Real Analysis Homework 1

import%20marimo%0A%0A__generated_with%20%3D%20%220.17.6%22%0Aapp%20%3D%20marimo.App()%0A%0A%0A%40app.cell%0Adef%20_()%3A%0A%20%20%20%20import%20marimo%20as%20mo%0A%20%20%20%20return%20(mo%2C)%0A%0A%0A%40app.cell%0Adef%20_(mo)%3A%0A%20%20%20%20nav_menu%20%3D%20mo.nav_menu(%0A%20%20%20%20%20%20%20%20%7B%0A%20%20%20%20%20%20%20%20%20%20%20%20%22%2Findex.html%22%3A%20f%22%7Bmo.icon('lucide%3Ahome')%7D%20Home%22%2C%0A%20%20%20%20%20%20%20%20%7D%2C%0A%20%20%20%20%20%20%20%20orientation%3D%22vertical%22%2C%0A%20%20%20%20)%0A%20%20%20%20nav_menu%0A%20%20%20%20return%0A%0A%0A%40app.cell%0Adef%20_(mo)%3A%0A%20%20%20%20mo.md(r%22%22%22%0A%20%20%20%20%23%20Fall%20Semester%20Homework%201%0A%20%20%20%20%22%22%22)%0A%20%20%20%20return%0A%0A%0A%40app.cell%0Adef%20_(mo)%3A%0A%20%20%20%20mo.md(r%22%22%22%0A%20%20%20%20%23%23%20Exercise%202%0A%0A%20%20%20%20**Prove%20an%20analogue%20of%20Theorem%201.8%20for%20%24n%24%20functions**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20The%20statement%20we%20want%20to%20prove%3A%0A%0A%20%20%20%20Let%20%24u_1%2Cu_2%2C%5Cdots%2Cu_n%24%20be%20%24n%24%20real%20measurable%20functions%20on%20a%20measurable%20space%20%24X%24%2C%20%24Y%24%20be%20some%20topological%20space%2C%20let%20%24%5CPhi%24%20be%20a%20continuous%20mapping%20of%20%24%5Cmathbb%7BR%7D%5En%20%5Cto%20Y%24.%20Define%20%24h(x)%20%3D%20%5CPhi(u_1(x)%2C%5Cdots%2Cu_n(x))%24%20for%20%24x%20%5Cin%20X%24%2C%20then%20%24h%3AX%5Cto%20Y%24%20is%20measurable.%0A%0A%20%20%20%20Let%20%24f(x)%20%3D%20(u_1(x)%2C%5Cdots%2Cu_n(x))%24%20then%20%24h%20%3D%20%5CPhi%20%5Ccirc%20f%24%2C%20if%20%24f%24%20is%20measurable%20then%20%24h%24%20is%20measurable%3A%0A%0A%20%20%20%20Let%20%24R%24%20be%20an%20arbitrary%20%24n%24%20dimension%20rectangle%20in%20the%20form%20%24R%20%3D%20I_1%20%5Ctimes%20I_2%20%5Ctimes%20%5Cdots%20%5Ctimes%20I_n%24%2C%20then%20%24f%5E%7B-1%7D(R)%3D%20u_1%5E%7B-1%7D(I_1)%20%5Ccap%20%5Cdots%5Ccap%20u_n%5E%7B-1%7D(I_n)%24%20(if%20%24x%20%5Cin%20f%5E%7B-1%7D(R)%24%20then%20%24f(x)%20%5Cin%20R%24%20then%20%24u_1(x)%20%5Cin%20I_1%2C%5Cdots%2Cu_n(x)%20%5Cin%20I_n%24%20then%20%24x%20%5Cin%20u_1%5E%7B-1%7D(I_1)%2C%5Cdots%2Cx%20%5Cin%20u_n%5E%7B-1%7D(I_n)%24%20thus%20%24x%24%20is%20in%20the%20intersection%2C%20similar%20the%20other%20way%20around).%0A%0A%20%20%20%20So%20%24f%5E%7B-1%7D(R)%24%20is%20measurable%20because%20it%20is%20a%20finite%20intersection%20of%20measurable%20set.%20Since%20the%20%24n%24-rectangles%20form%20a%20basis%20for%20%24%5Cmathbb%7BR%7D%5En%24%2C%20any%20%24V%24%20open%20in%20%24%5Cmathbb%7BR%7D%5En%24%20(which%20is%20second%20countable)%20is%20a%20countable%20union%20of%20such%20%24R%24's%2C%20so%20we%20can%20write%20%24f%5E%7B-1%7D(V)%20%3D%20f%5E%7B-1%7D(%5Cbigcup%5Climits_%7Bi%20%3D%201%7D%5E%7B%5Cinfty%7DR_i)%20%3D%20%5Cbigcup%5Climits_%7Bi%20%3D%201%7D%5E%7B%5Cinfty%7Df%5E%7B-1%7D(R_i)%24%20is%20a%20countable%20union%20of%20measurable%20sets%20thus%20is%20measurable.%0A%0A%20%20%20%20Since%20%24V%24%20can%20be%20chose%20arbitrarily%2C%20%24f%24%20thus%20%24h%24%20is%20measurable.%0A%0A%20%20%20%20%23%23%20Exercise%203%0A%0A%20%20%20%20**Prove%20that%20if%20%24f%24%20is%20a%20real%20function%20on%20a%20measurable%20space%20%24X%24%20such%20that%20%24%5Clbrace%20x%3Af(x)%20%5Cge%20r%5Crbrace%24%20is%20measurable%20for%20every%20rational%20%24r%24%2C%20then%20%24f%24%20is%20measurable**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20By%20taking%20set%20subtraction%20we%20can%20see%20that%20%24%5Clbrace%20x%3Aa%5Cle%20f(x)%20%3Cb%20%5Crbrace%24%20is%20measurable%20for%20any%20rational%20%24a%2Cb%24.%20Every%20open%20interval%20in%20%24%5Cmathbb%7BR%7D%24%20can%20be%20written%20as%20a%20countable%20union%20of%20half-open%20intervals%20of%20the%20form%20%24%5Ba%2Cb)%2Ca%2Cb%5Cin%5Cmathbb%7BQ%7D%24%2C%20and%20every%20open%20set%20in%20%24%5Cmathbb%7BR%7D%24%20can%20be%20written%20as%20a%20countable%20union%20of%20open%20intervals%2C%20i.e.%20%24%5Clbrace%20x%20%3A%20f(x)%20%5Cin%20U%5Crbrace%24%20is%20measurable%20for%20any%20%24U%24%20open%20in%20%24%5Cmathbb%7BR%7D%24%2C%20meaning%20%24f%24%20is%20measurable.%0A%0A%20%20%20%20%23%23%20Exercise%204%0A%0A%20%20%20%20**Let%20%24%5Clbrace%20a_n%5Crbrace%24%20and%20%24%5Clbrace%20b_n%5Crbrace%24%20be%20sequences%20in%20%24%5B-%5Cinfty%2C%5Cinfty%5D%24%2C%20and%20prove%20the%20following%20assertions**%3A%0A%0A%20%20%20%201.%20**%24%5Climsup_%7Bn%5Cto%5Cinfty%7D(-a_n)%20%3D%20-%5Climinf_%7Bn%5Cto%20%5Cinfty%7Da_n%24**%3B%0A%20%20%20%202.%20**%24%5Climsup_%7Bn%5Cto%5Cinfty%7D(a_n%2Bb_n)%20%5Cle%20%5Climsup_%7Bn%5Cto%5Cinfty%7D(a_n)%20%2B%20%5Climsup_%7Bn%5Cto%5Cinfty%7D(b_n)%24%20provided%20none%20of%20the%20sums%20is%20of%20the%20form%20%24%5Cinfty%20-%20%5Cinfty%24**%3B%0A%20%20%20%203.%20**If%20%24a_n%20%5Cle%20b_n%24%20for%20all%20%24n%24%2C%20then%20%24%5Climinf_%7Bn%5Cto%20%5Cinfty%7Da_n%20%5Cle%20%5Climinf_%7Bn%5Cto%20%5Cinfty%7Db_n%24**.%0A%0A%20%20%20%20**Show%20by%20an%20example%20that%20strict%20inequality%20can%20hold%20in%202**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%201.%20%24%5Climinf%20a_n%24%20%24%5Coverset%7B%5Ctext%7Bdef%7D%7D%7B%3D%7D%20%5Clim%5Climits_%7Bk%20%5Cto%20%5Cinfty%7D%5Cinf%5Climits_%7Bn%3Ek%7Da_n%24%20%24%3D%20%5Clim%5Climits_%7Bk%20%5Cto%20%5Cinfty%7D(-%5Csup%5Climits_%7Bn%3Ek%7D(-a_n))%24%20%24%5Coverset%7B%5Ctext%7Bdef%7D%7D%7B%3D%7D%20-%5Climsup(-a_n)%24%20(if%20%24b%24%20is%20a%20lower%20bound%20of%20%24a_n%24%20then%20%24-b%24%20is%20an%20upper%20bound%20of%20%24-a_n%24%20because%20%24b%20%5Cle%20a_n%20%5Cimplies%20-a_n%20%5Cle%20-b_n%24)%3B%0A%20%20%20%202.%20Consider%20the%20inequality%20%24%5Csup%5Climits_%7Bn%20%3E%20k%7Da_n%20%2B%20%5Csup%5Climits_%7Bn%20%3E%20k%7Db_n%20%5Cge%20a_l%20%2B%20b_l%24%20for%20any%20%24l%20%3E%20k%24%2C%20thus%20by%20definition%20%24%5Csup%5Climits_%7Bn%20%3E%20k%7Da_n%20%2B%20%5Csup%5Climits_%7Bn%20%3E%20k%7Db_n%24%20is%20an%20upper%20bound%20for%20%24a_n%20%2B%20b_n%24%20thus%20we%20have%20%24%5Csup%5Climits_%7Bn%20%3E%20k%7Da_n%20%2B%20%5Csup%5Climits_%7Bn%20%3E%20k%7Db_n%20%5Cge%20%5Csup%5Climits_%7Bn%20%3E%20k%7D(a_n%20%2B%20b_n)%24.%20Now%20pass%20limit%20each%20side%2C%20we%20have%20%24%5Climsup%7B(a_n%20%2B%20b_n)%7D%20%5Cle%20%5Climsup%7B(a_n)%7D%20%2B%20%5Climsup%7B(b_n)%7D%24%20(as%20long%20as%20we%20do%20not%20have%20the%20undefined%20form%20%24%5Cinfty-%5Cinfty%24)%3B%0A%20%20%20%203.%20If%20%24a_n%20%5Cle%20b_n%24%20for%20all%20%24n%24%20then%20any%20lower%20bound%20for%20%24a_n%24's%20is%20a%20lower%20bound%20for%20%24b_n%24's%2C%20i.e.%20%24%5Cinf%5Climits_%7Bn%3Ek%7Da_n%20%5Cle%20%5Cinf%5Climits_%7Bn%3Ek%7Db_n%24%2C%20now%20pass%20limit%20we%20have%20%24%5Climinf_%7Bn%5Cto%20%5Cinfty%7Da_n%20%5Cle%20%5Climinf_%7Bn%5Cto%20%5Cinfty%7Db_n%24.%0A%0A%20%20%20%20For%20example%20if%20%24%5Clbrace%20a_n%5Crbrace%20%3D%20-1%2C1%2C-1%2C1%2C%5Cdots%24%20and%20%24%5Clbrace%20b_n%5Crbrace%20%3D%201%2C-1%2C1%2C-1%2C%5Cdots%24%20then%20%24a_n%2Bb_n%20%5Cequiv%200%24%20so%20%24%5Climsup(a_n%2Bb_n)%20%3D%200%24%2C%20yet%20%24%5Climsup(a_n)%20%3D%20%5Climsup(b_n)%20%3D%201%24.%0A%0A%20%20%20%20%23%23%20Exercise%205%0A%0A%20%20%20%201.%20**Suppose%20%24f%3AX%5Cto%20%5B-%5Cinfty%2C%5Cinfty%5D%24%20and%20%24g%3AX%5Cto%20%5B-%5Cinfty%2C%5Cinfty%5D%24%20are%20measurable.%20Prove%20that%20the%20sets%20%24%5Clbrace%20x%3Af(x)%20%3C%20g(x)%5Crbrace%2C%5Clbrace%20x%3Af(x)%20%3D%20g(x)%5Crbrace%24%20are%20measurable**%3B%0A%20%20%20%202.%20**Prove%20that%20the%20set%20of%20points%20at%20which%20a%20sequence%20of%20measurable%20real-valued%20functions%20converges%20to%20a%20finite%20limit%20is%20measurable**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%201.%20Since%20%24f%2C%20g%24%20are%20measurable%2C%20so%20is%20%24f-g%3AX%5Cto%20%5B-%5Cinfty%2C%5Cinfty%5D%24%2C%20so%20%24%5Clbrace%20x%3Af(x)%20%3C%20g(x)%5Crbrace%20%3D%20(f-g)%5E%7B-1%7D(%5B-%5Cinfty%2C0))%24%20is%20preimage%20of%20open%20set%20is%20measurable.%20Thus%20%24%5Clbrace%20x%3Af(x)%20%5Cge%20g(x)%5Crbrace%20%3D%20%5Clbrace%20x%3Af(x)%20%3C%20g(x)%5Crbrace%5Ec%24%20is%20measurable%3B%20interchange%20the%20role%20of%20%24f%2Cg%24%20we%20get%20%24%5Clbrace%20x%3Af(x)%20%5Cle%20g(x)%5Crbrace%24%20is%20measurable%2C%20so%20%24%5Clbrace%20x%3Af(x)%3D%20g(x)%5Crbrace%24%20%24%3D%20%5Clbrace%20x%3Af(x)%20%5Cge%20g(x)%5Crbrace%20%5Ccap%20%5Clbrace%20x%3Af(x)%20%5Cle%20g(x)%5Crbrace%24%20is%20measurable%3B%0A%20%20%20%202.%20If%20I%20understand%20this%20right%20the%20question%20is%20asking%20to%20show%20%24%5Clbrace%20x%3A%20%5Clim%5Climits%20f_n(x)%24%20exists%20and%20finite%20%24%5Crbrace%24%20is%20measurable.%20Notice%20that%20%24f_n%24%20converge%20if%20and%20only%20if%20%24%5Climsup%20f_n%20%3D%20%5Climinf%20f_n%24%20and%20we%20proved%20that%20if%20%24%5Clbrace%20f_n%5Crbrace%24%20is%20a%20sequence%20of%20measurable%20function%20then%20%24%5Climsup%20f_n%24%20and%20%24%5Climinf%20f_n%24%20are%20measurable%2C%20so%20%24%5Clbrace%20x%3A%20%5Clim%5Climits%20f_n(x)%24%20exists%20%24%5Crbrace%24%20%24%3D%5Clbrace%20x%3A%20%5Climsup%20f_n(x)%20%3D%20%5Climinf%20f_n(x)%5Crbrace%24%20is%20measurable%20by%20part%201.%20Now%20notice%20that%20%24(-%5Cinfty%2C%20%5Cinfty%5D%24%20and%20%24%5B-%5Cinfty%2C%5Cinfty)%24%20are%20open%2C%20so%20taking%20preimage%20and%20intersection%20with%20the%20above%20set%20and%20we%20get%20the%20desired%20statement.%0A%0A%20%20%20%20%23%23%20Exercise%206%0A%0A%20%20%20%20**Let%20%24X%24%20be%20an%20uncountable%20set%2C%20let%20%24M%24%20be%20the%20collection%20of%20all%20sets%20%24E%20%5Csubseteq%20X%24%20such%20that%20either%20%24E%24%20or%20%24E%5Ec%24%20is%20at%20most%20countable%2C%20and%20define%20%24%5Cmu(E)%20%3D%200%24%20in%20the%20first%20case%2C%20%24%5Cmu(E)%20%3D%201%24%20in%20the%20second.%20Prove%20that%20%24M%24%20is%20a%20%24%5Csigma%24-algebra%20in%20%24X%24%20and%20that%20%24%5Cmu%24%20is%20a%20measure%20on%20%24M%24.%20Describe%20the%20corresponding%20measurable%20functions%20and%20their%20integrals**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20%24M%24%20is%20a%20%24%5Csigma%24-algebra%3A%0A%0A%20%20%20%201.%20%24X%5Ec%20%3D%20%5Cvarnothing%24%20is%20finite%20so%20%24X%20%5Cin%20M%24%3B%0A%20%20%20%202.%20If%20%24A%20%5Cin%20M%24%20then%20either%20%24A%20%3D%20(A%5Ec)%5Ec%24%20is%20at%20most%20countable%20or%20%24A%5Ec%24%20is%20at%20most%20countable%2C%20which%20by%20definition%20means%20%24A%5Ec%20%5Cin%20M%24%3B%0A%20%20%20%203.%20If%20%24A%20%3D%20%5Ccup%20A_i%24%20and%20%24A_i%20%5Cin%20M%24%20for%20all%20%24i%24%3A%0A%20%20%20%20%091.%20If%20%24A_i%24%20countable%20for%20all%20%24i%24%20then%20%24A%24%20is%20countable%20thus%20in%20%24M%24%3B%0A%20%20%20%20%092.%20If%20for%20at%20least%20one%20%24i%24%2C%20%24A_i%20%5Cin%20M%24%20because%20%24A_i%5Ec%24%20is%20countable%2C%20WLOG%20say%20%24A_1%5Ec%24%20is%20countable%2C%20then%20%24A%5Ec%20%3D%20(%5Ccup%20A_i)%5Ec%20%3D%20%5Ccap%20A_i%5Ec%20%5Csubseteq%20A_1%5Ec%24%20so%20%24A%5Ec%24%20is%20at%20most%20countable%2C%20thus%20%24A%20%5Csubset%20M%24.%0A%0A%20%20%20%20%24%5Cmu%24%20is%20a%20measure%20on%20%24M%24%3A%0A%0A%20%20%20%201.%20The%20range%20of%20%24%5Cmu%24%20is%20clearly%20in%20%24%5B0%2C%5Cinfty%5D%24%20and%20%24%5Cmu%24%20is%20non-trivial%3B%0A%20%20%20%202.%20Suppose%20%24%5Clbrace%20A_i%5Crbrace%24%20is%20a%20disjoint%20countable%20collection%20of%20members%20of%20%24M%24%2C%20then%20at%20most%20one%20of%20%24A_i%24%20is%20in%20%24M%24%20because%20%24A_i%5Ec%24%20is%20countable%20(if%20%24A_1%5Ec%24%20and%20%24A_2%5Ec%24%20are%20both%20countable%2C%20since%20%24A_1%2CA_2%24%20are%20disjoint%2C%20%24A_1%20%5Csubset%20A_2%5Ec%24%2C%20we%20have%20a%20contradiction%20because%20we%20assumed%20%24A_1%24%20is%20uncountable).%0A%20%20%20%20%09Now%20it%20is%20easy%20to%20see%20that%20if%20one%20%24A_i%5Ec%24%20is%20countable%20then%20%24%5Cmu(%5Ccup%20A_i)%20%3D%201%20%3D%20%5Csum%20%5Cmu(A_i)%24%20and%20if%20no%20%24A_i%5Ec%24%20is%20countable%20then%20%24%5Cmu(%5Ccup%20A_i)%20%3D%200%20%3D%20%5Csum%20%5Cmu(A_i)%24%2C%20so%20in%20either%20case%20we%20have%20countably%20additive.%0A%0A%20%20%20%20Suppose%20now%20%24f%3AX%5Cto%20%5Cmathbb%7BR%7D%24%20is%20a%20measurable%20function.%20For%20any%20%24a%20%5Cin%20%5Cmathbb%7BR%7D%24%2C%20since%20the%20set%20%24%5Clbrace%20a%5Crbrace%24%20can%20be%20written%20as%20intersection%20of%20countably%20many%20open%20sets%20(i.e.%20it%20is%20a%20%24G_%7B%5Cdelta%7D%24%2C%20in%20particular%2C%20it%20is%20a%20Borel%20set)%2C%20so%20%24f%5E%7B-1%7D(a)%20%3D%20E%24%20is%20in%20%24M%24%20by%20Theorem%201.12%2C%20now%3A%0A%0A%20%20%20%201.%20Suppose%20this%20%24E%24%20is%20uncountable%2C%20then%20%24E%5Ec%24%20is%20at%20most%20countable.%20We%20can%20see%20that%20%24%5Cint_X%20f%20d%5Cmu%20%3D%20%5Cint_Esd%5Cmu%24%20where%20%24s%24%20takes%20value%20%24a%24%20on%20%24E%24%20and%20any%20value%20on%20%24E%5Ec%24.%20In%20other%20words%2C%20as%20long%20as%20there%20is%20one%20%24a%24%20such%20that%20its%20preimage%20is%20uncountable%2C%20the%20integral%20of%20%24f%24%20over%20%24X%24%20is%20%24a%24%3B%0A%20%20%20%202.%20Suppose%20then%20for%20any%20%24a%20%5Cin%20%5Cmathbb%7BR%7D%24%2C%20%24f%5E%7B-1%7D(a)%20%3D%20E_a%24%20is%20countable%2C%20then%20we%20can%20write%20%24X%20%3D%20%5Cbigcup%5Climits_%7Ba%20%5Cin%20%5Cmathbb%7BR%7D%7DE_a%20%3D%20%5Csqcup_%7Ba%20%5Cin%20%5Cmathbb%7BR%7D%7DE_a%24%20(disjoint%20because%20otherwise%20%24f%24%20is%20not%20a%20function)%2C%20by%20removing%20all%20empty%20%24E_a%24%20we%20get%20%24X%20%3D%20%5Csqcup_%7Ba%20%5Cin%20I%7DE_a%24.%20Since%20%24E_a%24's%20are%20all%20countable%20yet%20%24X%24%20is%20uncountable%2C%20there%20must%20be%20uncountably%20many%20such%20%24a%24%2C%20i.e.%20%24f(X)%20%5Csubseteq%20%5Cmathbb%7BR%7D%24%20is%20uncountable.%20Choose%20some%20%24a%24%20such%20that%20%24f(X)%20%5Ccap%20(-%5Cinfty%2Ca)%24%20and%20%24f(X)%20%5Ccap%20(a%2C%5Cinfty)%24%20are%20both%20uncountable%2C%20then%20%24f%5E%7B-1%7D(-%5Cinfty%2Ca)%24%20and%20%24f%5E%7B-1%7D(a%2C%5Cinfty)%24%20are%20both%20uncountable%2C%20measurable%20and%20disjoint%2C%20we%20proved%20above%20that%20this%20cannot%20happen.%0A%0A%20%20%20%20%23%23%20Exercise%207%0A%0A%20%20%20%20**Suppose%20%24f_n%3AX%5Cto%20%5B0%2C%5Cinfty%5D%24%20is%20measurable%20for%20%24n%20%3D%201%2C2%2C3%2C%5Cdots%24%2C%20%24f_1%20%5Cge%20f_2%20%5Cge%20f_3%5Cge%20%5Cdots%20%5Cge%200%24%2C%20%24f_n(x)%20%5Cto%20f(x)%24%20as%20%24n%20%5Cto%20%5Cinfty%24%20for%20every%20%24x%5Cin%20X%24%2C%20and%20%24f_1%20%5Cin%20L%5E1(%5Cmu)%24.%20Prove%20that%20then%20%24%5Clim%5Climits_%7Bn%5Cto%20%5Cinfty%7D%5Cint_Xf_nd%5Cmu%20%3D%20%5Cint_Xfd%5Cmu%24%20and%20show%20that%20this%20conclusion%20does%20not%20follow%20if%20the%20condition%20%24f_1%20%5Cin%20L%5E1(%5Cmu)%24%20is%20omitted**.%0A%0A%20%20%20%20Proof%0A%0A%0A%20%20%20%20This%20is%20an%20immediate%20result%20of%20the%20Lebesgue%20Dominated%20Convergence%20Theorem.%0A%0A%20%20%20%20Consider%20%24X%20%3D%20%5B0%2C%5Cinfty)%24%20and%20%24%5Cmu%24%20be%20the%20Lebesgue%20measure.%20Consider%20the%20sequence%20of%20functions%3A%0A%0A%20%20%20%201.%20%24f_1%20%3D%201%5Cchi_%7B%5B0%2C1)%7D%2B%5Cfrac%7B1%7D%7B2%7D%20%5Cchi_%7B%5B1%2C2)%7D%2B%5Cfrac%7B1%7D%7B3%7D%5Cchi_%7B%5B2%2C3)%7D%2B%5Cdots%24%3B%0A%20%20%20%202.%20%24f_2%20%3D%20%5Cfrac%7B1%7D%7B2%7D%5Cchi_%7B%5B0%2C1)%7D%2B%5Cfrac%7B1%7D%7B4%7D%20%5Cchi_%7B%5B1%2C2)%7D%2B%5Cfrac%7B1%7D%7B6%7D%5Cchi_%7B%5B2%2C3)%7D%2B%5Cdots%24%3B%0A%20%20%20%203.%20%24f_3%20%3D%20%5Cfrac%7B1%7D%7B3%7D%5Cchi_%7B%5B0%2C1)%7D%2B%5Cfrac%7B1%7D%7B6%7D%20%5Cchi_%7B%5B1%2C2)%7D%2B%5Cfrac%7B1%7D%7B9%7D%5Cchi_%7B%5B2%2C3)%7D%2B%5Cdots%24%3B%0A%20%20%20%204.%20%24%5Cdots%24%0A%0A%20%20%20%20Each%20%24f_n%24%20is%20measurable%20and%20%24f_1%5Cge%20f_2%5Cge%20%5Cdots%24.%20It%20is%20easy%20to%20see%20%24f_n%20%5Cto%200%24.