Linear Algebra for Decisions - Full Manual

Learner Promise: Learn undergraduate linear algebra in a way you can actually use: read vectors and matrices with meaning, solve systems, and reach projection, eigen ideas, and least squares without getting lost in notation.

Course Orientation

Why It Matters: Linear algebra sits underneath data, optimization, economics, machine learning, and later math. Learn the objects now and later formulas stop feeling arbitrary.

What You Get: You will read vectors and matrices with meaning, solve systems, and work with span, basis, rank, projection, determinants, eigenvectors, and least squares.

How To Use: Look at the visual first. Say the idea in plain words. Open the formal note when you need precision.

Prerequisites:

• basic algebra manipulation
• reading coordinates and simple graphs
• comfort with short word problems
• willingness to reason from story to symbols and back again

Mastery Artifact

Output: A set of vector and matrix interpretations, linear-system notes, subspace and rank analysis, orthogonality and projection work, eigen and singular-direction summaries, and one final synthesis of the course.

Course Spine

1. Vectors and Coordinates - A vector is useful when each coordinate keeps a fixed interpretation, direction describes the pattern of change, and magnitude describes size.
2. Matrices as Linear Maps - A matrix is best read as a linear map: rows and columns carry structural meaning, and the output is another vector.
3. Checkpoint: Vectors and Matrices Together - This checkpoint combines vector and matrix reading: identify the input, the map, and the output, then state the relationship correctly.
4. Balance Equations and Feasibility - Linear equations express constraints. A candidate vector is feasible only when it satisfies all of them.
5. Span, Basis, and Reach - Span describes all reachable linear combinations, basis removes redundancy, and rank measures independent directions.
6. Repeated Matrix Updates and Stability - Repeated matrix application can reveal convergence, steady patterns, or growth depending on the map.
7. Factors, Compression, and Part I Synthesis - Part I ends by identifying dominant patterns, comparing repeated behavior, and preparing for the formal second half.
8. Linear Systems, Elimination, and Invertibility - Day 8 formalizes the first half of the course: elimination simplifies a system without changing the answers, inverses recover inputs when the map is reversible, and solvability depends on the geometry of the constraints.
9. Subspaces, Column Space, and Null Space - Day 9 turns reachability into formal spaces: subspaces are closed under linear combination, column space names what outputs are achievable, and null space names the input freedom the map cannot detect.
10. Basis Coordinates, Rank, and Dimension - Day 10 ties basis and rank together formally: the same vector can wear different coordinates in different bases, rank counts the independent directions the map keeps alive, and nullity counts what disappears from view.
11. Dot Product, Orthogonality, and Projection - Day 11 adds the geometry of inner products: the dot product measures alignment, orthogonality marks genuinely separate directions, and projection keeps the closest signal the model can express.
12. Determinant, Orientation, and Volume - Day 12 turns determinant into a geometric signal: its magnitude describes how volume scales, its sign tracks orientation, and its nonzero status tells whether the map stays invertible.
13. Eigenvectors, Eigenvalues, and Diagonalization - Day 13 gives the formal version of the dynamics arc: eigenvectors identify steady directions, eigenvalues explain growth and decay on those directions, and diagonalization rewrites the map in a basis where the modes stop interfering with each other.
14. Least Squares, Singular Directions, and Final Synthesis - Day 14 closes the course by connecting least squares, residuals, and singular directions into one final synthesis.

Chapter Map

1. Coordinate meaning and vector geometry: read vectors as states, compare direction and magnitude, and keep coordinate meaning attached to every calculation.
2. Matrices and systems: read matrices as linear maps, solve balance constraints, and decide when a system is feasible or invertible.
3. Subspaces, bases, rank, and dimension: describe reachable outputs, hidden input directions, independent coordinates, rank, nullity, and basis changes.
4. Orthogonality, projection, and determinants: use dot products, projections, residuals, determinant sign, determinant scale, and nonzero determinant tests.
5. Eigen, least squares, and synthesis: connect steady directions, diagonalization, best-fit models, singular directions, and final decision summaries.

Detailed Knowledge Manual

The day sections below form the detailed study manual. Each day gives the core idea, the formal point, examples, traps, practice checks, and repair notes needed for final-exam-level fluency.

Day 01: Vectors and Coordinates

Day 1: Vectors as State and Meaning

The learner starts with a three-coordinate vector. Each coordinate in the vector represents a meaningful quantity, not just a slot in an array.

Today the learner should answer one question clearly:

How do we represent a change as a vector so that each coordinate keeps a clear mathematical meaning?

Use the main explanation first. If a plain-language detour or coordinate refresher is needed, pause for the supporting note and then return to the main idea.

Story Anchor: Day 1 makes vectors readable. Later linear algebra only helps once each coordinate feels like a real quantity instead of a mysterious slot.

Focus Question: How do coordinates compress a full state into one readable mathematical object?

Daily Synthesis: A vector is useful when each coordinate keeps a fixed interpretation, direction describes the pattern of change, and magnitude describes size.

Sessions

Interpret a vector as a real system state instead of a bare list of numbers.

Hook: A three-coordinate vector changes, and the learner must read it entry by entry.

Mastery Signals:

• can map each coordinate to a system meaning
• can compare two state vectors in words

Study Guide

Vectors return everywhere later

Vectors are the first major object in linear algebra, and they return later in systems, optimization, data, machine learning, and advanced mathematics. If the slots do not feel meaningful now, later procedures stay mechanical.

A vector is one named snapshot

Treat the vector as one compact state description. The first slot is \(x_1\) , the second is \(x_2\) , and the third is \(x_3\) , so the order itself carries meaning.

Order is part of the definition

If \(\mathbf{v}=(a,b,c)\) , then \(x_1=a\) , \(x_2=b\) , and \(x_3=c\) . Swapping the same numbers into a different order usually produces a different vector because the coordinate labels stay fixed.

Read one vector in order

For \(\mathbf{v}=(3,1,-2)\) , read \(x_1=3\) , \(x_2=1\) , and \(x_3=-2\) in that order. If you swap it to \((1,3,-2)\) , the numbers are the same but the vector now says something different because the coordinate labels did not move.

Do not detach a value from its coordinate

Many beginner mistakes come from reading the numbers but ignoring the slot they occupy.

• The second entry is meaningful because it is second, not just because of its value.
• Two vectors with the same numbers in a different order usually mean different states.
• You should be able to say each coordinate in words before doing any computation.

Interpret vector direction and magnitude as pattern and size.

Hook: One example vector points partly upward and another changes more strongly.

Mastery Signals:

• can describe a change vector in words
• can compare larger versus smaller vectors without losing interpretation

Study Guide

Direction is the pattern of change

Direction tells which coordinates rise, which fall, and in what relative proportions. Magnitude tells how large the whole change is once you view all coordinates together.

Scalar multiples preserve direction

If \(\mathbf{u}=(1,2,-1)\) , then \(2\mathbf{u}=(2,4,-2)\) points in the same direction but has larger magnitude. Multiplying by a positive scalar keeps the pattern while changing the size.

Compare pattern and size separately

Let \(\mathbf{u}=(1,-1,2)\) and \(3\mathbf{u}=(3,-3,6)\) . The signs and coordinate proportions stay the same, so the direction is unchanged, but the second vector has three times the magnitude. This is the basic pattern-versus-size distinction.

Do not judge a vector by one coordinate only

A larger first entry does not automatically mean a larger vector or a different direction.

• Direction depends on the whole coordinate pattern, not one entry.
• Magnitude compares overall size, not just the biggest coordinate.
• Ask separately: what is the pattern, and how large is it?

Choose the interpretation that best matches a new vector.

Hook: A new comparison vector appears, and the learner must decide which interpretation reads it correctly.

Mastery Signals:

• can choose the interpretation that preserves coordinate meaning
• can compare one vector against an earlier vector

Study Guide

Use a fixed reading order

When a new vector appears, read it in the same sequence every time: name each coordinate, note the sign pattern, then compare size or direction with an earlier vector.

Comparison means coordinatewise comparison

To compare vectors, line up the same coordinates. If v and w describe the same labeled system, then comparing \(v_1\) with \(w_1\) , \(v_2\) with \(w_2\) , and \(v_3\) with \(w_3\) preserves meaning.

Use one fixed reading pass

If \(\mathbf{w}=(2,-1,0)\) , first read the coordinates in order, then note the sign pattern, then compare with an earlier vector. That fixed reading order prevents you from mixing labels, missing direction, or comparing the wrong coordinates.

Reject answers that blur the labels

A wrong interpretation often sounds plausible because it talks about the numbers without respecting the coordinate system.

• If an option swaps coordinate roles, discard it immediately.
• If an option ignores sign or direction, it is incomplete.
• The right interpretation should read the vector as one ordered object.

Repair confusion about which coordinate belongs to which label and quantity.

Hook: Two coordinates get swapped, and the learner has to spot why the interpretation breaks.

Mastery Signals:

• can match coordinate labels reliably
• can explain why swapping coordinates changes the interpretation

Study Guide

A coordinate has meaning because its position is fixed

The first entry, second entry, and third entry are not interchangeable. Each slot refers to one labeled quantity, so swapping entries changes the vector’s meaning immediately.

Vectors use ordered coordinates

In an ordered coordinate system, \(\mathbf{x}=(x_{1},x_{2},x_{3})\) assigns one value to each labeled coordinate. Reordering the same numbers usually creates a different vector because the labels stay attached to their positions.

Do not treat the entries as an unlabeled pile of numbers

The values matter only together with their coordinate labels.

• Position carries meaning.
• Swapping entries changes interpretation.
• Read the label and the value together.

Retell the vector in words before compressing it back into coordinate notation.

Hook: The learner describes the vector without notation first, then returns to symbols.

Mastery Signals:

• can narrate each coordinate without symbolic shorthand
• can restate the vector as one labeled object

Study Guide

You should be able to say what the vector means before writing symbols

Vector notation is useful because it compresses a labeled state into one object. If you can explain the coordinates in words, the notation becomes easier to trust.

Coordinate notation summarizes a labeled state

A vector such as \(\mathbf{x}=(x_{1},x_{2},x_{3})\) records several quantities at once. The symbols are compact, but each coordinate still refers to one specific measured component.

Do not let the notation hide the labels

If you cannot restate the entries in plain language, you are reading the vector too mechanically.

• Say each coordinate in words.
• Then compress back to symbols.
• Keep the labels attached throughout.

Day 02: Matrices as Linear Maps

Day 2: Matrices as System Maps

Today the learner stops treating matrices as static grids and starts reading them as reusable maps.

The through-line is direct:

An input vector enters the matrix map, the transformation acts on it, and the result becomes a new vector the learner can interpret.

Story Anchor: This day moves from single vectors to matrices as rules that send one input vector to an output vector.

Focus Question: How does a matrix act on an input vector to produce an output vector?

Daily Synthesis: A matrix is best read as a linear map: rows and columns carry structural meaning, and the output is another vector.

Sessions

See a matrix as a reusable linear map instead of a static table of numbers.

Hook: A matrix takes an input vector and redistributes its coordinates into a new output vector.

Mastery Signals:

• can explain a matrix as a map from one vector to another
• can read the output vector as a new vector

Study Guide

A matrix is a reusable machine

Think of a matrix as a rule that takes one input vector and produces one output vector. The entries of the matrix tell how the input coordinates are combined and redistributed.

Write the relation as \(\mathbf{y}=A\mathbf{x}\)

The vector x is the input, A is the linear map, and y is the output. The important idea is not the multiplication symbol by itself, but the fact that the same matrix A can act on many different inputs.

One rule, one input, one output

Suppose a matrix A sends \(\mathbf{x}=(1,2)\) to \(\mathbf{y}=(3,1)\) . The important point is not only that a multiplication happened, but that the same map A turned one input vector into a new output vector you can still interpret.

A matrix is not just a table to stare at

If you only see a rectangle of numbers, you miss the main point of linear algebra here: the matrix describes a transformation.

• Always ask what vector goes in and what vector comes out.
• The output is another vector with its own interpretation.
• A matrix becomes meaningful when you read it as an action.

Read rows and columns as different parts of the linear map.

Hook: The learner needs to explain what one row does and what one column collects in the matrix.

Mastery Signals:

• can explain one row as an output equation
• can explain one column as one input-coordinate contribution

Study Guide

Rows build outputs and columns track inputs

A row tells you how one output coordinate is assembled from the full input vector. A column tells you where one input coordinate sends its influence across all outputs.

Rows and columns answer different questions

The i-th row gives the coefficients for yi, so yi is the dot product of row i with x. The j-th column is Aej, which shows the output created by one unit placed entirely in input direction j.

Use one-hot inputs to read a column

If \(\mathbf{x}=\mathbf{e}_2\) , then \(A\mathbf{x}\) equals the second column of A. This is why columns tell you what each single input coordinate contributes to the total output.

Do not swap the jobs of rows and columns

A row answers how one output is formed; a column answers what one input direction contributes.

• Rows build outputs.
• Columns trace input contributions.
• Reading one as the other confuses the map.

Interpret the output vector as a new vector after the matrix acts.

Hook: A different example changes the matrix behavior, and the learner must explain the resulting output vector.

Mastery Signals:

• can explain why the output vector is still meaningful
• can compare two matrix-driven outputs qualitatively

Study Guide

The output is still a meaningful vector

After the matrix acts, you should not throw away interpretation. The output vector is a new state, and each of its coordinates still has to be read as a real quantity in the output space.

The map sends one vector space into another

If A maps inputs x to outputs y, then y belongs to the output space of the map. The coordinates of y may describe a different kind of state, but they are still coordinates that must be interpreted consistently.

Read the output as a new vector

If A sends \(\mathbf{x}=(1,0)\) to \(\mathbf{y}=(2,1)\) , the work is not finished at the multiplication sign. You still have to read y coordinate by coordinate as the new state created by the map.

Do not stop at the multiplication result

Computing \(\mathbf{y}=A\mathbf{x}\) is not the end of the reasoning. You still have to read what the entries of y actually say.

• Output coordinates may mean different things than input coordinates.
• Two different matrices can send similar inputs to very different outputs.
• An answer is incomplete if it gives products without interpretation.

Tell the input-output picture in plain language before returning to the matrix view.

Hook: The learner explains the map without matrix notation first.

Mastery Signals:

• can describe what the matrix map does before writing symbols

Study Guide

A matrix describes how one vector becomes another

Before you focus on entries, read the matrix as a rule from input vector to output vector. The numbers matter because they describe that transformation.

Rows and columns support the input-output view

Each output coordinate is formed from the input coordinates by the matrix entries. Columns show how one input direction contributes across outputs, while rows show how one output coordinate is assembled.

Do not read a matrix as a static table only

The matrix becomes meaningful when you connect it to the vectors it acts on.

• Name the input vector.
• Name the output vector.
• Use the entries to explain the map between them.

Day 03: Checkpoint: Vectors and Matrices Together

Day 3: Chapter Checkpoint

This day is the first synthesis checkpoint.

The learner should no longer treat vectors and matrices as separate topics. The task is to read a shock, a routing rule, and the resulting output together and then support one decision with that combined evidence.

Story Anchor: This checkpoint asks the learner to read an input vector, a matrix, and the resulting output vector together.

Focus Question: Can you combine vector and matrix reasoning in one coherent interpretation?

Daily Synthesis: This checkpoint combines vector and matrix reading: identify the input, the map, and the output, then state the relationship correctly.

Sessions

Connect the input vector to the matrix output in one explanation.

Hook: The learner needs one explanation that describes both the input vector and the mapped output.

Mastery Signals:

• can describe how the input enters the map and becomes an output vector
• can reject explanations that describe only the input or only the output

Study Guide

Linear algebra statements are relations, not fragments

You are no longer studying isolated objects. The main idea now is the relation between an input vector, a matrix rule, and the output vector that the rule creates.

Keep the full sentence \(\mathbf{y}=A\mathbf{x}\) in view

The clean explanation names x, then names A, then names y. If you describe only x or only y, you have dropped part of the mathematical relation.

Say the whole linear sentence

If \(\mathbf{x}=(1,0)\) enters the map and becomes \(\mathbf{y}=(2,1)\) , a complete explanation says what x represented, how A acted, and what y became. Leaving out any one of those three parts breaks the relation.

Do not confuse data description with map description

A common error is to describe the input carefully but say nothing about how the map changed it, or to describe the output without explaining what produced it.

• Input answers: what state went in?
• Matrix answers: what rule acted on it?
• Output answers: what new state came out?

Choose the best interpretation after reading how the matrix changes the output vector.

Hook: A new example changes the output, and the learner must choose the interpretation best supported by it.

Mastery Signals:

• can choose an interpretation supported by the output vector
• can reject interpretations that ignore what the matrix produced

Study Guide

Let the output constrain the story

When several interpretations are possible, the output vector should eliminate the ones that do not fit. Read what the map actually produced before you choose an explanation.

Interpretation follows from the coordinates of y

If \(\mathbf{y}=A\mathbf{x}\) , then the entries of y are evidence about the transformed state. An interpretation is mathematically weak if it ignores which output coordinates became larger, smaller, positive, or negative.

Let the output rule out bad stories

If the output vector is \(\mathbf{y}=(1,4,2)\) , then any explanation claiming the first coordinate dominates is already contradicted by the data. The second coordinate is largest, so the interpretation has to respond to that visible fact.

Before choosing an interpretation

Use a short checklist so the story stays tied to the algebra.

• Name the largest or most important output coordinates.
• Notice whether the pattern changed relative to earlier outputs.
• Reject any option that talks past the actual output vector.

Choose the final synthesis that integrates the input, the map, and the output.

Hook: The learner now has to pick one final explanation that keeps the whole linear relation intact.

Mastery Signals:

• can choose a synthesis that combines input, map, and output into one explanation
• can reject explanations that collapse the relation to one isolated number

Study Guide

A good synthesis has three parts

A strong linear algebra explanation says what the input vector represented, how the matrix acted on it, and what the resulting output vector became. Leaving out one part weakens the whole account.

Synthesis means preserving the full map

Mathematically, the point is to preserve the relation \(\mathbf{y}=A\mathbf{x}\) while translating it into words. A synthesis should connect all three symbols rather than collapsing the relation into one isolated number.

Build one three-part synthesis

A good synthesis can sound like this: the input vector x records the starting state, the matrix A transforms that state by one fixed rule, and the output vector y records the resulting state. That is the full relation preserved in words.

Do not replace the relation with one catchy sentence

An explanation can sound smooth and still be wrong if it drops the input or the transformation rule.

• Name the starting vector.
• Name the action of the matrix.
• Name the output vector that the action created.

Day 04: Balance Equations and Feasibility

Day 4

Balance comes before preference. A candidate vector that sounds attractive but breaks the system’s equations is not a real option.

Story Anchor: This day treats linear equations as balance conditions that a candidate solution must satisfy.

Focus Question: What makes a linear system balanced, consistent, or impossible?

Daily Synthesis: Linear equations express constraints. A candidate vector is feasible only when it satisfies all of them.

Sessions

Interpret equations as statements about what must balance in a linear system.

Hook: A candidate state vector is proposed, and the learner has to decide what the balance equation is really protecting.

