Chapter map

1. 1. Matrix
2. 2. Condition
3. 3. Pattern
4. 4. First pivot
5. 5. First edge
6. 6. Second pivot
7. 7. Next edge
8. 8. Third pivot
9. 9. Fourth edge
10. 10. Last pivot
11. 11. Reconstruct
12. 12. Preconditioned eigenvalues
13. 13. New condition
14. 14. Conclusion

Visit the How Mathematics Works hub for the wider series and its connections to education, reasoning, science, computing, design and careers.

---

1. Matrix

Here the geometry becomes operational. Use the four-cycle SPD matrix.

A concrete situation

Eigenvalues are 2,4,4,6.

What the mathematics does not license

Record ordering. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Write A. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

2. Condition

This step deserves a careful pause. Take 6/2.

A concrete situation

kappa2(A)=3.

What the mathematics does not license

Use spectral norm. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record kappa. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

3. Pattern

The useful question is not only how to calculate, but what the calculation assumes. Keep original lower nonzeros.

A concrete situation

This defines IC(0).

What the mathematics does not license

No threshold rule. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record pattern. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

4. First pivot

A small example makes the mechanism visible. Take sqrt4.

A concrete situation

L11=2.

What the mathematics does not license

Positive pivot. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record entry. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

5. First edge

This is where modelling choices begin to affect evidence. Divide -1 by 2.

A concrete situation

L21=L41=-0.5.

What the mathematics does not license

Respect pattern. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record entries. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

6. Second pivot

The central mathematical idea is concrete. Subtract 0.25 from 4.

A concrete situation

L22=sqrt15/2.

What the mathematics does not license

Check positive. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record pivot. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

7. Next edge

Here the geometry becomes operational. Divide adjusted -1.

A concrete situation

L32=-1/sqrt15.

What the mathematics does not license

Use zero dropped entry. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record entry. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

8. Third pivot

This step deserves a careful pause. Subtract L32 squared.

A concrete situation

L33=sqrt(56/15).

What the mathematics does not license

Check arithmetic. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record pivot. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

9. Fourth edge

The useful question is not only how to calculate, but what the calculation assumes. Compute retained L43.

A concrete situation

It is about -0.51755.

What the mathematics does not license

Dropped fill changes numerator. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record entry. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

10. Last pivot

A small example makes the mechanism visible. Subtract retained squares.

A concrete situation

L44 is about 1.86605.

What the mathematics does not license

Check positive. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record pivot. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

11. Reconstruct

This is where modelling choices begin to affect evidence. Form M=LL transpose.

A concrete situation

M24=M42=0.25 while A24=0.

What the mathematics does not license

This exposes approximation. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record M. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

12. Preconditioned eigenvalues

The central mathematical idea is concrete. Solve generalized eigenproblem.

A concrete situation

Values are about .92308,1,1,1.06667.

What the mathematics does not license

Use stable symmetric computation. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record spectrum. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

13. New condition

Here the geometry becomes operational. Divide extremes.

A concrete situation

Ratio is about 1.15556.

What the mathematics does not license

Example-specific. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record ratio. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

14. Conclusion

This step deserves a careful pause. State ordering,pattern,pivots and costs.

A concrete situation

Condition ratio improves from 3 to about 1.15556.

What the mathematics does not license

No universal improvement promise. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Write conclusion. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

A seven-day practice lab

1. Day 1: define the decision, observation unit, X variables and Y or reference condition.
2. Day 2: reproduce the smallest worked example and check every dimension, unit and sign.
3. Day 3: build a transparent baseline and write what it assumes.
4. Day 4: fit the main method inside a training-only pipeline; record every parameter.
5. Day 5: challenge one assumption with a deliberately difficult dataset or operating regime.
6. Day 6: compare validation metrics with the practical workload created by errors.
7. Day 7: present one supported conclusion, one unresolved uncertainty and one safe next experiment.

