Computational Quantum Physics

Quantum Basics

Hilbert space : H=C2n\mathcal H = \mathbb C^{2^n}

wave function : ϕH\ket \phi\in \mathcal H

  • a spin-12\frac{1}{2} system, H=C2\mathcal H=\mathbb C^2, ϕ=α+βα2+β2=1\phi = \alpha\ket\uparrow+\beta \ket \downarrow\quad |\alpha|^2+|\beta|^2=1
  • basic state : =[10]=[01]=12[11]\ket\uparrow =\begin{bmatrix}1\\0\end{bmatrix}\quad \ket \downarrow =\begin{bmatrix}0\\1\end{bmatrix}\quad \ket \rightarrow=\frac{1}{\sqrt 2}\begin{bmatrix}1\\1\end{bmatrix}

pauli matrices : σx=[0110] σy=[0ii0] σz=[1001]\sigma_x=\begin{bmatrix}0&1\\1&0\end{bmatrix}~\sigma_y=\begin{bmatrix}0&-i\\i&0\end{bmatrix}~\sigma_z = \begin{bmatrix}1&0\\0&-1\end{bmatrix}

  • σi=σi\sigma_i = \sigma_i^\dagger : Hermitian
  • σi=σi1\sigma_i = \sigma_i^{-1} : involutory
  • σi2=I\sigma_i^2 = I
  • σi=1|\sigma_i|=-1 : determinant
  • Tr(σi)=0\text{Tr}(\sigma_i) = 0 : trace
  • λ=±1\lambda = \pm1 : eigen values, eigen vectors are positive negative axes in Bloch sphere
  • [σi,σj]=2iϵijkσk[\sigma_i,\sigma_j]=2i\epsilon_{ijk}\sigma_k : commutation, ϵijk\epsilon_{ijk} : Levi-Civita symbol
  • {σi,σj}=2δijI\{\sigma_i,\sigma_j\} = 2\delta_{ij}I : anti-commutation, δij\delta_{ij} : kronecker delta

Operators :

  • spin operator : S^x=2σxS^y=2σyS^z=2σz\hat S_x = \frac{\hbar}{2}\sigma_x\quad\hat S_y=\frac{\hbar}{2}\sigma_y\quad \hat S_z=\frac{\hbar}{2}\sigma_z

    • spin pointing along direction e=[ex,ey,ez]\vec e =\begin{bmatrix}e_x,e_y,e_z\end{bmatrix} eS^=2[ezexieyex+ieyez]\vec e\cdot\hat {\vec S}=\frac{\hbar}{2}\begin{bmatrix}e_z&e_x-ie_y\\e_x+ie_y&-e_z\end{bmatrix}
  • position operator : q^ψ(q)=qψ(q)\hat q\ket {\psi(q)} = q\psi(q)

  • momentum operator : p^=iddq\hat p = -i\hbar \frac{\text d}{\text d q}

  • hamiltonian operator : H^=22m2+V\hat H = -\frac{\hbar^2}{2m}\nabla^2 + V

  • kinetic operator : T^=(p)^22m=i22m2\hat T = \frac{\hat{(\vec p)}^2}{2m} = \frac{-i\hbar ^2}{2m}\nabla^2

  • potential operator : V^=V\hat V = V

Schrödinger equationitψ(t)=H^ψ(t)i\hbar \partial_t \ket{\psi(t)}=\hat H\ket {\psi(t)}

  • for stationary problem : H^ψ=Eψψ(t)=eiEt/ψ(0)\hat H \ket \psi = E\ket\psi\quad \psi(t)=e^{-iEt/\hbar}\ket {\psi(0)}
  • external potential : itψ(r^)=22m2ψ(r)+V(r)ψ(r)i\hbar\partial_t\psi(\hat r)=-\frac{\hbar^2}{2m}\nabla^2\psi(\vec r)+V(\vec r)\psi(\vec r)

Notation

  • H^\hat H Hamilton operator
  • EE energy of the system
  • VV potential

Example

  • harmonic oscillator : 12(p^2+q^2)ψ=Eψ\frac{1}{2}(\hat p^2+\hat q^2)\ket \psi = E \ket \psi
    • V(q^)=12q^2V(\hat q) = \frac{1}{2}\hat q^2
    • ψ(q)=12nn!πeq2/2Hn(1q)\psi(q)=\frac{1}{\sqrt{2^nn!\sqrt{\hbar\pi}}}e^{-q^2/2}H_n\left(\frac{1}{\sqrt \hbar}q\right)
    • E=(n+12)E = \hbar (n+\frac{1}{2})

Density matrix : ρ^=i,jpi,jψiψj\hat \rho = \sum_{i,j}p_{i,j}\ket{\psi_i}\bra{\psi_j}

  • purity of the system Tr(ρ^2)\text{Tr}(\hat \rho^2)

  • for a pure state without noise : ρ^pure=ψψ\hat \rho_{\text{pure}}=\ket\psi\bra\psi

    Example

    ρ^==[12121212]\hat\rho_{\rightarrow} = \ket{\rightarrow}\bra{\rightarrow}=\begin{bmatrix}\frac{1}{2}&\frac{1}{2}\\\frac{1}{2}&\frac{1}{2}\end{bmatrix}

    Tr(ρ^)=1\text{Tr}(\hat \rho_\rightarrow)=1

    ρ^==[120012]\hat \rho_{\uparrow\downarrow}= \ket{\uparrow}\bra{\downarrow}=\begin{bmatrix}\frac{1}{2}&0\\0&\frac{1}{2}\end{bmatrix}

    Tr(ρ^)=12\text{Tr}(\hat \rho_{\uparrow\downarrow})=\frac{1}{2}

  • unitary time evolution : itρ^(t)=[H^,ρ^(t)]i\hbar \partial_t\hat\rho(t)=[\hat H,\hat\rho(t)]

  • thermal density matrix : ρ^β=1ieβEiieβEiii=1Tr(eβH^)eβH^\hat \rho_\beta = \frac{1}{\sum_i e^{-\beta E_i}}\sum_i e^{-\beta E_i}\ket i\bra i = \frac{1}{\text{Tr}(e^{-\beta \hat H})}e^{-\beta \hat H}

measurement : ψA^ψ=Tr(ρ^A^)\bra \psi \hat A \ket \psi = \text{Tr}(\hat \rho \hat A)

  • measure non commute operator : [A^,B^]=iΔAΔB2[\hat A,\hat B]=i\hbar\Leftrightarrow \Delta A\cdot \Delta B \ge \frac{\hbar}{2}

Quantum 1-body problem

given VV ,we want to know ψ\psi according to Schrödinger equation 22mx2ψ(x)+V(x)ψ(x)=Eψ(x)-\frac{\hbar^2}{2m}\partial_x^2\psi(x)+V(x)\psi(x)=E\psi(x)

Time-Independent 1 D Schrödinger equation

​ stationary assumption : ψ(t)=eiEt/ψ(0)\psi(t)=e^{-iEt/\hbar}\ket {\psi(0)}

22m2ψ(x)+V(x)ψ(x)=Eψ(x)Hψ(x)=Eψ(x)-\frac{\hbar^2}{2m}\nabla^2\psi(x) + V(x)\psi(x) = E\psi(x) \to H\psi(x) = E\psi(x)

​ for special form

ψ(x)+2m2(EV(x))ψ(x)=0\psi''(x) + \frac{2m}{\hbar^2}\left(E-V(x)\right)\psi(x) = 0

​ given V,m,EV,m,E we want to know ψ\psi

Numerov algorithm

(1+(Δx)212kn+1)ψn+1=2(15(Δx)212kn)ψn(1+(Δx)212kn1)ψn1+O[(Δx)6]k=2m2(EV(x))\begin{aligned} \left(1+\frac{(\Delta x)^2}{12}k_{n+1}\right)\psi_{n+1} &= 2\left(1-\frac{5(\Delta x)^2}{12}k_n\right)\psi_n - \left(1+\frac{(\Delta x)^2}{12}k_{n-1}\right)\psi_{n-1} + O[(\Delta x)^6] \\ k &= \frac{2m}{\hbar^2}(E-V(x)) \end{aligned}

  • initial problem for symmetry V(x)=V(x)V(x)=V(-x)
    • ψ(x)=ψ(x)\psi(x)=-\psi(x) : half integer mesh with ψ(12Δx)=ψ(12Δx)=1\psi(-\frac{1}{2}\Delta x) = \psi(\frac{1}{2}\Delta x) = 1
    • ψ(x)=ψ(x)\psi(-x) = -\psi(x) : integer mesh with ψ(0)=0ψ(Δx)=1\psi(0) = 0 \quad \psi(\Delta x) = 1
  • general V(x)=0V(x)=0 for xa|x|\ge a
    • ψ(aΔx)=exp(Δx2mE/)ψ(a)=1\psi(-a-\Delta x) = \text{exp}(-\Delta x\sqrt {-2mE}/\hbar)\quad \psi(-a)=1

1D scattering problem

a particle approaching the potential barrier V(x){0x[0,a]=0othersV(x)\begin{cases} \neq 0 & x\in[0,a]\\ = 0 & \text{others} \end{cases} from the left

1D scattering potential

  • wave function assumptions :

    • left (x<0x<0) wave function : ψL(x)=Aeiqxorigin wave+Beiqxreflection\psi_L(x) = \underbrace{Ae^{iqx}}_{\text{origin wave}}+\underbrace{Be^{-iqx}}_{\text{reflection}}

    • right (x>ax>a) wave function : ψR(x)=Ceiqxtransmission\psi_R(x) = \underbrace{Ce^{iqx}}_{\text{transmission}}

    • where qq is the wave number q2=2m[EV(x)]2q^2 = \frac{2m[E-V(x)]}{\hbar^2}

  • solution :

    Algorithm

    1. set C=1C=1, use Numerov algorithm starting at a+Δxa+\Delta x from right to left
    2. match the numerical solution on the left x<0x<0 to determine AA and BB
  • probability :

    • reflection probaility : R=B2A2R = \frac{|B|^2}{|A|^2}

    • transition probability : T=C2A2T = \frac{|C|^2}{|A|^2}

Bound state

particles are confined due to potential V(x){<0x[0,a]=0othersV(x)\begin{cases}<0&x\in[0,a]\\= 0&\text{others}\end{cases}

1D bound state potential

  • shooting method for eigen solver

    Algorithm

    1. try a energy EE
    2. use numerov algorithm from x=0x=0 to xfax_f\gg a
    3. satisfy ψE(xf)0\psi_E(x_f)\approx 0 then EE is a eigenvalue else try another EE
  • Improved Method - Integration from Both Sides

