A Riemannian metric turns tangent vectors and covectors into measurable geometric quantities
Objective: Why is grad f not simply the coordinate pair of df?
A manifold supplies directions but no length until a positive-definite inner product is chosen smoothly at every point. The metric also identifies gradients, one-forms, volume, and operator norms, so every later estimate depends on this first declaration. Before calculating, name the manifold, metric, dimension, sign convention, regularity, completeness or compactness assumption, and whether the claim is local, global, pointwise, integral, or only valid away from a cut or singular set.
Read the notation in words: A Riemannian metric g is a smooth positive-definite section of Sym²(T*M). For v∈T_pM, |v|_g=sqrt(g_p(v,v)); for α∈T_p*M, α^♯ is defined by g(α^♯,v)=α(v), and X^♭=g(X,·). The exact result used in this chapter is: The musical maps ♭:TM→T*M and ♯:T*M→TM are smooth bundle isomorphisms inverse to one another, and the gradient is grad f=(df)^♯.
The complete proof route is: 1. Positive definiteness makes v↦g(v,·) injective on each finite-dimensional tangent space. 2. Equal dimensions make the injective linear map an isomorphism, defining ♯ as its inverse. 3. Smoothness of the inverse metric matrix g^{ij} makes both bundle maps and grad f smooth. Keep coordinate identities separate from invariant statements, and record every curvature bound, endpoint condition, equality case, convergence topology, and radius restriction.
Work the model “On R² with g=4dx²+dy², compute |(1,2)|_g and grad f for f(x,y)=x+y.” from definitions. Its checked result is The norm is 2sqrt2 and grad f=(1/4)∂_x+∂_y. The nearest failure is: A nondegenerate indefinite metric also gives ♭ and ♯ but not a positive norm; a merely continuous positive tensor need not support the smooth Levi–Civita calculations used later. A plotted geodesic, finite mesh, floating-point eigenvalue, or small residual is evidence only after the metric, chart, step size, tolerance, normalization, and unresolved global hypotheses are recorded.
The musical maps ♭:TM→T*M and ♯:T*M→TM are smooth bundle isomorphisms inverse to one another, and the gradient is grad f=(df)^♯.
Positive definiteness makes v↦g(v,·) injective on each finite-dimensional tangent space.
Equal dimensions make the injective linear map an isomorphism, defining ♯ as its inverse.
Smoothness of the inverse metric matrix g^{ij} makes both bundle maps and grad f smooth.
On R² with g=4dx²+dy², compute |(1,2)|_g and grad f for f(x,y)=x+y.
- g((1,2),(1,2))=4·1²+2²=8, so the norm is 2sqrt2.
- df=dx+dy has coefficient vector (1,1), while the inverse metric is diag(1/4,1).
- Therefore grad f=(1/4)∂_x+∂_y.
Result: The norm is 2sqrt2 and grad f=(1/4)∂_x+∂_y.