Degrees of Freedom: Why Beginners Move Stiff and Experts Move Loose
A beginner and an expert doing the same skill are solving two different control problems. The beginner is trying to manage a body that has more independent parts than they can handle, so they lock most of them down. The expert has stopped managing most of those parts directly and lets them organize themselves. That shift, from rigid control to organized freedom, is a lot of what “getting better” physically is.
This is Bernstein’s degrees of freedom problem, and it’s the useful lens here.
The problem
Count the things that can move independently in a freestyle stroke. Shoulder, elbow, wrist, and the joints of the hand, times two arms. Hips, knees, ankles. Spine rotation and flexion. That’s already dozens of joint degrees of freedom. Below that, each joint is driven by multiple muscles, and each muscle by hundreds or thousands of motor units. The controllable variables run into the thousands.
Bernstein’s point, from the 1960s, was that the nervous system cannot possibly send a separate, moment-to-moment command to each of those. The bandwidth isn’t there. So skilled movement can’t be “the brain controlling every part.” It has to be something else. Learning a motor skill is largely the process of solving how to govern all those parts without micromanaging them.
The commonly described solution runs in stages. I’ll lay them out, then flag what’s solid and what isn’t.
Stage 1: Freezing
Early on, learners reduce the problem by rigidly fixing joints or locking them to move together as a block. Fewer independent parts means fewer things to control.
The reference study is Vereijken, van Emmerik, Whiting, and Newell (1992), Journal of Motor Behavior. Adults learned a ski-simulator task (side-to-side oscillation on a platform) over about a week of practice. Early on, joint range of motion was small and cross-correlations between joints were high, meaning segments moved together rather than independently. That’s freezing, measured two ways: low range of motion at individual joints, and high coupling between joints.
In the pool this is the novice who swims like a plank. Stiff arms, rigid trunk, head locked, everything moving as one unit. It looks like a fault. Mechanically it’s a sensible first move by the control system: collapse a thousand-variable problem into a handful.
Stage 2: Freeing
As control improves, learners release those frozen joints. Range of motion at individual joints increases, the rigid couplings loosen, and the parts start contributing independently. In the Vereijken study, by the end of the week joint range of motion had increased and the movement had become larger-amplitude and more effective at the task.
In swimming this is the swimmer whose hips start rotating on their own, whose arm bends to find a catch instead of swinging through straight. More parts are live and doing distinct jobs.
Stage 3: Exploiting passive forces
The third stage, as usually described, is that skilled performers stop fighting external and passive forces and start using them: gravity, momentum, the elastic recoil of stretched muscle and tendon, and for us, the water. The result is that the same outcome is produced with less active muscular force, because passive dynamics carry part of the load.
Two honesty flags here.
First, stages 1 and 2 (freeze then free) are reasonably well supported. The freezing-to-freeing pattern has been replicated across several tasks beyond the ski simulator, including dart throwing, soccer kicking, and racquet skills, mostly measured by joint range of motion and inter-joint coupling. It is not perfectly universal. Some studies don’t find a clean freeze-then-free sequence, and the pattern can depend on the task and how you measure it. So treat it as a strong general tendency, not a law.
Second, the “exploiting passive forces” stage is the most intuitive and the least cleanly nailed down of the three. The claim that active muscular force drops as people learn to use passive dynamics is accepted in principle and shown in specific cases, but I don’t have a single clean swimming EMG number for you, and I’d want to verify a specific citation before we put a figure on it. If you want to make that claim hard, that’s the thing to look up: EMG or metabolic-cost studies showing reduced active force with expertise. Sparrow’s work on metabolic cost of learning is one place to start.
How the system governs the parts: synergies
The mechanism that makes this possible is that the nervous system doesn’t control the parts one at a time. It groups them into units that are controlled together, so a single command drives a coordinated set of joints and muscles in fixed proportions. These go by “coordinative structures” or “muscle synergies.” That’s how a thousand-variable problem becomes a few-variable one.
This is well established as a description. There’s ongoing debate about whether synergies are a real control primitive in the nervous system or mostly a pattern that falls out of the mechanics and the way we analyze it (usually with dimensionality reduction like PCA on EMG). Established: movement can be described with far fewer dimensions than there are muscles. Contested: what that low-dimensionality actually proves about how the brain controls movement.
Which variability matters: the uncontrolled manifold
Related and worth knowing because it reframes “consistency.” The uncontrolled manifold idea (Scholz and Schoner, 1999, Experimental Brain Research) is that a skilled system tightly controls the combinations of variables that affect the task outcome, and lets the combinations that don’t affect the outcome vary freely.
Concretely: what stays stable across reps is the outcome-relevant variable (say, getting the hand into a strong catch position and moving water), while the specific joint configuration used to get there can differ rep to rep. So a good stroke can look slightly different each time in ways that don’t matter, while the part that matters stays locked. Variability in the movement is not automatically error. Some of it is the system correctly not wasting control on things that don’t change the result.
This connects to the “repetition without repetition” point: the target isn’t one fixed pattern, it’s a stable outcome reached through variable means.
What this implies for coaching
I’m labeling this as my inference from the above, not established experimental fact. The research describes how coordination develops; it does not prove these specific coaching moves are optimal.
• Stiffness in a beginner is expected, not a defect to attack immediately. It’s stage 1. Forcing “relax” before they have control fights the process.
• Cue load is a real constraint. If freezing is what the system does when it has too much to manage, adding many simultaneous technical cues gives it more to freeze against. Fewer cues at once is the mechanistically consistent move.
• You probably can’t instruct freeing directly. “Relax your shoulders” is an internal-focus instruction and often backfires. Setting up task conditions (overspeed with fins, drills that make the frozen pattern impossible, constraints) is more likely to produce freeing than telling them to loosen up. This links to the constraints-led approach.
• Don’t over-standardize the stroke. If some rep-to-rep variability is outcome-irrelevant, sanding off every visible difference is wasted effort and may suppress the system’s normal way of solving the movement. Identify the outcome-relevant thing and hold that stable.
• Freezing recurs under stress, fatigue, and novelty, even in advanced athletes. A new skill or a high-pressure race can push a good swimmer back toward rigid coupling. Expect regression when you add difficulty.
