Tensors and Exterior Calculus: A Comprehensive Guide for Differential Geometry
Imagine you're trying to describe the shape of a crumpled piece of paper. You could try to give coordinates for every point, but that would be messy. Better would be to describe how it bends and twists at each point. This is the world of differential geometry, and the language it speaks is tensors and differential forms.
Think of tensors as mathematical Swiss Army knives-tools that can handle multiple directions at once. And exterior calculus? That's the art of doing calculus without getting stuck in coordinates.
Chapter 1: The Space Between Spaces
1.1 What Even Is a Vector Space?
Let's start simple. A vector space is a collection of "arrows" (vectors) where you can:
- Add two arrows (put them tip to tail)
- Stretch or shrink an arrow (multiply by a number)
- There's a zero arrow (no movement)
Example: The set of all possible velocities at a point on a surface. At each point on a sphere, you can only move in directions tangent to the sphere-that's a 2D vector space.
1.2 Measuring Angles and Lengths
An inner product is just a fancy way to measure angles and lengths. For vectors \(u\) and \(v\):
- Symmetry: \( \langle u, v \rangle = \langle v, u \rangle \) (order doesn't matter)
- Bilinearity: Scales nicely: \( \langle 2u, v \rangle = 2\langle u, v \rangle \)
- Positive: \( \langle v, v \rangle \geq 0 \), and equals 0 only if \(v = 0\)
From this we get length: \( \|v\| = \sqrt{\langle v, v \rangle} \)
And the angle between vectors: \( \cos\theta = \frac{\langle u, v \rangle}{\|u\|\|v\|} \)
1.3 Projecting Shadows
Ever shine a flashlight to see a shadow? That's orthogonal projection.
Given a subspace \(S\) (like a plane), any vector \(v\) can be split:
- \(v^\parallel\) = the part in \(S\) (the shadow)
- \(v^\perp\) = the part perpendicular to \(S\)
Mathematically: \(v = v^\parallel + v^\perp\), with \(\langle v^\perp, w \rangle = 0\) for all \(w\) in \(S\)
Example: On a hillside, your velocity splits into:
- Along the slope (you slide)
- Into the hill (you dig in)
Chapter 2: The Dual World-Measuring Vectors
2.1 Functionals: Measuring Tapes for Vectors
The dual space \(V^*\) contains all linear measuring tapes. Each measuring tape (called a functional) takes a vector and gives back a number.
Think of it like this:
- A vector = a direction you can move
- A dual vector = a way to measure movement in that direction
Theorem: There are exactly as many independent measuring tapes as there are directions to measure.
Proof: If your space has basis vectors \(\{E_1, E_2, \ldots, E_n\}\), you can create special measuring tapes \(\{\omega^1, \omega^2, \ldots, \omega^n\}\) where:
- \(\omega^i(E_j) = 1\) if \(i = j\) (perfectly measures the \(i\)-th direction)
- \(\omega^i(E_j) = 0\) if \(i \neq j\) (ignores other directions)
2.2 The Musical Isomorphisms: ♭ and ♯
With an inner product, vectors and dual vectors can dance together:
Flat (♭): Turns a vector into a measuring tape
\[v^\flat(w) = \langle v, w \rangle\]
Translation: "How much does \(w\) align with \(v\)?"
Sharp (♯): Turns a measuring tape back into a vector
\[\xi(w) = \langle \xi^\sharp, w \rangle\]
Translation: "Find the vector that gives the same measurements as \(\xi\)"
Why musical notation? Think of lowering the index (♭ lowers the pitch) and raising the index (♯ raises the pitch).
2.3 Coordinates Without Orthonormality
Most bases aren't perfectly square to each other. We track this with the metric tensor \(g_{ij}\):
\[g_{ij} = \langle E_i, E_j \rangle\]
These numbers tell us:
- If \(g_{ij} = 0\) for \(i \neq j\): vectors are perpendicular
- If \(g_{ij} \neq 0\): vectors are skewed
The magic formulas:
Dual basis: \(\omega^i = \sum_j g^{ij} E_j^\flat\)
(The inverse metric \(g^{ij}\) untangles the skew)
Lowering indices: \(v^\flat = \sum_i v_i \omega^i\) where \(v_i = \sum_j g_{ij} v^j\)
(Multiply by \(g\) to go from vector components to dual components)
Raising indices: \(\xi^\sharp = \sum_i \xi^i E_i\) where \(\xi^i = \sum_j g^{ij} \xi_j\)
(Multiply by \(g^{-1}\) to go from dual components to vector components)
Special case: Orthonormal basis → \(g_{ij} = \delta_{ij}\) (identity matrix) → \(v_i = v^i\), indices don't care about position!
