Calogero–Moser–Sutherland systems

The Calogero–Moser–Sutherland system begins with a very concrete mechanical problem. We have $N$ particles on a line, with positions $x_1,\ldots,x_N$ and momenta $p_1,\ldots,p_N$ . Their phase space has canonical Poisson brackets $\{x_i,p_j\}=\delta_{ij}$ , while brackets among two positions or two momenta vanish. The rational Calogero–Moser Hamiltonian is

$\displaystyle H=\frac12\sum_{i=1}^N p_i^2+g^2\sum_{1\leq i<j\leq N}\frac1{(x_i-x_j)^2}.$

At first sight this is simply a many-body system with a repulsive inverse-square interaction. The potential becomes infinite when two particles meet, so collisions are forbidden. But this alone does not make the system special. There are many singular repulsive interactions, and almost all of them lead to nonlinear systems that cannot be solved exactly.

What makes Calogero–Moser exceptional is that the inverse-square interaction is not merely chosen because it is physically interesting. It is forced by a hidden geometric construction. The particles are what one sees after starting with a completely free particle moving in the vector space of Hermitian matrices, fixing an appropriate matrix-valued angular momentum, forgetting the choice of matrix basis, and then recording only the eigenvalues of the matrix position. Thus the ordinary particle coordinates $x_i$ are not the most fundamental variables. They are spectral coordinates: they are eigenvalues. The inverse-square force is not inserted from outside. It is the kinetic energy of matrix directions that disappear when one diagonalizes the matrix position. This is the main point of the whole story. The system looks nonlinear in the particle coordinates because diagonalizing a moving matrix is nonlinear. In the larger matrix space, the motion is just

$\displaystyle Q(t)=Q(0)+tP(0),\qquad P(t)=P(0).$

The particles are the eigenvalues of the freely moving matrix $Q(t)$ . Before explaining that construction, however, we should first understand the particle system directly. Otherwise the matrix reduction can seem like an abstract machine rather than an explanation of the actual force law.

The rational particle system

The Hamiltonian contains two pieces. The kinetic energy is $\frac12\sum_i p_i^2$ . The potential energy is $g^2\sum_{i<j}(x_i-x_j)^{-2}$ . The constant $g$ controls the strength of the repulsion. Since the potential depends only on differences $x_i-x_j$ , shifting every particle by the same amount changes nothing. This translation symmetry will imply conservation of total momentum. Hamilton’s equations say that $\dot x_i=\partial H/\partial p_i$ and $\dot p_i=-\partial H/\partial x_i$ . Since only the kinetic term depends on $p_i$ , we immediately obtain $\dot x_i=p_i$ . The second equation requires differentiating the potential. For a fixed pair $i,j$ , differentiation gives $\partial_{x_i}(x_i-x_j)^{-2}=-2(x_i-x_j)^{-3}$ . Therefore

$\displaystyle \dot p_i=2g^2\sum_{j\ne i}\frac1{(x_i-x_j)^3}, \quad \ddot x_i=2g^2\sum_{j\ne i}\frac1{(x_i-x_j)^3}.$

The sign is important. Suppose $x_i>x_j$ . Then $x_i-x_j>0$ , so the contribution of particle $j$ to $\dot p_i$ is positive. Particle $i$ is pushed right, away from particle $j$ . Conversely, particle $j$ is pushed left. Thus the interaction is repulsive. The singularity at $x_i=x_j$ has a simple energetic consequence. If the total energy is finite and $g\neq0$ , then no denominator $(x_i-x_j)^2$ can become zero. So particles never collide. If initially $x_1<x_2<\cdots<x_N$ , this order remains true for all time. The region $x_1<x_2<\cdots<x_N$ is one chamber of configuration space, separated from the others by the singular collision walls $x_i=x_j$ .

The total momentum $P=\sum_i p_i$ is conserved. Indeed, differentiating gives $\dot P=2g^2\sum_i\sum_{j\ne i}(x_i-x_j)^{-3}$ . Each ordered pair $(i,j)$ is cancelled by $(j,i)$ , because $(x_j-x_i)^{-3}=-(x_i-x_j)^{-3}$ . Hence $\dot P=0$ . This is the direct computation; geometrically, it is Noether’s theorem for translation invariance.

