F and G be twice differentiable on \mathbb{R}. Prove that u(x, t) = F(x + ct) + G(x - ct) satisfies u_{tt} = c^2 u_{xx}.
SolutionHold t fixed. The inner functions x + ct and x - ct have derivative 1 in x, so by the chain rule
Hold x fixed. The inner derivatives in t are c and -c, so
The profile F moves left at speed c and G moves right: waves.
Mixed partials that differ
Let f(x, y) = \dfrac{xy(x^2 - 4y^2)}{x^2 + y^2} for (x, y) \ne (0, 0) and f(0, 0) = 0. Show that f is continuous and has both partials everywhere, but f_{xy}(0, 0) = -4 and f_{yx}(0, 0) = 1.
SolutionOff the origin f is a rational function with nonzero denominator. With \displaystyle r = \sqrt{x^2 + y^2}, \lvert x^2 - 4y^2 \rvert \le 4r^2 and \lvert xy \rvert \le r^2, so \lvert f \rvert \le 4r^2, and f is continuous at the origin by the distance bound of Limits and continuity in several variables. The partials exist off the origin by Result 1, and at it f(h, 0) = 0 = f(0, k) gives f_x(0, 0) = f_y(0, 0) = 0.
f_{xy}(0, 0) needs f_x on the y-axis. For y \ne 0, f(0, y) = 0, so
With f_x(0, 0) = 0, f_x(0, y) = -4y for every y, and f_{xy}(0, 0) = -4. On the x-axis, for x \ne 0,
and f_y(0, 0) = 0, so f_y(x, 0) = x for every x and f_{yx}(0, 0) = 1.
Both mixed partials exist everywhere, yet -4 \ne 1, so by Result 3 they are not both continuous at the origin.
Which pairs are partial derivatives
(a) Find every f on \mathbb{R}^2 with f_x = y\cos(xy) + 2x and f_y = x\cos(xy) - 3y^2. (b) Show that no f has f_x = x^2 y and f_y = xy^2.
Solution(a) The test of Result 3 passes: both (y\cos(xy) + 2x)_y and (x\cos(xy) - 3y^2)_x equal \cos(xy) - xy\sin(xy). F(x, y) = \sin(xy) + x^2 has F_x = f_x, so (f - F)_x = 0, and Result 2 on disks of every radius gives f = \sin(xy) + x^2 + C(y). Then f_y = x\cos(xy) + C'(y), which is x\cos(xy) - 3y^2 exactly when C'(y) = -3y^2, so
K any constant; differentiating confirms each.
(b) If f existed, then f_{xy} = (x^2 y)_y = x^2 and f_{yx} = (xy^2)_x = y^2 would be continuous on \mathbb{R}^2, and Result 3 would give x^2 = y^2 everywhere. At (1, 0) that says 1 = 0, so no such f exists.
Boundaries
Partials see two lines only. f_x(a, b) and f_y(a, b) use the values of f on the two lines through (a, b) parallel to the axes. The function equal to 1 where xy \ne 0 and 0 on the axes has f_x(0, 0) = f_y(0, 0) = 0, yet equals 1 at every (t, t), t \ne 0, so it is not continuous at the origin. The condition that does give continuity, differentiability, is the subject of Tangent planes and linear approximation.
Clairaut’s theorem needs continuity (Example 6), and Result 2 needs segments: on an open set that is not convex, f_x = 0 does not force f to ignore x (Exercise 22).
Where this goes next
The traces of Figure 1 span the tangent plane of Tangent planes and linear approximation, where continuous partials are shown to give differentiability. The The multivariable chain rule differentiates compositions through partials, and Gradient and directional derivatives collects them in \nabla f to give the rate of change in every direction. The second partials form the Hessian of Extrema and the second derivative test, symmetric by Result 3. The test of Example 7 returns in Exact equations and Conservative fields and potential functions, and partials fill the Jacobian of Change of variables and the Jacobian and the equations of Complex differentiability and the Cauchy-Riemann equations.
Exercises
Computing partial derivatives
- Find f_x and f_y for f(x, y) = x^4 y - 3x^2 y^3 + 7y.
- Let g(x, y) = \dfrac{x - y}{x + 2y}. Find g_x(1, 1) and g_y(1, 1).
- Let h(x, y) = \arctan\left(\dfrac{y}{x}\right) for x > 0. Find h_x and h_y.
- Let w = xyz^2 + \ln(x + y^2 + z^2). Find w_x, w_y and w_z at (1, 1, 1).
- Let F(x, y) = \displaystyle\int_x^y e^{-t^2}\,dt. Find F_x and F_y.
From the definition
- Let \displaystyle f(x, y) = \sqrt{x^2 + y^2}. Show that f_x(0, 0) does not exist, and find f_x(3, 4).
- Let \displaystyle f(x, y) = \sqrt{\lvert xy \rvert}. Find f_x(0, 0) and f_y(0, 0), and show that f_x(0, 1) does not exist.