    Algorithm

    1. try a position b(0,a)b\in (0,a) , that E=V(b)E=V(b), 2ψEx2=0\frac{\partial^2 \psi_E}{\partial x^2} = 0
    2. use numerov from aa to bb as ψL\psi_L and from 00 to bb as ψR\psi_R
    3. satisfy ψL(b)ψL(b)=ψR(b)ψR(b)\frac{\psi'_L(b)}{\psi_L(b)}=\frac{\psi'_R(b)}{\psi_R(b)} (ψL,ψR\psi_L,\psi_R are not normalized ), then EE is a eigenvalue else try another bb

Time-independent nD Schrödinger equation

Factorization techniques :

  • along coordinate axes : ψ(r)=ψx(x)ψy(y)ψz(z)\psi(\vec r) =\psi_x(x)\psi_y(y)\psi_z(z)

  • spherical symmetry : ψ(r)=u(r)rYlm(θ,ϕ)lN0,mZ,ml\psi(\vec r) = \frac{u(r)}{r}Y_{lm}(\theta,\phi)\quad l\in\N_0,m\in\Z,|m|\le l

    apply to the Schrodinger equation : (22μ2+2l(l+1)2μr2+V(r))u(r)=Eu(r)\left(-\frac{\hbar^2}{2\mu}\nabla^2+\frac{\hbar^2l(l+1)}{2\mu r^2}+V(r)\right)u(r) = Eu(r)

    Notation

    • ll : azimuthal quantum number, magnitude of the orbital angular momentum
    • mm : magnetic quantum number, projection of the angular momentum vector along a chosen axis
    • μ\mu : mass

Solving methods :

  • finite difference

    Example : three dimensional Schrodinger

    2ψ(r)+2m[EV(r)]ψ(r)=00=1(Δx)2[ψ(xn+1,yn,zn)+ψ(xn1,yn,zn)+ψ(xn,yn+1,zn)+ψ(xn,yn1,zn)+ψ(xn,yn,zn+1)+ψ(xn,yn,zn1)]+{2m[EV(r)]6(Δx)2}ψ(xn,yn,zn)\nabla^2\psi(\vec r) + 2m[E-V(\vec r)]\psi(\vec r) = 0 \\ \Downarrow \\ \begin{aligned} 0 = \frac{1}{(\Delta x)^2}&[\psi(x_{n+1},y_n,z_n)+\psi(x_{n-1},y_n,z_n)\\ &+\psi(x_n,y_{n+1},z_n)+\psi(x_n,y_{n-1},z_n)\\ &+\psi(x_n,y_n,z_{n+1})+\psi(x_n,y_n,z_{n-1}) ]\\ &+\left\{ 2m[E-V(\vec r)]-\frac{6}{(\Delta x)^2} \right\}\psi(x_n,y_n,z_n) \end{aligned}

  • variational approaches : ϕ=iNaiui\ket \phi = \sum_i^N a_i\ket {u_i}

    ϕE=ϕH^ϕuiujSijE=uiH^ujHijUHUb^=Eb\braket{\phi}E^* = \bra \phi \hat H \ket \phi \to \underbrace{\bra{u_i}\ket{u_j}}_{S_{ij}} E^* = \underbrace{\bra{u_i}\hat H\ket{u_j}}_{H_{ij}} \to U^\top HU\hat b = E^*\vec b

    more basis more accurate

    Notation

    • ui\ket{u_i} basis
    • aia_i : basis coefficient, a=[a1,,an]\vec a = \begin{bmatrix}a_1,\cdots,a_n\end{bmatrix}^\top
    • UU : nomalization matrix for SS that USU=IU^\top SU = I
    • b\vec b : eigen vector, b=U1a\vec b = U^{-1}\vec a
  • finite element method

    • irregular geometries
    • higher accuracy

Time dependent Schrödinger Equation

itψ=H^ψi \partial_t \ket \psi = \hat H\ket \psi

Spectral methodψt=ncneiεn(tt0)/ϕn\ket{\psi_t} = \sum_n c_ne^{-i\varepsilon_n(t-t_0)/\hbar}\ket {\phi_n}

Algorithm

  1. eigen value εn\varepsilon_n and eigen vector ϕn\ket {\phi_n} for stationary problem H^ϕ=Eϕ\hat H\ket \phi = E\ket \phi
  2. represent initial wave function in eigen vectors ψ0=ncnϕn\ket {\psi_0} = \sum_n c_n \ket{\phi_n}
  3. the evolution state ψt=ncneiεn(tt0)/ϕn\ket{\psi_t} = \sum_n c_ne^{-i\varepsilon_n(t-t_0)/\hbar}\ket {\phi_n}

limitations :

  • the diagonalization of HH is complex, so this method is only useful for small problems

Notation

  • ϕn\ket {\phi_n} : eigenvector of Hϕ=EϕH |ϕ⟩ = E |ϕ⟩, ψ0=cnϕn\ket {\psi_0} = \sum c_n\ket{\phi_n}
  • εn\varepsilon_n : eigenvalue of Hϕ=EϕH |ϕ⟩ = E |ϕ⟩

Direct numerical integration : (1+iΔt2H)ψ(r,t+Δt)=(1iΔt2H)ψ(r,t)\left(\mathbb 1+ \frac{i\Delta t}{2\hbar}H\right)\psi(\vec r, t+\Delta t) = \left(\mathbb 1 - \frac{i\Delta t}{2\hbar}H\right)\psi(\vec r, t)

  • forward euler : ψ(tn+1)=ψ(tn)iΔtH^ψ(tn)\ket {\psi(t_{n+1})}=\ket {\psi(t_n)} - \frac{i\Delta t}{\hbar}\hat H\ket{\psi(t_n)}
    • numerically unstable
    • violet conservation of ϕ\braket \phi
  • implicit method : (1+iΔt2H)ψ(r,t+Δt)=(1iΔt2H)ψ(r,t)\left(\mathbb 1+ \frac{i\Delta t}{2\hbar}H\right)\psi(\vec r, t+\Delta t) = \left(\mathbb 1 - \frac{i\Delta t}{2\hbar}H\right)\psi(\vec r, t)
    • HH is sparse matrix, using iterative solver (e.g. biconjugate gradient)

Split-operator method : ψ(q)F1Fψ(p)H^=T^(p)+V^(q)\psi(\vec q) \xrightleftharpoons[\mathcal F^{-1}]{\mathcal F}\psi(\vec p)\Rightarrow \hat H =\textcolor{cyan}{\hat T(\vec p)}+\textcolor{magenta}{\hat V(\vec q)}

eitH^/=eiΔtV^/2[eiΔtT^/eiΔtV^/]N1eiΔtT^/eiΔtV^/2e^{-i t\hat H /\hbar} = \textcolor{magenta}{e^{-i\Delta t\hat V/2\hbar}}\left[\textcolor{cyan}{e^{-i\Delta t\hat T/\hbar}}\textcolor{magenta}{e^{-i\Delta t\hat V/\hbar}} \right]^{N-1}\textcolor{cyan}{e^{-i\Delta t\hat T/\hbar}}\textcolor{magenta}{e^{-i\Delta t\hat V/2\hbar}}

Algorithm

  1. ψ(q)eiΔtV(q)/2ψ0(q)\textcolor{magenta}{\psi(\vec q)\gets e^{-i\Delta tV(\vec q)/2\hbar}\psi_0(\vec q)}
  2. loop N-1 timesteps
    1. ψ(p)Fψ(q)\textcolor{cyan}{\psi(\vec p)} \overset{ \mathcal F}{\gets} \textcolor{magenta}{\psi(\vec q)}
    2. ψ(p)eiΔtp2/2mψ(p)\textcolor{cyan}{\psi(\vec p) \gets e^{-i\Delta t \hbar\Vert\vec p\Vert^2/2m}\psi(\vec p)}
    3. ψ(q)F1ψ(p)\textcolor{magenta}{\psi(\vec q)} \overset{\mathcal F^{-1}}\gets \textcolor{cyan}{\psi(\vec p)}
    4. ψ(q)eiΔtV(q)/ψ(q)\textcolor{magenta}{\psi(\vec q)\gets e^{-i\Delta t V(\vec q)/\hbar}\psi(\vec q)}
  3. ψ(p)Fψ(q)\textcolor{cyan}{\psi(\vec p)} \overset{\mathcal F}{\gets}\textcolor{magenta}{\psi(\vec q)}
  4. ψ(p)eiΔtp2/2mψ(p)\textcolor{cyan}{\psi(\vec p) \gets e^{-i\Delta t \hbar\Vert\vec p\Vert^2/2m}\psi(\vec p)}
  5. ψ(q)F1ψ(p)\textcolor{magenta}{\psi(\vec q)} \overset{\mathcal F^{-1}}{\gets} \textcolor{cyan}{\psi(\vec p)}
  6. ψ(q)eiΔtV(q)/2\textcolor{magenta}{\psi(\vec q)\gets e^{-i\Delta t V(\vec q)/2\hbar}}

Notation

  • p\vec p : momentum in hamilton expression
  • q\vec q : position in hamilton expression
  • T^\hat T : kinetic operator , T^=(p)^22m=i22m2\hat T = \frac{\hat{(\vec p)}^2}{2m} = \frac{-i\hbar ^2}{2m}\nabla^2
  • V^\hat V : potential operator , V^=V\hat V = V
  • F\mathcal F : fourier operator , Fψ(q)=(12π)d+ψ(q)eipqdq\mathcal F \psi(\vec q) = \left(\frac{1}{\sqrt {2\pi}}\right)^d\int_{-\infin}^{+\infin}\psi(\vec q)e^{-i\vec p\cdot \vec q}\text d\vec q
  • F1\mathcal F^{-1} : inverse fourier operator , F1ψ(p)=(12π)d+ψ(p)eipqdp\mathcal F^{-1} \psi(\vec p) = \left(\frac{1}{\sqrt {2\pi}}\right)^d\int_{-\infin}^{+\infin}\psi(\vec p)e^{-i\vec p\cdot \vec q}\text d\vec p

Quantum n-body problem

Hilbert space for n particles: HN=HN\mathcal H^N = \mathcal H^{\otimes N}

Indistinguishable Particles

Bosons and Fermions :

  • fermions : ψ(r1,r2)=ψ(r2,r1)\psi(\vec r_1,\vec r_2) = -\psi(\vec r_2, \vec r_1)

    ΨA=NApsign(p)ψ(rp(1),,rp(N))\Psi^A = \mathcal N_A\sum_p\text{sign}(p)\psi(\vec r_{p(1)},\cdots,\vec r_{p(N)})