The case $N=2$ is worth solving fully because it reveals in elementary terms what the inverse-square force does. Introduce the center-of-mass coordinate $R=(x_1+x_2)/2$ , the relative coordinate $r=x_1-x_2$ , the total momentum $P=p_1+p_2$ , and the relative momentum $p=(p_1-p_2)/2$ . These variables are canonical: ${R,P}=1$ , ${r,p}=1$ , and the other brackets vanish. Since $p_1=P/2+p$ and $p_2=P/2-p$ , the Hamiltonian becomes

$\displaystyle H=\frac{P^2}{4}+\left(p^2+\frac{g^2}{r^2}\right).$

The center of mass is free. Since $P$ is constant, $\dot R=P/2$ , so $R(t)=R(0)+Pt/2$ . Every nontrivial feature is concentrated in the relative motion. Write $E=p^2+g^2/r^2$ for the relative energy. Since $\dot r=\dot x_1-\dot x_2=p_1-p_2=2p$ , one has $p=\dot r/2$ . Therefore $E=\dot r^2/4+g^2/r^2$ , or equivalently $\dot r^2=4(E-g^2/r^2)$ . At the closest approach, $\dot r=0$ , so $r_{\min}=|g|/\sqrt E$ . The particles cannot reach $r=0$ . The equation can be integrated directly. Put $y=r^2$ . Then $\dot y=2r\dot r$ , and therefore $\dot y^2=16(Ey-g^2)$ . Solving gives

$\displaystyle r(t)^2=\frac{g^2}{E}+4E(t-t_*)^2.$

The time $t_*$ is the moment of closest approach. At large positive or negative time, $r(t)\sim2\sqrt E |t|$ , so the particles behave asymptotically like free particles. The interaction only matters when they come close. In the ordered-particle description, they approach, slow down, reach a nonzero minimum separation, and move apart again. If one ignores labels and instead follows asymptotic straight lines, the effect is equivalent to the two particles passing through one another while exchanging velocities.

Already here one sees the basic Calogero–Moser scattering picture: far in the past and far in the future the system looks free, but the interaction produces a highly organized, exactly calculable rearrangement of the asymptotic data.

Lax Equation

The Hamiltonian suggests an unexpected construction. Its kinetic term is quadratic in $p_i$ , while its potential is quadratic in $1/(x_i-x_j)$ . This invites us to seek a matrix whose diagonal entries are the momenta and whose off-diagonal entries are proportional to $1/(x_i-x_j)$ . Squaring such a matrix might produce both the kinetic and potential terms simultaneously. Define an $N\times N$ matrix $L$ by

$\displaystyle L_{ij}=p_i\delta_{ij}+ig(1-\delta_{ij})\frac1{x_i-x_j}.$

Thus $L_{ii}=p_i$ , while $L_{ij}=ig/(x_i-x_j)$ for $i\ne j$ . If the $x_i$ and $p_i$ are real, then $L$ is Hermitian. Indeed, $L_{ji}=ig/(x_j-x_i)=-ig/(x_i-x_j)=\overline{L_{ij}}$ . Hence its eigenvalues are real. The trace of $L$ is just $\sum_i p_i$ , namely total momentum. More importantly, ${\text{tr}}(L^2)$ gives the Hamiltonian. The diagonal term in $(L^2){ii}=\sum_jL{ij}L_{ji}$ is $p_i^2$ . For $j\ne i$ , one has $L_{ij}L_{ji}=(ig/(x_i-x_j))(ig/(x_j-x_i))=g^2/(x_i-x_j)^2$ . Therefore

$\displaystyle \frac12 {\text{tr}}(L^2) = \frac12\sum_i p_i^2+g^2\sum_{i<j}\frac1{(x_i-x_j)^2} =H.$

So the Hamiltonian is not merely related to $L$ . It is literally the quadratic spectral invariant of $L$ . At this stage, however, $L$ still looks engineered. We have guessed a matrix that works. Later we will see that its off-diagonal entry $ig/(x_i-x_j)$ is forced by a moment-map constraint coming from free matrix motion.

Define a second matrix $M$ by taking its off-diagonal entries to be $M_{ij}=ig/(x_i-x_j)^2$ for $i\ne j$ , and its diagonal entries to be $M_{ii}=-ig\sum_{k\ne i}(x_i-x_k)^{-2}$ . Then the particle equations are equivalent to the matrix equation