- Use the definition to find f_y(2, 1) for f(x, y) = \dfrac{x}{y}.
Rates at the forge
- From the table of Example 3, estimate T_y(8, 6) and T_y(12, 6) by centered quotients, then T_{yx}(10, 6). Compare with the estimate of T_{xy}(10, 6).
- Use the estimate of T_y(12, 6) from Exercise 9 to estimate T(12, 6.6).
- A bar Samir has pulled from the fire has temperature T(t, d) degrees Fahrenheit, t minutes later and d inches from its hot tip, with T_t(3, 2) = -40 and T_d(3, 2) = -25. Say what each number means, with units, and estimate T(3.1, 2) - T(3, 2) and T(3, 2.2) - T(3, 2).
Second-order partial derivatives
- Find the four second-order partials of f(x, y) = x^2 \sin y + y e^{2x}, and confirm that f_{xy} = f_{yx}.
- Find the constant c for which u(x, t) = e^{ct} \sin 3x satisfies u_t = 2u_{xx}.
- Show that u(x, y) = \ln(x^2 + y^2) satisfies u_{xx} + u_{yy} = 0 at every point other than the origin.
- For f(x, y, z) = x^2 y^3 z^4, compute f_{xyz} and f_{zyx}.
Is it a pair of partial derivatives?
- Find every f with f_x = 3x^2 + 2y^2 and f_y = 4xy + e^y, or show there is none.
- Find every f with f_x = x\cos y and f_y = x\sin y, or show there is none.
- Find the constant a for which some f has f_x = 2xy + ay^2 and f_y = x^2 + 6xy, and find every such f.
Proofs
- Refute: if f_x(a, b) and f_y(a, b) exist, then f is continuous at (a, b). Use f(x, y) = \dfrac{xy^2}{x^2 + y^4} with f(0, 0) = 0.
- Prove that if f_x = f_y = 0 at every point of an open disk B, then f is constant on B.
- Prove that if f is of class C^3 on an open set U, then f_{xyy} = f_{yxy} = f_{yyx} on U.
- Let U be the plane without the ray \{(0, y) : y \ge 0\}. Refute: if f_x = 0 at every point of an open set U, then f(x_1, y) = f(x_2, y) whenever both points lie in U.
- Prove that if f is of class C^2 on \mathbb{R}^2 and f_{xy} = 0 everywhere, then f(x, y) = g(x) + h(y) for some functions g and h of one variable.
Answers
- f_x = 4x^3 y - 6xy^3 and f_y = x^4 - 9x^2 y^2 + 7, by the power rule with the other variable constant.
- \displaystyle g_x(1, 1) = \frac13 and \displaystyle g_y(1, 1) = -\frac13. By the quotient rule \displaystyle g_x = \frac{3y}{(x + 2y)^2} and \displaystyle g_y = \frac{-(x + 2y) - 2(x - y)}{(x + 2y)^2} = \frac{-3x}{(x + 2y)^2}.
- \displaystyle h_x = -\frac{y}{x^2 + y^2} and \displaystyle h_y = \frac{x}{x^2 + y^2}, by the chain rule: \displaystyle h_x = \frac{1}{1 + y^2/x^2} \cdot \left(-\frac{y}{x^2}\right) and \displaystyle h_y = \frac{1}{1 + y^2/x^2} \cdot \frac1x.
- \displaystyle w_x = \frac43, \displaystyle w_y = \frac53, \displaystyle w_z = \frac83. \displaystyle w_x = yz^2 + \frac{1}{x + y^2 + z^2}, \displaystyle w_y = xz^2 + \frac{2y}{x + y^2 + z^2}, \displaystyle w_z = 2xyz + \frac{2z}{x + y^2 + z^2}, with denominator 3 at the point.
- F_x = -e^{-x^2} and F_y = e^{-y^2}. With E(s) = \int_0^s e^{-t^2}\,dt, F(x, y) = E(y) - E(x) and E'(s) = e^{-s^2} by the fundamental theorem of calculus.
- f(h, 0) = \lvert h \rvert, so the quotient \displaystyle \frac{\lvert h \rvert}{h} is 1 for h > 0 and -1 for h < 0: no limit. Away from the origin Result 1 gives \displaystyle f_x = \frac{x}{\sqrt{x^2 + y^2}}, so \displaystyle f_x(3, 4) = \frac35.
- f_x(0, 0) = f_y(0, 0) = 0, since f(h, 0) = 0 = f(0, k). At (0, 1) the quotient is \displaystyle \frac{\sqrt{\lvert h \rvert}}{h}, which tends to \infty as h \to 0^+ and to -\infty as h \to 0^-: no limit.
- f_y(2, 1) = -2, since \displaystyle \frac{1}{k}\left(\frac{2}{1 + k} - 2\right) = \frac{-2}{1 + k} \to -2.
