This paper studies a class of coupled oscillators whose interactions conserve the phase volume. Conserving phase volume means that, as the system evolves, the total “amount of phase space” occupied by nearby states does not shrink or grow. The author, Arkady Pikovsky, derives a simple structural condition for pairwise oscillator couplings that guarantees this conservation. He calls the generating function for a pairwise conservative interaction a pair‑Hamiltonian.

For two oscillators the condition reduces to a single partial‑derivative identity. If x and y are the two phases and f(x,y), g(x,y) are their coupling terms, phase‑volume conservation implies ∂xf+∂yg=0. This relation can be solved by introducing a function h(x,y) with f=∂yh and g=−∂xh. The function h is the pair‑Hamiltonian. Two commonly used coupling types fit this form. In Winfree‑type coupling h factors as A(x)B(y). In Kuramoto‑Daido type coupling h depends only on the phase difference, h(x,y)=H(x−y).

The paper then generalizes the idea to networks of N oscillators. Each pair of oscillators can have its own pair‑Hamiltonian hkj(xk,xj), so the network has N(N−1)/2 such functions. For Winfree‑type networks an antisymmetric coupling matrix is needed to keep the whole network conservative. For Kuramoto‑type networks the coupling splits into even and odd parts in the phase difference; this leads to representations with symmetric and antisymmetric matrices or with complex anti‑Hermitian coupling matrices. The author also shows how simple regular networks fit this frame: a globally coupled cosine interaction is conservative, and nearest‑neighbor lattice couplings can be written in Hamiltonian form.

A key dynamical consequence is that conservative phase dynamics cannot have attractors or repellers. Instead the systems display quasiperiodic motion and robust chaos, depending on initial conditions. The paper illustrates this with two‑oscillator and three‑oscillator examples. A Poincaré map for three Winfree‑type oscillators shows coexistence of regular tori and chaotic sets. For large networks the Lyapunov spectrum (which measures the rates of separation of nearby trajectories) is nearly symmetric, resembling Hamiltonian systems, although exact pairwise symmetry of exponents does not hold here.