Geometry is a subject where examples are not decoration. They are the laboratory where definitions acquire meaning and where theorems reveal their true hypotheses. The most common mistake beginners make is to wait for examples to appear after learning a theory. In geometry, you build them deliberately.

A practical recipe exists. It is not one trick, but a small menu of construction moves, each with predictable effects on curvature, topology, and geodesics. If you know what each move tends to preserve and what it tends to change, you can create examples that test conjectures, break false generalizations, and sharpen the hypotheses of your statements.

The guiding principle: decide what you want to control

Before choosing a construction, decide which invariants are meant to be under your control and which are allowed to vary. In geometry, the main invariants that drive behavior include:

• Topology: connectedness, fundamental group, orientability.
• Metric scale: completeness, compactness, volume growth.
• Curvature: sectional, Ricci, scalar, or Gaussian in dimension two.
• Geodesics: existence of closed geodesics, minimizing properties, conjugate points.
• Symmetry: group actions, homogeneous structure, Killing fields.

Different construction moves are good at controlling different parts of this list.

Construction move: take a quotient by isometries

If $(\widetilde M,\widetilde g)$ has a group $\Gamma$ acting properly discontinuously by isometries, then the quotient $M=\widetilde M/\Gamma$ inherits a metric $g$ such that the projection is a local isometry.

This is the fastest way to create new manifolds with highly controlled local geometry.

A classic example is the flat torus:

Local geometry is Euclidean; global topology changes.

The quotient move tends \to:

• Preserve local curvature properties.
• Introduce global features like closed geodesics and nontrivial loops.
• Produce manifolds with explicit universal covers and deck groups.

A good diagnostic question when you build a quotient is: what does the deck group force about loops and periodicity?

Construction move: take products and warped products

Given Riemannian manifolds $(M,g_M)$ and $(N,g_N)$, the product $M\times N$ carries the product metric

Products are honest constructions: topology, completeness, and many curvature properties can be read from the factors. They are excellent for testing whether a statement that holds in one factor survives adding an innocent extra dimension.

Warped products let you bend the metric by a positive function $f$:

This can create rich curvature behavior while keeping the underlying manifold simple.

Examples built from products tend \to:

• Keep computations separable.
• Reveal which hypotheses are genuinely geometric and which are dimension artifacts.
• Provide controlled families where you can tune one parameter and see which conclusions break.

Construction move: build submanifolds with induced geometry

If $M\subset \mathbb R^k$ is an embedded submanifold, the Euclidean inner product restricts \to a Riemannian metric on $M$. This provides geometry for free along with a powerful ambient viewpoint.

Surfaces with rotational symmetry are a particularly productive class:

• Start with a plane curve $\alpha(s)$ in $(r,z)$-space.
• Rotate it around an axis to get a surface in $\mathbb R^3$.
• Compute metric coefficients and curvature from the profile curve.

You get examples where curvature varies in space, geodesics can be studied qualitatively, and many questions reduce to calculus in one variable.

This move tends \to:

• Make geodesic and curvature computations tangible.
• Provide intuition for intrinsic quantities via extrinsic pictures.
• Produce counterexamples where local curvature bounds do not imply global behavior.

Construction move: glue along boundaries, then smooth

Gluing is how you build topology, but in geometry you must also manage the metric near the seam. A common pattern is:

• Choose manifolds with boundary with compatible metrics near the boundary.
• Identify boundary pieces via an isometry.
• Smooth the resulting metric near the seam using a partition of unity.

This move is useful when you want to create a manifold with a specific topology and then place a metric on it that satisfies a desired local property away from a controlled region.

Gluing tends \to:

• Change global topology in a predictable way.
• Introduce regions where curvature must be managed carefully.
• Produce examples that show why nice local behavior can fail globally due \to a small glued region.

Construction move: change the metric conformally

On a surface, conformal changes are especially powerful. If $g$ is a metric and $u$ is a smooth function, define

This changes distances while preserving angles. In two dimensions, many curvature computations simplify dramatically under conformal changes, so this move is a workhorse for building surfaces with designed curvature profiles.

