TheoremBase

Sketch: each clause unfolds the definition of the simplex and computes with the recursion, splitting and extraction rules for finite sums; membership follows by induction on the number of points from the splitting clause and the definition of convexity.

Proof

Throughout, sums of points are formed in the commutative ring Rn\mathbb{R}^{n} as in Convex Sets in Rn\mathbb{R}^n: Standing Notation (iii), a scalar multiple cxcx being the product with the constant function cc (Rules for Pointwise Operations on Real-Valued Maps: a Commutative Ring, Scalar Multiples and Evaluation (iv)), so the rules for finite sums in a commutative ring apply to sums of points as well as to sums of reals; membership in ΔN\Delta_{N} is unfolded by Convex Weights, the Standard Simplex, and Convex Combinations in Rn\mathbb{R}^n (i).

Proof of (i). For t∈R1t\in\mathbb{R}^{1} the recursion Iterated Operations: Recursion, Splitting, Reordering, Termwise Combination and Homomorphisms (i) gives ∑k=11tk=t1\sum_{k=1}^{1}t_{k}=t_{1}, so t∈Δ1t\in\Delta_{1} if and only if t1≥0t_{1}\ge0 and t1=1t_{1}=1, that is, if and only if t1=1t_{1}=1. For such tt, again by Iterated Operations: Recursion, Splitting, Reordering, Termwise Combination and Homomorphisms (i),

∑k=11tkxk=t1x1=1x1=x1.\sum_{k=1}^{1}t_{k}x_{k}=t_{1}x_{1}=1x_{1}=x_{1}.

Proof of (ii). For t∈R2t\in\mathbb{R}^{2}, Iterated Operations: Recursion, Splitting, Reordering, Termwise Combination and Homomorphisms (i) gives

∑k=12tk=t1+t2,∑k=12tkxk=t1x1+t2x2.\sum_{k=1}^{2}t_{k}=t_{1}+t_{2},\qquad\sum_{k=1}^{2}t_{k}x_{k}=t_{1}x_{1}+t_{2}x_{2}.

Hence t∈Δ2t\in\Delta_{2} if and only if t1,t2≥0t_{1},t_{2}\ge0 and t1+t2=1t_{1}+t_{2}=1, which holds if and only if t=(1−s,s)t=(1-s,s) with s=t2∈[0,1]s=t_{2}\in[0,1]; substituting into the second sum gives the displayed formula.

Proof of (iii). By Iterated Operations: Recursion, Splitting, Reordering, Termwise Combination and Homomorphisms (i),

1=∑k=1N+1tk=∑k=1Ntk+tN+1,1=\sum_{k=1}^{N+1}t_{k}=\sum_{k=1}^{N}t_{k}+t_{N+1},

so ∑k=1Ntk=1−tN+1\sum_{k=1}^{N}t_{k}=1-t_{N+1}. A finite sum of nonnegative reals is nonnegative and bounds each of its terms (Finite Sums in a Commutative Ring and in an Ordered Field: Distributivity, Differences, Telescoping, Constant Terms, Comparison and the Triangle Inequality (vi)); hence 1−tN+1≥01-t_{N+1}\ge0, and together with tN+1≥0t_{N+1}\ge0 this gives 0≤tN+1≤10\le t_{N+1}\le1.

If tN+1=1t_{N+1}=1, then 0≤tk≤∑k′=1Ntk′=00\le t_{k}\le\sum_{k'=1}^{N}t_{k'}=0 for every k∈[N]k\in[N], so every term tkxkt_{k}x_{k} with k≤Nk\le N is 00 and ∑k=1Ntkxk=0\sum_{k=1}^{N}t_{k}x_{k}=0 (Finite Sums in a Commutative Ring and in an Ordered Field: Distributivity, Differences, Telescoping, Constant Terms, Comparison and the Triangle Inequality (iv)). By Iterated Operations: Recursion, Splitting, Reordering, Termwise Combination and Homomorphisms (i) we then have z=0+1xN+1=xN+1z=0+1x_{N+1}=x_{N+1}.

If tN+1<1t_{N+1}<1, then s=1−tN+1>0s=1-t_{N+1}>0, so the weights τk=tk/s\tau_{k}=t_{k}/s are nonnegative, and by Finite Sums in a Commutative Ring and in an Ordered Field: Distributivity, Differences, Telescoping, Constant Terms, Comparison and the Triangle Inequality (i)

∑k=1Nτk=s−1∑k=1Ntk=1;\sum_{k=1}^{N}\tau_{k}=s^{-1}\sum_{k=1}^{N}t_{k}=1;

thus τ∈ΔN\tau\in\Delta_{N}. Writing tk=sτkt_{k}=s\tau_{k} for k≤Nk\le N, Iterated Operations: Recursion, Splitting, Reordering, Termwise Combination and Homomorphisms (i) and Finite Sums in a Commutative Ring and in an Ordered Field: Distributivity, Differences, Telescoping, Constant Terms, Comparison and the Triangle Inequality (i), now applied in Rn\mathbb{R}^{n}, give

z=∑k=1Nsτkxk+tN+1xN+1=s y+tN+1xN+1,z=\sum_{k=1}^{N}s\tau_{k}x_{k}+t_{N+1}x_{N+1}=s\,y+t_{N+1}x_{N+1},

which is the display, since s=1−tN+1s=1-t_{N+1}.