Questions for a student, parent or mentor

• Which part of the model is a mathematical definition, and which part is a choice?
• What small calculation would catch a sign, scale or dimension error?
• What evidence comes from training, validation and the untouched test?
• What benign condition could look like a fault or prediction failure?
• How would the answer change for another response, asset, school project or time period?
• What should a human review before acting on the score?

A method-specific audit before deployment

Incomplete Cholesky preconditioning approximates a symmetric positive definite matrix A by a sparse product L L transpose. It follows Cholesky-style elimination but drops selected fill entries so storage and triangular solves remain manageable.

Consider the four-cycle matrix with diagonal 4 and -1 on edges (1,2),(2,3),(3,4),(4,1). Its eigenvalues are 2,4,4,6, so it is symmetric positive definite and its 2-norm condition number is 3.

IC(0) keeps only the original lower-triangular nonzero pattern. The computed diagonal entries are 2, sqrt(15)/2, sqrt(56/15), and approximately 1.86605; retained off-diagonals begin -0.5, -1/sqrt15, -0.5 and approximately -0.51755.

Multiplying the incomplete factor by its transpose reproduces every original entry except that it creates a 0.25 coupling between variables 2 and 4. That position represents the dropped-fill effect of closing the cycle.

For this example, the eigenvalues of M^{-1}A are approximately 0.92308,1,1 and 1.06667. Their ratio is about 1.15556, substantially smaller than the original condition number 3.

The condition-number improvement is example-specific. Dropping can produce a weak preconditioner, and incomplete factorization may encounter a nonpositive pivot even when the original matrix is positive definite under an unsuitable ordering or variant.

Did You Know? A zero in the original sparse matrix does not mean elimination will keep it zero. Fill-in records indirect coupling created as variables are eliminated, so dropping fill is a deliberate approximation.

The label IC(0) refers to a level-of-fill rule, not a guarantee of zero error. Threshold dropping, modified incomplete Cholesky, diagonal shifts and reordered variants create different factors and different numerical behaviour.

A preconditioner is useful when the time saved in fewer Krylov iterations exceeds the cost of constructing and applying it. Setup time, memory and repeated right-hand sides belong in the comparison.

Ordering can change fill, pivots and parallelism. Two mathematically equivalent permutations of a system may lead to different incomplete factors, so the ordering algorithm is part of a reproducible result.

Triangular solves can be harder to parallelise than sparse matrix-vector products. A preconditioner that reduces iterations may still perform poorly on hardware with limited sequential throughput.

Students can run the IC(0) recurrence entry by entry, reconstruct M=L L transpose, and compare A-M. Then they can compute eigenvalues before and after preconditioning to connect local dropping with global convergence geometry.

Incomplete Cholesky does not change a poor physical model into a good one. It changes the numerical route to the solution, while discretisation error, measurement error and model validity remain separate questions.

A small unpreconditioned residual can still correspond to a sensitive solution if A is ill-conditioned. Preconditioning targets numerical work; accuracy must be judged against the original system and scientific units.

For parents and students, the transferable lesson is that a useful approximation keeps the structural connections that matter most while making repeated checking faster. The dropped pieces must still be measured.

A responsible report states matrix properties, ordering, fill or drop rule, shifts, pivot safeguards, factor nonzeros, setup cost, application cost, preconditioned spectrum or iteration evidence, and final residual for the original system.

---

Frequently asked questions

What is incomplete Cholesky?

A sparse approximate factorization A approximately equals L L transpose.

What does IC(0) keep?

The original sparsity pattern under a level-zero rule.

Why drop fill?

To limit storage and triangular-solve cost.

Can incomplete factorization fail?

Yes. Ordering, dropping and pivots can cause breakdown or poor quality.

Does a lower condition number guarantee faster wall time?

No. Setup, triangular solves, memory and hardware also matter.

What must be reported?

Matrix properties, ordering, drop rule, shifts, pivots, nonzeros, costs, spectrum or iterations, and original residual.