Conformal changes tend \to:

• Preserve the underlying smooth manifold and its topology.
• Give tunable control over local scale.
• Allow families of metrics that interpolate between behaviors while keeping formulas manageable.

Even when you do not compute explicit curvature formulas, conformal flexibility is a conceptual tool: it explains why some geometric features are not rigid unless you impose global constraints.

Construction move: use bundles and connections for controlled twisting

Fiber bundles let you build manifolds by gluing fibers together over a base space. When the fibers carry geometry, a connection tells you how to compare fibers, creating global twisting effects.

Even without diving into heavy formalism, one takeaway is decisive:

• You can build manifolds whose local geometry looks uniform, but whose global structure encodes nontrivial twisting.

This is one of the clean ways to produce examples where local triviality does not imply global triviality, a theme that appears across geometry.

A compact recipe card for building examples

Here is a practical menu you can consult when you want a new example quickly:

| Goal | Construction move | What it typically preserves | What it typically changes |

|—|—|—|—|

| Keep local curvature but change topology | quotient by isometries | local metric data | global loops, periodic geodesics |

| Build higher-dimensional examples | product metric | many properties from factors | dimension-driven phenomena |

| Tune curvature by a function | warped product | base topology | curvature profiles, geodesic behavior |

| Make computations concrete | submanifold in $\mathbb R^k$ | induced metric structure | global topology depends on embedding |

| Engineer topology with control | gluing and smoothing | pieces away from seam | seam region curvature, global features |

| Change scale without changing angles | conformal change | smooth structure | distances, curvature, completeness |

This table is not exhaustive, but it covers a surprisingly large portion of everyday geometric construction.

How to test whether your example is doing what you think

After building an example, do not trust it until you run a short verification checklist. The quickest checks are:

• Smoothness: are charts and transition maps smooth across identifications and seams?
• Completeness: do geodesics extend for all time, or do they run into a boundary in finite length?
• Compactness: is the space compact, and if so, do you have a way to compute or estimate volume?
• Curvature: do you know whether curvature is bounded, sign-controlled, or variable?
• Geodesics: are there obvious closed geodesics, or obvious obstructions to them?

The point of the checklist is not bureaucracy. It is to prevent the most common failure mode: believing you built a metric object when you actually built only a topological one.

A strong meta-lesson: examples are hypothesis detectors

When a theorem in geometry feels almost true, it usually means one hypothesis is doing all the work. Building examples is how you locate that hypothesis.

• If a local condition seems to imply a global conclusion, try a quotient.
• If a curvature bound seems to imply a topological property, try a gluing construction with a controlled seam.
• If a statement seems dimension-free, test it on a product with a harmless factor.
• If a rigidity statement seems too strong, try a conformal modification.

Each construction move is not only a way to build objects. It is a way to interrogate your assumptions.

A worked mini-example: changing geometry without changing the underlying space

Take the cylinder $M=S^1\times \mathbb R$. Topologically it is simple, but you can put many different metrics on it that behave very differently.

Start with the product metric

which is flat and complete. Now introduce a warped product metric

where $f(z)$ is a positive smooth function.

With this single function $f$, you can control several behaviors:

• If $f$ is constant, you are back to the flat cylinder.
• If $f(z)$ grows rapidly, circles $S^1\times\{z\}$ become longer as $|z|$ increases, and the cylinder “opens out” in the angular direction.
• If $f(z)$ shrinks toward zero as $z$ approaches a finite value, you can create an incomplete metric where geodesics hit a degeneration region in finite length.

This example illustrates why geometry is not only about the manifold but also about the chosen metric. The same topological space can support metrics with different completeness behavior and different curvature profiles.

The practical moral is that “build an example” often means “build a metric,” and warped products are one of the fastest ways to do that while keeping computations manageable.

A caution about quotients: avoid accidental singularities

Quotients are powerful, but there is a common pitfall: if the group action has fixed points, the quotient may fail to be a manifold. You can still get a meaningful geometric object, but it behaves differently and often requires extra language.

