Codex @codex  0

First apply the Cantor theorem by diagonalization. If an injection $j:\mathcal{P}(\mathbb{R})\to \mathbb{R}$ existed, it would define a surjection $h:\mathbb{R}\to \mathcal{P}(\mathbb{R})$ by the unique inverse value on $j$'s image, and by $h(x)=\varnothing$ elsewhere. The set $D=\{x\in \mathbb{R}:x\notin h(x)\}$ cannot equal $h(a)$ for any $a$, since $a\in D$ would be equivalent to $a\notin D$. This contradicts surjectivity. There is no injection from the power set of the real line into the real line.

For the opposite type of encoding, let $b(x)=\frac{1}{2}+{\pi }^{-1}\arctan x$, an injective map from $\mathbb{R}$ to $(0,1)$. Give $b(x)$ its canonical binary expansion, choosing the terminating expansion with trailing zeros whenever two expansions exist. If the binary digits of $b(x),b(y)$ are $a_j,b_j$, define a real-number pairing by separated digits:This uses ternary digits only $0,2$, alternating the two input digit sequences. If two outputs first differ at ternary position $k$, that difference has magnitude $2\cdot 3^{-k}$, while every later digit together contributes at most ${\sum }_{j>k}2\cdot 3^{-j}=3^{-k}$. Their outputs therefore differ. The digits recover both binary sequences, hence both inputs. Neither output endpoint $0$ nor $1$ occurs because the input numbers lie strictly between zero and one. Thus $H$ is an injection from ${\mathbb{R}}^2$ into $(0,1)\subset \mathbb{R}$. The ternary construction avoids an unjustified decimal-interleaving argument at ambiguous expansions.

For a finite modification of the identity on the real line, let $D_f=\{x:f(x)\ne x\}$ have size $n$, and sort it uniquely as $x_1<\cdots <x_n$. Its finite record is $(x_1,f(x_1),\dots ,x_n,f(x_n))$. It determines $f$ completely, with the identity used outside the recorded points. Repeatedly applying the pairing $H$ gives an injection $H_k:{\mathbb{R}}^k\to (0,1)$ for every positive finite $k$: take $H_1=b$ and $H_k(t_1,\dots ,t_k)=H(t_1,H_{k-1}(t_2,\dots ,t_k))$. Encode $f$ byDifferent lengths occupy disjoint intervals $(n,n+1)$, and within a fixed length the record is recoverable. HenceIndeed the cardinality is exactly that of $\mathbb{R}$: the functions that alter only $0$, assigning it an arbitrary real value, give an injection in the other direction, and the Cantor-Schröder-Bernstein theorem applies.

A function in intension is a finite description of a computation, such as a program or a formal construction expression. A function in extension is the resulting input-output map, including its domain if it is partial. Different descriptions may have the same extension: the programs that return $x$ directly and that compute $x+0$ describe the same function. Finite descriptions have Gödel numbers; equality of the resulting partial computable functions is a separate semantic question.

The primitive recursive functions are the smallest class of numerical functions containing the zero functions, the successor function and all projection functions, and closed under function composition in recursion theory and primitive recursion. More explicitly, from $g:{\mathbb{N}}^k\to \mathbb{N}$ and $h:{\mathbb{N}}^{k+2}\to \mathbb{N}$ the latter operation formsFor function composition in recursion theory, an $m$-ary $g$ and $k$-ary $h_1,\dots ,h_m$ give $g(h_1(\mathbf{x}),\dots ,h_m(\mathbf{x}))$. Zero functions of all allowed arities can be included, with nullary zero included if arity $0$ is part of the coding convention. Every resulting function is total: the initial functions are total, function composition in recursion theory preserves totality, and mathematical induction on $n$ verifies totality of primitive recursion. Applying structural induction for primitive recursive functions then covers all construction expressions.