Proof of (iv). Apply Iterated Operations over an Interval: Extracting One Term and Reversing the Order (i), with the extracted index jj and the bijection gjg_{j}, twice: once to the reals tkt_{k}, and once to the points tkxkt_{k}x_{k} (the addition of Rn\mathbb{R}^{n} is associative and commutative, with neutral element 00). This gives

∑k=1N+1tk=∑k=1Ntgj(k)+tj,∑k=1N+1tkxk=∑k=1Ntgj(k)xgj(k)+tjxj.\sum_{k=1}^{N+1}t_{k}=\sum_{k=1}^{N}t_{g_{j}(k)}+t_{j},\qquad\sum_{k=1}^{N+1}t_{k}x_{k}=\sum_{k=1}^{N}t_{g_{j}(k)}x_{g_{j}(k)}+t_{j}x_{j}.

Since tj=0t_{j}=0 and 0xj=00x_{j}=0, the first sum over [N][N] equals 11 and the second equals the original combination. The weights tgj(k)t_{g_{j}(k)} are nonnegative, so they form a point of ΔN\Delta_{N}.

Proof of (v). The points wkw_{k} and the weights tkt_{k} are well defined, since by Concatenation Identifies a Product of Euclidean Spaces with a Euclidean Space (i) every k∈[N+P]k\in[N+P] either lies in [N][N] or has the form N+jN+j for exactly one j∈[P]j\in[P]. By the same splitting of indices, every tkt_{k} is nonnegative, being (1−λ)τk(1-\lambda)\tau_{k} or λσj\lambda\sigma_{j} with nonnegative factors. Splitting the sums over [N+P][N+P] at NN by Iterated Operations: Recursion, Splitting, Reordering, Termwise Combination and Homomorphisms (iii), for the associative addition of R\mathbb{R} and of Rn\mathbb{R}^{n} respectively, gives

∑k=1N+Ptk=∑k=1N(1−λ)τk+∑j=1Pλσj,∑k=1N+Ptkwk=∑k=1N(1−λ)τkxk+∑j=1Pλσjxj′.\sum_{k=1}^{N+P}t_{k}=\sum_{k=1}^{N}(1-\lambda)\tau_{k}+\sum_{j=1}^{P}\lambda\sigma_{j},\qquad\sum_{k=1}^{N+P}t_{k}w_{k}=\sum_{k=1}^{N}(1-\lambda)\tau_{k}x_{k}+\sum_{j=1}^{P}\lambda\sigma_{j}x'_{j}.

By Finite Sums in a Commutative Ring and in an Ordered Field: Distributivity, Differences, Telescoping, Constant Terms, Comparison and the Triangle Inequality (i), the first is (1−λ)⋅1+λ⋅1=1(1-\lambda)\cdot1+\lambda\cdot1=1, so t∈ΔN+Pt\in\Delta_{N+P}, and the second is the left side of the display.

Proof of (vi). We argue by induction on N∈NN\in\mathbb{N}, for the property that ∑k=1Ntkxk∈C\sum_{k=1}^{N}t_{k}x_{k}\in C for all x1,…,xN∈Cx_{1},\dots,x_{N}\in C and all t∈ΔNt\in\Delta_{N}. For N=1N=1 the combination is x1∈Cx_{1}\in C, by (i). Suppose the property holds for NN, and let z=∑k=1N+1tkxkz=\sum_{k=1}^{N+1}t_{k}x_{k} with x1,…,xN+1∈Cx_{1},\dots,x_{N+1}\in C and t∈ΔN+1t\in\Delta_{N+1}. By (iii), either z=xN+1∈Cz=x_{N+1}\in C, or

z=(1−λ)y+λxN+1,λ=tN+1∈[0,1),z=(1-\lambda)y+\lambda x_{N+1},\qquad\lambda=t_{N+1}\in[0,1),

where yy is a convex combination of x1,…,xNx_{1},\dots,x_{N} and therefore lies in CC by the induction hypothesis. In the second case zz lies in the segment [y,xN+1][y,x_{N+1}] (Segments and Convex Subsets of Rn\mathbb{R}^n (i)), and this segment is contained in CC because CC is convex (Segments and Convex Subsets of Rn\mathbb{R}^n (ii)).

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…