Skip to content

How does an identity on three small spaces constrain an entire spin chain? One auxiliary spin visits every physical site in a fixed order. The resulting monodromy obeys the same exchange relation as each local factor, and tracing out the auxiliary spin produces a family of operators on the physical chain. You will construct this family and show explicitly why one special spectral value gives translation.

Required background. Use the rational Yang–Baxter identity and tensor-leg notation from Check a rational R-matrix. The XXX model defines the physical spin spaces and periodic bonds.

Take a finite homogeneous periodic spin-1/21/2 chain with N≥3N\geq3 and

H=V1⊗⋯⊗VN,Vn=C2.\mathcal H=V_1\otimes\cdots\otimes V_N, \qquad V_n=\mathbb C^2.

The auxiliary space Va=C2V_a=\mathbb C^2 is an additional tensor factor used to construct operators; it is not an extra physical site. Define

Lan(λ)=Ran(λ−i/2)=(λ−i/2)I+iPan.L_{an}(\lambda)=R_{an}(\lambda-i/2) =(\lambda-i/2)I+iP_{an}.

Every site uses the same spectral shift; this is the homogeneous choice. The monodromy and transfer matrix are

Ta(λ)=LaN(λ)⋯La1(λ),τ(λ)=tr⁡aTa(λ).T_a(\lambda)=L_{aN}(\lambda)\cdots L_{a1}(\lambda), \qquad \tau(\lambda)=\operatorname{tr}_a T_a(\lambda).

The rightmost factor acts first, so the auxiliary spin visits site 11, then 22, and so on. The trace closes that auxiliary path. TaT_a acts on a space of dimension 2N+12^{N+1}; τ\tau acts on the physical space of dimension 2N2^N.

These definitions follow Faddeev 1996, § 3, equations (31)–(43), PDF. The convention reference fixes their order and spectral origin.

An operator XX on Va⊗HV_a\otimes\mathcal H can be written as a 2×22\times2 block matrix, whose blocks act on H\mathcal H. What is tr⁡aX\operatorname{tr}_aX? Is it a scalar?

Repair. If X=∑r,s=↑,↓∣r⟩⟨s∣a⊗XrsX=\sum_{r,s=\uparrow,\downarrow}|r\rangle\langle s|_a\otimes X_{rs}, then

tr⁡aX=X↑↑+X↓↓.\operatorname{tr}_aX=X_{\uparrow\uparrow}+X_{\downarrow\downarrow}.

This is still an operator on the full physical Hilbert space. Taking the trace over the physical spins as well would discard the operator we want to study. For example, tr⁡aIaH=2IH\operatorname{tr}_aI_{a\mathcal H}=2I_{\mathcal H}, while tr⁡aPan=IH\operatorname{tr}_aP_{an}=I_{\mathcal H}.

Introduce a second auxiliary space VbV_b. Substitute the three factors a,b,na,b,n into the Yang–Baxter identity to get

Rab(λ−μ)Lan(λ)Lbn(μ)=Lbn(μ)Lan(λ)Rab(λ−μ).\begin{aligned} &R_{ab}(\lambda-\mu)L_{an}(\lambda)L_{bn}(\mu)\\ &\hspace{1.5em}=L_{bn}(\mu)L_{an}(\lambda)R_{ab}(\lambda-\mu). \end{aligned}

The spectral arguments work because (λ−μ)+(μ−i/2)=λ−i/2(\lambda-\mu)+(\mu-i/2)=\lambda-i/2.

The corresponding chain relation is

Rab(λ−μ)Ta(λ)Tb(μ)=Tb(μ)Ta(λ)Rab(λ−μ).R_{ab}(\lambda-\mu)T_a(\lambda)T_b(\mu) =T_b(\mu)T_a(\lambda)R_{ab}(\lambda-\mu).

