Norm Landscapes Teacher Guide Vector Norm Landscapes
Teacher Facilitation Guide | Lesson 1 of 5
Graduate Level
Duration
90 Minutes
Format
Workshop / Skill-Building
Prerequisites
Basic Linear Algebra, R^2/R^3 intuition
Instructional Narrative
This lesson moves students from a purely Euclidean view of "size" (the standard hypotenuse) to a functional view of norms . In graduate-level data science and optimization, the choice of norm acts as a prior on the model. We aren't just measuring length; we are defining what it means for a vector to be "large" or "small" in a way that penalizes specific behaviors (e.g., non-sparsity in L1 vs. high variance in L2).
Learning Objectives
Define and compute the \(L_p\) norm for \(p \in \{1, 2, \infty, p\}\).
Visualize and interpret the geometric "unit ball" for different norms in \(\mathbb{R}^2\).
Explain how \(L_1\) vs \(L_2\) norms impact optimization (the "Lasso vs Ridge" intuition).
Connect the concept of distance (\(d(x,y) = ||x-y||\)) to the underlying norm.
The Hook: Manhattan Logic
"In Manhattan, you can't walk through buildings. To go from 1st and 1st to 2nd and 2nd, you travel 2 blocks, not \(\sqrt{2}\). In this world, a 'circle' is actually a diamond. If our data lives in a 'Manhattan' space, our standard Euclidean geometry fails us. Today, we redefine what 'straight' and 'short' mean."
Instructional Sequence
15
The Axiomatic Approach (Slides 1-5)
Define a norm axiomatically: Non-negativity, Scalability, and Triangle Inequality. Ask: "Why is the Triangle Inequality the most important part of being a 'norm'?"
Key Formula: ||x||_p = (\sum |x_i|^p)^{1/p}
30
Unit Ball Workshop (Worksheet Activity)
Students sketch unit balls for \(L_1\), \(L_2\), and \(L_\infty\). Use the prompt: "If every vector on this line has length 1, what does that tell us about the geometry of the space?"
Teacher Note: Look for students realizing that \(L_\infty\) is a square and \(L_1\) is a diamond. This is the 'aha' moment.
25
Application: Regularization Landscapes
Show how the 'pointy' corners of the \(L_1\) diamond lead to sparse solutions (Lasso) in regression. Contrast with the 'smooth' \(L_2\) circle (Ridge).
20
Group Debrief & Exit Ticket
Discussion: "In what real-world data scenario would measuring distance with \(L_1\) be more honest than \(L_2\)? (e.g., circuit board wiring, city blocks, categorical differences)."
Common Misconceptions
Students thinking that \(L_0\) is a true norm (it violates scalability). Clarify it's a "pseudo-norm".
Confusion that \(L_\infty\) "ignores" other components. It only cares about the maximum, but it still defines a valid metric.
Pro Tips
"Encourage students to use the term 'Unit Ball' early and often. It helps them think of vectors as points on a boundary rather than just arrows from the origin."
Norm Space Slides Vector Quantities I
NORM
LANDSCAPES
Redefining magnitude and distance in high-dimensional data spaces.
The Euclidean Habit
We are hard-wired to perceive distance as "as the crow flies."
\[ ||x||_2 = \sqrt{\sum x_i^2} \]
The $L_2$ norm is rotationally invariant and smooth. But data isn't always physical.
r = 1
The Lp Family
\[ ||x||_p = \left( \sum_{i=1}^n |x_i|^p \right)^{1/p} \]
p = 1
Manhattan / Taxicab
p = 2
Euclidean
p = \(\infty\)
Chebyshev / Max
Why does the shape matter?
LASSO (L1)
Hits the axes at the "corners," setting some coefficients to exactly zero.
RIDGE (L2)
Shrinks coefficients uniformly but rarely hits zero.
"Is a vector's size a property of the vector, or a property of the observer?"
Think about signal noise vs. physical displacement.
Discussion: 5 minutes
Unit Ball Workshop Unit Ball Workshop
Vector Modeling | Lesson 1 Activity
Name: ________________________________
Date: ________________________________
1 Norm Definitions
Consider the vector \(\vec{v} = [3, -4, 0, 12]^T\). Calculate the following norms:
L1 Norm (||v||_1):
L2 Norm (||v||_2):
L-infinity Norm (||v||_inf):
2 Mapping the Unit Ball in \(\mathbb{R}^2\)
The Unit Ball \(B = \{x \in \mathbb{R}^2 : ||x|| \leq 1\}\). On the grids below, sketch the boundary of the unit ball for each specified norm.
