TheoremBase

Proof sketch: the vectors u1u_1, ..., umu_m are exchanged one at a time into the spanning family, each time removing one w by the linear dependence lemma, so that a spanning family of length n is kept and m never exceeds n.

Proof

Routine facts about nonempty segments and equality of tuples are used without comment (Toolkit for AL2). The argument follows Axler (2.22): the vectors u1,…,umu_{1},\dots,u_{m} are brought into the spanning family one at a time, each time pushing out one of the ww's.

For k∈{0,…,m}k\in\{0,\dots,m\} let P(k)P(k) be the statement: k≀nk\leq n, and there is a strictly increasing map Οƒ:[nβˆ’k]β†’[n]\sigma:[n-k]\to[n] (that is, Οƒ(l)<Οƒ(lβ€²)\sigma(l)<\sigma(l') whenever l<lβ€²l<l') such that the nn-tuple

Bk=(u1,…,uk,wΟƒ(1),…,wΟƒ(nβˆ’k)),B_{k}=(u_{1},\dots,u_{k},w_{\sigma(1)},\dots,w_{\sigma(n-k)}),

whose ii-th component is uiu_{i} for i≀ki\leq k and wΟƒ(iβˆ’k)w_{\sigma(i-k)} for k<i≀nk<i\leq n, spans VV (Span of a Finite Family; Spanning Families (iii)). We prove, by induction on k∈N0k\in\mathbb{N}_{0} (Induction on β„•β‚€ from 0 for a Condition Expressed by a Predicative Formula (ii)), that k≀mk\leq m implies P(k)P(k). Since P(m)P(m) includes m≀nm\leq n, this proves (i).

For k=0k=0 we have 0≀n0\leq n, and we take Οƒ=id[n]\sigma=\mathrm{id}_{[n]}, so that B0=(w1,…,wn)B_{0}=(w_{1},\dots,w_{n}), which spans VV by hypothesis.

For the induction step, let k∈[m]k\in[m] and assume P(kβˆ’1)P(k-1), with Οƒ:[nβˆ’k+1]β†’[n]\sigma:[n-k+1]\to[n]. Adjoining uku_{k} gives the (n+1)(n+1)-tuple

C=(u1,…,uk,wΟƒ(1),…,wΟƒ(nβˆ’k+1))∈Vn+1.C=(u_{1},\dots,u_{k},w_{\sigma(1)},\dots,w_{\sigma(n-k+1)})\in V^{n+1}.

Removing its kk-th component gives C(k)=Bkβˆ’1C^{(k)}=B_{k-1}, in the notation of Dropping a Redundant Vector Does Not Change the Span. Every component of Bkβˆ’1B_{k-1} is a component of CC and so lies in span⁑(C)\operatorname{span}(C); hence The Span of a Finite Family is the Smallest Subspace Containing It (ii) and The Span of a Finite Family is the Smallest Subspace Containing It (iii) give V=span⁑(Bkβˆ’1)βŠ†span⁑(C)V=\operatorname{span}(B_{k-1})\subseteq\operatorname{span}(C), and CC spans VV.

The family CC is linearly dependent: its component Ck=ukC_{k}=u_{k} lies in V=span⁑(Bkβˆ’1)=span⁑(C(k))V=\operatorname{span}(B_{k-1})=\operatorname{span}(C^{(k)}), so Elementary Properties of Linear Independence (v) applies.

We next show that the vector to be removed is one of the ww's. By Elementary Properties of Linear Independence (iv) there is j∈[n+1]j\in[n+1] with Cj∈span⁑(C1,…,Cjβˆ’1)C_{j}\in\operatorname{span}(C_{1},\dots,C_{j-1}). Suppose that j≀kj\leq k. Then uj∈span⁑(u1,…,ujβˆ’1)u_{j}\in\operatorname{span}(u_{1},\dots,u_{j-1}), and (u1,…,ujβˆ’1)(u_{1},\dots,u_{j-1}) is the family of the other vectors of (u1,…,uj)(u_{1},\dots,u_{j}), so (u1,…,uj)(u_{1},\dots,u_{j}) is linearly dependent by Elementary Properties of Linear Independence (v); but it is linearly independent by Elementary Properties of Linear Independence (i), applied to the inclusion [j]β†’[m][j]\to[m]. Hence k<j≀n+1k<j\leq n+1; in particular k≀nk\leq n, and Cj=wΟƒ(jβˆ’k)C_{j}=w_{\sigma(j-k)}.

We remove this vector. Since Ci=Ci(j)C_{i}=C^{(j)}_{i} for i<ji<j, The Span of a Finite Family is the Smallest Subspace Containing It (ii) and The Span of a Finite Family is the Smallest Subspace Containing It (iii) give

Cj∈span⁑(C1,…,Cjβˆ’1)βŠ†span⁑(C(j)),C_{j}\in\operatorname{span}(C_{1},\dots,C_{j-1})\subseteq\operatorname{span}(C^{(j)}),

so by Dropping a Redundant Vector Does Not Change the Span (i) and the fact that CC spans VV, span⁑(C(j))=span⁑(C)=V\operatorname{span}(C^{(j)})=\operatorname{span}(C)=V.

Finally, C(j)=(u1,…,uk,wΟƒβ€²(1),…,wΟƒβ€²(nβˆ’k))C^{(j)}=(u_{1},\dots,u_{k},w_{\sigma'(1)},\dots,w_{\sigma'(n-k)}) with Οƒβ€²=Οƒβˆ˜gjβˆ’k:[nβˆ’k]β†’[n]\sigma'=\sigma\circ g_{j-k}:[n-k]\to[n], where gjβˆ’k:[nβˆ’k]β†’[nβˆ’k+1]g_{j-k}:[n-k]\to[n-k+1] is the gap map of Bijections between Intervals: the Gap Map and the Reversal (i); and Οƒβ€²\sigma' is strictly increasing as a composite of strictly increasing maps. Taking Bk=C(j)B_{k}=C^{(j)}, together with k≀nk\leq n shown above, this is P(k)P(k), which completes the induction.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…