Mastery Signals:

• can explain an equation as a system balance rule
• can connect symbolic equality to a concrete quantitative constraint

Study Guide

An equation states a balance that must hold

A linear equation is not just a line of symbols. It says some combination of quantities must match a required total, so any acceptable state has to respect that balance.

A solution satisfies the equation exactly

If an equation has the form \(a_1x_1+a_2x_2+a_3x_3=b\) , then a vector x is a solution only when substituting its coordinates makes the left-hand side equal b. Equality is the rule.

Test one candidate vector against the equation

For the equation \(x_1+x_2=5\) , the vector \((2,3)\) works because \(2+3=5\) . The vector \((4,3)\) fails because \(4+3=7\) . This is the simplest form of feasibility: substitute and check the balance.

Do not read the equation as a suggestion

A candidate vector may look reasonable and still fail the balance rule.

• Feasibility comes before preference.
• Changing one coordinate usually changes the balance.
• If the equality fails, the state is not allowed.

Separate feasible candidate vectors from vectors that merely sound desirable.

Hook: The learner now compares several candidate states and has to eliminate any vector that breaks the equations.

Mastery Signals:

• can reject infeasible candidate vectors even when they sound attractive
• can explain feasibility as a filter before preference

Study Guide

A good-sounding vector can still be impossible

Linear algebra forces you to distinguish what is desirable from what is feasible. The system can only realize states that satisfy its defining equations and relations.

Feasibility means satisfying all constraints together

A vector is feasible only if it satisfies every relevant equation or condition at the same time. One correct-looking coordinate does not rescue a vector that breaks another condition.

Filter before you compare

If the rule is \(x_1+x_2=5\) , then \((2,3)\) and \((4,1)\) stay in the feasible set, but \((4,2)\) is ruled out immediately. Only the first two belong in any further comparison, because the third breaks the equation.

How to test a candidate state

Use a short elimination process before you compare candidate vectors.

• Substitute the candidate into the defining relation.
• Check whether every coordinate still has a valid interpretation.
• Discard impossible options before discussing which one you prefer.

Choose the synthesis statement that uses balance as the first mathematical gate.

Hook: The learner has to pick the short synthesis statement that respects what the equations actually allow.

Mastery Signals:

• can explain why feasibility comes before ranking options
• can choose a synthesis statement that preserves the balance logic

Study Guide

A balance-based synthesis starts with what is possible

The right synthesis does not open with enthusiasm or preference. It first says which states the equations allow, and only then moves toward interpretation.

Equations are the gatekeepers of admissible states

In linear systems language, the feasible set is defined by the equations. Any summary that ignores those equations misstates the problem before it even begins to reason.

Write a balance-first synthesis

A correct summary sounds like this: vectors satisfying the equation are admissible, vectors that break it are removed, and only then do we compare the remaining options. That is the mathematical order a strong synthesis should preserve.

Do not turn balance into vague prose

A synthesis statement can sound polished while hiding the mathematical gate.

• The first question is whether the proposed state satisfies the system.
• Feasibility language should be explicit, not implied.
• Interpretation should come after the equations have been respected.

Day 05: Span, Basis, and Reach

Day 5

Reachability turns linear algebra into a control question: which operating states can this system build from the levers it really has?

Story Anchor: This day studies which vectors can be built from a given set and which generators are redundant.

Focus Question: Which vectors lie in the span, and which generators are actually needed?

Daily Synthesis: Span describes all reachable linear combinations, basis removes redundancy, and rank measures independent directions.

Sessions

Explain span as the set of vectors reachable from the available generators.

Hook: The learner has several generating vectors but cannot assume every imagined target is reachable.

Mastery Signals:

• can explain span as reachable combinations
• can reject the idea that one generator must do all the work

Study Guide

Span is the set you can build from the available directions

If you are given a few driver vectors, span means every vector you can produce by scaling them and adding the results. Reachability is a construction question, not an imagination question.

Span consists of all linear combinations

The span of vectors \(v_1\) , …, vk is the set of all vectors c1v1 + … + ckvk. A target lies in the span only if some choice of scalars produces it exactly.

Build a target from two generators

If \(\mathbf{v}_1=(1,0)\) and \(\mathbf{v}_2=(0,1)\) , then \(2\mathbf{v}_1-\mathbf{v}_2=(2,-1)\) . The target vector \((2,-1)\) is in the span because it can be built exactly from the available generators.

Do not confuse one example with the whole span

Seeing one reachable vector does not tell you the whole reachable set.

• Span is a set, not one favorite combination.
• You have to ask whether a target can be generated exactly.
• The available directions matter more than the state you wish existed.

Treat a basis as the smallest sufficient set of drivers rather than any convenient list.

Hook: The learner now strips away redundant levers and keeps only the drivers that still preserve reach.

Mastery Signals:

• can explain why redundancy weakens a candidate basis
• can keep minimality and sufficiency in the same explanation

Study Guide

A basis is the smallest language that still says everything

You do not want extra vectors if they add no new reach. A basis keeps exactly enough independent directions to describe every vector in the span without redundancy.

Basis means spanning plus linear independence

A set is a basis if it spans the space and none of its vectors can be written as a linear combination of the others. Both conditions matter: enough vectors, but no unnecessary ones.

Drop the redundant vector

The set {\((1,0)\) , \((0,1)\) , \((1,1)\) } spans the plane, but the third vector is already the sum of the first two. So the first two vectors are a basis and the longer list is redundant.

A convenient list is not automatically a basis

Beginners often keep redundant vectors because they are easy to describe.

• If one vector is built from the others, the set is not independent.
• Minimal does not mean too small to span the target set.
• A basis changes coordinates, not the underlying space.

Use rank to explain why some targets remain unreachable.

Hook: The learner now has to choose the synthesis statement that correctly treats rank as a limit on independent reach.

Mastery Signals:

• can relate rank to independent reach dimension
• can reject the idea that redundant vectors always expand reach

Study Guide

Rank measures how many independent output directions survive

Rank tells you how much genuine independent reach the system has. A matrix may list many vectors, but if several repeat the same direction, the real reach is smaller.

Rank is the dimension of the column space

Formally, rank counts the number of linearly independent columns, which is the same as the dimension of the space of outputs the matrix can produce.

Rank one blocks a two-direction target

If the columns are \((1,1)\) and \((2,2)\) , then the rank is 1 because the second column repeats the first direction. A target like \((1,-1)\) cannot be reached, because it needs an independent direction the matrix does not have.

Do not count columns without checking independence

More columns do not automatically mean more independent reach.

• Repeated directions do not increase rank.
• Rank is about independent structure, not visual size.
• Unreachable targets often fail because rank is too small.

Day 06: Repeated Matrix Updates and Stability

Day 6

Dynamics asks a longer question: if the same rule keeps acting, does the vector sequence settle into a stable pattern or keep amplifying?

Story Anchor: This day studies what happens when the same matrix is applied repeatedly.

Focus Question: What does repeated multiplication by the same matrix do to a vector?

Daily Synthesis: Repeated matrix application can reveal convergence, steady patterns, or growth depending on the map.

Sessions

Explain why repeated application matters more than a single output alone.

Hook: The learner sees an update rule applied multiple times and has to stop reading the first output as the full story.

Mastery Signals:

• can explain why one step is not enough to judge a repeated rule
• can read the update as one unfolding process

Study Guide

Repeated application creates a process

One matrix output is only one frame of the movie. When the same rule acts again and again, the real question becomes how the whole sequence behaves over time.

Iteration means powers of the matrix

If \(\mathbf{x}_1=A\mathbf{x}_0\) , then \(\mathbf{x}_2=A\mathbf{x}_1=A^2\mathbf{x}_0\) , and so on. Repeated updates are described by matrix powers rather than by one isolated multiplication.

A shrinking rule only reveals itself through repetition

Let \(A=\tfrac12I\) and start from \(\mathbf{x}_0=(4,2)\) . Then \(\mathbf{x}_1=(2,1)\) , \(\mathbf{x}_2=(1,1/2)\) , and \(\mathbf{x}_3=(1/2,1/4)\) . One multiplication shows shrinkage once, but the repeated sequence is what justifies saying the rule damps the state over time.

Do not infer long-run behavior from one step

A first update can look calm even when later iterations explode, or look large even when the process later settles.

• One output is evidence, not the whole trajectory.
• Iteration can amplify patterns hidden in a single step.
• Always ask what repeated application will do.

Recognize the difference between a stabilizing pattern and an amplifying one.

Hook: The learner compares repeated-update scenarios and has to decide whether the rule is damping the vector sequence or amplifying it.

Mastery Signals:

• can explain convergence in plain language
• can separate stabilizing behavior from raw coordinate size

Study Guide

Stable rules dampen, unstable rules amplify

A stabilizing update rule pushes the system toward a calmer or bounded pattern. An amplifying rule makes some components grow, oscillate, or dominate more strongly over time.

Stability is about long-run behavior under iteration

The important question is what happens to \(A^k\mathbf{x}\) as k grows. If iterates remain controlled or approach a steady pattern, the rule behaves stably; if they blow up or keep magnifying, it does not.

How to compare repeated-update scenarios

Look for the trend in the trajectory, not just one output.

• Are magnitudes shrinking, staying bounded, or growing?
• Does the pattern approach a steady shape?
• Does one direction begin to dominate the others?

Compare a damping rule to an amplifying rule

Start from \(\mathbf{x}_0=(1,1)\) . Under \(A=\tfrac12I\) , the iterates are \((1/2,1/2)\) , then \((1/4,1/4)\) , so the sequence shrinks toward zero. Under \(B=2I\) , the iterates are \((2,2)\) , then \((4,4)\) , so the same starting state grows step after step. Stability is about this trend, not about which first output looks bigger.

Choose the synthesis statement that treats the update rule as a longer-run process.

Hook: The learner now has to summarize why one iterated rule stabilizes while another keeps amplifying the sequence.

Mastery Signals:

• can choose the synthesis statement that explains a repeated rule as a process
• can keep convergence and amplification distinct

Study Guide

A dynamics synthesis summarizes the path, not the snapshot

A strong synthesis says how the rule behaves over repeated steps. It explains whether the process settles, amplifies, or keeps cycling instead of pretending the first output told the full story.

The object of study is the sequence x, \(A\mathbf{x}\) , \(A^2\mathbf{x}\) , …

Dynamic reasoning is about the entire generated sequence. The synthesis should therefore describe long-run behavior of iterates, not only one vector in isolation.

A correct memo names the full trajectory

If \(\mathbf{x}_0=(1,1)\) and \(A=2I\) , then the sequence is \((1,1)\) , \((2,2)\) , \((4,4)\) , \((8,8)\) , …. The right synthesis is not ’the output is \((2,2)\) .’ It is ’each step doubles the state, so the rule amplifies rather than settles.'

Do not confuse a calm first step with stability

The language of the synthesis should match the whole process.

• Stability claims need repeated-update evidence.
• Amplification can emerge only after several steps.
• The best summary names the direction of the trajectory.

Day 07: Factors, Compression, and Part I Synthesis

Day 7

The last step is compression with judgment: keep the important pattern, preserve the governing constraints, and write one synthesis note that carries the intuition arc into the formal half.

Story Anchor: This synthesis day closes the intuition-first half of the course by identifying the broadest linear patterns in the data.

Focus Question: What broad pattern is worth carrying into the formal second half of the course?

Daily Synthesis: Part I ends by identifying dominant patterns, comparing repeated behavior, and preparing for the formal second half.

Sessions

Recognize a dominant factor as a shared system pattern rather than a single standout coordinate.

Hook: The learner now looks across several changing coordinates and has to identify the broad shared pattern behind them.

Mastery Signals:

• can describe a broad shared pattern across several coordinates
• can reject the idea that a factor is just the largest number

Study Guide

A factor is a shared pattern across many coordinates

Instead of explaining every coordinate separately, a dominant factor captures the main way several coordinates move together. It is a system-level pattern, not one unusually large entry.

Factors summarize correlated structure

In matrix language, a dominant factor points toward a direction that explains much of the variation or signal. It is useful because several observed coordinates can be approximated through one common mode.

Several vectors can share one direction

The vectors \((1,2)\) , \((2,4)\) , and \((3,6)\) all point in the same direction, because each is a multiple of \((1,2)\) . The dominant factor is therefore the shared pattern \((1,2)\) , not the single largest coordinate 6.

Do not confuse a factor with the biggest coordinate

A factor describes a pattern across coordinates, not a single winner.

• Shared movement matters more than one standout entry.
• A factor should explain several coordinates at once.
• Local noise should not distract you from the common structure.

Use compression to keep the main mathematical pattern while reducing noisy detail.

Hook: The learner now has to pick the summary that keeps the big pattern without drowning the synthesis in every local fluctuation.

Mastery Signals:

• can explain why some detail can be dropped without losing the dominant pattern
• can reject summaries that lose the mathematical structure

Study Guide

Compression keeps the signal while discarding noise

A useful summary is shorter than the full data but still preserves the main pattern. Compression is not deletion for its own sake; it is a controlled reduction.

A lower-dimensional description can preserve dominant structure

The formal idea is that a complex vector or matrix can often be approximated by fewer directions if those directions capture most of the important variation or action.

A clean direction can summarize noisy data

Suppose the observed states are \((1.0,2.1)\) , \((2.0,3.9)\) , and \((3.0,6.2)\) . They are not exact multiples, but they stay close to the direction \((1,2)\) . Using that one direction as a compressed summary preserves the main pattern while ignoring small local noise.

Compression fails if the mathematical meaning disappears

A short summary is only mathematically useful if the core structure survives.

• Dropping detail is acceptable only when the dominant pattern remains.
• Compression should simplify interpretation, not distort it.
• A concise summary can still be wrong if it loses the signal.

Choose the synthesis note that integrates Part I before the course turns formal.

Hook: The learner now closes the intuition arc by selecting the integrated explanation that best carries the early ideas into the formal half of the course.

Mastery Signals:

• can integrate the major Part I ideas into one mathematical story
• can choose the synthesis note that preserves feasibility, reach, dynamics, and summary

Study Guide

Part I synthesis should preserve the main idea chain

The best carry-forward note ties together state vectors, matrix maps, reachability, repeated updates, and dominant patterns. It should feel like one mathematical picture, not a bag of disconnected topics.

The formal half will refine, not replace, these ideas

Elimination, subspaces, orthogonality, eigen ideas, and least squares all deepen the same objects you already met: vectors, linear maps, reach, long-run modes, and compressed structure.

One example can connect the whole intuition arc

Let \(\mathbf{x}_0=(2,1)\) record a state and let \(A=\begin{bmatrix}1 & 0 \\ 0 & 1/2\end{bmatrix}\) . Then \(\mathbf{x}_1=(2,1/2)\) and \(\mathbf{x}_2=(2,1/4)\) . The vector is the state, the matrix is the rule, repeated application shows the second coordinate fading, and a compressed summary keeps the dominant first direction instead of every tiny fluctuation.

Do not treat the intuition arc as disposable

The coming formalism should sharpen the intuition, not erase it.

• State, map, and output should stay connected.
• Reach and independence are already formal ideas in disguise.
• Compression and dominant patterns are previews of later linear algebra.

Day 08: Linear Systems, Elimination, and Invertibility

Day 8: Linear Systems, Elimination, and Invertibility

The intuition arc now turns formal. The same decision story is solved with elimination, reversibility, and solution-set reasoning.

Story Anchor: The course now turns formal: elimination and inverses solve linear systems and reveal when a map is reversible.

Focus Question: How do elimination and inverses solve a linear system?

Daily Synthesis: Day 8 formalizes the first half of the course: elimination simplifies a system without changing the answers, inverses recover inputs when the map is reversible, and solvability depends on the geometry of the constraints.

Sessions

Understand elimination as a structure-preserving rewrite of the same system.

Hook: A coupled linear system has to be untangled without changing what counts as a valid solution.

Mastery Signals:

• can explain why valid row operations preserve the solution set
• can describe pivots as structure, not guesswork

Study Guide

Elimination rewrites without changing the real problem

Gaussian elimination is useful because it untangles a coupled system while preserving which states actually solve it. The syntax changes, but the solution set should not.

Elementary row operations preserve the solution set

Swapping equations, rescaling a nonzero row, and adding a multiple of one row to another all produce an equivalent system. They change the representation, not the underlying set of solutions.

Rewrite the system without changing its solution

Start with \(x+y=5\) and \(x-y=1\) . Add the second equation to the first to get \(2x=6\) , so \(x=3\) and then \(y=2\) . The rewritten system is easier to solve, but the solution set stayed the same the whole time.

What an undergraduate test usually asks here

Be ready to name the valid row operations, explain why they preserve the solution set, and carry a small system to echelon form without losing the meaning of the equations.

• State that equivalent row operations keep the same solutions.
• Use pivots to track leading variables and structure.
• Do not present row reduction as random arithmetic.

Show why a row operation preserves solutions

A proof-style answer starts with an arbitrary solution x of the original system and shows that the row-rewritten equations are still true for that same x, and conversely. The arithmetic move matters only because the truth set is unchanged.

• Explain swap, rescale, and row-add as equivalent rewrites of equations.
• Use the phrase same solution set or equivalent system.
• Keep the logic of preservation separate from the later computation.

Do not treat pivots as arbitrary arithmetic moves

The point of elimination is structural simplification, not random row fiddling.

• Ask what relation is being preserved.
• A cleaner row form should still describe the same solutions.
• Pivots reveal dependence and freedom in the system.

Treat an inverse as the map that undoes the original transformation.

Hook: An output vector is given, and the learner wants to recover the unique input vector that produced it.

Mastery Signals:

• can explain inverse as whole-map undoing
• can connect invertibility to unique recovery

Study Guide

An inverse is the rule that undoes the original map

If a transformation preserves enough information, you can run the logic backward from output to input. That is the job of the inverse.

Invertibility means unique recovery

A matrix A is invertible when there exists \(A^{-1}\) such that \(A^{-1}(A\mathbf{x})=\mathbf{x}\) and \(A(A^{-1}\mathbf{y})=\mathbf{y}\) . This means each output comes from exactly one input.

Undo one output back to one input

If an invertible matrix sends the input vector \((2,1)\) to the output vector \((5,1)\) , then applying \(A^{-1}\) to \((5,1)\) must return \((2,1)\) . That is the practical meaning of invertibility: one output corresponds to one recoverable input.

Inverse questions are verification questions

On a test, you may need to decide whether A is invertible, compute \(A^{-1}\) in simple cases, or verify a proposed inverse by checking that composition really gives the identity.

• An inverse undoes the whole map, not one coordinate at a time.
• Check \(AA^{-1}=I\) and \(A^{-1}A=I\) when verifying.
• Use \(A^{-1}\mathbf{b}\) only after invertibility is justified.

An inverse is not entrywise reciprocals

The inverse undoes the whole linear map, not each entry separately.

• If two inputs collapse to one output, no inverse can recover them uniquely.
• Invertibility is a property of the full transformation.
• Recovery only makes sense when information has not been lost.

Connect elimination outcomes to one-solution, no-solution, and many-solution stories.

