Skip to content

What does the Toda Lax equation tell us about conserved quantities? You will turn L˙=[B,L]\dot L=[B,L] into a proof that traces of powers are constant, then compute the three-particle characteristic polynomial from three such traces. The result identifies conserved spectral data; independence and Poisson commutativity remain separate questions.

Required background. Construct and check the matrices in Build the Toda Lax pair. You should be able to multiply small matrices. The entry repair explains the trace identity used below.

Helpful background. The Lax-equation reference summarizes the finite-matrix argument and its scope.

Use the dimensionless finite open chain with real canonical (q,p)(q,p), ai=e(qi−qi+1)/2>0a_i=e^{(q_i-q_{i+1})/2}\gt0, and bi=pib_i=p_i. The symmetric matrix LL has diagonal bib_i and nearest off-diagonal entries aia_i. Its skew partner has Bi,i+1=−ai/2B_{i,i+1}=-a_i/2 and Bi+1,i=ai/2B_{i+1,i}=a_i/2. The previous lesson established L˙=BL−LB\dot L=BL-LB with the missing bonds a0=aN=0a_0=a_N=0.

For every positive integer kk, define

Ik=1ktr⁡(Lk).I_k=\frac1k\operatorname{tr}(L^k).

Here the trace is the sum of the diagonal entries. For N=3N=3, we will use the names

P=I1,H=I2,J=I3.P=I_1,\qquad H=I_2,\qquad J=I_3.

The factor 1/k1/k is part of the definition, so JJ means tr⁡(L3)/3\operatorname{tr}(L^3)/3, not tr⁡(L3)\operatorname{tr}(L^3). The finite Toda characteristic polynomial as a source of constants of motion appears in Moser 1975, § 2, p. 473, equation (2.8), PDF; the normalization above is the one used in this course.

For

X=(0100),Y=(0010),X=\begin{pmatrix}0&1\\0&0\end{pmatrix}, \qquad Y=\begin{pmatrix}0&0\\1&0\end{pmatrix},

compute XYXY, YXYX, and their traces. Are equal traces sufficient to conclude that the two matrices are equal?

Repair. XY=diag⁡(1,0)XY=\operatorname{diag}(1,0) and YX=diag⁡(0,1)YX=\operatorname{diag}(0,1). Each has trace 11, but the matrices differ. More generally, for finite square matrices,

tr⁡(XY)=∑i,jXijYji=∑j,iYjiXij=tr⁡(YX).\operatorname{tr}(XY)=\sum_{i,j}X_{ij}Y_{ji} =\sum_{j,i}Y_{ji}X_{ij}=\operatorname{tr}(YX).

Repeated use gives the cyclic rule tr⁡(XYZ)=tr⁡(ZXY)\operatorname{tr}(XYZ)=\operatorname{tr}(ZXY). This permits cyclic rotations inside a trace, not arbitrary rearrangements of factors.

The derivative of a matrix power must initially retain every factor order:

ddtLk=∑r=0k−1LrL˙Lk−1−r.\frac{d}{dt}L^k =\sum_{r=0}^{k-1}L^r\dot L L^{k-1-r}.

After taking the trace, cyclicity makes all kk terms equal. Therefore

dIkdt=tr⁡(Lk−1L˙).\frac{dI_k}{dt}=\operatorname{tr}(L^{k-1}\dot L).

Insert the Lax equation:

dIkdt=tr⁡(Lk−1BL−LkB)=tr⁡(BLk−LkB)=0.\begin{aligned} \frac{dI_k}{dt} &=\operatorname{tr}(L^{k-1}BL-L^kB)\\ &=\operatorname{tr}(BL^k-L^kB)=0. \end{aligned}

We used neither LL˙=L˙LL\dot L=\dot L L nor LB=BLLB=BL. Those equalities generally fail. The argument applies to differentiable finite matrices satisfying the stated Lax equation; infinite-dimensional operator traces need additional hypotheses and are outside this calculation.

For three particles,

L=(p1a10a1p2a20a2p3).L=\begin{pmatrix} p_1&a_1&0\\ a_1&p_2&a_2\\ 0&a_2&p_3 \end{pmatrix}.

The first trace gives

P=p1+p2+p3.P=p_1+p_2+p_3.

Each off-diagonal square occurs twice in tr⁡L2\operatorname{tr}L^2, so

H=12(p12+p22+p32)+a12+a22.H=\frac12(p_1^2+p_2^2+p_3^2)+a_1^2+a_2^2.

