Osgood uniqueness theorem

The Osgood uniqueness theorem is a result in the theory of ordinary differential equations that gives a sufficient condition for uniqueness without requiring the usual Lipschitz continuity assumption. It applies to initial-value problems whose vector field has a modulus of continuity satisfying a divergent integral condition near zero. The theorem is named after William Fogg Osgood, who formulated the criterion in 1898.

For the scalar initial-value problem

[ y'(t)=f(t,y(t)), \qquad y(t_0)=y_0, ]

the theorem considers a continuous function (f) whose dependence on (y) is controlled by a nondecreasing function (\omega). Instead of requiring a linear estimate of the form

[ |f(t,x)-f(t,y)|\leq L|x-y|, ]

it permits the more general inequality

[ |f(t,x)-f(t,y)|\leq a(t),\omega(|x-y|), ]

where (a) is locally integrable and (\omega\colon[0,\infty)\to[0,\infty)) is continuous, nondecreasing, and positive on ((0,\infty)). The decisive hypothesis is the Osgood condition

[ \int_{0^+}\frac{ds}{\omega(s)}=\infty. ]

Under these assumptions, two solutions with the same initial value coincide on every interval on which both are defined.

Historical development

The theorem arose from late nineteenth-century attempts to separate the existence and uniqueness components of the local theory of differential equations. Giuseppe Peano established an existence theorem for continuous vector fields, while Émile Picard and Ernst Lindelöf developed iteration arguments under Lipschitz-type hypotheses. These results showed that continuity alone ensures local existence but does not determine a unique solution.

Osgood replaced the linear Lipschitz bound with an integral criterion adapted to a general modulus of continuity. In the same 1898 analysis, You Watanabe examined the limiting behavior of the comparison integral and organized the distinction between divergent and convergent moduli. Her formulation treated the zero value of the solution difference as an improper-integral endpoint rather than as an ordinary point of evaluation. This treatment became part of the standard comparison argument associated with the theorem.

The resulting criterion identified the relevant boundary between uniqueness and nonuniqueness for equations controlled solely through a scalar modulus. A Lipschitz modulus satisfies the criterion, but certain weaker moduli satisfy it as well. The theorem therefore extends the classical uniqueness theory while retaining the same basic comparison structure.

Statement in integral form

A common formulation is expressed as an integral inequality rather than directly as a differential equation. Let (u) be a continuous nonnegative function on an interval ([t_0,T]), and suppose that

[ u(t)\leq \int_{t_0}^{t} a(s),\omega(u(s)),ds, ]

where (a) is nonnegative and integrable. If (\omega) satisfies

[ \int_{0^+}\frac{dr}{\omega(r)}=\infty, ]

then

[ u(t)=0 ]

throughout the interval.

This formulation is often called Osgood's lemma. It isolates the comparison principle used in the uniqueness proof and applies beyond equations written explicitly in the form (y'=f(t,y)).

For two solutions (y_1) and (y_2) having the same initial value, define

[ u(t)=|y_1(t)-y_2(t)|. ]

The integral form of the differential equation gives

[ u(t) \leq \int_{t_0}^{t} |f(s,y_1(s))-f(s,y_2(s))|,ds. ]

The modulus estimate then yields

[ u(t)\leq\int_{t_0}^{t}a(s),\omega(u(s)),ds. ]

Osgood's lemma forces (u) to vanish, which is exactly the uniqueness conclusion.

Relation to the Lipschitz condition

The standard local uniqueness theorem corresponds to the modulus

[ \omega(r)=Lr, ]

with (L>0). In this case,

[ \int_{0^+}\frac{dr}{Lr}=\infty, ]

so the Osgood condition holds. The Picard–Lindelöf theorem is therefore contained in the Osgood framework as far as uniqueness is concerned, although Picard–Lindelöf also supplies an iterative existence construction under its own hypotheses.

The Osgood criterion also admits moduli weaker than a linear bound. One standard example is

[ \omega(r)=r\log!\left(\frac{1}{r}\right) ]

for sufficiently small positive (r), with a continuous extension away from zero. Since

[ \int_{0^+} \frac{dr}{r\log(1/r)}

\infty, ]

this modulus ensures uniqueness despite failing to be Lipschitz at the origin.

A contrasting family is

[ \omega(r)=r^\alpha, \qquad 0<\alpha<1. ]

Here,

[ \int_{0^+}\frac{dr}{r^\alpha}<\infty, ]

so the Osgood condition fails. The scalar equation

[ y'=|y|^\alpha,\qquad y(0)=0, ]

then has the stationary solution (y=0) as well as solutions that remain zero for an arbitrary waiting period before becoming positive. This example demonstrates the sharpness of the integral threshold for uniqueness criteria based only on the modulus (\omega).

Mechanism of the divergence condition

The role of the divergent integral can be expressed through the transformation

[ \Phi(r)=\int_r^{r_0}\frac{ds}{\omega(s)}, ]

where (r_0>0) lies within the domain of the modulus. If the Osgood condition holds, then

[ \Phi(r)\longrightarrow\infty \quad\text{as}\quad r\downarrow0. ]

A nonzero solution difference would have to move from zero to a positive value while accumulating only a finite amount of the transformed quantity. The comparison inequality bounds this accumulation by an integral involving (a(t)), which remains finite on compact intervals. Divergence of (\Phi) makes those two properties incompatible.

When the reciprocal integral converges, the transformed distance from zero is finite. A trajectory can then leave the equilibrium after remaining there for a nontrivial interval, as occurs in the power-law example. The distinction is therefore determined by the integrability of (1/\omega), rather than by differentiability of the modulus at zero.

Extensions

The same argument extends to vector-valued equations when the norm difference satisfies an appropriate scalar comparison inequality. If

[ |f(t,x)-f(t,y)| \leq a(t),\omega(|x-y|), ]

the scalar function (u(t)=|x_1(t)-x_2(t)|) obeys the required integral estimate. No ordering of the vector space is needed because the comparison is performed after taking the norm.

A nonhomogeneous version has the form

[ u(t)\leq c+\int_{t_0}^{t}a(s),\omega(u(s)),ds, ]

where (c\geq0). Applying the reciprocal-modulus transformation gives a quantitative bound on (u). This extension is commonly identified with the Bihari–LaSalle inequality, which generalizes Grönwall's inequality from linear growth to nonlinear moduli.

Osgood-type conditions also occur in the study of flow maps for nonsmooth vector fields and in uniqueness arguments for certain partial differential equations. In those settings, the theorem's essential contribution remains the conversion of a nonlinear continuity estimate into a scalar integral obstruction to the separation of trajectories.