Hook: The learner now has to read the geometry implied by the reduced system instead of just finishing the arithmetic.

Mastery Signals:

• can interpret contradiction as no solution
• can distinguish unique from underdetermined systems

Study Guide

Row-reduced form tells a story about the solution set

After elimination, the point is to interpret the outcome: one consistent intersection, a contradiction, or leftover freedom. Solving still ends in geometry and meaning.

Unique, none, or infinitely many solutions

A contradictory row signals no solution. A pivot in every variable can signal a unique solution. Free variables signal infinitely many solutions inside a consistent system.

Read three different reduced outcomes

Rows \(\begin{bmatrix}1 & 0 & & & \mid & & & 2\end{bmatrix}\) and \(\begin{bmatrix}0 & 1 & & & \mid & & & 1\end{bmatrix}\) give one unique solution. A row \(\begin{bmatrix}0 & 0 & & & \mid & & & 1\end{bmatrix}\) gives a contradiction and therefore no solution. A consistent reduced system with one free variable gives infinitely many solutions because one coordinate can still vary.

Reduced rows classify the whole system

A common exam move is to row reduce and then classify the system as unique, inconsistent, or underdetermined. Read that from the reduced form directly instead of guessing from the original equations.

• A contradictory row like \(0=1\) means no solution.
• A pivot in every variable gives a unique solution.
• A free variable gives infinitely many solutions.

Review set: systems and invertibility

Try these without looking back at the worked examples first.

• Name the three elementary row operations and say what they preserve.
• Classify a reduced system as unique, inconsistent, or infinitely solvable from its pivots and free variables.
• Explain why a reversible matrix gives one unique input for each output.

Checkpoint problems: elimination and solvability

Work these as if they were short exam questions before reading the worked solution.

• Row reduce the system \(x+y=2,\quad x-y=0\) and classify the solution set.
• Explain what a row \(\begin{bmatrix}0 & 0 & & & \mid & & & 1\end{bmatrix}\) means for consistency.
• State one reduced-form signal that a square system is invertible.

Solution sketch: read the reduced form, then conclude

A full-credit answer should show the reduced structure and then interpret it.

• Adding the equations gives \(2x=2\) , so \(x=1\) and then \(y=1\) ; the system has one unique solution.
• A row \(\begin{bmatrix}0 & 0 & & & \mid & & & 1\end{bmatrix}\) says \(0=1\) , so the system is inconsistent and has no solution.
• A square system is invertible when the reduced form has a pivot in every variable, equivalently no free variables and no contradiction.

Clean rows are not the final answer by themselves

A reduced matrix is only useful if you read what it implies about the set of possible states.

• Contradiction means the constraints cannot all hold together.
• Free variables mean the system still has genuine flexibility.
• Interpret the whole reduced system, not one row in isolation.

Day 09: Subspaces, Column Space, and Null Space

Day 9: Subspaces, Column Space, and Null Space

The course now studies the spaces created by a linear map: what stays stable, what can be reached, and what disappears from view.

Story Anchor: This day studies subspaces and the fundamental spaces created by a matrix.

Focus Question: Which sets stay closed under linear combination, and which directions map to zero?

Daily Synthesis: Day 9 turns reachability into formal spaces: subspaces are closed under linear combination, column space names what outputs are achievable, and null space names the input freedom the map cannot detect.

Sessions

Treat a subspace as a set closed under linear combination, not just a collection of examples.

Hook: A family of allowed adjustments only deserves to be called linear if mixing and rescaling still keeps you inside it.

Mastery Signals:

• can test closure instead of relying on examples
• can explain why scaling matters as much as addition

Study Guide

A subspace is a family that stays linear when you work inside it

A true linear family is stable under the operations linear algebra cares about. If you mix two allowed directions or rescale one, you should remain inside the family.

Closure under addition and scalar multiplication

A subset is a subspace if it contains the zero vector and is closed under vector addition and scalar multiplication. Examples alone are never enough.

Test closure on the line \(y=2x\)

Take the vectors \((1,2)\) and \((2,4)\) . Their sum is \((3,6)\) , which still satisfies \(y=2x\) , and \(3(1,2)=(3,6)\) also stays on the same line. This is why the line through the origin given by \(y=2x\) is a subspace.

Subspace proofs need the definition

Expect to prove that a set is or is not a subspace using the actual closure conditions. A few examples may help intuition, but they are not a proof.

• Check that the zero vector is included.
• Test closure under addition and scalar multiplication.
• Use one failed condition to build a counterexample.

Prove subspace or disprove it with one counterexample

A full-credit answer either checks zero, addition, and scalar multiplication for the whole set, or it gives one clean violation showing the set is not closed.

• If disproving, choose vectors already in the set and show the result leaves it.
• If proving, write the closure step with arbitrary elements, not isolated examples.
• Keep the definition visible all the way through the argument.

Several good examples do not prove subspace status

The definition is about all allowed combinations, not a few representative points.

• Check closure, not just membership.
• Scaling matters as much as addition.
• A set can look structured and still fail to be a subspace.

Read the column space as the outputs the map can actually produce.

Hook: The learner compares what the map can generate with what lies outside its genuine reach.

Mastery Signals:

• can describe column space as achievable outputs
• can connect output reach to span

Study Guide

Column space is the set of outputs the map can really produce

Some target outputs are achievable and some are not. Column space is the honest answer to what the matrix can generate from some input.

Column space is the span of the columns of A

Every output Ax is a linear combination of the columns of A, with coefficients coming from x. Therefore the achievable outputs are exactly the span of those columns.

Dependent columns restrict the outputs

If the columns of A are \((1,1)\) and \((2,2)\) , then every output is some multiple of \((1,1)\) . A target like \((1,-1)\) lies outside that span, so it is not in the column space and cannot be produced by the map.

Column-space questions are reachability questions

Typical exam problems ask whether a target vector b lies in Col(A), what vectors span the column space, or why some outputs are impossible for the map.

• Remember that Col(A) is the span of the columns of A.
• b lies in Col(A) exactly when \(A\mathbf{x}=\mathbf{b}\) is consistent.
• Column dependence shrinks the set of achievable outputs.

Do not confuse a desired target with an achievable one

A target can be operationally attractive and still lie outside the column space.

• Ask whether the output can be written as a combination of columns.
• Column space lives in the output space, not the input space.
• Reachability is about the map, not your preference.

Understand the null space as hidden input freedom that the map erases.

Hook: The learner now sees two different inputs produce the same output and has to name the hidden direction behind that duplication.

Mastery Signals:

• can describe null space as invisible input variation
• can connect repeated outputs to hidden directions

Study Guide

Null space is hidden input freedom

If two different inputs create the same output, the map has ignored their difference. Null space captures exactly those invisible directions.

Null space consists of all x such that \(A\mathbf{x}=\mathbf{0}\)

A vector lies in the null space when the map sends it to the zero vector. Differences between inputs that fall in the null space leave the visible output unchanged.

Find a hidden direction

For the map \(A(x,y)=(x+y,x+y)\) , the vector \((1,-1)\) is in the null space because \(A(1,-1)=(0,0)\) . Adding any multiple of \((1,-1)\) to an input changes the input, but not the visible output.

Null-space work means solving \(A\mathbf{x}=\mathbf{0}\)

On a test, compute the null space by row reducing the homogeneous system and writing the free variables as parameters. The answer is usually a whole set, not one vector.

• Nonpivot columns create free variables.
• Parameterize the full family of null vectors.
• A nontrivial null space signals hidden dependence.

Write the null space as a justified family

A strong exam answer does more than name one hidden direction. It row reduces \(A\mathbf{x}=\mathbf{0}\) , solves in parameters, and then writes the full null space as a span or parametric family.

• Identify pivot and free variables before solving.
• Express the answer as all vectors of the form t v when possible.
• Check that each basis vector really maps to zero.

Review set: subspace, column space, null space

Use these prompts to rehearse the whole Day 9 chain.

• Decide whether a set like \(\{(x,y):x+y=1\}\) is a subspace and justify the answer.
• Given matrix columns, decide whether a target b lies in the column space.
• Solve a small homogeneous system and write a basis for its null space.

Checkpoint problems: subspace and hidden directions

Treat these like short written checkpoint problems.

• Find the null space of the matrix with rows \(\begin{bmatrix}1 & 1\end{bmatrix}\) and \(\begin{bmatrix}1 & 1\end{bmatrix}\) .
• Decide whether \(\mathbf{b}=(1,-1)\) lies in the column space of columns \((1,1)\) and \((2,2)\) .
• Explain in one sentence why null space lives in the input space.

Solution sketch: solve \(A\mathbf{x}=\mathbf{0}\) , then interpret

Each answer should separate the algebra from the geometric meaning.

• The system \(x+y=0\) gives \(y=-x\) , so the null space is all t\((1,-1)\) ; a basis is {\((1,-1)\) }.
• No, because the column space is the span of \((1,1)\) , so every achievable output is a multiple of \((1,1)\) , not \((1,-1)\) .
• Null space lives in the input because it consists of input vectors x that the map sends to zero output.

Null space is not ’nothing'

It describes real input directions, but ones the output cannot detect.

• Null directions are hidden, not meaningless.
• They explain duplicate outputs from different causes.
• Null space lives in the input side of the map.

Day 10: Basis Coordinates, Rank, and Dimension

Day 10: Basis Coordinates, Rank, and Dimension

This day formalizes how basis choice changes coordinates while rank and nullity explain what the map preserves and what it hides.

Story Anchor: This day formalizes basis, coordinates, rank, and nullity.

Focus Question: How does a change of basis alter coordinates, and how do rank and nullity count independent directions?

Daily Synthesis: Day 10 ties basis and rank together formally: the same vector can wear different coordinates in different bases, rank counts the independent directions the map keeps alive, and nullity counts what disappears from view.

Sessions

Separate the vector itself from the coordinates used to describe it.

Hook: The same vector can be described in two different bases, and the learner has to keep the object separate from its encoding.

Mastery Signals:

• can explain why coordinates depend on basis
• can keep the underlying vector fixed while descriptions change

Study Guide

The vector stays the same even when the coordinates change

A basis is a language for describing vectors. Changing the basis changes the coordinate description, not the underlying state you are describing.

Coordinates depend on the chosen basis

The same vector v can have one coordinate list in basis B and a different coordinate list in basis C. The object v is fixed; only its representation changes.

Keep the vector fixed while the coordinates change

The vector \(\mathbf{v}=(3,1)\) has standard coordinates \((3,1)\) . In the basis \(\mathbf{b}_1=(1,1)\) , \(\mathbf{b}_2=(1,-1)\) , the same vector can be written as \(\mathbf{v}=2\mathbf{b}_1+\mathbf{b}_2\) , so its new coordinate list is \((2,1)\) . The object stayed the same; only the description changed.

Coordinate problems separate object from description

Be ready to write one vector in a chosen basis and to explain why the coordinates change even though the underlying vector does not.

• Solve for the coefficients in the given basis vectors.
• Check that the listed basis vectors are independent and spanning.
• Do not confuse the geometric vector with one coordinate list.

Why basis coordinates are unique

If v has two coordinate representations in the same basis, subtract them. The difference becomes a linear combination of basis vectors equaling zero, and linear independence forces every coefficient difference to vanish.

• Set \(c_1\mathbf{b}_1+\cdots+c_n\mathbf{b}_n=d_1\mathbf{b}_1+\cdots+d_n\mathbf{b}_n\) .
• Move all terms to one side and use independence.
• Conclude \(c_i=d_i\) for every coordinate.

Do not identify a vector with one coordinate list

Coordinates are basis-relative encodings, not the vector itself.

• A new basis can change every coordinate while preserving the same state.
• Comparing coordinate lists only makes sense relative to a basis.
• The vector is the object; coordinates are the description.

Interpret rank as the number of independent output directions that survive the map.

Hook: The learner now asks how many independent output directions remain once the matrix has collapsed redundant directions.

Mastery Signals:

• can explain rank as surviving independent output directions
• can distinguish raw matrix size from actual surviving dimension

Study Guide

Rank counts surviving independent output directions

A map can look large on the page and still collapse several directions together. Rank measures how many genuinely different output directions remain.

Rank is a dimension, not a raw size count

Rank is the dimension of the column space, equivalently the number of pivot columns. It tells how much independent output structure the matrix preserves.

Two columns can still give rank one

A matrix with columns \((1,0)\) and \((2,0)\) looks like it has two output directions written down, but the second column is just twice the first. The column space is one-dimensional, so the rank is 1, not 2.

Rank is a pivot count, not a shape count

Most exam problems compute rank from row reduction and then interpret what that number means for the dimension of the output space.

• Rank equals the number of pivots.
• Rank also equals the dimension of the column space.
• Extra dependent columns do not increase rank.

A wide matrix can still have low rank

Visual width or height does not guarantee independent dimension.

• Dependent columns do not add rank.
• Rank measures surviving structure, not decoration.
• Actual output reach can be much smaller than the matrix shape suggests.

Use rank-nullity to relate visible dimension and hidden freedom.

Hook: The learner now has to keep both sides of the count in mind: what the map shows and what it hides.

Mastery Signals:

• can explain rank plus nullity as a full accounting of the input space
• can relate hidden directions to lost output dimension

Study Guide

Rank-nullity balances visible dimension and hidden freedom

If the map hides some input directions, those lost degrees of freedom have to be accounted for. Rank-nullity is the bookkeeping rule for what the map shows and what it hides.

Dimension of domain equals rank plus nullity

For a linear map \(A:\mathbb{R}^n\to\mathbb{R}^m\) , \(n=\operatorname{rank}(A)+\operatorname{nullity}(A)\) . The full input space splits into directions the map sends visibly outward and directions it collapses into zero.

Account for every input direction

For \(A=\begin{bmatrix}1 & 0 & 0 \\ 0 & 1 & 0\end{bmatrix}\) , the map sends \((x,y,z)\) to \((x,y)\) . Rank is 2 because two output directions remain visible, and nullity is 1 because the z-direction disappears. The identity \(3=2+1\) is rank-nullity in action.

Rank-nullity is a bookkeeping theorem

If A has n input columns, then \(\operatorname{rank}(A)+\operatorname{nullity}(A)=n\) . Use the pivot count to recover the missing dimension quickly on exam questions.

• Nullity equals number of columns minus rank.
• Free variables account for the nullity.
• The theorem balances visible output directions and hidden input freedom.

Explain rank-nullity from free variables

A proof sketch usually row reduces A, counts pivot columns for the rank, then counts free variables for the nullity. Those two counts together exhaust the total number of input columns.

• Pivot columns count independent visible directions.
• Nonpivot columns create free variables and null directions.
• Rank plus nullity equals the total input dimension because every column is either pivot or free.

Review set: basis, rank, and nullity

Use these to rehearse the coordinate-and-dimension week.

• Rewrite one vector in a nonstandard basis and explain why the vector itself does not change.
• Find rank from a row-reduced matrix by counting pivots.
• Use rank-nullity to recover the number of free variables.

Checkpoint problems: coordinates and dimension counts

Answer these as compact written exam responses.

• A matrix has 5 columns and rank 3. What is its nullity?
• If a basis is \(\mathbf{b}_1=(1,1)\) , \(\mathbf{b}_2=(1,-1)\) , write \((3,1)\) in that basis.
• Explain why free variables and nullity count the same hidden freedom.

Solution sketch: count pivots, then solve for coordinates

Keep the counting logic explicit in each answer.

• \(\operatorname{nullity}=5-3=2\) by rank-nullity.
• \((3,1)\) = 2\((1,1)\) + 1\((1,-1)\) , so the basis coordinates are \((2,1)\) .
• Each free variable can vary independently and produces a null-space direction, so the number of free variables equals the nullity.

Do not memorize the equation without the counts

Each term counts something different and meaningful.

• Rank counts visible independent outputs.
• Nullity counts hidden input directions.
• Together they recover the full dimension of the input space.

Day 11: Dot Product, Orthogonality, and Projection

Day 11: Dot Product, Orthogonality, and Projection

This day turns geometry into a working tool: alignment, separation, and best-fit projection inside a simpler model.

Story Anchor: This day introduces inner-product geometry: alignment, perpendicularity, and best approximation.

Focus Question: How do the dot product and projection measure alignment and best fit?

Daily Synthesis: Day 11 adds the geometry of inner products: the dot product measures alignment, orthogonality marks genuinely separate directions, and projection keeps the closest signal the model can express.

Sessions

Interpret the dot product as one alignment score between two directions.

Hook: One vector may point with another direction or away from it, and the learner needs one number that captures that difference.

Mastery Signals:

• can explain dot product as alignment
• can use sign and size to interpret response quality

Study Guide

Dot product is one score for directional alignment

The dot product condenses two direction patterns into one number. Positive means aligned, negative means opposed, and zero means no directional overlap.

The sign and size both matter

For vectors u and v, \(\mathbf{u}\cdot\mathbf{v}=\sum_i u_iv_i\) . Its sign shows directional agreement or opposition, while its magnitude reflects how strongly the directions line up.

Compute one alignment score

For \(\mathbf{u}=(1,2)\) and \(\mathbf{v}=(2,4)\) , the dot product is \(1\cdot2+2\cdot4=10\) . The score is large and positive, so the vectors point strongly in the same direction even though they are not required to be identical.

Dot-product questions mix computation and interpretation

Expect to compute u·v and then explain alignment from the sign and size instead of stopping at the arithmetic.

• Positive means aligned, negative means opposed.
• Zero signals orthogonality, not merely small entries.
• A complete answer interprets the number geometrically.

The dot product is not just coordinate matching

It measures whole-vector alignment, not isolated coordinate coincidences.

• A large positive score means strong directional agreement.
• Zero means orthogonality, not necessarily small coordinates.
• Interpret the result geometrically, not only arithmetically.

Understand why orthogonal residuals certify that a projection is the best fit.

Hook: The learner now fits a noisy signal with a simpler model and has to explain what the leftover error means.

Mastery Signals:

• can explain why orthogonal residual means no better fit exists inside the subspace
• can distinguish leftover error from supported signal

Study Guide

A best fit leaves error in a different direction

If the fitted point is truly the closest one in the model, the leftover error cannot point along the model again. If it did, you could move further and improve the fit.

Best-fit residuals are orthogonal to the model subspace

When p is the projection of y onto a subspace S, the residual y - p is orthogonal to every vector in S. Orthogonality certifies that no closer point inside S exists.

Project onto the x-axis

Project \(\mathbf{y}=(2,1)\) onto the x-axis. The projection is \(\mathbf{p}=(2,0)\) , so the residual is \(\mathbf{y}-\mathbf{p}=(0,1)\) . That residual is orthogonal to the x-axis, which is why no other point on the axis can be closer.

Best-fit proofs use orthogonality

For projection problems, the test-worthy fact is that the error b - p is orthogonal to the subspace or chosen direction.

• Set the residual dot the model direction to zero.
• Orthogonality is what certifies the closest point.
• A small error alone is not the full justification.