---

Primary source and fact-check note

J. A. Meijerink and H. A. van der Vorst, “An Iterative Solution Method for Linear Systems of Which the Coefficient Matrix is a Symmetric M-Matrix,” Mathematics of Computation 31, 1977. Definitions and core claims in this article were checked against this primary publication. Calculations shown here were independently recomputed. Examples are educational illustrations; they are not reported industrial performance results.

Useful next reading

• Conjugate gradient method
• Randomized Kaczmarz
• Frequent Directions
• How Mathematics Works
• Study and learning methods

Keep the options open. Mathematics education supports clearer questions, transferable problem-solving skills and informed choices across science, engineering, computing, AI, statistics, design and many other careers. It does not guarantee intelligence, income, admission or employment; those outcomes depend on far more than one subject or model.

Method audit: assumptions, geometry and numerical evidence

Pattern audit

Record ordering and retained entries.

Pivot audit

Check every square root and safeguard.

Cost audit

Compare setup and apply work with saved iterations.

Accuracy audit

Check the original residual, not a transformed surrogate.

Mechanism workshop: calculate, challenge and transfer

Manual reconstruction

Build the smallest example for sparse approximate factors. Label every vector, matrix, probability, unit and assumption.

Assumption map

Place convexity, unbiasedness, positive definiteness or sparsity conditions beside the step they permit.

Counterexample

Remove one condition and locate the first conclusion that fails.

Sensitivity

Change a coordinate constant, learning rate, eigenvalue or drop rule one at a time.

Independent verification

Recompute gradients, residuals, factors and spectra by a second route.

Simulation

Repeat random samples or coordinate choices and separate observed frequency from proof.

Engineering transfer

Count memory traffic, sparse operations, setup work and stopping checks.

AI transfer

Explain an optimization use without promising generalisation, fairness or safety.

Parent explanation

Describe local updates and structural approximations in plain language.

Limit statement

Write one supported claim, one overclaim and one repaired conclusion.

Evidence notebook: reconstruct every step

Evidence notebook 1: Matrix

Step 1 isolates matrix. Use the four-cycle SPD matrix. The checked result is: Eigenvalues are 2,4,4,6. Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: Record ordering. Write A. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 2: Condition

Step 2 isolates condition. Take 6/2. The checked result is: kappa2(A)=3. Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: Use spectral norm. Record kappa. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 3: Pattern

Step 3 isolates pattern. Keep original lower nonzeros. The checked result is: This defines IC(0). Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: No threshold rule. Record pattern. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 4: First pivot

Step 4 isolates first pivot. Take sqrt4. The checked result is: L11=2. Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: Positive pivot. Record entry. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 5: First edge

Step 5 isolates first edge. Divide -1 by 2. The checked result is: L21=L41=-0.5. Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: Respect pattern. Record entries. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 6: Second pivot

Step 6 isolates second pivot. Subtract 0.25 from 4. The checked result is: L22=sqrt15/2. Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: Check positive. Record pivot. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 7: Next edge

Step 7 isolates next edge. Divide adjusted -1. The checked result is: L32=-1/sqrt15. Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: Use zero dropped entry. Record entry. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 8: Third pivot

Step 8 isolates third pivot. Subtract L32 squared. The checked result is: L33=sqrt(56/15). Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: Check arithmetic. Record pivot. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 9: Fourth edge

Step 9 isolates fourth edge. Compute retained L43. The checked result is: It is about -0.51755. Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: Dropped fill changes numerator. Record entry. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 10: Last pivot

Step 10 isolates last pivot. Subtract retained squares. The checked result is: L44 is about 1.86605. Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: Check positive. Record pivot. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 11: Reconstruct

Step 11 isolates reconstruct. Form M=LL transpose. The checked result is: M24=M42=0.25 while A24=0. Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: This exposes approximation. Record M. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 12: Preconditioned eigenvalues

