S.-T. Yau College Student Mathematics Contests 2022

Analysis and Differential Equations

Solve every problem.

Problem 1. For n 1 n 1 n >= 1n \geq 1n1, we consider the integral
I n = [ 0 , 1 ] n n 1 x 1 + 1 x 2 + + 1 x n d x 1 d x n . I n = [ 0 , 1 ] n n 1 x 1 + 1 x 2 + + 1 x n d x 1 d x n . I_(n)=int_([0,1]^(n))(n)/((1)/(x_(1))+(1)/(x_(2))+cdots+(1)/(x_(n)))dx_(1)cdots dx_(n).I_{n}=\int_{[0,1]^{n}} \frac{n}{\frac{1}{x_{1}}+\frac{1}{x_{2}}+\cdots+\frac{1}{x_{n}}} d x_{1} \cdots d x_{n} .In=[0,1]nn1x1+1x2++1xndx1dxn.
Prove that lim n I n lim n I n lim_(n rarr oo)I_(n)\lim _{n \rightarrow \infty} I_{n}limnIn exists.
Solution: For all positive integers m m mmm and n n nnn, for all x , y > 0 x , y > 0 x,y > 0x, y>0x,y>0, we check that
( m + n ) 2 x + y m 2 x + n 2 y ( m + n ) 2 x + y m 2 x + n 2 y ((m+n)^(2))/(x+y) <= (m^(2))/(x)+(n^(2))/(y)\frac{(m+n)^{2}}{x+y} \leq \frac{m^{2}}{x}+\frac{n^{2}}{y}(m+n)2x+ym2x+n2y
Thus, J n = n I n J n = n I n J_(n)=nI_(n)J_{n}=n I_{n}Jn=nIn satisfies
J m + n J m + J n J m + n J m + J n J_(m+n) <= J_(m)+J_(n)J_{m+n} \leq J_{m}+J_{n}Jm+nJm+Jn
It is well-known that lim n J n n lim n J n n lim_(n rarr oo)(J_(n))/(n)\lim _{n \rightarrow \infty} \frac{J_{n}}{n}limnJnn exists.
Problem 2. Let U C U C U subCU \subset \mathbf{C}UC be a non-empty open set and f : U U f : U U f:U rarr Uf: U \rightarrow Uf:UU be a non-constant holomorphic function. Prove that, if f f = f f f = f f@f=ff \circ f=fff=f, then f ( z ) z f ( z ) z f(z)-=zf(z) \equiv zf(z)z for all z U z U z in Uz \in UzU.
Solution: Since f f fff is not a constant map, V := f ( U ) U V := f ( U ) U V:=f(U)sub UV:=f(U) \subset UV:=f(U)U is an nonempty open set (the open mapping property). Thus, for z V z V z in Vz \in VzV, we have z = f ( z ) z = f ( z ) z=f(z)z=f(z)z=f(z). This implies that f ( z ) z f ( z ) z f(z)-=zf(z) \equiv zf(z)z.
Problem 3. Let X R X R X subRX \subset \mathbf{R}XR be a set with positive (Lebesgue) measure. Show that we can find an arithmetic progression of 2022 terms in X X XXX, i.e., there exists x 1 , , x 2022 X x 1 , , x 2022 X x_(1),dots,x_(2022)in Xx_{1}, \ldots, x_{2022} \in Xx1,,x2022X so that the x i + 1 x i x i + 1 x i x_(i+1)-x_(i)x_{i+1}-x_{i}xi+1xi 's are all equal and positive, i = 1 , , 2021 i = 1 , , 2021 i=1,dots,2021i=1, \ldots, 2021i=1,,2021.