    Notation

    • ΨA\Psi^A : n particle fermions wave function
    • NA\mathcal N_A : normalization factor
    • pp : permutation
    • Pauli exclusion principle : ΨA(r1,r2)=ψ(r1,r2)ψ(r2,r1)0\Psi^A(\vec r_1,\vec r_2) = \psi(\vec r_1,\vec r_2)-\psi(\vec r_2,\vec r_1) \neq 0
    • spinful, generalized coordinate r=(r,σ)r=(\vec r, \sigma)
  • bosons : ψ(r1,r2)=ψ(r2,r1)\psi(\vec r_1, \vec r_2) = \psi(\vec r_2, \vec r_1)

    ΨS=NSpψ(rp(1),,rp(N))\Psi^{S} = \mathcal N_S\sum_p \psi(\vec r_{p(1)},\cdots,\vec r_{p(N)})

    Notation

    • ΨS\Psi^S : n particle bosons wave function
    • NS\mathcal N_S : normalization factor

Fock space: F=N=0S±HN\mathcal F = \bigoplus_{N=0}^\infin \mathcal S_{\pm}\mathcal H^N

​ possible particle configurations for a given type of particle

  • F=F00 particlesF11 particles\mathcal F = \underbrace{\mathcal F_0}_{\text{0 particles}} \oplus \underbrace{\mathcal F_1}_{\text{1 particles}} \oplus \cdots

Notation

  • \oplus : direct sum, e.g. AB=[A00B]\textbf A \oplus \textbf B = \begin{bmatrix}\textbf A&\textbf 0\\\textbf 0 &\textbf B\end{bmatrix}
  • S±S_\pm : symmetrization for bosons S+=NSp\mathcal S_+ =\mathcal N_S\sum_p / antisymmetrization operator for fermions S=NApsgn(p)\mathcal S_- = \mathcal N_A\sum_p\text{sgn(p)}

Example

Bosons Spinless Fermions Spinful Fermions Spin-12\frac{1}{2}
Fock space dimension \infin(bosons can take same position) 2N2^N 4N4^N 2N2^N

Slater determinant :Ψ(r1,,rN)=1N!ϕ1(r1)ϕN(r1)ϕr(rN)ϕN(rN) \Psi(r_1,\cdots,r_N) = \frac{1}{\sqrt {N!}}\left|\begin{matrix} \phi_1(r_1)&\cdots&\phi_N(r_1)\\ \vdots & \ddots&\vdots\\ \phi_r(r_N)&\cdots&\phi_N(r_N) \end{matrix}\right|

​ anti-symmetrized and normalized NN single particle wave function product

Notation

  • ϕi(rj)\phi_i(r_j) : wave function of fermion ii at position rjr_j

Creation and annihilation operators

  • a^\hat a annihilation operator : remove particle a^iϕj=δij0\hat a_i\ket {\phi_j} = \delta_{ij}\ket {0}

  • a^\hat a^\dagger creation operator : add particle ϕi=a^i0\ket{\phi_i} = \hat a_i^\dagger \ket {0}

Notation

  • 0\ket {0} : vacuum state with no particles, 0=[00]\ket {0} = \begin{bmatrix}0\\0\end{bmatrix}
  • [,][\cdot,\cdot] : commute, [A,B]=ABBA[A,B]=AB-BA
  • {,}\{\cdot,\cdot\} : anti-commute, {A,B}=AB+BA\{A,B\}=AB+BA
  • Bosons : commute

    • a^ini=nini1a^ini=ni+1ni+1\hat a_i\ket{n_i}=\sqrt{n_i}\ket{n_i-1} \quad \hat a_i^\dagger \ket {n_i} = \sqrt {n_i+1}\ket{n_i+1}

    • a^ia^i=ni\hat a_i^\dagger\hat a_i = n_i

    • [a^i,a^j]=δij[a^i,a^j]=[a^i,a^j]=0[\hat a_i, \hat a_j^\dagger] = \delta _{ij}\quad [\hat a_i, \hat a_j] = [\hat a_i^\dagger, \hat a_j^\dagger] = 0

    • 0↚a^0a^a^1a^a^20\underset{\hat a}{\not\leftarrow}\ket 0 \xrightleftharpoons[\hat a]{\hat a^\dagger}\ket 1\xrightleftharpoons[\hat a]{\hat a^\dagger}\ket 2\cdots

  • Fermions : anti-commute

    • c^uiui,uj,=uj,c^uiuj,=ui,uj,\hat c_{u_i}\ket{u_i,u_j,\cdots}=\ket{u_j,\cdots} \quad \hat c_{u_i}\ket{u_j,\cdots} = \ket{u_i,u_j,\cdots}

    • c^ic^i=n^i\hat c_i^\dagger\hat c_i =\hat n_i

      • ni=0n_i = 0 : c^ic^iuj,=0\hat c_i^\dagger\hat c_i\ket{u_j,\cdots} = 0
      • ni=1n_i=1 : c^uic^uiui,uj,=ui,uj,\hat c_{u_i}^\dagger\hat c_{u_i}\ket{u_i,u_j,\cdots}=\ket{u_i,u_j,\cdots}
    • {c^i,c^j}=δij{c^i,c^j}={c^i,c^j}=0\{\hat c_i, \hat c_j^\dagger\} = \delta_{ij}\quad \{\hat c_i, \hat c_j\} = \{\hat c_i^\dagger, \hat c_j^\dagger\} = 0

    • 0↚c^ui0c^uic^uiui↛c^ui00 \underset{\hat c^\dagger_{u_i}}{\not \leftarrow}\ket 0 \xrightleftharpoons[\hat c_{u_i}]{\hat c^\dagger_{u_i}}\ket {u_i}\overset{\hat c^\dagger_{u_i}}{\not\rightarrow} 0

Quantum Spin Model

(TFIM)Transverse field Ising model

H^=<ij>JijS^izS^jzihi2S^ix\hat H = \sum_{<ij>} J_{ij} \hat S_i^z\hat S_j^z - \sum_{i} \frac{h_i}{2}\hat S_i^x

S^izS^jz=IS^zn=iS^zn=j1I\hat S_i^z\hat S_{j}^z = I\otimes\cdots \otimes \underbrace{\hat S^z}_{n=i}\otimes\cdots\otimes \underbrace{\hat S^z}_{n=j}\otimes\cdots\otimes \mathbb 1I

  • quantum phase transition between a spontaneously symmetry-broken and a disordered phase
  • extension of the classical Ising model by adding a magnetic field in the xx direction

Notation

  • <ij><ij> : connection between particle ii and particle jj
  • JijJ_{ij} : interacting constant between particle ii and particle jj
  • hih_i : external magenatic field on particle ii
  • S^x\hat S^x : spin operator in xx direction, S^x=12σx=12[0110]\hat S^x = \frac{1}{2}\hbar\sigma_x = \frac{1}{2}\hbar\begin{bmatrix}0&1\\1&0\end{bmatrix}
  • S^z\hat S^z : spin operator in zz direction, S^z=12σz=12[1001]\hat S^z = \frac{1}{2}\hbar \sigma_z = \frac{1}{2}\hbar\begin{bmatrix}1&0\\0&-1\end{bmatrix}

Heisenberg model

H^=<ij>JijSi^Sj^=<ij>Jij(S^ixS^jx+S^iyS^jy+S^izS^zj)=<ij>Jij[12(S^i+S^j+S^iS^j+)+S^izS^jz]\begin{aligned} \hat H &= \sum_{<ij>} J_{ij}\hat{\vec S_i}\cdot \hat{\vec S_j} = \sum_{<ij>}J_{ij}\left(\hat S^x_i \hat S^x_j +\hat S^y_i\hat S^y_j+\hat S^z_i\hat S_z^j\right) \\ &= \sum_{<ij>}J_{ij}\left[ \frac{1}{2} \left(\hat S_i^+\hat S_j^- + \hat S_i^- \hat S_j^+\right) + \hat S_i^z \hat S_j^z \right] \end{aligned}

Notation

  • S^±\hat S^\pm : raising/lowering operator , S^±=σ±=(σx±iσy)\hat S^\pm = \hbar \sigma^\pm =\hbar(\sigma_x\pm i \sigma_y)
    • S^+S^+=S^+=null=[10]=[01]null=[00]\hat S^+ \hat S^+ \ket \downarrow = \hat S^+\ket \uparrow = \ket {\text{null}} \quad \ket{\uparrow} = \begin{bmatrix}1\\0\end{bmatrix}\qquad \ket{\downarrow} = \begin{bmatrix}0\\1\end{bmatrix}\qquad\ket{\text{null}} = \begin{bmatrix}0\\0\end{bmatrix}
    • (σ±)2=0(\sigma^\pm)^2= 0 : a spin can be flipped only only once
  • M^z\hat M^z : total magnetization , M^z=iS^iz\hat M^z=\sum_i \hat S_i^z
  • conserve total magentization
  • Hamitonian has SU(2)SU(2) symmetry

Example : two particles ({,,,}\{\ket{\uparrow\uparrow}, \ket{\uparrow\downarrow},\ket {\downarrow\uparrow}, \ket{\downarrow\downarrow}\})

H^=[14Jij000014Jij12Jij0012Jij14Jij000014Jij]\hat H = \begin{bmatrix}\frac{1}{4}J_{ij}&0&0&0\\0&-\frac{1}{4}J_{ij}&\frac{1}{2}J_{ij}&0\\0&\frac{1}{2}J_{ij}&-\frac{1}{4}J_{ij}&0\\0&0&0&\frac{1}{4}J_{ij}\end{bmatrix}

XXZXXZ model

H^=<ij>Jij(S^ixS^jx+S^iyS^jy+ΔS^izS^jz)\hat H = \sum_{<ij>}J_{ij}\left( \hat S_i^x\hat S_j^x + \hat S_i^y\hat S_j^y + \Delta\hat S_i^z\hat S_j^z \right)

  • conserve total magentization M^z\hat M^z

Notation

  • Δ\Delta : hyperparameter

    Δ=0\Delta = 0 Δ=1\Delta = 1 Δ\Delta\to \infin
    XYXY model Heisenberg model Ising model

Jordan-Wigner Transformation

​ mapping spin models to spinless fermions, derive from XXZXXZ model

H^=12<ij>Jij(c^ic^j+c^jc^i+2Δn^in^j)\hat H = \frac{1}{2}\sum_{<ij>}J_{ij}\left( \hat c_i^\dagger\hat c_j +\hat c_j^\dagger\hat c_i+ 2\Delta\hat n_i\hat n_j \right)