L1 (Manhattan)
x1
L2 (Euclidean)
L-infinity (Max)
The Curse of Small P
Observe what happens to the shape of the unit ball as \(p \to 0\). (Note: This is technically a "quasi-norm").
1. As p decreases from 2 to 1 and then to 0.5, do the unit balls become more convex or more concave?
2. Why does the "pointiness" of the L1 norm at the axes lead to sparse solutions in regression models? Explain the geometric intuition of a constraint line intersecting the L1 ball vs the L2 ball.
Computational Vector Series: Module 1.1 | Developed for Graduate Advanced Analytics
Similarity Metrics Teacher Guide Directional Intelligence
Teacher Facilitation Guide | Lesson 2 of 5
Graduate Level
Time
90 Min
Method
Case Study / Discussion
Concept
Cosine Similarity
Instructional Narrative
In this lesson, we shift from measuring "how much" (magnitude) to "where" (direction). For many high-dimensional datasets—particularly Natural Language Processing (NLP)—magnitude is often a nuisance variable (e.g., document length). Cosine similarity allows us to project vectors onto the unit hypersphere and measure semantic closeness by the angle between them.
Learning Objectives
Mathematically derive cosine similarity from the dot product formula.
Contrast Euclidean distance and Cosine similarity in a 2D text-vector example.
Explain why orthogonality implies semantic independence in high dimensions.
Interpret a similarity matrix for a multi-document dataset.
The Case Study
The "Length" Trap
"Document A is a 5-sentence summary of Quantum Mechanics. Document B is a 500-page textbook on the same topic. Their Euclidean distance is massive. Their Cosine similarity is nearly 1. Which metric 'understands' the content better?"
Instructional Sequence
0-20
The Angular View (Slides 1-4)
Review the Dot Product. Pivot to \(\cos(\theta) = \frac{A \cdot B}{||A|| ||B||}\). Emphasize that this is equivalent to the dot product of normalized vectors.
20-50
Lab: The Term-Document Matrix
Students work through the "Document Similarity Lab". They compute similarities between vectors representing word counts. Challenge: "Find two documents that are semantically similar but distant in Euclidean space."
50-75
Orthogonality as Uniqueness
Discussion: In a 10,000-word vocabulary, most document pairs have a similarity of 0. Why? (Sparsity and Orthogonality). Introduce the idea of "nearly orthogonal" random vectors.
75-90
Synthesis & Wrap-up
"If cosine similarity is 1, are the vectors identical?" (No, they are collinear). "If cosine similarity is -1, what does that mean in a text context?" (This is rare with counts, but common in embeddings like Word2Vec).
Discussion Prompts
Metric Choice
"When would you prefer Euclidean distance over Cosine similarity for data vectors? Hint: Think about whether the 'magnitude' represents intensity (e.g., sensor data) or frequency (e.g., text)."
Semantic Vector Slides Vector Quantities II
SEMANTIC
DIRECTION
"It's not how far you go, but which way you point."
The Magnitude Nuisance
In many datasets, Magnitude is irrelevant to Class .
Example: Text Analysis
Document A: "Physics is great." (3 words)
Euclidean Distance is high, but the information is identical.
A
B
Large Euclidean Distance
Cosine Similarity
\[ \text{sim}(A, B) = \cos(\theta) = \frac{\mathbf{A} \cdot \mathbf{B}}{||\mathbf{A}|| ||\mathbf{B}||} \]
1.0
Same Direction
0.0
Orthogonal
-1.0
Opposite
The Unit Sphere View
Normalizing a vector removes magnitude bias. When vectors have unit length (\(||x||=1\)), the dot product and cosine similarity are identical.
High-dimensional similarity search is essentially finding points close together on a hypersphere.
\(\theta\)
The "90-Degree" Rule
In high-dimensional sparse spaces (like text), most vectors are orthogonal to each other. They share no common terms.
Orthogonality = Semantic Disconnection
Document Vector Lab Document Similarity Lab
Vector Modeling | Lesson 2 Activity
ID: VEC-202-SIM
Name: ________________________________
Objective
Using a simplified Term-Document Matrix, calculate the Euclidean Distance and Cosine Similarity between three documents to observe how magnitude biases results.
