Images in sets of module coordinates
Let \(R\) be a commutative ring, \(N\) an \(R\)-module, and \(f_x:N\to X_x\) a family of linear maps indexed by a set \(V\). For any subset \(U\subseteq V\), define \[ E_U:N\to\prod_{x\in U}X_x,\qquad E_U(n)=(f_x(n))_{x\in U}. \] Its coordinate image is \[ L(U)=\operatorname{im}E_U. \] Thus \(z\in L(U)\) means that a single \(n\in N\) satisfies \(f_x(n)=z_x\) for every \(x\in U\). The source calls these maps and modules eval and image.
Restricting the codomain to the image gives the surjection \[ p_U:N\twoheadrightarrow L(U),\qquad p_U(n)=E_U(n), \] called project. Its kernel is \[ \ker p_U=\{n:f_x(n)=0\text{ for every }x\in U\}. \]
Restriction to fewer coordinates
For \(U\subseteq K\), define \[ r_{K,U}:L(K)\to L(U),\qquad z\mapsto(z_x)_{x\in U}. \] This is well defined: a common preimage \(n\) for all coordinates in \(K\) is also a preimage for the selected coordinates in \(U\). It is linear and satisfies \[ r_{K,U}p_K=p_U. \] It is surjective. Given \(z=p_U(n)\), use \(p_K(n)\) as its preimage. Its kernel consists precisely of the image vectors on \(K\) whose coordinates at every \(x\in U\) vanish.
These statements allow arbitrary index sets and subsets; the products consist of coordinate assignments.
Recovering an image from two restrictions
If \(K=U\cup U'\), the pair of restrictions \[ L(K)\longrightarrow L(U)\oplus L(U') \] is injective. Equal restrictions mean equal coordinates on \(U\) and on \(U'\). Every coordinate of \(K\) occurs in one of these two sets, so the two original assignments agree.
This is the source's restrict_joint_injective lemma. It will identify kernels through their remaining coordinates.
Word-module coordinate images
For a finite-generator Coxeter system, take \(N=M_{\mathbf t}\), \(R=A\) its symmetric coefficient algebra, and \(f_x=\pi_x:M_{\mathbf t}\to M_{\mathbf t}^x\). Then \[ L_{\mathbf t}(U) =\{(\pi_x(m))_{x\in U}:m\in M_{\mathbf t}\}. \] The source names this module bottSetImage and its projection bottSetProjection. The restriction maps are bottSetRestrict.
Projecting the finite word basis gives a finite generating family for every image, even when \(U\) is infinite. Indeed, from \(m=\sum_i a_i b_i\), linearity gives \[ p_U(m)=\sum_i a_i p_U(b_i). \] Thus all image vectors are linear combinations of the finitely many projected basis vectors.
For the one-letter word \(s\) over \(A=\mathbb R[\alpha]\), \[ L_s(\{1,s\})=\{(f,g)\in A^2:f-g\in\alpha A\},\qquad L_s(\{s\})=A. \] Restriction is \((f,g)\mapsto g\), and the common-preimage description explains its surjectivity.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.CoordinateProjection.eval: IntegralPairings.lean:321.
OAI.KLInvariance.CoordinateProjection.image: IntegralPairings.lean:322.
OAI.KLInvariance.CoordinateProjection.project: IntegralPairings.lean:323.
OAI.KLInvariance.CoordinateProjection.project_surjective: IntegralPairings.lean:325.
OAI.KLInvariance.CoordinateProjection.project_apply: IntegralPairings.lean:330.
OAI.KLInvariance.CoordinateProjection.project_zero_iff: IntegralPairings.lean:332.
OAI.KLInvariance.CoordinateProjection.restrict: IntegralPairings.lean:341.
OAI.KLInvariance.CoordinateProjection.restrict_project: IntegralPairings.lean:350.
OAI.KLInvariance.CoordinateProjection.restrict_surjective: IntegralPairings.lean:353.
OAI.KLInvariance.CoordinateProjection.restrict_zero_iff: IntegralPairings.lean:358.
OAI.KLInvariance.CoordinateProjection.restrict_joint_injective: IntegralPairings.lean:1062.
OAI.KLInvariance.TitsSpace.bottSetImage: IntegralPairings.lean:453.
OAI.KLInvariance.TitsSpace.bottSetProjection: IntegralPairings.lean:459.
OAI.KLInvariance.TitsSpace.bottSetRestrict: IntegralPairings.lean:898.