Skip to content

Why does a single flipped spin behave like a wave? In the periodic XXX chain, the Hamiltonian moves that down spin to neighboring sites. You will diagonalize this discrete hopping equation and obtain both its dispersion E(k)=J(1−cos⁡k)E(k)=J(1-\cos k) and its allowed wave numbers k=2πm/Nk=2\pi m/N. In this one-magnon sector, a finite Fourier basis supplies all NN states.

Required background. Use the bond exchange rule and fixed-magnetization basis from Construct a small spin-chain Hamiltonian. The entry repair reviews the complex exponentials used below.

Helpful background. The spin-chain convention reference explains why a translation eigenvalue depends on the direction chosen for the translation operator.

Use N≥3N\geq3, ℏ=1\hbar=1, lattice spacing one, and the model Hamiltonian

H=J∑n=1N(14−Sn⋅Sn+1),J>0,SN+1=S1.H=J\sum_{n=1}^N\left(\frac14-\mathbf S_n\cdot\mathbf S_{n+1}\right), \qquad J\gt0,\qquad \mathbf S_{N+1}=\mathbf S_1.

The all-up state ∣F⟩|F\rangle has energy zero. The orthonormal basis of the M=1M=1 sector is

∣n⟩=Sn−∣F⟩,n=1,…,N.|n\rangle=S_n^-|F\rangle,\qquad n=1,\ldots,N.

Only two bonds touch the down spin. Each contributes J/2J/2 on the diagonal and −J/2-J/2 for moving it one site:

H∣n⟩=J∣n⟩−J2(∣n−1⟩+∣n+1⟩),H|n\rangle =J|n\rangle-\frac J2\bigl(|n-1\rangle+|n+1\rangle\bigr),

with site labels read modulo NN. For ∣ψ⟩=∑nψ(n)∣n⟩|\psi\rangle=\sum_n\psi(n)|n\rangle, the coefficient of ∣n⟩|n\rangle in H∣ψ⟩=E∣ψ⟩H|\psi\rangle=E|\psi\rangle gives

Eψ(n)=Jψ(n)−J2(ψ(n−1)+ψ(n+1)).E\psi(n)=J\psi(n)-\frac J2\bigl(\psi(n-1)+\psi(n+1)\bigr).

This is a discrete second-difference operator. It conserves the number of down spins; a magnon is a delocalized excitation within this sector, not an extra site or an extra spin added to the chain.

For ψ(n)=eikn\psi(n)=e^{ikn}, find ψ(n+1)/ψ(n)\psi(n+1)/\psi(n) and ψ(n−1)/ψ(n)\psi(n-1)/\psi(n). When is ψ(n+N)=ψ(n)\psi(n+N)=\psi(n)?

Repair. Exponential addition gives the ratios eike^{ik} and e−ike^{-ik}. Euler’s identity implies eik+e−ik=2cos⁡ke^{ik}+e^{-ik}=2\cos k. Since the amplitude is never zero, periodicity requires eikN=1e^{ikN}=1, or kN=2πmkN=2\pi m for an integer mm. Wave numbers differing by 2π2\pi give identical amplitudes at integer sites.

Insert the plane wave in the lattice equation and divide by eikne^{ikn}:

E=J−J2(e−ik+eik)=J(1−cos⁡k).E=J-\frac J2(e^{-ik}+e^{ik})=J(1-\cos k).

The bulk equation determines energy as a function of kk. The boundary condition then selects

km=2πmN,m=0,1,…,N−1.k_m=\frac{2\pi m}{N},\qquad m=0,1,\ldots,N-1.

These are different steps. A plane wave with arbitrary real kk satisfies the bulk difference equation, but it generally fails at the closing bond of a finite ring.

The normalized eigenstates are

∣km⟩=1N∑n=1Neikmn∣n⟩.|k_m\rangle=\frac1{\sqrt N}\sum_{n=1}^N e^{ik_mn}|n\rangle.

For m≠m′m\ne m' modulo NN, the finite geometric sum gives

⟨km′∣km⟩=1N∑n=1Ne2πi(m−m′)n/N=0.\langle k_{m'}|k_m\rangle =\frac1N\sum_{n=1}^N e^{2\pi i(m-m')n/N}=0.

Each vector has norm one, and there are NN orthogonal vectors in an NN-dimensional sector. This proves completeness here. It does not prove completeness of a two-magnon or many-magnon Bethe construction.

The lattice equation and dispersion agree with Karbach and Müller 1997, p. 2, equations (3)–(8), arXiv v1 PDF, after replacing their E−E0E-E_0 by our vacuum-shifted EE.

For N=6N=6, the spectrum is

mmkmk_mE/JE/J
000000
1,51,5π/3, 5π/3\pi/3,\ 5\pi/31/21/2
2,42,42π/3, 4π/32\pi/3,\ 4\pi/33/23/2
33π\pi22

At k=πk=\pi, neighboring amplitudes have opposite signs. Consequently each off-diagonal term adds to the positive diagonal contribution:

J(−1)n−J2((−1)n−1+(−1)n+1)=2J(−1)n.J(-1)^n-\frac J2\bigl((-1)^{n-1}+(-1)^{n+1}\bigr) =2J(-1)^n.

At k=0k=0, all amplitudes are equal and the terms cancel. This zero mode is

∣0⟩wave=1NStot−∣F⟩,Stot−=∑nSn−.|0\rangle_{\rm wave}=\frac1{\sqrt N}S_{\rm tot}^-|F\rangle, \qquad S_{\rm tot}^-=\sum_nS_n^-.

