12.13.4 Tensor Projection Component Effect

The Tensor Projection Component Effect describes how tensor components project onto specific directions, revealing structural insights in multi-dimensional algebra.

Tensor Projection Component Effect is the change induced in a specific coordinate, index slot, or factor of a tensor when a projection operation is applied, isolated from the overall result structure by tracking exactly how each individual component transforms rather than treating the projected tensor as an undifferentiated whole. Where the result structure describes the shared shape of every possible output, the component effect describes, component by component, which entries of the input are kept unchanged, which are set to zero, and which are recombined with others as a direct consequence of the specific projection formula being applied.

Component-Level Description of a Projection

Effect on a Single Coordinate

π {(t)}^{i} = \sum_{j} P_{j}^{i} t^{j}

Relative to a fixed basis, the i-th component of the projected tensor is a specific linear combination of the components tʲ of the input, weighted by the entries Pⱼⁱ of the projection's matrix. The component effect is exactly this rule, applied index by index, showing precisely how each output coordinate depends on the input coordinates.

Coordinates That Pass Through Unchanged

For coordinates corresponding to basis vectors already lying in im(π), the projection's matrix entries satisfy Pᵢⁱ = 1 and Pⱼⁱ = 0 for j ≠ i, so the component effect on such a coordinate is the identity: it passes through the projection completely unaltered. Identifying which coordinates behave this way is equivalent to identifying a basis adapted to the image of the projection.

Coordinates That Are Annihilated

For coordinates corresponding to basis vectors lying entirely in ker(π), every relevant matrix entry Pⱼⁱ vanishes, so the component effect sets that coordinate's contribution to zero in the output. These are exactly the coordinates the projection is designed to discard, and their systematic annihilation is the component-level manifestation of the projection's overall image-kernel decomposition.

Component Effect Under Common Projection Types

Factor-Extraction Projections

π (v_{1} \otimes v_{2}) = φ (v_{2}) \cdot v_{1}

When projecting V₁ ⊗ V₂ onto V₁ using a fixed functional φ on V₂, the component effect on a simple tensor's coordinates is to multiply every coordinate of the V₁ factor by the single scalar φ(v₂), while every coordinate associated with the V₂ factor is absorbed entirely into that scalar and does not appear independently in the output. This shows the component effect can mix a whole block of input coordinates into a single scalar multiplier rather than acting coordinate-by-coordinate in a diagonal fashion.

Symmetrization Component Effect

π {(T)}^{i j} = \frac{1}{2} (T^{i j} + T^{j i})

For the projection onto the symmetric part of a rank-2 tensor, the component effect on entry (i,j) averages it with the entry (j,i). Diagonal entries Tⁱⁱ are left unchanged by this averaging, since Tⁱⁱ averaged with itself is itself, while off-diagonal entries are each replaced by the average of a symmetric pair, illustrating that the component effect can depend jointly on more than one coordinate of the input at once.

Index Contraction Component Effect

For a projection realized as contraction of index k against a fixed covector φ, the component effect removes the k-th index entirely: π(T)^{i₁...î_k...i_n} = Σⱼ φⱼ T^{i₁...j...i_n}. Every remaining component of the output is a sum over the contracted index's range, so the component effect here reduces the number of free indices while summing over the eliminated one.

Diagram of Component-Level Effects

Consequences of the Component Effect

Sparsity of the Projection Matrix

Many tensor projections have a sparse component effect: most output coordinates depend on only a small number of input coordinates, often exactly one. This sparsity is what allows a computed projection to be evaluated efficiently on large tensors, since the full dense matrix P need not be applied when its component effect is known to be diagonal or block-diagonal in an appropriate basis.

Sensitivity of the Effect to Basis Choice

The component effect is described relative to a specific basis, and changing basis changes which coordinates are unchanged, annihilated, or mixed, even though the underlying abstract projection π is unaffected. Two bases related by a change-of-basis transformation that does not respect the image-kernel decomposition of π will generally report very different, seemingly unrelated component effects for the same projection.

Reconstructing the Global Projection from Component Effects

Once the component effect is known for every basis vector, the full projection matrix P is determined, and hence the projection π itself as an abstract linear map is completely specified. The component effect is therefore not merely a description of behavior but a complete, equivalent encoding of the projection operation relative to the chosen basis.