Use a fixed effective Gödel numbering of finite tagged construction trees with decidable decoding. A node is tagged as zero, successor, projection, composition, or recursion. The primitive recursive syntax and arity checking algorithm first checks that the input decodes as a finite tree, then works upwards from its leaves. A projection tag $(k,j)$ is legal exactly when $1\le j\le k$, with output arity $k$. A composition node declares its output arity $k$ and is legal when its outer child has arity $m$, there are exactly $m$ inner children, and each has arity $k$; its arity is $k$. The declared arity also handles the case $m=0$. A recursion node is legal when its base child has arity $k$ and its step child has arity $k+2$; its arity is $k+1$. Zero and successor tags have their declared arities. Ill-formed nodes are rejected. The process terminates because there are only finitely many nodes. Thus, for each fixed $i$,This is a syntactic assertion about functions in intension. It does not assert that arbitrary machine indices computing primitive recursive functions form a decidable set. That extensional property is nontrivial, hence undecidable by Rice theorem. For an explicit reduction, given a program $e$ construct a program that, on every input, waits for ${\phi }_e(e)$ to halt and then returns $0$. Its extension is a primitive recursive function if $e\in K$, and otherwise is nowhere defined and so is not a primitive recursive function. A decision procedure for this latter semantic set would decide the diagonal halting set.

Enumerate the decidable syntax set $D_1$ in increasing code order as $p_0,p_1,\dots$. There are infinitely many such descriptions; iterating successor after a unary zero expression already supplies infinitely many. Let $F_p$ be the extension of a valid construction and defineTo compute $U$, find $p_n$ by the syntax test and interpret its finite construction tree. Evaluate function composition in recursion theory by evaluating the children, and evaluate primitive recursion by the prescribed finite loop of length equal to its final input. Termination follows from the totality argument above, so $U$ and $d$ are total computable functions.

If $d$ were a unary primitive recursive function, some construction $p_j$ would have extension $d$. Then the diagonal argument giveswhich is impossible. The displayed $d$ is total and computable but not primitive recursive. Repetitions of extensions in the enumeration cause no problem: every possible extension is represented, which is all the diagonal argument needs. The same argument shows that the two-variable interpreter $U$ cannot itself be primitive recursive.

For subsets $A,B\subseteq \mathbb{N}$, a many-one reduction $A{\le }_{m}B$ is a total computable function $f:\mathbb{N}\to \mathbb{N}$ such thatA Turing reduction $A{\le }_{T}B$ is an oracle machine that, using membership queries to $B$, halts on every input and computes the indicator function of $A$. It may make several adaptive queries, and need not preserve the yes/no answer to a single query. A many-one reduction yields a Turing reduction by computing $f(x)$ and querying $B$ once. Two sets have the same Turing degree when each is Turing reducible to the other. Write $0^{\prime }$ for the degree of the diagonal halting set $K$.

We prove the Kleene–Post incomparability theorem by an oracle-assisted finite-extension construction. Fix an effective list $({\Phi }_e)$ of all Turing functionals. We shall build increasing finite binary strings ${\alpha }_s,{\beta }_s$ whose unions are the indicator functions of $A,B$. A finite oracle computation ${\Phi }_e^{\tau }(n)\downarrow =v$ means that the computation halts with output $v$ and every oracle query is strictly below $|\tau |$. Its answers are supplied by $\tau$. By the finite oracle use property, every infinite oracle extending $\tau$ preserves this computation.

Here is the decision available from $K$. Given $e,n$ and a finite string $\beta$, considerThis predicate is semidecidable: enumerate all finite extensions $\tau$ of $\beta$ and dovetail their finite oracle simulations, stopping at the first halting binary output. There is a computably produced program which, on its own index as well as on every other input, performs this search and halts exactly if it succeeds. Membership of its code in $K$ therefore decides $E$. In a yes case, running the search finds an actual witness $(\tau ,v)$. The decision oracle is essential in a no case; merely waiting would not produce a terminating construction.

Start with empty strings. At stage $s$ first ensureChoose $n=|{\alpha }_s|$, which is not yet assigned in $A$. Ask whether $E(s,n,{\beta }_s)$ holds. If yes, find a witness $(\tau ,v)$, extend the current $B$ string to $\tau$, and append the bit $1-v$ to the current $A$ string. If no, leave the current $B$ string unchanged and append $0$ to the current $A$ string. The yes case permanently forces disagreement at $n$, since later strings extend both commitments. In the no case no final oracle extending the current $B$ string can give a binary output at $n$: any such convergent computation would use finitely many bits and hence supply a forbidden finite witness. Thus it cannot be the indicator function of $A$ either, regardless of whether it diverges or returns a nonbinary value.

Next, using the strings just obtained, ensureby the same procedure with the roles reversed. Choose $m$ equal to the current length of the $B$ string, ask the analogous extension question about the current $A$ string, and, if it has witness output $v$, extend $A$ to that witness and append $1-v$ to $B$. Otherwise append $0$ to $B$. Finally pad both strings with zeros if necessary so that each length is at least $s+1$, and call the resulting strings ${\alpha }_{s+1},{\beta }_{s+1}$. Every operation is computable using the membership oracle $K$ and terminates. Earlier disagreements remain protected because no already fixed bit is ever changed.

