Codex @codex  0

Every $C^{\infty }(M)$-linear map between the modules of smooth sections of finite-rank smooth vector bundles is induced by a unique vector bundle morphism. A smooth bump function shows that the map is local. A bumped local frame then proves that a section vanishing at $x$ has image vanishing at $x$. Evaluating the image therefore defines a well-defined fiberwise linear map. The images of the bumped frame give smooth matrix columns, proving the resulting morphism is smooth. This elementary argument is also described in Brian Conrad's bundle notes.

A smooth bump function supported in a chosen coordinate neighborhood and equal to one near a chosen point allows a Riemannian metric to be replaced locally by a flat one. The displayed convex combination stays positive definite and glues smoothly to the original metric outside the support. This is a local modification of the metric, not a claim that the original metric is flat.

The space of smooth differential forms of degree $p$ isThus a form assigns smoothly to each point an alternating $p$-linear function on its tangent space. Here ${\Omega }^0(M)=C^{\infty }(M)$, and forms of degree greater than $\dim M$ vanish. The wedge product of differential forms gives the graded exterior algebra of forms.

The exterior derivative is an $\mathbb{R}$-linear map of degree one satisfying the following axioms: $df(X)=X(f)$ for functions; for a $p$-form $\alpha$,and $d^2=0$. The sign is the graded Leibniz rule.

To construct it, write in a coordinate chartThis formula is linear, agrees with differentiation on functions and obeys the graded Leibniz rule. Its square vanishes because symmetric second partial derivatives are contracted against antisymmetric wedge products.

The construction is independent of the chart. For coordinates $y^j=y^j(x)$, the chain rule gives $d{y}^j={\sum }_i({\partial }_i{y}^j)d{x}^i$, and the coordinate formula gives $d(d{y}^j)=0$ by equality of mixed partial derivatives. Consequently, expressing a form as ${\sum }_I{b}_{I}d{y}^I$ and applying the $x$-coordinate formula gives exactly ${\sum }_{I}d{b}_I\wedge d{y}^I$, the $y$-coordinate formula. The local constructions agree on overlaps and therefore define a global exterior derivative.

For uniqueness, first derive locality from the axioms. If $\alpha$ vanishes near $p$, choose a smooth bump function $\chi$ supported in that neighborhood and equal to one near $p$. Then $\chi \alpha =0$, andimplies $(d\alpha {)}_p=0$. Thus the value depends only on the local germ of $\alpha$. Local coordinate functions can be extended using bump functions, so the same axioms apply to their germs. Since $d(d{x}^i)=d^2{x}^i=0$, the graded Leibniz rule forces precisely the coordinate formula above. The exterior derivative exists and is unique, as expressed by uniqueness of the exterior derivative from its axioms.

Define the pointwise representation of one-forms by vector-field functionals byIt is $\mathbb{R}$-linear, its value is a smooth function, and $X_p=0$ implies $\theta (\omega )(X)(p)=0$. Moreoverso it is a map of $C^{\infty }(M)$-modules. This is the properly typed scalar-multiplication property: the same symbol for a form and its associated functional is understood through $\theta$.

To construct its inverse, let $\alpha \in W$ and $v\in T_{p}M$. Choose a global smooth vector field $X$ with $X_p=v$; a local coordinate field multiplied by a smooth bump function supplies such an extension. SetThis is well-defined: if $X_p=Y_p$, then $X-Y$ vanishes at $p$, so $\alpha (X-Y)(p)=0$. Real linearity of $\alpha$ makes ${\omega }_p$ a covector. In fact the pointwise vanishing assumption also forcesbecause $(f-f(p))X$ vanishes at $p$. Thus $\alpha (fX)=f\alpha (X)$ for every smooth $f$, even though only real linearity was initially imposed.

It remains to prove smoothness, rather than assume continuity of an abstract linear map. Near a fixed point, choose global fields $E_1,\dots ,E_n$ equal to the coordinate basis on a smaller neighborhood, using a bump function equal to one there. On that neighborhood,All coefficients are smooth by the definition of $W$, so $\omega$ is a smooth differential one-form. The construction holds near every point, hence gives $\omega \in {\Omega }^1(M)$ with $\alpha (X)=\omega (X)$ globally.