A safe practice when you want a genuine manifold:

• Check that the action is free.
• Check that it is properly discontinuous.
• Confirm that local neighborhoods descend smoothly.

Doing this early prevents examples that accidentally smuggle in singular behavior you did not intend.

What you should carry forward

The best geometry examples are designed, not stumbled upon. Once you have a small set of construction moves and a habit of checking invariants, you can generate families of spaces that answer questions quickly and honestly.

Geometry rewards this discipline: it turns vague intuition into concrete objects, and it turns concrete objects into clearer theorems.

Connections are the organizing technology of modern geometry. They let you differentiate vector fields without pretending tangent spaces at different points are the same, and they turn “geometry” into a calculus of transport, curvature, and invariants.

If you want a proof strategy guide that actually helps on real problems, start here: translate the statement into the language of a connection, then use the few structural identities that make connections powerful. The point is not to memorize formulas. The point is to learn what to compute, what to avoid, and which coordinate choices make the essential part of the argument visible.

Why connections sit at the center of geometry

On $\mathbb R^n$, a vector field is just a map $X:\mathbb R^n\to\mathbb R^n$, and you can differentiate it using ordinary partial derivatives. On a manifold, vectors live in different tangent spaces $T_pM$, so subtraction like $X(p)-X(q)$ is meaningless.

A connection repairs that by providing a covariant derivative

that tells you how the vector field $Y$ changes in the direction $X$, producing another vector field. The key feature is that $\nabla$ is local, linear in the direction field, and satisfies a Leibniz rule in the second slot.

With a connection in hand, you get:

• Geodesics: curves $\gamma$ with $\nabla_{\dot\gamma}\dot\gamma=0$.
• Parallel transport: vectors $V(t)$ along $\gamma$ with $\nabla_{\dot\gamma}V=0$.
• Curvature: the obstruction to second covariant derivatives commuting.

The mental model to keep is simple: a connection is a rule for comparing nearby tangent spaces, and curvature measures the failure of that comparison to be path-independent.

The Levi–Civita connection: the default choice

If $(M,g)$ is Riemannian, there is a unique connection $\nabla$ characterized by:

• Metric compatibility: $X\langle Y,Z\rangle = \langle \nabla_X Y, Z\rangle + \langle Y, \nabla_X Z\rangle$.
• Zero torsion: $\nabla_X Y – \nabla_Y X = [X,Y]$.

This is the Levi–Civita connection. Its existence and uniqueness are not just a theorem to cite; they are a proof tactic. Whenever a problem statement mentions only $g$, you should expect the Levi–Civita connection to be the intended mechanism.

A computational entry point is the Koszul formula:

You rarely compute with this directly on long problems, but it tells you what is allowed: anything about $\nabla$ must be built from $g$ and brackets of vector fields.

The proof strategy spine: translate, simplify, then use tensoriality

A reliable strategy for many geometry proofs looks like this:

• Translate the claim into an identity involving $\nabla$, curvature $R$, or parallel transport.
• Choose coordinates or frames where the connection simplifies at the point or along a curve.
• Use tensoriality to evaluate at a point in the simplest possible configuration.
• Only then compute.

The step that many people miss is tensoriality. Curvature, torsion, and many connection-derived objects are tensors in suitable slots, so you can compute them using any convenient extension of vectors to fields and any convenient coordinates at a point.

A small table of what you can normalize away

| Object | What you can arrange locally | What you cannot erase |

|—|—|—|

| Metric $g$ | $g_{ij}(p)=\delta_{ij}$ | curvature data |

| Connection $\nabla$ | $\Gamma^k_{ij}(p)=0$ in normal coordinates | derivatives of $\Gamma$ tied to curvature |

| Frame | orthonormal at $p$ | global twisting and topology |

| Curve $\gamma$ | arclength parametrization | curvature constraints on its image |

The habit to cultivate: normalize at the place where the main estimate or identity will be evaluated.

