Pais-Uhlenbeck Scalar Models

• Pais-Uhlenbeck scalar models are higher-derivative theories defined by fourth-order equations that factor into two harmonic oscillators with distinct frequencies.
• They serve as a standard framework for studying Ostrogradsky instabilities, ghost degrees of freedom, and alternative Hamiltonian formulations in both mechanical and field contexts.
• Applied in cosmology, these models elucidate phenomena such as phantom-divide crossing and de Sitter stability by linking fast higher-derivative dynamics to slow effective flows.

Pais–Uhlenbeck scalar models are higher-derivative scalar theories whose equations of motion are fourth order in time or, covariantly, involve higher powers of the d’Alembertian. In their standard non-degenerate form they are governed by operators such as $(\Box+m_1^2)(\Box+m_2^2)$, so that each spatial Fourier mode behaves as a Pais–Uhlenbeck (PU) oscillator with two frequencies. They were introduced as toy models for higher-derivative field theory, but they have become a standard arena for analyzing Ostrogradsky instability, ghost degrees of freedom, alternative Hamiltonian structures, non-standard quantization, and cosmological dynamics (Turner, 30 Sep 2025, Jiménez et al., 2012).

1. Mechanical core and defining equations

The basic mechanical prototype is the fourth-order PU oscillator with Lagrangian

$$L=\frac{1}{2}\Big[\ddot q^2-(\omega_1^2+\omega_2^2)\dot q^2+\omega_1^2\omega_2^2 q^2\Big],$$

or equivalently $L=\frac12(\beta q^2-\alpha \dot q^2+\ddot q^2)$ with $\alpha=\omega_1^2+\omega_2^2$ and $\beta=\omega_1^2\omega_2^2$. Its Euler–Lagrange equation is

$q^{(4)}+\alpha \ddot q+\beta q=0$

or, in factorized form,

$$\left(\frac{d^2}{dt^2}+\omega_1^2\right)\left(\frac{d^2}{dt^2}+\omega_2^2\right)q=0.$$

The general non-degenerate solution is a superposition of two harmonic modes with frequencies $\omega_1$ and $\omega_2$, and the free classical motion is bounded and purely oscillatory despite the higher derivatives (Turner, 30 Sep 2025).

This fourth-order system is the canonical mechanical model for higher-order dynamics in the Hamilton–Ostrogradski sense. For a Lagrangian $L(q,\dot q,\ddot q)$, one introduces $$L=\frac{1}{2}\Big[\ddot q^2-(\omega_1^2+\omega_2^2)\dot q^2+\omega_1^2\omega_2^2 q^2\Big],$$0, $$L=\frac{1}{2}\Big[\ddot q^2-(\omega_1^2+\omega_2^2)\dot q^2+\omega_1^2\omega_2^2 q^2\Big],$$1, $$L=\frac{1}{2}\Big[\ddot q^2-(\omega_1^2+\omega_2^2)\dot q^2+\omega_1^2\omega_2^2 q^2\Big],$$2, and $$L=\frac{1}{2}\Big[\ddot q^2-(\omega_1^2+\omega_2^2)\dot q^2+\omega_1^2\omega_2^2 q^2\Big],$$3, obtaining a four-dimensional phase space whose Hamilton equations reproduce the original fourth-order equation (Melo et al., 19 May 2026). In the degenerate case $$L=\frac{1}{2}\Big[\ddot q^2-(\omega_1^2+\omega_2^2)\dot q^2+\omega_1^2\omega_2^2 q^2\Big],$$4, solutions acquire terms linear in time,

$$L=\frac{1}{2}\Big[\ddot q^2-(\omega_1^2+\omega_2^2)\dot q^2+\omega_1^2\omega_2^2 q^2\Big],$$5

reflecting the double-pole structure characteristic of the equal-frequency limit (Jiménez et al., 2012).