Every tangent vector can be realized by a global field, so $\theta (\omega )=0$ forces $\omega =0$. The inverse just constructed gives surjectivity. The evaluation map is a natural $C^{\infty }(M)$-module isomorphism ${\Omega }^1(M)\cong W$. No auxiliary metric or connection was chosen.

A Riemannian metric is a smooth field of positive-definite symmetric bilinear forms $g_x$ on the tangent spaces, equivalently a smooth section of ${\mathrm{Sym}}^2{T}^{\ast }M$ with that positivity. Its musical isomorphism isNondegeneracy makes every fiber map an isomorphism. In coordinates the map has matrix $g_{ij}$, and its inverse ${♯}_g$ has matrix $g^{ij}$. Smoothness of the inverse follows from the inverse-matrix formula and the nonvanishing determinant. Both maps cover the identity on $M$, giving a smooth vector bundle isomorphism.

For local flattening of a Riemannian metric, choose a manifold chart neighborhood $V$ around $p$ with closure contained in $U$, and let $h={\varphi }^{\ast }\delta$ on $V$. Choose a smooth bump function $0\le \chi \le 1$, with compact support in $V$ and equal to one on a smaller neighborhood $W$ of $p$. SetThe formulas glue smoothly because the modification is supported strictly inside $V$. A convex combination of positive-definite forms is positive definite. On $W$ the metric is exactly $h$, so $\varphi$ gives an isometry to a Euclidean open subset. Outside $U$ the original metric remains unchanged.

Geodesic completeness means that every geodesic with arbitrary initial point and tangent vector extends for every real affine parameter. On each connected component, the Hopf-Rinow theorem equates this with completeness of the Riemannian distance.

The Euclidean-end claim needs a compact-core interpretation that is absent from its literal hypotheses. Indeed, $M=\{x\in {\mathbb{R}}^n:|x|>1\}$ with $n\ge 2$, $X=\varnothing$ and $g=\delta$ satisfy those hypotheses. The geodesic $\gamma (t)=(2-t)e_1$ reaches the missing unit sphere at $t=1$ and cannot continue within $M$. Thus the printed assumptions alone do not prove completeness.

Here is the intended compact-core completeness for a Euclidean end. Write $\Phi :M\setminus X\to \{|x|>1\}$ for the end coordinates and assume additionally thatis compact for every sufficiently large $R$. Choose such an $R$ beyond the metric transition and large enough to include the initial point of a geodesic. Its speed $s$ is constant. On every segment outside $K_R$, the Euclidean radius changes at a rate at most $s$, so over a finite parameter interval it cannot exceed $R+sT$. The full segment therefore stays in the compact set $K_{R+sT}$. Its velocities also stay in a compact subset of $TM$, since their metric norm is fixed and the base set is compact. The smooth geodesic ordinary differential equation consequently extends past any supposed finite endpoint. The reversed-time argument is identical. This proves completeness under the compact-core condition, and explains exactly what the exterior-of-a-ball counterexample lacks.

For upward stability of Riemannian completeness, lengths of all curves satisfy $L_{\tilde{g}}\ge L_g$, hence $d_{\tilde{g}}\ge d_g$. A $d_{\tilde{g}}$-Cauchy sequence is therefore $d_g$-Cauchy and has a limit $p$ because $g$ is complete. Smooth positive-definite metrics induce the manifold topology. More explicitly, on a small coordinate ball around $p$, $\tilde{g}$ has a bounded largest matrix eigenvalue, so the coordinate straight segment gives $d_{\tilde{g}}(p,q)\le C|\varphi (q)-\varphi (p)|$. Thus convergence to $p$ is also in $d_{\tilde{g}}$. That distance is complete, and Hopf-Rinow yieldsFor disconnected $M$, apply this argument on each component.

The exterior derivative is an $\mathbb{R}$-linear map $d:{\Omega }^p(M)\to {\Omega }^{p+1}(M)$, agrees with the differential $df$ of a smooth function, satisfies the graded Leibniz rule $d(\alpha \wedge \beta )=d\alpha \wedge \beta +(-1{)}^p\alpha \wedge d\beta$ for $\alpha \in {\Omega }^p(M)$, and has $d^2=0$. These properties determine it locally and hence globally. To see locality directly, if a differential form $\alpha$ vanishes near $x$, choose a smooth bump function $\chi$ equal to one near $x$ and supported where $\alpha =0$. The identity $d(\chi \alpha )=d\chi \wedge \alpha +\chi d\alpha$ gives $(d\alpha {)}_x=0$. Thus global forms can be computed using local extensions in a manifold chart.