This is exactly the Hamiltonian because ai2=eqi−qi+1a_i^2=e^{q_i-q_{i+1}}. For the cubic trace, set u=a12u=a_1^2 and v=a22v=a_2^2. Direct multiplication gives

(L3)11=p13+u(2p1+p2),(L3)22=p23+u(p1+2p2)+v(2p2+p3),(L3)33=p33+v(p2+2p3).\begin{aligned} (L^3)_{11}&=p_1^3+u(2p_1+p_2),\\ (L^3)_{22}&=p_2^3+u(p_1+2p_2)+v(2p_2+p_3),\\ (L^3)_{33}&=p_3^3+v(p_2+2p_3). \end{aligned}

Summing and dividing by three yields

J=13(p13+p23+p33)+u(p1+p2)+v(p2+p3).J=\frac13(p_1^3+p_2^3+p_3^3) +u(p_1+p_2)+v(p_2+p_3).

The mixed terms matter. The sum of momentum cubes alone is not generally conserved by the interacting chain.

At q=(0,0,0)q=(0,0,0) and p=(1,0,−1)p=(1,0,-1), the initial matrix is

L0=(11010101−1),L02=(21112−11−12).L_0=\begin{pmatrix}1&1&0\\1&0&1\\0&1&-1\end{pmatrix}, \qquad L_0^2=\begin{pmatrix}2&1&1\\1&2&-1\\1&-1&2\end{pmatrix}.

Thus P=0P=0 and H=3H=3. The cubic formula gives J=0J=0: the momentum-cube contribution vanishes, and the two bond contributions are 11 and −1-1.

For an arbitrary three-particle state, expansion of the determinant gives

χL(λ)=det⁡(λId3−L)=(λ−p1)(λ−p2)(λ−p3)−u(λ−p3)−v(λ−p1).\begin{aligned} \chi_L(\lambda) &=\det(\lambda\mathrm{Id}_3-L)\\ &=(\lambda-p_1)(\lambda-p_2)(\lambda-p_3)\\ &\quad-u(\lambda-p_3)-v(\lambda-p_1). \end{aligned}

At the initial state this becomes

χL0(λ)=λ3−3λ=λ(λ−3)(λ+3).\chi_{L_0}(\lambda)=\lambda^3-3\lambda =\lambda(\lambda-\sqrt3)(\lambda+\sqrt3).

The eigenvalues are −3-\sqrt3, 00, and 3\sqrt3. They remain the spectrum along this exact Toda trajectory even though the matrix entries change. In the guided exercise you will show that P,H,JP,H,J determine all three characteristic-polynomial coefficients, making that conclusion independent of any numerical diagonalization.

The values P=J=0P=J=0 at this state do not say that PP and JJ are the same function, or that their gradients vanish. Also, a fixed spectrum does not specify all positions and momenta along a trajectory.

Guided practice: recover the characteristic polynomial

Section titled “Guided practice: recover the characteristic polynomial”

Let the eigenvalues be λ1,λ2,λ3\lambda_1,\lambda_2,\lambda_3. Write

χL(λ)=λ3−e1λ2+e2λ−e3.\chi_L(\lambda)=\lambda^3-e_1\lambda^2+e_2\lambda-e_3.

Use ∑jλj=P\sum_j\lambda_j=P, ∑jλj2=2H\sum_j\lambda_j^2=2H, and ∑jλj3=3J\sum_j\lambda_j^3=3J to derive

e1=P,e2=P22−H,e3=P36−PH+J.e_1=P,\qquad e_2=\frac{P^2}{2}-H, \qquad e_3=\frac{P^3}{6}-PH+J.

Begin by expanding (λ1+λ2+λ3)2(\lambda_1+\lambda_2+\lambda_3)^2. For the cubic step, use

∑jλj3=e1∑jλj2−e2∑jλj+3e3.\sum_j\lambda_j^3=e_1\sum_j\lambda_j^2-e_2\sum_j\lambda_j+3e_3.

Explain why conservation of these traces fixes the spectrum as a multiset, including multiplicities.

Hint · Full solution

Independent practice: two routes to one polynomial

Section titled “Independent practice: two routes to one polynomial”

Use the unequal-bond state q=(log⁡4,0,−log⁡9)q=(\log4,0,-\log9) and p=(2,−1,0)p=(2,-1,0). Calculate P,H,JP,H,J and obtain χL\chi_L from the formulas above. Independently expand det⁡(λId3−L)\det(\lambda\mathrm{Id}_3-L) and compare all coefficients. What is det⁡L\det L? Is the conserved cubic trace equal to that determinant?

Hint · Full solution

Let L(t)L(t) satisfy the original Lax equation and let c(t)c(t) be a differentiable real function. Define a new matrix

L~(t)=L(t)+c(t)Id3.\widetilde L(t)=L(t)+c(t)\mathrm{Id}_3.