Solution: We use m m mmm to denote the Lebesgue measure. Let x X x X x in Xx \in XxX be a Lebesgue point; therefore, there exists an interval I I III so that x I x I x in Ix \in IxI and m ( I n X ) m ( I ) 1 ε m ( I n X ) m ( I ) 1 ε (m(InX))/(m(I)) >= 1-epsi\frac{m(I n X)}{m(I)} \geq 1-\varepsilonm(InX)m(I)1ε and ϵ ϵ epsilon\epsilonϵ will be determined at the end of the proof. By translating and rescaling, we may assume that I = [ 0 , 1 ] I = [ 0 , 1 ] I=[0,1]I=[0,1]I=[0,1]. We divide I I III into 2022 intervals:
I = I 1 I 2 I 2022 , I k = [ k 1 2022 , k 2022 ] , k = 1 , 2 , , 2022 . I = I 1 I 2 I 2022 , I k = k 1 2022 , k 2022 , k = 1 , 2 , , 2022 . I=I_(1)uuI_(2)uu cdots uuI_(2022),I_(k)=[(k-1)/(2022),(k)/( 2022)],k=1,2,dots,2022.I=I_{1} \cup I_{2} \cup \cdots \cup I_{2022}, I_{k}=\left[\frac{k-1}{2022}, \frac{k}{2022}\right], k=1,2, \ldots, 2022 .I=I1I2I2022,Ik=[k12022,k2022],k=1,2,,2022.
Let X k = ( I k X ) k 1 2022 X k = I k X k 1 2022 X_(k)=(I_(k)nn X)-(k-1)/(2022)X_{k}=\left(I_{k} \cap X\right)-\frac{k-1}{2022}Xk=(IkX)k12022 be the translation of I k X I k X I_(k)nn XI_{k} \cap XIkX and X k I 1 , k = 1 , , 2022 X k I 1 , k = 1 , , 2022 X_(k)subI_(1),k=1,dots,2022X_{k} \subset I_{1}, k=1, \ldots, 2022XkI1,k=1,,2022. We know that
k = 1 2022 m ( X k ) 1 ϵ k = 1 2022 m X k 1 ϵ sum_(k=1)^(2022)m(X_(k)) >= 1-epsilon\sum_{k=1}^{2022} m\left(X_{k}\right) \geq 1-\epsilonk=12022m(Xk)1ϵ
Thus,
m ( 1 k 2022 X k ) 1 2022 2022 ϵ m 1 k 2022 X k 1 2022 2022 ϵ m(nnn_(1 <= k <= 2022)X_(k)) >= (1)/(2022)-2022 epsilonm\left(\bigcap_{1 \leq k \leq 2022} X_{k}\right) \geq \frac{1}{2022}-2022 \epsilonm(1k2022Xk)120222022ϵ
We may take ϵ = 2023 ϵ = 2023 epsilon=2023\epsilon=2023ϵ=2023, thus, 1 k 2022 X k 1 k 2022 X k nnn_(1 <= k <= 2022)X_(k)!=O/\bigcap_{1 \leq k \leq 2022} X_{k} \neq \varnothing1k2022Xk. Let x 1 1 k 2022 X k x 1 1 k 2022 X k x_(1)innnn_(1 <= k <= 2022)X_(k)x_{1} \in \bigcap_{1 \leq k \leq 2022} X_{k}x11k2022Xk. Then x k = x 1 + k 1 2022 x k = x 1 + k 1 2022 x_(k)=x_(1)+(k-1)/(2022)x_{k}=x_{1}+\frac{k-1}{2022}xk=x1+k12022 is the arithmetic progression.
Problem 4. Let C ( [ 0 , 1 ] ) C ( [ 0 , 1 ] ) C([0,1])C([0,1])C([0,1]) be the space of all continuous C C C\mathbf{C}C-valued functions equipped with L L L^(oo)L^{\infty}L-norm. Let P P Psub\mathbf{P} \subsetP C ( [ 0 , 1 ] ) C ( [ 0 , 1 ] ) C([0,1])C([0,1])C([0,1]) be a closed linear subspace. Assume that the elements of P P P\mathbf{P}P are polynomials. Prove that dim P < dim P < dim P < oo\operatorname{dim} \mathbf{P}<\inftydimP<.
