Find your next learning step
When eigenvalues form a cluster, the physically stable object can be the whole subspace. Projectors and principal angles let recovery be measured without pretending that arbitrary basis vectors are unique.
Open the chapter contents · Polymer Science root · Gold Standard of Education · How X Works
Chapter contents
Choose the route closest to the question, or read from Chapter 1 for the complete argument. The chapters remain open on the page; these grouped routes are an index.
Route 1 · Define the cluster
Route 2 · Work the projector ledger
Route 3 · Test recovery and gaps
Route 4 · Check limits and evidence
Route 5 · Ask and continue
CHAPTER 1 OF 12 · DEFINE THE CLUSTER
1. Treat a clustered eigenspace as the estimand
The preceding recovery problem begins when tensor eigenvalues become nearly equal and individual eigenvectors become sensitive. The next move is constructive: define a cluster of eigenvalues separated from the rest of the spectrum, then estimate the subspace attached to that cluster.
For a symmetric tensor C with orthonormal eigenvectors u_i, the projector onto cluster I is P_I = Σ_{i in I} u_i u_i^T. Rotating or swapping the basis inside the cluster leaves P_I unchanged. The projector represents the plane or higher-dimensional subspace without pretending that one basis is physically preferred.
CHAPTER 2 OF 12 · DEFINE THE CLUSTER
2. Declare the cluster before comparing times
A cluster needs a rule: a fixed eigenvalue interval, a relative-gap threshold, or a model-based grouping that is applied consistently. Choosing the grouping after seeing a desired direction can turn uncertainty into confirmation bias.
Track two separations. Internal gaps describe how distinguishable vectors are inside the cluster. The external gap from the cluster to the remaining spectrum controls stability of the cluster subspace. A tight internal pair can still define a stable plane when that plane is well separated from the leading direction.
CHAPTER 3 OF 12 · WORK THE PROJECTOR LEDGER
3. Work a projector ledger
Let C_A = diag(4, 2.01, 2.00). Cluster the last two eigenvalues. Their basis vectors span the yz-plane and their projector is P_A = diag(0,1,1). Rotate the two in-plane eigenvectors by 45 degrees to form C_B while preserving the eigenvalues. The individual vectors move, but P_B = P_A.
Now tilt the plane by 6 degrees toward the x-direction to form a third estimate. Its two principal angles from the original plane are 0 and 6 degrees. The projector distance is no longer zero and describes actual subspace movement rather than a basis rotation within the plane.
| Comparison | Largest basis-vector change | Largest principal angle | Projector reading |
|---|---|---|---|
| A to in-plane rotation B | 45° | 0° | same clustered subspace |
| A to tilted plane C | basis dependent | 6° | subspace has moved |
CHAPTER 4 OF 12 · WORK THE PROJECTOR LEDGER
4. Measure subspace angles and projector distance
If U and V contain orthonormal bases for two r-dimensional subspaces, the singular values of U^T V are cosines of the principal angles. These angles are invariant to a change of basis within either subspace. The largest principal angle reports the weakest alignment direction.
The spectral norm of P_U − P_V equals the sine of the largest principal angle. The Frobenius norm combines all angle contributions and therefore grows when several dimensions move. State which norm is used; identical words such as “subspace error” can hide different summaries.
| Measure | Emphasis | Useful question |
|---|---|---|
| largest principal angle | worst aligned direction | Did any part of the cluster rotate strongly? |
| projector spectral norm | same worst-case geometry | How large is the maximum subspace deviation? |
| projector Frobenius norm | combined movement | Did several directions change together? |
CHAPTER 5 OF 12 · WORK THE PROJECTOR LEDGER
5. Use the external spectral gap in the error ledger
Davis–Kahan theory bounds angles between invariant subspaces in terms of perturbation size and spectral separation. The lesson is not to quote one universal number. It is to record the tensor perturbation norm, the exact cluster, and the gap separating that cluster from unwanted eigenvalues.
In C_A, the pair near 2 is separated from λ1 = 4 by about 1.99. A perturbation norm of 0.02 is small relative to that external gap even though it is larger than the 0.01 internal gap. The plane can therefore be well determined while its two chosen axes are not.
