58 Chapter 2. <strong>Surfaces</strong>: Local TheoryFor example, the plane <strong>and</strong> cyl<strong>in</strong>der are locally isometric, <strong>and</strong> hence the cyl<strong>in</strong>der (as we wellknow) is flat. We now conclude that s<strong>in</strong>ce the Gaussian curvature of a sphere is nonzero, a spherecannot be locally isometric to a plane. Thus, there is no way to map the earth “faithfully” (preserv<strong>in</strong>gdistance)—even locally—on a piece of paper. In some sense, the Mercator projection (seeExercise 2.1.9) is the best we can do, for, although it distorts distances, it does preserve angles.Let’s now apply the Codazzi equations to prove a rather strik<strong>in</strong>g result about surfaces withK =0everywhere, called flat surfaces.Proposition 3.3. Suppose M is a flat surface with no planar po<strong>in</strong>ts. Then M is a ruled surfacewhose tangent plane is constant along the rul<strong>in</strong>gs.Proof. S<strong>in</strong>ce M has no planar po<strong>in</strong>ts, we can choose k 1 =0<strong>and</strong> k 2 ≠0everywhere. Thenby Theorem 3.3 of the Appendix, there is a local parametrization of M so that the u-curves arethe first l<strong>in</strong>es of curvature <strong>and</strong> the v-curves are the second l<strong>in</strong>es of curvature. This means first ofall that F = m =0. Now, s<strong>in</strong>ce k 1 =0,for any P ∈ M we have S P (x u )=0, <strong>and</strong> so n u = 0everywhere <strong>and</strong> n is constant along the u-curves.We now want to show that the u-curves are <strong>in</strong> fact l<strong>in</strong>es. <strong>First</strong>, l =II(x u , x u )=−S P (x u )·x u =0<strong>and</strong> n ≠0,s<strong>in</strong>ce k 2 ≠0.From the first of the Codazzi equations we now deduce that<strong>and</strong> so Γ v uu =0. This means that0=l v − m u = lΓ u uv + m ( Γ v uv − Γ u uu)− nΓvuu = −nΓ v uu,x uu =Γ u uux u +Γ v uux v + ln =Γ u uux uis just a multiple of x u .Thus, the tangent vector x u never changes direction as we move along theu-curves, <strong>and</strong> this means that the u-curves must be l<strong>in</strong>es. In conclusion, we have a ruled surfacewhose tangent plane is constant along rul<strong>in</strong>gs. □Remark. Such ruled surfaces are called developable. (See Exercise 10 <strong>and</strong> Exercise 2.1.8.) Theterm<strong>in</strong>ology comes from the fact that they can be rolled out—or “developed”—onto a plane.Next we prove a strik<strong>in</strong>g global result about compact surfaces. (Recall that a subset of R 3 iscompact if it is closed <strong>and</strong> bounded. The salient feature of compact sets is the maximum valuetheorem: A cont<strong>in</strong>uous real-valued function on a compact set achieves its maximum <strong>and</strong> m<strong>in</strong>imumvalues.) We beg<strong>in</strong> with a straightforwardProposition 3.4. Suppose M ⊂ R 3 is a compact surface. Then there is a po<strong>in</strong>t P ∈ M withK(P ) > 0.Proof. Because M is compact, the cont<strong>in</strong>uous function f(x) =‖x‖ achieves its maximum atsome po<strong>in</strong>t of M, <strong>and</strong> so there is a po<strong>in</strong>t P ∈ M farthest from the orig<strong>in</strong> (which may or may not be<strong>in</strong>side M), as <strong>in</strong>dicated <strong>in</strong> Figure 3.2. Let f(P )=R. AsExercise 1.2.7 shows, the curvature of anycurve α ⊂ M at P is at least 1/R, so—if we choose the unit normal n to be <strong>in</strong>ward-po<strong>in</strong>t<strong>in</strong>g—everynormal curvature of M at P is at least 1/R. Itfollows that K(P ) ≥ 1/R 2 > 0. (That is, M is atleast as curved at P as the circumscribed sphere of radius R tangent to M at P .) □
§3. The Codazzi <strong>and</strong> Gauss Equations <strong>and</strong> the Fundamental Theorem of Surface Theory 590MPFigure 3.2The reader is asked <strong>in</strong> Exercise 16 to f<strong>in</strong>d surfaces of revolution of constant curvature. Thereare, <strong>in</strong>terest<strong>in</strong>gly, many nonobvious examples. However, if we restrict ourselves to smooth surfaces,we have the follow<strong>in</strong>g beautifulTheorem 3.5 (Liebmann). If M is a smooth, compact surface of constant Gaussian curvatureK, then K>0 <strong>and</strong> M must be a sphere of radius 1/ √ K.We will need the follow<strong>in</strong>gLemma 3.6 (Hilbert). Suppose P is not an umbilic po<strong>in</strong>t <strong>and</strong> k 1 (P ) >k 2 (P ). Suppose k 1 hasalocal maximum at P <strong>and</strong> k 2 has a local m<strong>in</strong>imum at P . Then K(P ) ≤ 0.Proof. We work <strong>in</strong> a “pr<strong>in</strong>cipal” coord<strong>in</strong>ate parametrization, 4 so that the u-curves are l<strong>in</strong>esof curvature with pr<strong>in</strong>cipal curvature k 1 <strong>and</strong> the v-curves are l<strong>in</strong>es of curvature with pr<strong>in</strong>cipalcurvature k 2 . Then we have k 1 = l/E, k 2 = n/G, <strong>and</strong> F = m =0.Bythe Codazzi equations <strong>and</strong>the equations (‡) onp.55,we have(⋆)l v = 1 2 E v(k 1 + k 2 ) <strong>and</strong> n u = 1 2 G u(k 1 + k 2 ).S<strong>in</strong>ce l = k 1 E,wehavel v =(k 1 ) v E + k 1 E v ; similarly, n u =(k 2 ) u G + k 2 G u . Us<strong>in</strong>g the equations(⋆), we then obta<strong>in</strong>(k 1 ) v = E v2E (k 2 − k 1 ) <strong>and</strong> (k 2 ) u = G u(⋆⋆)2G (k 1 − k 2 );s<strong>in</strong>ce k 1 ≠ k 2 <strong>and</strong> (k 1 ) v =(k 2 ) u =0atP ,we<strong>in</strong>fer that E v = G u =0atP .Differentiat<strong>in</strong>g the equations (⋆⋆), <strong>and</strong> remember<strong>in</strong>g that (k 1 ) u =(k 2 ) v =0atP as well, wehave at P :(k 1 ) vv = E vv2E (k 2 − k 1 ) ≤ 0 (because k 1 has a local maximum at P )(k 2 ) uu = G uu2G (k 1 − k 2 ) ≥ 0 (because k 2 has a local m<strong>in</strong>imum at P ),4 S<strong>in</strong>ce locally there are no umbilic po<strong>in</strong>ts, the existence of such a parametrization is an immediate consequenceof Theorem 3.3 of the Appendix.