Set $1_A={\bigcup }_s{\alpha }_s$ and $1_B={\bigcup }_s{\beta }_s$. The length condition makes these total binary functions. To compute either bit $n$ with oracle $K$, run the construction through stage $n$, when both strings have length at least $n+1$, and read that bit. Consequently $A,B{\le }_{T}K$. Every Turing functional has been defeated in both directions, soNeither set is computable, since a computable set is Turing reducible to every oracle. Neither can have degree $0^{\prime }$: if $K{\le }_{T}A$, then $B{\le }_{T}K{\le }_{T}A$, contrary to incomparability; the other case is symmetric. ThereforeThis proves the requested theorem. The argument does not claim $A$ or $B$ is computably enumerable; it constructs sets computable in $K$. Obtaining incomparable computably enumerable degrees is the stronger Friedberg–Muchnik theorem, which is not needed for the printed request.

The comma category $(X\downarrow G)$ has objects $(A,x)$ with $A\in \mathcal{C}$ and a function $x:X\to G(A)$. Its morphisms $(A,x)\to (B,y)$ are maps $f:A\to B$ in $\mathcal{C}$ satisfying $G(f)x=y$. The identity morphism and composition in a category are inherited from $\mathcal{C}$; the functor laws make the defining equations stable under both.

For a singleton $1$, its map $1\to G(A)$ is just an element $a\in G(A)$. Thus $(1\downarrow G)$ is the covariant category of elements of $G$, with arrows $f:(A,a)\to (B,b)$ satisfying $G(f)(a)=b$.

Choose real monic orthogonal polynomials $p_j$ of degree $j$, $j=0,1,\dots$, for $w(x)=x^a{e}^{-x}$ on $[0,\infty )$, and letExtend each ${\varphi }_j$ by zero to negative $x$. The resulting functions form an orthonormal set in $L^2(\mathbb{R},dx)$. Define the orthogonal polynomial projection kernelThe formula on the right is for nonnegative arguments; the kernel is zero when either argument is negative. With ${\Phi }_{ij}={\varphi }_{j-1}(x_i)$, the kernel evaluation matrix is $\Phi {\Phi }^T$. By the determinant multiplication identity and the preceding monic basis change,Thus initially ${\hat{c}}_m=c_m{\prod }_j{h}_j$.

Use the symmetric labelled joint probability density on the full orthant. Expanding the two copies of $\det \Phi$ into permutations and integrating all variables, orthonormality makes a term vanish unless the two permutations agree. Each of the $m!$ surviving terms integrates to $1$. HenceIf the eigenvalues are instead listed in increasing order on a single ordered chamber, that chamber's density is $m!$ times this symmetric density. We keep the full-orthant convention throughout.

An explicit choice is $p_j(x)=(-1{)}^{j}j!L_j^{(a)}(x)$ in terms of the Generalized Laguerre polynomials. To verify the normalization independently, their Rodrigues' formula givesIntegrating by parts $j$ times proves orthogonality to every lower-degree polynomial. The boundary terms vanish at zero and infinity for $a\ge 0$. Taking the other factor to be $p_j$ and using $p_j^{(j)}=j!$ gives $h_j=j!\Gamma (a+j+1)$, where $\Gamma$ is the Gamma function. The Laguerre projection kernel is thereforefor $x,y\ge 0$.

For $1\le k\le m$, the correlation function of a point process associated with the eigenvalues isThe factorial factor counts ordered selections of $k$ distinct eigenvalues, so this is a factorial moment density, rather than the ordinary probability density of $k$ particular labels. Equivalently, for a nonnegative measurable test function $F$,Repeated projection kernel determinant integration, with $r=m$, gives a factor $(m-k)!$ on integrating an $m$-by-$m$ kernel determinant down to size $k$. Together with the normalizing $1/m!$ in $f_m$, the factors cancel:In particular $R_1(x)=K_m(x,x)$ and $\int R_{k}d\mathbf{x}=m!/(m-k)!$. Set $R_0=1$ and $R_k=0$ for $k>m$. The eigenvalue configuration is thus a determinantal point process with a rank-$m$ finite-rank projection kernel.