%0A%0A%20%20%20%20We%20know%20that%20%24%5Csum%5Climits_%7Bn%3D1%7D%5E%7B%5Cinfty%7D%5Cfrac%7B1%7D%7Bn%7D%20%3D%20%5Cinfty%24%2C%20thus%20%24%5Cint_Xf_nd%5Cmu%20%3D%20%5Cinfty%24%20for%20any%20%24n%24%20(so%20in%20particular%2C%20%24f_1%24%20is%20not%20%24L%5E1%24)%2C%20and%20so%20does%20its%20limit.%20On%20the%20other%20hand%2C%20%24%5Cint_X%200%20d%5Cmu%20%3D%200%20%5Cne%20%5Cinfty%24.%0A%0A%20%20%20%20%23%23%20Exercise%208%0A%0A%20%20%20%20**Put%20%24f_n%20%3D%20%5Cchi_E%24%20if%20%24n%24%20is%20odd%2C%20%24f_n%20%3D%201-%5Cchi_E%24%20if%20%24n%24%20is%20even.%20What%20is%20the%20relevance%20of%20this%20example%20to%20Fatou's%20lemma**%3F%0A%0A%20%20%20%20Solution%0A%0A%20%20%20%20For%20example%2C%20consider%20%24X%20%3D%20%5B0%2C1%5D%24%2C%20%24E%20%3D%20%5B0%2C1%2F2%5D%24%20with%20the%20Lebesgue%20measure%2C%20then%20%24%5Cint_X%20f_%7Bodd%7Dd%5Cmu%20%3D%20%5Cint_X%20f_%7Beven%7Dd%20%5Cmu%20%3D%201%2F2%24%20so%20%24%5Climinf%20%5Cint_X%20f_n%20d%5Cmu%20%3D%201%2F2%24.%20On%20the%20other%20hand%20for%20each%20%24x%20%5Cin%20X%24%2C%20%24%5Climinf%20f_n(x)%20%3D%200%24%20thus%20%24%5Cint_X%20(%5Climinf%20f_n)d%5Cmu%20%3D%200%24%20so%20we%20have%20strict%20inequality.%0A%0A%20%20%20%20%23%23%20Exercise%209%0A%0A%20%20%20%20**Suppose%20%24%5Cmu%24%20is%20a%20positive%20measure%20on%20%24X%24%2C%20%24f%3AX%5Cto%20%5B0%2C%5Cinfty%5D%24%20is%20measurable%2C%20%24%5Cint_Xfd%5Cmu%20%3D%20c%24%20where%20%240%20%3Cc%3C%5Cinfty%24%2C%20and%20%24%5Calpha%24%20is%20a%20constant.%20Prove%20that%20%24%5Clim%5Climits_%7Bn%5Cto%20%5Cinfty%7D%5Cint_Xn%5Clog%5B1%2B(f%2Fn)%5E%7B%5Calpha%7D%5Dd%5Cmu%20%3D%20%5Cbegin%7Bcases%7D%5Cinfty%2C%260%3C%5Calpha%3C1%5C%5Cc%2C%26%5Calpha%20%3D%201%5C%5C0%2C%261%3C%5Calpha%3C%5Cinfty%5Cend%7Bcases%7D%24**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20First%20notice%20that%20%24%5Cint_Xfd%5Cmu%20%3D%20c%24%20implies%20that%20%24%5Cmu(X)%20%5Cne%200%24%20so%20the%20integral%20is%20not%20trivial.%0A%0A%20%20%20%20Suppose%20%24%5Calpha%5Cge%201%24%2C%20if%20%24x%20%3D%200%24%20then%20%24%5Calpha%20x%20%3D%20n%5Clog%5B1%2B(x%2Fn)%5E%7B%5Calpha%7D%5D%24%2C%20and%20%24(%5Calpha%20x%20-%20n%5Clog%5B1%2B(x%2Fn)%5E%7B%5Calpha%7D%5D)'%20%3D%20%5Calpha%20-%20%5Cfrac%7B%5Calpha(x%2Fn)%5E%7B%5Calpha-1%7D%7D%7B(x%2Fn)%5E%7B%5Calpha%7D%2B1%7D%3E0%24%20for%20%24x%3E0%24%2C%20and%20any%20%24n%5Cge%201%24%2C%20thus%20we%20may%20conclude%20that%20%24n%5Clog%5B1%2B(f%2Fn)%5E%7B%5Calpha%7D%5D%20%3C%5Calpha%20f%24.%0A%0A%20%20%20%20Now%20if%20%24%5Calpha%20%3D%201%24%20then%20%24%5Clim%5Climits_%7Bn%5Cto%20%5Cinfty%7D%20n%5Clog%5B1%2B(x%2Fn)%5D%20%3D%20%5Clim%5Climits%20%5Cfrac%7B%5Clog(1%2Bx%2Fn)%7D%7B1%2Fn%7D%20%3D%20%5Clim%5Climits%20%5Cfrac%7Bnx%7D%7Bn%2Bx%7D%20%3D%20x%24%20by%20L'Hospital%3B%0A%0A%20%20%20%20If%20%24%5Calpha%3E1%24%2C%20similarly%20we%20have%20%24%5Clim%5Climits_%7Bn%5Cto%20%5Cinfty%7D%20n%5Clog%5B1%2B(x%2Fn)%5E%7B%5Calpha%7D%5D%20%3D%20%5Clim%5Climits%20%5Cfrac%7B-%5Cfrac%7B%5Calpha(x%2Fn)%5E%7B%5Calpha%7D%7D%7Bn%2Bn(x%2Fn)%5E%7B%5Calpha%7D%7D%7D%7B-1%2Fn%5E2%7D%20%3D%20%5Clim%5Climits%20%5Cfrac%7B%5Calpha%20n%7D%7B(n%2Fx)%5E%7B%5Calpha%7D%2B1%7D%24.%20We%20can%20actually%20see%20here%20that%20if%20%24%5Calpha%3E1%24%20then%20we%20can%20cancel%20the%20%24n%24%20on%20the%20numerator%20and%20get%20the%20limit%20is%20%240%24%2C%20while%20if%20%24%5Calpha%3C1%24%20then%20we%20cancel%20the%20%24n%24%20on%20the%20denominator%20and%20the%20limit%20is%20%24%5Cinfty%24.%0A%0A%20%20%20%20So%20by%20Dominated%20Convergence%20Theorem%2C%20we%20get%20the%20parts%3A%20%24%5Clim%5Climits_%7Bn%5Cto%20%5Cinfty%7D%5Cint_%7BX%7Dn%5Clog%5B1%2B(f%2Fn)%5E%7B%5Calpha%7D%5Dd%5Cmu%20%3D%20%5Cbegin%7Bcases%7D%5Cint_%7BX%20-%20f%5E%7B-1%7D(%5Cinfty)%7Dfd%5Cmu%3Dc%2C%26%5Calpha%20%3D%201%5C%5C%5Cint_%7BX%20-%20f%5E%7B-1%7D(%5Cinfty)%7D0d%5Cmu%3D0%2C%261%3C%5Calpha%3C%5Cinfty%5Cend%7Bcases%7D%24%20(notice%20we%20integrate%20over%20%24X%20-%20f%5E%7B-1%7D(%5Cinfty)%24%2C%20but%20it%20does%20not%20really%20matter%20because%20we%20have%20the%20assumption%20%24%5Cint_Xfd%5Cmu%20%3D%20c%24%20so%20%24%5Cmu(f%5E%7B-1%7D(%5Cinfty))%24%20must%20be%20%240%24).%0A%0A%20%20%20%20Otherwise%20if%20%24%5Calpha%3C1%24%2C%20then%20consider%20Fatou's%20Lemma%3A%20%24%5Clim%5Climits_%7Bn%5Cto%20%5Cinfty%7D%5Cint_Xn%5Clog%5B1%2B(f%2Fn)%5E%7B%5Calpha%7D%5Dd%5Cmu%24%20%24%3D%20%5Climinf_%7Bn%5Cto%20%5Cinfty%7D%5Cint_Xn%5Clog%5B1%2B(f%2Fn)%5E%7B%5Calpha%7D%5Dd%5Cmu%24%20%24%5Cge%20%5Cint_X%20(%5Climinf_%7Bn%5Cto%20%5Cinfty%7Dn%5Clog%5B1%2B(f%2Fn)%5E%7B%5Calpha%7D%5D)d%5Cmu%20%3D%20%5Cinfty%24%20because%20by%20above%20we%20have%20%24%5Clim%5Climits_%7Bn%5Cto%20%5Cinfty%7Dn%5Clog%5B1%2B(f%2Fn)%5E%7B%5Calpha%7D%5D%20%3D%20%5Cinfty%24%20if%20%24%5Calpha%20%3C%201%24%2C%20and%20we%20complete%20the%20proof.%0A%0A%20%20%20%20%23%23%20Exercise%2010%0A%0A%20%20%20%20**Suppose%20%24%5Cmu(X)%3C%5Cinfty%24%2C%20%24%5Clbrace%20f_n%5Crbrace%24%20is%20a%20sequence%20of%20bounded%20complex%20measurable%20functions%20on%20%24X%24%2C%20and%20%24f_n%20%5Cto%20f%24%20uniformly%20on%20%24X%24.%20Prove%20that%20%24%5Clim%5Climits_%7Bn%20%5Cto%20%5Cinfty%7D%5Cint_Xf_nd%5Cmu%20%3D%20%5Cint_Xfd%5Cmu%24%20and%20show%20that%20the%20hypothesis%20%24%5Cmu(X)%20%3C%5Cinfty%24%20cannot%20be%20omitted**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20Since%20we%20have%20uniform%20convergence%2C%20for%20%24%5Cvarepsilon%24%20choose%20%24N%24%20such%20that%20%24%7C%20f_N(x)%20-%20f(x)%20%7C%20%3C%201%20%5Cimplies%20%7C%20f(x)%20%7C%20%3C%20%7C%20f_N(x)%20%7C%20%2B%201%24%20for%20all%20%24x%20%5Cin%20X%24%2C%20so%20%24%5Cint_X%20%7C%20f%20%7C%20d%5Cmu%24%20%24%3C%5Cint_X%20%7C%20f_N(x)%20%7C%20d%5Cmu%2B%20%5Cint_X%201d%5Cmu%24%20%24%3D%5Cint_X%20%7C%20f_N(x)%20%7C%20d%5Cmu%20%2B%20%5Cmu(X)%24%20%24%3C%5Cinfty%24%20by%20the%20assumptions%2C%20so%20%24f%24%20is%20also%20%24L%5E1%24.%0A%0A%20%20%20%20For%20each%20%24n%24%2C%20%24%7C%20%5Cint_X%20f_n%20d%5Cmu%20-%20%5Cint_X%20f%20d%5Cmu%20%7C%24%20%24%5Cle%20%5Cint_X%20%7C%20f_n-f%20%7C%20d%5Cmu%24%20%24%5Cle%20%5Cint_X%5Csup%5Climits_%7Bx%7D(%7C%20f_n-f%20%7C)d%5Cmu%24%20%24%3D%5Csup%5Climits_%7Bx%7D(%7C%20f_n-f%20%7C)%5Cmu(X)%24.%20If%20we%20pass%20limit%20then%20by%20definition%20of%20uniform%20convergence%20%24%5Clim%5Climits_%7Bn%5Cto%20%5Cinfty%7D%5Csup%5Climits_%7Bx%7D(%7C%20f_n-f%20%7C)%20%3D%200%24%2C%20thus%20we%20have%20%24%5Clim%5Climits%20%7C%20%5Cint_X%20f_n%20d%5Cmu%20-%20%5Cint_X%20f%20d%5Cmu%20%7C%20%3D%200%24%20%24%5Cimplies%20%5Clim%5Climits%20%5Cint_Xf_nd%5Cmu%20%3D%20%5Cint_Xfd%5Cmu%24.%0A%0A%20%20%20%20The%20example%20from%20exercise%207%20also%20works%20here%3A%0A%0A%20%20%20%20Consider%20%24X%20%3D%20%5B0%2C%5Cinfty)%24%20and%20%24%5Cmu%24%20be%20the%20Lebesgue%20measure%2C%20then%20we%20do%20not%20have%20%24%5Cmu(X)%20%3C%20%5Cinfty%24.%20Consider%20the%20sequence%20of%20functions%3A%0A%0A%20%20%20%201.%20%24f_1%20%3D%201%5Cchi_%7B%5B0%2C1)%7D%2B%5Cfrac%7B1%7D%7B2%7D%20%5Cchi_%7B%5B1%2C2)%7D%2B%5Cfrac%7B1%7D%7B3%7D%5Cchi_%7B%5B2%2C3)%7D%2B%5Cdots%24%3B%0A%20%20%20%202.%20%24f_2%20%3D%20%5Cfrac%7B1%7D%7B2%7D%5Cchi_%7B%5B0%2C1)%7D%2B%5Cfrac%7B1%7D%7B4%7D%20%5Cchi_%7B%5B1%2C2)%7D%2B%5Cfrac%7B1%7D%7B6%7D%5Cchi_%7B%5B2%2C3)%7D%2B%5Cdots%24%3B%0A%20%20%20%203.%20%24f_3%20%3D%20%5Cfrac%7B1%7D%7B3%7D%5Cchi_%7B%5B0%2C1)%7D%2B%5Cfrac%7B1%7D%7B6%7D%20%5Cchi_%7B%5B1%2C2)%7D%2B%5Cfrac%7B1%7D%7B9%7D%5Cchi_%7B%5B2%2C3)%7D%2B%5Cdots%24%3B%0A%20%20%20%204.%20%24%5Cdots%24%0A%0A%20%20%20%20We%20know%20that%20%24%5Csum%5Climits_%7Bn%3D1%7D%5E%7B%5Cinfty%7D%5Cfrac%7B1%7D%7Bn%7D%20%3D%20%5Cinfty%24%2C%20thus%20%24%5Cint_Xf_nd%5Cmu%20%3D%20%5Cinfty%24%20for%20any%20%24n%24%2C%20and%20so%20does%20its%20limit.%20On%20the%20other%20hand%2C%20for%20any%20%24%5Cvarepsilon%3E0%24%2C%20choose%20%24N%24%20large%20enough%20so%20that%20%241%2FN%3C%5Cvarepsilon%24%2C%20then%20for%20any%20%24n%3EN%24%2C%20%24%7C%20f_n(x)%20-%200%20%7C%20%3C%20%5Cvarepsilon%24%2C%20so%20%24f_n%24%20converges%20uniformly%20to%20%240%24%2C%20but%20%24%5Cint_X%200%20d%5Cmu%20%3D%200%20%5Cne%20%5Cinfty%24.%0A%0A%20%20%20%20%23%23%20Exercise%2011%0A%0A%20%20%20%20**Show%20that%20%24A%20%3D%20%5Cbigcap%5Climits_%7Bn%20%3D%201%7D%5E%7B%5Cinfty%7D%5Cbigcup%5Climits_%7Bk%20%3D%20n%7D%5E%7B%5Cinfty%7DE_k%24%20in%20Theorem%201.41%2C%20and%20hence%20prove%20the%20theorem%20without%20any%20reference%20to%20integration.**%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20(%24A%24%20is%20defined%20to%20be%20the%20set%20of%20all%20%24x%24%20which%20lie%20in%20infinitely%20many%20%24E_k%24%20where%20%24E_k%24%20is%20a%20sequence%20of%20measurable%20sets%20in%20%24X%24)%0A%0A%20%20%20%20Suppose%20%24x%5Cin%20A%24%2C%20then%20%24x%5Cin%20E_k%24%20for%20infinitely%20many%20%24k%24%2C%20namely%20%24k_1%3Ck_2%3C%5Cdots%24.%20We%20must%20have%20%24k_1%5Cge%201%2C%20k_2%5Cge%202%2C%20%5Cdots%24%2C%20so%20%24x%20%5Cin%20E_%7Bk_1%7D%20%5Cimplies%20x%5Cin%20%5Cbigcup%5Climits_%7Bk%3D1%7D%5E%7B%5Cinfty%7DE_k%24%2C%20%24x%20%5Cin%20E_%7Bk_2%7D%20%5Cimplies%20x%5Cin%20%5Cbigcup%5Climits_%7Bk%3D2%7D%5E%7B%5Cinfty%7DE_k%24%2C%20%24%5Cdots%24%2C%20so%20%24x%24%20is%20in%20the%20intersection.%0A%0A%20%20%20%20For%20the%20other%20inclusion%2C%20suppose%20%24x%5Cin%20%5Cbigcap%5Climits_%7Bn%20%3D%201%7D%5E%7B%5Cinfty%7D%5Cbigcup%5Climits_%7Bk%20%3D%20n%7D%5E%7B%5Cinfty%7DE_k%24%2C%20then%20%24x%20%5Cin%20%5Cbigcup%5Climits_%7Bk%20%3D%20n%7D%5E%7B%5Cinfty%7DE_k%24%20for%20each%20%24n%24.%20Suppose%20%24x%5Cin%20%5Cbigcup%5Climits_%7Bk%20%3D%201%7D%5E%7B%5Cinfty%7DE_k%24%20because%20%24x%20%5Cin%20E_%7Bk_1%7D%24%2C%20then%20in%20order%20for%20%24x%24%20to%20lie%20in%20%24%5Cbigcup%5Climits_%7Bk%20%3D%20k_1%2B1%7D%5E%7B%5Cinfty%7DE_k%24%2C%20we%20need%20to%20have%20%24x%20%5Cin%20E_%7Bk_2%7D%24%20for%20some%20%24k_2%3Ek_1%24.%20Continue%20and%20we%20have%20a%20sequence%20%24k_1%3Ck_2%3C%5Cdots%24%20and%20%24x%24%20lies%20in%20all%20of%20%24E_%7Bk_i%7D%24's%2C%20thus%20by%20definition%20%24x%20%5Cin%20A%24.%0A%0A%20%20%20%20Now%20by%20the%20assumption%20we%20have%20%24%5Cmu(E_k)%20%5Cto%200%24%2C%20thus%20%24%5Cmu(%5Cbigcup%5Climits_%7Bk%3Dn%7D%5E%7B%5Cinfty%7DE_k)%20%5Cto%200%24%20and%20%24%5Cmu(A)%20%3D%200%24%2C%20and%20the%20theorem%20is%20proved.%0A%0A%20%20%20%20%23%23%20Exercise%2012%0A%0A%20%20%20%20**Suppose%20%24f%20%5Cin%20L%5E1(%5Cmu)%24.%20Prove%20that%20to%20each%20%24%5Cvarepsilon%3E0%24%20there%20exists%20a%20%24%5Cdelta%3E0%24%20such%20that%20%24%5Cint_E%7C%20f%20%7C%20d%5Cmu%20%3C%20%5Cvarepsilon%24%20whenever%20%24%5Cmu(E)%20%3C%20%5Cdelta%24**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20Fix%20some%20%24%5Cvarepsilon%3E0%24.%0A%0A%20%20%20%20For%20any%20(integer)%20%24n%24%20we%20may%20write%20%24%5Cint_E%20%7C%20f%20%7C%20d%5Cmu%20%5Cle%20n%5Cmu(E)%2B%20%5Cint_%7B%7C%20f%20%7C%5E%7B-1%7D(n%2C%5Cinfty)%5Ccap%20E%7D%7C%20f%20%7C%20d%5Cmu%24.%0A%0A%20%20%20%20First%20we%20look%20at%20the%20second%20part%20%24%5Cint_%7B%7C%20f%20%7C%5E%7B-1%7D(n%2C%5Cinfty)%5Ccap%20E%7D%7C%20f%20%7C%20d%5Cmu%24.%20For%20different%20values%20of%20%24n%24%20we%20can%20see%20the%20integrals%20as%20integrals%20of%20a%20sequence%20%24f_n%24%20(which%20is%20now%20nonnegative)%20over%20%24E%24%2C%20where%20each%20%24f_n%20%3D%20f%24%20if%20%24f(x)%3En%24%20and%20%24%3D0%24%20otherwise%2C%20then%20%24%5Clim%5Climits_%7Bn%5Cto%20%5Cinfty%7Df_n(x)%20%3D%200%24%20for%20any%20%24x%20%5Cin%20E%24%20and%20%24%7C%20f_n(x)%20%7C%20%3D%20f_n(x)%24%20is%20less%20than%20or%20equal%20to%20the%20original%20%24%7C%20f(x)%20%7C%24%20which%20is%20%24L%5E1%24%20by%20the%20assumption%2C%20so%20we%20can%20apply%20the%20Lebesgue%20Dominated%20Convergence%20Theorem%20and%20see%20that%20%24%5Clim%5Climits_%7Bn%5Cto%20%5Cinfty%7D%5Cint_E%20f_n%20d%5Cmu%20%3D%200%24.%20By%20definition%20of%20a%20limit%2C%20we%20can%20find%20large%20enough%20%24n%24%20so%20that%20%24%5Cint_%7B%7C%20f%20%7C%5E%7B-1%7D(n%2C%5Cinfty)%5Ccap%20E%7D%7C%20f%20%7C%20d%5Cmu%20%3D%20%5Cint_Ef_nd%5Cmu%20%3C%20%5Cvarepsilon%2F2%24.%0A%0A%20%20%20%20For%20such%20%24n%24%2C%20assign%20%24%5Cdelta%20%3D%20%5Cvarepsilon%2F(2n)%24%2C%20this%20is%20a%20%24%5Cdelta%24%20we%20are%20looking%20for%3A%20as%20long%20as%20%24%5Cmu(E)%20%3C%20%5Cdelta%24%20we%20will%20have%20%24%5Cint_E%20%7C%20f%20%7C%20d%5Cmu%20%5Cle%20n%5Cmu(E)%2B%20%5Cint_%7B%7C%20f%20%7C%5E%7B-1%7D(n%2C%5Cinfty)%5Ccap%20E%7D%7C%20f%20%7C%20d%5Cmu%20%3C%20%5Cvarepsilon%2F2%2B%5Cvarepsilon%2F2%20%3D%20%5Cvarepsilon%24.%0A%20%20%20%20%22%22%22)%0A%20%20%20%20return%0A%0A%0A%40app.cell%0Adef%20_(mo)%3A%0A%20%20%20%20mo.md(r%22%22%22%0A%20%20%20%20%23%20Fall%20Semester%20Homework%202%0A%20%20%20%20%22%22%22)%0A%20%20%20%20return%0A%0A%0A%40app.cell%0Adef%20_(mo)%3A%0A%20%20%20%20mo.md(r%22%22%22%0A%20%20%20%20%23%23%20Exercise%201%0A%0A%20%20%20%20**Let%20%24%5Clbrace%20f_n%5Crbrace%24%20be%20a%20sequence%20of%20real%20nonnegative%20functions%20on%20%24%5Cmathbb%7BR%7D%24%2C%20and%20consider%20the%20following%20four%20statements**%3A%0A%0A%20%20%20%201.%20**If%20%24f_1%24%20and%20%24f_2%24%20are%20upper%20semicontinuous%2C%20then%20%24f_1%2Bf_2%24%20is%20upper%20semicontinuous**%3B%0A%20%20%20%202.%20**If%20%24f_1%24%20and%20%24f_2%24%20are%20lower%20semicontinuous%2C%20then%20%24f_1%2Bf_2%24%20is%20lower%20semicontinuous**%3B%0A%20%20%20%203.%20**If%20each%20%24f_n%24%20is%20upper%20semicontinuous%2C%20then%20%24%5Csum%5Climits_1%5E%7B%5Cinfty%7Df_n%24%20is%20upper%20semicontinuous**%3B%0A%20%20%20%204.%20**If%20each%20%24f_n%24%20is%20lower%20semicontinuous%2C%20then%20%24%5Csum%5Climits_1%5E%7B%5Cinfty%7Df_n%24%20is%20lower%20semicontinuous**.%0A%0A%20%20%20%20**Show%20that%20three%20of%20these%20are%20true%20and%20that%20one%20is%20false.%20What%20happens%20if%20the%20word%20%22nonnegative%22%20is%20omitted%3F%20Is%20the%20truth%20of%20the%20statements%20affected%20if%20%24%5Cmathbb%7BR%7D%24%20is%20replaced%20by%20a%20general%20topological%20space**%3F%0A%0A%20%20%20%20Solution%0A%0A%20%20%20%20For%20the%20first%20part%2C%20the%20answer%20will%20be%20that%201%2C2%2C4%20true%20and%203%20false%3B%20so%20I%20will%20answer%20in%20this%20order%20to%20illustrate%20why%203%20is%20false.%0A%0A%20%20%20%20Let%20%24a%24%20be%20an%20arbitrary%20real%20number%3A%0A%0A%20%20%20%201.%20The%20set%20%24%5Clbrace%20x%7C%20(f_1%2Bf_2)(x)%20%3C%20a%20%5Crbrace%24%20can%20be%20written%20as%20%24%3D%20%5Cbigcup%5Climits_%7Bb%20%5Cin%20%5Cmathbb%7BR%7D%7D(%5Clbrace%20x%20%7C%20f_1(x)%20%3C%20b%5Crbrace%20%5Ccap%20%5Clbrace%20x%20%7C%20f_2(x)%20%3C%20a%20-%20b%20%5Crbrace)%24%2C%20the%20union%20can%20be%20taken%20over%20%24b%20%5Cin%20%5Cmathbb%7BQ%7D%24%20instead%3A%20%24%3D%20%5Cbigcup%5Climits_%7Bb%20%5Cin%20%5Cmathbb%7BQ%7D%7D(%5Clbrace%20x%20%7C%20f_1(x)%20%3C%20b%5Crbrace%20%5Ccap%20%5Clbrace%20x%20%7C%20f_2(x)%20%3C%20a%20-%20b%20%5Crbrace)%24%2C%20this%20is%20a%20countable%20union%20of%20finite%20intersection%20of%20open%20sets%20thus%20open%2C%20and%20thus%20%24f_1%2Bf_2%24%20is%20upper%20semicontinuous%3B%0A%20%20%20%202.%20The%20idea%20is%20the%20same%2C%20write%20%24%5Clbrace%20x%7C%20(f_1%2Bf_2)(x)%20%3E%20a%20%5Crbrace%24%20as%20%24%3D%20%5Cbigcup%5Climits_%7Bb%20%5Cin%20%5Cmathbb%7BQ%7D%7D(%5Clbrace%20x%20%7C%20f_1(x)%20%3E%20b%5Crbrace%20%5Ccap%20%5Clbrace%20x%20%7C%20f_2(x)%20%3E%20a%20-%20b%20%5Crbrace)%24%20and%20realize%20this%20is%20a%20countable%20union%20of%20finite%20intersection%20of%20open%20sets%20thus%20open%2C%20and%20thus%20%24f_1%2Bf_2%24%20is%20lower%20semicontinuous%3B%0A%20%20%20%203.%20(part%204)%20For%20each%20positive%20integer%20%24n%24%2C%20write%20%24%5Clbrace%20x%7C%20(f_1%2B%5Cdots%20%2B%20f_n)(x)%20%3E%20%5Calpha%20%5Crbrace%24%20as%20%24%3D%20%5Cbigcup%5Climits_%7Bb_1%2C%5Cdots%2Cb_%7Bn-1%7D%20%5Cin%20%5Cmathbb%7BQ%7D%7D(%5Clbrace%20x%20%7C%20f_1(x)%20%3E%20b_1%20%5Crbrace%20%5Cbigcap%5Climits%24%20%24%5Cdots%20%5Ccap%20%5Clbrace%20x%20%7C%20f_%7Bn-1%7D(x)%20%3E%20b_%7Bn-1%7D%20%5Crbrace%24%20%24%5Ccap%20%5Clbrace%20x%20%7C%20f_n(x)%20%3E%20a%20-%20(b_1%20%2B%20%5Cdots%20%2B%20b_%7Bn-1%7D)%20%5Crbrace)%24.