$\displaystyle \dot L=[M,L]=ML-LM.$

It is useful to verify this rather than treating it as a miraculous identity. On the diagonal, $\dot L_{ii}=\dot p_i$ . The $k$ -th term in $[M,L]_{ii}$ , for $k\ne i$ , is $M_{ik}L_{ki}-L_{ik}M_{ki}$ . If $q=x_i-x_k$ , then $M_{ik}=ig/q^2$ , $L_{ki}=-ig/q$ , $L_{ik}=ig/q$ , and $M_{ki}=ig/q^2$ . Hence the first product is $g^2/q^3$ , the second is $-g^2/q^3$ , and their difference is $2g^2/q^3$ . Summing over $k\ne i$ gives exactly $\dot p_i$ . For an off-diagonal entry $i\ne j$ , differentiation of $L_{ij}=ig/(x_i-x_j)$ gives $\dot L_{ij}=-ig(p_i-p_j)/(x_i-x_j)^2$ . The terms involving $p_i$ and $p_j$ in the commutator give this expression. The remaining terms involve a third index $k$ . Their cancellation is the algebraic heart of the rational model. If $a=x_i-x_k$ , $b=x_k-x_j$ , then $x_i-x_j=a+b$ , and the terms cancel because the rational identities involving $1/a$ , $1/b$ , and $1/(a+b)$ fit together exactly. This is why a generic pair potential does not produce a Lax equation: most functions do not satisfy the necessary three-point identities.

The meaning of $\dot L=[M,L]$ is that $L$ changes by infinitesimal conjugation. If $L$ were replaced at a finite time by $L(t)=U(t)L(0)U(t)^{-1}$ , its eigenvalues would remain unchanged. A commutator is the infinitesimal form of precisely that operation. Therefore the eigenvalues of $L$ are constant. Equivalently, every trace ${\text{tr}}(L^k)$ is conserved. Indeed, differentiating gives $\frac{d}{dt} {\text{tr}}(L^k)=k{\text{tr}}(L^{k-1}[M,L])$ . The two terms produced by the commutator cancel by cyclicity of trace. Thus the quantities $I_k={\text{tr}}(L^k)/k$ are conserved.

The first invariant is total momentum. The second is the Hamiltonian. The remaining invariants are hidden conservation laws. For an $N\times N$ matrix, only the first $N$ are generically independent, because all higher powers are constrained by the characteristic polynomial. Liouville integrability requires not only that these quantities be conserved and independent, but also that they Poisson commute. The Lax equation alone proves conservation; the matrix-reduction construction below explains the stronger Poisson-commutativity statement.

A Lax pair proves that the system is isospectral, but it does not explain why the particular matrices $L$ and $M$ exist. The expressions $ig/(x_i-x_j)$ and $ig/(x_i-x_j)^2$ can still look like a clever trick. The deeper explanation comes from a familiar elementary analogy. Consider a free particle in the plane with Cartesian coordinates $(x,y)$ and momentum $(p_x,p_y)$ . Its Hamiltonian is $H=(p_x^2+p_y^2)/2$ . In polar coordinates, the same Hamiltonian becomes $H=(p_r^2+\ell^2/r^2)/2$ , where $\ell=xp_y-yp_x$ is angular momentum. If we fix $\ell$ , then the radial coordinate behaves as though it feels an inverse-square potential $\ell^2/(2r^2)$ . But no external inverse-square force was originally introduced. The term arises because angular motion costs kinetic energy. When we remove the angular variable while keeping angular momentum fixed, the missing angular kinetic energy reappears as an effective radial potential. The Calogero–Moser construction is a noncommutative many-dimensional version of this same idea. The ordinary radial coordinate $r$ is replaced by the eigenvalues of a Hermitian matrix. Ordinary angular momentum is replaced by a matrix commutator. The inverse-square particle potential is the kinetic energy of “angular” matrix directions after they are eliminated.

Free motion on Hermition Matrices

Let ${\text{Herm}}_N$ be the real vector space of $N\times N$ Hermitian matrices. A point of its cotangent bundle can be represented by a pair $(Q,P)$ of Hermitian matrices. Think of $Q$ as matrix position and $P$ as matrix momentum. The canonical one-form is $\theta={\text{tr}}(PdQ)$ , and the symplectic form is $\omega=d\theta={\text{tr}}(dP\wedge dQ)$ . This is the matrix analogue of the familiar form $\sum_i dp_i\wedge dx_i$ . Now place a completely free particle in this matrix space. Its Hamiltonian is

$\displaystyle H=\frac12{\text{tr}}(P^2).$

Hamilton’s equations are simply $\dot Q=P$ and $\dot P=0$ . Thus $P(t)=P_0$ is constant, and $Q(t)=Q_0+tP_0$ . Every matrix entry moves linearly in time. This free matrix system has $2N^2$ real dimensions, much more than the $2N$ -dimensional phase space of $N$ particles on a line. We must therefore identify a large symmetry and reduce by it. The unitary group $U(N)$ acts by simultaneous conjugation: $(Q,P)\mapsto(UQU^{-1},UPU^{-1})$ . This is simply a change of orthonormal basis in $\mathbb C^N$ . It does not change the intrinsic matrix motion. The free Hamiltonian is invariant because trace is invariant under conjugation.

