Which details can we leave out?
A small object attached to a spring moves through a viscous liquid. Pull it aside and release it. With little resistance it oscillates; with sufficiently strong resistance it returns without crossing the spring’s resting position. If we film the strongly resisted motion, most frames show a slow approach. The brief interval immediately after release may occupy only a few frames, yet it contains a change that matters to our description: the object starts at rest and acquires a velocity.
Suppose we want to predict its position several seconds later. Must we reconstruct that initial adjustment? A complete account within classical mechanics keeps track of both position and velocity. A simpler account might use position alone. Whether this is possible depends on how the two quantities evolve, on the preparation of the object, and on which part of the motion we intend to resolve.
Two processes in the same motion
Let be displacement from the spring’s resting position and the velocity. Assume the spring exerts a force , with positive stiffness , and the liquid exerts a resistance , with positive drag coefficient . The minus signs express the directions: the spring pulls toward its resting position, while resistance opposes velocity. We restrict attention to conditions where these linear force laws are adequate, neglect random thermal motion, and hold the properties of the liquid fixed.
The object’s mass is . Newton’s equation and the definition of velocity give
Here a derivative means an instantaneous rate of change. The first equation says how the forces change velocity; the second says how velocity changes position. The spring provides a restoring feedback through position, and drag opposes motion through velocity. Their combined action transfers mechanical energy into the surrounding liquid.
First imagine removing the spring. The remaining equation is : drag removes the same fraction of velocity in each sufficiently short interval. Velocity therefore decays exponentially, with a relaxation time
After this interval, approximately 37 per cent of the initial velocity remains. This is a useful measure of how quickly drag erases an initial motion. With the spring present, velocity need not decay toward zero on this time scale, because the spring continues to pull. If position hardly changes during the adjustment, velocity instead approaches the value at which spring force and resistance balance:
At a given displacement, a velocity more negative than this value produces excessive resistance in the positive direction and is pushed back toward balance. A less negative velocity leaves a net spring force that makes it more negative. The balance is thus approached dynamically. To use it as an approximation throughout the return, we must check whether it can be approached before the displacement changes appreciably.
Comparing the clocks
Substituting the balanced velocity into the position equation gives
This reduced rule predicts another exponential decay, now of displacement, with a time scale . It gives a precise form to the proposed slow return. The approximation asks that velocity adjust quickly compared with this return:
The double inequality means that one scale is much smaller than the other. The symbol names their ratio. Because both numerator and denominator are times, the ratio has no units. It is unchanged if we describe the same experiment in seconds or minutes.
This is why a statement such as “the drag coefficient is large” is incomplete. Its numerical value depends on the units, and its effect also depends on the mass and spring stiffness. The ratio compares the processes that compete within the experiment. A heavier object retains its initial velocity longer; a stiffer spring changes its position more quickly. Either change can spoil the separation even if the drag coefficient stays fixed.
Increasing drag has two effects that initially seem difficult to reconcile. It shortens , so velocity adjusts more rapidly, while it lengthens , so position returns more slowly. Strong resistance both suppresses the memory of the initial velocity and prolongs the visible motion. These refer to different processes, measured by different clocks.
For a numerical example, take kilogram, newton per metre and newton-seconds per metre. The estimated time scales are 0.1 seconds and 10 seconds, with . The full equation has two exponential components, whose decay times are approximately 0.101 and 9.899 seconds. The simple estimates recover both scales closely; the small discrepancy is part of the approximation.
Calling a motion overdamped means that its two decay rates are real and it does not have oscillatory modes. For this linear model that requires . A useful separation demands more: a system only just beyond the boundary can have two quite similar decay times. The absence of oscillations alone does not justify discarding velocity.
What happens to the initial velocity?
Release the object at displacement with zero velocity. The full model starts with a horizontal tangent on a graph of position against time, because the initial rate of position change is zero. The reduced model predicts an immediate velocity . They disagree at the instant of release, however small we make the ratio of time scales.