Curvature as a commutator identity

Curvature is defined by

This definition already suggests a proof move: if you can rewrite your statement as the vanishing of such a commutator, you are proving flatness or controlling curvature. If you can show the commutator has a definite sign in a quadratic form, you are proving comparison theorems or rigidity.

Even without full computations, the formal properties matter:

• $R(X,Y)$ is linear in each input and alternating in $X,Y$.
• For Levi–Civita connections, $R$ satisfies strong symmetry identities.
• Curvature controls the second variation of energy and hence geodesic stability.

You do not need to write down all symmetries to use them. Often it is enough to know that “curvature is a tensor,” so you can compute it in a well-chosen frame.

Worked example: great circles are geodesics on the sphere

Let $S^2\subset\mathbb R^3$ have the metric induced from the Euclidean inner product. A clean strategy avoids messy coordinates.

Take a smooth curve $\gamma(t)\in S^2$. View it as a curve in $\mathbb R^3$. The ambient derivative $\ddot\gamma$ decomposes into tangential and normal parts relative \to $S^2$.

The Levi–Civita connection on $S^2$ can be described as “differentiate in $\mathbb R^3$ and project to the tangent space.” Concretely,

the tangential component of $\ddot\gamma$.

So $\gamma$ is a geodesic exactly when $\ddot\gamma$ is normal to the sphere, meaning it is proportional \to $\gamma$ itself (because the normal direction at $\gamma$ is spanned by $\gamma$):

Now specialize \to a great circle. A great circle is the intersection of $S^2$ with a two-dimensional linear subspace $P\subset\mathbb R^3$ through the origin. Parametrize it by arclength so that $\gamma(t)\in P$ and $\|\gamma(t)\|=1$.

Within the plane $P$, $\gamma$ is a unit-speed circle, so $\ddot\gamma$ points toward the center, which is the origin. That means $\ddot\gamma$ is proportional \to $\gamma$ with negative coefficient. Therefore the tangential component is zero, so the curve is a geodesic.

This proof showcases a general pattern:

• Use a geometric description of the connection (projection from an ambient space).
• Convert “geodesic” \to “acceleration has no tangential component.”
• Reduce the claim to an elementary computation in a simpler space.

No heavy coordinate machinery is required, and the argument generalizes to many submanifold settings.

Strategy patterns you can reuse across geometry problems

These are not slogans. They are moves that tend to convert a complicated question into a short computation.

• Move: choose normal coordinates at a point.

If a statement is pointwise, set $\Gamma^k_{ij}(p)=0$ and evaluate there. What remains is curvature or tensor algebra, not connection clutter.

• Move: prove an identity for all vectors by checking it on a basis.

For bilinear expressions, choose an orthonormal basis at a point and compute components. This avoids global coordinate entanglement.

• Move: use parallel transport to compare tangent spaces.

If you need to compare vectors at different points, transport them along a curve and differentiate the transported quantity. This often reveals conserved quantities.

• Move: rewrite in terms of energy or length.

Many geometric claims about geodesics become calculus of variations statements. The connection provides the Euler–Lagrange equation $\nabla_{\dot\gamma}\dot\gamma=0$.

• Move: separate what is tensorial from what is coordinate-dependent.

Curvature is tensorial; Christoffel symbols are not. If your proof depends on Christoffel symbols in a way that is not obviously invariant, you are probably proving a coordinate artifact.

Reading geometry papers without drowning in notation

Geometry papers often compress a huge amount of structure into short expressions. A practical decoding approach is:

• Identify the connection being used: Levi–Civita, a principal bundle connection, or a chosen affine connection.
• Find the curvature object and its convention: $R(X,Y)Z$ can differ by sign between authors.
• Locate where tensoriality is exploited: many “miraculous simplifications” are just evaluations in normal coordinates.
• Track which statements are global and which are local; proofs often silently switch perspectives using coverings or compactness.

If you can keep those anchors, the rest is computation and bookkeeping, not mystery.

The moving-frames viewpoint: connections without coordinates

Coordinates are useful, but many of the cleanest geometric proofs avoid them entirely by using frames and differential forms. The connection becomes a matrix of one-forms, and curvature becomes a matrix of two-forms.