The conserved quantity associated with this symmetry is the commutator $[Q,P]$ . Indeed, under conjugation it transforms as $[Q,P]\mapsto U[Q,P]U^{-1}$ , and along free motion one has $\frac{d}{dt}[Q,P]=[\dot Q,P]+[Q,\dot P]=[P,P]=0$ . Thus $[Q,P]$ is constant. This commutator is the matrix analogue of angular momentum. If $[Q,P]=0$ , then generically $Q$ and $P$ can be diagonalized simultaneously. After quotienting by unitary conjugation, the result is just $N$ noninteracting free particles. Nonzero matrix angular momentum is what produces interaction.

Let $e=(1,\ldots,1)^T\in\mathbb C^N$ , and define the anti-Hermitian matrix $\xi=ig(ee^\ast-I)$ . Its diagonal entries vanish and all off-diagonal entries equal $ig$ . We do not impose the basis-dependent equation $[Q,P]=\xi$ globally, because changing basis would replace $\xi$ by $U\xi U^{-1}$ . The invariant condition is that $[Q,P]$ lie in the conjugacy orbit of $\xi$ . This is the reduction step. We first restrict to matrix pairs whose matrix angular momentum belongs to that orbit, and then identify pairs differing by unitary conjugation. The parameter $g$ is no longer merely a coupling inserted into a particle potential. It measures the fixed amount of noncommutative angular momentum.

Now take a generic matrix $Q$ with distinct eigenvalues. We can choose a unitary matrix $U$ so that $X=U^{-1}QU$ is diagonal: $X={\text{diag}}(x_1,\ldots,x_N)$ . These eigenvalues become the particle positions. At the same time define $L=U^{-1}PU$ , the matrix momentum in this eigenbasis. The remaining freedom after diagonalizing $Q$ consists of diagonal unitary matrices. Using this residual freedom, one may arrange that the commutator takes the representative $[X,L]=ig(ee^\ast-I)$ . The key point is that the diagonal entries of $[X,L]$ always vanish. If $[X,L]=ig(vv^\ast-I)$ for some vector $v$ , this forces $|v_i|=1$ for every $i$ . Diagonal unitary phase changes can then make every $v_i$ equal to $1$ .

Now look at an off-diagonal entry of $[X,L]$ . Since $X$ is diagonal, $[X,L]_{ij}=(x_i-x_j)L_{ij}$ . The constraint says that this equals $ig$ whenever $i\ne j$ . Thus

$\displaystyle L_{ij}=\frac{ig}{x_i-x_j}\quad(i\ne j).$

The diagonal entries of $L$ are unconstrained; call them $p_i$ . We have recovered the rational Calogero–Moser Lax matrix exactly. The denominator $x_i-x_j$ has not been guessed. It is forced by the equation $[X,L]=\xi$ .

The free matrix Hamiltonian is $\frac12{\text{tr}}(P^2)$ . Since $P$ and $L$ are related by conjugation, ${\text{tr}}(P^2)={\text{tr}}(L^2)$ . But the diagonal and off-diagonal entries of $L$ have just been determined by the reduction. The diagonal entries contribute $\frac12\sum_i p_i^2$ . The off-diagonal entries contribute $\frac12\sum_{i\ne j}g^2/(x_i-x_j)^2$ , which is $g^2\sum_{i<j}(x_i-x_j)^{-2}$ . Therefore the reduced free Hamiltonian is exactly

$\displaystyle \frac12{\text{tr}}(P^2) \longrightarrow \frac12\sum_i p_i^2+g^2\sum_{i<j}\frac1{(x_i-x_j)^2}.$

This is the conceptual origin of the Calogero–Moser force. The particles repel because if two eigenvalues $x_i$ and $x_j$ approach one another while the commutator remains fixed and nonzero, then $L_{ij}=ig/(x_i-x_j)$ must become large. But $L_{ij}$ contributes to the kinetic energy. Thus finite energy prevents eigenvalue collision.

The particle singularity is therefore a geometric shadow of fixed noncommutativity. A matrix with nonzero angular momentum cannot remain diagonal as two of its eigenvalues collide unless the off-diagonal momentum diverges.

The reduced symplectic form also becomes the usual particle symplectic form. On the diagonal gauge slice, $dX$ is diagonal, so in $\theta=\text{tr}(LdX)$ only the diagonal entries of $L$ contribute. Hence $\theta=\sum_i p_i dx_i$ , and $\omega=\sum_i dp_i\wedge dx_i$ . The reduced variables are therefore genuinely canonical particle coordinates. The reduction explains $L$ , but it also gives a much more transparent interpretation of the Lax equation.