Notation

  • c^i/c^i\hat c_i/\hat c_i^\dagger : Jordan-Wigner transformation operator, c^i=j<i(σjz)σi+c^i=j<i(σjz)σi\hat c_i = \prod_{j<i}\left(\sigma_j^z\right)\sigma_i^+ \qquad \hat c_i^\dagger = \prod_{j<i}\left(\sigma_j^z\right)\sigma_i^-
    • {c^i,c^j}=δij\{\hat c_i,\hat c_j^\dagger\}=\delta_{ij}
    • {c^i,c^j}={c^i,c^j}=0\{\hat c_i, \hat c_j\}=\{\hat c_i^\dagger, \hat c_j^\dagger\}=0
  • n^i\hat n_i : number operator, n^i=c^ic^i\hat n_i=\hat c_i^\dagger \hat c_i

Brute-force method

[ED] Exact Diagonalization

diagonalizing the Hamiltonian matrix

  • full spectrum N20N\approx 20
  • Lanczos algorithm N40N\approx 40

Lanczos algorithm

  • storage complexity O(2N)\mathcal O(2^N) compared to dense matrix eigen solvers of O(2N)2\mathcal O(2^N)^2
  • ghost state : low-lying eigen values result from round of error that rn\vec r_n is not fully orthogonal

Algorithm

  1. find the orthogonalized basis ri\vec r_i using Gram-Schmidt orthogonalization

    r0=vvβmrm=Hrm1αm1rm1βm1rm2αn=rnHrnβn=rnHrn1\begin{aligned} \vec r_0 = \frac{\vec v}{\Vert \vec v\Vert} \quad \beta_m\vec r_m = H\vec r_{m-1} - \alpha_{m-1}\vec r_{m-1} - \beta_{m-1}\vec{r}_{m-2} \end{aligned}\quad \alpha_n =\vec r_n^\dagger H\vec r_n \quad \beta_n = |\vec r_n^\dagger H\vec r_{n-1}|

  2. express Hamiltonian HH in tridiagonal matrix

    HM=[α0β100β1α10000αM1βM00βMαM]H^M = \begin{bmatrix} \alpha_0 & \beta_1 & \cdots & 0 & 0 \\ \beta_1 & \alpha_1 & \cdots & 0 & 0 \\ \vdots & \vdots & \ddots & \vdots & \vdots\\ 0 & 0 & \cdots & \alpha_{M-1}&\beta_M \\ 0 & 0 & \cdots & \beta_M & \alpha_M \end{bmatrix}

  3. eigendecomposite the HMH^M

  4. transform the eigenvectors to the original basis

    • for memory constraint, only store the last three rn\vec r_n and recompute rn\vec r_n iteratively to perform basis transformation

Spin-12\frac{1}{2} hamitonians

two possible state \ket\uparrow and \ket\downarrow bitwise operation (xor) rather than vector

  • S^izS^i+1z\hat S_i^z\hat S_{i+1}^z : s = s ^ (s>>1)
  • S^i+S^i+1\hat S_i^+\hat S_{i+1}^- : s = s ^ (3<<i)

Example

assume state s=0112s=011_2

then for heisenberg model s~=01120102=0102\tilde s = 011 _2\oplus 010_2 = 010_2 where \oplus is bitwise xor here.

Notation

  • S^±\hat S^\pm : S^±=σ±=(σx±iσy)\hat S^\pm = \hbar \sigma^\pm =\hbar(\sigma_x\pm i \sigma_y)

symmetries

block diagonalize the Hamitonian and solve within the symmetries’ eigenspaces.

Example : Transverse Field Ising Model

  1. parity operator : P^=iσix\hat P = \bigotimes_i \sigma_i^x, the eigen values are ±1\pm 1
  2. ψ=P^Mψ\ket \psi = \hat P^M \ket \psi : for random state ψ\psi, apply operator for MM times we find the initial state again
  3. eigen state becomes : i=0MP^iψ\sum_{i=0}^M\hat P^i\ket\psi
  4. construct hamiltonian HH from eigen state and eigen vector

Time evolution

Trotter-Suzuki decomposition: H^=k=1Kh^keiH^Δt/=k=1Keih^kΔt/+O(Δt2)\hat H = \sum_{k=1}^K \hat h_k\to e^{-i\hat H\Delta t/\hbar}= \prod_{k=1}^K e^{-i\hat h_k\Delta t/\hbar}+\mathcal O(\Delta t^2)

  • time-indepedent assumption : ψ(t+Δt)=eiH^Δt/ψ(t)\ket{\psi(t+\Delta t)}= e^{-i\hat H\Delta t/\hbar}\ket{\psi(t)}
  • non-commuting decomposition : H^=k=1Kh^k[h^i,h^j]0ij\hat H = \sum_{k=1}^K\hat h_k\quad [\hat h_i,\hat h_j]\neq 0\quad i\neq j
  • second order version : eiH^Δt/=(k=1Keih^kΔt/2)(k=K1eih^kΔt/2)+O(Δt3)e^{-i\hat H\Delta t/\hbar} = \left(\prod_{k=1}^K e^{-i\hat h_k\Delta t/2\hbar}\right)\left(\prod_{k=K}^1 e^{-i\hat h_k\Delta t/2\hbar}\right)+\mathcal O(\Delta t^3)

Example : K=2K=2

ψ(t+Δt)=eih^1Δt/2eih^2Δt/eih^1Δt/2ψ\ket{\psi(t+\Delta t)} = e^{-i\hat h_1 \Delta t/2\hbar}e^{-i\hat h_2\Delta t/\hbar}e^{-i\hat h_1\Delta t/2\hbar}\ket{\psi}

Example : Transverse Field Ising Model

H^=<ij>Jijσizσjzh^1ihiσixh^2\hat H = \underbrace{\sum_{<ij>}J_{ij}\sigma_i^z\sigma_j^z}_{\hat h_1} - \underbrace{\sum_i h_i\sigma_i^x}_{\hat h_2}

eih^1Δt/=<ij>eiΔtJijsizsj6z/e^{-i\hat h_1\Delta t/\hbar} = \bigotimes_{<ij>}e^{-i\Delta tJ_{ij}s_i^z s_j6z/\hbar}

eih^2Δt/=[cos(Δthi/)isin(Δthi/)isin(Δthi/)cos(Δthi/)]e^{-i\hat h_2\Delta t/\hbar} = \begin{bmatrix}\text{cos}(\Delta th_i/\hbar)&i\text{sin}(\Delta t h_i/\hbar)\\ i\text{sin}(\Delta th_i/\hbar)&\text{cos}(\Delta t h_i/\hbar)\end{bmatrix}

(since eA=1+A+A22!+e^A = 1 + A + \frac{A^2}{2!}+\cdots )

Notation

  • [,][\cdot,\cdot] : commute operator, [A,B]=ABBA=0A,B[A,B]=AB-BA=0\to A,B commute
  • <i,j><i,j> : means i,ji,j are neighbors
  • JijJ_{ij} : connection between site ii and jj
  • hih_i : magenatic field at site ii
  • h^k\hat h_k : non-commuting term
  • sis_i : eigen value for σiz\sigma^z_i

Imaginary-time evlotion: itτit\to \tau

  • time-indepedent assumption : ψ(t)=eiH^t/ψ(0)ψ(t)=eτH^ψ(0)\ket{\psi(t)}= e^{-i\hat Ht/\hbar}\ket{\psi(0)}\to \ket{\psi(t)}=e^{-\tau\hat H}\ket{\psi(0)}
  • converges to the ground state by suppressing the amplitudes of excited states
    exponentially fast in the product ΔEkτ\Delta E_k\tau .

Magnus expansian : U^(Δt)=eiHˉtΔt/+O(Δt2)Ht=Hˉt1+Hˉt2+\hat U(\Delta t) = e^{-i\bar H_t\Delta t/\hbar}+\mathcal O(\Delta t^2)\quad H_t = \bar H_t^1 +\bar H_t^2+\cdots

  • time-depdent assumption : ψ(t)=U(t,t)ψ(t)\ket{\psi(t')}=U(t',t)\ket{\psi(t)}
  • Hˉt1=1Δttt+ΔtH^(s)ds\bar H^1_t = \frac{1}{\Delta t}\int_{t}^{t+\Delta t} \hat H(s)ds and Ht2=iΔttt+Δtdstsdl[H^(s),H^(l)]H^2_t = -\frac{i}{\Delta t}\int_{t}^{t+\Delta t}ds\int_{t}^sdl\left[\hat H(s),\hat H(l)\right]

Notation

  • UU : evolution operator, U^(t,t)=eittH^(s)ds/t>t\hat U(t',t) = e^{-i\int_t^{t'}\hat H(s)\text ds/\hbar}\quad t'>t

Matrix Product States

Bipartite entanglement

Reduced density matrix : ρA=TrB(ψψ)ψH=HAHB\rho_A =\text{Tr}_B(\ket \psi \bra\psi)\quad \ket\psi \in\mathcal H =\mathcal H_A\otimes \mathcal H_B

Notation

  • \otimes : kronecker product, e.g. [1001][1111]=[1100110000110011]\begin{bmatrix}1&0\\0&1\end{bmatrix}\otimes \begin{bmatrix}1&1\\1&1\end{bmatrix}=\begin{bmatrix}1&1&0&0\\1&1&0&0\\0&0&1&1\\0&0&1&1\end{bmatrix}
  • TrB\text{Tr}_B : partial trace over subsystem BB , e.g. TrB[A11B11A11B12A12B11A12B12A11B21A11B22A12B21A12B22A21B11A21B12A22B11A22B12A21B21A21B22A22B21A22B22]=[iiBiiA11iiBiiA12iiBiiA21iiBiiA22]\text{Tr}_B\begin{bmatrix}A_{11}B_{11}&A_{11}B_{12}&A_{12}B_{11}&A_{12}B_{12}\\A_{11}B_{21}&A_{11}B_{22}&A_{12}B_{21}&A_{12}B_{22}\\A_{21}B_{11}&A_{21}B_{12}&A_{22}B_{11}&A_{22}B_{12}\\A_{21}B_{21}&A_{21}B_{22}&A_{22}B_{21}&A_{22}B_{22}\end{bmatrix}=\begin{bmatrix}\sum_{ii}B_{ii}A_{11}&\sum_{ii}B_{ii}A_{12}\\\sum_{ii}B_{ii}A_{21}&\sum_{ii}B_{ii}A_{22}\end{bmatrix}

Entanglement : S=Tr(ρA log ρA)=Tr(ρB log ρB)S=-\text{Tr}(\rho_A~ \text{log}~\rho_A)=-\text{Tr}(\rho_B~\text{log}~\rho_B)

using Schmidt decomposition ρA=αλα2ϕαAϕαASαλα2logλα2\rho_A = \sum_\alpha \lambda_\alpha^2 \ket{\phi_\alpha}_A\bra{\phi_\alpha}_A\to S-\sum_\alpha \lambda_\alpha^2\text{log}\lambda_\alpha^2