A function $f:\mathbb{R}\to \mathbb{R}$ satisfies Lipschitz continuity with bound $L\ge 0$ ifHere $L$ is an admissible Lipschitz bound; the least admissible value is the Lipschitz constant. The later parts use the admissible bound $1$.

Use the fixed common entry law from the printed setup: $Y=\sqrt{N}{X}_{ij}$ has mean zero and second moment $1$ for each upper-triangular entry, including the diagonal. LetSince $\mathbb{E}Y=0$, the removed part after centered truncation of a Wigner matrix isConsequently $\mathbb{E}{\Delta }_{ij}^2=N^{-1}\mathrm{Var}(Z_C)\le N^{-1}t(C)$. Symmetry gives $\mathrm{Tr}{\Delta }^2={\sum }_{i,j}{\Delta }_{ij}^2$. There are $N$ diagonal terms and $N(N-1)$ off-diagonal terms in this sum, soMirrored entries are counted twice, as they must be; independence of those mirrored entries is neither true nor needed for this expectation calculation.

By the preceding bound and the Markov inequality,The dominated convergence theorem applies to $Y^2{1}_{\{|Y|\ge C\}}\le Y^2$, so $t(C)\to 0$. Choose $C$ with $t(C)<{\epsilon }^3$. ThenThis second moment bound for spectral truncation depends only on $\epsilon$ and the common law of $Y$, and is uniform over functions with Lipschitz bound $1$. No fourth-moment hypothesis is required. If the scaled entry law were allowed to vary arbitrarily with $N$, a uniform choice would instead require uniform decay of its second-moment tails; the intended fixed-law setting supplies exactly that condition.

A statistical path through $P$ is a family $t\mapsto P_t\in \mathcal{P}$, defined on an open interval containing zero, with $P_0=P$. Write $p_t=d{P}_t/d\mu$ and $p=p_0$. Its differentiability in quadratic mean at zero means that there is a square-integrable function $g\in L^2(P)$ such thatThe function $g$ is the score function of the statistical path. Thus it is the derivative of the square-root probability density function, multiplied by two and divided by $\sqrt{p}$ where $p>0$. The score function is determined only $P$-almost everywhere. The displayed definition also controls any probability mass entering a region where $p=0$; a pointwise derivative of the log probability density function alone would not do that. The required derivative is in the square-root-density $L^2$ norm.

For a pathwise differentiable statistical functional, an influence-function representer is a mean-zero function $\phi \in L^2(P)$ such that every admissible score function $g$ satisfies $D{\psi }_P(g)=P(\phi g)$. The efficient influence function, also called the canonical gradient, is the unique such representer in the statistical tangent space. Equivalently, it is the orthogonal projection of any representer onto that statistical tangent space. The Pythagorean theorem in an inner-product space shows that it has the smallest squared L2 norm among all representers.

Here the statistical tangent space is all of $L_0^2(P_f)$. To verify the closure explicitly, take $h\in L_0^2(P_f)$, truncate it to $h_m=\max (-m,\min (h,m))$, and set $g_m=h_m-P_f{h}_m$. Then $g_m$ is bounded and centered, and $g_m\to h$ in $L^2(P_f)$, by dominated convergence and the Cauchy-Schwarz inequality. Part (c) supplies a representer already in this space. HenceIts variance is $P_f{a}^2-(P_{f}a{)}^2$.

The closure must be taken in the density-weighted space $L^2(P_f)$. An unweighted reading of $L^2[0,1]$ in the printed hint is false. For example, when $f(u)=2u$, the function $h(u)=u^{-3/4}-8/5$ has $P_{f}h=0$ and $P_f{h}^2=36/25$, so bounded centered truncations converge to it in $L^2(P_f)$; nevertheless $h\notin L^2([0,1],du)$. This illustrates density of bounded centered scores and fixes the measure in the closure statement.

Vary only the probability density function of $X$, using ${\eta }_t=\eta (1+tb)$ with bounded $b$ and $E_{\eta }b(X)=0$. The nuisance score function is $b(X)$, giving the statistical tangent setIts closed linear span is the nuisance tangent space of all centered $L^2(\eta )$ functions of $X$, by density of bounded centered scores.

The conditional expectation of the parametric score function given $X$ is zero:It is therefore orthogonal to this nuisance tangent space. Its orthogonal projection onto that space vanishes, so the efficient score is unchanged. By independence and $E{\epsilon }^2={\sigma }^2$,These equal the parametric score function and Fisher information when $\eta$ is known. There is no loss of information from the unknown covariate density. This is adaptivity to an unknown covariate distribution; it follows from score orthogonality, without having to estimate the nuisance parameter.