1 The Term-Document Matrix
Document Term: "Algorithm" Term: "Complexity" Magnitude (L2) Doc A (Snippet) 1 1 ? Doc B (Textbook) 10 10 ? Doc C (Review) 1 10 ?
Step A: Calculate L2 Magnitudes
Compute \(||Doc||_2\) for all three documents. Fill in the table above.
Step B: Visualization
Briefly sketch these three vectors in 2D space on a mental or margin grid. Which two look "aligned"?
2 Distance vs. Similarity
Compare Doc A and Doc B
Euclidean Distance d(A,B):
Cosine Similarity sim(A,B):
Compare Doc A and Doc C
Euclidean Distance d(A,C):
Cosine Similarity sim(A,C):
Interpretation
1. Which pair of documents (A,B or A,C) is more semantically similar? Does the Euclidean metric agree with the Cosine metric?
2. If you were building a search engine, why might a high cosine similarity be more useful than a low Euclidean distance for finding relevant documents?
[ENCODING SEMANTIC SPACE :: 02.B]
Dimensionality Curse Teacher Guide The Curse of Dimensionality
Teacher Facilitation Guide | Lesson 3 of 5
Conceptual Deep Dive
Time
90 Min
Method
Simulation / Inquiry
Level
Advanced Theory
Instructional Narrative
In low dimensions (2D, 3D), our geometric intuition is reliable. In high dimensions (\(\mathbb{R}^{1000}\)), geometry becomes alien . The "Curse of Dimensionality" refers to the fact that volume concentration, distance distributions, and vector orientations behave in ways that break traditional machine learning algorithms (like K-Nearest Neighbors). This lesson is about building a new, non-spatial intuition for "data space."
Learning Objectives
Explain why volume in a hypersphere concentrates in the "shell" (the thin skin paradox).
Analyze why random vectors in high dimensions tend to be nearly orthogonal.
Evaluate the failure of distance-based clustering as \(n \to \infty\).
Differentiate between "local" and "global" properties in high-dimensional manifolds.
The Hook: The Hyperspace Orange
"Imagine an orange with a thin peel. In 3D, most of the orange is fruit. In 100 dimensions, if you peel a 1cm layer off a 10cm orange, you have removed over 99% of the volume. In hyperspace, there is no interior—only the shell. What does this mean for 'representative' data samples?"
Instructional Sequence
0-20
The Shell Paradox (Slides 1-3)
Walk through the math of volume: \(V(r) \propto r^n\). Show that \(\frac{V(0.9r)}{V(r)} = 0.9^n\). As \(n\) grows, this ratio goes to 0. All the mass is at the boundary.
20-55
Inquiry: Distance Distributions
Students analyze the "Orange Peel Simulation" worksheet. They interpret histograms showing the ratio of \(\frac{d_{max} - d_{min}}{d_{min}}\). Conclusion: In high dimensions, every point is equally far from every other point. "Nearest Neighbor" becomes a meaningless concept.
55-80
The Spiky Sphere (Slides 4-6)
Visualization of the "spikiness" of hyperspheres. Discuss why random vectors are nearly orthogonal. It’s not that they are special; it’s that the "equator" of the hypersphere contains almost all its surface area from any perspective.
80-90
Exit Reflection
"If distance fails us, how do we find structure in high-dimensional data?" (Tease Lesson 5: Principal Components and Manifolds).
Facilitation Notes
Visualization Tip
"Stop trying to draw 4D. Instead, draw 2D/3D and describe the mathematical trend as \(n \to \infty\). Focus on the
Hyperspace Paradox Slides Vector Quantities III
HYPERSPACE
PARADOXES
Where intuition goes to die.
The "Thin Skin" Effect
In \(\mathbb{R}^n\), volume is an exponential function of radius. As \(n\) increases, almost all volume of a sphere lies in an arbitrarily thin outer shell.
\[ \frac{V(r - \epsilon)}{V(r)} = \left( 1 - \frac{\epsilon}{r} \right)^n \]
As \(n \to \infty\), this ratio \(\to 0\).
99% VOLUME
The "Fruit" is empty. The "Peel" is everything.