Chapter 3: Tensors-Multilinear Superpowers
3.1 What's a Tensor, Really?
A \((k,\ell)\)-tensor is a machine that:
- Takes \(k\) dual vectors and \(\ell\) vectors as input
- Spits out a number
- Is linear in every slot
Notation: \(T: \underbrace{V^* \times \cdots \times V^*}_{k} \times \underbrace{V \times \cdots \times V}_{\ell} \to \mathbb{R}\)
Dimension: If \(V\) has dimension \(n\), there are \(n^{k+\ell}\) independent tensors.
3.2 Everyday Tensors
- (0,0)-tensor: A number (scalar field)
- (0,1)-tensor: A vector field
- (1,0)-tensor: A dual vector field (1-form)
- (2,0)-tensor: A bilinear form (takes two vectors → number)
The Metric Tensor (our first important example):
\[g(v,w) = \langle v, w \rangle\]
In coordinates: \(g = \sum_{i,j} g_{ij} \omega^i \otimes \omega^j\)
The symbol \(\otimes\) (tensor product) means: "apply \(\omega^i\) to first input, \(\omega^j\) to second, multiply results."
3.3 Transformation Law: Why Tensors Are Geometric
Change your coordinate system? Tensors transform consistently:
Old basis: \(E_i\)
New basis: \(E'_i = \sum_j A_i^j E_j\)
Tensor components transform as:
\[T'^{i_1\cdots i_k}_{j_1\cdots j_\ell} = \sum (A^{-1})^{i_1}_{p_1} \cdots (A^{-1})^{i_k}_{p_k} A^{q_1}_{j_1} \cdots A^{q_\ell}_{j_\ell} T^{p_1\cdots p_k}_{q_1\cdots q_\ell}\]
This ensures: The geometric object doesn't change, only our description of it.
3.4 Contraction: Tracing Over
Like taking the trace of a matrix:
\[\text{Contraction}(T^i_j) = \sum_i T^i_i\]
Geometrically: You feed the same vector into one upper and one lower slot.
Chapter 4: Symmetric vs. Alternating-Two Personalities
4.1 Symmetric Tensors: Order Doesn't Matter
A symmetric bilinear tensor satisfies:
\[A(v,w) = A(w,v)\]
Examples:
- The metric tensor \(g\)
- The second fundamental form \(II\) (measures curvature)
Matrix representation: \(A_{ij} = A_{ji}\) (symmetric matrix)
4.2 The Second Fundamental Form: Measuring Curvature
On a surface in \(\mathbb{R}^3\):
\[II_p(v,w) = \langle -dN_p(v), w \rangle\]
Where \(N\) is the unit normal.
Interpretation: How much does the normal tilt when you move in direction \(v\)? Project that tilt onto direction \(w\).
Sphere example: \(N(p) = p\) (normal points outward)
\[dN_p(v) = v \quad \text{(derivative of identity)}\]
So \(II_p(v,w) = \langle -v, w \rangle = -\langle v, w \rangle\)
The sphere curves equally in all directions!
4.3 Alternating Tensors (Differential Forms): Antisymmetric
A \(k\)-form satisfies:
\[\sigma(\ldots, v, \ldots, w, \ldots) = -\sigma(\ldots, w, \ldots, v, \ldots)\]
Swap two inputs → change sign.
Key fact in 2D:
- 0-forms: Scalars (\(f: S \to \mathbb{R}\))
- 1-forms: Dual vectors (\(\omega: T_pS \to \mathbb{R}\))
- 2-forms: All multiples of the area form
Why? Let \(\sigma\) be a 2-form. For any vectors \(v,w\):
\[\sigma(v,w) = (v^1 w^2 - v^2 w^1) \sigma(E_1, E_2)\]
That determinant \((v^1 w^2 - v^2 w^1)\) is the oriented area of the parallelogram. So \(\sigma\) is completely determined by one number: \(\sigma(E_1, E_2)\).