%20So%20we%20can%20see%20that%20%24f_1%20%2B%20%5Cdots%20%2B%20f_n%24%20is%20lower%20semicontinuous.%20Write%20%24%5Csum%5Climits_1%5E%7B%5Cinfty%7Df_n%24%20%24%3D%5Clim%5Climits_%7Bn%5Cto%20%5Cinfty%7D(f_1%20%2B%20%5Cdots%20%2B%20f_n)%24%20and%20use%20the%20assumption%20that%20the%20functions%20are%20all%20non-negative%2C%20we%20can%20continue%20writing%20%24%3D%20%5Csup%5Climits_n%20(f_1%20%2B%20%5Cdots%20%2B%20f_n)%24.%20We%20may%20conclude%20using%20the%20fact%20that%20supremum%20of%20lower%20semicontinuous%20function%20is%20lower%20semicontinuous%3B%0A%20%20%20%204.%20(part%203)%20The%20above%20argument%20doesn't%20work%20with%20upper%20semicontinuous%20because%20of%20the%20last%20few%20steps%3A%20with%20the%20non-negative%20assumption%2C%20apparently%20the%20limit%20%24%5Clim%5Climits_%7Bn%5Cto%20%5Cinfty%7D(f_1%20%2B%20%5Cdots%20%2B%20f_n)%24%20is%20in%20general%20not%20the%20infimum%20%24%3D%20%5Cinf%5Climits_n%20(f_1%20%2B%20%5Cdots%20%2B%20f_n)%20%3D%20f_1%24.%20To%20see%20the%20statement%20is%20indeed%20not%20true%2C%20consider%20the%20series%20of%20functions%20%24f_n%20%3D%20%5Cchi_%7B%5B%5Cfrac%7B1%7D%7Bn%2B1%7D%2C%5Cfrac%7B1%7D%7Bn%7D%5D%7D%24%20so%20that%20each%20%24f_n%24%20is%20upper%20semicontinuous%20yet%20the%20infinite%20summation%20%24%5Csum%20f_n%20%3D%20%5Cchi_%7B(0%2C1%5D%7D%24%20is%20not.%0A%0A%20%20%20%20If%20we%20lift%20the%20assumption%20of%20non-negative%2C%20part%201%2C2%20won't%20be%20affect%20because%20we%20never%20use%20this%20assumption.%20part%203%20is%20of%20course%20still%20false%2C%20and%20from%20the%20construction%20of%20the%20counterexample%20for%20it%2C%20we%20should%20be%20able%20to%20see%20that%20the%20non-negative%20assumption%20is%20crucial%20for%20part%204.%20Here%20is%20a%20counterexample%3A%0A%0A%20%20%20%20Let%20%24f_n%24%20be%20%24%5Cchi_%7B(%5Cfrac%7B1%7D%7Bn%2B1%7D%2C%5Cfrac%7B1%7D%7Bn%7D)%7D%24%20if%20%24n%24%20is%20odd%2C%20and%20%24-%5Cchi_%7B%5B%5Cfrac%7B1%7D%7Bn%2B1%7D%2C%5Cfrac%7B1%7D%7Bn%7D%5D%7D%24%20if%20%24n%24%20is%20even.%20Then%20for%20any%20%24-1%3Ca%3C0%24%2C%20%24%5Clbrace%20x%20%7C%20%5Csum%20f_n%20%3E%20a%5Crbrace%24%20%24%3D%20%5Clbrace%20x%20%7C%20%5Csum%20f_n%20%3D%200%5Crbrace%20%5Ccup%20%5Clbrace%20x%20%7C%20%5Csum%20f_n%20%3D%201%5Crbrace%24%20%24%3D(-%5Cinfty%2C0%5D%20%5Ccup%20(1%2F3%2C1%2F2)%20%5Ccup%20(1%2F5%2C1%2F4)%20%5Ccup%20%5Cdots%24%20is%20not%20open.%0A%0A%20%20%20%20This%20argument%20depends%20on%20the%20fact%20that%20codomain%20%24%5Cmathbb%7BR%7D%24%20is%20separable%20(having%20a%20countable%20dense%20subset%20%24%5Cmathbb%7BQ%7D%24)%2C%20but%20not%20the%20domain.%20As%20long%20as%20%24f%24%20are%20still%20real%20valued%20functions%20(i.e.%20codomain%20not%20changing)%2C%20the%20choice%20of%20the%20domain%20is%20not%20important.%0A%0A%20%20%20%20%23%23%20Exercise%202%0A%0A%20%20%20%20**Let%20%24f%24%20be%20an%20arbitrary%20complex%20function%20on%20%24%5Cmathbb%7BR%7D%24%2C%20and%20define%20%24%5Cvarphi(x%2C%5Cdelta)%20%3D%20%5Csup%5Climits%5Clbrace%20%7C%20f(s)%20-%20f(t)%20%7C%3As%2Ct%20%5Cin(x-%5Cdelta%2Cx%2B%5Cdelta)%5Crbrace%24%2C%20%24%5Cvarphi(x)%20%3D%20%5Cinf%5Clbrace%20%5Cvarphi(x%2C%5Cdelta)%3A%20%5Cdelta%3E0%5Crbrace%24.%20Prove%20that%20%24%5Cvarphi%24%20is%20upper%20semicontinuous%2C%20that%20%24f%24%20is%20continuous%20at%20a%20point%20%24x%24%20if%20and%20only%20if%20%24%5Cvarphi(x)%20%3D%200%24%2C%20and%20hence%20that%20the%20set%20of%20points%20of%20continuity%20of%20an%20arbitrary%20complex%20function%20is%20a%20%24G_%7B%5Cdelta%7D%24**.%0A%0A%20%20%20%20**Formulate%20and%20prove%20an%20analogous%20statement%20for%20general%20topological%20spaces%20in%20place%20of%20%24%5Cmathbb%7BR%7D%24**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%201.%20%24%5Cvarphi%24%20is%20upper%20semicontinuous%3A%0A%20%20%20%20%09First%20notice%20that%20by%20construction%2C%20%24%5Cvarphi(x%2C%5Cdelta)%20%3C%20%5Cvarphi(x%2C%5Cdelta')%24%20whenever%20%24%5Cdelta%20%3C%20%5Cdelta'%24%3B%20thus%20%24%5Cvarphi(x)%20%3D%20%5Cinf%5Clbrace%20%5Cvarphi(x%2C%5Cdelta)%3A%5Cdelta%20%3E%200%5Crbrace%20%3D%20%5Clim%5Climits_%7B%5Cdelta%20%5Cto%200%7D%5Cvarphi(x%2C%5Cdelta)%24.%0A%20%20%20%20%09Let%20%24a%20%5Cin%20%5Cmathbb%7BR%7D%24%20and%20consider%20the%20set%20%24A%20%3D%20%5Clbrace%20x%20%7C%20%5Cvarphi(x)%20%3C%20a%20%5Crbrace%24%2C%20we%20want%20to%20show%20this%20set%20is%20open.%20If%20%24A%24%20is%20empty%20then%20it%20is%20automatically%20true%2C%20otherwise%20let%20%24x_0%20%5Cin%20A%24%2C%20it%20suffices%20to%20find%20a%20neighborhood%20of%20%24x_0%24%20contained%20in%20%24A%24.%0A%20%20%20%20%09We%20have%20%24a%3E%20%5Cvarphi(x_0)%20%3D%20%5Clim%5Climits_%7B%5Cdelta%20%5Cto%200%7D%20%5Cvarphi(x_0%2C%5Cdelta)%24.%20Pick%20an%20%24%5Cdelta_0%24%20so%20that%20%24%5Cvarphi(x_0%2C%5Cdelta_0)%20%3C%20a%24%2C%20then%20%24N(x_0)%20%3D%20(x_0%20-%20%5Cdelta_0%2F2%2C%20x_0%20%2B%20%5Cdelta_0%2F2)%24%20is%20a%20neighborhood%20of%20%24x_0%24%20we%20want%3A%0A%20%20%20%20%09Suppose%20%24x_1%20%5Cin%20N(x_0)%24%2C%20then%20%24%24%5Cbegin%7Baligned%7D%5Cvarphi(x_1)%20%3D%26%20%5Clim%5Climits_%7B%5Cdelta%5Cto%200%7D%5Cvarphi(x_1%2C%5Cdelta)%20%5C%5C%5Cle%26%20%5Cvarphi(x_1%2C%5Cdelta_0%2F2)%5C%5C%3D%26%5E%7Bdef%7D%20%5Csup%5Climits%5Clbrace%20%7C%20f(s)%20-%20f(t)%20%7C%3As%2Ct%20%5Cin(x_1-%5Cdelta_0%2F2%2Cx_1%2B%5Cdelta_0%2F2)%5Crbrace%5Cend%7Baligned%7D.%24%24%20Now%20notice%20%24(x_1-%5Cdelta_0%2F2%2Cx_1%2B%5Cdelta_0%2F2)%24%20is%20entirely%20contained%20in%20%24(x_0%20-%20%5Cdelta_0%2C%20x_0%20%2B%20%5Cdelta_0)%24%2C%20thus%20we%20may%20continue%20write%20%24%24%5Cbegin%7Baligned%7D%5Cle%26%5Csup%5Climits%5Clbrace%20%7C%20f(s)%20-%20f(t)%20%7C%3As%2Ct%20%5Cin(x_0-%5Cdelta_0%2Cx_0%2B%5Cdelta_0)%5Crbrace%5C%5C%3D%26%5E%7Bdef%7D%5Cvarphi(x_0%2C%5Cdelta_0)%5C%5C%3C%26a%5Cend%7Baligned%7D%24%24%0A%20%20%20%20%09by%20our%20construction.%20Thus%20%24x_1%20%5Cin%20A%24%20and%20thus%20%24N(x_0)%20%5Csubset%20A%24%3B%0A%20%20%20%202.%20%24f%24%20is%20continuous%20at%20%24x%24%20if%20and%20only%20if%20%24%5Cvarphi(x)%20%3D%200%24%3A%0A%20%20%20%20%091.%20(%24%5Cimplies%24)%20Suppose%20%24f%24%20is%20continuous%20at%20%24x%24%2C%20then%20for%20any%20%24%5Cvarepsilon0%24%20we%20may%20find%20%24%5Cdelta%20%3E%200%24%20such%20that%20%24%7C%20x%20-%20y%20%7C%20%3C%20%5Cdelta%24%20implies%20that%20%24%7C%20f(x)%20-%20f(y)%20%7C%20%3C%20%5Cvarepsilon%2F2%24.%20Use%20triangle%20inequality%20we%20have%20%24%7C%20f(s)%20-%20f(t)%20%7C%20%3C%20%5Cvarepsilon%2C%20%5Cforall%20s%2Ct%20%5Cin%20(x%20-%20%5Cdelta%2C%20x%20%2B%20%5Cdelta)%24%3B%20pass%20supremum%20we%20have%20%24%5Csup%5Climits%5Clbrace%20%7C%20f(s)%20-%20f(t)%20%7C%3As%2Ct%20%5Cin(x-%5Cdelta%2Cx%2B%5Cdelta)%5Crbrace%20%5Cle%20%5Cvarepsilon%24.%20Thus%20%24%5Cvarphi(x)%24%20%24%5Cle%20%5Cvarphi(x%2C%5Cdelta)%24%20%24%3D%20%5Csup%5Climits%5Clbrace%20%7C%20f(s)%20-%20f(t)%20%7C%3As%2Ct%20%5Cin(x-%5Cdelta%2Cx%2B%5Cdelta)%5Crbrace%24%20%24%5Cle%20%5Cvarepsilon%24%20which%20means%20it%20is%20%240%24%3B%0A%20%20%20%20%092.%20(%24%5Cimpliedby%24)%20Suppose%20now%20%24%5Cvarphi(x)%20%3D%200%24%2C%20then%20by%20above%20we%20have%20%24%5Clim%5Climits_%7B%5Cdelta%20%5Cto%200%7D%5Cvarphi(x%2C%5Cdelta)%20%3D%200%24.%20Thus%20for%20any%20%24%5Cvarepsilon%3E0%24%20we%20may%20find%20%24%5Cdelta%3E0%24%20such%20that%20%24%5Cvarepsilon%3E%5Cvarphi(x%2C%5Cdelta)%24%20%24%3D%20%5Csup%5Climits%5Clbrace%20%7C%20f(s)%20-%20f(t)%20%7C%3As%2Ct%20%5Cin%20(x-%5Cdelta%2Cx%2B%5Cdelta)%5Crbrace%24%20by%20taking%20%24s%20%3D%20x%24%20we%20have%20that%20%24%5Cge%20%7C%20f(x)%20-f(t)%7C%24%20for%20any%20%24t%20%5Cin%20(x-%5Cdelta%2Cx%2B%5Cdelta)%24%2C%20thus%20by%20definition%20%24f%24%20is%20continuous%20at%20%24x%24.%0A%0A%20%20%20%20%09For%20the%20'hence'%20statement%2C%20by%20above%20we%20know%20that%20the%20set%20of%20points%20of%20continuity%20of%20%24f%24%20is%20%24%5Clbrace%20x%20%3A%20%5Cvarphi(x)%20%3D%200%5Crbrace%24%20which%20can%20be%20written%20as%20%24%3D%20%5Cbigcap%5Climits_%7Bn%3D1%7D%5E%7B%5Cinfty%7D%5Clbrace%20x%20%3A%20%5Cvarphi(x)%20%3C%201%2Fn%5Crbrace%24.%20Since%20%24%5Cvarphi%24%20is%20upper%20semicontinuous%2C%20this%20is%20a%20countable%20intersection%20of%20open%20sets%2C%20i.e.%20a%20%24G_%7B%5Cdelta%7D%24%3B%0A%20%20%20%203.%20General%20version%20statement%3A%0A%20%20%20%20%09Suppose%20%24f%3AX%5Cto%20%5Cmathbb%7BC%7D%24%20where%20%24(X%2C%5Ctau)%24%20is%20a%20topological%20space%2C%20define%20%24%5Cvarphi(x%2CU)%20%3D%20%5Csup%5Climits%5Clbrace%20%7C%20f(s)%20-%20f(t)%20%7C%3As%2Ct%20%5Cin%20U%5Crbrace%24%2C%20%24%5Cvarphi(x)%20%3D%20%5Cinf%5Clbrace%20%5Cvarphi(x%2CU)%3A%20U%5Cin%20%5Ctau%2C%20x%5Cin%20U%5Crbrace%24%2C%20then%20%24%5Cvarphi%24%20is%20upper%20semicontinuous%20and%20%24f%24%20is%20continuous%20at%20a%20point%20%24x%24%20if%20and%20only%20if%20%24%5Cvarphi(x)%20%3D%200%24.%20The%20proof%20is%20pretty%20much%20the%20same%3A%0A%20%20%20%20%091.%20%24%5Cvarphi%24%20is%20upper%20semicontinuous%3A%0A%20%20%20%20%09%09Let%20%24a%20%5Cin%20%5Cmathbb%7BR%7D%24%20and%20consider%20the%20set%20%24A%20%3D%20%5Clbrace%20x%20%7C%20%5Cvarphi(x)%20%3C%20a%20%5Crbrace%24.%20If%20%24A%24%20is%20empty%20then%20it%20is%20empty%2C%20otherwise%20let%20%24x_0%20%5Cin%20A%24.%0A%20%20%20%20%09%09We%20have%20%24a%3E%20%5Cvarphi(x_0)%24%2C%20thus%20we%20can%20find%20%24%5Cvarepsilon%3E0%24%20such%20that%20%24%5Cvarphi(x_0)%20%2B%20%5Cvarepsilon%20%3C%20a%24.%20By%20property%20of%20infimum%20and%20supremum%20it%20follows%20that%20there%20is%20an%20%24U%24%20neighborhood%20of%20%24x_0%24%20such%20that%20%24%5Clbrace%20%7C%20f(s)%20-%20f(t)%20%7C%3As%2Ct%20%5Cin%20U%5Crbrace%20%3C%20a%24.%20In%20this%20formulation%20%24x_0%24%20is%20not%20really%20important%20and%20we%20automatically%20have%20%24U%20%5Csubset%20A%24%3B%0A%20%20%20%20%092.%20%24f%24%20is%20continuous%20at%20%24x%24%20if%20and%20only%20if%20%24%5Cvarphi(x)%20%3D%200%24%3A%0A%20%20%20%20%09%09Suppose%20%24f%24%20is%20continuous%20at%20%24x%24%2C%20then%20for%20any%20%24%5Cvarepsilon%20%3E%200%24%20we%20may%20find%20%24U%24%20neighborhood%20of%20%24x%24%20such%20that%20%24f(U)%20%5Csubset%20B(f(x)%2C%5Cvarepsilon)%24%2C%20thus%20for%20any%20%24s%2Ct%20%5Cin%20U%2C%20%7C%20f(s)%20-%20f(t)%20%7C%20%3C%202%20%5Cvarepsilon%24%20thus%20%24%5Cvarphi(x)%20%3C%202%20%5Cvarepsilon%24%20thus%20%24%5Cvarphi(x)%20%3D%200%24.%20It%20also%20works%20the%20other%20way%20around%3B%0A%20%20%20%20%093.%20The%20'hence'%20part%20is%20still%20the%20same%2C%20because%20(as%20being%20used%20above)%20the%20codomain%20of%20%24%5Cvarphi%24%20which%20is%20%24%5Cmathbb%7BR%7D%24%20is%20unchanged.%0A%0A%20%20%20%20%23%23%20Exercise%203%0A%0A%20%20%20%20**Let%20%24X%24%20be%20a%20metric%20space%2C%20with%20metric%20%24%5Crho%24.%20For%20any%20nonempty%20%24E%20%5Csubset%20X%24%2C%20define%20%24%5Crho_E(x)%20%3D%20%5Cinf%5Clbrace%20%5Crho(x%2Cy)%3A%20y%20%5Cin%20E%5Crbrace%24.%20Show%20that%20%24%5Crho_E%24%20is%20a%20uniformly%20continuous%20function%20on%20%24X%24.%20If%20%24A%2C%20B%24%20are%20disjoint%20nonempty%20closed%20subsets%20of%20%24X%24%2C%20examine%20the%20relevance%20of%20the%20function%20%24f(x)%20%3D%20%5Cfrac%7B%5Crho_A(x)%7D%7B%5Crho_A(x)%20%2B%20%5Crho_B(x)%7D%24%20to%20Urysohn's%20lemma**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20Fix%20some%20%24%5Cvarepsilon%20%3E%200%24%20and%20some%20%24E%24.%20Let%20%24%5Cdelta%20%3D%20%5Cvarepsilon%24.%20For%20any%20two%20points%20%24x%2Cy%20%5Cin%20X%24%20such%20that%20%24%5Crho(x%2Cy)%20%3C%20%5Cdelta%24%2C%20by%20triangle%20inequality%20we%20have%20%24%24%5Cbegin%7Baligned%7D%5Crho_E(x)%20%5Cle%26%20%5Crho(x%2Ca)%2C%5Cforall%20a%20%5Cin%20E%5C%5C%5Crho_E(x)%20%5Cle%26%20%5Crho(x%2Cy)%20%2B%20%5Crho(y%2Ca)%2C%5Cforall%20a%20%5Cin%20E%5C%5C%5Crho_E(x)%20-%20%5Crho(x%2Cy)%20%5Cle%26%20%5Crho(y%2Ca)%2C%5Cforall%20a%5Cin%20E%20%5C%5C%20%5Crho_E(x)%20-%20%5Crho(x%2Cy)%20%5Cle%26%20%5Crho_E(y)%5C%5C%20%5Crho_E(x)%20-%20%5Crho_E(y)%20%5Cle%26%20%5Crho(x%2Cy).%5Cend%7Baligned%7D%24%24%20Swap%20%24x%2Cy%24%20we%20will%20have%20%24%5Crho_E(y)%20-%20%5Crho_E(x)%20%3D%20-(%5Crho_E(x)%20-%20%5Crho_E(y))%20%5Cle%20%5Crho(x%2Cy)%24%20thus%20%24%7C%20%5Crho_E(x)%20-%20%5Crho_E(y)%20%7C%20%5Cle%20%5Crho(x%2Cy)%20%3C%20%5Cdelta%20%3D%20%5Cvarepsilon%24.%20Since%20%24%5Cdelta%24%20only%20depends%20on%20%24%5Cvarepsilon%24%2C%20by%20definition%20%24%5Crho_E%24%20is%20uniformly%20continuous.%0A%0A%20%20%20%20This%20is%20a%20function%20that%20can%20be%20used%20to%20prove%20Urysohn's%20Lemma%20(in%20a%20much%20easier%20way)%20in%20a%20metric%20space.%20Urysohn's%20Lemma%20says%20we%20can%20always%20find%20a%20function%20such%20is%3A%0A%0A%20%20%20%201.%20Well-defined%3A%20The%20denominator%20is%20never%20%240%24%3B%0A%20%20%20%202.%20Continuous%3A%20From%20above%20(also%20that%20denominator%20is%20never%20%240%24)%3B%0A%20%20%20%203.%20%24K%20%5Cprec%20f%20%5Cprec%20V%24%3A%20Notice%20that%20a%20metric%20space%20needs%20not%20to%20be%20locally%20compact%20(for%20example%2C%20%24%5Cmathbb%7BQ%7D%24)%2C%20thus%20the%20function%20may%20not%20have%20the%20compact%20support%20property.%20%24f%24%20takes%20value%20%240%24%20on%20%24A%24%20and%20value%20%241%24%20on%20%24B%24%2C%20consider%20%24A%20%3D%20X-V%24%20and%20%24B%20%3D%20K%24%20then%20we%20have%20an%20analogous%20property.%0A%0A%20%20%20%20%23%23%20Exercise%204%0A%0A%20%20%20%20**Examine%20the%20proof%20of%20the%20Riesz%20theorem%20and%20prove%20the%20following%20two%20statements**%3A%0A%0A%20%20%20%201.%20**If%20%24E_1%20%5Csubset%20V_1%24%20and%20%24E_2%20%5Csubset%20V_2%24%20where%20%24V_1%2C%20V_2%24%20are%20disjoint%20open%20sets%2C%20then%20%24%5Cmu(E_1%20%5Ccup%20E_2)%20%3D%20%5Cmu(E_1)%20%2B%20%5Cmu(E_2)%24%20even%20if%20%24E_1%2C%20E_2%24%20are%20not%20in%20%24M%24**%3B%0A%20%20%20%202.%20**If%20%24E%20%5Cin%20M_F%24%2C%20then%20%24E%20%3D%20N%20%5Ccup%20K_1%20%5Ccup%20K_2%20%5Ccup%20%5Cdots%24%20where%20%24%5Clbrace%20K_i%5Crbrace%24%20is%20a%20disjoint%20countable%20collection%20of%20compact%20sets%20and%20%24%5Cmu(N)%20%3D%200%24**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%201.%20By%20step%201.%20Of%20the%20proof%20of%20Riesz%20Representation%20Theorem%2C%20%24%5Cmu(E_1%20%5Ccup%20E_2)%20%5Cle%20%5Cmu(E_1)%20%2B%20%5Cmu(E_2)%24.%20Now%20recall%20that%20we%20defined%20(no%20matter%20if%20%24E%20%5Cin%20M%24)%20that%20%24%5Cmu(E)%20%3D%20%5Cinf%5Clbrace%20%5Cmu(V)%3A%20V%24%20open%2C%20%24V%20%5Csupset%20E%5Crbrace%24.%20Let%20%24U_1%2CU_2%24%20be%20disjoint%20open%20sets%20containing%20%24E_1%2CE_2%24%20respectively%2C%20then%20%24U_1%20%5Ccap%20V_1%2C%20U_2%20%5Ccap%20V_2%24%20are%20also%20disjoint%20open%20sets%20containing%20%24E_1%2CE_2%24%20respectively%2C%20so%20%24E_1%20%5Ccup%20E_2%20%5Csubset%20(V_1%20%5Ccap%20U_1)%20%5Ccup%20(V_2%20%5Ccap%20U_2)%24.%20Now%20%24%5Cmu((V_1%20%5Ccap%20U_1)%20%5Ccup%20(V_2%20%5Ccap%20U_2))%24%20%24%3D%20%5Cmu(V_1%20%5Ccap%20U_1)%20%2B%20%5Cmu(V_2%20%5Ccap%20U_2)%20%5Cge%20%5Cmu(E_1)%20%2B%20%5Cmu(E_2)%24%20for%20any%20such%20%24U_1%2CU_2%24%2C%20pass%20to%20infimum%2C%20we%20have%20%24%5Cmu(E_1%20%5Ccup%20E_2)%24%20%24%5Coverset%7B%5Ctext%7Bdef%7D%7D%7B%3D%7D%5Cinf%5Clbrace%20%5Cmu((V_1%20%5Ccap%20U_1)%20%5Ccup%20(V_2%20%5Ccap%20U_2))%5Crbrace%20%5Cge%20%5Cmu(E_1)%20%2B%20%5Cmu(E_2)%24.%20Put%20things%20together%20we%20have%20equality%3B%0A%20%20%20%202.%20Let%20%24E%20%5Cin%20M_F%24%20and%20set%20%24E_1%20%3D%20E%24%2C%20by%20step%205.%2C%20we%20can%20find%20%24K_1%20%5Csubset%20E_1%20%5Csubset%20V_1%24%20with%20%24%5Cmu(V_1%20-%20K_1)%20%3C%201%2F1%24.%20Set%20%24E_2%20%3D%20E_1%20-%20K_1%24%2C%20by%20definition%20of%20%24M_F%24%20(so%20%24K_1%20%5Cin%20M_F%24)%20and%20step%206%20(so%20%24V_1%20-%20K_1%20%5Cin%20M_F%24)%2C%20%24E_2%20%5Cin%20M_F%24%2C%20so%20we%20can%20find%20%24K_2%20%5Csubset%20E_2%20%5Csubset%20V_2%24%20with%20%24%5Cmu(V_2%20-%20K_2)%20%3C%201%2F2%24.