What Symmetry Preserves

How invariance generates conservation, from physics to computation

In 1918, Emmy Noether proved something that changed physics: every continuous symmetry corresponds to a conservation law. This wasn't just a technical result. It revealed that the conservation laws we took as fundamental — energy, momentum, angular momentum — weren't separate principles at all. They were consequences of deeper facts about symmetry.

Time translation invariance means energy conservation. Space translation invariance means momentum conservation. Rotational invariance means angular momentum conservation. The conservation law emerges from the symmetry. It couldn't be otherwise.

This pattern extends beyond physics. Wherever systems have structure, invariants follow from transformations that leave that structure unchanged.

What Symmetry Is

Hermann Weyl offered the cleanest definition: a thing is symmetrical if there's something you can do to it such that it looks the same after the operation.

This definition has two parts. A symmetry is always a transformation (something you can do) combined with invariance (nothing observable changes). No transformation, no symmetry. No invariance, no symmetry.

Rotate a square by 90 degrees and it looks the same. The transformation is rotation; the invariant is appearance. Flip a symmetric graph and it maps onto itself. The transformation is reflection; the invariant is structure.

Continuous symmetries are the ones Noether's theorem addresses. These are transformations that can be made arbitrarily small. You can translate by a nanometer or by a mile. You can rotate by one degree or by ninety. Discrete symmetries — reflection, exchange, parity — have different properties. Noether requires smoothness.

How Conservation Emerges

The derivation works like this: take a system with Lagrangian L. If the action S = ∫L dt is invariant under a continuous transformation (meaning L changes by at most a total derivative), then there exists a quantity Q that remains constant along any valid trajectory.

The logic is: because nothing distinguishes one value of the transformed coordinate from another, the system has no way to "choose" how that coordinate changes. Energy cannot be created or destroyed because shifting time cannot change what the system is. A translation in time is just relabeling moments — if the laws don't distinguish moments, there's no mechanism to favor one energy over another.

The conservation law isn't stated and then derived. It's generated by the structure of the theory. The symmetry is the generator; the conserved quantity is the consequence.

Gauge Symmetries and Redundancy

Some symmetries are "real" — they reflect physical invariances. Others are descriptive redundancies.

Gauge symmetries in electromagnetism don't correspond to anything physical. The vector potential A_μ has more degrees of freedom than the electromagnetic field. We can shift A_μ by a gradient without changing E and B. This isn't a physical symmetry — it's a redundancy in our description.

The physics must be gauge-invariant because we've introduced extra unphysical degrees of freedom. The conservation law that follows (charge conservation) is real, but it's generated by the requirement that physics cannot depend on our arbitrary choice of gauge.

This distinction matters: not every symmetry is a deep insight about the world. Some symmetries are insights about our descriptions.

Symmetry Breaking and Structure

If symmetries generate conservation laws, broken symmetries generate structure.

Spontaneous symmetry breaking occurs when the laws are symmetric but the ground state isn't. A ferromagnet has rotationally symmetric dynamics — there's no preferred direction in the physics. But below the Curie temperature, the spins align in some direction. The system "chooses" a configuration. The symmetry is broken.

This is how phases emerge. Critical points. Phase transitions. Order from disorder, because symmetry breaking gives the system a way to be different.

The broken symmetry doesn't disappear — it gives rise to Nambu-Goldstone modes, massless excitations that reflect the underlying invariance even though it's hidden.

Invariants as Conserved Quantities

The correspondence is direct: conservation laws in physics are invariants in computation. They're things that don't change while everything else does.

Loop invariants in Hoare logic: conditions that hold before and after each iteration. Type invariants: properties guaranteed by the type system across all operations. Memory safety invariants: properties preserved by allocation and deallocation.

These aren't arbitrary requirements imposed from outside. They're what the structure of the system generates. If your program has a loop, there's something that stays the same across iterations. That something is the invariant. If your type system is well-designed, there are values that cannot be constructed in certain ways. Those constraints are invariants.

Symmetry in Data

Data structures exploit symmetry. A balanced tree uses rotations that preserve the in-order traversal. The tree shape changes; the sorted sequence stays the same. A hash table maps keys to bins in a way that's symmetric in the sense of determinism: same key, same bin, every time.

Canonical forms are the ultimate symmetry breaking. Take a set of equivalent configurations and choose one representative. The canonical form breaks the symmetry — it picks a specific element from each equivalence class. But then you can compare equivalence by comparing canonical forms. You've traded symmetric structure for efficient comparison.

Sorting breaks permutation symmetry. Once sorted, the list is invariant under fewer transformations. But you've gained efficient search. You've given up symmetry for structure.

What This Reveals

Noether's theorem tells us that conservation laws aren't arbitrary. They follow from the structure of the theory. The same pattern appears in computation.

Invariants aren't things we impose because we want them. They're things that follow from the structure of our systems. A loop has an invariant because the loop structure is invariant under iteration — each iteration transforms the state, but something about the transformation pattern remains fixed.

Symmetry constraints what can happen. If a system is invariant under some transformation, then there are things that transformation cannot change. Conservation is what the symmetry preserves.

The insight for computation: when you find an invariant, you've found a kind of conservation law. When you build a system with symmetry, you're generating invariants whether you know it or not. The question isn't "how do I constrain this system?" but "what symmetries does my system already have, and what do they preserve?"

The deepest symmetries in physics generate the most fundamental conservation laws. Energy, momentum, charge — these are the things that must be conserved because the structure of the universe cannot distinguish. In computation, the deepest invariants come from the most fundamental symmetries of our representations. Pointers preserve the invariant of identity. Functions preserve the invariant of input-output mapping.

Symmetry tells us what doesn't change when everything else does. The invariant is the shadow of the symmetry.