MathDB

2021 Alibaba Global Math Competition

Part of Alibaba Global Math Competition

Subcontests

(20)
6
2

AGMC 2021 Prelim Q6

When a company releases a new social media software, the marketing development of the company researches and analyses the characteristics of the customer group apart from paying attention to the active customer depending on the change of the time. We use n(t,x)n(t, x) to express the customer density (which will be abbreviated as density). Here tt is the time and xx is the time of the customer spent on the social media software. In the instant time tt, for 0<x1<x20<x_1<x_2, the number of customers of spending time between x1x_1 and x2x_2 is x1x2n(t,x)dx\int_{x_1}^{x_2}n(t,x)dx. We assume the density n(t,x)n(t,x) depends on the time and the following factors: Assumption 1. When the customer keeps using that social media software, their time spent on social media increases linearly. Assumption 2. During the time that the customer uses the social media software, they may stop using it. We assumption the speed of stopping using it d(x)>0d(x)>0 only depends on xx. Assumption 3. There are two sources of new customer. (i) The promotion from the company: A function of time that expresses the increase of number of people in a time unit, expressed by c(t)c(t). (ii) The promotion from previous customer: Previous customer actively promotes this social media software to their colleagues and friends actively. The speed of promoting sucessfully depends on xx, denoted as b(x)b(x). Assume if in an instant time, denoted as t=0t=0, the density function is known and n(0,x)=n0(x)n(0,x)=n_0(x). We can derive. The change of time n(t,x)n(t,x) can satisfy the equation: {tn(t,x)+xn(t,x)+d(x)n(t,x)=0,t0,x0N(t):=n(t,x=0)=c(t)+0b(y)n(t,y)dy\begin{cases} \frac{\partial}{\partial t}n(t,x)+\frac{\partial}{\partial x}n(t,x)+d(x)n(t,x)=0, t\ge 0, x\ge 0 \\ N(t):=n(t,x=0)=c(t)+\int_{0}^{\infty}b(y)n(t,y)dy \end{cases}\, where N(t)N(t) iis the speed of the increase of new customers. We assume b,dL(0,)b, d \in L^\infty_-(0, \infty). b(x)b(x) and d(x)d(x) is bounded in essence. The following, we first make a simplified assumption: c(t)0c(t)\equiv 0, i.e. the increase of new customer depends only on the promotion of previous customer. (a) According to assumption 1 and 2, formally derive the PDE that n(t,x)n(t, x) satisfies in the two simtaneous equation above. You are required to show the assumption of model and the relationship between the Maths expression. Furthermore, according to assumption 3, explain the definition and meaning of N(t)N(t) in the simtaneous equation above. (b) We want to research the relationship of the speed of the increase of the new customers N(t)N(t) and the speed of promoting sucessfully b(x)b(x). Derive an equation that N(t)N(t) satisfies in terms of N(t),n0(x),b(x),d(x)N(t), n_0(x), b(x), d(x) only and does not include n(t,x)n(t, x). Prove that N(t)N(t) satifies the estimation N(t)bebt0n0(x)dx|N(t)|\le ||b||_\infty e^{||b||_\infty t}\int_{0}^{\infty}|n_0(x)|dx, where ||\cdot||_\infty is the norm of LL^\infty. (c) Finally, we want to research, after sufficiently long time, what trend of number density function n(t,x)n(t, x) \frac{d} has. As the total number of customers may keep increasing so it is not comfortable for us to research the number density function n(t,x)n(t, x). We should try to find a density function which is renormalized. Hence, we first assume there is one only solution (λ0,φ(x))(\lambda_0,\varphi(x)) of the following eigenvalue problem: {φ(x)+(λ0+d(x))φ(x)=0,x0φ(x)>0,φ(0)=0b(x)φ(x)dx=1\begin{cases} \varphi'(x)+(\lambda_0+d(x))\varphi(x)=0, x\ge 0 \\ \varphi(x)>0,\varphi(0)=\int_{0}^{\infty}b(x)\varphi(x)dx=1 \end{cases}\, and its dual problem has only solution ψ(x)\psi(x): {φ(x)+(λ0+d(x))ψ(x)=ψ(0)b(x),x0ψ(x)>0,0ψ(x)φ(x)dx=1\begin{cases} -\varphi'(x)+(\lambda_0+d(x))\psi(x)=\psi(0)b(x), x\ge 0 \\ \psi(x)>0,\int_{0}^{\infty}\psi(x)\varphi(x)dx=1 \end{cases}\, Prove that for any convex function H:R+R+H:\mathbb{R}^+\to \mathbb{R}^+ which satisfies H(0)=0H(0)=0. We have ddx0ψ(x)φ(x)H(n~(t,x)φ(x))dx0,t0\frac{d}{dx}\int_{0}^{\infty}\psi(x)\varphi(x)H(\frac{\tilde{n}(t,x)}{\varphi(x)})dx\le 0, \forall t\ge 0. Furthermore, prove that 0ψ(x)n(t,x)dx=eλ0t0ψ(x)n0(x)dx\int_{0}^{\infty}\psi(x)n(t,x)dx=e^{\lambda_0t}\int_{0}^{\infty}\psi(x)n_0(x)dx To simplify the proof, the contribution of boundary terms in \infty is negligible.
4
2