Why orthogonal residual means closest point

If p is the projection and q is any other point in the subspace, then \(\mathbf{y}-\mathbf{q}=(\mathbf{y}-\mathbf{p})+(\mathbf{p}-\mathbf{q})\) . Orthogonality makes these two pieces perpendicular, so the Pythagorean theorem shows \(\lVert\mathbf{y}-\mathbf{q}\rVert^2=\lVert\mathbf{y}-\mathbf{p}\rVert^2+\lVert\mathbf{p}-\mathbf{q}\rVert^2\) , which is never smaller than \(\lVert\mathbf{y}-\mathbf{p}\rVert^2\) .

• Decompose the error into residual plus in-subspace difference.
• Use orthogonality to justify the squared-norm split.
• Conclude that every other subspace point is at least as far away.

Residual is not just a small mistake term

Its direction matters, not just its size.

• Residual describes what the model cannot explain.
• Orthogonality is what makes the fit optimal.
• Best fit is geometric, not just procedural.

Treat projection as the closest point in the model subspace.

Hook: The learner now has to name the precise job projection is doing when the data does not land exactly on the model line.

Mastery Signals:

• can explain projection as closest fit
• can connect projection directly to model restriction

Study Guide

Projection keeps the closest point the model can support

When the data does not lie exactly inside the chosen model space, projection keeps the nearest supported signal instead of inventing an arbitrary simplification.

Projection is the closest point in the subspace

The projection of y onto a subspace S is the vector p in S minimizing ||y - p||. It is the disciplined way to approximate data under a model restriction.

Closest point on the line \(y=x\)

Project the vector \((2,0)\) onto the line spanned by \((1,1)\) . The closest point is \((1,1)\) , not \((2,0)\) with a coordinate deleted. Projection depends on the geometry of the model subspace, not on arbitrary truncation.

Projection is the closest-vector tool

You may be asked to project onto a line or subspace and justify why the answer is the closest point available inside that model.

• For a line spanned by u, use \(\operatorname{proj}_{\mathbf{u}}(\mathbf{v})=\dfrac{\mathbf{v}\cdot\mathbf{u}}{\mathbf{u}\cdot\mathbf{u}}\mathbf{u}\) .
• The leftover residual must be orthogonal to the model space.
• Projection solves a distance problem, not a truncation trick.

Review set: dot product, orthogonality, projection

These are the core Day 11 exam moves.

• Compute a dot product and interpret the sign and size.
• Decide whether two vectors are orthogonal from their dot product.
• Project a vector onto a line and explain why the residual must be perpendicular.

Checkpoint problems: alignment and projection

Work these as if they were a short orthogonality section on an exam.

• Compute \((1,2)\) · \((2,-1)\) and interpret the result.
• Project \((2,0)\) onto the line spanned by \((1,1)\) .
• Explain why the residual from a projection must be orthogonal to the model line.

Solution sketch: compute, project, justify

A strong answer finishes the arithmetic with geometric interpretation.

• \((1,2)\cdot(2,-1)=2-2=0\) , so the vectors are orthogonal.
• \(\operatorname{proj}_{(1,1)}(2,0)=\dfrac{(2,0)\cdot(1,1)}{(1,1)\cdot(1,1)}(1,1)=\dfrac22(1,1)=(1,1)\) .
• The residual must be perpendicular because otherwise moving along the line could reduce the distance further, so the point was not truly closest.

Projection is not truncation

You do not simply delete coordinates and hope for the best.

• Projection depends on the chosen subspace.
• The best fit is defined by distance minimization.
• If the data already lies in the subspace, projection changes nothing.

Day 12: Determinant, Orientation, and Volume

Day 12: Determinant, Orientation, and Volume

Determinant is introduced as a structural signal of scaling, reversal, and collapse in a linear map.

Story Anchor: This day reads determinant as a geometric measure of scaling, orientation, and collapse.

Focus Question: What does the determinant say about volume, orientation, and invertibility?

Daily Synthesis: Day 12 turns determinant into a geometric signal: its magnitude describes how volume scales, its sign tracks orientation, and its nonzero status tells whether the map stays invertible.

Sessions

Interpret determinant as signed volume scaling under the map.

Hook: The learner now asks what the whole transformation does to independent volume instead of staring at individual entries.

Mastery Signals:

• can explain determinant as a whole-map quantity
• can connect determinant magnitude to preserved or scaled volume

Study Guide

Determinant summarizes how the whole map scales volume

Determinant answers a global question: what happens to independent volume when the transformation acts? It is about the whole map, not one entry.

Magnitude of determinant gives signed volume scaling

For a square matrix, |det(A)| tells how n-dimensional volume is scaled. A nonzero determinant means independent directions still survive after the map acts.

Scale the unit square

The matrix \(\begin{bmatrix}2 & 0 \\ 0 & 3\end{bmatrix}\) sends the unit square to a 2-by-3 rectangle. Area grows from 1 to 6, so the determinant is 6. The determinant is reporting whole-shape scaling, not one entry at a time.

Determinant problems ask about scaling and orientation

For 2x2 and 3x3 exam questions, compute the determinant and then interpret area or volume scaling together with possible orientation reversal.

• Absolute value gives the scaling factor.
• The sign tells whether orientation flips.
• Zero means collapse to lower dimension.

Determinant is not one more arbitrary formula

Its meaning comes from the transformation, not from memorizing a calculation scheme.

• Ask what the map does globally.
• A nonzero determinant signals preserved dimension.
• Magnitude and sign each carry geometric meaning.

Use determinant zero to recognize collapse and lost invertibility.

Hook: A map that sends different inputs to the same output has already revealed that some independent direction has collapsed.

Mastery Signals:

• can connect determinant zero to collapse of dimension
• can explain why singular maps are not invertible

Study Guide

Determinant zero means some independent direction has collapsed

When different inputs land on the same output, the map has already lost a dimension of distinction. Determinant zero is the compact signal of that collapse.

Singular maps are not invertible

If \(\det(A)=0\) , then A is singular. The map collapses dimension, so it cannot have an inverse and cannot recover inputs uniquely from outputs.

Dependent columns force determinant zero

For \(A=\begin{bmatrix}1 & 1 \\ 2 & 2\end{bmatrix}\) , the second column is a multiple of the first. The map collapses the plane onto one line, so different inputs can land at the same output. That geometric collapse is exactly why \(\det(A)=0\) .

Determinant zero has many consequences

On a test, \(\det(A)=0\) is rarely the end of the story. It signals singularity, dependence, lost inverse, and collapse of dimension all at once.

• \(\det(A)=0\) means A is not invertible.
• Dependent columns or rows force determinant zero.
• Zero determinant often predicts failed uniqueness in \(A\mathbf{x}=\mathbf{b}\) .

Connect determinant zero to singularity

A proof-style explanation links the algebra to the structure: if \(\det(A)=0\) , the columns are dependent, so the map collapses dimension and cannot be invertible. In the opposite direction, invertibility forces nonzero determinant.

• Use dependence or dimension collapse, not just a memorized slogan.
• Tie singularity to failure of unique recovery.
• Say explicitly why an inverse cannot exist after collapse.

Do not think determinant zero is just a number fact

It reports a structural failure of the transformation.

• A lost direction means duplicate outputs become possible.
• Collapse and non-invertibility are the same story here.
• Read the geometry before the formula.

Separate determinant sign from determinant nonzero status.

Hook: The learner now has to keep two determinant jobs apart: sign for orientation, nonzero status for invertibility.

Mastery Signals:

• can explain orientation sign separately from nonzero status
• can avoid collapsing all determinant meaning into one fact

Study Guide

Determinant carries two messages at once

The sign tells whether orientation flips, while nonzero status tells whether dimension survives. These are related but distinct jobs.

Sign tracks orientation, nonzero tracks invertibility

A negative determinant indicates orientation reversal. A nonzero determinant indicates the map remains invertible. Sign and zero-status should not be merged into one idea.

Negative can still be invertible

The matrix \(\begin{bmatrix}-1 & 0 \\ 0 & 1\end{bmatrix}\) has determinant -1. The negative sign means orientation flips, but the determinant is still nonzero, so the map is invertible. This is why sign and invertibility have to be read separately.

Separate sign from invertibility

A negative determinant can still mean the matrix is invertible. The sign answers an orientation question; zero versus nonzero answers a solvability question.

• Negative and positive determinants can both be nonzero.
• Nonzero determinant implies invertibility.
• Do not let the sign distract from the zero/nonzero test.

Review set: determinant and invertibility

Use this short set to rehearse determinant interpretation.

• Compute a 2x2 determinant and interpret its absolute value as area scaling.
• Explain what determinant zero says about invertibility and dependence.
• Separate orientation sign from the zero versus nonzero question.

Checkpoint problems: determinant and orientation

Treat these like a short determinant section from an undergraduate midterm.

• Compute det(\(\begin{bmatrix}2 & 1 \\ 0 & 3\end{bmatrix}\) ) and interpret the area scaling.
• State what \(\det(A)=0\) implies about invertibility.
• Explain what a negative determinant tells you that a positive one does not.

Solution sketch: number first, then structure

Do not stop after computing the determinant.

• \(\det\begin{pmatrix}2&1\\0&3\end{pmatrix}=6\) , so areas are scaled by a factor of 6 and orientation is preserved because the determinant is positive.
• If \(\det(A)=0\) , the matrix is singular and not invertible.
• A negative determinant means orientation flips; zero versus nonzero still controls invertibility.

Do not collapse all determinant meaning into ‘positive or negative’

Orientation and invertibility answer different questions.

• Negative does not mean non-invertible.
• Zero is about collapse, not merely sign.
• Keep the determinant’s two roles conceptually separate.

Day 13: Eigenvectors, Eigenvalues, and Diagonalization

Day 13: Eigenvectors, Eigenvalues, and Diagonalization

The dynamics story becomes formal: steady directions, scaling factors, and the basis change that decouples them.

Story Anchor: This day studies directions that keep their line under a map and the basis that decouples them.

Focus Question: Which vectors keep their direction under the map, and what does diagonalization buy you?

Daily Synthesis: Day 13 gives the formal version of the dynamics arc: eigenvectors identify steady directions, eigenvalues explain growth and decay on those directions, and diagonalization rewrites the map in a basis where the modes stop interfering with each other.

Sessions

Recognize an eigenvector as a direction that keeps its shape under the map.

Hook: The learner now looks for directions that do not rotate into something qualitatively new when the matrix acts.

Mastery Signals:

• can explain eigenvectors as direction-preserving
• can avoid confusing eigenvectors with fixed points

Study Guide

An eigenvector keeps its direction under the map

Most vectors get bent into a new direction, but an eigenvector comes back pointing along the same line. Only its size may change.

Eigenvector means \(A\mathbf{v}=\lambda\mathbf{v}\)

A nonzero vector v is an eigenvector of A if applying A produces a scalar multiple of v. Direction is preserved even though magnitude may grow, shrink, or flip.

Find one steady direction in a diagonal map

For \(A=\begin{bmatrix}3 & 0 \\ 0 & 1/2\end{bmatrix}\) , the vector \(\mathbf{e}_1=(1,0)\) satisfies \(A\mathbf{e}_1=(3,0)=3\mathbf{e}_1\) . The direction stays on the x-axis, so \(\mathbf{e}_1\) is an eigenvector even though its size changes.

Eigenvector questions start from \(A\mathbf{v}=\lambda\mathbf{v}\)

Typical problems ask you to test a candidate vector, solve for eigenvalues, or find an eigenspace. Keep the scalar-multiple definition in front.

• An eigenvector must be nonzero.
• A negative eigenvalue flips direction along the same line.
• A fixed point is only the special case \(\lambda=1\) .

Eigenvectors are not fixed points unless \(\lambda=1\)

The key property is preserved direction, not unchanged value.

• A negative eigenvalue flips orientation along the same line.
• A larger magnitude can still preserve eigenvector status.
• Direction preservation is the defining idea.

Use the eigenvalue to explain growth, decay, or flip along a steady direction.

Hook: The learner now has to say not just which direction stays steady, but what the matrix does to its size.

Mastery Signals:

• can interpret eigenvalues as mode scaling
• can connect scaling size to long-run importance

Study Guide

Eigenvalues tell what the map does to a steady mode

Once a direction is preserved, the next question is whether the system grows it, shrinks it, or flips it. The eigenvalue gives that answer.

Magnitude controls growth or decay of the mode

If \(A\mathbf{v}=\lambda\mathbf{v}\) , then repeated application gives \(A^k\mathbf{v}=\lambda^k\mathbf{v}\) . So \(|\lambda|\) greater than 1 amplifies, less than 1 damps, and equal to 1 preserves size.

Compare a growing mode and a decaying mode

If \(A\mathbf{v}=2\mathbf{v}\) , then \(A^3\mathbf{v}=8\mathbf{v}\) , so that mode grows rapidly. If \(A\mathbf{w}=\tfrac12\mathbf{w}\) , then \(A^3\mathbf{w}=\tfrac18\mathbf{w}\) , so that mode fades away. This is why eigenvalue magnitude predicts long-run importance.

Eigenvalues control repeated behavior

On exams, interpret \(\lambda\) as growth, decay, or sign flip along an eigenvector under repeated multiplication.

• If \(|\lambda|\) > 1, the mode grows.
• If \(|\lambda|\) < 1, the mode decays.
• If \(\lambda<0\) , the direction alternates sign each step.

Direction alone is not enough

Two eigenvectors can both preserve direction while behaving very differently over time.

• The eigenvalue controls long-run importance.
• Large-magnitude modes dominate repeated dynamics.
• Read size change after direction preservation.

Explain why diagonalization makes a coupled system easier to analyze.

Hook: The learner now has to connect the eigen stories back to one larger formal move: rewriting the problem in a basis where the modes stop mixing.

Mastery Signals:

• can explain why the right basis decouples the dynamics
• can connect diagonalization to a coordinate change, not a new system

Study Guide

Diagonalization untangles coupled dynamics into separate modes

The same system can become much easier to study if you express it in the right basis. Then each important mode evolves on its own instead of staying mixed with the others.

A change of basis can make the matrix diagonal

When \(A=PDP^{-1}\) , the matrix D records independent eigen-mode scalings in the eigen-basis. The transformation is the same; only the coordinate description becomes simpler.

See the decoupled coordinates directly

If the eigen-basis gives \(D=\operatorname{diag}(2,\tfrac12)\) , then a coordinate vector \((a,b)\) becomes (2a, b/2) after one step. The two coordinates no longer mix, so one mode can grow while the other fades without extra coupling.

Diagonalization is useful only with enough eigenvectors

A matrix is diagonalizable when you can build a basis of eigenvectors. Then \(A=PDP^{-1}\) turns repeated powers into easy scalar powers on D.

• You need enough independent eigenvectors to form P.
• Columns of P are eigenvectors and entries of D are eigenvalues.
• Diagonalization simplifies analysis; it does not change the underlying map.

Show how independent eigenvectors produce \(A=PDP^{-1}\)

If \(v_1\) , …, vn are independent eigenvectors with eigenvalues \(\lambda_1\) , …, \(\lambda_n\) , put them into \(P=[\mathbf{v}_1\;\cdots\;\mathbf{v}_n]\) and \(D=\operatorname{diag}(\lambda_1,\ldots,\lambda_n)\) . Then \(AP=PD\) because each column satisfies \(A\mathbf{v}_i=\lambda_i\mathbf{v}_i\) , and multiplying by \(P^{-1}\) gives the diagonalization formula.

• State why independence makes P invertible.
• Use the column-by-column identity \(A\mathbf{v}_i=\lambda_i\mathbf{v}_i\) .
• Conclude with \(A=PDP^{-1}\) , not just \(AP=PD\) .

Review set: eigenvectors, eigenvalues, diagonalization

These are the main Day 13 revision prompts.

• Test whether a given nonzero vector is an eigenvector of A.
• Interpret an eigenvalue as growth, decay, or sign flip under iteration.
• State the condition that makes a matrix diagonalizable.

Checkpoint problems: modes and diagonalization

Work these as concise eigen-analysis questions.

• Show that \((1,0)\) is an eigenvector of \(\begin{bmatrix}3 & 0 \\ 0 & 1/2\end{bmatrix}\) and state its eigenvalue.
• What repeated behavior does eigenvalue 1/2 predict along its eigenvector?
• State one condition that guarantees diagonalization.

Solution sketch: verify, interpret, then state the theorem

Keep the definition \(A\mathbf{v}=\lambda\mathbf{v}\) visible in the answer.

• \(A(1,0)=(3,0)=3(1,0)\) , so \((1,0)\) is an eigenvector with eigenvalue 3.
• Eigenvalue 1/2 means that mode shrinks by a factor of 1/2 each step, so it decays under iteration.
• A matrix is diagonalizable when it has enough linearly independent eigenvectors to form a basis.

Diagonalization does not create a new system

It rewrites the same map in a better basis.

• P changes coordinates, not the underlying transformation.
• D is simpler because the modes stop mixing.
• The payoff is cleaner analysis of repeated behavior.

Day 14: Least Squares, Singular Directions, and Final Synthesis

Day 14: Least Squares, Singular Directions, and Final Synthesis

The course ends with best-fit modeling, low-rank summary, and a final synthesis that uses the whole chain from intuition to formal analysis.

Story Anchor: The course ends with best-fit approximation, dominant singular directions, and one final synthesis of the main ideas.

Focus Question: How do least squares and singular directions summarize high-dimensional data?

Daily Synthesis: Day 14 closes the course by connecting least squares, residuals, and singular directions into one final synthesis.

Sessions

Use least squares when the data is noisy and exact fit is impossible.

Hook: A simplified model cannot hit every noisy data point exactly, so the learner needs the best fit instead of a fake exact solve.

Mastery Signals:

• can explain why best fit replaces exact solve in noisy settings
• can connect least squares to projection and residual error

Study Guide

Least squares replaces exact solving when the data is noisy

Real data often does not land exactly on the model. Least squares asks for the model-supported point that comes closest instead of pretending an exact solution exists.

Minimize the squared residual norm

Least squares chooses x minimizing ||\(A\mathbf{x}\) - b||^2. This turns impossible exact matching into an optimization problem whose answer is the best fit within the model.

An overdetermined system chooses the average compromise

For \(A=\begin{bmatrix}1 \\ 1\end{bmatrix}\) and \(\mathbf{b}=(1,2)\) , no single x can make both equations exact. Least squares picks \(x=1.5\) because it minimizes \((x-1)^2+(x-2)^2\) . The best fit is honest about the remaining residual.

Least squares replaces inconsistent solving

When \(A\mathbf{x}=\mathbf{b}\) has no exact solution, exam problems often ask for the least-squares solution via the normal equations \(A^{\mathsf T}A\mathbf{x}=A^{\mathsf T}\mathbf{b}\) .

• Least squares minimizes the squared residual norm.
• The residual is generally nonzero but orthogonal to the column space.
• An exact solve happens only when b already lies in the column space.

Derive the normal equations from orthogonality

At a least-squares solution x*, the residual \(\mathbf{r}=\mathbf{b}-A\mathbf{x}^{*}\) must be orthogonal to every column of A. Writing those orthogonality conditions together gives \(A^{\mathsf T}\mathbf{r}=\mathbf{0}\) , which becomes \(A^{\mathsf T}A\mathbf{x}^{*}=A^{\mathsf T}\mathbf{b}\) .