- \displaystyle T_y(8, 6) \approx \frac{834 - 812}{4} = 5.5 and \displaystyle T_y(12, 6) \approx \frac{872 - 858}{4} = 3.5 degrees per inch, so \displaystyle T_{yx}(10, 6) \approx \frac{3.5 - 5.5}{4} = -0.5. It equals the estimate of T_{xy}(10, 6) exactly: both are the double difference \displaystyle \frac{-8}{16} of the same four corners (Figure 2).
- T(12, 6.6) \approx 868 + 0.6(3.5) = 870.1 degrees.
- At 3 minutes and 2 inches from the tip, the temperature is falling about 40 degrees per minute, and at that moment it drops about 25 degrees per inch moving away from the tip. So T(3.1, 2) - T(3, 2) \approx 0.1(-40) = -4 degrees and T(3, 2.2) - T(3, 2) \approx 0.2(-25) = -5 degrees.
- f_x = 2x\sin y + 2ye^{2x} and f_y = x^2\cos y + e^{2x}, so f_{xx} = 2\sin y + 4ye^{2x}, f_{yy} = -x^2\sin y and f_{xy} = 2x\cos y + 2e^{2x} = f_{yx}.
- c = -18, since u_t = ce^{ct}\sin 3x and u_{xx} = -9e^{ct}\sin 3x.
- \displaystyle u_x = \frac{2x}{x^2 + y^2} and by the quotient rule \displaystyle u_{xx} = \frac{2(y^2 - x^2)}{(x^2 + y^2)^2}. By symmetry \displaystyle u_{yy} = \frac{2(x^2 - y^2)}{(x^2 + y^2)^2}, and the sum is 0.
- Both are 24xy^2z^3: f_{xy} = 6xy^2z^4 and f_{zy} = 12x^2y^2z^3 on the way, as Result 3 predicts.
- f = x^3 + 2xy^2 + e^y + K, K any constant. The test passes, (3x^2 + 2y^2)_y = 4y = (4xy + e^y)_x; as in Example 7, f = x^3 + 2xy^2 + C(y) and f_y = 4xy + C'(y) forces C' = e^y.
- None. (x\cos y)_y = -x\sin y and (x\sin y)_x = \sin y are continuous on \mathbb{R}^2, so Result 3 would make them equal; at \displaystyle \left(2, \frac{\pi}{2}\right) they are -2 and 1.
- a = 3, and f = x^2 y + 3xy^2 + K. Result 3 needs (2xy + ay^2)_y = 2x + 2ay to equal (x^2 + 6xy)_x = 2x + 6y everywhere, so a = 3. Then f = x^2 y + 3xy^2 + C(y), and f_y = x^2 + 6xy + C'(y) forces C constant.
- f(x, 0) = 0 and f(0, y) = 0 for all x and y, so both difference quotients at the origin are 0 and f_x(0, 0) = f_y(0, 0) = 0. On the path x = y^2, y \ne 0, \displaystyle f = \frac{y^4}{y^4 + y^4} = \frac12, while (y^2, y) \to (0, 0). By the path result of Limits and continuity in several variables, f does not tend to f(0, 0) = 0, so it is not continuous at the origin.
- Let (a, b) be the center and r the radius of B, and let (x, y) \in B. The horizontal segment from (a, b) to (x, b) and the vertical one from (x, b) to (x, y) lie in B, since moving t from y toward b only shrinks (x - a)^2 + (t - b)^2. Result 2 with f_x = 0 gives f(x, b) = f(a, b), and Result 2 with the roles exchanged and f_y = 0 gives f(x, y) = f(x, b). So f(x, y) = f(a, b).
- f is of class C^2, so f_{xy} = f_{yx} on U by Result 3, and differentiating these equal functions in y gives f_{xyy} = f_{yxy}. Next, g = f_y has as second partials third partials of f, so g is of class C^2 on U, and Result 3 gives g_{xy} = g_{yx}, that is f_{yxy} = f_{yyx}.
- Let f(x, y) = y^2 when x > 0 and y > 0, and f(x, y) = 0 otherwise. Where y \le 0, f is 0 along the whole horizontal line, so f_x = 0. At a point of U with y > 0, x \ne 0, and x + h has the sign of x for small h, so f is constant on a short horizontal segment and f_x = 0. Yet f(1, 1) = 1 and f(-1, 1) = 0. The segment between them crosses the removed ray, where Result 2’s proof breaks.
- Since (f_x)_y = 0 on \mathbb{R}^2, Result 2 with the roles exchanged, applied to f_x on disks of every radius, gives f_x(x, y) = f_x(x, 0) = p(x), and p is continuous because f_x is. Let g(x) = \int_0^x p(s)\,ds, so g' = p by the fundamental theorem of calculus. Then (f - g)_x = 0 on \mathbb{R}^2, so by Result 2, f(x, y) - g(x) = f(0, y) - g(0) = f(0, y). With h(y) = f(0, y), f = g(x) + h(y).