Step 12 isolates preconditioned eigenvalues. Solve generalized eigenproblem. The checked result is: Values are about .92308,1,1,1.06667. Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: Use stable symmetric computation. Record spectrum. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 13: New condition

Step 13 isolates new condition. Divide extremes. The checked result is: Ratio is about 1.15556. Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: Example-specific. Record ratio. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 14: Conclusion

Step 14 isolates conclusion. State ordering,pattern,pivots and costs. The checked result is: Condition ratio improves from 3 to about 1.15556. Preserve objective definitions, dimensions, probabilities, scaling, sparsity, matrix properties and units beside every number. The safeguard is: No universal improvement promise. Write conclusion. Change one input while holding the others fixed, recalculate the update, residual, objective gap, factor reconstruction or spectrum, and explain whether the difference comes from sampling, curvature, conditioning, dropping, rounding or a changed target. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Failure cases and responsible recovery

Exact-factor claim

Dropped fill makes an approximation. Build a small counterexample, restore the missing condition and state the evidence before calculating again.

Pattern ambiguity

IC(0), threshold and modified rules differ. Build a small counterexample, restore the missing condition and state the evidence before calculating again.

Pivot optimism

Breakdown is possible. Build a small counterexample, restore the missing condition and state the evidence before calculating again.

Ordering silence

Permutation changes fill and quality. Build a small counterexample, restore the missing condition and state the evidence before calculating again.

Iteration-only win

Setup and application cost matter. Build a small counterexample, restore the missing condition and state the evidence before calculating again.

Transformed residual

Check Ax-b. Build a small counterexample, restore the missing condition and state the evidence before calculating again.

Condition guarantee

Improvement is problem-dependent. Build a small counterexample, restore the missing condition and state the evidence before calculating again.

Model leap

Preconditioning does not fix scientific error. Build a small counterexample, restore the missing condition and state the evidence before calculating again.

Transfer checks: from a worked example to trustworthy use

Definition check

Explain incomplete factors without relying on software names. Identify the mathematical object, the update and the quantity the method is trying to improve. Then compare the explanation with the displayed formula and repair any word that claims more than the formula proves.

Units check

Attach units to inputs, parameters, gradients, residuals and tolerances. Rescale one variable and repeat the calculation. State which numbers change because the problem changed and which change only because the numerical representation changed.

Baseline check

Compare the method with a simple full update, direct solve or unpreconditioned iteration on the same small example. Use the same starting point, stopping rule and accuracy measure so the comparison answers a real question.

Cost check

Count matrix-vector products, partial derivatives, sampled examples, triangular solves, memory reads and stored nonzeros. Report setup separately from repeated application and distinguish arithmetic counts from elapsed time.

Precision check

Repeat the worked arithmetic after rounding intermediate values. Measure the effect on objective, residual, conjugacy, factor reconstruction or spectrum, and explain why exact-arithmetic theorems need numerical safeguards on computers.

Stopping check

Write the actual condition that ends the algorithm. Test it on a deliberately misleading state where one local quantity looks small but the global objective, residual or validation evidence remains poor.

Reproducibility check

Record data order, coordinate or sample seed, matrix ordering, initialization, parameter schedule, software precision and all transformations. Another reader should be able to reconstruct both the intended mathematics and the realised path.

Communication check

Write a student explanation, a parent explanation and a technical report sentence. Keep the mathematical conclusion identical while changing vocabulary. Each version should name the main limit and avoid guaranteed performance claims.

Boundary check

Construct one input at the edge of the assumptions: a nearly zero coordinate constant, a highly variable sampled gradient, a nearly singular positive definite matrix or a marginal incomplete pivot. Do not run blindly. Predict which calculation becomes sensitive, choose a safeguard and record how the conclusion narrows.

Decision check

Turn the analysis into a practical choice. State when the method is suitable, when a simpler baseline is sufficient, what evidence would make you change method, and which result must be reviewed by a teacher, engineer or domain expert before real-world use.