• Start from residual orthogonal to the column space.
• Translate orthogonality into dot-product equations with the columns of A.
• Collect those conditions into the matrix equation \(A^{\mathsf T}A\mathbf{x}=A^{\mathsf T}\mathbf{b}\) .

Best fit is not a fake exact solution

The residual still matters and should be interpreted honestly.

• A nonzero residual means the model does not explain everything.
• Least squares is about closeness, not perfect recovery.
• Projection and residual language remain central.

Interpret a leading singular direction as the strongest shared mode in the data.

Hook: The learner now compresses many coordinates into one dominant mode that still preserves the most important structure.

Mastery Signals:

• can explain a leading singular direction as a shared pattern
• can connect compression to preserved dominant structure

Study Guide

A leading singular direction captures the strongest shared mode

When many coordinates move together, the leading singular direction extracts that dominant pattern. It is a principled compression of the data’s strongest structure.

Singular directions identify dominant input-output patterns

In singular value language, the leading direction is the one most strongly stretched by the map or best supported by the data matrix. It captures the principal mode before weaker variation is considered.

Extract one dominant trend from noisy data

If several observed vectors look like \((1,2.1)\) , \((2,4.0)\) , and \((3,5.9)\) , they all lie close to the same line through the origin. The leading singular direction points near \((1,2)\) , because that shared mode explains most of the variation.

Singular directions find the strongest mode

A test may ask which direction is stretched most or what one-direction summary best captures the data. Singular values rank those modes.

• The leading singular direction captures the strongest shared action.
• It is not the same as the largest single coordinate.
• Compression keeps dominant structure while discarding weaker variation.

A singular direction is not just the biggest coordinate

It is a whole-pattern summary, not one dominant entry.

• Read coordinated structure across all coordinates.
• Compression should preserve the strongest mode.
• One large number alone does not define the principal direction.

Choose the final synthesis statement that integrates intuition, formal structure, best fit, and compression into one mathematical story.

Hook: The learner now closes the full course by choosing the synthesis statement that best uses the entire chain from states and maps to fit and modes.

Mastery Signals:

• can integrate the intuition and formal halves of the course
• can choose the synthesis statement that keeps feasibility, fit quality, dynamics, and compression in one argument

Study Guide

The final synthesis should keep the whole linear algebra chain intact

A complete synthesis begins with interpretable states and maps, applies the right formal tool, respects feasibility and hidden structure, then compresses the result into one clear conclusion.

The course now connects states, maps, subspaces, fit, and modes

The final argument should show that these were never isolated topics. Least squares, singular directions, eigen ideas, rank, null space, and elimination are all ways of reading and controlling linear structure.

Combine fit and compression in one closing argument

Start with noisy outputs b and a model \(A\mathbf{x}\) ≈ b. Use least squares to find the best-fit x, read the residual as the part the model misses, and then summarize the data by its leading singular direction. One coherent synthesis now says what the model explains well, what remains as error, and which dominant mode is worth carrying forward.

Final exam habit: choose the right tool

Strong exam answers start by naming the linear-algebra object and then choosing the right tool: elimination, subspace and rank logic, orthogonal projection, determinant, eigen analysis, or least squares.

• Use elimination to solve and classify systems.
• Use rank, nullity, and subspaces for dimension questions.
• Use eigenvalues for dynamics and least squares for noisy fit.

Review set: full-course revision prompts

Use these as a compact undergraduate-style revision bank for the formal half.

• Row reduce a small system and classify its solutions.
• Test a set for subspace status and compute rank or nullity from pivots.
• Project a vector onto a line and explain the orthogonal residual.
• Interpret determinant, eigenvalue behavior, and least-squares fit in one sentence each.

Checkpoint problems: full-course mixed practice

Use this as a compact final review set across the formal half.

• Classify the system represented by a reduced form with one free variable.
• Find the least-squares fit for \(A=\begin{bmatrix}1 \\ 1\end{bmatrix}\) and \(\mathbf{b}=(1,2)\) .
• Choose the right tool for each task: solvability, dimension counting, orthogonal fit, or repeated dynamics.

Solution sketch: pick the object, then the method

Strong final-review answers name the structure before doing the work.

• One free variable in a consistent reduced system means infinitely many solutions.
• The least-squares fit is \(x=1.5\) because it minimizes \((x-1)^2+(x-2)^2\) , equivalently solves \(A^{\mathsf T}A\mathbf{x}=A^{\mathsf T}\mathbf{b}\) .
• Use elimination for solvability, rank-nullity for dimension, projection or least squares for fit, and eigen analysis for repeated-update behavior.

Do not use advanced terms without a coherent argument

The best capstone synthesis is not the one with the most jargon; it is the one that keeps the mathematical chain logically connected.

• Name the model and the state clearly.
• Choose the right formal tool for the situation.
• End with a conclusion that follows from the linear algebra.

Concept Forest

Forest ID: decision-math-forest

Concepts

Vector as State

Summary: A vector can represent one meaningful snapshot of a system instead of a list with no story.

Story Anchor: Regional demand, inventory, or supply status at one moment.

Mastery Hint: Can explain each coordinate as part of one system view.

Coordinate Meaning

Summary: Each coordinate carries a role, unit, and location in the scenario.

Story Anchor: A coordinate only helps if it still points to a region and quantity.

Mastery Hint: Can recover what each slot means without guessing.

Direction of Change

Summary: Direction describes which way the system moves when a policy or shock changes the state.

Story Anchor: Some regions rise while another falls after a policy move.

Mastery Hint: Can compare two change vectors qualitatively.

Magnitude of Change

Summary: Magnitude tracks how large a system move is, apart from its direction.

Story Anchor: A small operational adjustment versus a severe shock.

Mastery Hint: Can separate size from direction in an explanation.

Matrix as Map

Summary: A matrix can act like a reusable rule that transforms one system state into another.

Story Anchor: A routing table sends an input shock through the network and creates a new output state.

Mastery Hint: Can explain a matrix as a system map, not just a grid of numbers.

Row Meaning

Summary: Rows tell you how one source or rule distributes influence across outputs.

Story Anchor: One warehouse or policy channel pushes influence into several destinations.

Mastery Hint: Can explain what one row does in the map.

Column Meaning

Summary: Columns show how each destination receives contributions from the mapped inputs.

Story Anchor: A store or region receives flow from several upstream sources.

Mastery Hint: Can explain what one column collects.

Output as New State

Summary: After a matrix acts on an input vector, the result is another interpretable system state.

Story Anchor: A shock vector passes through the routing map and becomes a new operational snapshot.

Mastery Hint: Can read the output vector as a new story about the system.

Constraint Balance

Summary: Equations can express what must stay balanced in the system before a plan counts as feasible.

Story Anchor: Supply, demand, and resource limits must reconcile.

Mastery Hint: Can describe what conditions the system must satisfy.

Feasible State

Summary: A feasible state satisfies the balance conditions and resource limits instead of only sounding desirable.

Story Anchor: An allocation plan only counts if the network can actually support it.

Mastery Hint: Can distinguish a feasible plan from a preferred but impossible one.

Span as Reachability

Summary: Span tells you which system states can be reached by combining the levers you actually control.

Story Anchor: Different policy levers can be mixed to reach some states, but not every state.

Mastery Hint: Can explain reachability as combination, not as a magic jump.

Basis as Minimal Drivers

Summary: A basis is the smallest dependable set of drivers needed to build the reachable states without redundancy.

Story Anchor: Too many levers can be redundant, while too few cannot cover the needed operating states.

Mastery Hint: Can explain why a basis is both sufficient and minimal.

Rank as Reach Capacity

Summary: Rank captures how much independent reach the system has when it tries to create new operating states.

Story Anchor: A low-rank system cannot independently steer every region it cares about.

Mastery Hint: Can relate rank to how many independent directions remain available.

Repeated Update

Summary: A repeated update shows what happens when the same rule keeps acting on the state again and again.

Story Anchor: A daily routing or policy rule can keep pushing the network long after the first shock.

Mastery Hint: Can compare one-step behavior to multi-step behavior.

Stability Pattern

Summary: Some repeated updates settle into a stable pattern, while others amplify imbalance or drift.

Story Anchor: A network can calm a disruption or keep magnifying it depending on the update rule.

Mastery Hint: Can explain why repeated action may settle or amplify.

Dominant Factor

Summary: A dominant factor is the big underlying pattern that explains much of the system movement without following every coordinate separately.

Story Anchor: Holiday demand and weather pressure can each create a broad pattern across all regions.

Mastery Hint: Can explain a pattern as one broad factor instead of only isolated numbers.

Compression Summary

Summary: Compression keeps the important pattern while throwing away enough detail to make the system easier to reason about.

Story Anchor: A planner may need a short summary of the biggest drivers instead of every noisy detail.

Mastery Hint: Can explain why a compressed summary still supports a decision.

Elimination as Structured Reduction

Summary: Gaussian elimination simplifies a coupled system while preserving the underlying solution set.

Story Anchor: Untangle several linked operating constraints without changing what counts as a valid plan.

Mastery Hint: Can explain why row operations simplify a system without changing the real answers.

Inverse as Undoing Map

Summary: An inverse map recovers the original inputs when the transformation is one-to-one and onto.

Story Anchor: Recover the original driver mix from the observed outcome when the rule is reversible.

Mastery Hint: Can explain why inverses only exist when the map never collapses distinct inputs.

Solution Set Geometry

Summary: A linear system may have one solution, none, or many depending on how the constraints intersect.

Story Anchor: Several policy requirements can agree perfectly, conflict, or leave slack directions free.

Mastery Hint: Can connect elimination outcomes to the geometry of the solution set.

Subspace Closure

Summary: A subspace keeps the structural rules of linear combination: add two allowed directions or scale one, and you stay inside.

Story Anchor: A family of feasible adjustments is only a real linear space if it survives mixing and rescaling.

Mastery Hint: Can test closure instead of trusting a few examples.

Column Space as Achievable Output

Summary: The column space is the set of outputs the map can actually produce.

Story Anchor: Some target states are reachable because they lie inside the outputs supported by the system map.

Mastery Hint: Can explain column space as all achievable outcomes, not just a list of columns.

Null Space as Hidden Direction

Summary: The null space contains input changes that the map cannot see because they leave the output unchanged.

Story Anchor: Two different driver mixes can create the same visible outcome when their difference lies in the null space.

Mastery Hint: Can describe null space as invisible variation, not as nothingness.

Basis Coordinate System

Summary: Coordinates depend on the basis you choose, even when the underlying vector stays the same.

Story Anchor: The same operating state can look different when you describe it in a lever basis versus a region basis.

Mastery Hint: Can separate the vector itself from the coordinates used to describe it.

Rank-Nullity Link

Summary: Input dimension splits into visible capacity and hidden freedom: rank plus nullity.

Story Anchor: Every direction is either visible in the outputs or hidden in the system’s blind spots.

Mastery Hint: Can explain why lost output capacity shows up as extra hidden input freedom.

Dot Product as Alignment

Summary: The dot product measures how strongly two directions reinforce one another.

Story Anchor: Compare whether a response vector actually points with the demand pressure or drifts away from it.

Mastery Hint: Can read sign and size of the dot product as alignment rather than as a mysterious formula.

Orthogonality as Separation

Summary: Orthogonal directions do not overlap in the sense relevant to the chosen inner product.

Story Anchor: A residual direction can stay separate from the fitted signal instead of smuggling the same error back in.

Mastery Hint: Can explain why orthogonal leftovers represent genuinely new information.

Projection as Best Approximation

Summary: Projection keeps the closest point in a chosen subspace and pushes the leftover into an orthogonal residual.

Story Anchor: Fit a noisy demand snapshot with the best simpler model supported by the business rule.

Mastery Hint: Can explain why projection is the best fit inside a limited model class.

Determinant as Volume Scale

Summary: The determinant tracks signed volume scaling under a linear map.

Story Anchor: A map can expand capacity, compress it, or collapse it entirely.

Mastery Hint: Can explain why determinant zero means a collapse of independent directions.

Orientation Flip

Summary: The sign of the determinant records whether a map preserves or reverses orientation.

Story Anchor: A transformation may preserve handedness or mirror the system across a new frame.

Mastery Hint: Can connect determinant sign to orientation rather than only size.

Invertibility Signal

Summary: A nonzero determinant signals that the map stays reversible because no dimension fully collapses.

Story Anchor: If the map keeps enough independent directions, the original drivers can still be recovered.

Mastery Hint: Can connect nonzero determinant to one-to-one solvability.

Eigenvector as Steady Direction

Summary: An eigenvector keeps its direction under the map, changing only by scale.

Story Anchor: Some patterns survive repeated updates without rotating into a different shape.

Mastery Hint: Can identify the idea of a steady direction without confusing it with a fixed point.

Eigenvalue as Scaling Factor

Summary: The eigenvalue tells how a map stretches, shrinks, or flips its eigenvector direction.

Story Anchor: A stable mode may decay while another amplifies under the same repeated rule.

Mastery Hint: Can explain what the scaling factor means for long-run behavior.

Diagonalization as Decoupling

Summary: Diagonalization rewrites a map in a basis where the important modes evolve independently.

Story Anchor: A tangled update rule becomes easier to forecast once the dominant modes are separated.

Mastery Hint: Can explain why a good basis can turn a coupled problem into independent scalings.

Least Squares Fit

Summary: Least squares chooses the model fit with the smallest squared residual when exact agreement is impossible.

Story Anchor: A noisy demand pattern rarely lands exactly in the model subspace, so the goal becomes the closest useful fit.

Mastery Hint: Can explain why best fit replaces exact solution in overdetermined settings.

Residual as Orthogonal Error

Summary: The least-squares residual is orthogonal to the fitted subspace, which is why the fit is optimal.

Story Anchor: The leftover error points outside the model you decided to keep.

Mastery Hint: Can connect orthogonal residuals to why no better fit exists inside the chosen subspace.

Singular Direction

Summary: A singular direction is one of the principal ways a map stretches information or data.

Story Anchor: The strongest shared demand mode explains more variation than the smaller leftover ripples.

Mastery Hint: Can explain why the top singular directions are the best summary axes.

Derivative as Change Rate

Summary: A future calculus course can connect vector change stories to local rates and marginal movement.

Story Anchor: How quickly the system responds when one input moves slightly.

Mastery Hint: Can connect direction of change to local response.

Marginal Response

Summary: Economics and OR learners later connect vector changes to marginal cost, demand, and payoff shifts.

Story Anchor: A one-unit change creates a measurable downstream effect.

Mastery Hint: Can interpret small changes as decision tradeoffs.

Concept Links

• vector-as-state \(\to\) coordinate-meaning (unpacks-into): State vectors only stay interpretable when each coordinate keeps a clear meaning.
• vector-as-state \(\to\) vector-direction (extends-to): Once the state has meaning, the learner can compare how different moves change it.
• vector-as-state \(\to\) matrix-as-map (prepares): Once a learner can read one state vector, the next step is understanding how a reusable map transforms it.
• matrix-as-map \(\to\) row-meaning (unpacks-into): Learners need to read each row as one source of influence or routing behavior.
• matrix-as-map \(\to\) column-meaning (unpacks-into): Columns make it clear how each destination receives flow from the mapped inputs.
• matrix-as-map \(\to\) matrix-output-as-new-state (produces): The output of the matrix map should still be read as a meaningful system state.
• matrix-output-as-new-state \(\to\) constraint-balance (prepares): Once outputs are interpretable, the learner can ask which ones satisfy the real balance rules.
• constraint-balance \(\to\) feasible-state (produces): Balance rules determine whether a proposed state is actually feasible.
• feasible-state \(\to\) span-as-reachability (prepares): After defining feasible states, the learner can ask which of those states are reachable from the available levers.
• span-as-reachability \(\to\) basis-as-minimal-set (unpacks-into): Reachability depends on a minimal set of independent drivers rather than redundant levers.
• basis-as-minimal-set \(\to\) rank-as-reach-capacity (quantifies): Rank measures how much independent reach the basis actually gives the system.
• matrix-as-map \(\to\) elimination-as-structured-reduction (formalizes): The formal study of linear systems makes the map solvable through structured reduction.
• elimination-as-structured-reduction \(\to\) inverse-as-undoing-map (prepares): Once the learner can reduce a system, it becomes clear when the map can be undone.
• elimination-as-structured-reduction \(\to\) solution-set-geometry (reveals): Reduction exposes whether the underlying constraints have one solution, none, or many.
• span-as-reachability \(\to\) subspace-closure (formalizes): Reachable directions become a true linear object when the learner understands closure.
• subspace-closure \(\to\) column-space-meaning (specializes): The set of all outputs is one of the key subspaces created by a matrix.
• subspace-closure \(\to\) null-space-meaning (specializes): The directions the map erases form another essential subspace.
• basis-as-minimal-set \(\to\) basis-coordinate-system (supports): Coordinates only make sense relative to a basis, and different bases produce different coordinate descriptions.
• rank-as-reach-capacity \(\to\) rank-nullity-link (balances-with): Visible output capacity and hidden null directions must add up to the full input dimension.
• vector-magnitude \(\to\) dot-product-alignment (supports): Magnitude and direction combine in the dot product’s measurement of alignment.
• dot-product-alignment \(\to\) orthogonality-as-separation (zero-case): Orthogonality is the special case where the alignment score drops to zero.
• orthogonality-as-separation \(\to\) projection-as-best-approximation (powers): Projection works because the leftover error is pushed into an orthogonal direction.
• projection-as-best-approximation \(\to\) least-squares-fit (formalizes): Least squares turns geometric projection into a full best-fit method.
• least-squares-fit \(\to\) residual-as-orthogonal-error (produces): The best fit leaves behind an orthogonal residual that cannot be improved away inside the model space.
• inverse-as-undoing-map \(\to\) invertibility-signal (equivalent-to): A reversible map is exactly the kind of map with a nonzero determinant.
• determinant-as-volume-scale \(\to\) orientation-flip (signed-by): The determinant’s sign records whether orientation is preserved or reversed.
• determinant-as-volume-scale \(\to\) invertibility-signal (tests): A zero determinant means independent directions have collapsed and the map is no longer invertible.
• repeated-update \(\to\) eigenvector-as-steady-direction (formalizes): Steady directions explain the patterns that survive repeated application of the same map.
• stability-pattern \(\to\) eigenvalue-as-scaling-factor (quantifies): Growth or decay under repeated updates is governed by the relevant scaling factors.
• eigenvector-as-steady-direction \(\to\) diagonalization-as-decoupling (enables): Enough steady directions let the learner rewrite the system in a basis of independent modes.
• dominant-factor \(\to\) singular-direction (formalizes): The idea of one dominant factor becomes the formal idea of a leading singular direction.
• singular-direction \(\to\) compression-summary (supports): Top singular directions provide the best low-dimensional summary of the system.
• compression-summary \(\to\) marginal-response (prepares): Good summaries later feed into marginal reasoning in optimization and economics.
• vector-direction \(\to\) derivative-as-change-rate (prepares): Change directions later connect naturally to local rates in calculus.