4.4 Wedge Product: Multiplying Forms
For 1-forms \(\omega\) and \(\eta\):
\[(\omega \wedge \eta)(v,w) = \omega(v)\eta(w) - \omega(w)\eta(v)\]
Properties:
- \(\omega \wedge \eta = -\eta \wedge \omega\) (antisymmetric)
- \(\omega \wedge (\eta \wedge \xi) = (\omega \wedge \eta) \wedge \xi\) (associative)
- Bilinear: \((a\omega + b\eta) \wedge \xi = a(\omega \wedge \xi) + b(\eta \wedge \xi)\)
Basis: If \(\{\omega^1, \omega^2\}\) is a basis for 1-forms, then \(\{\omega^1 \wedge \omega^2\}\) is a basis for 2-forms.
Chapter 5: Area and the Hodge Star
5.1 The Area Form: Measuring Oriented Area
The area form \(dA\) is the standard way to measure oriented area:
\[dA(v,w) = \text{signed area of parallelogram spanned by } v \text{ and } w\]
Formula: \(dA(v,w) = \|v\|\|w\|\sin\theta\), where \(\theta\) is oriented angle from \(v\) to \(w\).
Coordinates: For orthonormal basis: \(dA = \omega^1 \wedge \omega^2\)
For general basis: \(dA = \sqrt{\det(g)} \omega^1 \wedge \omega^2\)
The \(\sqrt{\det(g)}\) factor accounts for non-square basis vectors.
5.2 Hodge Star: Rotating Forms
The Hodge star \(\ast\) is a 90° rotation for forms:
Definition: \(\omega \wedge (\ast \eta) = \langle \omega, \eta \rangle dA\)
In 2D:
- \(\ast 1 = dA\) (rotate a scalar? Get area!)
- \(\ast dA = 1\) (rotate area? Get back to scalar!)
- For 1-forms: \(\ast \omega(v) = \omega(R_{\pi/2}(v))\) (rotate the input vector)
Example: On \(\mathbb{R}^2\), \(\omega = a dx + b dy\)
Then \(\ast \omega = -b dx + a dy\)
Why? Because rotating the vector \((a,b)\) by 90° gives \((-b,a)\).
Check: Let \(\omega = dx\), \(\eta = dy\):
- \(\langle dx, dy \rangle = 0\) (orthogonal)
- \(dx \wedge (\ast dy) = dx \wedge (-dx) = 0\) ✓
- And indeed, \(dx\) and \(dy\) are perpendicular.
Chapter 6: Bundles-Families of Vector Spaces
6.1 The Tangent Bundle: All Tangent Spaces Together
For a surface \(S\):
\[TS = \bigcup_{p \in S} T_pS\]
Think: Attach a tangent plane at every point.
Tensor bundles: Attach a tensor space at every point:
\[T^{(k,\ell)}S = \bigcup_{p \in S} \underbrace{T_p^*S \otimes \cdots \otimes T_p^*S}_{k} \otimes \underbrace{T_pS \otimes \cdots \otimes T_pS}_{\ell}\]
6.2 Sections: Fields Over the Surface
A section chooses one tensor per point:
- Scalar field: \(f: S \to \mathbb{R}\) (temperature at each point)
- Vector field: \(X: S \to TS\) (wind velocity at each point)
- 1-form field: \(\omega: S \to T^*S\) (measuring tape at each point)
- Metric field: \(g: S \to \text{Sym}^2 T^*S\) (ruler at each point)
6.3 Pushforward and Pullback: Moving Between Spaces
Given \(\phi: S \to M\) (map between surfaces):
Pushforward \(d\phi_p\): Takes tangent vectors from \(S\) to \(M\)
- Like: "If I move this way on \(S\), how do I move on \(M\)?"
Pullback \(\phi^*\): Brings forms from \(M\) back to \(S\)
- Like: "A measuring tape on \(M\) becomes a measuring tape on \(S\)"
Formula: \((\phi^*\omega)_p(v) = \omega_{\phi(p)}(d\phi_p(v))\)
Example: Cylinder mapping: \(\phi(u,v) = (\cos u, \sin u, v)\)
Pullback of \(dz\) (height form) = \(dv\) (just the \(v\)-coordinate)
Chapter 7: Covariant Derivative-Derivatives That Stay Tangent
7.1 The Problem with Ordinary Derivatives
Take a vector field on a sphere. The ordinary derivative in \(\mathbb{R}^3\) points off the sphere! We want a derivative that stays tangent.