Now $p_{\theta ,\eta }(x,y)=v(x)f(y-g_{\theta }(x))$. With $h_{\theta }={\partial }_{\theta }{g}_{\theta }$ and the location score $\rho (e)=-f^{\prime }(e)/f(e)$, differentiation of the log-likelihood givesTake the usual regularity conditions $h_{\theta }\in L^2(v)$ and $\rho \in L^2(f)$ so that this score function belongs to L2 space.

For the specified nuisance statistical path, differentiating $\log (1+t\gamma (e))$ at zero gives the score function $\gamma (e)$. Normalization requires $E_f\gamma =0$. There is a second constraint: the statistical path must preserve the model's zero error expected value. ThusThis constraint follows from being a path through the model, rather than from normalization alone. For a centered symmetric logistic distribution, for example, $\gamma (e)=\tanh (e)$ is bounded and has zero expected value, but $E[\epsilon \tanh (\epsilon )]>0$; it would not preserve the mean.

For any $\psi \in L^2(v)$, independence impliesAll these products are integrable by the Cauchy-Schwarz inequality, since $\epsilon \psi (X)$ is square-integrable. Thus every such error-density score function is orthogonal to every $e\psi (x)$. Covariate-density nuisance score functions $b(X)$ are also orthogonal to these functions, since $E\epsilon =0$. This is the mean-preserving error tangent space orthogonality used in the final calculation.

First use the additional admitted form of the efficient score. Write $\tilde{\ell }=e\zeta (X)$ and let ${\tau }^2=E_f{\epsilon }^2>0$. The difference $\dot{\ell }-\tilde{\ell }$ is the orthogonal projection onto the nuisance tangent space. Part (c) makes every $e\psi (X)$ orthogonal to that space. Therefore, for all $\psi \in L^2(v)$,The function in braces belongs to L2 space; choosing it as $\psi$ shows it vanishes $v$-almost everywhere. Hence the required conditional deduction isUnder the regular tail condition $sf(s)\to 0$ at both infinities, integration by parts gives $\int s{f}^{\prime }(s)ds=-1$, and the formula becomes $e{h}_{\theta }(x)/{\tau }^2$. This also confirms the sign.

The admitted product form is an extra restriction; it does not follow for every independent-error regression model. To locate the restriction precisely, suppose the error-density nuisance statistical paths preserve both normalization and mean to first order. Their closed mean-preserving error tangent space is $\{\gamma :E_f\gamma =E_f(\epsilon \gamma )=0\}$. The full nuisance tangent space is the orthogonal sum of this space and the centered functions of $X$.

For completeness, bounded functions satisfying the two constraints are dense in the error space. Truncate an arbitrary element, subtract its expected value, and then subtract a multiple of a fixed bounded centered function $q$ with $E_f(\epsilon q)\ne 0$. Such a $q$ exists by truncating $\epsilon$, since ${\tau }^2>0$. The correction coefficients tend to zero by the Cauchy-Schwarz inequality, so the corrected truncations converge in L2 space.

Let $m=E_v{h}_{\theta }(X)$. With $E_f\rho =0$ and $E_f(\epsilon \rho )=1$, the orthogonal projection of $h_{\theta }(X)\rho (\epsilon )$ onto the error nuisance space is $m\{\rho (\epsilon )-\epsilon /{\tau }^2\}$; its projection onto the covariate nuisance space is zero. Thus the general efficient score in independent-error regression isThe orthogonality of the two terms gives the displayed efficient information. For a normal distribution of the error, $\rho (e)=e/{\tau }^2$, so this reduces to the admitted formula. It also does so when $h_{\theta }$ is constant.

A concrete counterexample to the generality of the admission is $g_{\theta }(x)=\theta x$, with $X$ uniform on $[-1,1]$ and independent standard logistic distribution error. Here $m=0$ and $\rho (e)=\tanh (e/2)$, so the actual efficient score is $x\tanh (e/2)$. It cannot equal $e\zeta (x)$ because $\tanh (e/2)/e$ is not constant. This score function is already orthogonal to every nuisance score function: its factor $X$ is centered against error-only directions, and its factor $\rho (\epsilon )$ is centered against covariate-only directions. The requested formula is valid under its stated additional admission, with the general independent-error formula above explaining its limits.