The End of "Nearest"
The Observation
In high dimensions, the difference between the minimum and maximum distance between points becomes negligible relative to the minimum distance.
\[ \lim_{n \to \infty} \frac{d_{max} - d_{min}}{d_{min}} = 0 \]
Algorithm Failure: KNN
Algorithm Failure: K-Means
Conclusion: All points are outliers.
The Spiky Sphere
If you pick two random vectors in high dimensions, they will almost certainly be orthogonal .
"The Equator Effect"
From the perspective of any vector, almost all of the hypersphere's surface area is concentrated around its 'equator'.
North Pole
South Pole
99.9% of mass
Survival Strategy
How do we solve the Curse?
1. Manifold Assumption
High-dimensional data actually lives on a low-dimensional sub-structure.
2. Dimensionality Reduction
Projecting down to "essential" dimensions using SVD/PCA.
Orange Peel Simulation Activity Orange Peel Simulation
Vector Modeling | Lesson 3 Inquiry
REF: DIM-303-CURSE
Name: ________________________________
1 Calculating the Inner Void
Let a hypersphere have radius \(r=1\). We define the "interior" as the region with radius \(r=0.9\) (the inner 90%). Let \(V_n(r)\) be the volume in \(n\) dimensions.
Ratio = \(\frac{Volume(0.9)}{Volume(1.0)} = (0.9)^n\)
n=2 (Circle)
81.0%
n=3 (Sphere)
72.9%
n=10
34.8%
n=100
0.002%
Q: If you are sampling data points uniformly from a 100-dimensional hypersphere, what is the probability that a point falls in the "center" (the inner 90% of the radius)?
2 Interpreting Distance Distributions
Pairwise Euclidean distances between 1,000 random points:
R^2 (Low Dimensions)
Distance spectrum is wide. Some points are close.
R^500 (High Dimensions)
Distance spectrum is extremely narrow. All points are "equally far."
Analysis Task:
Consider a Nearest Neighbor algorithm in \(\mathbb{R}^{500}\). If the distance to the nearest neighbor is 12.4 and the distance to the farthest point is 12.6, does "nearest" carry any meaningful discriminatory power for classification? Why or why not?
Finding the Equator
Explain why the concentration of measure in high-dimensional spheres implies that two random vectors will likely have a dot product near zero. Use the concept of the "hypersphere equator" in your explanation.
Module 3: Stochastic Geometry Dimensionality Inquiry V3.1
Orthogonal Basis Teacher Guide Orthonormal Foundations
Teacher Facilitation Guide | Lesson 4 of 5
Skill Building
Time
90 Min
Concept
Projections & GS
Application
Signal Decomposition
Instructional Narrative
This lesson transitions students from "seeing" vectors to "reconstructing" them. We define a basis not just as a set of spanning vectors, but as a coordinate system. The Gram-Schmidt process is the algorithmic tool that turns any messy set of variables into a clean, orthogonal coordinate system, which is the bedrock of numerical stability in computing.
Learning Objectives
Mathematically define the projection of a vector \(\vec{v}\) onto a subspace \(\mathcal{W}\).
Execute the Gram-Schmidt process manually for small sets in \(\mathbb{R}^3\).
Explain why orthonormal bases are preferred for numerical computation (stability).
Apply projections to "de-noise" a signal by removing components along undesired axes.
The Hook: Noise Cancellation
"If you're recording a voice in a room with a hum, how do you remove the hum without losing the voice? You treat the hum as a vector direction. If you can project the audio vector onto the 'Not-Hum' subspace, you've cleaned the data. Gram-Schmidt is the engine that finds that subspace."
Instructional Sequence
20
The Geometry of Shadow (Slides 1-3)
Visualizing \(\text{proj}_u(v) = \frac{v \cdot u}{u \cdot u} u\). Ask: "Why do we divide by \(u \cdot u\)? (Normalization)". Show the shadow analogy.
40
Gram-Schmidt Workshop
Students follow the "Signal Decomposer" worksheet. Walk through the iterative subtraction: \(e_2 = v_2 - \text{proj}_{e_1}(v_2)\). Point out the recursive nature of the algorithm.
Watch for: Calculation errors in the second subtraction. It's the most common failure point.
30
Orthonormality & PCA Teaser
Introduce Orthonormality (\(Q^T Q = I\)). Discuss why rotation matrices and orthonormal bases 'preserve' length and angle, making them the gold standard for data transformation.