AGMC 2021 Prelim Q4

Let nn be a positive integer. For any positive integer kk, let 0k=diag{0,...,0k}0_k=diag\{\underbrace{0, ...,0}_{k}\} be a k×kk \times k zero matrix. Let Y=(0nAAt0n+1)Y=\begin{pmatrix} 0_n & A \\ A^t & 0_{n+1} \end{pmatrix} be a (2n+1)×(2n+1)(2n+1) \times (2n+1) where A=(xi,j)1in,1jn+1A=(x_{i, j})_{1\leq i \leq n, 1\leq j \leq n+1} is a n×(n+1)n \times (n+1) real matrix. Let ATA^T be transpose matrix of AA i.e. (n+1)×n(n+1) \times n matrix, the element of (j,i)(j, i) is xi,jx_{i, j}. (a) Let complex number λ\lambda be an eigenvalue of k×kk \times k matrix XX. If there exists nonzero column vectors v=(x1,...,xk)tv=(x_1, ..., x_k)^t such that Xv=λvXv=\lambda v. Prove that 0 is the eigenvalue of YY and the other eigenvalues of YY can be expressed as a form of ±λ\pm \sqrt{\lambda} where nonnegative real number λ\lambda is the eigenvalue of AAtAA^t. (b) Let n=3n=3 and a1a_1, a2a_2, a3a_3, a4a_4 are 44 distinct positive real numbers. Let a=1i4ai2a=\sqrt[]{\sum_{1\leq i \leq 4}^{}a^{2}_{i}} and xi,j=aiδi,j+ajδ4,j1a2(ai2+a42)ajx_{i,j}=a_i\delta_{i,j}+a_j\delta_{4,j}-\frac{1}{a^2}(a^2_{i}+a^2_{4})a_j where 1i3,1j41\leq i \leq 3, 1\leq j \leq 4, δi,j={1 if i=j0 if ij\delta_{i, j}= \begin{cases} 1 \text{ if } i=j\\ 0 \text{ if } i\neq j\\ \end{cases}\,. Prove that YY has 7 distinct eigenvalue.
3
2