%20Continue%20this%20process%20and%20we%20get%20a%20countable%20collections%20of%20disjoint%20compact%20sets%20%24%5Clbrace%20K_i%5Crbrace%24.%20Set%20%24N%20%3D%20E%20-%20%5Ccup%20K_i%24%20then%20%24N%24%20always%20lies%20in%20%24V_n%20-%20K_n%24%20for%20any%20%24n%24%2C%20thus%20%24%5Cmu(N)%20%5Cle%201%2Fn%24%20for%20any%20%24n%24%2C%20which%20means%20%24%5Cmu(N)%20%3D%200%24.%0A%0A%20%20%20%20%23%23%20Exercise%205%0A%0A%20%20%20%20**Let%20%24E%24%20be%20Cantor's%20familiar%20middle%20thirds%20set.%20Show%20that%20%24m(E)%20%3D%200%24%20even%20though%20%24E%24%20and%20%24%5Cmathbb%7BR%7D%24%20have%20the%20same%20cardinality**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20From%20the%20construction%20of%20the%20Cantor%20set%2C%20we%20can%20write%20%24E%20%3D%20%5Cbigcap%5Climits_%7Bi%3D0%7D%5E%7B%5Cinfty%7D%20E_i%24%20where%20%24m(E_i)%20%3D%20(%5Cfrac%7B2%7D%7B3%7D)%5Ei%24%2C%20then%20%24m(E)%20%5Cle%20(%5Cfrac%7B2%7D%7B3%7D)%5Ei%24%20for%20arbitrary%20%24i%24%20thus%20is%20clearly%20%240%24.%0A%0A%20%20%20%20By%20an%20argument%20from%20topology%20(non-empty%20compact%20Hausdorff%20without%20isolated%20point%20%24%5Cimplies%24%20uncountable)%20we%20can%20show%20%24E%24%20is%20uncountable.%20But%20technically%20this%20does%20not%20say%20%24E%24%20and%20%24%5Cmathbb%7BR%7D%24%20has%20the%20same%20cardinality.%0A%0A%20%20%20%20To%20show%20that%2C%20realize%20that%20each%20number%20%24%5Calpha%24%20in%20%24%5B0%2C1%5D%24%20can%20be%20expressed%20in%20the%20following%20way%3A%20divides%20%24%5B0%2C1%5D%24%20to%20three%20parts%2C%20%24%5B0%2C1%2F3%5D%2C%20%5B1%2F3%2C2%2F3%5D%2C%20%5B2%2F3%2C1%5D%24%2C%20if%20%24%5Calpha%24%20lies%20in%20the%20first%2Fsecond%2Fthird%20part%2C%20record%20a%20%240%2F1%2F2%24%3B%20divides%20the%20part%20that%20%24%5Calpha%24%20lied%20in%20into%20three%20new%20parts%2C%20if%20%24%5Calpha%24%20lies%20in%20the%20first%2Fsecond%2Fthird%20new%20part%2C%20record%20a%20%240%2F1%2F2%24%2C%20and%20so%20on.%20This%20will%20not%20give%20a%20unique%20expression%2C%20but%20it%20is%20okay%3A%20for%20example%2C%20%241%2F3%20%3D%200.1%24%20or%20%24%3D0.0%5Cdot%7B2%7D%24%20because%20it%20lies%20in%20both%20%24%5B0%2C1%2F3%5D%24%20and%20%24%5B1%2F3%2C2%2F3%5D%24.%0A%0A%20%20%20%20Now%20notice%20that%20in%20the%20construction%20of%20Cantor%20set%2C%20at%20step%20%24n%24%20we%20eliminate%20numbers%20with%20expression%20with%20%241%24%20on%20the%20%24n%24-th%20place.%20And%20thus%20the%20numbers%20in%20Cantor%20set%20are%20exactly%20those%20with%20some%20expression%20with%20only%20number%20%240%24%20and%20%242%24%20(so%20for%20example%20%241%2F3%24%20is%20in%20the%20set).%20Now%20observe%20that%20numbers%20with%20expressions%20consist%20of%20only%20%240%24%20and%20%242%24%20are%20different%20if%20and%20only%20if%20the%20expressions%20are%20different.%20Replace%20all%20%242%24%20by%20%241%24%20and%20realize%20that%20it%20can%20now%20be%20viewed%20as%20a%20number%20in%20%24%5B0%2C1%5D%24%20written%20in%20base%20%242%24.%20So%20the%20cardinality%20of%20%24E%24%20is%20no%20less%20than%20the%20cardinality%20of%20%24%5B0%2C1%5D%24.%20Since%20%24E%20%5Csubset%20%5B0%2C1%5D%24%20the%20cardinality%20of%20%24E%24%20is%20no%20more%20than%20the%20cardinality%20of%20%24%5B0%2C1%5D%24%2C%20and%20we%20have%20equality.%0A%0A%20%20%20%20%23%23%20Exercise%206%0A%0A%20%20%20%20**Construct%20a%20totally%20disconnected%20compact%20set%20%24K%20%5Csubset%20%5Cmathbb%7BR%7D%24%20such%20that%20%24m(K)%20%3E%200%24.%20If%20%24v%24%20is%20lower%20semicontinuous%20and%20%24v%20%5Cle%20%5Cchi_K%24%2C%20show%20that%20actually%20%24v%20%5Cle%200%24.%20Hence%20%24%5Cchi_K%24%20cannot%20be%20approximated%20from%20below%20by%20lower%20semicontinuous%20functions%2C%20in%20the%20sense%20of%20the%20Vitali-Caratheodory%20Theorem**.%0A%0A%20%20%20%20Solution%0A%0A%20%20%20%20We%20do%20a%20construction%20that%20is%20analogous%20to%20the%20Cantor%20set%20with%20some%20modification%3A%0A%0A%20%20%20%20Let%20%24K_0%20%3D%20%5B0%2C1%5D%24%2C%20let%20%24K_1%24%20be%20%24K_0%24%20with%20the%20middle%20%24%5Cfrac%7B1%7D%7B3%5E1%7D%24-'length'%20open%20subset%20removed.%0A%0A%20%20%20%20Let%20%24K_2%24%20be%20%24K_1%24%20with%20a%20%24%5Cfrac%7B1%7D%7B3%5E2%7D%5Cfrac%7B1%7D%7B2%5E1%7D%20%3D%20%5Cfrac%7B1%7D%7B18%7D%24-'length'%20open%20subset%20removed%20in%20each%20remaining%20part.%0A%0A%20%20%20%20Continue%20as%20that%2C%20each%20%24K_n%24%20is%20%24K_%7Bn-1%7D%24%20with%20a%20%24%5Cfrac%7B1%7D%7B3%5En%7D%5Cfrac%7B1%7D%7B2%5E%7Bn-1%7D%7D%24-'length'%20open%20subset%20removed%20in%20each%20remaining%20part.%0A%0A%20%20%20%20In%20other%20words%2C%20we%20basically%20do%20the%20same%20steps%20as%20in%20the%20construction%20of%20a%20Cantor%20set%2C%20but%20we%20remove%20(much)%20less%20numbers%20each%20step.%0A%0A%20%20%20%20Let%20%24K%20%3D%20%5Cbigcap%5Climits_%7Bi%20%3D%200%7D%5E%7B%5Cinfty%7DK_i%24%20and%20this%20is%20our%20%24K%24.%20%24K%24%20is%3A%0A%0A%20%20%20%201.%20Compact%3A%20Each%20%24K_i%24%20is%20closed%2C%20so%20%24K%24%20is%20closed%3B%20%24K%24%20is%20also%20bounded%2C%20so%20%24K%24%20is%20compact%3B%0A%20%20%20%202.%20%24m(K)%3E0%24%3A%20For%20each%20%24n%24%2C%20notice%20that%20%24K_n%24%20is%20constructed%20from%20%24K_%7Bn-1%7D%24%20with%20%242%5E%7Bn-1%7D%24%20many%20of%20%24%5Cfrac%7B1%7D%7B3%5En%7D%5Cfrac%7B1%7D%7B2%5E%7Bn-1%7D%7D%24-'length'%20open%20sets%20removed%2C%20so%20%24m(K_n)%20%3D%20m(K_%7Bn-1%7D)%20-%20%5Cfrac%7B1%7D%7B3%5E%7Bn%7D%7D%24.%20I.e.%20%24m(K_n)%20%3D%201%20-%20%5Cfrac%7B1%7D%7B3%5E1%7D-%20%5Cfrac%7B1%7D%7B3%5E2%7D-%20%5Cfrac%7B1%7D%7B3%5E3%7D-%5Cdots%20-%20%5Cfrac%7B1%7D%7B3%5E%7Bn%7D%7D%24%20and%20that%20%24m(K)%20%3D%20%5Clim%5Climits%20m(K_n)%20%3D%201%2F2%3E0%24%2C%20which%20can%20be%20calculated%20as%20%241%24%20minus%20an%20infinite%20summation%20of%20a%20geometric%20sequence%3B%0A%20%20%20%203.%20%24K%24%20is%20totally%20disconnected%3A%20Notice%20that%20each%20%24K_n%24%20is%20by%20construction%20a%20disjoint%20union%20of%20%242%5En%24%20many%20closed%20set%20of%20the%20same%20certain%20length%2C%20which%20must%20go%20to%20zero%20as%20%24n%24%20goes%20to%20infinity%2C%20thus%20%24K%24%20cannot%20contain%20any%20open%20interval%2C%20i.e.%20%24K%24%20is%20totally%20disconnected.%0A%0A%20%20%20%20If%20%24v%24%20is%20lower%20semicontinuous%20then%20by%20definition%20the%20set%20%24A%20%3D%20%5Clbrace%20x%20%7C%20v(x)%3E0%5Crbrace%24%20has%20to%20be%20open.%20Since%20%24v%20%5Cle%20%5Cchi_K%24%20we%20must%20have%20%24A%20%5Csubset%20K%24.%20But%20we%20just%20proved%20that%20%24K%24%20does%20not%20contain%20any%20open%20interval%2C%20so%20%24A%20%3D%20%5Cvarnothing%24%2C%20and%20thus%20%24v%20%5Cle%200%24.%0A%0A%20%20%20%20%23%23%20Exercise%207%0A%0A%20%20%20%20**If%20%240%20%3C%20%5Cvarepsilon%24%2C%20construct%20an%20open%20set%20%24E%20%5Csubset%20%5B0%2C1%5D%24%20which%20is%20dense%20in%20%24%5B0%2C1%5D%24%20such%20that%20%24m(E)%20%3D%20%5Cvarepsilon%24**.%0A%0A%20%20%20%20Solution%0A%0A%20%20%20%20We%20can%20just%20take%20%24E%24%20be%20the%20complement%20of%20%24K%24%20as%20in%20the%20last%20problem%2C%20then%20this%20%24E%24%20will%20be%20open%20(complement%20of%20closed%20set)%2C%20dense%20(as%20argued%20above%2C%20interior%20of%20%24K%24%20is%20empty)%2C%20and%20%24m(E)%24%20will%20be%20a%20certain%20value.%20For%20a%20given%20%24%5Cvarepsilon%24%2C%20essentially%20we%20just%20need%20to%20solve%20%24%5Cvarepsilon%20%3D%20%5Csum%20(%5Cfrac%7B1%7D%7Bc%7D)%5En%24%20(because%20%24m(E)%20%3D%201%20-%20m(K)%24%20)%20%24%3D%20%5Cfrac%7B1%2Fc%7D%7B1-1%2Fc%7D%24%20for%20%24c%24%2C%20which%20obviously%20always%20have%20a%20solution.%20Use%20this%20%24c%24%20to%20substitute%20every%20%243%24%20in%20the%20last%20problem%2C%20get%20%24K%24%2C%20take%20complement%20and%20we%20get%20desired%20%24E%24.%0A%0A%20%20%20%20%23%23%20Exercise%209%0A%0A%20%20%20%20**Construct%20a%20sequence%20of%20continuous%20functions%20%24f_n%24%20on%20%24%5B0%2C1%5D%24%20such%20that%20%240%5Cle%20f_n%20%5Cle%201%24%20and%20%24%5Clim%5Climits_%7Bn%20%5Cto%20%5Cinfty%7D%5Cint_0%5E1f_n(x)dx%20%3D%200%24%20but%20such%20that%20the%20sequence%20%24%5Clbrace%20f_n(x)%5Crbrace%24%20converges%20for%20no%20%24x%20%5Cin%20%5B0%2C1%5D%24**.%0A%0A%20%20%20%20Solution%0A%0A%20%20%20%20A%20point%20that%20will%20help%20about%20the%20construction%3A%0A%0A%20%20%20%20If%20we%20just%20think%20of%20'simple'%20functions%20then%20the%20continuity%20requirement%20is%20not%20really%20a%20constraint%20because%20we%20can%20always%20'smooth'%20the%20part%20that%20is%20not%20continuous%2C%20the%20value%20of%20(Riemann)%20integral%20of%20the%20original%20and%20smoothed%20functions%20are%20the%20same%3A%0A%0A%20%20%20%20%3Cimg%20src%3D%22public%2FPasted%20image%2020210930172033.png%22%20width%3D%22600%22%20%2F%3E%0A%0A%20%20%20%20Since%20the%20construction%20below%20will%20only%20use%20simple%20functions%2C%20we%20just%20assume%20every%20function%20is%20already%20smoothed%20thus%20continuous.%0A%0A%20%20%20%20Now%2C%20consider%20the%20list%20of%20all%20fractional%20expressions%20in%20%24(0%2C1%5D%24%20which%20labeled%20in%20the%20order%20%24%5Cfrac%7B1%7D%7B1%7D%2C%20%5Cfrac%7B1%7D%7B2%7D%2C%20%5Cfrac%7B2%7D%7B2%7D%2C%20%5Cfrac%7B1%7D%7B3%7D%2C%5Cfrac%7B2%7D%7B3%7D%2C%5Cfrac%7B3%7D%7B3%7D%2C%20%5Cdots%2C%20%5Cfrac%7B1%7D%7Bn%7D%2C%5Cfrac%7B2%7D%7Bn%7D%2C%20%5Cdots%2C%20%5Cfrac%7Bn%7D%7Bn%7D%2C%5Cdots%24.%20For%20each%20%24p%2Fq%24%2C%20assign%20it%20a%20function%20%24%24f_%7Bp%2Cq%7D(x)%20%3D%20%5Cbegin%7Bcases%7D1%2C%26x%20%5Cin%20%5B%5Cfrac%7Bp%7D%7Bq%7D%20-%20%5Cfrac%7B1%7D%7Bq%7D%2C%20%5Cfrac%7Bp%7D%7Bq%7D%5D%5C%5C0%2C%26%5Ctext%7Botherwise%7D%5Cend%7Bcases%7D.%24%24%20Since%20the%20cardinality%20of%20all%20fractional%20expressions%20is%20countable%2C%20we%20can%20now%20label%20%24f_%7B1%2C1%7D%2Cf_%7B1%2C2%7D%2C%5Cdots%24%20as%20our%20%24f_1%2Cf_2%2C%5Cdots%24%3A%0A%0A%20%20%20%201.%20By%20construction%2C%20%24%5Cint_0%5E1f_%7Bp%2Cq%7Ddx%20%3D%20%5Cfrac%7B1%7D%7Bq%7D%24%2C%20thus%20%24%5Clim%5Climits_%7Bn%20%5Cto%20%5Cinfty%7D%5Cint_0%5E1f_n(x)dx%20%3D%20%5Clim%5Climits_%7Bq%20%5Cto%20%5Cinfty%7D%5Cint_0%5E1f_%7Bp%2Cq%7D(x)dx%20%3D%200%24%3B%0A%0A%20%20%20%20%3Cimg%20src%3D%22public%2FPasted%20image%2020210930174955.png%22%20width%3D%22600%22%20%2F%3E%0A%0A%20%20%20%203.%20On%20the%20other%20hand%2C%20notice%20that%20each%20number%20in%20%24%5B0%2C1%5D%24%20lies%20inside%20of%20infinitely%20many%20intervals%20of%20the%20form%20%24%5B%5Cfrac%7Bp%7D%7Bq%7D%20-%20%5Cfrac%7B1%7D%7Bq%7D%2C%20%5Cfrac%7Bp%7D%7Bq%7D%5D%24%20(and%20apparently%2C%20it%20also%20lies%20outside%20of%20infinitely%20many%20such%20intervals).%20For%20example%2C%20%240.1%24%20lies%20in%20%24%5B0%2C1%2F2%5D%2C%5B0%2C1%2F3%5D%2C%5Cdots%2C%5B0%2C1%2F10%5D%2C%5B1%2F11%2C2%2F11%5D%2C%5Cdots%24.%20Thus%20for%20any%20%24x%20%5Cin%20%5B0%2C1%5D%24%2C%20%24x%24%20takes%20value%20%240%24%20for%20infinitely%20many%20%24f_n%24%2C%20and%20takes%20value%20%241%24%20also%20for%20infinitely%20many%20%24f_n%24%2C%20thus%20%24f_n%24%20does%20not%20converge%20at%20%24x%24.%0A%0A%20%20%20%20%23%23%20Exercise%2010%0A%0A%20%20%20%20**If%20%24%5Clbrace%20f_n%5Crbrace%24%20is%20a%20sequence%20of%20continuous%20functions%20on%20%24%5B0%2C1%5D%24%20such%20that%20%240%5Cle%20f_n%20%5Cle%201%24%20and%20such%20that%20%24f_n(x)%20%5Cto%200%24%20as%20%24n%5Cto%20%5Cinfty%24%20for%20every%20%24x%20%5Cin%20%5B0%2C1%5D%24%2C%20then%20%24%5Clim%5Climits_%7Bn%20%5Cto%20%5Cinfty%7D%5Cint_0%5E1f_n(x)dx%20%3D%200%24.%20Try%20to%20prove%20this%20without%20using%20any%20measure%20theory%20or%20any%20theorems%20about%20Lebesgue%20integration**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20It%20is%20an%20immediate%20result%20of%20Lebesgue's%20Dominated%20Convergence%20Theorem%2C%20but%20I%20don't%20know%20how%20to%20do%20it%20without%20using%20measure%20theory.%0A%0A%20%20%20%20%23%23%20Exercise%2011%0A%0A%20%20%20%20**Let%20%24%5Cmu%24%20be%20a%20regular%20Borel%20measure%20on%20a%20compact%20Hausdorff%20space%20%24X%24.%20Assume%20%24%5Cmu(X)%20%3D%201%24.%20Prove%20that%20there%20is%20a%20compact%20set%20%24K%5Csubset%20X%24%20(the%20carrier%20or%20support%20of%20%24%5Cmu%24)%20such%20that%20%24%5Cmu(K)%20%3D%201%24%20but%20%24%5Cmu(H)%20%3C%201%24%20for%20every%20proper%20compact%20subset%20%24H%24%20of%20%24K%24**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20Define%20%24K%24%20to%20be%20the%20intersection%20of%20all%20compact%20sets%20%24K_i%24%20(%24X%24%20is%20a%20such%20set)%20with%20measure%20%241%24.%20Then%3A%0A%0A%20%20%20%201.%20%24K%24%20is%20compact%3A%20each%20%24K_i%24%20is%20compact%20in%20Hausdorff%20%24X%24%20thus%20closed%2C%20the%20intersection%20of%20all%20%24K_i%24's%20is%20then%20closed%2C%20and%20%24%5Ccap%20K_i%20%5Csubset%20K_1%24%20is%20a%20closed%20subset%20of%20compact%20set%20thus%20compact%3B%0A%20%20%20%202.%20%24%5Cmu(K)%20%3D%201%24%3A%20Let%20%24V%24%20be%20an%20open%20set%20containing%20%24K%24%2C%20then%20%24V%5Ec%24%20is%20closed%20(lies%20in%20compact%20%24X%24)%20thus%20compact.%20Notice%20that%20the%20set%20%24K_i%5Ec%20%5Ccap%20V%5Ec%24%20is%20open%20in%20%24V%5Ec%24%20(with%20subspace%20topology)%2C%20and%20that%20%24%5Ccup%20(K_i%5Ec%20%5Ccap%20V%5Ec)%20%3D%20(%5Ccup%20K_i%5Ec)%20%5Ccap%20V%5Ec%24%20(using%20%24%5Ccup%20K_i%5Ec%20%3D%20(%5Ccap%20K_i)%5Ec%20%3D%20K%5Ec%20%5Csupset%20V%5Ec%24)%20%24%3D%20V%5Ec%24%2C%20so%20this%20is%20an%20open%20covering%2C%20so%20it%20admits%20a%20finite%20sub-covering%2C%20say%20%24V%5Ec%20%3D%20(K_1%5Ec%20%5Ccap%20V%5Ec)%20%5Ccup%20%5Cdots%20%5Ccup%20(K_n%5Ec%20%5Ccap%20V%5Ec)%20%3D%20(K_1%5Ec%20%5Ccup%20%5Cdots%20%5Ccup%20K_n%5Ec)%20%5Ccap%20V%5Ec%24.%20Now%20since%20%24%5Cmu(K_i)%20%3D%201%2C%20%5Cmu(X)%20%3D%201%24%2C%20%24%5Cmu(K_i%5Ec)%20%3D%200%24%20for%20any%20%24i%24%2C%20thus%20%24%5Cmu(K_1%5Ec%20%5Ccup%20%5Cdots%20%5Ccup%20K_n%5Ec)%20%3D%200%24%20thus%20%24%5Cmu(V%5Ec)%20%5Cle%200%20%5Cimplies%20%5Cmu(V%5Ec)%20%3D%200%24%20thus%20%24%5Cmu(V)%20%3D%201%24.%20Since%20%24V%24%20is%20chosen%20arbitrarily%2C%20it%20follows%20that%20any%20open%20set%20containing%20%24K%24%20has%20measure%20%241%24%3B%20since%20%24%5Cmu%24%20is%20regular%2C%20%24K%24%20must%20have%20measure%20%241%24%3B%0A%20%20%20%203.%20%24%5Cmu(H)%3C%201%24%3A%20If%20not%2C%20then%20%24H%24%20is%20some%20%24K_i%24%20in%20the%20construction%20of%20%24K%24%2C%20but%20then%20it%20is%20not%20a%20proper%20subset%20of%20%24K%24.%0A%0A%20%20%20%20%23%23%20Exercise%2012%0A%0A%20%20%20%20**Show%20that%20every%20compact%20subset%20of%20%24%5Cmathbb%7BR%7D%24%20is%20the%20support%20of%20a%20Borel%20measure**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20Let%20%24K%24%20be%20a%20compact%20subset%20of%20%24%5Cmathbb%7BR%7D%24%2C%20consider%20the%20measure%20%24%5Cmu_K(E)%20%3D%20%5Cmu(E)%2F%5Cmu(K)%24%20where%20%24%5Cmu%24%20denote%20the%20Lebesgue%20measure.%20Since%20%24%5Cmu%24%20is%20a%20Borel%20measure%20and%20%24%5Cmu(K)%24%20is%20now%20just%20a%20fixed%20number%2C%20%24%5Cmu_K%24%20is%20a%20Borel%20measure.%20Now%20%24%5Cmu_K(K)%24%20is%20apparently%20%241%24%3B%20suppose%20%24H%24%20is%20a%20proper%20compact%20subset%20of%20%24K%24%2C%20since%20we%20work%20in%20%24%5Cmathbb%7BR%7D%24%20then%20%24K%24%20and%20%24H%24%20are%20just%20closed%20and%20bounded%20(union%20of)%20intervals%2C%20and%20there%20must%20be%20certain%20%24(a%2Cb)%24%20contained%20in%20%24K-H%24%2C%20thus%20%24%5Cmu(H)%20%3C%20%5Cmu(K)%24%2C%20thus%20%24%5Cmu_K(H)%20%3D%20%5Cmu(H)%2F%5Cmu(K)%20%3C%201%24.%20By%20definition%20%24K%24%20is%20the%20support%20of%20this%20Borel%20measure.