Common Pitfalls
Forgetting to normalize at the end (Orthogonal vs Orthonormal).
Confusion about 'reusing' the previous vectors in the calculation of the next.
Graduate Extension
Mention the Modified Gram-Schmidt (MGS) algorithm. Why is it more numerically stable than the classical version in floating-point arithmetic? (A teaser for Numerical Analysis/SVD).
Subspace Projection Slides Vector Quantities IV
SUBSPACE
PROJECTION
Orthogonalizing the Information Foundation
The Projection Shadow
Finding the best approximation of vector \(\mathbf{v}\) in direction \(\mathbf{u}\).
\[ \text{proj}_\mathbf{u}(\mathbf{v}) = \frac{\mathbf{v} \cdot \mathbf{u}}{\mathbf{u} \cdot \mathbf{u}} \mathbf{u} \]
The "error" vector (\(\mathbf{v} - \hat{\mathbf{v}}\)) is always orthogonal to the subspace.
v
proj(v)
Gram-Schmidt Workflow
Step 1: Anchor
\(u_1 = v_1\)
Pick an initial direction to define the axis.
Step 2: Subtract
\(u_2 = v_2 - \text{proj}_{u_1}(v_2)\)
Remove everything from \(v_2\) that points in direction \(u_1\).
Step 3: Repeat
\(u_3 = v_3 - \text{proj}_{u_1}(v_3) - \dots\)
Ensure the new vector is orthogonal to all previous.
Computational Gold
Numerical Stability
Orthogonal matrices (\(Q^T Q = I\)) do not amplify noise. They preserve Euclidean distance and norm. They are Isometries .
Inversion for Free
If Q is orthogonal, \(Q^{-1} = Q^T\).
STABLE
Wait... why are we doing this?
Because any dataset can be decomposed into an Orthonormal Basis (Q) and an Upper Triangular Matrix (R). This is the QR Decomposition .
A = QR
Signal Decomposer Worksheet Signal Decomposer
Vector Modeling | Lesson 4 Workshop
LOG: GS-404-ORTHO
Name: ________________________________
"You are given two noisy data signals as vectors. To extract independent features, you must construct an orthonormal basis from them using the Gram-Schmidt process."
1 The Projection Core
Let \(\vec{v} = [4, 3]^T\) and \(\vec{u} = [1, 0]^T\).
A. Compute the projection of \(\vec{v}\) onto \(\vec{u}\).
Result: [ ___, ___ ]
B. Compute the error vector \(\vec{w} = \vec{v} - \text{proj}_u(v)\).
Result: [ ___, ___ ]
Verification: Is \(\vec{w} \cdot \vec{u} = 0\)? (Yes/No)
2 Full Orthogonalization
Given the basis \(\{\vec{v_1}, \vec{v_2}\}\) in \(\mathbb{R}^3\):
v1 = [1, 1, 0]
v2 = [1, 0, 1]
Step I: First Basis Vector
Assign \(\vec{u_1} = \vec{v_1}\). Compute the norm \(||\vec{u_1}||\).
Step II: Second Basis Vector
Compute \(\vec{u_2} = \vec{v_2} - \frac{\vec{v_2} \cdot \vec{u_1}}{\vec{u_1} \cdot \vec{u_1}} \vec{u_1}\).
Step III: Normalization
Convert the orthogonal basis \(\{\vec{u_1}, \vec{u_2}\}\) into an orthonormal basis \(\{\vec{q_1}, \vec{q_2}\}\).
Structural Integrity
Why is it mathematically impossible for the Gram-Schmidt process to produce an orthogonal basis if the input vectors were linearly dependent?
Gram-Schmidt Signal Reconstruction Core Alpha
PCA Strategy Teacher Guide Principal Vector Directions
Teacher Facilitation Guide | Lesson 5 of 5
Capstone Lesson
Duration
120 Min
Method
Project-Based
Key Tool
PCA
Instructional Narrative
In this capstone, we synthesize the sequence by viewing matrices as operators . An eigenvector is a direction in high-dimensional space that remains invariant under a linear transformation—only its magnitude (the eigenvalue) changes. In the context of data (the Covariance Matrix), these directions represent the principal axes of variance . Solving the Curse of Dimensionality means finding these few "active" directions and ignoring the rest.