  • product state (zero entanglement) : S=0λ1=1,λα>1=0S=0\Leftrightarrow\lambda_1 = 1,\lambda_{\alpha>1} = 0
  • maximally entangled state : S=N2logdλi=1/dN/2S=\frac{N}{2}\text{log}d\Leftrightarrow \lambda_i = 1/\sqrt{d^{N/2}}
  • random state : SN2logd12S\approx \frac{N}{2}\text{log}d-\frac{1}{2}

Area law of entanglement : entanglement entropy scales as SLD1S\propto L^{D-1}

Example : 1-D entanglement

SconstS\sim\text{const}

Notation

  • SS : entanglement entropy
  • dd : Hilbert space dimension
  • λ\lambda : eigen value
  • NN : number of sites
  • DD : dimension of the entanglement system
  • LL : linear dimension of system

Matrix Product state

[MPS] Matrix Product State : ψ=sTr(A1s1ANsN)s1sN\ket{\psi}= \sum_s \text{Tr}(A_1^{s_1}\cdots A^{s_N}_N)\ket {s_1\cdots s_N}

mps

  • canonical form (normalization) : A=ΛΓA=\Lambda \Gamma

mps canonical

Example : GHZ or ‘cat’ state

GHZ=12(N+N)=1Z(Tr((A)N)N+Tr((A)N)N)\ket {GHZ} = \frac{1}{\sqrt 2}(\ket{\downarrow}^{\otimes N}+\ket{\uparrow}^{\otimes N}) = \frac{1}{Z}\left(\text{Tr}\left((A^{\uparrow})^N\right)\ket{\uparrow}^{\otimes N}+\text{Tr}\left((A^\downarrow)^N\right)\ket{\downarrow}^{\otimes N} \right)

where ZZ is a norm and Ai=A=[1000]Ai=A=[0001]A^\downarrow_i = A^\downarrow = \begin{bmatrix}1&0\\0&0\end{bmatrix}\quad A^\uparrow_i = A^\uparrow = \begin{bmatrix}0&0\\0&1\end{bmatrix}

Example : AKLT state

ground state of spin-1 Hamiltonian : H^=jSj^S^j+1+13(S^jS^j+1)2\hat H = \sum_j \hat {\vec S_j}\cdot \hat {\vec S}_{j+1} + \frac{1}{3}\left(\hat{\vec S}_j \cdot \hat{\vec S}_{j+1}\right)^2

with matrices : Ai+=A+=23σ+=[02300]Ai0=A0=13σz=[130013]Ai=A=23σ[00230]A^+_i = A^+=\sqrt{\frac{2}{3}}\sigma^+ = \begin{bmatrix}0&\sqrt{\frac{2}{3}}\\0&0\end{bmatrix}\quad A^0_i = A^0 =\frac{-1}{\sqrt 3}\sigma^z = \begin{bmatrix}-\frac{1}{\sqrt{3}} & 0 \\0 & \frac{1}{\sqrt 3}\end{bmatrix}\quad A^-_i = A^- = - \sqrt{\frac{2}{3}}\sigma^- \begin{bmatrix}0&0\\-\sqrt{\frac{2}{3}}&0\end{bmatrix}

the corresponding +\ket +, \ket - , 0\ket 0 are three states for spin-1 particle not for spin-12\frac{1}{2} particle

Notation

  • AiA_i : a rank-3 tensor, AiCD×2×2A_i\in \mathbb C^{D\times 2\times 2}, DD is the number of basis state for single site.

    AisiA_i^{s_i} means when the site ii is in state sis_i, there is a 2×22\times2 matrix for product

    For translationally symmetric Ai=AA_i=A

  • σ+\sigma^+ : creation / raising operator , σ+=[0010]\sigma^+ = \begin{bmatrix}0&0\\1&0\end{bmatrix}

    • σ+=0\sigma^+ \ket \downarrow = 0
    • σ+=\sigma^+ \ket \uparrow =\ket \downarrow
  • σ\sigma^- : annihilation / lowering operator, σ=[0100]\sigma^- = \begin{bmatrix}0&1\\0&0\end{bmatrix}

    • σ=\sigma^- \ket \downarrow = \ket \uparrow
    • σ=0\sigma^- \ket \uparrow = 0
  • +\ket + : for spin-1 , +=\ket + = \ket {\uparrow\uparrow}

  • \ket - : for spin-1 , =\ket - = \ket{\downarrow\downarrow}

  • 0\ket 0 : for spin-1 , 0=12(+)\ket 0 = \frac{1}{\sqrt 2}\left(\ket{\uparrow\downarrow}+\ket{\downarrow\uparrow}\right)

[MPO] Matrix Product Operator : O^=σi,σi[W1σ1σ1WNσNσN]σ1σNσ1σN\hat O = \sum_{\sigma_i,\sigma_i'}\left[W_1^{\sigma_1\sigma_1'}\cdots W_N^{\sigma_N\sigma_N'}\right]\ket{\sigma_1\dots\sigma_N}\bra{\sigma_1'\cdots\sigma_N'}

mpo

  • computing ψO^ψ\bra \psi \hat O\ket \psi :

    mps mpo computation

Example : single site operator

O^j=IO^site jI\hat O_j = I\otimes\cdots\otimes \underbrace{\hat O}_{\text{site}~j}\otimes\cdots\otimes I

Wiσi,σi=σiO^σiW_i^{\sigma_i,\sigma_i'}=\bra {\sigma_i}\hat O\ket {\sigma_i'}

Example : paramagnetic system H^=ihS^iz\hat H = -\sum_i h\hat S_i^z

H^=(hS^zII)++(IIhS^z)\hat H = (-h\hat S^z\otimes I\otimes \cdots \otimes I)+\cdots + (I\otimes \cdots\otimes I\otimes -h\hat S^z)

W1=[hSzI]Wi=[I0hSzI]WN=[IhSz]W_1 = \begin{bmatrix}-hS^z&I\end{bmatrix}\quad W_i=\begin{bmatrix}I&0\\-hS^z&I\end{bmatrix}\quad W_N=\begin{bmatrix}I\\-hS^z\end{bmatrix}

Example : Transverse field Ising model H^=iS^izS^i+1z+hiS^ix\hat H =-\sum_i \hat S_i^z\hat S_{i+1}^z + h\sum_i\hat S_i^x

W1=[hSxSzI]Wi=[I00Sz00hSxSzI]WN=[ISzhSx]W_1 = \begin{bmatrix}hS^x&-S^z&I\end{bmatrix}\quad W_i =\begin{bmatrix}I&0&0\\S^z&0&0\\hS^x&-S^z&I\end{bmatrix}\quad W_N = \begin{bmatrix}I\\S^z\\hS^x\end{bmatrix}

Notation

  • WiW_i a rank-4 tensor, WiCD×D×2×2W_i\in \mathbb C^{D\times D\times 2\times 2}

    WiσiσjW_i^{\sigma_i\sigma_j} means for site ii when the left state is σi\sigma_i and right state σj\sigma_j there is a 2×22\times2 matrix for product

[DMRG] Density matrix renormalization group

find the ground state that argminψψH^ψψ\underset{\ket \psi}{\text{argmin}}\frac{\bra \psi \hat H \ket \psi}{\braket{\psi}}

  • left normalization : AA=IA=UΣVA=UA^\dagger A = I\quad A'=U\Sigma V^\dagger\to A = U
  • right normalization : BB=IB=UΣVB=VBB^\dagger = I\quad B'=U\Sigma V^\dagger\to B=V^\dagger
  • substitution algorithm : imaginary time evolution, but converge slower

Algorithm

  1. random initialize ψ\ket \psi as right-normalized

  2. build R1R_1

  3. repeat until energy converge Var(H)<ϵ\text{Var}(H)<\epsilon

    1. right sweep for l=1,,L1l=1,\dots,L-1

      1. solve eigen value for MlM_l

      sweep

      1. left normalize MlM_l

      2. build LlL_l

    2. Left sweep for l=L,,2l=L,\dots,2

      1. solve eigen value for MlM_l
      2. right normalize MlM_l
      3. build RlR_l

[TEBD] Time evolving block decimation

TEBD

Algorithm

  1. two site tensor contraction

two site tensor contraction

  1. apply evolution gate

apply evolution gate

  1. split into single site tensor

split into single site tensor

  1. truncation : keep χmax\chi_{\text{max}} eigen value and renormalize iΛii2=1\sum_i\Lambda_ii^2=1

Computation errors :

  • truncation error : main error, grows exponentially
  • Trotter error : can be avoid reducing Δt\Delta t and higher expansion
  • small eigen value : at step 3 Λ1A\Lambda^{-1} A and BΛ1B\Lambda^{-1}
  • imaginary time evolution : canonical form only retrained when Δτ0\Delta\tau\to 0

Further topics

Two-dimensional system

  • converting two dimension system as chain

    2D as chain

  • [PEPS] projected entangled pair state

    pep

    • challenging computationally
    • lack a canonical form

Mixed state and open quantum system dynamics

  • mixed state unitary time evolution is governed by H^\hat H : tρ^(t)=i[H^,ρ^(t)]\partial_t \hat \rho(t)=-i[\hat H,\hat \rho(t)]

  • open quantum system : coupled to an environment or bath

    which can be described by Lindblad equation : tρ^=L^ρ^=i[H^,ρ^]+iγi(L^iρ^L^i12{L^iL^i,ρ^})\partial _t \hat \rho =\hat{\mathcal L}\hat \rho = -i[\hat H,\hat \rho] + \sum_i\gamma_i\left(\hat L_i\hat \rho \hat L_i^\dagger - \frac{1}{2}\{\hat L_i^\dagger\hat L_i,\hat \rho\}\right)

    lindbladian

    Notation

    • L^i\hat L_i : jump operator, the system operators directly coupled to the bath, e.g. creation, annilation
    • L^\hat {\mathcal L} : Lindbladian, could be considered as a linear operator ρ(t)=eiL^tρ0\ket {\rho(t)}=e^{-i\hat {\mathcal L}t}\ket{\rho_0}

Symmetries

schmidt eigenstates belong to a fixed magnetization sector

symmetry schmidt

[TDVP]Time-dependent variational principle

  • action function : S=t1t2ψ(t)itH^ψ(t)dttAi=iHiAiS=\int_{t_1}^{t_2}\bra {\psi(t)}i\partial_t-\hat H\ket {\psi(t)}\text dt \to \partial_tA_i=-iH_iA_i
  • analogue to DMRG algorithm, but better at simulate long-ranged