It is often called the RTT relation, after the order of its factors. Here is the two-site step of its proof. Suppress spectral arguments and write R=Rab(λ−μ)R=R_{ab}(\lambda-\mu):

RLa2La1Lb2Lb1=RLa2Lb2La1Lb1=Lb2La2RLa1Lb1=Lb2La2Lb1La1R=Lb2Lb1La2La1R.\begin{aligned} R L_{a2}L_{a1}L_{b2}L_{b1} &=R L_{a2}L_{b2}L_{a1}L_{b1}\\ &=L_{b2}L_{a2}R L_{a1}L_{b1}\\ &=L_{b2}L_{a2}L_{b1}L_{a1}R\\ &=L_{b2}L_{b1}L_{a2}L_{a1}R. \end{aligned}

The first and last equalities exchange operators acting on disjoint pairs of tensor factors. The middle equalities use the local relation at one site each. Repeating the same step moves RR through all NN sites; no assumption that neighboring physical spins commute inside a common local factor is required. This is the induction of Faddeev 1996, § 3, equations (44)–(48), PDF.

The next lesson will prove carefully why the auxiliary trace of RTT gives commuting transfer matrices. Generic cyclicity of a partial trace would be an invalid shortcut.

At λ0=i/2\lambda_0=i/2, every local factor is iPaniP_{an}. For N=3N=3, temporarily omit the overall scalar i3i^3 and track an arbitrary auxiliary state rr and physical states s1,s2,s3s_1,s_2,s_3.

StepAuxiliary factorPhysical factors 1,2,31,2,3
Initial tensorrr(s1,s2,s3)(s_1,s_2,s_3)
After Pa1P_{a1}s1s_1(r,s2,s3)(r,s_2,s_3)
After Pa2P_{a2}s2s_2(r,s1,s3)(r,s_1,s_3)
After Pa3P_{a3}s3s_3(r,s1,s2)(r,s_1,s_2)

In the partial trace, the outgoing auxiliary value must equal its initial value: r=s3r=s_3. Thus

τ3(i/2)∣s1,s2,s3⟩=i3∣s3,s1,s2⟩.\tau_3(i/2)|s_1,s_2,s_3\rangle =i^3|s_3,s_1,s_2\rangle.

The same bookkeeping on NN sites gives

τ(i/2)=iNU,U∣s1,…,sN⟩=∣sN,s1,…,sN−1⟩.\tau(i/2)=i^N U, \qquad U|s_1,\ldots,s_N\rangle=|s_N,s_1,\ldots,s_{N-1}\rangle.

Equivalently U=P12P23⋯PN−1,NU=P_{12}P_{23}\cdots P_{N-1,N}. It is unitary and invertible, since it permutes an orthonormal basis. This agrees with Faddeev 1996, § 3, equations (49)–(60), PDF.

For a down spin at site xx, U∣x⟩=∣x+1⟩U|x\rangle=|x+1\rangle. Hence the earlier coefficient convention ∣k⟩=N−1/2∑xeikx∣x⟩|k\rangle=N^{-1/2}\sum_xe^{ikx}|x\rangle gives U∣k⟩=e−ik∣k⟩U|k\rangle=e^{-ik}|k\rangle. The sign of the translation eigenvalue follows from the action just derived.

Put z=λ−i/2z=\lambda-i/2 and S=P12+P13+P23S=P_{12}+P_{13}+P_{23} on three physical sites. Expanding the three local factors before taking the trace yields

τ3(λ)=(2z3+3iz2)I−zS−iU.\tau_3(\lambda)=(2z^3+3iz^2)I-zS-iU.

The terms have a direct interpretation. Zero swaps give 2z3I2z^3I because the auxiliary identity has trace 22. One swap gives iz2Iiz^2I at each of the three sites. Two swaps give the physical exchange of the two visited sites, multiplied by i2z=−zi^2z=-z. Three swaps give i3U=−iUi^3U=-iU.

This polynomial supplies checks at generic complex λ\lambda, not just the regular point. In particular the all-up state has eigenvalue

2z3+3iz2−3z−i=(z+i)3+z3.2z^3+3iz^2-3z-i=(z+i)^3+z^3.

The equality follows either from this expansion or from the triangular local action on the all-up state.

