Chapter map

1. 1. Matrix
2. 2. Singular values
3. 3. First norm
4. 4. Second norm
5. 5. Third norm
6. 6. Frames
7. 7. Trace pairing
8. 8. Variational check
9. 9. Rotation
10. 10. Partial sums
11. 11. Variance warning
12. 12. Triangle check
13. 13. Dominance
14. 14. Cutoff
15. 15. Application
16. 16. Conclusion

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

---

1. Matrix

This is where modelling choices begin to affect evidence. Set A=diag(5,3,1).

A concrete situation

A is square and diagonal.

What the mathematics does not license

Record A. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Write matrix. 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. Singular values

The central mathematical idea is concrete. Take absolute diagonal entries.

A concrete situation

They are 5,3,1.

What the mathematics does not license

Order descending. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record list. 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. First norm

Here the geometry becomes operational. Sum the largest one.

A concrete situation

Ky Fan 1-norm is 5.

What the mathematics does not license

This equals operator norm. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record 5. 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. Second norm

This step deserves a careful pause. Sum the largest two.

A concrete situation

Ky Fan 2-norm is 8.

What the mathematics does not license

Use 5+3. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record 8. 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. Third norm

The useful question is not only how to calculate, but what the calculation assumes. Sum all three.

A concrete situation

Ky Fan 3-norm is 9.

What the mathematics does not license

This equals nuclear norm. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record 9. 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. Frames

A small example makes the mechanism visible. Choose U=V=[e1,e2].

A concrete situation

Columns are orthonormal.

What the mathematics does not license

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

Try it, check it, transfer it

Record frames. 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. Trace pairing

This is where modelling choices begin to affect evidence. Compute tr(U* A V).

A concrete situation

It equals 8.

What the mathematics does not license

Keep conjugate transpose. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record value. 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. Variational check

The central mathematical idea is concrete. Compare with Ky Fan 2-norm.

A concrete situation

The chosen frames attain 8.

What the mathematics does not license

This is an optimum example. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record equality. 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. Rotation

Here the geometry becomes operational. Multiply A by unitary matrices.

A concrete situation

Singular values do not change.

What the mathematics does not license

Do not use entrywise logic. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record invariance. 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. Partial sums

This step deserves a careful pause. List 5,8,9.

A concrete situation

They are cumulative leading magnitudes.

What the mathematics does not license

Do not square them. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record curve. 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. Variance warning

The useful question is not only how to calculate, but what the calculation assumes. Compute squares separately if needed.

A concrete situation

Explained variance uses another summary.

What the mathematics does not license

Do not call 8/9 variance. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record distinction. 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. Triangle check

A small example makes the mechanism visible. Choose a second diagonal matrix.

A concrete situation

Verify one Ky Fan triangle inequality.

What the mathematics does not license

This is an example. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record comparison. 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. Dominance

This is where modelling choices begin to affect evidence. Compare all partial singular sums.

A concrete situation

This can imply norm inequalities.

What the mathematics does not license

State theorem used. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record relation. 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. Cutoff

The central mathematical idea is concrete. Give an external reason for k.

A concrete situation

The norm does not select k.

What the mathematics does not license

Keep choice open. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record reason. 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.

---

15. Application

Here the geometry becomes operational. Connect to approximation or optimization.

A concrete situation

State matrix estimation error.

What the mathematics does not license

Do not promise model quality. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record evidence. 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.

---

16. Conclusion

This step deserves a careful pause. State singular values, k and norm.

A concrete situation

The checked Ky Fan 2-norm is 8.

What the mathematics does not license

Keep interpretation precise. That distinction matters because a correct formula can still answer the wrong operational question.

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

The Ky Fan k-norm of a matrix is the sum of its k largest singular values. It interpolates between the operator norm at k=1 and the nuclear norm when k reaches the full number of singular values. The definition focuses attention on the leading k directions without discarding the rest from the mathematics.

For A=diag(5,3,1), the singular values are 5,3,1. The Ky Fan 1-norm is 5, the Ky Fan 2-norm is 8, and the Ky Fan 3-norm is 9. Each value answers how much leading singular magnitude is accumulated through that cutoff.

Ky Fan’s variational principle expresses the sum of the leading singular values as a maximum of a trace pairing over orthonormal k-frame matrices. In the diagonal example, selecting the first two coordinate directions gives trace 5+3=8, which reaches the Ky Fan 2-norm.