In a fixed matrix basis, free motion is $Q(t)=Q_0+tP_0$ and $P(t)=P_0$ . At every time, choose a unitary matrix $U(t)$ which diagonalizes $Q(t)$ . Then $X(t)=U(t)^{-1}Q(t)U(t)$ is diagonal, with diagonal entries equal to the particle positions, and $L(t)=U(t)^{-1}P_0U(t)$ is the matrix momentum expressed in the moving eigenbasis.

The original matrix $P_0$ is constant. But a constant matrix appears time-dependent when written in a rotating basis. Set $M=-U^{-1}\dot U$ . Differentiating $L=U^{-1}P_0U$ gives $\dot L=[M,L]$ . Thus the Lax equation says nothing mysterious: it is the equation for a constant free momentum viewed in the changing eigenbasis of the matrix position.

Differentiating $X=U^{-1}QU$ gives $\dot X=L+[M,X]$ . Since $X$ remains diagonal, its off-diagonal derivative must vanish. Thus, for $i\ne j$ , $0=L_{ij}+(x_j-x_i)M_{ij}$ . Substituting $L_{ij}=ig/(x_i-x_j)$ gives $M_{ij}=ig/(x_i-x_j)^2$ . This explains the square denominator in $M$ : it is the angular velocity required to keep the freely moving matrix $Q(t)$ diagonal.

The diagonal part of $M$ is a gauge choice reflecting the freedom to rephase eigenvectors. The common choice $M_{ii}=-ig\sum_{k\ne i}(x_i-x_k)^{-2}$ makes $Me=0$ , preserving the preferred representative of the angular-momentum constraint.

Thus $L$ is the free matrix momentum in the moving eigenbasis, while $M$ is the angular velocity of that basis.

Explicit solution

The preceding discussion gives an exact solution formula. At time zero choose the basis in which $Q_0$ is diagonal. Then $Q_0=X(0)$ and $P_0=L(0)$ . The free matrix trajectory is therefore $\displaystyle Q(t)=X(0)+tL(0).$ The particle positions $x_1(t),\ldots,x_N(t)$ are the eigenvalues of this matrix.

This is one of the strongest forms of solvability in classical mechanics. The original system is nonlinear in the particle coordinates, but the motion is reduced to finding eigenvalues of a matrix whose entries depend linearly on time. The nonlinearity has not vanished; it has moved into the operation of diagonalization. Eigenvalues of a noncommuting matrix $X(0)+tL(0)$ are nonlinear functions of time even though every matrix entry is affine-linear in $t$ .

This also explains scattering. Dividing by $t$ , one has $t^{-1}Q(t)=t^{-1}X(0)+L(0)$ , which tends to $L(0)$ as $|t|\to\infty$ . Hence the asymptotic particle velocities are the eigenvalues of the conserved Lax matrix. The incoming and outgoing velocities are the same set, though their assignment to ordered particles changes. This is the many-body version of momentum exchange in the two-particle problem.

Conserved Quantities

The matrix reduction also explains the deepest part of Liouville integrability. On the unreduced free matrix phase space, define $F_k(P)={\text{tr}}(P^k)/k$ . Each $F_k$ depends only on momentum $P$ , not on position $Q$ . Therefore their canonical Poisson brackets vanish: $\{F_k,F_\ell\}=0$ .

These functions are also invariant under unitary conjugation, because trace is conjugation-invariant. Hence they descend through the reduction. After reduction, $P$ becomes $L$ , and the descended functions are $I_k={\text{tr}}(L^k)/k$ . Their Poisson brackets still vanish.

Thus the commuting Calogero–Moser integrals are inherited from trivially commuting free-matrix observables. The hard-looking result $\{I_k,I_\ell\}=0$ becomes natural once one remembers that the system came from free motion. This is an important lesson about Hamiltonian reduction. Reduction can turn a simple system into a nonlinear-looking one, but its integrability may survive because the commuting observables of the original system descend to the quotient.

General coadjoint orbits

The scalar rational Calogero–Moser model came from a special angular-momentum orbit. For a general orbit, the reduction produces extra internal variables. Let $S=[X,L]$ . For $i\ne j$ , the relation $[X,L]_{ij}=(x_i-x_j)L_{ij}$ gives $L_{ij}=S_{ij}/(x_i-x_j)$ . Now $S$ need not be fixed to the simple constant matrix $ig(ee^\ast-I)$ . Instead, it moves on a coadjoint orbit and has its own Poisson structure. Substituting into the kinetic energy gives a Hamiltonian of the form

$\displaystyle H=\frac12\sum_i p_i^2-\sum_{i<j}\frac{S_{ij}S_{ji}}{(x_i-x_j)^2}.$

If $S$ is anti-Hermitian, then $-S_{ij}S_{ji}=|S_{ij}|^2$ , so the interaction remains repulsive. The internal orbit variables are traditionally called spins, although they need not be physical spins. The ordinary Calogero–Moser coupling $g$ is the special case in which these spin variables are frozen to a particular constant orbit configuration. The scalar model is therefore not isolated. It is one particularly symmetric member of a larger family of spin Calogero–Moser systems.