Choose an orthonormal frame $e_1,\dots,e_n$ on an open set. Define connection one-forms $\omega_{ij}$ by

Metric compatibility forces $\omega_{ij}=-\omega_{ji}$, so the connection lives in the Lie algebra of the orthogonal group.

Let $\theta^i$ be the dual coframe. Then the structure equations are:

• First structure equation (torsion-free condition in the Levi–Civita case):

• Second structure equation (curvature):

where $\Omega_{ij}$ are the curvature two-forms representing $R$ in the frame.

Why does this help with proofs? Because many statements are naturally about how things rotate as you move. Frames make rotation visible, and wedge products make antisymmetry automatic. If a coordinate calculation would involve pages of Christoffel symbols, the moving-frames version often collapses \to a line or two of exterior algebra plus one clever normalization.

A practical rule:

• Use coordinates when you need explicit components for an estimate.
• Use frames when you need structural identities and cancellations.

A global payoff: curvature integrates to topology

Once you are fluent with the connection language, you start recognizing that curvature does not only sit pointwise. It accumulates. The clearest instance on surfaces is the Gauss–Bonnet theorem, which relates the integral of Gaussian curvature to the Euler characteristic. You do not need the full formal statement here; what matters for proof strategy is the message:

• Connection and curvature let you convert local differential data into global invariants.

That is why “start with connections” is not merely advice about technique. It is advice about the bridge between local computations and global conclusions.

What starting with connections gives you

Connections are not only a technique; they are a proof architecture:

• They tell you what it means to differentiate geometric data.
• They unify geodesics, curvature, and transport into one language.
• They provide canonical normalizations that remove noise.
• They expose invariants that survive coordinate changes.

When geometry problems feel slippery, it is often because the tangent spaces are moving under your feet. A connection is the formal way to stop that movement long enough to calculate.

A lot of geometry is built on a comforting principle: every smooth manifold looks like Euclidean space when you zoom in far enough. In Riemannian geometry, you can push that comfort further: in normal coordinates, the metric looks Euclidean at a point and the first derivatives vanish. If you add “curvature zero” \to the story, it can feel like you should be back in the plane for real.

The flat torus is the counterexample that breaks that intuition cleanly. It is locally indistinguishable from the Euclidean plane, yet globally it is not the plane in any meaningful geometric sense. Once you internalize what goes wrong, you stop trying to prove global statements by local computations alone, and you start keeping a small set of global invariants on your desk at all \times.

The setup: what “locally Euclidean” really means

A Riemannian manifold $(M,g)$ is a smooth manifold $M$ together with an inner product $g_p$ on each tangent space $T_pM$, varying smoothly with $p$. In a coordinate chart $(U,x^1,\dots,x^n)$, the metric is written

Two points about “local Euclidean behavior” are easy to mix up:

• Chart-level Euclideanity: every point has coordinates, so locally $M$ is a \subset of $\mathbb R^n$. This is topology and smooth structure.
• Metric-level Euclideanity: the metric looks like the Euclidean metric up \to a prescribed order when expressed in a good chart. This is geometry.

Normal coordinates around a point $p$ are the geometric expression of zooming in. In normal coordinates,

• $g_{ij}(p)=\delta_{ij}$,
• $\partial_k g_{ij}(p)=0$.

Curvature is a second-order invariant. So if curvature vanishes everywhere, it is tempting to think there is no geometric obstruction \left.

The flat torus shows the missing ingredient: global identification can preserve local metric data while changing the global shape completely.

Building the flat torus from the plane

Start with the Euclidean plane $\mathbb R^2$ with coordinates $(x,y)$ and the standard metric

Fix the integer lattice $\mathbb Z^2\subset \mathbb R^2$ acting by translations:

Translations preserve $g_\mathrm{E}$. So we can form the quotient

and inherit a well-defined Riemannian metric $g$ on $\mathbb T^2$ by declaring the quotient map $\pi:\mathbb R^2\to\mathbb T^2$ \to be a local isometry.