Solution: Project onto the tangent space:
\[\nabla_X Y = \text{(tangential component of } D_X Y\text{)}\]
Where \(D\) is the ordinary derivative.
7.2 Christoffel Symbols: Correction Factors
In coordinates \((x^1, x^2)\):
\[\nabla_{\partial_i} \partial_j = \sum_k \Gamma^k_{ij} \partial_k\]
The \(\Gamma^k_{ij}\) track how the basis vectors twist as we move.
Formula from metric:
\[\Gamma^k_{ij} = \frac{1}{2} \sum_\ell g^{k\ell} \left(\frac{\partial g_{j\ell}}{\partial x^i} + \frac{\partial g_{i\ell}}{\partial x^j} - \frac{\partial g_{ij}}{\partial x^\ell}\right)\]
Interpretation: How the measurement of angles/lengths changes.
7.3 Covariant Derivative of Tensors
Extend via Leibniz rule: "Derivative of product = sum of (derivative × other)"
For a 1-form \(\omega\):
\[(\nabla_j \omega)_i = \frac{\partial \omega_i}{\partial x^j} - \sum_k \Gamma^k_{ji} \omega_k\]
Why minus sign? Remember: \(\omega(Y)\) should differentiate as:
\[X(\omega(Y)) = (\nabla_X \omega)(Y) + \omega(\nabla_X Y)\]
The connection must be "compatible" with evaluating forms on vectors.
7.4 Parallel Transport: Carrying Without Twisting
A vector field is parallel transported along a curve \(\gamma\) if:
\[\nabla_{\dot{\gamma}(t)} V = 0 \quad \text{along the curve}\]
Sphere example: Transport a tangent vector along a great circle-keep constant angle with the circle.
Chapter 8: Exterior Derivative-Calculus Without Coordinates
8.1 Exterior Derivative Defined
For forms:
0-forms (functions): \(df(V) = V(f)\) (directional derivative)
In coordinates: \(df = \sum_i \frac{\partial f}{\partial x^i} dx^i\)
1-forms: \(d(a dx + b dy) = \left(\frac{\partial b}{\partial x} - \frac{\partial a}{\partial y}\right) dx \wedge dy\)
2-forms: \(d(\text{anything}) = 0\) in 2D
Key property: \(d^2 = 0\) (taking derivative twice gives zero)
8.2 Relation to Covariant Derivative
For a 1-form \(\omega\):
\[d\omega = \text{Alt}(\nabla \omega)\]
Where Alt antisymmetrizes: \(\text{Alt}(T)_{ij} = \frac{1}{2}(T_{ij} - T_{ji})\)
Magic: The Christoffel symbols cancel! Why?
\[(\nabla \omega)_{ij} = \partial_i \omega_j - \Gamma^k_{ij} \omega_k\]
Antisymmetrize: \(\frac{1}{2}[(\partial_i \omega_j - \Gamma^k_{ij} \omega_k) - (\partial_j \omega_i - \Gamma^k_{ji} \omega_k)]\)
Since \(\Gamma^k_{ij} = \Gamma^k_{ji}\) (torsion-free), the \(\Gamma\)'s cancel, leaving:
\[d\omega_{ij} = \partial_i \omega_j - \partial_j \omega_i\]
8.3 Closed vs. Exact Forms
- Closed: \(d\omega = 0\) (like a conservative field)
- Exact: \(\omega = d\eta\) (like a gradient)
Every exact form is closed (since \(d^2 = 0\)).
But not every closed form is exact-depends on topology!
Example: On punctured plane \(\mathbb{R}^2 \setminus \{0\}\):
\[\omega = \frac{-y dx + x dy}{x^2 + y^2}\]
Check: \(d\omega = 0\) (closed)
But \(\omega \neq d(\text{anything})\) globally (not exact)—integrating around origin gives \(2\pi\), not 0.
Chapter 9: Co-differential and Laplacian
9.1 Co-differential: The Adjoint of \(d\)
Definition: \(\delta\) is the \(L^2\)-adjoint of \(d\):
\[\int_S \langle d\omega, \tau \rangle dA = \int_S \langle \omega, \delta\tau \rangle dA\]
Formula: \(\delta = -\ast d \ast\) (in 2D)
Interpretation: While \(d\) measures circulation, \(\delta\) measures flux.