The maximum is basis independent. Rotating the original matrix with unitary matrices changes coordinates but not singular values or any Ky Fan norm. This makes the norms suitable for matrix approximation, optimization and spectral summaries.

The Ky Fan norms are unitarily invariant and satisfy the triangle inequality. They provide a family of lenses: k=1 cares only about the largest stretch, intermediate k values measure a leading block, and the full nuclear norm aggregates every singular value linearly.

These norms do not measure explained variance in the same way as squared singular values. In principal component analysis, variance accounting typically uses squares. A student should not call 8/9 an explained-variance fraction for the example; that would mix two different summaries.

The variational view supports best-subspace reasoning, matrix inequalities and convex optimization. It can help compare models or approximations, but it does not choose k automatically or guarantee that leading singular directions are scientifically meaningful.

Ky Fan dominance connects inequalities for partial sums of singular values with inequalities for all unitarily invariant norms. The idea is powerful because a family of leading-sum comparisons can control many norm statements at once.

Did You Know? The nuclear norm is the Ky Fan norm at the largest possible k. It is widely used as a convex surrogate for rank, but minimizing it is not the same as directly minimizing rank. The surrogate changes the optimization problem and should be named honestly.

A practical audit lists the singular values, every partial sum, the selected k and the reason for that cutoff. If the matrix is estimated, attach uncertainty to the singular values before drawing a stable ranking conclusion.

For students, Ky Fan norms illustrate how one sequence can answer several questions. The first value records the worst stretch, intermediate sums record leading capacity, and the final sum records total singular magnitude. Choosing the lens is part of the reasoning, not an afterthought.

A second example clarifies why partial sums are stronger than a single largest value. Matrices with singular values (5,1,1) and (4,4,0) have operator norms 5 and 4, so the first looks larger at k=1. At k=2 their Ky Fan values are 6 and 8, so the second carries more combined magnitude across two leading directions. Neither ordering alone is universally “better”; the chosen cutoff determines the question. Reporting the whole partial-sum curve avoids hiding this crossover.

Ky Fan dominance formalises such comparisons. If every leading singular-value partial sum of A is no larger than the corresponding partial sum of B, then every unitarily invariant norm of A is no larger than the corresponding norm of B. This is a powerful transfer principle, but it requires all relevant cutoffs, not one convenient k. Students should calculate the complete sequence, preserve multiplicities and state whether equality holds at the full sum before invoking a dominance conclusion.

Because singular values are nonnegative, the partial sums must be nondecreasing in k. Their increments reproduce the ordered singular values. A simple quality check therefore differences consecutive Ky Fan norms and compares the results with the SVD. If the recovered sequence is not nonincreasing, the ordering or arithmetic is wrong. This small audit connects a family of norms back to the spectrum it summarises.

---

Frequently asked questions

What is the Ky Fan k-norm?

The sum of the k largest singular values.

What are the endpoints?

k=1 gives the operator norm; full k gives the nuclear norm.

What does the variational principle say?

A maximum trace pairing over orthonormal k-frames reaches the leading singular-value sum.

Is it explained variance?

No. Variance accounting typically uses squared singular values.

Does the norm choose k?

No. The cutoff needs an external purpose.

What must be reported?

Singular values, k, partial sum, variational constraints and intended interpretation.

---

Primary source and fact-check note

Ky Fan, “Maximum Properties and Inequalities for the Eigenvalues of Completely Continuous Operators,” Proceedings of the National Academy of Sciences 37, 1951. 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

• Eckart–Young low-rank source
• Singular-value perturbation bounds
• Hermitian dilation
• 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.

Spectral audit: assumptions, norms and optimization logic

Order audit

Sort singular values descending.

Cutoff audit

State k and dimensions.

Variational audit

Use orthonormal k-frames.

Interpretation audit

Do not substitute squared-energy language.

Mechanism workshop: calculate, challenge and transfer

Manual reconstruction

Build the smallest matrix example for leading singular-value sums. Label spectra, dimensions, norms and assumptions.

Assumption map

Place normality, rank, orthogonality, separation and norm conditions beside the step they permit.

Counterexample

Remove one hypothesis and identify the first invalid conclusion.

Sensitivity

Change one singular value, pairing, gap or cutoff at a time.

Independent verification

Recompute spectra, norms and residuals through a second route.

Simulation

Generate bounded perturbations while keeping theorem and observation separate.

Engineering transfer

Map the mechanism to compression, signal analysis, numerical solvers or spectral data.