2. From oscillator to scalar field

PU scalar models are obtained by promoting the oscillator coordinate to a field. A standard quadratic action is

$$L=\frac{1}{2}\Big[\ddot q^2-(\omega_1^2+\omega_2^2)\dot q^2+\omega_1^2\omega_2^2 q^2\Big],$$6

with field equation

$$L=\frac{1}{2}\Big[\ddot q^2-(\omega_1^2+\omega_2^2)\dot q^2+\omega_1^2\omega_2^2 q^2\Big],$$7

After spatial Fourier expansion,

$$L=\frac{1}{2}\Big[\ddot q^2-(\omega_1^2+\omega_2^2)\dot q^2+\omega_1^2\omega_2^2 q^2\Big],$$8

each mode satisfies

$$L=\frac{1}{2}\Big[\ddot q^2-(\omega_1^2+\omega_2^2)\dot q^2+\omega_1^2\omega_2^2 q^2\Big],$$9

so every Fourier mode is a PU oscillator with shifted frequencies (Jiménez et al., 2012). This is why the one-dimensional PU model is repeatedly described as the “mechanical core” of higher-derivative scalar field theory (Turner, 30 Sep 2025).

A particularly important degenerate field model is

$L=\frac12(\beta q^2-\alpha \dot q^2+\ddot q^2)$0

whose field equation is $L=\frac12(\beta q^2-\alpha \dot q^2+\ddot q^2)$1 (Jiménez et al., 2012). In cosmological applications, a closely related higher-derivative scalar action is

$L=\frac12(\beta q^2-\alpha \dot q^2+\ddot q^2)$2

which is explicitly described as inspired by the PU oscillator and leads, after field redefinitions, to a two-field quintom-like system containing one normal mode and one ghostlike mode (Pulgar et al., 2014).

3. Ostrogradsky instability, ghosts, and quantization

In the standard Ostrogradsky formulation, the PU Hamiltonian is

$L=\frac12(\beta q^2-\alpha \dot q^2+\ddot q^2)$3

Because it is linear in $L=\frac12(\beta q^2-\alpha \dot q^2+\ddot q^2)$4, it is unbounded from below. In equivalent oscillator variables, the system can be written as a sum of two decoupled oscillators with alternating sign, so that one mode carries positive energy and the other negative energy. This is the classical Ostrogradsky instability and the standard source of the ghost problem in PU scalar models (Turner, 30 Sep 2025).

At the quantum level, standard canonical quantization yields the familiar dilemma. In one operator realization, both modes have positive energy but one-particle states in one sector acquire negative norm. In another, the norm is positive but the spectrum is unbounded from below. For the fourth-order PU model this appears explicitly in the commutators

$L=\frac12(\beta q^2-\alpha \dot q^2+\ddot q^2)$5

and in a Hamiltonian of the form

$L=\frac12(\beta q^2-\alpha \dot q^2+\ddot q^2)$6

so one must choose between negative norms and negative energies in the standard framework (Turner, 30 Sep 2025).

A different line of work argues that this conclusion is representation-dependent. In the $L=\frac12(\beta q^2-\alpha \dot q^2+\ddot q^2)$7-symmetric formulation of the non-degenerate fourth-order PU oscillator, the Hamiltonian is treated as non-Dirac-Hermitian but $L=\frac12(\beta q^2-\alpha \dot q^2+\ddot q^2)$8-symmetric, an exact $L=\frac12(\beta q^2-\alpha \dot q^2+\ddot q^2)$9 operator is constructed, and a similarity transformation maps the model to a manifestly Hermitian Hamiltonian

$\alpha=\omega_1^2+\omega_2^2$0

With the inner product $\alpha=\omega_1^2+\omega_2^2$1, the spectrum is real and bounded below, and time evolution is unitary (0706.0207). This is one of the principal controversies surrounding PU scalar models: whether the ghost is an unavoidable physical degree of freedom or an artifact of the canonical structure and inner product.

4. Bi-Hamiltonian structure and equivalent representations

Recent work has emphasized that the free non-degenerate PU oscillator is bi-Hamiltonian. Besides the Ostrogradsky pair $\alpha=\omega_1^2+\omega_2^2$2, there exists a second Poisson tensor $\alpha=\omega_1^2+\omega_2^2$3 and a second Hamiltonian $\alpha=\omega_1^2+\omega_2^2$4 generating the same dynamical vector field. One explicit choice is

$\alpha=\omega_1^2+\omega_2^2$5

with

$\alpha=\omega_1^2+\omega_2^2$6

This yields a one-parameter family of dynamically equivalent Hamiltonian structures,

$\alpha=\omega_1^2+\omega_2^2$7

with a corresponding $\alpha=\omega_1^2+\omega_2^2$8, all producing the same classical equation of motion (Turner, 30 Sep 2025).