Interaction Appendix

Eigen Mode Explorer

Description: Presents a map already written in an eigen-basis so the learner can see each mode scale independently.

• mode_scaled
• steady_direction_seen
• diagonal_basis_named

Row Labels:

• Mode 1
• Mode 2
• Mode 3

Column Labels:

• Mode 1 next
• Mode 2 next
• Mode 3 next

Matrix:

• steady-mode
• mixed-modes
• dominant-mode

Scenarios

Steady Mode

Input Vector:

• 4
• 0
• 0

Output Vector:

• 4.8
• 0
• 0

Narrative: One pure mode keeps its direction and only changes scale, which is the hallmark of an eigenvector.

Insight: Eigenvectors do not rotate under the map; they only get stretched, shrunk, or flipped.

Mixed Modes

Input Vector:

• 4
• 3
• 2

Output Vector:

• 4.8
• 2.4
• 1

Narrative: Once the system is expressed in an eigen-basis, each component evolves on its own scale.

Insight: Diagonalization decouples the behavior into independent modes.

Dominant Mode

Input Vector:

• 5
• 1
• 1

Output Vector:

• 6
• 0.8
• 0.5

Narrative: The first mode grows while the others decay, so long-run behavior becomes dominated by one direction.

Insight: The largest-magnitude eigenvalue controls which mode eventually matters most.

Formal Map Explorer

Description: Shows a reversible linear operator acting on several input vectors so elimination and inversion stay visible.

• operator_compared
• driver_mix_read
• recovered_state_explained

Row Labels:

• Input 1
• Input 2
• Input 3

Column Labels:

• Output 1
• Output 2
• Output 3

Matrix:

• coupled-system-a
• coupled-system-b
• recovered-driver-mix

Scenarios

Coupled System A

Input Vector:

• 2
• 1
• 0

Output Vector:

• 2
• 3
• 1

Narrative: A small input vector produces three linked outputs. Elimination simplifies the equations without changing the solution.

Insight: Row operations preserve the solution set while untangling the structure.

Coupled System B

Input Vector:

• 1
• 2
• 1

Output Vector:

• 2
• 3
• 3

Narrative: A second input makes the coupling clearer: each variable contributes to more than one equation.

Insight: A linear system is coupled precisely because one variable influences several equations together.

Recovered Driver Mix

Input Vector:

• 3
• 1
• 2

Output Vector:

• 5
• 4
• 3

Narrative: When the operator is invertible, the output can be traced back to one unique input vector.

Insight: An inverse map recovers the original input from the output.

Matrix Map Explorer

Description: Autoplays how a matrix maps an input vector into an output vector.

• scenario_changed
• matrix_read
• output_compared

Row Labels:

• Input 1
• Input 2
• Input 3

Column Labels:

• Output 1
• Output 2
• Output 3

Matrix:

• storm-forward
• targeted-reroute
• capacity-shift
• repeat-step-1
• repeat-step-2
• repeat-step-3
• amplified-loop

Scenarios

Example A

Input Vector:

• 47
• 26
• 19

Output Vector:

• 40
• 29
• 23

Narrative: An input vector is multiplied by the matrix to produce a new output vector.

Insight: The matrix is the rule that turns one vector into another.

Example B

Input Vector:

• 45
• 41
• 24

Output Vector:

• 38
• 35
• 37

Narrative: A second input makes it easier to compare row behavior with column behavior.

Insight: Rows describe how input coordinates contribute; columns describe how output coordinates collect those contributions.

Example C

Input Vector:

• 50
• 39
• 31

Output Vector:

• 43
• 37
• 40

Narrative: The matrix changes the output while the input remains interpretable.

Insight: The output of a matrix map is still a vector that needs interpretation.

Iteration 1

Input Vector:

• 40
• 29
• 23

Output Vector:

• 37
• 30
• 24

Narrative: The first repeated application begins to change the vector toward a new pattern.

Insight: One multiplication is only the start of a repeated-update process.

Iteration 2

Input Vector:

• 37
• 30
• 24

Output Vector:

• 35
• 30
• 26

Narrative: A second pass reveals whether the vector is moving toward a steadier pattern.

Insight: Repeated application can reveal stability.

Iteration 3

Input Vector:

• 35
• 30
• 26

Output Vector:

• 34
• 30
• 27

Narrative: By the third pass the change is smaller, suggesting convergence.

Insight: Smaller changes from one step to the next can signal convergence.

Growth Example

Input Vector:

• 52
• 44
• 30

Output Vector:

• 50
• 46
• 36

Narrative: A different repeated rule grows instead of settling.

Insight: Not every repeated rule stabilizes; some amplify.

Projection Map Explorer

Description: Shows how a projection keeps the closest vector in a model subspace and pushes the leftover into a residual.

• projection_compared
• fit_chosen
• residual_named

Row Labels:

• Input 1
• Input 2
• Input 3

Column Labels:

• Output 1
• Output 2
• Output 3

Matrix:

• noisy-fit
• balanced-fit
• already-on-line

Scenarios

Noisy Fit

Input Vector:

• 6
• 2
• 3

Output Vector:

• 4
• 4
• 3

Narrative: The fit keeps the component inside the model subspace and drops the part the model cannot represent.

Insight: Projection chooses the closest point in the model subspace.

Balanced Fit

Input Vector:

• 5
• 3
• 4

Output Vector:

• 4
• 4
• 4

Narrative: The model averages the first two coordinates and carries the third forward, leaving a smaller residual.

Insight: Least squares improves fit by minimizing the squared leftover error.

Already on Line

Input Vector:

• 4
• 4
• 2

Output Vector:

• 4
• 4
• 2

Narrative: When the input already lies inside the model subspace, projection leaves it unchanged.

Insight: A point already in the subspace is its own best approximation.

Singular Map Explorer

Description: Contrasts a singular map where different inputs produce the same output, making column space and null space visible.

• collapsed_output_seen
• duplicate_output_compared
• hidden_direction_named

Row Labels:

• Input 1
• Input 2
• Input 3

Column Labels:

• Output 1
• Output 2
• Output 3

Matrix:

• same-output-a
• same-output-b
• collapsed-output

Scenarios

Same Output A

Input Vector:

• 2
• 0
• 0

Output Vector:

• 2
• 2
• 0

Narrative: One input vector lands at a visible output in the column space.

Insight: The column space is the set of outputs the map can actually reach.

Same Output B

Input Vector:

• 0
• 1
• 0

Output Vector:

• 2
• 2
• 0

Narrative: A different input vector lands at the exact same output, revealing that the map has hidden directions.

Insight: When different inputs produce the same output, their difference points into the null space.

Collapsed Output

Input Vector:

• 1
• 1
• 1

Output Vector:

• 3
• 4
• 1

Narrative: The map still produces outputs, but one independent direction has collapsed, so invertibility is lost.

Insight: A singular map loses dimension: determinant zero, reduced rank, and nontrivial null space all arrive together.

Vector Explorer

Description: Cycles through preset vectors so the learner can compare coordinates, direction, and changes from a reference vector.

• dimension_changed
• state_compared
• vector_delta_explained

Region Labels:

• \(x_1\)
• \(x_2\)
• \(x_3\)

Baseline State:

• 42

• 35

• 28

• baseline

• storm-shock

• policy-boost

• double-boost

• reroute-response

• balanced-plan

• reachable-plan

• factor-weather

• factor-holiday

• compressed-summary

• aligned-response

• misaligned-response

• orthogonal-residual

• best-fit-signal

• principal-mode

Scenarios

Reference Vector

Values:

• 42
• 35
• 28

Narrative: Start from one reference vector with three labeled coordinates.

Insight: Each coordinate keeps a fixed position and meaning.

Shifted Vector

Values:

• 47
• 26
• 19

Narrative: The first coordinate increases while the second and third decrease.

Insight: Changing coordinates changes the vector itself.

Positive Shift

Values:

• 50
• 39
• 31

Narrative: All three coordinates increase relative to the reference vector.

Insight: Direction describes the overall pattern of change.

Larger Same Pattern

Values:

• 58
• 43
• 34

Narrative: The change is larger but keeps the same coordinate pattern as the positive shift.

Insight: A larger scalar multiple keeps direction while increasing magnitude.

Comparison Vector

Values:

• 45
• 41
• 24

Narrative: The second coordinate grows most while the third remains comparatively smaller.

Insight: Vectors can be compared coordinate by coordinate.

Nearby Vector

Values:

• 44
• 36
• 27

Narrative: This vector stays close to the reference while changing all three coordinates slightly.

Insight: Vectors can be close coordinate-wise without being identical.

Linear Combination Example

Values:

• 46
• 39
• 30

Narrative: This example can be read as one vector formed from a combination of simpler directions.

Insight: Later in the course, span explains which vectors can be built from a set of generators.

Pattern 1

Values:

• 49
• 33
• 24

Narrative: This vector follows one broad coordinate pattern.

Insight: A dominant factor captures one shared pattern across several coordinates.

Pattern 2

Values:

• 55
• 43
• 34

Narrative: All three coordinates increase, but not by the same amount.

Insight: Some vectors are easier to describe through a shared pattern than by listing every coordinate.

Compressed Vector

Values:

• 52
• 40
• 31

Narrative: This vector keeps the main pattern while smoothing smaller coordinate differences.

Insight: Compression keeps the dominant structure while discarding some detail.

Aligned Vector

Values:

• 48
• 38
• 30

Narrative: This vector points in nearly the same direction as the target pattern.

Insight: A large positive dot product signals strong alignment.

Misaligned Vector

Values:

• 34
• 46
• 31

Narrative: This vector points in a noticeably different direction from the target pattern.

Insight: Weak alignment means the vectors do not point the same way.

Residual Vector

Values:

• 42
• 39
• 24

Narrative: After fitting one direction, this leftover vector records what the fit does not capture.

Insight: An orthogonal residual carries different information from the fitted direction.

Projected Vector

Values:

• 44
• 44
• 28

Narrative: The projection keeps the component that lies inside the chosen model subspace.

Insight: Projection preserves the supported part and leaves the residual outside it.

Dominant Direction

Values:

• 58
• 46
• 34

Narrative: One dominant pattern explains most of the variation in the coordinates.

Insight: A leading singular direction is the strongest shared pattern in the data.

Exercise Appendix

In one short sentence, explain what a balance equation means in this linear system

Prompt: In one short sentence, explain what a balance equation means in this linear system.

Correct Feedback: Good. You treated the equation as a real balance rule that a valid vector must satisfy.

Retry Feedback: Not yet. Keep the equation tied to exact balance, not a loose preference.

Which synthesis statement respects the balance logic of Day 4?

Prompt: Which synthesis statement respects the balance logic of Day 4?

Correct Feedback: Good. You preserved the idea that feasibility comes before ranking options.

Retry Feedback: Choose the synthesis statement that uses balance as a gate before preference.

Choices

• Correct: Start with the candidate vector that satisfies the equations, then compare options inside the feasible set.
  • Feedback: Correct. Balance is the first gate, not an optional detail.
• Not correct: Ignore the balance equations because the most urgent coordinate should always get whatever value you want.
  • Feedback: Urgency matters, but a vector that breaks the balance rules is not a real solution.
• Not correct: Pick the statement with the largest gains on paper even if the equations cannot all be satisfied.
  • Feedback: That treats feasibility as optional. The synthesis has to respect what the system can support.

What changes when you describe a vector in a different basis?

Prompt: What changes when you describe a vector in a different basis?

Correct Feedback: Good. You separated the object from its coordinate description.

Retry Feedback: Choose the option that keeps the vector fixed and the coordinates relative.

Choices

• Correct: The coordinates change, but the underlying vector does not.
  • Feedback: Correct. Basis choice changes the description, not the state itself.
• Not correct: The vector changes into a different state because the numbers are different.
  • Feedback: Different coordinates can still describe the same vector in a new basis.
• Not correct: Nothing changes, because a basis is only a naming preference.
  • Feedback: A basis is not cosmetic. It determines how the vector is encoded.

A plane in \(\mathbb{R}^{3}\) needs how many independent vectors for a basis? Enter the minimum number

Prompt: A plane in \(\mathbb{R}^{3}\) needs how many independent vectors for a basis? Enter the minimum number.

Accepted Responses:

• 2

Correct Feedback: Good. A plane is two-dimensional, so it needs exactly two independent basis vectors.

Retry Feedback: Not yet. Count the number of independent directions a plane really needs.

Which final synthesis statement best uses the full bachelor-level course logic?

Prompt: Which final synthesis statement best uses the full bachelor-level course logic?

Correct Feedback: Good. You chose the synthesis that integrates the whole course from intuitive state reading through formal analysis and compression.

Retry Feedback: Choose the synthesis that keeps intuition, formal tools, and constraints in the same chain.

Choices

• Correct: Read the system as states and maps, solve or fit the formal model that matches the situation, keep feasibility and rank limits in view, then summarize the dominant modes clearly enough to support one coherent conclusion.
  • Feedback: Correct. This synthesis ties intuition, formal structure, and conclusion into one argument.
• Not correct: Use whichever advanced tool sounds most impressive and skip the earlier intuition because formal symbols should replace the story once the course becomes serious.
  • Feedback: The course is built around connecting the formal tools back to the original story, not discarding it.
• Not correct: Keep only the compression result and ignore solvability, projection error, determinant, and eigen behavior because the biggest factor always decides everything alone.
  • Feedback: The strongest factor matters, but a real synthesis still has to respect feasibility, fit quality, and system dynamics.

Which interpretation is best supported by the output vector?

Prompt: Which interpretation is best supported by the output vector?

Correct Feedback: Good. You used the output vector instead of guessing from the raw input alone.

Retry Feedback: Choose the interpretation that follows from what the matrix output actually shows.

Choices

• Correct: Read the largest output coordinates first, because the output vector shows where the matrix sends the strongest contribution.
  • Feedback: Correct. The output vector should drive the interpretation because it shows what the matrix produced.
• Not correct: Ignore the output and read only the original input, because the matrix does not change what matters.
  • Feedback: The map changes the result. The interpretation should respond to the output, not only the input.
• Not correct: Treat every output coordinate the same, because any matrix output is just a technical artifact.
  • Feedback: That loses the meaning of the output. Different output coordinates can carry different sizes.

Which final synthesis best explains the vector-matrix relation?

Prompt: Which final synthesis best explains the vector-matrix relation?

Correct Feedback: Good. You chose the synthesis that keeps the input, the map, and the output in one explanation.

Retry Feedback: Choose the synthesis that keeps all three layers of evidence in view.

Choices

• Correct: An input vector entered the matrix, the map redistributed that input across output coordinates, and the resulting output vector records the final pattern.
  • Feedback: Correct. This synthesis integrates the input vector, the map, and the resulting output vector into one explanation.
• Not correct: The checkpoint mainly proves that matrices are harder than vectors, so the safest answer is the one with the least interpretation.
  • Feedback: That avoids the real task. The point is to interpret the evidence, not retreat from it.
• Not correct: Because one output coordinate ends higher than the others, the whole result should be described using only that coordinate.
  • Feedback: A single coordinate is not the whole explanation. The synthesis should explain the full output pattern.

In one short sentence, explain how the input vector and the output vector belong in the same story

Prompt: In one short sentence, explain how the input vector and the output vector belong in the same story.

Correct Feedback: Good. You kept the input, the map, and the output in one coherent explanation.

Retry Feedback: Not yet. Keep the starting vector, the map, and the resulting vector in the same sentence.

What is the best interpretation of the column space of a matrix?

Prompt: What is the best interpretation of the column space of a matrix?

Correct Feedback: Good. You treated column space as achievable outputs.

Retry Feedback: Choose the option that describes reachable outputs.

Choices

• Correct: It is the set of outputs the matrix can actually produce from some input.
  • Feedback: Correct. Column space is the reachable output space of the map.
• Not correct: It is the set of input vectors that make the matrix vanish.
  • Feedback: That description belongs to the null space, not the column space.
• Not correct: It is just the literal collection of columns, without any combinations.
  • Feedback: The space is made from their span, not from the uncombined list alone.

In one short sentence, explain what good compression should preserve when you summarize many coordinates

Prompt: In one short sentence, explain what good compression should preserve when you summarize many coordinates.

Correct Feedback: Good. You kept the dominant structure while letting weaker detail drop away.

Retry Feedback: Not yet. Name both what compression keeps and what it is willing to reduce.

A linear map has determinant -3. Enter the signed volume scale factor

Prompt: A linear map has determinant -3. Enter the signed volume scale factor.

Accepted Responses:

• -3

Correct Feedback: Good. A determinant of -3 means volumes are scaled by 3 with an orientation flip, so the signed factor is -3.

Retry Feedback: Not yet. The determinant itself is the signed scale factor.

If the determinant is zero, which conclusion is justified?

Prompt: If the determinant is zero, which conclusion is justified?

Correct Feedback: Good. You connected zero determinant to collapse and lost invertibility.

Retry Feedback: Choose the option that describes dimension loss.

Choices

• Correct: At least one independent direction has collapsed, so the map is not invertible.
  • Feedback: Correct. Zero determinant means lost dimension and no inverse.
• Not correct: The matrix is diagonal, so the system is easier to solve.
  • Feedback: Determinant zero says nothing about being diagonal.
• Not correct: All outputs must be zero for every input.
  • Feedback: A singular matrix can still produce many nonzero outputs.

Why is diagonalization powerful when it is possible?

Prompt: Why is diagonalization powerful when it is possible?

Correct Feedback: Good. You connected diagonalization to decoupled modes.

Retry Feedback: Choose the option that explains why the right basis matters.

Choices

• Correct: It changes to a basis where the important modes evolve independently instead of staying coupled.
  • Feedback: Correct. Diagonalization decouples the dynamics in the right basis.
• Not correct: It erases the original matrix and replaces it with a simpler but unrelated one.
  • Feedback: The diagonal form represents the same transformation in a better basis.
• Not correct: It guarantees every matrix can be solved by inspection.
  • Feedback: Not every matrix is diagonalizable, and even diagonalizable ones still require interpretation.

Enter the dot product of \((2,1)\) and \((1,2)\)

Prompt: Enter the dot product of \((2,1)\) and \((1,2)\) .

Accepted Responses:

• 4

Correct Feedback: Good. The dot product is \(2\cdot1+1\cdot2=4\) , which is positive and signals alignment.

Retry Feedback: Not yet. Multiply matching coordinates and add the results.

Which synthesis statement best explains the dynamics lesson?

Prompt: Which synthesis statement best explains the dynamics lesson?

Correct Feedback: Good. You framed the update rule as a process that can stabilize or amplify the sequence.

Retry Feedback: Choose the synthesis statement that explains the repeated path, not just one output.

Choices

• Correct: Judge the rule by the path it creates over repeated steps: some iterates settle, others keep amplifying.
  • Feedback: Correct. The synthesis explains dynamics as a repeated-process judgment.
• Not correct: Only the first output matters, because later steps are mathematically repetitive and irrelevant.
  • Feedback: That misses the whole point of dynamics. Later steps reveal the behavior of the rule.
• Not correct: Whichever iterate has the biggest numbers is automatically the unstable one.
  • Feedback: Instability is about the repeated pattern, not simply the largest coordinate on one step.