In coordinates write $\alpha ={\sum }_I{a}_{I}d{x}^{i_1}\wedge \cdots \wedge d{x}^{i_p}$. Since $d(d{x}^i)=d^2{x}^i=0$, the graded Leibniz rule forcesThis is the uniqueness of the exterior derivative from its axioms. The coordinate formula also establishes existence: it has the stated properties, and the chain rule shows that coordinate changes give the same operator.

The de Rham cohomology is the real quotient vector spaceThus a cohomology class records a closed differential form modulo an exact differential form. In degree zero there are no exact forms; closed functions are locally constant. The Poincare lemma says that every closed form of positive degree on a star-shaped open subset of ${\mathbb{R}}^n$ is exact, and hence that positive-degree closed forms are locally exact on a smooth manifold.

For the first de Rham cohomology of the two-sphere, let $U=S^2\setminus \{N\}$ and $V=S^2\setminus \{S\}$. Stereographic projection identifies each with ${\mathbb{R}}^2$, so a closed one-form $\alpha$ has primitives $f_U,f_V$. On the connected overlap $U\cap V$, the derivative of $f_U-f_V$ is zero, so this difference is a constant. Subtracting that constant from $f_U$ makes the primitives agree. They glue to a global smooth primitive. Therefore $H_{\mathrm{dR}}^1(S^2)=0$.

Let $q:S^2\to {\mathbb{RP}}^2$ be the double covering map and $a$ the antipodal map. The pullback of a differential form identifies forms downstairs with $a$-invariant forms upstairs. Averaging $(\eta +a^{\ast }\eta )/2$ commutes with $d$. If an invariant form is exact upstairs, averaging its primitive proves it exact downstairs. Conversely an invariant cohomology class has an invariant representative by the same averaging. This proves the de Rham cohomology of a finite quotient identification $H_{\mathrm{dR}}^p({\mathbb{RP}}^2)\cong H_{\mathrm{dR}}^p(S^2{)}^{a^{\ast }}$. In degree zero the sphere is connected and $a^{\ast }$ fixes constants; degree one is zero; in degree two the granted action is multiplication by $-1$, whose invariant real subspace is zero. Forms of degree greater than two vanish. HenceThis is real de Rham cohomology; it does not detect the integral two-torsion of the real projective plane.

A fiber metric on a real vector bundle $E\to M$ is a smoothly varying positive-definite inner product on each fiber. Choose a trivializing open cover $\{U_i\}$ and a smooth partition of unity $\{{\rho }_i\}$ subordinate to it. The partition theorem gives nonnegative functions summing to one, a locally finite family of subsets of supports, and $\mathrm{supp}{\rho }_i\subset U_i$; the usual Hausdorff second-countable smooth manifold hypotheses ensure this theorem applies. Transfer the Euclidean inner product to each local vector bundle trivialization, obtaining $h_i$, and setEach weighted term extends smoothly by zero outside $U_i$, and local finiteness makes the sum smooth in every vector bundle trivialization. At each $x$ some weight is positive, so $h_x(v,v)>0$ for every nonzero $v$. This proves existence of a fiber metric, with no orientability or triviality assumption.

A vector bundle morphism covering the identity is a smooth map $F:E^{\prime }\to E^{\prime \prime }$ that preserves base points and is a linear map on each fiber. Its induced map on the module of smooth sections is $s\mapsto F\circ s$, and is $C^{\infty }(M)$-linear. We prove the converse by constructing bundle morphisms from maps of smooth sections.

First the given map $\alpha$ is local. If a global section $s$ vanishes on a neighborhood of $x$, take a smooth bump function $\chi$ supported there with $\chi =1$ near $x$. Then $\chi s=0$, so $\chi \alpha (s)=\alpha (\chi s)=0$, and hence $\alpha (s)(x)=0$. Thus sections agreeing near $x$ have images agreeing at $x$.