A crucial consequence is that $\alpha=\omega_1^2+\omega_2^2$9 can be written as a sum of squares,

$\beta=\omega_1^2\omega_2^2$0

so for non-degenerate frequencies and appropriate $\beta=\omega_1^2\omega_2^2$1, the free PU dynamics admits a positive-definite Hamiltonian description (Turner, 30 Sep 2025). The same work classifies all linear embeddings of the PU model into two-dimensional first-order systems, identifying four families $\beta=\omega_1^2\omega_2^2$2, $\beta=\omega_1^2\omega_2^2$3, $\beta=\omega_1^2\omega_2^2$4, and $\beta=\omega_1^2\omega_2^2$5; among these, $\beta=\omega_1^2\omega_2^2$6 and $\beta=\omega_1^2\omega_2^2$7 are the useful embeddings, while $\beta=\omega_1^2\omega_2^2$8 and $\beta=\omega_1^2\omega_2^2$9 are singular or degenerate (Turner, 30 Sep 2025).

A complementary symmetry-based treatment derives the relevant Hamiltonians from Lie symmetries of the fourth-order equation. In that construction the PU flow admits four commuting symmetry generators $q^{(4)}+\alpha \ddot q+\beta q=0$0, and $q^{(4)}+\alpha \ddot q+\beta q=0$1 acts as a ladder on the hierarchy of conserved Hamiltonians. The resulting picture is that the free PU model possesses several Poisson-bracket formulations preserving the same dynamics, including positive-definite representations, but these are nontrivially tied to the choice of symplectic structure (Felski et al., 9 May 2025). A plausible implication is that “ghost-free” formulations of free PU scalar models are classical-equivalence statements unless the associated nonstandard Poisson structure is also adopted at the quantum level.

5. Interactions, gauge embeddings, and stability

For generic interactions, the favorable free-theory structure is fragile. When a potential $q^{(4)}+\alpha \ddot q+\beta q=0$2 is added to the PU Lagrangian, the interacting vector field

$q^{(4)}+\alpha \ddot q+\beta q=0$3

typically admits only the Ostrogradsky Poisson structure $q^{(4)}+\alpha \ddot q+\beta q=0$4; the second Poisson tensor $q^{(4)}+\alpha \ddot q+\beta q=0$5 does not survive. In this sense the bi-Hamiltonian structure of the free PU oscillator is broken by generic interactions, and with it the freedom to choose a positive-definite dynamically equivalent Hamiltonian (Turner, 30 Sep 2025). Closely related symmetry analysis reaches the same conclusion: interactions usually destroy the alternative Hamiltonian realizations that stabilize the free theory (Felski et al., 9 May 2025).

This does not mean that every interacting PU model is uniformly catastrophic. In the quartically self-interacting mechanical model,

$q^{(4)}+\alpha \ddot q+\beta q=0$6

the Hamiltonian necessarily becomes an indefinite two-oscillator Hamiltonian with interaction, and positive-definite free-theory representations no longer apply (Pavšič, 2013). Even so, numerical analysis shows “islands of stability” in parameter space. Moreover, replacing the quartic interaction by a bounded sine interaction, or allowing unequal masses in the two-oscillator formulation, produces “continents” of stability extending from zero to infinity in parameter space (Pavšič, 2013). This suggests that higher-derivative instability in PU-type dynamics is sensitive to the detailed interaction sector rather than determined solely by the sign structure of the free Hamiltonian.

A distinct route to consistency arises in the degenerate field model $q^{(4)}+\alpha \ddot q+\beta q=0$7 when $q^{(4)}+\alpha \ddot q+\beta q=0$8 is interpreted as a Stückelberg scalar for a vector theory,