Quantum Monte Carlo

Monte Carlo Basics

Monte Carlo

  • error 1N\frac{1}{\sqrt N}

Markov Chain : PXY=T(XY)A(XY)A(XY)=min{1,W(Y)W(X)}P_{XY} = T(X\to Y)A(X\to Y)\quad A(X\to Y) = \text{min}\left\{1,\frac{W(Y)}{W(X)}\right\}

  • Ergodicity : T(XY)>0X,YT(X\to Y)>0\quad \forall X,Y
  • Normalization : YT(XY)=1\sum_Y T(X\to Y)=1
  • Reversibility : T(XY)=T(YX)T(X\to Y) = T(Y\to X), if TT not satisfy this, then A(XY)=min{1,W(Y)T(YX)W(X)T(XY)}A(X\to Y) = \text{min}\left\{1,\frac{W(Y)T(Y\to X)}{W(X)T(X\to Y)}\right\}

Notation

  • TT : transition probability
  • WW : static distribution
  • AA : accept probability

Classical Ising model

symmetry-braking phase transition at a finite temperature

H=<i,j>Jijσiσjihσiσi=±1H = -\sum_{<i,j>} J_{ij}\sigma_i\sigma_j - \sum_i h\sigma_i\quad \sigma_i=\pm 1

Notation

  • JijJ_{ij} : coupling constant
    • Jij0J_{ij} \ge 0 : symmetry-broken state
  • hih_i : external field
  • <i,j><i,j> : means i,ji,j are connected
  • cc : cluster, c|c| means the number of spins inside a cluster
  • β\beta : inverse temperature, β=1κBT\beta = \frac{1}{\kappa_B T}
  • mm : magnetization

Algorithm : Swendsen-Wang

  1. two neighboring parallel spins connected with probability p=1e2βJp=1-e^{-2\beta J}
  2. cluster labeling. e.g., Hoshen-Kopelman algorithm
  3. measurement : m2C=1N2cc2\langle m^2\rangle_{C'} = \frac{1}{N^2}\sum_c |c|^2
  4. cluster flipped with probability 12\frac{1}{2}

Algorithm : Wolff

  1. random site
  2. recursive find parallel neighbor add it to the cluster with p=1e2βJp=1-e^{-2\beta J}
  3. measurement : m2C=1Nc0\langle m^2\rangle_{C'}=\frac{1}{N}|c_0|, since only one cluster
  4. flip all spins in the clster
  • Swendsen-Wang will result in many small clusters in high dimension, but Wolff will result in one large cluster

Quantum spin system thermodynamics

m^=1ZTr(m^eβH^)=1ZCm(C)W(C)Z=Tr(eβH^)=CW(C)\langle \hat m \rangle = \frac{1}{Z}\text{Tr}\left(\hat m e^{-\beta \hat H}\right)=\frac{1}{Z}\sum_C m(C)W(C)\quad Z = \text{Tr}\left(e^{-\beta\hat H}\right)=\sum_C W(C)

Notation

  • m^\hat m : magnetization
  • β\beta : reverse of temperature β=1κBT\beta = \frac{1}{\kappa_B T}
  • ZZ : partition sum
  • m(C)m(C) : magnetization of a configuration CC
  • W(C)W(C) : weight of a configuration CC

spin-12\frac{1}{2} in a magnetic field : H^=hS^zΓS^x=[h2Γ2Γ2h2]\hat H = -h \hat S^z -\Gamma \hat S^x=\begin{bmatrix}-\frac{h}{2}&-\frac{\Gamma}{2}\\-\frac{\Gamma}{2}&\frac{h}{2}\end{bmatrix}

Notation

  • hh : longitudinal field
  • Γ\Gamma : transverse field

Discrete-time path integral : β=ΔτM\beta = \Delta\tau M

expand to first order eΔτH^=U^+O(Δτ2)ZTr(U^M)e^{-\Delta \tau \hat H} = \hat U +\mathcal O(\Delta\tau^2)\to Z \approx \text{Tr}\left(\hat U^M\right)

Notation

  • U^\hat U : transfer matrix : U^=IΔτH^=[1+Δτh2ΔτΓ2ΔτΓ21Δτh2]\hat U = I - \Delta \tau \hat H = \begin{bmatrix}1+\frac{\Delta\tau h}{2}&\frac{\Delta\tau\Gamma}{2}\\\frac{\Delta\tau \Gamma}{2}&1-\frac{\Delta \tau h}{2}\end{bmatrix}
  • Δτ\Delta \tau : discrete time step
  • MM : resolution
  • E0E_0 : ground energy

Example : 1D classical Ising model (0D transverse field Ising model)

H=JiMσiσi+1hiσiH = -J\sum_i^M\sigma_i\sigma_{i+1}-h\sum_i \sigma_i with periodic boundary condition σM+1=σ1\sigma_{M+1}=\sigma_1

  • βJ=12log(ΔτΓ/2)\beta J = -\frac{1}{2}\text{log}(\Delta \tau \Gamma/2) : off diagonal
  • βh=log(1+Δτh/2)\beta h = \text{log}(1+\Delta \tau h/2) : diagonal
  • βE0=MβJ\beta E_0 = M\beta J

Continous-time path integral: Δτ0\Delta\tau \to 0 ???

dd -dimensional quantum spin model \Leftrightarrow d+1d+1- dimensional classical Ising model

Example : 1D classical Ising model (0D transverse field Ising model)

quantum XYXY model : H^=<i,j>Jxy2(S^i+S^j+S^iS^j+)\hat H = -\sum_{<i,j>}\frac{J_{xy}}{2}(\hat S_i^+\hat S_j^- + \hat S_i^-\hat S_j^+)

spin flip-flops (blue line) proportional to β\beta which is a constant not grow bigger as Δτ0\Delta \tau \to 0

spin conservation

negative sign problem : positive off diagonal lead to negative probabilities

  • solution : A^W=CA(C)W(C)CW(C)=CA(C)sign(W)W(C)/CW(C)Csign(W)W(C)/CW(C)\langle \hat A\rangle_W = \frac{\sum_C A(C)W(C)}{\sum_C W(C)} = \frac{\sum_C A(C)\text{sign}(W)|W(C)|/\sum_C |W(C)|}{\sum_C \text{sign}(W)|W(C)|/\sum_C|W(C)|}
  • error : β,Lϵ\beta\uparrow ,L\uparrow\to \epsilon\uparrow error ϵ\epsilon increase with inverse temperature β\beta and system size LL

Variational Monte Carlo

variational principle : ψ(θ)=nψn(θ)n\ket {\psi(\theta)} = \sum_n \psi_n(\theta)\ket n

energy expectation(MCMC) : Eθ=nψn(θ)2E1(n)nψn(θ)2E_\theta = \frac{\sum_n |\psi_n(\theta)|^2E_1(n)}{\sum_n |\psi_n(\theta)|^2}

Notation

  • E1(n)E_1(n) : local energy E1(n)=mnH^mψm(θ)ψn(θ)E_1(n)=\sum_m \bra n\hat H\ket m\psi_m(\theta)\psi_n(\theta)
  • GklG_{kl} : metric tensor Gkl=O^kO^lθO^kθO^lθG_{kl} = \langle\hat O_k^*\hat O_l\rangle_\theta - \langle \hat O_k^*\rangle_\theta\langle\hat O_l\rangle_\theta
  • OO : logarithm wave-function derivative O=θψn(θ)/ψn(θ)O = \nabla_\theta \psi_n(\theta)/\psi_n(\theta)
    • O^=nO(n)nn\hat O = \sum_n O(n)\ket n \bra n

[SGD]Stochastic Graident Descent : θθλθEθ\theta\gets \theta - \lambda \nabla_\theta E_\theta

  • θEθ=2Re{nW(n)[E1(n)Eθ]O(n)}\nabla_\theta\langle E\rangle_\theta = 2\text{Re}\left\{\sum_n W(n)[E_1(n)-E_\theta]O(n)\right\}

Stochastic Reconfiguration : θθΔτG1θEθ\theta\gets \theta - \Delta \tau G^{-1}\nabla_\theta E_\theta

  • to avoid small value inverse : G=β2I+GGβRG' = \sqrt{\beta^2I + G^\dagger G}\quad \beta\in \R

Jastrow States : ψn(θ)=exp(iaiσi+<ij>Jijσiσj)θ={a,J}\psi_n(\theta ) = \text{exp}\left(\sum_i a_i\sigma_i + \sum_{<ij>}J_{ij}\sigma_i\sigma_j \right)\quad \theta= \{a,J\}

​ wave function form for spin system

[NQS]Neural Quantum States : ψn(θ)=MLP({σ1,,σN})\psi_n(\theta)=\text{MLP}(\{\sigma_1,\cdots,\sigma_N\})

[MFPWF]Mean-field projected wave function : ψ(θ)=PG[i,js,sFijssc^i,sc^j,s]N/20θ=FijssR2N×2N\ket {\psi(\theta)} = \mathcal P_G \left[\sum_{i,j}\sum_{s,s'}F_{ij}^{ss'}\hat c_{i,s}^\dagger\hat c_{j,s'}^\dagger\right]^{N/2}\ket 0\quad \theta = F_{ij}^{ss'}\in\R^{2N\times 2N}

  • ψn(θ)=(N/2)!Pf(X)\psi_n(\theta) = (N/2)!\text{Pf}(X)

    represent spin as pesudo-fermions : S^i{x,y,z}=12ssc^i,sσssαc^i,s\hat S_i^{\{x,y,z\}} = \frac{1}{2}\sum_{ss'}\hat c_{i,s}\sigma_{ss'}^\alpha\hat c_{i,s'}

Notation

  • PG\mathcal P_G : Gutzwilller projection operator
  • c^i,s,c^i,s\hat c_{i,s},\hat c_{i,s}^\dagger : fermionic annihlation/creation operator
  • s,ss,s' : spin of the site, \uparrow or \downarrow
  • i,ji,j : index of the site

Path integrals in quantum statistical mechanics

ρfree(R,R,Δτ)=ReΔτT^R=(2π2Δτm)Nd/2exp(RR222Δτ/m)\rho_{\text{free}}(\vec R,\vec R',\Delta\tau) = \bra {\vec R} e^{-\Delta\tau\hat T}\ket{\vec R'}=\left(\frac{2\pi\hbar^2 \Delta\tau}{m}\right)^{-Nd/2}\text{exp}\left({-\frac{|\vec R-\vec R'|^2}{2\hbar^2\Delta\tau/m}}\right)