Solution: Let I = { ( x , y ) [ 0 , 1 ] 2 x y } I = ( x , y ) [ 0 , 1 ] 2 x y I={(x,y)in[0,1]^(2)∣x!=y}I=\left\{(x, y) \in[0,1]^{2} \mid x \neq y\right\}I={(x,y)[0,1]2xy}. For each ( x , y ) I ( x , y ) I (x,y)in I(x, y) \in I(x,y)I, we define a mapping
T ( x , y ) : P C , u u ( x ) u ( y ) | x y | T ( x , y ) : P C , u u ( x ) u ( y ) | x y | T_((x,y)):PrarrC,u|->(u(x)-u(y))/(|x-y|)T_{(x, y)}: \mathbf{P} \rightarrow \mathbf{C}, u \mapsto \frac{u(x)-u(y)}{|x-y|}T(x,y):PC,uu(x)u(y)|xy|
Therefore, we have
sup ( x , y ) I | T ( x , y ) u | u L sup ( x , y ) I T ( x , y ) u u L s u p_((x,y)in I)|T_((x,y))u| <= ||u^(')||_(L^(oo))\sup _{(x, y) \in I}\left|T_{(x, y)} u\right| \leq\left\|u^{\prime}\right\|_{L^{\infty}}sup(x,y)I|T(x,y)u|uL
Since P P P\mathbf{P}P is closed, we can apply the Banach-Steinhaus Theorem: there exists C > 0 C > 0 C > 0C>0C>0, so that
sup ( x , y ) I T ( x , y ) P C C sup ( x , y ) I T ( x , y ) P C C s u p_((x,y)in I)||T_((x,y))||_(PrarrC) <= C\sup _{(x, y) \in I}\left\|T_{(x, y)}\right\|_{\mathrm{P} \rightarrow \mathrm{C}} \leq Csup(x,y)IT(x,y)PCC
We consider the unit ball of P P P\mathbf{P}P :
B = { u P u L 1 } B = u P u L 1 B={u inP∣||u||_(L^(oo)) <= 1}B=\left\{u \in \mathbf{P} \mid\|u\|_{L^{\infty}} \leq 1\right\}B={uPuL1}
Hence, for all u B u B u in Bu \in BuB, we have
| u ( x ) u ( y ) | C | x y | | u ( x ) u ( y ) | C | x y | |u(x)-u(y)| <= C|x-y||u(x)-u(y)| \leq C|x-y||u(x)u(y)|C|xy|
Thus, the family B B BBB is equicontinuous. By the Arzelà-Ascoli Theorem, it is compact. Thus, P P P\mathbf{P}P is finite-dimensional.
Problem 5. Let Ω R 3 Ω R 3 Omega subR^(3)\Omega \subset \mathbf{R}^{3}ΩR3 be a bounded domain with smooth boundary. Assume that u C ( R 3 Ω ¯ ) u C R 3 Ω ¯ u in C( bar(R^(3)-Omega))u \in C\left(\overline{\mathbf{R}^{3}-\Omega}\right)uC(R3Ω¯) is a harmonic function on R 3 Ω R 3 Ω R^(3)-Omega\mathbf{R}^{3}-\OmegaR3Ω so that u | Ω = 1 u Ω = 1 u|_(Omega)=1\left.u\right|_{\Omega}=1u|Ω=1 and lim | x | | u ( x ) | = 0 lim | x | | u ( x ) | = 0 lim_(|x|rarr oo)|u(x)|=0\lim _{|x| \rightarrow \infty}|u(x)|=0lim|x||u(x)|=0.
Prove that for such u , lim | x | | x | u ( x ) u , lim | x | | x | u ( x ) u,lim_(|x|rarr oo)|x|u(x)u, \lim _{|x| \rightarrow \infty}|x| u(x)u,lim|x||x|u(x) exists.