AI transfer

Explain a learning use without promising intelligence, accuracy or employment.

Parent explanation

Describe the reasoning habit in plain language without guaranteed results.

Limit statement

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

Evidence notebook: reconstruct every spectral step

Evidence notebook 1: Matrix

Step 1 isolates matrix. Set A=diag(5,3,1). The checked result is: A is square and diagonal. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: Record A. Write matrix. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 2: Singular values

Step 2 isolates singular values. Take absolute diagonal entries. The checked result is: They are 5,3,1. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: Order descending. Record list. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 3: First norm

Step 3 isolates first norm. Sum the largest one. The checked result is: Ky Fan 1-norm is 5. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: This equals operator norm. Record 5. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 4: Second norm

Step 4 isolates second norm. Sum the largest two. The checked result is: Ky Fan 2-norm is 8. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: Use 5+3. Record 8. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 5: Third norm

Step 5 isolates third norm. Sum all three. The checked result is: Ky Fan 3-norm is 9. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: This equals nuclear norm. Record 9. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 6: Frames

Step 6 isolates frames. Choose U=V=[e1,e2]. The checked result is: Columns are orthonormal. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: Check dimensions. Record frames. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 7: Trace pairing

Step 7 isolates trace pairing. Compute tr(U* A V). The checked result is: It equals 8. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: Keep conjugate transpose. Record value. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 8: Variational check

Step 8 isolates variational check. Compare with Ky Fan 2-norm. The checked result is: The chosen frames attain 8. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: This is an optimum example. Record equality. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 9: Rotation

Step 9 isolates rotation. Multiply A by unitary matrices. The checked result is: Singular values do not change. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: Do not use entrywise logic. Record invariance. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 10: Partial sums

Step 10 isolates partial sums. List 5,8,9. The checked result is: They are cumulative leading magnitudes. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: Do not square them. Record curve. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 11: Variance warning

Step 11 isolates variance warning. Compute squares separately if needed. The checked result is: Explained variance uses another summary. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: Do not call 8/9 variance. Record distinction. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 12: Triangle check

Step 12 isolates triangle check. Choose a second diagonal matrix. The checked result is: Verify one Ky Fan triangle inequality. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: This is an example. Record comparison. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 13: Dominance

Step 13 isolates dominance. Compare all partial singular sums. The checked result is: This can imply norm inequalities. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: State theorem used. Record relation. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 14: Cutoff

Step 14 isolates cutoff. Give an external reason for k. The checked result is: The norm does not select k. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: Keep choice open. Record reason. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 15: Application

Step 15 isolates application. Connect to approximation or optimization. The checked result is: State matrix estimation error. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: Do not promise model quality. Record evidence. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Evidence notebook 16: Conclusion

Step 16 isolates conclusion. State singular values, k and norm. The checked result is: The checked Ky Fan 2-norm is 8. Preserve dimensions, ordering, norm, permutation, rank or gap beside every number. The safeguard is: Keep interpretation precise. Write conclusion. Change one input while holding the others fixed, recalculate the full spectrum or singular spectrum, and explain whether the change comes from matching, discarded components, separation, basis geometry or rounding. Ask a second reader to reproduce the calculation. Finish with the narrow conclusion and the evidence still missing.

Failure cases and responsible recovery

Squared-value error

Ky Fan norms sum singular values, not squares. Build a small counterexample, choose an eligible theorem and state the missing evidence before calculating again.

Wrong endpoint

The full sum is nuclear norm, not Frobenius norm. Build a small counterexample, choose an eligible theorem and state the missing evidence before calculating again.

Nonorthogonal frames

Variational constraints require orthonormal columns. Build a small counterexample, choose an eligible theorem and state the missing evidence before calculating again.

Automatic cutoff

The norm does not choose k. Build a small counterexample, choose an eligible theorem and state the missing evidence before calculating again.

Variance language

Partial singular sums are not explained variance. Build a small counterexample, choose an eligible theorem and state the missing evidence before calculating again.

Rank surrogate confusion

Nuclear norm is not rank itself. Build a small counterexample, choose an eligible theorem and state the missing evidence before calculating again.

Uncertainty omission

Estimated singular values may be unstable. Build a small counterexample, choose an eligible theorem and state the missing evidence before calculating again.

Model guarantee

A norm comparison does not prove usefulness. Build a small counterexample, choose an eligible theorem and state the missing evidence before calculating again.