If \(A\mathbf{v}=2\mathbf{v}\) for a nonzero eigenvector v, enter the eigenvalue

Prompt: If \(A\mathbf{v}=2\mathbf{v}\) for a nonzero eigenvector v, enter the eigenvalue.

Accepted Responses:

• 2

Correct Feedback: Good. The scalar multiplying v is the eigenvalue, so \(\lambda=2\) .

Retry Feedback: Not yet. Read the scalar directly from \(A\mathbf{v}=\lambda\mathbf{v}\) .

What makes a vector an eigenvector of a matrix?

Prompt: What makes a vector an eigenvector of a matrix?

Correct Feedback: Good. You centered the idea on preserved direction.

Retry Feedback: Choose the option that keeps scale allowed but direction fixed.

Choices

• Correct: The matrix maps it to a scalar multiple of itself, so the direction stays the same.
  • Feedback: Correct. The defining feature is directional stability under the map.
• Not correct: The matrix leaves it numerically unchanged entry by entry.
  • Feedback: That would only happen for a special eigenvalue of 1.
• Not correct: The vector is orthogonal to every other vector in the space.
  • Feedback: Orthogonality is not the definition of an eigenvector.

Elimination leaves you with \(y=3\) and \(x+2y=7\) . Enter the value of x

Prompt: Elimination leaves you with \(y=3\) and \(x+2y=7\) . Enter the value of x.

Accepted Responses:

• 1

Correct Feedback: Good. Substituting \(y=3\) into \(x+2y=7\) gives \(x+6=7\) , so \(x=1\) .

Retry Feedback: Not yet. Back-substitute the pivot value into the earlier equation.

What is the best reason to describe the pattern as one dominant factor?

Prompt: What is the best reason to describe the pattern as one dominant factor?

Correct Feedback: Good. You treated the factor as a shared driver behind several coordinated movements.

Retry Feedback: Choose the answer that keeps the factor tied to a common system pattern.

Choices

• Correct: Because the same broad shared direction explains several coordinates at once, even though each coordinate still varies a little.
  • Feedback: Correct. A factor captures a shared driver across the system.
• Not correct: Because one coordinate is the largest, so the whole pattern should be renamed after that one number.
  • Feedback: A factor is not just the largest coordinate. It is a shared pattern across coordinates.
• Not correct: Because factor language lets you stop interpreting the coordinates altogether.
  • Feedback: Factor language should help interpretation, not replace it.

Which candidate vector sounds feasible rather than merely attractive?

Prompt: Which candidate vector sounds feasible rather than merely attractive?

Correct Feedback: Good. You filtered for a vector the system can actually support.

Retry Feedback: Choose the vector that stays inside the system’s limits, not just the one that sounds boldest.

Choices

• Correct: The candidate vector that changes all three coordinates while still satisfying the stated equations and limits.
  • Feedback: Correct. Feasibility comes from satisfying the limits and balance conditions.
• Not correct: The candidate vector that fixes one coordinate but breaks the allowed bounds or equations elsewhere.
  • Feedback: That may sound attractive, but it breaks feasibility by exceeding the stated limits.
• Not correct: The candidate vector with the most dramatic change, because bigger changes always show stronger control.
  • Feedback: Bigger is not the same as feasible. The vector still has to satisfy the system.

Which synthesis note best closes Part I of the course?

Prompt: Which synthesis note best closes Part I of the course?

Correct Feedback: Good. You chose the synthesis note that integrates state reading, mapping, balance, reach, dynamics, and compression into one carry-forward chain.

Retry Feedback: Choose the note that combines the Part I logic instead of isolating one topic.

Choices

• Correct: Read the system as states transformed by rules, check feasibility and reach, then carry forward the dominant pattern as the idea the formal tools should sharpen next.
  • Feedback: Correct. This note integrates the major layers of Part I and points naturally into the formal half.
• Not correct: Treat each topic as separate vocabulary and avoid combining them, because formal math should begin only after the intuition is forgotten.
  • Feedback: The course is built as a second pass, not a reset. The formal tools should sharpen the earlier ideas, not replace them.
• Not correct: Ignore balance and reach once you find a strong factor, because the biggest pattern always determines the right move by itself.
  • Feedback: A dominant pattern matters, but it does not cancel feasibility, reach, or dynamics.

Which statement best captures what an inverse matrix does?

Prompt: Which statement best captures what an inverse matrix does?

Correct Feedback: Good. You kept the inverse tied to undoing the full map.

Retry Feedback: Choose the option that connects inverse to recoverability.

Choices

• Correct: It is the map that undoes the original transformation when outputs still determine inputs uniquely.
  • Feedback: Correct. Invertibility is exactly the recoverability of the original input.
• Not correct: It flips every matrix entry into its reciprocal so the grid points backward.
  • Feedback: An inverse is a whole transformation, not entry-by-entry reciprocals.
• Not correct: It exists for any square matrix because every square grid can be reversed.
  • Feedback: Some square matrices collapse directions and are not invertible.

In one short sentence, explain when least squares is the right tool

Prompt: In one short sentence, explain when least squares is the right tool.

Correct Feedback: Good. You used least squares for the best-fit situation where exact solve is no longer honest.

Retry Feedback: Not yet. Say why exact fit fails and what least squares minimizes instead.

What is the best description of what the matrix does?

Prompt: What is the best description of what the matrix does?

Correct Feedback: Good. You treated the matrix as a map rather than a static table.

Retry Feedback: Pick the answer that preserves both the map idea and the output-vector idea.

Choices

• Correct: It acts like a reusable map that turns one input vector into a new output vector.
  • Feedback: Correct. The matrix is a rule for transforming one vector into another.
• Not correct: It is only a storage grid for numbers, so the output no longer needs interpretation.
  • Feedback: That loses the key idea. The output vector still needs interpretation.
• Not correct: It just renames the coordinates without changing the vector.
  • Feedback: Not quite. The matrix changes how input coordinates contribute to the output.

How should you read the output vector after the matrix acts?

Prompt: How should you read the output vector after the matrix acts?

Correct Feedback: Good. You kept the output attached to vector meaning.

Retry Feedback: Pick the answer that still treats the output vector as meaningful.

Choices

• Correct: As a new vector created by the matrix map.
  • Feedback: Correct. The output is still a meaningful vector.
• Not correct: As a list of products that no longer refers to the original coordinate structure.
  • Feedback: That drops the interpretation. The output still describes a structured vector.
• Not correct: As proof that matrices replace vectors once the math gets harder.
  • Feedback: Not quite. The matrix acts on a vector and produces another vector.

Which statement best separates row meaning from column meaning?

Prompt: Which statement best separates row meaning from column meaning?

Correct Feedback: Good. You kept source behavior and destination collection separate.

Retry Feedback: Choose the answer that gives rows and columns different jobs.

Choices

• Correct: A row shows how an output coordinate is formed; a column shows how one input coordinate contributes.
  • Feedback: Correct. That keeps rows and columns tied to distinct parts of the map.
• Not correct: Rows and columns mean the same thing, so you can read either one however you want.
  • Feedback: That erases the structure learners need to read the map correctly.
• Not correct: Only columns matter because outputs are the only numbers that count.
  • Feedback: Rows still matter because they help define the output coordinates.

In one short sentence, explain what one row tells you and what one column tells you

Prompt: In one short sentence, explain what one row tells you and what one column tells you.

Correct Feedback: Good. You kept the output-building job of rows separate from the input-contribution job of columns.

Retry Feedback: Not yet. Keep row meaning and column meaning separate.

Which statement best explains the null space?

Prompt: Which statement best explains the null space?

Correct Feedback: Good. You read null space as invisible input directions.

Retry Feedback: Choose the option that keeps the focus on hidden inputs.

Choices

• Correct: It contains input changes that the map cannot see because they leave the output unchanged.
  • Feedback: Correct. Null space is hidden variation in the input.
• Not correct: It lists the outputs that are impossible for the matrix to produce.
  • Feedback: That is a column-space question, not a null-space one.
• Not correct: It means the matrix sends every vector to zero.
  • Feedback: A nontrivial null space only says some directions vanish, not all of them.

Which statement correctly links determinant sign and invertibility?

Prompt: Which statement correctly links determinant sign and invertibility?

Correct Feedback: Good. You kept sign and nonzero status doing different jobs.

Retry Feedback: Choose the option that separates orientation from reversibility.

Choices

• Correct: A negative determinant means orientation flips, and any nonzero determinant still signals invertibility.
  • Feedback: Correct. Sign tracks orientation; nonzero magnitude preserves reversibility.
• Not correct: A negative determinant means the map is never invertible.
  • Feedback: Negative sign alone does not destroy invertibility. Zero does.
• Not correct: A positive determinant guarantees the matrix is orthogonal.
  • Feedback: Positive sign does not imply orthogonality.

Why is an orthogonal residual important after projection?

Prompt: Why is an orthogonal residual important after projection?

Correct Feedback: Good. You connected orthogonality to optimality.

Retry Feedback: Choose the option that explains why orthogonal error certifies best fit.

Choices

• Correct: It shows the leftover error carries no component along the fitted subspace, which is why the fit is optimal there.
  • Feedback: Correct. Orthogonality is the certificate that no better fit exists inside the subspace.
• Not correct: It means the residual is always zero, so the fit is exact.
  • Feedback: The residual can be nonzero. Orthogonal does not mean vanished.
• Not correct: It means any projected fit would have worked equally well.
  • Feedback: Projection is special because it gives the closest point, not just any point in the subspace.

What is the best interpretation of projecting a noisy signal onto a model subspace?

Prompt: What is the best interpretation of projecting a noisy signal onto a model subspace?

Correct Feedback: Good. You kept projection tied to closest fit.

Retry Feedback: Choose the option that describes a closest-point approximation.

Choices

• Correct: You keep the closest signal the model can express and leave the mismatch in the residual.
  • Feedback: Correct. Projection preserves the best available fit inside the model space.
• Not correct: You arbitrarily delete the coordinates the model does not like.
  • Feedback: Projection is not arbitrary deletion. It is the closest-point rule.
• Not correct: You force the data to become exact by changing the original signal.
  • Feedback: Projection does not make the original data exact. It finds the nearest modeled approximation.

Which statement best distinguishes rank from dimension?

Prompt: Which statement best distinguishes rank from dimension?

Correct Feedback: Good. You kept the questions behind rank and dimension separate.

Retry Feedback: Choose the option that distinguishes space size from visible output capacity.

Choices

• Correct: Dimension counts how many basis directions a space has, while rank counts how many independent output directions a map preserves.
  • Feedback: Correct. The two counts are related, but they answer different questions.
• Not correct: Rank and dimension are always identical because both just count coordinates.
  • Feedback: They can coincide in some cases, but they are not the same concept.
• Not correct: Rank counts rows, while dimension counts columns.
  • Feedback: Raw row and column counts are not the definition of either concept.

A linear map from \(\mathbb{R}^{5}\) has rank 3. Enter the nullity

Prompt: A linear map from \(\mathbb{R}^{5}\) has rank 3. Enter the nullity.

Accepted Responses:

• 2

Correct Feedback: Good. \(\operatorname{rank}+\operatorname{nullity}=\text{input dimension}\) , so \(3+\operatorname{nullity}=5\) gives \(\operatorname{nullity}=2\) .

Retry Feedback: Not yet. Use \(\text{input dimension}=\operatorname{rank}+\operatorname{nullity}\) .

Which conclusion best uses rank to reason about reachability?

Prompt: Which conclusion best uses rank to reason about reachability?

Correct Feedback: Good. You connected rank to how much independent reach the system actually has.

Retry Feedback: Choose the answer that ties rank to independent reachable directions.

Choices

• Correct: If the system lacks enough independent directions, some target states stay unreachable no matter how many redundant levers you add.
  • Feedback: Correct. Rank limits independent reach.
• Not correct: Any target becomes reachable as long as you list enough vectors, even if they repeat the same direction.
  • Feedback: Repeated directions do not increase rank or reach.
• Not correct: Rank only counts matrix size, so it does not affect which states you can build.
  • Feedback: Rank matters because it measures independent reach, not just printed size.

A state is halved at each update. Starting from 8, enter the value after three updates

Prompt: A state is halved at each update. Starting from 8, enter the value after three updates.

Accepted Responses:

• 1

Correct Feedback: Good. Halving three times gives 8 \(\to\) 4 \(\to\) 2 \(\to\) 1, which shows the repeated process directly.

Retry Feedback: Not yet. Apply the same update three times in sequence.

What is the most useful interpretation of a leading singular direction?

Prompt: What is the most useful interpretation of a leading singular direction?

Correct Feedback: Good. You kept the singular direction tied to the strongest shared mode.

Retry Feedback: Choose the option that treats the direction as a pattern summary rather than one coordinate.

Choices

• Correct: It is the strongest shared pattern the map or data stretches, so it captures the most variation or energy.
  • Feedback: Correct. A leading singular direction is a best summary axis, not just a big coordinate.
• Not correct: It is simply the coordinate with the largest raw value in the original data table.
  • Feedback: Singular directions summarize a whole pattern, not one isolated coordinate.
• Not correct: It only matters when the determinant is zero and otherwise can be ignored.
  • Feedback: Singular directions matter broadly for compression and best low-rank summaries.

If elimination reveals a contradictory row, what is the right story?

Prompt: If elimination reveals a contradictory row, what is the right story?

Correct Feedback: Good. You read the contradiction as conflicting conditions.

Retry Feedback: Choose the option that matches incompatible equations.

Choices

• Correct: The constraints conflict, so the system has no solution.
  • Feedback: Correct. A contradiction means no state satisfies all the equations at once.
• Not correct: The system must have infinitely many solutions because one row disappeared.
  • Feedback: A contradictory row signals inconsistency, not extra freedom.
• Not correct: The contradiction means the inverse still exists but was applied in the wrong order.
  • Feedback: Contradiction is about incompatible constraints, not inverse order.

What is the best interpretation of span in this course?

Prompt: What is the best interpretation of span in this course?

Correct Feedback: Good. You kept span tied to reachable combinations of real levers.

Retry Feedback: Choose the answer that treats span as reachable combination, not imagination or one isolated vector.

Choices

• Correct: Span tells you which operating states can be built by combining the levers you actually control.
  • Feedback: Correct. Span is about reachable combinations.
• Not correct: Span is the one best action vector, so you only need to test each lever separately.
  • Feedback: Span is not one action. It is the space created by combining available actions.
• Not correct: Span means any state you can imagine, even if no lever combination can produce it.
  • Feedback: Imagining a state is not enough. Span is limited by the levers you have.

Which observation best signals a stabilizing pattern?

Prompt: Which observation best signals a stabilizing pattern?

Correct Feedback: Good. You looked for the repeated pattern, not just the absolute size of one state.

Retry Feedback: Choose the answer that describes convergence rather than raw magnitude.

Choices

• Correct: The outputs keep changing less from one step to the next and move toward a steadier shape.
  • Feedback: Correct. Shrinking changes are a useful sign of convergence.
• Not correct: One coordinate stays large, so the system must be unstable.
  • Feedback: A large coordinate alone does not prove instability. The step-to-step pattern matters more.
• Not correct: Any movement from one output to the next proves the rule is unstable.
  • Feedback: Movement can still be part of a stabilizing path. What matters is whether the pattern is settling or amplifying.

In one short sentence, explain what closure conditions make a set of vectors a subspace

Prompt: In one short sentence, explain what closure conditions make a set of vectors a subspace.

Correct Feedback: Good. You defined a subspace by closure under the two linear operations.

Retry Feedback: Not yet. A subspace answer should name both closure tests.

Type the shifted vector you just saw by entering \(x_1\) , \(x_2\) , and \(x_3\) in order

Prompt: Type the shifted vector you just saw by entering \(x_1\) , \(x_2\) , and \(x_3\) in order.

Expected Vector:

• 47
• 26
• 19

Coordinate Labels:

• \(x_1\)
• \(x_2\)
• \(x_3\)

Correct Feedback: Good. You rebuilt the shifted vector coordinate by coordinate.

Retry Feedback: Keep the order fixed and rebuild the vector slot by slot.

In [47, 26, 19], what does the second coordinate represent?

Prompt: In [47, 26, 19], what does the second coordinate represent?

Correct Feedback: Good. One coordinate, one label.

Retry Feedback: Pick the answer that keeps one fixed coordinate label.

Choices

• Correct: \(x_2\) has value 26 units.
  • Feedback: Correct. The second coordinate keeps its label and unit.
• Not correct: 26 means this entry is ranked second in size.
  • Feedback: No. Position marks the coordinate label, not a ranking.
• Not correct: 26 is the total of the whole vector.
  • Feedback: No. It refers to one coordinate only.

What changes from the reference vector to the comparison vector?

Prompt: What changes from the reference vector to the comparison vector?

Correct Feedback: Good. You separated pattern from size.

Retry Feedback: Choose the answer that separates pattern from size.

Choices

• Correct: All three coordinates increase, and the overall change is larger than in the earlier shifted vector.
  • Feedback: Correct. Direction is the pattern; magnitude is the size.
• Not correct: Because all coordinates are positive, the vector has no direction at all.
  • Feedback: No. Positive coordinates still have direction.
• Not correct: Only the biggest coordinate matters, so the other coordinates do not affect the direction.
  • Feedback: No. Direction comes from the whole vector.

Which interpretation reads the comparison vector best?

Prompt: Which interpretation reads the comparison vector best?

Correct Feedback: Good. You kept coordinate meaning intact.

Retry Feedback: Pick the interpretation that keeps coordinate meaning intact.

Choices

• memo-a (correct): The second coordinate is largest, the first stays close to the reference size, and the third is still smaller.
  • Feedback: Correct. The interpretation keeps each coordinate tied to its position.
• memo-b (not correct): The second number is just a bigger symbol.
  • Feedback: No. Each number still belongs to a fixed coordinate position.
• memo-c (not correct): One lower coordinate means the whole vector has the same behavior everywhere.
  • Feedback: No. One coordinate does not determine the whole vector.

Repair Appendix

The vector stays the same when coordinates change

Diagnosis Prompt: Which confusion sounds familiar?

Bridge Prompt: Go back and separate the underlying state from the coordinate system used to describe it.

Repair Choices

• I thought new coordinates meant a new vector.
  • Changing basis changes the description, not the underlying state.
• I kept one basis in mind and forgot the coordinates are relative.
  • Coordinates only make sense after you name the basis they belong to.
• I treated the coordinate list as absolute data.
  • A coordinate list is meaningful only inside a chosen frame.

Basis means minimal and sufficient

Diagnosis Prompt: Which statement fits your confusion best?

Bridge Prompt: Return to the check and test both conditions: enough reach, but no redundant driver.

Repair Choices

• I keep treating any useful list of vectors as a basis.
  • A basis is not just useful. It must cover the reachable space and avoid redundancy at the same time.
• I understand coverage, but not why redundancy matters.
  • Redundancy matters because a basis should keep only the independent drivers that really add new reach.
• I understand minimality, but I forget the set still has to reach the targets.
  • Too small is not enough. A basis has to remain sufficient while staying minimal.