%0A%0A%20%20%20%20%23%23%20Exercise%2013%0A%0A%20%20%20%20**Is%20it%20true%20that%20every%20compact%20subset%20of%20%24%5Cmathbb%7BR%7D%24%20is%20the%20support%20of%20a%20continuous%20function%3F%20if%20not%2C%20can%20you%20describe%20the%20class%20of%20all%20compact%20sets%20in%20%24%5Cmathbb%7BR%7D%24%20which%20are%20supports%20of%20continuous%20functions%3F%20Is%20your%20description%20valid%20in%20other%20topological%20spaces**%3F%0A%0A%20%20%20%20Solution%0A%0A%20%20%20%201.%20No%2C%20for%20example%2C%20a%20single-point%20set%20cannot%20be%20support%20of%20a%20continuous%20function%3A%20otherwise%20the%20function%20takes%20some%20non-zero%20value%20at%20that%20point%2C%20and%20zero%20on%20all%20other%20points%2C%20making%20the%20function%20not%20continuous%3B%0A%20%20%20%202.%20The%20class%20of%20all%20compact%20sets%20in%20%24%5Cmathbb%7BR%7D%24%20which%20are%20supports%20of%20continuous%20functions%20are%20those%20compact%20sets%20whose%20each%20connect%20component%20has%20non-empty%20interior%3A%0A%20%20%20%20%091.%20If%20a%20compact%20set%20is%20of%20this%20type%3A%20since%20we%20do%20not%20need%20differentiability%20it%20is%20very%20easy%20to%20construct%20a%20continuous%20function%20has%20support%20of%20each%20of%20the%20compact%20set's%20connect%20component%3B%0A%20%20%20%20%092.%20If%20not%2C%20then%20this%20compact%20set%20has%20some%20isolated%20points%2C%20and%20it%20cannot%20be%20the%20support%20of%20a%20continuous%20function%20as%20argued%20above%3B%0A%20%20%20%203.%20Not%20in%20general%2C%20for%20example%20if%20we%20take%20%24X%20%3D%20%5Clbrace%20a%2C%20b%2C%20c%5Crbrace%24%20with%20%24%5Ctau%20%3D%20%5Clbrace%20%5Cvarnothing%2C%20X%2C%20%5Clbrace%20a%5Crbrace%5Crbrace%24%20and%20consider%20%24f%3AX%20%5Cto%20X%24%2C%20%24f(X)%20%3D%20%5Clbrace%20b%5Crbrace%24.%20%24f%24%20is%20continuous.%20The%20support%20of%20%24f%24%20is%20%24%5Clbrace%20b%2C%20c%5Crbrace%24%20(the%20smallest%20closed%20set%20contains%20%24%5Clbrace%20b%5Crbrace%24)%2C%20it%20is%20connected%2C%20yet%20has%20an%20empty%20interior%20(the%20only%20open%20set%20contained%20in%20it%20is%20the%20empty%20set).%0A%0A%20%20%20%20%23%23%20Exercise%2014%0A%0A%20%20%20%20**Let%20%24f%24%20be%20a%20real-valued%20Lebesgue%20measurable%20function%20on%20%24%5Cmathbb%7BR%7D%5Ek%24.%20Prove%20that%20there%20exist%20Borel%20functions%20%24g%2Ch%24%20such%20that%20%24g(x)%20%3D%20h(x)%24%20a.e.%20%24%5Bm%5D%24%2C%20and%20%24g(x)%5Cle%20f(x)%20%5Cle%20h(x)%24%20for%20every%20%24x%20%5Cin%20%5Cmathbb%7BR%7D%5Ek%24**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20Split%20%24%5Cmathbb%7BR%7D%5Ek%24%20into%20%24k%24-unit%20cubes%20%24(c_1%2Cc_1%2B1%5D%20%5Ctimes%20%5Cdots%20%5Ctimes%20(c_k%2C%20c_k%2B1%5D%24.%20Note%20there%20are%20only%20countably%20many%20of%20such%20cubes.%20Let's%20fix%20one%20of%20such%20cube%20%24C%24%2C%20and%20consider%20for%20now%20%24f%24%20on%20%24C%24%20only%2C%20then%3A%0A%0A%20%20%20%201.%20%24m(C)%20%3D%201%20%3C%20%5Cinfty%24%3B%0A%20%20%20%202.%20Let%20%24n%20%5Cin%20%5Cmathbb%7BZ%7D%24%20and%20consider%20%24C_n%20%3D%20f%5E%7B-1%7D((n%2Cn%2B1%5D)%20%5Ccap%20C%24%2C%20this%20is%20a%20countable%20partition%20of%20%24f(C)%24.%20Then%20by%20this%20construction%20%24n%3Cf(C_n)%5Cle%20n%2B1%24.%20Since%20we%20work%20with%20Lebesgue%20measure%2C%20we%20can%20apply%20the%20corollary%20of%20Lusin's%20Theorem%20(in%20that%20corollary%20we%20assumed%20%24%7C%20f%20%7C%20%5Cle%201%24%2C%20but%20that%20is%20not%20a%20big%20problem%20-%20we%20can%20just%20think%20our%20%24f%24%20of%20a%20constant%20map%20add%20a%20map%20bounded%20by%20%241%24)%20to%20get%20a%20sequence%20%24%5Clbrace%20%5Csigma_%7Bn%2Cm%7D%5Crbrace%24%20such%20that%20%24f(x)%20%3D%20%5Clim%5Climits_%7Bm%20%5Cto%20%5Cinfty%7D%20%5Csigma_%7Bn%2Cm%7D%20%3D%20%5Csigma_n%24%20a.e.%20on%20%24C_n%24.%20Each%20%24%5Csigma_%7Bn%2Cm%7D%24%20is%20by%20the%20theorem%20continuous%20thus%20Borel%20measurable%2C%20thus%20%24%5Csigma_n%24%20is%20Borel%20measurable.%20Define%20%24%5Csigma%24%20on%20%24C%24%20where%20%24%5Csigma(x)%20%3D%20%5Csigma_n(x)%24%20for%20%24x%24%20in%20corresponding%20%24C_n%24.%20Then%20%24%5Csigma%24%20is%20Borel%20measurable%20and%20%24f%20%3D%20%5Csigma%24%20a.e.%20(countable%20union%20of%20measure%20zero%20is%20still%20measure%20zero)%20on%20%24C%24.%0A%0A%20%20%20%20Do%20this%20for%20each%20%24C%24%20cube%20and%20extend%20the%20domain%20for%20%24f%24%20and%20%24%5Csigma%24%2C%20then%20%24%5Csigma%24%20is%20a%20Borel%20function%20on%20%24%5Cmathbb%7BR%7D%5Ek%24%20such%20that%20%24f%20%3D%20%5Csigma%24%20a.e.%0A%0A%20%20%20%20Consider%20the%20set%20%24X%20%3D%20%5Clbrace%20x%20%7C%20f(x)%20%5Cne%20%5Csigma(x)%5Crbrace%24%2C%20this%20set%20has%20measure%20%240%24%2C%20define%20a%20function%20%24%5Cvarphi%24%20on%20%24%5Cmathbb%7BR%7D%5Ek%24%20as%20following%3A%20if%20%24x%20%5Cnotin%20X%24%2C%20%24%5Cvarphi(x)%20%3D%200%24%2C%20otherwise%2C%20%24%5Cvarphi(x)%20%3D%20%5Clceil%20%5Csigma(x)%20-%20f(x)%20%5Crceil%24%20(the%20smallest%20integer%20that%20larger%20than%20or%20equal%20to%20%24%5Csigma(x)%20-%20f(x)%24).%20This%20%24%5Cvarphi%24%20is%20easy%20to%20see%20to%20be%20a%20Borel%20function.%0A%0A%20%20%20%20Define%20our%20%24g(x)%20%3D%20%5Csigma(x)%20-%20%5Cvarphi(x)%24%2C%20it%20is%20a%20Borel%20function%20by%20construction%2C%20and%20%24g(x)%20%3D%20%5Csigma(x)%20-%20%5Clceil%20%5Csigma(x)%20-%20f(x)%20%5Crceil%24%20%24%5Cle%20f(x)%24%20for%20all%20%24x%24.%20Measure%20of%20the%20set%20that%20%24g(x)%20%5Cne%20f(x)%24%20is%20by%20construction%20%240%24.%0A%0A%20%20%20%20Since%20this%20works%20for%20any%20real-valued%20Lebesgue%20measurable%20%24f%24%2C%20it%20also%20work%20for%20%24-f%24%20(which%20is%20apparently%20also%20real-valued%20Lebesgue%20measurable)%2C%20thus%20we%20can%20find%20%24-h%20%5Cle%20-f%24%20thus%20we%20can%20find%20%24f%20%5Cle%20h%24.%20%24g(x)%20%5Cne%20h(x)%24%20only%20if%20%24g(x)%20%5Cne%20f(x)%24%20and%20%24h(x)%20%5Cne%20f(x)%24%20thus%20have%20measure%20zero%2C%20so%20%24g(x)%20%3D%20h(x)%24%20a.e.%0A%0A%20%20%20%20%23%23%20Exercise%2015%0A%0A%20%20%20%20**It%20is%20easy%20to%20guess%20the%20limits%20of%20%24%5Cint_0%5En%20(1-%5Cfrac%7Bx%7D%7Bn%7D)%5Ene%5E%7Bx%2F2%7Ddx%24%20and%20%24%5Cint_0%5En%20(1%2B%5Cfrac%7Bx%7D%7Bn%7D)%5Ene%5E%7B-2x%7Ddx%24%20as%20%24n%5Cto%20%5Cinfty%24.%20Prove%20that%20your%20guesses%20are%20correct**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20For%20the%20first%20one%3A%20define%20a%20sequence%20%24%5Clbrace%20f_n%5Crbrace%24%20where%20%24f_n(x)%20%3D%20(1-%5Cfrac%7Bx%7D%7Bn%7D)%5Ene%5E%7Bx%2F2%7D%24%20on%20%24%5B0%2Cn%5D%24%20and%20%240%24%20otherwise.%20This%20is%20an%20important%20limit%20and%20we%20know%20that%20%24%5Clim%5Climits%20f_n%20%3D%20e%5E%7Bx%2F2%20-%20x%7D%20%3D%20e%5E%7B-x%2F2%7D%24.%20For%20any%20%24n%24%20we%20can%20see%20from%20the%20derivative%20that%20%24(1-%5Cfrac%7Bx%7D%7Bn%7D)%5En%20%5Cle%20e%5Ex%24%20on%20%24%5B0%2Cn%5D%24%20(equality%20only%20at%20%240%24)%2C%20so%20%24%5Clim%5Climits%20f_n%20%3E%20f_n%24%20for%20all%20%24n%24%20and%20%24x%24.%20Since%20%24e%5E%7B-x%2F2%7D%24%20is%20%24L%5E1%24%20we%20can%20apply%20the%20Lebesgue's%20Dominated%20Convergence%20Theorem%20to%20get%20%24%5Clim%5Climits_%7Bn%5Cto%20%5Cinfty%7D%5Cint_0%5En%20(1-%5Cfrac%7Bx%7D%7Bn%7D)%5Ene%5E%7Bx%2F2%7Ddx%24%20%24%3D%20%5Clim%5Climits%20%5Cint_0%5En%20e%5E%7B-x%2F2%7Ddx%20%3D%20%5Clim%5Climits%20(2e%5E0%20-%202e%5E%7B-n%7D)%20%3D%202%24%2C%20just%20use%20the%20common%20Riemann%20integral.%0A%0A%20%20%20%20The%20second%20one%20can%20be%20evaluated%20in%20a%20similar%20way%2C%20%24%5Clim%5Climits%20%5Cint_0%5En%20(1%2B%5Cfrac%7Bx%7D%7Bn%7D)%5Ene%5E%7B-2x%7Ddx%24%20%24%3D%20%5Clim%5Climits%20%5Cint_0%5En%20e%5E%7B-x%7Ddx%20%3D%201%24.%0A%20%20%20%20%22%22%22)%0A%20%20%20%20return%0A%0A%0A%40app.cell%0Adef%20_(mo)%3A%0A%20%20%20%20mo.md(r%22%22%22%0A%20%20%20%20%23%20Fall%20Semester%20Homework%203%0A%20%20%20%20%22%22%22)%0A%20%20%20%20return%0A%0A%0A%40app.cell%0Adef%20_(mo)%3A%0A%20%20%20%20mo.md(r%22%22%22%0A%20%20%20%20%23%23%20Exercise%201%0A%0A%20%20%20%20**Prove%20that%20the%20supremum%20of%20any%20collection%20of%20convex%20functions%20on%20%24(a%2Cb)%24%20is%20convex%20on%20%24(a%2Cb)%24%20if%20it%20is%20finite%2C%20and%20that%20point-wise%20limits%20of%20sequences%20of%20convex%20functions%20are%20convex.%20What%20can%20you%20say%20about%20upper%20and%20lower%20limits%20of%20sequence%20of%20convex%20functions**%3F%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20Suppose%20%24%5Clbrace%20f_i%20%5Crbrace%24%20is%20a%20collection%20of%20convex%20functions%20on%20%24(a%2Cb)%24%20and%20%24f%24%20is%20the%20supremum.%20By%20definition%20we%20have%20%24f_i((1-%5Clambda)x%20%2B%20%5Clambda%20y)%20%5Cle%20(1-%5Clambda)f_i(x)%20%2B%20%5Clambda%20f_i(y)%24%2C%20since%20%24f%24%20is%20the%20supremum%20we%20have%20%24%5Cle%20(1-%5Clambda)f(x)%20%2B%20%5Clambda%20f(y)%24.%20But%20since%20this%20works%20for%20any%20%24i%24%2C%20so%20we%20have%20%24%5Csup(f_i((1-%5Clambda)x%20%2B%20%5Clambda%20y))%20%5Cle%20(1-%5Clambda)f(x)%20%2B%20%5Clambda%20f(y)%24%2C%20i.e.%20%24f((1-%5Clambda)x%20%2B%20%5Clambda%20y)%20%5Cle%20(1-%5Clambda)f(x)%20%2B%20%5Clambda%20f(y)%24%20which%20by%20definition%20means%20convexity.%0A%0A%20%20%20%20Suppose%20%24%5Clbrace%20f_i%20%5Crbrace%24%20is%20a%20collection%20of%20convex%20functions%20and%20%24f%24%20is%20the%20point-wise%20limit%2C%20if%20we%20do%20not%20have%20convexity%20for%20%24f%24%2C%20i.e.%20%24(1-%5Clambda)f(x)%20%2B%20%5Clambda%20f(y)%20%3C%20f((1-%5Clambda)x%20%2B%20%5Clambda%20y)%24%20for%20certain%20%24x%2Cy%2C%20%5Clambda%24%2C%20then%20by%20definition%20of%20limit%2C%20we%20will%20have%20%24(1-%5Clambda)f_i(x)%20%2B%20%5Clambda%20f_i(y)%20%3C%20f_i((1-%5Clambda)x%20%2B%20%5Clambda%20y)%24%20for%20some%20%24f_i%24%2C%20so%20%24f_i%24%20is%20not%20convex.%0A%0A%20%20%20%20The%20difference%20between%20the%20second%20and%20third%20statement%20is%20the%20second%20statement%20assumes%20the%20existence%20of%20the%20point-wise%20limit.%20For%20the%20third%20statement%2C%20we%20may%20have%20the%20situation%20that%20we%20have%20some%20arbitrary%20upper%20or%20lower%20bound%20of%20the%20functions%2C%20yet%20no%20'infinity'%20comes%20into%20play.%0A%0A%20%20%20%20Nevertheless%2C%20the%20upper%20limits%20of%20sequence%20of%20convex%20functions%20are%20still%20convex%3A%20if%20%24%5Clbrace%20f_i%20%5Crbrace%24%20is%20a%20sequence%20of%20convex%20functions%20and%20%24f%24%20is%20the%20upper%20limit%2C%20then%20for%20contradiction%20if%20%24f((1-%5Clambda)x%20%2B%20%5Clambda%20y)%20%3E%20(1-%5Clambda)f(x)%20%2B%20%5Clambda%20f(y)%24%20then%20%24f((1-%5Clambda)x%20%2B%20%5Clambda%20y)%20%3E%20(1-%5Clambda)f_i(x)%20%2B%20%5Clambda%20f_i(y)%24%20for%20any%20%24i%24%2C%20but%20then%20there%20will%20be%20some%20%24i%24%20with%20%24f_i((1-%5Clambda)x%20%2B%20%5Clambda%20y)%20%3E%20(1-%5Clambda)f_i(x)%20%2B%20%5Clambda%20f_i(y)%24.%0A%0A%20%20%20%20The%20inequality%20does%20not%20work%20for%20the%20lower%20limit.%20Take%20%24f_1(x)%20%3D%20x%24%20on%20%24(0%2C1)%24%2C%20%24f_2(x)%20%3D%201-x%24%20on%20%24(0%2C1)%24%2C%20and%20%24f_i%20%3D%200.5%24%20for%20any%20%24i%20%3E%202%24%2C%20then%20we%20have%20a%20counterexample.%0A%0A%20%20%20%20%23%23%20Exercise%202%0A%0A%20%20%20%20**If%20%24%5Cvarphi%24%20is%20convex%20on%20%24(a%2Cb)%24%20and%20if%20%24%5Cpsi%24%20is%20convex%20and%20non-decreasing%20on%20the%20range%20of%20%24%5Cvarphi%24%2C%20prove%20%24%5Cpsi%20%5Ccirc%20%5Cvarphi%24%20is%20convex%20on%20%24(a%2Cb)%24.%20For%20%24%5Cvarphi%3E0%24%2C%20show%20that%20the%20convexity%20of%20%24%5Clog%7B%5Cvarphi%7D%24%20implies%20the%20convexity%20of%20%24%5Cvarphi%24%2C%20but%20not%20vice%20versa**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20We%20have%20%24%24%5Cbegin%7Baligned%7D%20%5Cpsi(%5Cvarphi((1-%5Clambda)x%20%2B%20%5Clambda%20y))%20%26%5Cle%20%5Cpsi((1-%5Clambda)%5Cvarphi(x)%20%2B%20%5Clambda%20%5Cvarphi(y))%20%5C%5C%20%26%5Cle%20(1-%5Clambda)%5Cpsi(%5Cvarphi(x))%20%2B%20%5Clambda%20%5Cpsi(%5Cvarphi(y))%20%5Cend%7Baligned%7D.%24%24%0A%20%20%20%20The%20first%20inequality%20comes%20from%20that%20%24%5Cpsi%24%20is%20non-decreasing%20and%20convexity%20of%20%24%5Cvarphi%24%2C%20and%20the%20second%20one%20comes%20from%20the%20convexity%20of%20%24%5Cpsi%24%3B%20by%20definition%20we%20have%20convexity%20of%20%24%5Cpsi%20%5Ccirc%20%5Cvarphi%24.%0A%0A%20%20%20%20For%20the%20second%2C%20we%20know%20%24%5Cexp%24%20is%20convex%20and%20non-decreasing%2C%20so%20%24%5Clog(%5Cvarphi)%24%20being%20convex%20implies%20%24%5Cexp(%5Clog(%5Cvarphi))%20%3D%20%5Cvarphi%24%20being%20convex%20by%20part%201.%0A%0A%20%20%20%20Converse%20not%20true%3A%20%24%5Cvarphi(x)%20%3D%20x%24%20the%20identity%20map%20is%20convex%20(say%20on%20%24(0%2C1)%24)%2C%20yet%20%24%5Clog(x)%24%20is%20certainly%20not%20convex.%0A%0A%20%20%20%20%23%23%20Exercise%203%0A%0A%20%20%20%20**Assume%20that%20%24%5Cvarphi%24%20is%20a%20continuous%20real%20function%20on%20%24(a%2Cb)%24%20such%20that%20%24%5Cvarphi(%5Cfrac%7Bx%2By%7D%7B2%7D)%20%5Cle%20%5Cfrac%7B1%7D%7B2%7D%5Cvarphi(x)%20%2B%20%5Cfrac%7B1%7D%7B2%7D%5Cvarphi(y)%24%20for%20all%20%24x%24%20and%20%24y%20%5Cin%20(a%2Cb)%24.%20Prove%20that%20%24%5Cvarphi%24%20is%20convex**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20Fix%20any%20%24%5Clambda%2C%20x%2C%20y%24%20in%20the%20setting%20of%20a%20convex%20function.%0A%0A%20%20%20%20If%20%24%5Clambda%24%20is%20a%20fraction%20of%20the%20form%20%24m%2F2%5En%24%20then%20things%20are%20easy%2C%20for%20example%20if%20%24%5Clambda%20%3D%201%2F4%24%20we%20can%20see%20the%20condition%20for%20convexity%20holds%20as%20follow%3A%20let%20%24x'%20%3D%20%5Cfrac%7B3%7D%7B4%7Dx%20%2B%20%5Cfrac%7B1%7D%7B4%7Dy%24%20and%20%24x''%20%3D%20%5Cfrac%7B1%7D%7B2%7Dx%20%2B%20%5Cfrac%7B1%7D%7B2%7Dy%24%2C%20then%20%24%5Cvarphi(x')%20%5Cle%20%5Cfrac%7B%5Cvarphi(x)%20%2B%20%5Cvarphi(x'')%7D%7B2%7D%24%20%24%5Cle%20%5Cfrac%7B%5Cvarphi(x)%20%2B%20%5Cfrac%7B%5Cvarphi(x)%20%2B%20%5Cvarphi(y)%7D%7B2%7D%7D%7B2%7D%24.%0A%0A%20%20%20%20If%20not%2C%20then%20we%20can%20approximate%20such%20%24%5Clambda%24%20by%20a%20fraction%20of%20the%20form%20%24r%20%3D%20m%2F2%5En%24%2C%20i.e.%20so%20that%20%24%7C%20%5Cvarphi((1-%5Clambda)x%20%2B%20%5Clambda%20y)%20-%20%5Cvarphi((1-r)x%20%2B%20r%20y)%20%7C%24%20and%20%24%7C%20(1-%5Clambda)%5Cvarphi(x)%20%2B%20%5Clambda%5Cvarphi(y)%20-%20(1-r)%5Cvarphi(x)%20-%20r%5Cvarphi(y)%20%7C%24%20are%20both%20arbitrarily%20small%2C%20and%20thus%20we%20have%20convexity.%0A%0A%20%20%20%20This%20approximation%20apparently%20requires%20continuity%2C%20so%20in%20general%20if%20we%20omit%20the%20continuity%20requirement%2C%20%24%5Cvarphi%24%20may%20not%20be%20convex.%0A%0A%20%20%20%20%23%23%20Exercise%206%0A%0A%20%20%20%20**Let%20%24m%24%20be%20Lebesgue%20measure%20on%20%24%5B0%2C1%5D%24%20and%20define%20%24%5C%7C%20f%20%5C%7C_p%24%20with%20respect%20to%20%24m%24.%20Find%20all%20functions%20%24%5CPhi%24%20on%20%24%5B0%2C%5Cinfty)%24%20such%20that%20the%20relation%20%24%5CPhi(%5Clim%5Climits_%7Bp%5Cto%200%7D%20%5C%7C%20f%20%5C%7C_p)%20%3D%20%5Cint%5Climits_0%5E1%20(%5CPhi%20%5Ccirc%20f)dm%24%20holds%20for%20every%20bounded%20measurable%20positive%20%24f%24.%20Show%20first%20that%20%24c%5CPhi(x)%20%2B%20(1-c)%5CPhi(1)%20%3D%20%5CPhi(x%5Ec)%2C%20x%3E0%2C%200%20%5Cle%20c%20%5Cle%201%24.%20Compare%20with%20Exercise%205(d)**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20Fix%20any%20%24x%20%3E%200%24%20and%20consider%20the%20function%20%24f%20%3D%20%5Cchi_Cx%20%2B%20%5Cchi_%7BC'%7D%24%20for%20some%20%24c%20%5Cin%20%5B0%2C1%5D%24%20and%20%24C%20%3D%20%5B0%2Cc%5D%2C%20C'%20%3D%20(c%2C1%5D%24.%20This%20%24f%24%20is%20by%20construction%20bounded%2C%20measurable%2C%20and%20positive.%20We%20have%2C%20by%20definition%2C%20%24%5C%7C%20f%20%5C%7C_p%20%3D%20(%5Cint_%7B%5B0%2C1%5D%7D%20%7C%20%5Cchi_Cx%20%2B%20%5Cchi_%7BC'%7D%20%7C%5Ep%20dm)%5E%7B1%2Fp%7D%24.