Learning Objectives
Identify eigenvectors as invariant directions of a linear transformation \(Ax = \lambda x\).
Construct a Covariance Matrix from a raw data vector set.
Perform Principal Component Analysis (PCA) to extract the top \(k\) directions of variance.
Reconstruct high-dimensional data from a low-dimensional projection (lossy compression).
The Hook: Eigenfaces
"A digital image of a face is a vector with 10,000 pixels (dimensions). But faces aren't random. They share structure: eyes, noses, mouth. If we can find the 'eigenvectors' of the human face, we can describe any person as just a combination of 50 'ghostly' base images. This is the power of vector quantities in the wild."
Instructional Sequence
25
The Invariant Direction (Slides 1-3)
Dynamic visualization: Rotate a vector by a matrix and see which one doesn't change angle. Solve the characteristic equation \(\det(A - \lambda I) = 0\). Emphasize: "Eigenvectors are the 'spine' of the transformation."
60
The Compression Project
Students work through the "Face Compression Project" worksheet. They interpret eigenvalue magnitudes as "percentage of information preserved." Challenge: "How many dimensions can we drop before the person is unrecognizable?"
35
Final Discussion: The Latent Space
Discussion: Connecting eigenvectors to Lesson 2 (Cosine Similarity) and Lesson 3 (The Curse). Conclusion: In high-dimensional data, most dimensions are noise; the eigenvectors are the truth.
Synthesis Prompt
"Students should leave with one core realization: A vector in a computer isn't an arrow in space; it's a coordinate in a semantic landscape. Our job is to find the most meaningful map of that landscape."
Invariant Direction Slides Vector Quantities V
INVARIANT
DIRECTIONS
Mastering the principal axes of high-dimensional variance.
Matrix as Operator
Think of a matrix not as a box of numbers, but as a Force Field that stretches and rotates space.
For most vectors \(\mathbf{x}\):
\(A\mathbf{x}\) points in a completely new direction.
x
Ax (Transformed)
The Special Directions
Some vectors are special. When you transform them, they stay on their own line. They only scale .
\[ \mathbf{A}\mathbf{x} = \lambda \mathbf{x} \]
x
The Eigenvector
\(\lambda\)
The Eigenvalue
Principal Component Analysis
PCA is the process of finding the eigenvectors of the Covariance Matrix of your data.
PC1
PC2
"Structure over Space"
The Curse of Dimensionality is only a curse if you assume all dimensions matter equally. Eigenvectors tell us which ones actually speak .
Face Compression Project Worksheet Face Compression Project
Vector Modeling | Lesson 5 Capstone
ID: CAPSTONE-505-PCA
Name: ________________________________
"Imagine a dataset of 1,000 face images. Each face is a vector in \(\mathbb{R}^{10,000}\). By applying PCA, we can compress these vectors into a 50-dimensional latent space. You will analyze the results of this compression."
1 Variance Explained
Below is a Scree Plot showing the eigenvalues (\(\lambda_i\)) of the covariance matrix, ordered by magnitude. The eigenvalue represents the amount of variance captured by its corresponding eigenvector.
Interpretation Task:
If the sum of all eigenvalues \(\sum \lambda_i = 100\), and the first three eigenvalues are \(\lambda_1 = 60, \lambda_2 = 15, \lambda_3 = 5\)...
What percentage of the total dataset variance is captured by the first TWO principal components?
2 Lossy Projection
Let \(\mathbf{Q_k}\) be a matrix whose columns are the top \(k\) eigenvectors. The compressed vector is \(\mathbf{y} = \mathbf{Q_k}^T \mathbf{x}\). The reconstructed face is \(\hat{\mathbf{x}} = \mathbf{Q_k} \mathbf{y}\).
A. Why is \(\hat{\mathbf{x}}\) not identical to \(\mathbf{x}\)?
B. In terms of "The Curse of Dimensionality," how does this projection solve the distance-indistinguishability problem?
Course Synthesis
You are clustering document vectors using Lesson 2's Cosine Similarity , but the documents have 50,000 dimensions (vocabulary size). Describe a pipeline that uses PCA (Lesson 5) to improve your clustering results. How would the Norms from Lesson 1 play a role in pre-processing?
Dimensionality Masters Program Final Report v5.5 End Sequence