Root systems and other Lie groups

The usual $N$ -particle Calogero–Moser system is associated with the root system $A_{N-1}$ . Remove the center of mass by imposing $x_1+\cdots+x_N=0$ . The remaining coordinates lie in the $(N-1)$ -dimensional vector space $\mathfrak h=\{(x_1,\ldots,x_N)\in\mathbb R^N:\sum_i x_i=0\}$ . The roots of $A_{N-1}$ are $\alpha_{ij}=e_i-e_j$ . Evaluating such a root on $x$ gives $\alpha_{ij}(x)=x_i-x_j$ . Thus the usual potential is simply the root-system expression $\sum_{\alpha\in R_+}g^2/\alpha(x)^2$ , where $R_+$ denotes the positive roots. The Weyl group of $A_{N-1}$ is the symmetric group $S_N$ , acting by permutations of coordinates. Thus the familiar statement that particles are indistinguishable is the Weyl-group symmetry of the root system. The region $x_1<x_2<\cdots<x_N$ is a Weyl chamber, and the collision walls $x_i=x_j$ are root hyperplanes $\alpha_{ij}(x)=0$ .

This is the correct abstraction. The fundamental objects are not necessarily individual particles on a literal line. They are coordinates in a Cartan subalgebra, modulo Weyl symmetry. Let $R$ be a finite root system in a Euclidean space $\mathfrak h$ . Let $R_+$ be a choice of positive roots. The rational Calogero–Moser Hamiltonian attached to $R$ is

$\displaystyle H_R^{\mathrm{rat}} = \frac12(p,p)+\sum_{\alpha\in R_+}\frac{g_\alpha^2}{\alpha(q)^2}.$

Here $q\in\mathfrak h$ is position, $p\in\mathfrak h$ is momentum, and the couplings $g_\alpha$ must be invariant under the Weyl group. In simply laced types, all roots have the same length, so one usually takes one coupling constant. In non-simply-laced types, short and long roots may carry different couplings. The singular walls are the hyperplanes $\alpha(q)=0$ . Their complement breaks into Weyl chambers. The force is obtained by differentiating the potential: $\dot q=p$ , while $\dot p=2\sum_{\alpha\in R_+}g_\alpha^2\alpha^\sharp/\alpha(q)^3$ , where $\alpha^\sharp$ is the vector corresponding to the linear functional $\alpha$ . The interpretation is identical to the $A$ -type case. The inverse-square terms keep the motion away from the reflection hyperplanes. The Weyl group identifies the different chambers as different coordinate descriptions of the same unordered or reflected configuration.

For type $D_N$ , the roots are $\pm e_i\pm e_j$ , with $i\ne j$ . The rational Hamiltonian becomes

$\displaystyle H_{D_N} = \frac12\sum_i p_i^2+ g^2\sum_{i<j} \left( \frac1{(x_i-x_j)^2}+ \frac1{(x_i+x_j)^2} \right).$

The first term is the usual interaction between $x_i$ and $x_j$ . The second is an interaction between $x_i$ and the reflected coordinate $-x_j$ . Thus the new singular walls are not only $x_i=x_j$ , but also $x_i=-x_j$ .

For $B_N$ , one also has roots $\pm e_i$ , producing an additional one-body term $g_0^2\sum_i x_i^{-2}$ . For $C_N$ , the additional roots are $\pm2e_i$ , which again produce a term proportional to $1/x_i^2$ , though with a different normalization of the coupling. Schematically,

$\displaystyle H_{B/C} = \frac12\sum_i p_i^2+ g^2\sum_{i<j} \left( \frac1{(x_i-x_j)^2}+ \frac1{(x_i+x_j)^2} \right) + g_0^2\sum_i\frac1{x_i^2}.$

A useful picture is to imagine every particle $x_i$ accompanied by a mirror image $-x_i$ . Pairwise differences among the combined set $x_1,\ldots,x_N,-x_1,\ldots,-x_N$ include $x_i-x_j$ , $x_i+x_j$ , and $2x_i$ . These are exactly the denominators that occur in the $B$ , $C$ , and $D$ models. The exceptional root systems $E_6,E_7,E_8,F_4,G_2$ have no equally simple description in terms of coordinates and mirror particles, but the root-system formula is unchanged. Every positive root produces one inverse-square term.