Choose a local frame $e_1,\dots ,e_r$ on $U$, and a bump function equal to one on a smaller neighborhood $V$ of $x$ and supported in $U$. Multiplying the frame by that bump and extending by zero gives global smooth sections $t_j$ whose restrictions to $V$ are the frame. If $s(x)=0$, write $s=\sum a_j{e}_j$ on $V$. A second bump extends each $a_j$ to a global smooth function ${\tilde{a}}_j$ agreeing near $x$. By locality and $C^{\infty }(M)$-linearity,Every fiber vector $v\in E_x^{\prime }$ is the value of a global smooth section, by the same bumped-frame construction. Define $F_x(v)=\alpha (s)(x)$ for any such section. The just-proved vanishing statement makes this well-defined. The maps $F_x$ are linear maps, and locally their matrix columns are the smooth sections $\alpha (t_j)$ in a frame of $E^{\prime \prime }$. Thus $F$ is smooth, is a vector bundle morphism, and satisfies $\alpha (s)=F\circ s$. Fiberwise evaluation also proves uniqueness.

Finally apply the fiber metric construction to the tangent bundle. A Riemannian metric $g$ defines the musical isomorphismPositive definiteness makes it a fiberwise bijection; its inverse is smooth because inverse metric matrices vary smoothly. Hence $TM$ and $T^{\ast }M$ are isomorphic as real smooth vector bundles on every such manifold. The isomorphism depends on the chosen metric and is not canonical.

For fixed-design nonparametric regression, let $x_i=i/n$ and $K_h(u)=h^{-1}K(u/h)$. The degree-$p$ local polynomial estimator takes the fitted intercept ${\hat{m}}_h(x;p)={\hat{\beta }}_0$, whereUse a nonnegative regression kernel and an invertible local polynomial Gram matrix to obtain a unique fit. For $p=0$, solving the single weighted least squares normal equation gives the local constant estimator:At the interior point $x_0$, use the standard unit-integral regression kernel convention ${\mu }_0(K)=1$. Otherwise the leading variance below is $R(K)/(nh{\mu }_0(K{)}^2)$; unlike a kernel density estimator, the local constant estimator itself is unchanged by rescaling $K$.

The local constant estimator is a linear estimator in nonparametric regression with weights $W_i=K_h(x_i-x_0)/(n{s}_{0,h})$. The independent unit-variance errors implyTo control the bias of an estimator, differentiability at this one point is enough. WriteSymmetry makes ${\mu }_1(K)=0$, and the supplied moment approximation gives $s_{1,h}=o(h)$. Therefore the linear contribution to the bias of an estimator is $m^{\prime }(x_0)s_{1,h}/s_{0,h}=o(h)$. The remaining contribution has absolute value at most $\eta (h)h$, because only $|x_i-x_0|\le h$ receive positive weight. This is the interior first-order bias cancellation for local constant regression, and gives squared bias of an estimator $o(h^2)$. Combining with the variance provesFor fixed $\epsilon >0$, take $h_{\epsilon }={\epsilon }^{-1/3}{n}^{-1/3}$. It satisfies both asymptotic smoothing bandwidth conditions, and the displayed remainder, multiplied by $n^{2/3}$, tends to zero for this fixed $\epsilon$. ThusLetting $\epsilon \downarrow 0$ proves the optimized scaled pointwise error tends to zero. The order of these limits matters: no remainder uniform in $\epsilon$ has been assumed.

For the lower bound over the whole function class, the absence of a uniform bound on derivatives permits a particularly direct argument. It also avoids assuming a normal distribution for errors when the model only supplies their first two moments. Fix $n$ and $x_0$. Choose a smooth bump function $b$ on the interval, with $b(x_0)=1$, whose support contains no design point except possibly $x_0$. Such a bump exists because the design is finite; at an endpoint use the restriction of a smooth bump on the real line. Every $m_a=ab$, $a\in \mathbb{R}$, belongs to the differentiability class.

If $x_0$ is not a design point, all observation means are zero for every $a$. The Le Cam two-point lemma applied to $a=\pm A$ has total variation distance zero and gives a lower bound tending to infinity with $A$. In fact, the supremum of the mean squared error is infinite at every such $x_0$.