Examples:
- For 1-form: \(\delta(a dx + b dy) = -\left(\frac{\partial a}{\partial x} + \frac{\partial b}{\partial y}\right)\) (negative divergence!)
- For function: \(\delta(df)^\sharp = -\text{div}(\nabla f)\)
9.2 Laplace-de Rham Operator
\[\Delta = d\delta + \delta d\]
For functions: \(\Delta f = \delta df = -\text{div}(\nabla f)\)
On \(\mathbb{R}^2\): \(\Delta f = -\left(\frac{\partial^2 f}{\partial x^2} + \frac{\partial^2 f}{\partial y^2}\right)\)
(The minus sign is convention-some authors include it, some don't.)
9.3 Harmonic Forms: Perfect Balance
A form \(\omega\) is harmonic if \(\Delta \omega = 0\).
Theorem: On closed surfaces, harmonic ⇔ (closed AND co-closed):
\[d\omega = 0 \quad \text{and} \quad \delta\omega = 0\]
Example: On flat torus, \(dx\) and \(dy\) are harmonic.
Chapter 10: Stokes' Theorem-Boundaries Tell the Story
10.1 The Grand Unification
For an \(n\)-dimensional manifold \(M\) with boundary \(\partial M\):
\[\int_M d\omega = \int_{\partial M} \omega\]
Special cases:
- Fundamental Theorem: \(\int_a^b f'(x) dx = f(b) - f(a)\)
- Green's Theorem: \(\iint_D \left(\frac{\partial Q}{\partial x} - \frac{\partial P}{\partial y}\right) dA = \oint_{\partial D} (P dx + Q dy)\)
- Divergence Theorem: \(\iiint_V \text{div}(F) dV = \iint_{\partial V} F \cdot n dS\)
10.2 Divergence Theorem on Surfaces
For vector field \(X\) on surface \(S\):
\[\int_S \text{div}(X) dA = \int_{\partial S} \langle N_{\partial S}, X \rangle d\ell\]
Where \(N_{\partial S}\) is outward unit normal along boundary.
Proof sketch: Take \(\omega = \ast X^\flat\). Then:
- \(d\omega = \text{div}(X) dA\)
- On boundary: \(\omega = \langle N_{\partial S}, X \rangle d\ell\)
Apply Stokes: \(\int_S d\omega = \int_{\partial S} \omega\) ✓
Example: Unit disk, \(X = (x,y)\):
- \(\text{div}(X) = 2\)
- Area = \(\pi\), so left side = \(2\pi\)
- On boundary: \(N = (x,y)\), \(\langle N, X \rangle = 1\)
- Boundary length = \(2\pi\), so right side = \(2\pi\) ✓
Chapter 11: Hodge Decomposition-Splitting Vector Fields
11.1 The Theorem
On closed surface \(S\):
\[\text{1-forms} = \text{(exact)} \oplus \text{(co-exact)} \oplus \text{(harmonic)}\]
Where:
- Exact: \(\omega = df\) (gradients)
- Co-exact: \(\omega = \ast dg\) (rotated gradients)
- Harmonic: \(d\omega = 0\) and \(\delta\omega = 0\)
11.2 For Vector Fields
Every vector field decomposes uniquely:
\[X = \nabla f + J(\nabla h) + H\]
Where:
- \(\nabla f\): gradient part (curl-free)
- \(J(\nabla h)\): rotated gradient (div-free, since \(\text{div}(J(\nabla h)) = \text{curl}(\nabla h) = 0\))
- \(H\): harmonic (both curl-free AND div-free)
Here \(J\) = rotate by 90°.
11.3 Topology Enters: Genus Matters
Theorem: \(\dim(\text{harmonic 1-forms}) = 2g\)
Where \(g\) = genus (number of holes).
Examples:
- Sphere (\(g=0\)): 0 harmonic fields
- Torus (\(g=1\)): 2 independent harmonic fields
- Double torus (\(g=2\)): 4 independent harmonic fields
Why? Harmonic 1-forms correspond to "irremovable circulations" around holes.
Chapter 12: Discrete World-Making It Computable
12.1 Discrete Exterior Calculus
Replace smooth with discrete:
🔹 Continuous (Smooth Setting)
This is the traditional mathematical framework used in differential geometry and physics.
- Differential forms – represent smooth fields (like velocity, flux, temperature gradients) that can be integrated over curves, surfaces, volumes.