$q^{(4)}+\alpha \ddot q+\beta q=0$9

In Lorenz gauge, the scalar obeys $$\left(\frac{d^2}{dt^2}+\omega_1^2\right)\left(\frac{d^2}{dt^2}+\omega_2^2\right)q=0.$$0, and the gauge-invariant quantity is $$\left(\frac{d^2}{dt^2}+\omega_1^2\right)\left(\frac{d^2}{dt^2}+\omega_2^2\right)q=0.$$1 rather than $$\left(\frac{d^2}{dt^2}+\omega_1^2\right)\left(\frac{d^2}{dt^2}+\omega_2^2\right)q=0.$$2 itself (Jiménez et al., 2012). The same framework defines “natural” interactions with charged scalars and fermions through covariant derivatives involving $$\left(\frac{d^2}{dt^2}+\omega_1^2\right)\left(\frac{d^2}{dt^2}+\omega_2^2\right)q=0.$$3; because the matter current is conserved, the scalar equation remains $$\left(\frac{d^2}{dt^2}+\omega_1^2\right)\left(\frac{d^2}{dt^2}+\omega_2^2\right)q=0.$$4 even in the interacting theory (Jiménez et al., 2012). In that restricted class, a Gupta–Bleuler–type condition removes the gauge mode, and the physical spectrum retains only the gauge-invariant sector.

Time dependence introduces another obstruction to simple decoupling. For the fourth-order PU oscillator with time-dependent frequencies, the usual Smilga transformation no longer gives two decoupled oscillators. The transformation becomes explicitly time dependent, extra interaction terms proportional to $$\left(\frac{d^2}{dt^2}+\omega_1^2\right)\left(\frac{d^2}{dt^2}+\omega_2^2\right)q=0.$$5 appear, and exact decomposition into independent modes fails (Kuwabara et al., 2015). A plausible implication is that PU scalar fields in time-dependent backgrounds, including cosmological ones, generically exhibit mode mixing even when the static theory is exactly factorizable.

6. Cosmological realizations

In FLRW cosmology, PU-inspired scalar models are commonly based on

$$\left(\frac{d^2}{dt^2}+\omega_1^2\right)\left(\frac{d^2}{dt^2}+\omega_2^2\right)q=0.$$6

which yields fourth-order background equations and can be recast as a two-field system by defining $$\left(\frac{d^2}{dt^2}+\omega_1^2\right)\left(\frac{d^2}{dt^2}+\omega_2^2\right)q=0.$$7 and $$\left(\frac{d^2}{dt^2}+\omega_1^2\right)\left(\frac{d^2}{dt^2}+\omega_2^2\right)q=0.$$8 (Pulgar et al., 2014). In that representation one field has the canonical sign and the other the opposite sign, so the model becomes a specific quintom system. For $$\left(\frac{d^2}{dt^2}+\omega_1^2\right)\left(\frac{d^2}{dt^2}+\omega_2^2\right)q=0.$$9, the analysis provides conditions for the stability of de Sitter solutions; the phantom divide $\omega_1$0 is crossed once, after which the equation-of-state parameter remains below $\omega_1$1 and asymptotically approaches de Sitter from below (Pulgar et al., 2014). For $\omega_1$2, cyclic behavior and multiple phantom crossings can occur, and the parameter space can be separated into regions with benign or malicious ghost behavior in Smilga’s sense (Pulgar et al., 2014).

A more recent slow–fast treatment rewrites the higher-order cosmological system in Hubble-normalized variables and identifies a singular surface

$\omega_1$3

For small $\omega_1$4, the dynamics splits into a fast sector and a slow manifold on which the effective evolution reduces to the usual Klein–Gordon equation $\omega_1$5 (Gonzalez-Espinoza et al., 25 Jul 2025). This means that the higher-derivative PU sector can be understood as a fast transient around a lower-order slow cosmological flow. The same work analyzes phase-space evolution for exponential and power-law potential families, studies center-manifold stability of de Sitter points, and emphasizes that de Sitter stability depends on the detailed behavior of $\omega_1$6 near $\omega_1$7 rather than on $\omega_1$8 alone (Gonzalez-Espinoza et al., 25 Jul 2025).

Within this cosmological literature, PU scalar models are used to model both early- and late-time acceleration. The inclusion of radiation and dust permits realistic scenarios with a transient matter-dominated era and late-time accelerated expansion, while the higher-order sector allows phantom-divide crossing, non-smooth transitions near singular surfaces, and slow–fast bifurcation phenomena (Gonzalez-Espinoza et al., 25 Jul 2025). This suggests that PU scalar cosmology is best viewed not as a single model class with a uniform ghost verdict, but as a structured family of higher-derivative systems whose physical viability depends on the free-versus-interacting distinction, the availability of alternative Hamiltonian structures, and the role of gauge or cosmological background dynamics.