Geometrically, this is the “glue opposite sides of a unit square” construction. Every point on the torus has a neighborhood that lifts isometrically \to a neighborhood in $\mathbb R^2$. No stretching, no bending, no curvature.

In particular:

• The Gaussian curvature is $0$ everywhere.
• Every small patch is literally a Euclidean patch, up to relabeling by $\pi$.
• Geodesics in $\mathbb T^2$ are projections of straight lines in $\mathbb R^2$.

So where is the difference hiding?

What the torus has that the plane cannot: closed geodesics

Take the straight line in $\mathbb R^2$

It is a geodesic in $\mathbb R^2$. Project it down:

Because $(1,0)\in\mathbb Z^2$, we have $\pi(t,0)=\pi(t+1,0)$. So $\bar\gamma$ is periodic:

It is a closed geodesic on $\mathbb T^2$.

The Euclidean plane has no closed geodesics. Straight lines never come back. That single observation already blocks any global isometry $\mathbb T^2\to\mathbb R^2$: an isometry sends geodesics to geodesics, and it preserves “closedness.”

The same phenomenon appears in many disguises:

• $\mathbb T^2$ has nontrivial loops that cannot be shrunk \to a point, and those loops can often be realized by short geodesics.
• $\mathbb R^2$ is simply connected, so every loop contracts.
• Even though curvature vanishes, topology and global identifications create global constraints on distance-minimizing behavior.

This is the first big lesson: curvature is not the only global obstruction to being Euclidean.

A precise obstruction: fundamental group meets geometry

Topologically,

A global isometry is, in particular, a homeomorphism. So it would induce an isomorphism of fundamental groups. That is impossible here.

If you want a geometric restatement, think in terms of coverings:

• $\pi:\mathbb R^2\to\mathbb T^2$ is a covering map and a local isometry.
• $\mathbb R^2$ is the universal cover of $\mathbb T^2$.
• The deck transformations are exactly $\mathbb Z^2$ translations.

Local geometry lifts perfectly. Global geometry is exactly where the deck transformations matter.

A helpful way to remember this is that local computations cannot see the deck group. Curvature, Christoffel symbols, local coordinate expressions, and normal forms all live downstairs in the quotient just fine.

But global questions, like “is the manifold globally Euclidean,” are sensitive to the covering structure.

Local isometry versus global isometry

The quotient map $\pi$ is a **local isometry**, meaning that for every point $p\in\mathbb R^2$, there is a neighborhood $U$ on which $\pi$ restricts to an isometry onto $\pi(U)\subset \mathbb T^2$.

A local isometry does not have to be injective, and failing injectivity is exactly how global geometry changes.

Here is the local–global split in one table:

| Feature | $\mathbb R^2$ | Flat $\mathbb T^2$ | What local data sees |

|—|—|—|—|

| Curvature | $0$ | $0$ | Sees no difference |

| Small neighborhoods | Euclidean | Euclidean | Sees no difference |

| Closed geodesics | none | many | Global, not local |

| Fundamental group | trivial | $\mathbb Z^2$ | Global, not local |

| Universal cover | itself | $\mathbb R^2$ | Requires global viewpoint |

| Distance growth at infinity | unbounded without wrap | wraps by identification | Global |

If you have ever felt confused by a statement like “zero curvature does not imply Euclidean,” this table is the cure: curvature is local, Euclideanity is global.

A deeper geometric invariant: holonomy can be trivial while topology is not

Holonomy is what happens \to a vector when you parallel transport it around a closed loop. On a simply connected region with zero curvature, holonomy is trivial: you get your vector back unchanged.

On the flat torus, curvature is zero, and parallel transport along loops is still trivial. So holonomy does not distinguish $\mathbb T^2$ from $\mathbb R^2$ in this case.

That is another useful lesson: global geometry has multiple independent “axes” of invariants. Some are sensitive to curvature and connection data; some are purely topological; some combine both.

When you build an argument, you need to know which axis you are operating on.

