Polynomial method

  • T-query algorithm: $|\psi_x^t\rangle=U_tO_xU_{t-1}O_x\cdots U_2O_xU_1O_x|\psi\rangle=\sum_{x,i}\alpha_{x,i}^{(t)}|i\rangle|\psi_{x,i}^t\rangle$
    • $\alpha_{x,i}^t$ is a deg $t$ polynomial
    • probability outputting $f(x)$ is a degree $2t$ polynomial
  • Adversary method - polynomial method
    • negative adversary method
    • learning graph
  • $\text{deg}_\epsilon(f)=\min{d:f \text{ is a degree }d\text{ polynomial and }\forall x\in{0,1}^n,|f(x)-p(x)|\leq\frac{1}{3}}$
  • $Q_\epsilon(f)=\min{t:\exists t$-query algorithm that computes $f$ correctly with probabiliy at least $1-\epsilon}$
  • Theorem: $Q_\epsilon(f)\geq\frac{1}{2}\text{deg}_\epsilon(f)$
  • Parity function: $f:{0,1}^n\rightarrow{0,1},f(x)=[2|\sum_i x_i]$
    • $\text{deg}_\epsilon(f)=\Omega(n)$
  • Symmetriazation
    • Given any $n$-variate multilinear polynomiial $p$, $P(k)=E_{|x|=k}[p(x)]$. Then $P$ is a polynomial with degree $\text{deg}(P)\leq\text{deg}(P)$
    • Symmetrizing + Symmetrizing the approximating function
  • Unstructured search: $\text{deg}(P)\geq\Omega(n)$
  • if the number of marked items is unknown, then there is no quantum speedup
  • Theorem(Paturi 92): $\text{deg}_\epsilon(f)=\Theta(\sqrt{n(n-\Gamma(f))})$
  • Theorem(Beals et.al. 95): for symmteric functions, $Q_\epsilon(f)=\Theta(\sqrt{n(n-\Gamma(f))})$
  • Theorem: for any total booleam function $f$, it holds that $D(f)\leq O(Q_\epsilon(f)^6)$
  • Theorem(2016): There exists a total Boolean functiion satisfying that $R_\epsilon(f)\geq\Omega(Q_\epsilon(f)^{2.5})$
    • Open Problam: gapp between $R_\epsilon(f)$ and $Q_\epsilon(f)$
  • Collision Problem

Adversary method for a pair of inputs

  • Adversary matrix $\Gamma$
    • $\Gamma_{xy}=\Gamma_{yx}\geq0$
    • if $f(x)=f(y)$ then $\Gamma_{xy}=0$
  • $W^j=\sum_{xy}\Gamma_{xy}a_x^**a_y\langle\psi_x^t|\psi_y^t\rangle$
    • $|W^0|=|\Gamma|$
      • $|\Gamma|$ is the largest eigenvalue in absolute value
      • $(a_x)$ is the priciple eigenvector of $T$
      • $(\Gamma_i){xy}=\Gamma{xy}$ if $x_i\neq y_i$
    • $|W^T|\leq2\sqrt{\epsilon(1-\epsilon)}|\Gamma|$
    • $|W^{j+1}-W^j|\leq2\max_{1\leq i\leq n}|\Gamma_i|$
  • Theorem: $Q_\epsilon(f)\geq\text{Adv}(f)=\max_{\Gamma}\frac{|\Gamma|}{\max_i|\Gamma_i|}$
  • Other adversary methods
    • Weighted adversary method [Ambainis 03]
    • Strong weighted adversary method
    • Kolmogorov complexity
    • Minimax method
    • Theorem: All adversary methods are equivalent[SS'04]
  • Limitation of adversary method
    • certificate complexity
      • $1$-certificate: $C_1(f)$
      • $0$-certificate: $C_0(f)$
    • $\text{Adv}(f)\leq O(\sqrt{C_0(f)C_1(f)})$
    • Element-Distinctness
    • 与 Polynomial method 不可比较
  • General adversary matrix $\Gamma$
    • $\Gamma_{xy}=\Gamma_{yx}\in\mathbb{R}$
    • if $f(x)=f(y)$ then $\Gamma_{xy}=0$
    • $Q_\epsilon(f)=\Theta(\text{Adv}^{\pm}(f))$ [Reichardt 09,11]
    • Learning graph [Belios]

Quantum Operations and Quantum Noise

  • Classical noise

$$\begin{bmatrix}q_0\newline q_1\end{bmatrix}=\begin{bmatrix}1-p & p’\newline p& 1-p’\end{bmatrix}\begin{bmatrix}p_0\newline p_1\end{bmatrix}$$

  • Quantum operations: $\rho\rightarrow\mathcal{E}(\rho)$
    • Unitary: $\rho\rightarrow U\rho U^\dagger$
    • Partial trace: $\rho_{AB}\rightarrow \rho_A$
    • Attachment: $\rho\rightarrow\rho\otimes\sigma_0$
  • $\mathcal{E}:M_{2^n}\rightarrow M_{2^m}$
    • Linearity
    • Trace-preserving
    • Positivity
    • Complete Positivity $\mathcal{E}(\rho)\otimes I_k$ positive for any integer $k>0$
  • System coupled to environment
    • $\mathcal{E}(X)=Tr_B U(\rho\otimes|0\rangle\langle 0|_B)U^\dagger$
    • $\mathcal{E}(\rho)=\sum_kE_k\rho E_k^\dagger$, $\sum_kE_k^\dagger E_k=I$
  • Theorem: Three conditions equivalent (合法量子操作)
  • Quantum Noise
    • Bit flip channel
    • Phase flip channel
    • Bit-phase flip channel
    • Depolarizing channel: $\mathcal{E}(\rho)=p\cdot\frac{I}{2}+(1-p)\rho=(1-p)\rho+\frac{p}{3}(X\rho X+Y\rho Y+Z\rho Z)$
      • $XMX+YMY+ZMZ+M=2TrM\cdot I$
    • Amplitude damping channel
  • Theorem: Let $\mathcal{E}:M_2\rightarrow M_2$ be a one-qubit quantum operation, then there exist real numbers $a,b,c,d$ such that $\mathcal{E}(\rho)=a\rho+bX\rho X+cY\rho Y+dZ\rho Z,a^2+b^2+c^2+d^2=1$