Spring 2009 - Problem 12

change of variables, conformal mappings

Let QQ be the closed unit square in the complex plane C\C and let RR be the closed rectangle in C\C with vertices {0,2,i,2+i}\set{0, 2, i, 2+i}. Prove that there does not exist a surjective homeomorphism f ⁣:QR\func{f}{Q}{R} that is conformal on the interior QoQ^\itr and that maps corners to corners.

Solution.

Suppose f ⁣:QR\func{f}{Q}{R} is such a mapping. Parametrize Q\partial{Q} counter-clockwise by γ\gamma. Since ff is continuous and injective, this means that fγf \circ \gamma is a simple closed curve. Suppose (fγ)Ro\p{f \circ \gamma} \cap R^{\itr} \neq \emptyset. Then by the Jordan curve theorem, this implies the existence of w1w_1 in the interior region of fγf \circ \gamma and w2w_2 in the exterior region. But the interior and exterior regions are disconnected, which implies that f(Qo)f\p{Q^{\itr}} is disconnected, which is impossible as it is the continuous image of a connected set. Thus, fγRf \circ \gamma \subseteq \partial{R}, and because ff maps corners to corners in an injective way, it follows that fγ=Rf \circ \gamma = \partial{R}.

Since ff is injective, it traverses the vertices of RR in a clockwise or counter-clockwise way. By reversing γ\gamma if necessary, we may assume that f(0)=0f\p{0} = 0, f(1)=2f\p{1} = 2, f(1+i)=2+if\p{1 + i} = 2 + i, and f(i)=if\p{i} = i. For y[0,1]y \in \br{0, 1} consider the segment Cy=[iy,1+iy]C_y = \br{iy, 1 + iy} parameterized by xx+iyx \mapsto x + iy. Because of how fγf \circ \gamma traverses R\partial{R}, it follows f(iy)f\p{iy} and f(1+iy)f\p{1 + iy} are on the sides of length 11 of RR, and it follows that diamf(Cγ)2\diam{f\p{C_\gamma}} \geq 2. Thus,

diamf(Cγ)=01f(x+iy)dx2.\diam{f\p{C_\gamma}} = \int_0^1 \abs{f'\p{x + iy}} \,\diff{x} \geq 2.

On the other hand, by Cauchy-Schwarz,

0101f(x+iy)dxdy=Qf(x+iy)dA(Qf(x+iy)2dA)1/2(QdA)1/2=(Qf(x+iy)2dA)1/2(QdA)1/2.\begin{aligned} \int_0^1 \int_0^1 \abs{f'\p{x + iy}} \,\diff{x} \,\diff{y} &= \int_Q \abs{f'\p{x + iy}} \,\diff{A} \\ &\leq \p{\int_Q \abs{f'\p{x + iy}}^2 \,\diff{A}}^{1/2} \p{\int_Q \diff{A}}^{1/2} \\ &= \p{\int_Q \abs{f'\p{x + iy}}^2 \,\diff{A}}^{1/2} \p{\int_Q \diff{A}}^{1/2}. \end{aligned}

The Jacobian of f=u+ivf = u + iv as a function R2R2\R^2 \to \R^2 is given by

df=(uxuyvxvy).\diff{f} = \begin{pmatrix} u_x & u_y \\ v_x & v_y \end{pmatrix}.

Combined with the Cauchy-Riemann equations, this has determinant

uxvyvxuy=(ux)2+(vx2)=fx2=f2.u_xv_y - v_xu_y = \p{u_x}^2 + \p{v_x^2} = \abs{\pder{f}{x}}^2 = \abs{f'}^2.

Thus, by the change of variables formula,

Qf(x+iy)2dA=f(Q)dA=areaR.\int_Q \abs{f'\p{x + iy}}^2 \,\diff{A} = \int_{f\p{Q}} \,\diff{A} = \area{R}.

Thus,

20101f(x+iy)dxdy(areaR)1/2(areaQ)1/2=2,\begin{aligned} 2 &\leq \int_0^1 \int_0^1 \abs{f'\p{x + iy}} \,\diff{x} \,\diff{y} \\ &\leq \p{\area{R}}^{1/2} \p{\area{Q}}^{1/2} \\ &= \sqrt{2}, \end{aligned}

which is impossible. Thus, no such ff exists, which completes the proof.