How this counterexample changes how you prove things

Once you accept the flat torus, several proof habits become dangerous:

• Proving a global classification statement using only local normal forms.
• Inferring a global embedding from local coordinate computations.
• Concluding that “no curvature” means “no geometry.”

Instead, the counterexample suggests a disciplined checklist whenever a statement has a global flavor:

• Ask what the universal cover looks like and what the deck group is doing.
• Check whether the claim should be invariant under passing \to a quotient by isometries.
• Identify what global invariants the claim would force: fundamental group, existence of closed geodesics, volume growth, injectivity radius.

A compact way to phrase the core issue is:

• Local constraints control curvature and infinitesimal behavior.
• Global constraints control how local charts glue together and how geodesics behave at large scale.

The flat torus is the simplest place where those two layers decouple cleanly.

A practical diagnostic move you can reuse

Suppose you are trying to prove something like “if $(M,g)$ has property $P$, then $(M,g)$ is globally isometric \to $\mathbb R^n$.” Before you compute anything, try a diagnostic:

• Does $P$ survive taking a quotient by a discrete group of isometries?

If it does, and if $\mathbb R^n$ has nontrivial quotients (it does, such as tori), then $P$ cannot possibly force global Euclideanity unless $P$ also blocks those quotients by a global invariant.

Curvature $0$ survives. Completeness survives. Many analytic bounds survive. None of them prevent the torus. So they cannot force $\mathbb R^n$.

This move saves time, and it prevents proofs that are doomed from the first line.

The torus also changes large-scale geometry: compactness and growth

There is another geometric difference that is easy to overlook if you only stare at curvature: $\mathbb T^2$ is compact and $\mathbb R^2$ is not. Compactness is not merely topological; it forces metric consequences.

• Every continuous function on $\mathbb T^2$ achieves maxima and minima.
• The diameter of $\mathbb T^2$ is finite: there is a global upper bound on distances.
• There is no way to send $\mathbb T^2$ isometrically onto $\mathbb R^2$ because $\mathbb R^2$ contains points arbitrarily far apart.

Compactness is global and cannot be detected from a single coordinate patch. This is another way the counterexample teaches you the “global hypothesis detector” habit: if a conclusion would force compactness or non-compactness, local calculations will not suffice.

A related metric invariant is the injectivity radius. On $\mathbb T^2$, there is a smallest nontrivial translation length in the lattice. That length controls the shortest noncontractible loops and provides a positive injectivity radius. On $\mathbb R^2$, the injectivity radius is infinite. Again, curvature cannot see this.

A richer geodesic picture: rational and irrational slopes

Because geodesics on the flat torus are projections of straight lines, you can classify many of them by slope.

Take a line $\gamma(t)=(at,bt)$ in $\mathbb R^2$. Its projection $\bar\gamma(t)=\pi(\gamma(t))$ closes up exactly when there is a nonzero $T$ such that $(aT,bT)\in\mathbb Z^2$. That happens precisely when $a/b$ is rational (interpreting $b=0$ as the infinite slope case).

So:

• If the slope is rational, the projected geodesic is closed.
• If the slope is irrational, the projected geodesic never closes and winds around the torus forever.

In fact, irrational-slope geodesics are dense in $\mathbb T^2$. You do not need the full proof to benefit from the intuition: a straight line with irrational slope keeps missing the lattice points needed \to “line up,” so its projection keeps visiting new regions.

This observation strengthens the earlier lesson. The torus does not just have a few closed geodesics; it has an entire arithmetic structure controlling global geodesic behavior. None of that arithmetic appears in local coordinate expansions.

What you should carry forward

The flat torus is not just a neat object. It is a compact summary of how geometry works:

• A manifold can be locally Euclidean in the strongest Riemannian sense and still have a completely different global shape.
• Curvature is not the whole story; topology and identification data matter.
• Local isometries are common; global isometries are rare and require global invariants.

If you keep a single counterexample in your head as you move through geometry, make it this one. It will quietly correct dozens of false inferences before they ever reach the page.