%20Notice%20that%20we%20have%20%24%7C%20%5Cchi_C%20x%20%2B%20%5Cchi_%7BC'%7D%20%7C%5Ep%20%3D%20%5Cchi_Cx%5Ep%20%2B%20%5Cchi_%7BC'%7D%24%2C%20so%20we%20can%20continue%20write%20%24%5C%7C%20f%20%5C%7C_p%20%3D%20(%5Cint_%7B%5B0%2C1%5D%7D%20%5Cchi_Cx%5Ep%20%2B%20%5Cchi_%7BC'%7D%20dm)%5E%7B1%2Fp%7D%24%20%24%3D%20(m(C)x%5Ep%20%2B%20m(C'))%5E%7B1%2Fp%7D%24%20%24%3D%20(cx%5Ep%20%2B%20(1-c))%5E%7B1%2Fp%7D%24.%0A%0A%20%20%20%20So%20now%20%24%5CPhi(%5Clim%5Climits_%7Bp%5Cto%200%7D%20%5C%7C%20f%20%5C%7C_p)%20%3D%20%5CPhi(%5Clim%5Climits_%7Bp%20%5Cto%200%7D(cx%5Ep%20%2B%20(1-c))%5E%7B1%2Fp%7D)%24%2C%20the%20limit%20inside%20can%20be%20calculated%20by%20taking%20log%20and%20apply%20the%20L'Hospital's%20rule%2C%20and%20the%20result%20is%20%24x%5Ec%24%2C%20so%20we%20have%20%24%5CPhi(%5Clim%5Climits_%7Bp%5Cto%200%7D%20%5C%7C%20f%20%5C%7C_p)%20%3D%20%5CPhi(x%5Ec)%24.%0A%0A%20%20%20%20On%20the%20other%20hand%2C%20%24%5CPhi%20%5Ccirc%20f%20%3D%20%5CPhi(%5Cchi_Cx%20%2B%20%5Cchi_%7BC'%7D)%20%3D%20%5Cchi_C%5CPhi(x)%20%2B%20%5Cchi_%7BC'%7D%5CPhi(1)%24%2C%20so%20%24%5Cint%5Climits_0%5E1%20(%5CPhi%20%5Ccirc%20f)dm%20%3D%20%5Cint%5Climits_0%5E1%20%5Cchi_C%5CPhi(x)%20%2B%20%5Cchi_%7BC'%7D%5CPhi(1)%20dm%24%20%24%3Dm(C)%5CPhi(x)%20%2B%20m(C')%5CPhi(1)%24%20%24%3Dc%5CPhi(x)%20%2B%20(1-c)%5CPhi(1)%24.%0A%0A%20%20%20%20Combine%20them%2C%20thus%20if%20by%20assumption%20%24%5CPhi(%5Clim%5Climits_%7Bp%5Cto%200%7D%20%5C%7C%20f%20%5C%7C_p)%20%3D%20%5Cint%5Climits_0%5E1%20(%5CPhi%20%5Ccirc%20f)dm%24%2C%20then%20we%20must%20have%20%24c%5CPhi(x)%20%2B%20(1-c)%5CPhi(1)%20%3D%20%5CPhi(x%5Ec)%24%2C%20%24x%24%20can%20be%20any%20positive%20number%2C%20and%20%240%20%5Cle%20c%20%5Cle%201%24.%0A%0A%20%20%20%20Now%20if%20%24c%3E1%24%2C%20then%20%240%5Cle%201%2Fc%20%5Cle%201%24%2C%20so%20%241%2Fc%5CPhi(x)%20%2B%20(1-1%2Fc)%5CPhi(1)%20%3D%20%5CPhi(x%5E%7B1%2Fc%7D)%24%2C%20substitute%20%24y%20%3D%20x%5E%7B1%2Fc%7D%24%2C%20we%20can%20get%20%241%2Fc%5CPhi(y%5Ec)%20%2B%20(1-1%2Fc)%5CPhi(1)%20%3D%20%5CPhi(y)%24%20%24%5Cimplies%20c%5CPhi(y)%20%2B%20(1-c)%5CPhi(1)%20%3D%20%5CPhi(y%5Ec)%24%2C%20so%20actually%20the%20equality%20holds%20for%20all%20%24c%20%5Cge%200%24.%0A%0A%20%20%20%20Consider%20the%20function%20%24%5Cvarphi%24%20defined%20as%20%24%5Cvarphi(x)%20%3D%20%5CPhi(x)%20-%20%5CPhi(1)%24%2C%20then%20%24%5Cvarphi(x%5Ec)%20%3D%20c%5Cvarphi(x)%24.%20In%20particular%2C%20note%20that%20any%20number%20%24y%20%3E%200%24%20can%20be%20written%20as%20%24y%20%3D%20e%5E%7B%5Cln%7By%7D%7D%24%2C%20thus%20%24%5Cvarphi(y)%20%3D%20%5Cvarphi(e)%5Cln%7By%7D%20%3D%20(%5CPhi(e)%20-%20%5CPhi(1))%5Cln%7By%7D%24.%20In%20case%20that%20%24y%3C1%24%2C%20we%20need%20to%20write%20it%20as%20%24%5Cvarphi(y)%20%3D%20%5Cvarphi(1%2Fe)%5Cln%7B1%2Fy%7D%20%3D%20-(%5CPhi(1%2Fe)-%5CPhi(1))%5Cln%7By%7D%24%20(because%20we%20require%20%24c%20%5Cge%200%24)%2C%20and%20thus%20%24%5CPhi(x)%24%20has%20the%20form%20%24%24%5CPhi(x)%20%3D%20%5Cbegin%7Bcases%7D%20(%5CPhi(e)%20-%20%5CPhi(1))%5Cln%7Bx%7D%20%2B%20%5CPhi(1)%2C%26%20x%5Cge%201%20%5C%5C%20%5CPhi(1)%20-(%5CPhi(1%2Fe)%20-%20%5CPhi(1))%5Cln%7Bx%7D%2C%26%200%3Cx%3C1%20%5Cend%7Bcases%7D.%24%24%0A%0A%20%20%20%20%23%23%20Exercise%2010%0A%0A%20%20%20%20**Suppose%20%24f_n%20%5Cin%20L%5Ep(%5Cmu)%24%20for%20%24n%20%3D%201%2C2%2C%5Cdots%24%20and%20%24%5C%7C%20f_n%20-%20f%20%5C%7C_p%20%5Cto%200%24%20and%20%24f_n%20%5Cto%20g%24%20a.e.%20as%20%24n%20%5Cto%20%5Cinfty%24.%20What%20relation%20exists%20between%20%24f%24%20and%20%24g%24**%3F%0A%0A%20%20%20%20Solution%0A%0A%20%20%20%20From%20%24%5C%7C%20f_n%20-%20f%20%5C%7C_p%20%5Cto%200%24%20we%20have%20%24f_n%20%5Cto%20f%24%20a.e.%20(the%20first%20part%20of%20proof%20of%20Theorem%203.11%20that%20%24L%5Ep(%5Cmu)%24%20is%20complete).%20Since%20%24f_n%20%5Cto%20g%24%20a.e.%20as%20well%2C%20it%20is%20easy%20to%20see%20%24f%20%3D%20g%24%20a.e.%20(union%20of%20two%20measure%20zero%20sets%20is%20of%20measure%20zero).%0A%0A%20%20%20%20%23%23%20Exercise%2011%0A%0A%20%20%20%20**Suppose%20%24%5Cmu(%5COmega)%20%3D%201%24%2C%20and%20suppose%20%24f%2Cg%24%20are%20positive%20measurable%20functions%20on%20%24%5COmega%24%20such%20that%20%24fg%20%5Cge%201%24.%20Prove%20that%20%24%5Cint_%7B%5COmega%7Dfd%5Cmu%20%5Ccdot%20%5Cint_%7B%5COmega%7Dgd%5Cmu%20%5Cge%201%24**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20If%20%24f%2Cg%24%20are%20positive%20measurable%20functions%20and%20%24fg%20%5Cge%201%24%2C%20then%20%24%5Csqrt%7Bf%7D%2C%5Csqrt%7Bg%7D%24%20are%20also%20positive%20measurable%20functions%20with%20%24%5Csqrt%7Bfg%7D%20%5Cge%201%24%20(measurable%20because%20taking%20square%20root%20is%20a%20continuous%20function).%0A%0A%20%20%20%20So%20now%20we%20can%20apply%20H%C3%B6lder's%20inequality%20on%20%24%5Csqrt%7Bf%7D%2C%20%5Csqrt%7Bg%7D%24%20with%20%24p%20%3D%20q%20%3D%202%24%2C%20then%20%24%24(%5Cint_%7B%5COmega%7D%5Csqrt%7Bf%7D%5E2d%5Cmu)%5E%7B1%2F2%7D(%5Cint_%7B%5COmega%7D%5Csqrt%7Bg%7D%5E2d%5Cmu)%5E%7B1%2F2%7D%20%5Cge%20%5Cint_%7B%5COmega%7D%5Csqrt%7Bfg%7Dd%5Cmu%20%5Cge%20%5Cint_%7B%5COmega%7D1%20d%5Cmu%20%3D%201.%24%24%20Taking%20square%20both%20side%20and%20we%20have%20the%20desired%20inequality.%0A%0A%20%20%20%20(A%20little%20remark%3A%20here%20is%20no%20need%20to%20argue%20if%20%24f%2C%20%5Csqrt%7Bf%7D%24%20are%20%24L%5E2%24%2C%20etc.%2C%20because%20the%20integrals%20are%20defined%20anyway%2C%20and%20it%20does%20not%20matter%20if%20it%20goes%20to%20infinity%20because%20we%20only%20want%20to%20show%20it%20is%20%24%5Cge%201%24)%0A%0A%20%20%20%20%23%23%20Exercise%2012%0A%0A%20%20%20%20**Suppose%20%24%5Cmu(%5COmega)%20%3D%201%24%20and%20%24h%3A%20%5COmega%20%5Cto%20%5B0%2C%5Cinfty%5D%24%20is%20measurable.%20If%20%24A%20%3D%20%5Cint_%7B%5COmega%7Dhd%5Cmu%24%2C%20prove%20that%20%24%5Csqrt%7B1%2BA%5E2%7D%20%5Cle%20%5Cint_%7B%5COmega%7D%5Csqrt%7B1%2Bh%5E2%7Dd%5Cmu%20%5Cle%201%2BA%24.%20If%20%24%5Cmu%24%20is%20Lebesgue%20measure%20on%20%24%5B0%2C1%5D%24%20and%20%24h%24%20is%20continuous%2C%20%24h%20%3D%20f'%24%2C%20the%20above%20inequality%20have%20a%20simple%20geometric%20interpretation.%20From%20this%2C%20conjecture%20for%20general%20%24%5COmega%24%20under%20what%20conditions%20on%20%24h%24%20equality%20can%20hold%20in%20either%20of%20the%20above%20inequalities%2C%20and%20prove%20your%20conjecture**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%201.%20Prove%20the%20inequality%3A%0A%20%20%20%20%091.%20For%20the%20first%20part%2C%20notice%20that%20it%20is%20in%20the%20form%20of%20the%20Jensen's%20Inequality%2C%20so%20this%20part%20is%20proven%20if%20%24%5Csqrt%7B1%2Bx%5E2%7D%24%20is%20convex%3A%20indeed%2C%20take%20derivative%20we%20have%20%24x(1%2Bx%5E2)%5E%7B-1%2F2%7D%24%20and%20second%20derivative%20we%20have%20%24(1%2Bx%5E2)%5E%7B-3%2F2%7D%24%20which%20is%20always%20positive%2C%20thus%20we%20have%20convexity%3B%0A%20%20%20%20%092.%20For%20the%20second%20part%2C%20write%20%241%2BA%20%3D%201%20%2B%20%5Cint_%7B%5COmega%7Dhd%5Cmu%24%20%24%3D%20%5Cint_%7B%5COmega%7D1%20%2B%20h%20d%5Cmu%24%20(because%20%24%5Cmu(%5COmega)%3D1%24)%2C%20then%20we%20can%20see%20if%20we%20can%20prove%20%24%5Csqrt%7B1%2Bh%5E2%7D%20%5Cle%201%20%2B%20h%24%20then%20this%20part%20is%20proven%3A%20this%20just%20comes%20from%20that%20%241%2Bh%5E2%20%5Cle%201%2B2h%2Bh%5E2%20%3D%20(1%2Bh)%5E2%24%20if%20%24h%5Cge%200%24%3B%0A%20%20%20%202.%20A%20geometric%20interpretation%3A%0A%20%20%20%20%09In%20this%20setting%20%24%5Cint_%7B%5COmega%7D%5Csqrt%7B1%2Bh%5E2%7Dd%5Cmu%20%3D%20%5Cint_0%5E1%20%5Csqrt%7B1%20%2B%20f'%5E2%7Ddx%24%2C%20recall%20that%20this%20integral%20calculates%20the%20distance%20between%20%24f(0)%24%20and%20%24f(1)%24%20along%20the%20curve%20%24f%24.%0A%20%20%20%20%09By%20Fundamental%20Theorem%20of%20Calculus%2C%20%24A%20%3D%20f(1)%20-%20f(0)%24%2C%20so%20%24%5Csqrt%7B1%2BA%5E2%7D%24%20can%20be%20thought%20as%20the%20straight-segment%20distance%20between%20%24f(0)%24%20and%20%24f(1)%24%3B%20%241%2BA%24%20can%20be%20thought%20as%20the%20two-straight-segment-passing-corner%20distance%20between%20%24f(0)%24%20and%20%24f(1)%24.%0A%20%20%20%20%09So%20this%20inequality%20just%20gives%20a%20natural%20range%20of%20the%20curve%20length%20of%20the%20graph%20of%20a%20non-decreasing%20(because%20%24f'%20%3D%20h%24%20is%20by%20assumption%20non-negative)%20function%3B%0A%20%20%20%203.%20From%20the%20geometric%20interpretation%20we%20could%20guess%20that%20the%20first%20equality%20holds%20if%20and%20only%20if%20%24f%24%20is%20a%20straight%20line%2C%20i.e.%20%24f'%20%3D%20h%24%20is%20a%20constant%20(with%20a%20little%20calculation%2C%20this%20constant%20is%20%24A%24)%20a.e.%2C%20and%20the%20second%20equality%20holds%20if%20and%20only%20if%20%24f%24%20is%20a%20horizontal%20line%20(it%20cannot%20have%20height%2C%20otherwise%20it%20is%20not%20a%20graph)%2C%20i.e.%20%24f'%20%3D%20h%20%3D%200%24%20a.e.%20(in%20this%20case%20the%20first%20equality%20also%20holds)%3A%0A%20%20%20%20%091.%20If%20%24h%20%3D%20A%24%20is%20then%20%24%5Cint_%7B%5COmega%7D%5Csqrt%7B1%2Bh%5E2%7Dd%5Cmu%20%3D%20%5Csqrt%7B1%2Bh%5E2%7D%20%3D%20%5Csqrt%7B1%2BA%5E2%7D%24%2C%20and%20if%20%24h%20%3D%200%20%3D%20A%24%20then%20%24%5Cint_%7B%5COmega%7D%5Csqrt%7B1%2Bh%5E2%7Dd%5Cmu%20%3D%20%5Csqrt%7B1%2Bh%5E2%7D%20%3D%201%20%3D%201%2B0%20%3D%20A%24%3B%0A%20%20%20%20%092.%20Converse%20for%20the%20first%20equality%20comes%20from%20Jensen's%20Inequality's%20proof%2C%20if%20%24%5Csqrt%7B1%2BA%5E2%7D%20%3D%20%5Cint_%7B%5COmega%7D%5Csqrt%7B1%2Bh%5E2%7Dd%5Cmu%24%20then%20%24%5Cvarphi(h)%20%3D%20%5Cvarphi(A)%20%2B%20%5Cbeta(h%20-A)%24%2C%20thus%20either%20%24%5Cvarphi%24%20is%20affine%20(not%20our%20case%3A%20we%20have%20%24%5Cvarphi(x)%20%3D%20%5Csqrt%7B1%2Bx%5E2%7D%24)%20or%20%24h%24%20is%20constant%20and%20must%20be%20%24A%24%20a.e%3B%0A%20%20%20%20%093.%20Converse%20for%20the%20second%20equality%3A%20we%20need%20to%20have%20%241%2Bh%5E2%20%3D%20(1%2Bh)%5E2%24%20a.e.%20thus%20%24h%20%3D%200%24%20a.e.%0A%0A%20%20%20%20(I%20am%20not%20sure%20what%20does%20the%20problem%20mean%20by%20'for%20general%20%24%5COmega%24'%2C%20apparently%20for%20different%20%24%5COmega%24%20thus%20different%20%24%5Cmu(%5COmega)%24%20the%20equalities%20are%20not%20the%20same%2C%20we%20need%20to%20change%20%241%24%20to%20%24%5Cmu(%5COmega)%24)%0A%0A%20%20%20%20%23%23%20Exercise%2013%0A%0A%20%20%20%20**Under%20what%20conditions%20on%20%24f%2Cg%24%20does%20equality%20hold%20in%20the%20conclusions%20of%20Theorem%203.8%20and%203.9%3F%20You%20may%20have%20to%20treat%20the%20cases%20%24p%20%3D%201%24%20and%20%24p%20%3D%20%5Cinfty%24%20separately**.%0A%0A%20%20%20%20Solution%0A%0A%20%20%20%20%23%23%23%20Part%201%0A%0A%20%20%20%20Theorem%203.8%20says%3A%20If%20%24p%2Cq%24%20are%20conjugate%20exponents%2C%20%241%20%5Cle%20p%20%5Cle%20%5Cinfty%24%2C%20and%20if%20%24f%20%5Cin%20L%5Ep(%5Cmu)%24%2C%20%24g%20%5Cin%20L%5Eq(%5Cmu)%24%2C%20then%20%24fg%20%5Cin%20L%5E1(%5Cmu)%24%20and%20%24%24%5C%7C%20fg%20%5C%7C_1%20%5Cle%20%5C%7C%20f%20%5C%7C_p%5C%7C%20g%20%5C%7C_q.%24%24%20If%20%241%20%3C%20p%20%3C%20%5Cinfty%24%20then%20this%20is%20H%C3%B6lder's%20inequality%2C%20so%20equality%20holds%20if%20equality%20in%20H%C3%B6lder's%20Inequality%20holds%2C%20which%20we%20know%20happens%20if%20and%20only%20if%20that%20%24af%5Ep%20%3D%20bg%5Eq%24%20a.e.%20for%20some%20non-zero%20constant%20%24a%2Cb%24%2C%20or%20that%20we%20have%20the%20trivial%20cases%20that%20at%20least%20one%20of%20%24f%2Cg%24%20is%20zero%20a.e.%0A%0A%20%20%20%20If%20%24p%20%3D%201%24%20(then%20%24q%20%3D%20%5Cinfty%24)%2C%20then%20we%20want%20%24%7C%20f(x)g(x)%20%7C%20%3D%20%5C%7C%20f%20%5C%7C_%7B%5Cinfty%7D%20%7C%20g(x)%20%7C%24%20for%20almost%20all%20%24x%24%2C%20which%20happens%20if%20and%20only%20if%20%24g%20%5Cequiv%200%24%20or%20%24%5C%7C%20f%20%5C%7C_%7B%5Cinfty%7D%20%3D%20%7C%20f(x)%20%7C%24%20(then%20product%20of%20absolute%20values%20is%20absolute%20value%20of%20products)%2C%20i.e.%20%24f%24%20is%20constant%20a.e..%20The%20case%20%24p%20%3D%20%5Cinfty%24%20is%20the%20analogous.%0A%0A%20%20%20%20%23%23%23%20Part%202%0A%0A%20%20%20%20Theorem%203.9%20says%3A%20If%20%241%20%5Cle%20p%20%5Cle%20%5Cinfty%24%20and%20%24f%2Cg%20%5Cin%20L%5Ep(%5Cmu)%24%2C%20then%20%24f%2Bg%20%5Cin%20L%5Ep(%5Cmu)%24%20and%20%24%24%5C%7C%20f%2Bg%20%5C%7C_p%20%5Cle%20%5C%7C%20f%20%5C%7C_p%2B%5C%7C%20g%20%5C%7C_p.%24%24%20If%20%241%3C%20p%20%3C%20%5Cinfty%24%20then%20this%20is%20Minkowski's%20Inequality.%20Look%20at%20the%20proof%20of%20Minkowski's%20Inequality%2C%20we%20can%20see%20that%20equality%20holds%20if%20and%20only%20if%20at%20least%20one%20of%20%24f%2Cg%24%20is%20%240%24%20a.e.%20or%20equality%20in%20H%C3%B6lder's%20Inequality%20holds%20for%20%24f%24%20and%20%24(f%2Bg)%5E%7Bp-1%7D%24%2C%20and%20for%20%24g%24%20and%20%24(f%2Bg)%5E%7Bp-1%7D%24%2C%20i.e.%20exists%20non-zero%20%24a%2Cb%24%20such%20that%20%24af%5Ep%20%3D%20b(f%2Bg)%5E%7Bqp-q%7D%24%20a.e.%20and%20non-zero%20%24c%2Cd%24%20such%20that%20%24cg%5Ep%20%3D%20d(f%2Bg)%5E%7Bqp-q%7D%24%20a.e.%0A%0A%20%20%20%20Put%20things%20together%2C%20equality%20holds%20if%20and%20only%20if%20that%20at%20least%20one%20of%20%24f%2Cg%24%20is%20zero%20a.e.%2C%20or%20that%20%24f%20%3D%20ag%24%20for%20some%20non-zero%20%24a%24%20a.e.%0A%0A%20%20%20%20If%20%24p%20%3D%201%24%2C%20equality%20holds%20if%20and%20only%20if%20%24%7C%20f%2B%20g%20%7C%20%3D%20%7C%20f%20%7C%20%2B%20%7C%20g%20%7C%24%20a.e.%2C%20which%20holds%20if%20and%20only%20if%20%24f%2Cg%24%20are%20both%20non-negative%20or%20both%20non-positive%2C%20or%20at%20least%20one%20of%20%24f%2Cg%24%20is%20%240%24%20a.e.%0A%0A%20%20%20%20If%20%24p%20%3D%20%5Cinfty%24%2C%20then%20equality%20holds%20if%20essential%20supreme%20of%20%24f%2Bg%24%20equals%20sum%20of%20essential%20supremes%2C%20i.e.%20%24f%2Cg%24%20achieve%2Fapproaches%20their%20essential%20supreme%20at%20the%20same%20time%20(this%20include%20the%20cases%20that%20at%20least%20on%20of%20%24f%2Cg%24%20is%20zero%20a.e.).%0A%0A%20%20%20%20%23%23%20Exercise%2014%0A%0A%20%20%20%20**Suppose%20%241%20%3C%20p%20%3C%20%5Cinfty%24%2C%20%24f%20%5Cin%20L%5Ep%20%3D%20L%5Ep((0%2C%5Cinfty))%24%20relative%20to%20Lebesgue%20measure%2C%20and%20%24F(x)%20%3D%20%5Cfrac%7B1%7D%7Bx%7D%5Cint_0%5Ex%20f(t)dt%2C%20(0%20%3C%20x%20%3C%20%5Cinfty)%24**%3A%0A%0A%20%20%20%201.%20**Prove%20Hardy's%20inequality%20%24%5C%7C%20F%20%5C%7C_p%20%5Cle%20%5Cfrac%7Bp%7D%7Bp-1%7D%5C%7C%20f%20%5C%7C_p%24%20which%20shows%20that%20the%20mapping%20%24f%20%5Cto%20F%24%20carries%20%24L%5Ep%24%20into%20%24L%5Ep%24**%3B%0A%20%20%20%202.%20**Prove%20that%20equality%20holds%20only%20if%20%24f%20%3D%200%24%20a.e**%3B%0A%20%20%20%203.%20**Prove%20that%20the%20constant%20%24p%2F(p-1)%24%20cannot%20be%20replaced%20by%20a%20smaller%20one**%3B%0A%20%20%20%204.%20**If%20%24f%20%3E%200%24%20and%20%24f%20%5Cin%20L%5E1%24%2C%20prove%20that%20%24F%20%5Cnotin%20L%5E1%24**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20%23%23%23%20Part%201%0A%0A%20%20%20%20Let%20%24q%24%20be%20the%20conjugate%20exponent%20of%20%24p%24%2C%20notice%20that%20then%20we%20can%20write%20%24q%20%3D%20%5Cfrac%7Bp%7D%7Bp-1%7D%24.%0A%0A%20%20%20%20Start%20by%20%24F(x)%24%20and%20we%20will%20'construct'%20%24%5C%7C%20F%20%5C%7C_p%24.%20Let%20%24a%24%20be%20some%20real%20number%2C%20we%20can%20write%20%24%24%7C%20x%20F(x)%20%7C%20%3A%3D%20%7C%20%5Cint%5Climits_0%5Exf(t)t%5Eat%5E%7B-a%7Ddt%7C.%24%24%0A%20%20%20%20Apply%20H%C3%B6lder's%20Inequality%20%24%24%5Cle(%5Cint%5Climits_0%5Ex%7C%20f(t)%7C%5Ept%5E%7Bpa%7Ddt)%5E%7B1%2Fp%7D(%5Cint%5Climits_0%5Ext%5E%7B-qa%7Ddt)%5E%7B1%2Fq%7D.%24%24%20Now%20just%20do%20some%20calculation%2C%20notice%20that%20we%20need%20%240%3Cqa%3C1%24%20for%20this%20to%20work%20%24%24%3D%20(%5Cint%5Climits_0%5Ex%7C%20f(t)%7C%5Ept%5E%7Bpa%7Ddt)%5E%7B1%2Fp%7D(%5Cfrac%7Bx%5E%7B1-qa%7D%7D%7B1-qa%7D)%5E%7B1%2Fq%7D.%24%24%20Since%20%24x%24%20is%20just%20viewed%20as%20a%20positive%20constant%2C%20we%20can%20write%20%24%24%5Cbegin%7Baligned%7D%7C%20F(x)%20%7C%20%26%5Cle%20(%5Cint%5Climits_0%5Ex%7C%20f(t)%7C%5Ept%5E%7Bpa%7Ddt)%5E%7B1%2Fp%7Dx%5E%7B1%2Fq-a-1%7D(1-qa)%5E%7B-1%2Fq%7D%20%5C%5C%20%7C%20F(x)%20%7C%5Ep%20%26%5Cle%20(%5Cint%5Climits_0%5Ex%7C%20f(t)%7C%5Ept%5E%7Bpa%7Ddt)~x%5E%7Bp%2Fq-pa-p%7D~(1-qa)%5E%7B-p%2Fq%7D%20%5C%5C%20%26%3D%20(%5Cint%5Climits_0%5Ex%7C%20f(t)%7C%5Ept%5E%7Bpa%7Ddt)~x%5E%7B-1-pa%7D~(1-qa)%5E%7B-p%2Fq%7D%20%5Cend%7Baligned%7D.