If $x_0$ is a design point, only its response depends on $a$. The resulting location family is the single observation $Y=a+ϵ$; all remaining observations are independent of $a$ and give no extra information. The single-observation location minimax bound isHere is a proof covering discrete as well as continuous errors. Give $a$ the prior distribution $U[-A,A]$, independently of the noise. Put $p_M=\mathbb{P}(|ϵ|\le M)$ and $v_M=\mathrm{Var}(ϵ\mid |ϵ|\le M)$. Consider an easier experiment in which an oracle also reveals whether $|ϵ|\le M$, and reveals $a$ itself on the complementary event. On the event $|ϵ|\le M$ and $|Y|\le A-M$, every possible truncated noise value places $a=Y-ϵ$ within $[-A,A]$. The flat prior distribution then makes the Bayesian posterior noise law exactly its truncated original law. The smallest conditional squared-error loss is its conditional variance $v_M$, attained by the posterior mean.

The probability of that event is at least $p_M(1-2M/A)$ for $A>2M$, by restricting further to $|a|\le A-2M$. Thus every rule's supremum risk function is at least the original Bayes risk, which is at least the easier experiment's Bayes risk, and therefore at least $p_M(1-2M/A)v_M$. Let $A\to \infty$ and then $M\to \infty$. Finite second moment and zero mean give $p_M{v}_M\to 1$. The rule $T(Y)=Y$ has constant mean squared error one, proving the equality.

Consequently, at every $x_0\in [0,1]$, the unrestricted differentiability class satisfies the stronger conclusionThus $c=1$ works for every $n\ge 1$. There is no conflict with the preceding fixed-function limit: the worst functions, including arbitrarily narrow bumps, can vary with $n$. This is the pointwise versus uniform risk distinction. A smoothness ball with a common derivative bound would pose a different minimax risk problem.

Put $v_p(u)=(1,u,\dots ,u^p{)}^T$ and $w_i(x)=K((x_i-x)/h)$, where $p$ is a nonnegative integer and $h$ is the smoothing bandwidth. The local polynomial estimator is the intercept of the weighted least squares fitWhen the local polynomial Gram matrix $G_p(x)={\sum }_i{w}_i(x)v_p((x_i-x)/h)v_p((x_i-x)/h{)}^T$ is invertible, the normal equations giveA singular local polynomial Gram matrix requires a selection convention and may leave the intercept unidentified; an arbitrary nonnegative regression kernel alone does not guarantee uniqueness. At $p=0$, provided ${\sum }_i{w}_i(x)>0$, the Nadaraya–Watson estimator isMultiplying every weight by $1/h$ changes neither fit.

For the uniform kernel, let $I_x=\{i:|i/n-x|\le h\}$ and $N_x=|I_x|$. Then ${\hat{m}}_h(x)=N_x^{-1}{\sum }_{i\in I_x}{Y}_i$. For $0<h<1/2$ and $x\in [h,1-h]$, the interval $[x-h,x+h]$ is inside $[0,1]$, and the regular design hasThe same lower bound holds when the interval starts at zero, despite the absence of the design point zero. In particular the Nadaraya–Watson estimator is defined throughout the integration interval.

For every Lipschitz continuous mean function in ${\Theta }_L$, the bias of an estimator and variance satisfyHere the variance uses the independence of the errors in the fixed-design nonparametric regression model. The bias-variance decomposition of mean squared error and Tonelli theorem give a slightly stronger integrated mean squared error bound than required:ThusThe printed integration interval represents a nonnegative risk function for $h\le 1/2$. For $1/2<h\le 1$, the written integral has reversed limits and is nonpositive, making the displayed upper bound trivial; it should not be interpreted as an integrated mean squared error in that range. For $h>1$, evaluation outside $[0,1]$ would additionally require a specified extension of the mean function. The subsequent optimization restricts to $h<1/3$.

Minimizing the stated right side givesFor sufficiently large $n$, $n{h}_{\ast }\ge 1$ and $h_{\ast }<1/3$. For example take $n_0=1+\lceil \max \{1,2L/(\sqrt{3}\sigma ),81{\sigma }^2/(4L^2)\}\rceil$. Therefore the upper bound holds with

For the lower bound, the PDF has the mean squared error ${\mathbb{E}}_m[(\tilde{m}(x_0)-m(x_0){)}^2]$; the TeX erroneously moves the square outside the expected value. The PDF's suggested smooth bump function is $\exp (-1/(1-t^2))1_{\{|t|\le 1\}}$, with value zero at the endpoints, rather than the corrupted TeX formula. A triangular alternative suffices because only Lipschitz continuity is needed.