Guided practice: identify the physical shift

Section titled “Guided practice: identify the physical shift”

For N=3N=3, calculate τ(i/2)∣↑↓↓⟩\tau(i/2)|\uparrow\downarrow\downarrow\rangle and U3U^3. For arbitrary NN, explain why U−1SnzU=Sn−1zU^{-1}S_n^zU=S_{n-1}^z, with site indices taken modulo NN.

Hint

Use the tuple action of UU, including the scalar i3i^3 only for τ\tau. The value arriving at site nn was previously at site n−1n-1.

Solution

The shifted state is ∣↓↑↓⟩|\downarrow\uparrow\downarrow\rangle, so τ(i/2)\tau(i/2) gives −i∣↓↑↓⟩-i|\downarrow\uparrow\downarrow\rangle. Three applications of UU return each spin to its original site, hence U3=IU^3=I on three sites. More generally SnzUS_n^zU reads the original spin at n−1n-1, so SnzU=USn−1zS_n^zU=US_{n-1}^z, proving the stated conjugation identity. Replacing U−1SnzUU^{-1}S_n^zU by USnzU−1US_n^zU^{-1} reverses the direction.

Independent practice: reconstruct the polynomial

Section titled “Independent practice: reconstruct the polynomial”

Prove tr⁡a(PajPai)=Pij\operatorname{tr}_a(P_{aj}P_{ai})=P_{ij} for distinct sites i,ji,j. Use this identity to rederive the two-swap coefficient in τ3(λ)\tau_3(\lambda) and check the all-up eigenvalue.

Hint

Starting from auxiliary state rr, the two swaps leave the auxiliary factor equal to the original state at jj. The trace equates that state with rr.

Solution

On the visited factors the sequence is (r;si,sj)↦(si;r,sj)↦(sj;r,si)(r;s_i,s_j)\mapsto(s_i;r,s_j)\mapsto(s_j;r,s_i). Tracing sets r=sjr=s_j, leaving (sj,si)(s_j,s_i) on the physical sites. Thus the trace is PijP_{ij}. The three possible pairs contribute −z(P12+P13+P23)-z(P_{12}+P_{13}+P_{23}).

On the all-up state, each permutation and UU act as the identity, giving 2z3+3iz2−3z−i2z^3+3iz^2-3z-i. Expanding (z+i)3+z3(z+i)^3+z^3 reproduces it. A trace over all four spins instead would give a scalar and could not pass this operator-level test.

Transfer: give the sites different spectral shifts

Section titled “Transfer: give the sites different spectral shifts”

Replace the local factor by Lan(λ)=Ran(λ−θn−i/2)L_{an}(\lambda)=R_{an}(\lambda-\theta_n-i/2) for fixed complex numbers θn\theta_n. Does the same local relation and RTT proof survive? Must there still be a single λ0\lambda_0 at which every local factor equals iPaniP_{an}?

Hint

Subtract the two spectral arguments at the same site. Then solve λ0−θn−i/2=0\lambda_0-\theta_n-i/2=0 simultaneously for all nn.

Solution

At site nn, the difference remains (λ−θn−i/2)−(μ−θn−i/2)=λ−μ(\lambda-\theta_n-i/2)-(\mu-\theta_n-i/2)=\lambda-\mu. The same R-matrix therefore intertwines the local factors, and disjoint-factor commutativity still permits the global induction. RTT survives.

A common regular point requires λ0=θn+i/2\lambda_0=\theta_n+i/2 for every site, so it exists in this form only when all θn\theta_n agree. Thus the commuting-family argument can remain valid while the simple translation and homogeneous nearest-neighbor Hamiltonian extraction no longer apply. These are separate proof steps.

You can now form the monodromy, take the correct trace and determine its shift orientation. Continue to Derive commuting charges to prove transfer commutativity and recover the periodic XXX Hamiltonian, including the closing bond.

  • Faddeev, L. D. How Algebraic Bethe Ansatz works for integrable model. Les Houches lectures, arXiv:hep-th/9605187v1, 1996, 59 pp. Version record. Open PDF. Section and equation numbers identify the cited locations in this version.