Reconnect the coordinate to its label

Diagnosis Prompt: What feels most likely to be slipping?

Bridge Prompt: Return to the question and attach the number to one coordinate label and one unit.

Repair Choices

• I forget which coordinate label the position refers to.
  • The position is not a ranking. It is a label slot. The second coordinate means \(x_2\) because that slot is reserved for \(x_2\) .
• I keep reading one coordinate as the whole system total.
  • One coordinate is only one piece of the system. The full vector is the whole state; each coordinate is one labeled component.
• I see the number, but not the unit.
  • Keep the unit attached every time you read the coordinate. The value is not abstract by itself; it is a measured quantity in one fixed slot.

Determinant is about the map, not the visible entries alone

Diagnosis Prompt: Which trap sounds familiar?

Bridge Prompt: Go back and read the determinant as what the whole transformation does to space: scale, flip, or collapse.

Repair Choices

• I treated determinant like multiplying a few entries together.
  • The determinant summarizes the global effect of the linear map on signed volume.
• I knew zero was important, but not why.
  • Zero means at least one independent direction collapses, so the map cannot be inverted.
• I ignored the sign and only looked at size.
  • The sign tells whether orientation is preserved or reversed.

Diagonalization needs the right basis

Diagnosis Prompt: Which issue sounds closest?

Bridge Prompt: Go back and remember that diagonalization is a change of basis that decouples the modes.

Repair Choices

• I thought every matrix is already diagonal if you stare at it the right way.
  • Diagonalization is not relabeling. It requires enough independent eigenvectors to form a basis.
• I forgot that a new basis changes the coordinates of the state too.
  • The whole description changes together: new basis, new coordinates, simpler action of the map.
• I knew the diagonal form was simpler, but not what became simpler.
  • In the eigen-basis, each mode scales independently instead of mixing with the others.

Separate pattern from size

Diagnosis Prompt: Which confusion is closest to yours?

Bridge Prompt: Go back and name the movement pattern first, then whether that movement is small or large.

Repair Choices

• I keep treating the largest coordinate as the whole direction.
  • Direction comes from the full pattern across coordinates, not just the biggest entry. Look at how all coordinates move together.
• If the coordinates are positive, I assume there is no direction question.
  • Positive coordinates still make a direction. They show a pattern across the vector even before you compare size.
• I mix up where the system moves with how strong the move is.
  • Direction answers where the change points. Magnitude answers how large that change is. Keep those as two separate sentences.

Dot product measures alignment

Diagnosis Prompt: Which issue sounds most likely?

Bridge Prompt: Go back and read the dot product as one alignment score, not as three unrelated coordinate checks.

Repair Choices

• I only compared coordinates one by one.
  • The dot product combines all coordinates into one overall alignment measure.
• I forgot that sign matters as well as size.
  • Positive, zero, and negative values tell very different stories about direction.
• I knew the formula but not what it says about action.
  • A strong positive score means the response points with the target direction rather than wasting effort away from it.

Eigenvectors keep direction, not necessarily value

Diagnosis Prompt: What sounds most likely?

Bridge Prompt: Return to the choice and ask whether the direction stays the same even if the size changes.

Repair Choices

• I thought an eigenvector had to stay exactly unchanged.
  • That would require eigenvalue 1. In general the vector can stretch or shrink while keeping its direction.
• I focused on the number and forgot the direction story.
  • The core idea is directional stability under the map.
• I mixed up repeated stability with immediate fixed points.
  • Eigenvectors identify the modes whose behavior is easy to track over repeated updates.

Treat the equation like a balance rule

Diagnosis Prompt: What sounds most like the gap?

Bridge Prompt: Return to the check and read each side of the equation as a resource promise, not just a symbol string.

Repair Choices

• I still read the equation like a calculation trick.
  • A balance equation is a rule about what must match in the real system. The symbols matter because they preserve that operational promise.
• I lose what each side of the equation stands for.
  • Give each side a job. One side names what enters or is available; the other names what must be satisfied or allocated.
• I can read the numbers, but not why equality matters.
  • Equality matters because imbalance means the plan either overuses resources or fails to meet one part of the system.

Look for the shared pattern

Diagnosis Prompt: Where does the confusion seem to be?

Bridge Prompt: Return to the check and describe the broad pattern first, then use coordinates only to support it.

Repair Choices

• I keep reading every coordinate separately and miss the shared driver.
  • A factor is useful because it explains several coordinates at once. Start with the common pattern before zooming into individual entries.
• I understand the pattern idea, but I cannot connect it back to the coordinates.
  • A factor does not erase the coordinates. It gives you a cleaner story for why several coordinates move together.
• I think factor language is just a fancy synonym for the biggest coordinate.
  • A dominant factor is a system-wide pattern, not the single largest number on the screen.

Separate feasible from merely attractive

Diagnosis Prompt: What confusion is closest to yours?

Bridge Prompt: Return to the choice and ask whether the candidate vector can actually be supported before you ask whether it sounds desirable.

Repair Choices

• I jump to the candidate that improves one coordinate most.
  • A candidate can improve one coordinate and still fail overall if the full system cannot support it. Feasibility comes before preference.
• I forget that limits can make a nice-sounding candidate impossible.
  • Constraints are not side notes. They define which vectors can exist at all, even before you rank them.
• I see the balance idea, but I do not use it to reject bad options.
  • Use balance like a filter. If a candidate breaks the balance rule, eliminate it before comparing softer tradeoffs.

An inverse is not entry-by-entry reciprocals

Diagnosis Prompt: What feels most likely?

Bridge Prompt: Go back and keep the whole map in view: the inverse undoes the transformation, not just each visible entry.

Repair Choices

• I treated the inverse like taking reciprocals of the matrix entries.
  • The inverse is another map whose composition with the original returns the identity. It is about undoing the whole transformation.
• I forgot that different inputs must stay distinguishable.
  • If distinct inputs collapse to one output, no inverse can recover which one came first.
• I know inverse means undo, but I could not connect that to uniqueness.
  • Undoing only works when every output comes from one input and every input leads to one recoverable output.

Best fit replaces exact solve when the model is too small

Diagnosis Prompt: What sounds most like the problem?

Bridge Prompt: Return to the check and ask what the best available fit is inside the chosen model, not whether the noisy data can be matched perfectly.

Repair Choices

• I kept insisting on an exact solution even when the data was noisy.
  • Least squares exists precisely because exact agreement is often impossible in real data.
• I forgot that the residual should be as small as possible in squared length.
  • The method compares fits by the size of the leftover error, not by arbitrary visual closeness.
• I could see the fit, but not why it was the best one.
  • The residual being orthogonal to the model space is the reason no better fit exists inside it.

Treat the matrix like a rule, not storage

Diagnosis Prompt: What sounds most like the blockage?

Bridge Prompt: Return to the check and describe what the matrix does to the input vector.

Repair Choices

• I still see the matrix as a table that stores numbers.
  • The matrix matters because it acts. It takes an input vector and transforms it into a new output vector.
• I lose track of why the output still needs interpretation.
  • The output is still a vector with labels. The map changes the vector, but it does not remove its meaning.
• I can follow the numbers, but not the role of the map.
  • Read the matrix as a reusable rule. It tells you how input coordinates are combined to make output coordinates.

Null space is hidden variation, not emptiness

Diagnosis Prompt: What sounds most like your mistake?

Bridge Prompt: Return to the check and think about which input changes the map cannot see in the output.

Repair Choices

• I read null space as meaning nothing happens anywhere.
  • Something can change in the input even when the visible output does not. Null space is invisible variation.
• I forgot that two different inputs can produce the same output.
  • When that happens, their difference lies in the null space.
• I mixed up null space with impossible outputs.
  • Impossible outputs belong to the story of column space. Null space is about hidden inputs, not missing outputs.

Keep the output attached to the vector

Diagnosis Prompt: What is most likely happening?

Bridge Prompt: Return to the check and read the output as a new vector created by the map.

Repair Choices

• Once multiplication happens, the output feels like a raw product only.
  • The output is still a labeled vector. Each coordinate still has a position and a meaning.
• I lose how the output relates back to the input vector.
  • The map transforms the input vector into a new one. Compare the coordinates before and after the map acts.
• I think the matrix replaces the vector, so I stop interpreting the result.
  • The matrix acts on a vector and produces another vector. The interpretation stays with the resulting state.

Elimination rewrites the same problem

Diagnosis Prompt: What sounds closest?

Bridge Prompt: Return to the check and treat pivots as a way to simplify the same system, not to invent a new one.

Repair Choices

• I thought each row operation changed the real answers.
  • Legal row operations preserve the solution set. They change the appearance of the equations so the structure is easier to read.
• I was chasing the final number and ignored the structure.
  • Elimination is first about exposing dependency and contradiction. The final numbers only make sense after the structure is clear.
• I lost track of which variable the pivot was isolating.
  • A pivot marks the direction currently being solved cleanly. Watch what becomes dependent on it and what stays free.

Projection is a best fit, not a crude cutoff

Diagnosis Prompt: What sounds closest?

Bridge Prompt: Return to the check and keep the geometry in view: projection keeps the closest point in the model space.

Repair Choices

• I treated projection like deleting coordinates.
  • Projection is chosen to minimize error under the given inner product, not to throw away coordinates arbitrarily.
• I forgot the residual must be orthogonal to the fit.
  • That orthogonality is what proves the chosen fit is best.
• I kept looking for an exact solution when the point was off the model line.
  • When exact fit is impossible, projection gives the closest useful approximation.

Rank and dimension answer different questions

Diagnosis Prompt: What sounds closest?

Bridge Prompt: Return to the choice and ask whether you are counting the size of a space, the visible outputs, or the hidden freedom.

Repair Choices

• I treated rank as just the number of rows or columns written down.
  • Rank counts independent directions that survive, not the raw size of the array.
• I mixed up the dimension of the input space with the rank of the map.
  • Input dimension counts available directions before the map acts. Rank counts independent output directions that remain visible.
• I knew rank and nullity were related, but not how.
  • They partition the input dimension: visible output capacity plus hidden freedom.

Use rank as a reach limit

Diagnosis Prompt: What is slipping?

Bridge Prompt: Return to the choice and interpret rank as the number of independent directions the system can still create.

Repair Choices

• Rank still feels like a technical count with no operational meaning.
  • Rank matters because it tells you how many independent adjustments the system can make, not just how many columns are printed.
• I forget that low rank means some targets stay unreachable.
  • If the system lacks independent directions, some desired states cannot be built no matter how hard you try.
• I mix up number of vectors with amount of reach.
  • More vectors do not always mean more reach. Redundant vectors can leave rank unchanged.

Do not stop at one step

Diagnosis Prompt: Which confusion sounds closest?

Bridge Prompt: Return to the check and compare what the first step does with what repeated application keeps doing afterwards.

Repair Choices

• I answer from the first output and ignore the later pattern.
  • One step can be misleading. Repeated updates reveal the direction the system keeps moving over time.
• I know the rule repeats, but I do not know what to watch for.
  • Watch whether the changes shrink, stay similar, or grow. That pattern tells you more than any single output alone.
• I treat each step like a separate story instead of one unfolding process.
  • The steps belong to one process. Read them as a sequence created by the same rule acting again and again.

Give rows and columns different jobs

Diagnosis Prompt: Which statement feels closest?

Bridge Prompt: Return to the check and say one row job and one column job out loud before choosing.

Repair Choices

• Rows and columns blur together for me.
  • Rows tell you how output coordinates are formed. Columns tell you how one input coordinate contributes across outputs.
• I can read rows, but columns feel abstract.
  • A column is an input-coordinate view. Read it as how one input coordinate contributes across multiple outputs.
• I can read columns, but rows feel abstract.
  • A row is an output-coordinate view. Read it as how one output coordinate is assembled from the input coordinates.

Think combination, not one lever

Diagnosis Prompt: What is the likely blockage?

Bridge Prompt: Return to the check and ask which state can be built by combining the available levers instead of expecting one lever to do everything.

Repair Choices

• I keep expecting one action vector to match the target by itself.
  • Span is about combination. A target can be reachable even if no single action matches it alone.
• I see several levers, but I do not know why adding them matters.
  • Adding levers matters because each one contributes a pattern of change. Together they can build states that none of them creates alone.
• I mix up ‘possible to imagine’ with ‘reachable from the given levers’.
  • Reachable means the system can actually build that state from the provided drivers, not just that the numbers look plausible.

Separate stability from raw size

Diagnosis Prompt: What is the likely mix-up?

Bridge Prompt: Return to the choice and ask whether the pattern is settling or amplifying, not just whether one coordinate is large.

Repair Choices

• I focus on whichever state has the biggest number.
  • A large number can still belong to a stabilizing process. Stability is about the repeated pattern, not only the absolute size.
• I see the repeated outputs, but I do not know how to judge convergence.
  • Look for whether the changes from one step to the next are shrinking and whether the pattern is settling into a steadier shape.
• I treat any movement as instability.
  • Movement alone is not instability. A system can move and still be converging toward a stable pattern.

A subspace needs closure, not a few good examples

Diagnosis Prompt: Which trap sounds familiar?

Bridge Prompt: Go back and test the rule itself: addition and scaling must keep you inside the set.

Repair Choices

• I saw two examples that worked and assumed the whole set was a subspace.
  • A subspace claim is structural. One or two examples do not prove closure.
• I checked addition but forgot scaling.
  • Linear structure requires both. If rescaling a valid vector breaks the set, it is not a subspace.
• I focused on whether the vectors looked useful instead of whether the set stayed stable.
  • Subspace questions are about the rule for membership, not whether the examples feel practical.

Compression should keep the mathematical story

Diagnosis Prompt: What sounds most likely?

Bridge Prompt: Return to the choice and keep only the summary that still preserves the mathematical story you need.

Repair Choices

• I think a compressed summary is allowed to drop the real meaning.
  • Compression is only useful when it keeps the core mathematical story intact. If the meaning disappears, the summary is too thin.
• I keep choosing the most detailed option because it feels safer.
  • More detail is not always better. A good summary keeps the decisive pattern without making the learner reread every coordinate.
• I can see the summary, but I do not know how it connects back to the main conclusion.
  • Ask whether the summary still tells you which pattern matters, where it appears, and what mathematical conclusion follows.

Make the vector one object again

Diagnosis Prompt: Which statement sounds closest to the gap right now?

Bridge Prompt: Go back and read the vector as one labeled object before choosing again.

Repair Choices

• I see three numbers, but not one shared vector.
  • Treat the vector like one object. Each coordinate is one labeled part of the same vector, not a separate problem.
• I know the labels, but I lose what comparison I should make.
  • Compare the same labeled coordinate across vectors. The labels let you compare \(x_1\) with \(x_1\) , \(x_2\) with \(x_2\) , and \(x_3\) with \(x_3\) .
• One low coordinate makes me think the whole vector behaves the same everywhere.
  • A single coordinate can fall while another rises. Read each labeled slot first, then describe the overall pattern across the full vector.

Final Exam Readiness

Summary: A stringent undergraduate linear-algebra readiness check. It is stricter than course completion and emphasizes interpretation, exact computation, proof/counterexample skill, and cumulative synthesis.

Candidate Profile: Use this after completing the course or equivalent prior study.

Time Limit Minutes: 135

Calculator Policy: Basic arithmetic calculator only on exact-computation sections. No CAS, graphing, symbolic algebra, or matrix automation tools.

Notes Policy: Closed notes and closed web. Formula sheets are not assumed; the benchmark expects recall of the standard undergraduate toolkit.

Retake Policy: After a serious attempt, wait at least 14 days and complete fresh review work before retaking.

Sections

Vectors And Linear Maps

Time Limit Minutes: 20

Focus: Interpret coordinates, rows, columns, and matrix maps with precise meaning; distinguish vector state, change, and map outputs without swapping roles.

Answer Modes:

• exact entry
• short explanation

Domains:

• interpretation
• structure

Systems, Subspaces, And Rank

Time Limit Minutes: 35

Focus: Solve and interpret linear systems, reason about span and null space, and connect pivot structure to rank and nullity.

Answer Modes:

• exact computation
• short explanation
• counterexample

Domains:

• computation
• structure
• interpretation

Orthogonality, Determinant, And Least Squares

Time Limit Minutes: 35

Focus: Work fluently with dot products, orthogonal projection, determinant meaning, and least-squares structure.

Answer Modes:

• exact computation
• short explanation

Domains:

• computation
• interpretation
• cumulative-synthesis

Eigen Structure And Proof

Time Limit Minutes: 25

Focus: Reason about eigenvectors, invariant directions, diagonalization conditions, and short proof or counterexample prompts.

Answer Modes:

• proof sketch
• counterexample
• short explanation

Domains:

• proof-counterexample
• structure

Cumulative Synthesis

Time Limit Minutes: 20

Focus: Solve one integrated problem that requires model reading, method choice, and written interpretation of the result.

Answer Modes:

• multi-step computation
• written interpretation

Domains:

• cumulative-synthesis
• interpretation
• computation

Domain Floors

Interpretation

Minimum Score Ratio: 0.7

Minimum Standard: Correctly assign meaning to vector entries, rows, columns, map outputs, and projections without role confusion or sign blindness.

Failure Rule: Falling below this floor blocks Ready or above, even if procedural scores are strong.

Computation

Minimum Score Ratio: 0.68

Minimum Standard: Carry out exact linear-algebra procedures reliably: elimination, basis coordinates, orthogonal projection, determinant work, and least-squares setup.

Failure Rule: Arithmetic noise is tolerated only when the method is otherwise correct; persistent procedure failure blocks Ready or above.

Proof / Counterexample

Minimum Score Ratio: 0.72

Minimum Standard: Produce short mathematically coherent justifications, recognize false universal statements, and give valid counterexamples when required.

Failure Rule: Weak proof or counterexample performance blocks Strong and A-Ready, even with strong raw totals elsewhere.

Structure

Minimum Score Ratio: 0.72

Minimum Standard: Connect pivots, span, basis, rank, nullity, eigen structure, and map properties as a linked system rather than isolated formulas.

Failure Rule: Shallow pattern matching without structural reasoning blocks Strong and A-Ready.

Cumulative Synthesis

Minimum Score Ratio: 0.78

Minimum Standard: Choose an appropriate method, carry it through, and explain the result in one coherent solution rather than disconnected steps.

Failure Rule: A weak synthesis task blocks A-Ready, even if the learner performs well on narrower items.

Readiness Bands

Not Ready

Minimum Total Score Ratio: 0.0

Description: Foundational gaps still block reliable undergraduate exam performance. More targeted repair is needed before a high-stakes attempt.

Ready

Minimum Total Score Ratio: 0.58

Description: Core undergraduate material is usable, but some domains still need tightening before a top-course grade is realistic.

Strong

Minimum Total Score Ratio: 0.76

Description: Performance is consistently solid across the syllabus with only limited remaining weakness. This is the first band that signals strong course readiness.

A-Ready

Minimum Total Score Ratio: 0.9

Description: Performance clears a stricter all-domain bar and looks compatible with top marks in a demanding undergraduate linear algebra course.

Anchor Problem Families

Row reduction with interpretation