%24%24%20Take%20integral%20we%20have%20%24%24%5Cbegin%7Baligned%7D%20%5Cint%5Climits_0%5E%5Cinfty%20%7C%20F(x)%20%7C%5Ep%20dx%20%26%5Cle%20%5Cint%5Climits_0%5E%5Cinfty%20%5B(%5Cint%5Climits_0%5Ex%7C%20f(t)%7C%5Ept%5E%7Bpa%7Ddt)~x%5E%7B-1-pa%7D~(1-qa)%5E%7B-p%2Fq%7D%5Ddx%20%5C%5C%20%26%3D%20(1-qa)%5E%7B-p%2Fq%7D%5Cint%5Climits_0%5E%5Cinfty%5Cint%5Climits_0%5Ex%7C%20f(t)%7C%5Ept%5E%7Bpa%7Dx%5E%7B-1-pa%7Ddtdx%5Cend%7Baligned%7D.%24%24%20We%20do%20a%20change%20of%20order%20of%20taking%20integral%2C%20it%20is%20viable%20because%20everything%20inside%20is%20positive.%20Then%20just%20do%20some%20simplification%20%24%24%5Cbegin%7Baligned%7D%26%3D%20(1-qa)%5E%7B-p%2Fq%7D%5Cint%5Climits_0%5E%5Cinfty%5Cint%5Climits_t%5E%7B%5Cinfty%7D%7C%20f(t)%7C%5Ept%5E%7Bpa%7Dx%5E%7B-1-pa%7Ddxdt%20%5C%5C%20%26%3D%20(1-qa)%5E%7B-p%2Fq%7D%5Cint%5Climits_0%5E%5Cinfty%7C%20f(t)%7C%5Ept%5E%7Bpa%7D(%5Cint%5Climits_t%5E%7B%5Cinfty%7Dx%5E%7B-1-pa%7Ddx)dt%20%5C%5C%20%26%3D%20(1-qa)%5E%7B-p%2Fq%7D%5Cint%5Climits_0%5E%5Cinfty%7C%20f(t)%7C%5Ept%5E%7Bpa%7D(%5Cfrac%7Bt%5E%7B-pa%7D%7D%7Bpa%7D)dt%20%5C%5C%20%26%3D%20(1-qa)%5E%7B-p%2Fq%7D%5Cfrac%7B1%7D%7Bpa%7D%5Cint%5Climits_0%5E%5Cinfty%7C%20f(t)%7C%5Epdt%5Cend%7Baligned%7D.%24%24%20Thus%20%24%24%5Cbegin%7Baligned%7D%5C%7C%20F(x)%20%5C%7C_p%20%26%5Cle%20((1-qa)%5E%7B-p%2Fq%7D%5Cfrac%7B1%7D%7Bpa%7D)%5E%7B1%2Fp%7D%5C%7C%20f(x)%20%5C%7C_p%5Cend%7Baligned%7D.%24%24%20If%20we%20now%20take%20%24a%20%3D%20%5Cfrac%7B1%7D%7Bpq%7D%24%20then%20%24((1-qa)%5E%7B-p%2Fq%7D%5Cfrac%7B1%7D%7Bpa%7D)%5E%7B1%2Fp%7D%24%20%24%3D%20((1-%5Cfrac%7B1%7D%7Bp%7D)%5E%7B-p%2Fq%7Dq)%5E%7B1%2Fp%7D%24%20%24%3D(q%5E%7Bp%2Fq%7Dq)%5E%7B1%2Fp%7D%20%3D%20q%24%2C%20and%20we%20have%20the%20desired%20inequality.%0A%0A%20%20%20%20%23%23%23%20Part%202%0A%0A%20%20%20%20Look%20at%20the%20above%20proof%20and%20we%20see%20the%20only%20inequality%20is%20the%20first%20step%20when%20we%20apply%20H%C3%B6lder%2C%20so%20equality%20in%20Hardy's%20Inequality%20holds%20if%20and%20only%20if%20the%20equality%20in%20H%C3%B6lder's%20Inequality%20holds%2C%20that%20is%20either%20we%20have%20the%20trivial%20case%20%24f%20%3D%200%24%20a.e.%20or%20there%20are%20non-zero%20%24m%2Cn%24%20such%20that%20%24m(f(t)t%5E%7B1%2Fpq%7D)%5Ep%20%3D%20n(t%5E%7B-1%2Fpq%7D)%5Eq%24%20%24%5Cimplies%20f%5Ep%20%3D%20%5Cfrac%7Bc%7D%7Bt%7D%24%2C%20but%20then%20%24%5C%7C%20f%20%5C%7C_p%20%3D%20(%5Cint%5Climits_0%5E%7B%5Cinfty%7D%5Cfrac%7Bc%7D%7Bt%7Ddt)%5E%7B1%2Fp%7D%24%20and%20it%20is%20known%20that%20%24%5Cint%5Climits_0%5E%7B%5Cinfty%7D%5Cfrac%7B1%7D%7Bt%7Ddt%24%20does%20not%20converge%2C%20so%20%24f%24%20is%20not%20%24L%5Ep%24%20thus%20this%20is%20not%20an%20viable%20option%2C%20so%20equality%20only%20holds%20at%20the%20trivial%20case%2C%20i.e.%20%24f%20%3D%200%24%20a.e.%0A%0A%20%20%20%20%23%23%23%20Part%203%0A%0A%20%20%20%20Prove%20by%20an%20example%3A%20consider%20%24f(x)%20%3D%20x%5E%7B-1%2Fp%7D%24%20on%20%24%5B1%2CA%5D%24%20and%20%240%24%20elsewhere%20for%20some%20%24A%3E1%24%2C%20then%20%24%5C%7C%20f%20%5C%7C_p%20%3D%20(%5Cint%5Climits_1%5EA%20x%5E%7B-1%7D%20dx)%5E%7B1%2Fp%7D%20%3D%20(%5Clog%7BA%7D)%5E%7B1%2Fp%7D%24.%20On%20the%20other%20hand%2C%20if%20%24x%3C1%24%20then%20%24F(x)%20%3D%200%24%2C%20if%20%241%3Cx%3CA%24%20then%20%24F(x)%20%3D%20%5Cfrac%7B1%7D%7Bx%7D%5Cint%5Climits_1%5Ext%5E%7B-1%2Fp%7Ddt%24%20%24%3D%20%5Cfrac%7Bpx%5E%7B-1%2Fp%7D-p%7D%7Bp-1%7D%24%2C%20and%20if%20%24A%3Cx%24%20then%20%24F(x)%20%3D%20%5Cfrac%7BpA%5E%7B1-1%2Fp%7D-p%7D%7Bx(p-1)%7D%24.%0A%0A%20%20%20%20(I%20think%20the%20idea%20here%20is%20then%20to%20calculate%20%24%5Clim%5Climits_%7BA%20%5Cto%20%5Cinfty%7D%5Cfrac%7B%5C%7C%20F%20%5C%7C_p%7D%7B%5C%7C%20f%20%5C%7C_p%7D%24.%20To%20calculate%20%24%5C%7C%20F%20%5C%7C_p%24%2C%20first%20evaluate%20from%20%241%24%20to%20%24A%24%2C%20then%20from%20%24A%24%20to%20%24%5Cinfty%24%2C%20which%20is%20%24%24%5C%7C%20F%20%5C%7C_p%5Ep%20%3D%20%5Cint%5Climits_1%5EA(%5Cfrac%7Bpx%5E%7B-1%2Fp%7D-p%7D%7Bp-1%7D)%5Epdx%20%2B%20%5Cint%5Climits_A%5E%7B%5Cinfty%7D(%5Cfrac%7BpA%5E%7B1-1%2Fp%7D-p%7D%7Bx(p-1)%7D)%5Epdx%2C%24%24%20but%20I'm%20not%20sure%20how%20to%20do%20that%20by%20hand%2C%20the%20result%20from%20WolframAlpha%20gives%3A%20the%20limit%20is%20%24%5Cfrac%7Bp%7D%7Bp-1%7D%24%2C%20and%20thus%20no%20constant%20smaller%20than%20this%20value%20could%20work%20in%20the%20Hardy's%20Inequality)%0A%0A%20%20%20%20%23%23%23%20Part%204%0A%0A%20%20%20%20Under%20the%20assumption%20we%20have%20%24F(x)%20%3D%20%5Cfrac%7B1%7D%7Bx%7D%5Cint_0%5Ex%20f(t)dt%24%2C%20choose%20some%20(measurable)%20set%20%24E%24%20such%20that%20%24f(E)%20%3E%20a%24%20for%20a%20positive%20%24a%24%20(existence%20is%20by%20that%20%24f%3E0%24)%2C%20then%20for%20%24x%3E%5Csup%7BE%7D%24%20we%20have%20%24F(x)%20%5Cge%20%5Cfrac%7B1%7D%7Bx%7D%20%5Cint_Ead%5Cmu%20%3D%20%5Cfrac%7Ba%5Cmu(E)%7D%7Bx%7D%20%3D%20%5Cfrac%7Bc%7D%7Bx%7D%24%2C%20yet%20it%20is%20known%20that%20%24%5Cint_n%5E%7B%5Cinfty%7D%5Cfrac%7Bc%7D%7Bx%7Ddx%24%20does%20not%20converge%20for%20any%20%24n%24%2C%20thus%20%24F%24%20is%20not%20%24L%5E1%24.%0A%0A%20%20%20%20%23%23%20Exercise%2018%0A%0A%20%20%20%20**Let%20%24%5Cmu%24%20be%20a%20positive%20measure%20on%20%24X%24.%20A%20sequence%20%24%5Clbrace%20f_n%20%5Crbrace%24%20of%20complex%20measurable%20functions%20on%20%24X%24%20is%20said%20to%20converge%20in%20measure%20to%20the%20measurable%20function%20%24f%24%20if%20to%20every%20%24%5Cvarepsilon%20there%20corresponds%20an%20%24N%24%20such%20that%20%24%5Cmu(%5Clbrace%20x%20%3A%20%7C%20f_n(x)%20-%20f(x)%20%7C%20%3E%20%5Cvarepsilon%5Crbrace)%20%3C%20%5Cvarepsilon%24%20for%20all%20%24n%20%3E%20N%24.%20Assume%20%24%5Cmu(X)%20%3C%5Cinfty%24%20and%20prove%20the%20following%20statements**%3A%0A%0A%20%20%20%201.%20**If%20%24f_n(x)%20%5Cto%20f(x)%24%20a.e.%2C%20then%20%24f_n%20%5Cto%20f%24%20in%20measure**%3B%0A%20%20%20%202.%20**If%20%24f_n%20%5Cin%20L%5Ep(%5Cmu)%24%20and%20%24%5C%7C%20f_n%20-%20f%20%5C%7C_p%20%5Cto%200%2C%201%20%5Cle%20p%20%5Cle%20%5Cinfty%24%2C%20then%20%24f_n%20%5Cto%20f%24%20in%20measure**%3B%0A%20%20%20%203.%20**If%20%24f_n%20%5Cto%20f%24%20in%20measure%2C%20then%20%24%5Clbrace%20f_n%20%5Crbrace%24%20has%20a%20subsequence%20which%20converges%20to%20%24f%24%20a.e**.%0A%0A%20%20%20%20**Investigate%20the%20converses%20of%201.%20And%202.%2C%20what%20happens%20to%201.%2C%202.%2C%203.%2C%20if%20%24%5Cmu(X)%20%3D%20%5Cinfty%24%2C%20for%20instance%2C%20if%20%24%5Cmu%24%20is%20the%20Lebesgue%20measure%20on%20%24%5Cmathbb%7BR%7D%24**%3F%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20%23%23%23%20Part%201%0A%0A%20%20%20%20Fix%20some%20%24%5Cvarepsilon%3E0%24.%0A%0A%20%20%20%20Suppose%20%24f_n(x)%20%5Cto%20f(x)%24%20a.e.%2C%20it%20means%20that%20for%20almost%20all%20%24x%24%2C%20%24f_n(x)%20%5Cto%20f(x)%24%2C%20which%20means%20that%20for%20almost%20all%20%24x%24%2C%20there%20exists%20some%20%24N_x%24%2C%20depending%20on%20%24x%24%2C%20such%20that%20%24%7C%20f_%7Bn_x%7D(x)%20-%20f(x)%20%7C%20%5Cle%20%5Cvarepsilon%24%20for%20all%20%24n_x%20%3E%20N_x%24.%20Now%20index%20on%20%24N_x%24%2C%20consider%20%24A_N%20%3D%20%5Clbrace%20x%20%3A%20%7C%20f_%7Bn_x%7D(x)%20-%20f(x)%20%7C%20%5Cle%20%5Cvarepsilon%24%20for%20all%20%24n_x%3EN%5Crbrace%24%2C%20then%20%24%5Clbrace%20x%20%7C%20f_n(x)%20%5Cto%20f(x)%20%5Crbrace%20%3D%20%5Ccup%20A_N%24.%20By%20assumption%20we%20have%20%24%5Cmu(%5Ccup%20A)%20%3D%20%5Cmu(X)%24.%0A%0A%20%20%20%20Write%20%24%5Clbrace%20x%20%3A%20%7C%20f_n(x)%20-%20f(x)%20%7C%20%3E%20%5Cvarepsilon%5Crbrace%24%20%24%3D%20X%20-%20%5Clbrace%20x%20%3A%20%7C%20f_n(x)%20-%20f(x)%20%7C%20%5Cle%20%5Cvarepsilon%24%20for%20some%20%24n%20%5Crbrace%24%2C%20so%20the%20measure%20%24%5Cmu(%5Clbrace%20x%20%3A%20%7C%20f_n(x)%20-%20f(x)%20%7C%20%3E%20%5Cvarepsilon%5Crbrace)%24%20%24%3D%20%5Cmu(X%20-%20%5Clbrace%20x%20%3A%20%7C%20f_n(x)%20-%20f(x)%20%7C%20%5Cle%20%5Cvarepsilon%24%20for%20some%20%24n%20%5Crbrace%20)%24.%20Since%20this%20should%20work%20for%20arbitrary%20%24%5Cvarepsilon%24%2C%20this%20value%20%24%3D%20%5Cmu(X%20-%20%5Ccup%20A)%20%3D%200%24%20as%20desired.%0A%0A%20%20%20%20%23%23%23%20Part%202%0A%0A%20%20%20%20Fix%20some%20%24%5Cvarepsilon%24.%0A%0A%20%20%20%20Suppose%20%24p%20%3D%20%5Cinfty%24%2C%20%24%5C%7C%20f_n(x)%20-%20f(x)%20%5C%7C_%7B%5Cinfty%7D%20%5Cto%200%24%20means%20%24%5C%7C%20f_n(x)%20-%20f(x)%20%5C%7C_%7B%5Cinfty%7D%20%5Cle%20%5Cvarepsilon%24%20for%20large%20enough%20%24n%24%2C%20which%20means%20%24%7C%20f_n(x)%20-%20f(x)%20%7C%20%5Cle%20%5Cvarepsilon%24%20a.e.%20by%20definition%20for%20large%20enough%20%24n%24.%20It%20is%20equivalent%20to%20say%20that%20%24%5Cmu(%5Clbrace%20x%20%3A%20%7C%20f_n(x)%20-%20f(x)%20%7C%20%3E%20%5Cvarepsilon%5Crbrace)%20%3D%200%24%20for%20large%20enough%20%24n%24.%0A%0A%20%20%20%20Suppose%20%241%20%5Cle%20p%20%3C%20%5Cinfty%24.%20Denote%20the%20set%20%24%5Clbrace%20x%20%3A%20%7C%20f_n(x)%20-%20f(x)%20%7C%20%3E%20%5Cvarepsilon%5Crbrace%20%3D%20B_n%24%2C%20then%20%24%5C%7C%20f_n%20-%20f%20%5C%7C_p%24%20%24%3A%3D%20(%5Cint_X%7C%20f_n%20-%20f%20%7C%5Ep)%5E%7B1%2Fp%7D%24%20%24%5Cge%20(%5Cint_%7BB_n%7D%5Cvarepsilon%5Ep)%5E%7B1%2Fp%7D%24%20%24%3D%20(%5Cvarepsilon%5Ep%5Cmu(B_n))%5E%7B1%2Fp%7D%24%2C%20thus%20%24%5Cmu(B_n)%20%5Cle%20(%5Cfrac%7B%5C%7C%20f_n%20-%20f%20%5C%7C_p%7D%7B%5Cvarepsilon%7D)%5Ep%24.%20Since%20%24%5C%7C%20f_n%20-%20f%20%5C%7C_p%20%5Cto%200%24%2C%20%24%5Cmu(B_n)%24%20thus%20%24%5Cmu(%5Clbrace%20x%20%3A%20%7C%20f_n(x)%20-%20f(x)%20%7C%20%3E%20%5Cvarepsilon%5Crbrace)%24%20takes%20arbitrarily%20small%20value%20for%20large%20enough%20%24n%24.%0A%0A%20%20%20%20%23%23%23%20Part%203%0A%0A%20%20%20%20For%20each%20%24i%24%2C%20there%20exists%20%24N_i%24%20such%20that%20%24%5Cmu(%5Clbrace%20x%20%3A%20%7C%20f_n(x)%20-%20f(x)%20%7C%20%3E%202%5E%7B-i%7D%5Crbrace)%20%3C%202%5E%7B-i%7D%24%20for%20all%20%24n%20%3E%20N_i%24.%20Pick%20%24n_i%20%3E%20N_i%24%20and%20%24n_i%20%3E%20n_%7Bi-1%7D%2C%5Cdots%2Cn_1%24%2C%20then%20we%20have%20a%20subsequence%20%24f_%7Bn_i%7D%24%20of%20%24f_n%24.%20Now%20use%20the%20fact%20that%20%24%5Csum%5Climits_%7Bi%7D%202%5E%7B-i%7D%20%3D%201%20%3C%20%5Cinfty%24%2C%20so%20almost%20all%20%24x%24%20lie%20in%20at%20most%20finitely%20many%20sets%20such%20that%20%24%5Clbrace%20x%20%3A%20%7C%20f_n(x)%20-%20f(x)%20%7C%20%3E%202%5E%7B-i%7D%5Crbrace%24%20(Theorem%201.41)%2C%20which%20is%20equivalent%20to%20say%20that%20for%20almost%20all%20%24x%24%20we%20have%20%24f_%7Bn_i%7D%20%5Cto%20f%24.%0A%0A%20%20%20%20%23%23%23%20Part%204%0A%0A%20%20%20%20The%20converse%20of%20part%201%20says%20convergence%20in%20measure%20implies%20convergence%20point-wise%20a.e.%20This%20is%20not%20true%2C%20Homework%202%20Exercise%209%20is%20a%20counterexample.%0A%0A%20%20%20%20The%20converse%20of%20part%202%20says%20convergence%20in%20measure%20implies%20convergence%20in%20%24L%5Ep%24%20norm.%20This%20is%20also%20not%20true%2C%20and%20we%20can%20use%20the%20same%20counterexample.%0A%0A%20%20%20%20%23%23%23%20Part%205%0A%0A%20%20%20%20If%20%24%5Cmu%20%3D%20m%24%20and%20%24X%20%3D%20%5Cmathbb%7BR%7D%24%2C%20then%3A%0A%0A%20%20%20%201.%20Part%201%20is%20false%2C%20for%20example%2C%20%24f_n%20%3D%20%5Cchi_%7B%5Bn%2Cn%2B1%5D%7D%20%5Cto%200%24%20for%20any%20%24x%24%20point-wisely%2C%20but%20not%20converge%20to%20%240%24%20in%20measure%3B%0A%20%20%20%202.%20Part%202%2C3%20are%20true%2C%20the%20proof%20does%20not%20depend%20on%20that%20%24%5Cmu(X)%3C%5Cinfty%24.%0A%20%20%20%20%22%22%22)%0A%20%20%20%20return%0A%0A%0A%40app.cell%0Adef%20_(mo)%3A%0A%20%20%20%20mo.md(r%22%22%22%0A%20%20%20%20%23%20Fall%20Semester%20Homework%204%0A%20%20%20%20%22%22%22)%0A%20%20%20%20return%0A%0A%0A%40app.cell%0Adef%20_(mo)%3A%0A%20%20%20%20mo.md(r%22%22%22%0A%20%20%20%20%23%23%20Exercise%201%0A%0A%20%20%20%20**If%20%24M%24%20is%20a%20closed%20subspace%20of%20%24H%24%2C%20prove%20that%20%24M%20%3D%20(M%5E%7B%5Cperp%7D)%5E%7B%5Cperp%7D%24.%20Is%20there%20a%20similar%20true%20statement%20for%20subspaces%20%24M%24%20which%20are%20not%20necessarily%20closed**%3F%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20If%20%24x%20%5Cin%20M%24%20then%20it%20is%20by%20definition%20perpendicular%20to%20any%20%24y%20%5Cin%20M%5E%7B%5Cperp%7D%24%20%2C%20thus%20by%20definition%20it%20is%20in%20%24(M%5E%7B%5Cperp%7D)%5E%7B%5Cperp%7D%24.%0A%0A%20%20%20%20Conversely%2C%20suppose%20%24x%20%5Cin%20(M%5E%7B%5Cperp%7D)%5E%7B%5Cperp%7D%24%2C%20clearly%20%24x%20%5Cin%20H%24%2C%20so%20we%20may%20write%20%24x%20%3D%20y%20%2B%20z%24%20with%20%24y%20%5Cin%20M%2C%20z%20%5Cin%20M%5E%7B%5Cperp%7D%24.%20In%20particular%20%24(x%2C%20z)%20%3D%20(y%2Bz%2Cz)%24.%20Now%20the%20LHS%20equals%20%240%24%20because%20%24x%2C%20z%24%20are%20perpendicular.%20RHS%20equals%20%24(y%2Cz)%20%2B%20(z%2Cz)%20%3D%20(z%2Cz)%24%20because%20%24y%2Cz%24%20are%20perpendicular%2C%20thus%20%24z%20%3D%200%24%2C%20thus%20%24x%20%3Dy%24%20with%20%24y%20%5Cin%20M%24%2C%20i.e.%20%24x%20%5Cin%20M%24.%0A%0A%20%20%20%20We%20know%20that%20%24M%5E%7B%5Cperp%7D%24%20is%20always%20closed%20(intersection%20of%20closed%20subset)%2C%20so%20%24M%5E%7B%5Cperp%7D%20%3D%20((M%5E%7B%5Cperp%7D)%5E%7B%5Cperp%7D)%5E%7B%5Cperp%7D%24%20always%20work.%0A%0A%20%20%20%20%23%23%20Exercise%202%0A%0A%20%20%20%20**Let%20%24%5Clbrace%20x_n%20%5Crbrace%24%20be%20a%20linearly%20independent%20set%20of%20vectors%20in%20%24H%24.%20Show%20that%20the%20following%20construction%20yields%20an%20orthonormal%20set%20%24%5Clbrace%20u_n%20%5Crbrace%24%20such%20that%20%24%5Clbrace%20x_1%2C%5Cdots%2C%20x_N%20%5Crbrace%24%20and%20%24%5Clbrace%20u_1%2C%5Cdots%2C%20u_N%20%5Crbrace%24%20have%20the%20same%20span%20for%20all%20%24N%24**%3A%0A%0A%20%20%20%20**Put%20%24u_1%20%3D%20x_1%2F%5C%7C%20x_1%20%5C%7C%24.%20Having%20%24u_1%2C%5Cdots%2C%20u_%7Bn-1%7D%24%20define**%20%24%24v_n%20%3D%20x_n%20-%20%5Csum%5Climits_%7Bi%20%3D%201%7D%5E%7Bn-1%7D(x_n%2C%20u_i)u_i%2C~~~~%20u_n%20%3D%20v_n%2F%5C%7C%20v_n%20%5C%7C.%24%24%20**This%20leads%20to%20a%20proof%20of%20the%20existence%20of%20a%20maximal%20orthonormal%20set%20in%20separable%20Hilbert%20spaces%20which%20makes%20no%20appeal%20to%20the%20Hausdorff%20maximality%20principle%20(a%20space%20is%20separable%20if%20it%20contains%20a%20countable%20dense%20subset)**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20Since%20all%20%24u_i%24%20is%20defined%20as%20a%20vector%20divided%20by%20its%20norm%2C%20clearly%20we%20have%20%24%5C%7C%20u_i%20%5C%7C%20%3D%201%24%20for%20all%20viable%20%24i%24.%0A%0A%20%20%20%20By%20induction%20on%20%24j%24%20we%20show%20%24(u_n%2C%20u_m)%20%3D%200%24%20for%20%24n%3Cm%24.%20If%20%24m%20%3D%201%24%20then%20there%20is%20no%20viable%20%24u_n%24.%20If%20%24m%20%3D%202%24%20then%20only%20possible%20%24u_n%24%20is%20%24u_1%24%20and%20%24v_2%24%20is%20defined%20as%20%24(x_2%20-%20(x_2%2C%20u_1)u_1)%24%2C%20then%20%24(u_1%2C%20v_2)%20%3D%20(u_1%2C%20x_2%20-%20(x_2%2C%20u_1)u_1)%24%20%24%3D(u_1%2Cx_2)%20-%20(x_2%2Cu_1)(u_1%2Cu_1)%24%20%24%3D0%24%2C%20so%20%24(u_1%2C%20u_2)%20%3D%200%24%20as%20well%2C%20so%20it%20works%20for%20%24m%20%3D%202%24.%0A%0A%20%20%20%20For%20a%20fixed%20%24m%24%20and%20%24n%3Cm%24%2C%20we%20can%20write%20%24(v_m%2C%20u_n)%24%20%24%3D%20(x_m%20-%20%5Csum%5Climits_%7Bi%3D1%7D%5E%7Bm-1%7D(x_m%2Cu_i)u_i%2C%20u_n)%24%20%24%3D%20(x_m%20-%20(x_m%2Cu_1)u_1%20-%20(x_m%2Cu_2)u_2%20-%20%5Cdots%20-%20(x_m%2Cu_%7Bm-1%7D)u_%7Bm-1%7D%2C%20u_n)%24%20%24%3D%20(x_m%2C%20u_n)%20-%20(x_m%2C%20u_n)%20%3D%200%24%20because%20by%20induction%20assumption%20%24(u_i%2C%20u_n)%20%3D%200%24%20for%20%24i%20%3Cm%2C%20i%20%5Cne%20n%24%2C%20and%20%24(u_n%2C%20u_n)%20%3D%201%24.%20Since%20%24u_m%24%20and%20%24v_m%24%20only%20different%20by%20a%20scalar%2C%20we%20have%20that%20%24(u_n%2C%20u_m)%20%3D%200%24%20for%20%24n%20%3C%20m%24.%0A%0A%20%20%20%20So%20%24%5Clbrace%20u_i%20%5Crbrace%24%20is%20an%20orthonormal%20set.