Find L~˙\dot{\widetilde L} and determine whether it still equals [B,L~][B,\widetilde L]. Compute tr⁡L~\operatorname{tr}\widetilde L and tr⁡(L~2)/2\operatorname{tr}(\widetilde L^2)/2. Explain what happens to individual eigenvalues and their pairwise differences when cc changes in time. How does the answer change if cc is constant?

Hint · Full solution

Polynomial coefficients. The cross terms in P2P^2 are twice e2e_2. Substitute the resulting e2e_2 into the supplied cubic identity and solve for e3e_3.

Unequal bonds. The two mixed contributions to JJ are 4(2−1)4(2-1) and 9(−1+0)9(-1+0). The constant coefficient of χL\chi_L is −det⁡L-\det L for a 3×33\times3 matrix.

Scalar shift. The identity matrix commutes with BB. If Lw=λwLw=\lambda w, compute L~w\widetilde Lw for the same nonzero vector ww.

Expansion gives P2=2H+2e2P^2=2H+2e_2, hence e2=P2/2−He_2=P^2/2-H. In the cubic identity,

3J=2PH−P(P22−H)+3e3=3PH−P32+3e3.\begin{aligned} 3J&=2PH-P\left(\frac{P^2}{2}-H\right)+3e_3\\ &=3PH-\frac{P^3}{2}+3e_3. \end{aligned}

Consequently e3=J−PH+P3/6e_3=J-PH+P^3/6, and

χL(λ)=λ3−Pλ2+(P22−H)λ−(P36−PH+J).\chi_L(\lambda)=\lambda^3-P\lambda^2 +\left(\frac{P^2}{2}-H\right)\lambda -\left(\frac{P^3}{6}-PH+J\right).

Every coefficient is a function of conserved quantities. The polynomial is therefore constant in time, so its roots, counted with their multiplicities, are unchanged. This is isospectrality. It is not yet a calculation of Poisson brackets between the conserved quantities.

The invariant values are

P=1,H=312,J=8−13+4−9=−83.P=1,\qquad H=\frac{31}{2},\qquad J=\frac{8-1}{3}+4-9=-\frac83.

Thus e2=−15e_2=-15 and e3=−18e_3=-18. The resulting polynomial is

χL(λ)=λ3−λ2−15λ+18.\chi_L(\lambda)=\lambda^3-\lambda^2-15\lambda+18.

Direct expansion of the matrix

L=(2202−13030)L=\begin{pmatrix}2&2&0\\2&-1&3\\0&3&0\end{pmatrix}

gives

χL(λ)=(λ−2)(λ+1)λ−4λ−9(λ−2)=λ3−λ2−15λ+18.\begin{aligned} \chi_L(\lambda) &=(\lambda-2)(\lambda+1)\lambda -4\lambda-9(\lambda-2)\\ &=\lambda^3-\lambda^2-15\lambda+18. \end{aligned}

The determinant is −18-18, which differs from J=−8/3J=-8/3. Both are conserved functions, related by det⁡L=P3/6−PH+J\det L=P^3/6-PH+J. They agree when P=0P=0, but that special equality must not be used at a general state.

Differentiating the new matrix gives

L~˙=[B,L~]+c˙ Id3.\dot{\widetilde L}=[B,\widetilde L]+\dot c\,\mathrm{Id}_3.

The extra term vanishes identically only when cc is constant. The shifted traces are

P~=P+3c,H~=H+cP+32c2,\widetilde P=P+3c,\qquad \widetilde H=H+cP+\frac32c^2,

and their derivatives are 3c˙3\dot c and c˙(P+3c)\dot c(P+3c), respectively. They are not generally constant. For example, at the course’s initial invariant values and with c(t)=tc(t)=t, they become 3t3t and 3+3t2/23+3t^2/2.

Each eigenvalue becomes λ~j(t)=λj+c(t)\widetilde\lambda_j(t)=\lambda_j+c(t), so its differences from the other eigenvalues stay fixed. Conserved spectral gaps alone do not imply an unchanged spectrum. If cc is constant, the shifted matrix again obeys the homogeneous Lax equation and all its traces are conserved.

You should now be able to prove trace conservation, identify the Hamiltonian among the traces, and reconcile trace invariants with characteristic-polynomial coefficients. Cyclicity is the key algebraic step; it does not let you commute matrices outside a trace.

Continue to test independence and Poisson commutativity. That lesson supplies the missing checks between isospectral evolution and Liouville integrability.

  • Moser, Jürgen. “Finitely many mass points on the line under the influence of an exponential potential—an integrable system.” In Dynamical Systems, Theory and Applications, edited by Jürgen Moser, Lecture Notes in Physics 38, pp. 467–497. Springer, 1975. DOI. Open PDF.