Solution: Let φ ( x ) C ( R 3 ) φ ( x ) C R 3 varphi(x)inC^(oo)(R^(3))\varphi(x) \in C^{\infty}\left(\mathbf{R}^{3}\right)φ(x)C(R3) so that φ 0 φ 0 varphi-=0\varphi \equiv 0φ0 on an open neighborhood of Ω Ω Omega\OmegaΩ and φ 1 φ 1 varphi-=1\varphi \equiv 1φ1 for | x | R | x | R |x| >= R|x| \geq R|x|R where R > 0 R > 0 R > 0R>0R>0 is a sufficiently large number. Therefore, we can regard φ u φ u varphi*u\varphi \cdot uφu as a smooth function defined on R 3 R 3 R^(3)\mathbf{R}^{3}R3. Hence,
Δ ( φ u ) = ρ Δ ( φ u ) = ρ Delta(varphi u)=rho\Delta(\varphi u)=\rhoΔ(φu)=ρ
where ρ 0 ρ 0 rho-=0\rho \equiv 0ρ0 for | x | R | x | R |x| >= R|x| \geq R|x|R. Therefore, for sufficiently large | x | | x | |x||x||x|, we have
u ( x ) = φ ( x ) u ( x ) = 1 4 π R 3 ρ ( y ) | y x | d y = 1 4 π | y | R ρ ( y ) | y x | d y u ( x ) = φ ( x ) u ( x ) = 1 4 π R 3 ρ ( y ) | y x | d y = 1 4 π | y | R ρ ( y ) | y x | d y {:[u(x)=varphi(x)u(x)],[=-(1)/(4pi)int_(R^(3))(rho(y))/(|y-x|)dy],[=-(1)/(4pi)int_(|y| <= R)(rho(y))/(|y-x|)dy]:}\begin{aligned} u(x) & =\varphi(x) u(x) \\ & =-\frac{1}{4 \pi} \int_{\mathbf{R}^{3}} \frac{\rho(y)}{|y-x|} d y \\ & =-\frac{1}{4 \pi} \int_{|y| \leq R} \frac{\rho(y)}{|y-x|} d y \end{aligned}u(x)=φ(x)u(x)=14πR3ρ(y)|yx|dy=14π|y|Rρ(y)|yx|dy
Therefore,
| x | u ( x ) = 1 4 π | y | R | x | | y x | ρ ( y ) d y | x | u ( x ) = 1 4 π | y | R | x | | y x | ρ ( y ) d y |x|u(x)=-(1)/(4pi)int_(|y| <= R)(|x|)/(|y-x|)rho(y)dy|x| u(x)=-\frac{1}{4 \pi} \int_{|y| \leq R} \frac{|x|}{|y-x|} \rho(y) d y|x|u(x)=14π|y|R|x||yx|ρ(y)dy
Since | y | R , | x | | y x | | y | R , | x | | y x | |y| <= R,(|x|)/(|y-x|)|y| \leq R, \frac{|x|}{|y-x|}|y|R,|x||yx| converges uniformly to 1 as | x | | x | |x|rarr oo|x| \rightarrow \infty|x|, the conclusion follows.
Problem 6. Let f ( x , y ) C 1 ( R 2 ) f ( x , y ) C 1 R 2 f(x,y)inC^(1)(R^(2))f(x, y) \in C^{1}\left(\mathbf{R}^{2}\right)f(x,y)C1(R2). We assume that there exists C > 0 C > 0 C > 0C>0C>0 so that for all ( x , y ) R 2 , | f y ( x , y ) | C ( x , y ) R 2 , f y ( x , y ) C (x,y)inR^(2),|(del f)/(del y)(x,y)| <= C(x, y) \in \mathbf{R}^{2},\left|\frac{\partial f}{\partial y}(x, y)\right| \leq C(x,y)R2,|fy(x,y)|C. Prove that the following ODE has a globally defined solution for all y ( 0 ) = y 0 R y ( 0 ) = y 0 R y(0)=y_(0)inRy(0)=y_{0} \in \mathbf{R}y(0)=y0R :
(1) { d d x y ( x ) = f ( x , y ( x ) ) y ( 0 ) = y 0 (1) d d x y ( x ) = f ( x , y ( x ) ) y ( 0 ) = y 0 {:(1){[(d)/(dx)y(x)=f(x","y(x))],[y(0)=y_(0)]:}:}\left\{\begin{array}{l} \frac{d}{d x} y(x)=f(x, y(x)) \tag{1}\\ y(0)=y_{0} \end{array}\right.(1){ddxy(x)=f(x,y(x))y(0)=y0
In addition, we assume that f f fff is 1-periodic in x x xxx, i.e., for all ( x , y ) R 2 ( x , y ) R 2 (x,y)inR^(2)(x, y) \in \mathbf{R}^{2}(x,y)R2, we have f ( x + 1 , y ) = f ( x , y ) f ( x + 1 , y ) = f ( x , y ) f(x+1,y)=f(x,y)f(x+1, y)=f(x, y)f(x+1,y)=f(x,y). Prove that if (1) admits a globally defined bounded solution, then (1) admits a periodic solution.