%0A%0A%20%20%20%20Span%20of%20%24%5Clbrace%20u_i%20%5Crbrace%24%20lies%20inside%20span%20of%20%24%5Clbrace%20x_i%20%5Crbrace%24%2C%20because%20%24%5Clbrace%20u_i%20%5Crbrace%24%20are%20constructed%20from%20%24%5Clbrace%20x_i%20%5Crbrace%24.%20It%20is%20also%20not%20hard%20to%20see%20%24%5Clbrace%20x_i%20%5Crbrace%24%20are%20constructible%20from%20%24%5Clbrace%20u_i%20%5Crbrace%24%20so%20the%20other%20inclusion%20also%20works.%0A%0A%20%20%20%20If%20%24H%24%20is%20separable%20then%20it%20admits%20a%20countable%20dense%20subset%2C%20so%20it%20admits%20a%20(at%20most)%20countable%20basis%2C%20which%20can%20be%20seen%20as%20%24%5Clbrace%20x_N%20%5Crbrace%24%2C%20now%20use%20the%20above%20construction%20to%20make%20%24%5Clbrace%20u_N%20%5Crbrace%24%2C%20then%20the%20set%20%24P%24%20of%20all%20finite%20linear%20combinations%20of%20members%20of%20%24%5Clbrace%20u_%7BN%7D%20%5Crbrace%24%20is%20by%20construction%20dense%2C%20thus%20%24%5Clbrace%20u_N%20%5Crbrace%24%20is%20maximal%20(and%20it%20is%20countable)%20by%20Theorem%204.18.%0A%0A%20%20%20%20%23%23%20Exercise%203%0A%0A%20%20%20%20**Show%20that%20%24L%5Ep(T)%24%20is%20separable%20if%20%241%5Cle%20p%20%3C%20%5Cinfty%24%2C%20but%20%24L%5E%7B%5Cinfty%7D(T)%24%20is%20not%20separable**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20We%20proved%20that%20trigonometry%20polynomials%20are%20dense%20in%20%24C(T)%24%2C%20since%20%24T%24%20is%20compact%2C%20every%20map%20in%20%24C(T)%24%20is%20also%20in%20%24C_C(T)%24%20(because%20any%20closed%20subset%20of%20compact%20space%20is%20compact%2C%20i.e.%20support%20is%20always%20compact)%2C%20and%20that%20we%20know%20for%20%241%20%5Cle%20p%20%3C%20%5Cinfty%24%2C%20%24C_C(T)%24%20is%20dense%20in%20%24L%5Ep(T)%24.%20Now%20the%20definition%20of%20a%20trigonometry%20polynomials%20uses%20real%20numbers%2C%20but%20since%20%24%5Cmathbb%7BQ%7D%24%20is%20dense%20in%20%24%5Cmathbb%7BR%7D%24%2C%20'rational'%20trigonometry%20polynomials%20are%20also%20dense%20in%20%24C(T)%24%2C%20there%20are%20countably%20many%20such%20polynomials%2C%20and%20result%20follows.%0A%0A%20%20%20%20For%20%24L%5E%7B%5Cinfty%7D(T)%24%3A%20consider%20the%20functions%20%24%5Clbrace%20f_a(t)%20%5Crbrace%24%20where%20%24%24f_a%20%3D%20%5Cbegin%7Bcases%7D%201%2C%20t%20%5Cin%20%5B0%2Ca)%20%5C%5C%200%2C%20t%20%5Cin%20%5Ba%2C2%5Cpi%5D%20%5Cend%7Bcases%7D%20%3D%20%5Cchi_%7B%5B0%2Ca)%7D%2C%20a%20%5Cin%20%5B0%2C2%5Cpi%5D.%24%24%20Apparently%20there%20are%20uncountably%20different%20%24f_a%24's%2C%20yet%20for%20any%20%24a%20%3C%20b%24%2C%20we%20have%20%24f_b%20-%20f_a%20%3D%20%5Cchi_%7B%5Ba%2Cb%5D%7D%24%2C%20so%20that%20%24%5C%7C%20f_b%20-%20f_a%20%5C%7C_p%20%3D%201%24.%20Thus%20there%20is%20no%20way%20to%20remove%20any%20%24f_a%24%20from%20the%20set%20and%20still%20have%20the%20rest%20of%20%24%5Clbrace%20f_a%20%5Crbrace%24%20be%20(potentially)%20dense.%20In%20particular%20%24L%5E%7B%5Cinfty%7D(T)%24%20cannot%20have%20a%20countable%20dense%20subset%2C%20i.e.%20it%20is%20not%20separable.%0A%0A%20%20%20%20%23%23%20Exercise%204%0A%0A%20%20%20%20**Show%20that%20%24H%24%20is%20separable%20if%20and%20only%20if%20%24H%24%20contains%20a%20maximal%20orthonormal%20system%20which%20is%20at%20most%20countable**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20We%20will%20use%20Theorem%204.18.%0A%0A%20%20%20%20If%20%24H%24%20contains%20a%20maximal%20orthonormal%20system%20which%20is%20at%20most%20countable%2C%20then%20the%20set%20%24P%24%20of%20all%20finite%20linear%20combinations%20of%20members%20of%20%24%5Clbrace%20u_%7B%5Calpha%7D%20%5Crbrace%24%20is%20dense%20in%20%24H%24%2C%20%24P%24%20is%20countable%2C%20thus%20%24H%24%20is%20separable.%0A%0A%20%20%20%20The%20converse%20is%20proved%20in%20Exercise%202.%0A%0A%20%20%20%20%23%23%20Exercise%205%0A%0A%20%20%20%20**If%20%24M%20%3D%20%5Clbrace%20x%20%7C%20Lx%20%3D%200%20%5Crbrace%24%2C%20where%20%24L%24%20is%20a%20continuous%20linear%20functional%20on%20%24H%24%2C%20prove%20that%20%24M%5E%7B%5Cperp%7D%24%20is%20a%20vector%20space%20of%20dimension%20%241%24%20unless%20%24M%20%3D%20H%24**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20If%20%24L%24%20is%20a%20continuous%20linear%20functional%20on%20%24H%24%2C%20then%20there%20is%20a%20unique%20%24y%20%5Cin%20H%24%20such%20that%20%24Lx%20%3D%20(x%2Cy)%2C%20%5Cforall%20x%20%5Cin%20H%24.%20So%20a%20given%20%24L%24%20implies%20a%20unique%20%24y%24%2C%20and%20%24M%24%20is%20equivalently%20defined%20as%20%24%5Clbrace%20x%20%7C%20(x%2Cy)%20%3D%200%20%5Crbrace%24.%20Denote%20the%20space%20spanned%20by%20%24y%24%20by%20%24N%24%2C%20then%20%24N%20%3D%20M%5E%7B%5Cperp%7D%24%3A%0A%0A%20%20%20%20%24N%24%20is%20closed%20because%20it%20contains%20all%20scalar%20multiplication%20of%20%24y%24.%20So%20by%20Exercise%201%2C%20%24N%20%3D%20(N%5E%7B%5Cperp%7D)%5E%7B%5Cperp%7D%24.%20Now%20it%20is%20from%20definition%20of%20%24N%24%20and%20%24M%24%20that%20%24N%5E%7B%5Cperp%7D%20%3D%20M%24%20because%20%24M%24%20consists%20of%20all%20element%20that%20perpendicular%20to%20%24y%24%20thus%20scalar%20multiplications%20of%20%24y%24.%20Plug%20into%20%24N%20%3D%20(N%5E%7B%5Cperp%7D)%5E%7B%5Cperp%7D%24%20we%20have%20that%20%24N%20%3D%20M%5E%7B%5Cperp%7D%24.%0A%0A%20%20%20%20If%20%24H%20%3D%20M%24%20then%20%24y%20%3D%200%24%20then%20%24M%5E%7B%5Cperp%7D%24%20is%20of%20dimension-zero%2C%20otherwise%20%24y%20%5Cne%200%24%20and%20%24M%5E%7B%5Cperp%7D%24%20be%20a%20space%20spanned%20by%20%241%24%20element%20is%20of%20dimension-one.%0A%0A%20%20%20%20%23%23%20Exercise%207%0A%0A%20%20%20%20**Suppose%20%24%5Clbrace%20a_n%20%5Crbrace%24%20is%20a%20sequence%20of%20positive%20numbers%20such%20that%20%24%5Csum%20a_n%20b_n%20%3C%20%5Cinfty%24%20whenever%20%24b_n%20%5Cge%200%24%20and%20%24%5Csum%20b_n%5E2%20%3C%20%5Cinfty%24%2C%20prove%20that%20%24%5Csum%20a_n%5E2%20%3C%20%5Cinfty%24**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20Since%20%24a_n%24%20are%20all%20positive%2C%20if%20in%20contrary%20that%20%24%5Csum%20a_n%5E2%24%20not%20converge%20then%20it%20must%20diverge%20to%20infinity.%20So%20suppose%20%24%5Csum%20a_n%5E2%20%3D%20%5Cinfty%24.%20By%20definition%20of%20limit%2C%20we%20can%20split%20%24%5Cmathbb%7BN%7D%24%20to%20infinite%20many%20disjoint%20sub-intervals%20%24N_i%24%20such%20that%20%24s_i%20%3D%20%5Csum%5Climits_%7Bn%20%5Cin%20N_i%7Da_n%5E2%20%3E%201%24.%20We%20can%20write%20%24%5Csum%5Climits_n%20a_n%5E2%20%3D%20%5Csum%5Climits_%7Bi%7D%20%5Csum%5Climits_%7Bn%20%5Cin%20N_i%7D%20a_n%5E2%20%3D%20%5Csum%5Climits_i%20s_i%24.%0A%0A%20%20%20%20Let%20%24b_n%20%3D%20%5Cfrac%7B1%7D%7Bi%5Csqrt%7Bs_i%7D%7Da_n%24%20for%20%24n%20%5Cin%20N_i%24%2C%20this%20is%20a%20positive%20sequence%2C%20and%20we%20have%3A%0A%0A%20%20%20%201.%20%24%5Csum%5Climits_na_nb_n%20%3D%20%5Csum%5Climits_n%20a_n%5E2%20%5Cfrac%7B1%7D%7Bi%5Csqrt%7Bs_i%7D%7D%24%20%24%3D%20%5Csum%5Climits_i%20%5Cfrac%7B1%7D%7Bi%5Csqrt%7Bs_i%7D%7D%20%5Csum%5Climits_%7Bn%20%5Cin%20N_i%7Da_n%5E2%24%20%24%3D%5Csum%5Climits_i%20%5Cfrac%7Bs_i%7D%7Bi%5Csqrt%7Bs_i%7D%7D%24%20%24%3E%20%5Csum%5Climits_i%20%5Cfrac%7B1%7D%7Bi%7D%24%20because%20%24s_i%20%3E%201%24%20and%20this%20sequence%20diverges%3B%0A%20%20%20%202.%20%24%5Csum%5Climits_n%20b_n%5E2%20%3D%20%5Csum%5Climits_n%20%5Cfrac%7B1%7D%7Bi%5E2s_i%7Da_n%5E2%24%20%24%3D%20%5Csum%5Climits_i%20%5Cfrac%7B1%7D%7Bi%5E2s_i%7D%20%5Csum%5Climits_%7Bn%20%5Cin%20N_i%7Da_n%5E2%24%20%24%3D%20%5Csum%5Climits_i%20%5Cfrac%7B1%7D%7Bi%5E2%7D%24%2C%20and%20this%20sequence%20is%20known%20to%20be%20convergent.%0A%0A%20%20%20%20So%20there%20is%20positive%20convergent%20sequence%20%24b_n%24%20makes%20%24a_nb_n%24%20divergent%2C%20this%20is%20the%20negation%20of%20the%20original%20statement%2C%20thus%20we%20are%20finished.%0A%0A%20%20%20%20%23%23%20Exercise%209%0A%0A%20%20%20%20**If%20%24A%20%5Csubset%20%5B0%2C2%20%5Cpi%5D%24%20and%20%24A%24%20is%20measurable%2C%20prove%20that%20%24%5Clim%5Climits_%7Bn%20%5Cto%20%5Cinfty%7D%20%5Cint_A%20%5Ccos(nx)dx%20%3D%20%5Clim%5Climits_%7Bn%20%5Cto%20%5Cinfty%7D%20%5Cint_A%20%5Csin(nx)dx%20%3D%200%24**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20Since%20the%20whole%20space%20%24%5B0%2C2%5Cpi%5D%24%20is%20of%20finite%20measure%2C%20result%20will%20follow%20if%20we%20show%20%24%5Clim%5Climits_%7Bn%20%5Cto%20%5Cinfty%7D%20%5Cint_0%5E%7B2%5Cpi%7D%20%5Ccos(nx)dx%20%3D%20%5Clim%5Climits_%7Bn%20%5Cto%20%5Cinfty%7D%20%5Cint_0%5E%7B2%5Cpi%7D%20%5Csin(nx)dx%20%3D%200%24.%0A%0A%20%20%20%20Now%20%24%5Cint_0%5E%7B2%5Cpi%7De%5E%7Binx%7Ddx%24%20%24%3D%5Cfrac%7Be%5E%7Binx%7D%7D%7Bin%7D%20%7C_0%5E%7B2%5Cpi%7D%24%20%24%3D%20%5Cfrac%7Be%5E%7B2in%20%5Cpi%7D-1%7D%7Bin%7D%24%2C%20it%20goes%20to%20%240%24%20as%20%24n%20%5Cto%20%5Cinfty%24%20because%20%24e%5E%7B2%20i%20n%20%5Cpi%7D%24%20always%20have%20length%20%241%24.%20So%20both%20real%20and%20imaginary%20parts%2C%20i.e.%20%24%5Cint_0%5E%7B2%5Cpi%7D%20%5Ccos(nx)dx%24%20and%20%24%5Cint_0%5E%7B2%5Cpi%7D%20%5Csin(nx)dx%24%20go%20to%20%240%24%20as%20%24n%20%5Cto%20%5Cinfty%24.%0A%0A%20%20%20%20%23%23%20Exercise%2014%0A%0A%20%20%20%20**Compute%20%24%5Cmin%5Climits_%7Ba%2Cb%2Cc%7D%5Cint%5Climits_%7B-1%7D%5E1%20%7C%20x%5E3%20-%20a%20-%20bx%20-%20cx%5E2%7C%5E2dx%24%20and%20find%20%24%5Cmax%5Cint%5Climits_%7B-1%7D%5E1x%5E3g(x)dx%24%2C%20where%20%24g%24%20is%20subject%20to%20the%20restrictions%20%24%5Cint%5Climits_%7B-1%7D%5E1g(x)dx%20%3D%20%5Cint%5Climits_%7B-1%7D%5E1xg(x)dx%20%3D%20%5Cint%5Climits_%7B-1%7D%5E1x%5E2g(x)dx%20%3D%200%24%20and%20%24%5Cint%5Climits_%7B-1%7D%5E1%7C%20g(x)%20%7C%5E2%20dx%20%3D%201%24**.%0A%0A%20%20%20%20Solution%0A%0A%20%20%20%20Consider%20%24f(x)%20%3D%20x%5E3%24%2C%20then%20in%20%24H%20%3D%20L%5E2(-1%2C1)%24%20with%20%24L%5E2%24%20norm%20we%20have%20%24%5C%7C%20f%20%5C%7C%5E2%20%3D%20%5Cint%5Climits_%7B-1%7D%5E1%20%7C%20x%5E3%20%7C%5E2%20dx%24.%20Consider%20%24M%24%20be%20span%20of%20%24%5Clangle%201%2Cx%2Cx%5E2%20%5Crangle%24%20lies%20insider%20%24H%24%2C%20we%20decomposite%20%24f%24%20into%20%24Pf%24%20and%20%24Qf%24.%20By%20our%20definition%20of%20%24M%24%2C%20%24Pf%24%20is%20of%20the%20form%20%24a%20%2B%20bx%20%2B%20cx%5E2%24.%20Since%20%24f%20%3D%20Pf%2BQf%24%20we%20must%20have%20%24Qf%20%3D%20x%5E3%20-%20a%20-%20bx%20-%20cx%5E2%24.%0A%0A%20%20%20%20We%20know%20%24Pf%20%5Cperp%20Qf%24%2C%20so%20%24(Pf%2C%20Qf)%20%3D%200%24%2C%20the%20left%20hand%20side%20is%20defined%20as%20%24%5Cint%5Climits_%7B-1%7D%5E1%20(a%2Bbx%2Bcx%5E2)(x%5E3-a-bx-cx%5E2)dx%24.%20Expand%20and%20some%20elementary%20calculus%20gives%20us%20(%24x%5E%7Bodd%7D%24%20terms%20just%20vanish)%20%24%3D%20-%5Cfrac%7B2%7D%7B15%7D(15a%5E2%20%2B%2010%20ac%20%2B%205b%5E2%20-%203b%20%2B%203c%5E2)%24%20and%20we%20want%20non-trivial%20%24a%2Cb%2Cc%24%20to%20make%20this%20%240%24.%20An%20relatively%20obvious%20solution%20is%20that%20%24a%20%3D%20c%20%3D%200%24%20and%20%24b%20%3D%20%5Cfrac%7B3%7D%7B5%7D%24.%0A%0A%20%20%20%20Now%20Theorem%204.11%20tells%20us%20%24%5C%7C%20Pf%20-%20f%20%5C%7C%5E2%24%20which%20is%20just%20our%20original%20%24%5Cint%5Climits_%7B-1%7D%5E1%20%7C%20x%5E3%20-%20a%20-%20bx%20-%20cx%5E2%7C%5E2dx%24%20is%20minimized%20at%20%24a%20%3D%20c%20%3D%200%2C%20b%20%3D%203%2F5%24%20(also%20that%20it%20is%20actually%20unique).%20At%20this%20point%20%24%5Cint%5Climits_%7B-1%7D%5E1%20%7C%20x%5E3%20-%20a%20-%20bx%20-%20cx%5E2%7C%5E2dx%24%20becomes%20%24%5Cint%5Climits_%7B-1%7D%5E1%20%7C%20x%5E3%20-%20%5Cfrac%7B3%7D%7B5%7Dx%20%7C%5E2dx%24%20and%20it%20is%20easy%20to%20be%20calculated%20with%20some%20elementary%20calculus%2C%20the%20answer%20is%20%24%3D%20%5Cfrac%7B8%7D%7B175%7D%24.%0A%0A%20%20%20%20For%20part%20two%2C%20we%20have%20%24%24%5Cbegin%7Baligned%7D%5Cint%5Climits_%7B-1%7D%5E1x%5E3g(x)dx%26%3D%20%5Cint%5Climits_%7B-1%7D%5E1(x%5E3-a-bx-cx%5E2%2Ba%2Bbx%2Bcx%5E2)g(x)dx%5C%5C%26%3D%20%5Cint%5Climits_%7B-1%7D%5E1(x%5E3-a-bx-cx%5E2)g(x)dx%20%2B%20%5Cint%5Climits_%7B-1%7D%5E1(-a-bx-cx%5E2)g(x)dx%5C%5C%26%3D%20%5Cint%5Climits_%7B-1%7D%5E1(x%5E3-a-bx-cx%5E2)g(x)dx%5Cend%7Baligned%7D%24%24%20because%20by%20assumption%20we%20have%20%24%24%5Cint%5Climits_%7B-1%7D%5E1g(x)dx%20%3D%20%5Cint%5Climits_%7B-1%7D%5E1xg(x)dx%20%3D%20%5Cint%5Climits_%7B-1%7D%5E1x%5E2g(x)dx%20%3D%200%24%24%20are%20all%20zero.%20Now%20by%20Cauchy-Schwarz%20we%20have%20%24%24%5Cbegin%7Baligned%7D%5Cint%5Climits_%7B-1%7D%5E1%20(x%5E3-a-bx-cx%5E2)g(x)dx%20%26%5Cle%20(%5Cint%5Climits_%7B-1%7D%5E1(x%5E3-a-bx-cx%5E2)%5E2dx)%5E%7B1%2F2%7D(%5Cint%5Climits_%7B-1%7D%5E1(g(x))%5E2dx)%5E%7B1%2F2%7D%20%5C%5C%20%26%3D(%5Cint%5Climits_%7B-1%7D%5E1(x%5E3-a-bx-cx%5E2)%5E2dx)%5E%7B1%2F2%7D%5Cend%7Baligned%7D%24%24%20because%20by%20assumption%20we%20have%20%24%24%5Cint%5Climits_%7B-1%7D%5E1%7C%20g(x)%20%7C%5E2%20dx%20%3D%201.%24%24%20So%20we%20see%20the%20%24%5Cmax%5Cint%5Climits_%7B-1%7D%5E1x%5E3g(x)dx%24%20is%20bounded%20above%20by%20the%20square-root%20of%20the%20answer%20of%20part%201%2C%20i.e.%20%24%5Cmax%5Cint%5Climits_%7B-1%7D%5E1x%5E3g(x)dx%20%5Cle%20%5Csqrt%7B%5Cfrac%7B8%7D%7B175%7D%7D%24.%20The%20equality%20in%20Cauchy-Schwarz%20in%20our%20case%20is%20achievable%20(equality%20is%20achieved%20if%20%24g(x)%24%20is%20a%20multiple%20of%20%24f(x)%24.%20In%20our%20case%20let%20%24g(x)%24%20be%20%24%5Csqrt%7B%5Cfrac%7B175%7D%7B8%7D%7D(x%5E3-%20%5Cfrac%7B3%7D%7B5%7Dx)%24%20will%20do%20the%20job)%2C%20so%20the%20maximum%20is%20indeed%20%24%5Csqrt%7B%5Cfrac%7B8%7D%7B175%7D%7D%24.%0A%0A%20%20%20%20%23%23%20Exercise%2016%0A%0A%20%20%20%20**If%20%24x_0%20%5Cin%20H%24%20and%20%24M%24%20is%20a%20closed%20linear%20subspace%20of%20%24H%24%2C%20prove%20that%20%24%5Cmin%20%5Clbrace%20%5C%7C%20x%20-%20x_0%20%5C%7C%20%3A%20x%20%5Cin%20M%20%5Crbrace%20%3D%20%5Cmax%20%5Clbrace%20%7C(x_0%2C%20y)%7C%20%3A%20y%20%5Cin%20M%5E%7B%5Cperp%7D%2C%20%5C%7C%20y%20%5C%7C%20%3D%201%5Crbrace%24**.%0A%0A%20%20%20%20Proof%0A%0A%20%20%20%20We%20can%20write%20%24x_0%20%3D%20Px_0%20%2B%20Qx_0%24%2C%20%24Px_0%20%5Cin%20M%24%20and%20%24Qx_0%20%5Cin%20M%5E%7B%5Cperp%7D%24.%20We%20know%20that%20%24Px_0%24%20is%20the%20nearest%20point%20to%20%24x_0%24%20in%20%24M%24%2C%20in%20other%20words%2C%20%24%5C%7C%20Px_0%20-%20x_0%20%5C%7C%20%3D%20%5Cmin%20%5Clbrace%20%5C%7C%20x-x_0%20%5C%7C%3Ax%5Cin%20M%20%5Crbrace%24%20which%20is%20the%20LHS.%20Move%20things%20around%20we%20have%20%24%5C%7C%20Px_0%20-%20x_0%20%5C%7C%20%3D%20%5C%7C%20Qx_0%20%5C%7C%24%2C%20so%20if%20RHS%20equals%20%24%5C%7C%20Qx_0%20%5C%7C%24%20then%20we%20are%20done.%0A%0A%20%20%20%20First%20notice%20that%20if%20we%20take%20%24y%20%3D%20%5Cfrac%7BQx_0%7D%7B%5C%7C%20Qx_0%20%5C%7C%7D%24%20then%20%24y%20%5Cin%20M%5E%7B%5Cperp%7D%24%2C%20%24%5C%7C%20y%20%5C%7C%20%3D%201%24%2C%20and%20%24%7C%20(x_0%2C%20y)%20%7C%24%20%24%3D%20%7C%20Qx_0%5Ccdot%20%5Coverline%7B%5Cfrac%7BQx_0%7D%7B%5C%7C%20Qx_0%20%5C%7C%7D%7D%20%7C%24%20(because%20%24y%20%5Cin%20M%5E%7B%5Cperp%7D%24%2C%20the%20part%20with%20%24Px_0%24%20equals%20%240%24.)%20%24%3D%20%5C%7C%20Qx_0%20%5C%7C%24.%20We%20want%20to%20show%20it%20is%20the%20maximum.%0A%0A%20%20%20%20Let%20%24y_0%20%5Cin%20M%5E%7B%5Cperp%7D%24%20and%20%24%5C%7C%20y_0%20%5C%7C%20%3D%201%24.%20Then%20%24%7C%20(x_0%2C%20y_0)%20%7C%24%20%24%3D%20%7C%20(Px_0%2BQx_0%2C%20y_0)%20%7C%24%20%24%3D%20%7C%20(Qx_0%2C%20y_0)%20%7C%24%20(again%20the%20part%20with%20%24Px_0%24%20equals%20%240%24)%20%24%3D%20%7C%20(%5Cfrac%7B%5C%7C%20Qx_0%20%5C%7C%7D%7B%5C%7C%20Qx_0%20%5C%7C%7DQx_0%2C%20y_0)%20%7C%24%20%24%3D%20%5C%7C%20Qx_0%20%5C%5C%7C%20(%5Cfrac%7BQx_0%7D%7B%5C%7C%20Qx_0%20%5C%7C%7D%2C%20y_0)%20%7C%24%20%24%5Cle%20%5C%7C%20Qx_0%20%5C%7C%20%5C%7C%20%5Cfrac%7BQx_0%7D%7B%5C%7C%20Qx_0%20%5C%7C%7D%20%5C%7C%20%5C%7C%20y_0%20%5C%7C%24%20%24%3D%20%5C%7C%20Qx_0%20%5C%7C%24.%20So%20%24%5C%7C%20Qx_0%20%5C%7C%24%20is%20indeed%20an%20upper-bound%2C%20and%20since%20we%20saw%20it%20is%20achievable%2C%20it%20is%20the%20maximum.%20The%20result%20follows.%0A%20%20%20%20%22%22%22)%0A%20%20%20%20return%0A%0A%0Aif%20__name__%20%3D%3D%20%22__main__%22%3A%0A%20%20%20%20app.run()%0A