The Le Cam two-point lemma says that for two data laws $P_0,P_1$ with parameter separation $\Delta =|{\theta }_1-{\theta }_0|$, every estimator $T$ satisfiesIndeed, classify according to the nearer parameter value. Misclassification incurs squared-error loss at least ${\Delta }^2/4$, and the sum of the two error probabilities is at least one minus the total variation distance. Averaging the two risk functions proves the stated form.

Let $d=\min (x_0,1-x_0)>0$, $m_0=0$, and use the two-point lower bound with a triangular bumpBoth functions are in ${\Theta }_L$, and $\Delta =Lh$. Assume first that $h\le d$ and $nh\ge 1$. At most $2nh+1$ design points meet the bump's support. The Kullback-Leibler divergence between normal distributions and chain rule for relative entropy giveThe Pinsker's inequality therefore gives $\Vert P_1-P_0{\Vert }_{\mathrm{TV}}\le \sqrt{3/32}<1/2$. The Le Cam two-point lemma yields the pointwise minimax rate for Lipschitz regressionThis universal constant is valid for $n\ge N:=\lceil \max \{1,{\sigma }^2/(8L^2{d}^3),2\sqrt{2}L/\sigma \}\rceil$.

If the printed lower bound is read as applying to all positive integers $n$, it also holds with a smaller constant depending on the fixed $L,\sigma ,x_0$. For $n<N$, use the constant mean functions $m_0=0$ and $m_1=\sigma /\sqrt{n}$. They have Kullback-Leibler divergence $1/2$, so Pinsker's inequality and the Le Cam two-point lemma give minimax risk at least ${\sigma }^2/(16n)$. Combining the two ranges gives the explicit all-$n$ choiceThe universal asymptotic constant and this parameter-dependent all-$n$ constant are distinct conventions; the latter supplies the assertion without inserting an unstated large-$n$ restriction.

Use the interior-domain definition of total variation seminorm on a domain, so there is no boundary-extension term at $\pm 1$. For any admissible scalar test function $\phi$,Thus $\mathrm{TV}(u_R)\le 2$.

Choose $0<\epsilon <\min \{R,1-R\}$. From the supplied smooth bump function, defineThe two supports are disjoint and inside $(-1,1)$. Each bump has maximum $e^{-1}$, so $\Vert \phi {\Vert }_{\infty }\le 1$, while $\phi (R)=1$ and $\phi (-R)=-1$. This attains the upper bound, provingEquivalently, its distributional derivative is $D{u}_R={\delta }_{-R}-{\delta }_R$, with total variation two. This is total variation of an interval indicator.

Choose a coordinate ball $W$ around $q_1$ and a smaller convex coordinate ball $U$ whose closure lies in $W$. For $q_2\in U$, let $a$ be the constant coordinate vector from $q_1$ to $q_2$. Choose a smooth bump function $\rho$ supported in $W$ and equal to one on $U$, and define in the chartextending it by zero outside $W$. This is a compactly supported smooth vector field. The segment $q_1+ta$ stays in $U$, where $w=a$, so uniqueness for ordinary differential equations givesIn particular, ${\Psi }^1(q_1)=q_2$.

Suppose the immersed subset is not embedded. Using the assumed embedded neighborhoods, there are $p\in N$, a relatively small coordinate neighborhood $U\ni p$, and points $p_j\in N\setminus U$ with $p_j\to p$ in $M$. Choose a bump function $g$ on $N$, supported in $U$, with $g(p)=1$. Then $g(p_j)=0$.

If $g=f{|}_N$ for some smooth $f$ on $M$, continuity gives both $f(p_j)\to f(p)=1$ and $f(p_j)=0$, a contradiction. Thus the extension hypothesis forces the subspace and manifold topologies to agree locally, and the immersion is an embedding. This proves the smooth extension criterion for an immersed submanifold.

Put $b=\pi (p)$ and $u=(d\pi {)}_{p}v$. Choose a vector field $X$ near $b$ with $X(b)=u$, multiply it by a bump function, and extend it by zero to $B$. At each $q\in E$, the restrictionis an isomorphism. Define $V(q)$ as the unique horizontal lift of $X(\pi (q))$. Smoothness of the horizontal distribution makes $V$ smooth, and uniqueness gives $V(p)=v$.