The reason can be traced to the information each description accepts. The full model permits independent choices of initial position and initial velocity. The reduced rule assigns a velocity to every position, so it cannot satisfy an arbitrary second choice. During the short initial adjustment, the full trajectory moves toward the relation between position and velocity that the reduced description assumes from the outset. This interval is often called an initial layer.
For the numerical example, a release from one metre at rest gives a displacement of approximately 0.9132 metres after one second according to the full model. The reduced prediction is approximately 0.9048 metres. After ten seconds the values are approximately 0.3679 and 0.3679 metres; their closeness at this particular instant should not be mistaken for an exact agreement at every later time. The two models have slightly different slow decay rates.
A small error in position can coexist with a substantial error in the early velocity. If our experiment measures the force immediately after release, or resolves the first tenth of a second, the omitted adjustment is precisely what we need to study. If it measures position on the scale of several seconds with a tolerance of a few centimetres, the reduced rule may be sufficient. An approximation has to be assessed against an observable, an interval and an acceptable error.
The preparation also matters. Give the object an exceptionally large initial velocity, and it can travel a substantial distance before drag suppresses that velocity. Over a fast adjustment during which the spring’s influence is small, integrating the drag-only velocity gives a displacement of order , where is the initial velocity. If that distance is comparable to the displacement we wish to predict, simply starting the reduced model at the original position loses essential information. We may need to follow the initial interval explicitly and then initialize the slow description at the position reached afterward. A small cannot by itself make every possible preparation harmless.
The trace of an omitted variable
There is another way to see why time-scale separation matters. Within the full linear model, velocity can be expressed exactly as
The integral adds contributions from earlier times . Each past displacement supplied a spring force; the exponential factor measures how much of its contribution to the present velocity survives drag. Recent forces retain more weight than distant ones. The first term carries the remaining contribution of the initial velocity.
This expression removes velocity as an independently evolved variable, but it makes the position rule depend on its history. Eliminating a variable exactly can therefore leave memory behind. A description written using fewer variables need not be simpler to use if it requires their entire past.
After several velocity relaxation times, the initial term is small. If position varies little over the short interval to which the integral gives appreciable weight, we can replace those recent values by the current value . The weighted interval then contributes approximately , leaving . The reduced rule follows because the relevant memory is short compared with the changes we are tracking. If the time scales are comparable, that replacement has no such justification.
This calculation also clarifies the nearly balanced forces. Exact cancellation would keep velocity constant by Newton’s equation. In the slow return, velocity is still changing, and a small residual force supplies that change. Neglecting inertia means accepting the error from omitting this residual at leading order. It does not mean that the object has literally ceased to possess mass.
A description with a stated range
We can now decide which experiment would test the simplification. Measure both the early adjustment and the slower return, vary mass or resistance, and compare the observed trajectories with the two predictions. The reduction should improve as the time-scale ratio decreases, provided the initial preparation and the observation window remain suitable. A discrepancy that persists might instead reveal a failure of the assumed drag law, a nonlinear spring, or an omitted force. Agreement between a reduced equation and its parent equation checks the reduction; agreement with measurements tests the physical assumptions they share.
The slow description is an effective model: it predicts selected behaviour in a specified regime while omitting processes that the more detailed model retains. Its success is supported by a mechanism for forgetting the omitted information, together with an estimate of how quickly that forgetting occurs. We can then use a single position to follow motion that originally required position and velocity, while retaining a clear account of when the extra variable must return.
In a system with many interacting parts, identifying such a separation can be harder than solving the resulting slow equations. Several relaxation times may remain comparable, or a process that was fast under one set of conditions may slow down as those conditions change. The practical question is which information continues to influence the observation on the time scale we care about. In the spring experiment, we can answer it by watching how an initial velocity fades and measuring how far the object moves before it does.
Letture collegate