AGMC 2021 Prelim Q3

Last year, Master Cheung is famous for multi-rotation. This year, he comes to DAMO to make noodles for sweeping monk. One day, software engineer Xiao Li talks with Master Cheung about his job. Xiao Li mainly researches and designs the algorithm to adjust the paramter of different kinds of products. These paramters can normally be obtainly by minimising loss function ff on Rn\mathbb{R}^n. In the recent project of Xiao Li, this loss function is obtained by other topics. For safety consideration and technique reasons, this topic makes Xiao Li difficult to find the interal details of the function. They only provide a port to calculate the value of f(x)f(\text x) for any xRn\text x\in\mathbb{R}^n. Therefore, Xiao Li must only use the value of the function to minimise ff. Also, every times calculating the value of ff will use a lot of calculating resources. It is good to know that the dimension nn is not very high (around 1010). Also, colleague who provides the function tells Xiao Li to assume ff is smooth first. This problem reminds Master Cheung of his antique radio. If you want to hear a programme from the radio, you need to turn the knob of the radio carefully. At the same time, you need to pay attention to the quality of the radio received, until the quality is the best. In this process, no one knows the relationship between the angle of turning the knob and the quality of the radio received. Master Cheung and Xiao Li realizes that minimising ff is same as adjusting the machine with multiple knobs: Assume every weight of x\text x is controlled by a knob. f(x)f(\text x) is a certain performance of the machine. We only need to adjust every knobs again and again and observes the value of ff in the same time. Maybe there is hope to find the best x\text x. As a result, two people suggest an iteration algorithm (named Automated Forward/Backward Tuning, AFBT\text{AFBT}, to minimise ff. In kk-th iteration, the algorithm adjusts the individual weight of xk\text{x}_k to 2n2n points {xk±tkei:i=1,...,n}\{\text x_k\pm t_k\text e^i:i=1,...,n\}, where tkt_k is the step size; then, make yky_k be the smallest one among the value of the function of thosse points. Then check if yk\text y_k sufficiently makes ff decrease; then, take xk+1=yk\text x_{k+1}=\text y_k, then make the step size doubled. Otherwise, make xk+1=xk\text x_{k+1}=\text x_k and makes the step size decrease in half. In the algorithm, ei\text e^i is the ii-th coordinate vector in Rn\mathbb{R}^n. The weight of ii-th is 11. Others are 00; 1()\mathbf{1}(\cdot) is indicator function. If f(xk)f(yk)f(\text x_k)-f(\text y_k) is at least the square of tkt_k, then take the value of 1(f(k)f(yk)tk2)\mathbf{1}(f(\text k)-f(y_k)\ge t^2_k) as 11. Otherwise, take it as 00.
AFBT\text{AFBT} algorithm Input x0Rn\text{x}_0\in \mathbb{R}^n, t0>0t_0>0. For k=0,1,2,...k=0, 1, 2, ..., perform the following loop: 1: #Calculate loss function. 2: sk:=1[f(xk)f(yk)tk2]s_k:=\mathbb{1}[f(\text{x}_k)-f(\text{y}_k)\ge t^2_k] #Is it sufficiently decreasing? Yes: sk=1s_k=1; No: sk=0s_k=0. 3: xk+1:=(1sk)xk+skyk\text{x}_{k+1}:=(1-s_k)\text{x}_k+s_k\text{y}_k #Update the point of iteration. 4: tk+1:=22Sk1tkt_{k+1}:=2^{2S_k-1}t_k #Update step size. sk=1s_k=1: Step size doubles; sk=0s_k=0: Step size decreases by half.
Now, we made assumption to the loss function f:RnRf:\mathbb{R}^n\to \mathbb{R}. Assumption 1. Let ff be a convex function. For any x,yRn\text{x}, \text{y}\in \mathbb{R}^n and α[0,1]\alpha \in [0, 1], we have f((1α)x+y)(1α)f(x)+αf(y)f((1-\alpha)\text{x}+\text{y})\le (1-\alpha)f(\text{x})+\alpha f(\text{y}). Assumption 2. ff is differentiable on Rn\mathbb{R}^n and f\nabla f is L-Lipschitz continuous on Rn\mathbb{R}^n. Assumption 3. The level set of ff is bounded. For any λR\lambda\in\mathbb{R}, set {xRn:f(x)λ}\{\text x\in \mathbb{R}^n:f(\text x)\le \lambda\} is all bounded. Based on assumption 1 and 2, we can prove that f(x),yxf(y)f(x)f(x),yx+L2xy2\left\langle \nabla f(\text x),\text y-\text x \right\rangle \le f(\text y)-f(\text x)\le \left\langle \nabla f(\text x),\text y-\text x\right\rangle+\frac{L}{2}||\text x-\text y||^2 You can refer to any convex analysis textbook for more properties of convex function. Prove that under the assumption 1-3, for AFBTAFBT, limkf(xk)=f\lim_{k \to \infty}f(\text{x}_k)=f^*
2
2
1
2