General Lie-algebraic Lax matrices

The matrix $L$ for type $A$ can be understood root-theoretically. The off-diagonal matrix unit $E_{ij}$ corresponds to the root $e_i-e_j$ . The denominator $x_i-x_j$ is the root evaluation $\alpha_{ij}(q)$ . For a general semisimple Lie algebra $\mathfrak g$ , choose a Cartan subalgebra $\mathfrak h$ , root spaces $\mathfrak g_\alpha$ , and root vectors $E_\alpha\in\mathfrak g_\alpha$ . The rational Lax matrix has the schematic form

$\displaystyle L=p\cdot H+i\sum_{\alpha\in R}\frac{g_\alpha}{\alpha(q)}E_\alpha.$

The auxiliary matrix $M$ contains coefficients proportional to $1/\alpha(q)^2$ . The exact signs and normalizations depend on the normalization of root vectors and the representation in which one writes the Lax pair. But the structure is universal: momenta live in the Cartan part, while inverse-root denominators live in root directions.

In type $A$ , the key cancellations in the Lax equation involve three indices $i,k,j$ . Root-theoretically, this is the relation $(e_i-e_k)+(e_k-e_j)=e_i-e_j$ . In a general Lie algebra, the corresponding mechanism is the existence of root sums $\alpha+\beta=\gamma$ together with the bracket relation $[E_\alpha,E_\beta]\propto E_{\alpha+\beta}$ . The Lax equation is therefore controlled by root geometry and Lie brackets, not by accidental index manipulations. The type $A$ free-matrix construction has a general Lie-algebraic version. Replace the space of Hermitian matrices by a Lie algebra $\mathfrak g$ , equipped with an invariant inner product. Consider free motion on $T^\ast\mathfrak g$ , with coordinates $(Q,P)$ , Hamiltonian $(P,P)/2$ , and adjoint symmetry under the Lie group $G$ . The moment map for the adjoint action is the commutator $[Q,P]$ . On the regular set, one can use conjugation to place $Q$ in a Cartan subalgebra: $Q=q\in\mathfrak h$ . Decompose the momentum into Cartan and root components, $P=p+\sum_\alpha P_\alpha E_\alpha$ . The moment-map constraint in a root direction becomes $\alpha(q)P_\alpha=S_\alpha$ , where $S_\alpha$ is the corresponding component of the fixed orbit variable. Thus $P_\alpha=S_\alpha/\alpha(q)$ . When one substitutes this into the free kinetic energy, the root-direction terms become inverse-square potentials. For a general orbit, the $S_\alpha$ are spin variables, producing spin Calogero–Moser systems. For special orbits and special choices of residual gauge, the spin degrees of freedom reduce to constants and one recovers scalar root-system models. This is the general form of the earlier $U(N)$ calculation. The denominator $x_i-x_j$ was the type $A$ instance of the universal factor $\alpha(q)$ .

Trigonometric Systems

The trigonometric Sutherland system is the compact-group analogue of rational Calogero–Moser. Instead of a free particle moving in a flat vector space $\mathfrak g$ , one begins with free geodesic motion on a compact Lie group $G$ . A point of $T^\ast G$ , in left-trivialized coordinates, may be written as $(g,J)$ , where $g\in G$ and $J\in\mathfrak g$ . The free Hamiltonian is $(J,J)/2$ . The equations are $\dot g=gJ$ and $\dot J=0$ , so $g(t)=g(0)e^{tJ_0}$ . Again the original motion is free.

One now reduces by conjugation. A generic group element can be conjugated into a maximal torus, written as $g=e^{iq}$ with $q\in\mathfrak h$ . In a root direction $\alpha$ , the adjoint action multiplies a root vector by $e^{i\alpha(q)}$ . The moment-map constraint therefore has a factor $1-e^{-i\alpha(q)}$ . Solving for the eliminated root-direction momentum produces denominators of this form. The crucial identity is $|1-e^{iu}|^2=4\sin^2(u/2)$ . Thus the kinetic energy of the removed directions becomes a trigonometric inverse-square potential:

$\displaystyle H_R^{\mathrm{trig}} = \frac12(p,p)+ \sum_{\alpha\in R_+} \frac{a^2g_\alpha^2} {4\sin^2(a\alpha(q)/2)}.$

The periodicity is now geometric. The variable $q$ lives on a torus rather than on a vector space. In type $A_{N-1}$ , this means particles naturally move on a circle, and the collision singularity occurs whenever two angular positions coincide modulo the circumference. The rational model is recovered near the identity of the group. Since $1-e^{iu}\sim-iu$ for small $u$ , the denominator $4\sin^2(u/2)$ becomes $u^2$ . Thus flat-space rational Calogero–Moser is the local, small-angle limit of trigonometric Sutherland motion.