- Exterior derivative (d) – generalizes gradient, curl, divergence; takes a \( k \)-form to a \((k+1)\)-form.
- Hodge star operator (\(\ast\)) – relates forms of complementary dimensions; encodes metric information (lengths, angles).
- Integral (\(\int\)) – physically meaningful quantity obtained by integrating a form over a domain.
🔹 Discrete (Mesh / Numerical Setting)
This is the discrete analogue used in computational geometry, finite elements, and simulations on meshes.
- Cochains – discrete counterpart to forms: they assign numbers to mesh elements (vertices, edges, faces, cells).
- Coboundary operator – discrete version of \(d\); encoded as an incidence matrix (e.g., edge-to-face connectivity).
- Dual mesh Hodge operator – discrete analogue of \(\ast\); often a diagonal matrix encoding metric info via volumes and lengths.
- Summation – replaces integration: sums cochain values weighted by geometric measures.
Discrete Stokes: \(\langle d\omega, c \rangle = \langle \omega, \partial c \rangle\)
(Just matrix multiplication!)
12.2 Hodge Decomposition Algorithm
# Pseudocode—real implementation would be more detailed
def hodge_decomposition(mesh, vector_field):
# Step 1: Convert vector field to discrete 1-form
# (Assign number to each edge = flow along edge)
omega = vector_field_to_1form(vector_field, mesh)
# Step 2: Build discrete d operators
d0 = build_d(0, mesh) # 0-forms → 1-forms (gradient)
d1 = build_d(1, mesh) # 1-forms → 2-forms (curl)
# Step 3: Build discrete δ operators (δ = -∗d∗)
delta1 = build_delta(1, mesh) # 1-forms → 0-forms (divergence)
delta2 = build_delta(2, mesh) # 2-forms → 1-forms
# Step 4: Solve for exact part (ω_exact = df)
# Solve Δf = δω, where Δ = δd on functions
f = solve_poisson(delta1 @ d0, delta1 @ omega)
exact_part = d0 @ f
# Step 5: Solve for co-exact part (ω_coexact = ∗dg)
# Solve Δg = d∗ω, but careful with signs...
g = solve_poisson(delta2 @ d1, something_with_star(omega))
co_exact_part = delta2 @ g # Actually need rotation...
# Step 6: Harmonic = remainder
harmonic_part = omega - exact_part - co_exact_part
return exact_part, co_exact_part, harmonic_part
12.3 Applications in Graphics and Simulation
- Surface parameterization: Flatten without distortion
- Vector field design: Create fields with desired properties
- Fluid simulation: Project to divergence-free
- Mesh processing: Denoise, smooth, edit
Chapter 13: Beyond-Where This Leads
13.1 Higher Dimensions
In \(n\) dimensions:
- Hodge star: \(k\)-forms ↔ \((n-k)\)-forms
- Codifferential: \(\delta = (-1)^{n(k+1)+1} \ast d \ast\)
- Laplacian: Still \(\Delta = d\delta + \delta d\)
13.2 Physics Connections
Maxwell's Equations (in vacuum):
\[\begin{aligned}dF &= 0 \\\delta F &= J\end{aligned}\]
Where \(F\) is electromagnetic 2-form combining E and B .
Fluid dynamics: Euler equations relate velocity, vorticity, and pressure via exterior calculus.
13.3 Modern Applications
- Computer graphics: Realistic simulation
- Medical imaging: Analyzing brain surfaces
- Machine learning: Geometric deep learning
- Computational physics: Lattice gauge theory
Conclusion: Why Bother?
Tensors and exterior calculus give us:
- Coordinate freedom: Describe geometry intrinsically
- Computational efficiency: Sparse matrices, stable algorithms
- Physical insight: Unified understanding of fields
- Mathematical elegance: Deep connections between analysis, geometry, topology
From the abstract purity of vector spaces to the messy reality of computer meshes, this language bridges theory and practice in beautiful ways.
The next time you see a vector field visualization or a CG fluid simulation, remember: underneath lies the elegant machinery of tensors and differential forms, turning deep mathematics into practical magic.
Further Reading:
- For intuition: Visual Differential Geometry by Needham
- For computation: Discrete Differential Geometry by Crane
- For physics: The Geometry of Physics by Frankel
- For math: Differential Forms and Applications by do Carmo
Start with simple examples, draw lots of pictures, and remember: even the most abstract math was invented to solve concrete problems.