CHAPTER 6 OF 12 · TEST RECOVERY AND GAPS
6. Changed context: the cluster splits
Suppose recovery changes the eigenvalues from 4, 2.01, 2.00 to 4, 2.8, 1.2. The former two-dimensional cluster no longer represents nearly tied directions. It may be appropriate to track three one-dimensional projectors, provided their gaps exceed perturbation uncertainty.
The time at which a cluster splits should be governed by a declared rule with uncertainty, not a visually attractive crossing. Near the threshold, report both the two-dimensional projector and the tentative individual directions so readers can see what is stable.
CHAPTER 7 OF 12 · TEST RECOVERY AND GAPS
7. Build a recovery curve without label swapping
Across time, compare projectors for the same declared cluster and use principal angles rather than greedily matching eigenvectors. If cluster dimension changes, mark the event; do not silently compare a plane with one line as if the metric were unchanged.
Combine subspace orientation with eigenvalue and invariant recovery. A plane can return to its original orientation while total stretch remains different. Conversely, trace and determinant can recover while a well-separated leading direction remains rotated.
CHAPTER 8 OF 12 · TEST RECOVERY AND GAPS
8. Connect the subspace to rheological meaning
Conformation and orientation tensors are reduced descriptions of molecular or microstructural distributions. A clustered transverse plane may express weak distinction between directions in that plane, while a separated leading eigenspace supports a robust directional comparison for restart loading.
Continue through Eigenvalue Degeneracy, Axis Uncertainty and Frame-Invariant Recovery Measures and Tensorial Entanglement Recovery, Eigenframes and Direction-Dependent Overshoot before assigning a microscopic mechanism to a projector curve.
CHAPTER 9 OF 12 · CHECK LIMITS AND EVIDENCE
9. Limits and common misconceptions
A stable projector does not prove that every molecule retains a common plane. Different distributions can share the same second-order tensor. A spectral gap supports mathematical stability under the chosen norm; it does not remove measurement bias, closure error or spatial averaging.
Projectors for different dimensions cannot be compared without stating the metric and interpretation. Temporal smoothing can make a subspace look stable. Positive-definiteness repair can alter eigenvalues and gaps. Degeneracy is not noise; sometimes it is a genuine symmetry of the state.
CHAPTER 10 OF 12 · CHECK LIMITS AND EVIDENCE
10. Build an evidence check
Archive the tensor before eigendecomposition, units, symmetrisation, positive-definiteness checks, spatial and temporal windows, eigenvalues, uncertainty estimate, cluster rule, basis, projector, principal angles and norms. Preserve raw observations and code versions.
Perturb each tensor within its error model and recompute clusters. Report the probability of each cluster dimension, projector-distance intervals and sensitivity to the gap threshold. Test whether projector-based recovery improves prediction of directional restart stress on held-out protocols.
CHAPTER 11 OF 12 · ASK AND CONTINUE
11. Questions readers should ask
Does a projector choose axes inside its subspace?
No. It identifies the subspace and deliberately remains unchanged when the internal orthonormal basis rotates.
Can a plane be stable when both axes inside it are unstable?
Yes. Stability depends on the external separation of the clustered plane from the rest of the spectrum, not the small internal gap.
Is a small principal angle proof of full material recovery?
No. It is one geometric component of a tensor-level description and must be paired with eigenvalues, invariants and independent mechanical evidence.
CHAPTER 12 OF 12 · ASK AND CONTINUE
12. Continue through verified routes
Begin with the Polymer Science root, Gold Standard of Education and How X Works. Continue through Orthogonal Double-Step Strain, Tensor Memory and Cross Components, Waiting-Time Spectra, Restart Superposition and Directional Recovery and Shear Cessation, Elastic Recoil and Structural Recovery.
Primary routes include Davis and Kahan’s rotation of invariant subspaces under perturbation, Hameduddin and colleagues’ geometric decomposition of the conformation tensor, and Advani and Tucker’s tensor descriptions of orientation. The projector ledger is an original teaching example.
Top · Contents · Previous · How X Works