Z=dRρ(R,R)=(j=1MdRj)j=1M[(2π2Δτm)Nd/2exp(RjRj+122Δτ/mΔτV(Rj))]Z = \int \text d\vec R \rho(\vec R,\vec R)=\int \left(\prod_{j=1}^M \text d\vec R_j\right)\prod_{j=1}^M \left[\left(\frac{2\pi\hbar^2 \Delta \tau}{m}\right)^{-Nd/2}\text{exp}\left(-\frac{\vec R_j-\vec R_{j+1}}{2\hbar^2\Delta\tau/m}-\Delta\tau V(\vec R_j)\right)\right]

Notation

  • ρfree\rho_{\text{free}} : density matrix of free particles
  • ZZ : partition function
  • Rj\vec R_j : (r1,r2,,rN)(\vec r_1,\vec r_2,\cdots,\vec r_N) , NN particles position at time jj
  • T^,V^\hat T,\hat V : kinetic, potential terms of Hamiltonian H^\hat H, T^=22mx2\hat T = -\frac{\hbar^2}{2m}\partial_x^2

path sampling method : A(XX)=min{1,exp(m[(rj1irji)2+(rjirj+1i)2]/22Δτ)exp(m[(rj1irji)2+(rjirj+1i)2]/22Δτ)exp(Δτ[V(Rj)V(Rj)])}A(X\to X' ) = \text{min}\left\{1, \frac{\text{exp}(-m[(\vec r_{j-1}^i - \vec r_j^{i'})^2+(\vec r_j^{i'}-\vec r^i_{j+1})^2]/2\hbar^2\Delta\tau)}{\text{exp}(-m[(\vec r_{j-1}^i-\vec r_j^i)^2+(\vec r_j^i - \vec r_{j+1}^i)^2]/2\hbar^2\Delta \tau)}\cdot \text{exp}(-\Delta \tau[V(\vec R'_j)-V(\vec R_j)])\right\}

The accept probability of Metropolis algorithm is defined above

H=jim2(Δτ)2(rjirj+1i)2+jV(Rj)H = \sum_j\sum_i \frac{m}{2(\hbar\Delta \tau)^2}(\vec r_j^i-\vec r_{j+1}^i)^2+\sum_j V(\vec R_j)

Notation

  • rji\vec r_j^i : position of particle ii at time jj
  • rji\vec r_j^{i'} : displaced position of particle ii at time jj
  • Rj\vec R_j : (r1,r2,,rN)(\vec r_1,\vec r_2,\cdots,\vec r_N) , NN particles position at time jj
  • VV : potential energy, in most cases it’s sum of single-particle and two-particle terms : V^=iNvext(r^i)+i<jv(r^ir^j)\hat V = \sum_i^N v_{\text{ext}}(\hat{\vec r}^i) + \sum_{i<j}v(\hat{\vec r}^i -\hat{\vec r}^j)

Boson symmetry :

ρBose=1N!Pρ(R1,PR2,β)\rho_{\text{Bose}} = \frac{1}{N!}\sum_P \rho(\vec R_1,P\vec R_2, \beta)

[DMC] Diffusion Monte Carlo

diffusion monte carlo

Algorithm

  1. w0α1,R0αR0w_0^\alpha \gets 1,\vec R_0^\alpha \gets \vec R_0
  2. update loop
    1. RkαN(Rk1α,Δτm)\vec R^\alpha_k \sim \mathcal N(\vec R^\alpha_{k-1},\frac{\Delta \tau}{m}) : diffusion update
    2. wkαwk1αeΔτ2[V(Rkα)+V(Rk1α)]w_k^\alpha\gets w_{k-1}^\alpha e^{-\frac{\Delta \tau}{2}[V(\vec R_k^\alpha)+V(\vec R^\alpha_{k-1})]}
    3. clone wkαEα[wkα]+r\lfloor \frac{w_k^\alpha}{\mathbb E_\alpha [w_k^\alpha]} + r\rfloor times for walker α\alpha
  • maximum clones \Leftrightarrow Δτ\Delta \tau too large
  • scale wαexp(EtΔτ)wαw^\alpha\to\text{exp}(E_t\Delta \tau)w^\alpha where EtE_t is trial energy V(R)V(R)EtV(\vec R)\gets V(\vec R)-E_t, when Et=E0E_t=E_0 stability will achieve.

Importance sampling : Rk1Rk1+2Δτ2m2ϕt(Rk1)ϕ(Rk1)\vec R_{k-1} \gets \vec R_{k-1}+\frac{\hbar^2\Delta\tau}{2m}\frac{2\nabla\phi_t(\vec R_{k-1})}{\phi(\vec R_{k-1})}

before update, add a dift : Rk1Rk1+2Δτ2m2ϕt(Rk1)ϕ(Rk1)\vec R_{k-1} \gets \vec R_{k-1}+\frac{\hbar^2\Delta\tau}{2m}\frac{2\nabla\phi_t(\vec R_{k-1})}{\phi(\vec R_{k-1})}

Notation

  • ϕt\phi_t : trial wavefunction ϕt(R)=i<jfz(rirj)\phi_t(\vec R)= \prod_{i<j}f_z(|\vec r_i -\vec r_j|)
  • fzf_z : Jastrow factor, two particle coorelations

Fermionic systems : ϕt(R)=ϕt(R)detl,n[eiklrn]\phi_{t'}(\vec R)=\phi_t(\vec R)\underset{l,n}{\text{det}}[e^{i\vec k_l\cdot \vec r_n}]

ϕ0\phi_0 could be negative, when ϕϕ\phi\to-\phi should be applied

Notation

  • nn : particle index
  • kl\vec k_l : wave vectors compatible with periodic boundary conditions

Electronic-structure problem

Full Hamiltonian of matter

H^=jNe22mrj2T^elNn22MlRl2T^n+12ijNee2rirjVee+12lmNnZlZme2RlRmVnnj=1Nel=1NnZle2rjRlVen+VSO\hat H = -\underbrace{\sum_j^{N_e}\frac{\hbar^2}{2m}\nabla^2_{\vec r_j}}_{\hat T_e}- \underbrace{\sum_l^{N_n}\frac{\hbar^2}{2M_l}\nabla^2_{\vec R_l}}_{\hat T_n} + \underbrace{\frac{1}{2}\sum_{i\neq j}^{N_e}\frac{e^2}{|\vec r_i -\vec r_j|}}_{V_{ee}}+\underbrace{\frac{1}{2}\sum_{l\neq m}^{N_n}\frac{Z_lZ_me^2}{|\vec R_l-\vec R_m|}}_{V_{nn}}-\underbrace{\sum_{j=1}^{N_e}\sum_{l=1}^{N_n}\frac{Z_le^2}{|\vec r_j -\vec R_l|}}_{V_{en}}+V_{SO}

Adiabatic (Born-Oppenheimer) approximation : MlmRlRmrirjM_l \gg m\quad |\vec R_l-\vec R_m| \ll |\vec r_i - \vec r_j|

H^=j=1Ne22mrj2T^ej=1Nej=1NnZle2rjRlVen+12ijNee2rirjVee\hat H = \underbrace{\sum_{j=1}^{N_e}\frac{\hbar^2}{2m}\nabla^2_{\vec r_j}}_{\hat T_e} - \underbrace{\sum_{j=1}^{N_e}\sum_{j=1}^{N_n}\frac{Z_le^2}{|\vec r_j - \vec R_l|}}_{V_{en}}+\underbrace{\frac{1}{2}\sum_{i\neq j}^{N_e}\frac{e^2}{|\vec r_i-\vec r_j|}}_{V_{ee}}

Non-interacing (mean-field) approximation : H^sp=22mr2+Veff(r)\hat H_{\text{sp}} = -\frac{\hbar^2}{2m}\nabla^2_{\vec r}+V_{\text{eff}}(\vec r)

non-interacting electrons assumption, Ven+VM+VeeVeffV_{en}+V_M + V_{ee}\to V_{\text{eff}}

Hatree-Fock approximation :

use a Slater determinant for non-interacting system

ΦH^Φ=i,σd3r ϕiσ(r)[22m2+Ven(r)]σiσ(r)+VMT^e+Ven+i,j,σ,σe2d3r d3r ϕiσ(r) ϕjσ(r)1rrϕiσ(r)ϕjσ(r)Hatree interaction}Vee\begin{aligned} \bra \Phi \hat H\ket\Phi &= \underbrace{\sum_{i,\sigma}\int \text d^3 \vec r~\phi_i^{\sigma*}(\vec r)[-\frac{\hbar^2}{2m}\nabla^2 + V_{en}(\vec r)]\sigma_i^\sigma(\vec r) + V_M}_{\hat T_e+ V_{en}} \\ &\begin{drcases} + \sum_{i,j,\sigma,\sigma'}e^2\int \text d^3\vec r~\text d^3\vec r'~\phi_i^{\sigma*}(\vec r)~\phi_j^{\sigma'*}(\vec r')\frac{1}{|\vec r-\vec r'|}\phi_i^\sigma(\vec r)\phi_j^{\sigma'}(\vec r')&\text{Hatree interaction} \\ % -\sum_{i,j,\sigma}e^2\int \text d^3\vec r~\text d^3\vec r'~\phi_i^{\sigma*}(\vec r)~\phi_j^{\sigma*}(\vec r)\frac{1}{|\vec r-\vec r'|}\phi_j^\sigma(\vec r)\phi_i^\sigma(\vec r')&\text{exchange interaction} \end{drcases} V_{ee} \end{aligned}

The integration shows that the wave function ψ\psi is defined in the whole space

Notation

  • i,ji,j : index of the single particle state
  • σ,σ\sigma,\sigma' : spin of the electron, \uparrow or \downarrow
  • ϕiσ(r)\phi_i^\sigma(\vec r) : dnotes for particle ii of state σ\sigma , the wave function value at position r\vec r

Configuration-Interaction : Φ0=(1+i,μαμic^ic^μ+i<j,μ<ναμ,νijc^ic^jc^μc^ν)ΦHF\ket {\Phi_0} = \left(1+\sum_{i,\mu}\alpha_\mu^i\hat c^\dagger_i\hat c_\mu +\sum_{i<j,\mu<\nu}\alpha_{\mu,\nu}^{ij}\hat c^\dagger_i\hat c^\dagger_j \hat c_\mu \hat c_\nu\right)\ket {\Phi_{\text{HF}}}

add interations between electrons correctly and allow calculation of excited state

Notation

  • ΦHF\ket {\Phi_{\text{HF}}} : Hartree-Fock ground state, which is from the Hartree-Fock approximation, ΦHF=μ=1Nc^μnulll\ket {\Phi_{\text{HF}}}=\prod_{\mu=1}^N \hat c^\dagger_\mu \ket {\text{nulll}}
  • c^,c^\hat c^\dagger, \hat c : creation / annihilation operator