Hyperbolic Systems

The hyperbolic version is obtained when compact torus geometry is replaced by noncompact Cartan geometry. The relevant multiplier in a root direction is no longer $e^{i\alpha(q)}$ , but $e^{\alpha(q)}$ . The difference of exponential factors becomes

$\displaystyle e^{\alpha(q)/2}-e^{-\alpha(q)/2} = 2\sinh(\alpha(q)/2).$

Consequently, the reduced kinetic energy produces the hyperbolic potential

$\displaystyle H_R^{\mathrm{hyp}} = \frac12(p,p)+ \sum_{\alpha\in R_+} \frac{a^2g_\alpha^2} {4\sinh^2(a\alpha(q)/2)}.$

For type $A$ , this is the $N$ -particle Hamiltonian with pair potential $a^2/[4\sinh^2(a(x_i-x_j)/2)]$ . Near a collision, $\sinh u\sim u$ , so the potential behaves like $1/(x_i-x_j)^2$ . At large separation, $\sinh(ax/2)\sim e^{a|x|/2}/2$ , so the potential decays exponentially like $a^2e^{-a|x|}$ . Thus the hyperbolic model shares the same local collision geometry as the rational system but has short-range interaction at infinity.

The rational, trigonometric, and hyperbolic denominators should therefore be viewed as one geometric pattern.

The elliptic model

The elliptic Calogero–Moser system is the doubly periodic master model. Its potential is built from the Weierstrass elliptic function $\wp(z)$ , which is periodic with respect to two independent complex periods and has local expansion $\wp(z)=z^{-2}+O(z^2)$ . The elliptic root-system Hamiltonian is

$\displaystyle H_R^{\mathrm{ell}} = \frac12(p,p)+ \sum_{\alpha\in R_+}g_\alpha^2\wp(\alpha(q)).$

Near every collision wall $\alpha(q)=0$ , the potential behaves like $1/\alpha(q)^2$ , so the local force law is again rational. But globally the coordinate $\alpha(q)$ lives on an elliptic curve, not on a line or circle. The elliptic system degenerates to the trigonometric system when one period becomes infinite. The trigonometric system degenerates to the rational system when its period becomes infinite, or equivalently when the parameter $a$ tends to zero. Indeed, $\sin(ax/2)\sim ax/2$ and $\sinh(ax/2)\sim ax/2$ , so both $a^2/[4\sin^2(ax/2)]$ and $a^2/[4\sinh^2(ax/2)]$ tend to $1/x^2$ .

The elliptic Lax matrix is more complicated because elliptic functions require an additional spectral parameter. Its off-diagonal entries are built from the Weierstrass sigma and zeta functions, or equivalently from the Kronecker function. The relevant addition identities replace the elementary rational identity used in the rational Lax calculation. This is why elliptic Calogero–Moser is integrable but does not generally admit the same simple finite-dimensional free-matrix linearization $Q(t)=Q_0+tP_0$ . Its hidden linearization takes place in a more sophisticated space: spectral curves, Jacobian varieties, moduli spaces of bundles, or Hitchin systems. The basic philosophy remains unchanged. The apparent nonlinear particle motion becomes linear after passing to the correct spectral geometry.

Conclusion

The rational $A_{N-1}$ Calogero–Moser system can be summarized in one chain of ideas. Start with a free particle moving on the space of Hermitian matrices. Its motion is $Q(t)=Q_0+tP_0$ . Fix nonzero matrix angular momentum, represented by the commutator $[Q,P]$ . Quotient by unitary changes of basis. Diagonalize $Q$ , so its eigenvalues become particle positions $x_i$ . The fixed commutator forces off-diagonal matrix momentum entries to be $ig/(x_i-x_j)$ . Squaring these entries turns free kinetic energy into the inverse-square Calogero potential. The constant free momentum, viewed in the rotating eigenbasis, becomes the Lax matrix $L$ , and the basis rotation becomes $M$ . The Lax equation is therefore a moving-frame equation. Finally, the particle positions are the eigenvalues of the free matrix $X(0)+tL(0)$ .

The root-system generalization replaces coordinate differences $x_i-x_j$ by root values $\alpha(q)$ . The classical types $A,B,C,D$ become different arrangements of collision, reflection, and boundary walls. The trigonometric and hyperbolic models arise when the original free motion takes place on compact or noncompact group geometry rather than on a flat matrix space. The elliptic model is the doubly periodic master version.

The unifying principle is not simply that there happen to be many conserved quantities. It is that Calogero–Moser systems are free or linear motions seen through nonlinear quotient and spectral coordinates.

Share this:

Related

Leave a comment Cancel reply