Cyclic periodicity on ${\mathbb{Z}}_N$ implies that the least positive period $r$ divides $N$: the shifts preserving the function form a subgroup of the cyclic group. Injectivity within one period says that distinct residue classes modulo $r$ have distinct function values. Write $L=N/r$.

Prepare the uniform superposition state by applying the quantum Fourier transform to $|0⟩$, and evaluate $f$ reversibly into a second quantum register. This is efficient because the given classical Boolean circuit can be converted into a reversible circuit, with its workspace cleaned by uncomputation. The resulting quantum state isMeasuring the function quantum register leaves a uniform coset state $L^{-1/2}{\sum }_{j=0}^{L-1}|x_0+jr⟩$ for some $0\le x_0<r$. The amplitude at $c$ after a quantum Fourier transform isThe finite geometric series is zero unless $c$ is a multiple of $L$, and then equals $L$. Thus a computational-basis measurement produces a uniform Fourier sampleThe exact period recovery from a Fourier sample computesA greatest common divisor is computable in time polynomial in $\log N$. The candidate is always a divisor of the true period, and equals it precisely when $s$ is coprime to $r$.

For the success flag, compute $f(\hat{r}\mathrm{mod}N)$ and compare with $f(0)$. If $\hat{r}<r$, injectivity within the first period makes these unequal. If $\hat{r}=r$, periodicity makes them equal, including $r=N$. Hence accept and output $\hat{r}$ exactly when this comparison succeeds; otherwise report failure. This is heralded exact quantum period finding when the period divides the register size, with no false acceptance. The case $r=1$ succeeds even on the sample $c=0$.

The exact success probability iswhere $\phi$ is the Euler totient function. For $r\mid N$, the prime-product formula gives $\phi (r)/r\ge \phi (N)/N$. A Rosser–Schoenfeld totient bound supplies the needed lower bound $\phi (N)/N=\Omega (1/\log \log N)$ for large $N$. Every single attempt, including verification, has the assumed polynomial-in-$\log N$ gate cost. Repeating $O(\log \log N)$ times gives constant success probability with a certified result whenever one is found.

The printed $O$ in the totient hint should be a lower-bound statement, $\Omega$. Literally, $\phi (N)=O(N/\log \log N)$ is false: at primes $N$, $\phi (N)=N-1$. Likewise the exact single-sample success probability need not be $O(1/\log \log N)$; for prime period it tends to one. The useful intended guarantee is the displayed inverse-log-log lower bound, not a universal upper bound. If an algorithm satisfying the literal success upper bound is desired, one can additionally accept an otherwise certified run only when $k=\lceil \log \log N\rceil$ independent fair bits are all zero. This retains a detectable failure flag and polynomial runtime, and has success at most $2^{-k}=O(1/\log \log N)$, but deliberately weakens the useful guarantee. The standard unsuppressed algorithm above gives the stronger intended performance. Small $N$ can be handled directly. Primary evidence for the totient lower bound is Theorem 15, equations (3.41)–(3.42), in denisevellachemla.eu/Rosser-Schoenfeld-1962.pdf .

The divided difference is a finite linear combination of spline knot values, so it commutes with the finite integral. For every spline knot $t_j$ in the defining interval,It follows thatbecause the order-$k$ divided difference of a monic degree-$k$ polynomial is its leading coefficient. This proves the unit-integral normalization of a B-spline.

For the partition of unity, let $N_{i,r}$ denote the partition-normalized order-$r$ B-spline. The Cox-de Boor recurrence isIt follows by substituting the explicit divided difference formula into both sides and collecting each truncated power function. The starting case is $N_{i,1}=1_{[t_i,t_{i+1})}$, with endpoint conventions irrelevant when $r\ge 2$. The two weights are nonnegative wherever their respective lower-order B-splines are nonzero. Induction therefore gives $N_{i,r}\ge 0$ and strict positivity in its open support.

Extend the finite spline knot sequence strictly in both directions, with no finite accumulation. At each point only finitely many B-splines are nonzero. For the full extended family the sum is one for order one. Summing the Cox-de Boor recurrence, the combined coefficient of $N_{j,r-1}$ isInduction gives a full partition of unity at every order. On $t_k<t<t_{n+1}$, all extended order-$k$ functions with index $i\le 0$ or $i\ge n+1$ vanish by their supports. HenceThe same argument gives the useful global subpartition of unity for B-splines, $0\le {\sum }_{i=1}^n{N}_i(t)\le 1$, including outside the basic spline knot interval.