The subscript distinguishes this wave state from a site label. It belongs to the same total-spin multiplet as the vacuum. Zero excitation energy does not make it the zero vector: its norm is one, and it lies in a different magnetization sector from ∣F⟩|F\rangle.

Define the active one-site translation by T∣n⟩=∣n+1⟩T|n\rangle=|n+1\rangle. Relabeling the sum gives

T∣k⟩=1N∑neik(n−1)∣n⟩=e−ik∣k⟩.T|k\rangle =\frac1{\sqrt N}\sum_n e^{ik(n-1)}|n\rangle =e^{-ik}|k\rangle.

Our coefficient convention is e+ikne^{+ikn}, so this active translation has eigenvalue e−ike^{-ik}. The inverse translation has eigenvalue e+ike^{+ik}. An energy calculation cannot detect a mistaken choice here because E(k)=E(−k)E(k)=E(-k); comparing translation eigenvalues can.

The wave numbers kk and −k-k are generally distinct states with equal energies. They coincide modulo 2π2\pi only for k=0k=0 and, when allowed, k=πk=\pi.

On N=5N=5 sites, take k=4π/5k=4\pi/5. Write the normalized amplitudes, find the exact energy, and check the equation at site 11, including the bond to site 55.

Hint

Use z=e4πi/5z=e^{4\pi i/5}, so z5=1z^5=1. In the site-11 equation, the amplitude at site 55 equals the continued amplitude at site 00.

Solution

The amplitudes are ψ(n)=zn/5\psi(n)=z^n/\sqrt5. Their norm is one because all five squared magnitudes equal 1/51/5. Since cos⁡(4π/5)=−(1+5)/4\cos(4\pi/5)=-(1+\sqrt5)/4,

E=5+54 J.E=\frac{5+\sqrt5}{4}\,J.

At site 11,

Jψ(1)−J2(ψ(5)+ψ(2))=J5(z−1+z22)=J(1−cos⁡k)z5.J\psi(1)-\frac J2\bigl(\psi(5)+\psi(2)\bigr) =\frac J{\sqrt5}\left(z-\frac{1+z^2}{2}\right) =J(1-\cos k)\frac z{\sqrt5}.

The equality ψ(5)=1/5\psi(5)=1/\sqrt5 is precisely where periodic quantization enters the endpoint check.

For N=6N=6, combine the k=π/3k=\pi/3 and −k-k states into normalized real cosine and sine states. Show that both have energy J/2J/2. Are they individually eigenstates of the active translation TT?

Hint

Take the sum and difference of the two orthonormal Fourier states. Translate amplitudes by replacing nn with n−1n-1.

Solution

Define

∣c⟩=∣k⟩+∣−k⟩2=26∑ncos⁡(kn)∣n⟩,∣s⟩=∣k⟩−∣−k⟩i2=26∑nsin⁡(kn)∣n⟩.\begin{aligned} |c\rangle&=\frac{|k\rangle+|-k\rangle}{\sqrt2} =\sqrt{\frac26}\sum_n\cos(kn)|n\rangle,\\ |s\rangle&=\frac{|k\rangle-|-k\rangle}{i\sqrt2} =\sqrt{\frac26}\sum_n\sin(kn)|n\rangle. \end{aligned}

The two states are normalized and orthogonal because this is a unitary change of basis within the Fourier pair. Both retain the common energy J/2J/2. Translation mixes them:

T∣c⟩=cos⁡k ∣c⟩+sin⁡k ∣s⟩,T∣s⟩=−sin⁡k ∣c⟩+cos⁡k ∣s⟩.\begin{aligned} T|c\rangle&=\cos k\,|c\rangle+\sin k\,|s\rangle,\\ T|s\rangle&=-\sin k\,|c\rangle+\cos k\,|s\rangle. \end{aligned}

Here sin⁡k≠0\sin k\ne0, so neither real state is a translation eigenstate. An arbitrary eigensolver can return this real basis of a degenerate energy eigenspace. To assign a definite wave number, diagonalize translation within that eigenspace.

Change the Hamiltonian to

Hh=H+h∑n(12−Snz),H_h=H+h\sum_n\left(\frac12-S_n^z\right),

where hh is a real parameter with units of energy. Determine the one-magnon eigenvectors, energies, and allowed wave numbers. For h>0h\gt0, does the k=0k=0 state still have zero energy?

Hint

The added operator counts down spins. Evaluate it throughout the fixed-MM sector before attempting a new Fourier calculation.

Solution

The added term is hMhM. In the M=1M=1 sector it is simply hh times the identity, so the Fourier eigenvectors and the condition k=2πm/Nk=2\pi m/N are unchanged:

Eh(k)=J(1−cos⁡k)+h.E_h(k)=J(1-\cos k)+h.

The vacuum still has energy zero, while the k=0k=0 one-magnon state has energy hh. For positive hh, its degeneracy with the vacuum is lifted. More generally, every fixed-MM block is shifted by hMhM; no claim about which sector is the ground state for arbitrary negative hh follows from the one-magnon calculation alone.

You have an exact and complete finite Fourier solution in the one-magnon sector. The quadratic small-kk behavior E(k)=Jk2/2+O(k4)E(k)=Jk^2/2+O(k^4) follows by Taylor expansion; on a finite ring, kk still takes discrete values. With two down spins, their adjacency changes the local equation. Continue to two-magnon scattering to find the extra matching condition.

  • Karbach, Michael, and Gerhard Müller. “Introduction to the Bethe Ansatz I.” Computers in Physics 11, 36–43 (1997). DOI. Author version arXiv:cond-mat/9809162v1 (1998); Open PDF.