Solution: The global existence is easy: fix an interval [ 0 , a ) [ 0 , a ) [0,a)[0, a)[0,a), we have
| y | | f ( x , y ( x ) ) f ( x , y ( 0 ) ) | + | f ( x , y ( 0 ) ) | C | y ( x ) y ( 0 ) | + M C | y | + M y | f ( x , y ( x ) ) f ( x , y ( 0 ) ) | + | f ( x , y ( 0 ) ) | C | y ( x ) y ( 0 ) | + M C | y | + M |y^(')| <= |f(x,y(x))-f(x,y(0))|+|f(x,y(0))| <= C|y(x)-y(0)|+M <= C|y|+M\left|y^{\prime}\right| \leq|f(x, y(x))-f(x, y(0))|+|f(x, y(0))| \leq C|y(x)-y(0)|+M \leq C|y|+M|y||f(x,y(x))f(x,y(0))|+|f(x,y(0))|C|y(x)y(0)|+MC|y|+M
where M = sup x [ 0 , a ] | f ( x , y ( 0 ) ) | M = sup x [ 0 , a ] | f ( x , y ( 0 ) ) | M=s u p_(x in[0,a])|f(x,y(0))|M=\sup _{x \in[0, a]}|f(x, y(0))|M=supx[0,a]|f(x,y(0))|. By Gronwall's inequality, y y yyy is bounded all the way up to [ 0 , a ] [ 0 , a ] [0,a][0, a][0,a]. We can then extend f f fff across a a aaa. This shows the solution can be defined globally.
Assume that φ φ varphi\varphiφ is a bounded solution. We may assume that φ ( 1 ) φ ( 0 ) φ ( 1 ) φ ( 0 ) varphi(1)!=varphi(0)\varphi(1) \neq \varphi(0)φ(1)φ(0). Otherwise, φ φ varphi\varphiφ is a periodic solution. Without loss of generality, we may assume that φ ( 1 ) > φ ( 0 ) φ ( 1 ) > φ ( 0 ) varphi(1) > varphi(0)\varphi(1)>\varphi(0)φ(1)>φ(0). By comparing two solutions φ ( x ) φ ( x ) varphi(x)\varphi(x)φ(x) and φ ( x + 1 ) φ ( x + 1 ) varphi(x+1)\varphi(x+1)φ(x+1) of (1), we see that φ ( 0 ) < φ ( 1 ) < < φ ( n ) < φ ( 0 ) < φ ( 1 ) < < φ ( n ) < varphi(0) < varphi(1) < cdots < varphi(n) < cdots\varphi(0)<\varphi(1)<\cdots<\varphi(n)<\cdotsφ(0)<φ(1)<<φ(n)<. Thus, by the boundedness of φ φ varphi\varphiφ, we may assume that
φ ( n ) y R , n φ ( n ) y R , n varphi(n)rarry_(**)inR,n rarr oo\varphi(n) \rightarrow y_{*} \in \mathbf{R}, n \rightarrow \inftyφ(n)yR,n
Therefore, the solution to (1) with y y y_(**)y_{*}y as the initial data is a 1-periodic solution.