[DFT] Density functional theory

Hohenberg-Kohn Theorem : for electron system H^=22mjj2T^e+12ije2rirjV^ee+vext(r)n(r)drV^ext\hat H = \underbrace{-\frac{\hbar^2}{2m}\sum_j\nabla_j^2}_{\hat T_e}+\underbrace{\frac{1}{2}\sum_{i\neq j}\frac{e^2}{|\vec r_i -\vec r_j|}}_{\hat V_{ee}}+\underbrace{\int v_{\text{ext}}(\vec r)n(\vec r)\text d\vec r}_{\hat V_{\text{ext}}}

  • Uniqueness : n0(r)vext(r)n_0(\vec r) \Leftrightarrow v_{\text{ext}}(\vec r)
  • Variational : n0=argminn E=argminnΨH^Ψn_0 = \underset{n}{\text{argmin}}~E = \underset{n}{\text{argmin}}\bra\Psi \hat H\ket \Psi

Notation

  • vext(r)v_{\text{ext}}(\vec r) : external potential density
  • n0(r)n_0(\vec r) : ground state electron density
  • n(r)n(\vec r) : electron density jϕj(r)2\sum_j|\phi_j(\vec r)|^2

Kohn-Sham solution :

find a non-interacting system that has the same particle density as the interacting one

Algorithm

  1. initial guess Veff0V^0_{\text{eff}}
  2. solve ϕj\phi_j (eigen vector) from KS1 : [22me2+Veff(r)]ϕj(r)=εjϕj(r)\left[-\frac{\hbar^2}{2m_e}\nabla^2+V_{\text{eff}}(\vec r)\right]\phi_j(\vec r) = \varepsilon_j \phi_j(\vec r)
  3. n(r)iϕj(r)2n(\vec r)\gets \sum_i |\phi_j(\vec r)|^2
  4. revise VeffV_{\text{eff}} from KS 2 : Veff(r)=d3rn(r)rr+μXC(r)+νext(r)V_{\text{eff}}(\vec r) = \int \text d^3 r'\frac{n(\vec r')}{|\vec r-\vec r'|}+\mu^{\text{XC}}(\vec r)+\nu_{\text{ext}}(\vec r)
  5. goto 2 if VeffnewVeffoddthreshold|V_{\text{eff}}^{\text{new}}-V_{\text{eff}}^{\text{odd}}|\ge \text{threshold}

Notation

  • μXC\mu^{\text{XC}} : functional derivative of the exchange-correlation energy, μXC=dEXCdn\mu^{\text{XC}} = \frac{\text d\text E^{\text{XC}}}{\text d n}
  • EXCE^{\text{XC}} : exchange-correlation energy, EXC=ΦT^ΦEk+ΦV^eeΦEcE^{\text{XC}}=\bra \Phi \hat T\ket \Phi - E_k +\bra \Phi \hat V_{ee}\ket \Phi -E_c with approximation (local density approximation) EXCn(r)εXC(n(r))dr=n(r)[εX(n(r))+εC(n(r))]drE^{\text{XC}}\approx \int n(\vec r) \varepsilon^{\text{XC}}(n(\vec r))\text d\vec r=\int n(\vec r)[\varepsilon^X(n(\vec r))+\varepsilon^C(n(\vec r))]\text d\vec r
    • uniform electron gas : ε(n(r))=34(3π)1/3n(r)1/3\varepsilon(n(\vec r)) = -\frac{3}{4}\left(\frac{3}{\pi}\right)^{1/3}n(\vec r)^{1/3}
    • Monte carlo -> interpolation
  • ECE_C : Hartree energy EC=12e2n(r)n(r)rrdrdrE_C= \frac{1}{2}\int e^2 \frac{n(\vec r)n(\vec r')}{|\vec r-\vec r'|}\text d\vec r\text d \vec r'
  • EKE_K : kinetic energy EK=22mjϕj2ϕjE_K = -\frac{\hbar^2}{2m}\sum_j \bra {\phi_j}\nabla^2\ket{\phi_j}

Basis functions

Atoms and molecules

  • [STO] Slater Type Orbitals : ψnlmi(r,θ,ϕ)rn1eξirYlm(θ,ϕ)\psi^i_{nlm}(r,\theta,\phi)\propto r^{n-1}e^{-\xi_ir}Y_{lm}(\theta,\phi)
    • two nuclei no closed form
  • [GTO] Gauss Type Orbitals : ψnlmi(r)xlymzneξir2\psi_{nlm}^i(\vec r) \propto x^ly^mz^n e^{-\xi_ir^2}
    • gaussian product still gaussian easy to integral

Free electron gas

H^=i=1Ne22mri2T^+e2i<j1rrVee\hat H = - \underbrace{\sum_{i=1}^{N_e}\frac{\hbar^2}{2m}\nabla^2_{\vec r_i}}_{\hat T}+\underbrace{e^2\sum_{i<j}\frac{1}{|\vec r-\vec r'|}}_{V_{ee}}

  • plane waves basis : ψk(r)=exp(ikr)\psi_{\vec k}(\vec r) = \text{exp}(-i\vec k\cdot \vec r)
  • low temperature : Wigner crystal
    • better basis will be eigenfunctions of harmonic oscillatiors centered around

Pseudo-potentials

  • only model outer shell with basis, use pseudo potential to model inner shell since they are not involved in chemical bond

Quantum Computing

Quantum Computer

quantum gates

  • one-qubit gate

    XX [0110]\begin{bmatrix}0&1\\1&0\end{bmatrix} YY [0ii0]\begin{bmatrix}0&-i\\i&0\end{bmatrix} ZZ [1001]\begin{bmatrix}1&0\\0&-1\end{bmatrix}
    name matrix name matrix name matrix
    H=X+Z2H=\frac{X+Z}{\sqrt 2} 12[1111]\frac{1}{\sqrt 2}\begin{bmatrix}1 & 1\\1&-1\end{bmatrix} TT [100eiπ/4]\begin{bmatrix}1&0\\0&e^{i\pi/4}\end{bmatrix} S=T2S=T^2 [100i]\begin{bmatrix}1&0\\0&i\end{bmatrix}
  • two-quitbit gate C(U)=[100100000000U]C(U)=\begin{bmatrix}\begin{matrix}1&0\\0&1\end{matrix}&\begin{matrix}0&0\\0&0\end{matrix}\\\begin{matrix}0&0\\0&0\end{matrix}&U\end{bmatrix}

measurement : z1z2ψ2|\bra{z_1z_2\dots}\ket{\psi}|^2

  • repeated O(1000)\mathcal O(1000) times

errors

  • coupling to environment \Rightarrow mixed density matrix
  • gate error
  • read out measurement error

Representing the Hilbert space

Spin-12\frac{1}{2} system : directly mapped to a qubit

Fermionic system

  • Jordan-Wigner Transformation : Ψ=nN1,,n0zN1,,z0ni=zi\ket \Psi = \ket {n_{N-1},\cdots,n_0}\leftrightarrow \ket {z_{N-1},\cdots,z_0}\quad n_i=z_i

    c^iAiZi1Z0c^iAiZi1Z0Ai=(Xi+iYi)2\hat c_i \leftrightarrow A_iZ_{i-1}\cdots Z_0\quad \hat c^\dagger_i \leftrightarrow A_i^\dagger Z_{i-1}\cdots Z_0\quad A_i = \frac{(X_i+iY_i)}{2}

    • measuring parity requires O(N)\mathcal O(N) operators
    • updating an occupation number requires O(1)\mathcal O(1) operators
  • Parity Encoding : Ψ=nN1,,n0zN1,,z0zi=[j=0ini]mod 2\ket \Psi = \ket {n_{N-1},\cdots,n_0}\leftrightarrow \ket {z_{N-1},\cdots,z_0}\quad z_i = \left[\sum_{j=0}^in_i\right]\text{mod}~2

    c^iXN1Xi+1(XiZi1+iYi)c^iXN1Xi+1(XiZi1iYi)\hat c_i \leftrightarrow X_{N-1}\cdots X_{i+1}(X_iZ_{i-1}+iY_i)\quad \hat c^\dagger _i \leftrightarrow X_{N-1}\cdots X_{i+1}(X_iZ_{i-1}-iY_i)

    • measuring parity requires O(1)\mathcal O(1) operators
    • updating an occupation requires O(N)\mathcal O(N) operators
  • Bravyi-Kitaev : a hypbrid of Parity and Jordan-Wigner

Notation

  • nin_i : fermionic orbitals/site occupation number ni{0,1}n_i\in \{0,1\}
  • c^i,c^i\hat c_i, \hat c_i^\dagger : creation operator, annihilation operator

Variational quantum solver

extract the spectrum of an operator

[QFT] Quantum fourier transform

  • exact solution
  • vast number of gate operations

[VQE] Variational Quantum Eigensolver : minθΨ(θ)H^Ψ(θ)Ψ(θ)Ψ(θ)\underset{\theta}{\text{min}}\frac{\bra{\Psi(\theta)}\hat H\ket{\Psi(\theta)}}{\bra{\Psi(\theta)}\ket{\Psi(\theta)}}

  • quantum computer : expectation evaluation
  • classical computer : optimization COBYLA (no SGD since no gradient)

[UCC] Unitary Coupled Cluster : Ψ(θ)=eT^(θ)T^(θ)Ψ0\ket {\Psi(\theta)} = e^{\hat T(\theta)-\hat T^\dagger(\theta)}\ket{\Psi_0}

  • a good choice for variational state
  • huge circuit depth
  • much depend on the choice of state Φ0\ket{\Phi_0}

[UCCSD] Unitary Coupled Cluster with Single and Double excitation

  • T^(θ)T^1(θ1)+T^2(θ2)\hat T(\theta) \approx \hat T_1(\theta_1)+\hat T_2(\theta_2)
    • T^1(θ1)=i,jθ1,i,jc^ic^j\hat T_1(\theta_1)=\sum_{i,j}\theta_{1,i,j}\hat c^\dagger_i\hat c_j
    • T^2(θ2)=i,jθ2,i,j,k,lc^ic^kc^jc^l\hat T_2(\theta_2) = \sum_{i,j}\theta_{2,i,j,k,l}\hat c^\dagger_i\hat c^\dagger_k\hat c_j\hat c_l

Notation

  • T^(θ)\hat T(\theta) : excitation operator
  • Ψ0\ket {\Psi_0} : Hartree-Fock/single Slater detereminant state

Computational Quantum Physics
https://walkerchi.github.io/2023/08/28/ETHz/ETHz-CQP/
Author
walkerchi
Posted on
August 28, 2023
Licensed under